Merge branch 'sched/for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip...
[linux-2.6] / net / core / neighbour.c
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/types.h>
19 #include <linux/kernel.h>
20 #include <linux/module.h>
21 #include <linux/socket.h>
22 #include <linux/netdevice.h>
23 #include <linux/proc_fs.h>
24 #ifdef CONFIG_SYSCTL
25 #include <linux/sysctl.h>
26 #endif
27 #include <linux/times.h>
28 #include <net/net_namespace.h>
29 #include <net/neighbour.h>
30 #include <net/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
39 #define NEIGH_DEBUG 1
40
41 #define NEIGH_PRINTK(x...) printk(x)
42 #define NEIGH_NOPRINTK(x...) do { ; } while(0)
43 #define NEIGH_PRINTK0 NEIGH_PRINTK
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
58 static void neigh_timer_handler(unsigned long arg);
59 static void __neigh_notify(struct neighbour *n, int type, int flags);
60 static void neigh_update_notify(struct neighbour *neigh);
61 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev);
62
63 static struct neigh_table *neigh_tables;
64 #ifdef CONFIG_PROC_FS
65 static 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
99 static DEFINE_RWLOCK(neigh_tbl_lock);
100
101 static int neigh_blackhole(struct sk_buff *skb)
102 {
103         kfree_skb(skb);
104         return -ENETDOWN;
105 }
106
107 static 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
122 unsigned long neigh_rand_reach_time(unsigned long base)
123 {
124         return (base ? (net_random() % base) + (base >> 1) : 0);
125 }
126 EXPORT_SYMBOL(neigh_rand_reach_time);
127
128
129 static int neigh_forced_gc(struct neigh_table *tbl)
130 {
131         int shrunk = 0;
132         int i;
133
134         NEIGH_CACHE_STAT_INC(tbl, forced_gc_runs);
135
136         write_lock_bh(&tbl->lock);
137         for (i = 0; i <= tbl->hash_mask; i++) {
138                 struct neighbour *n, **np;
139
140                 np = &tbl->hash_buckets[i];
141                 while ((n = *np) != NULL) {
142                         /* Neighbour record may be discarded if:
143                          * - nobody refers to it.
144                          * - it is not permanent
145                          */
146                         write_lock(&n->lock);
147                         if (atomic_read(&n->refcnt) == 1 &&
148                             !(n->nud_state & NUD_PERMANENT)) {
149                                 *np     = n->next;
150                                 n->dead = 1;
151                                 shrunk  = 1;
152                                 write_unlock(&n->lock);
153                                 neigh_cleanup_and_release(n);
154                                 continue;
155                         }
156                         write_unlock(&n->lock);
157                         np = &n->next;
158                 }
159         }
160
161         tbl->last_flush = jiffies;
162
163         write_unlock_bh(&tbl->lock);
164
165         return shrunk;
166 }
167
168 static void neigh_add_timer(struct neighbour *n, unsigned long when)
169 {
170         neigh_hold(n);
171         if (unlikely(mod_timer(&n->timer, when))) {
172                 printk("NEIGH: BUG, double timer add, state is %x\n",
173                        n->nud_state);
174                 dump_stack();
175         }
176 }
177
178 static int neigh_del_timer(struct neighbour *n)
179 {
180         if ((n->nud_state & NUD_IN_TIMER) &&
181             del_timer(&n->timer)) {
182                 neigh_release(n);
183                 return 1;
184         }
185         return 0;
186 }
187
188 static void pneigh_queue_purge(struct sk_buff_head *list)
189 {
190         struct sk_buff *skb;
191
192         while ((skb = skb_dequeue(list)) != NULL) {
193                 dev_put(skb->dev);
194                 kfree_skb(skb);
195         }
196 }
197
198 static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev)
199 {
200         int i;
201
202         for (i = 0; i <= tbl->hash_mask; i++) {
203                 struct neighbour *n, **np = &tbl->hash_buckets[i];
204
205                 while ((n = *np) != NULL) {
206                         if (dev && n->dev != dev) {
207                                 np = &n->next;
208                                 continue;
209                         }
210                         *np = n->next;
211                         write_lock(&n->lock);
212                         neigh_del_timer(n);
213                         n->dead = 1;
214
215                         if (atomic_read(&n->refcnt) != 1) {
216                                 /* The most unpleasant situation.
217                                    We must destroy neighbour entry,
218                                    but someone still uses it.
219
220                                    The destroy will be delayed until
221                                    the last user releases us, but
222                                    we must kill timers etc. and move
223                                    it to safe state.
224                                  */
225                                 skb_queue_purge(&n->arp_queue);
226                                 n->output = neigh_blackhole;
227                                 if (n->nud_state & NUD_VALID)
228                                         n->nud_state = NUD_NOARP;
229                                 else
230                                         n->nud_state = NUD_NONE;
231                                 NEIGH_PRINTK2("neigh %p is stray.\n", n);
232                         }
233                         write_unlock(&n->lock);
234                         neigh_cleanup_and_release(n);
235                 }
236         }
237 }
238
239 void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev)
240 {
241         write_lock_bh(&tbl->lock);
242         neigh_flush_dev(tbl, dev);
243         write_unlock_bh(&tbl->lock);
244 }
245 EXPORT_SYMBOL(neigh_changeaddr);
246
247 int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
248 {
249         write_lock_bh(&tbl->lock);
250         neigh_flush_dev(tbl, dev);
251         pneigh_ifdown(tbl, dev);
252         write_unlock_bh(&tbl->lock);
253
254         del_timer_sync(&tbl->proxy_timer);
255         pneigh_queue_purge(&tbl->proxy_queue);
256         return 0;
257 }
258 EXPORT_SYMBOL(neigh_ifdown);
259
260 static struct neighbour *neigh_alloc(struct neigh_table *tbl)
261 {
262         struct neighbour *n = NULL;
263         unsigned long now = jiffies;
264         int entries;
265
266         entries = atomic_inc_return(&tbl->entries) - 1;
267         if (entries >= tbl->gc_thresh3 ||
268             (entries >= tbl->gc_thresh2 &&
269              time_after(now, tbl->last_flush + 5 * HZ))) {
270                 if (!neigh_forced_gc(tbl) &&
271                     entries >= tbl->gc_thresh3)
272                         goto out_entries;
273         }
274
275         n = kmem_cache_zalloc(tbl->kmem_cachep, GFP_ATOMIC);
276         if (!n)
277                 goto out_entries;
278
279         skb_queue_head_init(&n->arp_queue);
280         rwlock_init(&n->lock);
281         n->updated        = n->used = now;
282         n->nud_state      = NUD_NONE;
283         n->output         = neigh_blackhole;
284         n->parms          = neigh_parms_clone(&tbl->parms);
285         setup_timer(&n->timer, neigh_timer_handler, (unsigned long)n);
286
287         NEIGH_CACHE_STAT_INC(tbl, allocs);
288         n->tbl            = tbl;
289         atomic_set(&n->refcnt, 1);
290         n->dead           = 1;
291 out:
292         return n;
293
294 out_entries:
295         atomic_dec(&tbl->entries);
296         goto out;
297 }
298
299 static struct neighbour **neigh_hash_alloc(unsigned int entries)
300 {
301         unsigned long size = entries * sizeof(struct neighbour *);
302         struct neighbour **ret;
303
304         if (size <= PAGE_SIZE) {
305                 ret = kzalloc(size, GFP_ATOMIC);
306         } else {
307                 ret = (struct neighbour **)
308                       __get_free_pages(GFP_ATOMIC|__GFP_ZERO, get_order(size));
309         }
310         return ret;
311 }
312
313 static void neigh_hash_free(struct neighbour **hash, unsigned int entries)
314 {
315         unsigned long size = entries * sizeof(struct neighbour *);
316
317         if (size <= PAGE_SIZE)
318                 kfree(hash);
319         else
320                 free_pages((unsigned long)hash, get_order(size));
321 }
322
323 static void neigh_hash_grow(struct neigh_table *tbl, unsigned long new_entries)
324 {
325         struct neighbour **new_hash, **old_hash;
326         unsigned int i, new_hash_mask, old_entries;
327
328         NEIGH_CACHE_STAT_INC(tbl, hash_grows);
329
330         BUG_ON(!is_power_of_2(new_entries));
331         new_hash = neigh_hash_alloc(new_entries);
332         if (!new_hash)
333                 return;
334
335         old_entries = tbl->hash_mask + 1;
336         new_hash_mask = new_entries - 1;
337         old_hash = tbl->hash_buckets;
338
339         get_random_bytes(&tbl->hash_rnd, sizeof(tbl->hash_rnd));
340         for (i = 0; i < old_entries; i++) {
341                 struct neighbour *n, *next;
342
343                 for (n = old_hash[i]; n; n = next) {
344                         unsigned int hash_val = tbl->hash(n->primary_key, n->dev);
345
346                         hash_val &= new_hash_mask;
347                         next = n->next;
348
349                         n->next = new_hash[hash_val];
350                         new_hash[hash_val] = n;
351                 }
352         }
353         tbl->hash_buckets = new_hash;
354         tbl->hash_mask = new_hash_mask;
355
356         neigh_hash_free(old_hash, old_entries);
357 }
358
359 struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
360                                struct net_device *dev)
361 {
362         struct neighbour *n;
363         int key_len = tbl->key_len;
364         u32 hash_val;
365
366         NEIGH_CACHE_STAT_INC(tbl, lookups);
367
368         read_lock_bh(&tbl->lock);
369         hash_val = tbl->hash(pkey, dev);
370         for (n = tbl->hash_buckets[hash_val & tbl->hash_mask]; n; n = n->next) {
371                 if (dev == n->dev && !memcmp(n->primary_key, pkey, key_len)) {
372                         neigh_hold(n);
373                         NEIGH_CACHE_STAT_INC(tbl, hits);
374                         break;
375                 }
376         }
377         read_unlock_bh(&tbl->lock);
378         return n;
379 }
380 EXPORT_SYMBOL(neigh_lookup);
381
382 struct neighbour *neigh_lookup_nodev(struct neigh_table *tbl, struct net *net,
383                                      const void *pkey)
384 {
385         struct neighbour *n;
386         int key_len = tbl->key_len;
387         u32 hash_val;
388
389         NEIGH_CACHE_STAT_INC(tbl, lookups);
390
391         read_lock_bh(&tbl->lock);
392         hash_val = tbl->hash(pkey, NULL);
393         for (n = tbl->hash_buckets[hash_val & tbl->hash_mask]; n; n = n->next) {
394                 if (!memcmp(n->primary_key, pkey, key_len) &&
395                     net_eq(dev_net(n->dev), net)) {
396                         neigh_hold(n);
397                         NEIGH_CACHE_STAT_INC(tbl, hits);
398                         break;
399                 }
400         }
401         read_unlock_bh(&tbl->lock);
402         return n;
403 }
404 EXPORT_SYMBOL(neigh_lookup_nodev);
405
406 struct neighbour *neigh_create(struct neigh_table *tbl, const void *pkey,
407                                struct net_device *dev)
408 {
409         u32 hash_val;
410         int key_len = tbl->key_len;
411         int error;
412         struct neighbour *n1, *rc, *n = neigh_alloc(tbl);
413
414         if (!n) {
415                 rc = ERR_PTR(-ENOBUFS);
416                 goto out;
417         }
418
419         memcpy(n->primary_key, pkey, key_len);
420         n->dev = dev;
421         dev_hold(dev);
422
423         /* Protocol specific setup. */
424         if (tbl->constructor && (error = tbl->constructor(n)) < 0) {
425                 rc = ERR_PTR(error);
426                 goto out_neigh_release;
427         }
428
429         /* Device specific setup. */
430         if (n->parms->neigh_setup &&
431             (error = n->parms->neigh_setup(n)) < 0) {
432                 rc = ERR_PTR(error);
433                 goto out_neigh_release;
434         }
435
436         n->confirmed = jiffies - (n->parms->base_reachable_time << 1);
437
438         write_lock_bh(&tbl->lock);
439
440         if (atomic_read(&tbl->entries) > (tbl->hash_mask + 1))
441                 neigh_hash_grow(tbl, (tbl->hash_mask + 1) << 1);
442
443         hash_val = tbl->hash(pkey, dev) & tbl->hash_mask;
444
445         if (n->parms->dead) {
446                 rc = ERR_PTR(-EINVAL);
447                 goto out_tbl_unlock;
448         }
449
450         for (n1 = tbl->hash_buckets[hash_val]; n1; n1 = n1->next) {
451                 if (dev == n1->dev && !memcmp(n1->primary_key, pkey, key_len)) {
452                         neigh_hold(n1);
453                         rc = n1;
454                         goto out_tbl_unlock;
455                 }
456         }
457
458         n->next = tbl->hash_buckets[hash_val];
459         tbl->hash_buckets[hash_val] = n;
460         n->dead = 0;
461         neigh_hold(n);
462         write_unlock_bh(&tbl->lock);
463         NEIGH_PRINTK2("neigh %p is created.\n", n);
464         rc = n;
465 out:
466         return rc;
467 out_tbl_unlock:
468         write_unlock_bh(&tbl->lock);
469 out_neigh_release:
470         neigh_release(n);
471         goto out;
472 }
473 EXPORT_SYMBOL(neigh_create);
474
475 static u32 pneigh_hash(const void *pkey, int key_len)
476 {
477         u32 hash_val = *(u32 *)(pkey + key_len - 4);
478         hash_val ^= (hash_val >> 16);
479         hash_val ^= hash_val >> 8;
480         hash_val ^= hash_val >> 4;
481         hash_val &= PNEIGH_HASHMASK;
482         return hash_val;
483 }
484
485 static struct pneigh_entry *__pneigh_lookup_1(struct pneigh_entry *n,
486                                               struct net *net,
487                                               const void *pkey,
488                                               int key_len,
489                                               struct net_device *dev)
490 {
491         while (n) {
492                 if (!memcmp(n->key, pkey, key_len) &&
493                     net_eq(pneigh_net(n), net) &&
494                     (n->dev == dev || !n->dev))
495                         return n;
496                 n = n->next;
497         }
498         return NULL;
499 }
500
501 struct pneigh_entry *__pneigh_lookup(struct neigh_table *tbl,
502                 struct net *net, const void *pkey, struct net_device *dev)
503 {
504         int key_len = tbl->key_len;
505         u32 hash_val = pneigh_hash(pkey, key_len);
506
507         return __pneigh_lookup_1(tbl->phash_buckets[hash_val],
508                                  net, pkey, key_len, dev);
509 }
510 EXPORT_SYMBOL_GPL(__pneigh_lookup);
511
512 struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl,
513                                     struct net *net, const void *pkey,
514                                     struct net_device *dev, int creat)
515 {
516         struct pneigh_entry *n;
517         int key_len = tbl->key_len;
518         u32 hash_val = pneigh_hash(pkey, key_len);
519
520         read_lock_bh(&tbl->lock);
521         n = __pneigh_lookup_1(tbl->phash_buckets[hash_val],
522                               net, pkey, key_len, dev);
523         read_unlock_bh(&tbl->lock);
524
525         if (n || !creat)
526                 goto out;
527
528         ASSERT_RTNL();
529
530         n = kmalloc(sizeof(*n) + key_len, GFP_KERNEL);
531         if (!n)
532                 goto out;
533
534 #ifdef CONFIG_NET_NS
535         n->net = hold_net(net);
536 #endif
537         memcpy(n->key, pkey, key_len);
538         n->dev = dev;
539         if (dev)
540                 dev_hold(dev);
541
542         if (tbl->pconstructor && tbl->pconstructor(n)) {
543                 if (dev)
544                         dev_put(dev);
545                 release_net(net);
546                 kfree(n);
547                 n = NULL;
548                 goto out;
549         }
550
551         write_lock_bh(&tbl->lock);
552         n->next = tbl->phash_buckets[hash_val];
553         tbl->phash_buckets[hash_val] = n;
554         write_unlock_bh(&tbl->lock);
555 out:
556         return n;
557 }
558 EXPORT_SYMBOL(pneigh_lookup);
559
560
561 int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
562                   struct net_device *dev)
563 {
564         struct pneigh_entry *n, **np;
565         int key_len = tbl->key_len;
566         u32 hash_val = pneigh_hash(pkey, key_len);
567
568         write_lock_bh(&tbl->lock);
569         for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
570              np = &n->next) {
571                 if (!memcmp(n->key, pkey, key_len) && n->dev == dev &&
572                     net_eq(pneigh_net(n), net)) {
573                         *np = n->next;
574                         write_unlock_bh(&tbl->lock);
575                         if (tbl->pdestructor)
576                                 tbl->pdestructor(n);
577                         if (n->dev)
578                                 dev_put(n->dev);
579                         release_net(pneigh_net(n));
580                         kfree(n);
581                         return 0;
582                 }
583         }
584         write_unlock_bh(&tbl->lock);
585         return -ENOENT;
586 }
587
588 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
589 {
590         struct pneigh_entry *n, **np;
591         u32 h;
592
593         for (h = 0; h <= PNEIGH_HASHMASK; h++) {
594                 np = &tbl->phash_buckets[h];
595                 while ((n = *np) != NULL) {
596                         if (!dev || n->dev == dev) {
597                                 *np = n->next;
598                                 if (tbl->pdestructor)
599                                         tbl->pdestructor(n);
600                                 if (n->dev)
601                                         dev_put(n->dev);
602                                 release_net(pneigh_net(n));
603                                 kfree(n);
604                                 continue;
605                         }
606                         np = &n->next;
607                 }
608         }
609         return -ENOENT;
610 }
611
612 static void neigh_parms_destroy(struct neigh_parms *parms);
613
614 static inline void neigh_parms_put(struct neigh_parms *parms)
615 {
616         if (atomic_dec_and_test(&parms->refcnt))
617                 neigh_parms_destroy(parms);
618 }
619
620 /*
621  *      neighbour must already be out of the table;
622  *
623  */
624 void neigh_destroy(struct neighbour *neigh)
625 {
626         struct hh_cache *hh;
627
628         NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
629
630         if (!neigh->dead) {
631                 printk(KERN_WARNING
632                        "Destroying alive neighbour %p\n", neigh);
633                 dump_stack();
634                 return;
635         }
636
637         if (neigh_del_timer(neigh))
638                 printk(KERN_WARNING "Impossible event.\n");
639
640         while ((hh = neigh->hh) != NULL) {
641                 neigh->hh = hh->hh_next;
642                 hh->hh_next = NULL;
643
644                 write_seqlock_bh(&hh->hh_lock);
645                 hh->hh_output = neigh_blackhole;
646                 write_sequnlock_bh(&hh->hh_lock);
647                 if (atomic_dec_and_test(&hh->hh_refcnt))
648                         kfree(hh);
649         }
650
651         skb_queue_purge(&neigh->arp_queue);
652
653         dev_put(neigh->dev);
654         neigh_parms_put(neigh->parms);
655
656         NEIGH_PRINTK2("neigh %p is destroyed.\n", neigh);
657
658         atomic_dec(&neigh->tbl->entries);
659         kmem_cache_free(neigh->tbl->kmem_cachep, neigh);
660 }
661 EXPORT_SYMBOL(neigh_destroy);
662
663 /* Neighbour state is suspicious;
664    disable fast path.
665
666    Called with write_locked neigh.
667  */
668 static void neigh_suspect(struct neighbour *neigh)
669 {
670         struct hh_cache *hh;
671
672         NEIGH_PRINTK2("neigh %p is suspected.\n", neigh);
673
674         neigh->output = neigh->ops->output;
675
676         for (hh = neigh->hh; hh; hh = hh->hh_next)
677                 hh->hh_output = neigh->ops->output;
678 }
679
680 /* Neighbour state is OK;
681    enable fast path.
682
683    Called with write_locked neigh.
684  */
685 static void neigh_connect(struct neighbour *neigh)
686 {
687         struct hh_cache *hh;
688
689         NEIGH_PRINTK2("neigh %p is connected.\n", neigh);
690
691         neigh->output = neigh->ops->connected_output;
692
693         for (hh = neigh->hh; hh; hh = hh->hh_next)
694                 hh->hh_output = neigh->ops->hh_output;
695 }
696
697 static void neigh_periodic_timer(unsigned long arg)
698 {
699         struct neigh_table *tbl = (struct neigh_table *)arg;
700         struct neighbour *n, **np;
701         unsigned long expire, now = jiffies;
702
703         NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
704
705         write_lock(&tbl->lock);
706
707         /*
708          *      periodically recompute ReachableTime from random function
709          */
710
711         if (time_after(now, tbl->last_rand + 300 * HZ)) {
712                 struct neigh_parms *p;
713                 tbl->last_rand = now;
714                 for (p = &tbl->parms; p; p = p->next)
715                         p->reachable_time =
716                                 neigh_rand_reach_time(p->base_reachable_time);
717         }
718
719         np = &tbl->hash_buckets[tbl->hash_chain_gc];
720         tbl->hash_chain_gc = ((tbl->hash_chain_gc + 1) & tbl->hash_mask);
721
722         while ((n = *np) != NULL) {
723                 unsigned int state;
724
725                 write_lock(&n->lock);
726
727                 state = n->nud_state;
728                 if (state & (NUD_PERMANENT | NUD_IN_TIMER)) {
729                         write_unlock(&n->lock);
730                         goto next_elt;
731                 }
732
733                 if (time_before(n->used, n->confirmed))
734                         n->used = n->confirmed;
735
736                 if (atomic_read(&n->refcnt) == 1 &&
737                     (state == NUD_FAILED ||
738                      time_after(now, n->used + n->parms->gc_staletime))) {
739                         *np = n->next;
740                         n->dead = 1;
741                         write_unlock(&n->lock);
742                         neigh_cleanup_and_release(n);
743                         continue;
744                 }
745                 write_unlock(&n->lock);
746
747 next_elt:
748                 np = &n->next;
749         }
750
751         /* Cycle through all hash buckets every base_reachable_time/2 ticks.
752          * ARP entry timeouts range from 1/2 base_reachable_time to 3/2
753          * base_reachable_time.
754          */
755         expire = tbl->parms.base_reachable_time >> 1;
756         expire /= (tbl->hash_mask + 1);
757         if (!expire)
758                 expire = 1;
759
760         if (expire>HZ)
761                 mod_timer(&tbl->gc_timer, round_jiffies(now + expire));
762         else
763                 mod_timer(&tbl->gc_timer, now + expire);
764
765         write_unlock(&tbl->lock);
766 }
767
768 static __inline__ int neigh_max_probes(struct neighbour *n)
769 {
770         struct neigh_parms *p = n->parms;
771         return (n->nud_state & NUD_PROBE ?
772                 p->ucast_probes :
773                 p->ucast_probes + p->app_probes + p->mcast_probes);
774 }
775
776 /* Called when a timer expires for a neighbour entry. */
777
778 static void neigh_timer_handler(unsigned long arg)
779 {
780         unsigned long now, next;
781         struct neighbour *neigh = (struct neighbour *)arg;
782         unsigned state;
783         int notify = 0;
784
785         write_lock(&neigh->lock);
786
787         state = neigh->nud_state;
788         now = jiffies;
789         next = now + HZ;
790
791         if (!(state & NUD_IN_TIMER)) {
792 #ifndef CONFIG_SMP
793                 printk(KERN_WARNING "neigh: timer & !nud_in_timer\n");
794 #endif
795                 goto out;
796         }
797
798         if (state & NUD_REACHABLE) {
799                 if (time_before_eq(now,
800                                    neigh->confirmed + neigh->parms->reachable_time)) {
801                         NEIGH_PRINTK2("neigh %p is still alive.\n", neigh);
802                         next = neigh->confirmed + neigh->parms->reachable_time;
803                 } else if (time_before_eq(now,
804                                           neigh->used + neigh->parms->delay_probe_time)) {
805                         NEIGH_PRINTK2("neigh %p is delayed.\n", neigh);
806                         neigh->nud_state = NUD_DELAY;
807                         neigh->updated = jiffies;
808                         neigh_suspect(neigh);
809                         next = now + neigh->parms->delay_probe_time;
810                 } else {
811                         NEIGH_PRINTK2("neigh %p is suspected.\n", neigh);
812                         neigh->nud_state = NUD_STALE;
813                         neigh->updated = jiffies;
814                         neigh_suspect(neigh);
815                         notify = 1;
816                 }
817         } else if (state & NUD_DELAY) {
818                 if (time_before_eq(now,
819                                    neigh->confirmed + neigh->parms->delay_probe_time)) {
820                         NEIGH_PRINTK2("neigh %p is now reachable.\n", neigh);
821                         neigh->nud_state = NUD_REACHABLE;
822                         neigh->updated = jiffies;
823                         neigh_connect(neigh);
824                         notify = 1;
825                         next = neigh->confirmed + neigh->parms->reachable_time;
826                 } else {
827                         NEIGH_PRINTK2("neigh %p is probed.\n", neigh);
828                         neigh->nud_state = NUD_PROBE;
829                         neigh->updated = jiffies;
830                         atomic_set(&neigh->probes, 0);
831                         next = now + neigh->parms->retrans_time;
832                 }
833         } else {
834                 /* NUD_PROBE|NUD_INCOMPLETE */
835                 next = now + neigh->parms->retrans_time;
836         }
837
838         if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
839             atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
840                 struct sk_buff *skb;
841
842                 neigh->nud_state = NUD_FAILED;
843                 neigh->updated = jiffies;
844                 notify = 1;
845                 NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
846                 NEIGH_PRINTK2("neigh %p is failed.\n", neigh);
847
848                 /* It is very thin place. report_unreachable is very complicated
849                    routine. Particularly, it can hit the same neighbour entry!
850
851                    So that, we try to be accurate and avoid dead loop. --ANK
852                  */
853                 while (neigh->nud_state == NUD_FAILED &&
854                        (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
855                         write_unlock(&neigh->lock);
856                         neigh->ops->error_report(neigh, skb);
857                         write_lock(&neigh->lock);
858                 }
859                 skb_queue_purge(&neigh->arp_queue);
860         }
861
862         if (neigh->nud_state & NUD_IN_TIMER) {
863                 if (time_before(next, jiffies + HZ/2))
864                         next = jiffies + HZ/2;
865                 if (!mod_timer(&neigh->timer, next))
866                         neigh_hold(neigh);
867         }
868         if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
869                 struct sk_buff *skb = skb_peek(&neigh->arp_queue);
870                 /* keep skb alive even if arp_queue overflows */
871                 if (skb)
872                         skb = skb_copy(skb, GFP_ATOMIC);
873                 write_unlock(&neigh->lock);
874                 neigh->ops->solicit(neigh, skb);
875                 atomic_inc(&neigh->probes);
876                 if (skb)
877                         kfree_skb(skb);
878         } else {
879 out:
880                 write_unlock(&neigh->lock);
881         }
882
883         if (notify)
884                 neigh_update_notify(neigh);
885
886         neigh_release(neigh);
887 }
888
889 int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
890 {
891         int rc;
892         unsigned long now;
893
894         write_lock_bh(&neigh->lock);
895
896         rc = 0;
897         if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
898                 goto out_unlock_bh;
899
900         now = jiffies;
901
902         if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
903                 if (neigh->parms->mcast_probes + neigh->parms->app_probes) {
904                         atomic_set(&neigh->probes, neigh->parms->ucast_probes);
905                         neigh->nud_state     = NUD_INCOMPLETE;
906                         neigh->updated = jiffies;
907                         neigh_add_timer(neigh, now + 1);
908                 } else {
909                         neigh->nud_state = NUD_FAILED;
910                         neigh->updated = jiffies;
911                         write_unlock_bh(&neigh->lock);
912
913                         if (skb)
914                                 kfree_skb(skb);
915                         return 1;
916                 }
917         } else if (neigh->nud_state & NUD_STALE) {
918                 NEIGH_PRINTK2("neigh %p is delayed.\n", neigh);
919                 neigh->nud_state = NUD_DELAY;
920                 neigh->updated = jiffies;
921                 neigh_add_timer(neigh,
922                                 jiffies + neigh->parms->delay_probe_time);
923         }
924
925         if (neigh->nud_state == NUD_INCOMPLETE) {
926                 if (skb) {
927                         if (skb_queue_len(&neigh->arp_queue) >=
928                             neigh->parms->queue_len) {
929                                 struct sk_buff *buff;
930                                 buff = neigh->arp_queue.next;
931                                 __skb_unlink(buff, &neigh->arp_queue);
932                                 kfree_skb(buff);
933                                 NEIGH_CACHE_STAT_INC(neigh->tbl, unres_discards);
934                         }
935                         __skb_queue_tail(&neigh->arp_queue, skb);
936                 }
937                 rc = 1;
938         }
939 out_unlock_bh:
940         write_unlock_bh(&neigh->lock);
941         return rc;
942 }
943 EXPORT_SYMBOL(__neigh_event_send);
944
945 static void neigh_update_hhs(struct neighbour *neigh)
946 {
947         struct hh_cache *hh;
948         void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *)
949                 = neigh->dev->header_ops->cache_update;
950
951         if (update) {
952                 for (hh = neigh->hh; hh; hh = hh->hh_next) {
953                         write_seqlock_bh(&hh->hh_lock);
954                         update(hh, neigh->dev, neigh->ha);
955                         write_sequnlock_bh(&hh->hh_lock);
956                 }
957         }
958 }
959
960
961
962 /* Generic update routine.
963    -- lladdr is new lladdr or NULL, if it is not supplied.
964    -- new    is new state.
965    -- flags
966         NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
967                                 if it is different.
968         NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
969                                 lladdr instead of overriding it
970                                 if it is different.
971                                 It also allows to retain current state
972                                 if lladdr is unchanged.
973         NEIGH_UPDATE_F_ADMIN    means that the change is administrative.
974
975         NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
976                                 NTF_ROUTER flag.
977         NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as
978                                 a router.
979
980    Caller MUST hold reference count on the entry.
981  */
982
983 int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
984                  u32 flags)
985 {
986         u8 old;
987         int err;
988         int notify = 0;
989         struct net_device *dev;
990         int update_isrouter = 0;
991
992         write_lock_bh(&neigh->lock);
993
994         dev    = neigh->dev;
995         old    = neigh->nud_state;
996         err    = -EPERM;
997
998         if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
999             (old & (NUD_NOARP | NUD_PERMANENT)))
1000                 goto out;
1001
1002         if (!(new & NUD_VALID)) {
1003                 neigh_del_timer(neigh);
1004                 if (old & NUD_CONNECTED)
1005                         neigh_suspect(neigh);
1006                 neigh->nud_state = new;
1007                 err = 0;
1008                 notify = old & NUD_VALID;
1009                 goto out;
1010         }
1011
1012         /* Compare new lladdr with cached one */
1013         if (!dev->addr_len) {
1014                 /* First case: device needs no address. */
1015                 lladdr = neigh->ha;
1016         } else if (lladdr) {
1017                 /* The second case: if something is already cached
1018                    and a new address is proposed:
1019                    - compare new & old
1020                    - if they are different, check override flag
1021                  */
1022                 if ((old & NUD_VALID) &&
1023                     !memcmp(lladdr, neigh->ha, dev->addr_len))
1024                         lladdr = neigh->ha;
1025         } else {
1026                 /* No address is supplied; if we know something,
1027                    use it, otherwise discard the request.
1028                  */
1029                 err = -EINVAL;
1030                 if (!(old & NUD_VALID))
1031                         goto out;
1032                 lladdr = neigh->ha;
1033         }
1034
1035         if (new & NUD_CONNECTED)
1036                 neigh->confirmed = jiffies;
1037         neigh->updated = jiffies;
1038
1039         /* If entry was valid and address is not changed,
1040            do not change entry state, if new one is STALE.
1041          */
1042         err = 0;
1043         update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1044         if (old & NUD_VALID) {
1045                 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
1046                         update_isrouter = 0;
1047                         if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
1048                             (old & NUD_CONNECTED)) {
1049                                 lladdr = neigh->ha;
1050                                 new = NUD_STALE;
1051                         } else
1052                                 goto out;
1053                 } else {
1054                         if (lladdr == neigh->ha && new == NUD_STALE &&
1055                             ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) ||
1056                              (old & NUD_CONNECTED))
1057                             )
1058                                 new = old;
1059                 }
1060         }
1061
1062         if (new != old) {
1063                 neigh_del_timer(neigh);
1064                 if (new & NUD_IN_TIMER)
1065                         neigh_add_timer(neigh, (jiffies +
1066                                                 ((new & NUD_REACHABLE) ?
1067                                                  neigh->parms->reachable_time :
1068                                                  0)));
1069                 neigh->nud_state = new;
1070         }
1071
1072         if (lladdr != neigh->ha) {
1073                 memcpy(&neigh->ha, lladdr, dev->addr_len);
1074                 neigh_update_hhs(neigh);
1075                 if (!(new & NUD_CONNECTED))
1076                         neigh->confirmed = jiffies -
1077                                       (neigh->parms->base_reachable_time << 1);
1078                 notify = 1;
1079         }
1080         if (new == old)
1081                 goto out;
1082         if (new & NUD_CONNECTED)
1083                 neigh_connect(neigh);
1084         else
1085                 neigh_suspect(neigh);
1086         if (!(old & NUD_VALID)) {
1087                 struct sk_buff *skb;
1088
1089                 /* Again: avoid dead loop if something went wrong */
1090
1091                 while (neigh->nud_state & NUD_VALID &&
1092                        (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1093                         struct neighbour *n1 = neigh;
1094                         write_unlock_bh(&neigh->lock);
1095                         /* On shaper/eql skb->dst->neighbour != neigh :( */
1096                         if (skb->dst && skb->dst->neighbour)
1097                                 n1 = skb->dst->neighbour;
1098                         n1->output(skb);
1099                         write_lock_bh(&neigh->lock);
1100                 }
1101                 skb_queue_purge(&neigh->arp_queue);
1102         }
1103 out:
1104         if (update_isrouter) {
1105                 neigh->flags = (flags & NEIGH_UPDATE_F_ISROUTER) ?
1106                         (neigh->flags | NTF_ROUTER) :
1107                         (neigh->flags & ~NTF_ROUTER);
1108         }
1109         write_unlock_bh(&neigh->lock);
1110
1111         if (notify)
1112                 neigh_update_notify(neigh);
1113
1114         return err;
1115 }
1116 EXPORT_SYMBOL(neigh_update);
1117
1118 struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1119                                  u8 *lladdr, void *saddr,
1120                                  struct net_device *dev)
1121 {
1122         struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1123                                                  lladdr || !dev->addr_len);
1124         if (neigh)
1125                 neigh_update(neigh, lladdr, NUD_STALE,
1126                              NEIGH_UPDATE_F_OVERRIDE);
1127         return neigh;
1128 }
1129 EXPORT_SYMBOL(neigh_event_ns);
1130
1131 static void neigh_hh_init(struct neighbour *n, struct dst_entry *dst,
1132                           __be16 protocol)
1133 {
1134         struct hh_cache *hh;
1135         struct net_device *dev = dst->dev;
1136
1137         for (hh = n->hh; hh; hh = hh->hh_next)
1138                 if (hh->hh_type == protocol)
1139                         break;
1140
1141         if (!hh && (hh = kzalloc(sizeof(*hh), GFP_ATOMIC)) != NULL) {
1142                 seqlock_init(&hh->hh_lock);
1143                 hh->hh_type = protocol;
1144                 atomic_set(&hh->hh_refcnt, 0);
1145                 hh->hh_next = NULL;
1146
1147                 if (dev->header_ops->cache(n, hh)) {
1148                         kfree(hh);
1149                         hh = NULL;
1150                 } else {
1151                         atomic_inc(&hh->hh_refcnt);
1152                         hh->hh_next = n->hh;
1153                         n->hh       = hh;
1154                         if (n->nud_state & NUD_CONNECTED)
1155                                 hh->hh_output = n->ops->hh_output;
1156                         else
1157                                 hh->hh_output = n->ops->output;
1158                 }
1159         }
1160         if (hh) {
1161                 atomic_inc(&hh->hh_refcnt);
1162                 dst->hh = hh;
1163         }
1164 }
1165
1166 /* This function can be used in contexts, where only old dev_queue_xmit
1167    worked, f.e. if you want to override normal output path (eql, shaper),
1168    but resolution is not made yet.
1169  */
1170
1171 int neigh_compat_output(struct sk_buff *skb)
1172 {
1173         struct net_device *dev = skb->dev;
1174
1175         __skb_pull(skb, skb_network_offset(skb));
1176
1177         if (dev_hard_header(skb, dev, ntohs(skb->protocol), NULL, NULL,
1178                             skb->len) < 0 &&
1179             dev->header_ops->rebuild(skb))
1180                 return 0;
1181
1182         return dev_queue_xmit(skb);
1183 }
1184 EXPORT_SYMBOL(neigh_compat_output);
1185
1186 /* Slow and careful. */
1187
1188 int neigh_resolve_output(struct sk_buff *skb)
1189 {
1190         struct dst_entry *dst = skb->dst;
1191         struct neighbour *neigh;
1192         int rc = 0;
1193
1194         if (!dst || !(neigh = dst->neighbour))
1195                 goto discard;
1196
1197         __skb_pull(skb, skb_network_offset(skb));
1198
1199         if (!neigh_event_send(neigh, skb)) {
1200                 int err;
1201                 struct net_device *dev = neigh->dev;
1202                 if (dev->header_ops->cache && !dst->hh) {
1203                         write_lock_bh(&neigh->lock);
1204                         if (!dst->hh)
1205                                 neigh_hh_init(neigh, dst, dst->ops->protocol);
1206                         err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1207                                               neigh->ha, NULL, skb->len);
1208                         write_unlock_bh(&neigh->lock);
1209                 } else {
1210                         read_lock_bh(&neigh->lock);
1211                         err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1212                                               neigh->ha, NULL, skb->len);
1213                         read_unlock_bh(&neigh->lock);
1214                 }
1215                 if (err >= 0)
1216                         rc = neigh->ops->queue_xmit(skb);
1217                 else
1218                         goto out_kfree_skb;
1219         }
1220 out:
1221         return rc;
1222 discard:
1223         NEIGH_PRINTK1("neigh_resolve_output: dst=%p neigh=%p\n",
1224                       dst, dst ? dst->neighbour : NULL);
1225 out_kfree_skb:
1226         rc = -EINVAL;
1227         kfree_skb(skb);
1228         goto out;
1229 }
1230 EXPORT_SYMBOL(neigh_resolve_output);
1231
1232 /* As fast as possible without hh cache */
1233
1234 int neigh_connected_output(struct sk_buff *skb)
1235 {
1236         int err;
1237         struct dst_entry *dst = skb->dst;
1238         struct neighbour *neigh = dst->neighbour;
1239         struct net_device *dev = neigh->dev;
1240
1241         __skb_pull(skb, skb_network_offset(skb));
1242
1243         read_lock_bh(&neigh->lock);
1244         err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1245                               neigh->ha, NULL, skb->len);
1246         read_unlock_bh(&neigh->lock);
1247         if (err >= 0)
1248                 err = neigh->ops->queue_xmit(skb);
1249         else {
1250                 err = -EINVAL;
1251                 kfree_skb(skb);
1252         }
1253         return err;
1254 }
1255 EXPORT_SYMBOL(neigh_connected_output);
1256
1257 static void neigh_proxy_process(unsigned long arg)
1258 {
1259         struct neigh_table *tbl = (struct neigh_table *)arg;
1260         long sched_next = 0;
1261         unsigned long now = jiffies;
1262         struct sk_buff *skb;
1263
1264         spin_lock(&tbl->proxy_queue.lock);
1265
1266         skb = tbl->proxy_queue.next;
1267
1268         while (skb != (struct sk_buff *)&tbl->proxy_queue) {
1269                 struct sk_buff *back = skb;
1270                 long tdif = NEIGH_CB(back)->sched_next - now;
1271
1272                 skb = skb->next;
1273                 if (tdif <= 0) {
1274                         struct net_device *dev = back->dev;
1275                         __skb_unlink(back, &tbl->proxy_queue);
1276                         if (tbl->proxy_redo && netif_running(dev))
1277                                 tbl->proxy_redo(back);
1278                         else
1279                                 kfree_skb(back);
1280
1281                         dev_put(dev);
1282                 } else if (!sched_next || tdif < sched_next)
1283                         sched_next = tdif;
1284         }
1285         del_timer(&tbl->proxy_timer);
1286         if (sched_next)
1287                 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1288         spin_unlock(&tbl->proxy_queue.lock);
1289 }
1290
1291 void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1292                     struct sk_buff *skb)
1293 {
1294         unsigned long now = jiffies;
1295         unsigned long sched_next = now + (net_random() % p->proxy_delay);
1296
1297         if (tbl->proxy_queue.qlen > p->proxy_qlen) {
1298                 kfree_skb(skb);
1299                 return;
1300         }
1301
1302         NEIGH_CB(skb)->sched_next = sched_next;
1303         NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1304
1305         spin_lock(&tbl->proxy_queue.lock);
1306         if (del_timer(&tbl->proxy_timer)) {
1307                 if (time_before(tbl->proxy_timer.expires, sched_next))
1308                         sched_next = tbl->proxy_timer.expires;
1309         }
1310         dst_release(skb->dst);
1311         skb->dst = NULL;
1312         dev_hold(skb->dev);
1313         __skb_queue_tail(&tbl->proxy_queue, skb);
1314         mod_timer(&tbl->proxy_timer, sched_next);
1315         spin_unlock(&tbl->proxy_queue.lock);
1316 }
1317 EXPORT_SYMBOL(pneigh_enqueue);
1318
1319 static inline struct neigh_parms *lookup_neigh_params(struct neigh_table *tbl,
1320                                                       struct net *net, int ifindex)
1321 {
1322         struct neigh_parms *p;
1323
1324         for (p = &tbl->parms; p; p = p->next) {
1325                 if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
1326                     (!p->dev && !ifindex))
1327                         return p;
1328         }
1329
1330         return NULL;
1331 }
1332
1333 struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1334                                       struct neigh_table *tbl)
1335 {
1336         struct neigh_parms *p, *ref;
1337         struct net *net;
1338
1339         net = dev_net(dev);
1340         ref = lookup_neigh_params(tbl, net, 0);
1341         if (!ref)
1342                 return NULL;
1343
1344         p = kmemdup(ref, sizeof(*p), GFP_KERNEL);
1345         if (p) {
1346                 p->tbl            = tbl;
1347                 atomic_set(&p->refcnt, 1);
1348                 INIT_RCU_HEAD(&p->rcu_head);
1349                 p->reachable_time =
1350                                 neigh_rand_reach_time(p->base_reachable_time);
1351
1352                 if (dev->neigh_setup && dev->neigh_setup(dev, p)) {
1353                         kfree(p);
1354                         return NULL;
1355                 }
1356
1357                 dev_hold(dev);
1358                 p->dev = dev;
1359 #ifdef CONFIG_NET_NS
1360                 p->net = hold_net(net);
1361 #endif
1362                 p->sysctl_table = NULL;
1363                 write_lock_bh(&tbl->lock);
1364                 p->next         = tbl->parms.next;
1365                 tbl->parms.next = p;
1366                 write_unlock_bh(&tbl->lock);
1367         }
1368         return p;
1369 }
1370 EXPORT_SYMBOL(neigh_parms_alloc);
1371
1372 static void neigh_rcu_free_parms(struct rcu_head *head)
1373 {
1374         struct neigh_parms *parms =
1375                 container_of(head, struct neigh_parms, rcu_head);
1376
1377         neigh_parms_put(parms);
1378 }
1379
1380 void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1381 {
1382         struct neigh_parms **p;
1383
1384         if (!parms || parms == &tbl->parms)
1385                 return;
1386         write_lock_bh(&tbl->lock);
1387         for (p = &tbl->parms.next; *p; p = &(*p)->next) {
1388                 if (*p == parms) {
1389                         *p = parms->next;
1390                         parms->dead = 1;
1391                         write_unlock_bh(&tbl->lock);
1392                         if (parms->dev)
1393                                 dev_put(parms->dev);
1394                         call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1395                         return;
1396                 }
1397         }
1398         write_unlock_bh(&tbl->lock);
1399         NEIGH_PRINTK1("neigh_parms_release: not found\n");
1400 }
1401 EXPORT_SYMBOL(neigh_parms_release);
1402
1403 static void neigh_parms_destroy(struct neigh_parms *parms)
1404 {
1405         release_net(neigh_parms_net(parms));
1406         kfree(parms);
1407 }
1408
1409 static struct lock_class_key neigh_table_proxy_queue_class;
1410
1411 void neigh_table_init_no_netlink(struct neigh_table *tbl)
1412 {
1413         unsigned long now = jiffies;
1414         unsigned long phsize;
1415
1416 #ifdef CONFIG_NET_NS
1417         tbl->parms.net = &init_net;
1418 #endif
1419         atomic_set(&tbl->parms.refcnt, 1);
1420         INIT_RCU_HEAD(&tbl->parms.rcu_head);
1421         tbl->parms.reachable_time =
1422                           neigh_rand_reach_time(tbl->parms.base_reachable_time);
1423
1424         if (!tbl->kmem_cachep)
1425                 tbl->kmem_cachep =
1426                         kmem_cache_create(tbl->id, tbl->entry_size, 0,
1427                                           SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1428                                           NULL);
1429         tbl->stats = alloc_percpu(struct neigh_statistics);
1430         if (!tbl->stats)
1431                 panic("cannot create neighbour cache statistics");
1432
1433 #ifdef CONFIG_PROC_FS
1434         tbl->pde = proc_create_data(tbl->id, 0, init_net.proc_net_stat,
1435                                     &neigh_stat_seq_fops, tbl);
1436         if (!tbl->pde)
1437                 panic("cannot create neighbour proc dir entry");
1438 #endif
1439
1440         tbl->hash_mask = 1;
1441         tbl->hash_buckets = neigh_hash_alloc(tbl->hash_mask + 1);
1442
1443         phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1444         tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1445
1446         if (!tbl->hash_buckets || !tbl->phash_buckets)
1447                 panic("cannot allocate neighbour cache hashes");
1448
1449         get_random_bytes(&tbl->hash_rnd, sizeof(tbl->hash_rnd));
1450
1451         rwlock_init(&tbl->lock);
1452         setup_timer(&tbl->gc_timer, neigh_periodic_timer, (unsigned long)tbl);
1453         tbl->gc_timer.expires  = now + 1;
1454         add_timer(&tbl->gc_timer);
1455
1456         setup_timer(&tbl->proxy_timer, neigh_proxy_process, (unsigned long)tbl);
1457         skb_queue_head_init_class(&tbl->proxy_queue,
1458                         &neigh_table_proxy_queue_class);
1459
1460         tbl->last_flush = now;
1461         tbl->last_rand  = now + tbl->parms.reachable_time * 20;
1462 }
1463 EXPORT_SYMBOL(neigh_table_init_no_netlink);
1464
1465 void neigh_table_init(struct neigh_table *tbl)
1466 {
1467         struct neigh_table *tmp;
1468
1469         neigh_table_init_no_netlink(tbl);
1470         write_lock(&neigh_tbl_lock);
1471         for (tmp = neigh_tables; tmp; tmp = tmp->next) {
1472                 if (tmp->family == tbl->family)
1473                         break;
1474         }
1475         tbl->next       = neigh_tables;
1476         neigh_tables    = tbl;
1477         write_unlock(&neigh_tbl_lock);
1478
1479         if (unlikely(tmp)) {
1480                 printk(KERN_ERR "NEIGH: Registering multiple tables for "
1481                        "family %d\n", tbl->family);
1482                 dump_stack();
1483         }
1484 }
1485 EXPORT_SYMBOL(neigh_table_init);
1486
1487 int neigh_table_clear(struct neigh_table *tbl)
1488 {
1489         struct neigh_table **tp;
1490
1491         /* It is not clean... Fix it to unload IPv6 module safely */
1492         del_timer_sync(&tbl->gc_timer);
1493         del_timer_sync(&tbl->proxy_timer);
1494         pneigh_queue_purge(&tbl->proxy_queue);
1495         neigh_ifdown(tbl, NULL);
1496         if (atomic_read(&tbl->entries))
1497                 printk(KERN_CRIT "neighbour leakage\n");
1498         write_lock(&neigh_tbl_lock);
1499         for (tp = &neigh_tables; *tp; tp = &(*tp)->next) {
1500                 if (*tp == tbl) {
1501                         *tp = tbl->next;
1502                         break;
1503                 }
1504         }
1505         write_unlock(&neigh_tbl_lock);
1506
1507         neigh_hash_free(tbl->hash_buckets, tbl->hash_mask + 1);
1508         tbl->hash_buckets = NULL;
1509
1510         kfree(tbl->phash_buckets);
1511         tbl->phash_buckets = NULL;
1512
1513         remove_proc_entry(tbl->id, init_net.proc_net_stat);
1514
1515         free_percpu(tbl->stats);
1516         tbl->stats = NULL;
1517
1518         kmem_cache_destroy(tbl->kmem_cachep);
1519         tbl->kmem_cachep = NULL;
1520
1521         return 0;
1522 }
1523 EXPORT_SYMBOL(neigh_table_clear);
1524
1525 static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1526 {
1527         struct net *net = sock_net(skb->sk);
1528         struct ndmsg *ndm;
1529         struct nlattr *dst_attr;
1530         struct neigh_table *tbl;
1531         struct net_device *dev = NULL;
1532         int err = -EINVAL;
1533
1534         if (nlmsg_len(nlh) < sizeof(*ndm))
1535                 goto out;
1536
1537         dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1538         if (dst_attr == NULL)
1539                 goto out;
1540
1541         ndm = nlmsg_data(nlh);
1542         if (ndm->ndm_ifindex) {
1543                 dev = dev_get_by_index(net, ndm->ndm_ifindex);
1544                 if (dev == NULL) {
1545                         err = -ENODEV;
1546                         goto out;
1547                 }
1548         }
1549
1550         read_lock(&neigh_tbl_lock);
1551         for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1552                 struct neighbour *neigh;
1553
1554                 if (tbl->family != ndm->ndm_family)
1555                         continue;
1556                 read_unlock(&neigh_tbl_lock);
1557
1558                 if (nla_len(dst_attr) < tbl->key_len)
1559                         goto out_dev_put;
1560
1561                 if (ndm->ndm_flags & NTF_PROXY) {
1562                         err = pneigh_delete(tbl, net, nla_data(dst_attr), dev);
1563                         goto out_dev_put;
1564                 }
1565
1566                 if (dev == NULL)
1567                         goto out_dev_put;
1568
1569                 neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1570                 if (neigh == NULL) {
1571                         err = -ENOENT;
1572                         goto out_dev_put;
1573                 }
1574
1575                 err = neigh_update(neigh, NULL, NUD_FAILED,
1576                                    NEIGH_UPDATE_F_OVERRIDE |
1577                                    NEIGH_UPDATE_F_ADMIN);
1578                 neigh_release(neigh);
1579                 goto out_dev_put;
1580         }
1581         read_unlock(&neigh_tbl_lock);
1582         err = -EAFNOSUPPORT;
1583
1584 out_dev_put:
1585         if (dev)
1586                 dev_put(dev);
1587 out:
1588         return err;
1589 }
1590
1591 static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1592 {
1593         struct net *net = sock_net(skb->sk);
1594         struct ndmsg *ndm;
1595         struct nlattr *tb[NDA_MAX+1];
1596         struct neigh_table *tbl;
1597         struct net_device *dev = NULL;
1598         int err;
1599
1600         err = nlmsg_parse(nlh, sizeof(*ndm), tb, NDA_MAX, NULL);
1601         if (err < 0)
1602                 goto out;
1603
1604         err = -EINVAL;
1605         if (tb[NDA_DST] == NULL)
1606                 goto out;
1607
1608         ndm = nlmsg_data(nlh);
1609         if (ndm->ndm_ifindex) {
1610                 dev = dev_get_by_index(net, ndm->ndm_ifindex);
1611                 if (dev == NULL) {
1612                         err = -ENODEV;
1613                         goto out;
1614                 }
1615
1616                 if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len)
1617                         goto out_dev_put;
1618         }
1619
1620         read_lock(&neigh_tbl_lock);
1621         for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1622                 int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE;
1623                 struct neighbour *neigh;
1624                 void *dst, *lladdr;
1625
1626                 if (tbl->family != ndm->ndm_family)
1627                         continue;
1628                 read_unlock(&neigh_tbl_lock);
1629
1630                 if (nla_len(tb[NDA_DST]) < tbl->key_len)
1631                         goto out_dev_put;
1632                 dst = nla_data(tb[NDA_DST]);
1633                 lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
1634
1635                 if (ndm->ndm_flags & NTF_PROXY) {
1636                         struct pneigh_entry *pn;
1637
1638                         err = -ENOBUFS;
1639                         pn = pneigh_lookup(tbl, net, dst, dev, 1);
1640                         if (pn) {
1641                                 pn->flags = ndm->ndm_flags;
1642                                 err = 0;
1643                         }
1644                         goto out_dev_put;
1645                 }
1646
1647                 if (dev == NULL)
1648                         goto out_dev_put;
1649
1650                 neigh = neigh_lookup(tbl, dst, dev);
1651                 if (neigh == NULL) {
1652                         if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
1653                                 err = -ENOENT;
1654                                 goto out_dev_put;
1655                         }
1656
1657                         neigh = __neigh_lookup_errno(tbl, dst, dev);
1658                         if (IS_ERR(neigh)) {
1659                                 err = PTR_ERR(neigh);
1660                                 goto out_dev_put;
1661                         }
1662                 } else {
1663                         if (nlh->nlmsg_flags & NLM_F_EXCL) {
1664                                 err = -EEXIST;
1665                                 neigh_release(neigh);
1666                                 goto out_dev_put;
1667                         }
1668
1669                         if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
1670                                 flags &= ~NEIGH_UPDATE_F_OVERRIDE;
1671                 }
1672
1673                 err = neigh_update(neigh, lladdr, ndm->ndm_state, flags);
1674                 neigh_release(neigh);
1675                 goto out_dev_put;
1676         }
1677
1678         read_unlock(&neigh_tbl_lock);
1679         err = -EAFNOSUPPORT;
1680
1681 out_dev_put:
1682         if (dev)
1683                 dev_put(dev);
1684 out:
1685         return err;
1686 }
1687
1688 static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
1689 {
1690         struct nlattr *nest;
1691
1692         nest = nla_nest_start(skb, NDTA_PARMS);
1693         if (nest == NULL)
1694                 return -ENOBUFS;
1695
1696         if (parms->dev)
1697                 NLA_PUT_U32(skb, NDTPA_IFINDEX, parms->dev->ifindex);
1698
1699         NLA_PUT_U32(skb, NDTPA_REFCNT, atomic_read(&parms->refcnt));
1700         NLA_PUT_U32(skb, NDTPA_QUEUE_LEN, parms->queue_len);
1701         NLA_PUT_U32(skb, NDTPA_PROXY_QLEN, parms->proxy_qlen);
1702         NLA_PUT_U32(skb, NDTPA_APP_PROBES, parms->app_probes);
1703         NLA_PUT_U32(skb, NDTPA_UCAST_PROBES, parms->ucast_probes);
1704         NLA_PUT_U32(skb, NDTPA_MCAST_PROBES, parms->mcast_probes);
1705         NLA_PUT_MSECS(skb, NDTPA_REACHABLE_TIME, parms->reachable_time);
1706         NLA_PUT_MSECS(skb, NDTPA_BASE_REACHABLE_TIME,
1707                       parms->base_reachable_time);
1708         NLA_PUT_MSECS(skb, NDTPA_GC_STALETIME, parms->gc_staletime);
1709         NLA_PUT_MSECS(skb, NDTPA_DELAY_PROBE_TIME, parms->delay_probe_time);
1710         NLA_PUT_MSECS(skb, NDTPA_RETRANS_TIME, parms->retrans_time);
1711         NLA_PUT_MSECS(skb, NDTPA_ANYCAST_DELAY, parms->anycast_delay);
1712         NLA_PUT_MSECS(skb, NDTPA_PROXY_DELAY, parms->proxy_delay);
1713         NLA_PUT_MSECS(skb, NDTPA_LOCKTIME, parms->locktime);
1714
1715         return nla_nest_end(skb, nest);
1716
1717 nla_put_failure:
1718         nla_nest_cancel(skb, nest);
1719         return -EMSGSIZE;
1720 }
1721
1722 static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl,
1723                               u32 pid, u32 seq, int type, int flags)
1724 {
1725         struct nlmsghdr *nlh;
1726         struct ndtmsg *ndtmsg;
1727
1728         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1729         if (nlh == NULL)
1730                 return -EMSGSIZE;
1731
1732         ndtmsg = nlmsg_data(nlh);
1733
1734         read_lock_bh(&tbl->lock);
1735         ndtmsg->ndtm_family = tbl->family;
1736         ndtmsg->ndtm_pad1   = 0;
1737         ndtmsg->ndtm_pad2   = 0;
1738
1739         NLA_PUT_STRING(skb, NDTA_NAME, tbl->id);
1740         NLA_PUT_MSECS(skb, NDTA_GC_INTERVAL, tbl->gc_interval);
1741         NLA_PUT_U32(skb, NDTA_THRESH1, tbl->gc_thresh1);
1742         NLA_PUT_U32(skb, NDTA_THRESH2, tbl->gc_thresh2);
1743         NLA_PUT_U32(skb, NDTA_THRESH3, tbl->gc_thresh3);
1744
1745         {
1746                 unsigned long now = jiffies;
1747                 unsigned int flush_delta = now - tbl->last_flush;
1748                 unsigned int rand_delta = now - tbl->last_rand;
1749
1750                 struct ndt_config ndc = {
1751                         .ndtc_key_len           = tbl->key_len,
1752                         .ndtc_entry_size        = tbl->entry_size,
1753                         .ndtc_entries           = atomic_read(&tbl->entries),
1754                         .ndtc_last_flush        = jiffies_to_msecs(flush_delta),
1755                         .ndtc_last_rand         = jiffies_to_msecs(rand_delta),
1756                         .ndtc_hash_rnd          = tbl->hash_rnd,
1757                         .ndtc_hash_mask         = tbl->hash_mask,
1758                         .ndtc_hash_chain_gc     = tbl->hash_chain_gc,
1759                         .ndtc_proxy_qlen        = tbl->proxy_queue.qlen,
1760                 };
1761
1762                 NLA_PUT(skb, NDTA_CONFIG, sizeof(ndc), &ndc);
1763         }
1764
1765         {
1766                 int cpu;
1767                 struct ndt_stats ndst;
1768
1769                 memset(&ndst, 0, sizeof(ndst));
1770
1771                 for_each_possible_cpu(cpu) {
1772                         struct neigh_statistics *st;
1773
1774                         st = per_cpu_ptr(tbl->stats, cpu);
1775                         ndst.ndts_allocs                += st->allocs;
1776                         ndst.ndts_destroys              += st->destroys;
1777                         ndst.ndts_hash_grows            += st->hash_grows;
1778                         ndst.ndts_res_failed            += st->res_failed;
1779                         ndst.ndts_lookups               += st->lookups;
1780                         ndst.ndts_hits                  += st->hits;
1781                         ndst.ndts_rcv_probes_mcast      += st->rcv_probes_mcast;
1782                         ndst.ndts_rcv_probes_ucast      += st->rcv_probes_ucast;
1783                         ndst.ndts_periodic_gc_runs      += st->periodic_gc_runs;
1784                         ndst.ndts_forced_gc_runs        += st->forced_gc_runs;
1785                 }
1786
1787                 NLA_PUT(skb, NDTA_STATS, sizeof(ndst), &ndst);
1788         }
1789
1790         BUG_ON(tbl->parms.dev);
1791         if (neightbl_fill_parms(skb, &tbl->parms) < 0)
1792                 goto nla_put_failure;
1793
1794         read_unlock_bh(&tbl->lock);
1795         return nlmsg_end(skb, nlh);
1796
1797 nla_put_failure:
1798         read_unlock_bh(&tbl->lock);
1799         nlmsg_cancel(skb, nlh);
1800         return -EMSGSIZE;
1801 }
1802
1803 static int neightbl_fill_param_info(struct sk_buff *skb,
1804                                     struct neigh_table *tbl,
1805                                     struct neigh_parms *parms,
1806                                     u32 pid, u32 seq, int type,
1807                                     unsigned int flags)
1808 {
1809         struct ndtmsg *ndtmsg;
1810         struct nlmsghdr *nlh;
1811
1812         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1813         if (nlh == NULL)
1814                 return -EMSGSIZE;
1815
1816         ndtmsg = nlmsg_data(nlh);
1817
1818         read_lock_bh(&tbl->lock);
1819         ndtmsg->ndtm_family = tbl->family;
1820         ndtmsg->ndtm_pad1   = 0;
1821         ndtmsg->ndtm_pad2   = 0;
1822
1823         if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 ||
1824             neightbl_fill_parms(skb, parms) < 0)
1825                 goto errout;
1826
1827         read_unlock_bh(&tbl->lock);
1828         return nlmsg_end(skb, nlh);
1829 errout:
1830         read_unlock_bh(&tbl->lock);
1831         nlmsg_cancel(skb, nlh);
1832         return -EMSGSIZE;
1833 }
1834
1835 static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = {
1836         [NDTA_NAME]             = { .type = NLA_STRING },
1837         [NDTA_THRESH1]          = { .type = NLA_U32 },
1838         [NDTA_THRESH2]          = { .type = NLA_U32 },
1839         [NDTA_THRESH3]          = { .type = NLA_U32 },
1840         [NDTA_GC_INTERVAL]      = { .type = NLA_U64 },
1841         [NDTA_PARMS]            = { .type = NLA_NESTED },
1842 };
1843
1844 static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = {
1845         [NDTPA_IFINDEX]                 = { .type = NLA_U32 },
1846         [NDTPA_QUEUE_LEN]               = { .type = NLA_U32 },
1847         [NDTPA_PROXY_QLEN]              = { .type = NLA_U32 },
1848         [NDTPA_APP_PROBES]              = { .type = NLA_U32 },
1849         [NDTPA_UCAST_PROBES]            = { .type = NLA_U32 },
1850         [NDTPA_MCAST_PROBES]            = { .type = NLA_U32 },
1851         [NDTPA_BASE_REACHABLE_TIME]     = { .type = NLA_U64 },
1852         [NDTPA_GC_STALETIME]            = { .type = NLA_U64 },
1853         [NDTPA_DELAY_PROBE_TIME]        = { .type = NLA_U64 },
1854         [NDTPA_RETRANS_TIME]            = { .type = NLA_U64 },
1855         [NDTPA_ANYCAST_DELAY]           = { .type = NLA_U64 },
1856         [NDTPA_PROXY_DELAY]             = { .type = NLA_U64 },
1857         [NDTPA_LOCKTIME]                = { .type = NLA_U64 },
1858 };
1859
1860 static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1861 {
1862         struct net *net = sock_net(skb->sk);
1863         struct neigh_table *tbl;
1864         struct ndtmsg *ndtmsg;
1865         struct nlattr *tb[NDTA_MAX+1];
1866         int err;
1867
1868         err = nlmsg_parse(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
1869                           nl_neightbl_policy);
1870         if (err < 0)
1871                 goto errout;
1872
1873         if (tb[NDTA_NAME] == NULL) {
1874                 err = -EINVAL;
1875                 goto errout;
1876         }
1877
1878         ndtmsg = nlmsg_data(nlh);
1879         read_lock(&neigh_tbl_lock);
1880         for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1881                 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
1882                         continue;
1883
1884                 if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0)
1885                         break;
1886         }
1887
1888         if (tbl == NULL) {
1889                 err = -ENOENT;
1890                 goto errout_locked;
1891         }
1892
1893         /*
1894          * We acquire tbl->lock to be nice to the periodic timers and
1895          * make sure they always see a consistent set of values.
1896          */
1897         write_lock_bh(&tbl->lock);
1898
1899         if (tb[NDTA_PARMS]) {
1900                 struct nlattr *tbp[NDTPA_MAX+1];
1901                 struct neigh_parms *p;
1902                 int i, ifindex = 0;
1903
1904                 err = nla_parse_nested(tbp, NDTPA_MAX, tb[NDTA_PARMS],
1905                                        nl_ntbl_parm_policy);
1906                 if (err < 0)
1907                         goto errout_tbl_lock;
1908
1909                 if (tbp[NDTPA_IFINDEX])
1910                         ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
1911
1912                 p = lookup_neigh_params(tbl, net, ifindex);
1913                 if (p == NULL) {
1914                         err = -ENOENT;
1915                         goto errout_tbl_lock;
1916                 }
1917
1918                 for (i = 1; i <= NDTPA_MAX; i++) {
1919                         if (tbp[i] == NULL)
1920                                 continue;
1921
1922                         switch (i) {
1923                         case NDTPA_QUEUE_LEN:
1924                                 p->queue_len = nla_get_u32(tbp[i]);
1925                                 break;
1926                         case NDTPA_PROXY_QLEN:
1927                                 p->proxy_qlen = nla_get_u32(tbp[i]);
1928                                 break;
1929                         case NDTPA_APP_PROBES:
1930                                 p->app_probes = nla_get_u32(tbp[i]);
1931                                 break;
1932                         case NDTPA_UCAST_PROBES:
1933                                 p->ucast_probes = nla_get_u32(tbp[i]);
1934                                 break;
1935                         case NDTPA_MCAST_PROBES:
1936                                 p->mcast_probes = nla_get_u32(tbp[i]);
1937                                 break;
1938                         case NDTPA_BASE_REACHABLE_TIME:
1939                                 p->base_reachable_time = nla_get_msecs(tbp[i]);
1940                                 break;
1941                         case NDTPA_GC_STALETIME:
1942                                 p->gc_staletime = nla_get_msecs(tbp[i]);
1943                                 break;
1944                         case NDTPA_DELAY_PROBE_TIME:
1945                                 p->delay_probe_time = nla_get_msecs(tbp[i]);
1946                                 break;
1947                         case NDTPA_RETRANS_TIME:
1948                                 p->retrans_time = nla_get_msecs(tbp[i]);
1949                                 break;
1950                         case NDTPA_ANYCAST_DELAY:
1951                                 p->anycast_delay = nla_get_msecs(tbp[i]);
1952                                 break;
1953                         case NDTPA_PROXY_DELAY:
1954                                 p->proxy_delay = nla_get_msecs(tbp[i]);
1955                                 break;
1956                         case NDTPA_LOCKTIME:
1957                                 p->locktime = nla_get_msecs(tbp[i]);
1958                                 break;
1959                         }
1960                 }
1961         }
1962
1963         if (tb[NDTA_THRESH1])
1964                 tbl->gc_thresh1 = nla_get_u32(tb[NDTA_THRESH1]);
1965
1966         if (tb[NDTA_THRESH2])
1967                 tbl->gc_thresh2 = nla_get_u32(tb[NDTA_THRESH2]);
1968
1969         if (tb[NDTA_THRESH3])
1970                 tbl->gc_thresh3 = nla_get_u32(tb[NDTA_THRESH3]);
1971
1972         if (tb[NDTA_GC_INTERVAL])
1973                 tbl->gc_interval = nla_get_msecs(tb[NDTA_GC_INTERVAL]);
1974
1975         err = 0;
1976
1977 errout_tbl_lock:
1978         write_unlock_bh(&tbl->lock);
1979 errout_locked:
1980         read_unlock(&neigh_tbl_lock);
1981 errout:
1982         return err;
1983 }
1984
1985 static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
1986 {
1987         struct net *net = sock_net(skb->sk);
1988         int family, tidx, nidx = 0;
1989         int tbl_skip = cb->args[0];
1990         int neigh_skip = cb->args[1];
1991         struct neigh_table *tbl;
1992
1993         family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
1994
1995         read_lock(&neigh_tbl_lock);
1996         for (tbl = neigh_tables, tidx = 0; tbl; tbl = tbl->next, tidx++) {
1997                 struct neigh_parms *p;
1998
1999                 if (tidx < tbl_skip || (family && tbl->family != family))
2000                         continue;
2001
2002                 if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).pid,
2003                                        cb->nlh->nlmsg_seq, RTM_NEWNEIGHTBL,
2004                                        NLM_F_MULTI) <= 0)
2005                         break;
2006
2007                 for (nidx = 0, p = tbl->parms.next; p; p = p->next) {
2008                         if (!net_eq(neigh_parms_net(p), net))
2009                                 continue;
2010
2011                         if (nidx++ < neigh_skip)
2012                                 continue;
2013
2014                         if (neightbl_fill_param_info(skb, tbl, p,
2015                                                      NETLINK_CB(cb->skb).pid,
2016                                                      cb->nlh->nlmsg_seq,
2017                                                      RTM_NEWNEIGHTBL,
2018                                                      NLM_F_MULTI) <= 0)
2019                                 goto out;
2020                 }
2021
2022                 neigh_skip = 0;
2023         }
2024 out:
2025         read_unlock(&neigh_tbl_lock);
2026         cb->args[0] = tidx;
2027         cb->args[1] = nidx;
2028
2029         return skb->len;
2030 }
2031
2032 static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
2033                            u32 pid, u32 seq, int type, unsigned int flags)
2034 {
2035         unsigned long now = jiffies;
2036         struct nda_cacheinfo ci;
2037         struct nlmsghdr *nlh;
2038         struct ndmsg *ndm;
2039
2040         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2041         if (nlh == NULL)
2042                 return -EMSGSIZE;
2043
2044         ndm = nlmsg_data(nlh);
2045         ndm->ndm_family  = neigh->ops->family;
2046         ndm->ndm_pad1    = 0;
2047         ndm->ndm_pad2    = 0;
2048         ndm->ndm_flags   = neigh->flags;
2049         ndm->ndm_type    = neigh->type;
2050         ndm->ndm_ifindex = neigh->dev->ifindex;
2051
2052         NLA_PUT(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key);
2053
2054         read_lock_bh(&neigh->lock);
2055         ndm->ndm_state   = neigh->nud_state;
2056         if ((neigh->nud_state & NUD_VALID) &&
2057             nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, neigh->ha) < 0) {
2058                 read_unlock_bh(&neigh->lock);
2059                 goto nla_put_failure;
2060         }
2061
2062         ci.ndm_used      = jiffies_to_clock_t(now - neigh->used);
2063         ci.ndm_confirmed = jiffies_to_clock_t(now - neigh->confirmed);
2064         ci.ndm_updated   = jiffies_to_clock_t(now - neigh->updated);
2065         ci.ndm_refcnt    = atomic_read(&neigh->refcnt) - 1;
2066         read_unlock_bh(&neigh->lock);
2067
2068         NLA_PUT_U32(skb, NDA_PROBES, atomic_read(&neigh->probes));
2069         NLA_PUT(skb, NDA_CACHEINFO, sizeof(ci), &ci);
2070
2071         return nlmsg_end(skb, nlh);
2072
2073 nla_put_failure:
2074         nlmsg_cancel(skb, nlh);
2075         return -EMSGSIZE;
2076 }
2077
2078 static void neigh_update_notify(struct neighbour *neigh)
2079 {
2080         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
2081         __neigh_notify(neigh, RTM_NEWNEIGH, 0);
2082 }
2083
2084 static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2085                             struct netlink_callback *cb)
2086 {
2087         struct net * net = sock_net(skb->sk);
2088         struct neighbour *n;
2089         int rc, h, s_h = cb->args[1];
2090         int idx, s_idx = idx = cb->args[2];
2091
2092         read_lock_bh(&tbl->lock);
2093         for (h = 0; h <= tbl->hash_mask; h++) {
2094                 if (h < s_h)
2095                         continue;
2096                 if (h > s_h)
2097                         s_idx = 0;
2098                 for (n = tbl->hash_buckets[h], idx = 0; n; n = n->next) {
2099                         int lidx;
2100                         if (dev_net(n->dev) != net)
2101                                 continue;
2102                         lidx = idx++;
2103                         if (lidx < s_idx)
2104                                 continue;
2105                         if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).pid,
2106                                             cb->nlh->nlmsg_seq,
2107                                             RTM_NEWNEIGH,
2108                                             NLM_F_MULTI) <= 0) {
2109                                 read_unlock_bh(&tbl->lock);
2110                                 rc = -1;
2111                                 goto out;
2112                         }
2113                 }
2114         }
2115         read_unlock_bh(&tbl->lock);
2116         rc = skb->len;
2117 out:
2118         cb->args[1] = h;
2119         cb->args[2] = idx;
2120         return rc;
2121 }
2122
2123 static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2124 {
2125         struct neigh_table *tbl;
2126         int t, family, s_t;
2127
2128         read_lock(&neigh_tbl_lock);
2129         family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2130         s_t = cb->args[0];
2131
2132         for (tbl = neigh_tables, t = 0; tbl; tbl = tbl->next, t++) {
2133                 if (t < s_t || (family && tbl->family != family))
2134                         continue;
2135                 if (t > s_t)
2136                         memset(&cb->args[1], 0, sizeof(cb->args) -
2137                                                 sizeof(cb->args[0]));
2138                 if (neigh_dump_table(tbl, skb, cb) < 0)
2139                         break;
2140         }
2141         read_unlock(&neigh_tbl_lock);
2142
2143         cb->args[0] = t;
2144         return skb->len;
2145 }
2146
2147 void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
2148 {
2149         int chain;
2150
2151         read_lock_bh(&tbl->lock);
2152         for (chain = 0; chain <= tbl->hash_mask; chain++) {
2153                 struct neighbour *n;
2154
2155                 for (n = tbl->hash_buckets[chain]; n; n = n->next)
2156                         cb(n, cookie);
2157         }
2158         read_unlock_bh(&tbl->lock);
2159 }
2160 EXPORT_SYMBOL(neigh_for_each);
2161
2162 /* The tbl->lock must be held as a writer and BH disabled. */
2163 void __neigh_for_each_release(struct neigh_table *tbl,
2164                               int (*cb)(struct neighbour *))
2165 {
2166         int chain;
2167
2168         for (chain = 0; chain <= tbl->hash_mask; chain++) {
2169                 struct neighbour *n, **np;
2170
2171                 np = &tbl->hash_buckets[chain];
2172                 while ((n = *np) != NULL) {
2173                         int release;
2174
2175                         write_lock(&n->lock);
2176                         release = cb(n);
2177                         if (release) {
2178                                 *np = n->next;
2179                                 n->dead = 1;
2180                         } else
2181                                 np = &n->next;
2182                         write_unlock(&n->lock);
2183                         if (release)
2184                                 neigh_cleanup_and_release(n);
2185                 }
2186         }
2187 }
2188 EXPORT_SYMBOL(__neigh_for_each_release);
2189
2190 #ifdef CONFIG_PROC_FS
2191
2192 static struct neighbour *neigh_get_first(struct seq_file *seq)
2193 {
2194         struct neigh_seq_state *state = seq->private;
2195         struct net *net = seq_file_net(seq);
2196         struct neigh_table *tbl = state->tbl;
2197         struct neighbour *n = NULL;
2198         int bucket = state->bucket;
2199
2200         state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
2201         for (bucket = 0; bucket <= tbl->hash_mask; bucket++) {
2202                 n = tbl->hash_buckets[bucket];
2203
2204                 while (n) {
2205                         if (!net_eq(dev_net(n->dev), net))
2206                                 goto next;
2207                         if (state->neigh_sub_iter) {
2208                                 loff_t fakep = 0;
2209                                 void *v;
2210
2211                                 v = state->neigh_sub_iter(state, n, &fakep);
2212                                 if (!v)
2213                                         goto next;
2214                         }
2215                         if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2216                                 break;
2217                         if (n->nud_state & ~NUD_NOARP)
2218                                 break;
2219                 next:
2220                         n = n->next;
2221                 }
2222
2223                 if (n)
2224                         break;
2225         }
2226         state->bucket = bucket;
2227
2228         return n;
2229 }
2230
2231 static struct neighbour *neigh_get_next(struct seq_file *seq,
2232                                         struct neighbour *n,
2233                                         loff_t *pos)
2234 {
2235         struct neigh_seq_state *state = seq->private;
2236         struct net *net = seq_file_net(seq);
2237         struct neigh_table *tbl = state->tbl;
2238
2239         if (state->neigh_sub_iter) {
2240                 void *v = state->neigh_sub_iter(state, n, pos);
2241                 if (v)
2242                         return n;
2243         }
2244         n = n->next;
2245
2246         while (1) {
2247                 while (n) {
2248                         if (!net_eq(dev_net(n->dev), net))
2249                                 goto next;
2250                         if (state->neigh_sub_iter) {
2251                                 void *v = state->neigh_sub_iter(state, n, pos);
2252                                 if (v)
2253                                         return n;
2254                                 goto next;
2255                         }
2256                         if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2257                                 break;
2258
2259                         if (n->nud_state & ~NUD_NOARP)
2260                                 break;
2261                 next:
2262                         n = n->next;
2263                 }
2264
2265                 if (n)
2266                         break;
2267
2268                 if (++state->bucket > tbl->hash_mask)
2269                         break;
2270
2271                 n = tbl->hash_buckets[state->bucket];
2272         }
2273
2274         if (n && pos)
2275                 --(*pos);
2276         return n;
2277 }
2278
2279 static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
2280 {
2281         struct neighbour *n = neigh_get_first(seq);
2282
2283         if (n) {
2284                 while (*pos) {
2285                         n = neigh_get_next(seq, n, pos);
2286                         if (!n)
2287                                 break;
2288                 }
2289         }
2290         return *pos ? NULL : n;
2291 }
2292
2293 static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
2294 {
2295         struct neigh_seq_state *state = seq->private;
2296         struct net *net = seq_file_net(seq);
2297         struct neigh_table *tbl = state->tbl;
2298         struct pneigh_entry *pn = NULL;
2299         int bucket = state->bucket;
2300
2301         state->flags |= NEIGH_SEQ_IS_PNEIGH;
2302         for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
2303                 pn = tbl->phash_buckets[bucket];
2304                 while (pn && !net_eq(pneigh_net(pn), net))
2305                         pn = pn->next;
2306                 if (pn)
2307                         break;
2308         }
2309         state->bucket = bucket;
2310
2311         return pn;
2312 }
2313
2314 static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
2315                                             struct pneigh_entry *pn,
2316                                             loff_t *pos)
2317 {
2318         struct neigh_seq_state *state = seq->private;
2319         struct net *net = seq_file_net(seq);
2320         struct neigh_table *tbl = state->tbl;
2321
2322         pn = pn->next;
2323         while (!pn) {
2324                 if (++state->bucket > PNEIGH_HASHMASK)
2325                         break;
2326                 pn = tbl->phash_buckets[state->bucket];
2327                 while (pn && !net_eq(pneigh_net(pn), net))
2328                         pn = pn->next;
2329                 if (pn)
2330                         break;
2331         }
2332
2333         if (pn && pos)
2334                 --(*pos);
2335
2336         return pn;
2337 }
2338
2339 static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
2340 {
2341         struct pneigh_entry *pn = pneigh_get_first(seq);
2342
2343         if (pn) {
2344                 while (*pos) {
2345                         pn = pneigh_get_next(seq, pn, pos);
2346                         if (!pn)
2347                                 break;
2348                 }
2349         }
2350         return *pos ? NULL : pn;
2351 }
2352
2353 static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
2354 {
2355         struct neigh_seq_state *state = seq->private;
2356         void *rc;
2357
2358         rc = neigh_get_idx(seq, pos);
2359         if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2360                 rc = pneigh_get_idx(seq, pos);
2361
2362         return rc;
2363 }
2364
2365 void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
2366         __acquires(tbl->lock)
2367 {
2368         struct neigh_seq_state *state = seq->private;
2369         loff_t pos_minus_one;
2370
2371         state->tbl = tbl;
2372         state->bucket = 0;
2373         state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
2374
2375         read_lock_bh(&tbl->lock);
2376
2377         pos_minus_one = *pos - 1;
2378         return *pos ? neigh_get_idx_any(seq, &pos_minus_one) : SEQ_START_TOKEN;
2379 }
2380 EXPORT_SYMBOL(neigh_seq_start);
2381
2382 void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2383 {
2384         struct neigh_seq_state *state;
2385         void *rc;
2386
2387         if (v == SEQ_START_TOKEN) {
2388                 rc = neigh_get_idx(seq, pos);
2389                 goto out;
2390         }
2391
2392         state = seq->private;
2393         if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
2394                 rc = neigh_get_next(seq, v, NULL);
2395                 if (rc)
2396                         goto out;
2397                 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2398                         rc = pneigh_get_first(seq);
2399         } else {
2400                 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
2401                 rc = pneigh_get_next(seq, v, NULL);
2402         }
2403 out:
2404         ++(*pos);
2405         return rc;
2406 }
2407 EXPORT_SYMBOL(neigh_seq_next);
2408
2409 void neigh_seq_stop(struct seq_file *seq, void *v)
2410         __releases(tbl->lock)
2411 {
2412         struct neigh_seq_state *state = seq->private;
2413         struct neigh_table *tbl = state->tbl;
2414
2415         read_unlock_bh(&tbl->lock);
2416 }
2417 EXPORT_SYMBOL(neigh_seq_stop);
2418
2419 /* statistics via seq_file */
2420
2421 static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
2422 {
2423         struct proc_dir_entry *pde = seq->private;
2424         struct neigh_table *tbl = pde->data;
2425         int cpu;
2426
2427         if (*pos == 0)
2428                 return SEQ_START_TOKEN;
2429
2430         for (cpu = *pos-1; cpu < NR_CPUS; ++cpu) {
2431                 if (!cpu_possible(cpu))
2432                         continue;
2433                 *pos = cpu+1;
2434                 return per_cpu_ptr(tbl->stats, cpu);
2435         }
2436         return NULL;
2437 }
2438
2439 static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2440 {
2441         struct proc_dir_entry *pde = seq->private;
2442         struct neigh_table *tbl = pde->data;
2443         int cpu;
2444
2445         for (cpu = *pos; cpu < NR_CPUS; ++cpu) {
2446                 if (!cpu_possible(cpu))
2447                         continue;
2448                 *pos = cpu+1;
2449                 return per_cpu_ptr(tbl->stats, cpu);
2450         }
2451         return NULL;
2452 }
2453
2454 static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
2455 {
2456
2457 }
2458
2459 static int neigh_stat_seq_show(struct seq_file *seq, void *v)
2460 {
2461         struct proc_dir_entry *pde = seq->private;
2462         struct neigh_table *tbl = pde->data;
2463         struct neigh_statistics *st = v;
2464
2465         if (v == SEQ_START_TOKEN) {
2466                 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");
2467                 return 0;
2468         }
2469
2470         seq_printf(seq, "%08x  %08lx %08lx %08lx  %08lx %08lx  %08lx  "
2471                         "%08lx %08lx  %08lx %08lx %08lx\n",
2472                    atomic_read(&tbl->entries),
2473
2474                    st->allocs,
2475                    st->destroys,
2476                    st->hash_grows,
2477
2478                    st->lookups,
2479                    st->hits,
2480
2481                    st->res_failed,
2482
2483                    st->rcv_probes_mcast,
2484                    st->rcv_probes_ucast,
2485
2486                    st->periodic_gc_runs,
2487                    st->forced_gc_runs,
2488                    st->unres_discards
2489                    );
2490
2491         return 0;
2492 }
2493
2494 static const struct seq_operations neigh_stat_seq_ops = {
2495         .start  = neigh_stat_seq_start,
2496         .next   = neigh_stat_seq_next,
2497         .stop   = neigh_stat_seq_stop,
2498         .show   = neigh_stat_seq_show,
2499 };
2500
2501 static int neigh_stat_seq_open(struct inode *inode, struct file *file)
2502 {
2503         int ret = seq_open(file, &neigh_stat_seq_ops);
2504
2505         if (!ret) {
2506                 struct seq_file *sf = file->private_data;
2507                 sf->private = PDE(inode);
2508         }
2509         return ret;
2510 };
2511
2512 static const struct file_operations neigh_stat_seq_fops = {
2513         .owner   = THIS_MODULE,
2514         .open    = neigh_stat_seq_open,
2515         .read    = seq_read,
2516         .llseek  = seq_lseek,
2517         .release = seq_release,
2518 };
2519
2520 #endif /* CONFIG_PROC_FS */
2521
2522 static inline size_t neigh_nlmsg_size(void)
2523 {
2524         return NLMSG_ALIGN(sizeof(struct ndmsg))
2525                + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2526                + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */
2527                + nla_total_size(sizeof(struct nda_cacheinfo))
2528                + nla_total_size(4); /* NDA_PROBES */
2529 }
2530
2531 static void __neigh_notify(struct neighbour *n, int type, int flags)
2532 {
2533         struct net *net = dev_net(n->dev);
2534         struct sk_buff *skb;
2535         int err = -ENOBUFS;
2536
2537         skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC);
2538         if (skb == NULL)
2539                 goto errout;
2540
2541         err = neigh_fill_info(skb, n, 0, 0, type, flags);
2542         if (err < 0) {
2543                 /* -EMSGSIZE implies BUG in neigh_nlmsg_size() */
2544                 WARN_ON(err == -EMSGSIZE);
2545                 kfree_skb(skb);
2546                 goto errout;
2547         }
2548         err = rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
2549 errout:
2550         if (err < 0)
2551                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
2552 }
2553
2554 #ifdef CONFIG_ARPD
2555 void neigh_app_ns(struct neighbour *n)
2556 {
2557         __neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST);
2558 }
2559 EXPORT_SYMBOL(neigh_app_ns);
2560 #endif /* CONFIG_ARPD */
2561
2562 #ifdef CONFIG_SYSCTL
2563
2564 static struct neigh_sysctl_table {
2565         struct ctl_table_header *sysctl_header;
2566         struct ctl_table neigh_vars[__NET_NEIGH_MAX];
2567         char *dev_name;
2568 } neigh_sysctl_template __read_mostly = {
2569         .neigh_vars = {
2570                 {
2571                         .ctl_name       = NET_NEIGH_MCAST_SOLICIT,
2572                         .procname       = "mcast_solicit",
2573                         .maxlen         = sizeof(int),
2574                         .mode           = 0644,
2575                         .proc_handler   = &proc_dointvec,
2576                 },
2577                 {
2578                         .ctl_name       = NET_NEIGH_UCAST_SOLICIT,
2579                         .procname       = "ucast_solicit",
2580                         .maxlen         = sizeof(int),
2581                         .mode           = 0644,
2582                         .proc_handler   = &proc_dointvec,
2583                 },
2584                 {
2585                         .ctl_name       = NET_NEIGH_APP_SOLICIT,
2586                         .procname       = "app_solicit",
2587                         .maxlen         = sizeof(int),
2588                         .mode           = 0644,
2589                         .proc_handler   = &proc_dointvec,
2590                 },
2591                 {
2592                         .procname       = "retrans_time",
2593                         .maxlen         = sizeof(int),
2594                         .mode           = 0644,
2595                         .proc_handler   = &proc_dointvec_userhz_jiffies,
2596                 },
2597                 {
2598                         .ctl_name       = NET_NEIGH_REACHABLE_TIME,
2599                         .procname       = "base_reachable_time",
2600                         .maxlen         = sizeof(int),
2601                         .mode           = 0644,
2602                         .proc_handler   = &proc_dointvec_jiffies,
2603                         .strategy       = &sysctl_jiffies,
2604                 },
2605                 {
2606                         .ctl_name       = NET_NEIGH_DELAY_PROBE_TIME,
2607                         .procname       = "delay_first_probe_time",
2608                         .maxlen         = sizeof(int),
2609                         .mode           = 0644,
2610                         .proc_handler   = &proc_dointvec_jiffies,
2611                         .strategy       = &sysctl_jiffies,
2612                 },
2613                 {
2614                         .ctl_name       = NET_NEIGH_GC_STALE_TIME,
2615                         .procname       = "gc_stale_time",
2616                         .maxlen         = sizeof(int),
2617                         .mode           = 0644,
2618                         .proc_handler   = &proc_dointvec_jiffies,
2619                         .strategy       = &sysctl_jiffies,
2620                 },
2621                 {
2622                         .ctl_name       = NET_NEIGH_UNRES_QLEN,
2623                         .procname       = "unres_qlen",
2624                         .maxlen         = sizeof(int),
2625                         .mode           = 0644,
2626                         .proc_handler   = &proc_dointvec,
2627                 },
2628                 {
2629                         .ctl_name       = NET_NEIGH_PROXY_QLEN,
2630                         .procname       = "proxy_qlen",
2631                         .maxlen         = sizeof(int),
2632                         .mode           = 0644,
2633                         .proc_handler   = &proc_dointvec,
2634                 },
2635                 {
2636                         .procname       = "anycast_delay",
2637                         .maxlen         = sizeof(int),
2638                         .mode           = 0644,
2639                         .proc_handler   = &proc_dointvec_userhz_jiffies,
2640                 },
2641                 {
2642                         .procname       = "proxy_delay",
2643                         .maxlen         = sizeof(int),
2644                         .mode           = 0644,
2645                         .proc_handler   = &proc_dointvec_userhz_jiffies,
2646                 },
2647                 {
2648                         .procname       = "locktime",
2649                         .maxlen         = sizeof(int),
2650                         .mode           = 0644,
2651                         .proc_handler   = &proc_dointvec_userhz_jiffies,
2652                 },
2653                 {
2654                         .ctl_name       = NET_NEIGH_RETRANS_TIME_MS,
2655                         .procname       = "retrans_time_ms",
2656                         .maxlen         = sizeof(int),
2657                         .mode           = 0644,
2658                         .proc_handler   = &proc_dointvec_ms_jiffies,
2659                         .strategy       = &sysctl_ms_jiffies,
2660                 },
2661                 {
2662                         .ctl_name       = NET_NEIGH_REACHABLE_TIME_MS,
2663                         .procname       = "base_reachable_time_ms",
2664                         .maxlen         = sizeof(int),
2665                         .mode           = 0644,
2666                         .proc_handler   = &proc_dointvec_ms_jiffies,
2667                         .strategy       = &sysctl_ms_jiffies,
2668                 },
2669                 {
2670                         .ctl_name       = NET_NEIGH_GC_INTERVAL,
2671                         .procname       = "gc_interval",
2672                         .maxlen         = sizeof(int),
2673                         .mode           = 0644,
2674                         .proc_handler   = &proc_dointvec_jiffies,
2675                         .strategy       = &sysctl_jiffies,
2676                 },
2677                 {
2678                         .ctl_name       = NET_NEIGH_GC_THRESH1,
2679                         .procname       = "gc_thresh1",
2680                         .maxlen         = sizeof(int),
2681                         .mode           = 0644,
2682                         .proc_handler   = &proc_dointvec,
2683                 },
2684                 {
2685                         .ctl_name       = NET_NEIGH_GC_THRESH2,
2686                         .procname       = "gc_thresh2",
2687                         .maxlen         = sizeof(int),
2688                         .mode           = 0644,
2689                         .proc_handler   = &proc_dointvec,
2690                 },
2691                 {
2692                         .ctl_name       = NET_NEIGH_GC_THRESH3,
2693                         .procname       = "gc_thresh3",
2694                         .maxlen         = sizeof(int),
2695                         .mode           = 0644,
2696                         .proc_handler   = &proc_dointvec,
2697                 },
2698                 {},
2699         },
2700 };
2701
2702 int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
2703                           int p_id, int pdev_id, char *p_name,
2704                           proc_handler *handler, ctl_handler *strategy)
2705 {
2706         struct neigh_sysctl_table *t;
2707         const char *dev_name_source = NULL;
2708
2709 #define NEIGH_CTL_PATH_ROOT     0
2710 #define NEIGH_CTL_PATH_PROTO    1
2711 #define NEIGH_CTL_PATH_NEIGH    2
2712 #define NEIGH_CTL_PATH_DEV      3
2713
2714         struct ctl_path neigh_path[] = {
2715                 { .procname = "net",     .ctl_name = CTL_NET, },
2716                 { .procname = "proto",   .ctl_name = 0, },
2717                 { .procname = "neigh",   .ctl_name = 0, },
2718                 { .procname = "default", .ctl_name = NET_PROTO_CONF_DEFAULT, },
2719                 { },
2720         };
2721
2722         t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL);
2723         if (!t)
2724                 goto err;
2725
2726         t->neigh_vars[0].data  = &p->mcast_probes;
2727         t->neigh_vars[1].data  = &p->ucast_probes;
2728         t->neigh_vars[2].data  = &p->app_probes;
2729         t->neigh_vars[3].data  = &p->retrans_time;
2730         t->neigh_vars[4].data  = &p->base_reachable_time;
2731         t->neigh_vars[5].data  = &p->delay_probe_time;
2732         t->neigh_vars[6].data  = &p->gc_staletime;
2733         t->neigh_vars[7].data  = &p->queue_len;
2734         t->neigh_vars[8].data  = &p->proxy_qlen;
2735         t->neigh_vars[9].data  = &p->anycast_delay;
2736         t->neigh_vars[10].data = &p->proxy_delay;
2737         t->neigh_vars[11].data = &p->locktime;
2738         t->neigh_vars[12].data  = &p->retrans_time;
2739         t->neigh_vars[13].data  = &p->base_reachable_time;
2740
2741         if (dev) {
2742                 dev_name_source = dev->name;
2743                 neigh_path[NEIGH_CTL_PATH_DEV].ctl_name = dev->ifindex;
2744                 /* Terminate the table early */
2745                 memset(&t->neigh_vars[14], 0, sizeof(t->neigh_vars[14]));
2746         } else {
2747                 dev_name_source = neigh_path[NEIGH_CTL_PATH_DEV].procname;
2748                 t->neigh_vars[14].data = (int *)(p + 1);
2749                 t->neigh_vars[15].data = (int *)(p + 1) + 1;
2750                 t->neigh_vars[16].data = (int *)(p + 1) + 2;
2751                 t->neigh_vars[17].data = (int *)(p + 1) + 3;
2752         }
2753
2754
2755         if (handler || strategy) {
2756                 /* RetransTime */
2757                 t->neigh_vars[3].proc_handler = handler;
2758                 t->neigh_vars[3].strategy = strategy;
2759                 t->neigh_vars[3].extra1 = dev;
2760                 if (!strategy)
2761                         t->neigh_vars[3].ctl_name = CTL_UNNUMBERED;
2762                 /* ReachableTime */
2763                 t->neigh_vars[4].proc_handler = handler;
2764                 t->neigh_vars[4].strategy = strategy;
2765                 t->neigh_vars[4].extra1 = dev;
2766                 if (!strategy)
2767                         t->neigh_vars[4].ctl_name = CTL_UNNUMBERED;
2768                 /* RetransTime (in milliseconds)*/
2769                 t->neigh_vars[12].proc_handler = handler;
2770                 t->neigh_vars[12].strategy = strategy;
2771                 t->neigh_vars[12].extra1 = dev;
2772                 if (!strategy)
2773                         t->neigh_vars[12].ctl_name = CTL_UNNUMBERED;
2774                 /* ReachableTime (in milliseconds) */
2775                 t->neigh_vars[13].proc_handler = handler;
2776                 t->neigh_vars[13].strategy = strategy;
2777                 t->neigh_vars[13].extra1 = dev;
2778                 if (!strategy)
2779                         t->neigh_vars[13].ctl_name = CTL_UNNUMBERED;
2780         }
2781
2782         t->dev_name = kstrdup(dev_name_source, GFP_KERNEL);
2783         if (!t->dev_name)
2784                 goto free;
2785
2786         neigh_path[NEIGH_CTL_PATH_DEV].procname = t->dev_name;
2787         neigh_path[NEIGH_CTL_PATH_NEIGH].ctl_name = pdev_id;
2788         neigh_path[NEIGH_CTL_PATH_PROTO].procname = p_name;
2789         neigh_path[NEIGH_CTL_PATH_PROTO].ctl_name = p_id;
2790
2791         t->sysctl_header =
2792                 register_net_sysctl_table(neigh_parms_net(p), neigh_path, t->neigh_vars);
2793         if (!t->sysctl_header)
2794                 goto free_procname;
2795
2796         p->sysctl_table = t;
2797         return 0;
2798
2799 free_procname:
2800         kfree(t->dev_name);
2801 free:
2802         kfree(t);
2803 err:
2804         return -ENOBUFS;
2805 }
2806 EXPORT_SYMBOL(neigh_sysctl_register);
2807
2808 void neigh_sysctl_unregister(struct neigh_parms *p)
2809 {
2810         if (p->sysctl_table) {
2811                 struct neigh_sysctl_table *t = p->sysctl_table;
2812                 p->sysctl_table = NULL;
2813                 unregister_sysctl_table(t->sysctl_header);
2814                 kfree(t->dev_name);
2815                 kfree(t);
2816         }
2817 }
2818 EXPORT_SYMBOL(neigh_sysctl_unregister);
2819
2820 #endif  /* CONFIG_SYSCTL */
2821
2822 static int __init neigh_init(void)
2823 {
2824         rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL);
2825         rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL);
2826         rtnl_register(PF_UNSPEC, RTM_GETNEIGH, NULL, neigh_dump_info);
2827
2828         rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info);
2829         rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL);
2830
2831         return 0;
2832 }
2833
2834 subsys_initcall(neigh_init);
2835