1 /* 2 * Generic address resolution entity 3 * 4 * Authors: 5 * Pedro Roque <roque@di.fc.ul.pt> 6 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> 7 * 8 * This program is free software; you can redistribute it and/or 9 * modify it under the terms of the GNU General Public License 10 * as published by the Free Software Foundation; either version 11 * 2 of the License, or (at your option) any later version. 12 * 13 * Fixes: 14 * Vitaly E. Lavrov releasing NULL neighbor in neigh_add. 15 * Harald Welte Add neighbour cache statistics like rtstat 16 */ 17 18 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 19 20 #include <linux/slab.h> 21 #include <linux/types.h> 22 #include <linux/kernel.h> 23 #include <linux/module.h> 24 #include <linux/socket.h> 25 #include <linux/netdevice.h> 26 #include <linux/proc_fs.h> 27 #ifdef CONFIG_SYSCTL 28 #include <linux/sysctl.h> 29 #endif 30 #include <linux/times.h> 31 #include <net/net_namespace.h> 32 #include <net/neighbour.h> 33 #include <net/dst.h> 34 #include <net/sock.h> 35 #include <net/netevent.h> 36 #include <net/netlink.h> 37 #include <linux/rtnetlink.h> 38 #include <linux/random.h> 39 #include <linux/string.h> 40 #include <linux/log2.h> 41 #include <linux/inetdevice.h> 42 #include <net/addrconf.h> 43 44 #define DEBUG 45 #define NEIGH_DEBUG 1 46 #define neigh_dbg(level, fmt, ...) \ 47 do { \ 48 if (level <= NEIGH_DEBUG) \ 49 pr_debug(fmt, ##__VA_ARGS__); \ 50 } while (0) 51 52 #define PNEIGH_HASHMASK 0xF 53 54 static void neigh_timer_handler(unsigned long arg); 55 static void __neigh_notify(struct neighbour *n, int type, int flags); 56 static void neigh_update_notify(struct neighbour *neigh); 57 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev); 58 59 static struct neigh_table *neigh_tables; 60 #ifdef CONFIG_PROC_FS 61 static const struct file_operations neigh_stat_seq_fops; 62 #endif 63 64 /* 65 Neighbour hash table buckets are protected with rwlock tbl->lock. 66 67 - All the scans/updates to hash buckets MUST be made under this lock. 68 - NOTHING clever should be made under this lock: no callbacks 69 to protocol backends, no attempts to send something to network. 70 It will result in deadlocks, if backend/driver wants to use neighbour 71 cache. 72 - If the entry requires some non-trivial actions, increase 73 its reference count and release table lock. 74 75 Neighbour entries are protected: 76 - with reference count. 77 - with rwlock neigh->lock 78 79 Reference count prevents destruction. 80 81 neigh->lock mainly serializes ll address data and its validity state. 82 However, the same lock is used to protect another entry fields: 83 - timer 84 - resolution queue 85 86 Again, nothing clever shall be made under neigh->lock, 87 the most complicated procedure, which we allow is dev->hard_header. 88 It is supposed, that dev->hard_header is simplistic and does 89 not make callbacks to neighbour tables. 90 91 The last lock is neigh_tbl_lock. It is pure SMP lock, protecting 92 list of neighbour tables. This list is used only in process context, 93 */ 94 95 static DEFINE_RWLOCK(neigh_tbl_lock); 96 97 static int neigh_blackhole(struct neighbour *neigh, struct sk_buff *skb) 98 { 99 kfree_skb(skb); 100 return -ENETDOWN; 101 } 102 103 static void neigh_cleanup_and_release(struct neighbour *neigh) 104 { 105 if (neigh->parms->neigh_cleanup) 106 neigh->parms->neigh_cleanup(neigh); 107 108 __neigh_notify(neigh, RTM_DELNEIGH, 0); 109 neigh_release(neigh); 110 } 111 112 /* 113 * It is random distribution in the interval (1/2)*base...(3/2)*base. 114 * It corresponds to default IPv6 settings and is not overridable, 115 * because it is really reasonable choice. 116 */ 117 118 unsigned long neigh_rand_reach_time(unsigned long base) 119 { 120 return base ? (prandom_u32() % base) + (base >> 1) : 0; 121 } 122 EXPORT_SYMBOL(neigh_rand_reach_time); 123 124 125 static int neigh_forced_gc(struct neigh_table *tbl) 126 { 127 int shrunk = 0; 128 int i; 129 struct neigh_hash_table *nht; 130 131 NEIGH_CACHE_STAT_INC(tbl, forced_gc_runs); 132 133 write_lock_bh(&tbl->lock); 134 nht = rcu_dereference_protected(tbl->nht, 135 lockdep_is_held(&tbl->lock)); 136 for (i = 0; i < (1 << nht->hash_shift); i++) { 137 struct neighbour *n; 138 struct neighbour __rcu **np; 139 140 np = &nht->hash_buckets[i]; 141 while ((n = rcu_dereference_protected(*np, 142 lockdep_is_held(&tbl->lock))) != NULL) { 143 /* Neighbour record may be discarded if: 144 * - nobody refers to it. 145 * - it is not permanent 146 */ 147 write_lock(&n->lock); 148 if (atomic_read(&n->refcnt) == 1 && 149 !(n->nud_state & NUD_PERMANENT)) { 150 rcu_assign_pointer(*np, 151 rcu_dereference_protected(n->next, 152 lockdep_is_held(&tbl->lock))); 153 n->dead = 1; 154 shrunk = 1; 155 write_unlock(&n->lock); 156 neigh_cleanup_and_release(n); 157 continue; 158 } 159 write_unlock(&n->lock); 160 np = &n->next; 161 } 162 } 163 164 tbl->last_flush = jiffies; 165 166 write_unlock_bh(&tbl->lock); 167 168 return shrunk; 169 } 170 171 static void neigh_add_timer(struct neighbour *n, unsigned long when) 172 { 173 neigh_hold(n); 174 if (unlikely(mod_timer(&n->timer, when))) { 175 printk("NEIGH: BUG, double timer add, state is %x\n", 176 n->nud_state); 177 dump_stack(); 178 } 179 } 180 181 static int neigh_del_timer(struct neighbour *n) 182 { 183 if ((n->nud_state & NUD_IN_TIMER) && 184 del_timer(&n->timer)) { 185 neigh_release(n); 186 return 1; 187 } 188 return 0; 189 } 190 191 static void pneigh_queue_purge(struct sk_buff_head *list) 192 { 193 struct sk_buff *skb; 194 195 while ((skb = skb_dequeue(list)) != NULL) { 196 dev_put(skb->dev); 197 kfree_skb(skb); 198 } 199 } 200 201 static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev) 202 { 203 int i; 204 struct neigh_hash_table *nht; 205 206 nht = rcu_dereference_protected(tbl->nht, 207 lockdep_is_held(&tbl->lock)); 208 209 for (i = 0; i < (1 << nht->hash_shift); i++) { 210 struct neighbour *n; 211 struct neighbour __rcu **np = &nht->hash_buckets[i]; 212 213 while ((n = rcu_dereference_protected(*np, 214 lockdep_is_held(&tbl->lock))) != NULL) { 215 if (dev && n->dev != dev) { 216 np = &n->next; 217 continue; 218 } 219 rcu_assign_pointer(*np, 220 rcu_dereference_protected(n->next, 221 lockdep_is_held(&tbl->lock))); 222 write_lock(&n->lock); 223 neigh_del_timer(n); 224 n->dead = 1; 225 226 if (atomic_read(&n->refcnt) != 1) { 227 /* The most unpleasant situation. 228 We must destroy neighbour entry, 229 but someone still uses it. 230 231 The destroy will be delayed until 232 the last user releases us, but 233 we must kill timers etc. and move 234 it to safe state. 235 */ 236 __skb_queue_purge(&n->arp_queue); 237 n->arp_queue_len_bytes = 0; 238 n->output = neigh_blackhole; 239 if (n->nud_state & NUD_VALID) 240 n->nud_state = NUD_NOARP; 241 else 242 n->nud_state = NUD_NONE; 243 neigh_dbg(2, "neigh %p is stray\n", n); 244 } 245 write_unlock(&n->lock); 246 neigh_cleanup_and_release(n); 247 } 248 } 249 } 250 251 void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev) 252 { 253 write_lock_bh(&tbl->lock); 254 neigh_flush_dev(tbl, dev); 255 write_unlock_bh(&tbl->lock); 256 } 257 EXPORT_SYMBOL(neigh_changeaddr); 258 259 int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev) 260 { 261 write_lock_bh(&tbl->lock); 262 neigh_flush_dev(tbl, dev); 263 pneigh_ifdown(tbl, dev); 264 write_unlock_bh(&tbl->lock); 265 266 del_timer_sync(&tbl->proxy_timer); 267 pneigh_queue_purge(&tbl->proxy_queue); 268 return 0; 269 } 270 EXPORT_SYMBOL(neigh_ifdown); 271 272 static struct neighbour *neigh_alloc(struct neigh_table *tbl, struct net_device *dev) 273 { 274 struct neighbour *n = NULL; 275 unsigned long now = jiffies; 276 int entries; 277 278 entries = atomic_inc_return(&tbl->entries) - 1; 279 if (entries >= tbl->gc_thresh3 || 280 (entries >= tbl->gc_thresh2 && 281 time_after(now, tbl->last_flush + 5 * HZ))) { 282 if (!neigh_forced_gc(tbl) && 283 entries >= tbl->gc_thresh3) 284 goto out_entries; 285 } 286 287 n = kzalloc(tbl->entry_size + dev->neigh_priv_len, GFP_ATOMIC); 288 if (!n) 289 goto out_entries; 290 291 __skb_queue_head_init(&n->arp_queue); 292 rwlock_init(&n->lock); 293 seqlock_init(&n->ha_lock); 294 n->updated = n->used = now; 295 n->nud_state = NUD_NONE; 296 n->output = neigh_blackhole; 297 seqlock_init(&n->hh.hh_lock); 298 n->parms = neigh_parms_clone(&tbl->parms); 299 setup_timer(&n->timer, neigh_timer_handler, (unsigned long)n); 300 301 NEIGH_CACHE_STAT_INC(tbl, allocs); 302 n->tbl = tbl; 303 atomic_set(&n->refcnt, 1); 304 n->dead = 1; 305 out: 306 return n; 307 308 out_entries: 309 atomic_dec(&tbl->entries); 310 goto out; 311 } 312 313 static void neigh_get_hash_rnd(u32 *x) 314 { 315 get_random_bytes(x, sizeof(*x)); 316 *x |= 1; 317 } 318 319 static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift) 320 { 321 size_t size = (1 << shift) * sizeof(struct neighbour *); 322 struct neigh_hash_table *ret; 323 struct neighbour __rcu **buckets; 324 int i; 325 326 ret = kmalloc(sizeof(*ret), GFP_ATOMIC); 327 if (!ret) 328 return NULL; 329 if (size <= PAGE_SIZE) 330 buckets = kzalloc(size, GFP_ATOMIC); 331 else 332 buckets = (struct neighbour __rcu **) 333 __get_free_pages(GFP_ATOMIC | __GFP_ZERO, 334 get_order(size)); 335 if (!buckets) { 336 kfree(ret); 337 return NULL; 338 } 339 ret->hash_buckets = buckets; 340 ret->hash_shift = shift; 341 for (i = 0; i < NEIGH_NUM_HASH_RND; i++) 342 neigh_get_hash_rnd(&ret->hash_rnd[i]); 343 return ret; 344 } 345 346 static void neigh_hash_free_rcu(struct rcu_head *head) 347 { 348 struct neigh_hash_table *nht = container_of(head, 349 struct neigh_hash_table, 350 rcu); 351 size_t size = (1 << nht->hash_shift) * sizeof(struct neighbour *); 352 struct neighbour __rcu **buckets = nht->hash_buckets; 353 354 if (size <= PAGE_SIZE) 355 kfree(buckets); 356 else 357 free_pages((unsigned long)buckets, get_order(size)); 358 kfree(nht); 359 } 360 361 static struct neigh_hash_table *neigh_hash_grow(struct neigh_table *tbl, 362 unsigned long new_shift) 363 { 364 unsigned int i, hash; 365 struct neigh_hash_table *new_nht, *old_nht; 366 367 NEIGH_CACHE_STAT_INC(tbl, hash_grows); 368 369 old_nht = rcu_dereference_protected(tbl->nht, 370 lockdep_is_held(&tbl->lock)); 371 new_nht = neigh_hash_alloc(new_shift); 372 if (!new_nht) 373 return old_nht; 374 375 for (i = 0; i < (1 << old_nht->hash_shift); i++) { 376 struct neighbour *n, *next; 377 378 for (n = rcu_dereference_protected(old_nht->hash_buckets[i], 379 lockdep_is_held(&tbl->lock)); 380 n != NULL; 381 n = next) { 382 hash = tbl->hash(n->primary_key, n->dev, 383 new_nht->hash_rnd); 384 385 hash >>= (32 - new_nht->hash_shift); 386 next = rcu_dereference_protected(n->next, 387 lockdep_is_held(&tbl->lock)); 388 389 rcu_assign_pointer(n->next, 390 rcu_dereference_protected( 391 new_nht->hash_buckets[hash], 392 lockdep_is_held(&tbl->lock))); 393 rcu_assign_pointer(new_nht->hash_buckets[hash], n); 394 } 395 } 396 397 rcu_assign_pointer(tbl->nht, new_nht); 398 call_rcu(&old_nht->rcu, neigh_hash_free_rcu); 399 return new_nht; 400 } 401 402 struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey, 403 struct net_device *dev) 404 { 405 struct neighbour *n; 406 int key_len = tbl->key_len; 407 u32 hash_val; 408 struct neigh_hash_table *nht; 409 410 NEIGH_CACHE_STAT_INC(tbl, lookups); 411 412 rcu_read_lock_bh(); 413 nht = rcu_dereference_bh(tbl->nht); 414 hash_val = tbl->hash(pkey, dev, nht->hash_rnd) >> (32 - nht->hash_shift); 415 416 for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]); 417 n != NULL; 418 n = rcu_dereference_bh(n->next)) { 419 if (dev == n->dev && !memcmp(n->primary_key, pkey, key_len)) { 420 if (!atomic_inc_not_zero(&n->refcnt)) 421 n = NULL; 422 NEIGH_CACHE_STAT_INC(tbl, hits); 423 break; 424 } 425 } 426 427 rcu_read_unlock_bh(); 428 return n; 429 } 430 EXPORT_SYMBOL(neigh_lookup); 431 432 struct neighbour *neigh_lookup_nodev(struct neigh_table *tbl, struct net *net, 433 const void *pkey) 434 { 435 struct neighbour *n; 436 int key_len = tbl->key_len; 437 u32 hash_val; 438 struct neigh_hash_table *nht; 439 440 NEIGH_CACHE_STAT_INC(tbl, lookups); 441 442 rcu_read_lock_bh(); 443 nht = rcu_dereference_bh(tbl->nht); 444 hash_val = tbl->hash(pkey, NULL, nht->hash_rnd) >> (32 - nht->hash_shift); 445 446 for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]); 447 n != NULL; 448 n = rcu_dereference_bh(n->next)) { 449 if (!memcmp(n->primary_key, pkey, key_len) && 450 net_eq(dev_net(n->dev), net)) { 451 if (!atomic_inc_not_zero(&n->refcnt)) 452 n = NULL; 453 NEIGH_CACHE_STAT_INC(tbl, hits); 454 break; 455 } 456 } 457 458 rcu_read_unlock_bh(); 459 return n; 460 } 461 EXPORT_SYMBOL(neigh_lookup_nodev); 462 463 struct neighbour *__neigh_create(struct neigh_table *tbl, const void *pkey, 464 struct net_device *dev, bool want_ref) 465 { 466 u32 hash_val; 467 int key_len = tbl->key_len; 468 int error; 469 struct neighbour *n1, *rc, *n = neigh_alloc(tbl, dev); 470 struct neigh_hash_table *nht; 471 472 if (!n) { 473 rc = ERR_PTR(-ENOBUFS); 474 goto out; 475 } 476 477 memcpy(n->primary_key, pkey, key_len); 478 n->dev = dev; 479 dev_hold(dev); 480 481 /* Protocol specific setup. */ 482 if (tbl->constructor && (error = tbl->constructor(n)) < 0) { 483 rc = ERR_PTR(error); 484 goto out_neigh_release; 485 } 486 487 if (dev->netdev_ops->ndo_neigh_construct) { 488 error = dev->netdev_ops->ndo_neigh_construct(n); 489 if (error < 0) { 490 rc = ERR_PTR(error); 491 goto out_neigh_release; 492 } 493 } 494 495 /* Device specific setup. */ 496 if (n->parms->neigh_setup && 497 (error = n->parms->neigh_setup(n)) < 0) { 498 rc = ERR_PTR(error); 499 goto out_neigh_release; 500 } 501 502 n->confirmed = jiffies - (NEIGH_VAR(n->parms, BASE_REACHABLE_TIME) << 1); 503 504 write_lock_bh(&tbl->lock); 505 nht = rcu_dereference_protected(tbl->nht, 506 lockdep_is_held(&tbl->lock)); 507 508 if (atomic_read(&tbl->entries) > (1 << nht->hash_shift)) 509 nht = neigh_hash_grow(tbl, nht->hash_shift + 1); 510 511 hash_val = tbl->hash(pkey, dev, nht->hash_rnd) >> (32 - nht->hash_shift); 512 513 if (n->parms->dead) { 514 rc = ERR_PTR(-EINVAL); 515 goto out_tbl_unlock; 516 } 517 518 for (n1 = rcu_dereference_protected(nht->hash_buckets[hash_val], 519 lockdep_is_held(&tbl->lock)); 520 n1 != NULL; 521 n1 = rcu_dereference_protected(n1->next, 522 lockdep_is_held(&tbl->lock))) { 523 if (dev == n1->dev && !memcmp(n1->primary_key, pkey, key_len)) { 524 if (want_ref) 525 neigh_hold(n1); 526 rc = n1; 527 goto out_tbl_unlock; 528 } 529 } 530 531 n->dead = 0; 532 if (want_ref) 533 neigh_hold(n); 534 rcu_assign_pointer(n->next, 535 rcu_dereference_protected(nht->hash_buckets[hash_val], 536 lockdep_is_held(&tbl->lock))); 537 rcu_assign_pointer(nht->hash_buckets[hash_val], n); 538 write_unlock_bh(&tbl->lock); 539 neigh_dbg(2, "neigh %p is created\n", n); 540 rc = n; 541 out: 542 return rc; 543 out_tbl_unlock: 544 write_unlock_bh(&tbl->lock); 545 out_neigh_release: 546 neigh_release(n); 547 goto out; 548 } 549 EXPORT_SYMBOL(__neigh_create); 550 551 static u32 pneigh_hash(const void *pkey, int key_len) 552 { 553 u32 hash_val = *(u32 *)(pkey + key_len - 4); 554 hash_val ^= (hash_val >> 16); 555 hash_val ^= hash_val >> 8; 556 hash_val ^= hash_val >> 4; 557 hash_val &= PNEIGH_HASHMASK; 558 return hash_val; 559 } 560 561 static struct pneigh_entry *__pneigh_lookup_1(struct pneigh_entry *n, 562 struct net *net, 563 const void *pkey, 564 int key_len, 565 struct net_device *dev) 566 { 567 while (n) { 568 if (!memcmp(n->key, pkey, key_len) && 569 net_eq(pneigh_net(n), net) && 570 (n->dev == dev || !n->dev)) 571 return n; 572 n = n->next; 573 } 574 return NULL; 575 } 576 577 struct pneigh_entry *__pneigh_lookup(struct neigh_table *tbl, 578 struct net *net, const void *pkey, struct net_device *dev) 579 { 580 int key_len = tbl->key_len; 581 u32 hash_val = pneigh_hash(pkey, key_len); 582 583 return __pneigh_lookup_1(tbl->phash_buckets[hash_val], 584 net, pkey, key_len, dev); 585 } 586 EXPORT_SYMBOL_GPL(__pneigh_lookup); 587 588 struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl, 589 struct net *net, const void *pkey, 590 struct net_device *dev, int creat) 591 { 592 struct pneigh_entry *n; 593 int key_len = tbl->key_len; 594 u32 hash_val = pneigh_hash(pkey, key_len); 595 596 read_lock_bh(&tbl->lock); 597 n = __pneigh_lookup_1(tbl->phash_buckets[hash_val], 598 net, pkey, key_len, dev); 599 read_unlock_bh(&tbl->lock); 600 601 if (n || !creat) 602 goto out; 603 604 ASSERT_RTNL(); 605 606 n = kmalloc(sizeof(*n) + key_len, GFP_KERNEL); 607 if (!n) 608 goto out; 609 610 write_pnet(&n->net, hold_net(net)); 611 memcpy(n->key, pkey, key_len); 612 n->dev = dev; 613 if (dev) 614 dev_hold(dev); 615 616 if (tbl->pconstructor && tbl->pconstructor(n)) { 617 if (dev) 618 dev_put(dev); 619 release_net(net); 620 kfree(n); 621 n = NULL; 622 goto out; 623 } 624 625 write_lock_bh(&tbl->lock); 626 n->next = tbl->phash_buckets[hash_val]; 627 tbl->phash_buckets[hash_val] = n; 628 write_unlock_bh(&tbl->lock); 629 out: 630 return n; 631 } 632 EXPORT_SYMBOL(pneigh_lookup); 633 634 635 int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey, 636 struct net_device *dev) 637 { 638 struct pneigh_entry *n, **np; 639 int key_len = tbl->key_len; 640 u32 hash_val = pneigh_hash(pkey, key_len); 641 642 write_lock_bh(&tbl->lock); 643 for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL; 644 np = &n->next) { 645 if (!memcmp(n->key, pkey, key_len) && n->dev == dev && 646 net_eq(pneigh_net(n), net)) { 647 *np = n->next; 648 write_unlock_bh(&tbl->lock); 649 if (tbl->pdestructor) 650 tbl->pdestructor(n); 651 if (n->dev) 652 dev_put(n->dev); 653 release_net(pneigh_net(n)); 654 kfree(n); 655 return 0; 656 } 657 } 658 write_unlock_bh(&tbl->lock); 659 return -ENOENT; 660 } 661 662 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev) 663 { 664 struct pneigh_entry *n, **np; 665 u32 h; 666 667 for (h = 0; h <= PNEIGH_HASHMASK; h++) { 668 np = &tbl->phash_buckets[h]; 669 while ((n = *np) != NULL) { 670 if (!dev || n->dev == dev) { 671 *np = n->next; 672 if (tbl->pdestructor) 673 tbl->pdestructor(n); 674 if (n->dev) 675 dev_put(n->dev); 676 release_net(pneigh_net(n)); 677 kfree(n); 678 continue; 679 } 680 np = &n->next; 681 } 682 } 683 return -ENOENT; 684 } 685 686 static void neigh_parms_destroy(struct neigh_parms *parms); 687 688 static inline void neigh_parms_put(struct neigh_parms *parms) 689 { 690 if (atomic_dec_and_test(&parms->refcnt)) 691 neigh_parms_destroy(parms); 692 } 693 694 /* 695 * neighbour must already be out of the table; 696 * 697 */ 698 void neigh_destroy(struct neighbour *neigh) 699 { 700 struct net_device *dev = neigh->dev; 701 702 NEIGH_CACHE_STAT_INC(neigh->tbl, destroys); 703 704 if (!neigh->dead) { 705 pr_warn("Destroying alive neighbour %p\n", neigh); 706 dump_stack(); 707 return; 708 } 709 710 if (neigh_del_timer(neigh)) 711 pr_warn("Impossible event\n"); 712 713 write_lock_bh(&neigh->lock); 714 __skb_queue_purge(&neigh->arp_queue); 715 write_unlock_bh(&neigh->lock); 716 neigh->arp_queue_len_bytes = 0; 717 718 if (dev->netdev_ops->ndo_neigh_destroy) 719 dev->netdev_ops->ndo_neigh_destroy(neigh); 720 721 dev_put(dev); 722 neigh_parms_put(neigh->parms); 723 724 neigh_dbg(2, "neigh %p is destroyed\n", neigh); 725 726 atomic_dec(&neigh->tbl->entries); 727 kfree_rcu(neigh, rcu); 728 } 729 EXPORT_SYMBOL(neigh_destroy); 730 731 /* Neighbour state is suspicious; 732 disable fast path. 733 734 Called with write_locked neigh. 735 */ 736 static void neigh_suspect(struct neighbour *neigh) 737 { 738 neigh_dbg(2, "neigh %p is suspected\n", neigh); 739 740 neigh->output = neigh->ops->output; 741 } 742 743 /* Neighbour state is OK; 744 enable fast path. 745 746 Called with write_locked neigh. 747 */ 748 static void neigh_connect(struct neighbour *neigh) 749 { 750 neigh_dbg(2, "neigh %p is connected\n", neigh); 751 752 neigh->output = neigh->ops->connected_output; 753 } 754 755 static void neigh_periodic_work(struct work_struct *work) 756 { 757 struct neigh_table *tbl = container_of(work, struct neigh_table, gc_work.work); 758 struct neighbour *n; 759 struct neighbour __rcu **np; 760 unsigned int i; 761 struct neigh_hash_table *nht; 762 763 NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs); 764 765 write_lock_bh(&tbl->lock); 766 nht = rcu_dereference_protected(tbl->nht, 767 lockdep_is_held(&tbl->lock)); 768 769 /* 770 * periodically recompute ReachableTime from random function 771 */ 772 773 if (time_after(jiffies, tbl->last_rand + 300 * HZ)) { 774 struct neigh_parms *p; 775 tbl->last_rand = jiffies; 776 for (p = &tbl->parms; p; p = p->next) 777 p->reachable_time = 778 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME)); 779 } 780 781 if (atomic_read(&tbl->entries) < tbl->gc_thresh1) 782 goto out; 783 784 for (i = 0 ; i < (1 << nht->hash_shift); i++) { 785 np = &nht->hash_buckets[i]; 786 787 while ((n = rcu_dereference_protected(*np, 788 lockdep_is_held(&tbl->lock))) != NULL) { 789 unsigned int state; 790 791 write_lock(&n->lock); 792 793 state = n->nud_state; 794 if (state & (NUD_PERMANENT | NUD_IN_TIMER)) { 795 write_unlock(&n->lock); 796 goto next_elt; 797 } 798 799 if (time_before(n->used, n->confirmed)) 800 n->used = n->confirmed; 801 802 if (atomic_read(&n->refcnt) == 1 && 803 (state == NUD_FAILED || 804 time_after(jiffies, n->used + NEIGH_VAR(n->parms, GC_STALETIME)))) { 805 *np = n->next; 806 n->dead = 1; 807 write_unlock(&n->lock); 808 neigh_cleanup_and_release(n); 809 continue; 810 } 811 write_unlock(&n->lock); 812 813 next_elt: 814 np = &n->next; 815 } 816 /* 817 * It's fine to release lock here, even if hash table 818 * grows while we are preempted. 819 */ 820 write_unlock_bh(&tbl->lock); 821 cond_resched(); 822 write_lock_bh(&tbl->lock); 823 nht = rcu_dereference_protected(tbl->nht, 824 lockdep_is_held(&tbl->lock)); 825 } 826 out: 827 /* Cycle through all hash buckets every BASE_REACHABLE_TIME/2 ticks. 828 * ARP entry timeouts range from 1/2 BASE_REACHABLE_TIME to 3/2 829 * BASE_REACHABLE_TIME. 830 */ 831 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work, 832 NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME) >> 1); 833 write_unlock_bh(&tbl->lock); 834 } 835 836 static __inline__ int neigh_max_probes(struct neighbour *n) 837 { 838 struct neigh_parms *p = n->parms; 839 int max_probes = NEIGH_VAR(p, UCAST_PROBES) + NEIGH_VAR(p, APP_PROBES); 840 if (!(n->nud_state & NUD_PROBE)) 841 max_probes += NEIGH_VAR(p, MCAST_PROBES); 842 return max_probes; 843 } 844 845 static void neigh_invalidate(struct neighbour *neigh) 846 __releases(neigh->lock) 847 __acquires(neigh->lock) 848 { 849 struct sk_buff *skb; 850 851 NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed); 852 neigh_dbg(2, "neigh %p is failed\n", neigh); 853 neigh->updated = jiffies; 854 855 /* It is very thin place. report_unreachable is very complicated 856 routine. Particularly, it can hit the same neighbour entry! 857 858 So that, we try to be accurate and avoid dead loop. --ANK 859 */ 860 while (neigh->nud_state == NUD_FAILED && 861 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) { 862 write_unlock(&neigh->lock); 863 neigh->ops->error_report(neigh, skb); 864 write_lock(&neigh->lock); 865 } 866 __skb_queue_purge(&neigh->arp_queue); 867 neigh->arp_queue_len_bytes = 0; 868 } 869 870 static void neigh_probe(struct neighbour *neigh) 871 __releases(neigh->lock) 872 { 873 struct sk_buff *skb = skb_peek_tail(&neigh->arp_queue); 874 /* keep skb alive even if arp_queue overflows */ 875 if (skb) 876 skb = skb_copy(skb, GFP_ATOMIC); 877 write_unlock(&neigh->lock); 878 neigh->ops->solicit(neigh, skb); 879 atomic_inc(&neigh->probes); 880 kfree_skb(skb); 881 } 882 883 /* Called when a timer expires for a neighbour entry. */ 884 885 static void neigh_timer_handler(unsigned long arg) 886 { 887 unsigned long now, next; 888 struct neighbour *neigh = (struct neighbour *)arg; 889 unsigned int state; 890 int notify = 0; 891 892 write_lock(&neigh->lock); 893 894 state = neigh->nud_state; 895 now = jiffies; 896 next = now + HZ; 897 898 if (!(state & NUD_IN_TIMER)) 899 goto out; 900 901 if (state & NUD_REACHABLE) { 902 if (time_before_eq(now, 903 neigh->confirmed + neigh->parms->reachable_time)) { 904 neigh_dbg(2, "neigh %p is still alive\n", neigh); 905 next = neigh->confirmed + neigh->parms->reachable_time; 906 } else if (time_before_eq(now, 907 neigh->used + 908 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) { 909 neigh_dbg(2, "neigh %p is delayed\n", neigh); 910 neigh->nud_state = NUD_DELAY; 911 neigh->updated = jiffies; 912 neigh_suspect(neigh); 913 next = now + NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME); 914 } else { 915 neigh_dbg(2, "neigh %p is suspected\n", neigh); 916 neigh->nud_state = NUD_STALE; 917 neigh->updated = jiffies; 918 neigh_suspect(neigh); 919 notify = 1; 920 } 921 } else if (state & NUD_DELAY) { 922 if (time_before_eq(now, 923 neigh->confirmed + 924 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) { 925 neigh_dbg(2, "neigh %p is now reachable\n", neigh); 926 neigh->nud_state = NUD_REACHABLE; 927 neigh->updated = jiffies; 928 neigh_connect(neigh); 929 notify = 1; 930 next = neigh->confirmed + neigh->parms->reachable_time; 931 } else { 932 neigh_dbg(2, "neigh %p is probed\n", neigh); 933 neigh->nud_state = NUD_PROBE; 934 neigh->updated = jiffies; 935 atomic_set(&neigh->probes, 0); 936 next = now + NEIGH_VAR(neigh->parms, RETRANS_TIME); 937 } 938 } else { 939 /* NUD_PROBE|NUD_INCOMPLETE */ 940 next = now + NEIGH_VAR(neigh->parms, RETRANS_TIME); 941 } 942 943 if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) && 944 atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) { 945 neigh->nud_state = NUD_FAILED; 946 notify = 1; 947 neigh_invalidate(neigh); 948 goto out; 949 } 950 951 if (neigh->nud_state & NUD_IN_TIMER) { 952 if (time_before(next, jiffies + HZ/2)) 953 next = jiffies + HZ/2; 954 if (!mod_timer(&neigh->timer, next)) 955 neigh_hold(neigh); 956 } 957 if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) { 958 neigh_probe(neigh); 959 } else { 960 out: 961 write_unlock(&neigh->lock); 962 } 963 964 if (notify) 965 neigh_update_notify(neigh); 966 967 neigh_release(neigh); 968 } 969 970 int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb) 971 { 972 int rc; 973 bool immediate_probe = false; 974 975 write_lock_bh(&neigh->lock); 976 977 rc = 0; 978 if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE)) 979 goto out_unlock_bh; 980 981 if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) { 982 if (NEIGH_VAR(neigh->parms, MCAST_PROBES) + 983 NEIGH_VAR(neigh->parms, APP_PROBES)) { 984 unsigned long next, now = jiffies; 985 986 atomic_set(&neigh->probes, 987 NEIGH_VAR(neigh->parms, UCAST_PROBES)); 988 neigh->nud_state = NUD_INCOMPLETE; 989 neigh->updated = now; 990 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME), 991 HZ/2); 992 neigh_add_timer(neigh, next); 993 immediate_probe = true; 994 } else { 995 neigh->nud_state = NUD_FAILED; 996 neigh->updated = jiffies; 997 write_unlock_bh(&neigh->lock); 998 999 kfree_skb(skb); 1000 return 1; 1001 } 1002 } else if (neigh->nud_state & NUD_STALE) { 1003 neigh_dbg(2, "neigh %p is delayed\n", neigh); 1004 neigh->nud_state = NUD_DELAY; 1005 neigh->updated = jiffies; 1006 neigh_add_timer(neigh, jiffies + 1007 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME)); 1008 } 1009 1010 if (neigh->nud_state == NUD_INCOMPLETE) { 1011 if (skb) { 1012 while (neigh->arp_queue_len_bytes + skb->truesize > 1013 NEIGH_VAR(neigh->parms, QUEUE_LEN_BYTES)) { 1014 struct sk_buff *buff; 1015 1016 buff = __skb_dequeue(&neigh->arp_queue); 1017 if (!buff) 1018 break; 1019 neigh->arp_queue_len_bytes -= buff->truesize; 1020 kfree_skb(buff); 1021 NEIGH_CACHE_STAT_INC(neigh->tbl, unres_discards); 1022 } 1023 skb_dst_force(skb); 1024 __skb_queue_tail(&neigh->arp_queue, skb); 1025 neigh->arp_queue_len_bytes += skb->truesize; 1026 } 1027 rc = 1; 1028 } 1029 out_unlock_bh: 1030 if (immediate_probe) 1031 neigh_probe(neigh); 1032 else 1033 write_unlock(&neigh->lock); 1034 local_bh_enable(); 1035 return rc; 1036 } 1037 EXPORT_SYMBOL(__neigh_event_send); 1038 1039 static void neigh_update_hhs(struct neighbour *neigh) 1040 { 1041 struct hh_cache *hh; 1042 void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *) 1043 = NULL; 1044 1045 if (neigh->dev->header_ops) 1046 update = neigh->dev->header_ops->cache_update; 1047 1048 if (update) { 1049 hh = &neigh->hh; 1050 if (hh->hh_len) { 1051 write_seqlock_bh(&hh->hh_lock); 1052 update(hh, neigh->dev, neigh->ha); 1053 write_sequnlock_bh(&hh->hh_lock); 1054 } 1055 } 1056 } 1057 1058 1059 1060 /* Generic update routine. 1061 -- lladdr is new lladdr or NULL, if it is not supplied. 1062 -- new is new state. 1063 -- flags 1064 NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr, 1065 if it is different. 1066 NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected" 1067 lladdr instead of overriding it 1068 if it is different. 1069 It also allows to retain current state 1070 if lladdr is unchanged. 1071 NEIGH_UPDATE_F_ADMIN means that the change is administrative. 1072 1073 NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing 1074 NTF_ROUTER flag. 1075 NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as 1076 a router. 1077 1078 Caller MUST hold reference count on the entry. 1079 */ 1080 1081 int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new, 1082 u32 flags) 1083 { 1084 u8 old; 1085 int err; 1086 int notify = 0; 1087 struct net_device *dev; 1088 int update_isrouter = 0; 1089 1090 write_lock_bh(&neigh->lock); 1091 1092 dev = neigh->dev; 1093 old = neigh->nud_state; 1094 err = -EPERM; 1095 1096 if (!(flags & NEIGH_UPDATE_F_ADMIN) && 1097 (old & (NUD_NOARP | NUD_PERMANENT))) 1098 goto out; 1099 1100 if (!(new & NUD_VALID)) { 1101 neigh_del_timer(neigh); 1102 if (old & NUD_CONNECTED) 1103 neigh_suspect(neigh); 1104 neigh->nud_state = new; 1105 err = 0; 1106 notify = old & NUD_VALID; 1107 if ((old & (NUD_INCOMPLETE | NUD_PROBE)) && 1108 (new & NUD_FAILED)) { 1109 neigh_invalidate(neigh); 1110 notify = 1; 1111 } 1112 goto out; 1113 } 1114 1115 /* Compare new lladdr with cached one */ 1116 if (!dev->addr_len) { 1117 /* First case: device needs no address. */ 1118 lladdr = neigh->ha; 1119 } else if (lladdr) { 1120 /* The second case: if something is already cached 1121 and a new address is proposed: 1122 - compare new & old 1123 - if they are different, check override flag 1124 */ 1125 if ((old & NUD_VALID) && 1126 !memcmp(lladdr, neigh->ha, dev->addr_len)) 1127 lladdr = neigh->ha; 1128 } else { 1129 /* No address is supplied; if we know something, 1130 use it, otherwise discard the request. 1131 */ 1132 err = -EINVAL; 1133 if (!(old & NUD_VALID)) 1134 goto out; 1135 lladdr = neigh->ha; 1136 } 1137 1138 if (new & NUD_CONNECTED) 1139 neigh->confirmed = jiffies; 1140 neigh->updated = jiffies; 1141 1142 /* If entry was valid and address is not changed, 1143 do not change entry state, if new one is STALE. 1144 */ 1145 err = 0; 1146 update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER; 1147 if (old & NUD_VALID) { 1148 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) { 1149 update_isrouter = 0; 1150 if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) && 1151 (old & NUD_CONNECTED)) { 1152 lladdr = neigh->ha; 1153 new = NUD_STALE; 1154 } else 1155 goto out; 1156 } else { 1157 if (lladdr == neigh->ha && new == NUD_STALE && 1158 ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) || 1159 (old & NUD_CONNECTED)) 1160 ) 1161 new = old; 1162 } 1163 } 1164 1165 if (new != old) { 1166 neigh_del_timer(neigh); 1167 if (new & NUD_IN_TIMER) 1168 neigh_add_timer(neigh, (jiffies + 1169 ((new & NUD_REACHABLE) ? 1170 neigh->parms->reachable_time : 1171 0))); 1172 neigh->nud_state = new; 1173 notify = 1; 1174 } 1175 1176 if (lladdr != neigh->ha) { 1177 write_seqlock(&neigh->ha_lock); 1178 memcpy(&neigh->ha, lladdr, dev->addr_len); 1179 write_sequnlock(&neigh->ha_lock); 1180 neigh_update_hhs(neigh); 1181 if (!(new & NUD_CONNECTED)) 1182 neigh->confirmed = jiffies - 1183 (NEIGH_VAR(neigh->parms, BASE_REACHABLE_TIME) << 1); 1184 notify = 1; 1185 } 1186 if (new == old) 1187 goto out; 1188 if (new & NUD_CONNECTED) 1189 neigh_connect(neigh); 1190 else 1191 neigh_suspect(neigh); 1192 if (!(old & NUD_VALID)) { 1193 struct sk_buff *skb; 1194 1195 /* Again: avoid dead loop if something went wrong */ 1196 1197 while (neigh->nud_state & NUD_VALID && 1198 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) { 1199 struct dst_entry *dst = skb_dst(skb); 1200 struct neighbour *n2, *n1 = neigh; 1201 write_unlock_bh(&neigh->lock); 1202 1203 rcu_read_lock(); 1204 1205 /* Why not just use 'neigh' as-is? The problem is that 1206 * things such as shaper, eql, and sch_teql can end up 1207 * using alternative, different, neigh objects to output 1208 * the packet in the output path. So what we need to do 1209 * here is re-lookup the top-level neigh in the path so 1210 * we can reinject the packet there. 1211 */ 1212 n2 = NULL; 1213 if (dst) { 1214 n2 = dst_neigh_lookup_skb(dst, skb); 1215 if (n2) 1216 n1 = n2; 1217 } 1218 n1->output(n1, skb); 1219 if (n2) 1220 neigh_release(n2); 1221 rcu_read_unlock(); 1222 1223 write_lock_bh(&neigh->lock); 1224 } 1225 __skb_queue_purge(&neigh->arp_queue); 1226 neigh->arp_queue_len_bytes = 0; 1227 } 1228 out: 1229 if (update_isrouter) { 1230 neigh->flags = (flags & NEIGH_UPDATE_F_ISROUTER) ? 1231 (neigh->flags | NTF_ROUTER) : 1232 (neigh->flags & ~NTF_ROUTER); 1233 } 1234 write_unlock_bh(&neigh->lock); 1235 1236 if (notify) 1237 neigh_update_notify(neigh); 1238 1239 return err; 1240 } 1241 EXPORT_SYMBOL(neigh_update); 1242 1243 /* Update the neigh to listen temporarily for probe responses, even if it is 1244 * in a NUD_FAILED state. The caller has to hold neigh->lock for writing. 1245 */ 1246 void __neigh_set_probe_once(struct neighbour *neigh) 1247 { 1248 neigh->updated = jiffies; 1249 if (!(neigh->nud_state & NUD_FAILED)) 1250 return; 1251 neigh->nud_state = NUD_INCOMPLETE; 1252 atomic_set(&neigh->probes, neigh_max_probes(neigh)); 1253 neigh_add_timer(neigh, 1254 jiffies + NEIGH_VAR(neigh->parms, RETRANS_TIME)); 1255 } 1256 EXPORT_SYMBOL(__neigh_set_probe_once); 1257 1258 struct neighbour *neigh_event_ns(struct neigh_table *tbl, 1259 u8 *lladdr, void *saddr, 1260 struct net_device *dev) 1261 { 1262 struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev, 1263 lladdr || !dev->addr_len); 1264 if (neigh) 1265 neigh_update(neigh, lladdr, NUD_STALE, 1266 NEIGH_UPDATE_F_OVERRIDE); 1267 return neigh; 1268 } 1269 EXPORT_SYMBOL(neigh_event_ns); 1270 1271 /* called with read_lock_bh(&n->lock); */ 1272 static void neigh_hh_init(struct neighbour *n, struct dst_entry *dst) 1273 { 1274 struct net_device *dev = dst->dev; 1275 __be16 prot = dst->ops->protocol; 1276 struct hh_cache *hh = &n->hh; 1277 1278 write_lock_bh(&n->lock); 1279 1280 /* Only one thread can come in here and initialize the 1281 * hh_cache entry. 1282 */ 1283 if (!hh->hh_len) 1284 dev->header_ops->cache(n, hh, prot); 1285 1286 write_unlock_bh(&n->lock); 1287 } 1288 1289 /* This function can be used in contexts, where only old dev_queue_xmit 1290 * worked, f.e. if you want to override normal output path (eql, shaper), 1291 * but resolution is not made yet. 1292 */ 1293 1294 int neigh_compat_output(struct neighbour *neigh, struct sk_buff *skb) 1295 { 1296 struct net_device *dev = skb->dev; 1297 1298 __skb_pull(skb, skb_network_offset(skb)); 1299 1300 if (dev_hard_header(skb, dev, ntohs(skb->protocol), NULL, NULL, 1301 skb->len) < 0 && 1302 dev_rebuild_header(skb)) 1303 return 0; 1304 1305 return dev_queue_xmit(skb); 1306 } 1307 EXPORT_SYMBOL(neigh_compat_output); 1308 1309 /* Slow and careful. */ 1310 1311 int neigh_resolve_output(struct neighbour *neigh, struct sk_buff *skb) 1312 { 1313 struct dst_entry *dst = skb_dst(skb); 1314 int rc = 0; 1315 1316 if (!dst) 1317 goto discard; 1318 1319 if (!neigh_event_send(neigh, skb)) { 1320 int err; 1321 struct net_device *dev = neigh->dev; 1322 unsigned int seq; 1323 1324 if (dev->header_ops->cache && !neigh->hh.hh_len) 1325 neigh_hh_init(neigh, dst); 1326 1327 do { 1328 __skb_pull(skb, skb_network_offset(skb)); 1329 seq = read_seqbegin(&neigh->ha_lock); 1330 err = dev_hard_header(skb, dev, ntohs(skb->protocol), 1331 neigh->ha, NULL, skb->len); 1332 } while (read_seqretry(&neigh->ha_lock, seq)); 1333 1334 if (err >= 0) 1335 rc = dev_queue_xmit(skb); 1336 else 1337 goto out_kfree_skb; 1338 } 1339 out: 1340 return rc; 1341 discard: 1342 neigh_dbg(1, "%s: dst=%p neigh=%p\n", __func__, dst, neigh); 1343 out_kfree_skb: 1344 rc = -EINVAL; 1345 kfree_skb(skb); 1346 goto out; 1347 } 1348 EXPORT_SYMBOL(neigh_resolve_output); 1349 1350 /* As fast as possible without hh cache */ 1351 1352 int neigh_connected_output(struct neighbour *neigh, struct sk_buff *skb) 1353 { 1354 struct net_device *dev = neigh->dev; 1355 unsigned int seq; 1356 int err; 1357 1358 do { 1359 __skb_pull(skb, skb_network_offset(skb)); 1360 seq = read_seqbegin(&neigh->ha_lock); 1361 err = dev_hard_header(skb, dev, ntohs(skb->protocol), 1362 neigh->ha, NULL, skb->len); 1363 } while (read_seqretry(&neigh->ha_lock, seq)); 1364 1365 if (err >= 0) 1366 err = dev_queue_xmit(skb); 1367 else { 1368 err = -EINVAL; 1369 kfree_skb(skb); 1370 } 1371 return err; 1372 } 1373 EXPORT_SYMBOL(neigh_connected_output); 1374 1375 int neigh_direct_output(struct neighbour *neigh, struct sk_buff *skb) 1376 { 1377 return dev_queue_xmit(skb); 1378 } 1379 EXPORT_SYMBOL(neigh_direct_output); 1380 1381 static void neigh_proxy_process(unsigned long arg) 1382 { 1383 struct neigh_table *tbl = (struct neigh_table *)arg; 1384 long sched_next = 0; 1385 unsigned long now = jiffies; 1386 struct sk_buff *skb, *n; 1387 1388 spin_lock(&tbl->proxy_queue.lock); 1389 1390 skb_queue_walk_safe(&tbl->proxy_queue, skb, n) { 1391 long tdif = NEIGH_CB(skb)->sched_next - now; 1392 1393 if (tdif <= 0) { 1394 struct net_device *dev = skb->dev; 1395 1396 __skb_unlink(skb, &tbl->proxy_queue); 1397 if (tbl->proxy_redo && netif_running(dev)) { 1398 rcu_read_lock(); 1399 tbl->proxy_redo(skb); 1400 rcu_read_unlock(); 1401 } else { 1402 kfree_skb(skb); 1403 } 1404 1405 dev_put(dev); 1406 } else if (!sched_next || tdif < sched_next) 1407 sched_next = tdif; 1408 } 1409 del_timer(&tbl->proxy_timer); 1410 if (sched_next) 1411 mod_timer(&tbl->proxy_timer, jiffies + sched_next); 1412 spin_unlock(&tbl->proxy_queue.lock); 1413 } 1414 1415 void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p, 1416 struct sk_buff *skb) 1417 { 1418 unsigned long now = jiffies; 1419 1420 unsigned long sched_next = now + (prandom_u32() % 1421 NEIGH_VAR(p, PROXY_DELAY)); 1422 1423 if (tbl->proxy_queue.qlen > NEIGH_VAR(p, PROXY_QLEN)) { 1424 kfree_skb(skb); 1425 return; 1426 } 1427 1428 NEIGH_CB(skb)->sched_next = sched_next; 1429 NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED; 1430 1431 spin_lock(&tbl->proxy_queue.lock); 1432 if (del_timer(&tbl->proxy_timer)) { 1433 if (time_before(tbl->proxy_timer.expires, sched_next)) 1434 sched_next = tbl->proxy_timer.expires; 1435 } 1436 skb_dst_drop(skb); 1437 dev_hold(skb->dev); 1438 __skb_queue_tail(&tbl->proxy_queue, skb); 1439 mod_timer(&tbl->proxy_timer, sched_next); 1440 spin_unlock(&tbl->proxy_queue.lock); 1441 } 1442 EXPORT_SYMBOL(pneigh_enqueue); 1443 1444 static inline struct neigh_parms *lookup_neigh_parms(struct neigh_table *tbl, 1445 struct net *net, int ifindex) 1446 { 1447 struct neigh_parms *p; 1448 1449 for (p = &tbl->parms; p; p = p->next) { 1450 if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) || 1451 (!p->dev && !ifindex && net_eq(net, &init_net))) 1452 return p; 1453 } 1454 1455 return NULL; 1456 } 1457 1458 struct neigh_parms *neigh_parms_alloc(struct net_device *dev, 1459 struct neigh_table *tbl) 1460 { 1461 struct neigh_parms *p; 1462 struct net *net = dev_net(dev); 1463 const struct net_device_ops *ops = dev->netdev_ops; 1464 1465 p = kmemdup(&tbl->parms, sizeof(*p), GFP_KERNEL); 1466 if (p) { 1467 p->tbl = tbl; 1468 atomic_set(&p->refcnt, 1); 1469 p->reachable_time = 1470 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME)); 1471 dev_hold(dev); 1472 p->dev = dev; 1473 write_pnet(&p->net, hold_net(net)); 1474 p->sysctl_table = NULL; 1475 1476 if (ops->ndo_neigh_setup && ops->ndo_neigh_setup(dev, p)) { 1477 release_net(net); 1478 dev_put(dev); 1479 kfree(p); 1480 return NULL; 1481 } 1482 1483 write_lock_bh(&tbl->lock); 1484 p->next = tbl->parms.next; 1485 tbl->parms.next = p; 1486 write_unlock_bh(&tbl->lock); 1487 1488 neigh_parms_data_state_cleanall(p); 1489 } 1490 return p; 1491 } 1492 EXPORT_SYMBOL(neigh_parms_alloc); 1493 1494 static void neigh_rcu_free_parms(struct rcu_head *head) 1495 { 1496 struct neigh_parms *parms = 1497 container_of(head, struct neigh_parms, rcu_head); 1498 1499 neigh_parms_put(parms); 1500 } 1501 1502 void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms) 1503 { 1504 struct neigh_parms **p; 1505 1506 if (!parms || parms == &tbl->parms) 1507 return; 1508 write_lock_bh(&tbl->lock); 1509 for (p = &tbl->parms.next; *p; p = &(*p)->next) { 1510 if (*p == parms) { 1511 *p = parms->next; 1512 parms->dead = 1; 1513 write_unlock_bh(&tbl->lock); 1514 if (parms->dev) 1515 dev_put(parms->dev); 1516 call_rcu(&parms->rcu_head, neigh_rcu_free_parms); 1517 return; 1518 } 1519 } 1520 write_unlock_bh(&tbl->lock); 1521 neigh_dbg(1, "%s: not found\n", __func__); 1522 } 1523 EXPORT_SYMBOL(neigh_parms_release); 1524 1525 static void neigh_parms_destroy(struct neigh_parms *parms) 1526 { 1527 release_net(neigh_parms_net(parms)); 1528 kfree(parms); 1529 } 1530 1531 static struct lock_class_key neigh_table_proxy_queue_class; 1532 1533 static void neigh_table_init_no_netlink(struct neigh_table *tbl) 1534 { 1535 unsigned long now = jiffies; 1536 unsigned long phsize; 1537 1538 write_pnet(&tbl->parms.net, &init_net); 1539 atomic_set(&tbl->parms.refcnt, 1); 1540 tbl->parms.reachable_time = 1541 neigh_rand_reach_time(NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME)); 1542 1543 tbl->stats = alloc_percpu(struct neigh_statistics); 1544 if (!tbl->stats) 1545 panic("cannot create neighbour cache statistics"); 1546 1547 #ifdef CONFIG_PROC_FS 1548 if (!proc_create_data(tbl->id, 0, init_net.proc_net_stat, 1549 &neigh_stat_seq_fops, tbl)) 1550 panic("cannot create neighbour proc dir entry"); 1551 #endif 1552 1553 RCU_INIT_POINTER(tbl->nht, neigh_hash_alloc(3)); 1554 1555 phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *); 1556 tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL); 1557 1558 if (!tbl->nht || !tbl->phash_buckets) 1559 panic("cannot allocate neighbour cache hashes"); 1560 1561 if (!tbl->entry_size) 1562 tbl->entry_size = ALIGN(offsetof(struct neighbour, primary_key) + 1563 tbl->key_len, NEIGH_PRIV_ALIGN); 1564 else 1565 WARN_ON(tbl->entry_size % NEIGH_PRIV_ALIGN); 1566 1567 rwlock_init(&tbl->lock); 1568 INIT_DEFERRABLE_WORK(&tbl->gc_work, neigh_periodic_work); 1569 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work, 1570 tbl->parms.reachable_time); 1571 setup_timer(&tbl->proxy_timer, neigh_proxy_process, (unsigned long)tbl); 1572 skb_queue_head_init_class(&tbl->proxy_queue, 1573 &neigh_table_proxy_queue_class); 1574 1575 tbl->last_flush = now; 1576 tbl->last_rand = now + tbl->parms.reachable_time * 20; 1577 } 1578 1579 void neigh_table_init(struct neigh_table *tbl) 1580 { 1581 struct neigh_table *tmp; 1582 1583 neigh_table_init_no_netlink(tbl); 1584 write_lock(&neigh_tbl_lock); 1585 for (tmp = neigh_tables; tmp; tmp = tmp->next) { 1586 if (tmp->family == tbl->family) 1587 break; 1588 } 1589 tbl->next = neigh_tables; 1590 neigh_tables = tbl; 1591 write_unlock(&neigh_tbl_lock); 1592 1593 if (unlikely(tmp)) { 1594 pr_err("Registering multiple tables for family %d\n", 1595 tbl->family); 1596 dump_stack(); 1597 } 1598 } 1599 EXPORT_SYMBOL(neigh_table_init); 1600 1601 int neigh_table_clear(struct neigh_table *tbl) 1602 { 1603 struct neigh_table **tp; 1604 1605 /* It is not clean... Fix it to unload IPv6 module safely */ 1606 cancel_delayed_work_sync(&tbl->gc_work); 1607 del_timer_sync(&tbl->proxy_timer); 1608 pneigh_queue_purge(&tbl->proxy_queue); 1609 neigh_ifdown(tbl, NULL); 1610 if (atomic_read(&tbl->entries)) 1611 pr_crit("neighbour leakage\n"); 1612 write_lock(&neigh_tbl_lock); 1613 for (tp = &neigh_tables; *tp; tp = &(*tp)->next) { 1614 if (*tp == tbl) { 1615 *tp = tbl->next; 1616 break; 1617 } 1618 } 1619 write_unlock(&neigh_tbl_lock); 1620 1621 call_rcu(&rcu_dereference_protected(tbl->nht, 1)->rcu, 1622 neigh_hash_free_rcu); 1623 tbl->nht = NULL; 1624 1625 kfree(tbl->phash_buckets); 1626 tbl->phash_buckets = NULL; 1627 1628 remove_proc_entry(tbl->id, init_net.proc_net_stat); 1629 1630 free_percpu(tbl->stats); 1631 tbl->stats = NULL; 1632 1633 return 0; 1634 } 1635 EXPORT_SYMBOL(neigh_table_clear); 1636 1637 static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh) 1638 { 1639 struct net *net = sock_net(skb->sk); 1640 struct ndmsg *ndm; 1641 struct nlattr *dst_attr; 1642 struct neigh_table *tbl; 1643 struct net_device *dev = NULL; 1644 int err = -EINVAL; 1645 1646 ASSERT_RTNL(); 1647 if (nlmsg_len(nlh) < sizeof(*ndm)) 1648 goto out; 1649 1650 dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST); 1651 if (dst_attr == NULL) 1652 goto out; 1653 1654 ndm = nlmsg_data(nlh); 1655 if (ndm->ndm_ifindex) { 1656 dev = __dev_get_by_index(net, ndm->ndm_ifindex); 1657 if (dev == NULL) { 1658 err = -ENODEV; 1659 goto out; 1660 } 1661 } 1662 1663 read_lock(&neigh_tbl_lock); 1664 for (tbl = neigh_tables; tbl; tbl = tbl->next) { 1665 struct neighbour *neigh; 1666 1667 if (tbl->family != ndm->ndm_family) 1668 continue; 1669 read_unlock(&neigh_tbl_lock); 1670 1671 if (nla_len(dst_attr) < tbl->key_len) 1672 goto out; 1673 1674 if (ndm->ndm_flags & NTF_PROXY) { 1675 err = pneigh_delete(tbl, net, nla_data(dst_attr), dev); 1676 goto out; 1677 } 1678 1679 if (dev == NULL) 1680 goto out; 1681 1682 neigh = neigh_lookup(tbl, nla_data(dst_attr), dev); 1683 if (neigh == NULL) { 1684 err = -ENOENT; 1685 goto out; 1686 } 1687 1688 err = neigh_update(neigh, NULL, NUD_FAILED, 1689 NEIGH_UPDATE_F_OVERRIDE | 1690 NEIGH_UPDATE_F_ADMIN); 1691 neigh_release(neigh); 1692 goto out; 1693 } 1694 read_unlock(&neigh_tbl_lock); 1695 err = -EAFNOSUPPORT; 1696 1697 out: 1698 return err; 1699 } 1700 1701 static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh) 1702 { 1703 struct net *net = sock_net(skb->sk); 1704 struct ndmsg *ndm; 1705 struct nlattr *tb[NDA_MAX+1]; 1706 struct neigh_table *tbl; 1707 struct net_device *dev = NULL; 1708 int err; 1709 1710 ASSERT_RTNL(); 1711 err = nlmsg_parse(nlh, sizeof(*ndm), tb, NDA_MAX, NULL); 1712 if (err < 0) 1713 goto out; 1714 1715 err = -EINVAL; 1716 if (tb[NDA_DST] == NULL) 1717 goto out; 1718 1719 ndm = nlmsg_data(nlh); 1720 if (ndm->ndm_ifindex) { 1721 dev = __dev_get_by_index(net, ndm->ndm_ifindex); 1722 if (dev == NULL) { 1723 err = -ENODEV; 1724 goto out; 1725 } 1726 1727 if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len) 1728 goto out; 1729 } 1730 1731 read_lock(&neigh_tbl_lock); 1732 for (tbl = neigh_tables; tbl; tbl = tbl->next) { 1733 int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE; 1734 struct neighbour *neigh; 1735 void *dst, *lladdr; 1736 1737 if (tbl->family != ndm->ndm_family) 1738 continue; 1739 read_unlock(&neigh_tbl_lock); 1740 1741 if (nla_len(tb[NDA_DST]) < tbl->key_len) 1742 goto out; 1743 dst = nla_data(tb[NDA_DST]); 1744 lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL; 1745 1746 if (ndm->ndm_flags & NTF_PROXY) { 1747 struct pneigh_entry *pn; 1748 1749 err = -ENOBUFS; 1750 pn = pneigh_lookup(tbl, net, dst, dev, 1); 1751 if (pn) { 1752 pn->flags = ndm->ndm_flags; 1753 err = 0; 1754 } 1755 goto out; 1756 } 1757 1758 if (dev == NULL) 1759 goto out; 1760 1761 neigh = neigh_lookup(tbl, dst, dev); 1762 if (neigh == NULL) { 1763 if (!(nlh->nlmsg_flags & NLM_F_CREATE)) { 1764 err = -ENOENT; 1765 goto out; 1766 } 1767 1768 neigh = __neigh_lookup_errno(tbl, dst, dev); 1769 if (IS_ERR(neigh)) { 1770 err = PTR_ERR(neigh); 1771 goto out; 1772 } 1773 } else { 1774 if (nlh->nlmsg_flags & NLM_F_EXCL) { 1775 err = -EEXIST; 1776 neigh_release(neigh); 1777 goto out; 1778 } 1779 1780 if (!(nlh->nlmsg_flags & NLM_F_REPLACE)) 1781 flags &= ~NEIGH_UPDATE_F_OVERRIDE; 1782 } 1783 1784 if (ndm->ndm_flags & NTF_USE) { 1785 neigh_event_send(neigh, NULL); 1786 err = 0; 1787 } else 1788 err = neigh_update(neigh, lladdr, ndm->ndm_state, flags); 1789 neigh_release(neigh); 1790 goto out; 1791 } 1792 1793 read_unlock(&neigh_tbl_lock); 1794 err = -EAFNOSUPPORT; 1795 out: 1796 return err; 1797 } 1798 1799 static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms) 1800 { 1801 struct nlattr *nest; 1802 1803 nest = nla_nest_start(skb, NDTA_PARMS); 1804 if (nest == NULL) 1805 return -ENOBUFS; 1806 1807 if ((parms->dev && 1808 nla_put_u32(skb, NDTPA_IFINDEX, parms->dev->ifindex)) || 1809 nla_put_u32(skb, NDTPA_REFCNT, atomic_read(&parms->refcnt)) || 1810 nla_put_u32(skb, NDTPA_QUEUE_LENBYTES, 1811 NEIGH_VAR(parms, QUEUE_LEN_BYTES)) || 1812 /* approximative value for deprecated QUEUE_LEN (in packets) */ 1813 nla_put_u32(skb, NDTPA_QUEUE_LEN, 1814 NEIGH_VAR(parms, QUEUE_LEN_BYTES) / SKB_TRUESIZE(ETH_FRAME_LEN)) || 1815 nla_put_u32(skb, NDTPA_PROXY_QLEN, NEIGH_VAR(parms, PROXY_QLEN)) || 1816 nla_put_u32(skb, NDTPA_APP_PROBES, NEIGH_VAR(parms, APP_PROBES)) || 1817 nla_put_u32(skb, NDTPA_UCAST_PROBES, 1818 NEIGH_VAR(parms, UCAST_PROBES)) || 1819 nla_put_u32(skb, NDTPA_MCAST_PROBES, 1820 NEIGH_VAR(parms, MCAST_PROBES)) || 1821 nla_put_msecs(skb, NDTPA_REACHABLE_TIME, parms->reachable_time) || 1822 nla_put_msecs(skb, NDTPA_BASE_REACHABLE_TIME, 1823 NEIGH_VAR(parms, BASE_REACHABLE_TIME)) || 1824 nla_put_msecs(skb, NDTPA_GC_STALETIME, 1825 NEIGH_VAR(parms, GC_STALETIME)) || 1826 nla_put_msecs(skb, NDTPA_DELAY_PROBE_TIME, 1827 NEIGH_VAR(parms, DELAY_PROBE_TIME)) || 1828 nla_put_msecs(skb, NDTPA_RETRANS_TIME, 1829 NEIGH_VAR(parms, RETRANS_TIME)) || 1830 nla_put_msecs(skb, NDTPA_ANYCAST_DELAY, 1831 NEIGH_VAR(parms, ANYCAST_DELAY)) || 1832 nla_put_msecs(skb, NDTPA_PROXY_DELAY, 1833 NEIGH_VAR(parms, PROXY_DELAY)) || 1834 nla_put_msecs(skb, NDTPA_LOCKTIME, 1835 NEIGH_VAR(parms, LOCKTIME))) 1836 goto nla_put_failure; 1837 return nla_nest_end(skb, nest); 1838 1839 nla_put_failure: 1840 nla_nest_cancel(skb, nest); 1841 return -EMSGSIZE; 1842 } 1843 1844 static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl, 1845 u32 pid, u32 seq, int type, int flags) 1846 { 1847 struct nlmsghdr *nlh; 1848 struct ndtmsg *ndtmsg; 1849 1850 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags); 1851 if (nlh == NULL) 1852 return -EMSGSIZE; 1853 1854 ndtmsg = nlmsg_data(nlh); 1855 1856 read_lock_bh(&tbl->lock); 1857 ndtmsg->ndtm_family = tbl->family; 1858 ndtmsg->ndtm_pad1 = 0; 1859 ndtmsg->ndtm_pad2 = 0; 1860 1861 if (nla_put_string(skb, NDTA_NAME, tbl->id) || 1862 nla_put_msecs(skb, NDTA_GC_INTERVAL, tbl->gc_interval) || 1863 nla_put_u32(skb, NDTA_THRESH1, tbl->gc_thresh1) || 1864 nla_put_u32(skb, NDTA_THRESH2, tbl->gc_thresh2) || 1865 nla_put_u32(skb, NDTA_THRESH3, tbl->gc_thresh3)) 1866 goto nla_put_failure; 1867 { 1868 unsigned long now = jiffies; 1869 unsigned int flush_delta = now - tbl->last_flush; 1870 unsigned int rand_delta = now - tbl->last_rand; 1871 struct neigh_hash_table *nht; 1872 struct ndt_config ndc = { 1873 .ndtc_key_len = tbl->key_len, 1874 .ndtc_entry_size = tbl->entry_size, 1875 .ndtc_entries = atomic_read(&tbl->entries), 1876 .ndtc_last_flush = jiffies_to_msecs(flush_delta), 1877 .ndtc_last_rand = jiffies_to_msecs(rand_delta), 1878 .ndtc_proxy_qlen = tbl->proxy_queue.qlen, 1879 }; 1880 1881 rcu_read_lock_bh(); 1882 nht = rcu_dereference_bh(tbl->nht); 1883 ndc.ndtc_hash_rnd = nht->hash_rnd[0]; 1884 ndc.ndtc_hash_mask = ((1 << nht->hash_shift) - 1); 1885 rcu_read_unlock_bh(); 1886 1887 if (nla_put(skb, NDTA_CONFIG, sizeof(ndc), &ndc)) 1888 goto nla_put_failure; 1889 } 1890 1891 { 1892 int cpu; 1893 struct ndt_stats ndst; 1894 1895 memset(&ndst, 0, sizeof(ndst)); 1896 1897 for_each_possible_cpu(cpu) { 1898 struct neigh_statistics *st; 1899 1900 st = per_cpu_ptr(tbl->stats, cpu); 1901 ndst.ndts_allocs += st->allocs; 1902 ndst.ndts_destroys += st->destroys; 1903 ndst.ndts_hash_grows += st->hash_grows; 1904 ndst.ndts_res_failed += st->res_failed; 1905 ndst.ndts_lookups += st->lookups; 1906 ndst.ndts_hits += st->hits; 1907 ndst.ndts_rcv_probes_mcast += st->rcv_probes_mcast; 1908 ndst.ndts_rcv_probes_ucast += st->rcv_probes_ucast; 1909 ndst.ndts_periodic_gc_runs += st->periodic_gc_runs; 1910 ndst.ndts_forced_gc_runs += st->forced_gc_runs; 1911 } 1912 1913 if (nla_put(skb, NDTA_STATS, sizeof(ndst), &ndst)) 1914 goto nla_put_failure; 1915 } 1916 1917 BUG_ON(tbl->parms.dev); 1918 if (neightbl_fill_parms(skb, &tbl->parms) < 0) 1919 goto nla_put_failure; 1920 1921 read_unlock_bh(&tbl->lock); 1922 return nlmsg_end(skb, nlh); 1923 1924 nla_put_failure: 1925 read_unlock_bh(&tbl->lock); 1926 nlmsg_cancel(skb, nlh); 1927 return -EMSGSIZE; 1928 } 1929 1930 static int neightbl_fill_param_info(struct sk_buff *skb, 1931 struct neigh_table *tbl, 1932 struct neigh_parms *parms, 1933 u32 pid, u32 seq, int type, 1934 unsigned int flags) 1935 { 1936 struct ndtmsg *ndtmsg; 1937 struct nlmsghdr *nlh; 1938 1939 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags); 1940 if (nlh == NULL) 1941 return -EMSGSIZE; 1942 1943 ndtmsg = nlmsg_data(nlh); 1944 1945 read_lock_bh(&tbl->lock); 1946 ndtmsg->ndtm_family = tbl->family; 1947 ndtmsg->ndtm_pad1 = 0; 1948 ndtmsg->ndtm_pad2 = 0; 1949 1950 if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 || 1951 neightbl_fill_parms(skb, parms) < 0) 1952 goto errout; 1953 1954 read_unlock_bh(&tbl->lock); 1955 return nlmsg_end(skb, nlh); 1956 errout: 1957 read_unlock_bh(&tbl->lock); 1958 nlmsg_cancel(skb, nlh); 1959 return -EMSGSIZE; 1960 } 1961 1962 static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = { 1963 [NDTA_NAME] = { .type = NLA_STRING }, 1964 [NDTA_THRESH1] = { .type = NLA_U32 }, 1965 [NDTA_THRESH2] = { .type = NLA_U32 }, 1966 [NDTA_THRESH3] = { .type = NLA_U32 }, 1967 [NDTA_GC_INTERVAL] = { .type = NLA_U64 }, 1968 [NDTA_PARMS] = { .type = NLA_NESTED }, 1969 }; 1970 1971 static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = { 1972 [NDTPA_IFINDEX] = { .type = NLA_U32 }, 1973 [NDTPA_QUEUE_LEN] = { .type = NLA_U32 }, 1974 [NDTPA_PROXY_QLEN] = { .type = NLA_U32 }, 1975 [NDTPA_APP_PROBES] = { .type = NLA_U32 }, 1976 [NDTPA_UCAST_PROBES] = { .type = NLA_U32 }, 1977 [NDTPA_MCAST_PROBES] = { .type = NLA_U32 }, 1978 [NDTPA_BASE_REACHABLE_TIME] = { .type = NLA_U64 }, 1979 [NDTPA_GC_STALETIME] = { .type = NLA_U64 }, 1980 [NDTPA_DELAY_PROBE_TIME] = { .type = NLA_U64 }, 1981 [NDTPA_RETRANS_TIME] = { .type = NLA_U64 }, 1982 [NDTPA_ANYCAST_DELAY] = { .type = NLA_U64 }, 1983 [NDTPA_PROXY_DELAY] = { .type = NLA_U64 }, 1984 [NDTPA_LOCKTIME] = { .type = NLA_U64 }, 1985 }; 1986 1987 static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh) 1988 { 1989 struct net *net = sock_net(skb->sk); 1990 struct neigh_table *tbl; 1991 struct ndtmsg *ndtmsg; 1992 struct nlattr *tb[NDTA_MAX+1]; 1993 int err; 1994 1995 err = nlmsg_parse(nlh, sizeof(*ndtmsg), tb, NDTA_MAX, 1996 nl_neightbl_policy); 1997 if (err < 0) 1998 goto errout; 1999 2000 if (tb[NDTA_NAME] == NULL) { 2001 err = -EINVAL; 2002 goto errout; 2003 } 2004 2005 ndtmsg = nlmsg_data(nlh); 2006 read_lock(&neigh_tbl_lock); 2007 for (tbl = neigh_tables; tbl; tbl = tbl->next) { 2008 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family) 2009 continue; 2010 2011 if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0) 2012 break; 2013 } 2014 2015 if (tbl == NULL) { 2016 err = -ENOENT; 2017 goto errout_locked; 2018 } 2019 2020 /* 2021 * We acquire tbl->lock to be nice to the periodic timers and 2022 * make sure they always see a consistent set of values. 2023 */ 2024 write_lock_bh(&tbl->lock); 2025 2026 if (tb[NDTA_PARMS]) { 2027 struct nlattr *tbp[NDTPA_MAX+1]; 2028 struct neigh_parms *p; 2029 int i, ifindex = 0; 2030 2031 err = nla_parse_nested(tbp, NDTPA_MAX, tb[NDTA_PARMS], 2032 nl_ntbl_parm_policy); 2033 if (err < 0) 2034 goto errout_tbl_lock; 2035 2036 if (tbp[NDTPA_IFINDEX]) 2037 ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]); 2038 2039 p = lookup_neigh_parms(tbl, net, ifindex); 2040 if (p == NULL) { 2041 err = -ENOENT; 2042 goto errout_tbl_lock; 2043 } 2044 2045 for (i = 1; i <= NDTPA_MAX; i++) { 2046 if (tbp[i] == NULL) 2047 continue; 2048 2049 switch (i) { 2050 case NDTPA_QUEUE_LEN: 2051 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES, 2052 nla_get_u32(tbp[i]) * 2053 SKB_TRUESIZE(ETH_FRAME_LEN)); 2054 break; 2055 case NDTPA_QUEUE_LENBYTES: 2056 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES, 2057 nla_get_u32(tbp[i])); 2058 break; 2059 case NDTPA_PROXY_QLEN: 2060 NEIGH_VAR_SET(p, PROXY_QLEN, 2061 nla_get_u32(tbp[i])); 2062 break; 2063 case NDTPA_APP_PROBES: 2064 NEIGH_VAR_SET(p, APP_PROBES, 2065 nla_get_u32(tbp[i])); 2066 break; 2067 case NDTPA_UCAST_PROBES: 2068 NEIGH_VAR_SET(p, UCAST_PROBES, 2069 nla_get_u32(tbp[i])); 2070 break; 2071 case NDTPA_MCAST_PROBES: 2072 NEIGH_VAR_SET(p, MCAST_PROBES, 2073 nla_get_u32(tbp[i])); 2074 break; 2075 case NDTPA_BASE_REACHABLE_TIME: 2076 NEIGH_VAR_SET(p, BASE_REACHABLE_TIME, 2077 nla_get_msecs(tbp[i])); 2078 break; 2079 case NDTPA_GC_STALETIME: 2080 NEIGH_VAR_SET(p, GC_STALETIME, 2081 nla_get_msecs(tbp[i])); 2082 break; 2083 case NDTPA_DELAY_PROBE_TIME: 2084 NEIGH_VAR_SET(p, DELAY_PROBE_TIME, 2085 nla_get_msecs(tbp[i])); 2086 break; 2087 case NDTPA_RETRANS_TIME: 2088 NEIGH_VAR_SET(p, RETRANS_TIME, 2089 nla_get_msecs(tbp[i])); 2090 break; 2091 case NDTPA_ANYCAST_DELAY: 2092 NEIGH_VAR_SET(p, ANYCAST_DELAY, 2093 nla_get_msecs(tbp[i])); 2094 break; 2095 case NDTPA_PROXY_DELAY: 2096 NEIGH_VAR_SET(p, PROXY_DELAY, 2097 nla_get_msecs(tbp[i])); 2098 break; 2099 case NDTPA_LOCKTIME: 2100 NEIGH_VAR_SET(p, LOCKTIME, 2101 nla_get_msecs(tbp[i])); 2102 break; 2103 } 2104 } 2105 } 2106 2107 err = -ENOENT; 2108 if ((tb[NDTA_THRESH1] || tb[NDTA_THRESH2] || 2109 tb[NDTA_THRESH3] || tb[NDTA_GC_INTERVAL]) && 2110 !net_eq(net, &init_net)) 2111 goto errout_tbl_lock; 2112 2113 if (tb[NDTA_THRESH1]) 2114 tbl->gc_thresh1 = nla_get_u32(tb[NDTA_THRESH1]); 2115 2116 if (tb[NDTA_THRESH2]) 2117 tbl->gc_thresh2 = nla_get_u32(tb[NDTA_THRESH2]); 2118 2119 if (tb[NDTA_THRESH3]) 2120 tbl->gc_thresh3 = nla_get_u32(tb[NDTA_THRESH3]); 2121 2122 if (tb[NDTA_GC_INTERVAL]) 2123 tbl->gc_interval = nla_get_msecs(tb[NDTA_GC_INTERVAL]); 2124 2125 err = 0; 2126 2127 errout_tbl_lock: 2128 write_unlock_bh(&tbl->lock); 2129 errout_locked: 2130 read_unlock(&neigh_tbl_lock); 2131 errout: 2132 return err; 2133 } 2134 2135 static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb) 2136 { 2137 struct net *net = sock_net(skb->sk); 2138 int family, tidx, nidx = 0; 2139 int tbl_skip = cb->args[0]; 2140 int neigh_skip = cb->args[1]; 2141 struct neigh_table *tbl; 2142 2143 family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family; 2144 2145 read_lock(&neigh_tbl_lock); 2146 for (tbl = neigh_tables, tidx = 0; tbl; tbl = tbl->next, tidx++) { 2147 struct neigh_parms *p; 2148 2149 if (tidx < tbl_skip || (family && tbl->family != family)) 2150 continue; 2151 2152 if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).portid, 2153 cb->nlh->nlmsg_seq, RTM_NEWNEIGHTBL, 2154 NLM_F_MULTI) <= 0) 2155 break; 2156 2157 for (nidx = 0, p = tbl->parms.next; p; p = p->next) { 2158 if (!net_eq(neigh_parms_net(p), net)) 2159 continue; 2160 2161 if (nidx < neigh_skip) 2162 goto next; 2163 2164 if (neightbl_fill_param_info(skb, tbl, p, 2165 NETLINK_CB(cb->skb).portid, 2166 cb->nlh->nlmsg_seq, 2167 RTM_NEWNEIGHTBL, 2168 NLM_F_MULTI) <= 0) 2169 goto out; 2170 next: 2171 nidx++; 2172 } 2173 2174 neigh_skip = 0; 2175 } 2176 out: 2177 read_unlock(&neigh_tbl_lock); 2178 cb->args[0] = tidx; 2179 cb->args[1] = nidx; 2180 2181 return skb->len; 2182 } 2183 2184 static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh, 2185 u32 pid, u32 seq, int type, unsigned int flags) 2186 { 2187 unsigned long now = jiffies; 2188 struct nda_cacheinfo ci; 2189 struct nlmsghdr *nlh; 2190 struct ndmsg *ndm; 2191 2192 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags); 2193 if (nlh == NULL) 2194 return -EMSGSIZE; 2195 2196 ndm = nlmsg_data(nlh); 2197 ndm->ndm_family = neigh->ops->family; 2198 ndm->ndm_pad1 = 0; 2199 ndm->ndm_pad2 = 0; 2200 ndm->ndm_flags = neigh->flags; 2201 ndm->ndm_type = neigh->type; 2202 ndm->ndm_ifindex = neigh->dev->ifindex; 2203 2204 if (nla_put(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key)) 2205 goto nla_put_failure; 2206 2207 read_lock_bh(&neigh->lock); 2208 ndm->ndm_state = neigh->nud_state; 2209 if (neigh->nud_state & NUD_VALID) { 2210 char haddr[MAX_ADDR_LEN]; 2211 2212 neigh_ha_snapshot(haddr, neigh, neigh->dev); 2213 if (nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, haddr) < 0) { 2214 read_unlock_bh(&neigh->lock); 2215 goto nla_put_failure; 2216 } 2217 } 2218 2219 ci.ndm_used = jiffies_to_clock_t(now - neigh->used); 2220 ci.ndm_confirmed = jiffies_to_clock_t(now - neigh->confirmed); 2221 ci.ndm_updated = jiffies_to_clock_t(now - neigh->updated); 2222 ci.ndm_refcnt = atomic_read(&neigh->refcnt) - 1; 2223 read_unlock_bh(&neigh->lock); 2224 2225 if (nla_put_u32(skb, NDA_PROBES, atomic_read(&neigh->probes)) || 2226 nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci)) 2227 goto nla_put_failure; 2228 2229 return nlmsg_end(skb, nlh); 2230 2231 nla_put_failure: 2232 nlmsg_cancel(skb, nlh); 2233 return -EMSGSIZE; 2234 } 2235 2236 static int pneigh_fill_info(struct sk_buff *skb, struct pneigh_entry *pn, 2237 u32 pid, u32 seq, int type, unsigned int flags, 2238 struct neigh_table *tbl) 2239 { 2240 struct nlmsghdr *nlh; 2241 struct ndmsg *ndm; 2242 2243 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags); 2244 if (nlh == NULL) 2245 return -EMSGSIZE; 2246 2247 ndm = nlmsg_data(nlh); 2248 ndm->ndm_family = tbl->family; 2249 ndm->ndm_pad1 = 0; 2250 ndm->ndm_pad2 = 0; 2251 ndm->ndm_flags = pn->flags | NTF_PROXY; 2252 ndm->ndm_type = NDA_DST; 2253 ndm->ndm_ifindex = pn->dev->ifindex; 2254 ndm->ndm_state = NUD_NONE; 2255 2256 if (nla_put(skb, NDA_DST, tbl->key_len, pn->key)) 2257 goto nla_put_failure; 2258 2259 return nlmsg_end(skb, nlh); 2260 2261 nla_put_failure: 2262 nlmsg_cancel(skb, nlh); 2263 return -EMSGSIZE; 2264 } 2265 2266 static void neigh_update_notify(struct neighbour *neigh) 2267 { 2268 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh); 2269 __neigh_notify(neigh, RTM_NEWNEIGH, 0); 2270 } 2271 2272 static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb, 2273 struct netlink_callback *cb) 2274 { 2275 struct net *net = sock_net(skb->sk); 2276 struct neighbour *n; 2277 int rc, h, s_h = cb->args[1]; 2278 int idx, s_idx = idx = cb->args[2]; 2279 struct neigh_hash_table *nht; 2280 2281 rcu_read_lock_bh(); 2282 nht = rcu_dereference_bh(tbl->nht); 2283 2284 for (h = s_h; h < (1 << nht->hash_shift); h++) { 2285 if (h > s_h) 2286 s_idx = 0; 2287 for (n = rcu_dereference_bh(nht->hash_buckets[h]), idx = 0; 2288 n != NULL; 2289 n = rcu_dereference_bh(n->next)) { 2290 if (!net_eq(dev_net(n->dev), net)) 2291 continue; 2292 if (idx < s_idx) 2293 goto next; 2294 if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid, 2295 cb->nlh->nlmsg_seq, 2296 RTM_NEWNEIGH, 2297 NLM_F_MULTI) <= 0) { 2298 rc = -1; 2299 goto out; 2300 } 2301 next: 2302 idx++; 2303 } 2304 } 2305 rc = skb->len; 2306 out: 2307 rcu_read_unlock_bh(); 2308 cb->args[1] = h; 2309 cb->args[2] = idx; 2310 return rc; 2311 } 2312 2313 static int pneigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb, 2314 struct netlink_callback *cb) 2315 { 2316 struct pneigh_entry *n; 2317 struct net *net = sock_net(skb->sk); 2318 int rc, h, s_h = cb->args[3]; 2319 int idx, s_idx = idx = cb->args[4]; 2320 2321 read_lock_bh(&tbl->lock); 2322 2323 for (h = s_h; h <= PNEIGH_HASHMASK; h++) { 2324 if (h > s_h) 2325 s_idx = 0; 2326 for (n = tbl->phash_buckets[h], idx = 0; n; n = n->next) { 2327 if (dev_net(n->dev) != net) 2328 continue; 2329 if (idx < s_idx) 2330 goto next; 2331 if (pneigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid, 2332 cb->nlh->nlmsg_seq, 2333 RTM_NEWNEIGH, 2334 NLM_F_MULTI, tbl) <= 0) { 2335 read_unlock_bh(&tbl->lock); 2336 rc = -1; 2337 goto out; 2338 } 2339 next: 2340 idx++; 2341 } 2342 } 2343 2344 read_unlock_bh(&tbl->lock); 2345 rc = skb->len; 2346 out: 2347 cb->args[3] = h; 2348 cb->args[4] = idx; 2349 return rc; 2350 2351 } 2352 2353 static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb) 2354 { 2355 struct neigh_table *tbl; 2356 int t, family, s_t; 2357 int proxy = 0; 2358 int err; 2359 2360 read_lock(&neigh_tbl_lock); 2361 family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family; 2362 2363 /* check for full ndmsg structure presence, family member is 2364 * the same for both structures 2365 */ 2366 if (nlmsg_len(cb->nlh) >= sizeof(struct ndmsg) && 2367 ((struct ndmsg *) nlmsg_data(cb->nlh))->ndm_flags == NTF_PROXY) 2368 proxy = 1; 2369 2370 s_t = cb->args[0]; 2371 2372 for (tbl = neigh_tables, t = 0; tbl; 2373 tbl = tbl->next, t++) { 2374 if (t < s_t || (family && tbl->family != family)) 2375 continue; 2376 if (t > s_t) 2377 memset(&cb->args[1], 0, sizeof(cb->args) - 2378 sizeof(cb->args[0])); 2379 if (proxy) 2380 err = pneigh_dump_table(tbl, skb, cb); 2381 else 2382 err = neigh_dump_table(tbl, skb, cb); 2383 if (err < 0) 2384 break; 2385 } 2386 read_unlock(&neigh_tbl_lock); 2387 2388 cb->args[0] = t; 2389 return skb->len; 2390 } 2391 2392 void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie) 2393 { 2394 int chain; 2395 struct neigh_hash_table *nht; 2396 2397 rcu_read_lock_bh(); 2398 nht = rcu_dereference_bh(tbl->nht); 2399 2400 read_lock(&tbl->lock); /* avoid resizes */ 2401 for (chain = 0; chain < (1 << nht->hash_shift); chain++) { 2402 struct neighbour *n; 2403 2404 for (n = rcu_dereference_bh(nht->hash_buckets[chain]); 2405 n != NULL; 2406 n = rcu_dereference_bh(n->next)) 2407 cb(n, cookie); 2408 } 2409 read_unlock(&tbl->lock); 2410 rcu_read_unlock_bh(); 2411 } 2412 EXPORT_SYMBOL(neigh_for_each); 2413 2414 /* The tbl->lock must be held as a writer and BH disabled. */ 2415 void __neigh_for_each_release(struct neigh_table *tbl, 2416 int (*cb)(struct neighbour *)) 2417 { 2418 int chain; 2419 struct neigh_hash_table *nht; 2420 2421 nht = rcu_dereference_protected(tbl->nht, 2422 lockdep_is_held(&tbl->lock)); 2423 for (chain = 0; chain < (1 << nht->hash_shift); chain++) { 2424 struct neighbour *n; 2425 struct neighbour __rcu **np; 2426 2427 np = &nht->hash_buckets[chain]; 2428 while ((n = rcu_dereference_protected(*np, 2429 lockdep_is_held(&tbl->lock))) != NULL) { 2430 int release; 2431 2432 write_lock(&n->lock); 2433 release = cb(n); 2434 if (release) { 2435 rcu_assign_pointer(*np, 2436 rcu_dereference_protected(n->next, 2437 lockdep_is_held(&tbl->lock))); 2438 n->dead = 1; 2439 } else 2440 np = &n->next; 2441 write_unlock(&n->lock); 2442 if (release) 2443 neigh_cleanup_and_release(n); 2444 } 2445 } 2446 } 2447 EXPORT_SYMBOL(__neigh_for_each_release); 2448 2449 #ifdef CONFIG_PROC_FS 2450 2451 static struct neighbour *neigh_get_first(struct seq_file *seq) 2452 { 2453 struct neigh_seq_state *state = seq->private; 2454 struct net *net = seq_file_net(seq); 2455 struct neigh_hash_table *nht = state->nht; 2456 struct neighbour *n = NULL; 2457 int bucket = state->bucket; 2458 2459 state->flags &= ~NEIGH_SEQ_IS_PNEIGH; 2460 for (bucket = 0; bucket < (1 << nht->hash_shift); bucket++) { 2461 n = rcu_dereference_bh(nht->hash_buckets[bucket]); 2462 2463 while (n) { 2464 if (!net_eq(dev_net(n->dev), net)) 2465 goto next; 2466 if (state->neigh_sub_iter) { 2467 loff_t fakep = 0; 2468 void *v; 2469 2470 v = state->neigh_sub_iter(state, n, &fakep); 2471 if (!v) 2472 goto next; 2473 } 2474 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP)) 2475 break; 2476 if (n->nud_state & ~NUD_NOARP) 2477 break; 2478 next: 2479 n = rcu_dereference_bh(n->next); 2480 } 2481 2482 if (n) 2483 break; 2484 } 2485 state->bucket = bucket; 2486 2487 return n; 2488 } 2489 2490 static struct neighbour *neigh_get_next(struct seq_file *seq, 2491 struct neighbour *n, 2492 loff_t *pos) 2493 { 2494 struct neigh_seq_state *state = seq->private; 2495 struct net *net = seq_file_net(seq); 2496 struct neigh_hash_table *nht = state->nht; 2497 2498 if (state->neigh_sub_iter) { 2499 void *v = state->neigh_sub_iter(state, n, pos); 2500 if (v) 2501 return n; 2502 } 2503 n = rcu_dereference_bh(n->next); 2504 2505 while (1) { 2506 while (n) { 2507 if (!net_eq(dev_net(n->dev), net)) 2508 goto next; 2509 if (state->neigh_sub_iter) { 2510 void *v = state->neigh_sub_iter(state, n, pos); 2511 if (v) 2512 return n; 2513 goto next; 2514 } 2515 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP)) 2516 break; 2517 2518 if (n->nud_state & ~NUD_NOARP) 2519 break; 2520 next: 2521 n = rcu_dereference_bh(n->next); 2522 } 2523 2524 if (n) 2525 break; 2526 2527 if (++state->bucket >= (1 << nht->hash_shift)) 2528 break; 2529 2530 n = rcu_dereference_bh(nht->hash_buckets[state->bucket]); 2531 } 2532 2533 if (n && pos) 2534 --(*pos); 2535 return n; 2536 } 2537 2538 static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos) 2539 { 2540 struct neighbour *n = neigh_get_first(seq); 2541 2542 if (n) { 2543 --(*pos); 2544 while (*pos) { 2545 n = neigh_get_next(seq, n, pos); 2546 if (!n) 2547 break; 2548 } 2549 } 2550 return *pos ? NULL : n; 2551 } 2552 2553 static struct pneigh_entry *pneigh_get_first(struct seq_file *seq) 2554 { 2555 struct neigh_seq_state *state = seq->private; 2556 struct net *net = seq_file_net(seq); 2557 struct neigh_table *tbl = state->tbl; 2558 struct pneigh_entry *pn = NULL; 2559 int bucket = state->bucket; 2560 2561 state->flags |= NEIGH_SEQ_IS_PNEIGH; 2562 for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) { 2563 pn = tbl->phash_buckets[bucket]; 2564 while (pn && !net_eq(pneigh_net(pn), net)) 2565 pn = pn->next; 2566 if (pn) 2567 break; 2568 } 2569 state->bucket = bucket; 2570 2571 return pn; 2572 } 2573 2574 static struct pneigh_entry *pneigh_get_next(struct seq_file *seq, 2575 struct pneigh_entry *pn, 2576 loff_t *pos) 2577 { 2578 struct neigh_seq_state *state = seq->private; 2579 struct net *net = seq_file_net(seq); 2580 struct neigh_table *tbl = state->tbl; 2581 2582 do { 2583 pn = pn->next; 2584 } while (pn && !net_eq(pneigh_net(pn), net)); 2585 2586 while (!pn) { 2587 if (++state->bucket > PNEIGH_HASHMASK) 2588 break; 2589 pn = tbl->phash_buckets[state->bucket]; 2590 while (pn && !net_eq(pneigh_net(pn), net)) 2591 pn = pn->next; 2592 if (pn) 2593 break; 2594 } 2595 2596 if (pn && pos) 2597 --(*pos); 2598 2599 return pn; 2600 } 2601 2602 static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos) 2603 { 2604 struct pneigh_entry *pn = pneigh_get_first(seq); 2605 2606 if (pn) { 2607 --(*pos); 2608 while (*pos) { 2609 pn = pneigh_get_next(seq, pn, pos); 2610 if (!pn) 2611 break; 2612 } 2613 } 2614 return *pos ? NULL : pn; 2615 } 2616 2617 static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos) 2618 { 2619 struct neigh_seq_state *state = seq->private; 2620 void *rc; 2621 loff_t idxpos = *pos; 2622 2623 rc = neigh_get_idx(seq, &idxpos); 2624 if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY)) 2625 rc = pneigh_get_idx(seq, &idxpos); 2626 2627 return rc; 2628 } 2629 2630 void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags) 2631 __acquires(rcu_bh) 2632 { 2633 struct neigh_seq_state *state = seq->private; 2634 2635 state->tbl = tbl; 2636 state->bucket = 0; 2637 state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH); 2638 2639 rcu_read_lock_bh(); 2640 state->nht = rcu_dereference_bh(tbl->nht); 2641 2642 return *pos ? neigh_get_idx_any(seq, pos) : SEQ_START_TOKEN; 2643 } 2644 EXPORT_SYMBOL(neigh_seq_start); 2645 2646 void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2647 { 2648 struct neigh_seq_state *state; 2649 void *rc; 2650 2651 if (v == SEQ_START_TOKEN) { 2652 rc = neigh_get_first(seq); 2653 goto out; 2654 } 2655 2656 state = seq->private; 2657 if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) { 2658 rc = neigh_get_next(seq, v, NULL); 2659 if (rc) 2660 goto out; 2661 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY)) 2662 rc = pneigh_get_first(seq); 2663 } else { 2664 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY); 2665 rc = pneigh_get_next(seq, v, NULL); 2666 } 2667 out: 2668 ++(*pos); 2669 return rc; 2670 } 2671 EXPORT_SYMBOL(neigh_seq_next); 2672 2673 void neigh_seq_stop(struct seq_file *seq, void *v) 2674 __releases(rcu_bh) 2675 { 2676 rcu_read_unlock_bh(); 2677 } 2678 EXPORT_SYMBOL(neigh_seq_stop); 2679 2680 /* statistics via seq_file */ 2681 2682 static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos) 2683 { 2684 struct neigh_table *tbl = seq->private; 2685 int cpu; 2686 2687 if (*pos == 0) 2688 return SEQ_START_TOKEN; 2689 2690 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) { 2691 if (!cpu_possible(cpu)) 2692 continue; 2693 *pos = cpu+1; 2694 return per_cpu_ptr(tbl->stats, cpu); 2695 } 2696 return NULL; 2697 } 2698 2699 static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2700 { 2701 struct neigh_table *tbl = seq->private; 2702 int cpu; 2703 2704 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) { 2705 if (!cpu_possible(cpu)) 2706 continue; 2707 *pos = cpu+1; 2708 return per_cpu_ptr(tbl->stats, cpu); 2709 } 2710 return NULL; 2711 } 2712 2713 static void neigh_stat_seq_stop(struct seq_file *seq, void *v) 2714 { 2715 2716 } 2717 2718 static int neigh_stat_seq_show(struct seq_file *seq, void *v) 2719 { 2720 struct neigh_table *tbl = seq->private; 2721 struct neigh_statistics *st = v; 2722 2723 if (v == SEQ_START_TOKEN) { 2724 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"); 2725 return 0; 2726 } 2727 2728 seq_printf(seq, "%08x %08lx %08lx %08lx %08lx %08lx %08lx " 2729 "%08lx %08lx %08lx %08lx %08lx\n", 2730 atomic_read(&tbl->entries), 2731 2732 st->allocs, 2733 st->destroys, 2734 st->hash_grows, 2735 2736 st->lookups, 2737 st->hits, 2738 2739 st->res_failed, 2740 2741 st->rcv_probes_mcast, 2742 st->rcv_probes_ucast, 2743 2744 st->periodic_gc_runs, 2745 st->forced_gc_runs, 2746 st->unres_discards 2747 ); 2748 2749 return 0; 2750 } 2751 2752 static const struct seq_operations neigh_stat_seq_ops = { 2753 .start = neigh_stat_seq_start, 2754 .next = neigh_stat_seq_next, 2755 .stop = neigh_stat_seq_stop, 2756 .show = neigh_stat_seq_show, 2757 }; 2758 2759 static int neigh_stat_seq_open(struct inode *inode, struct file *file) 2760 { 2761 int ret = seq_open(file, &neigh_stat_seq_ops); 2762 2763 if (!ret) { 2764 struct seq_file *sf = file->private_data; 2765 sf->private = PDE_DATA(inode); 2766 } 2767 return ret; 2768 }; 2769 2770 static const struct file_operations neigh_stat_seq_fops = { 2771 .owner = THIS_MODULE, 2772 .open = neigh_stat_seq_open, 2773 .read = seq_read, 2774 .llseek = seq_lseek, 2775 .release = seq_release, 2776 }; 2777 2778 #endif /* CONFIG_PROC_FS */ 2779 2780 static inline size_t neigh_nlmsg_size(void) 2781 { 2782 return NLMSG_ALIGN(sizeof(struct ndmsg)) 2783 + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */ 2784 + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */ 2785 + nla_total_size(sizeof(struct nda_cacheinfo)) 2786 + nla_total_size(4); /* NDA_PROBES */ 2787 } 2788 2789 static void __neigh_notify(struct neighbour *n, int type, int flags) 2790 { 2791 struct net *net = dev_net(n->dev); 2792 struct sk_buff *skb; 2793 int err = -ENOBUFS; 2794 2795 skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC); 2796 if (skb == NULL) 2797 goto errout; 2798 2799 err = neigh_fill_info(skb, n, 0, 0, type, flags); 2800 if (err < 0) { 2801 /* -EMSGSIZE implies BUG in neigh_nlmsg_size() */ 2802 WARN_ON(err == -EMSGSIZE); 2803 kfree_skb(skb); 2804 goto errout; 2805 } 2806 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC); 2807 return; 2808 errout: 2809 if (err < 0) 2810 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err); 2811 } 2812 2813 void neigh_app_ns(struct neighbour *n) 2814 { 2815 __neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST); 2816 } 2817 EXPORT_SYMBOL(neigh_app_ns); 2818 2819 #ifdef CONFIG_SYSCTL 2820 static int zero; 2821 static int int_max = INT_MAX; 2822 static int unres_qlen_max = INT_MAX / SKB_TRUESIZE(ETH_FRAME_LEN); 2823 2824 static int proc_unres_qlen(struct ctl_table *ctl, int write, 2825 void __user *buffer, size_t *lenp, loff_t *ppos) 2826 { 2827 int size, ret; 2828 struct ctl_table tmp = *ctl; 2829 2830 tmp.extra1 = &zero; 2831 tmp.extra2 = &unres_qlen_max; 2832 tmp.data = &size; 2833 2834 size = *(int *)ctl->data / SKB_TRUESIZE(ETH_FRAME_LEN); 2835 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos); 2836 2837 if (write && !ret) 2838 *(int *)ctl->data = size * SKB_TRUESIZE(ETH_FRAME_LEN); 2839 return ret; 2840 } 2841 2842 static struct neigh_parms *neigh_get_dev_parms_rcu(struct net_device *dev, 2843 int family) 2844 { 2845 switch (family) { 2846 case AF_INET: 2847 return __in_dev_arp_parms_get_rcu(dev); 2848 case AF_INET6: 2849 return __in6_dev_nd_parms_get_rcu(dev); 2850 } 2851 return NULL; 2852 } 2853 2854 static void neigh_copy_dflt_parms(struct net *net, struct neigh_parms *p, 2855 int index) 2856 { 2857 struct net_device *dev; 2858 int family = neigh_parms_family(p); 2859 2860 rcu_read_lock(); 2861 for_each_netdev_rcu(net, dev) { 2862 struct neigh_parms *dst_p = 2863 neigh_get_dev_parms_rcu(dev, family); 2864 2865 if (dst_p && !test_bit(index, dst_p->data_state)) 2866 dst_p->data[index] = p->data[index]; 2867 } 2868 rcu_read_unlock(); 2869 } 2870 2871 static void neigh_proc_update(struct ctl_table *ctl, int write) 2872 { 2873 struct net_device *dev = ctl->extra1; 2874 struct neigh_parms *p = ctl->extra2; 2875 struct net *net = neigh_parms_net(p); 2876 int index = (int *) ctl->data - p->data; 2877 2878 if (!write) 2879 return; 2880 2881 set_bit(index, p->data_state); 2882 if (!dev) /* NULL dev means this is default value */ 2883 neigh_copy_dflt_parms(net, p, index); 2884 } 2885 2886 static int neigh_proc_dointvec_zero_intmax(struct ctl_table *ctl, int write, 2887 void __user *buffer, 2888 size_t *lenp, loff_t *ppos) 2889 { 2890 struct ctl_table tmp = *ctl; 2891 int ret; 2892 2893 tmp.extra1 = &zero; 2894 tmp.extra2 = &int_max; 2895 2896 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos); 2897 neigh_proc_update(ctl, write); 2898 return ret; 2899 } 2900 2901 int neigh_proc_dointvec(struct ctl_table *ctl, int write, 2902 void __user *buffer, size_t *lenp, loff_t *ppos) 2903 { 2904 int ret = proc_dointvec(ctl, write, buffer, lenp, ppos); 2905 2906 neigh_proc_update(ctl, write); 2907 return ret; 2908 } 2909 EXPORT_SYMBOL(neigh_proc_dointvec); 2910 2911 int neigh_proc_dointvec_jiffies(struct ctl_table *ctl, int write, 2912 void __user *buffer, 2913 size_t *lenp, loff_t *ppos) 2914 { 2915 int ret = proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos); 2916 2917 neigh_proc_update(ctl, write); 2918 return ret; 2919 } 2920 EXPORT_SYMBOL(neigh_proc_dointvec_jiffies); 2921 2922 static int neigh_proc_dointvec_userhz_jiffies(struct ctl_table *ctl, int write, 2923 void __user *buffer, 2924 size_t *lenp, loff_t *ppos) 2925 { 2926 int ret = proc_dointvec_userhz_jiffies(ctl, write, buffer, lenp, ppos); 2927 2928 neigh_proc_update(ctl, write); 2929 return ret; 2930 } 2931 2932 int neigh_proc_dointvec_ms_jiffies(struct ctl_table *ctl, int write, 2933 void __user *buffer, 2934 size_t *lenp, loff_t *ppos) 2935 { 2936 int ret = proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos); 2937 2938 neigh_proc_update(ctl, write); 2939 return ret; 2940 } 2941 EXPORT_SYMBOL(neigh_proc_dointvec_ms_jiffies); 2942 2943 static int neigh_proc_dointvec_unres_qlen(struct ctl_table *ctl, int write, 2944 void __user *buffer, 2945 size_t *lenp, loff_t *ppos) 2946 { 2947 int ret = proc_unres_qlen(ctl, write, buffer, lenp, ppos); 2948 2949 neigh_proc_update(ctl, write); 2950 return ret; 2951 } 2952 2953 #define NEIGH_PARMS_DATA_OFFSET(index) \ 2954 (&((struct neigh_parms *) 0)->data[index]) 2955 2956 #define NEIGH_SYSCTL_ENTRY(attr, data_attr, name, mval, proc) \ 2957 [NEIGH_VAR_ ## attr] = { \ 2958 .procname = name, \ 2959 .data = NEIGH_PARMS_DATA_OFFSET(NEIGH_VAR_ ## data_attr), \ 2960 .maxlen = sizeof(int), \ 2961 .mode = mval, \ 2962 .proc_handler = proc, \ 2963 } 2964 2965 #define NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(attr, name) \ 2966 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_zero_intmax) 2967 2968 #define NEIGH_SYSCTL_JIFFIES_ENTRY(attr, name) \ 2969 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_jiffies) 2970 2971 #define NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(attr, name) \ 2972 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_userhz_jiffies) 2973 2974 #define NEIGH_SYSCTL_MS_JIFFIES_ENTRY(attr, name) \ 2975 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_ms_jiffies) 2976 2977 #define NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(attr, data_attr, name) \ 2978 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_ms_jiffies) 2979 2980 #define NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(attr, data_attr, name) \ 2981 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_unres_qlen) 2982 2983 static struct neigh_sysctl_table { 2984 struct ctl_table_header *sysctl_header; 2985 struct ctl_table neigh_vars[NEIGH_VAR_MAX + 1]; 2986 } neigh_sysctl_template __read_mostly = { 2987 .neigh_vars = { 2988 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_PROBES, "mcast_solicit"), 2989 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(UCAST_PROBES, "ucast_solicit"), 2990 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(APP_PROBES, "app_solicit"), 2991 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(RETRANS_TIME, "retrans_time"), 2992 NEIGH_SYSCTL_JIFFIES_ENTRY(BASE_REACHABLE_TIME, "base_reachable_time"), 2993 NEIGH_SYSCTL_JIFFIES_ENTRY(DELAY_PROBE_TIME, "delay_first_probe_time"), 2994 NEIGH_SYSCTL_JIFFIES_ENTRY(GC_STALETIME, "gc_stale_time"), 2995 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(QUEUE_LEN_BYTES, "unres_qlen_bytes"), 2996 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(PROXY_QLEN, "proxy_qlen"), 2997 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(ANYCAST_DELAY, "anycast_delay"), 2998 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(PROXY_DELAY, "proxy_delay"), 2999 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(LOCKTIME, "locktime"), 3000 NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(QUEUE_LEN, QUEUE_LEN_BYTES, "unres_qlen"), 3001 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(RETRANS_TIME_MS, RETRANS_TIME, "retrans_time_ms"), 3002 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(BASE_REACHABLE_TIME_MS, BASE_REACHABLE_TIME, "base_reachable_time_ms"), 3003 [NEIGH_VAR_GC_INTERVAL] = { 3004 .procname = "gc_interval", 3005 .maxlen = sizeof(int), 3006 .mode = 0644, 3007 .proc_handler = proc_dointvec_jiffies, 3008 }, 3009 [NEIGH_VAR_GC_THRESH1] = { 3010 .procname = "gc_thresh1", 3011 .maxlen = sizeof(int), 3012 .mode = 0644, 3013 .extra1 = &zero, 3014 .extra2 = &int_max, 3015 .proc_handler = proc_dointvec_minmax, 3016 }, 3017 [NEIGH_VAR_GC_THRESH2] = { 3018 .procname = "gc_thresh2", 3019 .maxlen = sizeof(int), 3020 .mode = 0644, 3021 .extra1 = &zero, 3022 .extra2 = &int_max, 3023 .proc_handler = proc_dointvec_minmax, 3024 }, 3025 [NEIGH_VAR_GC_THRESH3] = { 3026 .procname = "gc_thresh3", 3027 .maxlen = sizeof(int), 3028 .mode = 0644, 3029 .extra1 = &zero, 3030 .extra2 = &int_max, 3031 .proc_handler = proc_dointvec_minmax, 3032 }, 3033 {}, 3034 }, 3035 }; 3036 3037 int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p, 3038 proc_handler *handler) 3039 { 3040 int i; 3041 struct neigh_sysctl_table *t; 3042 const char *dev_name_source; 3043 char neigh_path[ sizeof("net//neigh/") + IFNAMSIZ + IFNAMSIZ ]; 3044 char *p_name; 3045 3046 t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL); 3047 if (!t) 3048 goto err; 3049 3050 for (i = 0; i < NEIGH_VAR_GC_INTERVAL; i++) { 3051 t->neigh_vars[i].data += (long) p; 3052 t->neigh_vars[i].extra1 = dev; 3053 t->neigh_vars[i].extra2 = p; 3054 } 3055 3056 if (dev) { 3057 dev_name_source = dev->name; 3058 /* Terminate the table early */ 3059 memset(&t->neigh_vars[NEIGH_VAR_GC_INTERVAL], 0, 3060 sizeof(t->neigh_vars[NEIGH_VAR_GC_INTERVAL])); 3061 } else { 3062 dev_name_source = "default"; 3063 t->neigh_vars[NEIGH_VAR_GC_INTERVAL].data = (int *)(p + 1); 3064 t->neigh_vars[NEIGH_VAR_GC_THRESH1].data = (int *)(p + 1) + 1; 3065 t->neigh_vars[NEIGH_VAR_GC_THRESH2].data = (int *)(p + 1) + 2; 3066 t->neigh_vars[NEIGH_VAR_GC_THRESH3].data = (int *)(p + 1) + 3; 3067 } 3068 3069 if (handler) { 3070 /* RetransTime */ 3071 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].proc_handler = handler; 3072 /* ReachableTime */ 3073 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler = handler; 3074 /* RetransTime (in milliseconds)*/ 3075 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].proc_handler = handler; 3076 /* ReachableTime (in milliseconds) */ 3077 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler = handler; 3078 } 3079 3080 /* Don't export sysctls to unprivileged users */ 3081 if (neigh_parms_net(p)->user_ns != &init_user_ns) 3082 t->neigh_vars[0].procname = NULL; 3083 3084 switch (neigh_parms_family(p)) { 3085 case AF_INET: 3086 p_name = "ipv4"; 3087 break; 3088 case AF_INET6: 3089 p_name = "ipv6"; 3090 break; 3091 default: 3092 BUG(); 3093 } 3094 3095 snprintf(neigh_path, sizeof(neigh_path), "net/%s/neigh/%s", 3096 p_name, dev_name_source); 3097 t->sysctl_header = 3098 register_net_sysctl(neigh_parms_net(p), neigh_path, t->neigh_vars); 3099 if (!t->sysctl_header) 3100 goto free; 3101 3102 p->sysctl_table = t; 3103 return 0; 3104 3105 free: 3106 kfree(t); 3107 err: 3108 return -ENOBUFS; 3109 } 3110 EXPORT_SYMBOL(neigh_sysctl_register); 3111 3112 void neigh_sysctl_unregister(struct neigh_parms *p) 3113 { 3114 if (p->sysctl_table) { 3115 struct neigh_sysctl_table *t = p->sysctl_table; 3116 p->sysctl_table = NULL; 3117 unregister_net_sysctl_table(t->sysctl_header); 3118 kfree(t); 3119 } 3120 } 3121 EXPORT_SYMBOL(neigh_sysctl_unregister); 3122 3123 #endif /* CONFIG_SYSCTL */ 3124 3125 static int __init neigh_init(void) 3126 { 3127 rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL, NULL); 3128 rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL, NULL); 3129 rtnl_register(PF_UNSPEC, RTM_GETNEIGH, NULL, neigh_dump_info, NULL); 3130 3131 rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info, 3132 NULL); 3133 rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL, NULL); 3134 3135 return 0; 3136 } 3137 3138 subsys_initcall(neigh_init); 3139 3140