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#include <linux/slab.h>
19#include <linux/types.h>
20#include <linux/kernel.h>
21#include <linux/module.h>
22#include <linux/socket.h>
23#include <linux/netdevice.h>
24#include <linux/proc_fs.h>
25#ifdef CONFIG_SYSCTL
26#include <linux/sysctl.h>
27#endif
28#include <linux/times.h>
29#include <net/net_namespace.h>
30#include <net/neighbour.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
40#define NEIGH_DEBUG 1
41
42#define NEIGH_PRINTK(x...) printk(x)
43#define NEIGH_NOPRINTK(x...) do { ; } while(0)
44#define NEIGH_PRINTK1 NEIGH_NOPRINTK
45#define NEIGH_PRINTK2 NEIGH_NOPRINTK
46
47#if NEIGH_DEBUG >= 1
48#undef NEIGH_PRINTK1
49#define NEIGH_PRINTK1 NEIGH_PRINTK
50#endif
51#if NEIGH_DEBUG >= 2
52#undef NEIGH_PRINTK2
53#define NEIGH_PRINTK2 NEIGH_PRINTK
54#endif
55
56#define PNEIGH_HASHMASK 0xF
57
58static void neigh_timer_handler(unsigned long arg);
59static void __neigh_notify(struct neighbour *n, int type, int flags);
60static void neigh_update_notify(struct neighbour *neigh);
61static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev);
62
63static struct neigh_table *neigh_tables;
64#ifdef CONFIG_PROC_FS
65static const struct file_operations neigh_stat_seq_fops;
66#endif
67
68/*
69 Neighbour hash table buckets are protected with rwlock tbl->lock.
70
71 - All the scans/updates to hash buckets MUST be made under this lock.
72 - NOTHING clever should be made under this lock: no callbacks
73 to protocol backends, no attempts to send something to network.
74 It will result in deadlocks, if backend/driver wants to use neighbour
75 cache.
76 - If the entry requires some non-trivial actions, increase
77 its reference count and release table lock.
78
79 Neighbour entries are protected:
80 - with reference count.
81 - with rwlock neigh->lock
82
83 Reference count prevents destruction.
84
85 neigh->lock mainly serializes ll address data and its validity state.
86 However, the same lock is used to protect another entry fields:
87 - timer
88 - resolution queue
89
90 Again, nothing clever shall be made under neigh->lock,
91 the most complicated procedure, which we allow is dev->hard_header.
92 It is supposed, that dev->hard_header is simplistic and does
93 not make callbacks to neighbour tables.
94
95 The last lock is neigh_tbl_lock. It is pure SMP lock, protecting
96 list of neighbour tables. This list is used only in process context,
97 */
98
99static DEFINE_RWLOCK(neigh_tbl_lock);
100
101static int neigh_blackhole(struct neighbour *neigh, struct sk_buff *skb)
102{
103 kfree_skb(skb);
104 return -ENETDOWN;
105}
106
107static void neigh_cleanup_and_release(struct neighbour *neigh)
108{
109 if (neigh->parms->neigh_cleanup)
110 neigh->parms->neigh_cleanup(neigh);
111
112 __neigh_notify(neigh, RTM_DELNEIGH, 0);
113 neigh_release(neigh);
114}
115
116/*
117 * It is random distribution in the interval (1/2)*base...(3/2)*base.
118 * It corresponds to default IPv6 settings and is not overridable,
119 * because it is really reasonable choice.
120 */
121
122unsigned long neigh_rand_reach_time(unsigned long base)
123{
124 return base ? (net_random() % base) + (base >> 1) : 0;
125}
126EXPORT_SYMBOL(neigh_rand_reach_time);
127
128
129static int neigh_forced_gc(struct neigh_table *tbl)
130{
131 int shrunk = 0;
132 int i;
133 struct neigh_hash_table *nht;
134
135 NEIGH_CACHE_STAT_INC(tbl, forced_gc_runs);
136
137 write_lock_bh(&tbl->lock);
138 nht = rcu_dereference_protected(tbl->nht,
139 lockdep_is_held(&tbl->lock));
140 for (i = 0; i < (1 << nht->hash_shift); i++) {
141 struct neighbour *n;
142 struct neighbour __rcu **np;
143
144 np = &nht->hash_buckets[i];
145 while ((n = rcu_dereference_protected(*np,
146 lockdep_is_held(&tbl->lock))) != NULL) {
147 /* Neighbour record may be discarded if:
148 * - nobody refers to it.
149 * - it is not permanent
150 */
151 write_lock(&n->lock);
152 if (atomic_read(&n->refcnt) == 1 &&
153 !(n->nud_state & NUD_PERMANENT)) {
154 rcu_assign_pointer(*np,
155 rcu_dereference_protected(n->next,
156 lockdep_is_held(&tbl->lock)));
157 n->dead = 1;
158 shrunk = 1;
159 write_unlock(&n->lock);
160 neigh_cleanup_and_release(n);
161 continue;
162 }
163 write_unlock(&n->lock);
164 np = &n->next;
165 }
166 }
167
168 tbl->last_flush = jiffies;
169
170 write_unlock_bh(&tbl->lock);
171
172 return shrunk;
173}
174
175static void neigh_add_timer(struct neighbour *n, unsigned long when)
176{
177 neigh_hold(n);
178 if (unlikely(mod_timer(&n->timer, when))) {
179 printk("NEIGH: BUG, double timer add, state is %x\n",
180 n->nud_state);
181 dump_stack();
182 }
183}
184
185static int neigh_del_timer(struct neighbour *n)
186{
187 if ((n->nud_state & NUD_IN_TIMER) &&
188 del_timer(&n->timer)) {
189 neigh_release(n);
190 return 1;
191 }
192 return 0;
193}
194
195static void pneigh_queue_purge(struct sk_buff_head *list)
196{
197 struct sk_buff *skb;
198
199 while ((skb = skb_dequeue(list)) != NULL) {
200 dev_put(skb->dev);
201 kfree_skb(skb);
202 }
203}
204
205static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev)
206{
207 int i;
208 struct neigh_hash_table *nht;
209
210 nht = rcu_dereference_protected(tbl->nht,
211 lockdep_is_held(&tbl->lock));
212
213 for (i = 0; i < (1 << nht->hash_shift); i++) {
214 struct neighbour *n;
215 struct neighbour __rcu **np = &nht->hash_buckets[i];
216
217 while ((n = rcu_dereference_protected(*np,
218 lockdep_is_held(&tbl->lock))) != NULL) {
219 if (dev && n->dev != dev) {
220 np = &n->next;
221 continue;
222 }
223 rcu_assign_pointer(*np,
224 rcu_dereference_protected(n->next,
225 lockdep_is_held(&tbl->lock)));
226 write_lock(&n->lock);
227 neigh_del_timer(n);
228 n->dead = 1;
229
230 if (atomic_read(&n->refcnt) != 1) {
231 /* The most unpleasant situation.
232 We must destroy neighbour entry,
233 but someone still uses it.
234
235 The destroy will be delayed until
236 the last user releases us, but
237 we must kill timers etc. and move
238 it to safe state.
239 */
240 skb_queue_purge(&n->arp_queue);
241 n->output = neigh_blackhole;
242 if (n->nud_state & NUD_VALID)
243 n->nud_state = NUD_NOARP;
244 else
245 n->nud_state = NUD_NONE;
246 NEIGH_PRINTK2("neigh %p is stray.\n", n);
247 }
248 write_unlock(&n->lock);
249 neigh_cleanup_and_release(n);
250 }
251 }
252}
253
254void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev)
255{
256 write_lock_bh(&tbl->lock);
257 neigh_flush_dev(tbl, dev);
258 write_unlock_bh(&tbl->lock);
259}
260EXPORT_SYMBOL(neigh_changeaddr);
261
262int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
263{
264 write_lock_bh(&tbl->lock);
265 neigh_flush_dev(tbl, dev);
266 pneigh_ifdown(tbl, dev);
267 write_unlock_bh(&tbl->lock);
268
269 del_timer_sync(&tbl->proxy_timer);
270 pneigh_queue_purge(&tbl->proxy_queue);
271 return 0;
272}
273EXPORT_SYMBOL(neigh_ifdown);
274
275static struct neighbour *neigh_alloc(struct neigh_table *tbl)
276{
277 struct neighbour *n = NULL;
278 unsigned long now = jiffies;
279 int entries;
280
281 entries = atomic_inc_return(&tbl->entries) - 1;
282 if (entries >= tbl->gc_thresh3 ||
283 (entries >= tbl->gc_thresh2 &&
284 time_after(now, tbl->last_flush + 5 * HZ))) {
285 if (!neigh_forced_gc(tbl) &&
286 entries >= tbl->gc_thresh3)
287 goto out_entries;
288 }
289
290 n = kmem_cache_zalloc(tbl->kmem_cachep, GFP_ATOMIC);
291 if (!n)
292 goto out_entries;
293
294 skb_queue_head_init(&n->arp_queue);
295 rwlock_init(&n->lock);
296 seqlock_init(&n->ha_lock);
297 n->updated = n->used = now;
298 n->nud_state = NUD_NONE;
299 n->output = neigh_blackhole;
300 seqlock_init(&n->hh.hh_lock);
301 n->parms = neigh_parms_clone(&tbl->parms);
302 setup_timer(&n->timer, neigh_timer_handler, (unsigned long)n);
303
304 NEIGH_CACHE_STAT_INC(tbl, allocs);
305 n->tbl = tbl;
306 atomic_set(&n->refcnt, 1);
307 n->dead = 1;
308out:
309 return n;
310
311out_entries:
312 atomic_dec(&tbl->entries);
313 goto out;
314}
315
316static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift)
317{
318 size_t size = (1 << shift) * sizeof(struct neighbour *);
319 struct neigh_hash_table *ret;
320 struct neighbour __rcu **buckets;
321
322 ret = kmalloc(sizeof(*ret), GFP_ATOMIC);
323 if (!ret)
324 return NULL;
325 if (size <= PAGE_SIZE)
326 buckets = kzalloc(size, GFP_ATOMIC);
327 else
328 buckets = (struct neighbour __rcu **)
329 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
330 get_order(size));
331 if (!buckets) {
332 kfree(ret);
333 return NULL;
334 }
335 ret->hash_buckets = buckets;
336 ret->hash_shift = shift;
337 get_random_bytes(&ret->hash_rnd, sizeof(ret->hash_rnd));
338 ret->hash_rnd |= 1;
339 return ret;
340}
341
342static void neigh_hash_free_rcu(struct rcu_head *head)
343{
344 struct neigh_hash_table *nht = container_of(head,
345 struct neigh_hash_table,
346 rcu);
347 size_t size = (1 << nht->hash_shift) * sizeof(struct neighbour *);
348 struct neighbour __rcu **buckets = nht->hash_buckets;
349
350 if (size <= PAGE_SIZE)
351 kfree(buckets);
352 else
353 free_pages((unsigned long)buckets, get_order(size));
354 kfree(nht);
355}
356
357static struct neigh_hash_table *neigh_hash_grow(struct neigh_table *tbl,
358 unsigned long new_shift)
359{
360 unsigned int i, hash;
361 struct neigh_hash_table *new_nht, *old_nht;
362
363 NEIGH_CACHE_STAT_INC(tbl, hash_grows);
364
365 old_nht = rcu_dereference_protected(tbl->nht,
366 lockdep_is_held(&tbl->lock));
367 new_nht = neigh_hash_alloc(new_shift);
368 if (!new_nht)
369 return old_nht;
370
371 for (i = 0; i < (1 << old_nht->hash_shift); i++) {
372 struct neighbour *n, *next;
373
374 for (n = rcu_dereference_protected(old_nht->hash_buckets[i],
375 lockdep_is_held(&tbl->lock));
376 n != NULL;
377 n = next) {
378 hash = tbl->hash(n->primary_key, n->dev,
379 new_nht->hash_rnd);
380
381 hash >>= (32 - new_nht->hash_shift);
382 next = rcu_dereference_protected(n->next,
383 lockdep_is_held(&tbl->lock));
384
385 rcu_assign_pointer(n->next,
386 rcu_dereference_protected(
387 new_nht->hash_buckets[hash],
388 lockdep_is_held(&tbl->lock)));
389 rcu_assign_pointer(new_nht->hash_buckets[hash], n);
390 }
391 }
392
393 rcu_assign_pointer(tbl->nht, new_nht);
394 call_rcu(&old_nht->rcu, neigh_hash_free_rcu);
395 return new_nht;
396}
397
398struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
399 struct net_device *dev)
400{
401 struct neighbour *n;
402 int key_len = tbl->key_len;
403 u32 hash_val;
404 struct neigh_hash_table *nht;
405
406 NEIGH_CACHE_STAT_INC(tbl, lookups);
407
408 rcu_read_lock_bh();
409 nht = rcu_dereference_bh(tbl->nht);
410 hash_val = tbl->hash(pkey, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
411
412 for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]);
413 n != NULL;
414 n = rcu_dereference_bh(n->next)) {
415 if (dev == n->dev && !memcmp(n->primary_key, pkey, key_len)) {
416 if (!atomic_inc_not_zero(&n->refcnt))
417 n = NULL;
418 NEIGH_CACHE_STAT_INC(tbl, hits);
419 break;
420 }
421 }
422
423 rcu_read_unlock_bh();
424 return n;
425}
426EXPORT_SYMBOL(neigh_lookup);
427
428struct neighbour *neigh_lookup_nodev(struct neigh_table *tbl, struct net *net,
429 const void *pkey)
430{
431 struct neighbour *n;
432 int key_len = tbl->key_len;
433 u32 hash_val;
434 struct neigh_hash_table *nht;
435
436 NEIGH_CACHE_STAT_INC(tbl, lookups);
437
438 rcu_read_lock_bh();
439 nht = rcu_dereference_bh(tbl->nht);
440 hash_val = tbl->hash(pkey, NULL, nht->hash_rnd) >> (32 - nht->hash_shift);
441
442 for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]);
443 n != NULL;
444 n = rcu_dereference_bh(n->next)) {
445 if (!memcmp(n->primary_key, pkey, key_len) &&
446 net_eq(dev_net(n->dev), net)) {
447 if (!atomic_inc_not_zero(&n->refcnt))
448 n = NULL;
449 NEIGH_CACHE_STAT_INC(tbl, hits);
450 break;
451 }
452 }
453
454 rcu_read_unlock_bh();
455 return n;
456}
457EXPORT_SYMBOL(neigh_lookup_nodev);
458
459struct neighbour *neigh_create(struct neigh_table *tbl, const void *pkey,
460 struct net_device *dev)
461{
462 u32 hash_val;
463 int key_len = tbl->key_len;
464 int error;
465 struct neighbour *n1, *rc, *n = neigh_alloc(tbl);
466 struct neigh_hash_table *nht;
467
468 if (!n) {
469 rc = ERR_PTR(-ENOBUFS);
470 goto out;
471 }
472
473 memcpy(n->primary_key, pkey, key_len);
474 n->dev = dev;
475 dev_hold(dev);
476
477 /* Protocol specific setup. */
478 if (tbl->constructor && (error = tbl->constructor(n)) < 0) {
479 rc = ERR_PTR(error);
480 goto out_neigh_release;
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 - (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 neigh_hold(n1);
513 rc = n1;
514 goto out_tbl_unlock;
515 }
516 }
517
518 n->dead = 0;
519 neigh_hold(n);
520 rcu_assign_pointer(n->next,
521 rcu_dereference_protected(nht->hash_buckets[hash_val],
522 lockdep_is_held(&tbl->lock)));
523 rcu_assign_pointer(nht->hash_buckets[hash_val], n);
524 write_unlock_bh(&tbl->lock);
525 NEIGH_PRINTK2("neigh %p is created.\n", n);
526 rc = n;
527out:
528 return rc;
529out_tbl_unlock:
530 write_unlock_bh(&tbl->lock);
531out_neigh_release:
532 neigh_release(n);
533 goto out;
534}
535EXPORT_SYMBOL(neigh_create);
536
537static u32 pneigh_hash(const void *pkey, int key_len)
538{
539 u32 hash_val = *(u32 *)(pkey + key_len - 4);
540 hash_val ^= (hash_val >> 16);
541 hash_val ^= hash_val >> 8;
542 hash_val ^= hash_val >> 4;
543 hash_val &= PNEIGH_HASHMASK;
544 return hash_val;
545}
546
547static struct pneigh_entry *__pneigh_lookup_1(struct pneigh_entry *n,
548 struct net *net,
549 const void *pkey,
550 int key_len,
551 struct net_device *dev)
552{
553 while (n) {
554 if (!memcmp(n->key, pkey, key_len) &&
555 net_eq(pneigh_net(n), net) &&
556 (n->dev == dev || !n->dev))
557 return n;
558 n = n->next;
559 }
560 return NULL;
561}
562
563struct pneigh_entry *__pneigh_lookup(struct neigh_table *tbl,
564 struct net *net, const void *pkey, struct net_device *dev)
565{
566 int key_len = tbl->key_len;
567 u32 hash_val = pneigh_hash(pkey, key_len);
568
569 return __pneigh_lookup_1(tbl->phash_buckets[hash_val],
570 net, pkey, key_len, dev);
571}
572EXPORT_SYMBOL_GPL(__pneigh_lookup);
573
574struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl,
575 struct net *net, const void *pkey,
576 struct net_device *dev, int creat)
577{
578 struct pneigh_entry *n;
579 int key_len = tbl->key_len;
580 u32 hash_val = pneigh_hash(pkey, key_len);
581
582 read_lock_bh(&tbl->lock);
583 n = __pneigh_lookup_1(tbl->phash_buckets[hash_val],
584 net, pkey, key_len, dev);
585 read_unlock_bh(&tbl->lock);
586
587 if (n || !creat)
588 goto out;
589
590 ASSERT_RTNL();
591
592 n = kmalloc(sizeof(*n) + key_len, GFP_KERNEL);
593 if (!n)
594 goto out;
595
596 write_pnet(&n->net, hold_net(net));
597 memcpy(n->key, pkey, key_len);
598 n->dev = dev;
599 if (dev)
600 dev_hold(dev);
601
602 if (tbl->pconstructor && tbl->pconstructor(n)) {
603 if (dev)
604 dev_put(dev);
605 release_net(net);
606 kfree(n);
607 n = NULL;
608 goto out;
609 }
610
611 write_lock_bh(&tbl->lock);
612 n->next = tbl->phash_buckets[hash_val];
613 tbl->phash_buckets[hash_val] = n;
614 write_unlock_bh(&tbl->lock);
615out:
616 return n;
617}
618EXPORT_SYMBOL(pneigh_lookup);
619
620
621int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
622 struct net_device *dev)
623{
624 struct pneigh_entry *n, **np;
625 int key_len = tbl->key_len;
626 u32 hash_val = pneigh_hash(pkey, key_len);
627
628 write_lock_bh(&tbl->lock);
629 for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
630 np = &n->next) {
631 if (!memcmp(n->key, pkey, key_len) && n->dev == dev &&
632 net_eq(pneigh_net(n), net)) {
633 *np = n->next;
634 write_unlock_bh(&tbl->lock);
635 if (tbl->pdestructor)
636 tbl->pdestructor(n);
637 if (n->dev)
638 dev_put(n->dev);
639 release_net(pneigh_net(n));
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 release_net(pneigh_net(n));
663 kfree(n);
664 continue;
665 }
666 np = &n->next;
667 }
668 }
669 return -ENOENT;
670}
671
672static void neigh_parms_destroy(struct neigh_parms *parms);
673
674static inline void neigh_parms_put(struct neigh_parms *parms)
675{
676 if (atomic_dec_and_test(&parms->refcnt))
677 neigh_parms_destroy(parms);
678}
679
680static void neigh_destroy_rcu(struct rcu_head *head)
681{
682 struct neighbour *neigh = container_of(head, struct neighbour, rcu);
683
684 kmem_cache_free(neigh->tbl->kmem_cachep, neigh);
685}
686/*
687 * neighbour must already be out of the table;
688 *
689 */
690void neigh_destroy(struct neighbour *neigh)
691{
692 NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
693
694 if (!neigh->dead) {
695 printk(KERN_WARNING
696 "Destroying alive neighbour %p\n", neigh);
697 dump_stack();
698 return;
699 }
700
701 if (neigh_del_timer(neigh))
702 printk(KERN_WARNING "Impossible event.\n");
703
704 skb_queue_purge(&neigh->arp_queue);
705
706 dev_put(neigh->dev);
707 neigh_parms_put(neigh->parms);
708
709 NEIGH_PRINTK2("neigh %p is destroyed.\n", neigh);
710
711 atomic_dec(&neigh->tbl->entries);
712 call_rcu(&neigh->rcu, neigh_destroy_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_PRINTK2("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_PRINTK2("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 for (p = &tbl->parms; p; p = p->next)
762 p->reachable_time =
763 neigh_rand_reach_time(p->base_reachable_time);
764 }
765
766 for (i = 0 ; i < (1 << nht->hash_shift); i++) {
767 np = &nht->hash_buckets[i];
768
769 while ((n = rcu_dereference_protected(*np,
770 lockdep_is_held(&tbl->lock))) != NULL) {
771 unsigned int state;
772
773 write_lock(&n->lock);
774
775 state = n->nud_state;
776 if (state & (NUD_PERMANENT | NUD_IN_TIMER)) {
777 write_unlock(&n->lock);
778 goto next_elt;
779 }
780
781 if (time_before(n->used, n->confirmed))
782 n->used = n->confirmed;
783
784 if (atomic_read(&n->refcnt) == 1 &&
785 (state == NUD_FAILED ||
786 time_after(jiffies, n->used + n->parms->gc_staletime))) {
787 *np = n->next;
788 n->dead = 1;
789 write_unlock(&n->lock);
790 neigh_cleanup_and_release(n);
791 continue;
792 }
793 write_unlock(&n->lock);
794
795next_elt:
796 np = &n->next;
797 }
798 /*
799 * It's fine to release lock here, even if hash table
800 * grows while we are preempted.
801 */
802 write_unlock_bh(&tbl->lock);
803 cond_resched();
804 write_lock_bh(&tbl->lock);
805 }
806 /* Cycle through all hash buckets every base_reachable_time/2 ticks.
807 * ARP entry timeouts range from 1/2 base_reachable_time to 3/2
808 * base_reachable_time.
809 */
810 schedule_delayed_work(&tbl->gc_work,
811 tbl->parms.base_reachable_time >> 1);
812 write_unlock_bh(&tbl->lock);
813}
814
815static __inline__ int neigh_max_probes(struct neighbour *n)
816{
817 struct neigh_parms *p = n->parms;
818 return (n->nud_state & NUD_PROBE) ?
819 p->ucast_probes :
820 p->ucast_probes + p->app_probes + p->mcast_probes;
821}
822
823static void neigh_invalidate(struct neighbour *neigh)
824 __releases(neigh->lock)
825 __acquires(neigh->lock)
826{
827 struct sk_buff *skb;
828
829 NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
830 NEIGH_PRINTK2("neigh %p is failed.\n", neigh);
831 neigh->updated = jiffies;
832
833 /* It is very thin place. report_unreachable is very complicated
834 routine. Particularly, it can hit the same neighbour entry!
835
836 So that, we try to be accurate and avoid dead loop. --ANK
837 */
838 while (neigh->nud_state == NUD_FAILED &&
839 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
840 write_unlock(&neigh->lock);
841 neigh->ops->error_report(neigh, skb);
842 write_lock(&neigh->lock);
843 }
844 skb_queue_purge(&neigh->arp_queue);
845}
846
847/* Called when a timer expires for a neighbour entry. */
848
849static void neigh_timer_handler(unsigned long arg)
850{
851 unsigned long now, next;
852 struct neighbour *neigh = (struct neighbour *)arg;
853 unsigned state;
854 int notify = 0;
855
856 write_lock(&neigh->lock);
857
858 state = neigh->nud_state;
859 now = jiffies;
860 next = now + HZ;
861
862 if (!(state & NUD_IN_TIMER)) {
863#ifndef CONFIG_SMP
864 printk(KERN_WARNING "neigh: timer & !nud_in_timer\n");
865#endif
866 goto out;
867 }
868
869 if (state & NUD_REACHABLE) {
870 if (time_before_eq(now,
871 neigh->confirmed + neigh->parms->reachable_time)) {
872 NEIGH_PRINTK2("neigh %p is still alive.\n", neigh);
873 next = neigh->confirmed + neigh->parms->reachable_time;
874 } else if (time_before_eq(now,
875 neigh->used + neigh->parms->delay_probe_time)) {
876 NEIGH_PRINTK2("neigh %p is delayed.\n", neigh);
877 neigh->nud_state = NUD_DELAY;
878 neigh->updated = jiffies;
879 neigh_suspect(neigh);
880 next = now + neigh->parms->delay_probe_time;
881 } else {
882 NEIGH_PRINTK2("neigh %p is suspected.\n", neigh);
883 neigh->nud_state = NUD_STALE;
884 neigh->updated = jiffies;
885 neigh_suspect(neigh);
886 notify = 1;
887 }
888 } else if (state & NUD_DELAY) {
889 if (time_before_eq(now,
890 neigh->confirmed + neigh->parms->delay_probe_time)) {
891 NEIGH_PRINTK2("neigh %p is now reachable.\n", neigh);
892 neigh->nud_state = NUD_REACHABLE;
893 neigh->updated = jiffies;
894 neigh_connect(neigh);
895 notify = 1;
896 next = neigh->confirmed + neigh->parms->reachable_time;
897 } else {
898 NEIGH_PRINTK2("neigh %p is probed.\n", neigh);
899 neigh->nud_state = NUD_PROBE;
900 neigh->updated = jiffies;
901 atomic_set(&neigh->probes, 0);
902 next = now + neigh->parms->retrans_time;
903 }
904 } else {
905 /* NUD_PROBE|NUD_INCOMPLETE */
906 next = now + neigh->parms->retrans_time;
907 }
908
909 if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
910 atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
911 neigh->nud_state = NUD_FAILED;
912 notify = 1;
913 neigh_invalidate(neigh);
914 }
915
916 if (neigh->nud_state & NUD_IN_TIMER) {
917 if (time_before(next, jiffies + HZ/2))
918 next = jiffies + HZ/2;
919 if (!mod_timer(&neigh->timer, next))
920 neigh_hold(neigh);
921 }
922 if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
923 struct sk_buff *skb = skb_peek(&neigh->arp_queue);
924 /* keep skb alive even if arp_queue overflows */
925 if (skb)
926 skb = skb_copy(skb, GFP_ATOMIC);
927 write_unlock(&neigh->lock);
928 neigh->ops->solicit(neigh, skb);
929 atomic_inc(&neigh->probes);
930 kfree_skb(skb);
931 } else {
932out:
933 write_unlock(&neigh->lock);
934 }
935
936 if (notify)
937 neigh_update_notify(neigh);
938
939 neigh_release(neigh);
940}
941
942int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
943{
944 int rc;
945 unsigned long now;
946
947 write_lock_bh(&neigh->lock);
948
949 rc = 0;
950 if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
951 goto out_unlock_bh;
952
953 now = jiffies;
954
955 if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
956 if (neigh->parms->mcast_probes + neigh->parms->app_probes) {
957 atomic_set(&neigh->probes, neigh->parms->ucast_probes);
958 neigh->nud_state = NUD_INCOMPLETE;
959 neigh->updated = jiffies;
960 neigh_add_timer(neigh, now + 1);
961 } else {
962 neigh->nud_state = NUD_FAILED;
963 neigh->updated = jiffies;
964 write_unlock_bh(&neigh->lock);
965
966 kfree_skb(skb);
967 return 1;
968 }
969 } else if (neigh->nud_state & NUD_STALE) {
970 NEIGH_PRINTK2("neigh %p is delayed.\n", neigh);
971 neigh->nud_state = NUD_DELAY;
972 neigh->updated = jiffies;
973 neigh_add_timer(neigh,
974 jiffies + neigh->parms->delay_probe_time);
975 }
976
977 if (neigh->nud_state == NUD_INCOMPLETE) {
978 if (skb) {
979 if (skb_queue_len(&neigh->arp_queue) >=
980 neigh->parms->queue_len) {
981 struct sk_buff *buff;
982 buff = __skb_dequeue(&neigh->arp_queue);
983 kfree_skb(buff);
984 NEIGH_CACHE_STAT_INC(neigh->tbl, unres_discards);
985 }
986 skb_dst_force(skb);
987 __skb_queue_tail(&neigh->arp_queue, skb);
988 }
989 rc = 1;
990 }
991out_unlock_bh:
992 write_unlock_bh(&neigh->lock);
993 return rc;
994}
995EXPORT_SYMBOL(__neigh_event_send);
996
997static void neigh_update_hhs(struct neighbour *neigh)
998{
999 struct hh_cache *hh;
1000 void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *)
1001 = NULL;
1002
1003 if (neigh->dev->header_ops)
1004 update = neigh->dev->header_ops->cache_update;
1005
1006 if (update) {
1007 hh = &neigh->hh;
1008 if (hh->hh_len) {
1009 write_seqlock_bh(&hh->hh_lock);
1010 update(hh, neigh->dev, neigh->ha);
1011 write_sequnlock_bh(&hh->hh_lock);
1012 }
1013 }
1014}
1015
1016
1017
1018/* Generic update routine.
1019 -- lladdr is new lladdr or NULL, if it is not supplied.
1020 -- new is new state.
1021 -- flags
1022 NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
1023 if it is different.
1024 NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
1025 lladdr instead of overriding it
1026 if it is different.
1027 It also allows to retain current state
1028 if lladdr is unchanged.
1029 NEIGH_UPDATE_F_ADMIN means that the change is administrative.
1030
1031 NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
1032 NTF_ROUTER flag.
1033 NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as
1034 a router.
1035
1036 Caller MUST hold reference count on the entry.
1037 */
1038
1039int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
1040 u32 flags)
1041{
1042 u8 old;
1043 int err;
1044 int notify = 0;
1045 struct net_device *dev;
1046 int update_isrouter = 0;
1047
1048 write_lock_bh(&neigh->lock);
1049
1050 dev = neigh->dev;
1051 old = neigh->nud_state;
1052 err = -EPERM;
1053
1054 if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
1055 (old & (NUD_NOARP | NUD_PERMANENT)))
1056 goto out;
1057
1058 if (!(new & NUD_VALID)) {
1059 neigh_del_timer(neigh);
1060 if (old & NUD_CONNECTED)
1061 neigh_suspect(neigh);
1062 neigh->nud_state = new;
1063 err = 0;
1064 notify = old & NUD_VALID;
1065 if ((old & (NUD_INCOMPLETE | NUD_PROBE)) &&
1066 (new & NUD_FAILED)) {
1067 neigh_invalidate(neigh);
1068 notify = 1;
1069 }
1070 goto out;
1071 }
1072
1073 /* Compare new lladdr with cached one */
1074 if (!dev->addr_len) {
1075 /* First case: device needs no address. */
1076 lladdr = neigh->ha;
1077 } else if (lladdr) {
1078 /* The second case: if something is already cached
1079 and a new address is proposed:
1080 - compare new & old
1081 - if they are different, check override flag
1082 */
1083 if ((old & NUD_VALID) &&
1084 !memcmp(lladdr, neigh->ha, dev->addr_len))
1085 lladdr = neigh->ha;
1086 } else {
1087 /* No address is supplied; if we know something,
1088 use it, otherwise discard the request.
1089 */
1090 err = -EINVAL;
1091 if (!(old & NUD_VALID))
1092 goto out;
1093 lladdr = neigh->ha;
1094 }
1095
1096 if (new & NUD_CONNECTED)
1097 neigh->confirmed = jiffies;
1098 neigh->updated = jiffies;
1099
1100 /* If entry was valid and address is not changed,
1101 do not change entry state, if new one is STALE.
1102 */
1103 err = 0;
1104 update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1105 if (old & NUD_VALID) {
1106 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
1107 update_isrouter = 0;
1108 if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
1109 (old & NUD_CONNECTED)) {
1110 lladdr = neigh->ha;
1111 new = NUD_STALE;
1112 } else
1113 goto out;
1114 } else {
1115 if (lladdr == neigh->ha && new == NUD_STALE &&
1116 ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) ||
1117 (old & NUD_CONNECTED))
1118 )
1119 new = old;
1120 }
1121 }
1122
1123 if (new != old) {
1124 neigh_del_timer(neigh);
1125 if (new & NUD_IN_TIMER)
1126 neigh_add_timer(neigh, (jiffies +
1127 ((new & NUD_REACHABLE) ?
1128 neigh->parms->reachable_time :
1129 0)));
1130 neigh->nud_state = new;
1131 }
1132
1133 if (lladdr != neigh->ha) {
1134 write_seqlock(&neigh->ha_lock);
1135 memcpy(&neigh->ha, lladdr, dev->addr_len);
1136 write_sequnlock(&neigh->ha_lock);
1137 neigh_update_hhs(neigh);
1138 if (!(new & NUD_CONNECTED))
1139 neigh->confirmed = jiffies -
1140 (neigh->parms->base_reachable_time << 1);
1141 notify = 1;
1142 }
1143 if (new == old)
1144 goto out;
1145 if (new & NUD_CONNECTED)
1146 neigh_connect(neigh);
1147 else
1148 neigh_suspect(neigh);
1149 if (!(old & NUD_VALID)) {
1150 struct sk_buff *skb;
1151
1152 /* Again: avoid dead loop if something went wrong */
1153
1154 while (neigh->nud_state & NUD_VALID &&
1155 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1156 struct dst_entry *dst = skb_dst(skb);
1157 struct neighbour *n2, *n1 = neigh;
1158 write_unlock_bh(&neigh->lock);
1159 /* On shaper/eql skb->dst->neighbour != neigh :( */
1160 if (dst && (n2 = dst_get_neighbour(dst)) != NULL)
1161 n1 = n2;
1162 n1->output(n1, skb);
1163 write_lock_bh(&neigh->lock);
1164 }
1165 skb_queue_purge(&neigh->arp_queue);
1166 }
1167out:
1168 if (update_isrouter) {
1169 neigh->flags = (flags & NEIGH_UPDATE_F_ISROUTER) ?
1170 (neigh->flags | NTF_ROUTER) :
1171 (neigh->flags & ~NTF_ROUTER);
1172 }
1173 write_unlock_bh(&neigh->lock);
1174
1175 if (notify)
1176 neigh_update_notify(neigh);
1177
1178 return err;
1179}
1180EXPORT_SYMBOL(neigh_update);
1181
1182struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1183 u8 *lladdr, void *saddr,
1184 struct net_device *dev)
1185{
1186 struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1187 lladdr || !dev->addr_len);
1188 if (neigh)
1189 neigh_update(neigh, lladdr, NUD_STALE,
1190 NEIGH_UPDATE_F_OVERRIDE);
1191 return neigh;
1192}
1193EXPORT_SYMBOL(neigh_event_ns);
1194
1195/* called with read_lock_bh(&n->lock); */
1196static void neigh_hh_init(struct neighbour *n, struct dst_entry *dst)
1197{
1198 struct net_device *dev = dst->dev;
1199 __be16 prot = dst->ops->protocol;
1200 struct hh_cache *hh = &n->hh;
1201
1202 write_lock_bh(&n->lock);
1203
1204 /* Only one thread can come in here and initialize the
1205 * hh_cache entry.
1206 */
1207 if (!hh->hh_len)
1208 dev->header_ops->cache(n, hh, prot);
1209
1210 write_unlock_bh(&n->lock);
1211}
1212
1213/* This function can be used in contexts, where only old dev_queue_xmit
1214 * worked, f.e. if you want to override normal output path (eql, shaper),
1215 * but resolution is not made yet.
1216 */
1217
1218int neigh_compat_output(struct neighbour *neigh, struct sk_buff *skb)
1219{
1220 struct net_device *dev = skb->dev;
1221
1222 __skb_pull(skb, skb_network_offset(skb));
1223
1224 if (dev_hard_header(skb, dev, ntohs(skb->protocol), NULL, NULL,
1225 skb->len) < 0 &&
1226 dev->header_ops->rebuild(skb))
1227 return 0;
1228
1229 return dev_queue_xmit(skb);
1230}
1231EXPORT_SYMBOL(neigh_compat_output);
1232
1233/* Slow and careful. */
1234
1235int neigh_resolve_output(struct neighbour *neigh, struct sk_buff *skb)
1236{
1237 struct dst_entry *dst = skb_dst(skb);
1238 int rc = 0;
1239
1240 if (!dst)
1241 goto discard;
1242
1243 __skb_pull(skb, skb_network_offset(skb));
1244
1245 if (!neigh_event_send(neigh, skb)) {
1246 int err;
1247 struct net_device *dev = neigh->dev;
1248 unsigned int seq;
1249
1250 if (dev->header_ops->cache && !neigh->hh.hh_len)
1251 neigh_hh_init(neigh, dst);
1252
1253 do {
1254 seq = read_seqbegin(&neigh->ha_lock);
1255 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1256 neigh->ha, NULL, skb->len);
1257 } while (read_seqretry(&neigh->ha_lock, seq));
1258
1259 if (err >= 0)
1260 rc = dev_queue_xmit(skb);
1261 else
1262 goto out_kfree_skb;
1263 }
1264out:
1265 return rc;
1266discard:
1267 NEIGH_PRINTK1("neigh_resolve_output: dst=%p neigh=%p\n",
1268 dst, neigh);
1269out_kfree_skb:
1270 rc = -EINVAL;
1271 kfree_skb(skb);
1272 goto out;
1273}
1274EXPORT_SYMBOL(neigh_resolve_output);
1275
1276/* As fast as possible without hh cache */
1277
1278int neigh_connected_output(struct neighbour *neigh, struct sk_buff *skb)
1279{
1280 struct net_device *dev = neigh->dev;
1281 unsigned int seq;
1282 int err;
1283
1284 __skb_pull(skb, skb_network_offset(skb));
1285
1286 do {
1287 seq = read_seqbegin(&neigh->ha_lock);
1288 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1289 neigh->ha, NULL, skb->len);
1290 } while (read_seqretry(&neigh->ha_lock, seq));
1291
1292 if (err >= 0)
1293 err = dev_queue_xmit(skb);
1294 else {
1295 err = -EINVAL;
1296 kfree_skb(skb);
1297 }
1298 return err;
1299}
1300EXPORT_SYMBOL(neigh_connected_output);
1301
1302int neigh_direct_output(struct neighbour *neigh, struct sk_buff *skb)
1303{
1304 return dev_queue_xmit(skb);
1305}
1306EXPORT_SYMBOL(neigh_direct_output);
1307
1308static void neigh_proxy_process(unsigned long arg)
1309{
1310 struct neigh_table *tbl = (struct neigh_table *)arg;
1311 long sched_next = 0;
1312 unsigned long now = jiffies;
1313 struct sk_buff *skb, *n;
1314
1315 spin_lock(&tbl->proxy_queue.lock);
1316
1317 skb_queue_walk_safe(&tbl->proxy_queue, skb, n) {
1318 long tdif = NEIGH_CB(skb)->sched_next - now;
1319
1320 if (tdif <= 0) {
1321 struct net_device *dev = skb->dev;
1322
1323 __skb_unlink(skb, &tbl->proxy_queue);
1324 if (tbl->proxy_redo && netif_running(dev)) {
1325 rcu_read_lock();
1326 tbl->proxy_redo(skb);
1327 rcu_read_unlock();
1328 } else {
1329 kfree_skb(skb);
1330 }
1331
1332 dev_put(dev);
1333 } else if (!sched_next || tdif < sched_next)
1334 sched_next = tdif;
1335 }
1336 del_timer(&tbl->proxy_timer);
1337 if (sched_next)
1338 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1339 spin_unlock(&tbl->proxy_queue.lock);
1340}
1341
1342void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1343 struct sk_buff *skb)
1344{
1345 unsigned long now = jiffies;
1346 unsigned long sched_next = now + (net_random() % p->proxy_delay);
1347
1348 if (tbl->proxy_queue.qlen > p->proxy_qlen) {
1349 kfree_skb(skb);
1350 return;
1351 }
1352
1353 NEIGH_CB(skb)->sched_next = sched_next;
1354 NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1355
1356 spin_lock(&tbl->proxy_queue.lock);
1357 if (del_timer(&tbl->proxy_timer)) {
1358 if (time_before(tbl->proxy_timer.expires, sched_next))
1359 sched_next = tbl->proxy_timer.expires;
1360 }
1361 skb_dst_drop(skb);
1362 dev_hold(skb->dev);
1363 __skb_queue_tail(&tbl->proxy_queue, skb);
1364 mod_timer(&tbl->proxy_timer, sched_next);
1365 spin_unlock(&tbl->proxy_queue.lock);
1366}
1367EXPORT_SYMBOL(pneigh_enqueue);
1368
1369static inline struct neigh_parms *lookup_neigh_parms(struct neigh_table *tbl,
1370 struct net *net, int ifindex)
1371{
1372 struct neigh_parms *p;
1373
1374 for (p = &tbl->parms; p; p = p->next) {
1375 if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
1376 (!p->dev && !ifindex))
1377 return p;
1378 }
1379
1380 return NULL;
1381}
1382
1383struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1384 struct neigh_table *tbl)
1385{
1386 struct neigh_parms *p, *ref;
1387 struct net *net = dev_net(dev);
1388 const struct net_device_ops *ops = dev->netdev_ops;
1389
1390 ref = lookup_neigh_parms(tbl, net, 0);
1391 if (!ref)
1392 return NULL;
1393
1394 p = kmemdup(ref, sizeof(*p), GFP_KERNEL);
1395 if (p) {
1396 p->tbl = tbl;
1397 atomic_set(&p->refcnt, 1);
1398 p->reachable_time =
1399 neigh_rand_reach_time(p->base_reachable_time);
1400
1401 if (ops->ndo_neigh_setup && ops->ndo_neigh_setup(dev, p)) {
1402 kfree(p);
1403 return NULL;
1404 }
1405
1406 dev_hold(dev);
1407 p->dev = dev;
1408 write_pnet(&p->net, hold_net(net));
1409 p->sysctl_table = NULL;
1410 write_lock_bh(&tbl->lock);
1411 p->next = tbl->parms.next;
1412 tbl->parms.next = p;
1413 write_unlock_bh(&tbl->lock);
1414 }
1415 return p;
1416}
1417EXPORT_SYMBOL(neigh_parms_alloc);
1418
1419static void neigh_rcu_free_parms(struct rcu_head *head)
1420{
1421 struct neigh_parms *parms =
1422 container_of(head, struct neigh_parms, rcu_head);
1423
1424 neigh_parms_put(parms);
1425}
1426
1427void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1428{
1429 struct neigh_parms **p;
1430
1431 if (!parms || parms == &tbl->parms)
1432 return;
1433 write_lock_bh(&tbl->lock);
1434 for (p = &tbl->parms.next; *p; p = &(*p)->next) {
1435 if (*p == parms) {
1436 *p = parms->next;
1437 parms->dead = 1;
1438 write_unlock_bh(&tbl->lock);
1439 if (parms->dev)
1440 dev_put(parms->dev);
1441 call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1442 return;
1443 }
1444 }
1445 write_unlock_bh(&tbl->lock);
1446 NEIGH_PRINTK1("neigh_parms_release: not found\n");
1447}
1448EXPORT_SYMBOL(neigh_parms_release);
1449
1450static void neigh_parms_destroy(struct neigh_parms *parms)
1451{
1452 release_net(neigh_parms_net(parms));
1453 kfree(parms);
1454}
1455
1456static struct lock_class_key neigh_table_proxy_queue_class;
1457
1458void neigh_table_init_no_netlink(struct neigh_table *tbl)
1459{
1460 unsigned long now = jiffies;
1461 unsigned long phsize;
1462
1463 write_pnet(&tbl->parms.net, &init_net);
1464 atomic_set(&tbl->parms.refcnt, 1);
1465 tbl->parms.reachable_time =
1466 neigh_rand_reach_time(tbl->parms.base_reachable_time);
1467
1468 if (!tbl->kmem_cachep)
1469 tbl->kmem_cachep =
1470 kmem_cache_create(tbl->id, tbl->entry_size, 0,
1471 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1472 NULL);
1473 tbl->stats = alloc_percpu(struct neigh_statistics);
1474 if (!tbl->stats)
1475 panic("cannot create neighbour cache statistics");
1476
1477#ifdef CONFIG_PROC_FS
1478 if (!proc_create_data(tbl->id, 0, init_net.proc_net_stat,
1479 &neigh_stat_seq_fops, tbl))
1480 panic("cannot create neighbour proc dir entry");
1481#endif
1482
1483 RCU_INIT_POINTER(tbl->nht, neigh_hash_alloc(3));
1484
1485 phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1486 tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1487
1488 if (!tbl->nht || !tbl->phash_buckets)
1489 panic("cannot allocate neighbour cache hashes");
1490
1491 rwlock_init(&tbl->lock);
1492 INIT_DELAYED_WORK_DEFERRABLE(&tbl->gc_work, neigh_periodic_work);
1493 schedule_delayed_work(&tbl->gc_work, tbl->parms.reachable_time);
1494 setup_timer(&tbl->proxy_timer, neigh_proxy_process, (unsigned long)tbl);
1495 skb_queue_head_init_class(&tbl->proxy_queue,
1496 &neigh_table_proxy_queue_class);
1497
1498 tbl->last_flush = now;
1499 tbl->last_rand = now + tbl->parms.reachable_time * 20;
1500}
1501EXPORT_SYMBOL(neigh_table_init_no_netlink);
1502
1503void neigh_table_init(struct neigh_table *tbl)
1504{
1505 struct neigh_table *tmp;
1506
1507 neigh_table_init_no_netlink(tbl);
1508 write_lock(&neigh_tbl_lock);
1509 for (tmp = neigh_tables; tmp; tmp = tmp->next) {
1510 if (tmp->family == tbl->family)
1511 break;
1512 }
1513 tbl->next = neigh_tables;
1514 neigh_tables = tbl;
1515 write_unlock(&neigh_tbl_lock);
1516
1517 if (unlikely(tmp)) {
1518 printk(KERN_ERR "NEIGH: Registering multiple tables for "
1519 "family %d\n", tbl->family);
1520 dump_stack();
1521 }
1522}
1523EXPORT_SYMBOL(neigh_table_init);
1524
1525int neigh_table_clear(struct neigh_table *tbl)
1526{
1527 struct neigh_table **tp;
1528
1529 /* It is not clean... Fix it to unload IPv6 module safely */
1530 cancel_delayed_work_sync(&tbl->gc_work);
1531 del_timer_sync(&tbl->proxy_timer);
1532 pneigh_queue_purge(&tbl->proxy_queue);
1533 neigh_ifdown(tbl, NULL);
1534 if (atomic_read(&tbl->entries))
1535 printk(KERN_CRIT "neighbour leakage\n");
1536 write_lock(&neigh_tbl_lock);
1537 for (tp = &neigh_tables; *tp; tp = &(*tp)->next) {
1538 if (*tp == tbl) {
1539 *tp = tbl->next;
1540 break;
1541 }
1542 }
1543 write_unlock(&neigh_tbl_lock);
1544
1545 call_rcu(&rcu_dereference_protected(tbl->nht, 1)->rcu,
1546 neigh_hash_free_rcu);
1547 tbl->nht = NULL;
1548
1549 kfree(tbl->phash_buckets);
1550 tbl->phash_buckets = NULL;
1551
1552 remove_proc_entry(tbl->id, init_net.proc_net_stat);
1553
1554 free_percpu(tbl->stats);
1555 tbl->stats = NULL;
1556
1557 kmem_cache_destroy(tbl->kmem_cachep);
1558 tbl->kmem_cachep = NULL;
1559
1560 return 0;
1561}
1562EXPORT_SYMBOL(neigh_table_clear);
1563
1564static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1565{
1566 struct net *net = sock_net(skb->sk);
1567 struct ndmsg *ndm;
1568 struct nlattr *dst_attr;
1569 struct neigh_table *tbl;
1570 struct net_device *dev = NULL;
1571 int err = -EINVAL;
1572
1573 ASSERT_RTNL();
1574 if (nlmsg_len(nlh) < sizeof(*ndm))
1575 goto out;
1576
1577 dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1578 if (dst_attr == NULL)
1579 goto out;
1580
1581 ndm = nlmsg_data(nlh);
1582 if (ndm->ndm_ifindex) {
1583 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1584 if (dev == NULL) {
1585 err = -ENODEV;
1586 goto out;
1587 }
1588 }
1589
1590 read_lock(&neigh_tbl_lock);
1591 for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1592 struct neighbour *neigh;
1593
1594 if (tbl->family != ndm->ndm_family)
1595 continue;
1596 read_unlock(&neigh_tbl_lock);
1597
1598 if (nla_len(dst_attr) < tbl->key_len)
1599 goto out;
1600
1601 if (ndm->ndm_flags & NTF_PROXY) {
1602 err = pneigh_delete(tbl, net, nla_data(dst_attr), dev);
1603 goto out;
1604 }
1605
1606 if (dev == NULL)
1607 goto out;
1608
1609 neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1610 if (neigh == NULL) {
1611 err = -ENOENT;
1612 goto out;
1613 }
1614
1615 err = neigh_update(neigh, NULL, NUD_FAILED,
1616 NEIGH_UPDATE_F_OVERRIDE |
1617 NEIGH_UPDATE_F_ADMIN);
1618 neigh_release(neigh);
1619 goto out;
1620 }
1621 read_unlock(&neigh_tbl_lock);
1622 err = -EAFNOSUPPORT;
1623
1624out:
1625 return err;
1626}
1627
1628static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1629{
1630 struct net *net = sock_net(skb->sk);
1631 struct ndmsg *ndm;
1632 struct nlattr *tb[NDA_MAX+1];
1633 struct neigh_table *tbl;
1634 struct net_device *dev = NULL;
1635 int err;
1636
1637 ASSERT_RTNL();
1638 err = nlmsg_parse(nlh, sizeof(*ndm), tb, NDA_MAX, NULL);
1639 if (err < 0)
1640 goto out;
1641
1642 err = -EINVAL;
1643 if (tb[NDA_DST] == NULL)
1644 goto out;
1645
1646 ndm = nlmsg_data(nlh);
1647 if (ndm->ndm_ifindex) {
1648 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1649 if (dev == NULL) {
1650 err = -ENODEV;
1651 goto out;
1652 }
1653
1654 if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len)
1655 goto out;
1656 }
1657
1658 read_lock(&neigh_tbl_lock);
1659 for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1660 int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE;
1661 struct neighbour *neigh;
1662 void *dst, *lladdr;
1663
1664 if (tbl->family != ndm->ndm_family)
1665 continue;
1666 read_unlock(&neigh_tbl_lock);
1667
1668 if (nla_len(tb[NDA_DST]) < tbl->key_len)
1669 goto out;
1670 dst = nla_data(tb[NDA_DST]);
1671 lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
1672
1673 if (ndm->ndm_flags & NTF_PROXY) {
1674 struct pneigh_entry *pn;
1675
1676 err = -ENOBUFS;
1677 pn = pneigh_lookup(tbl, net, dst, dev, 1);
1678 if (pn) {
1679 pn->flags = ndm->ndm_flags;
1680 err = 0;
1681 }
1682 goto out;
1683 }
1684
1685 if (dev == NULL)
1686 goto out;
1687
1688 neigh = neigh_lookup(tbl, dst, dev);
1689 if (neigh == NULL) {
1690 if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
1691 err = -ENOENT;
1692 goto out;
1693 }
1694
1695 neigh = __neigh_lookup_errno(tbl, dst, dev);
1696 if (IS_ERR(neigh)) {
1697 err = PTR_ERR(neigh);
1698 goto out;
1699 }
1700 } else {
1701 if (nlh->nlmsg_flags & NLM_F_EXCL) {
1702 err = -EEXIST;
1703 neigh_release(neigh);
1704 goto out;
1705 }
1706
1707 if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
1708 flags &= ~NEIGH_UPDATE_F_OVERRIDE;
1709 }
1710
1711 if (ndm->ndm_flags & NTF_USE) {
1712 neigh_event_send(neigh, NULL);
1713 err = 0;
1714 } else
1715 err = neigh_update(neigh, lladdr, ndm->ndm_state, flags);
1716 neigh_release(neigh);
1717 goto out;
1718 }
1719
1720 read_unlock(&neigh_tbl_lock);
1721 err = -EAFNOSUPPORT;
1722out:
1723 return err;
1724}
1725
1726static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
1727{
1728 struct nlattr *nest;
1729
1730 nest = nla_nest_start(skb, NDTA_PARMS);
1731 if (nest == NULL)
1732 return -ENOBUFS;
1733
1734 if (parms->dev)
1735 NLA_PUT_U32(skb, NDTPA_IFINDEX, parms->dev->ifindex);
1736
1737 NLA_PUT_U32(skb, NDTPA_REFCNT, atomic_read(&parms->refcnt));
1738 NLA_PUT_U32(skb, NDTPA_QUEUE_LEN, parms->queue_len);
1739 NLA_PUT_U32(skb, NDTPA_PROXY_QLEN, parms->proxy_qlen);
1740 NLA_PUT_U32(skb, NDTPA_APP_PROBES, parms->app_probes);
1741 NLA_PUT_U32(skb, NDTPA_UCAST_PROBES, parms->ucast_probes);
1742 NLA_PUT_U32(skb, NDTPA_MCAST_PROBES, parms->mcast_probes);
1743 NLA_PUT_MSECS(skb, NDTPA_REACHABLE_TIME, parms->reachable_time);
1744 NLA_PUT_MSECS(skb, NDTPA_BASE_REACHABLE_TIME,
1745 parms->base_reachable_time);
1746 NLA_PUT_MSECS(skb, NDTPA_GC_STALETIME, parms->gc_staletime);
1747 NLA_PUT_MSECS(skb, NDTPA_DELAY_PROBE_TIME, parms->delay_probe_time);
1748 NLA_PUT_MSECS(skb, NDTPA_RETRANS_TIME, parms->retrans_time);
1749 NLA_PUT_MSECS(skb, NDTPA_ANYCAST_DELAY, parms->anycast_delay);
1750 NLA_PUT_MSECS(skb, NDTPA_PROXY_DELAY, parms->proxy_delay);
1751 NLA_PUT_MSECS(skb, NDTPA_LOCKTIME, parms->locktime);
1752
1753 return nla_nest_end(skb, nest);
1754
1755nla_put_failure:
1756 nla_nest_cancel(skb, nest);
1757 return -EMSGSIZE;
1758}
1759
1760static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl,
1761 u32 pid, u32 seq, int type, int flags)
1762{
1763 struct nlmsghdr *nlh;
1764 struct ndtmsg *ndtmsg;
1765
1766 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1767 if (nlh == NULL)
1768 return -EMSGSIZE;
1769
1770 ndtmsg = nlmsg_data(nlh);
1771
1772 read_lock_bh(&tbl->lock);
1773 ndtmsg->ndtm_family = tbl->family;
1774 ndtmsg->ndtm_pad1 = 0;
1775 ndtmsg->ndtm_pad2 = 0;
1776
1777 NLA_PUT_STRING(skb, NDTA_NAME, tbl->id);
1778 NLA_PUT_MSECS(skb, NDTA_GC_INTERVAL, tbl->gc_interval);
1779 NLA_PUT_U32(skb, NDTA_THRESH1, tbl->gc_thresh1);
1780 NLA_PUT_U32(skb, NDTA_THRESH2, tbl->gc_thresh2);
1781 NLA_PUT_U32(skb, NDTA_THRESH3, tbl->gc_thresh3);
1782
1783 {
1784 unsigned long now = jiffies;
1785 unsigned int flush_delta = now - tbl->last_flush;
1786 unsigned int rand_delta = now - tbl->last_rand;
1787 struct neigh_hash_table *nht;
1788 struct ndt_config ndc = {
1789 .ndtc_key_len = tbl->key_len,
1790 .ndtc_entry_size = tbl->entry_size,
1791 .ndtc_entries = atomic_read(&tbl->entries),
1792 .ndtc_last_flush = jiffies_to_msecs(flush_delta),
1793 .ndtc_last_rand = jiffies_to_msecs(rand_delta),
1794 .ndtc_proxy_qlen = tbl->proxy_queue.qlen,
1795 };
1796
1797 rcu_read_lock_bh();
1798 nht = rcu_dereference_bh(tbl->nht);
1799 ndc.ndtc_hash_rnd = nht->hash_rnd;
1800 ndc.ndtc_hash_mask = ((1 << nht->hash_shift) - 1);
1801 rcu_read_unlock_bh();
1802
1803 NLA_PUT(skb, NDTA_CONFIG, sizeof(ndc), &ndc);
1804 }
1805
1806 {
1807 int cpu;
1808 struct ndt_stats ndst;
1809
1810 memset(&ndst, 0, sizeof(ndst));
1811
1812 for_each_possible_cpu(cpu) {
1813 struct neigh_statistics *st;
1814
1815 st = per_cpu_ptr(tbl->stats, cpu);
1816 ndst.ndts_allocs += st->allocs;
1817 ndst.ndts_destroys += st->destroys;
1818 ndst.ndts_hash_grows += st->hash_grows;
1819 ndst.ndts_res_failed += st->res_failed;
1820 ndst.ndts_lookups += st->lookups;
1821 ndst.ndts_hits += st->hits;
1822 ndst.ndts_rcv_probes_mcast += st->rcv_probes_mcast;
1823 ndst.ndts_rcv_probes_ucast += st->rcv_probes_ucast;
1824 ndst.ndts_periodic_gc_runs += st->periodic_gc_runs;
1825 ndst.ndts_forced_gc_runs += st->forced_gc_runs;
1826 }
1827
1828 NLA_PUT(skb, NDTA_STATS, sizeof(ndst), &ndst);
1829 }
1830
1831 BUG_ON(tbl->parms.dev);
1832 if (neightbl_fill_parms(skb, &tbl->parms) < 0)
1833 goto nla_put_failure;
1834
1835 read_unlock_bh(&tbl->lock);
1836 return nlmsg_end(skb, nlh);
1837
1838nla_put_failure:
1839 read_unlock_bh(&tbl->lock);
1840 nlmsg_cancel(skb, nlh);
1841 return -EMSGSIZE;
1842}
1843
1844static int neightbl_fill_param_info(struct sk_buff *skb,
1845 struct neigh_table *tbl,
1846 struct neigh_parms *parms,
1847 u32 pid, u32 seq, int type,
1848 unsigned int flags)
1849{
1850 struct ndtmsg *ndtmsg;
1851 struct nlmsghdr *nlh;
1852
1853 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1854 if (nlh == NULL)
1855 return -EMSGSIZE;
1856
1857 ndtmsg = nlmsg_data(nlh);
1858
1859 read_lock_bh(&tbl->lock);
1860 ndtmsg->ndtm_family = tbl->family;
1861 ndtmsg->ndtm_pad1 = 0;
1862 ndtmsg->ndtm_pad2 = 0;
1863
1864 if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 ||
1865 neightbl_fill_parms(skb, parms) < 0)
1866 goto errout;
1867
1868 read_unlock_bh(&tbl->lock);
1869 return nlmsg_end(skb, nlh);
1870errout:
1871 read_unlock_bh(&tbl->lock);
1872 nlmsg_cancel(skb, nlh);
1873 return -EMSGSIZE;
1874}
1875
1876static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = {
1877 [NDTA_NAME] = { .type = NLA_STRING },
1878 [NDTA_THRESH1] = { .type = NLA_U32 },
1879 [NDTA_THRESH2] = { .type = NLA_U32 },
1880 [NDTA_THRESH3] = { .type = NLA_U32 },
1881 [NDTA_GC_INTERVAL] = { .type = NLA_U64 },
1882 [NDTA_PARMS] = { .type = NLA_NESTED },
1883};
1884
1885static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = {
1886 [NDTPA_IFINDEX] = { .type = NLA_U32 },
1887 [NDTPA_QUEUE_LEN] = { .type = NLA_U32 },
1888 [NDTPA_PROXY_QLEN] = { .type = NLA_U32 },
1889 [NDTPA_APP_PROBES] = { .type = NLA_U32 },
1890 [NDTPA_UCAST_PROBES] = { .type = NLA_U32 },
1891 [NDTPA_MCAST_PROBES] = { .type = NLA_U32 },
1892 [NDTPA_BASE_REACHABLE_TIME] = { .type = NLA_U64 },
1893 [NDTPA_GC_STALETIME] = { .type = NLA_U64 },
1894 [NDTPA_DELAY_PROBE_TIME] = { .type = NLA_U64 },
1895 [NDTPA_RETRANS_TIME] = { .type = NLA_U64 },
1896 [NDTPA_ANYCAST_DELAY] = { .type = NLA_U64 },
1897 [NDTPA_PROXY_DELAY] = { .type = NLA_U64 },
1898 [NDTPA_LOCKTIME] = { .type = NLA_U64 },
1899};
1900
1901static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1902{
1903 struct net *net = sock_net(skb->sk);
1904 struct neigh_table *tbl;
1905 struct ndtmsg *ndtmsg;
1906 struct nlattr *tb[NDTA_MAX+1];
1907 int err;
1908
1909 err = nlmsg_parse(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
1910 nl_neightbl_policy);
1911 if (err < 0)
1912 goto errout;
1913
1914 if (tb[NDTA_NAME] == NULL) {
1915 err = -EINVAL;
1916 goto errout;
1917 }
1918
1919 ndtmsg = nlmsg_data(nlh);
1920 read_lock(&neigh_tbl_lock);
1921 for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1922 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
1923 continue;
1924
1925 if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0)
1926 break;
1927 }
1928
1929 if (tbl == NULL) {
1930 err = -ENOENT;
1931 goto errout_locked;
1932 }
1933
1934 /*
1935 * We acquire tbl->lock to be nice to the periodic timers and
1936 * make sure they always see a consistent set of values.
1937 */
1938 write_lock_bh(&tbl->lock);
1939
1940 if (tb[NDTA_PARMS]) {
1941 struct nlattr *tbp[NDTPA_MAX+1];
1942 struct neigh_parms *p;
1943 int i, ifindex = 0;
1944
1945 err = nla_parse_nested(tbp, NDTPA_MAX, tb[NDTA_PARMS],
1946 nl_ntbl_parm_policy);
1947 if (err < 0)
1948 goto errout_tbl_lock;
1949
1950 if (tbp[NDTPA_IFINDEX])
1951 ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
1952
1953 p = lookup_neigh_parms(tbl, net, ifindex);
1954 if (p == NULL) {
1955 err = -ENOENT;
1956 goto errout_tbl_lock;
1957 }
1958
1959 for (i = 1; i <= NDTPA_MAX; i++) {
1960 if (tbp[i] == NULL)
1961 continue;
1962
1963 switch (i) {
1964 case NDTPA_QUEUE_LEN:
1965 p->queue_len = nla_get_u32(tbp[i]);
1966 break;
1967 case NDTPA_PROXY_QLEN:
1968 p->proxy_qlen = nla_get_u32(tbp[i]);
1969 break;
1970 case NDTPA_APP_PROBES:
1971 p->app_probes = nla_get_u32(tbp[i]);
1972 break;
1973 case NDTPA_UCAST_PROBES:
1974 p->ucast_probes = nla_get_u32(tbp[i]);
1975 break;
1976 case NDTPA_MCAST_PROBES:
1977 p->mcast_probes = nla_get_u32(tbp[i]);
1978 break;
1979 case NDTPA_BASE_REACHABLE_TIME:
1980 p->base_reachable_time = nla_get_msecs(tbp[i]);
1981 break;
1982 case NDTPA_GC_STALETIME:
1983 p->gc_staletime = nla_get_msecs(tbp[i]);
1984 break;
1985 case NDTPA_DELAY_PROBE_TIME:
1986 p->delay_probe_time = nla_get_msecs(tbp[i]);
1987 break;
1988 case NDTPA_RETRANS_TIME:
1989 p->retrans_time = nla_get_msecs(tbp[i]);
1990 break;
1991 case NDTPA_ANYCAST_DELAY:
1992 p->anycast_delay = nla_get_msecs(tbp[i]);
1993 break;
1994 case NDTPA_PROXY_DELAY:
1995 p->proxy_delay = nla_get_msecs(tbp[i]);
1996 break;
1997 case NDTPA_LOCKTIME:
1998 p->locktime = nla_get_msecs(tbp[i]);
1999 break;
2000 }
2001 }
2002 }
2003
2004 if (tb[NDTA_THRESH1])
2005 tbl->gc_thresh1 = nla_get_u32(tb[NDTA_THRESH1]);
2006
2007 if (tb[NDTA_THRESH2])
2008 tbl->gc_thresh2 = nla_get_u32(tb[NDTA_THRESH2]);
2009
2010 if (tb[NDTA_THRESH3])
2011 tbl->gc_thresh3 = nla_get_u32(tb[NDTA_THRESH3]);
2012
2013 if (tb[NDTA_GC_INTERVAL])
2014 tbl->gc_interval = nla_get_msecs(tb[NDTA_GC_INTERVAL]);
2015
2016 err = 0;
2017
2018errout_tbl_lock:
2019 write_unlock_bh(&tbl->lock);
2020errout_locked:
2021 read_unlock(&neigh_tbl_lock);
2022errout:
2023 return err;
2024}
2025
2026static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2027{
2028 struct net *net = sock_net(skb->sk);
2029 int family, tidx, nidx = 0;
2030 int tbl_skip = cb->args[0];
2031 int neigh_skip = cb->args[1];
2032 struct neigh_table *tbl;
2033
2034 family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2035
2036 read_lock(&neigh_tbl_lock);
2037 for (tbl = neigh_tables, tidx = 0; tbl; tbl = tbl->next, tidx++) {
2038 struct neigh_parms *p;
2039
2040 if (tidx < tbl_skip || (family && tbl->family != family))
2041 continue;
2042
2043 if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).pid,
2044 cb->nlh->nlmsg_seq, RTM_NEWNEIGHTBL,
2045 NLM_F_MULTI) <= 0)
2046 break;
2047
2048 for (nidx = 0, p = tbl->parms.next; p; p = p->next) {
2049 if (!net_eq(neigh_parms_net(p), net))
2050 continue;
2051
2052 if (nidx < neigh_skip)
2053 goto next;
2054
2055 if (neightbl_fill_param_info(skb, tbl, p,
2056 NETLINK_CB(cb->skb).pid,
2057 cb->nlh->nlmsg_seq,
2058 RTM_NEWNEIGHTBL,
2059 NLM_F_MULTI) <= 0)
2060 goto out;
2061 next:
2062 nidx++;
2063 }
2064
2065 neigh_skip = 0;
2066 }
2067out:
2068 read_unlock(&neigh_tbl_lock);
2069 cb->args[0] = tidx;
2070 cb->args[1] = nidx;
2071
2072 return skb->len;
2073}
2074
2075static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
2076 u32 pid, u32 seq, int type, unsigned int flags)
2077{
2078 unsigned long now = jiffies;
2079 struct nda_cacheinfo ci;
2080 struct nlmsghdr *nlh;
2081 struct ndmsg *ndm;
2082
2083 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2084 if (nlh == NULL)
2085 return -EMSGSIZE;
2086
2087 ndm = nlmsg_data(nlh);
2088 ndm->ndm_family = neigh->ops->family;
2089 ndm->ndm_pad1 = 0;
2090 ndm->ndm_pad2 = 0;
2091 ndm->ndm_flags = neigh->flags;
2092 ndm->ndm_type = neigh->type;
2093 ndm->ndm_ifindex = neigh->dev->ifindex;
2094
2095 NLA_PUT(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key);
2096
2097 read_lock_bh(&neigh->lock);
2098 ndm->ndm_state = neigh->nud_state;
2099 if (neigh->nud_state & NUD_VALID) {
2100 char haddr[MAX_ADDR_LEN];
2101
2102 neigh_ha_snapshot(haddr, neigh, neigh->dev);
2103 if (nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, haddr) < 0) {
2104 read_unlock_bh(&neigh->lock);
2105 goto nla_put_failure;
2106 }
2107 }
2108
2109 ci.ndm_used = jiffies_to_clock_t(now - neigh->used);
2110 ci.ndm_confirmed = jiffies_to_clock_t(now - neigh->confirmed);
2111 ci.ndm_updated = jiffies_to_clock_t(now - neigh->updated);
2112 ci.ndm_refcnt = atomic_read(&neigh->refcnt) - 1;
2113 read_unlock_bh(&neigh->lock);
2114
2115 NLA_PUT_U32(skb, NDA_PROBES, atomic_read(&neigh->probes));
2116 NLA_PUT(skb, NDA_CACHEINFO, sizeof(ci), &ci);
2117
2118 return nlmsg_end(skb, nlh);
2119
2120nla_put_failure:
2121 nlmsg_cancel(skb, nlh);
2122 return -EMSGSIZE;
2123}
2124
2125static void neigh_update_notify(struct neighbour *neigh)
2126{
2127 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
2128 __neigh_notify(neigh, RTM_NEWNEIGH, 0);
2129}
2130
2131static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2132 struct netlink_callback *cb)
2133{
2134 struct net *net = sock_net(skb->sk);
2135 struct neighbour *n;
2136 int rc, h, s_h = cb->args[1];
2137 int idx, s_idx = idx = cb->args[2];
2138 struct neigh_hash_table *nht;
2139
2140 rcu_read_lock_bh();
2141 nht = rcu_dereference_bh(tbl->nht);
2142
2143 for (h = 0; h < (1 << nht->hash_shift); h++) {
2144 if (h < s_h)
2145 continue;
2146 if (h > s_h)
2147 s_idx = 0;
2148 for (n = rcu_dereference_bh(nht->hash_buckets[h]), idx = 0;
2149 n != NULL;
2150 n = rcu_dereference_bh(n->next)) {
2151 if (!net_eq(dev_net(n->dev), net))
2152 continue;
2153 if (idx < s_idx)
2154 goto next;
2155 if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).pid,
2156 cb->nlh->nlmsg_seq,
2157 RTM_NEWNEIGH,
2158 NLM_F_MULTI) <= 0) {
2159 rc = -1;
2160 goto out;
2161 }
2162next:
2163 idx++;
2164 }
2165 }
2166 rc = skb->len;
2167out:
2168 rcu_read_unlock_bh();
2169 cb->args[1] = h;
2170 cb->args[2] = idx;
2171 return rc;
2172}
2173
2174static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2175{
2176 struct neigh_table *tbl;
2177 int t, family, s_t;
2178
2179 read_lock(&neigh_tbl_lock);
2180 family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2181 s_t = cb->args[0];
2182
2183 for (tbl = neigh_tables, t = 0; tbl; tbl = tbl->next, t++) {
2184 if (t < s_t || (family && tbl->family != family))
2185 continue;
2186 if (t > s_t)
2187 memset(&cb->args[1], 0, sizeof(cb->args) -
2188 sizeof(cb->args[0]));
2189 if (neigh_dump_table(tbl, skb, cb) < 0)
2190 break;
2191 }
2192 read_unlock(&neigh_tbl_lock);
2193
2194 cb->args[0] = t;
2195 return skb->len;
2196}
2197
2198void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
2199{
2200 int chain;
2201 struct neigh_hash_table *nht;
2202
2203 rcu_read_lock_bh();
2204 nht = rcu_dereference_bh(tbl->nht);
2205
2206 read_lock(&tbl->lock); /* avoid resizes */
2207 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2208 struct neighbour *n;
2209
2210 for (n = rcu_dereference_bh(nht->hash_buckets[chain]);
2211 n != NULL;
2212 n = rcu_dereference_bh(n->next))
2213 cb(n, cookie);
2214 }
2215 read_unlock(&tbl->lock);
2216 rcu_read_unlock_bh();
2217}
2218EXPORT_SYMBOL(neigh_for_each);
2219
2220/* The tbl->lock must be held as a writer and BH disabled. */
2221void __neigh_for_each_release(struct neigh_table *tbl,
2222 int (*cb)(struct neighbour *))
2223{
2224 int chain;
2225 struct neigh_hash_table *nht;
2226
2227 nht = rcu_dereference_protected(tbl->nht,
2228 lockdep_is_held(&tbl->lock));
2229 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2230 struct neighbour *n;
2231 struct neighbour __rcu **np;
2232
2233 np = &nht->hash_buckets[chain];
2234 while ((n = rcu_dereference_protected(*np,
2235 lockdep_is_held(&tbl->lock))) != NULL) {
2236 int release;
2237
2238 write_lock(&n->lock);
2239 release = cb(n);
2240 if (release) {
2241 rcu_assign_pointer(*np,
2242 rcu_dereference_protected(n->next,
2243 lockdep_is_held(&tbl->lock)));
2244 n->dead = 1;
2245 } else
2246 np = &n->next;
2247 write_unlock(&n->lock);
2248 if (release)
2249 neigh_cleanup_and_release(n);
2250 }
2251 }
2252}
2253EXPORT_SYMBOL(__neigh_for_each_release);
2254
2255#ifdef CONFIG_PROC_FS
2256
2257static struct neighbour *neigh_get_first(struct seq_file *seq)
2258{
2259 struct neigh_seq_state *state = seq->private;
2260 struct net *net = seq_file_net(seq);
2261 struct neigh_hash_table *nht = state->nht;
2262 struct neighbour *n = NULL;
2263 int bucket = state->bucket;
2264
2265 state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
2266 for (bucket = 0; bucket < (1 << nht->hash_shift); bucket++) {
2267 n = rcu_dereference_bh(nht->hash_buckets[bucket]);
2268
2269 while (n) {
2270 if (!net_eq(dev_net(n->dev), net))
2271 goto next;
2272 if (state->neigh_sub_iter) {
2273 loff_t fakep = 0;
2274 void *v;
2275
2276 v = state->neigh_sub_iter(state, n, &fakep);
2277 if (!v)
2278 goto next;
2279 }
2280 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2281 break;
2282 if (n->nud_state & ~NUD_NOARP)
2283 break;
2284next:
2285 n = rcu_dereference_bh(n->next);
2286 }
2287
2288 if (n)
2289 break;
2290 }
2291 state->bucket = bucket;
2292
2293 return n;
2294}
2295
2296static struct neighbour *neigh_get_next(struct seq_file *seq,
2297 struct neighbour *n,
2298 loff_t *pos)
2299{
2300 struct neigh_seq_state *state = seq->private;
2301 struct net *net = seq_file_net(seq);
2302 struct neigh_hash_table *nht = state->nht;
2303
2304 if (state->neigh_sub_iter) {
2305 void *v = state->neigh_sub_iter(state, n, pos);
2306 if (v)
2307 return n;
2308 }
2309 n = rcu_dereference_bh(n->next);
2310
2311 while (1) {
2312 while (n) {
2313 if (!net_eq(dev_net(n->dev), net))
2314 goto next;
2315 if (state->neigh_sub_iter) {
2316 void *v = state->neigh_sub_iter(state, n, pos);
2317 if (v)
2318 return n;
2319 goto next;
2320 }
2321 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2322 break;
2323
2324 if (n->nud_state & ~NUD_NOARP)
2325 break;
2326next:
2327 n = rcu_dereference_bh(n->next);
2328 }
2329
2330 if (n)
2331 break;
2332
2333 if (++state->bucket >= (1 << nht->hash_shift))
2334 break;
2335
2336 n = rcu_dereference_bh(nht->hash_buckets[state->bucket]);
2337 }
2338
2339 if (n && pos)
2340 --(*pos);
2341 return n;
2342}
2343
2344static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
2345{
2346 struct neighbour *n = neigh_get_first(seq);
2347
2348 if (n) {
2349 --(*pos);
2350 while (*pos) {
2351 n = neigh_get_next(seq, n, pos);
2352 if (!n)
2353 break;
2354 }
2355 }
2356 return *pos ? NULL : n;
2357}
2358
2359static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
2360{
2361 struct neigh_seq_state *state = seq->private;
2362 struct net *net = seq_file_net(seq);
2363 struct neigh_table *tbl = state->tbl;
2364 struct pneigh_entry *pn = NULL;
2365 int bucket = state->bucket;
2366
2367 state->flags |= NEIGH_SEQ_IS_PNEIGH;
2368 for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
2369 pn = tbl->phash_buckets[bucket];
2370 while (pn && !net_eq(pneigh_net(pn), net))
2371 pn = pn->next;
2372 if (pn)
2373 break;
2374 }
2375 state->bucket = bucket;
2376
2377 return pn;
2378}
2379
2380static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
2381 struct pneigh_entry *pn,
2382 loff_t *pos)
2383{
2384 struct neigh_seq_state *state = seq->private;
2385 struct net *net = seq_file_net(seq);
2386 struct neigh_table *tbl = state->tbl;
2387
2388 pn = pn->next;
2389 while (!pn) {
2390 if (++state->bucket > PNEIGH_HASHMASK)
2391 break;
2392 pn = tbl->phash_buckets[state->bucket];
2393 while (pn && !net_eq(pneigh_net(pn), net))
2394 pn = pn->next;
2395 if (pn)
2396 break;
2397 }
2398
2399 if (pn && pos)
2400 --(*pos);
2401
2402 return pn;
2403}
2404
2405static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
2406{
2407 struct pneigh_entry *pn = pneigh_get_first(seq);
2408
2409 if (pn) {
2410 --(*pos);
2411 while (*pos) {
2412 pn = pneigh_get_next(seq, pn, pos);
2413 if (!pn)
2414 break;
2415 }
2416 }
2417 return *pos ? NULL : pn;
2418}
2419
2420static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
2421{
2422 struct neigh_seq_state *state = seq->private;
2423 void *rc;
2424 loff_t idxpos = *pos;
2425
2426 rc = neigh_get_idx(seq, &idxpos);
2427 if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2428 rc = pneigh_get_idx(seq, &idxpos);
2429
2430 return rc;
2431}
2432
2433void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
2434 __acquires(rcu_bh)
2435{
2436 struct neigh_seq_state *state = seq->private;
2437
2438 state->tbl = tbl;
2439 state->bucket = 0;
2440 state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
2441
2442 rcu_read_lock_bh();
2443 state->nht = rcu_dereference_bh(tbl->nht);
2444
2445 return *pos ? neigh_get_idx_any(seq, pos) : SEQ_START_TOKEN;
2446}
2447EXPORT_SYMBOL(neigh_seq_start);
2448
2449void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2450{
2451 struct neigh_seq_state *state;
2452 void *rc;
2453
2454 if (v == SEQ_START_TOKEN) {
2455 rc = neigh_get_first(seq);
2456 goto out;
2457 }
2458
2459 state = seq->private;
2460 if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
2461 rc = neigh_get_next(seq, v, NULL);
2462 if (rc)
2463 goto out;
2464 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2465 rc = pneigh_get_first(seq);
2466 } else {
2467 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
2468 rc = pneigh_get_next(seq, v, NULL);
2469 }
2470out:
2471 ++(*pos);
2472 return rc;
2473}
2474EXPORT_SYMBOL(neigh_seq_next);
2475
2476void neigh_seq_stop(struct seq_file *seq, void *v)
2477 __releases(rcu_bh)
2478{
2479 rcu_read_unlock_bh();
2480}
2481EXPORT_SYMBOL(neigh_seq_stop);
2482
2483/* statistics via seq_file */
2484
2485static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
2486{
2487 struct neigh_table *tbl = seq->private;
2488 int cpu;
2489
2490 if (*pos == 0)
2491 return SEQ_START_TOKEN;
2492
2493 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
2494 if (!cpu_possible(cpu))
2495 continue;
2496 *pos = cpu+1;
2497 return per_cpu_ptr(tbl->stats, cpu);
2498 }
2499 return NULL;
2500}
2501
2502static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2503{
2504 struct neigh_table *tbl = seq->private;
2505 int cpu;
2506
2507 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
2508 if (!cpu_possible(cpu))
2509 continue;
2510 *pos = cpu+1;
2511 return per_cpu_ptr(tbl->stats, cpu);
2512 }
2513 return NULL;
2514}
2515
2516static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
2517{
2518
2519}
2520
2521static int neigh_stat_seq_show(struct seq_file *seq, void *v)
2522{
2523 struct neigh_table *tbl = seq->private;
2524 struct neigh_statistics *st = v;
2525
2526 if (v == SEQ_START_TOKEN) {
2527 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\n");
2528 return 0;
2529 }
2530
2531 seq_printf(seq, "%08x %08lx %08lx %08lx %08lx %08lx %08lx "
2532 "%08lx %08lx %08lx %08lx %08lx\n",
2533 atomic_read(&tbl->entries),
2534
2535 st->allocs,
2536 st->destroys,
2537 st->hash_grows,
2538
2539 st->lookups,
2540 st->hits,
2541
2542 st->res_failed,
2543
2544 st->rcv_probes_mcast,
2545 st->rcv_probes_ucast,
2546
2547 st->periodic_gc_runs,
2548 st->forced_gc_runs,
2549 st->unres_discards
2550 );
2551
2552 return 0;
2553}
2554
2555static const struct seq_operations neigh_stat_seq_ops = {
2556 .start = neigh_stat_seq_start,
2557 .next = neigh_stat_seq_next,
2558 .stop = neigh_stat_seq_stop,
2559 .show = neigh_stat_seq_show,
2560};
2561
2562static int neigh_stat_seq_open(struct inode *inode, struct file *file)
2563{
2564 int ret = seq_open(file, &neigh_stat_seq_ops);
2565
2566 if (!ret) {
2567 struct seq_file *sf = file->private_data;
2568 sf->private = PDE(inode)->data;
2569 }
2570 return ret;
2571};
2572
2573static const struct file_operations neigh_stat_seq_fops = {
2574 .owner = THIS_MODULE,
2575 .open = neigh_stat_seq_open,
2576 .read = seq_read,
2577 .llseek = seq_lseek,
2578 .release = seq_release,
2579};
2580
2581#endif /* CONFIG_PROC_FS */
2582
2583static inline size_t neigh_nlmsg_size(void)
2584{
2585 return NLMSG_ALIGN(sizeof(struct ndmsg))
2586 + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2587 + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */
2588 + nla_total_size(sizeof(struct nda_cacheinfo))
2589 + nla_total_size(4); /* NDA_PROBES */
2590}
2591
2592static void __neigh_notify(struct neighbour *n, int type, int flags)
2593{
2594 struct net *net = dev_net(n->dev);
2595 struct sk_buff *skb;
2596 int err = -ENOBUFS;
2597
2598 skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC);
2599 if (skb == NULL)
2600 goto errout;
2601
2602 err = neigh_fill_info(skb, n, 0, 0, type, flags);
2603 if (err < 0) {
2604 /* -EMSGSIZE implies BUG in neigh_nlmsg_size() */
2605 WARN_ON(err == -EMSGSIZE);
2606 kfree_skb(skb);
2607 goto errout;
2608 }
2609 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
2610 return;
2611errout:
2612 if (err < 0)
2613 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
2614}
2615
2616#ifdef CONFIG_ARPD
2617void neigh_app_ns(struct neighbour *n)
2618{
2619 __neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST);
2620}
2621EXPORT_SYMBOL(neigh_app_ns);
2622#endif /* CONFIG_ARPD */
2623
2624#ifdef CONFIG_SYSCTL
2625
2626#define NEIGH_VARS_MAX 19
2627
2628static struct neigh_sysctl_table {
2629 struct ctl_table_header *sysctl_header;
2630 struct ctl_table neigh_vars[NEIGH_VARS_MAX];
2631 char *dev_name;
2632} neigh_sysctl_template __read_mostly = {
2633 .neigh_vars = {
2634 {
2635 .procname = "mcast_solicit",
2636 .maxlen = sizeof(int),
2637 .mode = 0644,
2638 .proc_handler = proc_dointvec,
2639 },
2640 {
2641 .procname = "ucast_solicit",
2642 .maxlen = sizeof(int),
2643 .mode = 0644,
2644 .proc_handler = proc_dointvec,
2645 },
2646 {
2647 .procname = "app_solicit",
2648 .maxlen = sizeof(int),
2649 .mode = 0644,
2650 .proc_handler = proc_dointvec,
2651 },
2652 {
2653 .procname = "retrans_time",
2654 .maxlen = sizeof(int),
2655 .mode = 0644,
2656 .proc_handler = proc_dointvec_userhz_jiffies,
2657 },
2658 {
2659 .procname = "base_reachable_time",
2660 .maxlen = sizeof(int),
2661 .mode = 0644,
2662 .proc_handler = proc_dointvec_jiffies,
2663 },
2664 {
2665 .procname = "delay_first_probe_time",
2666 .maxlen = sizeof(int),
2667 .mode = 0644,
2668 .proc_handler = proc_dointvec_jiffies,
2669 },
2670 {
2671 .procname = "gc_stale_time",
2672 .maxlen = sizeof(int),
2673 .mode = 0644,
2674 .proc_handler = proc_dointvec_jiffies,
2675 },
2676 {
2677 .procname = "unres_qlen",
2678 .maxlen = sizeof(int),
2679 .mode = 0644,
2680 .proc_handler = proc_dointvec,
2681 },
2682 {
2683 .procname = "proxy_qlen",
2684 .maxlen = sizeof(int),
2685 .mode = 0644,
2686 .proc_handler = proc_dointvec,
2687 },
2688 {
2689 .procname = "anycast_delay",
2690 .maxlen = sizeof(int),
2691 .mode = 0644,
2692 .proc_handler = proc_dointvec_userhz_jiffies,
2693 },
2694 {
2695 .procname = "proxy_delay",
2696 .maxlen = sizeof(int),
2697 .mode = 0644,
2698 .proc_handler = proc_dointvec_userhz_jiffies,
2699 },
2700 {
2701 .procname = "locktime",
2702 .maxlen = sizeof(int),
2703 .mode = 0644,
2704 .proc_handler = proc_dointvec_userhz_jiffies,
2705 },
2706 {
2707 .procname = "retrans_time_ms",
2708 .maxlen = sizeof(int),
2709 .mode = 0644,
2710 .proc_handler = proc_dointvec_ms_jiffies,
2711 },
2712 {
2713 .procname = "base_reachable_time_ms",
2714 .maxlen = sizeof(int),
2715 .mode = 0644,
2716 .proc_handler = proc_dointvec_ms_jiffies,
2717 },
2718 {
2719 .procname = "gc_interval",
2720 .maxlen = sizeof(int),
2721 .mode = 0644,
2722 .proc_handler = proc_dointvec_jiffies,
2723 },
2724 {
2725 .procname = "gc_thresh1",
2726 .maxlen = sizeof(int),
2727 .mode = 0644,
2728 .proc_handler = proc_dointvec,
2729 },
2730 {
2731 .procname = "gc_thresh2",
2732 .maxlen = sizeof(int),
2733 .mode = 0644,
2734 .proc_handler = proc_dointvec,
2735 },
2736 {
2737 .procname = "gc_thresh3",
2738 .maxlen = sizeof(int),
2739 .mode = 0644,
2740 .proc_handler = proc_dointvec,
2741 },
2742 {},
2743 },
2744};
2745
2746int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
2747 char *p_name, proc_handler *handler)
2748{
2749 struct neigh_sysctl_table *t;
2750 const char *dev_name_source = NULL;
2751
2752#define NEIGH_CTL_PATH_ROOT 0
2753#define NEIGH_CTL_PATH_PROTO 1
2754#define NEIGH_CTL_PATH_NEIGH 2
2755#define NEIGH_CTL_PATH_DEV 3
2756
2757 struct ctl_path neigh_path[] = {
2758 { .procname = "net", },
2759 { .procname = "proto", },
2760 { .procname = "neigh", },
2761 { .procname = "default", },
2762 { },
2763 };
2764
2765 t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL);
2766 if (!t)
2767 goto err;
2768
2769 t->neigh_vars[0].data = &p->mcast_probes;
2770 t->neigh_vars[1].data = &p->ucast_probes;
2771 t->neigh_vars[2].data = &p->app_probes;
2772 t->neigh_vars[3].data = &p->retrans_time;
2773 t->neigh_vars[4].data = &p->base_reachable_time;
2774 t->neigh_vars[5].data = &p->delay_probe_time;
2775 t->neigh_vars[6].data = &p->gc_staletime;
2776 t->neigh_vars[7].data = &p->queue_len;
2777 t->neigh_vars[8].data = &p->proxy_qlen;
2778 t->neigh_vars[9].data = &p->anycast_delay;
2779 t->neigh_vars[10].data = &p->proxy_delay;
2780 t->neigh_vars[11].data = &p->locktime;
2781 t->neigh_vars[12].data = &p->retrans_time;
2782 t->neigh_vars[13].data = &p->base_reachable_time;
2783
2784 if (dev) {
2785 dev_name_source = dev->name;
2786 /* Terminate the table early */
2787 memset(&t->neigh_vars[14], 0, sizeof(t->neigh_vars[14]));
2788 } else {
2789 dev_name_source = neigh_path[NEIGH_CTL_PATH_DEV].procname;
2790 t->neigh_vars[14].data = (int *)(p + 1);
2791 t->neigh_vars[15].data = (int *)(p + 1) + 1;
2792 t->neigh_vars[16].data = (int *)(p + 1) + 2;
2793 t->neigh_vars[17].data = (int *)(p + 1) + 3;
2794 }
2795
2796
2797 if (handler) {
2798 /* RetransTime */
2799 t->neigh_vars[3].proc_handler = handler;
2800 t->neigh_vars[3].extra1 = dev;
2801 /* ReachableTime */
2802 t->neigh_vars[4].proc_handler = handler;
2803 t->neigh_vars[4].extra1 = dev;
2804 /* RetransTime (in milliseconds)*/
2805 t->neigh_vars[12].proc_handler = handler;
2806 t->neigh_vars[12].extra1 = dev;
2807 /* ReachableTime (in milliseconds) */
2808 t->neigh_vars[13].proc_handler = handler;
2809 t->neigh_vars[13].extra1 = dev;
2810 }
2811
2812 t->dev_name = kstrdup(dev_name_source, GFP_KERNEL);
2813 if (!t->dev_name)
2814 goto free;
2815
2816 neigh_path[NEIGH_CTL_PATH_DEV].procname = t->dev_name;
2817 neigh_path[NEIGH_CTL_PATH_PROTO].procname = p_name;
2818
2819 t->sysctl_header =
2820 register_net_sysctl_table(neigh_parms_net(p), neigh_path, t->neigh_vars);
2821 if (!t->sysctl_header)
2822 goto free_procname;
2823
2824 p->sysctl_table = t;
2825 return 0;
2826
2827free_procname:
2828 kfree(t->dev_name);
2829free:
2830 kfree(t);
2831err:
2832 return -ENOBUFS;
2833}
2834EXPORT_SYMBOL(neigh_sysctl_register);
2835
2836void neigh_sysctl_unregister(struct neigh_parms *p)
2837{
2838 if (p->sysctl_table) {
2839 struct neigh_sysctl_table *t = p->sysctl_table;
2840 p->sysctl_table = NULL;
2841 unregister_sysctl_table(t->sysctl_header);
2842 kfree(t->dev_name);
2843 kfree(t);
2844 }
2845}
2846EXPORT_SYMBOL(neigh_sysctl_unregister);
2847
2848#endif /* CONFIG_SYSCTL */
2849
2850static int __init neigh_init(void)
2851{
2852 rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL, NULL);
2853 rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL, NULL);
2854 rtnl_register(PF_UNSPEC, RTM_GETNEIGH, NULL, neigh_dump_info, NULL);
2855
2856 rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info,
2857 NULL);
2858 rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL, NULL);
2859
2860 return 0;
2861}
2862
2863subsys_initcall(neigh_init);
2864
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/types.h>
18#include <linux/kernel.h>
19#include <linux/module.h>
20#include <linux/socket.h>
21#include <linux/netdevice.h>
22#include <linux/proc_fs.h>
23#ifdef CONFIG_SYSCTL
24#include <linux/sysctl.h>
25#endif
26#include <linux/times.h>
27#include <net/net_namespace.h>
28#include <net/neighbour.h>
29#include <net/arp.h>
30#include <net/dst.h>
31#include <net/sock.h>
32#include <net/netevent.h>
33#include <net/netlink.h>
34#include <linux/rtnetlink.h>
35#include <linux/random.h>
36#include <linux/string.h>
37#include <linux/log2.h>
38#include <linux/inetdevice.h>
39#include <net/addrconf.h>
40
41#include <trace/events/neigh.h>
42
43#define NEIGH_DEBUG 1
44#define neigh_dbg(level, fmt, ...) \
45do { \
46 if (level <= NEIGH_DEBUG) \
47 pr_debug(fmt, ##__VA_ARGS__); \
48} while (0)
49
50#define PNEIGH_HASHMASK 0xF
51
52static void neigh_timer_handler(struct timer_list *t);
53static void __neigh_notify(struct neighbour *n, int type, int flags,
54 u32 pid);
55static void neigh_update_notify(struct neighbour *neigh, u32 nlmsg_pid);
56static int pneigh_ifdown_and_unlock(struct neigh_table *tbl,
57 struct net_device *dev);
58
59#ifdef CONFIG_PROC_FS
60static const struct seq_operations neigh_stat_seq_ops;
61#endif
62
63static struct hlist_head *neigh_get_dev_table(struct net_device *dev, int family)
64{
65 int i;
66
67 switch (family) {
68 default:
69 DEBUG_NET_WARN_ON_ONCE(1);
70 fallthrough; /* to avoid panic by null-ptr-deref */
71 case AF_INET:
72 i = NEIGH_ARP_TABLE;
73 break;
74 case AF_INET6:
75 i = NEIGH_ND_TABLE;
76 break;
77 }
78
79 return &dev->neighbours[i];
80}
81
82/*
83 Neighbour hash table buckets are protected with rwlock tbl->lock.
84
85 - All the scans/updates to hash buckets MUST be made under this lock.
86 - NOTHING clever should be made under this lock: no callbacks
87 to protocol backends, no attempts to send something to network.
88 It will result in deadlocks, if backend/driver wants to use neighbour
89 cache.
90 - If the entry requires some non-trivial actions, increase
91 its reference count and release table lock.
92
93 Neighbour entries are protected:
94 - with reference count.
95 - with rwlock neigh->lock
96
97 Reference count prevents destruction.
98
99 neigh->lock mainly serializes ll address data and its validity state.
100 However, the same lock is used to protect another entry fields:
101 - timer
102 - resolution queue
103
104 Again, nothing clever shall be made under neigh->lock,
105 the most complicated procedure, which we allow is dev->hard_header.
106 It is supposed, that dev->hard_header is simplistic and does
107 not make callbacks to neighbour tables.
108 */
109
110static int neigh_blackhole(struct neighbour *neigh, struct sk_buff *skb)
111{
112 kfree_skb(skb);
113 return -ENETDOWN;
114}
115
116static void neigh_cleanup_and_release(struct neighbour *neigh)
117{
118 trace_neigh_cleanup_and_release(neigh, 0);
119 __neigh_notify(neigh, RTM_DELNEIGH, 0, 0);
120 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
121 neigh_release(neigh);
122}
123
124/*
125 * It is random distribution in the interval (1/2)*base...(3/2)*base.
126 * It corresponds to default IPv6 settings and is not overridable,
127 * because it is really reasonable choice.
128 */
129
130unsigned long neigh_rand_reach_time(unsigned long base)
131{
132 return base ? get_random_u32_below(base) + (base >> 1) : 0;
133}
134EXPORT_SYMBOL(neigh_rand_reach_time);
135
136static void neigh_mark_dead(struct neighbour *n)
137{
138 n->dead = 1;
139 if (!list_empty(&n->gc_list)) {
140 list_del_init(&n->gc_list);
141 atomic_dec(&n->tbl->gc_entries);
142 }
143 if (!list_empty(&n->managed_list))
144 list_del_init(&n->managed_list);
145}
146
147static void neigh_update_gc_list(struct neighbour *n)
148{
149 bool on_gc_list, exempt_from_gc;
150
151 write_lock_bh(&n->tbl->lock);
152 write_lock(&n->lock);
153 if (n->dead)
154 goto out;
155
156 /* remove from the gc list if new state is permanent or if neighbor
157 * is externally learned; otherwise entry should be on the gc list
158 */
159 exempt_from_gc = n->nud_state & NUD_PERMANENT ||
160 n->flags & NTF_EXT_LEARNED;
161 on_gc_list = !list_empty(&n->gc_list);
162
163 if (exempt_from_gc && on_gc_list) {
164 list_del_init(&n->gc_list);
165 atomic_dec(&n->tbl->gc_entries);
166 } else if (!exempt_from_gc && !on_gc_list) {
167 /* add entries to the tail; cleaning removes from the front */
168 list_add_tail(&n->gc_list, &n->tbl->gc_list);
169 atomic_inc(&n->tbl->gc_entries);
170 }
171out:
172 write_unlock(&n->lock);
173 write_unlock_bh(&n->tbl->lock);
174}
175
176static void neigh_update_managed_list(struct neighbour *n)
177{
178 bool on_managed_list, add_to_managed;
179
180 write_lock_bh(&n->tbl->lock);
181 write_lock(&n->lock);
182 if (n->dead)
183 goto out;
184
185 add_to_managed = n->flags & NTF_MANAGED;
186 on_managed_list = !list_empty(&n->managed_list);
187
188 if (!add_to_managed && on_managed_list)
189 list_del_init(&n->managed_list);
190 else if (add_to_managed && !on_managed_list)
191 list_add_tail(&n->managed_list, &n->tbl->managed_list);
192out:
193 write_unlock(&n->lock);
194 write_unlock_bh(&n->tbl->lock);
195}
196
197static void neigh_update_flags(struct neighbour *neigh, u32 flags, int *notify,
198 bool *gc_update, bool *managed_update)
199{
200 u32 ndm_flags, old_flags = neigh->flags;
201
202 if (!(flags & NEIGH_UPDATE_F_ADMIN))
203 return;
204
205 ndm_flags = (flags & NEIGH_UPDATE_F_EXT_LEARNED) ? NTF_EXT_LEARNED : 0;
206 ndm_flags |= (flags & NEIGH_UPDATE_F_MANAGED) ? NTF_MANAGED : 0;
207
208 if ((old_flags ^ ndm_flags) & NTF_EXT_LEARNED) {
209 if (ndm_flags & NTF_EXT_LEARNED)
210 neigh->flags |= NTF_EXT_LEARNED;
211 else
212 neigh->flags &= ~NTF_EXT_LEARNED;
213 *notify = 1;
214 *gc_update = true;
215 }
216 if ((old_flags ^ ndm_flags) & NTF_MANAGED) {
217 if (ndm_flags & NTF_MANAGED)
218 neigh->flags |= NTF_MANAGED;
219 else
220 neigh->flags &= ~NTF_MANAGED;
221 *notify = 1;
222 *managed_update = true;
223 }
224}
225
226bool neigh_remove_one(struct neighbour *n)
227{
228 bool retval = false;
229
230 write_lock(&n->lock);
231 if (refcount_read(&n->refcnt) == 1) {
232 hlist_del_rcu(&n->hash);
233 hlist_del_rcu(&n->dev_list);
234 neigh_mark_dead(n);
235 retval = true;
236 }
237 write_unlock(&n->lock);
238 if (retval)
239 neigh_cleanup_and_release(n);
240 return retval;
241}
242
243static int neigh_forced_gc(struct neigh_table *tbl)
244{
245 int max_clean = atomic_read(&tbl->gc_entries) -
246 READ_ONCE(tbl->gc_thresh2);
247 u64 tmax = ktime_get_ns() + NSEC_PER_MSEC;
248 unsigned long tref = jiffies - 5 * HZ;
249 struct neighbour *n, *tmp;
250 int shrunk = 0;
251 int loop = 0;
252
253 NEIGH_CACHE_STAT_INC(tbl, forced_gc_runs);
254
255 write_lock_bh(&tbl->lock);
256
257 list_for_each_entry_safe(n, tmp, &tbl->gc_list, gc_list) {
258 if (refcount_read(&n->refcnt) == 1) {
259 bool remove = false;
260
261 write_lock(&n->lock);
262 if ((n->nud_state == NUD_FAILED) ||
263 (n->nud_state == NUD_NOARP) ||
264 (tbl->is_multicast &&
265 tbl->is_multicast(n->primary_key)) ||
266 !time_in_range(n->updated, tref, jiffies))
267 remove = true;
268 write_unlock(&n->lock);
269
270 if (remove && neigh_remove_one(n))
271 shrunk++;
272 if (shrunk >= max_clean)
273 break;
274 if (++loop == 16) {
275 if (ktime_get_ns() > tmax)
276 goto unlock;
277 loop = 0;
278 }
279 }
280 }
281
282 WRITE_ONCE(tbl->last_flush, jiffies);
283unlock:
284 write_unlock_bh(&tbl->lock);
285
286 return shrunk;
287}
288
289static void neigh_add_timer(struct neighbour *n, unsigned long when)
290{
291 /* Use safe distance from the jiffies - LONG_MAX point while timer
292 * is running in DELAY/PROBE state but still show to user space
293 * large times in the past.
294 */
295 unsigned long mint = jiffies - (LONG_MAX - 86400 * HZ);
296
297 neigh_hold(n);
298 if (!time_in_range(n->confirmed, mint, jiffies))
299 n->confirmed = mint;
300 if (time_before(n->used, n->confirmed))
301 n->used = n->confirmed;
302 if (unlikely(mod_timer(&n->timer, when))) {
303 printk("NEIGH: BUG, double timer add, state is %x\n",
304 n->nud_state);
305 dump_stack();
306 }
307}
308
309static int neigh_del_timer(struct neighbour *n)
310{
311 if ((n->nud_state & NUD_IN_TIMER) &&
312 del_timer(&n->timer)) {
313 neigh_release(n);
314 return 1;
315 }
316 return 0;
317}
318
319static struct neigh_parms *neigh_get_dev_parms_rcu(struct net_device *dev,
320 int family)
321{
322 switch (family) {
323 case AF_INET:
324 return __in_dev_arp_parms_get_rcu(dev);
325 case AF_INET6:
326 return __in6_dev_nd_parms_get_rcu(dev);
327 }
328 return NULL;
329}
330
331static void neigh_parms_qlen_dec(struct net_device *dev, int family)
332{
333 struct neigh_parms *p;
334
335 rcu_read_lock();
336 p = neigh_get_dev_parms_rcu(dev, family);
337 if (p)
338 p->qlen--;
339 rcu_read_unlock();
340}
341
342static void pneigh_queue_purge(struct sk_buff_head *list, struct net *net,
343 int family)
344{
345 struct sk_buff_head tmp;
346 unsigned long flags;
347 struct sk_buff *skb;
348
349 skb_queue_head_init(&tmp);
350 spin_lock_irqsave(&list->lock, flags);
351 skb = skb_peek(list);
352 while (skb != NULL) {
353 struct sk_buff *skb_next = skb_peek_next(skb, list);
354 struct net_device *dev = skb->dev;
355
356 if (net == NULL || net_eq(dev_net(dev), net)) {
357 neigh_parms_qlen_dec(dev, family);
358 __skb_unlink(skb, list);
359 __skb_queue_tail(&tmp, skb);
360 }
361 skb = skb_next;
362 }
363 spin_unlock_irqrestore(&list->lock, flags);
364
365 while ((skb = __skb_dequeue(&tmp))) {
366 dev_put(skb->dev);
367 kfree_skb(skb);
368 }
369}
370
371static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev,
372 bool skip_perm)
373{
374 struct hlist_head *dev_head;
375 struct hlist_node *tmp;
376 struct neighbour *n;
377
378 dev_head = neigh_get_dev_table(dev, tbl->family);
379
380 hlist_for_each_entry_safe(n, tmp, dev_head, dev_list) {
381 if (skip_perm && n->nud_state & NUD_PERMANENT)
382 continue;
383
384 hlist_del_rcu(&n->hash);
385 hlist_del_rcu(&n->dev_list);
386 write_lock(&n->lock);
387 neigh_del_timer(n);
388 neigh_mark_dead(n);
389 if (refcount_read(&n->refcnt) != 1) {
390 /* The most unpleasant situation.
391 * We must destroy neighbour entry,
392 * but someone still uses it.
393 *
394 * The destroy will be delayed until
395 * the last user releases us, but
396 * we must kill timers etc. and move
397 * it to safe state.
398 */
399 __skb_queue_purge(&n->arp_queue);
400 n->arp_queue_len_bytes = 0;
401 WRITE_ONCE(n->output, neigh_blackhole);
402 if (n->nud_state & NUD_VALID)
403 n->nud_state = NUD_NOARP;
404 else
405 n->nud_state = NUD_NONE;
406 neigh_dbg(2, "neigh %p is stray\n", n);
407 }
408 write_unlock(&n->lock);
409 neigh_cleanup_and_release(n);
410 }
411}
412
413void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev)
414{
415 write_lock_bh(&tbl->lock);
416 neigh_flush_dev(tbl, dev, false);
417 write_unlock_bh(&tbl->lock);
418}
419EXPORT_SYMBOL(neigh_changeaddr);
420
421static int __neigh_ifdown(struct neigh_table *tbl, struct net_device *dev,
422 bool skip_perm)
423{
424 write_lock_bh(&tbl->lock);
425 neigh_flush_dev(tbl, dev, skip_perm);
426 pneigh_ifdown_and_unlock(tbl, dev);
427 pneigh_queue_purge(&tbl->proxy_queue, dev ? dev_net(dev) : NULL,
428 tbl->family);
429 if (skb_queue_empty_lockless(&tbl->proxy_queue))
430 del_timer_sync(&tbl->proxy_timer);
431 return 0;
432}
433
434int neigh_carrier_down(struct neigh_table *tbl, struct net_device *dev)
435{
436 __neigh_ifdown(tbl, dev, true);
437 return 0;
438}
439EXPORT_SYMBOL(neigh_carrier_down);
440
441int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
442{
443 __neigh_ifdown(tbl, dev, false);
444 return 0;
445}
446EXPORT_SYMBOL(neigh_ifdown);
447
448static struct neighbour *neigh_alloc(struct neigh_table *tbl,
449 struct net_device *dev,
450 u32 flags, bool exempt_from_gc)
451{
452 struct neighbour *n = NULL;
453 unsigned long now = jiffies;
454 int entries, gc_thresh3;
455
456 if (exempt_from_gc)
457 goto do_alloc;
458
459 entries = atomic_inc_return(&tbl->gc_entries) - 1;
460 gc_thresh3 = READ_ONCE(tbl->gc_thresh3);
461 if (entries >= gc_thresh3 ||
462 (entries >= READ_ONCE(tbl->gc_thresh2) &&
463 time_after(now, READ_ONCE(tbl->last_flush) + 5 * HZ))) {
464 if (!neigh_forced_gc(tbl) && entries >= gc_thresh3) {
465 net_info_ratelimited("%s: neighbor table overflow!\n",
466 tbl->id);
467 NEIGH_CACHE_STAT_INC(tbl, table_fulls);
468 goto out_entries;
469 }
470 }
471
472do_alloc:
473 n = kzalloc(tbl->entry_size + dev->neigh_priv_len, GFP_ATOMIC);
474 if (!n)
475 goto out_entries;
476
477 __skb_queue_head_init(&n->arp_queue);
478 rwlock_init(&n->lock);
479 seqlock_init(&n->ha_lock);
480 n->updated = n->used = now;
481 n->nud_state = NUD_NONE;
482 n->output = neigh_blackhole;
483 n->flags = flags;
484 seqlock_init(&n->hh.hh_lock);
485 n->parms = neigh_parms_clone(&tbl->parms);
486 timer_setup(&n->timer, neigh_timer_handler, 0);
487
488 NEIGH_CACHE_STAT_INC(tbl, allocs);
489 n->tbl = tbl;
490 refcount_set(&n->refcnt, 1);
491 n->dead = 1;
492 INIT_LIST_HEAD(&n->gc_list);
493 INIT_LIST_HEAD(&n->managed_list);
494
495 atomic_inc(&tbl->entries);
496out:
497 return n;
498
499out_entries:
500 if (!exempt_from_gc)
501 atomic_dec(&tbl->gc_entries);
502 goto out;
503}
504
505static void neigh_get_hash_rnd(u32 *x)
506{
507 *x = get_random_u32() | 1;
508}
509
510static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift)
511{
512 size_t size = (1 << shift) * sizeof(struct hlist_head);
513 struct hlist_head *hash_heads;
514 struct neigh_hash_table *ret;
515 int i;
516
517 ret = kmalloc(sizeof(*ret), GFP_ATOMIC);
518 if (!ret)
519 return NULL;
520
521 hash_heads = kvzalloc(size, GFP_ATOMIC);
522 if (!hash_heads) {
523 kfree(ret);
524 return NULL;
525 }
526 ret->hash_heads = hash_heads;
527 ret->hash_shift = shift;
528 for (i = 0; i < NEIGH_NUM_HASH_RND; i++)
529 neigh_get_hash_rnd(&ret->hash_rnd[i]);
530 return ret;
531}
532
533static void neigh_hash_free_rcu(struct rcu_head *head)
534{
535 struct neigh_hash_table *nht = container_of(head,
536 struct neigh_hash_table,
537 rcu);
538
539 kvfree(nht->hash_heads);
540 kfree(nht);
541}
542
543static struct neigh_hash_table *neigh_hash_grow(struct neigh_table *tbl,
544 unsigned long new_shift)
545{
546 unsigned int i, hash;
547 struct neigh_hash_table *new_nht, *old_nht;
548
549 NEIGH_CACHE_STAT_INC(tbl, hash_grows);
550
551 old_nht = rcu_dereference_protected(tbl->nht,
552 lockdep_is_held(&tbl->lock));
553 new_nht = neigh_hash_alloc(new_shift);
554 if (!new_nht)
555 return old_nht;
556
557 for (i = 0; i < (1 << old_nht->hash_shift); i++) {
558 struct hlist_node *tmp;
559 struct neighbour *n;
560
561 neigh_for_each_in_bucket_safe(n, tmp, &old_nht->hash_heads[i]) {
562 hash = tbl->hash(n->primary_key, n->dev,
563 new_nht->hash_rnd);
564
565 hash >>= (32 - new_nht->hash_shift);
566
567 hlist_del_rcu(&n->hash);
568 hlist_add_head_rcu(&n->hash, &new_nht->hash_heads[hash]);
569 }
570 }
571
572 rcu_assign_pointer(tbl->nht, new_nht);
573 call_rcu(&old_nht->rcu, neigh_hash_free_rcu);
574 return new_nht;
575}
576
577struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
578 struct net_device *dev)
579{
580 struct neighbour *n;
581
582 NEIGH_CACHE_STAT_INC(tbl, lookups);
583
584 rcu_read_lock();
585 n = __neigh_lookup_noref(tbl, pkey, dev);
586 if (n) {
587 if (!refcount_inc_not_zero(&n->refcnt))
588 n = NULL;
589 NEIGH_CACHE_STAT_INC(tbl, hits);
590 }
591
592 rcu_read_unlock();
593 return n;
594}
595EXPORT_SYMBOL(neigh_lookup);
596
597static struct neighbour *
598___neigh_create(struct neigh_table *tbl, const void *pkey,
599 struct net_device *dev, u32 flags,
600 bool exempt_from_gc, bool want_ref)
601{
602 u32 hash_val, key_len = tbl->key_len;
603 struct neighbour *n1, *rc, *n;
604 struct neigh_hash_table *nht;
605 int error;
606
607 n = neigh_alloc(tbl, dev, flags, exempt_from_gc);
608 trace_neigh_create(tbl, dev, pkey, n, exempt_from_gc);
609 if (!n) {
610 rc = ERR_PTR(-ENOBUFS);
611 goto out;
612 }
613
614 memcpy(n->primary_key, pkey, key_len);
615 n->dev = dev;
616 netdev_hold(dev, &n->dev_tracker, GFP_ATOMIC);
617
618 /* Protocol specific setup. */
619 if (tbl->constructor && (error = tbl->constructor(n)) < 0) {
620 rc = ERR_PTR(error);
621 goto out_neigh_release;
622 }
623
624 if (dev->netdev_ops->ndo_neigh_construct) {
625 error = dev->netdev_ops->ndo_neigh_construct(dev, n);
626 if (error < 0) {
627 rc = ERR_PTR(error);
628 goto out_neigh_release;
629 }
630 }
631
632 /* Device specific setup. */
633 if (n->parms->neigh_setup &&
634 (error = n->parms->neigh_setup(n)) < 0) {
635 rc = ERR_PTR(error);
636 goto out_neigh_release;
637 }
638
639 n->confirmed = jiffies - (NEIGH_VAR(n->parms, BASE_REACHABLE_TIME) << 1);
640
641 write_lock_bh(&tbl->lock);
642 nht = rcu_dereference_protected(tbl->nht,
643 lockdep_is_held(&tbl->lock));
644
645 if (atomic_read(&tbl->entries) > (1 << nht->hash_shift))
646 nht = neigh_hash_grow(tbl, nht->hash_shift + 1);
647
648 hash_val = tbl->hash(n->primary_key, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
649
650 if (n->parms->dead) {
651 rc = ERR_PTR(-EINVAL);
652 goto out_tbl_unlock;
653 }
654
655 neigh_for_each_in_bucket(n1, &nht->hash_heads[hash_val]) {
656 if (dev == n1->dev && !memcmp(n1->primary_key, n->primary_key, key_len)) {
657 if (want_ref)
658 neigh_hold(n1);
659 rc = n1;
660 goto out_tbl_unlock;
661 }
662 }
663
664 n->dead = 0;
665 if (!exempt_from_gc)
666 list_add_tail(&n->gc_list, &n->tbl->gc_list);
667 if (n->flags & NTF_MANAGED)
668 list_add_tail(&n->managed_list, &n->tbl->managed_list);
669 if (want_ref)
670 neigh_hold(n);
671 hlist_add_head_rcu(&n->hash, &nht->hash_heads[hash_val]);
672
673 hlist_add_head_rcu(&n->dev_list,
674 neigh_get_dev_table(dev, tbl->family));
675
676 write_unlock_bh(&tbl->lock);
677 neigh_dbg(2, "neigh %p is created\n", n);
678 rc = n;
679out:
680 return rc;
681out_tbl_unlock:
682 write_unlock_bh(&tbl->lock);
683out_neigh_release:
684 if (!exempt_from_gc)
685 atomic_dec(&tbl->gc_entries);
686 neigh_release(n);
687 goto out;
688}
689
690struct neighbour *__neigh_create(struct neigh_table *tbl, const void *pkey,
691 struct net_device *dev, bool want_ref)
692{
693 bool exempt_from_gc = !!(dev->flags & IFF_LOOPBACK);
694
695 return ___neigh_create(tbl, pkey, dev, 0, exempt_from_gc, want_ref);
696}
697EXPORT_SYMBOL(__neigh_create);
698
699static u32 pneigh_hash(const void *pkey, unsigned int key_len)
700{
701 u32 hash_val = *(u32 *)(pkey + key_len - 4);
702 hash_val ^= (hash_val >> 16);
703 hash_val ^= hash_val >> 8;
704 hash_val ^= hash_val >> 4;
705 hash_val &= PNEIGH_HASHMASK;
706 return hash_val;
707}
708
709static struct pneigh_entry *__pneigh_lookup_1(struct pneigh_entry *n,
710 struct net *net,
711 const void *pkey,
712 unsigned int key_len,
713 struct net_device *dev)
714{
715 while (n) {
716 if (!memcmp(n->key, pkey, key_len) &&
717 net_eq(pneigh_net(n), net) &&
718 (n->dev == dev || !n->dev))
719 return n;
720 n = n->next;
721 }
722 return NULL;
723}
724
725struct pneigh_entry *__pneigh_lookup(struct neigh_table *tbl,
726 struct net *net, const void *pkey, struct net_device *dev)
727{
728 unsigned int key_len = tbl->key_len;
729 u32 hash_val = pneigh_hash(pkey, key_len);
730
731 return __pneigh_lookup_1(tbl->phash_buckets[hash_val],
732 net, pkey, key_len, dev);
733}
734EXPORT_SYMBOL_GPL(__pneigh_lookup);
735
736struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl,
737 struct net *net, const void *pkey,
738 struct net_device *dev, int creat)
739{
740 struct pneigh_entry *n;
741 unsigned int key_len = tbl->key_len;
742 u32 hash_val = pneigh_hash(pkey, key_len);
743
744 read_lock_bh(&tbl->lock);
745 n = __pneigh_lookup_1(tbl->phash_buckets[hash_val],
746 net, pkey, key_len, dev);
747 read_unlock_bh(&tbl->lock);
748
749 if (n || !creat)
750 goto out;
751
752 ASSERT_RTNL();
753
754 n = kzalloc(sizeof(*n) + key_len, GFP_KERNEL);
755 if (!n)
756 goto out;
757
758 write_pnet(&n->net, net);
759 memcpy(n->key, pkey, key_len);
760 n->dev = dev;
761 netdev_hold(dev, &n->dev_tracker, GFP_KERNEL);
762
763 if (tbl->pconstructor && tbl->pconstructor(n)) {
764 netdev_put(dev, &n->dev_tracker);
765 kfree(n);
766 n = NULL;
767 goto out;
768 }
769
770 write_lock_bh(&tbl->lock);
771 n->next = tbl->phash_buckets[hash_val];
772 tbl->phash_buckets[hash_val] = n;
773 write_unlock_bh(&tbl->lock);
774out:
775 return n;
776}
777EXPORT_SYMBOL(pneigh_lookup);
778
779
780int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
781 struct net_device *dev)
782{
783 struct pneigh_entry *n, **np;
784 unsigned int key_len = tbl->key_len;
785 u32 hash_val = pneigh_hash(pkey, key_len);
786
787 write_lock_bh(&tbl->lock);
788 for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
789 np = &n->next) {
790 if (!memcmp(n->key, pkey, key_len) && n->dev == dev &&
791 net_eq(pneigh_net(n), net)) {
792 *np = n->next;
793 write_unlock_bh(&tbl->lock);
794 if (tbl->pdestructor)
795 tbl->pdestructor(n);
796 netdev_put(n->dev, &n->dev_tracker);
797 kfree(n);
798 return 0;
799 }
800 }
801 write_unlock_bh(&tbl->lock);
802 return -ENOENT;
803}
804
805static int pneigh_ifdown_and_unlock(struct neigh_table *tbl,
806 struct net_device *dev)
807{
808 struct pneigh_entry *n, **np, *freelist = NULL;
809 u32 h;
810
811 for (h = 0; h <= PNEIGH_HASHMASK; h++) {
812 np = &tbl->phash_buckets[h];
813 while ((n = *np) != NULL) {
814 if (!dev || n->dev == dev) {
815 *np = n->next;
816 n->next = freelist;
817 freelist = n;
818 continue;
819 }
820 np = &n->next;
821 }
822 }
823 write_unlock_bh(&tbl->lock);
824 while ((n = freelist)) {
825 freelist = n->next;
826 n->next = NULL;
827 if (tbl->pdestructor)
828 tbl->pdestructor(n);
829 netdev_put(n->dev, &n->dev_tracker);
830 kfree(n);
831 }
832 return -ENOENT;
833}
834
835static void neigh_parms_destroy(struct neigh_parms *parms);
836
837static inline void neigh_parms_put(struct neigh_parms *parms)
838{
839 if (refcount_dec_and_test(&parms->refcnt))
840 neigh_parms_destroy(parms);
841}
842
843/*
844 * neighbour must already be out of the table;
845 *
846 */
847void neigh_destroy(struct neighbour *neigh)
848{
849 struct net_device *dev = neigh->dev;
850
851 NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
852
853 if (!neigh->dead) {
854 pr_warn("Destroying alive neighbour %p\n", neigh);
855 dump_stack();
856 return;
857 }
858
859 if (neigh_del_timer(neigh))
860 pr_warn("Impossible event\n");
861
862 write_lock_bh(&neigh->lock);
863 __skb_queue_purge(&neigh->arp_queue);
864 write_unlock_bh(&neigh->lock);
865 neigh->arp_queue_len_bytes = 0;
866
867 if (dev->netdev_ops->ndo_neigh_destroy)
868 dev->netdev_ops->ndo_neigh_destroy(dev, neigh);
869
870 netdev_put(dev, &neigh->dev_tracker);
871 neigh_parms_put(neigh->parms);
872
873 neigh_dbg(2, "neigh %p is destroyed\n", neigh);
874
875 atomic_dec(&neigh->tbl->entries);
876 kfree_rcu(neigh, rcu);
877}
878EXPORT_SYMBOL(neigh_destroy);
879
880/* Neighbour state is suspicious;
881 disable fast path.
882
883 Called with write_locked neigh.
884 */
885static void neigh_suspect(struct neighbour *neigh)
886{
887 neigh_dbg(2, "neigh %p is suspected\n", neigh);
888
889 WRITE_ONCE(neigh->output, neigh->ops->output);
890}
891
892/* Neighbour state is OK;
893 enable fast path.
894
895 Called with write_locked neigh.
896 */
897static void neigh_connect(struct neighbour *neigh)
898{
899 neigh_dbg(2, "neigh %p is connected\n", neigh);
900
901 WRITE_ONCE(neigh->output, neigh->ops->connected_output);
902}
903
904static void neigh_periodic_work(struct work_struct *work)
905{
906 struct neigh_table *tbl = container_of(work, struct neigh_table, gc_work.work);
907 struct neigh_hash_table *nht;
908 struct hlist_node *tmp;
909 struct neighbour *n;
910 unsigned int i;
911
912 NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
913
914 write_lock_bh(&tbl->lock);
915 nht = rcu_dereference_protected(tbl->nht,
916 lockdep_is_held(&tbl->lock));
917
918 /*
919 * periodically recompute ReachableTime from random function
920 */
921
922 if (time_after(jiffies, tbl->last_rand + 300 * HZ)) {
923 struct neigh_parms *p;
924
925 WRITE_ONCE(tbl->last_rand, jiffies);
926 list_for_each_entry(p, &tbl->parms_list, list)
927 p->reachable_time =
928 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
929 }
930
931 if (atomic_read(&tbl->entries) < READ_ONCE(tbl->gc_thresh1))
932 goto out;
933
934 for (i = 0 ; i < (1 << nht->hash_shift); i++) {
935 neigh_for_each_in_bucket_safe(n, tmp, &nht->hash_heads[i]) {
936 unsigned int state;
937
938 write_lock(&n->lock);
939
940 state = n->nud_state;
941 if ((state & (NUD_PERMANENT | NUD_IN_TIMER)) ||
942 (n->flags & NTF_EXT_LEARNED)) {
943 write_unlock(&n->lock);
944 continue;
945 }
946
947 if (time_before(n->used, n->confirmed) &&
948 time_is_before_eq_jiffies(n->confirmed))
949 n->used = n->confirmed;
950
951 if (refcount_read(&n->refcnt) == 1 &&
952 (state == NUD_FAILED ||
953 !time_in_range_open(jiffies, n->used,
954 n->used + NEIGH_VAR(n->parms, GC_STALETIME)))) {
955 hlist_del_rcu(&n->hash);
956 hlist_del_rcu(&n->dev_list);
957 neigh_mark_dead(n);
958 write_unlock(&n->lock);
959 neigh_cleanup_and_release(n);
960 continue;
961 }
962 write_unlock(&n->lock);
963 }
964 /*
965 * It's fine to release lock here, even if hash table
966 * grows while we are preempted.
967 */
968 write_unlock_bh(&tbl->lock);
969 cond_resched();
970 write_lock_bh(&tbl->lock);
971 nht = rcu_dereference_protected(tbl->nht,
972 lockdep_is_held(&tbl->lock));
973 }
974out:
975 /* Cycle through all hash buckets every BASE_REACHABLE_TIME/2 ticks.
976 * ARP entry timeouts range from 1/2 BASE_REACHABLE_TIME to 3/2
977 * BASE_REACHABLE_TIME.
978 */
979 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
980 NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME) >> 1);
981 write_unlock_bh(&tbl->lock);
982}
983
984static __inline__ int neigh_max_probes(struct neighbour *n)
985{
986 struct neigh_parms *p = n->parms;
987 return NEIGH_VAR(p, UCAST_PROBES) + NEIGH_VAR(p, APP_PROBES) +
988 (n->nud_state & NUD_PROBE ? NEIGH_VAR(p, MCAST_REPROBES) :
989 NEIGH_VAR(p, MCAST_PROBES));
990}
991
992static void neigh_invalidate(struct neighbour *neigh)
993 __releases(neigh->lock)
994 __acquires(neigh->lock)
995{
996 struct sk_buff *skb;
997
998 NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
999 neigh_dbg(2, "neigh %p is failed\n", neigh);
1000 neigh->updated = jiffies;
1001
1002 /* It is very thin place. report_unreachable is very complicated
1003 routine. Particularly, it can hit the same neighbour entry!
1004
1005 So that, we try to be accurate and avoid dead loop. --ANK
1006 */
1007 while (neigh->nud_state == NUD_FAILED &&
1008 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1009 write_unlock(&neigh->lock);
1010 neigh->ops->error_report(neigh, skb);
1011 write_lock(&neigh->lock);
1012 }
1013 __skb_queue_purge(&neigh->arp_queue);
1014 neigh->arp_queue_len_bytes = 0;
1015}
1016
1017static void neigh_probe(struct neighbour *neigh)
1018 __releases(neigh->lock)
1019{
1020 struct sk_buff *skb = skb_peek_tail(&neigh->arp_queue);
1021 /* keep skb alive even if arp_queue overflows */
1022 if (skb)
1023 skb = skb_clone(skb, GFP_ATOMIC);
1024 write_unlock(&neigh->lock);
1025 if (neigh->ops->solicit)
1026 neigh->ops->solicit(neigh, skb);
1027 atomic_inc(&neigh->probes);
1028 consume_skb(skb);
1029}
1030
1031/* Called when a timer expires for a neighbour entry. */
1032
1033static void neigh_timer_handler(struct timer_list *t)
1034{
1035 unsigned long now, next;
1036 struct neighbour *neigh = from_timer(neigh, t, timer);
1037 unsigned int state;
1038 int notify = 0;
1039
1040 write_lock(&neigh->lock);
1041
1042 state = neigh->nud_state;
1043 now = jiffies;
1044 next = now + HZ;
1045
1046 if (!(state & NUD_IN_TIMER))
1047 goto out;
1048
1049 if (state & NUD_REACHABLE) {
1050 if (time_before_eq(now,
1051 neigh->confirmed + neigh->parms->reachable_time)) {
1052 neigh_dbg(2, "neigh %p is still alive\n", neigh);
1053 next = neigh->confirmed + neigh->parms->reachable_time;
1054 } else if (time_before_eq(now,
1055 neigh->used +
1056 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
1057 neigh_dbg(2, "neigh %p is delayed\n", neigh);
1058 WRITE_ONCE(neigh->nud_state, NUD_DELAY);
1059 neigh->updated = jiffies;
1060 neigh_suspect(neigh);
1061 next = now + NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME);
1062 } else {
1063 neigh_dbg(2, "neigh %p is suspected\n", neigh);
1064 WRITE_ONCE(neigh->nud_state, NUD_STALE);
1065 neigh->updated = jiffies;
1066 neigh_suspect(neigh);
1067 notify = 1;
1068 }
1069 } else if (state & NUD_DELAY) {
1070 if (time_before_eq(now,
1071 neigh->confirmed +
1072 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
1073 neigh_dbg(2, "neigh %p is now reachable\n", neigh);
1074 WRITE_ONCE(neigh->nud_state, NUD_REACHABLE);
1075 neigh->updated = jiffies;
1076 neigh_connect(neigh);
1077 notify = 1;
1078 next = neigh->confirmed + neigh->parms->reachable_time;
1079 } else {
1080 neigh_dbg(2, "neigh %p is probed\n", neigh);
1081 WRITE_ONCE(neigh->nud_state, NUD_PROBE);
1082 neigh->updated = jiffies;
1083 atomic_set(&neigh->probes, 0);
1084 notify = 1;
1085 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
1086 HZ/100);
1087 }
1088 } else {
1089 /* NUD_PROBE|NUD_INCOMPLETE */
1090 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME), HZ/100);
1091 }
1092
1093 if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
1094 atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
1095 WRITE_ONCE(neigh->nud_state, NUD_FAILED);
1096 notify = 1;
1097 neigh_invalidate(neigh);
1098 goto out;
1099 }
1100
1101 if (neigh->nud_state & NUD_IN_TIMER) {
1102 if (time_before(next, jiffies + HZ/100))
1103 next = jiffies + HZ/100;
1104 if (!mod_timer(&neigh->timer, next))
1105 neigh_hold(neigh);
1106 }
1107 if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
1108 neigh_probe(neigh);
1109 } else {
1110out:
1111 write_unlock(&neigh->lock);
1112 }
1113
1114 if (notify)
1115 neigh_update_notify(neigh, 0);
1116
1117 trace_neigh_timer_handler(neigh, 0);
1118
1119 neigh_release(neigh);
1120}
1121
1122int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb,
1123 const bool immediate_ok)
1124{
1125 int rc;
1126 bool immediate_probe = false;
1127
1128 write_lock_bh(&neigh->lock);
1129
1130 rc = 0;
1131 if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
1132 goto out_unlock_bh;
1133 if (neigh->dead)
1134 goto out_dead;
1135
1136 if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
1137 if (NEIGH_VAR(neigh->parms, MCAST_PROBES) +
1138 NEIGH_VAR(neigh->parms, APP_PROBES)) {
1139 unsigned long next, now = jiffies;
1140
1141 atomic_set(&neigh->probes,
1142 NEIGH_VAR(neigh->parms, UCAST_PROBES));
1143 neigh_del_timer(neigh);
1144 WRITE_ONCE(neigh->nud_state, NUD_INCOMPLETE);
1145 neigh->updated = now;
1146 if (!immediate_ok) {
1147 next = now + 1;
1148 } else {
1149 immediate_probe = true;
1150 next = now + max(NEIGH_VAR(neigh->parms,
1151 RETRANS_TIME),
1152 HZ / 100);
1153 }
1154 neigh_add_timer(neigh, next);
1155 } else {
1156 WRITE_ONCE(neigh->nud_state, NUD_FAILED);
1157 neigh->updated = jiffies;
1158 write_unlock_bh(&neigh->lock);
1159
1160 kfree_skb_reason(skb, SKB_DROP_REASON_NEIGH_FAILED);
1161 return 1;
1162 }
1163 } else if (neigh->nud_state & NUD_STALE) {
1164 neigh_dbg(2, "neigh %p is delayed\n", neigh);
1165 neigh_del_timer(neigh);
1166 WRITE_ONCE(neigh->nud_state, NUD_DELAY);
1167 neigh->updated = jiffies;
1168 neigh_add_timer(neigh, jiffies +
1169 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME));
1170 }
1171
1172 if (neigh->nud_state == NUD_INCOMPLETE) {
1173 if (skb) {
1174 while (neigh->arp_queue_len_bytes + skb->truesize >
1175 NEIGH_VAR(neigh->parms, QUEUE_LEN_BYTES)) {
1176 struct sk_buff *buff;
1177
1178 buff = __skb_dequeue(&neigh->arp_queue);
1179 if (!buff)
1180 break;
1181 neigh->arp_queue_len_bytes -= buff->truesize;
1182 kfree_skb_reason(buff, SKB_DROP_REASON_NEIGH_QUEUEFULL);
1183 NEIGH_CACHE_STAT_INC(neigh->tbl, unres_discards);
1184 }
1185 skb_dst_force(skb);
1186 __skb_queue_tail(&neigh->arp_queue, skb);
1187 neigh->arp_queue_len_bytes += skb->truesize;
1188 }
1189 rc = 1;
1190 }
1191out_unlock_bh:
1192 if (immediate_probe)
1193 neigh_probe(neigh);
1194 else
1195 write_unlock(&neigh->lock);
1196 local_bh_enable();
1197 trace_neigh_event_send_done(neigh, rc);
1198 return rc;
1199
1200out_dead:
1201 if (neigh->nud_state & NUD_STALE)
1202 goto out_unlock_bh;
1203 write_unlock_bh(&neigh->lock);
1204 kfree_skb_reason(skb, SKB_DROP_REASON_NEIGH_DEAD);
1205 trace_neigh_event_send_dead(neigh, 1);
1206 return 1;
1207}
1208EXPORT_SYMBOL(__neigh_event_send);
1209
1210static void neigh_update_hhs(struct neighbour *neigh)
1211{
1212 struct hh_cache *hh;
1213 void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *)
1214 = NULL;
1215
1216 if (neigh->dev->header_ops)
1217 update = neigh->dev->header_ops->cache_update;
1218
1219 if (update) {
1220 hh = &neigh->hh;
1221 if (READ_ONCE(hh->hh_len)) {
1222 write_seqlock_bh(&hh->hh_lock);
1223 update(hh, neigh->dev, neigh->ha);
1224 write_sequnlock_bh(&hh->hh_lock);
1225 }
1226 }
1227}
1228
1229/* Generic update routine.
1230 -- lladdr is new lladdr or NULL, if it is not supplied.
1231 -- new is new state.
1232 -- flags
1233 NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
1234 if it is different.
1235 NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
1236 lladdr instead of overriding it
1237 if it is different.
1238 NEIGH_UPDATE_F_ADMIN means that the change is administrative.
1239 NEIGH_UPDATE_F_USE means that the entry is user triggered.
1240 NEIGH_UPDATE_F_MANAGED means that the entry will be auto-refreshed.
1241 NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
1242 NTF_ROUTER flag.
1243 NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as
1244 a router.
1245
1246 Caller MUST hold reference count on the entry.
1247 */
1248static int __neigh_update(struct neighbour *neigh, const u8 *lladdr,
1249 u8 new, u32 flags, u32 nlmsg_pid,
1250 struct netlink_ext_ack *extack)
1251{
1252 bool gc_update = false, managed_update = false;
1253 int update_isrouter = 0;
1254 struct net_device *dev;
1255 int err, notify = 0;
1256 u8 old;
1257
1258 trace_neigh_update(neigh, lladdr, new, flags, nlmsg_pid);
1259
1260 write_lock_bh(&neigh->lock);
1261
1262 dev = neigh->dev;
1263 old = neigh->nud_state;
1264 err = -EPERM;
1265
1266 if (neigh->dead) {
1267 NL_SET_ERR_MSG(extack, "Neighbor entry is now dead");
1268 new = old;
1269 goto out;
1270 }
1271 if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
1272 (old & (NUD_NOARP | NUD_PERMANENT)))
1273 goto out;
1274
1275 neigh_update_flags(neigh, flags, ¬ify, &gc_update, &managed_update);
1276 if (flags & (NEIGH_UPDATE_F_USE | NEIGH_UPDATE_F_MANAGED)) {
1277 new = old & ~NUD_PERMANENT;
1278 WRITE_ONCE(neigh->nud_state, new);
1279 err = 0;
1280 goto out;
1281 }
1282
1283 if (!(new & NUD_VALID)) {
1284 neigh_del_timer(neigh);
1285 if (old & NUD_CONNECTED)
1286 neigh_suspect(neigh);
1287 WRITE_ONCE(neigh->nud_state, new);
1288 err = 0;
1289 notify = old & NUD_VALID;
1290 if ((old & (NUD_INCOMPLETE | NUD_PROBE)) &&
1291 (new & NUD_FAILED)) {
1292 neigh_invalidate(neigh);
1293 notify = 1;
1294 }
1295 goto out;
1296 }
1297
1298 /* Compare new lladdr with cached one */
1299 if (!dev->addr_len) {
1300 /* First case: device needs no address. */
1301 lladdr = neigh->ha;
1302 } else if (lladdr) {
1303 /* The second case: if something is already cached
1304 and a new address is proposed:
1305 - compare new & old
1306 - if they are different, check override flag
1307 */
1308 if ((old & NUD_VALID) &&
1309 !memcmp(lladdr, neigh->ha, dev->addr_len))
1310 lladdr = neigh->ha;
1311 } else {
1312 /* No address is supplied; if we know something,
1313 use it, otherwise discard the request.
1314 */
1315 err = -EINVAL;
1316 if (!(old & NUD_VALID)) {
1317 NL_SET_ERR_MSG(extack, "No link layer address given");
1318 goto out;
1319 }
1320 lladdr = neigh->ha;
1321 }
1322
1323 /* Update confirmed timestamp for neighbour entry after we
1324 * received ARP packet even if it doesn't change IP to MAC binding.
1325 */
1326 if (new & NUD_CONNECTED)
1327 neigh->confirmed = jiffies;
1328
1329 /* If entry was valid and address is not changed,
1330 do not change entry state, if new one is STALE.
1331 */
1332 err = 0;
1333 update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1334 if (old & NUD_VALID) {
1335 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
1336 update_isrouter = 0;
1337 if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
1338 (old & NUD_CONNECTED)) {
1339 lladdr = neigh->ha;
1340 new = NUD_STALE;
1341 } else
1342 goto out;
1343 } else {
1344 if (lladdr == neigh->ha && new == NUD_STALE &&
1345 !(flags & NEIGH_UPDATE_F_ADMIN))
1346 new = old;
1347 }
1348 }
1349
1350 /* Update timestamp only once we know we will make a change to the
1351 * neighbour entry. Otherwise we risk to move the locktime window with
1352 * noop updates and ignore relevant ARP updates.
1353 */
1354 if (new != old || lladdr != neigh->ha)
1355 neigh->updated = jiffies;
1356
1357 if (new != old) {
1358 neigh_del_timer(neigh);
1359 if (new & NUD_PROBE)
1360 atomic_set(&neigh->probes, 0);
1361 if (new & NUD_IN_TIMER)
1362 neigh_add_timer(neigh, (jiffies +
1363 ((new & NUD_REACHABLE) ?
1364 neigh->parms->reachable_time :
1365 0)));
1366 WRITE_ONCE(neigh->nud_state, new);
1367 notify = 1;
1368 }
1369
1370 if (lladdr != neigh->ha) {
1371 write_seqlock(&neigh->ha_lock);
1372 memcpy(&neigh->ha, lladdr, dev->addr_len);
1373 write_sequnlock(&neigh->ha_lock);
1374 neigh_update_hhs(neigh);
1375 if (!(new & NUD_CONNECTED))
1376 neigh->confirmed = jiffies -
1377 (NEIGH_VAR(neigh->parms, BASE_REACHABLE_TIME) << 1);
1378 notify = 1;
1379 }
1380 if (new == old)
1381 goto out;
1382 if (new & NUD_CONNECTED)
1383 neigh_connect(neigh);
1384 else
1385 neigh_suspect(neigh);
1386 if (!(old & NUD_VALID)) {
1387 struct sk_buff *skb;
1388
1389 /* Again: avoid dead loop if something went wrong */
1390
1391 while (neigh->nud_state & NUD_VALID &&
1392 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1393 struct dst_entry *dst = skb_dst(skb);
1394 struct neighbour *n2, *n1 = neigh;
1395 write_unlock_bh(&neigh->lock);
1396
1397 rcu_read_lock();
1398
1399 /* Why not just use 'neigh' as-is? The problem is that
1400 * things such as shaper, eql, and sch_teql can end up
1401 * using alternative, different, neigh objects to output
1402 * the packet in the output path. So what we need to do
1403 * here is re-lookup the top-level neigh in the path so
1404 * we can reinject the packet there.
1405 */
1406 n2 = NULL;
1407 if (dst && dst->obsolete != DST_OBSOLETE_DEAD) {
1408 n2 = dst_neigh_lookup_skb(dst, skb);
1409 if (n2)
1410 n1 = n2;
1411 }
1412 READ_ONCE(n1->output)(n1, skb);
1413 if (n2)
1414 neigh_release(n2);
1415 rcu_read_unlock();
1416
1417 write_lock_bh(&neigh->lock);
1418 }
1419 __skb_queue_purge(&neigh->arp_queue);
1420 neigh->arp_queue_len_bytes = 0;
1421 }
1422out:
1423 if (update_isrouter)
1424 neigh_update_is_router(neigh, flags, ¬ify);
1425 write_unlock_bh(&neigh->lock);
1426 if (((new ^ old) & NUD_PERMANENT) || gc_update)
1427 neigh_update_gc_list(neigh);
1428 if (managed_update)
1429 neigh_update_managed_list(neigh);
1430 if (notify)
1431 neigh_update_notify(neigh, nlmsg_pid);
1432 trace_neigh_update_done(neigh, err);
1433 return err;
1434}
1435
1436int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
1437 u32 flags, u32 nlmsg_pid)
1438{
1439 return __neigh_update(neigh, lladdr, new, flags, nlmsg_pid, NULL);
1440}
1441EXPORT_SYMBOL(neigh_update);
1442
1443/* Update the neigh to listen temporarily for probe responses, even if it is
1444 * in a NUD_FAILED state. The caller has to hold neigh->lock for writing.
1445 */
1446void __neigh_set_probe_once(struct neighbour *neigh)
1447{
1448 if (neigh->dead)
1449 return;
1450 neigh->updated = jiffies;
1451 if (!(neigh->nud_state & NUD_FAILED))
1452 return;
1453 WRITE_ONCE(neigh->nud_state, NUD_INCOMPLETE);
1454 atomic_set(&neigh->probes, neigh_max_probes(neigh));
1455 neigh_add_timer(neigh,
1456 jiffies + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
1457 HZ/100));
1458}
1459EXPORT_SYMBOL(__neigh_set_probe_once);
1460
1461struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1462 u8 *lladdr, void *saddr,
1463 struct net_device *dev)
1464{
1465 struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1466 lladdr || !dev->addr_len);
1467 if (neigh)
1468 neigh_update(neigh, lladdr, NUD_STALE,
1469 NEIGH_UPDATE_F_OVERRIDE, 0);
1470 return neigh;
1471}
1472EXPORT_SYMBOL(neigh_event_ns);
1473
1474/* called with read_lock_bh(&n->lock); */
1475static void neigh_hh_init(struct neighbour *n)
1476{
1477 struct net_device *dev = n->dev;
1478 __be16 prot = n->tbl->protocol;
1479 struct hh_cache *hh = &n->hh;
1480
1481 write_lock_bh(&n->lock);
1482
1483 /* Only one thread can come in here and initialize the
1484 * hh_cache entry.
1485 */
1486 if (!hh->hh_len)
1487 dev->header_ops->cache(n, hh, prot);
1488
1489 write_unlock_bh(&n->lock);
1490}
1491
1492/* Slow and careful. */
1493
1494int neigh_resolve_output(struct neighbour *neigh, struct sk_buff *skb)
1495{
1496 int rc = 0;
1497
1498 if (!neigh_event_send(neigh, skb)) {
1499 int err;
1500 struct net_device *dev = neigh->dev;
1501 unsigned int seq;
1502
1503 if (dev->header_ops->cache && !READ_ONCE(neigh->hh.hh_len))
1504 neigh_hh_init(neigh);
1505
1506 do {
1507 __skb_pull(skb, skb_network_offset(skb));
1508 seq = read_seqbegin(&neigh->ha_lock);
1509 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1510 neigh->ha, NULL, skb->len);
1511 } while (read_seqretry(&neigh->ha_lock, seq));
1512
1513 if (err >= 0)
1514 rc = dev_queue_xmit(skb);
1515 else
1516 goto out_kfree_skb;
1517 }
1518out:
1519 return rc;
1520out_kfree_skb:
1521 rc = -EINVAL;
1522 kfree_skb(skb);
1523 goto out;
1524}
1525EXPORT_SYMBOL(neigh_resolve_output);
1526
1527/* As fast as possible without hh cache */
1528
1529int neigh_connected_output(struct neighbour *neigh, struct sk_buff *skb)
1530{
1531 struct net_device *dev = neigh->dev;
1532 unsigned int seq;
1533 int err;
1534
1535 do {
1536 __skb_pull(skb, skb_network_offset(skb));
1537 seq = read_seqbegin(&neigh->ha_lock);
1538 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1539 neigh->ha, NULL, skb->len);
1540 } while (read_seqretry(&neigh->ha_lock, seq));
1541
1542 if (err >= 0)
1543 err = dev_queue_xmit(skb);
1544 else {
1545 err = -EINVAL;
1546 kfree_skb(skb);
1547 }
1548 return err;
1549}
1550EXPORT_SYMBOL(neigh_connected_output);
1551
1552int neigh_direct_output(struct neighbour *neigh, struct sk_buff *skb)
1553{
1554 return dev_queue_xmit(skb);
1555}
1556EXPORT_SYMBOL(neigh_direct_output);
1557
1558static void neigh_managed_work(struct work_struct *work)
1559{
1560 struct neigh_table *tbl = container_of(work, struct neigh_table,
1561 managed_work.work);
1562 struct neighbour *neigh;
1563
1564 write_lock_bh(&tbl->lock);
1565 list_for_each_entry(neigh, &tbl->managed_list, managed_list)
1566 neigh_event_send_probe(neigh, NULL, false);
1567 queue_delayed_work(system_power_efficient_wq, &tbl->managed_work,
1568 NEIGH_VAR(&tbl->parms, INTERVAL_PROBE_TIME_MS));
1569 write_unlock_bh(&tbl->lock);
1570}
1571
1572static void neigh_proxy_process(struct timer_list *t)
1573{
1574 struct neigh_table *tbl = from_timer(tbl, t, proxy_timer);
1575 long sched_next = 0;
1576 unsigned long now = jiffies;
1577 struct sk_buff *skb, *n;
1578
1579 spin_lock(&tbl->proxy_queue.lock);
1580
1581 skb_queue_walk_safe(&tbl->proxy_queue, skb, n) {
1582 long tdif = NEIGH_CB(skb)->sched_next - now;
1583
1584 if (tdif <= 0) {
1585 struct net_device *dev = skb->dev;
1586
1587 neigh_parms_qlen_dec(dev, tbl->family);
1588 __skb_unlink(skb, &tbl->proxy_queue);
1589
1590 if (tbl->proxy_redo && netif_running(dev)) {
1591 rcu_read_lock();
1592 tbl->proxy_redo(skb);
1593 rcu_read_unlock();
1594 } else {
1595 kfree_skb(skb);
1596 }
1597
1598 dev_put(dev);
1599 } else if (!sched_next || tdif < sched_next)
1600 sched_next = tdif;
1601 }
1602 del_timer(&tbl->proxy_timer);
1603 if (sched_next)
1604 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1605 spin_unlock(&tbl->proxy_queue.lock);
1606}
1607
1608static unsigned long neigh_proxy_delay(struct neigh_parms *p)
1609{
1610 /* If proxy_delay is zero, do not call get_random_u32_below()
1611 * as it is undefined behavior.
1612 */
1613 unsigned long proxy_delay = NEIGH_VAR(p, PROXY_DELAY);
1614
1615 return proxy_delay ?
1616 jiffies + get_random_u32_below(proxy_delay) : jiffies;
1617}
1618
1619void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1620 struct sk_buff *skb)
1621{
1622 unsigned long sched_next = neigh_proxy_delay(p);
1623
1624 if (p->qlen > NEIGH_VAR(p, PROXY_QLEN)) {
1625 kfree_skb(skb);
1626 return;
1627 }
1628
1629 NEIGH_CB(skb)->sched_next = sched_next;
1630 NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1631
1632 spin_lock(&tbl->proxy_queue.lock);
1633 if (del_timer(&tbl->proxy_timer)) {
1634 if (time_before(tbl->proxy_timer.expires, sched_next))
1635 sched_next = tbl->proxy_timer.expires;
1636 }
1637 skb_dst_drop(skb);
1638 dev_hold(skb->dev);
1639 __skb_queue_tail(&tbl->proxy_queue, skb);
1640 p->qlen++;
1641 mod_timer(&tbl->proxy_timer, sched_next);
1642 spin_unlock(&tbl->proxy_queue.lock);
1643}
1644EXPORT_SYMBOL(pneigh_enqueue);
1645
1646static inline struct neigh_parms *lookup_neigh_parms(struct neigh_table *tbl,
1647 struct net *net, int ifindex)
1648{
1649 struct neigh_parms *p;
1650
1651 list_for_each_entry(p, &tbl->parms_list, list) {
1652 if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
1653 (!p->dev && !ifindex && net_eq(net, &init_net)))
1654 return p;
1655 }
1656
1657 return NULL;
1658}
1659
1660struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1661 struct neigh_table *tbl)
1662{
1663 struct neigh_parms *p;
1664 struct net *net = dev_net(dev);
1665 const struct net_device_ops *ops = dev->netdev_ops;
1666
1667 p = kmemdup(&tbl->parms, sizeof(*p), GFP_KERNEL);
1668 if (p) {
1669 p->tbl = tbl;
1670 refcount_set(&p->refcnt, 1);
1671 p->reachable_time =
1672 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
1673 p->qlen = 0;
1674 netdev_hold(dev, &p->dev_tracker, GFP_KERNEL);
1675 p->dev = dev;
1676 write_pnet(&p->net, net);
1677 p->sysctl_table = NULL;
1678
1679 if (ops->ndo_neigh_setup && ops->ndo_neigh_setup(dev, p)) {
1680 netdev_put(dev, &p->dev_tracker);
1681 kfree(p);
1682 return NULL;
1683 }
1684
1685 write_lock_bh(&tbl->lock);
1686 list_add(&p->list, &tbl->parms.list);
1687 write_unlock_bh(&tbl->lock);
1688
1689 neigh_parms_data_state_cleanall(p);
1690 }
1691 return p;
1692}
1693EXPORT_SYMBOL(neigh_parms_alloc);
1694
1695static void neigh_rcu_free_parms(struct rcu_head *head)
1696{
1697 struct neigh_parms *parms =
1698 container_of(head, struct neigh_parms, rcu_head);
1699
1700 neigh_parms_put(parms);
1701}
1702
1703void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1704{
1705 if (!parms || parms == &tbl->parms)
1706 return;
1707 write_lock_bh(&tbl->lock);
1708 list_del(&parms->list);
1709 parms->dead = 1;
1710 write_unlock_bh(&tbl->lock);
1711 netdev_put(parms->dev, &parms->dev_tracker);
1712 call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1713}
1714EXPORT_SYMBOL(neigh_parms_release);
1715
1716static void neigh_parms_destroy(struct neigh_parms *parms)
1717{
1718 kfree(parms);
1719}
1720
1721static struct lock_class_key neigh_table_proxy_queue_class;
1722
1723static struct neigh_table __rcu *neigh_tables[NEIGH_NR_TABLES] __read_mostly;
1724
1725void neigh_table_init(int index, struct neigh_table *tbl)
1726{
1727 unsigned long now = jiffies;
1728 unsigned long phsize;
1729
1730 INIT_LIST_HEAD(&tbl->parms_list);
1731 INIT_LIST_HEAD(&tbl->gc_list);
1732 INIT_LIST_HEAD(&tbl->managed_list);
1733
1734 list_add(&tbl->parms.list, &tbl->parms_list);
1735 write_pnet(&tbl->parms.net, &init_net);
1736 refcount_set(&tbl->parms.refcnt, 1);
1737 tbl->parms.reachable_time =
1738 neigh_rand_reach_time(NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME));
1739 tbl->parms.qlen = 0;
1740
1741 tbl->stats = alloc_percpu(struct neigh_statistics);
1742 if (!tbl->stats)
1743 panic("cannot create neighbour cache statistics");
1744
1745#ifdef CONFIG_PROC_FS
1746 if (!proc_create_seq_data(tbl->id, 0, init_net.proc_net_stat,
1747 &neigh_stat_seq_ops, tbl))
1748 panic("cannot create neighbour proc dir entry");
1749#endif
1750
1751 RCU_INIT_POINTER(tbl->nht, neigh_hash_alloc(3));
1752
1753 phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1754 tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1755
1756 if (!tbl->nht || !tbl->phash_buckets)
1757 panic("cannot allocate neighbour cache hashes");
1758
1759 if (!tbl->entry_size)
1760 tbl->entry_size = ALIGN(offsetof(struct neighbour, primary_key) +
1761 tbl->key_len, NEIGH_PRIV_ALIGN);
1762 else
1763 WARN_ON(tbl->entry_size % NEIGH_PRIV_ALIGN);
1764
1765 rwlock_init(&tbl->lock);
1766
1767 INIT_DEFERRABLE_WORK(&tbl->gc_work, neigh_periodic_work);
1768 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
1769 tbl->parms.reachable_time);
1770 INIT_DEFERRABLE_WORK(&tbl->managed_work, neigh_managed_work);
1771 queue_delayed_work(system_power_efficient_wq, &tbl->managed_work, 0);
1772
1773 timer_setup(&tbl->proxy_timer, neigh_proxy_process, 0);
1774 skb_queue_head_init_class(&tbl->proxy_queue,
1775 &neigh_table_proxy_queue_class);
1776
1777 tbl->last_flush = now;
1778 tbl->last_rand = now + tbl->parms.reachable_time * 20;
1779
1780 rcu_assign_pointer(neigh_tables[index], tbl);
1781}
1782EXPORT_SYMBOL(neigh_table_init);
1783
1784/*
1785 * Only called from ndisc_cleanup(), which means this is dead code
1786 * because we no longer can unload IPv6 module.
1787 */
1788int neigh_table_clear(int index, struct neigh_table *tbl)
1789{
1790 RCU_INIT_POINTER(neigh_tables[index], NULL);
1791 synchronize_rcu();
1792
1793 /* It is not clean... Fix it to unload IPv6 module safely */
1794 cancel_delayed_work_sync(&tbl->managed_work);
1795 cancel_delayed_work_sync(&tbl->gc_work);
1796 del_timer_sync(&tbl->proxy_timer);
1797 pneigh_queue_purge(&tbl->proxy_queue, NULL, tbl->family);
1798 neigh_ifdown(tbl, NULL);
1799 if (atomic_read(&tbl->entries))
1800 pr_crit("neighbour leakage\n");
1801
1802 call_rcu(&rcu_dereference_protected(tbl->nht, 1)->rcu,
1803 neigh_hash_free_rcu);
1804 tbl->nht = NULL;
1805
1806 kfree(tbl->phash_buckets);
1807 tbl->phash_buckets = NULL;
1808
1809 remove_proc_entry(tbl->id, init_net.proc_net_stat);
1810
1811 free_percpu(tbl->stats);
1812 tbl->stats = NULL;
1813
1814 return 0;
1815}
1816EXPORT_SYMBOL(neigh_table_clear);
1817
1818static struct neigh_table *neigh_find_table(int family)
1819{
1820 struct neigh_table *tbl = NULL;
1821
1822 switch (family) {
1823 case AF_INET:
1824 tbl = rcu_dereference_rtnl(neigh_tables[NEIGH_ARP_TABLE]);
1825 break;
1826 case AF_INET6:
1827 tbl = rcu_dereference_rtnl(neigh_tables[NEIGH_ND_TABLE]);
1828 break;
1829 }
1830
1831 return tbl;
1832}
1833
1834const struct nla_policy nda_policy[NDA_MAX+1] = {
1835 [NDA_UNSPEC] = { .strict_start_type = NDA_NH_ID },
1836 [NDA_DST] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1837 [NDA_LLADDR] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1838 [NDA_CACHEINFO] = { .len = sizeof(struct nda_cacheinfo) },
1839 [NDA_PROBES] = { .type = NLA_U32 },
1840 [NDA_VLAN] = { .type = NLA_U16 },
1841 [NDA_PORT] = { .type = NLA_U16 },
1842 [NDA_VNI] = { .type = NLA_U32 },
1843 [NDA_IFINDEX] = { .type = NLA_U32 },
1844 [NDA_MASTER] = { .type = NLA_U32 },
1845 [NDA_PROTOCOL] = { .type = NLA_U8 },
1846 [NDA_NH_ID] = { .type = NLA_U32 },
1847 [NDA_FLAGS_EXT] = NLA_POLICY_MASK(NLA_U32, NTF_EXT_MASK),
1848 [NDA_FDB_EXT_ATTRS] = { .type = NLA_NESTED },
1849};
1850
1851static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh,
1852 struct netlink_ext_ack *extack)
1853{
1854 struct net *net = sock_net(skb->sk);
1855 struct ndmsg *ndm;
1856 struct nlattr *dst_attr;
1857 struct neigh_table *tbl;
1858 struct neighbour *neigh;
1859 struct net_device *dev = NULL;
1860 int err = -EINVAL;
1861
1862 ASSERT_RTNL();
1863 if (nlmsg_len(nlh) < sizeof(*ndm))
1864 goto out;
1865
1866 dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1867 if (!dst_attr) {
1868 NL_SET_ERR_MSG(extack, "Network address not specified");
1869 goto out;
1870 }
1871
1872 ndm = nlmsg_data(nlh);
1873 if (ndm->ndm_ifindex) {
1874 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1875 if (dev == NULL) {
1876 err = -ENODEV;
1877 goto out;
1878 }
1879 }
1880
1881 tbl = neigh_find_table(ndm->ndm_family);
1882 if (tbl == NULL)
1883 return -EAFNOSUPPORT;
1884
1885 if (nla_len(dst_attr) < (int)tbl->key_len) {
1886 NL_SET_ERR_MSG(extack, "Invalid network address");
1887 goto out;
1888 }
1889
1890 if (ndm->ndm_flags & NTF_PROXY) {
1891 err = pneigh_delete(tbl, net, nla_data(dst_attr), dev);
1892 goto out;
1893 }
1894
1895 if (dev == NULL)
1896 goto out;
1897
1898 neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1899 if (neigh == NULL) {
1900 err = -ENOENT;
1901 goto out;
1902 }
1903
1904 err = __neigh_update(neigh, NULL, NUD_FAILED,
1905 NEIGH_UPDATE_F_OVERRIDE | NEIGH_UPDATE_F_ADMIN,
1906 NETLINK_CB(skb).portid, extack);
1907 write_lock_bh(&tbl->lock);
1908 neigh_release(neigh);
1909 neigh_remove_one(neigh);
1910 write_unlock_bh(&tbl->lock);
1911
1912out:
1913 return err;
1914}
1915
1916static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh,
1917 struct netlink_ext_ack *extack)
1918{
1919 int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE |
1920 NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1921 struct net *net = sock_net(skb->sk);
1922 struct ndmsg *ndm;
1923 struct nlattr *tb[NDA_MAX+1];
1924 struct neigh_table *tbl;
1925 struct net_device *dev = NULL;
1926 struct neighbour *neigh;
1927 void *dst, *lladdr;
1928 u8 protocol = 0;
1929 u32 ndm_flags;
1930 int err;
1931
1932 ASSERT_RTNL();
1933 err = nlmsg_parse_deprecated(nlh, sizeof(*ndm), tb, NDA_MAX,
1934 nda_policy, extack);
1935 if (err < 0)
1936 goto out;
1937
1938 err = -EINVAL;
1939 if (!tb[NDA_DST]) {
1940 NL_SET_ERR_MSG(extack, "Network address not specified");
1941 goto out;
1942 }
1943
1944 ndm = nlmsg_data(nlh);
1945 ndm_flags = ndm->ndm_flags;
1946 if (tb[NDA_FLAGS_EXT]) {
1947 u32 ext = nla_get_u32(tb[NDA_FLAGS_EXT]);
1948
1949 BUILD_BUG_ON(sizeof(neigh->flags) * BITS_PER_BYTE <
1950 (sizeof(ndm->ndm_flags) * BITS_PER_BYTE +
1951 hweight32(NTF_EXT_MASK)));
1952 ndm_flags |= (ext << NTF_EXT_SHIFT);
1953 }
1954 if (ndm->ndm_ifindex) {
1955 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1956 if (dev == NULL) {
1957 err = -ENODEV;
1958 goto out;
1959 }
1960
1961 if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len) {
1962 NL_SET_ERR_MSG(extack, "Invalid link address");
1963 goto out;
1964 }
1965 }
1966
1967 tbl = neigh_find_table(ndm->ndm_family);
1968 if (tbl == NULL)
1969 return -EAFNOSUPPORT;
1970
1971 if (nla_len(tb[NDA_DST]) < (int)tbl->key_len) {
1972 NL_SET_ERR_MSG(extack, "Invalid network address");
1973 goto out;
1974 }
1975
1976 dst = nla_data(tb[NDA_DST]);
1977 lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
1978
1979 if (tb[NDA_PROTOCOL])
1980 protocol = nla_get_u8(tb[NDA_PROTOCOL]);
1981 if (ndm_flags & NTF_PROXY) {
1982 struct pneigh_entry *pn;
1983
1984 if (ndm_flags & NTF_MANAGED) {
1985 NL_SET_ERR_MSG(extack, "Invalid NTF_* flag combination");
1986 goto out;
1987 }
1988
1989 err = -ENOBUFS;
1990 pn = pneigh_lookup(tbl, net, dst, dev, 1);
1991 if (pn) {
1992 pn->flags = ndm_flags;
1993 if (protocol)
1994 pn->protocol = protocol;
1995 err = 0;
1996 }
1997 goto out;
1998 }
1999
2000 if (!dev) {
2001 NL_SET_ERR_MSG(extack, "Device not specified");
2002 goto out;
2003 }
2004
2005 if (tbl->allow_add && !tbl->allow_add(dev, extack)) {
2006 err = -EINVAL;
2007 goto out;
2008 }
2009
2010 neigh = neigh_lookup(tbl, dst, dev);
2011 if (neigh == NULL) {
2012 bool ndm_permanent = ndm->ndm_state & NUD_PERMANENT;
2013 bool exempt_from_gc = ndm_permanent ||
2014 ndm_flags & NTF_EXT_LEARNED;
2015
2016 if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
2017 err = -ENOENT;
2018 goto out;
2019 }
2020 if (ndm_permanent && (ndm_flags & NTF_MANAGED)) {
2021 NL_SET_ERR_MSG(extack, "Invalid NTF_* flag for permanent entry");
2022 err = -EINVAL;
2023 goto out;
2024 }
2025
2026 neigh = ___neigh_create(tbl, dst, dev,
2027 ndm_flags &
2028 (NTF_EXT_LEARNED | NTF_MANAGED),
2029 exempt_from_gc, true);
2030 if (IS_ERR(neigh)) {
2031 err = PTR_ERR(neigh);
2032 goto out;
2033 }
2034 } else {
2035 if (nlh->nlmsg_flags & NLM_F_EXCL) {
2036 err = -EEXIST;
2037 neigh_release(neigh);
2038 goto out;
2039 }
2040
2041 if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
2042 flags &= ~(NEIGH_UPDATE_F_OVERRIDE |
2043 NEIGH_UPDATE_F_OVERRIDE_ISROUTER);
2044 }
2045
2046 if (protocol)
2047 neigh->protocol = protocol;
2048 if (ndm_flags & NTF_EXT_LEARNED)
2049 flags |= NEIGH_UPDATE_F_EXT_LEARNED;
2050 if (ndm_flags & NTF_ROUTER)
2051 flags |= NEIGH_UPDATE_F_ISROUTER;
2052 if (ndm_flags & NTF_MANAGED)
2053 flags |= NEIGH_UPDATE_F_MANAGED;
2054 if (ndm_flags & NTF_USE)
2055 flags |= NEIGH_UPDATE_F_USE;
2056
2057 err = __neigh_update(neigh, lladdr, ndm->ndm_state, flags,
2058 NETLINK_CB(skb).portid, extack);
2059 if (!err && ndm_flags & (NTF_USE | NTF_MANAGED)) {
2060 neigh_event_send(neigh, NULL);
2061 err = 0;
2062 }
2063 neigh_release(neigh);
2064out:
2065 return err;
2066}
2067
2068static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
2069{
2070 struct nlattr *nest;
2071
2072 nest = nla_nest_start_noflag(skb, NDTA_PARMS);
2073 if (nest == NULL)
2074 return -ENOBUFS;
2075
2076 if ((parms->dev &&
2077 nla_put_u32(skb, NDTPA_IFINDEX, parms->dev->ifindex)) ||
2078 nla_put_u32(skb, NDTPA_REFCNT, refcount_read(&parms->refcnt)) ||
2079 nla_put_u32(skb, NDTPA_QUEUE_LENBYTES,
2080 NEIGH_VAR(parms, QUEUE_LEN_BYTES)) ||
2081 /* approximative value for deprecated QUEUE_LEN (in packets) */
2082 nla_put_u32(skb, NDTPA_QUEUE_LEN,
2083 NEIGH_VAR(parms, QUEUE_LEN_BYTES) / SKB_TRUESIZE(ETH_FRAME_LEN)) ||
2084 nla_put_u32(skb, NDTPA_PROXY_QLEN, NEIGH_VAR(parms, PROXY_QLEN)) ||
2085 nla_put_u32(skb, NDTPA_APP_PROBES, NEIGH_VAR(parms, APP_PROBES)) ||
2086 nla_put_u32(skb, NDTPA_UCAST_PROBES,
2087 NEIGH_VAR(parms, UCAST_PROBES)) ||
2088 nla_put_u32(skb, NDTPA_MCAST_PROBES,
2089 NEIGH_VAR(parms, MCAST_PROBES)) ||
2090 nla_put_u32(skb, NDTPA_MCAST_REPROBES,
2091 NEIGH_VAR(parms, MCAST_REPROBES)) ||
2092 nla_put_msecs(skb, NDTPA_REACHABLE_TIME, parms->reachable_time,
2093 NDTPA_PAD) ||
2094 nla_put_msecs(skb, NDTPA_BASE_REACHABLE_TIME,
2095 NEIGH_VAR(parms, BASE_REACHABLE_TIME), NDTPA_PAD) ||
2096 nla_put_msecs(skb, NDTPA_GC_STALETIME,
2097 NEIGH_VAR(parms, GC_STALETIME), NDTPA_PAD) ||
2098 nla_put_msecs(skb, NDTPA_DELAY_PROBE_TIME,
2099 NEIGH_VAR(parms, DELAY_PROBE_TIME), NDTPA_PAD) ||
2100 nla_put_msecs(skb, NDTPA_RETRANS_TIME,
2101 NEIGH_VAR(parms, RETRANS_TIME), NDTPA_PAD) ||
2102 nla_put_msecs(skb, NDTPA_ANYCAST_DELAY,
2103 NEIGH_VAR(parms, ANYCAST_DELAY), NDTPA_PAD) ||
2104 nla_put_msecs(skb, NDTPA_PROXY_DELAY,
2105 NEIGH_VAR(parms, PROXY_DELAY), NDTPA_PAD) ||
2106 nla_put_msecs(skb, NDTPA_LOCKTIME,
2107 NEIGH_VAR(parms, LOCKTIME), NDTPA_PAD) ||
2108 nla_put_msecs(skb, NDTPA_INTERVAL_PROBE_TIME_MS,
2109 NEIGH_VAR(parms, INTERVAL_PROBE_TIME_MS), NDTPA_PAD))
2110 goto nla_put_failure;
2111 return nla_nest_end(skb, nest);
2112
2113nla_put_failure:
2114 nla_nest_cancel(skb, nest);
2115 return -EMSGSIZE;
2116}
2117
2118static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl,
2119 u32 pid, u32 seq, int type, int flags)
2120{
2121 struct nlmsghdr *nlh;
2122 struct ndtmsg *ndtmsg;
2123
2124 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
2125 if (nlh == NULL)
2126 return -EMSGSIZE;
2127
2128 ndtmsg = nlmsg_data(nlh);
2129
2130 read_lock_bh(&tbl->lock);
2131 ndtmsg->ndtm_family = tbl->family;
2132 ndtmsg->ndtm_pad1 = 0;
2133 ndtmsg->ndtm_pad2 = 0;
2134
2135 if (nla_put_string(skb, NDTA_NAME, tbl->id) ||
2136 nla_put_msecs(skb, NDTA_GC_INTERVAL, READ_ONCE(tbl->gc_interval),
2137 NDTA_PAD) ||
2138 nla_put_u32(skb, NDTA_THRESH1, READ_ONCE(tbl->gc_thresh1)) ||
2139 nla_put_u32(skb, NDTA_THRESH2, READ_ONCE(tbl->gc_thresh2)) ||
2140 nla_put_u32(skb, NDTA_THRESH3, READ_ONCE(tbl->gc_thresh3)))
2141 goto nla_put_failure;
2142 {
2143 unsigned long now = jiffies;
2144 long flush_delta = now - READ_ONCE(tbl->last_flush);
2145 long rand_delta = now - READ_ONCE(tbl->last_rand);
2146 struct neigh_hash_table *nht;
2147 struct ndt_config ndc = {
2148 .ndtc_key_len = tbl->key_len,
2149 .ndtc_entry_size = tbl->entry_size,
2150 .ndtc_entries = atomic_read(&tbl->entries),
2151 .ndtc_last_flush = jiffies_to_msecs(flush_delta),
2152 .ndtc_last_rand = jiffies_to_msecs(rand_delta),
2153 .ndtc_proxy_qlen = READ_ONCE(tbl->proxy_queue.qlen),
2154 };
2155
2156 rcu_read_lock();
2157 nht = rcu_dereference(tbl->nht);
2158 ndc.ndtc_hash_rnd = nht->hash_rnd[0];
2159 ndc.ndtc_hash_mask = ((1 << nht->hash_shift) - 1);
2160 rcu_read_unlock();
2161
2162 if (nla_put(skb, NDTA_CONFIG, sizeof(ndc), &ndc))
2163 goto nla_put_failure;
2164 }
2165
2166 {
2167 int cpu;
2168 struct ndt_stats ndst;
2169
2170 memset(&ndst, 0, sizeof(ndst));
2171
2172 for_each_possible_cpu(cpu) {
2173 struct neigh_statistics *st;
2174
2175 st = per_cpu_ptr(tbl->stats, cpu);
2176 ndst.ndts_allocs += READ_ONCE(st->allocs);
2177 ndst.ndts_destroys += READ_ONCE(st->destroys);
2178 ndst.ndts_hash_grows += READ_ONCE(st->hash_grows);
2179 ndst.ndts_res_failed += READ_ONCE(st->res_failed);
2180 ndst.ndts_lookups += READ_ONCE(st->lookups);
2181 ndst.ndts_hits += READ_ONCE(st->hits);
2182 ndst.ndts_rcv_probes_mcast += READ_ONCE(st->rcv_probes_mcast);
2183 ndst.ndts_rcv_probes_ucast += READ_ONCE(st->rcv_probes_ucast);
2184 ndst.ndts_periodic_gc_runs += READ_ONCE(st->periodic_gc_runs);
2185 ndst.ndts_forced_gc_runs += READ_ONCE(st->forced_gc_runs);
2186 ndst.ndts_table_fulls += READ_ONCE(st->table_fulls);
2187 }
2188
2189 if (nla_put_64bit(skb, NDTA_STATS, sizeof(ndst), &ndst,
2190 NDTA_PAD))
2191 goto nla_put_failure;
2192 }
2193
2194 BUG_ON(tbl->parms.dev);
2195 if (neightbl_fill_parms(skb, &tbl->parms) < 0)
2196 goto nla_put_failure;
2197
2198 read_unlock_bh(&tbl->lock);
2199 nlmsg_end(skb, nlh);
2200 return 0;
2201
2202nla_put_failure:
2203 read_unlock_bh(&tbl->lock);
2204 nlmsg_cancel(skb, nlh);
2205 return -EMSGSIZE;
2206}
2207
2208static int neightbl_fill_param_info(struct sk_buff *skb,
2209 struct neigh_table *tbl,
2210 struct neigh_parms *parms,
2211 u32 pid, u32 seq, int type,
2212 unsigned int flags)
2213{
2214 struct ndtmsg *ndtmsg;
2215 struct nlmsghdr *nlh;
2216
2217 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
2218 if (nlh == NULL)
2219 return -EMSGSIZE;
2220
2221 ndtmsg = nlmsg_data(nlh);
2222
2223 read_lock_bh(&tbl->lock);
2224 ndtmsg->ndtm_family = tbl->family;
2225 ndtmsg->ndtm_pad1 = 0;
2226 ndtmsg->ndtm_pad2 = 0;
2227
2228 if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 ||
2229 neightbl_fill_parms(skb, parms) < 0)
2230 goto errout;
2231
2232 read_unlock_bh(&tbl->lock);
2233 nlmsg_end(skb, nlh);
2234 return 0;
2235errout:
2236 read_unlock_bh(&tbl->lock);
2237 nlmsg_cancel(skb, nlh);
2238 return -EMSGSIZE;
2239}
2240
2241static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = {
2242 [NDTA_NAME] = { .type = NLA_STRING },
2243 [NDTA_THRESH1] = { .type = NLA_U32 },
2244 [NDTA_THRESH2] = { .type = NLA_U32 },
2245 [NDTA_THRESH3] = { .type = NLA_U32 },
2246 [NDTA_GC_INTERVAL] = { .type = NLA_U64 },
2247 [NDTA_PARMS] = { .type = NLA_NESTED },
2248};
2249
2250static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = {
2251 [NDTPA_IFINDEX] = { .type = NLA_U32 },
2252 [NDTPA_QUEUE_LEN] = { .type = NLA_U32 },
2253 [NDTPA_PROXY_QLEN] = { .type = NLA_U32 },
2254 [NDTPA_APP_PROBES] = { .type = NLA_U32 },
2255 [NDTPA_UCAST_PROBES] = { .type = NLA_U32 },
2256 [NDTPA_MCAST_PROBES] = { .type = NLA_U32 },
2257 [NDTPA_MCAST_REPROBES] = { .type = NLA_U32 },
2258 [NDTPA_BASE_REACHABLE_TIME] = { .type = NLA_U64 },
2259 [NDTPA_GC_STALETIME] = { .type = NLA_U64 },
2260 [NDTPA_DELAY_PROBE_TIME] = { .type = NLA_U64 },
2261 [NDTPA_RETRANS_TIME] = { .type = NLA_U64 },
2262 [NDTPA_ANYCAST_DELAY] = { .type = NLA_U64 },
2263 [NDTPA_PROXY_DELAY] = { .type = NLA_U64 },
2264 [NDTPA_LOCKTIME] = { .type = NLA_U64 },
2265 [NDTPA_INTERVAL_PROBE_TIME_MS] = { .type = NLA_U64, .min = 1 },
2266};
2267
2268static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh,
2269 struct netlink_ext_ack *extack)
2270{
2271 struct net *net = sock_net(skb->sk);
2272 struct neigh_table *tbl;
2273 struct ndtmsg *ndtmsg;
2274 struct nlattr *tb[NDTA_MAX+1];
2275 bool found = false;
2276 int err, tidx;
2277
2278 err = nlmsg_parse_deprecated(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
2279 nl_neightbl_policy, extack);
2280 if (err < 0)
2281 goto errout;
2282
2283 if (tb[NDTA_NAME] == NULL) {
2284 err = -EINVAL;
2285 goto errout;
2286 }
2287
2288 ndtmsg = nlmsg_data(nlh);
2289
2290 for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
2291 tbl = rcu_dereference_rtnl(neigh_tables[tidx]);
2292 if (!tbl)
2293 continue;
2294 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
2295 continue;
2296 if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0) {
2297 found = true;
2298 break;
2299 }
2300 }
2301
2302 if (!found)
2303 return -ENOENT;
2304
2305 /*
2306 * We acquire tbl->lock to be nice to the periodic timers and
2307 * make sure they always see a consistent set of values.
2308 */
2309 write_lock_bh(&tbl->lock);
2310
2311 if (tb[NDTA_PARMS]) {
2312 struct nlattr *tbp[NDTPA_MAX+1];
2313 struct neigh_parms *p;
2314 int i, ifindex = 0;
2315
2316 err = nla_parse_nested_deprecated(tbp, NDTPA_MAX,
2317 tb[NDTA_PARMS],
2318 nl_ntbl_parm_policy, extack);
2319 if (err < 0)
2320 goto errout_tbl_lock;
2321
2322 if (tbp[NDTPA_IFINDEX])
2323 ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
2324
2325 p = lookup_neigh_parms(tbl, net, ifindex);
2326 if (p == NULL) {
2327 err = -ENOENT;
2328 goto errout_tbl_lock;
2329 }
2330
2331 for (i = 1; i <= NDTPA_MAX; i++) {
2332 if (tbp[i] == NULL)
2333 continue;
2334
2335 switch (i) {
2336 case NDTPA_QUEUE_LEN:
2337 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2338 nla_get_u32(tbp[i]) *
2339 SKB_TRUESIZE(ETH_FRAME_LEN));
2340 break;
2341 case NDTPA_QUEUE_LENBYTES:
2342 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2343 nla_get_u32(tbp[i]));
2344 break;
2345 case NDTPA_PROXY_QLEN:
2346 NEIGH_VAR_SET(p, PROXY_QLEN,
2347 nla_get_u32(tbp[i]));
2348 break;
2349 case NDTPA_APP_PROBES:
2350 NEIGH_VAR_SET(p, APP_PROBES,
2351 nla_get_u32(tbp[i]));
2352 break;
2353 case NDTPA_UCAST_PROBES:
2354 NEIGH_VAR_SET(p, UCAST_PROBES,
2355 nla_get_u32(tbp[i]));
2356 break;
2357 case NDTPA_MCAST_PROBES:
2358 NEIGH_VAR_SET(p, MCAST_PROBES,
2359 nla_get_u32(tbp[i]));
2360 break;
2361 case NDTPA_MCAST_REPROBES:
2362 NEIGH_VAR_SET(p, MCAST_REPROBES,
2363 nla_get_u32(tbp[i]));
2364 break;
2365 case NDTPA_BASE_REACHABLE_TIME:
2366 NEIGH_VAR_SET(p, BASE_REACHABLE_TIME,
2367 nla_get_msecs(tbp[i]));
2368 /* update reachable_time as well, otherwise, the change will
2369 * only be effective after the next time neigh_periodic_work
2370 * decides to recompute it (can be multiple minutes)
2371 */
2372 p->reachable_time =
2373 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
2374 break;
2375 case NDTPA_GC_STALETIME:
2376 NEIGH_VAR_SET(p, GC_STALETIME,
2377 nla_get_msecs(tbp[i]));
2378 break;
2379 case NDTPA_DELAY_PROBE_TIME:
2380 NEIGH_VAR_SET(p, DELAY_PROBE_TIME,
2381 nla_get_msecs(tbp[i]));
2382 call_netevent_notifiers(NETEVENT_DELAY_PROBE_TIME_UPDATE, p);
2383 break;
2384 case NDTPA_INTERVAL_PROBE_TIME_MS:
2385 NEIGH_VAR_SET(p, INTERVAL_PROBE_TIME_MS,
2386 nla_get_msecs(tbp[i]));
2387 break;
2388 case NDTPA_RETRANS_TIME:
2389 NEIGH_VAR_SET(p, RETRANS_TIME,
2390 nla_get_msecs(tbp[i]));
2391 break;
2392 case NDTPA_ANYCAST_DELAY:
2393 NEIGH_VAR_SET(p, ANYCAST_DELAY,
2394 nla_get_msecs(tbp[i]));
2395 break;
2396 case NDTPA_PROXY_DELAY:
2397 NEIGH_VAR_SET(p, PROXY_DELAY,
2398 nla_get_msecs(tbp[i]));
2399 break;
2400 case NDTPA_LOCKTIME:
2401 NEIGH_VAR_SET(p, LOCKTIME,
2402 nla_get_msecs(tbp[i]));
2403 break;
2404 }
2405 }
2406 }
2407
2408 err = -ENOENT;
2409 if ((tb[NDTA_THRESH1] || tb[NDTA_THRESH2] ||
2410 tb[NDTA_THRESH3] || tb[NDTA_GC_INTERVAL]) &&
2411 !net_eq(net, &init_net))
2412 goto errout_tbl_lock;
2413
2414 if (tb[NDTA_THRESH1])
2415 WRITE_ONCE(tbl->gc_thresh1, nla_get_u32(tb[NDTA_THRESH1]));
2416
2417 if (tb[NDTA_THRESH2])
2418 WRITE_ONCE(tbl->gc_thresh2, nla_get_u32(tb[NDTA_THRESH2]));
2419
2420 if (tb[NDTA_THRESH3])
2421 WRITE_ONCE(tbl->gc_thresh3, nla_get_u32(tb[NDTA_THRESH3]));
2422
2423 if (tb[NDTA_GC_INTERVAL])
2424 WRITE_ONCE(tbl->gc_interval, nla_get_msecs(tb[NDTA_GC_INTERVAL]));
2425
2426 err = 0;
2427
2428errout_tbl_lock:
2429 write_unlock_bh(&tbl->lock);
2430errout:
2431 return err;
2432}
2433
2434static int neightbl_valid_dump_info(const struct nlmsghdr *nlh,
2435 struct netlink_ext_ack *extack)
2436{
2437 struct ndtmsg *ndtm;
2438
2439 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndtm))) {
2440 NL_SET_ERR_MSG(extack, "Invalid header for neighbor table dump request");
2441 return -EINVAL;
2442 }
2443
2444 ndtm = nlmsg_data(nlh);
2445 if (ndtm->ndtm_pad1 || ndtm->ndtm_pad2) {
2446 NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor table dump request");
2447 return -EINVAL;
2448 }
2449
2450 if (nlmsg_attrlen(nlh, sizeof(*ndtm))) {
2451 NL_SET_ERR_MSG(extack, "Invalid data after header in neighbor table dump request");
2452 return -EINVAL;
2453 }
2454
2455 return 0;
2456}
2457
2458static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2459{
2460 const struct nlmsghdr *nlh = cb->nlh;
2461 struct net *net = sock_net(skb->sk);
2462 int family, tidx, nidx = 0;
2463 int tbl_skip = cb->args[0];
2464 int neigh_skip = cb->args[1];
2465 struct neigh_table *tbl;
2466
2467 if (cb->strict_check) {
2468 int err = neightbl_valid_dump_info(nlh, cb->extack);
2469
2470 if (err < 0)
2471 return err;
2472 }
2473
2474 family = ((struct rtgenmsg *)nlmsg_data(nlh))->rtgen_family;
2475
2476 for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
2477 struct neigh_parms *p;
2478
2479 tbl = rcu_dereference_rtnl(neigh_tables[tidx]);
2480 if (!tbl)
2481 continue;
2482
2483 if (tidx < tbl_skip || (family && tbl->family != family))
2484 continue;
2485
2486 if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).portid,
2487 nlh->nlmsg_seq, RTM_NEWNEIGHTBL,
2488 NLM_F_MULTI) < 0)
2489 break;
2490
2491 nidx = 0;
2492 p = list_next_entry(&tbl->parms, list);
2493 list_for_each_entry_from(p, &tbl->parms_list, list) {
2494 if (!net_eq(neigh_parms_net(p), net))
2495 continue;
2496
2497 if (nidx < neigh_skip)
2498 goto next;
2499
2500 if (neightbl_fill_param_info(skb, tbl, p,
2501 NETLINK_CB(cb->skb).portid,
2502 nlh->nlmsg_seq,
2503 RTM_NEWNEIGHTBL,
2504 NLM_F_MULTI) < 0)
2505 goto out;
2506 next:
2507 nidx++;
2508 }
2509
2510 neigh_skip = 0;
2511 }
2512out:
2513 cb->args[0] = tidx;
2514 cb->args[1] = nidx;
2515
2516 return skb->len;
2517}
2518
2519static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
2520 u32 pid, u32 seq, int type, unsigned int flags)
2521{
2522 u32 neigh_flags, neigh_flags_ext;
2523 unsigned long now = jiffies;
2524 struct nda_cacheinfo ci;
2525 struct nlmsghdr *nlh;
2526 struct ndmsg *ndm;
2527
2528 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2529 if (nlh == NULL)
2530 return -EMSGSIZE;
2531
2532 neigh_flags_ext = neigh->flags >> NTF_EXT_SHIFT;
2533 neigh_flags = neigh->flags & NTF_OLD_MASK;
2534
2535 ndm = nlmsg_data(nlh);
2536 ndm->ndm_family = neigh->ops->family;
2537 ndm->ndm_pad1 = 0;
2538 ndm->ndm_pad2 = 0;
2539 ndm->ndm_flags = neigh_flags;
2540 ndm->ndm_type = neigh->type;
2541 ndm->ndm_ifindex = neigh->dev->ifindex;
2542
2543 if (nla_put(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key))
2544 goto nla_put_failure;
2545
2546 read_lock_bh(&neigh->lock);
2547 ndm->ndm_state = neigh->nud_state;
2548 if (neigh->nud_state & NUD_VALID) {
2549 char haddr[MAX_ADDR_LEN];
2550
2551 neigh_ha_snapshot(haddr, neigh, neigh->dev);
2552 if (nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, haddr) < 0) {
2553 read_unlock_bh(&neigh->lock);
2554 goto nla_put_failure;
2555 }
2556 }
2557
2558 ci.ndm_used = jiffies_to_clock_t(now - neigh->used);
2559 ci.ndm_confirmed = jiffies_to_clock_t(now - neigh->confirmed);
2560 ci.ndm_updated = jiffies_to_clock_t(now - neigh->updated);
2561 ci.ndm_refcnt = refcount_read(&neigh->refcnt) - 1;
2562 read_unlock_bh(&neigh->lock);
2563
2564 if (nla_put_u32(skb, NDA_PROBES, atomic_read(&neigh->probes)) ||
2565 nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
2566 goto nla_put_failure;
2567
2568 if (neigh->protocol && nla_put_u8(skb, NDA_PROTOCOL, neigh->protocol))
2569 goto nla_put_failure;
2570 if (neigh_flags_ext && nla_put_u32(skb, NDA_FLAGS_EXT, neigh_flags_ext))
2571 goto nla_put_failure;
2572
2573 nlmsg_end(skb, nlh);
2574 return 0;
2575
2576nla_put_failure:
2577 nlmsg_cancel(skb, nlh);
2578 return -EMSGSIZE;
2579}
2580
2581static int pneigh_fill_info(struct sk_buff *skb, struct pneigh_entry *pn,
2582 u32 pid, u32 seq, int type, unsigned int flags,
2583 struct neigh_table *tbl)
2584{
2585 u32 neigh_flags, neigh_flags_ext;
2586 struct nlmsghdr *nlh;
2587 struct ndmsg *ndm;
2588
2589 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2590 if (nlh == NULL)
2591 return -EMSGSIZE;
2592
2593 neigh_flags_ext = pn->flags >> NTF_EXT_SHIFT;
2594 neigh_flags = pn->flags & NTF_OLD_MASK;
2595
2596 ndm = nlmsg_data(nlh);
2597 ndm->ndm_family = tbl->family;
2598 ndm->ndm_pad1 = 0;
2599 ndm->ndm_pad2 = 0;
2600 ndm->ndm_flags = neigh_flags | NTF_PROXY;
2601 ndm->ndm_type = RTN_UNICAST;
2602 ndm->ndm_ifindex = pn->dev ? pn->dev->ifindex : 0;
2603 ndm->ndm_state = NUD_NONE;
2604
2605 if (nla_put(skb, NDA_DST, tbl->key_len, pn->key))
2606 goto nla_put_failure;
2607
2608 if (pn->protocol && nla_put_u8(skb, NDA_PROTOCOL, pn->protocol))
2609 goto nla_put_failure;
2610 if (neigh_flags_ext && nla_put_u32(skb, NDA_FLAGS_EXT, neigh_flags_ext))
2611 goto nla_put_failure;
2612
2613 nlmsg_end(skb, nlh);
2614 return 0;
2615
2616nla_put_failure:
2617 nlmsg_cancel(skb, nlh);
2618 return -EMSGSIZE;
2619}
2620
2621static void neigh_update_notify(struct neighbour *neigh, u32 nlmsg_pid)
2622{
2623 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
2624 __neigh_notify(neigh, RTM_NEWNEIGH, 0, nlmsg_pid);
2625}
2626
2627static bool neigh_master_filtered(struct net_device *dev, int master_idx)
2628{
2629 struct net_device *master;
2630
2631 if (!master_idx)
2632 return false;
2633
2634 master = dev ? netdev_master_upper_dev_get_rcu(dev) : NULL;
2635
2636 /* 0 is already used to denote NDA_MASTER wasn't passed, therefore need another
2637 * invalid value for ifindex to denote "no master".
2638 */
2639 if (master_idx == -1)
2640 return !!master;
2641
2642 if (!master || master->ifindex != master_idx)
2643 return true;
2644
2645 return false;
2646}
2647
2648static bool neigh_ifindex_filtered(struct net_device *dev, int filter_idx)
2649{
2650 if (filter_idx && (!dev || dev->ifindex != filter_idx))
2651 return true;
2652
2653 return false;
2654}
2655
2656struct neigh_dump_filter {
2657 int master_idx;
2658 int dev_idx;
2659};
2660
2661static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2662 struct netlink_callback *cb,
2663 struct neigh_dump_filter *filter)
2664{
2665 struct net *net = sock_net(skb->sk);
2666 struct neighbour *n;
2667 int err = 0, h, s_h = cb->args[1];
2668 int idx, s_idx = idx = cb->args[2];
2669 struct neigh_hash_table *nht;
2670 unsigned int flags = NLM_F_MULTI;
2671
2672 if (filter->dev_idx || filter->master_idx)
2673 flags |= NLM_F_DUMP_FILTERED;
2674
2675 nht = rcu_dereference(tbl->nht);
2676
2677 for (h = s_h; h < (1 << nht->hash_shift); h++) {
2678 if (h > s_h)
2679 s_idx = 0;
2680 idx = 0;
2681 neigh_for_each_in_bucket_rcu(n, &nht->hash_heads[h]) {
2682 if (idx < s_idx || !net_eq(dev_net(n->dev), net))
2683 goto next;
2684 if (neigh_ifindex_filtered(n->dev, filter->dev_idx) ||
2685 neigh_master_filtered(n->dev, filter->master_idx))
2686 goto next;
2687 err = neigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2688 cb->nlh->nlmsg_seq,
2689 RTM_NEWNEIGH, flags);
2690 if (err < 0)
2691 goto out;
2692next:
2693 idx++;
2694 }
2695 }
2696out:
2697 cb->args[1] = h;
2698 cb->args[2] = idx;
2699 return err;
2700}
2701
2702static int pneigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2703 struct netlink_callback *cb,
2704 struct neigh_dump_filter *filter)
2705{
2706 struct pneigh_entry *n;
2707 struct net *net = sock_net(skb->sk);
2708 int err = 0, h, s_h = cb->args[3];
2709 int idx, s_idx = idx = cb->args[4];
2710 unsigned int flags = NLM_F_MULTI;
2711
2712 if (filter->dev_idx || filter->master_idx)
2713 flags |= NLM_F_DUMP_FILTERED;
2714
2715 read_lock_bh(&tbl->lock);
2716
2717 for (h = s_h; h <= PNEIGH_HASHMASK; h++) {
2718 if (h > s_h)
2719 s_idx = 0;
2720 for (n = tbl->phash_buckets[h], idx = 0; n; n = n->next) {
2721 if (idx < s_idx || pneigh_net(n) != net)
2722 goto next;
2723 if (neigh_ifindex_filtered(n->dev, filter->dev_idx) ||
2724 neigh_master_filtered(n->dev, filter->master_idx))
2725 goto next;
2726 err = pneigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2727 cb->nlh->nlmsg_seq,
2728 RTM_NEWNEIGH, flags, tbl);
2729 if (err < 0) {
2730 read_unlock_bh(&tbl->lock);
2731 goto out;
2732 }
2733 next:
2734 idx++;
2735 }
2736 }
2737
2738 read_unlock_bh(&tbl->lock);
2739out:
2740 cb->args[3] = h;
2741 cb->args[4] = idx;
2742 return err;
2743}
2744
2745static int neigh_valid_dump_req(const struct nlmsghdr *nlh,
2746 bool strict_check,
2747 struct neigh_dump_filter *filter,
2748 struct netlink_ext_ack *extack)
2749{
2750 struct nlattr *tb[NDA_MAX + 1];
2751 int err, i;
2752
2753 if (strict_check) {
2754 struct ndmsg *ndm;
2755
2756 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndm))) {
2757 NL_SET_ERR_MSG(extack, "Invalid header for neighbor dump request");
2758 return -EINVAL;
2759 }
2760
2761 ndm = nlmsg_data(nlh);
2762 if (ndm->ndm_pad1 || ndm->ndm_pad2 || ndm->ndm_ifindex ||
2763 ndm->ndm_state || ndm->ndm_type) {
2764 NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor dump request");
2765 return -EINVAL;
2766 }
2767
2768 if (ndm->ndm_flags & ~NTF_PROXY) {
2769 NL_SET_ERR_MSG(extack, "Invalid flags in header for neighbor dump request");
2770 return -EINVAL;
2771 }
2772
2773 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct ndmsg),
2774 tb, NDA_MAX, nda_policy,
2775 extack);
2776 } else {
2777 err = nlmsg_parse_deprecated(nlh, sizeof(struct ndmsg), tb,
2778 NDA_MAX, nda_policy, extack);
2779 }
2780 if (err < 0)
2781 return err;
2782
2783 for (i = 0; i <= NDA_MAX; ++i) {
2784 if (!tb[i])
2785 continue;
2786
2787 /* all new attributes should require strict_check */
2788 switch (i) {
2789 case NDA_IFINDEX:
2790 filter->dev_idx = nla_get_u32(tb[i]);
2791 break;
2792 case NDA_MASTER:
2793 filter->master_idx = nla_get_u32(tb[i]);
2794 break;
2795 default:
2796 if (strict_check) {
2797 NL_SET_ERR_MSG(extack, "Unsupported attribute in neighbor dump request");
2798 return -EINVAL;
2799 }
2800 }
2801 }
2802
2803 return 0;
2804}
2805
2806static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2807{
2808 const struct nlmsghdr *nlh = cb->nlh;
2809 struct neigh_dump_filter filter = {};
2810 struct neigh_table *tbl;
2811 int t, family, s_t;
2812 int proxy = 0;
2813 int err;
2814
2815 family = ((struct rtgenmsg *)nlmsg_data(nlh))->rtgen_family;
2816
2817 /* check for full ndmsg structure presence, family member is
2818 * the same for both structures
2819 */
2820 if (nlmsg_len(nlh) >= sizeof(struct ndmsg) &&
2821 ((struct ndmsg *)nlmsg_data(nlh))->ndm_flags == NTF_PROXY)
2822 proxy = 1;
2823
2824 err = neigh_valid_dump_req(nlh, cb->strict_check, &filter, cb->extack);
2825 if (err < 0 && cb->strict_check)
2826 return err;
2827 err = 0;
2828
2829 s_t = cb->args[0];
2830
2831 rcu_read_lock();
2832 for (t = 0; t < NEIGH_NR_TABLES; t++) {
2833 tbl = rcu_dereference(neigh_tables[t]);
2834
2835 if (!tbl)
2836 continue;
2837 if (t < s_t || (family && tbl->family != family))
2838 continue;
2839 if (t > s_t)
2840 memset(&cb->args[1], 0, sizeof(cb->args) -
2841 sizeof(cb->args[0]));
2842 if (proxy)
2843 err = pneigh_dump_table(tbl, skb, cb, &filter);
2844 else
2845 err = neigh_dump_table(tbl, skb, cb, &filter);
2846 if (err < 0)
2847 break;
2848 }
2849 rcu_read_unlock();
2850
2851 cb->args[0] = t;
2852 return err;
2853}
2854
2855static int neigh_valid_get_req(const struct nlmsghdr *nlh,
2856 struct neigh_table **tbl,
2857 void **dst, int *dev_idx, u8 *ndm_flags,
2858 struct netlink_ext_ack *extack)
2859{
2860 struct nlattr *tb[NDA_MAX + 1];
2861 struct ndmsg *ndm;
2862 int err, i;
2863
2864 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndm))) {
2865 NL_SET_ERR_MSG(extack, "Invalid header for neighbor get request");
2866 return -EINVAL;
2867 }
2868
2869 ndm = nlmsg_data(nlh);
2870 if (ndm->ndm_pad1 || ndm->ndm_pad2 || ndm->ndm_state ||
2871 ndm->ndm_type) {
2872 NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor get request");
2873 return -EINVAL;
2874 }
2875
2876 if (ndm->ndm_flags & ~NTF_PROXY) {
2877 NL_SET_ERR_MSG(extack, "Invalid flags in header for neighbor get request");
2878 return -EINVAL;
2879 }
2880
2881 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct ndmsg), tb,
2882 NDA_MAX, nda_policy, extack);
2883 if (err < 0)
2884 return err;
2885
2886 *ndm_flags = ndm->ndm_flags;
2887 *dev_idx = ndm->ndm_ifindex;
2888 *tbl = neigh_find_table(ndm->ndm_family);
2889 if (*tbl == NULL) {
2890 NL_SET_ERR_MSG(extack, "Unsupported family in header for neighbor get request");
2891 return -EAFNOSUPPORT;
2892 }
2893
2894 for (i = 0; i <= NDA_MAX; ++i) {
2895 if (!tb[i])
2896 continue;
2897
2898 switch (i) {
2899 case NDA_DST:
2900 if (nla_len(tb[i]) != (int)(*tbl)->key_len) {
2901 NL_SET_ERR_MSG(extack, "Invalid network address in neighbor get request");
2902 return -EINVAL;
2903 }
2904 *dst = nla_data(tb[i]);
2905 break;
2906 default:
2907 NL_SET_ERR_MSG(extack, "Unsupported attribute in neighbor get request");
2908 return -EINVAL;
2909 }
2910 }
2911
2912 return 0;
2913}
2914
2915static inline size_t neigh_nlmsg_size(void)
2916{
2917 return NLMSG_ALIGN(sizeof(struct ndmsg))
2918 + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2919 + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */
2920 + nla_total_size(sizeof(struct nda_cacheinfo))
2921 + nla_total_size(4) /* NDA_PROBES */
2922 + nla_total_size(4) /* NDA_FLAGS_EXT */
2923 + nla_total_size(1); /* NDA_PROTOCOL */
2924}
2925
2926static int neigh_get_reply(struct net *net, struct neighbour *neigh,
2927 u32 pid, u32 seq)
2928{
2929 struct sk_buff *skb;
2930 int err = 0;
2931
2932 skb = nlmsg_new(neigh_nlmsg_size(), GFP_KERNEL);
2933 if (!skb)
2934 return -ENOBUFS;
2935
2936 err = neigh_fill_info(skb, neigh, pid, seq, RTM_NEWNEIGH, 0);
2937 if (err) {
2938 kfree_skb(skb);
2939 goto errout;
2940 }
2941
2942 err = rtnl_unicast(skb, net, pid);
2943errout:
2944 return err;
2945}
2946
2947static inline size_t pneigh_nlmsg_size(void)
2948{
2949 return NLMSG_ALIGN(sizeof(struct ndmsg))
2950 + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2951 + nla_total_size(4) /* NDA_FLAGS_EXT */
2952 + nla_total_size(1); /* NDA_PROTOCOL */
2953}
2954
2955static int pneigh_get_reply(struct net *net, struct pneigh_entry *neigh,
2956 u32 pid, u32 seq, struct neigh_table *tbl)
2957{
2958 struct sk_buff *skb;
2959 int err = 0;
2960
2961 skb = nlmsg_new(pneigh_nlmsg_size(), GFP_KERNEL);
2962 if (!skb)
2963 return -ENOBUFS;
2964
2965 err = pneigh_fill_info(skb, neigh, pid, seq, RTM_NEWNEIGH, 0, tbl);
2966 if (err) {
2967 kfree_skb(skb);
2968 goto errout;
2969 }
2970
2971 err = rtnl_unicast(skb, net, pid);
2972errout:
2973 return err;
2974}
2975
2976static int neigh_get(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2977 struct netlink_ext_ack *extack)
2978{
2979 struct net *net = sock_net(in_skb->sk);
2980 struct net_device *dev = NULL;
2981 struct neigh_table *tbl = NULL;
2982 struct neighbour *neigh;
2983 void *dst = NULL;
2984 u8 ndm_flags = 0;
2985 int dev_idx = 0;
2986 int err;
2987
2988 err = neigh_valid_get_req(nlh, &tbl, &dst, &dev_idx, &ndm_flags,
2989 extack);
2990 if (err < 0)
2991 return err;
2992
2993 if (dev_idx) {
2994 dev = __dev_get_by_index(net, dev_idx);
2995 if (!dev) {
2996 NL_SET_ERR_MSG(extack, "Unknown device ifindex");
2997 return -ENODEV;
2998 }
2999 }
3000
3001 if (!dst) {
3002 NL_SET_ERR_MSG(extack, "Network address not specified");
3003 return -EINVAL;
3004 }
3005
3006 if (ndm_flags & NTF_PROXY) {
3007 struct pneigh_entry *pn;
3008
3009 pn = pneigh_lookup(tbl, net, dst, dev, 0);
3010 if (!pn) {
3011 NL_SET_ERR_MSG(extack, "Proxy neighbour entry not found");
3012 return -ENOENT;
3013 }
3014 return pneigh_get_reply(net, pn, NETLINK_CB(in_skb).portid,
3015 nlh->nlmsg_seq, tbl);
3016 }
3017
3018 if (!dev) {
3019 NL_SET_ERR_MSG(extack, "No device specified");
3020 return -EINVAL;
3021 }
3022
3023 neigh = neigh_lookup(tbl, dst, dev);
3024 if (!neigh) {
3025 NL_SET_ERR_MSG(extack, "Neighbour entry not found");
3026 return -ENOENT;
3027 }
3028
3029 err = neigh_get_reply(net, neigh, NETLINK_CB(in_skb).portid,
3030 nlh->nlmsg_seq);
3031
3032 neigh_release(neigh);
3033
3034 return err;
3035}
3036
3037void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
3038{
3039 int chain;
3040 struct neigh_hash_table *nht;
3041
3042 rcu_read_lock();
3043 nht = rcu_dereference(tbl->nht);
3044
3045 read_lock_bh(&tbl->lock); /* avoid resizes */
3046 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
3047 struct neighbour *n;
3048
3049 neigh_for_each_in_bucket(n, &nht->hash_heads[chain])
3050 cb(n, cookie);
3051 }
3052 read_unlock_bh(&tbl->lock);
3053 rcu_read_unlock();
3054}
3055EXPORT_SYMBOL(neigh_for_each);
3056
3057/* The tbl->lock must be held as a writer and BH disabled. */
3058void __neigh_for_each_release(struct neigh_table *tbl,
3059 int (*cb)(struct neighbour *))
3060{
3061 struct neigh_hash_table *nht;
3062 int chain;
3063
3064 nht = rcu_dereference_protected(tbl->nht,
3065 lockdep_is_held(&tbl->lock));
3066 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
3067 struct hlist_node *tmp;
3068 struct neighbour *n;
3069
3070 neigh_for_each_in_bucket_safe(n, tmp, &nht->hash_heads[chain]) {
3071 int release;
3072
3073 write_lock(&n->lock);
3074 release = cb(n);
3075 if (release) {
3076 hlist_del_rcu(&n->hash);
3077 hlist_del_rcu(&n->dev_list);
3078 neigh_mark_dead(n);
3079 }
3080 write_unlock(&n->lock);
3081 if (release)
3082 neigh_cleanup_and_release(n);
3083 }
3084 }
3085}
3086EXPORT_SYMBOL(__neigh_for_each_release);
3087
3088int neigh_xmit(int index, struct net_device *dev,
3089 const void *addr, struct sk_buff *skb)
3090{
3091 int err = -EAFNOSUPPORT;
3092
3093 if (likely(index < NEIGH_NR_TABLES)) {
3094 struct neigh_table *tbl;
3095 struct neighbour *neigh;
3096
3097 rcu_read_lock();
3098 tbl = rcu_dereference(neigh_tables[index]);
3099 if (!tbl)
3100 goto out_unlock;
3101 if (index == NEIGH_ARP_TABLE) {
3102 u32 key = *((u32 *)addr);
3103
3104 neigh = __ipv4_neigh_lookup_noref(dev, key);
3105 } else {
3106 neigh = __neigh_lookup_noref(tbl, addr, dev);
3107 }
3108 if (!neigh)
3109 neigh = __neigh_create(tbl, addr, dev, false);
3110 err = PTR_ERR(neigh);
3111 if (IS_ERR(neigh)) {
3112 rcu_read_unlock();
3113 goto out_kfree_skb;
3114 }
3115 err = READ_ONCE(neigh->output)(neigh, skb);
3116out_unlock:
3117 rcu_read_unlock();
3118 }
3119 else if (index == NEIGH_LINK_TABLE) {
3120 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
3121 addr, NULL, skb->len);
3122 if (err < 0)
3123 goto out_kfree_skb;
3124 err = dev_queue_xmit(skb);
3125 }
3126out:
3127 return err;
3128out_kfree_skb:
3129 kfree_skb(skb);
3130 goto out;
3131}
3132EXPORT_SYMBOL(neigh_xmit);
3133
3134#ifdef CONFIG_PROC_FS
3135
3136static struct neighbour *neigh_get_valid(struct seq_file *seq,
3137 struct neighbour *n,
3138 loff_t *pos)
3139{
3140 struct neigh_seq_state *state = seq->private;
3141 struct net *net = seq_file_net(seq);
3142
3143 if (!net_eq(dev_net(n->dev), net))
3144 return NULL;
3145
3146 if (state->neigh_sub_iter) {
3147 loff_t fakep = 0;
3148 void *v;
3149
3150 v = state->neigh_sub_iter(state, n, pos ? pos : &fakep);
3151 if (!v)
3152 return NULL;
3153 if (pos)
3154 return v;
3155 }
3156
3157 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
3158 return n;
3159
3160 if (READ_ONCE(n->nud_state) & ~NUD_NOARP)
3161 return n;
3162
3163 return NULL;
3164}
3165
3166static struct neighbour *neigh_get_first(struct seq_file *seq)
3167{
3168 struct neigh_seq_state *state = seq->private;
3169 struct neigh_hash_table *nht = state->nht;
3170 struct neighbour *n, *tmp;
3171
3172 state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
3173
3174 while (++state->bucket < (1 << nht->hash_shift)) {
3175 neigh_for_each_in_bucket(n, &nht->hash_heads[state->bucket]) {
3176 tmp = neigh_get_valid(seq, n, NULL);
3177 if (tmp)
3178 return tmp;
3179 }
3180 }
3181
3182 return NULL;
3183}
3184
3185static struct neighbour *neigh_get_next(struct seq_file *seq,
3186 struct neighbour *n,
3187 loff_t *pos)
3188{
3189 struct neigh_seq_state *state = seq->private;
3190 struct neighbour *tmp;
3191
3192 if (state->neigh_sub_iter) {
3193 void *v = state->neigh_sub_iter(state, n, pos);
3194
3195 if (v)
3196 return n;
3197 }
3198
3199 hlist_for_each_entry_continue(n, hash) {
3200 tmp = neigh_get_valid(seq, n, pos);
3201 if (tmp) {
3202 n = tmp;
3203 goto out;
3204 }
3205 }
3206
3207 n = neigh_get_first(seq);
3208out:
3209 if (n && pos)
3210 --(*pos);
3211
3212 return n;
3213}
3214
3215static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
3216{
3217 struct neighbour *n = neigh_get_first(seq);
3218
3219 if (n) {
3220 --(*pos);
3221 while (*pos) {
3222 n = neigh_get_next(seq, n, pos);
3223 if (!n)
3224 break;
3225 }
3226 }
3227 return *pos ? NULL : n;
3228}
3229
3230static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
3231{
3232 struct neigh_seq_state *state = seq->private;
3233 struct net *net = seq_file_net(seq);
3234 struct neigh_table *tbl = state->tbl;
3235 struct pneigh_entry *pn = NULL;
3236 int bucket;
3237
3238 state->flags |= NEIGH_SEQ_IS_PNEIGH;
3239 for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
3240 pn = tbl->phash_buckets[bucket];
3241 while (pn && !net_eq(pneigh_net(pn), net))
3242 pn = pn->next;
3243 if (pn)
3244 break;
3245 }
3246 state->bucket = bucket;
3247
3248 return pn;
3249}
3250
3251static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
3252 struct pneigh_entry *pn,
3253 loff_t *pos)
3254{
3255 struct neigh_seq_state *state = seq->private;
3256 struct net *net = seq_file_net(seq);
3257 struct neigh_table *tbl = state->tbl;
3258
3259 do {
3260 pn = pn->next;
3261 } while (pn && !net_eq(pneigh_net(pn), net));
3262
3263 while (!pn) {
3264 if (++state->bucket > PNEIGH_HASHMASK)
3265 break;
3266 pn = tbl->phash_buckets[state->bucket];
3267 while (pn && !net_eq(pneigh_net(pn), net))
3268 pn = pn->next;
3269 if (pn)
3270 break;
3271 }
3272
3273 if (pn && pos)
3274 --(*pos);
3275
3276 return pn;
3277}
3278
3279static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
3280{
3281 struct pneigh_entry *pn = pneigh_get_first(seq);
3282
3283 if (pn) {
3284 --(*pos);
3285 while (*pos) {
3286 pn = pneigh_get_next(seq, pn, pos);
3287 if (!pn)
3288 break;
3289 }
3290 }
3291 return *pos ? NULL : pn;
3292}
3293
3294static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
3295{
3296 struct neigh_seq_state *state = seq->private;
3297 void *rc;
3298 loff_t idxpos = *pos;
3299
3300 rc = neigh_get_idx(seq, &idxpos);
3301 if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
3302 rc = pneigh_get_idx(seq, &idxpos);
3303
3304 return rc;
3305}
3306
3307void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
3308 __acquires(tbl->lock)
3309 __acquires(rcu)
3310{
3311 struct neigh_seq_state *state = seq->private;
3312
3313 state->tbl = tbl;
3314 state->bucket = -1;
3315 state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
3316
3317 rcu_read_lock();
3318 state->nht = rcu_dereference(tbl->nht);
3319 read_lock_bh(&tbl->lock);
3320
3321 return *pos ? neigh_get_idx_any(seq, pos) : SEQ_START_TOKEN;
3322}
3323EXPORT_SYMBOL(neigh_seq_start);
3324
3325void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3326{
3327 struct neigh_seq_state *state;
3328 void *rc;
3329
3330 if (v == SEQ_START_TOKEN) {
3331 rc = neigh_get_first(seq);
3332 goto out;
3333 }
3334
3335 state = seq->private;
3336 if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
3337 rc = neigh_get_next(seq, v, NULL);
3338 if (rc)
3339 goto out;
3340 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
3341 rc = pneigh_get_first(seq);
3342 } else {
3343 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
3344 rc = pneigh_get_next(seq, v, NULL);
3345 }
3346out:
3347 ++(*pos);
3348 return rc;
3349}
3350EXPORT_SYMBOL(neigh_seq_next);
3351
3352void neigh_seq_stop(struct seq_file *seq, void *v)
3353 __releases(tbl->lock)
3354 __releases(rcu)
3355{
3356 struct neigh_seq_state *state = seq->private;
3357 struct neigh_table *tbl = state->tbl;
3358
3359 read_unlock_bh(&tbl->lock);
3360 rcu_read_unlock();
3361}
3362EXPORT_SYMBOL(neigh_seq_stop);
3363
3364/* statistics via seq_file */
3365
3366static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
3367{
3368 struct neigh_table *tbl = pde_data(file_inode(seq->file));
3369 int cpu;
3370
3371 if (*pos == 0)
3372 return SEQ_START_TOKEN;
3373
3374 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
3375 if (!cpu_possible(cpu))
3376 continue;
3377 *pos = cpu+1;
3378 return per_cpu_ptr(tbl->stats, cpu);
3379 }
3380 return NULL;
3381}
3382
3383static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3384{
3385 struct neigh_table *tbl = pde_data(file_inode(seq->file));
3386 int cpu;
3387
3388 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
3389 if (!cpu_possible(cpu))
3390 continue;
3391 *pos = cpu+1;
3392 return per_cpu_ptr(tbl->stats, cpu);
3393 }
3394 (*pos)++;
3395 return NULL;
3396}
3397
3398static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
3399{
3400
3401}
3402
3403static int neigh_stat_seq_show(struct seq_file *seq, void *v)
3404{
3405 struct neigh_table *tbl = pde_data(file_inode(seq->file));
3406 struct neigh_statistics *st = v;
3407
3408 if (v == SEQ_START_TOKEN) {
3409 seq_puts(seq, "entries allocs destroys hash_grows lookups hits res_failed rcv_probes_mcast rcv_probes_ucast periodic_gc_runs forced_gc_runs unresolved_discards table_fulls\n");
3410 return 0;
3411 }
3412
3413 seq_printf(seq, "%08x %08lx %08lx %08lx %08lx %08lx %08lx "
3414 "%08lx %08lx %08lx "
3415 "%08lx %08lx %08lx\n",
3416 atomic_read(&tbl->entries),
3417
3418 st->allocs,
3419 st->destroys,
3420 st->hash_grows,
3421
3422 st->lookups,
3423 st->hits,
3424
3425 st->res_failed,
3426
3427 st->rcv_probes_mcast,
3428 st->rcv_probes_ucast,
3429
3430 st->periodic_gc_runs,
3431 st->forced_gc_runs,
3432 st->unres_discards,
3433 st->table_fulls
3434 );
3435
3436 return 0;
3437}
3438
3439static const struct seq_operations neigh_stat_seq_ops = {
3440 .start = neigh_stat_seq_start,
3441 .next = neigh_stat_seq_next,
3442 .stop = neigh_stat_seq_stop,
3443 .show = neigh_stat_seq_show,
3444};
3445#endif /* CONFIG_PROC_FS */
3446
3447static void __neigh_notify(struct neighbour *n, int type, int flags,
3448 u32 pid)
3449{
3450 struct sk_buff *skb;
3451 int err = -ENOBUFS;
3452 struct net *net;
3453
3454 rcu_read_lock();
3455 net = dev_net_rcu(n->dev);
3456 skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC);
3457 if (skb == NULL)
3458 goto errout;
3459
3460 err = neigh_fill_info(skb, n, pid, 0, type, flags);
3461 if (err < 0) {
3462 /* -EMSGSIZE implies BUG in neigh_nlmsg_size() */
3463 WARN_ON(err == -EMSGSIZE);
3464 kfree_skb(skb);
3465 goto errout;
3466 }
3467 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
3468 goto out;
3469errout:
3470 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
3471out:
3472 rcu_read_unlock();
3473}
3474
3475void neigh_app_ns(struct neighbour *n)
3476{
3477 __neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST, 0);
3478}
3479EXPORT_SYMBOL(neigh_app_ns);
3480
3481#ifdef CONFIG_SYSCTL
3482static int unres_qlen_max = INT_MAX / SKB_TRUESIZE(ETH_FRAME_LEN);
3483
3484static int proc_unres_qlen(const struct ctl_table *ctl, int write,
3485 void *buffer, size_t *lenp, loff_t *ppos)
3486{
3487 int size, ret;
3488 struct ctl_table tmp = *ctl;
3489
3490 tmp.extra1 = SYSCTL_ZERO;
3491 tmp.extra2 = &unres_qlen_max;
3492 tmp.data = &size;
3493
3494 size = *(int *)ctl->data / SKB_TRUESIZE(ETH_FRAME_LEN);
3495 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
3496
3497 if (write && !ret)
3498 *(int *)ctl->data = size * SKB_TRUESIZE(ETH_FRAME_LEN);
3499 return ret;
3500}
3501
3502static void neigh_copy_dflt_parms(struct net *net, struct neigh_parms *p,
3503 int index)
3504{
3505 struct net_device *dev;
3506 int family = neigh_parms_family(p);
3507
3508 rcu_read_lock();
3509 for_each_netdev_rcu(net, dev) {
3510 struct neigh_parms *dst_p =
3511 neigh_get_dev_parms_rcu(dev, family);
3512
3513 if (dst_p && !test_bit(index, dst_p->data_state))
3514 dst_p->data[index] = p->data[index];
3515 }
3516 rcu_read_unlock();
3517}
3518
3519static void neigh_proc_update(const struct ctl_table *ctl, int write)
3520{
3521 struct net_device *dev = ctl->extra1;
3522 struct neigh_parms *p = ctl->extra2;
3523 struct net *net = neigh_parms_net(p);
3524 int index = (int *) ctl->data - p->data;
3525
3526 if (!write)
3527 return;
3528
3529 set_bit(index, p->data_state);
3530 if (index == NEIGH_VAR_DELAY_PROBE_TIME)
3531 call_netevent_notifiers(NETEVENT_DELAY_PROBE_TIME_UPDATE, p);
3532 if (!dev) /* NULL dev means this is default value */
3533 neigh_copy_dflt_parms(net, p, index);
3534}
3535
3536static int neigh_proc_dointvec_zero_intmax(const struct ctl_table *ctl, int write,
3537 void *buffer, size_t *lenp,
3538 loff_t *ppos)
3539{
3540 struct ctl_table tmp = *ctl;
3541 int ret;
3542
3543 tmp.extra1 = SYSCTL_ZERO;
3544 tmp.extra2 = SYSCTL_INT_MAX;
3545
3546 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
3547 neigh_proc_update(ctl, write);
3548 return ret;
3549}
3550
3551static int neigh_proc_dointvec_ms_jiffies_positive(const struct ctl_table *ctl, int write,
3552 void *buffer, size_t *lenp, loff_t *ppos)
3553{
3554 struct ctl_table tmp = *ctl;
3555 int ret;
3556
3557 int min = msecs_to_jiffies(1);
3558
3559 tmp.extra1 = &min;
3560 tmp.extra2 = NULL;
3561
3562 ret = proc_dointvec_ms_jiffies_minmax(&tmp, write, buffer, lenp, ppos);
3563 neigh_proc_update(ctl, write);
3564 return ret;
3565}
3566
3567int neigh_proc_dointvec(const struct ctl_table *ctl, int write, void *buffer,
3568 size_t *lenp, loff_t *ppos)
3569{
3570 int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
3571
3572 neigh_proc_update(ctl, write);
3573 return ret;
3574}
3575EXPORT_SYMBOL(neigh_proc_dointvec);
3576
3577int neigh_proc_dointvec_jiffies(const struct ctl_table *ctl, int write, void *buffer,
3578 size_t *lenp, loff_t *ppos)
3579{
3580 int ret = proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
3581
3582 neigh_proc_update(ctl, write);
3583 return ret;
3584}
3585EXPORT_SYMBOL(neigh_proc_dointvec_jiffies);
3586
3587static int neigh_proc_dointvec_userhz_jiffies(const struct ctl_table *ctl, int write,
3588 void *buffer, size_t *lenp,
3589 loff_t *ppos)
3590{
3591 int ret = proc_dointvec_userhz_jiffies(ctl, write, buffer, lenp, ppos);
3592
3593 neigh_proc_update(ctl, write);
3594 return ret;
3595}
3596
3597int neigh_proc_dointvec_ms_jiffies(const struct ctl_table *ctl, int write,
3598 void *buffer, size_t *lenp, loff_t *ppos)
3599{
3600 int ret = proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
3601
3602 neigh_proc_update(ctl, write);
3603 return ret;
3604}
3605EXPORT_SYMBOL(neigh_proc_dointvec_ms_jiffies);
3606
3607static int neigh_proc_dointvec_unres_qlen(const struct ctl_table *ctl, int write,
3608 void *buffer, size_t *lenp,
3609 loff_t *ppos)
3610{
3611 int ret = proc_unres_qlen(ctl, write, buffer, lenp, ppos);
3612
3613 neigh_proc_update(ctl, write);
3614 return ret;
3615}
3616
3617static int neigh_proc_base_reachable_time(const struct ctl_table *ctl, int write,
3618 void *buffer, size_t *lenp,
3619 loff_t *ppos)
3620{
3621 struct neigh_parms *p = ctl->extra2;
3622 int ret;
3623
3624 if (strcmp(ctl->procname, "base_reachable_time") == 0)
3625 ret = neigh_proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
3626 else if (strcmp(ctl->procname, "base_reachable_time_ms") == 0)
3627 ret = neigh_proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
3628 else
3629 ret = -1;
3630
3631 if (write && ret == 0) {
3632 /* update reachable_time as well, otherwise, the change will
3633 * only be effective after the next time neigh_periodic_work
3634 * decides to recompute it
3635 */
3636 p->reachable_time =
3637 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
3638 }
3639 return ret;
3640}
3641
3642#define NEIGH_PARMS_DATA_OFFSET(index) \
3643 (&((struct neigh_parms *) 0)->data[index])
3644
3645#define NEIGH_SYSCTL_ENTRY(attr, data_attr, name, mval, proc) \
3646 [NEIGH_VAR_ ## attr] = { \
3647 .procname = name, \
3648 .data = NEIGH_PARMS_DATA_OFFSET(NEIGH_VAR_ ## data_attr), \
3649 .maxlen = sizeof(int), \
3650 .mode = mval, \
3651 .proc_handler = proc, \
3652 }
3653
3654#define NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(attr, name) \
3655 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_zero_intmax)
3656
3657#define NEIGH_SYSCTL_JIFFIES_ENTRY(attr, name) \
3658 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_jiffies)
3659
3660#define NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(attr, name) \
3661 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_userhz_jiffies)
3662
3663#define NEIGH_SYSCTL_MS_JIFFIES_POSITIVE_ENTRY(attr, name) \
3664 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_ms_jiffies_positive)
3665
3666#define NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(attr, data_attr, name) \
3667 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_ms_jiffies)
3668
3669#define NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(attr, data_attr, name) \
3670 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_unres_qlen)
3671
3672static struct neigh_sysctl_table {
3673 struct ctl_table_header *sysctl_header;
3674 struct ctl_table neigh_vars[NEIGH_VAR_MAX];
3675} neigh_sysctl_template __read_mostly = {
3676 .neigh_vars = {
3677 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_PROBES, "mcast_solicit"),
3678 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(UCAST_PROBES, "ucast_solicit"),
3679 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(APP_PROBES, "app_solicit"),
3680 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_REPROBES, "mcast_resolicit"),
3681 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(RETRANS_TIME, "retrans_time"),
3682 NEIGH_SYSCTL_JIFFIES_ENTRY(BASE_REACHABLE_TIME, "base_reachable_time"),
3683 NEIGH_SYSCTL_JIFFIES_ENTRY(DELAY_PROBE_TIME, "delay_first_probe_time"),
3684 NEIGH_SYSCTL_MS_JIFFIES_POSITIVE_ENTRY(INTERVAL_PROBE_TIME_MS,
3685 "interval_probe_time_ms"),
3686 NEIGH_SYSCTL_JIFFIES_ENTRY(GC_STALETIME, "gc_stale_time"),
3687 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(QUEUE_LEN_BYTES, "unres_qlen_bytes"),
3688 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(PROXY_QLEN, "proxy_qlen"),
3689 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(ANYCAST_DELAY, "anycast_delay"),
3690 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(PROXY_DELAY, "proxy_delay"),
3691 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(LOCKTIME, "locktime"),
3692 NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(QUEUE_LEN, QUEUE_LEN_BYTES, "unres_qlen"),
3693 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(RETRANS_TIME_MS, RETRANS_TIME, "retrans_time_ms"),
3694 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(BASE_REACHABLE_TIME_MS, BASE_REACHABLE_TIME, "base_reachable_time_ms"),
3695 [NEIGH_VAR_GC_INTERVAL] = {
3696 .procname = "gc_interval",
3697 .maxlen = sizeof(int),
3698 .mode = 0644,
3699 .proc_handler = proc_dointvec_jiffies,
3700 },
3701 [NEIGH_VAR_GC_THRESH1] = {
3702 .procname = "gc_thresh1",
3703 .maxlen = sizeof(int),
3704 .mode = 0644,
3705 .extra1 = SYSCTL_ZERO,
3706 .extra2 = SYSCTL_INT_MAX,
3707 .proc_handler = proc_dointvec_minmax,
3708 },
3709 [NEIGH_VAR_GC_THRESH2] = {
3710 .procname = "gc_thresh2",
3711 .maxlen = sizeof(int),
3712 .mode = 0644,
3713 .extra1 = SYSCTL_ZERO,
3714 .extra2 = SYSCTL_INT_MAX,
3715 .proc_handler = proc_dointvec_minmax,
3716 },
3717 [NEIGH_VAR_GC_THRESH3] = {
3718 .procname = "gc_thresh3",
3719 .maxlen = sizeof(int),
3720 .mode = 0644,
3721 .extra1 = SYSCTL_ZERO,
3722 .extra2 = SYSCTL_INT_MAX,
3723 .proc_handler = proc_dointvec_minmax,
3724 },
3725 },
3726};
3727
3728int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
3729 proc_handler *handler)
3730{
3731 int i;
3732 struct neigh_sysctl_table *t;
3733 const char *dev_name_source;
3734 char neigh_path[ sizeof("net//neigh/") + IFNAMSIZ + IFNAMSIZ ];
3735 char *p_name;
3736 size_t neigh_vars_size;
3737
3738 t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL_ACCOUNT);
3739 if (!t)
3740 goto err;
3741
3742 for (i = 0; i < NEIGH_VAR_GC_INTERVAL; i++) {
3743 t->neigh_vars[i].data += (long) p;
3744 t->neigh_vars[i].extra1 = dev;
3745 t->neigh_vars[i].extra2 = p;
3746 }
3747
3748 neigh_vars_size = ARRAY_SIZE(t->neigh_vars);
3749 if (dev) {
3750 dev_name_source = dev->name;
3751 /* Terminate the table early */
3752 neigh_vars_size = NEIGH_VAR_BASE_REACHABLE_TIME_MS + 1;
3753 } else {
3754 struct neigh_table *tbl = p->tbl;
3755 dev_name_source = "default";
3756 t->neigh_vars[NEIGH_VAR_GC_INTERVAL].data = &tbl->gc_interval;
3757 t->neigh_vars[NEIGH_VAR_GC_THRESH1].data = &tbl->gc_thresh1;
3758 t->neigh_vars[NEIGH_VAR_GC_THRESH2].data = &tbl->gc_thresh2;
3759 t->neigh_vars[NEIGH_VAR_GC_THRESH3].data = &tbl->gc_thresh3;
3760 }
3761
3762 if (handler) {
3763 /* RetransTime */
3764 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].proc_handler = handler;
3765 /* ReachableTime */
3766 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler = handler;
3767 /* RetransTime (in milliseconds)*/
3768 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].proc_handler = handler;
3769 /* ReachableTime (in milliseconds) */
3770 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler = handler;
3771 } else {
3772 /* Those handlers will update p->reachable_time after
3773 * base_reachable_time(_ms) is set to ensure the new timer starts being
3774 * applied after the next neighbour update instead of waiting for
3775 * neigh_periodic_work to update its value (can be multiple minutes)
3776 * So any handler that replaces them should do this as well
3777 */
3778 /* ReachableTime */
3779 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler =
3780 neigh_proc_base_reachable_time;
3781 /* ReachableTime (in milliseconds) */
3782 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler =
3783 neigh_proc_base_reachable_time;
3784 }
3785
3786 switch (neigh_parms_family(p)) {
3787 case AF_INET:
3788 p_name = "ipv4";
3789 break;
3790 case AF_INET6:
3791 p_name = "ipv6";
3792 break;
3793 default:
3794 BUG();
3795 }
3796
3797 snprintf(neigh_path, sizeof(neigh_path), "net/%s/neigh/%s",
3798 p_name, dev_name_source);
3799 t->sysctl_header = register_net_sysctl_sz(neigh_parms_net(p),
3800 neigh_path, t->neigh_vars,
3801 neigh_vars_size);
3802 if (!t->sysctl_header)
3803 goto free;
3804
3805 p->sysctl_table = t;
3806 return 0;
3807
3808free:
3809 kfree(t);
3810err:
3811 return -ENOBUFS;
3812}
3813EXPORT_SYMBOL(neigh_sysctl_register);
3814
3815void neigh_sysctl_unregister(struct neigh_parms *p)
3816{
3817 if (p->sysctl_table) {
3818 struct neigh_sysctl_table *t = p->sysctl_table;
3819 p->sysctl_table = NULL;
3820 unregister_net_sysctl_table(t->sysctl_header);
3821 kfree(t);
3822 }
3823}
3824EXPORT_SYMBOL(neigh_sysctl_unregister);
3825
3826#endif /* CONFIG_SYSCTL */
3827
3828static const struct rtnl_msg_handler neigh_rtnl_msg_handlers[] __initconst = {
3829 {.msgtype = RTM_NEWNEIGH, .doit = neigh_add},
3830 {.msgtype = RTM_DELNEIGH, .doit = neigh_delete},
3831 {.msgtype = RTM_GETNEIGH, .doit = neigh_get, .dumpit = neigh_dump_info,
3832 .flags = RTNL_FLAG_DUMP_UNLOCKED},
3833 {.msgtype = RTM_GETNEIGHTBL, .dumpit = neightbl_dump_info},
3834 {.msgtype = RTM_SETNEIGHTBL, .doit = neightbl_set},
3835};
3836
3837static int __init neigh_init(void)
3838{
3839 rtnl_register_many(neigh_rtnl_msg_handlers);
3840 return 0;
3841}
3842
3843subsys_initcall(neigh_init);