Loading...
1/*
2 * Generic address resolution entity
3 *
4 * Authors:
5 * Pedro Roque <roque@di.fc.ul.pt>
6 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
7 *
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
12 *
13 * Fixes:
14 * Vitaly E. Lavrov releasing NULL neighbor in neigh_add.
15 * Harald Welte Add neighbour cache statistics like rtstat
16 */
17
18#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
19
20#include <linux/slab.h>
21#include <linux/types.h>
22#include <linux/kernel.h>
23#include <linux/module.h>
24#include <linux/socket.h>
25#include <linux/netdevice.h>
26#include <linux/proc_fs.h>
27#ifdef CONFIG_SYSCTL
28#include <linux/sysctl.h>
29#endif
30#include <linux/times.h>
31#include <net/net_namespace.h>
32#include <net/neighbour.h>
33#include <net/dst.h>
34#include <net/sock.h>
35#include <net/netevent.h>
36#include <net/netlink.h>
37#include <linux/rtnetlink.h>
38#include <linux/random.h>
39#include <linux/string.h>
40#include <linux/log2.h>
41#include <linux/inetdevice.h>
42#include <net/addrconf.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(unsigned long arg);
55static void __neigh_notify(struct neighbour *n, int type, int flags);
56static void neigh_update_notify(struct neighbour *neigh);
57static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev);
58
59#ifdef CONFIG_PROC_FS
60static const struct file_operations neigh_stat_seq_fops;
61#endif
62
63/*
64 Neighbour hash table buckets are protected with rwlock tbl->lock.
65
66 - All the scans/updates to hash buckets MUST be made under this lock.
67 - NOTHING clever should be made under this lock: no callbacks
68 to protocol backends, no attempts to send something to network.
69 It will result in deadlocks, if backend/driver wants to use neighbour
70 cache.
71 - If the entry requires some non-trivial actions, increase
72 its reference count and release table lock.
73
74 Neighbour entries are protected:
75 - with reference count.
76 - with rwlock neigh->lock
77
78 Reference count prevents destruction.
79
80 neigh->lock mainly serializes ll address data and its validity state.
81 However, the same lock is used to protect another entry fields:
82 - timer
83 - resolution queue
84
85 Again, nothing clever shall be made under neigh->lock,
86 the most complicated procedure, which we allow is dev->hard_header.
87 It is supposed, that dev->hard_header is simplistic and does
88 not make callbacks to neighbour tables.
89 */
90
91static int neigh_blackhole(struct neighbour *neigh, struct sk_buff *skb)
92{
93 kfree_skb(skb);
94 return -ENETDOWN;
95}
96
97static void neigh_cleanup_and_release(struct neighbour *neigh)
98{
99 if (neigh->parms->neigh_cleanup)
100 neigh->parms->neigh_cleanup(neigh);
101
102 __neigh_notify(neigh, RTM_DELNEIGH, 0);
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
118
119static int neigh_forced_gc(struct neigh_table *tbl)
120{
121 int shrunk = 0;
122 int i;
123 struct neigh_hash_table *nht;
124
125 NEIGH_CACHE_STAT_INC(tbl, forced_gc_runs);
126
127 write_lock_bh(&tbl->lock);
128 nht = rcu_dereference_protected(tbl->nht,
129 lockdep_is_held(&tbl->lock));
130 for (i = 0; i < (1 << nht->hash_shift); i++) {
131 struct neighbour *n;
132 struct neighbour __rcu **np;
133
134 np = &nht->hash_buckets[i];
135 while ((n = rcu_dereference_protected(*np,
136 lockdep_is_held(&tbl->lock))) != NULL) {
137 /* Neighbour record may be discarded if:
138 * - nobody refers to it.
139 * - it is not permanent
140 */
141 write_lock(&n->lock);
142 if (atomic_read(&n->refcnt) == 1 &&
143 !(n->nud_state & NUD_PERMANENT)) {
144 rcu_assign_pointer(*np,
145 rcu_dereference_protected(n->next,
146 lockdep_is_held(&tbl->lock)));
147 n->dead = 1;
148 shrunk = 1;
149 write_unlock(&n->lock);
150 neigh_cleanup_and_release(n);
151 continue;
152 }
153 write_unlock(&n->lock);
154 np = &n->next;
155 }
156 }
157
158 tbl->last_flush = jiffies;
159
160 write_unlock_bh(&tbl->lock);
161
162 return shrunk;
163}
164
165static void neigh_add_timer(struct neighbour *n, unsigned long when)
166{
167 neigh_hold(n);
168 if (unlikely(mod_timer(&n->timer, when))) {
169 printk("NEIGH: BUG, double timer add, state is %x\n",
170 n->nud_state);
171 dump_stack();
172 }
173}
174
175static int neigh_del_timer(struct neighbour *n)
176{
177 if ((n->nud_state & NUD_IN_TIMER) &&
178 del_timer(&n->timer)) {
179 neigh_release(n);
180 return 1;
181 }
182 return 0;
183}
184
185static void pneigh_queue_purge(struct sk_buff_head *list)
186{
187 struct sk_buff *skb;
188
189 while ((skb = skb_dequeue(list)) != NULL) {
190 dev_put(skb->dev);
191 kfree_skb(skb);
192 }
193}
194
195static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev)
196{
197 int i;
198 struct neigh_hash_table *nht;
199
200 nht = rcu_dereference_protected(tbl->nht,
201 lockdep_is_held(&tbl->lock));
202
203 for (i = 0; i < (1 << nht->hash_shift); i++) {
204 struct neighbour *n;
205 struct neighbour __rcu **np = &nht->hash_buckets[i];
206
207 while ((n = rcu_dereference_protected(*np,
208 lockdep_is_held(&tbl->lock))) != NULL) {
209 if (dev && n->dev != dev) {
210 np = &n->next;
211 continue;
212 }
213 rcu_assign_pointer(*np,
214 rcu_dereference_protected(n->next,
215 lockdep_is_held(&tbl->lock)));
216 write_lock(&n->lock);
217 neigh_del_timer(n);
218 n->dead = 1;
219
220 if (atomic_read(&n->refcnt) != 1) {
221 /* The most unpleasant situation.
222 We must destroy neighbour entry,
223 but someone still uses it.
224
225 The destroy will be delayed until
226 the last user releases us, but
227 we must kill timers etc. and move
228 it to safe state.
229 */
230 __skb_queue_purge(&n->arp_queue);
231 n->arp_queue_len_bytes = 0;
232 n->output = neigh_blackhole;
233 if (n->nud_state & NUD_VALID)
234 n->nud_state = NUD_NOARP;
235 else
236 n->nud_state = NUD_NONE;
237 neigh_dbg(2, "neigh %p is stray\n", n);
238 }
239 write_unlock(&n->lock);
240 neigh_cleanup_and_release(n);
241 }
242 }
243}
244
245void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev)
246{
247 write_lock_bh(&tbl->lock);
248 neigh_flush_dev(tbl, dev);
249 write_unlock_bh(&tbl->lock);
250}
251EXPORT_SYMBOL(neigh_changeaddr);
252
253int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
254{
255 write_lock_bh(&tbl->lock);
256 neigh_flush_dev(tbl, dev);
257 pneigh_ifdown(tbl, dev);
258 write_unlock_bh(&tbl->lock);
259
260 del_timer_sync(&tbl->proxy_timer);
261 pneigh_queue_purge(&tbl->proxy_queue);
262 return 0;
263}
264EXPORT_SYMBOL(neigh_ifdown);
265
266static struct neighbour *neigh_alloc(struct neigh_table *tbl, struct net_device *dev)
267{
268 struct neighbour *n = NULL;
269 unsigned long now = jiffies;
270 int entries;
271
272 entries = atomic_inc_return(&tbl->entries) - 1;
273 if (entries >= tbl->gc_thresh3 ||
274 (entries >= tbl->gc_thresh2 &&
275 time_after(now, tbl->last_flush + 5 * HZ))) {
276 if (!neigh_forced_gc(tbl) &&
277 entries >= tbl->gc_thresh3) {
278 net_info_ratelimited("%s: neighbor table overflow!\n",
279 tbl->id);
280 NEIGH_CACHE_STAT_INC(tbl, table_fulls);
281 goto out_entries;
282 }
283 }
284
285 n = kzalloc(tbl->entry_size + dev->neigh_priv_len, GFP_ATOMIC);
286 if (!n)
287 goto out_entries;
288
289 __skb_queue_head_init(&n->arp_queue);
290 rwlock_init(&n->lock);
291 seqlock_init(&n->ha_lock);
292 n->updated = n->used = now;
293 n->nud_state = NUD_NONE;
294 n->output = neigh_blackhole;
295 seqlock_init(&n->hh.hh_lock);
296 n->parms = neigh_parms_clone(&tbl->parms);
297 setup_timer(&n->timer, neigh_timer_handler, (unsigned long)n);
298
299 NEIGH_CACHE_STAT_INC(tbl, allocs);
300 n->tbl = tbl;
301 atomic_set(&n->refcnt, 1);
302 n->dead = 1;
303out:
304 return n;
305
306out_entries:
307 atomic_dec(&tbl->entries);
308 goto out;
309}
310
311static void neigh_get_hash_rnd(u32 *x)
312{
313 get_random_bytes(x, sizeof(*x));
314 *x |= 1;
315}
316
317static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift)
318{
319 size_t size = (1 << shift) * sizeof(struct neighbour *);
320 struct neigh_hash_table *ret;
321 struct neighbour __rcu **buckets;
322 int i;
323
324 ret = kmalloc(sizeof(*ret), GFP_ATOMIC);
325 if (!ret)
326 return NULL;
327 if (size <= PAGE_SIZE)
328 buckets = kzalloc(size, GFP_ATOMIC);
329 else
330 buckets = (struct neighbour __rcu **)
331 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
332 get_order(size));
333 if (!buckets) {
334 kfree(ret);
335 return NULL;
336 }
337 ret->hash_buckets = buckets;
338 ret->hash_shift = shift;
339 for (i = 0; i < NEIGH_NUM_HASH_RND; i++)
340 neigh_get_hash_rnd(&ret->hash_rnd[i]);
341 return ret;
342}
343
344static void neigh_hash_free_rcu(struct rcu_head *head)
345{
346 struct neigh_hash_table *nht = container_of(head,
347 struct neigh_hash_table,
348 rcu);
349 size_t size = (1 << nht->hash_shift) * sizeof(struct neighbour *);
350 struct neighbour __rcu **buckets = nht->hash_buckets;
351
352 if (size <= PAGE_SIZE)
353 kfree(buckets);
354 else
355 free_pages((unsigned long)buckets, get_order(size));
356 kfree(nht);
357}
358
359static struct neigh_hash_table *neigh_hash_grow(struct neigh_table *tbl,
360 unsigned long new_shift)
361{
362 unsigned int i, hash;
363 struct neigh_hash_table *new_nht, *old_nht;
364
365 NEIGH_CACHE_STAT_INC(tbl, hash_grows);
366
367 old_nht = rcu_dereference_protected(tbl->nht,
368 lockdep_is_held(&tbl->lock));
369 new_nht = neigh_hash_alloc(new_shift);
370 if (!new_nht)
371 return old_nht;
372
373 for (i = 0; i < (1 << old_nht->hash_shift); i++) {
374 struct neighbour *n, *next;
375
376 for (n = rcu_dereference_protected(old_nht->hash_buckets[i],
377 lockdep_is_held(&tbl->lock));
378 n != NULL;
379 n = next) {
380 hash = tbl->hash(n->primary_key, n->dev,
381 new_nht->hash_rnd);
382
383 hash >>= (32 - new_nht->hash_shift);
384 next = rcu_dereference_protected(n->next,
385 lockdep_is_held(&tbl->lock));
386
387 rcu_assign_pointer(n->next,
388 rcu_dereference_protected(
389 new_nht->hash_buckets[hash],
390 lockdep_is_held(&tbl->lock)));
391 rcu_assign_pointer(new_nht->hash_buckets[hash], n);
392 }
393 }
394
395 rcu_assign_pointer(tbl->nht, new_nht);
396 call_rcu(&old_nht->rcu, neigh_hash_free_rcu);
397 return new_nht;
398}
399
400struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
401 struct net_device *dev)
402{
403 struct neighbour *n;
404
405 NEIGH_CACHE_STAT_INC(tbl, lookups);
406
407 rcu_read_lock_bh();
408 n = __neigh_lookup_noref(tbl, pkey, dev);
409 if (n) {
410 if (!atomic_inc_not_zero(&n->refcnt))
411 n = NULL;
412 NEIGH_CACHE_STAT_INC(tbl, hits);
413 }
414
415 rcu_read_unlock_bh();
416 return n;
417}
418EXPORT_SYMBOL(neigh_lookup);
419
420struct neighbour *neigh_lookup_nodev(struct neigh_table *tbl, struct net *net,
421 const void *pkey)
422{
423 struct neighbour *n;
424 int key_len = tbl->key_len;
425 u32 hash_val;
426 struct neigh_hash_table *nht;
427
428 NEIGH_CACHE_STAT_INC(tbl, lookups);
429
430 rcu_read_lock_bh();
431 nht = rcu_dereference_bh(tbl->nht);
432 hash_val = tbl->hash(pkey, NULL, nht->hash_rnd) >> (32 - nht->hash_shift);
433
434 for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]);
435 n != NULL;
436 n = rcu_dereference_bh(n->next)) {
437 if (!memcmp(n->primary_key, pkey, key_len) &&
438 net_eq(dev_net(n->dev), net)) {
439 if (!atomic_inc_not_zero(&n->refcnt))
440 n = NULL;
441 NEIGH_CACHE_STAT_INC(tbl, hits);
442 break;
443 }
444 }
445
446 rcu_read_unlock_bh();
447 return n;
448}
449EXPORT_SYMBOL(neigh_lookup_nodev);
450
451struct neighbour *__neigh_create(struct neigh_table *tbl, const void *pkey,
452 struct net_device *dev, bool want_ref)
453{
454 u32 hash_val;
455 int key_len = tbl->key_len;
456 int error;
457 struct neighbour *n1, *rc, *n = neigh_alloc(tbl, dev);
458 struct neigh_hash_table *nht;
459
460 if (!n) {
461 rc = ERR_PTR(-ENOBUFS);
462 goto out;
463 }
464
465 memcpy(n->primary_key, pkey, key_len);
466 n->dev = dev;
467 dev_hold(dev);
468
469 /* Protocol specific setup. */
470 if (tbl->constructor && (error = tbl->constructor(n)) < 0) {
471 rc = ERR_PTR(error);
472 goto out_neigh_release;
473 }
474
475 if (dev->netdev_ops->ndo_neigh_construct) {
476 error = dev->netdev_ops->ndo_neigh_construct(n);
477 if (error < 0) {
478 rc = ERR_PTR(error);
479 goto out_neigh_release;
480 }
481 }
482
483 /* Device specific setup. */
484 if (n->parms->neigh_setup &&
485 (error = n->parms->neigh_setup(n)) < 0) {
486 rc = ERR_PTR(error);
487 goto out_neigh_release;
488 }
489
490 n->confirmed = jiffies - (NEIGH_VAR(n->parms, BASE_REACHABLE_TIME) << 1);
491
492 write_lock_bh(&tbl->lock);
493 nht = rcu_dereference_protected(tbl->nht,
494 lockdep_is_held(&tbl->lock));
495
496 if (atomic_read(&tbl->entries) > (1 << nht->hash_shift))
497 nht = neigh_hash_grow(tbl, nht->hash_shift + 1);
498
499 hash_val = tbl->hash(pkey, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
500
501 if (n->parms->dead) {
502 rc = ERR_PTR(-EINVAL);
503 goto out_tbl_unlock;
504 }
505
506 for (n1 = rcu_dereference_protected(nht->hash_buckets[hash_val],
507 lockdep_is_held(&tbl->lock));
508 n1 != NULL;
509 n1 = rcu_dereference_protected(n1->next,
510 lockdep_is_held(&tbl->lock))) {
511 if (dev == n1->dev && !memcmp(n1->primary_key, pkey, key_len)) {
512 if (want_ref)
513 neigh_hold(n1);
514 rc = n1;
515 goto out_tbl_unlock;
516 }
517 }
518
519 n->dead = 0;
520 if (want_ref)
521 neigh_hold(n);
522 rcu_assign_pointer(n->next,
523 rcu_dereference_protected(nht->hash_buckets[hash_val],
524 lockdep_is_held(&tbl->lock)));
525 rcu_assign_pointer(nht->hash_buckets[hash_val], n);
526 write_unlock_bh(&tbl->lock);
527 neigh_dbg(2, "neigh %p is created\n", n);
528 rc = n;
529out:
530 return rc;
531out_tbl_unlock:
532 write_unlock_bh(&tbl->lock);
533out_neigh_release:
534 neigh_release(n);
535 goto out;
536}
537EXPORT_SYMBOL(__neigh_create);
538
539static u32 pneigh_hash(const void *pkey, int key_len)
540{
541 u32 hash_val = *(u32 *)(pkey + key_len - 4);
542 hash_val ^= (hash_val >> 16);
543 hash_val ^= hash_val >> 8;
544 hash_val ^= hash_val >> 4;
545 hash_val &= PNEIGH_HASHMASK;
546 return hash_val;
547}
548
549static struct pneigh_entry *__pneigh_lookup_1(struct pneigh_entry *n,
550 struct net *net,
551 const void *pkey,
552 int key_len,
553 struct net_device *dev)
554{
555 while (n) {
556 if (!memcmp(n->key, pkey, key_len) &&
557 net_eq(pneigh_net(n), net) &&
558 (n->dev == dev || !n->dev))
559 return n;
560 n = n->next;
561 }
562 return NULL;
563}
564
565struct pneigh_entry *__pneigh_lookup(struct neigh_table *tbl,
566 struct net *net, const void *pkey, struct net_device *dev)
567{
568 int key_len = tbl->key_len;
569 u32 hash_val = pneigh_hash(pkey, key_len);
570
571 return __pneigh_lookup_1(tbl->phash_buckets[hash_val],
572 net, pkey, key_len, dev);
573}
574EXPORT_SYMBOL_GPL(__pneigh_lookup);
575
576struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl,
577 struct net *net, const void *pkey,
578 struct net_device *dev, int creat)
579{
580 struct pneigh_entry *n;
581 int key_len = tbl->key_len;
582 u32 hash_val = pneigh_hash(pkey, key_len);
583
584 read_lock_bh(&tbl->lock);
585 n = __pneigh_lookup_1(tbl->phash_buckets[hash_val],
586 net, pkey, key_len, dev);
587 read_unlock_bh(&tbl->lock);
588
589 if (n || !creat)
590 goto out;
591
592 ASSERT_RTNL();
593
594 n = kmalloc(sizeof(*n) + key_len, GFP_KERNEL);
595 if (!n)
596 goto out;
597
598 write_pnet(&n->net, net);
599 memcpy(n->key, pkey, key_len);
600 n->dev = dev;
601 if (dev)
602 dev_hold(dev);
603
604 if (tbl->pconstructor && tbl->pconstructor(n)) {
605 if (dev)
606 dev_put(dev);
607 kfree(n);
608 n = NULL;
609 goto out;
610 }
611
612 write_lock_bh(&tbl->lock);
613 n->next = tbl->phash_buckets[hash_val];
614 tbl->phash_buckets[hash_val] = n;
615 write_unlock_bh(&tbl->lock);
616out:
617 return n;
618}
619EXPORT_SYMBOL(pneigh_lookup);
620
621
622int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
623 struct net_device *dev)
624{
625 struct pneigh_entry *n, **np;
626 int key_len = tbl->key_len;
627 u32 hash_val = pneigh_hash(pkey, key_len);
628
629 write_lock_bh(&tbl->lock);
630 for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
631 np = &n->next) {
632 if (!memcmp(n->key, pkey, key_len) && n->dev == dev &&
633 net_eq(pneigh_net(n), net)) {
634 *np = n->next;
635 write_unlock_bh(&tbl->lock);
636 if (tbl->pdestructor)
637 tbl->pdestructor(n);
638 if (n->dev)
639 dev_put(n->dev);
640 kfree(n);
641 return 0;
642 }
643 }
644 write_unlock_bh(&tbl->lock);
645 return -ENOENT;
646}
647
648static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
649{
650 struct pneigh_entry *n, **np;
651 u32 h;
652
653 for (h = 0; h <= PNEIGH_HASHMASK; h++) {
654 np = &tbl->phash_buckets[h];
655 while ((n = *np) != NULL) {
656 if (!dev || n->dev == dev) {
657 *np = n->next;
658 if (tbl->pdestructor)
659 tbl->pdestructor(n);
660 if (n->dev)
661 dev_put(n->dev);
662 kfree(n);
663 continue;
664 }
665 np = &n->next;
666 }
667 }
668 return -ENOENT;
669}
670
671static void neigh_parms_destroy(struct neigh_parms *parms);
672
673static inline void neigh_parms_put(struct neigh_parms *parms)
674{
675 if (atomic_dec_and_test(&parms->refcnt))
676 neigh_parms_destroy(parms);
677}
678
679/*
680 * neighbour must already be out of the table;
681 *
682 */
683void neigh_destroy(struct neighbour *neigh)
684{
685 struct net_device *dev = neigh->dev;
686
687 NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
688
689 if (!neigh->dead) {
690 pr_warn("Destroying alive neighbour %p\n", neigh);
691 dump_stack();
692 return;
693 }
694
695 if (neigh_del_timer(neigh))
696 pr_warn("Impossible event\n");
697
698 write_lock_bh(&neigh->lock);
699 __skb_queue_purge(&neigh->arp_queue);
700 write_unlock_bh(&neigh->lock);
701 neigh->arp_queue_len_bytes = 0;
702
703 if (dev->netdev_ops->ndo_neigh_destroy)
704 dev->netdev_ops->ndo_neigh_destroy(neigh);
705
706 dev_put(dev);
707 neigh_parms_put(neigh->parms);
708
709 neigh_dbg(2, "neigh %p is destroyed\n", neigh);
710
711 atomic_dec(&neigh->tbl->entries);
712 kfree_rcu(neigh, rcu);
713}
714EXPORT_SYMBOL(neigh_destroy);
715
716/* Neighbour state is suspicious;
717 disable fast path.
718
719 Called with write_locked neigh.
720 */
721static void neigh_suspect(struct neighbour *neigh)
722{
723 neigh_dbg(2, "neigh %p is suspected\n", neigh);
724
725 neigh->output = neigh->ops->output;
726}
727
728/* Neighbour state is OK;
729 enable fast path.
730
731 Called with write_locked neigh.
732 */
733static void neigh_connect(struct neighbour *neigh)
734{
735 neigh_dbg(2, "neigh %p is connected\n", neigh);
736
737 neigh->output = neigh->ops->connected_output;
738}
739
740static void neigh_periodic_work(struct work_struct *work)
741{
742 struct neigh_table *tbl = container_of(work, struct neigh_table, gc_work.work);
743 struct neighbour *n;
744 struct neighbour __rcu **np;
745 unsigned int i;
746 struct neigh_hash_table *nht;
747
748 NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
749
750 write_lock_bh(&tbl->lock);
751 nht = rcu_dereference_protected(tbl->nht,
752 lockdep_is_held(&tbl->lock));
753
754 /*
755 * periodically recompute ReachableTime from random function
756 */
757
758 if (time_after(jiffies, tbl->last_rand + 300 * HZ)) {
759 struct neigh_parms *p;
760 tbl->last_rand = jiffies;
761 list_for_each_entry(p, &tbl->parms_list, list)
762 p->reachable_time =
763 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
764 }
765
766 if (atomic_read(&tbl->entries) < tbl->gc_thresh1)
767 goto out;
768
769 for (i = 0 ; i < (1 << nht->hash_shift); i++) {
770 np = &nht->hash_buckets[i];
771
772 while ((n = rcu_dereference_protected(*np,
773 lockdep_is_held(&tbl->lock))) != NULL) {
774 unsigned int state;
775
776 write_lock(&n->lock);
777
778 state = n->nud_state;
779 if (state & (NUD_PERMANENT | NUD_IN_TIMER)) {
780 write_unlock(&n->lock);
781 goto next_elt;
782 }
783
784 if (time_before(n->used, n->confirmed))
785 n->used = n->confirmed;
786
787 if (atomic_read(&n->refcnt) == 1 &&
788 (state == NUD_FAILED ||
789 time_after(jiffies, n->used + NEIGH_VAR(n->parms, GC_STALETIME)))) {
790 *np = n->next;
791 n->dead = 1;
792 write_unlock(&n->lock);
793 neigh_cleanup_and_release(n);
794 continue;
795 }
796 write_unlock(&n->lock);
797
798next_elt:
799 np = &n->next;
800 }
801 /*
802 * It's fine to release lock here, even if hash table
803 * grows while we are preempted.
804 */
805 write_unlock_bh(&tbl->lock);
806 cond_resched();
807 write_lock_bh(&tbl->lock);
808 nht = rcu_dereference_protected(tbl->nht,
809 lockdep_is_held(&tbl->lock));
810 }
811out:
812 /* Cycle through all hash buckets every BASE_REACHABLE_TIME/2 ticks.
813 * ARP entry timeouts range from 1/2 BASE_REACHABLE_TIME to 3/2
814 * BASE_REACHABLE_TIME.
815 */
816 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
817 NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME) >> 1);
818 write_unlock_bh(&tbl->lock);
819}
820
821static __inline__ int neigh_max_probes(struct neighbour *n)
822{
823 struct neigh_parms *p = n->parms;
824 return NEIGH_VAR(p, UCAST_PROBES) + NEIGH_VAR(p, APP_PROBES) +
825 (n->nud_state & NUD_PROBE ? NEIGH_VAR(p, MCAST_REPROBES) :
826 NEIGH_VAR(p, MCAST_PROBES));
827}
828
829static void neigh_invalidate(struct neighbour *neigh)
830 __releases(neigh->lock)
831 __acquires(neigh->lock)
832{
833 struct sk_buff *skb;
834
835 NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
836 neigh_dbg(2, "neigh %p is failed\n", neigh);
837 neigh->updated = jiffies;
838
839 /* It is very thin place. report_unreachable is very complicated
840 routine. Particularly, it can hit the same neighbour entry!
841
842 So that, we try to be accurate and avoid dead loop. --ANK
843 */
844 while (neigh->nud_state == NUD_FAILED &&
845 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
846 write_unlock(&neigh->lock);
847 neigh->ops->error_report(neigh, skb);
848 write_lock(&neigh->lock);
849 }
850 __skb_queue_purge(&neigh->arp_queue);
851 neigh->arp_queue_len_bytes = 0;
852}
853
854static void neigh_probe(struct neighbour *neigh)
855 __releases(neigh->lock)
856{
857 struct sk_buff *skb = skb_peek_tail(&neigh->arp_queue);
858 /* keep skb alive even if arp_queue overflows */
859 if (skb)
860 skb = skb_clone(skb, GFP_ATOMIC);
861 write_unlock(&neigh->lock);
862 neigh->ops->solicit(neigh, skb);
863 atomic_inc(&neigh->probes);
864 kfree_skb(skb);
865}
866
867/* Called when a timer expires for a neighbour entry. */
868
869static void neigh_timer_handler(unsigned long arg)
870{
871 unsigned long now, next;
872 struct neighbour *neigh = (struct neighbour *)arg;
873 unsigned int state;
874 int notify = 0;
875
876 write_lock(&neigh->lock);
877
878 state = neigh->nud_state;
879 now = jiffies;
880 next = now + HZ;
881
882 if (!(state & NUD_IN_TIMER))
883 goto out;
884
885 if (state & NUD_REACHABLE) {
886 if (time_before_eq(now,
887 neigh->confirmed + neigh->parms->reachable_time)) {
888 neigh_dbg(2, "neigh %p is still alive\n", neigh);
889 next = neigh->confirmed + neigh->parms->reachable_time;
890 } else if (time_before_eq(now,
891 neigh->used +
892 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
893 neigh_dbg(2, "neigh %p is delayed\n", neigh);
894 neigh->nud_state = NUD_DELAY;
895 neigh->updated = jiffies;
896 neigh_suspect(neigh);
897 next = now + NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME);
898 } else {
899 neigh_dbg(2, "neigh %p is suspected\n", neigh);
900 neigh->nud_state = NUD_STALE;
901 neigh->updated = jiffies;
902 neigh_suspect(neigh);
903 notify = 1;
904 }
905 } else if (state & NUD_DELAY) {
906 if (time_before_eq(now,
907 neigh->confirmed +
908 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
909 neigh_dbg(2, "neigh %p is now reachable\n", neigh);
910 neigh->nud_state = NUD_REACHABLE;
911 neigh->updated = jiffies;
912 neigh_connect(neigh);
913 notify = 1;
914 next = neigh->confirmed + neigh->parms->reachable_time;
915 } else {
916 neigh_dbg(2, "neigh %p is probed\n", neigh);
917 neigh->nud_state = NUD_PROBE;
918 neigh->updated = jiffies;
919 atomic_set(&neigh->probes, 0);
920 notify = 1;
921 next = now + NEIGH_VAR(neigh->parms, RETRANS_TIME);
922 }
923 } else {
924 /* NUD_PROBE|NUD_INCOMPLETE */
925 next = now + NEIGH_VAR(neigh->parms, RETRANS_TIME);
926 }
927
928 if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
929 atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
930 neigh->nud_state = NUD_FAILED;
931 notify = 1;
932 neigh_invalidate(neigh);
933 goto out;
934 }
935
936 if (neigh->nud_state & NUD_IN_TIMER) {
937 if (time_before(next, jiffies + HZ/2))
938 next = jiffies + HZ/2;
939 if (!mod_timer(&neigh->timer, next))
940 neigh_hold(neigh);
941 }
942 if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
943 neigh_probe(neigh);
944 } else {
945out:
946 write_unlock(&neigh->lock);
947 }
948
949 if (notify)
950 neigh_update_notify(neigh);
951
952 neigh_release(neigh);
953}
954
955int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
956{
957 int rc;
958 bool immediate_probe = false;
959
960 write_lock_bh(&neigh->lock);
961
962 rc = 0;
963 if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
964 goto out_unlock_bh;
965 if (neigh->dead)
966 goto out_dead;
967
968 if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
969 if (NEIGH_VAR(neigh->parms, MCAST_PROBES) +
970 NEIGH_VAR(neigh->parms, APP_PROBES)) {
971 unsigned long next, now = jiffies;
972
973 atomic_set(&neigh->probes,
974 NEIGH_VAR(neigh->parms, UCAST_PROBES));
975 neigh->nud_state = NUD_INCOMPLETE;
976 neigh->updated = now;
977 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
978 HZ/2);
979 neigh_add_timer(neigh, next);
980 immediate_probe = true;
981 } else {
982 neigh->nud_state = NUD_FAILED;
983 neigh->updated = jiffies;
984 write_unlock_bh(&neigh->lock);
985
986 kfree_skb(skb);
987 return 1;
988 }
989 } else if (neigh->nud_state & NUD_STALE) {
990 neigh_dbg(2, "neigh %p is delayed\n", neigh);
991 neigh->nud_state = NUD_DELAY;
992 neigh->updated = jiffies;
993 neigh_add_timer(neigh, jiffies +
994 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME));
995 }
996
997 if (neigh->nud_state == NUD_INCOMPLETE) {
998 if (skb) {
999 while (neigh->arp_queue_len_bytes + skb->truesize >
1000 NEIGH_VAR(neigh->parms, QUEUE_LEN_BYTES)) {
1001 struct sk_buff *buff;
1002
1003 buff = __skb_dequeue(&neigh->arp_queue);
1004 if (!buff)
1005 break;
1006 neigh->arp_queue_len_bytes -= buff->truesize;
1007 kfree_skb(buff);
1008 NEIGH_CACHE_STAT_INC(neigh->tbl, unres_discards);
1009 }
1010 skb_dst_force(skb);
1011 __skb_queue_tail(&neigh->arp_queue, skb);
1012 neigh->arp_queue_len_bytes += skb->truesize;
1013 }
1014 rc = 1;
1015 }
1016out_unlock_bh:
1017 if (immediate_probe)
1018 neigh_probe(neigh);
1019 else
1020 write_unlock(&neigh->lock);
1021 local_bh_enable();
1022 return rc;
1023
1024out_dead:
1025 if (neigh->nud_state & NUD_STALE)
1026 goto out_unlock_bh;
1027 write_unlock_bh(&neigh->lock);
1028 kfree_skb(skb);
1029 return 1;
1030}
1031EXPORT_SYMBOL(__neigh_event_send);
1032
1033static void neigh_update_hhs(struct neighbour *neigh)
1034{
1035 struct hh_cache *hh;
1036 void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *)
1037 = NULL;
1038
1039 if (neigh->dev->header_ops)
1040 update = neigh->dev->header_ops->cache_update;
1041
1042 if (update) {
1043 hh = &neigh->hh;
1044 if (hh->hh_len) {
1045 write_seqlock_bh(&hh->hh_lock);
1046 update(hh, neigh->dev, neigh->ha);
1047 write_sequnlock_bh(&hh->hh_lock);
1048 }
1049 }
1050}
1051
1052
1053
1054/* Generic update routine.
1055 -- lladdr is new lladdr or NULL, if it is not supplied.
1056 -- new is new state.
1057 -- flags
1058 NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
1059 if it is different.
1060 NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
1061 lladdr instead of overriding it
1062 if it is different.
1063 It also allows to retain current state
1064 if lladdr is unchanged.
1065 NEIGH_UPDATE_F_ADMIN means that the change is administrative.
1066
1067 NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
1068 NTF_ROUTER flag.
1069 NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as
1070 a router.
1071
1072 Caller MUST hold reference count on the entry.
1073 */
1074
1075int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
1076 u32 flags)
1077{
1078 u8 old;
1079 int err;
1080 int notify = 0;
1081 struct net_device *dev;
1082 int update_isrouter = 0;
1083
1084 write_lock_bh(&neigh->lock);
1085
1086 dev = neigh->dev;
1087 old = neigh->nud_state;
1088 err = -EPERM;
1089
1090 if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
1091 (old & (NUD_NOARP | NUD_PERMANENT)))
1092 goto out;
1093 if (neigh->dead)
1094 goto out;
1095
1096 if (!(new & NUD_VALID)) {
1097 neigh_del_timer(neigh);
1098 if (old & NUD_CONNECTED)
1099 neigh_suspect(neigh);
1100 neigh->nud_state = new;
1101 err = 0;
1102 notify = old & NUD_VALID;
1103 if ((old & (NUD_INCOMPLETE | NUD_PROBE)) &&
1104 (new & NUD_FAILED)) {
1105 neigh_invalidate(neigh);
1106 notify = 1;
1107 }
1108 goto out;
1109 }
1110
1111 /* Compare new lladdr with cached one */
1112 if (!dev->addr_len) {
1113 /* First case: device needs no address. */
1114 lladdr = neigh->ha;
1115 } else if (lladdr) {
1116 /* The second case: if something is already cached
1117 and a new address is proposed:
1118 - compare new & old
1119 - if they are different, check override flag
1120 */
1121 if ((old & NUD_VALID) &&
1122 !memcmp(lladdr, neigh->ha, dev->addr_len))
1123 lladdr = neigh->ha;
1124 } else {
1125 /* No address is supplied; if we know something,
1126 use it, otherwise discard the request.
1127 */
1128 err = -EINVAL;
1129 if (!(old & NUD_VALID))
1130 goto out;
1131 lladdr = neigh->ha;
1132 }
1133
1134 if (new & NUD_CONNECTED)
1135 neigh->confirmed = jiffies;
1136 neigh->updated = jiffies;
1137
1138 /* If entry was valid and address is not changed,
1139 do not change entry state, if new one is STALE.
1140 */
1141 err = 0;
1142 update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1143 if (old & NUD_VALID) {
1144 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
1145 update_isrouter = 0;
1146 if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
1147 (old & NUD_CONNECTED)) {
1148 lladdr = neigh->ha;
1149 new = NUD_STALE;
1150 } else
1151 goto out;
1152 } else {
1153 if (lladdr == neigh->ha && new == NUD_STALE &&
1154 ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) ||
1155 (old & NUD_CONNECTED))
1156 )
1157 new = old;
1158 }
1159 }
1160
1161 if (new != old) {
1162 neigh_del_timer(neigh);
1163 if (new & NUD_PROBE)
1164 atomic_set(&neigh->probes, 0);
1165 if (new & NUD_IN_TIMER)
1166 neigh_add_timer(neigh, (jiffies +
1167 ((new & NUD_REACHABLE) ?
1168 neigh->parms->reachable_time :
1169 0)));
1170 neigh->nud_state = new;
1171 notify = 1;
1172 }
1173
1174 if (lladdr != neigh->ha) {
1175 write_seqlock(&neigh->ha_lock);
1176 memcpy(&neigh->ha, lladdr, dev->addr_len);
1177 write_sequnlock(&neigh->ha_lock);
1178 neigh_update_hhs(neigh);
1179 if (!(new & NUD_CONNECTED))
1180 neigh->confirmed = jiffies -
1181 (NEIGH_VAR(neigh->parms, BASE_REACHABLE_TIME) << 1);
1182 notify = 1;
1183 }
1184 if (new == old)
1185 goto out;
1186 if (new & NUD_CONNECTED)
1187 neigh_connect(neigh);
1188 else
1189 neigh_suspect(neigh);
1190 if (!(old & NUD_VALID)) {
1191 struct sk_buff *skb;
1192
1193 /* Again: avoid dead loop if something went wrong */
1194
1195 while (neigh->nud_state & NUD_VALID &&
1196 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1197 struct dst_entry *dst = skb_dst(skb);
1198 struct neighbour *n2, *n1 = neigh;
1199 write_unlock_bh(&neigh->lock);
1200
1201 rcu_read_lock();
1202
1203 /* Why not just use 'neigh' as-is? The problem is that
1204 * things such as shaper, eql, and sch_teql can end up
1205 * using alternative, different, neigh objects to output
1206 * the packet in the output path. So what we need to do
1207 * here is re-lookup the top-level neigh in the path so
1208 * we can reinject the packet there.
1209 */
1210 n2 = NULL;
1211 if (dst) {
1212 n2 = dst_neigh_lookup_skb(dst, skb);
1213 if (n2)
1214 n1 = n2;
1215 }
1216 n1->output(n1, skb);
1217 if (n2)
1218 neigh_release(n2);
1219 rcu_read_unlock();
1220
1221 write_lock_bh(&neigh->lock);
1222 }
1223 __skb_queue_purge(&neigh->arp_queue);
1224 neigh->arp_queue_len_bytes = 0;
1225 }
1226out:
1227 if (update_isrouter) {
1228 neigh->flags = (flags & NEIGH_UPDATE_F_ISROUTER) ?
1229 (neigh->flags | NTF_ROUTER) :
1230 (neigh->flags & ~NTF_ROUTER);
1231 }
1232 write_unlock_bh(&neigh->lock);
1233
1234 if (notify)
1235 neigh_update_notify(neigh);
1236
1237 return err;
1238}
1239EXPORT_SYMBOL(neigh_update);
1240
1241/* Update the neigh to listen temporarily for probe responses, even if it is
1242 * in a NUD_FAILED state. The caller has to hold neigh->lock for writing.
1243 */
1244void __neigh_set_probe_once(struct neighbour *neigh)
1245{
1246 if (neigh->dead)
1247 return;
1248 neigh->updated = jiffies;
1249 if (!(neigh->nud_state & NUD_FAILED))
1250 return;
1251 neigh->nud_state = NUD_INCOMPLETE;
1252 atomic_set(&neigh->probes, neigh_max_probes(neigh));
1253 neigh_add_timer(neigh,
1254 jiffies + NEIGH_VAR(neigh->parms, RETRANS_TIME));
1255}
1256EXPORT_SYMBOL(__neigh_set_probe_once);
1257
1258struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1259 u8 *lladdr, void *saddr,
1260 struct net_device *dev)
1261{
1262 struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1263 lladdr || !dev->addr_len);
1264 if (neigh)
1265 neigh_update(neigh, lladdr, NUD_STALE,
1266 NEIGH_UPDATE_F_OVERRIDE);
1267 return neigh;
1268}
1269EXPORT_SYMBOL(neigh_event_ns);
1270
1271/* called with read_lock_bh(&n->lock); */
1272static void neigh_hh_init(struct neighbour *n)
1273{
1274 struct net_device *dev = n->dev;
1275 __be16 prot = n->tbl->protocol;
1276 struct hh_cache *hh = &n->hh;
1277
1278 write_lock_bh(&n->lock);
1279
1280 /* Only one thread can come in here and initialize the
1281 * hh_cache entry.
1282 */
1283 if (!hh->hh_len)
1284 dev->header_ops->cache(n, hh, prot);
1285
1286 write_unlock_bh(&n->lock);
1287}
1288
1289/* Slow and careful. */
1290
1291int neigh_resolve_output(struct neighbour *neigh, struct sk_buff *skb)
1292{
1293 int rc = 0;
1294
1295 if (!neigh_event_send(neigh, skb)) {
1296 int err;
1297 struct net_device *dev = neigh->dev;
1298 unsigned int seq;
1299
1300 if (dev->header_ops->cache && !neigh->hh.hh_len)
1301 neigh_hh_init(neigh);
1302
1303 do {
1304 __skb_pull(skb, skb_network_offset(skb));
1305 seq = read_seqbegin(&neigh->ha_lock);
1306 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1307 neigh->ha, NULL, skb->len);
1308 } while (read_seqretry(&neigh->ha_lock, seq));
1309
1310 if (err >= 0)
1311 rc = dev_queue_xmit(skb);
1312 else
1313 goto out_kfree_skb;
1314 }
1315out:
1316 return rc;
1317out_kfree_skb:
1318 rc = -EINVAL;
1319 kfree_skb(skb);
1320 goto out;
1321}
1322EXPORT_SYMBOL(neigh_resolve_output);
1323
1324/* As fast as possible without hh cache */
1325
1326int neigh_connected_output(struct neighbour *neigh, struct sk_buff *skb)
1327{
1328 struct net_device *dev = neigh->dev;
1329 unsigned int seq;
1330 int err;
1331
1332 do {
1333 __skb_pull(skb, skb_network_offset(skb));
1334 seq = read_seqbegin(&neigh->ha_lock);
1335 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1336 neigh->ha, NULL, skb->len);
1337 } while (read_seqretry(&neigh->ha_lock, seq));
1338
1339 if (err >= 0)
1340 err = dev_queue_xmit(skb);
1341 else {
1342 err = -EINVAL;
1343 kfree_skb(skb);
1344 }
1345 return err;
1346}
1347EXPORT_SYMBOL(neigh_connected_output);
1348
1349int neigh_direct_output(struct neighbour *neigh, struct sk_buff *skb)
1350{
1351 return dev_queue_xmit(skb);
1352}
1353EXPORT_SYMBOL(neigh_direct_output);
1354
1355static void neigh_proxy_process(unsigned long arg)
1356{
1357 struct neigh_table *tbl = (struct neigh_table *)arg;
1358 long sched_next = 0;
1359 unsigned long now = jiffies;
1360 struct sk_buff *skb, *n;
1361
1362 spin_lock(&tbl->proxy_queue.lock);
1363
1364 skb_queue_walk_safe(&tbl->proxy_queue, skb, n) {
1365 long tdif = NEIGH_CB(skb)->sched_next - now;
1366
1367 if (tdif <= 0) {
1368 struct net_device *dev = skb->dev;
1369
1370 __skb_unlink(skb, &tbl->proxy_queue);
1371 if (tbl->proxy_redo && netif_running(dev)) {
1372 rcu_read_lock();
1373 tbl->proxy_redo(skb);
1374 rcu_read_unlock();
1375 } else {
1376 kfree_skb(skb);
1377 }
1378
1379 dev_put(dev);
1380 } else if (!sched_next || tdif < sched_next)
1381 sched_next = tdif;
1382 }
1383 del_timer(&tbl->proxy_timer);
1384 if (sched_next)
1385 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1386 spin_unlock(&tbl->proxy_queue.lock);
1387}
1388
1389void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1390 struct sk_buff *skb)
1391{
1392 unsigned long now = jiffies;
1393
1394 unsigned long sched_next = now + (prandom_u32() %
1395 NEIGH_VAR(p, PROXY_DELAY));
1396
1397 if (tbl->proxy_queue.qlen > NEIGH_VAR(p, PROXY_QLEN)) {
1398 kfree_skb(skb);
1399 return;
1400 }
1401
1402 NEIGH_CB(skb)->sched_next = sched_next;
1403 NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1404
1405 spin_lock(&tbl->proxy_queue.lock);
1406 if (del_timer(&tbl->proxy_timer)) {
1407 if (time_before(tbl->proxy_timer.expires, sched_next))
1408 sched_next = tbl->proxy_timer.expires;
1409 }
1410 skb_dst_drop(skb);
1411 dev_hold(skb->dev);
1412 __skb_queue_tail(&tbl->proxy_queue, skb);
1413 mod_timer(&tbl->proxy_timer, sched_next);
1414 spin_unlock(&tbl->proxy_queue.lock);
1415}
1416EXPORT_SYMBOL(pneigh_enqueue);
1417
1418static inline struct neigh_parms *lookup_neigh_parms(struct neigh_table *tbl,
1419 struct net *net, int ifindex)
1420{
1421 struct neigh_parms *p;
1422
1423 list_for_each_entry(p, &tbl->parms_list, list) {
1424 if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
1425 (!p->dev && !ifindex && net_eq(net, &init_net)))
1426 return p;
1427 }
1428
1429 return NULL;
1430}
1431
1432struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1433 struct neigh_table *tbl)
1434{
1435 struct neigh_parms *p;
1436 struct net *net = dev_net(dev);
1437 const struct net_device_ops *ops = dev->netdev_ops;
1438
1439 p = kmemdup(&tbl->parms, sizeof(*p), GFP_KERNEL);
1440 if (p) {
1441 p->tbl = tbl;
1442 atomic_set(&p->refcnt, 1);
1443 p->reachable_time =
1444 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
1445 dev_hold(dev);
1446 p->dev = dev;
1447 write_pnet(&p->net, net);
1448 p->sysctl_table = NULL;
1449
1450 if (ops->ndo_neigh_setup && ops->ndo_neigh_setup(dev, p)) {
1451 dev_put(dev);
1452 kfree(p);
1453 return NULL;
1454 }
1455
1456 write_lock_bh(&tbl->lock);
1457 list_add(&p->list, &tbl->parms.list);
1458 write_unlock_bh(&tbl->lock);
1459
1460 neigh_parms_data_state_cleanall(p);
1461 }
1462 return p;
1463}
1464EXPORT_SYMBOL(neigh_parms_alloc);
1465
1466static void neigh_rcu_free_parms(struct rcu_head *head)
1467{
1468 struct neigh_parms *parms =
1469 container_of(head, struct neigh_parms, rcu_head);
1470
1471 neigh_parms_put(parms);
1472}
1473
1474void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1475{
1476 if (!parms || parms == &tbl->parms)
1477 return;
1478 write_lock_bh(&tbl->lock);
1479 list_del(&parms->list);
1480 parms->dead = 1;
1481 write_unlock_bh(&tbl->lock);
1482 if (parms->dev)
1483 dev_put(parms->dev);
1484 call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1485}
1486EXPORT_SYMBOL(neigh_parms_release);
1487
1488static void neigh_parms_destroy(struct neigh_parms *parms)
1489{
1490 kfree(parms);
1491}
1492
1493static struct lock_class_key neigh_table_proxy_queue_class;
1494
1495static struct neigh_table *neigh_tables[NEIGH_NR_TABLES] __read_mostly;
1496
1497void neigh_table_init(int index, struct neigh_table *tbl)
1498{
1499 unsigned long now = jiffies;
1500 unsigned long phsize;
1501
1502 INIT_LIST_HEAD(&tbl->parms_list);
1503 list_add(&tbl->parms.list, &tbl->parms_list);
1504 write_pnet(&tbl->parms.net, &init_net);
1505 atomic_set(&tbl->parms.refcnt, 1);
1506 tbl->parms.reachable_time =
1507 neigh_rand_reach_time(NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME));
1508
1509 tbl->stats = alloc_percpu(struct neigh_statistics);
1510 if (!tbl->stats)
1511 panic("cannot create neighbour cache statistics");
1512
1513#ifdef CONFIG_PROC_FS
1514 if (!proc_create_data(tbl->id, 0, init_net.proc_net_stat,
1515 &neigh_stat_seq_fops, tbl))
1516 panic("cannot create neighbour proc dir entry");
1517#endif
1518
1519 RCU_INIT_POINTER(tbl->nht, neigh_hash_alloc(3));
1520
1521 phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1522 tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1523
1524 if (!tbl->nht || !tbl->phash_buckets)
1525 panic("cannot allocate neighbour cache hashes");
1526
1527 if (!tbl->entry_size)
1528 tbl->entry_size = ALIGN(offsetof(struct neighbour, primary_key) +
1529 tbl->key_len, NEIGH_PRIV_ALIGN);
1530 else
1531 WARN_ON(tbl->entry_size % NEIGH_PRIV_ALIGN);
1532
1533 rwlock_init(&tbl->lock);
1534 INIT_DEFERRABLE_WORK(&tbl->gc_work, neigh_periodic_work);
1535 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
1536 tbl->parms.reachable_time);
1537 setup_timer(&tbl->proxy_timer, neigh_proxy_process, (unsigned long)tbl);
1538 skb_queue_head_init_class(&tbl->proxy_queue,
1539 &neigh_table_proxy_queue_class);
1540
1541 tbl->last_flush = now;
1542 tbl->last_rand = now + tbl->parms.reachable_time * 20;
1543
1544 neigh_tables[index] = tbl;
1545}
1546EXPORT_SYMBOL(neigh_table_init);
1547
1548int neigh_table_clear(int index, struct neigh_table *tbl)
1549{
1550 neigh_tables[index] = NULL;
1551 /* It is not clean... Fix it to unload IPv6 module safely */
1552 cancel_delayed_work_sync(&tbl->gc_work);
1553 del_timer_sync(&tbl->proxy_timer);
1554 pneigh_queue_purge(&tbl->proxy_queue);
1555 neigh_ifdown(tbl, NULL);
1556 if (atomic_read(&tbl->entries))
1557 pr_crit("neighbour leakage\n");
1558
1559 call_rcu(&rcu_dereference_protected(tbl->nht, 1)->rcu,
1560 neigh_hash_free_rcu);
1561 tbl->nht = NULL;
1562
1563 kfree(tbl->phash_buckets);
1564 tbl->phash_buckets = NULL;
1565
1566 remove_proc_entry(tbl->id, init_net.proc_net_stat);
1567
1568 free_percpu(tbl->stats);
1569 tbl->stats = NULL;
1570
1571 return 0;
1572}
1573EXPORT_SYMBOL(neigh_table_clear);
1574
1575static struct neigh_table *neigh_find_table(int family)
1576{
1577 struct neigh_table *tbl = NULL;
1578
1579 switch (family) {
1580 case AF_INET:
1581 tbl = neigh_tables[NEIGH_ARP_TABLE];
1582 break;
1583 case AF_INET6:
1584 tbl = neigh_tables[NEIGH_ND_TABLE];
1585 break;
1586 case AF_DECnet:
1587 tbl = neigh_tables[NEIGH_DN_TABLE];
1588 break;
1589 }
1590
1591 return tbl;
1592}
1593
1594static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh)
1595{
1596 struct net *net = sock_net(skb->sk);
1597 struct ndmsg *ndm;
1598 struct nlattr *dst_attr;
1599 struct neigh_table *tbl;
1600 struct neighbour *neigh;
1601 struct net_device *dev = NULL;
1602 int err = -EINVAL;
1603
1604 ASSERT_RTNL();
1605 if (nlmsg_len(nlh) < sizeof(*ndm))
1606 goto out;
1607
1608 dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1609 if (dst_attr == NULL)
1610 goto out;
1611
1612 ndm = nlmsg_data(nlh);
1613 if (ndm->ndm_ifindex) {
1614 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1615 if (dev == NULL) {
1616 err = -ENODEV;
1617 goto out;
1618 }
1619 }
1620
1621 tbl = neigh_find_table(ndm->ndm_family);
1622 if (tbl == NULL)
1623 return -EAFNOSUPPORT;
1624
1625 if (nla_len(dst_attr) < tbl->key_len)
1626 goto out;
1627
1628 if (ndm->ndm_flags & NTF_PROXY) {
1629 err = pneigh_delete(tbl, net, nla_data(dst_attr), dev);
1630 goto out;
1631 }
1632
1633 if (dev == NULL)
1634 goto out;
1635
1636 neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1637 if (neigh == NULL) {
1638 err = -ENOENT;
1639 goto out;
1640 }
1641
1642 err = neigh_update(neigh, NULL, NUD_FAILED,
1643 NEIGH_UPDATE_F_OVERRIDE |
1644 NEIGH_UPDATE_F_ADMIN);
1645 neigh_release(neigh);
1646
1647out:
1648 return err;
1649}
1650
1651static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh)
1652{
1653 int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE;
1654 struct net *net = sock_net(skb->sk);
1655 struct ndmsg *ndm;
1656 struct nlattr *tb[NDA_MAX+1];
1657 struct neigh_table *tbl;
1658 struct net_device *dev = NULL;
1659 struct neighbour *neigh;
1660 void *dst, *lladdr;
1661 int err;
1662
1663 ASSERT_RTNL();
1664 err = nlmsg_parse(nlh, sizeof(*ndm), tb, NDA_MAX, NULL);
1665 if (err < 0)
1666 goto out;
1667
1668 err = -EINVAL;
1669 if (tb[NDA_DST] == NULL)
1670 goto out;
1671
1672 ndm = nlmsg_data(nlh);
1673 if (ndm->ndm_ifindex) {
1674 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1675 if (dev == NULL) {
1676 err = -ENODEV;
1677 goto out;
1678 }
1679
1680 if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len)
1681 goto out;
1682 }
1683
1684 tbl = neigh_find_table(ndm->ndm_family);
1685 if (tbl == NULL)
1686 return -EAFNOSUPPORT;
1687
1688 if (nla_len(tb[NDA_DST]) < tbl->key_len)
1689 goto out;
1690 dst = nla_data(tb[NDA_DST]);
1691 lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
1692
1693 if (ndm->ndm_flags & NTF_PROXY) {
1694 struct pneigh_entry *pn;
1695
1696 err = -ENOBUFS;
1697 pn = pneigh_lookup(tbl, net, dst, dev, 1);
1698 if (pn) {
1699 pn->flags = ndm->ndm_flags;
1700 err = 0;
1701 }
1702 goto out;
1703 }
1704
1705 if (dev == NULL)
1706 goto out;
1707
1708 neigh = neigh_lookup(tbl, dst, dev);
1709 if (neigh == NULL) {
1710 if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
1711 err = -ENOENT;
1712 goto out;
1713 }
1714
1715 neigh = __neigh_lookup_errno(tbl, dst, dev);
1716 if (IS_ERR(neigh)) {
1717 err = PTR_ERR(neigh);
1718 goto out;
1719 }
1720 } else {
1721 if (nlh->nlmsg_flags & NLM_F_EXCL) {
1722 err = -EEXIST;
1723 neigh_release(neigh);
1724 goto out;
1725 }
1726
1727 if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
1728 flags &= ~NEIGH_UPDATE_F_OVERRIDE;
1729 }
1730
1731 if (ndm->ndm_flags & NTF_USE) {
1732 neigh_event_send(neigh, NULL);
1733 err = 0;
1734 } else
1735 err = neigh_update(neigh, lladdr, ndm->ndm_state, flags);
1736 neigh_release(neigh);
1737
1738out:
1739 return err;
1740}
1741
1742static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
1743{
1744 struct nlattr *nest;
1745
1746 nest = nla_nest_start(skb, NDTA_PARMS);
1747 if (nest == NULL)
1748 return -ENOBUFS;
1749
1750 if ((parms->dev &&
1751 nla_put_u32(skb, NDTPA_IFINDEX, parms->dev->ifindex)) ||
1752 nla_put_u32(skb, NDTPA_REFCNT, atomic_read(&parms->refcnt)) ||
1753 nla_put_u32(skb, NDTPA_QUEUE_LENBYTES,
1754 NEIGH_VAR(parms, QUEUE_LEN_BYTES)) ||
1755 /* approximative value for deprecated QUEUE_LEN (in packets) */
1756 nla_put_u32(skb, NDTPA_QUEUE_LEN,
1757 NEIGH_VAR(parms, QUEUE_LEN_BYTES) / SKB_TRUESIZE(ETH_FRAME_LEN)) ||
1758 nla_put_u32(skb, NDTPA_PROXY_QLEN, NEIGH_VAR(parms, PROXY_QLEN)) ||
1759 nla_put_u32(skb, NDTPA_APP_PROBES, NEIGH_VAR(parms, APP_PROBES)) ||
1760 nla_put_u32(skb, NDTPA_UCAST_PROBES,
1761 NEIGH_VAR(parms, UCAST_PROBES)) ||
1762 nla_put_u32(skb, NDTPA_MCAST_PROBES,
1763 NEIGH_VAR(parms, MCAST_PROBES)) ||
1764 nla_put_u32(skb, NDTPA_MCAST_REPROBES,
1765 NEIGH_VAR(parms, MCAST_REPROBES)) ||
1766 nla_put_msecs(skb, NDTPA_REACHABLE_TIME, parms->reachable_time) ||
1767 nla_put_msecs(skb, NDTPA_BASE_REACHABLE_TIME,
1768 NEIGH_VAR(parms, BASE_REACHABLE_TIME)) ||
1769 nla_put_msecs(skb, NDTPA_GC_STALETIME,
1770 NEIGH_VAR(parms, GC_STALETIME)) ||
1771 nla_put_msecs(skb, NDTPA_DELAY_PROBE_TIME,
1772 NEIGH_VAR(parms, DELAY_PROBE_TIME)) ||
1773 nla_put_msecs(skb, NDTPA_RETRANS_TIME,
1774 NEIGH_VAR(parms, RETRANS_TIME)) ||
1775 nla_put_msecs(skb, NDTPA_ANYCAST_DELAY,
1776 NEIGH_VAR(parms, ANYCAST_DELAY)) ||
1777 nla_put_msecs(skb, NDTPA_PROXY_DELAY,
1778 NEIGH_VAR(parms, PROXY_DELAY)) ||
1779 nla_put_msecs(skb, NDTPA_LOCKTIME,
1780 NEIGH_VAR(parms, LOCKTIME)))
1781 goto nla_put_failure;
1782 return nla_nest_end(skb, nest);
1783
1784nla_put_failure:
1785 nla_nest_cancel(skb, nest);
1786 return -EMSGSIZE;
1787}
1788
1789static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl,
1790 u32 pid, u32 seq, int type, int flags)
1791{
1792 struct nlmsghdr *nlh;
1793 struct ndtmsg *ndtmsg;
1794
1795 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1796 if (nlh == NULL)
1797 return -EMSGSIZE;
1798
1799 ndtmsg = nlmsg_data(nlh);
1800
1801 read_lock_bh(&tbl->lock);
1802 ndtmsg->ndtm_family = tbl->family;
1803 ndtmsg->ndtm_pad1 = 0;
1804 ndtmsg->ndtm_pad2 = 0;
1805
1806 if (nla_put_string(skb, NDTA_NAME, tbl->id) ||
1807 nla_put_msecs(skb, NDTA_GC_INTERVAL, tbl->gc_interval) ||
1808 nla_put_u32(skb, NDTA_THRESH1, tbl->gc_thresh1) ||
1809 nla_put_u32(skb, NDTA_THRESH2, tbl->gc_thresh2) ||
1810 nla_put_u32(skb, NDTA_THRESH3, tbl->gc_thresh3))
1811 goto nla_put_failure;
1812 {
1813 unsigned long now = jiffies;
1814 unsigned int flush_delta = now - tbl->last_flush;
1815 unsigned int rand_delta = now - tbl->last_rand;
1816 struct neigh_hash_table *nht;
1817 struct ndt_config ndc = {
1818 .ndtc_key_len = tbl->key_len,
1819 .ndtc_entry_size = tbl->entry_size,
1820 .ndtc_entries = atomic_read(&tbl->entries),
1821 .ndtc_last_flush = jiffies_to_msecs(flush_delta),
1822 .ndtc_last_rand = jiffies_to_msecs(rand_delta),
1823 .ndtc_proxy_qlen = tbl->proxy_queue.qlen,
1824 };
1825
1826 rcu_read_lock_bh();
1827 nht = rcu_dereference_bh(tbl->nht);
1828 ndc.ndtc_hash_rnd = nht->hash_rnd[0];
1829 ndc.ndtc_hash_mask = ((1 << nht->hash_shift) - 1);
1830 rcu_read_unlock_bh();
1831
1832 if (nla_put(skb, NDTA_CONFIG, sizeof(ndc), &ndc))
1833 goto nla_put_failure;
1834 }
1835
1836 {
1837 int cpu;
1838 struct ndt_stats ndst;
1839
1840 memset(&ndst, 0, sizeof(ndst));
1841
1842 for_each_possible_cpu(cpu) {
1843 struct neigh_statistics *st;
1844
1845 st = per_cpu_ptr(tbl->stats, cpu);
1846 ndst.ndts_allocs += st->allocs;
1847 ndst.ndts_destroys += st->destroys;
1848 ndst.ndts_hash_grows += st->hash_grows;
1849 ndst.ndts_res_failed += st->res_failed;
1850 ndst.ndts_lookups += st->lookups;
1851 ndst.ndts_hits += st->hits;
1852 ndst.ndts_rcv_probes_mcast += st->rcv_probes_mcast;
1853 ndst.ndts_rcv_probes_ucast += st->rcv_probes_ucast;
1854 ndst.ndts_periodic_gc_runs += st->periodic_gc_runs;
1855 ndst.ndts_forced_gc_runs += st->forced_gc_runs;
1856 ndst.ndts_table_fulls += st->table_fulls;
1857 }
1858
1859 if (nla_put(skb, NDTA_STATS, sizeof(ndst), &ndst))
1860 goto nla_put_failure;
1861 }
1862
1863 BUG_ON(tbl->parms.dev);
1864 if (neightbl_fill_parms(skb, &tbl->parms) < 0)
1865 goto nla_put_failure;
1866
1867 read_unlock_bh(&tbl->lock);
1868 nlmsg_end(skb, nlh);
1869 return 0;
1870
1871nla_put_failure:
1872 read_unlock_bh(&tbl->lock);
1873 nlmsg_cancel(skb, nlh);
1874 return -EMSGSIZE;
1875}
1876
1877static int neightbl_fill_param_info(struct sk_buff *skb,
1878 struct neigh_table *tbl,
1879 struct neigh_parms *parms,
1880 u32 pid, u32 seq, int type,
1881 unsigned int flags)
1882{
1883 struct ndtmsg *ndtmsg;
1884 struct nlmsghdr *nlh;
1885
1886 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1887 if (nlh == NULL)
1888 return -EMSGSIZE;
1889
1890 ndtmsg = nlmsg_data(nlh);
1891
1892 read_lock_bh(&tbl->lock);
1893 ndtmsg->ndtm_family = tbl->family;
1894 ndtmsg->ndtm_pad1 = 0;
1895 ndtmsg->ndtm_pad2 = 0;
1896
1897 if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 ||
1898 neightbl_fill_parms(skb, parms) < 0)
1899 goto errout;
1900
1901 read_unlock_bh(&tbl->lock);
1902 nlmsg_end(skb, nlh);
1903 return 0;
1904errout:
1905 read_unlock_bh(&tbl->lock);
1906 nlmsg_cancel(skb, nlh);
1907 return -EMSGSIZE;
1908}
1909
1910static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = {
1911 [NDTA_NAME] = { .type = NLA_STRING },
1912 [NDTA_THRESH1] = { .type = NLA_U32 },
1913 [NDTA_THRESH2] = { .type = NLA_U32 },
1914 [NDTA_THRESH3] = { .type = NLA_U32 },
1915 [NDTA_GC_INTERVAL] = { .type = NLA_U64 },
1916 [NDTA_PARMS] = { .type = NLA_NESTED },
1917};
1918
1919static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = {
1920 [NDTPA_IFINDEX] = { .type = NLA_U32 },
1921 [NDTPA_QUEUE_LEN] = { .type = NLA_U32 },
1922 [NDTPA_PROXY_QLEN] = { .type = NLA_U32 },
1923 [NDTPA_APP_PROBES] = { .type = NLA_U32 },
1924 [NDTPA_UCAST_PROBES] = { .type = NLA_U32 },
1925 [NDTPA_MCAST_PROBES] = { .type = NLA_U32 },
1926 [NDTPA_MCAST_REPROBES] = { .type = NLA_U32 },
1927 [NDTPA_BASE_REACHABLE_TIME] = { .type = NLA_U64 },
1928 [NDTPA_GC_STALETIME] = { .type = NLA_U64 },
1929 [NDTPA_DELAY_PROBE_TIME] = { .type = NLA_U64 },
1930 [NDTPA_RETRANS_TIME] = { .type = NLA_U64 },
1931 [NDTPA_ANYCAST_DELAY] = { .type = NLA_U64 },
1932 [NDTPA_PROXY_DELAY] = { .type = NLA_U64 },
1933 [NDTPA_LOCKTIME] = { .type = NLA_U64 },
1934};
1935
1936static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh)
1937{
1938 struct net *net = sock_net(skb->sk);
1939 struct neigh_table *tbl;
1940 struct ndtmsg *ndtmsg;
1941 struct nlattr *tb[NDTA_MAX+1];
1942 bool found = false;
1943 int err, tidx;
1944
1945 err = nlmsg_parse(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
1946 nl_neightbl_policy);
1947 if (err < 0)
1948 goto errout;
1949
1950 if (tb[NDTA_NAME] == NULL) {
1951 err = -EINVAL;
1952 goto errout;
1953 }
1954
1955 ndtmsg = nlmsg_data(nlh);
1956
1957 for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
1958 tbl = neigh_tables[tidx];
1959 if (!tbl)
1960 continue;
1961 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
1962 continue;
1963 if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0) {
1964 found = true;
1965 break;
1966 }
1967 }
1968
1969 if (!found)
1970 return -ENOENT;
1971
1972 /*
1973 * We acquire tbl->lock to be nice to the periodic timers and
1974 * make sure they always see a consistent set of values.
1975 */
1976 write_lock_bh(&tbl->lock);
1977
1978 if (tb[NDTA_PARMS]) {
1979 struct nlattr *tbp[NDTPA_MAX+1];
1980 struct neigh_parms *p;
1981 int i, ifindex = 0;
1982
1983 err = nla_parse_nested(tbp, NDTPA_MAX, tb[NDTA_PARMS],
1984 nl_ntbl_parm_policy);
1985 if (err < 0)
1986 goto errout_tbl_lock;
1987
1988 if (tbp[NDTPA_IFINDEX])
1989 ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
1990
1991 p = lookup_neigh_parms(tbl, net, ifindex);
1992 if (p == NULL) {
1993 err = -ENOENT;
1994 goto errout_tbl_lock;
1995 }
1996
1997 for (i = 1; i <= NDTPA_MAX; i++) {
1998 if (tbp[i] == NULL)
1999 continue;
2000
2001 switch (i) {
2002 case NDTPA_QUEUE_LEN:
2003 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2004 nla_get_u32(tbp[i]) *
2005 SKB_TRUESIZE(ETH_FRAME_LEN));
2006 break;
2007 case NDTPA_QUEUE_LENBYTES:
2008 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2009 nla_get_u32(tbp[i]));
2010 break;
2011 case NDTPA_PROXY_QLEN:
2012 NEIGH_VAR_SET(p, PROXY_QLEN,
2013 nla_get_u32(tbp[i]));
2014 break;
2015 case NDTPA_APP_PROBES:
2016 NEIGH_VAR_SET(p, APP_PROBES,
2017 nla_get_u32(tbp[i]));
2018 break;
2019 case NDTPA_UCAST_PROBES:
2020 NEIGH_VAR_SET(p, UCAST_PROBES,
2021 nla_get_u32(tbp[i]));
2022 break;
2023 case NDTPA_MCAST_PROBES:
2024 NEIGH_VAR_SET(p, MCAST_PROBES,
2025 nla_get_u32(tbp[i]));
2026 break;
2027 case NDTPA_MCAST_REPROBES:
2028 NEIGH_VAR_SET(p, MCAST_REPROBES,
2029 nla_get_u32(tbp[i]));
2030 break;
2031 case NDTPA_BASE_REACHABLE_TIME:
2032 NEIGH_VAR_SET(p, BASE_REACHABLE_TIME,
2033 nla_get_msecs(tbp[i]));
2034 /* update reachable_time as well, otherwise, the change will
2035 * only be effective after the next time neigh_periodic_work
2036 * decides to recompute it (can be multiple minutes)
2037 */
2038 p->reachable_time =
2039 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
2040 break;
2041 case NDTPA_GC_STALETIME:
2042 NEIGH_VAR_SET(p, GC_STALETIME,
2043 nla_get_msecs(tbp[i]));
2044 break;
2045 case NDTPA_DELAY_PROBE_TIME:
2046 NEIGH_VAR_SET(p, DELAY_PROBE_TIME,
2047 nla_get_msecs(tbp[i]));
2048 break;
2049 case NDTPA_RETRANS_TIME:
2050 NEIGH_VAR_SET(p, RETRANS_TIME,
2051 nla_get_msecs(tbp[i]));
2052 break;
2053 case NDTPA_ANYCAST_DELAY:
2054 NEIGH_VAR_SET(p, ANYCAST_DELAY,
2055 nla_get_msecs(tbp[i]));
2056 break;
2057 case NDTPA_PROXY_DELAY:
2058 NEIGH_VAR_SET(p, PROXY_DELAY,
2059 nla_get_msecs(tbp[i]));
2060 break;
2061 case NDTPA_LOCKTIME:
2062 NEIGH_VAR_SET(p, LOCKTIME,
2063 nla_get_msecs(tbp[i]));
2064 break;
2065 }
2066 }
2067 }
2068
2069 err = -ENOENT;
2070 if ((tb[NDTA_THRESH1] || tb[NDTA_THRESH2] ||
2071 tb[NDTA_THRESH3] || tb[NDTA_GC_INTERVAL]) &&
2072 !net_eq(net, &init_net))
2073 goto errout_tbl_lock;
2074
2075 if (tb[NDTA_THRESH1])
2076 tbl->gc_thresh1 = nla_get_u32(tb[NDTA_THRESH1]);
2077
2078 if (tb[NDTA_THRESH2])
2079 tbl->gc_thresh2 = nla_get_u32(tb[NDTA_THRESH2]);
2080
2081 if (tb[NDTA_THRESH3])
2082 tbl->gc_thresh3 = nla_get_u32(tb[NDTA_THRESH3]);
2083
2084 if (tb[NDTA_GC_INTERVAL])
2085 tbl->gc_interval = nla_get_msecs(tb[NDTA_GC_INTERVAL]);
2086
2087 err = 0;
2088
2089errout_tbl_lock:
2090 write_unlock_bh(&tbl->lock);
2091errout:
2092 return err;
2093}
2094
2095static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2096{
2097 struct net *net = sock_net(skb->sk);
2098 int family, tidx, nidx = 0;
2099 int tbl_skip = cb->args[0];
2100 int neigh_skip = cb->args[1];
2101 struct neigh_table *tbl;
2102
2103 family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2104
2105 for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
2106 struct neigh_parms *p;
2107
2108 tbl = neigh_tables[tidx];
2109 if (!tbl)
2110 continue;
2111
2112 if (tidx < tbl_skip || (family && tbl->family != family))
2113 continue;
2114
2115 if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).portid,
2116 cb->nlh->nlmsg_seq, RTM_NEWNEIGHTBL,
2117 NLM_F_MULTI) < 0)
2118 break;
2119
2120 nidx = 0;
2121 p = list_next_entry(&tbl->parms, list);
2122 list_for_each_entry_from(p, &tbl->parms_list, list) {
2123 if (!net_eq(neigh_parms_net(p), net))
2124 continue;
2125
2126 if (nidx < neigh_skip)
2127 goto next;
2128
2129 if (neightbl_fill_param_info(skb, tbl, p,
2130 NETLINK_CB(cb->skb).portid,
2131 cb->nlh->nlmsg_seq,
2132 RTM_NEWNEIGHTBL,
2133 NLM_F_MULTI) < 0)
2134 goto out;
2135 next:
2136 nidx++;
2137 }
2138
2139 neigh_skip = 0;
2140 }
2141out:
2142 cb->args[0] = tidx;
2143 cb->args[1] = nidx;
2144
2145 return skb->len;
2146}
2147
2148static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
2149 u32 pid, u32 seq, int type, unsigned int flags)
2150{
2151 unsigned long now = jiffies;
2152 struct nda_cacheinfo ci;
2153 struct nlmsghdr *nlh;
2154 struct ndmsg *ndm;
2155
2156 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2157 if (nlh == NULL)
2158 return -EMSGSIZE;
2159
2160 ndm = nlmsg_data(nlh);
2161 ndm->ndm_family = neigh->ops->family;
2162 ndm->ndm_pad1 = 0;
2163 ndm->ndm_pad2 = 0;
2164 ndm->ndm_flags = neigh->flags;
2165 ndm->ndm_type = neigh->type;
2166 ndm->ndm_ifindex = neigh->dev->ifindex;
2167
2168 if (nla_put(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key))
2169 goto nla_put_failure;
2170
2171 read_lock_bh(&neigh->lock);
2172 ndm->ndm_state = neigh->nud_state;
2173 if (neigh->nud_state & NUD_VALID) {
2174 char haddr[MAX_ADDR_LEN];
2175
2176 neigh_ha_snapshot(haddr, neigh, neigh->dev);
2177 if (nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, haddr) < 0) {
2178 read_unlock_bh(&neigh->lock);
2179 goto nla_put_failure;
2180 }
2181 }
2182
2183 ci.ndm_used = jiffies_to_clock_t(now - neigh->used);
2184 ci.ndm_confirmed = jiffies_to_clock_t(now - neigh->confirmed);
2185 ci.ndm_updated = jiffies_to_clock_t(now - neigh->updated);
2186 ci.ndm_refcnt = atomic_read(&neigh->refcnt) - 1;
2187 read_unlock_bh(&neigh->lock);
2188
2189 if (nla_put_u32(skb, NDA_PROBES, atomic_read(&neigh->probes)) ||
2190 nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
2191 goto nla_put_failure;
2192
2193 nlmsg_end(skb, nlh);
2194 return 0;
2195
2196nla_put_failure:
2197 nlmsg_cancel(skb, nlh);
2198 return -EMSGSIZE;
2199}
2200
2201static int pneigh_fill_info(struct sk_buff *skb, struct pneigh_entry *pn,
2202 u32 pid, u32 seq, int type, unsigned int flags,
2203 struct neigh_table *tbl)
2204{
2205 struct nlmsghdr *nlh;
2206 struct ndmsg *ndm;
2207
2208 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2209 if (nlh == NULL)
2210 return -EMSGSIZE;
2211
2212 ndm = nlmsg_data(nlh);
2213 ndm->ndm_family = tbl->family;
2214 ndm->ndm_pad1 = 0;
2215 ndm->ndm_pad2 = 0;
2216 ndm->ndm_flags = pn->flags | NTF_PROXY;
2217 ndm->ndm_type = RTN_UNICAST;
2218 ndm->ndm_ifindex = pn->dev ? pn->dev->ifindex : 0;
2219 ndm->ndm_state = NUD_NONE;
2220
2221 if (nla_put(skb, NDA_DST, tbl->key_len, pn->key))
2222 goto nla_put_failure;
2223
2224 nlmsg_end(skb, nlh);
2225 return 0;
2226
2227nla_put_failure:
2228 nlmsg_cancel(skb, nlh);
2229 return -EMSGSIZE;
2230}
2231
2232static void neigh_update_notify(struct neighbour *neigh)
2233{
2234 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
2235 __neigh_notify(neigh, RTM_NEWNEIGH, 0);
2236}
2237
2238static bool neigh_master_filtered(struct net_device *dev, int master_idx)
2239{
2240 struct net_device *master;
2241
2242 if (!master_idx)
2243 return false;
2244
2245 master = netdev_master_upper_dev_get(dev);
2246 if (!master || master->ifindex != master_idx)
2247 return true;
2248
2249 return false;
2250}
2251
2252static bool neigh_ifindex_filtered(struct net_device *dev, int filter_idx)
2253{
2254 if (filter_idx && dev->ifindex != filter_idx)
2255 return true;
2256
2257 return false;
2258}
2259
2260static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2261 struct netlink_callback *cb)
2262{
2263 struct net *net = sock_net(skb->sk);
2264 const struct nlmsghdr *nlh = cb->nlh;
2265 struct nlattr *tb[NDA_MAX + 1];
2266 struct neighbour *n;
2267 int rc, h, s_h = cb->args[1];
2268 int idx, s_idx = idx = cb->args[2];
2269 struct neigh_hash_table *nht;
2270 int filter_master_idx = 0, filter_idx = 0;
2271 unsigned int flags = NLM_F_MULTI;
2272 int err;
2273
2274 err = nlmsg_parse(nlh, sizeof(struct ndmsg), tb, NDA_MAX, NULL);
2275 if (!err) {
2276 if (tb[NDA_IFINDEX])
2277 filter_idx = nla_get_u32(tb[NDA_IFINDEX]);
2278
2279 if (tb[NDA_MASTER])
2280 filter_master_idx = nla_get_u32(tb[NDA_MASTER]);
2281
2282 if (filter_idx || filter_master_idx)
2283 flags |= NLM_F_DUMP_FILTERED;
2284 }
2285
2286 rcu_read_lock_bh();
2287 nht = rcu_dereference_bh(tbl->nht);
2288
2289 for (h = s_h; h < (1 << nht->hash_shift); h++) {
2290 if (h > s_h)
2291 s_idx = 0;
2292 for (n = rcu_dereference_bh(nht->hash_buckets[h]), idx = 0;
2293 n != NULL;
2294 n = rcu_dereference_bh(n->next)) {
2295 if (!net_eq(dev_net(n->dev), net))
2296 continue;
2297 if (neigh_ifindex_filtered(n->dev, filter_idx))
2298 continue;
2299 if (neigh_master_filtered(n->dev, filter_master_idx))
2300 continue;
2301 if (idx < s_idx)
2302 goto next;
2303 if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2304 cb->nlh->nlmsg_seq,
2305 RTM_NEWNEIGH,
2306 flags) < 0) {
2307 rc = -1;
2308 goto out;
2309 }
2310next:
2311 idx++;
2312 }
2313 }
2314 rc = skb->len;
2315out:
2316 rcu_read_unlock_bh();
2317 cb->args[1] = h;
2318 cb->args[2] = idx;
2319 return rc;
2320}
2321
2322static int pneigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2323 struct netlink_callback *cb)
2324{
2325 struct pneigh_entry *n;
2326 struct net *net = sock_net(skb->sk);
2327 int rc, h, s_h = cb->args[3];
2328 int idx, s_idx = idx = cb->args[4];
2329
2330 read_lock_bh(&tbl->lock);
2331
2332 for (h = s_h; h <= PNEIGH_HASHMASK; h++) {
2333 if (h > s_h)
2334 s_idx = 0;
2335 for (n = tbl->phash_buckets[h], idx = 0; n; n = n->next) {
2336 if (pneigh_net(n) != net)
2337 continue;
2338 if (idx < s_idx)
2339 goto next;
2340 if (pneigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2341 cb->nlh->nlmsg_seq,
2342 RTM_NEWNEIGH,
2343 NLM_F_MULTI, tbl) < 0) {
2344 read_unlock_bh(&tbl->lock);
2345 rc = -1;
2346 goto out;
2347 }
2348 next:
2349 idx++;
2350 }
2351 }
2352
2353 read_unlock_bh(&tbl->lock);
2354 rc = skb->len;
2355out:
2356 cb->args[3] = h;
2357 cb->args[4] = idx;
2358 return rc;
2359
2360}
2361
2362static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2363{
2364 struct neigh_table *tbl;
2365 int t, family, s_t;
2366 int proxy = 0;
2367 int err;
2368
2369 family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2370
2371 /* check for full ndmsg structure presence, family member is
2372 * the same for both structures
2373 */
2374 if (nlmsg_len(cb->nlh) >= sizeof(struct ndmsg) &&
2375 ((struct ndmsg *) nlmsg_data(cb->nlh))->ndm_flags == NTF_PROXY)
2376 proxy = 1;
2377
2378 s_t = cb->args[0];
2379
2380 for (t = 0; t < NEIGH_NR_TABLES; t++) {
2381 tbl = neigh_tables[t];
2382
2383 if (!tbl)
2384 continue;
2385 if (t < s_t || (family && tbl->family != family))
2386 continue;
2387 if (t > s_t)
2388 memset(&cb->args[1], 0, sizeof(cb->args) -
2389 sizeof(cb->args[0]));
2390 if (proxy)
2391 err = pneigh_dump_table(tbl, skb, cb);
2392 else
2393 err = neigh_dump_table(tbl, skb, cb);
2394 if (err < 0)
2395 break;
2396 }
2397
2398 cb->args[0] = t;
2399 return skb->len;
2400}
2401
2402void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
2403{
2404 int chain;
2405 struct neigh_hash_table *nht;
2406
2407 rcu_read_lock_bh();
2408 nht = rcu_dereference_bh(tbl->nht);
2409
2410 read_lock(&tbl->lock); /* avoid resizes */
2411 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2412 struct neighbour *n;
2413
2414 for (n = rcu_dereference_bh(nht->hash_buckets[chain]);
2415 n != NULL;
2416 n = rcu_dereference_bh(n->next))
2417 cb(n, cookie);
2418 }
2419 read_unlock(&tbl->lock);
2420 rcu_read_unlock_bh();
2421}
2422EXPORT_SYMBOL(neigh_for_each);
2423
2424/* The tbl->lock must be held as a writer and BH disabled. */
2425void __neigh_for_each_release(struct neigh_table *tbl,
2426 int (*cb)(struct neighbour *))
2427{
2428 int chain;
2429 struct neigh_hash_table *nht;
2430
2431 nht = rcu_dereference_protected(tbl->nht,
2432 lockdep_is_held(&tbl->lock));
2433 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2434 struct neighbour *n;
2435 struct neighbour __rcu **np;
2436
2437 np = &nht->hash_buckets[chain];
2438 while ((n = rcu_dereference_protected(*np,
2439 lockdep_is_held(&tbl->lock))) != NULL) {
2440 int release;
2441
2442 write_lock(&n->lock);
2443 release = cb(n);
2444 if (release) {
2445 rcu_assign_pointer(*np,
2446 rcu_dereference_protected(n->next,
2447 lockdep_is_held(&tbl->lock)));
2448 n->dead = 1;
2449 } else
2450 np = &n->next;
2451 write_unlock(&n->lock);
2452 if (release)
2453 neigh_cleanup_and_release(n);
2454 }
2455 }
2456}
2457EXPORT_SYMBOL(__neigh_for_each_release);
2458
2459int neigh_xmit(int index, struct net_device *dev,
2460 const void *addr, struct sk_buff *skb)
2461{
2462 int err = -EAFNOSUPPORT;
2463 if (likely(index < NEIGH_NR_TABLES)) {
2464 struct neigh_table *tbl;
2465 struct neighbour *neigh;
2466
2467 tbl = neigh_tables[index];
2468 if (!tbl)
2469 goto out;
2470 neigh = __neigh_lookup_noref(tbl, addr, dev);
2471 if (!neigh)
2472 neigh = __neigh_create(tbl, addr, dev, false);
2473 err = PTR_ERR(neigh);
2474 if (IS_ERR(neigh))
2475 goto out_kfree_skb;
2476 err = neigh->output(neigh, skb);
2477 }
2478 else if (index == NEIGH_LINK_TABLE) {
2479 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
2480 addr, NULL, skb->len);
2481 if (err < 0)
2482 goto out_kfree_skb;
2483 err = dev_queue_xmit(skb);
2484 }
2485out:
2486 return err;
2487out_kfree_skb:
2488 kfree_skb(skb);
2489 goto out;
2490}
2491EXPORT_SYMBOL(neigh_xmit);
2492
2493#ifdef CONFIG_PROC_FS
2494
2495static struct neighbour *neigh_get_first(struct seq_file *seq)
2496{
2497 struct neigh_seq_state *state = seq->private;
2498 struct net *net = seq_file_net(seq);
2499 struct neigh_hash_table *nht = state->nht;
2500 struct neighbour *n = NULL;
2501 int bucket = state->bucket;
2502
2503 state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
2504 for (bucket = 0; bucket < (1 << nht->hash_shift); bucket++) {
2505 n = rcu_dereference_bh(nht->hash_buckets[bucket]);
2506
2507 while (n) {
2508 if (!net_eq(dev_net(n->dev), net))
2509 goto next;
2510 if (state->neigh_sub_iter) {
2511 loff_t fakep = 0;
2512 void *v;
2513
2514 v = state->neigh_sub_iter(state, n, &fakep);
2515 if (!v)
2516 goto next;
2517 }
2518 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2519 break;
2520 if (n->nud_state & ~NUD_NOARP)
2521 break;
2522next:
2523 n = rcu_dereference_bh(n->next);
2524 }
2525
2526 if (n)
2527 break;
2528 }
2529 state->bucket = bucket;
2530
2531 return n;
2532}
2533
2534static struct neighbour *neigh_get_next(struct seq_file *seq,
2535 struct neighbour *n,
2536 loff_t *pos)
2537{
2538 struct neigh_seq_state *state = seq->private;
2539 struct net *net = seq_file_net(seq);
2540 struct neigh_hash_table *nht = state->nht;
2541
2542 if (state->neigh_sub_iter) {
2543 void *v = state->neigh_sub_iter(state, n, pos);
2544 if (v)
2545 return n;
2546 }
2547 n = rcu_dereference_bh(n->next);
2548
2549 while (1) {
2550 while (n) {
2551 if (!net_eq(dev_net(n->dev), net))
2552 goto next;
2553 if (state->neigh_sub_iter) {
2554 void *v = state->neigh_sub_iter(state, n, pos);
2555 if (v)
2556 return n;
2557 goto next;
2558 }
2559 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2560 break;
2561
2562 if (n->nud_state & ~NUD_NOARP)
2563 break;
2564next:
2565 n = rcu_dereference_bh(n->next);
2566 }
2567
2568 if (n)
2569 break;
2570
2571 if (++state->bucket >= (1 << nht->hash_shift))
2572 break;
2573
2574 n = rcu_dereference_bh(nht->hash_buckets[state->bucket]);
2575 }
2576
2577 if (n && pos)
2578 --(*pos);
2579 return n;
2580}
2581
2582static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
2583{
2584 struct neighbour *n = neigh_get_first(seq);
2585
2586 if (n) {
2587 --(*pos);
2588 while (*pos) {
2589 n = neigh_get_next(seq, n, pos);
2590 if (!n)
2591 break;
2592 }
2593 }
2594 return *pos ? NULL : n;
2595}
2596
2597static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
2598{
2599 struct neigh_seq_state *state = seq->private;
2600 struct net *net = seq_file_net(seq);
2601 struct neigh_table *tbl = state->tbl;
2602 struct pneigh_entry *pn = NULL;
2603 int bucket = state->bucket;
2604
2605 state->flags |= NEIGH_SEQ_IS_PNEIGH;
2606 for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
2607 pn = tbl->phash_buckets[bucket];
2608 while (pn && !net_eq(pneigh_net(pn), net))
2609 pn = pn->next;
2610 if (pn)
2611 break;
2612 }
2613 state->bucket = bucket;
2614
2615 return pn;
2616}
2617
2618static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
2619 struct pneigh_entry *pn,
2620 loff_t *pos)
2621{
2622 struct neigh_seq_state *state = seq->private;
2623 struct net *net = seq_file_net(seq);
2624 struct neigh_table *tbl = state->tbl;
2625
2626 do {
2627 pn = pn->next;
2628 } while (pn && !net_eq(pneigh_net(pn), net));
2629
2630 while (!pn) {
2631 if (++state->bucket > PNEIGH_HASHMASK)
2632 break;
2633 pn = tbl->phash_buckets[state->bucket];
2634 while (pn && !net_eq(pneigh_net(pn), net))
2635 pn = pn->next;
2636 if (pn)
2637 break;
2638 }
2639
2640 if (pn && pos)
2641 --(*pos);
2642
2643 return pn;
2644}
2645
2646static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
2647{
2648 struct pneigh_entry *pn = pneigh_get_first(seq);
2649
2650 if (pn) {
2651 --(*pos);
2652 while (*pos) {
2653 pn = pneigh_get_next(seq, pn, pos);
2654 if (!pn)
2655 break;
2656 }
2657 }
2658 return *pos ? NULL : pn;
2659}
2660
2661static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
2662{
2663 struct neigh_seq_state *state = seq->private;
2664 void *rc;
2665 loff_t idxpos = *pos;
2666
2667 rc = neigh_get_idx(seq, &idxpos);
2668 if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2669 rc = pneigh_get_idx(seq, &idxpos);
2670
2671 return rc;
2672}
2673
2674void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
2675 __acquires(rcu_bh)
2676{
2677 struct neigh_seq_state *state = seq->private;
2678
2679 state->tbl = tbl;
2680 state->bucket = 0;
2681 state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
2682
2683 rcu_read_lock_bh();
2684 state->nht = rcu_dereference_bh(tbl->nht);
2685
2686 return *pos ? neigh_get_idx_any(seq, pos) : SEQ_START_TOKEN;
2687}
2688EXPORT_SYMBOL(neigh_seq_start);
2689
2690void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2691{
2692 struct neigh_seq_state *state;
2693 void *rc;
2694
2695 if (v == SEQ_START_TOKEN) {
2696 rc = neigh_get_first(seq);
2697 goto out;
2698 }
2699
2700 state = seq->private;
2701 if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
2702 rc = neigh_get_next(seq, v, NULL);
2703 if (rc)
2704 goto out;
2705 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2706 rc = pneigh_get_first(seq);
2707 } else {
2708 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
2709 rc = pneigh_get_next(seq, v, NULL);
2710 }
2711out:
2712 ++(*pos);
2713 return rc;
2714}
2715EXPORT_SYMBOL(neigh_seq_next);
2716
2717void neigh_seq_stop(struct seq_file *seq, void *v)
2718 __releases(rcu_bh)
2719{
2720 rcu_read_unlock_bh();
2721}
2722EXPORT_SYMBOL(neigh_seq_stop);
2723
2724/* statistics via seq_file */
2725
2726static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
2727{
2728 struct neigh_table *tbl = seq->private;
2729 int cpu;
2730
2731 if (*pos == 0)
2732 return SEQ_START_TOKEN;
2733
2734 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
2735 if (!cpu_possible(cpu))
2736 continue;
2737 *pos = cpu+1;
2738 return per_cpu_ptr(tbl->stats, cpu);
2739 }
2740 return NULL;
2741}
2742
2743static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2744{
2745 struct neigh_table *tbl = seq->private;
2746 int cpu;
2747
2748 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
2749 if (!cpu_possible(cpu))
2750 continue;
2751 *pos = cpu+1;
2752 return per_cpu_ptr(tbl->stats, cpu);
2753 }
2754 return NULL;
2755}
2756
2757static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
2758{
2759
2760}
2761
2762static int neigh_stat_seq_show(struct seq_file *seq, void *v)
2763{
2764 struct neigh_table *tbl = seq->private;
2765 struct neigh_statistics *st = v;
2766
2767 if (v == SEQ_START_TOKEN) {
2768 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");
2769 return 0;
2770 }
2771
2772 seq_printf(seq, "%08x %08lx %08lx %08lx %08lx %08lx %08lx "
2773 "%08lx %08lx %08lx %08lx %08lx %08lx\n",
2774 atomic_read(&tbl->entries),
2775
2776 st->allocs,
2777 st->destroys,
2778 st->hash_grows,
2779
2780 st->lookups,
2781 st->hits,
2782
2783 st->res_failed,
2784
2785 st->rcv_probes_mcast,
2786 st->rcv_probes_ucast,
2787
2788 st->periodic_gc_runs,
2789 st->forced_gc_runs,
2790 st->unres_discards,
2791 st->table_fulls
2792 );
2793
2794 return 0;
2795}
2796
2797static const struct seq_operations neigh_stat_seq_ops = {
2798 .start = neigh_stat_seq_start,
2799 .next = neigh_stat_seq_next,
2800 .stop = neigh_stat_seq_stop,
2801 .show = neigh_stat_seq_show,
2802};
2803
2804static int neigh_stat_seq_open(struct inode *inode, struct file *file)
2805{
2806 int ret = seq_open(file, &neigh_stat_seq_ops);
2807
2808 if (!ret) {
2809 struct seq_file *sf = file->private_data;
2810 sf->private = PDE_DATA(inode);
2811 }
2812 return ret;
2813};
2814
2815static const struct file_operations neigh_stat_seq_fops = {
2816 .owner = THIS_MODULE,
2817 .open = neigh_stat_seq_open,
2818 .read = seq_read,
2819 .llseek = seq_lseek,
2820 .release = seq_release,
2821};
2822
2823#endif /* CONFIG_PROC_FS */
2824
2825static inline size_t neigh_nlmsg_size(void)
2826{
2827 return NLMSG_ALIGN(sizeof(struct ndmsg))
2828 + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2829 + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */
2830 + nla_total_size(sizeof(struct nda_cacheinfo))
2831 + nla_total_size(4); /* NDA_PROBES */
2832}
2833
2834static void __neigh_notify(struct neighbour *n, int type, int flags)
2835{
2836 struct net *net = dev_net(n->dev);
2837 struct sk_buff *skb;
2838 int err = -ENOBUFS;
2839
2840 skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC);
2841 if (skb == NULL)
2842 goto errout;
2843
2844 err = neigh_fill_info(skb, n, 0, 0, type, flags);
2845 if (err < 0) {
2846 /* -EMSGSIZE implies BUG in neigh_nlmsg_size() */
2847 WARN_ON(err == -EMSGSIZE);
2848 kfree_skb(skb);
2849 goto errout;
2850 }
2851 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
2852 return;
2853errout:
2854 if (err < 0)
2855 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
2856}
2857
2858void neigh_app_ns(struct neighbour *n)
2859{
2860 __neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST);
2861}
2862EXPORT_SYMBOL(neigh_app_ns);
2863
2864#ifdef CONFIG_SYSCTL
2865static int zero;
2866static int int_max = INT_MAX;
2867static int unres_qlen_max = INT_MAX / SKB_TRUESIZE(ETH_FRAME_LEN);
2868
2869static int proc_unres_qlen(struct ctl_table *ctl, int write,
2870 void __user *buffer, size_t *lenp, loff_t *ppos)
2871{
2872 int size, ret;
2873 struct ctl_table tmp = *ctl;
2874
2875 tmp.extra1 = &zero;
2876 tmp.extra2 = &unres_qlen_max;
2877 tmp.data = &size;
2878
2879 size = *(int *)ctl->data / SKB_TRUESIZE(ETH_FRAME_LEN);
2880 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2881
2882 if (write && !ret)
2883 *(int *)ctl->data = size * SKB_TRUESIZE(ETH_FRAME_LEN);
2884 return ret;
2885}
2886
2887static struct neigh_parms *neigh_get_dev_parms_rcu(struct net_device *dev,
2888 int family)
2889{
2890 switch (family) {
2891 case AF_INET:
2892 return __in_dev_arp_parms_get_rcu(dev);
2893 case AF_INET6:
2894 return __in6_dev_nd_parms_get_rcu(dev);
2895 }
2896 return NULL;
2897}
2898
2899static void neigh_copy_dflt_parms(struct net *net, struct neigh_parms *p,
2900 int index)
2901{
2902 struct net_device *dev;
2903 int family = neigh_parms_family(p);
2904
2905 rcu_read_lock();
2906 for_each_netdev_rcu(net, dev) {
2907 struct neigh_parms *dst_p =
2908 neigh_get_dev_parms_rcu(dev, family);
2909
2910 if (dst_p && !test_bit(index, dst_p->data_state))
2911 dst_p->data[index] = p->data[index];
2912 }
2913 rcu_read_unlock();
2914}
2915
2916static void neigh_proc_update(struct ctl_table *ctl, int write)
2917{
2918 struct net_device *dev = ctl->extra1;
2919 struct neigh_parms *p = ctl->extra2;
2920 struct net *net = neigh_parms_net(p);
2921 int index = (int *) ctl->data - p->data;
2922
2923 if (!write)
2924 return;
2925
2926 set_bit(index, p->data_state);
2927 if (!dev) /* NULL dev means this is default value */
2928 neigh_copy_dflt_parms(net, p, index);
2929}
2930
2931static int neigh_proc_dointvec_zero_intmax(struct ctl_table *ctl, int write,
2932 void __user *buffer,
2933 size_t *lenp, loff_t *ppos)
2934{
2935 struct ctl_table tmp = *ctl;
2936 int ret;
2937
2938 tmp.extra1 = &zero;
2939 tmp.extra2 = &int_max;
2940
2941 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2942 neigh_proc_update(ctl, write);
2943 return ret;
2944}
2945
2946int neigh_proc_dointvec(struct ctl_table *ctl, int write,
2947 void __user *buffer, size_t *lenp, loff_t *ppos)
2948{
2949 int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
2950
2951 neigh_proc_update(ctl, write);
2952 return ret;
2953}
2954EXPORT_SYMBOL(neigh_proc_dointvec);
2955
2956int neigh_proc_dointvec_jiffies(struct ctl_table *ctl, int write,
2957 void __user *buffer,
2958 size_t *lenp, loff_t *ppos)
2959{
2960 int ret = proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
2961
2962 neigh_proc_update(ctl, write);
2963 return ret;
2964}
2965EXPORT_SYMBOL(neigh_proc_dointvec_jiffies);
2966
2967static int neigh_proc_dointvec_userhz_jiffies(struct ctl_table *ctl, int write,
2968 void __user *buffer,
2969 size_t *lenp, loff_t *ppos)
2970{
2971 int ret = proc_dointvec_userhz_jiffies(ctl, write, buffer, lenp, ppos);
2972
2973 neigh_proc_update(ctl, write);
2974 return ret;
2975}
2976
2977int neigh_proc_dointvec_ms_jiffies(struct ctl_table *ctl, int write,
2978 void __user *buffer,
2979 size_t *lenp, loff_t *ppos)
2980{
2981 int ret = proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
2982
2983 neigh_proc_update(ctl, write);
2984 return ret;
2985}
2986EXPORT_SYMBOL(neigh_proc_dointvec_ms_jiffies);
2987
2988static int neigh_proc_dointvec_unres_qlen(struct ctl_table *ctl, int write,
2989 void __user *buffer,
2990 size_t *lenp, loff_t *ppos)
2991{
2992 int ret = proc_unres_qlen(ctl, write, buffer, lenp, ppos);
2993
2994 neigh_proc_update(ctl, write);
2995 return ret;
2996}
2997
2998static int neigh_proc_base_reachable_time(struct ctl_table *ctl, int write,
2999 void __user *buffer,
3000 size_t *lenp, loff_t *ppos)
3001{
3002 struct neigh_parms *p = ctl->extra2;
3003 int ret;
3004
3005 if (strcmp(ctl->procname, "base_reachable_time") == 0)
3006 ret = neigh_proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
3007 else if (strcmp(ctl->procname, "base_reachable_time_ms") == 0)
3008 ret = neigh_proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
3009 else
3010 ret = -1;
3011
3012 if (write && ret == 0) {
3013 /* update reachable_time as well, otherwise, the change will
3014 * only be effective after the next time neigh_periodic_work
3015 * decides to recompute it
3016 */
3017 p->reachable_time =
3018 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
3019 }
3020 return ret;
3021}
3022
3023#define NEIGH_PARMS_DATA_OFFSET(index) \
3024 (&((struct neigh_parms *) 0)->data[index])
3025
3026#define NEIGH_SYSCTL_ENTRY(attr, data_attr, name, mval, proc) \
3027 [NEIGH_VAR_ ## attr] = { \
3028 .procname = name, \
3029 .data = NEIGH_PARMS_DATA_OFFSET(NEIGH_VAR_ ## data_attr), \
3030 .maxlen = sizeof(int), \
3031 .mode = mval, \
3032 .proc_handler = proc, \
3033 }
3034
3035#define NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(attr, name) \
3036 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_zero_intmax)
3037
3038#define NEIGH_SYSCTL_JIFFIES_ENTRY(attr, name) \
3039 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_jiffies)
3040
3041#define NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(attr, name) \
3042 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_userhz_jiffies)
3043
3044#define NEIGH_SYSCTL_MS_JIFFIES_ENTRY(attr, name) \
3045 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_ms_jiffies)
3046
3047#define NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(attr, data_attr, name) \
3048 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_ms_jiffies)
3049
3050#define NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(attr, data_attr, name) \
3051 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_unres_qlen)
3052
3053static struct neigh_sysctl_table {
3054 struct ctl_table_header *sysctl_header;
3055 struct ctl_table neigh_vars[NEIGH_VAR_MAX + 1];
3056} neigh_sysctl_template __read_mostly = {
3057 .neigh_vars = {
3058 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_PROBES, "mcast_solicit"),
3059 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(UCAST_PROBES, "ucast_solicit"),
3060 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(APP_PROBES, "app_solicit"),
3061 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_REPROBES, "mcast_resolicit"),
3062 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(RETRANS_TIME, "retrans_time"),
3063 NEIGH_SYSCTL_JIFFIES_ENTRY(BASE_REACHABLE_TIME, "base_reachable_time"),
3064 NEIGH_SYSCTL_JIFFIES_ENTRY(DELAY_PROBE_TIME, "delay_first_probe_time"),
3065 NEIGH_SYSCTL_JIFFIES_ENTRY(GC_STALETIME, "gc_stale_time"),
3066 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(QUEUE_LEN_BYTES, "unres_qlen_bytes"),
3067 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(PROXY_QLEN, "proxy_qlen"),
3068 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(ANYCAST_DELAY, "anycast_delay"),
3069 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(PROXY_DELAY, "proxy_delay"),
3070 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(LOCKTIME, "locktime"),
3071 NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(QUEUE_LEN, QUEUE_LEN_BYTES, "unres_qlen"),
3072 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(RETRANS_TIME_MS, RETRANS_TIME, "retrans_time_ms"),
3073 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(BASE_REACHABLE_TIME_MS, BASE_REACHABLE_TIME, "base_reachable_time_ms"),
3074 [NEIGH_VAR_GC_INTERVAL] = {
3075 .procname = "gc_interval",
3076 .maxlen = sizeof(int),
3077 .mode = 0644,
3078 .proc_handler = proc_dointvec_jiffies,
3079 },
3080 [NEIGH_VAR_GC_THRESH1] = {
3081 .procname = "gc_thresh1",
3082 .maxlen = sizeof(int),
3083 .mode = 0644,
3084 .extra1 = &zero,
3085 .extra2 = &int_max,
3086 .proc_handler = proc_dointvec_minmax,
3087 },
3088 [NEIGH_VAR_GC_THRESH2] = {
3089 .procname = "gc_thresh2",
3090 .maxlen = sizeof(int),
3091 .mode = 0644,
3092 .extra1 = &zero,
3093 .extra2 = &int_max,
3094 .proc_handler = proc_dointvec_minmax,
3095 },
3096 [NEIGH_VAR_GC_THRESH3] = {
3097 .procname = "gc_thresh3",
3098 .maxlen = sizeof(int),
3099 .mode = 0644,
3100 .extra1 = &zero,
3101 .extra2 = &int_max,
3102 .proc_handler = proc_dointvec_minmax,
3103 },
3104 {},
3105 },
3106};
3107
3108int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
3109 proc_handler *handler)
3110{
3111 int i;
3112 struct neigh_sysctl_table *t;
3113 const char *dev_name_source;
3114 char neigh_path[ sizeof("net//neigh/") + IFNAMSIZ + IFNAMSIZ ];
3115 char *p_name;
3116
3117 t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL);
3118 if (!t)
3119 goto err;
3120
3121 for (i = 0; i < NEIGH_VAR_GC_INTERVAL; i++) {
3122 t->neigh_vars[i].data += (long) p;
3123 t->neigh_vars[i].extra1 = dev;
3124 t->neigh_vars[i].extra2 = p;
3125 }
3126
3127 if (dev) {
3128 dev_name_source = dev->name;
3129 /* Terminate the table early */
3130 memset(&t->neigh_vars[NEIGH_VAR_GC_INTERVAL], 0,
3131 sizeof(t->neigh_vars[NEIGH_VAR_GC_INTERVAL]));
3132 } else {
3133 struct neigh_table *tbl = p->tbl;
3134 dev_name_source = "default";
3135 t->neigh_vars[NEIGH_VAR_GC_INTERVAL].data = &tbl->gc_interval;
3136 t->neigh_vars[NEIGH_VAR_GC_THRESH1].data = &tbl->gc_thresh1;
3137 t->neigh_vars[NEIGH_VAR_GC_THRESH2].data = &tbl->gc_thresh2;
3138 t->neigh_vars[NEIGH_VAR_GC_THRESH3].data = &tbl->gc_thresh3;
3139 }
3140
3141 if (handler) {
3142 /* RetransTime */
3143 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].proc_handler = handler;
3144 /* ReachableTime */
3145 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler = handler;
3146 /* RetransTime (in milliseconds)*/
3147 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].proc_handler = handler;
3148 /* ReachableTime (in milliseconds) */
3149 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler = handler;
3150 } else {
3151 /* Those handlers will update p->reachable_time after
3152 * base_reachable_time(_ms) is set to ensure the new timer starts being
3153 * applied after the next neighbour update instead of waiting for
3154 * neigh_periodic_work to update its value (can be multiple minutes)
3155 * So any handler that replaces them should do this as well
3156 */
3157 /* ReachableTime */
3158 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler =
3159 neigh_proc_base_reachable_time;
3160 /* ReachableTime (in milliseconds) */
3161 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler =
3162 neigh_proc_base_reachable_time;
3163 }
3164
3165 /* Don't export sysctls to unprivileged users */
3166 if (neigh_parms_net(p)->user_ns != &init_user_ns)
3167 t->neigh_vars[0].procname = NULL;
3168
3169 switch (neigh_parms_family(p)) {
3170 case AF_INET:
3171 p_name = "ipv4";
3172 break;
3173 case AF_INET6:
3174 p_name = "ipv6";
3175 break;
3176 default:
3177 BUG();
3178 }
3179
3180 snprintf(neigh_path, sizeof(neigh_path), "net/%s/neigh/%s",
3181 p_name, dev_name_source);
3182 t->sysctl_header =
3183 register_net_sysctl(neigh_parms_net(p), neigh_path, t->neigh_vars);
3184 if (!t->sysctl_header)
3185 goto free;
3186
3187 p->sysctl_table = t;
3188 return 0;
3189
3190free:
3191 kfree(t);
3192err:
3193 return -ENOBUFS;
3194}
3195EXPORT_SYMBOL(neigh_sysctl_register);
3196
3197void neigh_sysctl_unregister(struct neigh_parms *p)
3198{
3199 if (p->sysctl_table) {
3200 struct neigh_sysctl_table *t = p->sysctl_table;
3201 p->sysctl_table = NULL;
3202 unregister_net_sysctl_table(t->sysctl_header);
3203 kfree(t);
3204 }
3205}
3206EXPORT_SYMBOL(neigh_sysctl_unregister);
3207
3208#endif /* CONFIG_SYSCTL */
3209
3210static int __init neigh_init(void)
3211{
3212 rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL, NULL);
3213 rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL, NULL);
3214 rtnl_register(PF_UNSPEC, RTM_GETNEIGH, NULL, neigh_dump_info, NULL);
3215
3216 rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info,
3217 NULL);
3218 rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL, NULL);
3219
3220 return 0;
3221}
3222
3223subsys_initcall(neigh_init);
3224
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 trace_neigh_cleanup_and_release(neigh, 0);
102 __neigh_notify(neigh, RTM_DELNEIGH, 0, 0);
103 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
104 neigh_release(neigh);
105}
106
107/*
108 * It is random distribution in the interval (1/2)*base...(3/2)*base.
109 * It corresponds to default IPv6 settings and is not overridable,
110 * because it is really reasonable choice.
111 */
112
113unsigned long neigh_rand_reach_time(unsigned long base)
114{
115 return base ? (prandom_u32() % base) + (base >> 1) : 0;
116}
117EXPORT_SYMBOL(neigh_rand_reach_time);
118
119static void neigh_mark_dead(struct neighbour *n)
120{
121 n->dead = 1;
122 if (!list_empty(&n->gc_list)) {
123 list_del_init(&n->gc_list);
124 atomic_dec(&n->tbl->gc_entries);
125 }
126}
127
128static void neigh_update_gc_list(struct neighbour *n)
129{
130 bool on_gc_list, exempt_from_gc;
131
132 write_lock_bh(&n->tbl->lock);
133 write_lock(&n->lock);
134
135 /* remove from the gc list if new state is permanent or if neighbor
136 * is externally learned; otherwise entry should be on the gc list
137 */
138 exempt_from_gc = n->nud_state & NUD_PERMANENT ||
139 n->flags & NTF_EXT_LEARNED;
140 on_gc_list = !list_empty(&n->gc_list);
141
142 if (exempt_from_gc && on_gc_list) {
143 list_del_init(&n->gc_list);
144 atomic_dec(&n->tbl->gc_entries);
145 } else if (!exempt_from_gc && !on_gc_list) {
146 /* add entries to the tail; cleaning removes from the front */
147 list_add_tail(&n->gc_list, &n->tbl->gc_list);
148 atomic_inc(&n->tbl->gc_entries);
149 }
150
151 write_unlock(&n->lock);
152 write_unlock_bh(&n->tbl->lock);
153}
154
155static bool neigh_update_ext_learned(struct neighbour *neigh, u32 flags,
156 int *notify)
157{
158 bool rc = false;
159 u8 ndm_flags;
160
161 if (!(flags & NEIGH_UPDATE_F_ADMIN))
162 return rc;
163
164 ndm_flags = (flags & NEIGH_UPDATE_F_EXT_LEARNED) ? NTF_EXT_LEARNED : 0;
165 if ((neigh->flags ^ ndm_flags) & NTF_EXT_LEARNED) {
166 if (ndm_flags & NTF_EXT_LEARNED)
167 neigh->flags |= NTF_EXT_LEARNED;
168 else
169 neigh->flags &= ~NTF_EXT_LEARNED;
170 rc = true;
171 *notify = 1;
172 }
173
174 return rc;
175}
176
177static bool neigh_del(struct neighbour *n, struct neighbour __rcu **np,
178 struct neigh_table *tbl)
179{
180 bool retval = false;
181
182 write_lock(&n->lock);
183 if (refcount_read(&n->refcnt) == 1) {
184 struct neighbour *neigh;
185
186 neigh = rcu_dereference_protected(n->next,
187 lockdep_is_held(&tbl->lock));
188 rcu_assign_pointer(*np, neigh);
189 neigh_mark_dead(n);
190 retval = true;
191 }
192 write_unlock(&n->lock);
193 if (retval)
194 neigh_cleanup_and_release(n);
195 return retval;
196}
197
198bool neigh_remove_one(struct neighbour *ndel, struct neigh_table *tbl)
199{
200 struct neigh_hash_table *nht;
201 void *pkey = ndel->primary_key;
202 u32 hash_val;
203 struct neighbour *n;
204 struct neighbour __rcu **np;
205
206 nht = rcu_dereference_protected(tbl->nht,
207 lockdep_is_held(&tbl->lock));
208 hash_val = tbl->hash(pkey, ndel->dev, nht->hash_rnd);
209 hash_val = hash_val >> (32 - nht->hash_shift);
210
211 np = &nht->hash_buckets[hash_val];
212 while ((n = rcu_dereference_protected(*np,
213 lockdep_is_held(&tbl->lock)))) {
214 if (n == ndel)
215 return neigh_del(n, np, tbl);
216 np = &n->next;
217 }
218 return false;
219}
220
221static int neigh_forced_gc(struct neigh_table *tbl)
222{
223 int max_clean = atomic_read(&tbl->gc_entries) - tbl->gc_thresh2;
224 unsigned long tref = jiffies - 5 * HZ;
225 struct neighbour *n, *tmp;
226 int shrunk = 0;
227
228 NEIGH_CACHE_STAT_INC(tbl, forced_gc_runs);
229
230 write_lock_bh(&tbl->lock);
231
232 list_for_each_entry_safe(n, tmp, &tbl->gc_list, gc_list) {
233 if (refcount_read(&n->refcnt) == 1) {
234 bool remove = false;
235
236 write_lock(&n->lock);
237 if ((n->nud_state == NUD_FAILED) ||
238 time_after(tref, n->updated))
239 remove = true;
240 write_unlock(&n->lock);
241
242 if (remove && neigh_remove_one(n, tbl))
243 shrunk++;
244 if (shrunk >= max_clean)
245 break;
246 }
247 }
248
249 tbl->last_flush = jiffies;
250
251 write_unlock_bh(&tbl->lock);
252
253 return shrunk;
254}
255
256static void neigh_add_timer(struct neighbour *n, unsigned long when)
257{
258 neigh_hold(n);
259 if (unlikely(mod_timer(&n->timer, when))) {
260 printk("NEIGH: BUG, double timer add, state is %x\n",
261 n->nud_state);
262 dump_stack();
263 }
264}
265
266static int neigh_del_timer(struct neighbour *n)
267{
268 if ((n->nud_state & NUD_IN_TIMER) &&
269 del_timer(&n->timer)) {
270 neigh_release(n);
271 return 1;
272 }
273 return 0;
274}
275
276static void pneigh_queue_purge(struct sk_buff_head *list)
277{
278 struct sk_buff *skb;
279
280 while ((skb = skb_dequeue(list)) != NULL) {
281 dev_put(skb->dev);
282 kfree_skb(skb);
283 }
284}
285
286static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev,
287 bool skip_perm)
288{
289 int i;
290 struct neigh_hash_table *nht;
291
292 nht = rcu_dereference_protected(tbl->nht,
293 lockdep_is_held(&tbl->lock));
294
295 for (i = 0; i < (1 << nht->hash_shift); i++) {
296 struct neighbour *n;
297 struct neighbour __rcu **np = &nht->hash_buckets[i];
298
299 while ((n = rcu_dereference_protected(*np,
300 lockdep_is_held(&tbl->lock))) != NULL) {
301 if (dev && n->dev != dev) {
302 np = &n->next;
303 continue;
304 }
305 if (skip_perm && n->nud_state & NUD_PERMANENT) {
306 np = &n->next;
307 continue;
308 }
309 rcu_assign_pointer(*np,
310 rcu_dereference_protected(n->next,
311 lockdep_is_held(&tbl->lock)));
312 write_lock(&n->lock);
313 neigh_del_timer(n);
314 neigh_mark_dead(n);
315 if (refcount_read(&n->refcnt) != 1) {
316 /* The most unpleasant situation.
317 We must destroy neighbour entry,
318 but someone still uses it.
319
320 The destroy will be delayed until
321 the last user releases us, but
322 we must kill timers etc. and move
323 it to safe state.
324 */
325 __skb_queue_purge(&n->arp_queue);
326 n->arp_queue_len_bytes = 0;
327 n->output = neigh_blackhole;
328 if (n->nud_state & NUD_VALID)
329 n->nud_state = NUD_NOARP;
330 else
331 n->nud_state = NUD_NONE;
332 neigh_dbg(2, "neigh %p is stray\n", n);
333 }
334 write_unlock(&n->lock);
335 neigh_cleanup_and_release(n);
336 }
337 }
338}
339
340void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev)
341{
342 write_lock_bh(&tbl->lock);
343 neigh_flush_dev(tbl, dev, false);
344 write_unlock_bh(&tbl->lock);
345}
346EXPORT_SYMBOL(neigh_changeaddr);
347
348static int __neigh_ifdown(struct neigh_table *tbl, struct net_device *dev,
349 bool skip_perm)
350{
351 write_lock_bh(&tbl->lock);
352 neigh_flush_dev(tbl, dev, skip_perm);
353 pneigh_ifdown_and_unlock(tbl, dev);
354
355 del_timer_sync(&tbl->proxy_timer);
356 pneigh_queue_purge(&tbl->proxy_queue);
357 return 0;
358}
359
360int neigh_carrier_down(struct neigh_table *tbl, struct net_device *dev)
361{
362 __neigh_ifdown(tbl, dev, true);
363 return 0;
364}
365EXPORT_SYMBOL(neigh_carrier_down);
366
367int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
368{
369 __neigh_ifdown(tbl, dev, false);
370 return 0;
371}
372EXPORT_SYMBOL(neigh_ifdown);
373
374static struct neighbour *neigh_alloc(struct neigh_table *tbl,
375 struct net_device *dev,
376 bool exempt_from_gc)
377{
378 struct neighbour *n = NULL;
379 unsigned long now = jiffies;
380 int entries;
381
382 if (exempt_from_gc)
383 goto do_alloc;
384
385 entries = atomic_inc_return(&tbl->gc_entries) - 1;
386 if (entries >= tbl->gc_thresh3 ||
387 (entries >= tbl->gc_thresh2 &&
388 time_after(now, tbl->last_flush + 5 * HZ))) {
389 if (!neigh_forced_gc(tbl) &&
390 entries >= tbl->gc_thresh3) {
391 net_info_ratelimited("%s: neighbor table overflow!\n",
392 tbl->id);
393 NEIGH_CACHE_STAT_INC(tbl, table_fulls);
394 goto out_entries;
395 }
396 }
397
398do_alloc:
399 n = kzalloc(tbl->entry_size + dev->neigh_priv_len, GFP_ATOMIC);
400 if (!n)
401 goto out_entries;
402
403 __skb_queue_head_init(&n->arp_queue);
404 rwlock_init(&n->lock);
405 seqlock_init(&n->ha_lock);
406 n->updated = n->used = now;
407 n->nud_state = NUD_NONE;
408 n->output = neigh_blackhole;
409 seqlock_init(&n->hh.hh_lock);
410 n->parms = neigh_parms_clone(&tbl->parms);
411 timer_setup(&n->timer, neigh_timer_handler, 0);
412
413 NEIGH_CACHE_STAT_INC(tbl, allocs);
414 n->tbl = tbl;
415 refcount_set(&n->refcnt, 1);
416 n->dead = 1;
417 INIT_LIST_HEAD(&n->gc_list);
418
419 atomic_inc(&tbl->entries);
420out:
421 return n;
422
423out_entries:
424 if (!exempt_from_gc)
425 atomic_dec(&tbl->gc_entries);
426 goto out;
427}
428
429static void neigh_get_hash_rnd(u32 *x)
430{
431 *x = get_random_u32() | 1;
432}
433
434static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift)
435{
436 size_t size = (1 << shift) * sizeof(struct neighbour *);
437 struct neigh_hash_table *ret;
438 struct neighbour __rcu **buckets;
439 int i;
440
441 ret = kmalloc(sizeof(*ret), GFP_ATOMIC);
442 if (!ret)
443 return NULL;
444 if (size <= PAGE_SIZE) {
445 buckets = kzalloc(size, GFP_ATOMIC);
446 } else {
447 buckets = (struct neighbour __rcu **)
448 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
449 get_order(size));
450 kmemleak_alloc(buckets, size, 1, GFP_ATOMIC);
451 }
452 if (!buckets) {
453 kfree(ret);
454 return NULL;
455 }
456 ret->hash_buckets = buckets;
457 ret->hash_shift = shift;
458 for (i = 0; i < NEIGH_NUM_HASH_RND; i++)
459 neigh_get_hash_rnd(&ret->hash_rnd[i]);
460 return ret;
461}
462
463static void neigh_hash_free_rcu(struct rcu_head *head)
464{
465 struct neigh_hash_table *nht = container_of(head,
466 struct neigh_hash_table,
467 rcu);
468 size_t size = (1 << nht->hash_shift) * sizeof(struct neighbour *);
469 struct neighbour __rcu **buckets = nht->hash_buckets;
470
471 if (size <= PAGE_SIZE) {
472 kfree(buckets);
473 } else {
474 kmemleak_free(buckets);
475 free_pages((unsigned long)buckets, get_order(size));
476 }
477 kfree(nht);
478}
479
480static struct neigh_hash_table *neigh_hash_grow(struct neigh_table *tbl,
481 unsigned long new_shift)
482{
483 unsigned int i, hash;
484 struct neigh_hash_table *new_nht, *old_nht;
485
486 NEIGH_CACHE_STAT_INC(tbl, hash_grows);
487
488 old_nht = rcu_dereference_protected(tbl->nht,
489 lockdep_is_held(&tbl->lock));
490 new_nht = neigh_hash_alloc(new_shift);
491 if (!new_nht)
492 return old_nht;
493
494 for (i = 0; i < (1 << old_nht->hash_shift); i++) {
495 struct neighbour *n, *next;
496
497 for (n = rcu_dereference_protected(old_nht->hash_buckets[i],
498 lockdep_is_held(&tbl->lock));
499 n != NULL;
500 n = next) {
501 hash = tbl->hash(n->primary_key, n->dev,
502 new_nht->hash_rnd);
503
504 hash >>= (32 - new_nht->hash_shift);
505 next = rcu_dereference_protected(n->next,
506 lockdep_is_held(&tbl->lock));
507
508 rcu_assign_pointer(n->next,
509 rcu_dereference_protected(
510 new_nht->hash_buckets[hash],
511 lockdep_is_held(&tbl->lock)));
512 rcu_assign_pointer(new_nht->hash_buckets[hash], n);
513 }
514 }
515
516 rcu_assign_pointer(tbl->nht, new_nht);
517 call_rcu(&old_nht->rcu, neigh_hash_free_rcu);
518 return new_nht;
519}
520
521struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
522 struct net_device *dev)
523{
524 struct neighbour *n;
525
526 NEIGH_CACHE_STAT_INC(tbl, lookups);
527
528 rcu_read_lock_bh();
529 n = __neigh_lookup_noref(tbl, pkey, dev);
530 if (n) {
531 if (!refcount_inc_not_zero(&n->refcnt))
532 n = NULL;
533 NEIGH_CACHE_STAT_INC(tbl, hits);
534 }
535
536 rcu_read_unlock_bh();
537 return n;
538}
539EXPORT_SYMBOL(neigh_lookup);
540
541struct neighbour *neigh_lookup_nodev(struct neigh_table *tbl, struct net *net,
542 const void *pkey)
543{
544 struct neighbour *n;
545 unsigned int key_len = tbl->key_len;
546 u32 hash_val;
547 struct neigh_hash_table *nht;
548
549 NEIGH_CACHE_STAT_INC(tbl, lookups);
550
551 rcu_read_lock_bh();
552 nht = rcu_dereference_bh(tbl->nht);
553 hash_val = tbl->hash(pkey, NULL, nht->hash_rnd) >> (32 - nht->hash_shift);
554
555 for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]);
556 n != NULL;
557 n = rcu_dereference_bh(n->next)) {
558 if (!memcmp(n->primary_key, pkey, key_len) &&
559 net_eq(dev_net(n->dev), net)) {
560 if (!refcount_inc_not_zero(&n->refcnt))
561 n = NULL;
562 NEIGH_CACHE_STAT_INC(tbl, hits);
563 break;
564 }
565 }
566
567 rcu_read_unlock_bh();
568 return n;
569}
570EXPORT_SYMBOL(neigh_lookup_nodev);
571
572static struct neighbour *___neigh_create(struct neigh_table *tbl,
573 const void *pkey,
574 struct net_device *dev,
575 bool exempt_from_gc, bool want_ref)
576{
577 struct neighbour *n1, *rc, *n = neigh_alloc(tbl, dev, exempt_from_gc);
578 u32 hash_val;
579 unsigned int key_len = tbl->key_len;
580 int error;
581 struct neigh_hash_table *nht;
582
583 trace_neigh_create(tbl, dev, pkey, n, exempt_from_gc);
584
585 if (!n) {
586 rc = ERR_PTR(-ENOBUFS);
587 goto out;
588 }
589
590 memcpy(n->primary_key, pkey, key_len);
591 n->dev = dev;
592 dev_hold(dev);
593
594 /* Protocol specific setup. */
595 if (tbl->constructor && (error = tbl->constructor(n)) < 0) {
596 rc = ERR_PTR(error);
597 goto out_neigh_release;
598 }
599
600 if (dev->netdev_ops->ndo_neigh_construct) {
601 error = dev->netdev_ops->ndo_neigh_construct(dev, n);
602 if (error < 0) {
603 rc = ERR_PTR(error);
604 goto out_neigh_release;
605 }
606 }
607
608 /* Device specific setup. */
609 if (n->parms->neigh_setup &&
610 (error = n->parms->neigh_setup(n)) < 0) {
611 rc = ERR_PTR(error);
612 goto out_neigh_release;
613 }
614
615 n->confirmed = jiffies - (NEIGH_VAR(n->parms, BASE_REACHABLE_TIME) << 1);
616
617 write_lock_bh(&tbl->lock);
618 nht = rcu_dereference_protected(tbl->nht,
619 lockdep_is_held(&tbl->lock));
620
621 if (atomic_read(&tbl->entries) > (1 << nht->hash_shift))
622 nht = neigh_hash_grow(tbl, nht->hash_shift + 1);
623
624 hash_val = tbl->hash(n->primary_key, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
625
626 if (n->parms->dead) {
627 rc = ERR_PTR(-EINVAL);
628 goto out_tbl_unlock;
629 }
630
631 for (n1 = rcu_dereference_protected(nht->hash_buckets[hash_val],
632 lockdep_is_held(&tbl->lock));
633 n1 != NULL;
634 n1 = rcu_dereference_protected(n1->next,
635 lockdep_is_held(&tbl->lock))) {
636 if (dev == n1->dev && !memcmp(n1->primary_key, n->primary_key, key_len)) {
637 if (want_ref)
638 neigh_hold(n1);
639 rc = n1;
640 goto out_tbl_unlock;
641 }
642 }
643
644 n->dead = 0;
645 if (!exempt_from_gc)
646 list_add_tail(&n->gc_list, &n->tbl->gc_list);
647
648 if (want_ref)
649 neigh_hold(n);
650 rcu_assign_pointer(n->next,
651 rcu_dereference_protected(nht->hash_buckets[hash_val],
652 lockdep_is_held(&tbl->lock)));
653 rcu_assign_pointer(nht->hash_buckets[hash_val], n);
654 write_unlock_bh(&tbl->lock);
655 neigh_dbg(2, "neigh %p is created\n", n);
656 rc = n;
657out:
658 return rc;
659out_tbl_unlock:
660 write_unlock_bh(&tbl->lock);
661out_neigh_release:
662 if (!exempt_from_gc)
663 atomic_dec(&tbl->gc_entries);
664 neigh_release(n);
665 goto out;
666}
667
668struct neighbour *__neigh_create(struct neigh_table *tbl, const void *pkey,
669 struct net_device *dev, bool want_ref)
670{
671 return ___neigh_create(tbl, pkey, dev, false, want_ref);
672}
673EXPORT_SYMBOL(__neigh_create);
674
675static u32 pneigh_hash(const void *pkey, unsigned int key_len)
676{
677 u32 hash_val = *(u32 *)(pkey + key_len - 4);
678 hash_val ^= (hash_val >> 16);
679 hash_val ^= hash_val >> 8;
680 hash_val ^= hash_val >> 4;
681 hash_val &= PNEIGH_HASHMASK;
682 return hash_val;
683}
684
685static struct pneigh_entry *__pneigh_lookup_1(struct pneigh_entry *n,
686 struct net *net,
687 const void *pkey,
688 unsigned int key_len,
689 struct net_device *dev)
690{
691 while (n) {
692 if (!memcmp(n->key, pkey, key_len) &&
693 net_eq(pneigh_net(n), net) &&
694 (n->dev == dev || !n->dev))
695 return n;
696 n = n->next;
697 }
698 return NULL;
699}
700
701struct pneigh_entry *__pneigh_lookup(struct neigh_table *tbl,
702 struct net *net, const void *pkey, struct net_device *dev)
703{
704 unsigned int key_len = tbl->key_len;
705 u32 hash_val = pneigh_hash(pkey, key_len);
706
707 return __pneigh_lookup_1(tbl->phash_buckets[hash_val],
708 net, pkey, key_len, dev);
709}
710EXPORT_SYMBOL_GPL(__pneigh_lookup);
711
712struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl,
713 struct net *net, const void *pkey,
714 struct net_device *dev, int creat)
715{
716 struct pneigh_entry *n;
717 unsigned int key_len = tbl->key_len;
718 u32 hash_val = pneigh_hash(pkey, key_len);
719
720 read_lock_bh(&tbl->lock);
721 n = __pneigh_lookup_1(tbl->phash_buckets[hash_val],
722 net, pkey, key_len, dev);
723 read_unlock_bh(&tbl->lock);
724
725 if (n || !creat)
726 goto out;
727
728 ASSERT_RTNL();
729
730 n = kmalloc(sizeof(*n) + key_len, GFP_KERNEL);
731 if (!n)
732 goto out;
733
734 n->protocol = 0;
735 write_pnet(&n->net, net);
736 memcpy(n->key, pkey, key_len);
737 n->dev = dev;
738 if (dev)
739 dev_hold(dev);
740
741 if (tbl->pconstructor && tbl->pconstructor(n)) {
742 if (dev)
743 dev_put(dev);
744 kfree(n);
745 n = NULL;
746 goto out;
747 }
748
749 write_lock_bh(&tbl->lock);
750 n->next = tbl->phash_buckets[hash_val];
751 tbl->phash_buckets[hash_val] = n;
752 write_unlock_bh(&tbl->lock);
753out:
754 return n;
755}
756EXPORT_SYMBOL(pneigh_lookup);
757
758
759int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
760 struct net_device *dev)
761{
762 struct pneigh_entry *n, **np;
763 unsigned int key_len = tbl->key_len;
764 u32 hash_val = pneigh_hash(pkey, key_len);
765
766 write_lock_bh(&tbl->lock);
767 for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
768 np = &n->next) {
769 if (!memcmp(n->key, pkey, key_len) && n->dev == dev &&
770 net_eq(pneigh_net(n), net)) {
771 *np = n->next;
772 write_unlock_bh(&tbl->lock);
773 if (tbl->pdestructor)
774 tbl->pdestructor(n);
775 if (n->dev)
776 dev_put(n->dev);
777 kfree(n);
778 return 0;
779 }
780 }
781 write_unlock_bh(&tbl->lock);
782 return -ENOENT;
783}
784
785static int pneigh_ifdown_and_unlock(struct neigh_table *tbl,
786 struct net_device *dev)
787{
788 struct pneigh_entry *n, **np, *freelist = NULL;
789 u32 h;
790
791 for (h = 0; h <= PNEIGH_HASHMASK; h++) {
792 np = &tbl->phash_buckets[h];
793 while ((n = *np) != NULL) {
794 if (!dev || n->dev == dev) {
795 *np = n->next;
796 n->next = freelist;
797 freelist = n;
798 continue;
799 }
800 np = &n->next;
801 }
802 }
803 write_unlock_bh(&tbl->lock);
804 while ((n = freelist)) {
805 freelist = n->next;
806 n->next = NULL;
807 if (tbl->pdestructor)
808 tbl->pdestructor(n);
809 if (n->dev)
810 dev_put(n->dev);
811 kfree(n);
812 }
813 return -ENOENT;
814}
815
816static void neigh_parms_destroy(struct neigh_parms *parms);
817
818static inline void neigh_parms_put(struct neigh_parms *parms)
819{
820 if (refcount_dec_and_test(&parms->refcnt))
821 neigh_parms_destroy(parms);
822}
823
824/*
825 * neighbour must already be out of the table;
826 *
827 */
828void neigh_destroy(struct neighbour *neigh)
829{
830 struct net_device *dev = neigh->dev;
831
832 NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
833
834 if (!neigh->dead) {
835 pr_warn("Destroying alive neighbour %p\n", neigh);
836 dump_stack();
837 return;
838 }
839
840 if (neigh_del_timer(neigh))
841 pr_warn("Impossible event\n");
842
843 write_lock_bh(&neigh->lock);
844 __skb_queue_purge(&neigh->arp_queue);
845 write_unlock_bh(&neigh->lock);
846 neigh->arp_queue_len_bytes = 0;
847
848 if (dev->netdev_ops->ndo_neigh_destroy)
849 dev->netdev_ops->ndo_neigh_destroy(dev, neigh);
850
851 dev_put(dev);
852 neigh_parms_put(neigh->parms);
853
854 neigh_dbg(2, "neigh %p is destroyed\n", neigh);
855
856 atomic_dec(&neigh->tbl->entries);
857 kfree_rcu(neigh, rcu);
858}
859EXPORT_SYMBOL(neigh_destroy);
860
861/* Neighbour state is suspicious;
862 disable fast path.
863
864 Called with write_locked neigh.
865 */
866static void neigh_suspect(struct neighbour *neigh)
867{
868 neigh_dbg(2, "neigh %p is suspected\n", neigh);
869
870 neigh->output = neigh->ops->output;
871}
872
873/* Neighbour state is OK;
874 enable fast path.
875
876 Called with write_locked neigh.
877 */
878static void neigh_connect(struct neighbour *neigh)
879{
880 neigh_dbg(2, "neigh %p is connected\n", neigh);
881
882 neigh->output = neigh->ops->connected_output;
883}
884
885static void neigh_periodic_work(struct work_struct *work)
886{
887 struct neigh_table *tbl = container_of(work, struct neigh_table, gc_work.work);
888 struct neighbour *n;
889 struct neighbour __rcu **np;
890 unsigned int i;
891 struct neigh_hash_table *nht;
892
893 NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
894
895 write_lock_bh(&tbl->lock);
896 nht = rcu_dereference_protected(tbl->nht,
897 lockdep_is_held(&tbl->lock));
898
899 /*
900 * periodically recompute ReachableTime from random function
901 */
902
903 if (time_after(jiffies, tbl->last_rand + 300 * HZ)) {
904 struct neigh_parms *p;
905 tbl->last_rand = jiffies;
906 list_for_each_entry(p, &tbl->parms_list, list)
907 p->reachable_time =
908 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
909 }
910
911 if (atomic_read(&tbl->entries) < tbl->gc_thresh1)
912 goto out;
913
914 for (i = 0 ; i < (1 << nht->hash_shift); i++) {
915 np = &nht->hash_buckets[i];
916
917 while ((n = rcu_dereference_protected(*np,
918 lockdep_is_held(&tbl->lock))) != NULL) {
919 unsigned int state;
920
921 write_lock(&n->lock);
922
923 state = n->nud_state;
924 if ((state & (NUD_PERMANENT | NUD_IN_TIMER)) ||
925 (n->flags & NTF_EXT_LEARNED)) {
926 write_unlock(&n->lock);
927 goto next_elt;
928 }
929
930 if (time_before(n->used, n->confirmed))
931 n->used = n->confirmed;
932
933 if (refcount_read(&n->refcnt) == 1 &&
934 (state == NUD_FAILED ||
935 time_after(jiffies, n->used + NEIGH_VAR(n->parms, GC_STALETIME)))) {
936 *np = n->next;
937 neigh_mark_dead(n);
938 write_unlock(&n->lock);
939 neigh_cleanup_and_release(n);
940 continue;
941 }
942 write_unlock(&n->lock);
943
944next_elt:
945 np = &n->next;
946 }
947 /*
948 * It's fine to release lock here, even if hash table
949 * grows while we are preempted.
950 */
951 write_unlock_bh(&tbl->lock);
952 cond_resched();
953 write_lock_bh(&tbl->lock);
954 nht = rcu_dereference_protected(tbl->nht,
955 lockdep_is_held(&tbl->lock));
956 }
957out:
958 /* Cycle through all hash buckets every BASE_REACHABLE_TIME/2 ticks.
959 * ARP entry timeouts range from 1/2 BASE_REACHABLE_TIME to 3/2
960 * BASE_REACHABLE_TIME.
961 */
962 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
963 NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME) >> 1);
964 write_unlock_bh(&tbl->lock);
965}
966
967static __inline__ int neigh_max_probes(struct neighbour *n)
968{
969 struct neigh_parms *p = n->parms;
970 return NEIGH_VAR(p, UCAST_PROBES) + NEIGH_VAR(p, APP_PROBES) +
971 (n->nud_state & NUD_PROBE ? NEIGH_VAR(p, MCAST_REPROBES) :
972 NEIGH_VAR(p, MCAST_PROBES));
973}
974
975static void neigh_invalidate(struct neighbour *neigh)
976 __releases(neigh->lock)
977 __acquires(neigh->lock)
978{
979 struct sk_buff *skb;
980
981 NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
982 neigh_dbg(2, "neigh %p is failed\n", neigh);
983 neigh->updated = jiffies;
984
985 /* It is very thin place. report_unreachable is very complicated
986 routine. Particularly, it can hit the same neighbour entry!
987
988 So that, we try to be accurate and avoid dead loop. --ANK
989 */
990 while (neigh->nud_state == NUD_FAILED &&
991 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
992 write_unlock(&neigh->lock);
993 neigh->ops->error_report(neigh, skb);
994 write_lock(&neigh->lock);
995 }
996 __skb_queue_purge(&neigh->arp_queue);
997 neigh->arp_queue_len_bytes = 0;
998}
999
1000static void neigh_probe(struct neighbour *neigh)
1001 __releases(neigh->lock)
1002{
1003 struct sk_buff *skb = skb_peek_tail(&neigh->arp_queue);
1004 /* keep skb alive even if arp_queue overflows */
1005 if (skb)
1006 skb = skb_clone(skb, GFP_ATOMIC);
1007 write_unlock(&neigh->lock);
1008 if (neigh->ops->solicit)
1009 neigh->ops->solicit(neigh, skb);
1010 atomic_inc(&neigh->probes);
1011 consume_skb(skb);
1012}
1013
1014/* Called when a timer expires for a neighbour entry. */
1015
1016static void neigh_timer_handler(struct timer_list *t)
1017{
1018 unsigned long now, next;
1019 struct neighbour *neigh = from_timer(neigh, t, timer);
1020 unsigned int state;
1021 int notify = 0;
1022
1023 write_lock(&neigh->lock);
1024
1025 state = neigh->nud_state;
1026 now = jiffies;
1027 next = now + HZ;
1028
1029 if (!(state & NUD_IN_TIMER))
1030 goto out;
1031
1032 if (state & NUD_REACHABLE) {
1033 if (time_before_eq(now,
1034 neigh->confirmed + neigh->parms->reachable_time)) {
1035 neigh_dbg(2, "neigh %p is still alive\n", neigh);
1036 next = neigh->confirmed + neigh->parms->reachable_time;
1037 } else if (time_before_eq(now,
1038 neigh->used +
1039 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
1040 neigh_dbg(2, "neigh %p is delayed\n", neigh);
1041 neigh->nud_state = NUD_DELAY;
1042 neigh->updated = jiffies;
1043 neigh_suspect(neigh);
1044 next = now + NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME);
1045 } else {
1046 neigh_dbg(2, "neigh %p is suspected\n", neigh);
1047 neigh->nud_state = NUD_STALE;
1048 neigh->updated = jiffies;
1049 neigh_suspect(neigh);
1050 notify = 1;
1051 }
1052 } else if (state & NUD_DELAY) {
1053 if (time_before_eq(now,
1054 neigh->confirmed +
1055 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
1056 neigh_dbg(2, "neigh %p is now reachable\n", neigh);
1057 neigh->nud_state = NUD_REACHABLE;
1058 neigh->updated = jiffies;
1059 neigh_connect(neigh);
1060 notify = 1;
1061 next = neigh->confirmed + neigh->parms->reachable_time;
1062 } else {
1063 neigh_dbg(2, "neigh %p is probed\n", neigh);
1064 neigh->nud_state = NUD_PROBE;
1065 neigh->updated = jiffies;
1066 atomic_set(&neigh->probes, 0);
1067 notify = 1;
1068 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
1069 HZ/100);
1070 }
1071 } else {
1072 /* NUD_PROBE|NUD_INCOMPLETE */
1073 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME), HZ/100);
1074 }
1075
1076 if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
1077 atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
1078 neigh->nud_state = NUD_FAILED;
1079 notify = 1;
1080 neigh_invalidate(neigh);
1081 goto out;
1082 }
1083
1084 if (neigh->nud_state & NUD_IN_TIMER) {
1085 if (time_before(next, jiffies + HZ/100))
1086 next = jiffies + HZ/100;
1087 if (!mod_timer(&neigh->timer, next))
1088 neigh_hold(neigh);
1089 }
1090 if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
1091 neigh_probe(neigh);
1092 } else {
1093out:
1094 write_unlock(&neigh->lock);
1095 }
1096
1097 if (notify)
1098 neigh_update_notify(neigh, 0);
1099
1100 trace_neigh_timer_handler(neigh, 0);
1101
1102 neigh_release(neigh);
1103}
1104
1105int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
1106{
1107 int rc;
1108 bool immediate_probe = false;
1109
1110 write_lock_bh(&neigh->lock);
1111
1112 rc = 0;
1113 if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
1114 goto out_unlock_bh;
1115 if (neigh->dead)
1116 goto out_dead;
1117
1118 if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
1119 if (NEIGH_VAR(neigh->parms, MCAST_PROBES) +
1120 NEIGH_VAR(neigh->parms, APP_PROBES)) {
1121 unsigned long next, now = jiffies;
1122
1123 atomic_set(&neigh->probes,
1124 NEIGH_VAR(neigh->parms, UCAST_PROBES));
1125 neigh_del_timer(neigh);
1126 neigh->nud_state = NUD_INCOMPLETE;
1127 neigh->updated = now;
1128 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
1129 HZ/100);
1130 neigh_add_timer(neigh, next);
1131 immediate_probe = true;
1132 } else {
1133 neigh->nud_state = NUD_FAILED;
1134 neigh->updated = jiffies;
1135 write_unlock_bh(&neigh->lock);
1136
1137 kfree_skb(skb);
1138 return 1;
1139 }
1140 } else if (neigh->nud_state & NUD_STALE) {
1141 neigh_dbg(2, "neigh %p is delayed\n", neigh);
1142 neigh_del_timer(neigh);
1143 neigh->nud_state = NUD_DELAY;
1144 neigh->updated = jiffies;
1145 neigh_add_timer(neigh, jiffies +
1146 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME));
1147 }
1148
1149 if (neigh->nud_state == NUD_INCOMPLETE) {
1150 if (skb) {
1151 while (neigh->arp_queue_len_bytes + skb->truesize >
1152 NEIGH_VAR(neigh->parms, QUEUE_LEN_BYTES)) {
1153 struct sk_buff *buff;
1154
1155 buff = __skb_dequeue(&neigh->arp_queue);
1156 if (!buff)
1157 break;
1158 neigh->arp_queue_len_bytes -= buff->truesize;
1159 kfree_skb(buff);
1160 NEIGH_CACHE_STAT_INC(neigh->tbl, unres_discards);
1161 }
1162 skb_dst_force(skb);
1163 __skb_queue_tail(&neigh->arp_queue, skb);
1164 neigh->arp_queue_len_bytes += skb->truesize;
1165 }
1166 rc = 1;
1167 }
1168out_unlock_bh:
1169 if (immediate_probe)
1170 neigh_probe(neigh);
1171 else
1172 write_unlock(&neigh->lock);
1173 local_bh_enable();
1174 trace_neigh_event_send_done(neigh, rc);
1175 return rc;
1176
1177out_dead:
1178 if (neigh->nud_state & NUD_STALE)
1179 goto out_unlock_bh;
1180 write_unlock_bh(&neigh->lock);
1181 kfree_skb(skb);
1182 trace_neigh_event_send_dead(neigh, 1);
1183 return 1;
1184}
1185EXPORT_SYMBOL(__neigh_event_send);
1186
1187static void neigh_update_hhs(struct neighbour *neigh)
1188{
1189 struct hh_cache *hh;
1190 void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *)
1191 = NULL;
1192
1193 if (neigh->dev->header_ops)
1194 update = neigh->dev->header_ops->cache_update;
1195
1196 if (update) {
1197 hh = &neigh->hh;
1198 if (READ_ONCE(hh->hh_len)) {
1199 write_seqlock_bh(&hh->hh_lock);
1200 update(hh, neigh->dev, neigh->ha);
1201 write_sequnlock_bh(&hh->hh_lock);
1202 }
1203 }
1204}
1205
1206
1207
1208/* Generic update routine.
1209 -- lladdr is new lladdr or NULL, if it is not supplied.
1210 -- new is new state.
1211 -- flags
1212 NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
1213 if it is different.
1214 NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
1215 lladdr instead of overriding it
1216 if it is different.
1217 NEIGH_UPDATE_F_ADMIN means that the change is administrative.
1218
1219 NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
1220 NTF_ROUTER flag.
1221 NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as
1222 a router.
1223
1224 Caller MUST hold reference count on the entry.
1225 */
1226
1227static int __neigh_update(struct neighbour *neigh, const u8 *lladdr,
1228 u8 new, u32 flags, u32 nlmsg_pid,
1229 struct netlink_ext_ack *extack)
1230{
1231 bool ext_learn_change = false;
1232 u8 old;
1233 int err;
1234 int notify = 0;
1235 struct net_device *dev;
1236 int update_isrouter = 0;
1237
1238 trace_neigh_update(neigh, lladdr, new, flags, nlmsg_pid);
1239
1240 write_lock_bh(&neigh->lock);
1241
1242 dev = neigh->dev;
1243 old = neigh->nud_state;
1244 err = -EPERM;
1245
1246 if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
1247 (old & (NUD_NOARP | NUD_PERMANENT)))
1248 goto out;
1249 if (neigh->dead) {
1250 NL_SET_ERR_MSG(extack, "Neighbor entry is now dead");
1251 goto out;
1252 }
1253
1254 ext_learn_change = neigh_update_ext_learned(neigh, flags, ¬ify);
1255
1256 if (!(new & NUD_VALID)) {
1257 neigh_del_timer(neigh);
1258 if (old & NUD_CONNECTED)
1259 neigh_suspect(neigh);
1260 neigh->nud_state = new;
1261 err = 0;
1262 notify = old & NUD_VALID;
1263 if ((old & (NUD_INCOMPLETE | NUD_PROBE)) &&
1264 (new & NUD_FAILED)) {
1265 neigh_invalidate(neigh);
1266 notify = 1;
1267 }
1268 goto out;
1269 }
1270
1271 /* Compare new lladdr with cached one */
1272 if (!dev->addr_len) {
1273 /* First case: device needs no address. */
1274 lladdr = neigh->ha;
1275 } else if (lladdr) {
1276 /* The second case: if something is already cached
1277 and a new address is proposed:
1278 - compare new & old
1279 - if they are different, check override flag
1280 */
1281 if ((old & NUD_VALID) &&
1282 !memcmp(lladdr, neigh->ha, dev->addr_len))
1283 lladdr = neigh->ha;
1284 } else {
1285 /* No address is supplied; if we know something,
1286 use it, otherwise discard the request.
1287 */
1288 err = -EINVAL;
1289 if (!(old & NUD_VALID)) {
1290 NL_SET_ERR_MSG(extack, "No link layer address given");
1291 goto out;
1292 }
1293 lladdr = neigh->ha;
1294 }
1295
1296 /* Update confirmed timestamp for neighbour entry after we
1297 * received ARP packet even if it doesn't change IP to MAC binding.
1298 */
1299 if (new & NUD_CONNECTED)
1300 neigh->confirmed = jiffies;
1301
1302 /* If entry was valid and address is not changed,
1303 do not change entry state, if new one is STALE.
1304 */
1305 err = 0;
1306 update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1307 if (old & NUD_VALID) {
1308 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
1309 update_isrouter = 0;
1310 if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
1311 (old & NUD_CONNECTED)) {
1312 lladdr = neigh->ha;
1313 new = NUD_STALE;
1314 } else
1315 goto out;
1316 } else {
1317 if (lladdr == neigh->ha && new == NUD_STALE &&
1318 !(flags & NEIGH_UPDATE_F_ADMIN))
1319 new = old;
1320 }
1321 }
1322
1323 /* Update timestamp only once we know we will make a change to the
1324 * neighbour entry. Otherwise we risk to move the locktime window with
1325 * noop updates and ignore relevant ARP updates.
1326 */
1327 if (new != old || lladdr != neigh->ha)
1328 neigh->updated = jiffies;
1329
1330 if (new != old) {
1331 neigh_del_timer(neigh);
1332 if (new & NUD_PROBE)
1333 atomic_set(&neigh->probes, 0);
1334 if (new & NUD_IN_TIMER)
1335 neigh_add_timer(neigh, (jiffies +
1336 ((new & NUD_REACHABLE) ?
1337 neigh->parms->reachable_time :
1338 0)));
1339 neigh->nud_state = new;
1340 notify = 1;
1341 }
1342
1343 if (lladdr != neigh->ha) {
1344 write_seqlock(&neigh->ha_lock);
1345 memcpy(&neigh->ha, lladdr, dev->addr_len);
1346 write_sequnlock(&neigh->ha_lock);
1347 neigh_update_hhs(neigh);
1348 if (!(new & NUD_CONNECTED))
1349 neigh->confirmed = jiffies -
1350 (NEIGH_VAR(neigh->parms, BASE_REACHABLE_TIME) << 1);
1351 notify = 1;
1352 }
1353 if (new == old)
1354 goto out;
1355 if (new & NUD_CONNECTED)
1356 neigh_connect(neigh);
1357 else
1358 neigh_suspect(neigh);
1359 if (!(old & NUD_VALID)) {
1360 struct sk_buff *skb;
1361
1362 /* Again: avoid dead loop if something went wrong */
1363
1364 while (neigh->nud_state & NUD_VALID &&
1365 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1366 struct dst_entry *dst = skb_dst(skb);
1367 struct neighbour *n2, *n1 = neigh;
1368 write_unlock_bh(&neigh->lock);
1369
1370 rcu_read_lock();
1371
1372 /* Why not just use 'neigh' as-is? The problem is that
1373 * things such as shaper, eql, and sch_teql can end up
1374 * using alternative, different, neigh objects to output
1375 * the packet in the output path. So what we need to do
1376 * here is re-lookup the top-level neigh in the path so
1377 * we can reinject the packet there.
1378 */
1379 n2 = NULL;
1380 if (dst) {
1381 n2 = dst_neigh_lookup_skb(dst, skb);
1382 if (n2)
1383 n1 = n2;
1384 }
1385 n1->output(n1, skb);
1386 if (n2)
1387 neigh_release(n2);
1388 rcu_read_unlock();
1389
1390 write_lock_bh(&neigh->lock);
1391 }
1392 __skb_queue_purge(&neigh->arp_queue);
1393 neigh->arp_queue_len_bytes = 0;
1394 }
1395out:
1396 if (update_isrouter)
1397 neigh_update_is_router(neigh, flags, ¬ify);
1398 write_unlock_bh(&neigh->lock);
1399
1400 if (((new ^ old) & NUD_PERMANENT) || ext_learn_change)
1401 neigh_update_gc_list(neigh);
1402
1403 if (notify)
1404 neigh_update_notify(neigh, nlmsg_pid);
1405
1406 trace_neigh_update_done(neigh, err);
1407
1408 return err;
1409}
1410
1411int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
1412 u32 flags, u32 nlmsg_pid)
1413{
1414 return __neigh_update(neigh, lladdr, new, flags, nlmsg_pid, NULL);
1415}
1416EXPORT_SYMBOL(neigh_update);
1417
1418/* Update the neigh to listen temporarily for probe responses, even if it is
1419 * in a NUD_FAILED state. The caller has to hold neigh->lock for writing.
1420 */
1421void __neigh_set_probe_once(struct neighbour *neigh)
1422{
1423 if (neigh->dead)
1424 return;
1425 neigh->updated = jiffies;
1426 if (!(neigh->nud_state & NUD_FAILED))
1427 return;
1428 neigh->nud_state = NUD_INCOMPLETE;
1429 atomic_set(&neigh->probes, neigh_max_probes(neigh));
1430 neigh_add_timer(neigh,
1431 jiffies + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
1432 HZ/100));
1433}
1434EXPORT_SYMBOL(__neigh_set_probe_once);
1435
1436struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1437 u8 *lladdr, void *saddr,
1438 struct net_device *dev)
1439{
1440 struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1441 lladdr || !dev->addr_len);
1442 if (neigh)
1443 neigh_update(neigh, lladdr, NUD_STALE,
1444 NEIGH_UPDATE_F_OVERRIDE, 0);
1445 return neigh;
1446}
1447EXPORT_SYMBOL(neigh_event_ns);
1448
1449/* called with read_lock_bh(&n->lock); */
1450static void neigh_hh_init(struct neighbour *n)
1451{
1452 struct net_device *dev = n->dev;
1453 __be16 prot = n->tbl->protocol;
1454 struct hh_cache *hh = &n->hh;
1455
1456 write_lock_bh(&n->lock);
1457
1458 /* Only one thread can come in here and initialize the
1459 * hh_cache entry.
1460 */
1461 if (!hh->hh_len)
1462 dev->header_ops->cache(n, hh, prot);
1463
1464 write_unlock_bh(&n->lock);
1465}
1466
1467/* Slow and careful. */
1468
1469int neigh_resolve_output(struct neighbour *neigh, struct sk_buff *skb)
1470{
1471 int rc = 0;
1472
1473 if (!neigh_event_send(neigh, skb)) {
1474 int err;
1475 struct net_device *dev = neigh->dev;
1476 unsigned int seq;
1477
1478 if (dev->header_ops->cache && !READ_ONCE(neigh->hh.hh_len))
1479 neigh_hh_init(neigh);
1480
1481 do {
1482 __skb_pull(skb, skb_network_offset(skb));
1483 seq = read_seqbegin(&neigh->ha_lock);
1484 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1485 neigh->ha, NULL, skb->len);
1486 } while (read_seqretry(&neigh->ha_lock, seq));
1487
1488 if (err >= 0)
1489 rc = dev_queue_xmit(skb);
1490 else
1491 goto out_kfree_skb;
1492 }
1493out:
1494 return rc;
1495out_kfree_skb:
1496 rc = -EINVAL;
1497 kfree_skb(skb);
1498 goto out;
1499}
1500EXPORT_SYMBOL(neigh_resolve_output);
1501
1502/* As fast as possible without hh cache */
1503
1504int neigh_connected_output(struct neighbour *neigh, struct sk_buff *skb)
1505{
1506 struct net_device *dev = neigh->dev;
1507 unsigned int seq;
1508 int err;
1509
1510 do {
1511 __skb_pull(skb, skb_network_offset(skb));
1512 seq = read_seqbegin(&neigh->ha_lock);
1513 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1514 neigh->ha, NULL, skb->len);
1515 } while (read_seqretry(&neigh->ha_lock, seq));
1516
1517 if (err >= 0)
1518 err = dev_queue_xmit(skb);
1519 else {
1520 err = -EINVAL;
1521 kfree_skb(skb);
1522 }
1523 return err;
1524}
1525EXPORT_SYMBOL(neigh_connected_output);
1526
1527int neigh_direct_output(struct neighbour *neigh, struct sk_buff *skb)
1528{
1529 return dev_queue_xmit(skb);
1530}
1531EXPORT_SYMBOL(neigh_direct_output);
1532
1533static void neigh_proxy_process(struct timer_list *t)
1534{
1535 struct neigh_table *tbl = from_timer(tbl, t, proxy_timer);
1536 long sched_next = 0;
1537 unsigned long now = jiffies;
1538 struct sk_buff *skb, *n;
1539
1540 spin_lock(&tbl->proxy_queue.lock);
1541
1542 skb_queue_walk_safe(&tbl->proxy_queue, skb, n) {
1543 long tdif = NEIGH_CB(skb)->sched_next - now;
1544
1545 if (tdif <= 0) {
1546 struct net_device *dev = skb->dev;
1547
1548 __skb_unlink(skb, &tbl->proxy_queue);
1549 if (tbl->proxy_redo && netif_running(dev)) {
1550 rcu_read_lock();
1551 tbl->proxy_redo(skb);
1552 rcu_read_unlock();
1553 } else {
1554 kfree_skb(skb);
1555 }
1556
1557 dev_put(dev);
1558 } else if (!sched_next || tdif < sched_next)
1559 sched_next = tdif;
1560 }
1561 del_timer(&tbl->proxy_timer);
1562 if (sched_next)
1563 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1564 spin_unlock(&tbl->proxy_queue.lock);
1565}
1566
1567void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1568 struct sk_buff *skb)
1569{
1570 unsigned long now = jiffies;
1571
1572 unsigned long sched_next = now + (prandom_u32() %
1573 NEIGH_VAR(p, PROXY_DELAY));
1574
1575 if (tbl->proxy_queue.qlen > NEIGH_VAR(p, PROXY_QLEN)) {
1576 kfree_skb(skb);
1577 return;
1578 }
1579
1580 NEIGH_CB(skb)->sched_next = sched_next;
1581 NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1582
1583 spin_lock(&tbl->proxy_queue.lock);
1584 if (del_timer(&tbl->proxy_timer)) {
1585 if (time_before(tbl->proxy_timer.expires, sched_next))
1586 sched_next = tbl->proxy_timer.expires;
1587 }
1588 skb_dst_drop(skb);
1589 dev_hold(skb->dev);
1590 __skb_queue_tail(&tbl->proxy_queue, skb);
1591 mod_timer(&tbl->proxy_timer, sched_next);
1592 spin_unlock(&tbl->proxy_queue.lock);
1593}
1594EXPORT_SYMBOL(pneigh_enqueue);
1595
1596static inline struct neigh_parms *lookup_neigh_parms(struct neigh_table *tbl,
1597 struct net *net, int ifindex)
1598{
1599 struct neigh_parms *p;
1600
1601 list_for_each_entry(p, &tbl->parms_list, list) {
1602 if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
1603 (!p->dev && !ifindex && net_eq(net, &init_net)))
1604 return p;
1605 }
1606
1607 return NULL;
1608}
1609
1610struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1611 struct neigh_table *tbl)
1612{
1613 struct neigh_parms *p;
1614 struct net *net = dev_net(dev);
1615 const struct net_device_ops *ops = dev->netdev_ops;
1616
1617 p = kmemdup(&tbl->parms, sizeof(*p), GFP_KERNEL);
1618 if (p) {
1619 p->tbl = tbl;
1620 refcount_set(&p->refcnt, 1);
1621 p->reachable_time =
1622 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
1623 dev_hold(dev);
1624 p->dev = dev;
1625 write_pnet(&p->net, net);
1626 p->sysctl_table = NULL;
1627
1628 if (ops->ndo_neigh_setup && ops->ndo_neigh_setup(dev, p)) {
1629 dev_put(dev);
1630 kfree(p);
1631 return NULL;
1632 }
1633
1634 write_lock_bh(&tbl->lock);
1635 list_add(&p->list, &tbl->parms.list);
1636 write_unlock_bh(&tbl->lock);
1637
1638 neigh_parms_data_state_cleanall(p);
1639 }
1640 return p;
1641}
1642EXPORT_SYMBOL(neigh_parms_alloc);
1643
1644static void neigh_rcu_free_parms(struct rcu_head *head)
1645{
1646 struct neigh_parms *parms =
1647 container_of(head, struct neigh_parms, rcu_head);
1648
1649 neigh_parms_put(parms);
1650}
1651
1652void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1653{
1654 if (!parms || parms == &tbl->parms)
1655 return;
1656 write_lock_bh(&tbl->lock);
1657 list_del(&parms->list);
1658 parms->dead = 1;
1659 write_unlock_bh(&tbl->lock);
1660 if (parms->dev)
1661 dev_put(parms->dev);
1662 call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1663}
1664EXPORT_SYMBOL(neigh_parms_release);
1665
1666static void neigh_parms_destroy(struct neigh_parms *parms)
1667{
1668 kfree(parms);
1669}
1670
1671static struct lock_class_key neigh_table_proxy_queue_class;
1672
1673static struct neigh_table *neigh_tables[NEIGH_NR_TABLES] __read_mostly;
1674
1675void neigh_table_init(int index, struct neigh_table *tbl)
1676{
1677 unsigned long now = jiffies;
1678 unsigned long phsize;
1679
1680 INIT_LIST_HEAD(&tbl->parms_list);
1681 INIT_LIST_HEAD(&tbl->gc_list);
1682 list_add(&tbl->parms.list, &tbl->parms_list);
1683 write_pnet(&tbl->parms.net, &init_net);
1684 refcount_set(&tbl->parms.refcnt, 1);
1685 tbl->parms.reachable_time =
1686 neigh_rand_reach_time(NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME));
1687
1688 tbl->stats = alloc_percpu(struct neigh_statistics);
1689 if (!tbl->stats)
1690 panic("cannot create neighbour cache statistics");
1691
1692#ifdef CONFIG_PROC_FS
1693 if (!proc_create_seq_data(tbl->id, 0, init_net.proc_net_stat,
1694 &neigh_stat_seq_ops, tbl))
1695 panic("cannot create neighbour proc dir entry");
1696#endif
1697
1698 RCU_INIT_POINTER(tbl->nht, neigh_hash_alloc(3));
1699
1700 phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1701 tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1702
1703 if (!tbl->nht || !tbl->phash_buckets)
1704 panic("cannot allocate neighbour cache hashes");
1705
1706 if (!tbl->entry_size)
1707 tbl->entry_size = ALIGN(offsetof(struct neighbour, primary_key) +
1708 tbl->key_len, NEIGH_PRIV_ALIGN);
1709 else
1710 WARN_ON(tbl->entry_size % NEIGH_PRIV_ALIGN);
1711
1712 rwlock_init(&tbl->lock);
1713 INIT_DEFERRABLE_WORK(&tbl->gc_work, neigh_periodic_work);
1714 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
1715 tbl->parms.reachable_time);
1716 timer_setup(&tbl->proxy_timer, neigh_proxy_process, 0);
1717 skb_queue_head_init_class(&tbl->proxy_queue,
1718 &neigh_table_proxy_queue_class);
1719
1720 tbl->last_flush = now;
1721 tbl->last_rand = now + tbl->parms.reachable_time * 20;
1722
1723 neigh_tables[index] = tbl;
1724}
1725EXPORT_SYMBOL(neigh_table_init);
1726
1727int neigh_table_clear(int index, struct neigh_table *tbl)
1728{
1729 neigh_tables[index] = NULL;
1730 /* It is not clean... Fix it to unload IPv6 module safely */
1731 cancel_delayed_work_sync(&tbl->gc_work);
1732 del_timer_sync(&tbl->proxy_timer);
1733 pneigh_queue_purge(&tbl->proxy_queue);
1734 neigh_ifdown(tbl, NULL);
1735 if (atomic_read(&tbl->entries))
1736 pr_crit("neighbour leakage\n");
1737
1738 call_rcu(&rcu_dereference_protected(tbl->nht, 1)->rcu,
1739 neigh_hash_free_rcu);
1740 tbl->nht = NULL;
1741
1742 kfree(tbl->phash_buckets);
1743 tbl->phash_buckets = NULL;
1744
1745 remove_proc_entry(tbl->id, init_net.proc_net_stat);
1746
1747 free_percpu(tbl->stats);
1748 tbl->stats = NULL;
1749
1750 return 0;
1751}
1752EXPORT_SYMBOL(neigh_table_clear);
1753
1754static struct neigh_table *neigh_find_table(int family)
1755{
1756 struct neigh_table *tbl = NULL;
1757
1758 switch (family) {
1759 case AF_INET:
1760 tbl = neigh_tables[NEIGH_ARP_TABLE];
1761 break;
1762 case AF_INET6:
1763 tbl = neigh_tables[NEIGH_ND_TABLE];
1764 break;
1765 case AF_DECnet:
1766 tbl = neigh_tables[NEIGH_DN_TABLE];
1767 break;
1768 }
1769
1770 return tbl;
1771}
1772
1773const struct nla_policy nda_policy[NDA_MAX+1] = {
1774 [NDA_UNSPEC] = { .strict_start_type = NDA_NH_ID },
1775 [NDA_DST] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1776 [NDA_LLADDR] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1777 [NDA_CACHEINFO] = { .len = sizeof(struct nda_cacheinfo) },
1778 [NDA_PROBES] = { .type = NLA_U32 },
1779 [NDA_VLAN] = { .type = NLA_U16 },
1780 [NDA_PORT] = { .type = NLA_U16 },
1781 [NDA_VNI] = { .type = NLA_U32 },
1782 [NDA_IFINDEX] = { .type = NLA_U32 },
1783 [NDA_MASTER] = { .type = NLA_U32 },
1784 [NDA_PROTOCOL] = { .type = NLA_U8 },
1785 [NDA_NH_ID] = { .type = NLA_U32 },
1786 [NDA_FDB_EXT_ATTRS] = { .type = NLA_NESTED },
1787};
1788
1789static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh,
1790 struct netlink_ext_ack *extack)
1791{
1792 struct net *net = sock_net(skb->sk);
1793 struct ndmsg *ndm;
1794 struct nlattr *dst_attr;
1795 struct neigh_table *tbl;
1796 struct neighbour *neigh;
1797 struct net_device *dev = NULL;
1798 int err = -EINVAL;
1799
1800 ASSERT_RTNL();
1801 if (nlmsg_len(nlh) < sizeof(*ndm))
1802 goto out;
1803
1804 dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1805 if (!dst_attr) {
1806 NL_SET_ERR_MSG(extack, "Network address not specified");
1807 goto out;
1808 }
1809
1810 ndm = nlmsg_data(nlh);
1811 if (ndm->ndm_ifindex) {
1812 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1813 if (dev == NULL) {
1814 err = -ENODEV;
1815 goto out;
1816 }
1817 }
1818
1819 tbl = neigh_find_table(ndm->ndm_family);
1820 if (tbl == NULL)
1821 return -EAFNOSUPPORT;
1822
1823 if (nla_len(dst_attr) < (int)tbl->key_len) {
1824 NL_SET_ERR_MSG(extack, "Invalid network address");
1825 goto out;
1826 }
1827
1828 if (ndm->ndm_flags & NTF_PROXY) {
1829 err = pneigh_delete(tbl, net, nla_data(dst_attr), dev);
1830 goto out;
1831 }
1832
1833 if (dev == NULL)
1834 goto out;
1835
1836 neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1837 if (neigh == NULL) {
1838 err = -ENOENT;
1839 goto out;
1840 }
1841
1842 err = __neigh_update(neigh, NULL, NUD_FAILED,
1843 NEIGH_UPDATE_F_OVERRIDE | NEIGH_UPDATE_F_ADMIN,
1844 NETLINK_CB(skb).portid, extack);
1845 write_lock_bh(&tbl->lock);
1846 neigh_release(neigh);
1847 neigh_remove_one(neigh, tbl);
1848 write_unlock_bh(&tbl->lock);
1849
1850out:
1851 return err;
1852}
1853
1854static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh,
1855 struct netlink_ext_ack *extack)
1856{
1857 int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE |
1858 NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1859 struct net *net = sock_net(skb->sk);
1860 struct ndmsg *ndm;
1861 struct nlattr *tb[NDA_MAX+1];
1862 struct neigh_table *tbl;
1863 struct net_device *dev = NULL;
1864 struct neighbour *neigh;
1865 void *dst, *lladdr;
1866 u8 protocol = 0;
1867 int err;
1868
1869 ASSERT_RTNL();
1870 err = nlmsg_parse_deprecated(nlh, sizeof(*ndm), tb, NDA_MAX,
1871 nda_policy, extack);
1872 if (err < 0)
1873 goto out;
1874
1875 err = -EINVAL;
1876 if (!tb[NDA_DST]) {
1877 NL_SET_ERR_MSG(extack, "Network address not specified");
1878 goto out;
1879 }
1880
1881 ndm = nlmsg_data(nlh);
1882 if (ndm->ndm_ifindex) {
1883 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1884 if (dev == NULL) {
1885 err = -ENODEV;
1886 goto out;
1887 }
1888
1889 if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len) {
1890 NL_SET_ERR_MSG(extack, "Invalid link address");
1891 goto out;
1892 }
1893 }
1894
1895 tbl = neigh_find_table(ndm->ndm_family);
1896 if (tbl == NULL)
1897 return -EAFNOSUPPORT;
1898
1899 if (nla_len(tb[NDA_DST]) < (int)tbl->key_len) {
1900 NL_SET_ERR_MSG(extack, "Invalid network address");
1901 goto out;
1902 }
1903
1904 dst = nla_data(tb[NDA_DST]);
1905 lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
1906
1907 if (tb[NDA_PROTOCOL])
1908 protocol = nla_get_u8(tb[NDA_PROTOCOL]);
1909
1910 if (ndm->ndm_flags & NTF_PROXY) {
1911 struct pneigh_entry *pn;
1912
1913 err = -ENOBUFS;
1914 pn = pneigh_lookup(tbl, net, dst, dev, 1);
1915 if (pn) {
1916 pn->flags = ndm->ndm_flags;
1917 if (protocol)
1918 pn->protocol = protocol;
1919 err = 0;
1920 }
1921 goto out;
1922 }
1923
1924 if (!dev) {
1925 NL_SET_ERR_MSG(extack, "Device not specified");
1926 goto out;
1927 }
1928
1929 if (tbl->allow_add && !tbl->allow_add(dev, extack)) {
1930 err = -EINVAL;
1931 goto out;
1932 }
1933
1934 neigh = neigh_lookup(tbl, dst, dev);
1935 if (neigh == NULL) {
1936 bool exempt_from_gc;
1937
1938 if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
1939 err = -ENOENT;
1940 goto out;
1941 }
1942
1943 exempt_from_gc = ndm->ndm_state & NUD_PERMANENT ||
1944 ndm->ndm_flags & NTF_EXT_LEARNED;
1945 neigh = ___neigh_create(tbl, dst, dev, exempt_from_gc, true);
1946 if (IS_ERR(neigh)) {
1947 err = PTR_ERR(neigh);
1948 goto out;
1949 }
1950 } else {
1951 if (nlh->nlmsg_flags & NLM_F_EXCL) {
1952 err = -EEXIST;
1953 neigh_release(neigh);
1954 goto out;
1955 }
1956
1957 if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
1958 flags &= ~(NEIGH_UPDATE_F_OVERRIDE |
1959 NEIGH_UPDATE_F_OVERRIDE_ISROUTER);
1960 }
1961
1962 if (protocol)
1963 neigh->protocol = protocol;
1964
1965 if (ndm->ndm_flags & NTF_EXT_LEARNED)
1966 flags |= NEIGH_UPDATE_F_EXT_LEARNED;
1967
1968 if (ndm->ndm_flags & NTF_ROUTER)
1969 flags |= NEIGH_UPDATE_F_ISROUTER;
1970
1971 if (ndm->ndm_flags & NTF_USE) {
1972 neigh_event_send(neigh, NULL);
1973 err = 0;
1974 } else
1975 err = __neigh_update(neigh, lladdr, ndm->ndm_state, flags,
1976 NETLINK_CB(skb).portid, extack);
1977
1978 neigh_release(neigh);
1979
1980out:
1981 return err;
1982}
1983
1984static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
1985{
1986 struct nlattr *nest;
1987
1988 nest = nla_nest_start_noflag(skb, NDTA_PARMS);
1989 if (nest == NULL)
1990 return -ENOBUFS;
1991
1992 if ((parms->dev &&
1993 nla_put_u32(skb, NDTPA_IFINDEX, parms->dev->ifindex)) ||
1994 nla_put_u32(skb, NDTPA_REFCNT, refcount_read(&parms->refcnt)) ||
1995 nla_put_u32(skb, NDTPA_QUEUE_LENBYTES,
1996 NEIGH_VAR(parms, QUEUE_LEN_BYTES)) ||
1997 /* approximative value for deprecated QUEUE_LEN (in packets) */
1998 nla_put_u32(skb, NDTPA_QUEUE_LEN,
1999 NEIGH_VAR(parms, QUEUE_LEN_BYTES) / SKB_TRUESIZE(ETH_FRAME_LEN)) ||
2000 nla_put_u32(skb, NDTPA_PROXY_QLEN, NEIGH_VAR(parms, PROXY_QLEN)) ||
2001 nla_put_u32(skb, NDTPA_APP_PROBES, NEIGH_VAR(parms, APP_PROBES)) ||
2002 nla_put_u32(skb, NDTPA_UCAST_PROBES,
2003 NEIGH_VAR(parms, UCAST_PROBES)) ||
2004 nla_put_u32(skb, NDTPA_MCAST_PROBES,
2005 NEIGH_VAR(parms, MCAST_PROBES)) ||
2006 nla_put_u32(skb, NDTPA_MCAST_REPROBES,
2007 NEIGH_VAR(parms, MCAST_REPROBES)) ||
2008 nla_put_msecs(skb, NDTPA_REACHABLE_TIME, parms->reachable_time,
2009 NDTPA_PAD) ||
2010 nla_put_msecs(skb, NDTPA_BASE_REACHABLE_TIME,
2011 NEIGH_VAR(parms, BASE_REACHABLE_TIME), NDTPA_PAD) ||
2012 nla_put_msecs(skb, NDTPA_GC_STALETIME,
2013 NEIGH_VAR(parms, GC_STALETIME), NDTPA_PAD) ||
2014 nla_put_msecs(skb, NDTPA_DELAY_PROBE_TIME,
2015 NEIGH_VAR(parms, DELAY_PROBE_TIME), NDTPA_PAD) ||
2016 nla_put_msecs(skb, NDTPA_RETRANS_TIME,
2017 NEIGH_VAR(parms, RETRANS_TIME), NDTPA_PAD) ||
2018 nla_put_msecs(skb, NDTPA_ANYCAST_DELAY,
2019 NEIGH_VAR(parms, ANYCAST_DELAY), NDTPA_PAD) ||
2020 nla_put_msecs(skb, NDTPA_PROXY_DELAY,
2021 NEIGH_VAR(parms, PROXY_DELAY), NDTPA_PAD) ||
2022 nla_put_msecs(skb, NDTPA_LOCKTIME,
2023 NEIGH_VAR(parms, LOCKTIME), NDTPA_PAD))
2024 goto nla_put_failure;
2025 return nla_nest_end(skb, nest);
2026
2027nla_put_failure:
2028 nla_nest_cancel(skb, nest);
2029 return -EMSGSIZE;
2030}
2031
2032static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl,
2033 u32 pid, u32 seq, int type, int flags)
2034{
2035 struct nlmsghdr *nlh;
2036 struct ndtmsg *ndtmsg;
2037
2038 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
2039 if (nlh == NULL)
2040 return -EMSGSIZE;
2041
2042 ndtmsg = nlmsg_data(nlh);
2043
2044 read_lock_bh(&tbl->lock);
2045 ndtmsg->ndtm_family = tbl->family;
2046 ndtmsg->ndtm_pad1 = 0;
2047 ndtmsg->ndtm_pad2 = 0;
2048
2049 if (nla_put_string(skb, NDTA_NAME, tbl->id) ||
2050 nla_put_msecs(skb, NDTA_GC_INTERVAL, tbl->gc_interval, NDTA_PAD) ||
2051 nla_put_u32(skb, NDTA_THRESH1, tbl->gc_thresh1) ||
2052 nla_put_u32(skb, NDTA_THRESH2, tbl->gc_thresh2) ||
2053 nla_put_u32(skb, NDTA_THRESH3, tbl->gc_thresh3))
2054 goto nla_put_failure;
2055 {
2056 unsigned long now = jiffies;
2057 long flush_delta = now - tbl->last_flush;
2058 long rand_delta = now - tbl->last_rand;
2059 struct neigh_hash_table *nht;
2060 struct ndt_config ndc = {
2061 .ndtc_key_len = tbl->key_len,
2062 .ndtc_entry_size = tbl->entry_size,
2063 .ndtc_entries = atomic_read(&tbl->entries),
2064 .ndtc_last_flush = jiffies_to_msecs(flush_delta),
2065 .ndtc_last_rand = jiffies_to_msecs(rand_delta),
2066 .ndtc_proxy_qlen = tbl->proxy_queue.qlen,
2067 };
2068
2069 rcu_read_lock_bh();
2070 nht = rcu_dereference_bh(tbl->nht);
2071 ndc.ndtc_hash_rnd = nht->hash_rnd[0];
2072 ndc.ndtc_hash_mask = ((1 << nht->hash_shift) - 1);
2073 rcu_read_unlock_bh();
2074
2075 if (nla_put(skb, NDTA_CONFIG, sizeof(ndc), &ndc))
2076 goto nla_put_failure;
2077 }
2078
2079 {
2080 int cpu;
2081 struct ndt_stats ndst;
2082
2083 memset(&ndst, 0, sizeof(ndst));
2084
2085 for_each_possible_cpu(cpu) {
2086 struct neigh_statistics *st;
2087
2088 st = per_cpu_ptr(tbl->stats, cpu);
2089 ndst.ndts_allocs += st->allocs;
2090 ndst.ndts_destroys += st->destroys;
2091 ndst.ndts_hash_grows += st->hash_grows;
2092 ndst.ndts_res_failed += st->res_failed;
2093 ndst.ndts_lookups += st->lookups;
2094 ndst.ndts_hits += st->hits;
2095 ndst.ndts_rcv_probes_mcast += st->rcv_probes_mcast;
2096 ndst.ndts_rcv_probes_ucast += st->rcv_probes_ucast;
2097 ndst.ndts_periodic_gc_runs += st->periodic_gc_runs;
2098 ndst.ndts_forced_gc_runs += st->forced_gc_runs;
2099 ndst.ndts_table_fulls += st->table_fulls;
2100 }
2101
2102 if (nla_put_64bit(skb, NDTA_STATS, sizeof(ndst), &ndst,
2103 NDTA_PAD))
2104 goto nla_put_failure;
2105 }
2106
2107 BUG_ON(tbl->parms.dev);
2108 if (neightbl_fill_parms(skb, &tbl->parms) < 0)
2109 goto nla_put_failure;
2110
2111 read_unlock_bh(&tbl->lock);
2112 nlmsg_end(skb, nlh);
2113 return 0;
2114
2115nla_put_failure:
2116 read_unlock_bh(&tbl->lock);
2117 nlmsg_cancel(skb, nlh);
2118 return -EMSGSIZE;
2119}
2120
2121static int neightbl_fill_param_info(struct sk_buff *skb,
2122 struct neigh_table *tbl,
2123 struct neigh_parms *parms,
2124 u32 pid, u32 seq, int type,
2125 unsigned int flags)
2126{
2127 struct ndtmsg *ndtmsg;
2128 struct nlmsghdr *nlh;
2129
2130 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
2131 if (nlh == NULL)
2132 return -EMSGSIZE;
2133
2134 ndtmsg = nlmsg_data(nlh);
2135
2136 read_lock_bh(&tbl->lock);
2137 ndtmsg->ndtm_family = tbl->family;
2138 ndtmsg->ndtm_pad1 = 0;
2139 ndtmsg->ndtm_pad2 = 0;
2140
2141 if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 ||
2142 neightbl_fill_parms(skb, parms) < 0)
2143 goto errout;
2144
2145 read_unlock_bh(&tbl->lock);
2146 nlmsg_end(skb, nlh);
2147 return 0;
2148errout:
2149 read_unlock_bh(&tbl->lock);
2150 nlmsg_cancel(skb, nlh);
2151 return -EMSGSIZE;
2152}
2153
2154static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = {
2155 [NDTA_NAME] = { .type = NLA_STRING },
2156 [NDTA_THRESH1] = { .type = NLA_U32 },
2157 [NDTA_THRESH2] = { .type = NLA_U32 },
2158 [NDTA_THRESH3] = { .type = NLA_U32 },
2159 [NDTA_GC_INTERVAL] = { .type = NLA_U64 },
2160 [NDTA_PARMS] = { .type = NLA_NESTED },
2161};
2162
2163static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = {
2164 [NDTPA_IFINDEX] = { .type = NLA_U32 },
2165 [NDTPA_QUEUE_LEN] = { .type = NLA_U32 },
2166 [NDTPA_PROXY_QLEN] = { .type = NLA_U32 },
2167 [NDTPA_APP_PROBES] = { .type = NLA_U32 },
2168 [NDTPA_UCAST_PROBES] = { .type = NLA_U32 },
2169 [NDTPA_MCAST_PROBES] = { .type = NLA_U32 },
2170 [NDTPA_MCAST_REPROBES] = { .type = NLA_U32 },
2171 [NDTPA_BASE_REACHABLE_TIME] = { .type = NLA_U64 },
2172 [NDTPA_GC_STALETIME] = { .type = NLA_U64 },
2173 [NDTPA_DELAY_PROBE_TIME] = { .type = NLA_U64 },
2174 [NDTPA_RETRANS_TIME] = { .type = NLA_U64 },
2175 [NDTPA_ANYCAST_DELAY] = { .type = NLA_U64 },
2176 [NDTPA_PROXY_DELAY] = { .type = NLA_U64 },
2177 [NDTPA_LOCKTIME] = { .type = NLA_U64 },
2178};
2179
2180static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh,
2181 struct netlink_ext_ack *extack)
2182{
2183 struct net *net = sock_net(skb->sk);
2184 struct neigh_table *tbl;
2185 struct ndtmsg *ndtmsg;
2186 struct nlattr *tb[NDTA_MAX+1];
2187 bool found = false;
2188 int err, tidx;
2189
2190 err = nlmsg_parse_deprecated(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
2191 nl_neightbl_policy, extack);
2192 if (err < 0)
2193 goto errout;
2194
2195 if (tb[NDTA_NAME] == NULL) {
2196 err = -EINVAL;
2197 goto errout;
2198 }
2199
2200 ndtmsg = nlmsg_data(nlh);
2201
2202 for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
2203 tbl = neigh_tables[tidx];
2204 if (!tbl)
2205 continue;
2206 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
2207 continue;
2208 if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0) {
2209 found = true;
2210 break;
2211 }
2212 }
2213
2214 if (!found)
2215 return -ENOENT;
2216
2217 /*
2218 * We acquire tbl->lock to be nice to the periodic timers and
2219 * make sure they always see a consistent set of values.
2220 */
2221 write_lock_bh(&tbl->lock);
2222
2223 if (tb[NDTA_PARMS]) {
2224 struct nlattr *tbp[NDTPA_MAX+1];
2225 struct neigh_parms *p;
2226 int i, ifindex = 0;
2227
2228 err = nla_parse_nested_deprecated(tbp, NDTPA_MAX,
2229 tb[NDTA_PARMS],
2230 nl_ntbl_parm_policy, extack);
2231 if (err < 0)
2232 goto errout_tbl_lock;
2233
2234 if (tbp[NDTPA_IFINDEX])
2235 ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
2236
2237 p = lookup_neigh_parms(tbl, net, ifindex);
2238 if (p == NULL) {
2239 err = -ENOENT;
2240 goto errout_tbl_lock;
2241 }
2242
2243 for (i = 1; i <= NDTPA_MAX; i++) {
2244 if (tbp[i] == NULL)
2245 continue;
2246
2247 switch (i) {
2248 case NDTPA_QUEUE_LEN:
2249 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2250 nla_get_u32(tbp[i]) *
2251 SKB_TRUESIZE(ETH_FRAME_LEN));
2252 break;
2253 case NDTPA_QUEUE_LENBYTES:
2254 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2255 nla_get_u32(tbp[i]));
2256 break;
2257 case NDTPA_PROXY_QLEN:
2258 NEIGH_VAR_SET(p, PROXY_QLEN,
2259 nla_get_u32(tbp[i]));
2260 break;
2261 case NDTPA_APP_PROBES:
2262 NEIGH_VAR_SET(p, APP_PROBES,
2263 nla_get_u32(tbp[i]));
2264 break;
2265 case NDTPA_UCAST_PROBES:
2266 NEIGH_VAR_SET(p, UCAST_PROBES,
2267 nla_get_u32(tbp[i]));
2268 break;
2269 case NDTPA_MCAST_PROBES:
2270 NEIGH_VAR_SET(p, MCAST_PROBES,
2271 nla_get_u32(tbp[i]));
2272 break;
2273 case NDTPA_MCAST_REPROBES:
2274 NEIGH_VAR_SET(p, MCAST_REPROBES,
2275 nla_get_u32(tbp[i]));
2276 break;
2277 case NDTPA_BASE_REACHABLE_TIME:
2278 NEIGH_VAR_SET(p, BASE_REACHABLE_TIME,
2279 nla_get_msecs(tbp[i]));
2280 /* update reachable_time as well, otherwise, the change will
2281 * only be effective after the next time neigh_periodic_work
2282 * decides to recompute it (can be multiple minutes)
2283 */
2284 p->reachable_time =
2285 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
2286 break;
2287 case NDTPA_GC_STALETIME:
2288 NEIGH_VAR_SET(p, GC_STALETIME,
2289 nla_get_msecs(tbp[i]));
2290 break;
2291 case NDTPA_DELAY_PROBE_TIME:
2292 NEIGH_VAR_SET(p, DELAY_PROBE_TIME,
2293 nla_get_msecs(tbp[i]));
2294 call_netevent_notifiers(NETEVENT_DELAY_PROBE_TIME_UPDATE, p);
2295 break;
2296 case NDTPA_RETRANS_TIME:
2297 NEIGH_VAR_SET(p, RETRANS_TIME,
2298 nla_get_msecs(tbp[i]));
2299 break;
2300 case NDTPA_ANYCAST_DELAY:
2301 NEIGH_VAR_SET(p, ANYCAST_DELAY,
2302 nla_get_msecs(tbp[i]));
2303 break;
2304 case NDTPA_PROXY_DELAY:
2305 NEIGH_VAR_SET(p, PROXY_DELAY,
2306 nla_get_msecs(tbp[i]));
2307 break;
2308 case NDTPA_LOCKTIME:
2309 NEIGH_VAR_SET(p, LOCKTIME,
2310 nla_get_msecs(tbp[i]));
2311 break;
2312 }
2313 }
2314 }
2315
2316 err = -ENOENT;
2317 if ((tb[NDTA_THRESH1] || tb[NDTA_THRESH2] ||
2318 tb[NDTA_THRESH3] || tb[NDTA_GC_INTERVAL]) &&
2319 !net_eq(net, &init_net))
2320 goto errout_tbl_lock;
2321
2322 if (tb[NDTA_THRESH1])
2323 tbl->gc_thresh1 = nla_get_u32(tb[NDTA_THRESH1]);
2324
2325 if (tb[NDTA_THRESH2])
2326 tbl->gc_thresh2 = nla_get_u32(tb[NDTA_THRESH2]);
2327
2328 if (tb[NDTA_THRESH3])
2329 tbl->gc_thresh3 = nla_get_u32(tb[NDTA_THRESH3]);
2330
2331 if (tb[NDTA_GC_INTERVAL])
2332 tbl->gc_interval = nla_get_msecs(tb[NDTA_GC_INTERVAL]);
2333
2334 err = 0;
2335
2336errout_tbl_lock:
2337 write_unlock_bh(&tbl->lock);
2338errout:
2339 return err;
2340}
2341
2342static int neightbl_valid_dump_info(const struct nlmsghdr *nlh,
2343 struct netlink_ext_ack *extack)
2344{
2345 struct ndtmsg *ndtm;
2346
2347 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndtm))) {
2348 NL_SET_ERR_MSG(extack, "Invalid header for neighbor table dump request");
2349 return -EINVAL;
2350 }
2351
2352 ndtm = nlmsg_data(nlh);
2353 if (ndtm->ndtm_pad1 || ndtm->ndtm_pad2) {
2354 NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor table dump request");
2355 return -EINVAL;
2356 }
2357
2358 if (nlmsg_attrlen(nlh, sizeof(*ndtm))) {
2359 NL_SET_ERR_MSG(extack, "Invalid data after header in neighbor table dump request");
2360 return -EINVAL;
2361 }
2362
2363 return 0;
2364}
2365
2366static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2367{
2368 const struct nlmsghdr *nlh = cb->nlh;
2369 struct net *net = sock_net(skb->sk);
2370 int family, tidx, nidx = 0;
2371 int tbl_skip = cb->args[0];
2372 int neigh_skip = cb->args[1];
2373 struct neigh_table *tbl;
2374
2375 if (cb->strict_check) {
2376 int err = neightbl_valid_dump_info(nlh, cb->extack);
2377
2378 if (err < 0)
2379 return err;
2380 }
2381
2382 family = ((struct rtgenmsg *)nlmsg_data(nlh))->rtgen_family;
2383
2384 for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
2385 struct neigh_parms *p;
2386
2387 tbl = neigh_tables[tidx];
2388 if (!tbl)
2389 continue;
2390
2391 if (tidx < tbl_skip || (family && tbl->family != family))
2392 continue;
2393
2394 if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).portid,
2395 nlh->nlmsg_seq, RTM_NEWNEIGHTBL,
2396 NLM_F_MULTI) < 0)
2397 break;
2398
2399 nidx = 0;
2400 p = list_next_entry(&tbl->parms, list);
2401 list_for_each_entry_from(p, &tbl->parms_list, list) {
2402 if (!net_eq(neigh_parms_net(p), net))
2403 continue;
2404
2405 if (nidx < neigh_skip)
2406 goto next;
2407
2408 if (neightbl_fill_param_info(skb, tbl, p,
2409 NETLINK_CB(cb->skb).portid,
2410 nlh->nlmsg_seq,
2411 RTM_NEWNEIGHTBL,
2412 NLM_F_MULTI) < 0)
2413 goto out;
2414 next:
2415 nidx++;
2416 }
2417
2418 neigh_skip = 0;
2419 }
2420out:
2421 cb->args[0] = tidx;
2422 cb->args[1] = nidx;
2423
2424 return skb->len;
2425}
2426
2427static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
2428 u32 pid, u32 seq, int type, unsigned int flags)
2429{
2430 unsigned long now = jiffies;
2431 struct nda_cacheinfo ci;
2432 struct nlmsghdr *nlh;
2433 struct ndmsg *ndm;
2434
2435 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2436 if (nlh == NULL)
2437 return -EMSGSIZE;
2438
2439 ndm = nlmsg_data(nlh);
2440 ndm->ndm_family = neigh->ops->family;
2441 ndm->ndm_pad1 = 0;
2442 ndm->ndm_pad2 = 0;
2443 ndm->ndm_flags = neigh->flags;
2444 ndm->ndm_type = neigh->type;
2445 ndm->ndm_ifindex = neigh->dev->ifindex;
2446
2447 if (nla_put(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key))
2448 goto nla_put_failure;
2449
2450 read_lock_bh(&neigh->lock);
2451 ndm->ndm_state = neigh->nud_state;
2452 if (neigh->nud_state & NUD_VALID) {
2453 char haddr[MAX_ADDR_LEN];
2454
2455 neigh_ha_snapshot(haddr, neigh, neigh->dev);
2456 if (nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, haddr) < 0) {
2457 read_unlock_bh(&neigh->lock);
2458 goto nla_put_failure;
2459 }
2460 }
2461
2462 ci.ndm_used = jiffies_to_clock_t(now - neigh->used);
2463 ci.ndm_confirmed = jiffies_to_clock_t(now - neigh->confirmed);
2464 ci.ndm_updated = jiffies_to_clock_t(now - neigh->updated);
2465 ci.ndm_refcnt = refcount_read(&neigh->refcnt) - 1;
2466 read_unlock_bh(&neigh->lock);
2467
2468 if (nla_put_u32(skb, NDA_PROBES, atomic_read(&neigh->probes)) ||
2469 nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
2470 goto nla_put_failure;
2471
2472 if (neigh->protocol && nla_put_u8(skb, NDA_PROTOCOL, neigh->protocol))
2473 goto nla_put_failure;
2474
2475 nlmsg_end(skb, nlh);
2476 return 0;
2477
2478nla_put_failure:
2479 nlmsg_cancel(skb, nlh);
2480 return -EMSGSIZE;
2481}
2482
2483static int pneigh_fill_info(struct sk_buff *skb, struct pneigh_entry *pn,
2484 u32 pid, u32 seq, int type, unsigned int flags,
2485 struct neigh_table *tbl)
2486{
2487 struct nlmsghdr *nlh;
2488 struct ndmsg *ndm;
2489
2490 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2491 if (nlh == NULL)
2492 return -EMSGSIZE;
2493
2494 ndm = nlmsg_data(nlh);
2495 ndm->ndm_family = tbl->family;
2496 ndm->ndm_pad1 = 0;
2497 ndm->ndm_pad2 = 0;
2498 ndm->ndm_flags = pn->flags | NTF_PROXY;
2499 ndm->ndm_type = RTN_UNICAST;
2500 ndm->ndm_ifindex = pn->dev ? pn->dev->ifindex : 0;
2501 ndm->ndm_state = NUD_NONE;
2502
2503 if (nla_put(skb, NDA_DST, tbl->key_len, pn->key))
2504 goto nla_put_failure;
2505
2506 if (pn->protocol && nla_put_u8(skb, NDA_PROTOCOL, pn->protocol))
2507 goto nla_put_failure;
2508
2509 nlmsg_end(skb, nlh);
2510 return 0;
2511
2512nla_put_failure:
2513 nlmsg_cancel(skb, nlh);
2514 return -EMSGSIZE;
2515}
2516
2517static void neigh_update_notify(struct neighbour *neigh, u32 nlmsg_pid)
2518{
2519 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
2520 __neigh_notify(neigh, RTM_NEWNEIGH, 0, nlmsg_pid);
2521}
2522
2523static bool neigh_master_filtered(struct net_device *dev, int master_idx)
2524{
2525 struct net_device *master;
2526
2527 if (!master_idx)
2528 return false;
2529
2530 master = dev ? netdev_master_upper_dev_get(dev) : NULL;
2531 if (!master || master->ifindex != master_idx)
2532 return true;
2533
2534 return false;
2535}
2536
2537static bool neigh_ifindex_filtered(struct net_device *dev, int filter_idx)
2538{
2539 if (filter_idx && (!dev || dev->ifindex != filter_idx))
2540 return true;
2541
2542 return false;
2543}
2544
2545struct neigh_dump_filter {
2546 int master_idx;
2547 int dev_idx;
2548};
2549
2550static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2551 struct netlink_callback *cb,
2552 struct neigh_dump_filter *filter)
2553{
2554 struct net *net = sock_net(skb->sk);
2555 struct neighbour *n;
2556 int rc, h, s_h = cb->args[1];
2557 int idx, s_idx = idx = cb->args[2];
2558 struct neigh_hash_table *nht;
2559 unsigned int flags = NLM_F_MULTI;
2560
2561 if (filter->dev_idx || filter->master_idx)
2562 flags |= NLM_F_DUMP_FILTERED;
2563
2564 rcu_read_lock_bh();
2565 nht = rcu_dereference_bh(tbl->nht);
2566
2567 for (h = s_h; h < (1 << nht->hash_shift); h++) {
2568 if (h > s_h)
2569 s_idx = 0;
2570 for (n = rcu_dereference_bh(nht->hash_buckets[h]), idx = 0;
2571 n != NULL;
2572 n = rcu_dereference_bh(n->next)) {
2573 if (idx < s_idx || !net_eq(dev_net(n->dev), net))
2574 goto next;
2575 if (neigh_ifindex_filtered(n->dev, filter->dev_idx) ||
2576 neigh_master_filtered(n->dev, filter->master_idx))
2577 goto next;
2578 if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2579 cb->nlh->nlmsg_seq,
2580 RTM_NEWNEIGH,
2581 flags) < 0) {
2582 rc = -1;
2583 goto out;
2584 }
2585next:
2586 idx++;
2587 }
2588 }
2589 rc = skb->len;
2590out:
2591 rcu_read_unlock_bh();
2592 cb->args[1] = h;
2593 cb->args[2] = idx;
2594 return rc;
2595}
2596
2597static int pneigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2598 struct netlink_callback *cb,
2599 struct neigh_dump_filter *filter)
2600{
2601 struct pneigh_entry *n;
2602 struct net *net = sock_net(skb->sk);
2603 int rc, h, s_h = cb->args[3];
2604 int idx, s_idx = idx = cb->args[4];
2605 unsigned int flags = NLM_F_MULTI;
2606
2607 if (filter->dev_idx || filter->master_idx)
2608 flags |= NLM_F_DUMP_FILTERED;
2609
2610 read_lock_bh(&tbl->lock);
2611
2612 for (h = s_h; h <= PNEIGH_HASHMASK; h++) {
2613 if (h > s_h)
2614 s_idx = 0;
2615 for (n = tbl->phash_buckets[h], idx = 0; n; n = n->next) {
2616 if (idx < s_idx || pneigh_net(n) != net)
2617 goto next;
2618 if (neigh_ifindex_filtered(n->dev, filter->dev_idx) ||
2619 neigh_master_filtered(n->dev, filter->master_idx))
2620 goto next;
2621 if (pneigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2622 cb->nlh->nlmsg_seq,
2623 RTM_NEWNEIGH, flags, tbl) < 0) {
2624 read_unlock_bh(&tbl->lock);
2625 rc = -1;
2626 goto out;
2627 }
2628 next:
2629 idx++;
2630 }
2631 }
2632
2633 read_unlock_bh(&tbl->lock);
2634 rc = skb->len;
2635out:
2636 cb->args[3] = h;
2637 cb->args[4] = idx;
2638 return rc;
2639
2640}
2641
2642static int neigh_valid_dump_req(const struct nlmsghdr *nlh,
2643 bool strict_check,
2644 struct neigh_dump_filter *filter,
2645 struct netlink_ext_ack *extack)
2646{
2647 struct nlattr *tb[NDA_MAX + 1];
2648 int err, i;
2649
2650 if (strict_check) {
2651 struct ndmsg *ndm;
2652
2653 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndm))) {
2654 NL_SET_ERR_MSG(extack, "Invalid header for neighbor dump request");
2655 return -EINVAL;
2656 }
2657
2658 ndm = nlmsg_data(nlh);
2659 if (ndm->ndm_pad1 || ndm->ndm_pad2 || ndm->ndm_ifindex ||
2660 ndm->ndm_state || ndm->ndm_type) {
2661 NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor dump request");
2662 return -EINVAL;
2663 }
2664
2665 if (ndm->ndm_flags & ~NTF_PROXY) {
2666 NL_SET_ERR_MSG(extack, "Invalid flags in header for neighbor dump request");
2667 return -EINVAL;
2668 }
2669
2670 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct ndmsg),
2671 tb, NDA_MAX, nda_policy,
2672 extack);
2673 } else {
2674 err = nlmsg_parse_deprecated(nlh, sizeof(struct ndmsg), tb,
2675 NDA_MAX, nda_policy, extack);
2676 }
2677 if (err < 0)
2678 return err;
2679
2680 for (i = 0; i <= NDA_MAX; ++i) {
2681 if (!tb[i])
2682 continue;
2683
2684 /* all new attributes should require strict_check */
2685 switch (i) {
2686 case NDA_IFINDEX:
2687 filter->dev_idx = nla_get_u32(tb[i]);
2688 break;
2689 case NDA_MASTER:
2690 filter->master_idx = nla_get_u32(tb[i]);
2691 break;
2692 default:
2693 if (strict_check) {
2694 NL_SET_ERR_MSG(extack, "Unsupported attribute in neighbor dump request");
2695 return -EINVAL;
2696 }
2697 }
2698 }
2699
2700 return 0;
2701}
2702
2703static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2704{
2705 const struct nlmsghdr *nlh = cb->nlh;
2706 struct neigh_dump_filter filter = {};
2707 struct neigh_table *tbl;
2708 int t, family, s_t;
2709 int proxy = 0;
2710 int err;
2711
2712 family = ((struct rtgenmsg *)nlmsg_data(nlh))->rtgen_family;
2713
2714 /* check for full ndmsg structure presence, family member is
2715 * the same for both structures
2716 */
2717 if (nlmsg_len(nlh) >= sizeof(struct ndmsg) &&
2718 ((struct ndmsg *)nlmsg_data(nlh))->ndm_flags == NTF_PROXY)
2719 proxy = 1;
2720
2721 err = neigh_valid_dump_req(nlh, cb->strict_check, &filter, cb->extack);
2722 if (err < 0 && cb->strict_check)
2723 return err;
2724
2725 s_t = cb->args[0];
2726
2727 for (t = 0; t < NEIGH_NR_TABLES; t++) {
2728 tbl = neigh_tables[t];
2729
2730 if (!tbl)
2731 continue;
2732 if (t < s_t || (family && tbl->family != family))
2733 continue;
2734 if (t > s_t)
2735 memset(&cb->args[1], 0, sizeof(cb->args) -
2736 sizeof(cb->args[0]));
2737 if (proxy)
2738 err = pneigh_dump_table(tbl, skb, cb, &filter);
2739 else
2740 err = neigh_dump_table(tbl, skb, cb, &filter);
2741 if (err < 0)
2742 break;
2743 }
2744
2745 cb->args[0] = t;
2746 return skb->len;
2747}
2748
2749static int neigh_valid_get_req(const struct nlmsghdr *nlh,
2750 struct neigh_table **tbl,
2751 void **dst, int *dev_idx, u8 *ndm_flags,
2752 struct netlink_ext_ack *extack)
2753{
2754 struct nlattr *tb[NDA_MAX + 1];
2755 struct ndmsg *ndm;
2756 int err, i;
2757
2758 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndm))) {
2759 NL_SET_ERR_MSG(extack, "Invalid header for neighbor get request");
2760 return -EINVAL;
2761 }
2762
2763 ndm = nlmsg_data(nlh);
2764 if (ndm->ndm_pad1 || ndm->ndm_pad2 || ndm->ndm_state ||
2765 ndm->ndm_type) {
2766 NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor get request");
2767 return -EINVAL;
2768 }
2769
2770 if (ndm->ndm_flags & ~NTF_PROXY) {
2771 NL_SET_ERR_MSG(extack, "Invalid flags in header for neighbor get request");
2772 return -EINVAL;
2773 }
2774
2775 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct ndmsg), tb,
2776 NDA_MAX, nda_policy, extack);
2777 if (err < 0)
2778 return err;
2779
2780 *ndm_flags = ndm->ndm_flags;
2781 *dev_idx = ndm->ndm_ifindex;
2782 *tbl = neigh_find_table(ndm->ndm_family);
2783 if (*tbl == NULL) {
2784 NL_SET_ERR_MSG(extack, "Unsupported family in header for neighbor get request");
2785 return -EAFNOSUPPORT;
2786 }
2787
2788 for (i = 0; i <= NDA_MAX; ++i) {
2789 if (!tb[i])
2790 continue;
2791
2792 switch (i) {
2793 case NDA_DST:
2794 if (nla_len(tb[i]) != (int)(*tbl)->key_len) {
2795 NL_SET_ERR_MSG(extack, "Invalid network address in neighbor get request");
2796 return -EINVAL;
2797 }
2798 *dst = nla_data(tb[i]);
2799 break;
2800 default:
2801 NL_SET_ERR_MSG(extack, "Unsupported attribute in neighbor get request");
2802 return -EINVAL;
2803 }
2804 }
2805
2806 return 0;
2807}
2808
2809static inline size_t neigh_nlmsg_size(void)
2810{
2811 return NLMSG_ALIGN(sizeof(struct ndmsg))
2812 + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2813 + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */
2814 + nla_total_size(sizeof(struct nda_cacheinfo))
2815 + nla_total_size(4) /* NDA_PROBES */
2816 + nla_total_size(1); /* NDA_PROTOCOL */
2817}
2818
2819static int neigh_get_reply(struct net *net, struct neighbour *neigh,
2820 u32 pid, u32 seq)
2821{
2822 struct sk_buff *skb;
2823 int err = 0;
2824
2825 skb = nlmsg_new(neigh_nlmsg_size(), GFP_KERNEL);
2826 if (!skb)
2827 return -ENOBUFS;
2828
2829 err = neigh_fill_info(skb, neigh, pid, seq, RTM_NEWNEIGH, 0);
2830 if (err) {
2831 kfree_skb(skb);
2832 goto errout;
2833 }
2834
2835 err = rtnl_unicast(skb, net, pid);
2836errout:
2837 return err;
2838}
2839
2840static inline size_t pneigh_nlmsg_size(void)
2841{
2842 return NLMSG_ALIGN(sizeof(struct ndmsg))
2843 + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2844 + nla_total_size(1); /* NDA_PROTOCOL */
2845}
2846
2847static int pneigh_get_reply(struct net *net, struct pneigh_entry *neigh,
2848 u32 pid, u32 seq, struct neigh_table *tbl)
2849{
2850 struct sk_buff *skb;
2851 int err = 0;
2852
2853 skb = nlmsg_new(pneigh_nlmsg_size(), GFP_KERNEL);
2854 if (!skb)
2855 return -ENOBUFS;
2856
2857 err = pneigh_fill_info(skb, neigh, pid, seq, RTM_NEWNEIGH, 0, tbl);
2858 if (err) {
2859 kfree_skb(skb);
2860 goto errout;
2861 }
2862
2863 err = rtnl_unicast(skb, net, pid);
2864errout:
2865 return err;
2866}
2867
2868static int neigh_get(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2869 struct netlink_ext_ack *extack)
2870{
2871 struct net *net = sock_net(in_skb->sk);
2872 struct net_device *dev = NULL;
2873 struct neigh_table *tbl = NULL;
2874 struct neighbour *neigh;
2875 void *dst = NULL;
2876 u8 ndm_flags = 0;
2877 int dev_idx = 0;
2878 int err;
2879
2880 err = neigh_valid_get_req(nlh, &tbl, &dst, &dev_idx, &ndm_flags,
2881 extack);
2882 if (err < 0)
2883 return err;
2884
2885 if (dev_idx) {
2886 dev = __dev_get_by_index(net, dev_idx);
2887 if (!dev) {
2888 NL_SET_ERR_MSG(extack, "Unknown device ifindex");
2889 return -ENODEV;
2890 }
2891 }
2892
2893 if (!dst) {
2894 NL_SET_ERR_MSG(extack, "Network address not specified");
2895 return -EINVAL;
2896 }
2897
2898 if (ndm_flags & NTF_PROXY) {
2899 struct pneigh_entry *pn;
2900
2901 pn = pneigh_lookup(tbl, net, dst, dev, 0);
2902 if (!pn) {
2903 NL_SET_ERR_MSG(extack, "Proxy neighbour entry not found");
2904 return -ENOENT;
2905 }
2906 return pneigh_get_reply(net, pn, NETLINK_CB(in_skb).portid,
2907 nlh->nlmsg_seq, tbl);
2908 }
2909
2910 if (!dev) {
2911 NL_SET_ERR_MSG(extack, "No device specified");
2912 return -EINVAL;
2913 }
2914
2915 neigh = neigh_lookup(tbl, dst, dev);
2916 if (!neigh) {
2917 NL_SET_ERR_MSG(extack, "Neighbour entry not found");
2918 return -ENOENT;
2919 }
2920
2921 err = neigh_get_reply(net, neigh, NETLINK_CB(in_skb).portid,
2922 nlh->nlmsg_seq);
2923
2924 neigh_release(neigh);
2925
2926 return err;
2927}
2928
2929void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
2930{
2931 int chain;
2932 struct neigh_hash_table *nht;
2933
2934 rcu_read_lock_bh();
2935 nht = rcu_dereference_bh(tbl->nht);
2936
2937 read_lock(&tbl->lock); /* avoid resizes */
2938 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2939 struct neighbour *n;
2940
2941 for (n = rcu_dereference_bh(nht->hash_buckets[chain]);
2942 n != NULL;
2943 n = rcu_dereference_bh(n->next))
2944 cb(n, cookie);
2945 }
2946 read_unlock(&tbl->lock);
2947 rcu_read_unlock_bh();
2948}
2949EXPORT_SYMBOL(neigh_for_each);
2950
2951/* The tbl->lock must be held as a writer and BH disabled. */
2952void __neigh_for_each_release(struct neigh_table *tbl,
2953 int (*cb)(struct neighbour *))
2954{
2955 int chain;
2956 struct neigh_hash_table *nht;
2957
2958 nht = rcu_dereference_protected(tbl->nht,
2959 lockdep_is_held(&tbl->lock));
2960 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2961 struct neighbour *n;
2962 struct neighbour __rcu **np;
2963
2964 np = &nht->hash_buckets[chain];
2965 while ((n = rcu_dereference_protected(*np,
2966 lockdep_is_held(&tbl->lock))) != NULL) {
2967 int release;
2968
2969 write_lock(&n->lock);
2970 release = cb(n);
2971 if (release) {
2972 rcu_assign_pointer(*np,
2973 rcu_dereference_protected(n->next,
2974 lockdep_is_held(&tbl->lock)));
2975 neigh_mark_dead(n);
2976 } else
2977 np = &n->next;
2978 write_unlock(&n->lock);
2979 if (release)
2980 neigh_cleanup_and_release(n);
2981 }
2982 }
2983}
2984EXPORT_SYMBOL(__neigh_for_each_release);
2985
2986int neigh_xmit(int index, struct net_device *dev,
2987 const void *addr, struct sk_buff *skb)
2988{
2989 int err = -EAFNOSUPPORT;
2990 if (likely(index < NEIGH_NR_TABLES)) {
2991 struct neigh_table *tbl;
2992 struct neighbour *neigh;
2993
2994 tbl = neigh_tables[index];
2995 if (!tbl)
2996 goto out;
2997 rcu_read_lock_bh();
2998 if (index == NEIGH_ARP_TABLE) {
2999 u32 key = *((u32 *)addr);
3000
3001 neigh = __ipv4_neigh_lookup_noref(dev, key);
3002 } else {
3003 neigh = __neigh_lookup_noref(tbl, addr, dev);
3004 }
3005 if (!neigh)
3006 neigh = __neigh_create(tbl, addr, dev, false);
3007 err = PTR_ERR(neigh);
3008 if (IS_ERR(neigh)) {
3009 rcu_read_unlock_bh();
3010 goto out_kfree_skb;
3011 }
3012 err = neigh->output(neigh, skb);
3013 rcu_read_unlock_bh();
3014 }
3015 else if (index == NEIGH_LINK_TABLE) {
3016 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
3017 addr, NULL, skb->len);
3018 if (err < 0)
3019 goto out_kfree_skb;
3020 err = dev_queue_xmit(skb);
3021 }
3022out:
3023 return err;
3024out_kfree_skb:
3025 kfree_skb(skb);
3026 goto out;
3027}
3028EXPORT_SYMBOL(neigh_xmit);
3029
3030#ifdef CONFIG_PROC_FS
3031
3032static struct neighbour *neigh_get_first(struct seq_file *seq)
3033{
3034 struct neigh_seq_state *state = seq->private;
3035 struct net *net = seq_file_net(seq);
3036 struct neigh_hash_table *nht = state->nht;
3037 struct neighbour *n = NULL;
3038 int bucket;
3039
3040 state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
3041 for (bucket = 0; bucket < (1 << nht->hash_shift); bucket++) {
3042 n = rcu_dereference_bh(nht->hash_buckets[bucket]);
3043
3044 while (n) {
3045 if (!net_eq(dev_net(n->dev), net))
3046 goto next;
3047 if (state->neigh_sub_iter) {
3048 loff_t fakep = 0;
3049 void *v;
3050
3051 v = state->neigh_sub_iter(state, n, &fakep);
3052 if (!v)
3053 goto next;
3054 }
3055 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
3056 break;
3057 if (n->nud_state & ~NUD_NOARP)
3058 break;
3059next:
3060 n = rcu_dereference_bh(n->next);
3061 }
3062
3063 if (n)
3064 break;
3065 }
3066 state->bucket = bucket;
3067
3068 return n;
3069}
3070
3071static struct neighbour *neigh_get_next(struct seq_file *seq,
3072 struct neighbour *n,
3073 loff_t *pos)
3074{
3075 struct neigh_seq_state *state = seq->private;
3076 struct net *net = seq_file_net(seq);
3077 struct neigh_hash_table *nht = state->nht;
3078
3079 if (state->neigh_sub_iter) {
3080 void *v = state->neigh_sub_iter(state, n, pos);
3081 if (v)
3082 return n;
3083 }
3084 n = rcu_dereference_bh(n->next);
3085
3086 while (1) {
3087 while (n) {
3088 if (!net_eq(dev_net(n->dev), net))
3089 goto next;
3090 if (state->neigh_sub_iter) {
3091 void *v = state->neigh_sub_iter(state, n, pos);
3092 if (v)
3093 return n;
3094 goto next;
3095 }
3096 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
3097 break;
3098
3099 if (n->nud_state & ~NUD_NOARP)
3100 break;
3101next:
3102 n = rcu_dereference_bh(n->next);
3103 }
3104
3105 if (n)
3106 break;
3107
3108 if (++state->bucket >= (1 << nht->hash_shift))
3109 break;
3110
3111 n = rcu_dereference_bh(nht->hash_buckets[state->bucket]);
3112 }
3113
3114 if (n && pos)
3115 --(*pos);
3116 return n;
3117}
3118
3119static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
3120{
3121 struct neighbour *n = neigh_get_first(seq);
3122
3123 if (n) {
3124 --(*pos);
3125 while (*pos) {
3126 n = neigh_get_next(seq, n, pos);
3127 if (!n)
3128 break;
3129 }
3130 }
3131 return *pos ? NULL : n;
3132}
3133
3134static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
3135{
3136 struct neigh_seq_state *state = seq->private;
3137 struct net *net = seq_file_net(seq);
3138 struct neigh_table *tbl = state->tbl;
3139 struct pneigh_entry *pn = NULL;
3140 int bucket = state->bucket;
3141
3142 state->flags |= NEIGH_SEQ_IS_PNEIGH;
3143 for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
3144 pn = tbl->phash_buckets[bucket];
3145 while (pn && !net_eq(pneigh_net(pn), net))
3146 pn = pn->next;
3147 if (pn)
3148 break;
3149 }
3150 state->bucket = bucket;
3151
3152 return pn;
3153}
3154
3155static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
3156 struct pneigh_entry *pn,
3157 loff_t *pos)
3158{
3159 struct neigh_seq_state *state = seq->private;
3160 struct net *net = seq_file_net(seq);
3161 struct neigh_table *tbl = state->tbl;
3162
3163 do {
3164 pn = pn->next;
3165 } while (pn && !net_eq(pneigh_net(pn), net));
3166
3167 while (!pn) {
3168 if (++state->bucket > PNEIGH_HASHMASK)
3169 break;
3170 pn = tbl->phash_buckets[state->bucket];
3171 while (pn && !net_eq(pneigh_net(pn), net))
3172 pn = pn->next;
3173 if (pn)
3174 break;
3175 }
3176
3177 if (pn && pos)
3178 --(*pos);
3179
3180 return pn;
3181}
3182
3183static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
3184{
3185 struct pneigh_entry *pn = pneigh_get_first(seq);
3186
3187 if (pn) {
3188 --(*pos);
3189 while (*pos) {
3190 pn = pneigh_get_next(seq, pn, pos);
3191 if (!pn)
3192 break;
3193 }
3194 }
3195 return *pos ? NULL : pn;
3196}
3197
3198static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
3199{
3200 struct neigh_seq_state *state = seq->private;
3201 void *rc;
3202 loff_t idxpos = *pos;
3203
3204 rc = neigh_get_idx(seq, &idxpos);
3205 if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
3206 rc = pneigh_get_idx(seq, &idxpos);
3207
3208 return rc;
3209}
3210
3211void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
3212 __acquires(tbl->lock)
3213 __acquires(rcu_bh)
3214{
3215 struct neigh_seq_state *state = seq->private;
3216
3217 state->tbl = tbl;
3218 state->bucket = 0;
3219 state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
3220
3221 rcu_read_lock_bh();
3222 state->nht = rcu_dereference_bh(tbl->nht);
3223 read_lock(&tbl->lock);
3224
3225 return *pos ? neigh_get_idx_any(seq, pos) : SEQ_START_TOKEN;
3226}
3227EXPORT_SYMBOL(neigh_seq_start);
3228
3229void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3230{
3231 struct neigh_seq_state *state;
3232 void *rc;
3233
3234 if (v == SEQ_START_TOKEN) {
3235 rc = neigh_get_first(seq);
3236 goto out;
3237 }
3238
3239 state = seq->private;
3240 if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
3241 rc = neigh_get_next(seq, v, NULL);
3242 if (rc)
3243 goto out;
3244 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
3245 rc = pneigh_get_first(seq);
3246 } else {
3247 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
3248 rc = pneigh_get_next(seq, v, NULL);
3249 }
3250out:
3251 ++(*pos);
3252 return rc;
3253}
3254EXPORT_SYMBOL(neigh_seq_next);
3255
3256void neigh_seq_stop(struct seq_file *seq, void *v)
3257 __releases(tbl->lock)
3258 __releases(rcu_bh)
3259{
3260 struct neigh_seq_state *state = seq->private;
3261 struct neigh_table *tbl = state->tbl;
3262
3263 read_unlock(&tbl->lock);
3264 rcu_read_unlock_bh();
3265}
3266EXPORT_SYMBOL(neigh_seq_stop);
3267
3268/* statistics via seq_file */
3269
3270static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
3271{
3272 struct neigh_table *tbl = PDE_DATA(file_inode(seq->file));
3273 int cpu;
3274
3275 if (*pos == 0)
3276 return SEQ_START_TOKEN;
3277
3278 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
3279 if (!cpu_possible(cpu))
3280 continue;
3281 *pos = cpu+1;
3282 return per_cpu_ptr(tbl->stats, cpu);
3283 }
3284 return NULL;
3285}
3286
3287static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3288{
3289 struct neigh_table *tbl = PDE_DATA(file_inode(seq->file));
3290 int cpu;
3291
3292 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
3293 if (!cpu_possible(cpu))
3294 continue;
3295 *pos = cpu+1;
3296 return per_cpu_ptr(tbl->stats, cpu);
3297 }
3298 (*pos)++;
3299 return NULL;
3300}
3301
3302static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
3303{
3304
3305}
3306
3307static int neigh_stat_seq_show(struct seq_file *seq, void *v)
3308{
3309 struct neigh_table *tbl = PDE_DATA(file_inode(seq->file));
3310 struct neigh_statistics *st = v;
3311
3312 if (v == SEQ_START_TOKEN) {
3313 seq_printf(seq, "entries allocs destroys hash_grows lookups hits res_failed rcv_probes_mcast rcv_probes_ucast periodic_gc_runs forced_gc_runs unresolved_discards table_fulls\n");
3314 return 0;
3315 }
3316
3317 seq_printf(seq, "%08x %08lx %08lx %08lx %08lx %08lx %08lx "
3318 "%08lx %08lx %08lx %08lx %08lx %08lx\n",
3319 atomic_read(&tbl->entries),
3320
3321 st->allocs,
3322 st->destroys,
3323 st->hash_grows,
3324
3325 st->lookups,
3326 st->hits,
3327
3328 st->res_failed,
3329
3330 st->rcv_probes_mcast,
3331 st->rcv_probes_ucast,
3332
3333 st->periodic_gc_runs,
3334 st->forced_gc_runs,
3335 st->unres_discards,
3336 st->table_fulls
3337 );
3338
3339 return 0;
3340}
3341
3342static const struct seq_operations neigh_stat_seq_ops = {
3343 .start = neigh_stat_seq_start,
3344 .next = neigh_stat_seq_next,
3345 .stop = neigh_stat_seq_stop,
3346 .show = neigh_stat_seq_show,
3347};
3348#endif /* CONFIG_PROC_FS */
3349
3350static void __neigh_notify(struct neighbour *n, int type, int flags,
3351 u32 pid)
3352{
3353 struct net *net = dev_net(n->dev);
3354 struct sk_buff *skb;
3355 int err = -ENOBUFS;
3356
3357 skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC);
3358 if (skb == NULL)
3359 goto errout;
3360
3361 err = neigh_fill_info(skb, n, pid, 0, type, flags);
3362 if (err < 0) {
3363 /* -EMSGSIZE implies BUG in neigh_nlmsg_size() */
3364 WARN_ON(err == -EMSGSIZE);
3365 kfree_skb(skb);
3366 goto errout;
3367 }
3368 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
3369 return;
3370errout:
3371 if (err < 0)
3372 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
3373}
3374
3375void neigh_app_ns(struct neighbour *n)
3376{
3377 __neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST, 0);
3378}
3379EXPORT_SYMBOL(neigh_app_ns);
3380
3381#ifdef CONFIG_SYSCTL
3382static int unres_qlen_max = INT_MAX / SKB_TRUESIZE(ETH_FRAME_LEN);
3383
3384static int proc_unres_qlen(struct ctl_table *ctl, int write,
3385 void *buffer, size_t *lenp, loff_t *ppos)
3386{
3387 int size, ret;
3388 struct ctl_table tmp = *ctl;
3389
3390 tmp.extra1 = SYSCTL_ZERO;
3391 tmp.extra2 = &unres_qlen_max;
3392 tmp.data = &size;
3393
3394 size = *(int *)ctl->data / SKB_TRUESIZE(ETH_FRAME_LEN);
3395 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
3396
3397 if (write && !ret)
3398 *(int *)ctl->data = size * SKB_TRUESIZE(ETH_FRAME_LEN);
3399 return ret;
3400}
3401
3402static struct neigh_parms *neigh_get_dev_parms_rcu(struct net_device *dev,
3403 int family)
3404{
3405 switch (family) {
3406 case AF_INET:
3407 return __in_dev_arp_parms_get_rcu(dev);
3408 case AF_INET6:
3409 return __in6_dev_nd_parms_get_rcu(dev);
3410 }
3411 return NULL;
3412}
3413
3414static void neigh_copy_dflt_parms(struct net *net, struct neigh_parms *p,
3415 int index)
3416{
3417 struct net_device *dev;
3418 int family = neigh_parms_family(p);
3419
3420 rcu_read_lock();
3421 for_each_netdev_rcu(net, dev) {
3422 struct neigh_parms *dst_p =
3423 neigh_get_dev_parms_rcu(dev, family);
3424
3425 if (dst_p && !test_bit(index, dst_p->data_state))
3426 dst_p->data[index] = p->data[index];
3427 }
3428 rcu_read_unlock();
3429}
3430
3431static void neigh_proc_update(struct ctl_table *ctl, int write)
3432{
3433 struct net_device *dev = ctl->extra1;
3434 struct neigh_parms *p = ctl->extra2;
3435 struct net *net = neigh_parms_net(p);
3436 int index = (int *) ctl->data - p->data;
3437
3438 if (!write)
3439 return;
3440
3441 set_bit(index, p->data_state);
3442 if (index == NEIGH_VAR_DELAY_PROBE_TIME)
3443 call_netevent_notifiers(NETEVENT_DELAY_PROBE_TIME_UPDATE, p);
3444 if (!dev) /* NULL dev means this is default value */
3445 neigh_copy_dflt_parms(net, p, index);
3446}
3447
3448static int neigh_proc_dointvec_zero_intmax(struct ctl_table *ctl, int write,
3449 void *buffer, size_t *lenp,
3450 loff_t *ppos)
3451{
3452 struct ctl_table tmp = *ctl;
3453 int ret;
3454
3455 tmp.extra1 = SYSCTL_ZERO;
3456 tmp.extra2 = SYSCTL_INT_MAX;
3457
3458 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
3459 neigh_proc_update(ctl, write);
3460 return ret;
3461}
3462
3463int neigh_proc_dointvec(struct ctl_table *ctl, int write, void *buffer,
3464 size_t *lenp, loff_t *ppos)
3465{
3466 int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
3467
3468 neigh_proc_update(ctl, write);
3469 return ret;
3470}
3471EXPORT_SYMBOL(neigh_proc_dointvec);
3472
3473int neigh_proc_dointvec_jiffies(struct ctl_table *ctl, int write, void *buffer,
3474 size_t *lenp, loff_t *ppos)
3475{
3476 int ret = proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
3477
3478 neigh_proc_update(ctl, write);
3479 return ret;
3480}
3481EXPORT_SYMBOL(neigh_proc_dointvec_jiffies);
3482
3483static int neigh_proc_dointvec_userhz_jiffies(struct ctl_table *ctl, int write,
3484 void *buffer, size_t *lenp,
3485 loff_t *ppos)
3486{
3487 int ret = proc_dointvec_userhz_jiffies(ctl, write, buffer, lenp, ppos);
3488
3489 neigh_proc_update(ctl, write);
3490 return ret;
3491}
3492
3493int neigh_proc_dointvec_ms_jiffies(struct ctl_table *ctl, int write,
3494 void *buffer, size_t *lenp, loff_t *ppos)
3495{
3496 int ret = proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
3497
3498 neigh_proc_update(ctl, write);
3499 return ret;
3500}
3501EXPORT_SYMBOL(neigh_proc_dointvec_ms_jiffies);
3502
3503static int neigh_proc_dointvec_unres_qlen(struct ctl_table *ctl, int write,
3504 void *buffer, size_t *lenp,
3505 loff_t *ppos)
3506{
3507 int ret = proc_unres_qlen(ctl, write, buffer, lenp, ppos);
3508
3509 neigh_proc_update(ctl, write);
3510 return ret;
3511}
3512
3513static int neigh_proc_base_reachable_time(struct ctl_table *ctl, int write,
3514 void *buffer, size_t *lenp,
3515 loff_t *ppos)
3516{
3517 struct neigh_parms *p = ctl->extra2;
3518 int ret;
3519
3520 if (strcmp(ctl->procname, "base_reachable_time") == 0)
3521 ret = neigh_proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
3522 else if (strcmp(ctl->procname, "base_reachable_time_ms") == 0)
3523 ret = neigh_proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
3524 else
3525 ret = -1;
3526
3527 if (write && ret == 0) {
3528 /* update reachable_time as well, otherwise, the change will
3529 * only be effective after the next time neigh_periodic_work
3530 * decides to recompute it
3531 */
3532 p->reachable_time =
3533 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
3534 }
3535 return ret;
3536}
3537
3538#define NEIGH_PARMS_DATA_OFFSET(index) \
3539 (&((struct neigh_parms *) 0)->data[index])
3540
3541#define NEIGH_SYSCTL_ENTRY(attr, data_attr, name, mval, proc) \
3542 [NEIGH_VAR_ ## attr] = { \
3543 .procname = name, \
3544 .data = NEIGH_PARMS_DATA_OFFSET(NEIGH_VAR_ ## data_attr), \
3545 .maxlen = sizeof(int), \
3546 .mode = mval, \
3547 .proc_handler = proc, \
3548 }
3549
3550#define NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(attr, name) \
3551 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_zero_intmax)
3552
3553#define NEIGH_SYSCTL_JIFFIES_ENTRY(attr, name) \
3554 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_jiffies)
3555
3556#define NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(attr, name) \
3557 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_userhz_jiffies)
3558
3559#define NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(attr, data_attr, name) \
3560 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_ms_jiffies)
3561
3562#define NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(attr, data_attr, name) \
3563 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_unres_qlen)
3564
3565static struct neigh_sysctl_table {
3566 struct ctl_table_header *sysctl_header;
3567 struct ctl_table neigh_vars[NEIGH_VAR_MAX + 1];
3568} neigh_sysctl_template __read_mostly = {
3569 .neigh_vars = {
3570 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_PROBES, "mcast_solicit"),
3571 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(UCAST_PROBES, "ucast_solicit"),
3572 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(APP_PROBES, "app_solicit"),
3573 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_REPROBES, "mcast_resolicit"),
3574 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(RETRANS_TIME, "retrans_time"),
3575 NEIGH_SYSCTL_JIFFIES_ENTRY(BASE_REACHABLE_TIME, "base_reachable_time"),
3576 NEIGH_SYSCTL_JIFFIES_ENTRY(DELAY_PROBE_TIME, "delay_first_probe_time"),
3577 NEIGH_SYSCTL_JIFFIES_ENTRY(GC_STALETIME, "gc_stale_time"),
3578 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(QUEUE_LEN_BYTES, "unres_qlen_bytes"),
3579 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(PROXY_QLEN, "proxy_qlen"),
3580 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(ANYCAST_DELAY, "anycast_delay"),
3581 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(PROXY_DELAY, "proxy_delay"),
3582 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(LOCKTIME, "locktime"),
3583 NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(QUEUE_LEN, QUEUE_LEN_BYTES, "unres_qlen"),
3584 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(RETRANS_TIME_MS, RETRANS_TIME, "retrans_time_ms"),
3585 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(BASE_REACHABLE_TIME_MS, BASE_REACHABLE_TIME, "base_reachable_time_ms"),
3586 [NEIGH_VAR_GC_INTERVAL] = {
3587 .procname = "gc_interval",
3588 .maxlen = sizeof(int),
3589 .mode = 0644,
3590 .proc_handler = proc_dointvec_jiffies,
3591 },
3592 [NEIGH_VAR_GC_THRESH1] = {
3593 .procname = "gc_thresh1",
3594 .maxlen = sizeof(int),
3595 .mode = 0644,
3596 .extra1 = SYSCTL_ZERO,
3597 .extra2 = SYSCTL_INT_MAX,
3598 .proc_handler = proc_dointvec_minmax,
3599 },
3600 [NEIGH_VAR_GC_THRESH2] = {
3601 .procname = "gc_thresh2",
3602 .maxlen = sizeof(int),
3603 .mode = 0644,
3604 .extra1 = SYSCTL_ZERO,
3605 .extra2 = SYSCTL_INT_MAX,
3606 .proc_handler = proc_dointvec_minmax,
3607 },
3608 [NEIGH_VAR_GC_THRESH3] = {
3609 .procname = "gc_thresh3",
3610 .maxlen = sizeof(int),
3611 .mode = 0644,
3612 .extra1 = SYSCTL_ZERO,
3613 .extra2 = SYSCTL_INT_MAX,
3614 .proc_handler = proc_dointvec_minmax,
3615 },
3616 {},
3617 },
3618};
3619
3620int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
3621 proc_handler *handler)
3622{
3623 int i;
3624 struct neigh_sysctl_table *t;
3625 const char *dev_name_source;
3626 char neigh_path[ sizeof("net//neigh/") + IFNAMSIZ + IFNAMSIZ ];
3627 char *p_name;
3628
3629 t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL);
3630 if (!t)
3631 goto err;
3632
3633 for (i = 0; i < NEIGH_VAR_GC_INTERVAL; i++) {
3634 t->neigh_vars[i].data += (long) p;
3635 t->neigh_vars[i].extra1 = dev;
3636 t->neigh_vars[i].extra2 = p;
3637 }
3638
3639 if (dev) {
3640 dev_name_source = dev->name;
3641 /* Terminate the table early */
3642 memset(&t->neigh_vars[NEIGH_VAR_GC_INTERVAL], 0,
3643 sizeof(t->neigh_vars[NEIGH_VAR_GC_INTERVAL]));
3644 } else {
3645 struct neigh_table *tbl = p->tbl;
3646 dev_name_source = "default";
3647 t->neigh_vars[NEIGH_VAR_GC_INTERVAL].data = &tbl->gc_interval;
3648 t->neigh_vars[NEIGH_VAR_GC_THRESH1].data = &tbl->gc_thresh1;
3649 t->neigh_vars[NEIGH_VAR_GC_THRESH2].data = &tbl->gc_thresh2;
3650 t->neigh_vars[NEIGH_VAR_GC_THRESH3].data = &tbl->gc_thresh3;
3651 }
3652
3653 if (handler) {
3654 /* RetransTime */
3655 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].proc_handler = handler;
3656 /* ReachableTime */
3657 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler = handler;
3658 /* RetransTime (in milliseconds)*/
3659 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].proc_handler = handler;
3660 /* ReachableTime (in milliseconds) */
3661 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler = handler;
3662 } else {
3663 /* Those handlers will update p->reachable_time after
3664 * base_reachable_time(_ms) is set to ensure the new timer starts being
3665 * applied after the next neighbour update instead of waiting for
3666 * neigh_periodic_work to update its value (can be multiple minutes)
3667 * So any handler that replaces them should do this as well
3668 */
3669 /* ReachableTime */
3670 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler =
3671 neigh_proc_base_reachable_time;
3672 /* ReachableTime (in milliseconds) */
3673 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler =
3674 neigh_proc_base_reachable_time;
3675 }
3676
3677 /* Don't export sysctls to unprivileged users */
3678 if (neigh_parms_net(p)->user_ns != &init_user_ns)
3679 t->neigh_vars[0].procname = NULL;
3680
3681 switch (neigh_parms_family(p)) {
3682 case AF_INET:
3683 p_name = "ipv4";
3684 break;
3685 case AF_INET6:
3686 p_name = "ipv6";
3687 break;
3688 default:
3689 BUG();
3690 }
3691
3692 snprintf(neigh_path, sizeof(neigh_path), "net/%s/neigh/%s",
3693 p_name, dev_name_source);
3694 t->sysctl_header =
3695 register_net_sysctl(neigh_parms_net(p), neigh_path, t->neigh_vars);
3696 if (!t->sysctl_header)
3697 goto free;
3698
3699 p->sysctl_table = t;
3700 return 0;
3701
3702free:
3703 kfree(t);
3704err:
3705 return -ENOBUFS;
3706}
3707EXPORT_SYMBOL(neigh_sysctl_register);
3708
3709void neigh_sysctl_unregister(struct neigh_parms *p)
3710{
3711 if (p->sysctl_table) {
3712 struct neigh_sysctl_table *t = p->sysctl_table;
3713 p->sysctl_table = NULL;
3714 unregister_net_sysctl_table(t->sysctl_header);
3715 kfree(t);
3716 }
3717}
3718EXPORT_SYMBOL(neigh_sysctl_unregister);
3719
3720#endif /* CONFIG_SYSCTL */
3721
3722static int __init neigh_init(void)
3723{
3724 rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL, 0);
3725 rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL, 0);
3726 rtnl_register(PF_UNSPEC, RTM_GETNEIGH, neigh_get, neigh_dump_info, 0);
3727
3728 rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info,
3729 0);
3730 rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL, 0);
3731
3732 return 0;
3733}
3734
3735subsys_initcall(neigh_init);