projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
neigh: do not modify unlinked entries
[linux-2.6-block.git] / net / core / neighbour.c
CommitLineData
1da177e4
LT		 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
e005d193
JP		 18#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
19
5a0e3ad6 20#include <linux/slab.h>
1da177e4
LT		 21#include <linux/types.h>
22#include <linux/kernel.h>
23#include <linux/module.h>
24#include <linux/socket.h>
1da177e4
LT		 25#include <linux/netdevice.h>
26#include <linux/proc_fs.h>
27#ifdef CONFIG_SYSCTL
28#include <linux/sysctl.h>
29#endif
30#include <linux/times.h>
457c4cbc 31#include <net/net_namespace.h>
1da177e4
LT		 32#include <net/neighbour.h>
33#include <net/dst.h>
34#include <net/sock.h>
8d71740c 35#include <net/netevent.h>
a14a49d2 36#include <net/netlink.h>
1da177e4
LT		 37#include <linux/rtnetlink.h>
38#include <linux/random.h>
543537bd 39#include <linux/string.h>
c3609d51 40#include <linux/log2.h>
1d4c8c29 41#include <linux/inetdevice.h>
bba24896 42#include <net/addrconf.h>
1da177e4 43
d5d427cd 44#define DEBUG
1da177e4 45#define NEIGH_DEBUG 1
d5d427cd
JP		 46#define neigh_dbg(level, fmt, ...) \
47do { \
48 if (level <= NEIGH_DEBUG) \
49 pr_debug(fmt, ##__VA_ARGS__); \
50} while (0)
1da177e4
LT		 51
52#define PNEIGH_HASHMASK 0xF
53
54static void neigh_timer_handler(unsigned long arg);
d961db35
TG		 55static void __neigh_notify(struct neighbour *n, int type, int flags);
56static void neigh_update_notify(struct neighbour *neigh);
1da177e4 57static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev);
1da177e4 58
45fc3b11 59#ifdef CONFIG_PROC_FS
9a32144e 60static const struct file_operations neigh_stat_seq_fops;
45fc3b11 61#endif
1da177e4
LT		 62
63/*
64 Neighbour hash table buckets are protected with rwlock tbl->lock.
65
66 - All the scans/updates to hash buckets MUST be made under this lock.
67 - NOTHING clever should be made under this lock: no callbacks
68 to protocol backends, no attempts to send something to network.
69 It will result in deadlocks, if backend/driver wants to use neighbour
70 cache.
71 - If the entry requires some non-trivial actions, increase
72 its reference count and release table lock.
73
74 Neighbour entries are protected:
75 - with reference count.
76 - with rwlock neigh->lock
77
78 Reference count prevents destruction.
79
80 neigh->lock mainly serializes ll address data and its validity state.
81 However, the same lock is used to protect another entry fields:
82 - timer
83 - resolution queue
84
85 Again, nothing clever shall be made under neigh->lock,
86 the most complicated procedure, which we allow is dev->hard_header.
87 It is supposed, that dev->hard_header is simplistic and does
88 not make callbacks to neighbour tables.
1da177e4
LT		 89 */
90
8f40b161 91static int neigh_blackhole(struct neighbour *neigh, struct sk_buff *skb)
1da177e4
LT		 92{
93 kfree_skb(skb);
94 return -ENETDOWN;
95}
96
4f494554
TG		 97static void neigh_cleanup_and_release(struct neighbour *neigh)
98{
99 if (neigh->parms->neigh_cleanup)
100 neigh->parms->neigh_cleanup(neigh);
101
d961db35 102 __neigh_notify(neigh, RTM_DELNEIGH, 0);
4f494554
TG		 103 neigh_release(neigh);
104}
105
1da177e4
LT		 106/*
107 * It is random distribution in the interval (1/2)*base...(3/2)*base.
108 * It corresponds to default IPv6 settings and is not overridable,
109 * because it is really reasonable choice.
110 */
111
112unsigned long neigh_rand_reach_time(unsigned long base)
113{
63862b5b 114 return base ? (prandom_u32() % base) + (base >> 1) : 0;
1da177e4 115}
0a204500 116EXPORT_SYMBOL(neigh_rand_reach_time);
1da177e4
LT		 117
118
119static int neigh_forced_gc(struct neigh_table *tbl)
120{
121 int shrunk = 0;
122 int i;
d6bf7817 123 struct neigh_hash_table *nht;
1da177e4
LT		 124
125 NEIGH_CACHE_STAT_INC(tbl, forced_gc_runs);
126
127 write_lock_bh(&tbl->lock);
d6bf7817
ED		 128 nht = rcu_dereference_protected(tbl->nht,
129 lockdep_is_held(&tbl->lock));
cd089336 130 for (i = 0; i < (1 << nht->hash_shift); i++) {
767e97e1
ED		 131 struct neighbour *n;
132 struct neighbour __rcu **np;
1da177e4 133
d6bf7817 134 np = &nht->hash_buckets[i];
767e97e1
ED		 135 while ((n = rcu_dereference_protected(*np,
136 lockdep_is_held(&tbl->lock))) != NULL) {
1da177e4
LT		 137 /* Neighbour record may be discarded if:
138 * - nobody refers to it.
139 * - it is not permanent
140 */
141 write_lock(&n->lock);
142 if (atomic_read(&n->refcnt) == 1 &&
143 !(n->nud_state & NUD_PERMANENT)) {
767e97e1
ED		 144 rcu_assign_pointer(*np,
145 rcu_dereference_protected(n->next,
146 lockdep_is_held(&tbl->lock)));
1da177e4
LT		 147 n->dead = 1;
148 shrunk = 1;
149 write_unlock(&n->lock);
4f494554 150 neigh_cleanup_and_release(n);
1da177e4
LT		 151 continue;
152 }
153 write_unlock(&n->lock);
154 np = &n->next;
155 }
156 }
157
158 tbl->last_flush = jiffies;
159
160 write_unlock_bh(&tbl->lock);
161
162 return shrunk;
163}
164
a43d8994
PE		 165static void neigh_add_timer(struct neighbour *n, unsigned long when)
166{
167 neigh_hold(n);
168 if (unlikely(mod_timer(&n->timer, when))) {
169 printk("NEIGH: BUG, double timer add, state is %x\n",
170 n->nud_state);
171 dump_stack();
172 }
173}
174
1da177e4
LT		 175static int neigh_del_timer(struct neighbour *n)
176{
177 if ((n->nud_state & NUD_IN_TIMER) &&
178 del_timer(&n->timer)) {
179 neigh_release(n);
180 return 1;
181 }
182 return 0;
183}
184
185static void pneigh_queue_purge(struct sk_buff_head *list)
186{
187 struct sk_buff *skb;
188
189 while ((skb = skb_dequeue(list)) != NULL) {
190 dev_put(skb->dev);
191 kfree_skb(skb);
192 }
193}
194
49636bb1 195static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev)
1da177e4
LT		 196{
197 int i;
d6bf7817 198 struct neigh_hash_table *nht;
1da177e4 199
d6bf7817
ED		 200 nht = rcu_dereference_protected(tbl->nht,
201 lockdep_is_held(&tbl->lock));
202
cd089336 203 for (i = 0; i < (1 << nht->hash_shift); i++) {
767e97e1
ED		 204 struct neighbour *n;
205 struct neighbour __rcu **np = &nht->hash_buckets[i];
1da177e4 206
767e97e1
ED		 207 while ((n = rcu_dereference_protected(*np,
208 lockdep_is_held(&tbl->lock))) != NULL) {
1da177e4
LT		 209 if (dev && n->dev != dev) {
210 np = &n->next;
211 continue;
212 }
767e97e1
ED		 213 rcu_assign_pointer(*np,
214 rcu_dereference_protected(n->next,
215 lockdep_is_held(&tbl->lock)));
1da177e4
LT		 216 write_lock(&n->lock);
217 neigh_del_timer(n);
218 n->dead = 1;
219
220 if (atomic_read(&n->refcnt) != 1) {
221 /* The most unpleasant situation.
222 We must destroy neighbour entry,
223 but someone still uses it.
224
225 The destroy will be delayed until
226 the last user releases us, but
227 we must kill timers etc. and move
228 it to safe state.
229 */
c9ab4d85 230 __skb_queue_purge(&n->arp_queue);
8b5c171b 231 n->arp_queue_len_bytes = 0;
1da177e4
LT		 232 n->output = neigh_blackhole;
233 if (n->nud_state & NUD_VALID)
234 n->nud_state = NUD_NOARP;
235 else
236 n->nud_state = NUD_NONE;
d5d427cd 237 neigh_dbg(2, "neigh %p is stray\n", n);
1da177e4
LT		 238 }
239 write_unlock(&n->lock);
4f494554 240 neigh_cleanup_and_release(n);
1da177e4
LT		 241 }
242 }
49636bb1 243}
1da177e4 244
49636bb1
HX		 245void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev)
246{
247 write_lock_bh(&tbl->lock);
248 neigh_flush_dev(tbl, dev);
249 write_unlock_bh(&tbl->lock);
250}
0a204500 251EXPORT_SYMBOL(neigh_changeaddr);
49636bb1
HX		 252
253int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
254{
255 write_lock_bh(&tbl->lock);
256 neigh_flush_dev(tbl, dev);
1da177e4
LT		 257 pneigh_ifdown(tbl, dev);
258 write_unlock_bh(&tbl->lock);
259
260 del_timer_sync(&tbl->proxy_timer);
261 pneigh_queue_purge(&tbl->proxy_queue);
262 return 0;
263}
0a204500 264EXPORT_SYMBOL(neigh_ifdown);
1da177e4 265
596b9b68 266static struct neighbour *neigh_alloc(struct neigh_table *tbl, struct net_device *dev)
1da177e4
LT		 267{
268 struct neighbour *n = NULL;
269 unsigned long now = jiffies;
270 int entries;
271
272 entries = atomic_inc_return(&tbl->entries) - 1;
273 if (entries >= tbl->gc_thresh3 ||
274 (entries >= tbl->gc_thresh2 &&
275 time_after(now, tbl->last_flush + 5 * HZ))) {
276 if (!neigh_forced_gc(tbl) &&
277 entries >= tbl->gc_thresh3)
278 goto out_entries;
279 }
280
08433eff 281 n = kzalloc(tbl->entry_size + dev->neigh_priv_len, GFP_ATOMIC);
1da177e4
LT		 282 if (!n)
283 goto out_entries;
284
c9ab4d85 285 __skb_queue_head_init(&n->arp_queue);
1da177e4 286 rwlock_init(&n->lock);
0ed8ddf4 287 seqlock_init(&n->ha_lock);
1da177e4
LT		 288 n->updated = n->used = now;
289 n->nud_state = NUD_NONE;
290 n->output = neigh_blackhole;
f6b72b62 291 seqlock_init(&n->hh.hh_lock);
1da177e4 292 n->parms = neigh_parms_clone(&tbl->parms);
b24b8a24 293 setup_timer(&n->timer, neigh_timer_handler, (unsigned long)n);
1da177e4
LT		 294
295 NEIGH_CACHE_STAT_INC(tbl, allocs);
296 n->tbl = tbl;
297 atomic_set(&n->refcnt, 1);
298 n->dead = 1;
299out:
300 return n;
301
302out_entries:
303 atomic_dec(&tbl->entries);
304 goto out;
305}
306
2c2aba6c
DM		 307static void neigh_get_hash_rnd(u32 *x)
308{
309 get_random_bytes(x, sizeof(*x));
310 *x |= 1;
311}
312
cd089336 313static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift)
1da177e4 314{
cd089336 315 size_t size = (1 << shift) * sizeof(struct neighbour *);
d6bf7817 316 struct neigh_hash_table *ret;
6193d2be 317 struct neighbour __rcu **buckets;
2c2aba6c 318 int i;
1da177e4 319
d6bf7817
ED		 320 ret = kmalloc(sizeof(*ret), GFP_ATOMIC);
321 if (!ret)
322 return NULL;
323 if (size <= PAGE_SIZE)
324 buckets = kzalloc(size, GFP_ATOMIC);
325 else
6193d2be 326 buckets = (struct neighbour __rcu **)
d6bf7817
ED		 327 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
328 get_order(size));
329 if (!buckets) {
330 kfree(ret);
331 return NULL;
1da177e4 332 }
6193d2be 333 ret->hash_buckets = buckets;
cd089336 334 ret->hash_shift = shift;
2c2aba6c
DM		 335 for (i = 0; i < NEIGH_NUM_HASH_RND; i++)
336 neigh_get_hash_rnd(&ret->hash_rnd[i]);
1da177e4
LT		 337 return ret;
338}
339
d6bf7817 340static void neigh_hash_free_rcu(struct rcu_head *head)
1da177e4 341{
d6bf7817
ED		 342 struct neigh_hash_table *nht = container_of(head,
343 struct neigh_hash_table,
344 rcu);
cd089336 345 size_t size = (1 << nht->hash_shift) * sizeof(struct neighbour *);
6193d2be 346 struct neighbour __rcu **buckets = nht->hash_buckets;
1da177e4
LT		 347
348 if (size <= PAGE_SIZE)
d6bf7817 349 kfree(buckets);
1da177e4 350 else
d6bf7817
ED		 351 free_pages((unsigned long)buckets, get_order(size));
352 kfree(nht);
1da177e4
LT		 353}
354
d6bf7817 355static struct neigh_hash_table *neigh_hash_grow(struct neigh_table *tbl,
cd089336 356 unsigned long new_shift)
1da177e4 357{
d6bf7817
ED		 358 unsigned int i, hash;
359 struct neigh_hash_table *new_nht, *old_nht;
1da177e4
LT		 360
361 NEIGH_CACHE_STAT_INC(tbl, hash_grows);
362
d6bf7817
ED		 363 old_nht = rcu_dereference_protected(tbl->nht,
364 lockdep_is_held(&tbl->lock));
cd089336 365 new_nht = neigh_hash_alloc(new_shift);
d6bf7817
ED		 366 if (!new_nht)
367 return old_nht;
1da177e4 368
cd089336 369 for (i = 0; i < (1 << old_nht->hash_shift); i++) {
1da177e4
LT		 370 struct neighbour *n, *next;
371
767e97e1
ED		 372 for (n = rcu_dereference_protected(old_nht->hash_buckets[i],
373 lockdep_is_held(&tbl->lock));
d6bf7817
ED		 374 n != NULL;
375 n = next) {
376 hash = tbl->hash(n->primary_key, n->dev,
377 new_nht->hash_rnd);
1da177e4 378
cd089336 379 hash >>= (32 - new_nht->hash_shift);
767e97e1
ED		 380 next = rcu_dereference_protected(n->next,
381 lockdep_is_held(&tbl->lock));
382
383 rcu_assign_pointer(n->next,
384 rcu_dereference_protected(
385 new_nht->hash_buckets[hash],
386 lockdep_is_held(&tbl->lock)));
387 rcu_assign_pointer(new_nht->hash_buckets[hash], n);
1da177e4
LT		 388 }
389 }
1da177e4 390
d6bf7817
ED		 391 rcu_assign_pointer(tbl->nht, new_nht);
392 call_rcu(&old_nht->rcu, neigh_hash_free_rcu);
393 return new_nht;
1da177e4
LT		 394}
395
396struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
397 struct net_device *dev)
398{
399 struct neighbour *n;
4ec93edb 400
1da177e4
LT		 401 NEIGH_CACHE_STAT_INC(tbl, lookups);
402
d6bf7817 403 rcu_read_lock_bh();
60395a20
EB		 404 n = __neigh_lookup_noref(tbl, pkey, dev);
405 if (n) {
406 if (!atomic_inc_not_zero(&n->refcnt))
407 n = NULL;
408 NEIGH_CACHE_STAT_INC(tbl, hits);
1da177e4 409 }
767e97e1 410
d6bf7817 411 rcu_read_unlock_bh();
1da177e4
LT		 412 return n;
413}
0a204500 414EXPORT_SYMBOL(neigh_lookup);
1da177e4 415
426b5303
EB		 416struct neighbour *neigh_lookup_nodev(struct neigh_table *tbl, struct net *net,
417 const void *pkey)
1da177e4
LT		 418{
419 struct neighbour *n;
420 int key_len = tbl->key_len;
bc4bf5f3 421 u32 hash_val;
d6bf7817 422 struct neigh_hash_table *nht;
1da177e4
LT		 423
424 NEIGH_CACHE_STAT_INC(tbl, lookups);
425
d6bf7817
ED		 426 rcu_read_lock_bh();
427 nht = rcu_dereference_bh(tbl->nht);
cd089336 428 hash_val = tbl->hash(pkey, NULL, nht->hash_rnd) >> (32 - nht->hash_shift);
767e97e1
ED		 429
430 for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]);
431 n != NULL;
432 n = rcu_dereference_bh(n->next)) {
426b5303 433 if (!memcmp(n->primary_key, pkey, key_len) &&
878628fb 434 net_eq(dev_net(n->dev), net)) {
767e97e1
ED		 435 if (!atomic_inc_not_zero(&n->refcnt))
436 n = NULL;
1da177e4
LT		 437 NEIGH_CACHE_STAT_INC(tbl, hits);
438 break;
439 }
440 }
767e97e1 441
d6bf7817 442 rcu_read_unlock_bh();
1da177e4
LT		 443 return n;
444}
0a204500 445EXPORT_SYMBOL(neigh_lookup_nodev);
1da177e4 446
a263b309
DM		 447struct neighbour *__neigh_create(struct neigh_table *tbl, const void *pkey,
448 struct net_device *dev, bool want_ref)
1da177e4
LT		 449{
450 u32 hash_val;
451 int key_len = tbl->key_len;
452 int error;
596b9b68 453 struct neighbour *n1, *rc, *n = neigh_alloc(tbl, dev);
d6bf7817 454 struct neigh_hash_table *nht;
1da177e4
LT		 455
456 if (!n) {
457 rc = ERR_PTR(-ENOBUFS);
458 goto out;
459 }
460
461 memcpy(n->primary_key, pkey, key_len);
462 n->dev = dev;
463 dev_hold(dev);
464
465 /* Protocol specific setup. */
466 if (tbl->constructor && (error = tbl->constructor(n)) < 0) {
467 rc = ERR_PTR(error);
468 goto out_neigh_release;
469 }
470
da6a8fa0
DM		 471 if (dev->netdev_ops->ndo_neigh_construct) {
472 error = dev->netdev_ops->ndo_neigh_construct(n);
473 if (error < 0) {
474 rc = ERR_PTR(error);
475 goto out_neigh_release;
476 }
477 }
478
447f2191
DM		 479 /* Device specific setup. */
480 if (n->parms->neigh_setup &&
481 (error = n->parms->neigh_setup(n)) < 0) {
482 rc = ERR_PTR(error);
483 goto out_neigh_release;
484 }
485
1f9248e5 486 n->confirmed = jiffies - (NEIGH_VAR(n->parms, BASE_REACHABLE_TIME) << 1);
1da177e4
LT		 487
488 write_lock_bh(&tbl->lock);
d6bf7817
ED		 489 nht = rcu_dereference_protected(tbl->nht,
490 lockdep_is_held(&tbl->lock));
1da177e4 491
cd089336
DM		 492 if (atomic_read(&tbl->entries) > (1 << nht->hash_shift))
493 nht = neigh_hash_grow(tbl, nht->hash_shift + 1);
1da177e4 494
cd089336 495 hash_val = tbl->hash(pkey, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
1da177e4
LT		 496
497 if (n->parms->dead) {
498 rc = ERR_PTR(-EINVAL);
499 goto out_tbl_unlock;
500 }
501
767e97e1
ED		 502 for (n1 = rcu_dereference_protected(nht->hash_buckets[hash_val],
503 lockdep_is_held(&tbl->lock));
504 n1 != NULL;
505 n1 = rcu_dereference_protected(n1->next,
506 lockdep_is_held(&tbl->lock))) {
1da177e4 507 if (dev == n1->dev && !memcmp(n1->primary_key, pkey, key_len)) {
a263b309
DM		 508 if (want_ref)
509 neigh_hold(n1);
1da177e4
LT		 510 rc = n1;
511 goto out_tbl_unlock;
512 }
513 }
514
1da177e4 515 n->dead = 0;
a263b309
DM		 516 if (want_ref)
517 neigh_hold(n);
767e97e1
ED		 518 rcu_assign_pointer(n->next,
519 rcu_dereference_protected(nht->hash_buckets[hash_val],
520 lockdep_is_held(&tbl->lock)));
521 rcu_assign_pointer(nht->hash_buckets[hash_val], n);
1da177e4 522 write_unlock_bh(&tbl->lock);
d5d427cd 523 neigh_dbg(2, "neigh %p is created\n", n);
1da177e4
LT		 524 rc = n;
525out:
526 return rc;
527out_tbl_unlock:
528 write_unlock_bh(&tbl->lock);
529out_neigh_release:
530 neigh_release(n);
531 goto out;
532}
a263b309 533EXPORT_SYMBOL(__neigh_create);
1da177e4 534
be01d655 535static u32 pneigh_hash(const void *pkey, int key_len)
fa86d322 536{
fa86d322 537 u32 hash_val = *(u32 *)(pkey + key_len - 4);
fa86d322
PE		 538 hash_val ^= (hash_val >> 16);
539 hash_val ^= hash_val >> 8;
540 hash_val ^= hash_val >> 4;
541 hash_val &= PNEIGH_HASHMASK;
be01d655
YH		 542 return hash_val;
543}
fa86d322 544
be01d655
YH		 545static struct pneigh_entry *__pneigh_lookup_1(struct pneigh_entry *n,
546 struct net *net,
547 const void *pkey,
548 int key_len,
549 struct net_device *dev)
550{
551 while (n) {
fa86d322 552 if (!memcmp(n->key, pkey, key_len) &&
be01d655 553 net_eq(pneigh_net(n), net) &&
fa86d322 554 (n->dev == dev || !n->dev))
be01d655
YH		 555 return n;
556 n = n->next;
fa86d322 557 }
be01d655
YH		 558 return NULL;
559}
fa86d322 560
be01d655
YH		 561struct pneigh_entry *__pneigh_lookup(struct neigh_table *tbl,
562 struct net *net, const void *pkey, struct net_device *dev)
563{
564 int key_len = tbl->key_len;
565 u32 hash_val = pneigh_hash(pkey, key_len);
566
567 return __pneigh_lookup_1(tbl->phash_buckets[hash_val],
568 net, pkey, key_len, dev);
fa86d322 569}
0a204500 570EXPORT_SYMBOL_GPL(__pneigh_lookup);
fa86d322 571
426b5303
EB		 572struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl,
573 struct net *net, const void *pkey,
1da177e4
LT		 574 struct net_device *dev, int creat)
575{
576 struct pneigh_entry *n;
577 int key_len = tbl->key_len;
be01d655 578 u32 hash_val = pneigh_hash(pkey, key_len);
1da177e4
LT		 579
580 read_lock_bh(&tbl->lock);
be01d655
YH		 581 n = __pneigh_lookup_1(tbl->phash_buckets[hash_val],
582 net, pkey, key_len, dev);
1da177e4 583 read_unlock_bh(&tbl->lock);
be01d655
YH		 584
585 if (n || !creat)
1da177e4
LT		 586 goto out;
587
4ae28944
PE		 588 ASSERT_RTNL();
589
1da177e4
LT		 590 n = kmalloc(sizeof(*n) + key_len, GFP_KERNEL);
591 if (!n)
592 goto out;
593
efd7ef1c 594 write_pnet(&n->net, net);
1da177e4
LT		 595 memcpy(n->key, pkey, key_len);
596 n->dev = dev;
597 if (dev)
598 dev_hold(dev);
599
600 if (tbl->pconstructor && tbl->pconstructor(n)) {
601 if (dev)
602 dev_put(dev);
603 kfree(n);
604 n = NULL;
605 goto out;
606 }
607
608 write_lock_bh(&tbl->lock);
609 n->next = tbl->phash_buckets[hash_val];
610 tbl->phash_buckets[hash_val] = n;
611 write_unlock_bh(&tbl->lock);
612out:
613 return n;
614}
0a204500 615EXPORT_SYMBOL(pneigh_lookup);
1da177e4
LT		 616
617
426b5303 618int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
1da177e4
LT		 619 struct net_device *dev)
620{
621 struct pneigh_entry *n, **np;
622 int key_len = tbl->key_len;
be01d655 623 u32 hash_val = pneigh_hash(pkey, key_len);
1da177e4
LT		 624
625 write_lock_bh(&tbl->lock);
626 for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
627 np = &n->next) {
426b5303 628 if (!memcmp(n->key, pkey, key_len) && n->dev == dev &&
878628fb 629 net_eq(pneigh_net(n), net)) {
1da177e4
LT		 630 *np = n->next;
631 write_unlock_bh(&tbl->lock);
632 if (tbl->pdestructor)
633 tbl->pdestructor(n);
634 if (n->dev)
635 dev_put(n->dev);
636 kfree(n);
637 return 0;
638 }
639 }
640 write_unlock_bh(&tbl->lock);
641 return -ENOENT;
642}
643
644static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
645{
646 struct pneigh_entry *n, **np;
647 u32 h;
648
649 for (h = 0; h <= PNEIGH_HASHMASK; h++) {
650 np = &tbl->phash_buckets[h];
651 while ((n = *np) != NULL) {
652 if (!dev || n->dev == dev) {
653 *np = n->next;
654 if (tbl->pdestructor)
655 tbl->pdestructor(n);
656 if (n->dev)
657 dev_put(n->dev);
658 kfree(n);
659 continue;
660 }
661 np = &n->next;
662 }
663 }
664 return -ENOENT;
665}
666
06f0511d
DL		 667static void neigh_parms_destroy(struct neigh_parms *parms);
668
669static inline void neigh_parms_put(struct neigh_parms *parms)
670{
671 if (atomic_dec_and_test(&parms->refcnt))
672 neigh_parms_destroy(parms);
673}
1da177e4
LT		 674
675/*
676 * neighbour must already be out of the table;
677 *
678 */
679void neigh_destroy(struct neighbour *neigh)
680{
da6a8fa0
DM		 681 struct net_device *dev = neigh->dev;
682
1da177e4
LT		 683 NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
684
685 if (!neigh->dead) {
e005d193 686 pr_warn("Destroying alive neighbour %p\n", neigh);
1da177e4
LT		 687 dump_stack();
688 return;
689 }
690
691 if (neigh_del_timer(neigh))
e005d193 692 pr_warn("Impossible event\n");
1da177e4 693
c9ab4d85
ED		 694 write_lock_bh(&neigh->lock);
695 __skb_queue_purge(&neigh->arp_queue);
696 write_unlock_bh(&neigh->lock);
8b5c171b 697 neigh->arp_queue_len_bytes = 0;
1da177e4 698
447f2191
DM		 699 if (dev->netdev_ops->ndo_neigh_destroy)
700 dev->netdev_ops->ndo_neigh_destroy(neigh);
701
da6a8fa0 702 dev_put(dev);
1da177e4
LT		 703 neigh_parms_put(neigh->parms);
704
d5d427cd 705 neigh_dbg(2, "neigh %p is destroyed\n", neigh);
1da177e4
LT		 706
707 atomic_dec(&neigh->tbl->entries);
5b8b0060 708 kfree_rcu(neigh, rcu);
1da177e4 709}
0a204500 710EXPORT_SYMBOL(neigh_destroy);
1da177e4
LT		 711
712/* Neighbour state is suspicious;
713 disable fast path.
714
715 Called with write_locked neigh.
716 */
717static void neigh_suspect(struct neighbour *neigh)
718{
d5d427cd 719 neigh_dbg(2, "neigh %p is suspected\n", neigh);
1da177e4
LT		 720
721 neigh->output = neigh->ops->output;
1da177e4
LT		 722}
723
724/* Neighbour state is OK;
725 enable fast path.
726
727 Called with write_locked neigh.
728 */
729static void neigh_connect(struct neighbour *neigh)
730{
d5d427cd 731 neigh_dbg(2, "neigh %p is connected\n", neigh);
1da177e4
LT		 732
733 neigh->output = neigh->ops->connected_output;
1da177e4
LT		 734}
735
e4c4e448 736static void neigh_periodic_work(struct work_struct *work)
1da177e4 737{
e4c4e448 738 struct neigh_table *tbl = container_of(work, struct neigh_table, gc_work.work);
767e97e1
ED		 739 struct neighbour *n;
740 struct neighbour __rcu **np;
e4c4e448 741 unsigned int i;
d6bf7817 742 struct neigh_hash_table *nht;
1da177e4
LT		 743
744 NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
745
e4c4e448 746 write_lock_bh(&tbl->lock);
d6bf7817
ED		 747 nht = rcu_dereference_protected(tbl->nht,
748 lockdep_is_held(&tbl->lock));
1da177e4
LT		 749
750 /*
751 * periodically recompute ReachableTime from random function
752 */
753
e4c4e448 754 if (time_after(jiffies, tbl->last_rand + 300 * HZ)) {
1da177e4 755 struct neigh_parms *p;
e4c4e448 756 tbl->last_rand = jiffies;
75fbfd33 757 list_for_each_entry(p, &tbl->parms_list, list)
1da177e4 758 p->reachable_time =
1f9248e5 759 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
1da177e4
LT		 760 }
761
feff9ab2
DJ		 762 if (atomic_read(&tbl->entries) < tbl->gc_thresh1)
763 goto out;
764
cd089336 765 for (i = 0 ; i < (1 << nht->hash_shift); i++) {
d6bf7817 766 np = &nht->hash_buckets[i];
1da177e4 767
767e97e1
ED		 768 while ((n = rcu_dereference_protected(*np,
769 lockdep_is_held(&tbl->lock))) != NULL) {
e4c4e448 770 unsigned int state;
1da177e4 771
e4c4e448 772 write_lock(&n->lock);
1da177e4 773
e4c4e448
ED		 774 state = n->nud_state;
775 if (state & (NUD_PERMANENT | NUD_IN_TIMER)) {
776 write_unlock(&n->lock);
777 goto next_elt;
778 }
1da177e4 779
e4c4e448
ED		 780 if (time_before(n->used, n->confirmed))
781 n->used = n->confirmed;
1da177e4 782
e4c4e448
ED		 783 if (atomic_read(&n->refcnt) == 1 &&
784 (state == NUD_FAILED ||
1f9248e5 785 time_after(jiffies, n->used + NEIGH_VAR(n->parms, GC_STALETIME)))) {
e4c4e448
ED		 786 *np = n->next;
787 n->dead = 1;
788 write_unlock(&n->lock);
789 neigh_cleanup_and_release(n);
790 continue;
791 }
1da177e4 792 write_unlock(&n->lock);
1da177e4
LT		 793
794next_elt:
e4c4e448
ED		 795 np = &n->next;
796 }
797 /*
798 * It's fine to release lock here, even if hash table
799 * grows while we are preempted.
800 */
801 write_unlock_bh(&tbl->lock);
802 cond_resched();
803 write_lock_bh(&tbl->lock);
84338a6c
MM		 804 nht = rcu_dereference_protected(tbl->nht,
805 lockdep_is_held(&tbl->lock));
1da177e4 806 }
2724680b 807out:
1f9248e5
JP		 808 /* Cycle through all hash buckets every BASE_REACHABLE_TIME/2 ticks.
809 * ARP entry timeouts range from 1/2 BASE_REACHABLE_TIME to 3/2
810 * BASE_REACHABLE_TIME.
1da177e4 811 */
f618002b 812 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
1f9248e5 813 NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME) >> 1);
e4c4e448 814 write_unlock_bh(&tbl->lock);
1da177e4
LT		 815}
816
817static __inline__ int neigh_max_probes(struct neighbour *n)
818{
819 struct neigh_parms *p = n->parms;
8da86466
YH		 820 return NEIGH_VAR(p, UCAST_PROBES) + NEIGH_VAR(p, APP_PROBES) +
821 (n->nud_state & NUD_PROBE ? NEIGH_VAR(p, MCAST_REPROBES) :
822 NEIGH_VAR(p, MCAST_PROBES));
1da177e4
LT		 823}
824
5ef12d98 825static void neigh_invalidate(struct neighbour *neigh)
0a141509
ED		 826 __releases(neigh->lock)
827 __acquires(neigh->lock)
5ef12d98
TT		 828{
829 struct sk_buff *skb;
830
831 NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
d5d427cd 832 neigh_dbg(2, "neigh %p is failed\n", neigh);
5ef12d98
TT		 833 neigh->updated = jiffies;
834
835 /* It is very thin place. report_unreachable is very complicated
836 routine. Particularly, it can hit the same neighbour entry!
837
838 So that, we try to be accurate and avoid dead loop. --ANK
839 */
840 while (neigh->nud_state == NUD_FAILED &&
841 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
842 write_unlock(&neigh->lock);
843 neigh->ops->error_report(neigh, skb);
844 write_lock(&neigh->lock);
845 }
c9ab4d85 846 __skb_queue_purge(&neigh->arp_queue);
8b5c171b 847 neigh->arp_queue_len_bytes = 0;
5ef12d98
TT		 848}
849
cd28ca0a
ED		 850static void neigh_probe(struct neighbour *neigh)
851 __releases(neigh->lock)
852{
4ed377e3 853 struct sk_buff *skb = skb_peek_tail(&neigh->arp_queue);
cd28ca0a
ED		 854 /* keep skb alive even if arp_queue overflows */
855 if (skb)
856 skb = skb_copy(skb, GFP_ATOMIC);
857 write_unlock(&neigh->lock);
858 neigh->ops->solicit(neigh, skb);
859 atomic_inc(&neigh->probes);
860 kfree_skb(skb);
861}
862
1da177e4
LT		 863/* Called when a timer expires for a neighbour entry. */
864
865static void neigh_timer_handler(unsigned long arg)
866{
867 unsigned long now, next;
868 struct neighbour *neigh = (struct neighbour *)arg;
95c96174 869 unsigned int state;
1da177e4
LT		 870 int notify = 0;
871
872 write_lock(&neigh->lock);
873
874 state = neigh->nud_state;
875 now = jiffies;
876 next = now + HZ;
877
045f7b3b 878 if (!(state & NUD_IN_TIMER))
1da177e4 879 goto out;
1da177e4
LT		 880
881 if (state & NUD_REACHABLE) {
4ec93edb 882 if (time_before_eq(now,
1da177e4 883 neigh->confirmed + neigh->parms->reachable_time)) {
d5d427cd 884 neigh_dbg(2, "neigh %p is still alive\n", neigh);
1da177e4
LT		 885 next = neigh->confirmed + neigh->parms->reachable_time;
886 } else if (time_before_eq(now,
1f9248e5
JP		 887 neigh->used +
888 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
d5d427cd 889 neigh_dbg(2, "neigh %p is delayed\n", neigh);
1da177e4 890 neigh->nud_state = NUD_DELAY;
955aaa2f 891 neigh->updated = jiffies;
1da177e4 892 neigh_suspect(neigh);
1f9248e5 893 next = now + NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME);
1da177e4 894 } else {
d5d427cd 895 neigh_dbg(2, "neigh %p is suspected\n", neigh);
1da177e4 896 neigh->nud_state = NUD_STALE;
955aaa2f 897 neigh->updated = jiffies;
1da177e4 898 neigh_suspect(neigh);
8d71740c 899 notify = 1;
1da177e4
LT		 900 }
901 } else if (state & NUD_DELAY) {
4ec93edb 902 if (time_before_eq(now,
1f9248e5
JP		 903 neigh->confirmed +
904 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
d5d427cd 905 neigh_dbg(2, "neigh %p is now reachable\n", neigh);
1da177e4 906 neigh->nud_state = NUD_REACHABLE;
955aaa2f 907 neigh->updated = jiffies;
1da177e4 908 neigh_connect(neigh);
8d71740c 909 notify = 1;
1da177e4
LT		 910 next = neigh->confirmed + neigh->parms->reachable_time;
911 } else {
d5d427cd 912 neigh_dbg(2, "neigh %p is probed\n", neigh);
1da177e4 913 neigh->nud_state = NUD_PROBE;
955aaa2f 914 neigh->updated = jiffies;
1da177e4 915 atomic_set(&neigh->probes, 0);
1f9248e5 916 next = now + NEIGH_VAR(neigh->parms, RETRANS_TIME);
1da177e4
LT		 917 }
918 } else {
919 /* NUD_PROBE|NUD_INCOMPLETE */
1f9248e5 920 next = now + NEIGH_VAR(neigh->parms, RETRANS_TIME);
1da177e4
LT		 921 }
922
923 if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
924 atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
1da177e4
LT		 925 neigh->nud_state = NUD_FAILED;
926 notify = 1;
5ef12d98 927 neigh_invalidate(neigh);
5e2c21dc 928 goto out;
1da177e4
LT		 929 }
930
931 if (neigh->nud_state & NUD_IN_TIMER) {
1da177e4
LT		 932 if (time_before(next, jiffies + HZ/2))
933 next = jiffies + HZ/2;
6fb9974f
HX		 934 if (!mod_timer(&neigh->timer, next))
935 neigh_hold(neigh);
1da177e4
LT		 936 }
937 if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
cd28ca0a 938 neigh_probe(neigh);
9ff56607 939 } else {
69cc64d8 940out:
9ff56607
DM		 941 write_unlock(&neigh->lock);
942 }
d961db35 943
8d71740c 944 if (notify)
d961db35 945 neigh_update_notify(neigh);
1da177e4 946
1da177e4
LT		 947 neigh_release(neigh);
948}
949
950int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
951{
952 int rc;
cd28ca0a 953 bool immediate_probe = false;
1da177e4
LT		 954
955 write_lock_bh(&neigh->lock);
956
957 rc = 0;
958 if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
959 goto out_unlock_bh;
2c51a97f
JA		 960 if (neigh->dead)
961 goto out_dead;
1da177e4 962
1da177e4 963 if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
1f9248e5
JP		 964 if (NEIGH_VAR(neigh->parms, MCAST_PROBES) +
965 NEIGH_VAR(neigh->parms, APP_PROBES)) {
cd28ca0a
ED		 966 unsigned long next, now = jiffies;
967
1f9248e5
JP		 968 atomic_set(&neigh->probes,
969 NEIGH_VAR(neigh->parms, UCAST_PROBES));
1da177e4 970 neigh->nud_state = NUD_INCOMPLETE;
cd28ca0a 971 neigh->updated = now;
1f9248e5
JP		 972 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
973 HZ/2);
cd28ca0a
ED		 974 neigh_add_timer(neigh, next);
975 immediate_probe = true;
1da177e4
LT		 976 } else {
977 neigh->nud_state = NUD_FAILED;
955aaa2f 978 neigh->updated = jiffies;
1da177e4
LT		 979 write_unlock_bh(&neigh->lock);
980
f3fbbe0f 981 kfree_skb(skb);
1da177e4
LT		 982 return 1;
983 }
984 } else if (neigh->nud_state & NUD_STALE) {
d5d427cd 985 neigh_dbg(2, "neigh %p is delayed\n", neigh);
1da177e4 986 neigh->nud_state = NUD_DELAY;
955aaa2f 987 neigh->updated = jiffies;
1f9248e5
JP		 988 neigh_add_timer(neigh, jiffies +
989 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME));
1da177e4
LT		 990 }
991
992 if (neigh->nud_state == NUD_INCOMPLETE) {
993 if (skb) {
8b5c171b 994 while (neigh->arp_queue_len_bytes + skb->truesize >
1f9248e5 995 NEIGH_VAR(neigh->parms, QUEUE_LEN_BYTES)) {
1da177e4 996 struct sk_buff *buff;
8b5c171b 997
f72051b0 998 buff = __skb_dequeue(&neigh->arp_queue);
8b5c171b
ED		 999 if (!buff)
1000 break;
1001 neigh->arp_queue_len_bytes -= buff->truesize;
1da177e4 1002 kfree_skb(buff);
9a6d276e 1003 NEIGH_CACHE_STAT_INC(neigh->tbl, unres_discards);
1da177e4 1004 }
a4731138 1005 skb_dst_force(skb);
1da177e4 1006 __skb_queue_tail(&neigh->arp_queue, skb);
8b5c171b 1007 neigh->arp_queue_len_bytes += skb->truesize;
1da177e4
LT		 1008 }
1009 rc = 1;
1010 }
1011out_unlock_bh:
cd28ca0a
ED		 1012 if (immediate_probe)
1013 neigh_probe(neigh);
1014 else
1015 write_unlock(&neigh->lock);
1016 local_bh_enable();
1da177e4 1017 return rc;
2c51a97f
JA		 1018
1019out_dead:
1020 if (neigh->nud_state & NUD_STALE)
1021 goto out_unlock_bh;
1022 write_unlock_bh(&neigh->lock);
1023 kfree_skb(skb);
1024 return 1;
1da177e4 1025}
0a204500 1026EXPORT_SYMBOL(__neigh_event_send);
1da177e4 1027
f6b72b62 1028static void neigh_update_hhs(struct neighbour *neigh)
1da177e4
LT		 1029{
1030 struct hh_cache *hh;
3b04ddde 1031 void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *)
91a72a70
DK		 1032 = NULL;
1033
1034 if (neigh->dev->header_ops)
1035 update = neigh->dev->header_ops->cache_update;
1da177e4
LT		 1036
1037 if (update) {
f6b72b62
DM		 1038 hh = &neigh->hh;
1039 if (hh->hh_len) {
3644f0ce 1040 write_seqlock_bh(&hh->hh_lock);
1da177e4 1041 update(hh, neigh->dev, neigh->ha);
3644f0ce 1042 write_sequnlock_bh(&hh->hh_lock);
1da177e4
LT		 1043 }
1044 }
1045}
1046
1047
1048
1049/* Generic update routine.
1050 -- lladdr is new lladdr or NULL, if it is not supplied.
1051 -- new is new state.
1052 -- flags
1053 NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
1054 if it is different.
1055 NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
4ec93edb 1056 lladdr instead of overriding it
1da177e4
LT		 1057 if it is different.
1058 It also allows to retain current state
1059 if lladdr is unchanged.
1060 NEIGH_UPDATE_F_ADMIN means that the change is administrative.
1061
4ec93edb 1062 NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
1da177e4
LT		 1063 NTF_ROUTER flag.
1064 NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as
1065 a router.
1066
1067 Caller MUST hold reference count on the entry.
1068 */
1069
1070int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
1071 u32 flags)
1072{
1073 u8 old;
1074 int err;
1da177e4 1075 int notify = 0;
1da177e4
LT		 1076 struct net_device *dev;
1077 int update_isrouter = 0;
1078
1079 write_lock_bh(&neigh->lock);
1080
1081 dev = neigh->dev;
1082 old = neigh->nud_state;
1083 err = -EPERM;
1084
4ec93edb 1085 if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
1da177e4
LT		 1086 (old & (NUD_NOARP | NUD_PERMANENT)))
1087 goto out;
2c51a97f
JA		 1088 if (neigh->dead)
1089 goto out;
1da177e4
LT		 1090
1091 if (!(new & NUD_VALID)) {
1092 neigh_del_timer(neigh);
1093 if (old & NUD_CONNECTED)
1094 neigh_suspect(neigh);
1095 neigh->nud_state = new;
1096 err = 0;
1da177e4 1097 notify = old & NUD_VALID;
5ef12d98
TT		 1098 if ((old & (NUD_INCOMPLETE | NUD_PROBE)) &&
1099 (new & NUD_FAILED)) {
1100 neigh_invalidate(neigh);
1101 notify = 1;
1102 }
1da177e4
LT		 1103 goto out;
1104 }
1105
1106 /* Compare new lladdr with cached one */
1107 if (!dev->addr_len) {
1108 /* First case: device needs no address. */
1109 lladdr = neigh->ha;
1110 } else if (lladdr) {
1111 /* The second case: if something is already cached
1112 and a new address is proposed:
1113 - compare new & old
1114 - if they are different, check override flag
1115 */
4ec93edb 1116 if ((old & NUD_VALID) &&
1da177e4
LT		 1117 !memcmp(lladdr, neigh->ha, dev->addr_len))
1118 lladdr = neigh->ha;
1119 } else {
1120 /* No address is supplied; if we know something,
1121 use it, otherwise discard the request.
1122 */
1123 err = -EINVAL;
1124 if (!(old & NUD_VALID))
1125 goto out;
1126 lladdr = neigh->ha;
1127 }
1128
1129 if (new & NUD_CONNECTED)
1130 neigh->confirmed = jiffies;
1131 neigh->updated = jiffies;
1132
1133 /* If entry was valid and address is not changed,
1134 do not change entry state, if new one is STALE.
1135 */
1136 err = 0;
1137 update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1138 if (old & NUD_VALID) {
1139 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
1140 update_isrouter = 0;
1141 if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
1142 (old & NUD_CONNECTED)) {
1143 lladdr = neigh->ha;
1144 new = NUD_STALE;
1145 } else
1146 goto out;
1147 } else {
1148 if (lladdr == neigh->ha && new == NUD_STALE &&
1149 ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) ||
1150 (old & NUD_CONNECTED))
1151 )
1152 new = old;
1153 }
1154 }
1155
1156 if (new != old) {
1157 neigh_del_timer(neigh);
a43d8994 1158 if (new & NUD_IN_TIMER)
4ec93edb
YH		 1159 neigh_add_timer(neigh, (jiffies +
1160 ((new & NUD_REACHABLE) ?
667347f1
DM		 1161 neigh->parms->reachable_time :
1162 0)));
1da177e4 1163 neigh->nud_state = new;
53385d2d 1164 notify = 1;
1da177e4
LT		 1165 }
1166
1167 if (lladdr != neigh->ha) {
0ed8ddf4 1168 write_seqlock(&neigh->ha_lock);
1da177e4 1169 memcpy(&neigh->ha, lladdr, dev->addr_len);
0ed8ddf4 1170 write_sequnlock(&neigh->ha_lock);
1da177e4
LT		 1171 neigh_update_hhs(neigh);
1172 if (!(new & NUD_CONNECTED))
1173 neigh->confirmed = jiffies -
1f9248e5 1174 (NEIGH_VAR(neigh->parms, BASE_REACHABLE_TIME) << 1);
1da177e4 1175 notify = 1;
1da177e4
LT		 1176 }
1177 if (new == old)
1178 goto out;
1179 if (new & NUD_CONNECTED)
1180 neigh_connect(neigh);
1181 else
1182 neigh_suspect(neigh);
1183 if (!(old & NUD_VALID)) {
1184 struct sk_buff *skb;
1185
1186 /* Again: avoid dead loop if something went wrong */
1187
1188 while (neigh->nud_state & NUD_VALID &&
1189 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
69cce1d1
DM		 1190 struct dst_entry *dst = skb_dst(skb);
1191 struct neighbour *n2, *n1 = neigh;
1da177e4 1192 write_unlock_bh(&neigh->lock);
e049f288 1193
1194 rcu_read_lock();
13a43d94
DM		 1195
1196 /* Why not just use 'neigh' as-is? The problem is that
1197 * things such as shaper, eql, and sch_teql can end up
1198 * using alternative, different, neigh objects to output
1199 * the packet in the output path. So what we need to do
1200 * here is re-lookup the top-level neigh in the path so
1201 * we can reinject the packet there.
1202 */
1203 n2 = NULL;
1204 if (dst) {
1205 n2 = dst_neigh_lookup_skb(dst, skb);
1206 if (n2)
1207 n1 = n2;
1208 }
8f40b161 1209 n1->output(n1, skb);
13a43d94
DM		 1210 if (n2)
1211 neigh_release(n2);
e049f288 1212 rcu_read_unlock();
1213
1da177e4
LT		 1214 write_lock_bh(&neigh->lock);
1215 }
c9ab4d85 1216 __skb_queue_purge(&neigh->arp_queue);
8b5c171b 1217 neigh->arp_queue_len_bytes = 0;
1da177e4
LT		 1218 }
1219out:
1220 if (update_isrouter) {
1221 neigh->flags = (flags & NEIGH_UPDATE_F_ISROUTER) ?
1222 (neigh->flags | NTF_ROUTER) :
1223 (neigh->flags & ~NTF_ROUTER);
1224 }
1225 write_unlock_bh(&neigh->lock);
8d71740c
TT		 1226
1227 if (notify)
d961db35
TG		 1228 neigh_update_notify(neigh);
1229
1da177e4
LT		 1230 return err;
1231}
0a204500 1232EXPORT_SYMBOL(neigh_update);
1da177e4 1233
7e980569
JB		 1234/* Update the neigh to listen temporarily for probe responses, even if it is
1235 * in a NUD_FAILED state. The caller has to hold neigh->lock for writing.
1236 */
1237void __neigh_set_probe_once(struct neighbour *neigh)
1238{
2c51a97f
JA		 1239 if (neigh->dead)
1240 return;
7e980569
JB		 1241 neigh->updated = jiffies;
1242 if (!(neigh->nud_state & NUD_FAILED))
1243 return;
2176d5d4
DJ		 1244 neigh->nud_state = NUD_INCOMPLETE;
1245 atomic_set(&neigh->probes, neigh_max_probes(neigh));
7e980569
JB		 1246 neigh_add_timer(neigh,
1247 jiffies + NEIGH_VAR(neigh->parms, RETRANS_TIME));
1248}
1249EXPORT_SYMBOL(__neigh_set_probe_once);
1250
1da177e4
LT		 1251struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1252 u8 *lladdr, void *saddr,
1253 struct net_device *dev)
1254{
1255 struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1256 lladdr || !dev->addr_len);
1257 if (neigh)
4ec93edb 1258 neigh_update(neigh, lladdr, NUD_STALE,
1da177e4
LT		 1259 NEIGH_UPDATE_F_OVERRIDE);
1260 return neigh;
1261}
0a204500 1262EXPORT_SYMBOL(neigh_event_ns);
1da177e4 1263
34d101dd 1264/* called with read_lock_bh(&n->lock); */
bdf53c58 1265static void neigh_hh_init(struct neighbour *n)
1da177e4 1266{
bdf53c58
EB		 1267 struct net_device *dev = n->dev;
1268 __be16 prot = n->tbl->protocol;
f6b72b62 1269 struct hh_cache *hh = &n->hh;
0ed8ddf4
ED		 1270
1271 write_lock_bh(&n->lock);
34d101dd 1272
f6b72b62
DM		 1273 /* Only one thread can come in here and initialize the
1274 * hh_cache entry.
1275 */
b23b5455
DM		 1276 if (!hh->hh_len)
1277 dev->header_ops->cache(n, hh, prot);
34d101dd 1278
0ed8ddf4 1279 write_unlock_bh(&n->lock);
1da177e4
LT		 1280}
1281
1da177e4
LT		 1282/* Slow and careful. */
1283
8f40b161 1284int neigh_resolve_output(struct neighbour *neigh, struct sk_buff *skb)
1da177e4 1285{
1da177e4
LT		 1286 int rc = 0;
1287
1da177e4
LT		 1288 if (!neigh_event_send(neigh, skb)) {
1289 int err;
1290 struct net_device *dev = neigh->dev;
0ed8ddf4 1291 unsigned int seq;
34d101dd 1292
f6b72b62 1293 if (dev->header_ops->cache && !neigh->hh.hh_len)
bdf53c58 1294 neigh_hh_init(neigh);
34d101dd 1295
0ed8ddf4 1296 do {
e1f16503 1297 __skb_pull(skb, skb_network_offset(skb));
0ed8ddf4
ED		 1298 seq = read_seqbegin(&neigh->ha_lock);
1299 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1300 neigh->ha, NULL, skb->len);
1301 } while (read_seqretry(&neigh->ha_lock, seq));
34d101dd 1302
1da177e4 1303 if (err >= 0)
542d4d68 1304 rc = dev_queue_xmit(skb);
1da177e4
LT		 1305 else
1306 goto out_kfree_skb;
1307 }
1308out:
1309 return rc;
1da177e4
LT		 1310out_kfree_skb:
1311 rc = -EINVAL;
1312 kfree_skb(skb);
1313 goto out;
1314}
0a204500 1315EXPORT_SYMBOL(neigh_resolve_output);
1da177e4
LT		 1316
1317/* As fast as possible without hh cache */
1318
8f40b161 1319int neigh_connected_output(struct neighbour *neigh, struct sk_buff *skb)
1da177e4 1320{
1da177e4 1321 struct net_device *dev = neigh->dev;
0ed8ddf4 1322 unsigned int seq;
8f40b161 1323 int err;
1da177e4 1324
0ed8ddf4 1325 do {
e1f16503 1326 __skb_pull(skb, skb_network_offset(skb));
0ed8ddf4
ED		 1327 seq = read_seqbegin(&neigh->ha_lock);
1328 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1329 neigh->ha, NULL, skb->len);
1330 } while (read_seqretry(&neigh->ha_lock, seq));
1331
1da177e4 1332 if (err >= 0)
542d4d68 1333 err = dev_queue_xmit(skb);
1da177e4
LT		 1334 else {
1335 err = -EINVAL;
1336 kfree_skb(skb);
1337 }
1338 return err;
1339}
0a204500 1340EXPORT_SYMBOL(neigh_connected_output);
1da177e4 1341
8f40b161
DM		 1342int neigh_direct_output(struct neighbour *neigh, struct sk_buff *skb)
1343{
1344 return dev_queue_xmit(skb);
1345}
1346EXPORT_SYMBOL(neigh_direct_output);
1347
1da177e4
LT		 1348static void neigh_proxy_process(unsigned long arg)
1349{
1350 struct neigh_table *tbl = (struct neigh_table *)arg;
1351 long sched_next = 0;
1352 unsigned long now = jiffies;
f72051b0 1353 struct sk_buff *skb, *n;
1da177e4
LT		 1354
1355 spin_lock(&tbl->proxy_queue.lock);
1356
f72051b0
DM		 1357 skb_queue_walk_safe(&tbl->proxy_queue, skb, n) {
1358 long tdif = NEIGH_CB(skb)->sched_next - now;
1da177e4 1359
1da177e4 1360 if (tdif <= 0) {
f72051b0 1361 struct net_device *dev = skb->dev;
20e6074e 1362
f72051b0 1363 __skb_unlink(skb, &tbl->proxy_queue);
20e6074e
ED		 1364 if (tbl->proxy_redo && netif_running(dev)) {
1365 rcu_read_lock();
f72051b0 1366 tbl->proxy_redo(skb);
20e6074e
ED		 1367 rcu_read_unlock();
1368 } else {
f72051b0 1369 kfree_skb(skb);
20e6074e 1370 }
1da177e4
LT		 1371
1372 dev_put(dev);
1373 } else if (!sched_next || tdif < sched_next)
1374 sched_next = tdif;
1375 }
1376 del_timer(&tbl->proxy_timer);
1377 if (sched_next)
1378 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1379 spin_unlock(&tbl->proxy_queue.lock);
1380}
1381
1382void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1383 struct sk_buff *skb)
1384{
1385 unsigned long now = jiffies;
63862b5b
AH		 1386
1387 unsigned long sched_next = now + (prandom_u32() %
1f9248e5 1388 NEIGH_VAR(p, PROXY_DELAY));
1da177e4 1389
1f9248e5 1390 if (tbl->proxy_queue.qlen > NEIGH_VAR(p, PROXY_QLEN)) {
1da177e4
LT		 1391 kfree_skb(skb);
1392 return;
1393 }
a61bbcf2
PM		 1394
1395 NEIGH_CB(skb)->sched_next = sched_next;
1396 NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1da177e4
LT		 1397
1398 spin_lock(&tbl->proxy_queue.lock);
1399 if (del_timer(&tbl->proxy_timer)) {
1400 if (time_before(tbl->proxy_timer.expires, sched_next))
1401 sched_next = tbl->proxy_timer.expires;
1402 }
adf30907 1403 skb_dst_drop(skb);
1da177e4
LT		 1404 dev_hold(skb->dev);
1405 __skb_queue_tail(&tbl->proxy_queue, skb);
1406 mod_timer(&tbl->proxy_timer, sched_next);
1407 spin_unlock(&tbl->proxy_queue.lock);
1408}
0a204500 1409EXPORT_SYMBOL(pneigh_enqueue);
1da177e4 1410
97fd5bc7 1411static inline struct neigh_parms *lookup_neigh_parms(struct neigh_table *tbl,
426b5303
EB		 1412 struct net *net, int ifindex)
1413{
1414 struct neigh_parms *p;
1415
75fbfd33 1416 list_for_each_entry(p, &tbl->parms_list, list) {
878628fb 1417 if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
170d6f99 1418 (!p->dev && !ifindex && net_eq(net, &init_net)))
426b5303
EB		 1419 return p;
1420 }
1421
1422 return NULL;
1423}
1da177e4
LT		 1424
1425struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1426 struct neigh_table *tbl)
1427{
cf89d6b2 1428 struct neigh_parms *p;
00829823
SH		 1429 struct net *net = dev_net(dev);
1430 const struct net_device_ops *ops = dev->netdev_ops;
426b5303 1431
cf89d6b2 1432 p = kmemdup(&tbl->parms, sizeof(*p), GFP_KERNEL);
1da177e4 1433 if (p) {
1da177e4
LT		 1434 p->tbl = tbl;
1435 atomic_set(&p->refcnt, 1);
1da177e4 1436 p->reachable_time =
1f9248e5 1437 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
63134803
VF		 1438 dev_hold(dev);
1439 p->dev = dev;
efd7ef1c 1440 write_pnet(&p->net, net);
63134803 1441 p->sysctl_table = NULL;
c7fb64db 1442
00829823 1443 if (ops->ndo_neigh_setup && ops->ndo_neigh_setup(dev, p)) {
63134803 1444 dev_put(dev);
486b51d3
DL		 1445 kfree(p);
1446 return NULL;
1da177e4 1447 }
486b51d3 1448
1da177e4 1449 write_lock_bh(&tbl->lock);
75fbfd33 1450 list_add(&p->list, &tbl->parms.list);
1da177e4 1451 write_unlock_bh(&tbl->lock);
1d4c8c29
JP		 1452
1453 neigh_parms_data_state_cleanall(p);
1da177e4
LT		 1454 }
1455 return p;
1456}
0a204500 1457EXPORT_SYMBOL(neigh_parms_alloc);
1da177e4
LT		 1458
1459static void neigh_rcu_free_parms(struct rcu_head *head)
1460{
1461 struct neigh_parms *parms =
1462 container_of(head, struct neigh_parms, rcu_head);
1463
1464 neigh_parms_put(parms);
1465}
1466
1467void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1468{
1da177e4
LT		 1469 if (!parms || parms == &tbl->parms)
1470 return;
1471 write_lock_bh(&tbl->lock);
75fbfd33
ND		 1472 list_del(&parms->list);
1473 parms->dead = 1;
1da177e4 1474 write_unlock_bh(&tbl->lock);
75fbfd33
ND		 1475 if (parms->dev)
1476 dev_put(parms->dev);
1477 call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1da177e4 1478}
0a204500 1479EXPORT_SYMBOL(neigh_parms_release);
1da177e4 1480
06f0511d 1481static void neigh_parms_destroy(struct neigh_parms *parms)
1da177e4
LT		 1482{
1483 kfree(parms);
1484}
1485
c2ecba71
PE		 1486static struct lock_class_key neigh_table_proxy_queue_class;
1487
d7480fd3
WC		 1488static struct neigh_table *neigh_tables[NEIGH_NR_TABLES] __read_mostly;
1489
1490void neigh_table_init(int index, struct neigh_table *tbl)
1da177e4
LT		 1491{
1492 unsigned long now = jiffies;
1493 unsigned long phsize;
1494
75fbfd33
ND		 1495 INIT_LIST_HEAD(&tbl->parms_list);
1496 list_add(&tbl->parms.list, &tbl->parms_list);
e42ea986 1497 write_pnet(&tbl->parms.net, &init_net);
1da177e4 1498 atomic_set(&tbl->parms.refcnt, 1);
1da177e4 1499 tbl->parms.reachable_time =
1f9248e5 1500 neigh_rand_reach_time(NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME));
1da177e4 1501
1da177e4
LT		 1502 tbl->stats = alloc_percpu(struct neigh_statistics);
1503 if (!tbl->stats)
1504 panic("cannot create neighbour cache statistics");
4ec93edb 1505
1da177e4 1506#ifdef CONFIG_PROC_FS
9b739ba5
AD		 1507 if (!proc_create_data(tbl->id, 0, init_net.proc_net_stat,
1508 &neigh_stat_seq_fops, tbl))
1da177e4 1509 panic("cannot create neighbour proc dir entry");
1da177e4
LT		 1510#endif
1511
cd089336 1512 RCU_INIT_POINTER(tbl->nht, neigh_hash_alloc(3));
1da177e4
LT		 1513
1514 phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
77d04bd9 1515 tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1da177e4 1516
d6bf7817 1517 if (!tbl->nht || !tbl->phash_buckets)
1da177e4
LT		 1518 panic("cannot allocate neighbour cache hashes");
1519
08433eff
YH		 1520 if (!tbl->entry_size)
1521 tbl->entry_size = ALIGN(offsetof(struct neighbour, primary_key) +
1522 tbl->key_len, NEIGH_PRIV_ALIGN);
1523 else
1524 WARN_ON(tbl->entry_size % NEIGH_PRIV_ALIGN);
1525
1da177e4 1526 rwlock_init(&tbl->lock);
203b42f7 1527 INIT_DEFERRABLE_WORK(&tbl->gc_work, neigh_periodic_work);
f618002b 1528 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
1529 tbl->parms.reachable_time);
b24b8a24 1530 setup_timer(&tbl->proxy_timer, neigh_proxy_process, (unsigned long)tbl);
c2ecba71
PE		 1531 skb_queue_head_init_class(&tbl->proxy_queue,
1532 &neigh_table_proxy_queue_class);
1da177e4
LT		 1533
1534 tbl->last_flush = now;
1535 tbl->last_rand = now + tbl->parms.reachable_time * 20;
bd89efc5 1536
d7480fd3 1537 neigh_tables[index] = tbl;
1da177e4 1538}
0a204500 1539EXPORT_SYMBOL(neigh_table_init);
1da177e4 1540
d7480fd3 1541int neigh_table_clear(int index, struct neigh_table *tbl)
1da177e4 1542{
d7480fd3 1543 neigh_tables[index] = NULL;
1da177e4 1544 /* It is not clean... Fix it to unload IPv6 module safely */
a5c30b34 1545 cancel_delayed_work_sync(&tbl->gc_work);
1da177e4
LT		 1546 del_timer_sync(&tbl->proxy_timer);
1547 pneigh_queue_purge(&tbl->proxy_queue);
1548 neigh_ifdown(tbl, NULL);
1549 if (atomic_read(&tbl->entries))
e005d193 1550 pr_crit("neighbour leakage\n");
1da177e4 1551
6193d2be
ED		 1552 call_rcu(&rcu_dereference_protected(tbl->nht, 1)->rcu,
1553 neigh_hash_free_rcu);
d6bf7817 1554 tbl->nht = NULL;
1da177e4
LT		 1555
1556 kfree(tbl->phash_buckets);
1557 tbl->phash_buckets = NULL;
1558
3f192b5c
AD		 1559 remove_proc_entry(tbl->id, init_net.proc_net_stat);
1560
3fcde74b
KK		 1561 free_percpu(tbl->stats);
1562 tbl->stats = NULL;
1563
1da177e4
LT		 1564 return 0;
1565}
0a204500 1566EXPORT_SYMBOL(neigh_table_clear);
1da177e4 1567
d7480fd3
WC		 1568static struct neigh_table *neigh_find_table(int family)
1569{
1570 struct neigh_table *tbl = NULL;
1571
1572 switch (family) {
1573 case AF_INET:
1574 tbl = neigh_tables[NEIGH_ARP_TABLE];
1575 break;
1576 case AF_INET6:
1577 tbl = neigh_tables[NEIGH_ND_TABLE];
1578 break;
1579 case AF_DECnet:
1580 tbl = neigh_tables[NEIGH_DN_TABLE];
1581 break;
1582 }
1583
1584 return tbl;
1585}
1586
661d2967 1587static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh)
1da177e4 1588{
3b1e0a65 1589 struct net *net = sock_net(skb->sk);
a14a49d2
TG		 1590 struct ndmsg *ndm;
1591 struct nlattr *dst_attr;
1da177e4 1592 struct neigh_table *tbl;
d7480fd3 1593 struct neighbour *neigh;
1da177e4 1594 struct net_device *dev = NULL;
a14a49d2 1595 int err = -EINVAL;
1da177e4 1596
110b2499 1597 ASSERT_RTNL();
a14a49d2 1598 if (nlmsg_len(nlh) < sizeof(*ndm))
1da177e4
LT		 1599 goto out;
1600
a14a49d2
TG		 1601 dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1602 if (dst_attr == NULL)
1603 goto out;
1604
1605 ndm = nlmsg_data(nlh);
1606 if (ndm->ndm_ifindex) {
110b2499 1607 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
a14a49d2
TG		 1608 if (dev == NULL) {
1609 err = -ENODEV;
1610 goto out;
1611 }
1612 }
1613
d7480fd3
WC		 1614 tbl = neigh_find_table(ndm->ndm_family);
1615 if (tbl == NULL)
1616 return -EAFNOSUPPORT;
1da177e4 1617
d7480fd3
WC		 1618 if (nla_len(dst_attr) < tbl->key_len)
1619 goto out;
1da177e4 1620
d7480fd3
WC		 1621 if (ndm->ndm_flags & NTF_PROXY) {
1622 err = pneigh_delete(tbl, net, nla_data(dst_attr), dev);
1623 goto out;
1624 }
1da177e4 1625
d7480fd3
WC		 1626 if (dev == NULL)
1627 goto out;
a14a49d2 1628
d7480fd3
WC		 1629 neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1630 if (neigh == NULL) {
1631 err = -ENOENT;
110b2499 1632 goto out;
1da177e4 1633 }
d7480fd3
WC		 1634
1635 err = neigh_update(neigh, NULL, NUD_FAILED,
1636 NEIGH_UPDATE_F_OVERRIDE |
1637 NEIGH_UPDATE_F_ADMIN);
1638 neigh_release(neigh);
a14a49d2 1639
1da177e4
LT		 1640out:
1641 return err;
1642}
1643
661d2967 1644static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh)
1da177e4 1645{
d7480fd3 1646 int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE;
3b1e0a65 1647 struct net *net = sock_net(skb->sk);
5208debd
TG		 1648 struct ndmsg *ndm;
1649 struct nlattr *tb[NDA_MAX+1];
1da177e4
LT		 1650 struct neigh_table *tbl;
1651 struct net_device *dev = NULL;
d7480fd3
WC		 1652 struct neighbour *neigh;
1653 void *dst, *lladdr;
5208debd 1654 int err;
1da177e4 1655
110b2499 1656 ASSERT_RTNL();
5208debd
TG		 1657 err = nlmsg_parse(nlh, sizeof(*ndm), tb, NDA_MAX, NULL);
1658 if (err < 0)
1da177e4
LT		 1659 goto out;
1660
5208debd
TG		 1661 err = -EINVAL;
1662 if (tb[NDA_DST] == NULL)
1663 goto out;
1664
1665 ndm = nlmsg_data(nlh);
1666 if (ndm->ndm_ifindex) {
110b2499 1667 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
5208debd
TG		 1668 if (dev == NULL) {
1669 err = -ENODEV;
1670 goto out;
1671 }
1672
1673 if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len)
110b2499 1674 goto out;
5208debd
TG		 1675 }
1676
d7480fd3
WC		 1677 tbl = neigh_find_table(ndm->ndm_family);
1678 if (tbl == NULL)
1679 return -EAFNOSUPPORT;
1da177e4 1680
d7480fd3
WC		 1681 if (nla_len(tb[NDA_DST]) < tbl->key_len)
1682 goto out;
1683 dst = nla_data(tb[NDA_DST]);
1684 lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
1da177e4 1685
d7480fd3
WC		 1686 if (ndm->ndm_flags & NTF_PROXY) {
1687 struct pneigh_entry *pn;
1688
1689 err = -ENOBUFS;
1690 pn = pneigh_lookup(tbl, net, dst, dev, 1);
1691 if (pn) {
1692 pn->flags = ndm->ndm_flags;
1693 err = 0;
1694 }
1695 goto out;
1696 }
1da177e4 1697
d7480fd3
WC		 1698 if (dev == NULL)
1699 goto out;
62dd9318 1700
d7480fd3
WC		 1701 neigh = neigh_lookup(tbl, dst, dev);
1702 if (neigh == NULL) {
1703 if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
1704 err = -ENOENT;
110b2499 1705 goto out;
1da177e4
LT		 1706 }
1707
d7480fd3
WC		 1708 neigh = __neigh_lookup_errno(tbl, dst, dev);
1709 if (IS_ERR(neigh)) {
1710 err = PTR_ERR(neigh);
1711 goto out;
1712 }
1713 } else {
1714 if (nlh->nlmsg_flags & NLM_F_EXCL) {
1715 err = -EEXIST;
1716 neigh_release(neigh);
110b2499 1717 goto out;
5208debd 1718 }
1da177e4 1719
d7480fd3
WC		 1720 if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
1721 flags &= ~NEIGH_UPDATE_F_OVERRIDE;
1da177e4
LT		 1722 }
1723
d7480fd3
WC		 1724 if (ndm->ndm_flags & NTF_USE) {
1725 neigh_event_send(neigh, NULL);
1726 err = 0;
1727 } else
1728 err = neigh_update(neigh, lladdr, ndm->ndm_state, flags);
1729 neigh_release(neigh);
1730
1da177e4
LT		 1731out:
1732 return err;
1733}
1734
c7fb64db
TG		 1735static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
1736{
ca860fb3
TG		 1737 struct nlattr *nest;
1738
1739 nest = nla_nest_start(skb, NDTA_PARMS);
1740 if (nest == NULL)
1741 return -ENOBUFS;
c7fb64db 1742
9a6308d7
DM		 1743 if ((parms->dev &&
1744 nla_put_u32(skb, NDTPA_IFINDEX, parms->dev->ifindex)) ||
1745 nla_put_u32(skb, NDTPA_REFCNT, atomic_read(&parms->refcnt)) ||
1f9248e5
JP		 1746 nla_put_u32(skb, NDTPA_QUEUE_LENBYTES,
1747 NEIGH_VAR(parms, QUEUE_LEN_BYTES)) ||
9a6308d7
DM		 1748 /* approximative value for deprecated QUEUE_LEN (in packets) */
1749 nla_put_u32(skb, NDTPA_QUEUE_LEN,
1f9248e5
JP		 1750 NEIGH_VAR(parms, QUEUE_LEN_BYTES) / SKB_TRUESIZE(ETH_FRAME_LEN)) ||
1751 nla_put_u32(skb, NDTPA_PROXY_QLEN, NEIGH_VAR(parms, PROXY_QLEN)) ||
1752 nla_put_u32(skb, NDTPA_APP_PROBES, NEIGH_VAR(parms, APP_PROBES)) ||
1753 nla_put_u32(skb, NDTPA_UCAST_PROBES,
1754 NEIGH_VAR(parms, UCAST_PROBES)) ||
1755 nla_put_u32(skb, NDTPA_MCAST_PROBES,
1756 NEIGH_VAR(parms, MCAST_PROBES)) ||
8da86466
YH		 1757 nla_put_u32(skb, NDTPA_MCAST_REPROBES,
1758 NEIGH_VAR(parms, MCAST_REPROBES)) ||
9a6308d7
DM		 1759 nla_put_msecs(skb, NDTPA_REACHABLE_TIME, parms->reachable_time) ||
1760 nla_put_msecs(skb, NDTPA_BASE_REACHABLE_TIME,
1f9248e5
JP		 1761 NEIGH_VAR(parms, BASE_REACHABLE_TIME)) ||
1762 nla_put_msecs(skb, NDTPA_GC_STALETIME,
1763 NEIGH_VAR(parms, GC_STALETIME)) ||
9a6308d7 1764 nla_put_msecs(skb, NDTPA_DELAY_PROBE_TIME,
1f9248e5
JP		 1765 NEIGH_VAR(parms, DELAY_PROBE_TIME)) ||
1766 nla_put_msecs(skb, NDTPA_RETRANS_TIME,
1767 NEIGH_VAR(parms, RETRANS_TIME)) ||
1768 nla_put_msecs(skb, NDTPA_ANYCAST_DELAY,
1769 NEIGH_VAR(parms, ANYCAST_DELAY)) ||
1770 nla_put_msecs(skb, NDTPA_PROXY_DELAY,
1771 NEIGH_VAR(parms, PROXY_DELAY)) ||
1772 nla_put_msecs(skb, NDTPA_LOCKTIME,
1773 NEIGH_VAR(parms, LOCKTIME)))
9a6308d7 1774 goto nla_put_failure;
ca860fb3 1775 return nla_nest_end(skb, nest);
c7fb64db 1776
ca860fb3 1777nla_put_failure:
bc3ed28c
TG		 1778 nla_nest_cancel(skb, nest);
1779 return -EMSGSIZE;
c7fb64db
TG		 1780}
1781
ca860fb3
TG		 1782static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl,
1783 u32 pid, u32 seq, int type, int flags)
c7fb64db
TG		 1784{
1785 struct nlmsghdr *nlh;
1786 struct ndtmsg *ndtmsg;
1787
ca860fb3
TG		 1788 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1789 if (nlh == NULL)
26932566 1790 return -EMSGSIZE;
c7fb64db 1791
ca860fb3 1792 ndtmsg = nlmsg_data(nlh);
c7fb64db
TG		 1793
1794 read_lock_bh(&tbl->lock);
1795 ndtmsg->ndtm_family = tbl->family;
9ef1d4c7
PM		 1796 ndtmsg->ndtm_pad1 = 0;
1797 ndtmsg->ndtm_pad2 = 0;
c7fb64db 1798
9a6308d7
DM		 1799 if (nla_put_string(skb, NDTA_NAME, tbl->id) ||
1800 nla_put_msecs(skb, NDTA_GC_INTERVAL, tbl->gc_interval) ||
1801 nla_put_u32(skb, NDTA_THRESH1, tbl->gc_thresh1) ||
1802 nla_put_u32(skb, NDTA_THRESH2, tbl->gc_thresh2) ||
1803 nla_put_u32(skb, NDTA_THRESH3, tbl->gc_thresh3))
1804 goto nla_put_failure;
c7fb64db
TG		 1805 {
1806 unsigned long now = jiffies;
1807 unsigned int flush_delta = now - tbl->last_flush;
1808 unsigned int rand_delta = now - tbl->last_rand;
d6bf7817 1809 struct neigh_hash_table *nht;
c7fb64db
TG		 1810 struct ndt_config ndc = {
1811 .ndtc_key_len = tbl->key_len,
1812 .ndtc_entry_size = tbl->entry_size,
1813 .ndtc_entries = atomic_read(&tbl->entries),
1814 .ndtc_last_flush = jiffies_to_msecs(flush_delta),
1815 .ndtc_last_rand = jiffies_to_msecs(rand_delta),
c7fb64db
TG		 1816 .ndtc_proxy_qlen = tbl->proxy_queue.qlen,
1817 };
1818
d6bf7817
ED		 1819 rcu_read_lock_bh();
1820 nht = rcu_dereference_bh(tbl->nht);
2c2aba6c 1821 ndc.ndtc_hash_rnd = nht->hash_rnd[0];
cd089336 1822 ndc.ndtc_hash_mask = ((1 << nht->hash_shift) - 1);
d6bf7817
ED		 1823 rcu_read_unlock_bh();
1824
9a6308d7
DM		 1825 if (nla_put(skb, NDTA_CONFIG, sizeof(ndc), &ndc))
1826 goto nla_put_failure;
c7fb64db
TG		 1827 }
1828
1829 {
1830 int cpu;
1831 struct ndt_stats ndst;
1832
1833 memset(&ndst, 0, sizeof(ndst));
1834
6f912042 1835 for_each_possible_cpu(cpu) {
c7fb64db
TG		 1836 struct neigh_statistics *st;
1837
c7fb64db
TG		 1838 st = per_cpu_ptr(tbl->stats, cpu);
1839 ndst.ndts_allocs += st->allocs;
1840 ndst.ndts_destroys += st->destroys;
1841 ndst.ndts_hash_grows += st->hash_grows;
1842 ndst.ndts_res_failed += st->res_failed;
1843 ndst.ndts_lookups += st->lookups;
1844 ndst.ndts_hits += st->hits;
1845 ndst.ndts_rcv_probes_mcast += st->rcv_probes_mcast;
1846 ndst.ndts_rcv_probes_ucast += st->rcv_probes_ucast;
1847 ndst.ndts_periodic_gc_runs += st->periodic_gc_runs;
1848 ndst.ndts_forced_gc_runs += st->forced_gc_runs;
1849 }
1850
9a6308d7
DM		 1851 if (nla_put(skb, NDTA_STATS, sizeof(ndst), &ndst))
1852 goto nla_put_failure;
c7fb64db
TG		 1853 }
1854
1855 BUG_ON(tbl->parms.dev);
1856 if (neightbl_fill_parms(skb, &tbl->parms) < 0)
ca860fb3 1857 goto nla_put_failure;
c7fb64db
TG		 1858
1859 read_unlock_bh(&tbl->lock);
053c095a
JB		 1860 nlmsg_end(skb, nlh);
1861 return 0;
c7fb64db 1862
ca860fb3 1863nla_put_failure:
c7fb64db 1864 read_unlock_bh(&tbl->lock);
26932566
PM		 1865 nlmsg_cancel(skb, nlh);
1866 return -EMSGSIZE;
c7fb64db
TG		 1867}
1868
ca860fb3
TG		 1869static int neightbl_fill_param_info(struct sk_buff *skb,
1870 struct neigh_table *tbl,
c7fb64db 1871 struct neigh_parms *parms,
ca860fb3
TG		 1872 u32 pid, u32 seq, int type,
1873 unsigned int flags)
c7fb64db
TG		 1874{
1875 struct ndtmsg *ndtmsg;
1876 struct nlmsghdr *nlh;
1877
ca860fb3
TG		 1878 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1879 if (nlh == NULL)
26932566 1880 return -EMSGSIZE;
c7fb64db 1881
ca860fb3 1882 ndtmsg = nlmsg_data(nlh);
c7fb64db
TG		 1883
1884 read_lock_bh(&tbl->lock);
1885 ndtmsg->ndtm_family = tbl->family;
9ef1d4c7
PM		 1886 ndtmsg->ndtm_pad1 = 0;
1887 ndtmsg->ndtm_pad2 = 0;
c7fb64db 1888
ca860fb3
TG		 1889 if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 ||
1890 neightbl_fill_parms(skb, parms) < 0)
1891 goto errout;
c7fb64db
TG		 1892
1893 read_unlock_bh(&tbl->lock);
053c095a
JB		 1894 nlmsg_end(skb, nlh);
1895 return 0;
ca860fb3 1896errout:
c7fb64db 1897 read_unlock_bh(&tbl->lock);
26932566
PM		 1898 nlmsg_cancel(skb, nlh);
1899 return -EMSGSIZE;
c7fb64db 1900}
4ec93edb 1901
ef7c79ed 1902static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = {
6b3f8674
TG		 1903 [NDTA_NAME] = { .type = NLA_STRING },
1904 [NDTA_THRESH1] = { .type = NLA_U32 },
1905 [NDTA_THRESH2] = { .type = NLA_U32 },
1906 [NDTA_THRESH3] = { .type = NLA_U32 },
1907 [NDTA_GC_INTERVAL] = { .type = NLA_U64 },
1908 [NDTA_PARMS] = { .type = NLA_NESTED },
1909};
1910
ef7c79ed 1911static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = {
6b3f8674
TG		 1912 [NDTPA_IFINDEX] = { .type = NLA_U32 },
1913 [NDTPA_QUEUE_LEN] = { .type = NLA_U32 },
1914 [NDTPA_PROXY_QLEN] = { .type = NLA_U32 },
1915 [NDTPA_APP_PROBES] = { .type = NLA_U32 },
1916 [NDTPA_UCAST_PROBES] = { .type = NLA_U32 },
1917 [NDTPA_MCAST_PROBES] = { .type = NLA_U32 },
8da86466 1918 [NDTPA_MCAST_REPROBES] = { .type = NLA_U32 },
6b3f8674
TG		 1919 [NDTPA_BASE_REACHABLE_TIME] = { .type = NLA_U64 },
1920 [NDTPA_GC_STALETIME] = { .type = NLA_U64 },
1921 [NDTPA_DELAY_PROBE_TIME] = { .type = NLA_U64 },
1922 [NDTPA_RETRANS_TIME] = { .type = NLA_U64 },
1923 [NDTPA_ANYCAST_DELAY] = { .type = NLA_U64 },
1924 [NDTPA_PROXY_DELAY] = { .type = NLA_U64 },
1925 [NDTPA_LOCKTIME] = { .type = NLA_U64 },
1926};
1927
661d2967 1928static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh)
c7fb64db 1929{
3b1e0a65 1930 struct net *net = sock_net(skb->sk);
c7fb64db 1931 struct neigh_table *tbl;
6b3f8674
TG		 1932 struct ndtmsg *ndtmsg;
1933 struct nlattr *tb[NDTA_MAX+1];
d7480fd3
WC		 1934 bool found = false;
1935 int err, tidx;
c7fb64db 1936
6b3f8674
TG		 1937 err = nlmsg_parse(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
1938 nl_neightbl_policy);
1939 if (err < 0)
1940 goto errout;
c7fb64db 1941
6b3f8674
TG		 1942 if (tb[NDTA_NAME] == NULL) {
1943 err = -EINVAL;
1944 goto errout;
1945 }
1946
1947 ndtmsg = nlmsg_data(nlh);
d7480fd3
WC		 1948
1949 for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
1950 tbl = neigh_tables[tidx];
1951 if (!tbl)
1952 continue;
c7fb64db
TG		 1953 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
1954 continue;
d7480fd3
WC		 1955 if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0) {
1956 found = true;
c7fb64db 1957 break;
d7480fd3 1958 }
c7fb64db
TG		 1959 }
1960
d7480fd3
WC		 1961 if (!found)
1962 return -ENOENT;
c7fb64db 1963
4ec93edb 1964 /*
c7fb64db
TG		 1965 * We acquire tbl->lock to be nice to the periodic timers and
1966 * make sure they always see a consistent set of values.
1967 */
1968 write_lock_bh(&tbl->lock);
1969
6b3f8674
TG		 1970 if (tb[NDTA_PARMS]) {
1971 struct nlattr *tbp[NDTPA_MAX+1];
c7fb64db 1972 struct neigh_parms *p;
6b3f8674 1973 int i, ifindex = 0;
c7fb64db 1974
6b3f8674
TG		 1975 err = nla_parse_nested(tbp, NDTPA_MAX, tb[NDTA_PARMS],
1976 nl_ntbl_parm_policy);
1977 if (err < 0)
1978 goto errout_tbl_lock;
c7fb64db 1979
6b3f8674
TG		 1980 if (tbp[NDTPA_IFINDEX])
1981 ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
c7fb64db 1982
97fd5bc7 1983 p = lookup_neigh_parms(tbl, net, ifindex);
c7fb64db
TG		 1984 if (p == NULL) {
1985 err = -ENOENT;
6b3f8674 1986 goto errout_tbl_lock;
c7fb64db 1987 }
c7fb64db 1988
6b3f8674
TG		 1989 for (i = 1; i <= NDTPA_MAX; i++) {
1990 if (tbp[i] == NULL)
1991 continue;
c7fb64db 1992
6b3f8674
TG		 1993 switch (i) {
1994 case NDTPA_QUEUE_LEN:
1f9248e5
JP		 1995 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
1996 nla_get_u32(tbp[i]) *
1997 SKB_TRUESIZE(ETH_FRAME_LEN));
8b5c171b
ED		 1998 break;
1999 case NDTPA_QUEUE_LENBYTES:
1f9248e5
JP		 2000 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2001 nla_get_u32(tbp[i]));
6b3f8674
TG		 2002 break;
2003 case NDTPA_PROXY_QLEN:
1f9248e5
JP		 2004 NEIGH_VAR_SET(p, PROXY_QLEN,
2005 nla_get_u32(tbp[i]));
6b3f8674
TG		 2006 break;
2007 case NDTPA_APP_PROBES:
1f9248e5
JP		 2008 NEIGH_VAR_SET(p, APP_PROBES,
2009 nla_get_u32(tbp[i]));
6b3f8674
TG		 2010 break;
2011 case NDTPA_UCAST_PROBES:
1f9248e5
JP		 2012 NEIGH_VAR_SET(p, UCAST_PROBES,
2013 nla_get_u32(tbp[i]));
6b3f8674
TG		 2014 break;
2015 case NDTPA_MCAST_PROBES:
1f9248e5
JP		 2016 NEIGH_VAR_SET(p, MCAST_PROBES,
2017 nla_get_u32(tbp[i]));
6b3f8674 2018 break;
8da86466
YH		 2019 case NDTPA_MCAST_REPROBES:
2020 NEIGH_VAR_SET(p, MCAST_REPROBES,
2021 nla_get_u32(tbp[i]));
2022 break;
6b3f8674 2023 case NDTPA_BASE_REACHABLE_TIME:
1f9248e5
JP		 2024 NEIGH_VAR_SET(p, BASE_REACHABLE_TIME,
2025 nla_get_msecs(tbp[i]));
4bf6980d
JFR		 2026 /* update reachable_time as well, otherwise, the change will
2027 * only be effective after the next time neigh_periodic_work
2028 * decides to recompute it (can be multiple minutes)
2029 */
2030 p->reachable_time =
2031 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
6b3f8674
TG		 2032 break;
2033 case NDTPA_GC_STALETIME:
1f9248e5
JP		 2034 NEIGH_VAR_SET(p, GC_STALETIME,
2035 nla_get_msecs(tbp[i]));
6b3f8674
TG		 2036 break;
2037 case NDTPA_DELAY_PROBE_TIME:
1f9248e5
JP		 2038 NEIGH_VAR_SET(p, DELAY_PROBE_TIME,
2039 nla_get_msecs(tbp[i]));
6b3f8674
TG		 2040 break;
2041 case NDTPA_RETRANS_TIME:
1f9248e5
JP		 2042 NEIGH_VAR_SET(p, RETRANS_TIME,
2043 nla_get_msecs(tbp[i]));
6b3f8674
TG		 2044 break;
2045 case NDTPA_ANYCAST_DELAY:
3977458c
JP		 2046 NEIGH_VAR_SET(p, ANYCAST_DELAY,
2047 nla_get_msecs(tbp[i]));
6b3f8674
TG		 2048 break;
2049 case NDTPA_PROXY_DELAY:
3977458c
JP		 2050 NEIGH_VAR_SET(p, PROXY_DELAY,
2051 nla_get_msecs(tbp[i]));
6b3f8674
TG		 2052 break;
2053 case NDTPA_LOCKTIME:
3977458c
JP		 2054 NEIGH_VAR_SET(p, LOCKTIME,
2055 nla_get_msecs(tbp[i]));
6b3f8674
TG		 2056 break;
2057 }
2058 }
2059 }
c7fb64db 2060
dc25c676
G		 2061 err = -ENOENT;
2062 if ((tb[NDTA_THRESH1] || tb[NDTA_THRESH2] ||
2063 tb[NDTA_THRESH3] || tb[NDTA_GC_INTERVAL]) &&
2064 !net_eq(net, &init_net))
2065 goto errout_tbl_lock;
2066
6b3f8674
TG		 2067 if (tb[NDTA_THRESH1])
2068 tbl->gc_thresh1 = nla_get_u32(tb[NDTA_THRESH1]);
c7fb64db 2069
6b3f8674
TG		 2070 if (tb[NDTA_THRESH2])
2071 tbl->gc_thresh2 = nla_get_u32(tb[NDTA_THRESH2]);
c7fb64db 2072
6b3f8674
TG		 2073 if (tb[NDTA_THRESH3])
2074 tbl->gc_thresh3 = nla_get_u32(tb[NDTA_THRESH3]);
c7fb64db 2075
6b3f8674
TG		 2076 if (tb[NDTA_GC_INTERVAL])
2077 tbl->gc_interval = nla_get_msecs(tb[NDTA_GC_INTERVAL]);
c7fb64db
TG		 2078
2079 err = 0;
2080
6b3f8674 2081errout_tbl_lock:
c7fb64db 2082 write_unlock_bh(&tbl->lock);
6b3f8674 2083errout:
c7fb64db
TG		 2084 return err;
2085}
2086
c8822a4e 2087static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
c7fb64db 2088{
3b1e0a65 2089 struct net *net = sock_net(skb->sk);
ca860fb3
TG		 2090 int family, tidx, nidx = 0;
2091 int tbl_skip = cb->args[0];
2092 int neigh_skip = cb->args[1];
c7fb64db
TG		 2093 struct neigh_table *tbl;
2094
ca860fb3 2095 family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
c7fb64db 2096
d7480fd3 2097 for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
c7fb64db
TG		 2098 struct neigh_parms *p;
2099
d7480fd3
WC		 2100 tbl = neigh_tables[tidx];
2101 if (!tbl)
2102 continue;
2103
ca860fb3 2104 if (tidx < tbl_skip || (family && tbl->family != family))
c7fb64db
TG		 2105 continue;
2106
15e47304 2107 if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).portid,
ca860fb3 2108 cb->nlh->nlmsg_seq, RTM_NEWNEIGHTBL,
7b46a644 2109 NLM_F_MULTI) < 0)
c7fb64db
TG		 2110 break;
2111
75fbfd33
ND		 2112 nidx = 0;
2113 p = list_next_entry(&tbl->parms, list);
2114 list_for_each_entry_from(p, &tbl->parms_list, list) {
878628fb 2115 if (!net_eq(neigh_parms_net(p), net))
426b5303
EB		 2116 continue;
2117
efc683fc
GK		 2118 if (nidx < neigh_skip)
2119 goto next;
c7fb64db 2120
ca860fb3 2121 if (neightbl_fill_param_info(skb, tbl, p,
15e47304 2122 NETLINK_CB(cb->skb).portid,
ca860fb3
TG		 2123 cb->nlh->nlmsg_seq,
2124 RTM_NEWNEIGHTBL,
7b46a644 2125 NLM_F_MULTI) < 0)
c7fb64db 2126 goto out;
efc683fc
GK		 2127 next:
2128 nidx++;
c7fb64db
TG		 2129 }
2130
ca860fb3 2131 neigh_skip = 0;
c7fb64db
TG		 2132 }
2133out:
ca860fb3
TG		 2134 cb->args[0] = tidx;
2135 cb->args[1] = nidx;
c7fb64db
TG		 2136
2137 return skb->len;
2138}
1da177e4 2139
8b8aec50
TG		 2140static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
2141 u32 pid, u32 seq, int type, unsigned int flags)
1da177e4
LT		 2142{
2143 unsigned long now = jiffies;
1da177e4 2144 struct nda_cacheinfo ci;
8b8aec50
TG		 2145 struct nlmsghdr *nlh;
2146 struct ndmsg *ndm;
2147
2148 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2149 if (nlh == NULL)
26932566 2150 return -EMSGSIZE;
1da177e4 2151
8b8aec50
TG		 2152 ndm = nlmsg_data(nlh);
2153 ndm->ndm_family = neigh->ops->family;
9ef1d4c7
PM		 2154 ndm->ndm_pad1 = 0;
2155 ndm->ndm_pad2 = 0;
8b8aec50
TG		 2156 ndm->ndm_flags = neigh->flags;
2157 ndm->ndm_type = neigh->type;
2158 ndm->ndm_ifindex = neigh->dev->ifindex;
1da177e4 2159
9a6308d7
DM		 2160 if (nla_put(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key))
2161 goto nla_put_failure;
8b8aec50
TG		 2162
2163 read_lock_bh(&neigh->lock);
2164 ndm->ndm_state = neigh->nud_state;
0ed8ddf4
ED		 2165 if (neigh->nud_state & NUD_VALID) {
2166 char haddr[MAX_ADDR_LEN];
2167
2168 neigh_ha_snapshot(haddr, neigh, neigh->dev);
2169 if (nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, haddr) < 0) {
2170 read_unlock_bh(&neigh->lock);
2171 goto nla_put_failure;
2172 }
8b8aec50
TG		 2173 }
2174
b9f5f52c
SH		 2175 ci.ndm_used = jiffies_to_clock_t(now - neigh->used);
2176 ci.ndm_confirmed = jiffies_to_clock_t(now - neigh->confirmed);
2177 ci.ndm_updated = jiffies_to_clock_t(now - neigh->updated);
8b8aec50
TG		 2178 ci.ndm_refcnt = atomic_read(&neigh->refcnt) - 1;
2179 read_unlock_bh(&neigh->lock);
2180
9a6308d7
DM		 2181 if (nla_put_u32(skb, NDA_PROBES, atomic_read(&neigh->probes)) ||
2182 nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
2183 goto nla_put_failure;
8b8aec50 2184
053c095a
JB		 2185 nlmsg_end(skb, nlh);
2186 return 0;
8b8aec50
TG		 2187
2188nla_put_failure:
26932566
PM		 2189 nlmsg_cancel(skb, nlh);
2190 return -EMSGSIZE;
1da177e4
LT		 2191}
2192
84920c14
TZ		 2193static int pneigh_fill_info(struct sk_buff *skb, struct pneigh_entry *pn,
2194 u32 pid, u32 seq, int type, unsigned int flags,
2195 struct neigh_table *tbl)
2196{
2197 struct nlmsghdr *nlh;
2198 struct ndmsg *ndm;
2199
2200 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2201 if (nlh == NULL)
2202 return -EMSGSIZE;
2203
2204 ndm = nlmsg_data(nlh);
2205 ndm->ndm_family = tbl->family;
2206 ndm->ndm_pad1 = 0;
2207 ndm->ndm_pad2 = 0;
2208 ndm->ndm_flags = pn->flags | NTF_PROXY;
545469f7 2209 ndm->ndm_type = RTN_UNICAST;
84920c14
TZ		 2210 ndm->ndm_ifindex = pn->dev->ifindex;
2211 ndm->ndm_state = NUD_NONE;
2212
9a6308d7
DM		 2213 if (nla_put(skb, NDA_DST, tbl->key_len, pn->key))
2214 goto nla_put_failure;
84920c14 2215
053c095a
JB		 2216 nlmsg_end(skb, nlh);
2217 return 0;
84920c14
TZ		 2218
2219nla_put_failure:
2220 nlmsg_cancel(skb, nlh);
2221 return -EMSGSIZE;
2222}
2223
d961db35
TG		 2224static void neigh_update_notify(struct neighbour *neigh)
2225{
2226 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
2227 __neigh_notify(neigh, RTM_NEWNEIGH, 0);
2228}
1da177e4
LT		 2229
2230static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2231 struct netlink_callback *cb)
2232{
767e97e1 2233 struct net *net = sock_net(skb->sk);
1da177e4
LT		 2234 struct neighbour *n;
2235 int rc, h, s_h = cb->args[1];
2236 int idx, s_idx = idx = cb->args[2];
d6bf7817 2237 struct neigh_hash_table *nht;
1da177e4 2238
d6bf7817
ED		 2239 rcu_read_lock_bh();
2240 nht = rcu_dereference_bh(tbl->nht);
2241
4bd6683b 2242 for (h = s_h; h < (1 << nht->hash_shift); h++) {
1da177e4
LT		 2243 if (h > s_h)
2244 s_idx = 0;
767e97e1
ED		 2245 for (n = rcu_dereference_bh(nht->hash_buckets[h]), idx = 0;
2246 n != NULL;
2247 n = rcu_dereference_bh(n->next)) {
09ad9bc7 2248 if (!net_eq(dev_net(n->dev), net))
426b5303 2249 continue;
efc683fc
GK		 2250 if (idx < s_idx)
2251 goto next;
15e47304 2252 if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
1da177e4 2253 cb->nlh->nlmsg_seq,
b6544c0b 2254 RTM_NEWNEIGH,
7b46a644 2255 NLM_F_MULTI) < 0) {
1da177e4
LT		 2256 rc = -1;
2257 goto out;
2258 }
767e97e1 2259next:
efc683fc 2260 idx++;
1da177e4 2261 }
1da177e4
LT		 2262 }
2263 rc = skb->len;
2264out:
d6bf7817 2265 rcu_read_unlock_bh();
1da177e4
LT		 2266 cb->args[1] = h;
2267 cb->args[2] = idx;
2268 return rc;
2269}
2270
84920c14
TZ		 2271static int pneigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2272 struct netlink_callback *cb)
2273{
2274 struct pneigh_entry *n;
2275 struct net *net = sock_net(skb->sk);
2276 int rc, h, s_h = cb->args[3];
2277 int idx, s_idx = idx = cb->args[4];
2278
2279 read_lock_bh(&tbl->lock);
2280
4bd6683b 2281 for (h = s_h; h <= PNEIGH_HASHMASK; h++) {
84920c14
TZ		 2282 if (h > s_h)
2283 s_idx = 0;
2284 for (n = tbl->phash_buckets[h], idx = 0; n; n = n->next) {
2285 if (dev_net(n->dev) != net)
2286 continue;
2287 if (idx < s_idx)
2288 goto next;
15e47304 2289 if (pneigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
84920c14
TZ		 2290 cb->nlh->nlmsg_seq,
2291 RTM_NEWNEIGH,
7b46a644 2292 NLM_F_MULTI, tbl) < 0) {
84920c14
TZ		 2293 read_unlock_bh(&tbl->lock);
2294 rc = -1;
2295 goto out;
2296 }
2297 next:
2298 idx++;
2299 }
2300 }
2301
2302 read_unlock_bh(&tbl->lock);
2303 rc = skb->len;
2304out:
2305 cb->args[3] = h;
2306 cb->args[4] = idx;
2307 return rc;
2308
2309}
2310
c8822a4e 2311static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
1da177e4
LT		 2312{
2313 struct neigh_table *tbl;
2314 int t, family, s_t;
84920c14 2315 int proxy = 0;
4bd6683b 2316 int err;
1da177e4 2317
8b8aec50 2318 family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
84920c14
TZ		 2319
2320 /* check for full ndmsg structure presence, family member is
2321 * the same for both structures
2322 */
2323 if (nlmsg_len(cb->nlh) >= sizeof(struct ndmsg) &&
2324 ((struct ndmsg *) nlmsg_data(cb->nlh))->ndm_flags == NTF_PROXY)
2325 proxy = 1;
2326
1da177e4
LT		 2327 s_t = cb->args[0];
2328
d7480fd3
WC		 2329 for (t = 0; t < NEIGH_NR_TABLES; t++) {
2330 tbl = neigh_tables[t];
2331
2332 if (!tbl)
2333 continue;
1da177e4
LT		 2334 if (t < s_t || (family && tbl->family != family))
2335 continue;
2336 if (t > s_t)
2337 memset(&cb->args[1], 0, sizeof(cb->args) -
2338 sizeof(cb->args[0]));
84920c14
TZ		 2339 if (proxy)
2340 err = pneigh_dump_table(tbl, skb, cb);
2341 else
2342 err = neigh_dump_table(tbl, skb, cb);
4bd6683b
ED		 2343 if (err < 0)
2344 break;
1da177e4 2345 }
1da177e4
LT		 2346
2347 cb->args[0] = t;
2348 return skb->len;
2349}
2350
2351void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
2352{
2353 int chain;
d6bf7817 2354 struct neigh_hash_table *nht;
1da177e4 2355
d6bf7817
ED		 2356 rcu_read_lock_bh();
2357 nht = rcu_dereference_bh(tbl->nht);
2358
767e97e1 2359 read_lock(&tbl->lock); /* avoid resizes */
cd089336 2360 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
1da177e4
LT		 2361 struct neighbour *n;
2362
767e97e1
ED		 2363 for (n = rcu_dereference_bh(nht->hash_buckets[chain]);
2364 n != NULL;
2365 n = rcu_dereference_bh(n->next))
1da177e4
LT		 2366 cb(n, cookie);
2367 }
d6bf7817
ED		 2368 read_unlock(&tbl->lock);
2369 rcu_read_unlock_bh();
1da177e4
LT		 2370}
2371EXPORT_SYMBOL(neigh_for_each);
2372
2373/* The tbl->lock must be held as a writer and BH disabled. */
2374void __neigh_for_each_release(struct neigh_table *tbl,
2375 int (*cb)(struct neighbour *))
2376{
2377 int chain;
d6bf7817 2378 struct neigh_hash_table *nht;
1da177e4 2379
d6bf7817
ED		 2380 nht = rcu_dereference_protected(tbl->nht,
2381 lockdep_is_held(&tbl->lock));
cd089336 2382 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
767e97e1
ED		 2383 struct neighbour *n;
2384 struct neighbour __rcu **np;
1da177e4 2385
d6bf7817 2386 np = &nht->hash_buckets[chain];
767e97e1
ED		 2387 while ((n = rcu_dereference_protected(*np,
2388 lockdep_is_held(&tbl->lock))) != NULL) {
1da177e4
LT		 2389 int release;
2390
2391 write_lock(&n->lock);
2392 release = cb(n);
2393 if (release) {
767e97e1
ED		 2394 rcu_assign_pointer(*np,
2395 rcu_dereference_protected(n->next,
2396 lockdep_is_held(&tbl->lock)));
1da177e4
LT		 2397 n->dead = 1;
2398 } else
2399 np = &n->next;
2400 write_unlock(&n->lock);
4f494554
TG		 2401 if (release)
2402 neigh_cleanup_and_release(n);
1da177e4
LT		 2403 }
2404 }
2405}
2406EXPORT_SYMBOL(__neigh_for_each_release);
2407
b79bda3d 2408int neigh_xmit(int index, struct net_device *dev,
4fd3d7d9
EB		 2409 const void *addr, struct sk_buff *skb)
2410{
b79bda3d
EB		 2411 int err = -EAFNOSUPPORT;
2412 if (likely(index < NEIGH_NR_TABLES)) {
4fd3d7d9
EB		 2413 struct neigh_table *tbl;
2414 struct neighbour *neigh;
2415
b79bda3d 2416 tbl = neigh_tables[index];
4fd3d7d9
EB		 2417 if (!tbl)
2418 goto out;
2419 neigh = __neigh_lookup_noref(tbl, addr, dev);
2420 if (!neigh)
2421 neigh = __neigh_create(tbl, addr, dev, false);
2422 err = PTR_ERR(neigh);
2423 if (IS_ERR(neigh))
2424 goto out_kfree_skb;
2425 err = neigh->output(neigh, skb);
2426 }
b79bda3d
EB		 2427 else if (index == NEIGH_LINK_TABLE) {
2428 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
2429 addr, NULL, skb->len);
2430 if (err < 0)
2431 goto out_kfree_skb;
2432 err = dev_queue_xmit(skb);
2433 }
4fd3d7d9
EB		 2434out:
2435 return err;
2436out_kfree_skb:
2437 kfree_skb(skb);
2438 goto out;
2439}
2440EXPORT_SYMBOL(neigh_xmit);
2441
1da177e4
LT		 2442#ifdef CONFIG_PROC_FS
2443
2444static struct neighbour *neigh_get_first(struct seq_file *seq)
2445{
2446 struct neigh_seq_state *state = seq->private;
1218854a 2447 struct net *net = seq_file_net(seq);
d6bf7817 2448 struct neigh_hash_table *nht = state->nht;
1da177e4
LT		 2449 struct neighbour *n = NULL;
2450 int bucket = state->bucket;
2451
2452 state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
cd089336 2453 for (bucket = 0; bucket < (1 << nht->hash_shift); bucket++) {
767e97e1 2454 n = rcu_dereference_bh(nht->hash_buckets[bucket]);
1da177e4
LT		 2455
2456 while (n) {
878628fb 2457 if (!net_eq(dev_net(n->dev), net))
426b5303 2458 goto next;
1da177e4
LT		 2459 if (state->neigh_sub_iter) {
2460 loff_t fakep = 0;
2461 void *v;
2462
2463 v = state->neigh_sub_iter(state, n, &fakep);
2464 if (!v)
2465 goto next;
2466 }
2467 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2468 break;
2469 if (n->nud_state & ~NUD_NOARP)
2470 break;
767e97e1
ED		 2471next:
2472 n = rcu_dereference_bh(n->next);
1da177e4
LT		 2473 }
2474
2475 if (n)
2476 break;
2477 }
2478 state->bucket = bucket;
2479
2480 return n;
2481}
2482
2483static struct neighbour *neigh_get_next(struct seq_file *seq,
2484 struct neighbour *n,
2485 loff_t *pos)
2486{
2487 struct neigh_seq_state *state = seq->private;
1218854a 2488 struct net *net = seq_file_net(seq);
d6bf7817 2489 struct neigh_hash_table *nht = state->nht;
1da177e4
LT		 2490
2491 if (state->neigh_sub_iter) {
2492 void *v = state->neigh_sub_iter(state, n, pos);
2493 if (v)
2494 return n;
2495 }
767e97e1 2496 n = rcu_dereference_bh(n->next);
1da177e4
LT		 2497
2498 while (1) {
2499 while (n) {
878628fb 2500 if (!net_eq(dev_net(n->dev), net))
426b5303 2501 goto next;
1da177e4
LT		 2502 if (state->neigh_sub_iter) {
2503 void *v = state->neigh_sub_iter(state, n, pos);
2504 if (v)
2505 return n;
2506 goto next;
2507 }
2508 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2509 break;
2510
2511 if (n->nud_state & ~NUD_NOARP)
2512 break;
767e97e1
ED		 2513next:
2514 n = rcu_dereference_bh(n->next);
1da177e4
LT		 2515 }
2516
2517 if (n)
2518 break;
2519
cd089336 2520 if (++state->bucket >= (1 << nht->hash_shift))
1da177e4
LT		 2521 break;
2522
767e97e1 2523 n = rcu_dereference_bh(nht->hash_buckets[state->bucket]);
1da177e4
LT		 2524 }
2525
2526 if (n && pos)
2527 --(*pos);
2528 return n;
2529}
2530
2531static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
2532{
2533 struct neighbour *n = neigh_get_first(seq);
2534
2535 if (n) {
745e2031 2536 --(*pos);
1da177e4
LT		 2537 while (*pos) {
2538 n = neigh_get_next(seq, n, pos);
2539 if (!n)
2540 break;
2541 }
2542 }
2543 return *pos ? NULL : n;
2544}
2545
2546static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
2547{
2548 struct neigh_seq_state *state = seq->private;
1218854a 2549 struct net *net = seq_file_net(seq);
1da177e4
LT		 2550 struct neigh_table *tbl = state->tbl;
2551 struct pneigh_entry *pn = NULL;
2552 int bucket = state->bucket;
2553
2554 state->flags |= NEIGH_SEQ_IS_PNEIGH;
2555 for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
2556 pn = tbl->phash_buckets[bucket];
878628fb 2557 while (pn && !net_eq(pneigh_net(pn), net))
426b5303 2558 pn = pn->next;
1da177e4
LT		 2559 if (pn)
2560 break;
2561 }
2562 state->bucket = bucket;
2563
2564 return pn;
2565}
2566
2567static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
2568 struct pneigh_entry *pn,
2569 loff_t *pos)
2570{
2571 struct neigh_seq_state *state = seq->private;
1218854a 2572 struct net *net = seq_file_net(seq);
1da177e4
LT		 2573 struct neigh_table *tbl = state->tbl;
2574
df07a94c
JBD		 2575 do {
2576 pn = pn->next;
2577 } while (pn && !net_eq(pneigh_net(pn), net));
2578
1da177e4
LT		 2579 while (!pn) {
2580 if (++state->bucket > PNEIGH_HASHMASK)
2581 break;
2582 pn = tbl->phash_buckets[state->bucket];
878628fb 2583 while (pn && !net_eq(pneigh_net(pn), net))
426b5303 2584 pn = pn->next;
1da177e4
LT		 2585 if (pn)
2586 break;
2587 }
2588
2589 if (pn && pos)
2590 --(*pos);
2591
2592 return pn;
2593}
2594
2595static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
2596{
2597 struct pneigh_entry *pn = pneigh_get_first(seq);
2598
2599 if (pn) {
745e2031 2600 --(*pos);
1da177e4
LT		 2601 while (*pos) {
2602 pn = pneigh_get_next(seq, pn, pos);
2603 if (!pn)
2604 break;
2605 }
2606 }
2607 return *pos ? NULL : pn;
2608}
2609
2610static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
2611{
2612 struct neigh_seq_state *state = seq->private;
2613 void *rc;
745e2031 2614 loff_t idxpos = *pos;
1da177e4 2615
745e2031 2616 rc = neigh_get_idx(seq, &idxpos);
1da177e4 2617 if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
745e2031 2618 rc = pneigh_get_idx(seq, &idxpos);
1da177e4
LT		 2619
2620 return rc;
2621}
2622
2623void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
d6bf7817 2624 __acquires(rcu_bh)
1da177e4
LT		 2625{
2626 struct neigh_seq_state *state = seq->private;
1da177e4
LT		 2627
2628 state->tbl = tbl;
2629 state->bucket = 0;
2630 state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
2631
d6bf7817
ED		 2632 rcu_read_lock_bh();
2633 state->nht = rcu_dereference_bh(tbl->nht);
767e97e1 2634
745e2031 2635 return *pos ? neigh_get_idx_any(seq, pos) : SEQ_START_TOKEN;
1da177e4
LT		 2636}
2637EXPORT_SYMBOL(neigh_seq_start);
2638
2639void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2640{
2641 struct neigh_seq_state *state;
2642 void *rc;
2643
2644 if (v == SEQ_START_TOKEN) {
bff69732 2645 rc = neigh_get_first(seq);
1da177e4
LT		 2646 goto out;
2647 }
2648
2649 state = seq->private;
2650 if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
2651 rc = neigh_get_next(seq, v, NULL);
2652 if (rc)
2653 goto out;
2654 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2655 rc = pneigh_get_first(seq);
2656 } else {
2657 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
2658 rc = pneigh_get_next(seq, v, NULL);
2659 }
2660out:
2661 ++(*pos);
2662 return rc;
2663}
2664EXPORT_SYMBOL(neigh_seq_next);
2665
2666void neigh_seq_stop(struct seq_file *seq, void *v)
d6bf7817 2667 __releases(rcu_bh)
1da177e4 2668{
d6bf7817 2669 rcu_read_unlock_bh();
1da177e4
LT		 2670}
2671EXPORT_SYMBOL(neigh_seq_stop);
2672
2673/* statistics via seq_file */
2674
2675static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
2676{
81c1ebfc 2677 struct neigh_table *tbl = seq->private;
1da177e4
LT		 2678 int cpu;
2679
2680 if (*pos == 0)
2681 return SEQ_START_TOKEN;
4ec93edb 2682
0f23174a 2683 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
1da177e4
LT		 2684 if (!cpu_possible(cpu))
2685 continue;
2686 *pos = cpu+1;
2687 return per_cpu_ptr(tbl->stats, cpu);
2688 }
2689 return NULL;
2690}
2691
2692static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2693{
81c1ebfc 2694 struct neigh_table *tbl = seq->private;
1da177e4
LT		 2695 int cpu;
2696
0f23174a 2697 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
1da177e4
LT		 2698 if (!cpu_possible(cpu))
2699 continue;
2700 *pos = cpu+1;
2701 return per_cpu_ptr(tbl->stats, cpu);
2702 }
2703 return NULL;
2704}
2705
2706static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
2707{
2708
2709}
2710
2711static int neigh_stat_seq_show(struct seq_file *seq, void *v)
2712{
81c1ebfc 2713 struct neigh_table *tbl = seq->private;
1da177e4
LT		 2714 struct neigh_statistics *st = v;
2715
2716 if (v == SEQ_START_TOKEN) {
9a6d276e 2717 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");
1da177e4
LT		 2718 return 0;
2719 }
2720
2721 seq_printf(seq, "%08x %08lx %08lx %08lx %08lx %08lx %08lx "
9a6d276e 2722 "%08lx %08lx %08lx %08lx %08lx\n",
1da177e4
LT		 2723 atomic_read(&tbl->entries),
2724
2725 st->allocs,
2726 st->destroys,
2727 st->hash_grows,
2728
2729 st->lookups,
2730 st->hits,
2731
2732 st->res_failed,
2733
2734 st->rcv_probes_mcast,
2735 st->rcv_probes_ucast,
2736
2737 st->periodic_gc_runs,
9a6d276e
NH		 2738 st->forced_gc_runs,
2739 st->unres_discards
1da177e4
LT		 2740 );
2741
2742 return 0;
2743}
2744
f690808e 2745static const struct seq_operations neigh_stat_seq_ops = {
1da177e4
LT		 2746 .start = neigh_stat_seq_start,
2747 .next = neigh_stat_seq_next,
2748 .stop = neigh_stat_seq_stop,
2749 .show = neigh_stat_seq_show,
2750};
2751
2752static int neigh_stat_seq_open(struct inode *inode, struct file *file)
2753{
2754 int ret = seq_open(file, &neigh_stat_seq_ops);
2755
2756 if (!ret) {
2757 struct seq_file *sf = file->private_data;
d9dda78b 2758 sf->private = PDE_DATA(inode);
1da177e4
LT		 2759 }
2760 return ret;
2761};
2762
9a32144e 2763static const struct file_operations neigh_stat_seq_fops = {
1da177e4
LT		 2764 .owner = THIS_MODULE,
2765 .open = neigh_stat_seq_open,
2766 .read = seq_read,
2767 .llseek = seq_lseek,
2768 .release = seq_release,
2769};
2770
2771#endif /* CONFIG_PROC_FS */
2772
339bf98f
TG		 2773static inline size_t neigh_nlmsg_size(void)
2774{
2775 return NLMSG_ALIGN(sizeof(struct ndmsg))
2776 + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2777 + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */
2778 + nla_total_size(sizeof(struct nda_cacheinfo))
2779 + nla_total_size(4); /* NDA_PROBES */
2780}
2781
b8673311 2782static void __neigh_notify(struct neighbour *n, int type, int flags)
1da177e4 2783{
c346dca1 2784 struct net *net = dev_net(n->dev);
8b8aec50 2785 struct sk_buff *skb;
b8673311 2786 int err = -ENOBUFS;
1da177e4 2787
339bf98f 2788 skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC);
8b8aec50 2789 if (skb == NULL)
b8673311 2790 goto errout;
1da177e4 2791
b8673311 2792 err = neigh_fill_info(skb, n, 0, 0, type, flags);
26932566
PM		 2793 if (err < 0) {
2794 /* -EMSGSIZE implies BUG in neigh_nlmsg_size() */
2795 WARN_ON(err == -EMSGSIZE);
2796 kfree_skb(skb);
2797 goto errout;
2798 }
1ce85fe4
PNA		 2799 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
2800 return;
b8673311
TG		 2801errout:
2802 if (err < 0)
426b5303 2803 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
1da177e4
LT		 2804}
2805
b8673311 2806void neigh_app_ns(struct neighbour *n)
1da177e4 2807{
b8673311
TG		 2808 __neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST);
2809}
0a204500 2810EXPORT_SYMBOL(neigh_app_ns);
1da177e4
LT		 2811
2812#ifdef CONFIG_SYSCTL
b93196dc 2813static int zero;
555445cd 2814static int int_max = INT_MAX;
b93196dc 2815static int unres_qlen_max = INT_MAX / SKB_TRUESIZE(ETH_FRAME_LEN);
1da177e4 2816
fe2c6338
JP		 2817static int proc_unres_qlen(struct ctl_table *ctl, int write,
2818 void __user *buffer, size_t *lenp, loff_t *ppos)
8b5c171b
ED		 2819{
2820 int size, ret;
fe2c6338 2821 struct ctl_table tmp = *ctl;
8b5c171b 2822
ce46cc64
SW		 2823 tmp.extra1 = &zero;
2824 tmp.extra2 = &unres_qlen_max;
8b5c171b 2825 tmp.data = &size;
ce46cc64
SW		 2826
2827 size = *(int *)ctl->data / SKB_TRUESIZE(ETH_FRAME_LEN);
2828 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2829
8b5c171b
ED		 2830 if (write && !ret)
2831 *(int *)ctl->data = size * SKB_TRUESIZE(ETH_FRAME_LEN);
2832 return ret;
2833}
2834
1d4c8c29
JP		 2835static struct neigh_parms *neigh_get_dev_parms_rcu(struct net_device *dev,
2836 int family)
2837{
bba24896
JP		 2838 switch (family) {
2839 case AF_INET:
1d4c8c29 2840 return __in_dev_arp_parms_get_rcu(dev);
bba24896
JP		 2841 case AF_INET6:
2842 return __in6_dev_nd_parms_get_rcu(dev);
2843 }
1d4c8c29
JP		 2844 return NULL;
2845}
2846
2847static void neigh_copy_dflt_parms(struct net *net, struct neigh_parms *p,
2848 int index)
2849{
2850 struct net_device *dev;
2851 int family = neigh_parms_family(p);
2852
2853 rcu_read_lock();
2854 for_each_netdev_rcu(net, dev) {
2855 struct neigh_parms *dst_p =
2856 neigh_get_dev_parms_rcu(dev, family);
2857
2858 if (dst_p && !test_bit(index, dst_p->data_state))
2859 dst_p->data[index] = p->data[index];
2860 }
2861 rcu_read_unlock();
2862}
2863
2864static void neigh_proc_update(struct ctl_table *ctl, int write)
2865{
2866 struct net_device *dev = ctl->extra1;
2867 struct neigh_parms *p = ctl->extra2;
77d47afb 2868 struct net *net = neigh_parms_net(p);
1d4c8c29
JP		 2869 int index = (int *) ctl->data - p->data;
2870
2871 if (!write)
2872 return;
2873
2874 set_bit(index, p->data_state);
2875 if (!dev) /* NULL dev means this is default value */
2876 neigh_copy_dflt_parms(net, p, index);
2877}
2878
1f9248e5
JP		 2879static int neigh_proc_dointvec_zero_intmax(struct ctl_table *ctl, int write,
2880 void __user *buffer,
2881 size_t *lenp, loff_t *ppos)
2882{
2883 struct ctl_table tmp = *ctl;
1d4c8c29 2884 int ret;
1f9248e5
JP		 2885
2886 tmp.extra1 = &zero;
2887 tmp.extra2 = &int_max;
2888
1d4c8c29
JP		 2889 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2890 neigh_proc_update(ctl, write);
2891 return ret;
1f9248e5
JP		 2892}
2893
cb5b09c1
JP		 2894int neigh_proc_dointvec(struct ctl_table *ctl, int write,
2895 void __user *buffer, size_t *lenp, loff_t *ppos)
2896{
1d4c8c29
JP		 2897 int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
2898
2899 neigh_proc_update(ctl, write);
2900 return ret;
cb5b09c1
JP		 2901}
2902EXPORT_SYMBOL(neigh_proc_dointvec);
2903
2904int neigh_proc_dointvec_jiffies(struct ctl_table *ctl, int write,
2905 void __user *buffer,
2906 size_t *lenp, loff_t *ppos)
2907{
1d4c8c29
JP		 2908 int ret = proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
2909
2910 neigh_proc_update(ctl, write);
2911 return ret;
cb5b09c1
JP		 2912}
2913EXPORT_SYMBOL(neigh_proc_dointvec_jiffies);
2914
2915static int neigh_proc_dointvec_userhz_jiffies(struct ctl_table *ctl, int write,
2916 void __user *buffer,
2917 size_t *lenp, loff_t *ppos)
2918{
1d4c8c29
JP		 2919 int ret = proc_dointvec_userhz_jiffies(ctl, write, buffer, lenp, ppos);
2920
2921 neigh_proc_update(ctl, write);
2922 return ret;
cb5b09c1
JP		 2923}
2924
2925int neigh_proc_dointvec_ms_jiffies(struct ctl_table *ctl, int write,
2926 void __user *buffer,
2927 size_t *lenp, loff_t *ppos)
2928{
1d4c8c29
JP		 2929 int ret = proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
2930
2931 neigh_proc_update(ctl, write);
2932 return ret;
cb5b09c1
JP		 2933}
2934EXPORT_SYMBOL(neigh_proc_dointvec_ms_jiffies);
2935
2936static int neigh_proc_dointvec_unres_qlen(struct ctl_table *ctl, int write,
2937 void __user *buffer,
2938 size_t *lenp, loff_t *ppos)
2939{
1d4c8c29
JP		 2940 int ret = proc_unres_qlen(ctl, write, buffer, lenp, ppos);
2941
2942 neigh_proc_update(ctl, write);
2943 return ret;
cb5b09c1
JP		 2944}
2945
4bf6980d
JFR		 2946static int neigh_proc_base_reachable_time(struct ctl_table *ctl, int write,
2947 void __user *buffer,
2948 size_t *lenp, loff_t *ppos)
2949{
2950 struct neigh_parms *p = ctl->extra2;
2951 int ret;
2952
2953 if (strcmp(ctl->procname, "base_reachable_time") == 0)
2954 ret = neigh_proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
2955 else if (strcmp(ctl->procname, "base_reachable_time_ms") == 0)
2956 ret = neigh_proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
2957 else
2958 ret = -1;
2959
2960 if (write && ret == 0) {
2961 /* update reachable_time as well, otherwise, the change will
2962 * only be effective after the next time neigh_periodic_work
2963 * decides to recompute it
2964 */
2965 p->reachable_time =
2966 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
2967 }
2968 return ret;
2969}
2970
1f9248e5
JP		 2971#define NEIGH_PARMS_DATA_OFFSET(index) \
2972 (&((struct neigh_parms *) 0)->data[index])
2973
2974#define NEIGH_SYSCTL_ENTRY(attr, data_attr, name, mval, proc) \
2975 [NEIGH_VAR_ ## attr] = { \
2976 .procname = name, \
2977 .data = NEIGH_PARMS_DATA_OFFSET(NEIGH_VAR_ ## data_attr), \
2978 .maxlen = sizeof(int), \
2979 .mode = mval, \
2980 .proc_handler = proc, \
2981 }
2982
2983#define NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(attr, name) \
2984 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_zero_intmax)
2985
2986#define NEIGH_SYSCTL_JIFFIES_ENTRY(attr, name) \
cb5b09c1 2987 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_jiffies)
1f9248e5
JP		 2988
2989#define NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(attr, name) \
cb5b09c1 2990 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_userhz_jiffies)
1f9248e5
JP		 2991
2992#define NEIGH_SYSCTL_MS_JIFFIES_ENTRY(attr, name) \
cb5b09c1 2993 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_ms_jiffies)
1f9248e5
JP		 2994
2995#define NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(attr, data_attr, name) \
cb5b09c1 2996 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_ms_jiffies)
1f9248e5
JP		 2997
2998#define NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(attr, data_attr, name) \
cb5b09c1 2999 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_unres_qlen)
54716e3b 3000
1da177e4
LT		 3001static struct neigh_sysctl_table {
3002 struct ctl_table_header *sysctl_header;
8b5c171b 3003 struct ctl_table neigh_vars[NEIGH_VAR_MAX + 1];
ab32ea5d 3004} neigh_sysctl_template __read_mostly = {
1da177e4 3005 .neigh_vars = {
1f9248e5
JP		 3006 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_PROBES, "mcast_solicit"),
3007 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(UCAST_PROBES, "ucast_solicit"),
3008 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(APP_PROBES, "app_solicit"),
8da86466 3009 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_REPROBES, "mcast_resolicit"),
1f9248e5
JP		 3010 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(RETRANS_TIME, "retrans_time"),
3011 NEIGH_SYSCTL_JIFFIES_ENTRY(BASE_REACHABLE_TIME, "base_reachable_time"),
3012 NEIGH_SYSCTL_JIFFIES_ENTRY(DELAY_PROBE_TIME, "delay_first_probe_time"),
3013 NEIGH_SYSCTL_JIFFIES_ENTRY(GC_STALETIME, "gc_stale_time"),
3014 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(QUEUE_LEN_BYTES, "unres_qlen_bytes"),
3015 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(PROXY_QLEN, "proxy_qlen"),
3016 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(ANYCAST_DELAY, "anycast_delay"),
3017 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(PROXY_DELAY, "proxy_delay"),
3018 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(LOCKTIME, "locktime"),
3019 NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(QUEUE_LEN, QUEUE_LEN_BYTES, "unres_qlen"),
3020 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(RETRANS_TIME_MS, RETRANS_TIME, "retrans_time_ms"),
3021 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(BASE_REACHABLE_TIME_MS, BASE_REACHABLE_TIME, "base_reachable_time_ms"),
8b5c171b 3022 [NEIGH_VAR_GC_INTERVAL] = {
1da177e4
LT		 3023 .procname = "gc_interval",
3024 .maxlen = sizeof(int),
3025 .mode = 0644,
6d9f239a 3026 .proc_handler = proc_dointvec_jiffies,
1da177e4 3027 },
8b5c171b 3028 [NEIGH_VAR_GC_THRESH1] = {
1da177e4
LT		 3029 .procname = "gc_thresh1",
3030 .maxlen = sizeof(int),
3031 .mode = 0644,
555445cd
FF		 3032 .extra1 = &zero,
3033 .extra2 = &int_max,
3034 .proc_handler = proc_dointvec_minmax,
1da177e4 3035 },
8b5c171b 3036 [NEIGH_VAR_GC_THRESH2] = {
1da177e4
LT		 3037 .procname = "gc_thresh2",
3038 .maxlen = sizeof(int),
3039 .mode = 0644,
555445cd
FF		 3040 .extra1 = &zero,
3041 .extra2 = &int_max,
3042 .proc_handler = proc_dointvec_minmax,
1da177e4 3043 },
8b5c171b 3044 [NEIGH_VAR_GC_THRESH3] = {
1da177e4
LT		 3045 .procname = "gc_thresh3",
3046 .maxlen = sizeof(int),
3047 .mode = 0644,
555445cd
FF		 3048 .extra1 = &zero,
3049 .extra2 = &int_max,
3050 .proc_handler = proc_dointvec_minmax,
1da177e4 3051 },
c3bac5a7 3052 {},
1da177e4
LT		 3053 },
3054};
3055
3056int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
73af614a 3057 proc_handler *handler)
1da177e4 3058{
1f9248e5 3059 int i;
3c607bbb 3060 struct neigh_sysctl_table *t;
1f9248e5 3061 const char *dev_name_source;
8f40a1f9 3062 char neigh_path[ sizeof("net//neigh/") + IFNAMSIZ + IFNAMSIZ ];
73af614a 3063 char *p_name;
1da177e4 3064
3c607bbb 3065 t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL);
1da177e4 3066 if (!t)
3c607bbb
PE		 3067 goto err;
3068
b194c1f1 3069 for (i = 0; i < NEIGH_VAR_GC_INTERVAL; i++) {
1f9248e5 3070 t->neigh_vars[i].data += (long) p;
cb5b09c1 3071 t->neigh_vars[i].extra1 = dev;
1d4c8c29 3072 t->neigh_vars[i].extra2 = p;
cb5b09c1 3073 }
1da177e4
LT		 3074
3075 if (dev) {
3076 dev_name_source = dev->name;
d12af679 3077 /* Terminate the table early */
8b5c171b
ED		 3078 memset(&t->neigh_vars[NEIGH_VAR_GC_INTERVAL], 0,
3079 sizeof(t->neigh_vars[NEIGH_VAR_GC_INTERVAL]));
1da177e4 3080 } else {
9ecf07a1 3081 struct neigh_table *tbl = p->tbl;
8f40a1f9 3082 dev_name_source = "default";
9ecf07a1
MK		 3083 t->neigh_vars[NEIGH_VAR_GC_INTERVAL].data = &tbl->gc_interval;
3084 t->neigh_vars[NEIGH_VAR_GC_THRESH1].data = &tbl->gc_thresh1;
3085 t->neigh_vars[NEIGH_VAR_GC_THRESH2].data = &tbl->gc_thresh2;
3086 t->neigh_vars[NEIGH_VAR_GC_THRESH3].data = &tbl->gc_thresh3;
1da177e4
LT		 3087 }
3088
f8572d8f 3089 if (handler) {
1da177e4 3090 /* RetransTime */
8b5c171b 3091 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].proc_handler = handler;
1da177e4 3092 /* ReachableTime */
8b5c171b 3093 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler = handler;
1da177e4 3094 /* RetransTime (in milliseconds)*/
8b5c171b 3095 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].proc_handler = handler;
1da177e4 3096 /* ReachableTime (in milliseconds) */
8b5c171b 3097 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler = handler;
4bf6980d
JFR		 3098 } else {
3099 /* Those handlers will update p->reachable_time after
3100 * base_reachable_time(_ms) is set to ensure the new timer starts being
3101 * applied after the next neighbour update instead of waiting for
3102 * neigh_periodic_work to update its value (can be multiple minutes)
3103 * So any handler that replaces them should do this as well
3104 */
3105 /* ReachableTime */
3106 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler =
3107 neigh_proc_base_reachable_time;
3108 /* ReachableTime (in milliseconds) */
3109 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler =
3110 neigh_proc_base_reachable_time;
1da177e4
LT		 3111 }
3112
464dc801
EB		 3113 /* Don't export sysctls to unprivileged users */
3114 if (neigh_parms_net(p)->user_ns != &init_user_ns)
3115 t->neigh_vars[0].procname = NULL;
3116
73af614a
JP		 3117 switch (neigh_parms_family(p)) {
3118 case AF_INET:
3119 p_name = "ipv4";
3120 break;
3121 case AF_INET6:
3122 p_name = "ipv6";
3123 break;
3124 default:
3125 BUG();
3126 }
3127
8f40a1f9
EB		 3128 snprintf(neigh_path, sizeof(neigh_path), "net/%s/neigh/%s",
3129 p_name, dev_name_source);
4ab438fc 3130 t->sysctl_header =
8f40a1f9 3131 register_net_sysctl(neigh_parms_net(p), neigh_path, t->neigh_vars);
3c607bbb 3132 if (!t->sysctl_header)
8f40a1f9 3133 goto free;
3c607bbb 3134
1da177e4
LT		 3135 p->sysctl_table = t;
3136 return 0;
3137
3c607bbb 3138free:
1da177e4 3139 kfree(t);
3c607bbb
PE		 3140err:
3141 return -ENOBUFS;
1da177e4 3142}
0a204500 3143EXPORT_SYMBOL(neigh_sysctl_register);
1da177e4
LT		 3144
3145void neigh_sysctl_unregister(struct neigh_parms *p)
3146{
3147 if (p->sysctl_table) {
3148 struct neigh_sysctl_table *t = p->sysctl_table;
3149 p->sysctl_table = NULL;
5dd3df10 3150 unregister_net_sysctl_table(t->sysctl_header);
1da177e4
LT		 3151 kfree(t);
3152 }
3153}
0a204500 3154EXPORT_SYMBOL(neigh_sysctl_unregister);
1da177e4
LT		 3155
3156#endif /* CONFIG_SYSCTL */
3157
c8822a4e
TG		 3158static int __init neigh_init(void)
3159{
c7ac8679
GR		 3160 rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL, NULL);
3161 rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL, NULL);
3162 rtnl_register(PF_UNSPEC, RTM_GETNEIGH, NULL, neigh_dump_info, NULL);
c8822a4e 3163
c7ac8679
GR		 3164 rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info,
3165 NULL);
3166 rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL, NULL);
c8822a4e
TG		 3167
3168 return 0;
3169}
3170
3171subsys_initcall(neigh_init);
3172
Linux 6.x block layer and io_uring tree(s)