Linux Kernel  3.7.1
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
ip6_fib.c
Go to the documentation of this file.
1 /*
2  * Linux INET6 implementation
3  * Forwarding Information Database
4  *
5  * Authors:
6  * Pedro Roque <[email protected]>
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 
14 /*
15  * Changes:
16  * Yuji SEKIYA @USAGI: Support default route on router node;
17  * remove ip6_null_entry from the top of
18  * routing table.
19  * Ville Nuorvala: Fixed routing subtrees.
20  */
21 
22 #define pr_fmt(fmt) "IPv6: " fmt
23 
24 #include <linux/errno.h>
25 #include <linux/types.h>
26 #include <linux/net.h>
27 #include <linux/route.h>
28 #include <linux/netdevice.h>
29 #include <linux/in6.h>
30 #include <linux/init.h>
31 #include <linux/list.h>
32 #include <linux/slab.h>
33 
34 #include <net/ipv6.h>
35 #include <net/ndisc.h>
36 #include <net/addrconf.h>
37 
38 #include <net/ip6_fib.h>
39 #include <net/ip6_route.h>
40 
41 #define RT6_DEBUG 2
42 
43 #if RT6_DEBUG >= 3
44 #define RT6_TRACE(x...) pr_debug(x)
45 #else
46 #define RT6_TRACE(x...) do { ; } while (0)
47 #endif
48 
49 static struct kmem_cache * fib6_node_kmem __read_mostly;
50 
52 {
53 #ifdef CONFIG_IPV6_SUBTREES
54  FWS_S,
55 #endif
60 };
61 
63 {
64  struct fib6_walker_t w;
65  struct net *net;
66  int (*func)(struct rt6_info *, void *arg);
67  void *arg;
68 };
69 
70 static DEFINE_RWLOCK(fib6_walker_lock);
71 
72 #ifdef CONFIG_IPV6_SUBTREES
73 #define FWS_INIT FWS_S
74 #else
75 #define FWS_INIT FWS_L
76 #endif
77 
78 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
79  struct rt6_info *rt);
80 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn);
81 static struct fib6_node *fib6_repair_tree(struct net *net, struct fib6_node *fn);
82 static int fib6_walk(struct fib6_walker_t *w);
83 static int fib6_walk_continue(struct fib6_walker_t *w);
84 
85 /*
86  * A routing update causes an increase of the serial number on the
87  * affected subtree. This allows for cached routes to be asynchronously
88  * tested when modifications are made to the destination cache as a
89  * result of redirects, path MTU changes, etc.
90  */
91 
92 static __u32 rt_sernum;
93 
94 static void fib6_gc_timer_cb(unsigned long arg);
95 
96 static LIST_HEAD(fib6_walkers);
97 #define FOR_WALKERS(w) list_for_each_entry(w, &fib6_walkers, lh)
98 
99 static inline void fib6_walker_link(struct fib6_walker_t *w)
100 {
101  write_lock_bh(&fib6_walker_lock);
102  list_add(&w->lh, &fib6_walkers);
103  write_unlock_bh(&fib6_walker_lock);
104 }
105 
106 static inline void fib6_walker_unlink(struct fib6_walker_t *w)
107 {
108  write_lock_bh(&fib6_walker_lock);
109  list_del(&w->lh);
110  write_unlock_bh(&fib6_walker_lock);
111 }
112 static __inline__ u32 fib6_new_sernum(void)
113 {
114  u32 n = ++rt_sernum;
115  if ((__s32)n <= 0)
116  rt_sernum = n = 1;
117  return n;
118 }
119 
120 /*
121  * Auxiliary address test functions for the radix tree.
122  *
123  * These assume a 32bit processor (although it will work on
124  * 64bit processors)
125  */
126 
127 /*
128  * test bit
129  */
130 #if defined(__LITTLE_ENDIAN)
131 # define BITOP_BE32_SWIZZLE (0x1F & ~7)
132 #else
133 # define BITOP_BE32_SWIZZLE 0
134 #endif
135 
136 static __inline__ __be32 addr_bit_set(const void *token, int fn_bit)
137 {
138  const __be32 *addr = token;
139  /*
140  * Here,
141  * 1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
142  * is optimized version of
143  * htonl(1 << ((~fn_bit)&0x1F))
144  * See include/asm-generic/bitops/le.h.
145  */
146  return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
147  addr[fn_bit >> 5];
148 }
149 
150 static __inline__ struct fib6_node * node_alloc(void)
151 {
152  struct fib6_node *fn;
153 
154  fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
155 
156  return fn;
157 }
158 
159 static __inline__ void node_free(struct fib6_node * fn)
160 {
161  kmem_cache_free(fib6_node_kmem, fn);
162 }
163 
164 static __inline__ void rt6_release(struct rt6_info *rt)
165 {
166  if (atomic_dec_and_test(&rt->rt6i_ref))
167  dst_free(&rt->dst);
168 }
169 
170 static void fib6_link_table(struct net *net, struct fib6_table *tb)
171 {
172  unsigned int h;
173 
174  /*
175  * Initialize table lock at a single place to give lockdep a key,
176  * tables aren't visible prior to being linked to the list.
177  */
178  rwlock_init(&tb->tb6_lock);
179 
180  h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
181 
182  /*
183  * No protection necessary, this is the only list mutatation
184  * operation, tables never disappear once they exist.
185  */
186  hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
187 }
188 
189 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
190 
191 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
192 {
193  struct fib6_table *table;
194 
195  table = kzalloc(sizeof(*table), GFP_ATOMIC);
196  if (table) {
197  table->tb6_id = id;
198  table->tb6_root.leaf = net->ipv6.ip6_null_entry;
199  table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
201  }
202 
203  return table;
204 }
205 
206 struct fib6_table *fib6_new_table(struct net *net, u32 id)
207 {
208  struct fib6_table *tb;
209 
210  if (id == 0)
211  id = RT6_TABLE_MAIN;
212  tb = fib6_get_table(net, id);
213  if (tb)
214  return tb;
215 
216  tb = fib6_alloc_table(net, id);
217  if (tb)
218  fib6_link_table(net, tb);
219 
220  return tb;
221 }
222 
223 struct fib6_table *fib6_get_table(struct net *net, u32 id)
224 {
225  struct fib6_table *tb;
226  struct hlist_head *head;
227  struct hlist_node *node;
228  unsigned int h;
229 
230  if (id == 0)
231  id = RT6_TABLE_MAIN;
232  h = id & (FIB6_TABLE_HASHSZ - 1);
233  rcu_read_lock();
234  head = &net->ipv6.fib_table_hash[h];
235  hlist_for_each_entry_rcu(tb, node, head, tb6_hlist) {
236  if (tb->tb6_id == id) {
237  rcu_read_unlock();
238  return tb;
239  }
240  }
241  rcu_read_unlock();
242 
243  return NULL;
244 }
245 
246 static void __net_init fib6_tables_init(struct net *net)
247 {
248  fib6_link_table(net, net->ipv6.fib6_main_tbl);
249  fib6_link_table(net, net->ipv6.fib6_local_tbl);
250 }
251 #else
252 
253 struct fib6_table *fib6_new_table(struct net *net, u32 id)
254 {
255  return fib6_get_table(net, id);
256 }
257 
258 struct fib6_table *fib6_get_table(struct net *net, u32 id)
259 {
260  return net->ipv6.fib6_main_tbl;
261 }
262 
263 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
265 {
266  return (struct dst_entry *) lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
267 }
268 
269 static void __net_init fib6_tables_init(struct net *net)
270 {
271  fib6_link_table(net, net->ipv6.fib6_main_tbl);
272 }
273 
274 #endif
275 
276 static int fib6_dump_node(struct fib6_walker_t *w)
277 {
278  int res;
279  struct rt6_info *rt;
280 
281  for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
282  res = rt6_dump_route(rt, w->args);
283  if (res < 0) {
284  /* Frame is full, suspend walking */
285  w->leaf = rt;
286  return 1;
287  }
288  WARN_ON(res == 0);
289  }
290  w->leaf = NULL;
291  return 0;
292 }
293 
294 static void fib6_dump_end(struct netlink_callback *cb)
295 {
296  struct fib6_walker_t *w = (void*)cb->args[2];
297 
298  if (w) {
299  if (cb->args[4]) {
300  cb->args[4] = 0;
301  fib6_walker_unlink(w);
302  }
303  cb->args[2] = 0;
304  kfree(w);
305  }
306  cb->done = (void*)cb->args[3];
307  cb->args[1] = 3;
308 }
309 
310 static int fib6_dump_done(struct netlink_callback *cb)
311 {
312  fib6_dump_end(cb);
313  return cb->done ? cb->done(cb) : 0;
314 }
315 
316 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
317  struct netlink_callback *cb)
318 {
319  struct fib6_walker_t *w;
320  int res;
321 
322  w = (void *)cb->args[2];
323  w->root = &table->tb6_root;
324 
325  if (cb->args[4] == 0) {
326  w->count = 0;
327  w->skip = 0;
328 
329  read_lock_bh(&table->tb6_lock);
330  res = fib6_walk(w);
331  read_unlock_bh(&table->tb6_lock);
332  if (res > 0) {
333  cb->args[4] = 1;
334  cb->args[5] = w->root->fn_sernum;
335  }
336  } else {
337  if (cb->args[5] != w->root->fn_sernum) {
338  /* Begin at the root if the tree changed */
339  cb->args[5] = w->root->fn_sernum;
340  w->state = FWS_INIT;
341  w->node = w->root;
342  w->skip = w->count;
343  } else
344  w->skip = 0;
345 
346  read_lock_bh(&table->tb6_lock);
347  res = fib6_walk_continue(w);
348  read_unlock_bh(&table->tb6_lock);
349  if (res <= 0) {
350  fib6_walker_unlink(w);
351  cb->args[4] = 0;
352  }
353  }
354 
355  return res;
356 }
357 
358 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
359 {
360  struct net *net = sock_net(skb->sk);
361  unsigned int h, s_h;
362  unsigned int e = 0, s_e;
363  struct rt6_rtnl_dump_arg arg;
364  struct fib6_walker_t *w;
365  struct fib6_table *tb;
366  struct hlist_node *node;
367  struct hlist_head *head;
368  int res = 0;
369 
370  s_h = cb->args[0];
371  s_e = cb->args[1];
372 
373  w = (void *)cb->args[2];
374  if (!w) {
375  /* New dump:
376  *
377  * 1. hook callback destructor.
378  */
379  cb->args[3] = (long)cb->done;
380  cb->done = fib6_dump_done;
381 
382  /*
383  * 2. allocate and initialize walker.
384  */
385  w = kzalloc(sizeof(*w), GFP_ATOMIC);
386  if (!w)
387  return -ENOMEM;
388  w->func = fib6_dump_node;
389  cb->args[2] = (long)w;
390  }
391 
392  arg.skb = skb;
393  arg.cb = cb;
394  arg.net = net;
395  w->args = &arg;
396 
397  rcu_read_lock();
398  for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
399  e = 0;
400  head = &net->ipv6.fib_table_hash[h];
401  hlist_for_each_entry_rcu(tb, node, head, tb6_hlist) {
402  if (e < s_e)
403  goto next;
404  res = fib6_dump_table(tb, skb, cb);
405  if (res != 0)
406  goto out;
407 next:
408  e++;
409  }
410  }
411 out:
412  rcu_read_unlock();
413  cb->args[1] = e;
414  cb->args[0] = h;
415 
416  res = res < 0 ? res : skb->len;
417  if (res <= 0)
418  fib6_dump_end(cb);
419  return res;
420 }
421 
422 /*
423  * Routing Table
424  *
425  * return the appropriate node for a routing tree "add" operation
426  * by either creating and inserting or by returning an existing
427  * node.
428  */
429 
430 static struct fib6_node * fib6_add_1(struct fib6_node *root, void *addr,
431  int addrlen, int plen,
432  int offset, int allow_create,
433  int replace_required)
434 {
435  struct fib6_node *fn, *in, *ln;
436  struct fib6_node *pn = NULL;
437  struct rt6key *key;
438  int bit;
439  __be32 dir = 0;
440  __u32 sernum = fib6_new_sernum();
441 
442  RT6_TRACE("fib6_add_1\n");
443 
444  /* insert node in tree */
445 
446  fn = root;
447 
448  do {
449  key = (struct rt6key *)((u8 *)fn->leaf + offset);
450 
451  /*
452  * Prefix match
453  */
454  if (plen < fn->fn_bit ||
455  !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
456  if (!allow_create) {
457  if (replace_required) {
458  pr_warn("Can't replace route, no match found\n");
459  return ERR_PTR(-ENOENT);
460  }
461  pr_warn("NLM_F_CREATE should be set when creating new route\n");
462  }
463  goto insert_above;
464  }
465 
466  /*
467  * Exact match ?
468  */
469 
470  if (plen == fn->fn_bit) {
471  /* clean up an intermediate node */
472  if (!(fn->fn_flags & RTN_RTINFO)) {
473  rt6_release(fn->leaf);
474  fn->leaf = NULL;
475  }
476 
477  fn->fn_sernum = sernum;
478 
479  return fn;
480  }
481 
482  /*
483  * We have more bits to go
484  */
485 
486  /* Try to walk down on tree. */
487  fn->fn_sernum = sernum;
488  dir = addr_bit_set(addr, fn->fn_bit);
489  pn = fn;
490  fn = dir ? fn->right: fn->left;
491  } while (fn);
492 
493  if (!allow_create) {
494  /* We should not create new node because
495  * NLM_F_REPLACE was specified without NLM_F_CREATE
496  * I assume it is safe to require NLM_F_CREATE when
497  * REPLACE flag is used! Later we may want to remove the
498  * check for replace_required, because according
499  * to netlink specification, NLM_F_CREATE
500  * MUST be specified if new route is created.
501  * That would keep IPv6 consistent with IPv4
502  */
503  if (replace_required) {
504  pr_warn("Can't replace route, no match found\n");
505  return ERR_PTR(-ENOENT);
506  }
507  pr_warn("NLM_F_CREATE should be set when creating new route\n");
508  }
509  /*
510  * We walked to the bottom of tree.
511  * Create new leaf node without children.
512  */
513 
514  ln = node_alloc();
515 
516  if (!ln)
517  return ERR_PTR(-ENOMEM);
518  ln->fn_bit = plen;
519 
520  ln->parent = pn;
521  ln->fn_sernum = sernum;
522 
523  if (dir)
524  pn->right = ln;
525  else
526  pn->left = ln;
527 
528  return ln;
529 
530 
531 insert_above:
532  /*
533  * split since we don't have a common prefix anymore or
534  * we have a less significant route.
535  * we've to insert an intermediate node on the list
536  * this new node will point to the one we need to create
537  * and the current
538  */
539 
540  pn = fn->parent;
541 
542  /* find 1st bit in difference between the 2 addrs.
543 
544  See comment in __ipv6_addr_diff: bit may be an invalid value,
545  but if it is >= plen, the value is ignored in any case.
546  */
547 
548  bit = __ipv6_addr_diff(addr, &key->addr, addrlen);
549 
550  /*
551  * (intermediate)[in]
552  * / \
553  * (new leaf node)[ln] (old node)[fn]
554  */
555  if (plen > bit) {
556  in = node_alloc();
557  ln = node_alloc();
558 
559  if (!in || !ln) {
560  if (in)
561  node_free(in);
562  if (ln)
563  node_free(ln);
564  return ERR_PTR(-ENOMEM);
565  }
566 
567  /*
568  * new intermediate node.
569  * RTN_RTINFO will
570  * be off since that an address that chooses one of
571  * the branches would not match less specific routes
572  * in the other branch
573  */
574 
575  in->fn_bit = bit;
576 
577  in->parent = pn;
578  in->leaf = fn->leaf;
579  atomic_inc(&in->leaf->rt6i_ref);
580 
581  in->fn_sernum = sernum;
582 
583  /* update parent pointer */
584  if (dir)
585  pn->right = in;
586  else
587  pn->left = in;
588 
589  ln->fn_bit = plen;
590 
591  ln->parent = in;
592  fn->parent = in;
593 
594  ln->fn_sernum = sernum;
595 
596  if (addr_bit_set(addr, bit)) {
597  in->right = ln;
598  in->left = fn;
599  } else {
600  in->left = ln;
601  in->right = fn;
602  }
603  } else { /* plen <= bit */
604 
605  /*
606  * (new leaf node)[ln]
607  * / \
608  * (old node)[fn] NULL
609  */
610 
611  ln = node_alloc();
612 
613  if (!ln)
614  return ERR_PTR(-ENOMEM);
615 
616  ln->fn_bit = plen;
617 
618  ln->parent = pn;
619 
620  ln->fn_sernum = sernum;
621 
622  if (dir)
623  pn->right = ln;
624  else
625  pn->left = ln;
626 
627  if (addr_bit_set(&key->addr, plen))
628  ln->right = fn;
629  else
630  ln->left = fn;
631 
632  fn->parent = ln;
633  }
634  return ln;
635 }
636 
637 /*
638  * Insert routing information in a node.
639  */
640 
641 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
642  struct nl_info *info)
643 {
644  struct rt6_info *iter = NULL;
645  struct rt6_info **ins;
646  int replace = (info->nlh &&
647  (info->nlh->nlmsg_flags & NLM_F_REPLACE));
648  int add = (!info->nlh ||
649  (info->nlh->nlmsg_flags & NLM_F_CREATE));
650  int found = 0;
651 
652  ins = &fn->leaf;
653 
654  for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) {
655  /*
656  * Search for duplicates
657  */
658 
659  if (iter->rt6i_metric == rt->rt6i_metric) {
660  /*
661  * Same priority level
662  */
663  if (info->nlh &&
664  (info->nlh->nlmsg_flags & NLM_F_EXCL))
665  return -EEXIST;
666  if (replace) {
667  found++;
668  break;
669  }
670 
671  if (iter->dst.dev == rt->dst.dev &&
672  iter->rt6i_idev == rt->rt6i_idev &&
673  ipv6_addr_equal(&iter->rt6i_gateway,
674  &rt->rt6i_gateway)) {
675  if (!(iter->rt6i_flags & RTF_EXPIRES))
676  return -EEXIST;
677  if (!(rt->rt6i_flags & RTF_EXPIRES))
678  rt6_clean_expires(iter);
679  else
680  rt6_set_expires(iter, rt->dst.expires);
681  return -EEXIST;
682  }
683  }
684 
685  if (iter->rt6i_metric > rt->rt6i_metric)
686  break;
687 
688  ins = &iter->dst.rt6_next;
689  }
690 
691  /* Reset round-robin state, if necessary */
692  if (ins == &fn->leaf)
693  fn->rr_ptr = NULL;
694 
695  /*
696  * insert node
697  */
698  if (!replace) {
699  if (!add)
700  pr_warn("NLM_F_CREATE should be set when creating new route\n");
701 
702 add:
703  rt->dst.rt6_next = iter;
704  *ins = rt;
705  rt->rt6i_node = fn;
706  atomic_inc(&rt->rt6i_ref);
707  inet6_rt_notify(RTM_NEWROUTE, rt, info);
708  info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
709 
710  if (!(fn->fn_flags & RTN_RTINFO)) {
711  info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
712  fn->fn_flags |= RTN_RTINFO;
713  }
714 
715  } else {
716  if (!found) {
717  if (add)
718  goto add;
719  pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
720  return -ENOENT;
721  }
722  *ins = rt;
723  rt->rt6i_node = fn;
724  rt->dst.rt6_next = iter->dst.rt6_next;
725  atomic_inc(&rt->rt6i_ref);
726  inet6_rt_notify(RTM_NEWROUTE, rt, info);
727  rt6_release(iter);
728  if (!(fn->fn_flags & RTN_RTINFO)) {
729  info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
730  fn->fn_flags |= RTN_RTINFO;
731  }
732  }
733 
734  return 0;
735 }
736 
737 static __inline__ void fib6_start_gc(struct net *net, struct rt6_info *rt)
738 {
739  if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
740  (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
741  mod_timer(&net->ipv6.ip6_fib_timer,
742  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
743 }
744 
745 void fib6_force_start_gc(struct net *net)
746 {
747  if (!timer_pending(&net->ipv6.ip6_fib_timer))
748  mod_timer(&net->ipv6.ip6_fib_timer,
749  jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
750 }
751 
752 /*
753  * Add routing information to the routing tree.
754  * <destination addr>/<source addr>
755  * with source addr info in sub-trees
756  */
757 
758 int fib6_add(struct fib6_node *root, struct rt6_info *rt, struct nl_info *info)
759 {
760  struct fib6_node *fn, *pn = NULL;
761  int err = -ENOMEM;
762  int allow_create = 1;
763  int replace_required = 0;
764 
765  if (info->nlh) {
766  if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
767  allow_create = 0;
768  if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
769  replace_required = 1;
770  }
771  if (!allow_create && !replace_required)
772  pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
773 
774  fn = fib6_add_1(root, &rt->rt6i_dst.addr, sizeof(struct in6_addr),
775  rt->rt6i_dst.plen, offsetof(struct rt6_info, rt6i_dst),
776  allow_create, replace_required);
777 
778  if (IS_ERR(fn)) {
779  err = PTR_ERR(fn);
780  goto out;
781  }
782 
783  pn = fn;
784 
785 #ifdef CONFIG_IPV6_SUBTREES
786  if (rt->rt6i_src.plen) {
787  struct fib6_node *sn;
788 
789  if (!fn->subtree) {
790  struct fib6_node *sfn;
791 
792  /*
793  * Create subtree.
794  *
795  * fn[main tree]
796  * |
797  * sfn[subtree root]
798  * \
799  * sn[new leaf node]
800  */
801 
802  /* Create subtree root node */
803  sfn = node_alloc();
804  if (!sfn)
805  goto st_failure;
806 
807  sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
808  atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
809  sfn->fn_flags = RTN_ROOT;
810  sfn->fn_sernum = fib6_new_sernum();
811 
812  /* Now add the first leaf node to new subtree */
813 
814  sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
815  sizeof(struct in6_addr), rt->rt6i_src.plen,
816  offsetof(struct rt6_info, rt6i_src),
817  allow_create, replace_required);
818 
819  if (IS_ERR(sn)) {
820  /* If it is failed, discard just allocated
821  root, and then (in st_failure) stale node
822  in main tree.
823  */
824  node_free(sfn);
825  err = PTR_ERR(sn);
826  goto st_failure;
827  }
828 
829  /* Now link new subtree to main tree */
830  sfn->parent = fn;
831  fn->subtree = sfn;
832  } else {
833  sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
834  sizeof(struct in6_addr), rt->rt6i_src.plen,
835  offsetof(struct rt6_info, rt6i_src),
836  allow_create, replace_required);
837 
838  if (IS_ERR(sn)) {
839  err = PTR_ERR(sn);
840  goto st_failure;
841  }
842  }
843 
844  if (!fn->leaf) {
845  fn->leaf = rt;
846  atomic_inc(&rt->rt6i_ref);
847  }
848  fn = sn;
849  }
850 #endif
851 
852  err = fib6_add_rt2node(fn, rt, info);
853  if (!err) {
854  fib6_start_gc(info->nl_net, rt);
855  if (!(rt->rt6i_flags & RTF_CACHE))
856  fib6_prune_clones(info->nl_net, pn, rt);
857  }
858 
859 out:
860  if (err) {
861 #ifdef CONFIG_IPV6_SUBTREES
862  /*
863  * If fib6_add_1 has cleared the old leaf pointer in the
864  * super-tree leaf node we have to find a new one for it.
865  */
866  if (pn != fn && pn->leaf == rt) {
867  pn->leaf = NULL;
868  atomic_dec(&rt->rt6i_ref);
869  }
870  if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
871  pn->leaf = fib6_find_prefix(info->nl_net, pn);
872 #if RT6_DEBUG >= 2
873  if (!pn->leaf) {
874  WARN_ON(pn->leaf == NULL);
875  pn->leaf = info->nl_net->ipv6.ip6_null_entry;
876  }
877 #endif
878  atomic_inc(&pn->leaf->rt6i_ref);
879  }
880 #endif
881  dst_free(&rt->dst);
882  }
883  return err;
884 
885 #ifdef CONFIG_IPV6_SUBTREES
886  /* Subtree creation failed, probably main tree node
887  is orphan. If it is, shoot it.
888  */
889 st_failure:
890  if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
891  fib6_repair_tree(info->nl_net, fn);
892  dst_free(&rt->dst);
893  return err;
894 #endif
895 }
896 
897 /*
898  * Routing tree lookup
899  *
900  */
901 
902 struct lookup_args {
903  int offset; /* key offset on rt6_info */
904  const struct in6_addr *addr; /* search key */
905 };
906 
907 static struct fib6_node * fib6_lookup_1(struct fib6_node *root,
908  struct lookup_args *args)
909 {
910  struct fib6_node *fn;
911  __be32 dir;
912 
913  if (unlikely(args->offset == 0))
914  return NULL;
915 
916  /*
917  * Descend on a tree
918  */
919 
920  fn = root;
921 
922  for (;;) {
923  struct fib6_node *next;
924 
925  dir = addr_bit_set(args->addr, fn->fn_bit);
926 
927  next = dir ? fn->right : fn->left;
928 
929  if (next) {
930  fn = next;
931  continue;
932  }
933  break;
934  }
935 
936  while (fn) {
937  if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
938  struct rt6key *key;
939 
940  key = (struct rt6key *) ((u8 *) fn->leaf +
941  args->offset);
942 
943  if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
944 #ifdef CONFIG_IPV6_SUBTREES
945  if (fn->subtree)
946  fn = fib6_lookup_1(fn->subtree, args + 1);
947 #endif
948  if (!fn || fn->fn_flags & RTN_RTINFO)
949  return fn;
950  }
951  }
952 
953  if (fn->fn_flags & RTN_ROOT)
954  break;
955 
956  fn = fn->parent;
957  }
958 
959  return NULL;
960 }
961 
962 struct fib6_node * fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
963  const struct in6_addr *saddr)
964 {
965  struct fib6_node *fn;
966  struct lookup_args args[] = {
967  {
968  .offset = offsetof(struct rt6_info, rt6i_dst),
969  .addr = daddr,
970  },
971 #ifdef CONFIG_IPV6_SUBTREES
972  {
973  .offset = offsetof(struct rt6_info, rt6i_src),
974  .addr = saddr,
975  },
976 #endif
977  {
978  .offset = 0, /* sentinel */
979  }
980  };
981 
982  fn = fib6_lookup_1(root, daddr ? args : args + 1);
983  if (!fn || fn->fn_flags & RTN_TL_ROOT)
984  fn = root;
985 
986  return fn;
987 }
988 
989 /*
990  * Get node with specified destination prefix (and source prefix,
991  * if subtrees are used)
992  */
993 
994 
995 static struct fib6_node * fib6_locate_1(struct fib6_node *root,
996  const struct in6_addr *addr,
997  int plen, int offset)
998 {
999  struct fib6_node *fn;
1000 
1001  for (fn = root; fn ; ) {
1002  struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
1003 
1004  /*
1005  * Prefix match
1006  */
1007  if (plen < fn->fn_bit ||
1008  !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1009  return NULL;
1010 
1011  if (plen == fn->fn_bit)
1012  return fn;
1013 
1014  /*
1015  * We have more bits to go
1016  */
1017  if (addr_bit_set(addr, fn->fn_bit))
1018  fn = fn->right;
1019  else
1020  fn = fn->left;
1021  }
1022  return NULL;
1023 }
1024 
1025 struct fib6_node * fib6_locate(struct fib6_node *root,
1026  const struct in6_addr *daddr, int dst_len,
1027  const struct in6_addr *saddr, int src_len)
1028 {
1029  struct fib6_node *fn;
1030 
1031  fn = fib6_locate_1(root, daddr, dst_len,
1032  offsetof(struct rt6_info, rt6i_dst));
1033 
1034 #ifdef CONFIG_IPV6_SUBTREES
1035  if (src_len) {
1036  WARN_ON(saddr == NULL);
1037  if (fn && fn->subtree)
1038  fn = fib6_locate_1(fn->subtree, saddr, src_len,
1039  offsetof(struct rt6_info, rt6i_src));
1040  }
1041 #endif
1042 
1043  if (fn && fn->fn_flags & RTN_RTINFO)
1044  return fn;
1045 
1046  return NULL;
1047 }
1048 
1049 
1050 /*
1051  * Deletion
1052  *
1053  */
1054 
1055 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
1056 {
1057  if (fn->fn_flags & RTN_ROOT)
1058  return net->ipv6.ip6_null_entry;
1059 
1060  while (fn) {
1061  if (fn->left)
1062  return fn->left->leaf;
1063  if (fn->right)
1064  return fn->right->leaf;
1065 
1066  fn = FIB6_SUBTREE(fn);
1067  }
1068  return NULL;
1069 }
1070 
1071 /*
1072  * Called to trim the tree of intermediate nodes when possible. "fn"
1073  * is the node we want to try and remove.
1074  */
1075 
1076 static struct fib6_node *fib6_repair_tree(struct net *net,
1077  struct fib6_node *fn)
1078 {
1079  int children;
1080  int nstate;
1081  struct fib6_node *child, *pn;
1082  struct fib6_walker_t *w;
1083  int iter = 0;
1084 
1085  for (;;) {
1086  RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1087  iter++;
1088 
1089  WARN_ON(fn->fn_flags & RTN_RTINFO);
1090  WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1091  WARN_ON(fn->leaf != NULL);
1092 
1093  children = 0;
1094  child = NULL;
1095  if (fn->right) child = fn->right, children |= 1;
1096  if (fn->left) child = fn->left, children |= 2;
1097 
1098  if (children == 3 || FIB6_SUBTREE(fn)
1099 #ifdef CONFIG_IPV6_SUBTREES
1100  /* Subtree root (i.e. fn) may have one child */
1101  || (children && fn->fn_flags & RTN_ROOT)
1102 #endif
1103  ) {
1104  fn->leaf = fib6_find_prefix(net, fn);
1105 #if RT6_DEBUG >= 2
1106  if (!fn->leaf) {
1107  WARN_ON(!fn->leaf);
1108  fn->leaf = net->ipv6.ip6_null_entry;
1109  }
1110 #endif
1111  atomic_inc(&fn->leaf->rt6i_ref);
1112  return fn->parent;
1113  }
1114 
1115  pn = fn->parent;
1116 #ifdef CONFIG_IPV6_SUBTREES
1117  if (FIB6_SUBTREE(pn) == fn) {
1118  WARN_ON(!(fn->fn_flags & RTN_ROOT));
1119  FIB6_SUBTREE(pn) = NULL;
1120  nstate = FWS_L;
1121  } else {
1122  WARN_ON(fn->fn_flags & RTN_ROOT);
1123 #endif
1124  if (pn->right == fn) pn->right = child;
1125  else if (pn->left == fn) pn->left = child;
1126 #if RT6_DEBUG >= 2
1127  else
1128  WARN_ON(1);
1129 #endif
1130  if (child)
1131  child->parent = pn;
1132  nstate = FWS_R;
1133 #ifdef CONFIG_IPV6_SUBTREES
1134  }
1135 #endif
1136 
1137  read_lock(&fib6_walker_lock);
1138  FOR_WALKERS(w) {
1139  if (!child) {
1140  if (w->root == fn) {
1141  w->root = w->node = NULL;
1142  RT6_TRACE("W %p adjusted by delroot 1\n", w);
1143  } else if (w->node == fn) {
1144  RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1145  w->node = pn;
1146  w->state = nstate;
1147  }
1148  } else {
1149  if (w->root == fn) {
1150  w->root = child;
1151  RT6_TRACE("W %p adjusted by delroot 2\n", w);
1152  }
1153  if (w->node == fn) {
1154  w->node = child;
1155  if (children&2) {
1156  RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1157  w->state = w->state>=FWS_R ? FWS_U : FWS_INIT;
1158  } else {
1159  RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1160  w->state = w->state>=FWS_C ? FWS_U : FWS_INIT;
1161  }
1162  }
1163  }
1164  }
1165  read_unlock(&fib6_walker_lock);
1166 
1167  node_free(fn);
1168  if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1169  return pn;
1170 
1171  rt6_release(pn->leaf);
1172  pn->leaf = NULL;
1173  fn = pn;
1174  }
1175 }
1176 
1177 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1178  struct nl_info *info)
1179 {
1180  struct fib6_walker_t *w;
1181  struct rt6_info *rt = *rtp;
1182  struct net *net = info->nl_net;
1183 
1184  RT6_TRACE("fib6_del_route\n");
1185 
1186  /* Unlink it */
1187  *rtp = rt->dst.rt6_next;
1188  rt->rt6i_node = NULL;
1189  net->ipv6.rt6_stats->fib_rt_entries--;
1190  net->ipv6.rt6_stats->fib_discarded_routes++;
1191 
1192  /* Reset round-robin state, if necessary */
1193  if (fn->rr_ptr == rt)
1194  fn->rr_ptr = NULL;
1195 
1196  /* Adjust walkers */
1197  read_lock(&fib6_walker_lock);
1198  FOR_WALKERS(w) {
1199  if (w->state == FWS_C && w->leaf == rt) {
1200  RT6_TRACE("walker %p adjusted by delroute\n", w);
1201  w->leaf = rt->dst.rt6_next;
1202  if (!w->leaf)
1203  w->state = FWS_U;
1204  }
1205  }
1206  read_unlock(&fib6_walker_lock);
1207 
1208  rt->dst.rt6_next = NULL;
1209 
1210  /* If it was last route, expunge its radix tree node */
1211  if (!fn->leaf) {
1212  fn->fn_flags &= ~RTN_RTINFO;
1213  net->ipv6.rt6_stats->fib_route_nodes--;
1214  fn = fib6_repair_tree(net, fn);
1215  }
1216 
1217  if (atomic_read(&rt->rt6i_ref) != 1) {
1218  /* This route is used as dummy address holder in some split
1219  * nodes. It is not leaked, but it still holds other resources,
1220  * which must be released in time. So, scan ascendant nodes
1221  * and replace dummy references to this route with references
1222  * to still alive ones.
1223  */
1224  while (fn) {
1225  if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) {
1226  fn->leaf = fib6_find_prefix(net, fn);
1227  atomic_inc(&fn->leaf->rt6i_ref);
1228  rt6_release(rt);
1229  }
1230  fn = fn->parent;
1231  }
1232  /* No more references are possible at this point. */
1233  BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
1234  }
1235 
1236  inet6_rt_notify(RTM_DELROUTE, rt, info);
1237  rt6_release(rt);
1238 }
1239 
1240 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1241 {
1242  struct net *net = info->nl_net;
1243  struct fib6_node *fn = rt->rt6i_node;
1244  struct rt6_info **rtp;
1245 
1246 #if RT6_DEBUG >= 2
1247  if (rt->dst.obsolete>0) {
1248  WARN_ON(fn != NULL);
1249  return -ENOENT;
1250  }
1251 #endif
1252  if (!fn || rt == net->ipv6.ip6_null_entry)
1253  return -ENOENT;
1254 
1255  WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1256 
1257  if (!(rt->rt6i_flags & RTF_CACHE)) {
1258  struct fib6_node *pn = fn;
1259 #ifdef CONFIG_IPV6_SUBTREES
1260  /* clones of this route might be in another subtree */
1261  if (rt->rt6i_src.plen) {
1262  while (!(pn->fn_flags & RTN_ROOT))
1263  pn = pn->parent;
1264  pn = pn->parent;
1265  }
1266 #endif
1267  fib6_prune_clones(info->nl_net, pn, rt);
1268  }
1269 
1270  /*
1271  * Walk the leaf entries looking for ourself
1272  */
1273 
1274  for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1275  if (*rtp == rt) {
1276  fib6_del_route(fn, rtp, info);
1277  return 0;
1278  }
1279  }
1280  return -ENOENT;
1281 }
1282 
1283 /*
1284  * Tree traversal function.
1285  *
1286  * Certainly, it is not interrupt safe.
1287  * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1288  * It means, that we can modify tree during walking
1289  * and use this function for garbage collection, clone pruning,
1290  * cleaning tree when a device goes down etc. etc.
1291  *
1292  * It guarantees that every node will be traversed,
1293  * and that it will be traversed only once.
1294  *
1295  * Callback function w->func may return:
1296  * 0 -> continue walking.
1297  * positive value -> walking is suspended (used by tree dumps,
1298  * and probably by gc, if it will be split to several slices)
1299  * negative value -> terminate walking.
1300  *
1301  * The function itself returns:
1302  * 0 -> walk is complete.
1303  * >0 -> walk is incomplete (i.e. suspended)
1304  * <0 -> walk is terminated by an error.
1305  */
1306 
1307 static int fib6_walk_continue(struct fib6_walker_t *w)
1308 {
1309  struct fib6_node *fn, *pn;
1310 
1311  for (;;) {
1312  fn = w->node;
1313  if (!fn)
1314  return 0;
1315 
1316  if (w->prune && fn != w->root &&
1317  fn->fn_flags & RTN_RTINFO && w->state < FWS_C) {
1318  w->state = FWS_C;
1319  w->leaf = fn->leaf;
1320  }
1321  switch (w->state) {
1322 #ifdef CONFIG_IPV6_SUBTREES
1323  case FWS_S:
1324  if (FIB6_SUBTREE(fn)) {
1325  w->node = FIB6_SUBTREE(fn);
1326  continue;
1327  }
1328  w->state = FWS_L;
1329 #endif
1330  case FWS_L:
1331  if (fn->left) {
1332  w->node = fn->left;
1333  w->state = FWS_INIT;
1334  continue;
1335  }
1336  w->state = FWS_R;
1337  case FWS_R:
1338  if (fn->right) {
1339  w->node = fn->right;
1340  w->state = FWS_INIT;
1341  continue;
1342  }
1343  w->state = FWS_C;
1344  w->leaf = fn->leaf;
1345  case FWS_C:
1346  if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1347  int err;
1348 
1349  if (w->skip) {
1350  w->skip--;
1351  continue;
1352  }
1353 
1354  err = w->func(w);
1355  if (err)
1356  return err;
1357 
1358  w->count++;
1359  continue;
1360  }
1361  w->state = FWS_U;
1362  case FWS_U:
1363  if (fn == w->root)
1364  return 0;
1365  pn = fn->parent;
1366  w->node = pn;
1367 #ifdef CONFIG_IPV6_SUBTREES
1368  if (FIB6_SUBTREE(pn) == fn) {
1369  WARN_ON(!(fn->fn_flags & RTN_ROOT));
1370  w->state = FWS_L;
1371  continue;
1372  }
1373 #endif
1374  if (pn->left == fn) {
1375  w->state = FWS_R;
1376  continue;
1377  }
1378  if (pn->right == fn) {
1379  w->state = FWS_C;
1380  w->leaf = w->node->leaf;
1381  continue;
1382  }
1383 #if RT6_DEBUG >= 2
1384  WARN_ON(1);
1385 #endif
1386  }
1387  }
1388 }
1389 
1390 static int fib6_walk(struct fib6_walker_t *w)
1391 {
1392  int res;
1393 
1394  w->state = FWS_INIT;
1395  w->node = w->root;
1396 
1397  fib6_walker_link(w);
1398  res = fib6_walk_continue(w);
1399  if (res <= 0)
1400  fib6_walker_unlink(w);
1401  return res;
1402 }
1403 
1404 static int fib6_clean_node(struct fib6_walker_t *w)
1405 {
1406  int res;
1407  struct rt6_info *rt;
1408  struct fib6_cleaner_t *c = container_of(w, struct fib6_cleaner_t, w);
1409  struct nl_info info = {
1410  .nl_net = c->net,
1411  };
1412 
1413  for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1414  res = c->func(rt, c->arg);
1415  if (res < 0) {
1416  w->leaf = rt;
1417  res = fib6_del(rt, &info);
1418  if (res) {
1419 #if RT6_DEBUG >= 2
1420  pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1421  __func__, rt, rt->rt6i_node, res);
1422 #endif
1423  continue;
1424  }
1425  return 0;
1426  }
1427  WARN_ON(res != 0);
1428  }
1429  w->leaf = rt;
1430  return 0;
1431 }
1432 
1433 /*
1434  * Convenient frontend to tree walker.
1435  *
1436  * func is called on each route.
1437  * It may return -1 -> delete this route.
1438  * 0 -> continue walking
1439  *
1440  * prune==1 -> only immediate children of node (certainly,
1441  * ignoring pure split nodes) will be scanned.
1442  */
1443 
1444 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1445  int (*func)(struct rt6_info *, void *arg),
1446  int prune, void *arg)
1447 {
1448  struct fib6_cleaner_t c;
1449 
1450  c.w.root = root;
1451  c.w.func = fib6_clean_node;
1452  c.w.prune = prune;
1453  c.w.count = 0;
1454  c.w.skip = 0;
1455  c.func = func;
1456  c.arg = arg;
1457  c.net = net;
1458 
1459  fib6_walk(&c.w);
1460 }
1461 
1462 void fib6_clean_all_ro(struct net *net, int (*func)(struct rt6_info *, void *arg),
1463  int prune, void *arg)
1464 {
1465  struct fib6_table *table;
1466  struct hlist_node *node;
1467  struct hlist_head *head;
1468  unsigned int h;
1469 
1470  rcu_read_lock();
1471  for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1472  head = &net->ipv6.fib_table_hash[h];
1473  hlist_for_each_entry_rcu(table, node, head, tb6_hlist) {
1474  read_lock_bh(&table->tb6_lock);
1475  fib6_clean_tree(net, &table->tb6_root,
1476  func, prune, arg);
1477  read_unlock_bh(&table->tb6_lock);
1478  }
1479  }
1480  rcu_read_unlock();
1481 }
1482 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *arg),
1483  int prune, void *arg)
1484 {
1485  struct fib6_table *table;
1486  struct hlist_node *node;
1487  struct hlist_head *head;
1488  unsigned int h;
1489 
1490  rcu_read_lock();
1491  for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1492  head = &net->ipv6.fib_table_hash[h];
1493  hlist_for_each_entry_rcu(table, node, head, tb6_hlist) {
1494  write_lock_bh(&table->tb6_lock);
1495  fib6_clean_tree(net, &table->tb6_root,
1496  func, prune, arg);
1497  write_unlock_bh(&table->tb6_lock);
1498  }
1499  }
1500  rcu_read_unlock();
1501 }
1502 
1503 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1504 {
1505  if (rt->rt6i_flags & RTF_CACHE) {
1506  RT6_TRACE("pruning clone %p\n", rt);
1507  return -1;
1508  }
1509 
1510  return 0;
1511 }
1512 
1513 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
1514  struct rt6_info *rt)
1515 {
1516  fib6_clean_tree(net, fn, fib6_prune_clone, 1, rt);
1517 }
1518 
1519 /*
1520  * Garbage collection
1521  */
1522 
1523 static struct fib6_gc_args
1524 {
1525  int timeout;
1526  int more;
1527 } gc_args;
1528 
1529 static int fib6_age(struct rt6_info *rt, void *arg)
1530 {
1531  unsigned long now = jiffies;
1532 
1533  /*
1534  * check addrconf expiration here.
1535  * Routes are expired even if they are in use.
1536  *
1537  * Also age clones. Note, that clones are aged out
1538  * only if they are not in use now.
1539  */
1540 
1541  if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
1542  if (time_after(now, rt->dst.expires)) {
1543  RT6_TRACE("expiring %p\n", rt);
1544  return -1;
1545  }
1546  gc_args.more++;
1547  } else if (rt->rt6i_flags & RTF_CACHE) {
1548  if (atomic_read(&rt->dst.__refcnt) == 0 &&
1549  time_after_eq(now, rt->dst.lastuse + gc_args.timeout)) {
1550  RT6_TRACE("aging clone %p\n", rt);
1551  return -1;
1552  } else if (rt->rt6i_flags & RTF_GATEWAY) {
1553  struct neighbour *neigh;
1554  __u8 neigh_flags = 0;
1555 
1556  neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway);
1557  if (neigh) {
1558  neigh_flags = neigh->flags;
1559  neigh_release(neigh);
1560  }
1561  if (!(neigh_flags & NTF_ROUTER)) {
1562  RT6_TRACE("purging route %p via non-router but gateway\n",
1563  rt);
1564  return -1;
1565  }
1566  }
1567  gc_args.more++;
1568  }
1569 
1570  return 0;
1571 }
1572 
1573 static DEFINE_SPINLOCK(fib6_gc_lock);
1574 
1575 void fib6_run_gc(unsigned long expires, struct net *net)
1576 {
1577  if (expires != ~0UL) {
1578  spin_lock_bh(&fib6_gc_lock);
1579  gc_args.timeout = expires ? (int)expires :
1580  net->ipv6.sysctl.ip6_rt_gc_interval;
1581  } else {
1582  if (!spin_trylock_bh(&fib6_gc_lock)) {
1583  mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1584  return;
1585  }
1586  gc_args.timeout = net->ipv6.sysctl.ip6_rt_gc_interval;
1587  }
1588 
1589  gc_args.more = icmp6_dst_gc();
1590 
1591  fib6_clean_all(net, fib6_age, 0, NULL);
1592 
1593  if (gc_args.more)
1594  mod_timer(&net->ipv6.ip6_fib_timer,
1596  + net->ipv6.sysctl.ip6_rt_gc_interval));
1597  else
1598  del_timer(&net->ipv6.ip6_fib_timer);
1599  spin_unlock_bh(&fib6_gc_lock);
1600 }
1601 
1602 static void fib6_gc_timer_cb(unsigned long arg)
1603 {
1604  fib6_run_gc(0, (struct net *)arg);
1605 }
1606 
1607 static int __net_init fib6_net_init(struct net *net)
1608 {
1609  size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1610 
1611  setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1612 
1613  net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1614  if (!net->ipv6.rt6_stats)
1615  goto out_timer;
1616 
1617  /* Avoid false sharing : Use at least a full cache line */
1618  size = max_t(size_t, size, L1_CACHE_BYTES);
1619 
1620  net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1621  if (!net->ipv6.fib_table_hash)
1622  goto out_rt6_stats;
1623 
1624  net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1625  GFP_KERNEL);
1626  if (!net->ipv6.fib6_main_tbl)
1627  goto out_fib_table_hash;
1628 
1629  net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1630  net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1631  net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1633  inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
1634 
1635 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1636  net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1637  GFP_KERNEL);
1638  if (!net->ipv6.fib6_local_tbl)
1639  goto out_fib6_main_tbl;
1640  net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1641  net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1642  net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1644  inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
1645 #endif
1646  fib6_tables_init(net);
1647 
1648  return 0;
1649 
1650 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1651 out_fib6_main_tbl:
1652  kfree(net->ipv6.fib6_main_tbl);
1653 #endif
1654 out_fib_table_hash:
1655  kfree(net->ipv6.fib_table_hash);
1656 out_rt6_stats:
1657  kfree(net->ipv6.rt6_stats);
1658 out_timer:
1659  return -ENOMEM;
1660  }
1661 
1662 static void fib6_net_exit(struct net *net)
1663 {
1664  rt6_ifdown(net, NULL);
1665  del_timer_sync(&net->ipv6.ip6_fib_timer);
1666 
1667 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1668  inetpeer_invalidate_tree(&net->ipv6.fib6_local_tbl->tb6_peers);
1669  kfree(net->ipv6.fib6_local_tbl);
1670 #endif
1671  inetpeer_invalidate_tree(&net->ipv6.fib6_main_tbl->tb6_peers);
1672  kfree(net->ipv6.fib6_main_tbl);
1673  kfree(net->ipv6.fib_table_hash);
1674  kfree(net->ipv6.rt6_stats);
1675 }
1676 
1677 static struct pernet_operations fib6_net_ops = {
1678  .init = fib6_net_init,
1679  .exit = fib6_net_exit,
1680 };
1681 
1683 {
1684  int ret = -ENOMEM;
1685 
1686  fib6_node_kmem = kmem_cache_create("fib6_nodes",
1687  sizeof(struct fib6_node),
1688  0, SLAB_HWCACHE_ALIGN,
1689  NULL);
1690  if (!fib6_node_kmem)
1691  goto out;
1692 
1693  ret = register_pernet_subsys(&fib6_net_ops);
1694  if (ret)
1695  goto out_kmem_cache_create;
1696 
1697  ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
1698  NULL);
1699  if (ret)
1700  goto out_unregister_subsys;
1701 out:
1702  return ret;
1703 
1704 out_unregister_subsys:
1705  unregister_pernet_subsys(&fib6_net_ops);
1706 out_kmem_cache_create:
1707  kmem_cache_destroy(fib6_node_kmem);
1708  goto out;
1709 }
1710 
1712 {
1713  unregister_pernet_subsys(&fib6_net_ops);
1714  kmem_cache_destroy(fib6_node_kmem);
1715 }