powerpc/powernv: Report size of OPAL memcons log
[linux/fpc-iii.git] / net / ipv6 / ip6_fib.c
blobef548520452253dd5a39f6c24388b8fa1ddaa580
1 /*
2 * Linux INET6 implementation
3 * Forwarding Information Database
5 * Authors:
6 * Pedro Roque <roque@di.fc.ul.pt>
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.
13 * Changes:
14 * Yuji SEKIYA @USAGI: Support default route on router node;
15 * remove ip6_null_entry from the top of
16 * routing table.
17 * Ville Nuorvala: Fixed routing subtrees.
20 #define pr_fmt(fmt) "IPv6: " fmt
22 #include <linux/errno.h>
23 #include <linux/types.h>
24 #include <linux/net.h>
25 #include <linux/route.h>
26 #include <linux/netdevice.h>
27 #include <linux/in6.h>
28 #include <linux/init.h>
29 #include <linux/list.h>
30 #include <linux/slab.h>
32 #include <net/ipv6.h>
33 #include <net/ndisc.h>
34 #include <net/addrconf.h>
35 #include <net/lwtunnel.h>
37 #include <net/ip6_fib.h>
38 #include <net/ip6_route.h>
40 #define RT6_DEBUG 2
42 #if RT6_DEBUG >= 3
43 #define RT6_TRACE(x...) pr_debug(x)
44 #else
45 #define RT6_TRACE(x...) do { ; } while (0)
46 #endif
48 static struct kmem_cache *fib6_node_kmem __read_mostly;
50 struct fib6_cleaner {
51 struct fib6_walker w;
52 struct net *net;
53 int (*func)(struct rt6_info *, void *arg);
54 int sernum;
55 void *arg;
58 #ifdef CONFIG_IPV6_SUBTREES
59 #define FWS_INIT FWS_S
60 #else
61 #define FWS_INIT FWS_L
62 #endif
64 static void fib6_prune_clones(struct net *net, struct fib6_node *fn);
65 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn);
66 static struct fib6_node *fib6_repair_tree(struct net *net, struct fib6_node *fn);
67 static int fib6_walk(struct net *net, struct fib6_walker *w);
68 static int fib6_walk_continue(struct fib6_walker *w);
71 * A routing update causes an increase of the serial number on the
72 * affected subtree. This allows for cached routes to be asynchronously
73 * tested when modifications are made to the destination cache as a
74 * result of redirects, path MTU changes, etc.
77 static void fib6_gc_timer_cb(unsigned long arg);
79 #define FOR_WALKERS(net, w) \
80 list_for_each_entry(w, &(net)->ipv6.fib6_walkers, lh)
82 static void fib6_walker_link(struct net *net, struct fib6_walker *w)
84 write_lock_bh(&net->ipv6.fib6_walker_lock);
85 list_add(&w->lh, &net->ipv6.fib6_walkers);
86 write_unlock_bh(&net->ipv6.fib6_walker_lock);
89 static void fib6_walker_unlink(struct net *net, struct fib6_walker *w)
91 write_lock_bh(&net->ipv6.fib6_walker_lock);
92 list_del(&w->lh);
93 write_unlock_bh(&net->ipv6.fib6_walker_lock);
96 static int fib6_new_sernum(struct net *net)
98 int new, old;
100 do {
101 old = atomic_read(&net->ipv6.fib6_sernum);
102 new = old < INT_MAX ? old + 1 : 1;
103 } while (atomic_cmpxchg(&net->ipv6.fib6_sernum,
104 old, new) != old);
105 return new;
108 enum {
109 FIB6_NO_SERNUM_CHANGE = 0,
113 * Auxiliary address test functions for the radix tree.
115 * These assume a 32bit processor (although it will work on
116 * 64bit processors)
120 * test bit
122 #if defined(__LITTLE_ENDIAN)
123 # define BITOP_BE32_SWIZZLE (0x1F & ~7)
124 #else
125 # define BITOP_BE32_SWIZZLE 0
126 #endif
128 static __be32 addr_bit_set(const void *token, int fn_bit)
130 const __be32 *addr = token;
132 * Here,
133 * 1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
134 * is optimized version of
135 * htonl(1 << ((~fn_bit)&0x1F))
136 * See include/asm-generic/bitops/le.h.
138 return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
139 addr[fn_bit >> 5];
142 static struct fib6_node *node_alloc(void)
144 struct fib6_node *fn;
146 fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
148 return fn;
151 static void node_free(struct fib6_node *fn)
153 kmem_cache_free(fib6_node_kmem, fn);
156 static void rt6_rcu_free(struct rt6_info *rt)
158 call_rcu(&rt->dst.rcu_head, dst_rcu_free);
161 static void rt6_free_pcpu(struct rt6_info *non_pcpu_rt)
163 int cpu;
165 if (!non_pcpu_rt->rt6i_pcpu)
166 return;
168 for_each_possible_cpu(cpu) {
169 struct rt6_info **ppcpu_rt;
170 struct rt6_info *pcpu_rt;
172 ppcpu_rt = per_cpu_ptr(non_pcpu_rt->rt6i_pcpu, cpu);
173 pcpu_rt = *ppcpu_rt;
174 if (pcpu_rt) {
175 rt6_rcu_free(pcpu_rt);
176 *ppcpu_rt = NULL;
180 free_percpu(non_pcpu_rt->rt6i_pcpu);
181 non_pcpu_rt->rt6i_pcpu = NULL;
184 static void rt6_release(struct rt6_info *rt)
186 if (atomic_dec_and_test(&rt->rt6i_ref)) {
187 rt6_free_pcpu(rt);
188 rt6_rcu_free(rt);
192 static void fib6_link_table(struct net *net, struct fib6_table *tb)
194 unsigned int h;
197 * Initialize table lock at a single place to give lockdep a key,
198 * tables aren't visible prior to being linked to the list.
200 rwlock_init(&tb->tb6_lock);
202 h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
205 * No protection necessary, this is the only list mutatation
206 * operation, tables never disappear once they exist.
208 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
211 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
213 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
215 struct fib6_table *table;
217 table = kzalloc(sizeof(*table), GFP_ATOMIC);
218 if (table) {
219 table->tb6_id = id;
220 table->tb6_root.leaf = net->ipv6.ip6_null_entry;
221 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
222 inet_peer_base_init(&table->tb6_peers);
225 return table;
228 struct fib6_table *fib6_new_table(struct net *net, u32 id)
230 struct fib6_table *tb;
232 if (id == 0)
233 id = RT6_TABLE_MAIN;
234 tb = fib6_get_table(net, id);
235 if (tb)
236 return tb;
238 tb = fib6_alloc_table(net, id);
239 if (tb)
240 fib6_link_table(net, tb);
242 return tb;
244 EXPORT_SYMBOL_GPL(fib6_new_table);
246 struct fib6_table *fib6_get_table(struct net *net, u32 id)
248 struct fib6_table *tb;
249 struct hlist_head *head;
250 unsigned int h;
252 if (id == 0)
253 id = RT6_TABLE_MAIN;
254 h = id & (FIB6_TABLE_HASHSZ - 1);
255 rcu_read_lock();
256 head = &net->ipv6.fib_table_hash[h];
257 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
258 if (tb->tb6_id == id) {
259 rcu_read_unlock();
260 return tb;
263 rcu_read_unlock();
265 return NULL;
267 EXPORT_SYMBOL_GPL(fib6_get_table);
269 static void __net_init fib6_tables_init(struct net *net)
271 fib6_link_table(net, net->ipv6.fib6_main_tbl);
272 fib6_link_table(net, net->ipv6.fib6_local_tbl);
274 #else
276 struct fib6_table *fib6_new_table(struct net *net, u32 id)
278 return fib6_get_table(net, id);
281 struct fib6_table *fib6_get_table(struct net *net, u32 id)
283 return net->ipv6.fib6_main_tbl;
286 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
287 int flags, pol_lookup_t lookup)
289 struct rt6_info *rt;
291 rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
292 if (rt->rt6i_flags & RTF_REJECT &&
293 rt->dst.error == -EAGAIN) {
294 ip6_rt_put(rt);
295 rt = net->ipv6.ip6_null_entry;
296 dst_hold(&rt->dst);
299 return &rt->dst;
302 static void __net_init fib6_tables_init(struct net *net)
304 fib6_link_table(net, net->ipv6.fib6_main_tbl);
307 #endif
309 static int fib6_dump_node(struct fib6_walker *w)
311 int res;
312 struct rt6_info *rt;
314 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
315 res = rt6_dump_route(rt, w->args);
316 if (res < 0) {
317 /* Frame is full, suspend walking */
318 w->leaf = rt;
319 return 1;
322 w->leaf = NULL;
323 return 0;
326 static void fib6_dump_end(struct netlink_callback *cb)
328 struct net *net = sock_net(cb->skb->sk);
329 struct fib6_walker *w = (void *)cb->args[2];
331 if (w) {
332 if (cb->args[4]) {
333 cb->args[4] = 0;
334 fib6_walker_unlink(net, w);
336 cb->args[2] = 0;
337 kfree(w);
339 cb->done = (void *)cb->args[3];
340 cb->args[1] = 3;
343 static int fib6_dump_done(struct netlink_callback *cb)
345 fib6_dump_end(cb);
346 return cb->done ? cb->done(cb) : 0;
349 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
350 struct netlink_callback *cb)
352 struct net *net = sock_net(skb->sk);
353 struct fib6_walker *w;
354 int res;
356 w = (void *)cb->args[2];
357 w->root = &table->tb6_root;
359 if (cb->args[4] == 0) {
360 w->count = 0;
361 w->skip = 0;
363 read_lock_bh(&table->tb6_lock);
364 res = fib6_walk(net, w);
365 read_unlock_bh(&table->tb6_lock);
366 if (res > 0) {
367 cb->args[4] = 1;
368 cb->args[5] = w->root->fn_sernum;
370 } else {
371 if (cb->args[5] != w->root->fn_sernum) {
372 /* Begin at the root if the tree changed */
373 cb->args[5] = w->root->fn_sernum;
374 w->state = FWS_INIT;
375 w->node = w->root;
376 w->skip = w->count;
377 } else
378 w->skip = 0;
380 read_lock_bh(&table->tb6_lock);
381 res = fib6_walk_continue(w);
382 read_unlock_bh(&table->tb6_lock);
383 if (res <= 0) {
384 fib6_walker_unlink(net, w);
385 cb->args[4] = 0;
389 return res;
392 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
394 struct net *net = sock_net(skb->sk);
395 unsigned int h, s_h;
396 unsigned int e = 0, s_e;
397 struct rt6_rtnl_dump_arg arg;
398 struct fib6_walker *w;
399 struct fib6_table *tb;
400 struct hlist_head *head;
401 int res = 0;
403 s_h = cb->args[0];
404 s_e = cb->args[1];
406 w = (void *)cb->args[2];
407 if (!w) {
408 /* New dump:
410 * 1. hook callback destructor.
412 cb->args[3] = (long)cb->done;
413 cb->done = fib6_dump_done;
416 * 2. allocate and initialize walker.
418 w = kzalloc(sizeof(*w), GFP_ATOMIC);
419 if (!w)
420 return -ENOMEM;
421 w->func = fib6_dump_node;
422 cb->args[2] = (long)w;
425 arg.skb = skb;
426 arg.cb = cb;
427 arg.net = net;
428 w->args = &arg;
430 rcu_read_lock();
431 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
432 e = 0;
433 head = &net->ipv6.fib_table_hash[h];
434 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
435 if (e < s_e)
436 goto next;
437 res = fib6_dump_table(tb, skb, cb);
438 if (res != 0)
439 goto out;
440 next:
441 e++;
444 out:
445 rcu_read_unlock();
446 cb->args[1] = e;
447 cb->args[0] = h;
449 res = res < 0 ? res : skb->len;
450 if (res <= 0)
451 fib6_dump_end(cb);
452 return res;
456 * Routing Table
458 * return the appropriate node for a routing tree "add" operation
459 * by either creating and inserting or by returning an existing
460 * node.
463 static struct fib6_node *fib6_add_1(struct fib6_node *root,
464 struct in6_addr *addr, int plen,
465 int offset, int allow_create,
466 int replace_required, int sernum)
468 struct fib6_node *fn, *in, *ln;
469 struct fib6_node *pn = NULL;
470 struct rt6key *key;
471 int bit;
472 __be32 dir = 0;
474 RT6_TRACE("fib6_add_1\n");
476 /* insert node in tree */
478 fn = root;
480 do {
481 key = (struct rt6key *)((u8 *)fn->leaf + offset);
484 * Prefix match
486 if (plen < fn->fn_bit ||
487 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
488 if (!allow_create) {
489 if (replace_required) {
490 pr_warn("Can't replace route, no match found\n");
491 return ERR_PTR(-ENOENT);
493 pr_warn("NLM_F_CREATE should be set when creating new route\n");
495 goto insert_above;
499 * Exact match ?
502 if (plen == fn->fn_bit) {
503 /* clean up an intermediate node */
504 if (!(fn->fn_flags & RTN_RTINFO)) {
505 rt6_release(fn->leaf);
506 fn->leaf = NULL;
509 fn->fn_sernum = sernum;
511 return fn;
515 * We have more bits to go
518 /* Try to walk down on tree. */
519 fn->fn_sernum = sernum;
520 dir = addr_bit_set(addr, fn->fn_bit);
521 pn = fn;
522 fn = dir ? fn->right : fn->left;
523 } while (fn);
525 if (!allow_create) {
526 /* We should not create new node because
527 * NLM_F_REPLACE was specified without NLM_F_CREATE
528 * I assume it is safe to require NLM_F_CREATE when
529 * REPLACE flag is used! Later we may want to remove the
530 * check for replace_required, because according
531 * to netlink specification, NLM_F_CREATE
532 * MUST be specified if new route is created.
533 * That would keep IPv6 consistent with IPv4
535 if (replace_required) {
536 pr_warn("Can't replace route, no match found\n");
537 return ERR_PTR(-ENOENT);
539 pr_warn("NLM_F_CREATE should be set when creating new route\n");
542 * We walked to the bottom of tree.
543 * Create new leaf node without children.
546 ln = node_alloc();
548 if (!ln)
549 return ERR_PTR(-ENOMEM);
550 ln->fn_bit = plen;
552 ln->parent = pn;
553 ln->fn_sernum = sernum;
555 if (dir)
556 pn->right = ln;
557 else
558 pn->left = ln;
560 return ln;
563 insert_above:
565 * split since we don't have a common prefix anymore or
566 * we have a less significant route.
567 * we've to insert an intermediate node on the list
568 * this new node will point to the one we need to create
569 * and the current
572 pn = fn->parent;
574 /* find 1st bit in difference between the 2 addrs.
576 See comment in __ipv6_addr_diff: bit may be an invalid value,
577 but if it is >= plen, the value is ignored in any case.
580 bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
583 * (intermediate)[in]
584 * / \
585 * (new leaf node)[ln] (old node)[fn]
587 if (plen > bit) {
588 in = node_alloc();
589 ln = node_alloc();
591 if (!in || !ln) {
592 if (in)
593 node_free(in);
594 if (ln)
595 node_free(ln);
596 return ERR_PTR(-ENOMEM);
600 * new intermediate node.
601 * RTN_RTINFO will
602 * be off since that an address that chooses one of
603 * the branches would not match less specific routes
604 * in the other branch
607 in->fn_bit = bit;
609 in->parent = pn;
610 in->leaf = fn->leaf;
611 atomic_inc(&in->leaf->rt6i_ref);
613 in->fn_sernum = sernum;
615 /* update parent pointer */
616 if (dir)
617 pn->right = in;
618 else
619 pn->left = in;
621 ln->fn_bit = plen;
623 ln->parent = in;
624 fn->parent = in;
626 ln->fn_sernum = sernum;
628 if (addr_bit_set(addr, bit)) {
629 in->right = ln;
630 in->left = fn;
631 } else {
632 in->left = ln;
633 in->right = fn;
635 } else { /* plen <= bit */
638 * (new leaf node)[ln]
639 * / \
640 * (old node)[fn] NULL
643 ln = node_alloc();
645 if (!ln)
646 return ERR_PTR(-ENOMEM);
648 ln->fn_bit = plen;
650 ln->parent = pn;
652 ln->fn_sernum = sernum;
654 if (dir)
655 pn->right = ln;
656 else
657 pn->left = ln;
659 if (addr_bit_set(&key->addr, plen))
660 ln->right = fn;
661 else
662 ln->left = fn;
664 fn->parent = ln;
666 return ln;
669 static bool rt6_qualify_for_ecmp(struct rt6_info *rt)
671 return (rt->rt6i_flags & (RTF_GATEWAY|RTF_ADDRCONF|RTF_DYNAMIC)) ==
672 RTF_GATEWAY;
675 static void fib6_copy_metrics(u32 *mp, const struct mx6_config *mxc)
677 int i;
679 for (i = 0; i < RTAX_MAX; i++) {
680 if (test_bit(i, mxc->mx_valid))
681 mp[i] = mxc->mx[i];
685 static int fib6_commit_metrics(struct dst_entry *dst, struct mx6_config *mxc)
687 if (!mxc->mx)
688 return 0;
690 if (dst->flags & DST_HOST) {
691 u32 *mp = dst_metrics_write_ptr(dst);
693 if (unlikely(!mp))
694 return -ENOMEM;
696 fib6_copy_metrics(mp, mxc);
697 } else {
698 dst_init_metrics(dst, mxc->mx, false);
700 /* We've stolen mx now. */
701 mxc->mx = NULL;
704 return 0;
707 static void fib6_purge_rt(struct rt6_info *rt, struct fib6_node *fn,
708 struct net *net)
710 if (atomic_read(&rt->rt6i_ref) != 1) {
711 /* This route is used as dummy address holder in some split
712 * nodes. It is not leaked, but it still holds other resources,
713 * which must be released in time. So, scan ascendant nodes
714 * and replace dummy references to this route with references
715 * to still alive ones.
717 while (fn) {
718 if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) {
719 fn->leaf = fib6_find_prefix(net, fn);
720 atomic_inc(&fn->leaf->rt6i_ref);
721 rt6_release(rt);
723 fn = fn->parent;
725 /* No more references are possible at this point. */
726 BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
731 * Insert routing information in a node.
734 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
735 struct nl_info *info, struct mx6_config *mxc)
737 struct rt6_info *iter = NULL;
738 struct rt6_info **ins;
739 struct rt6_info **fallback_ins = NULL;
740 int replace = (info->nlh &&
741 (info->nlh->nlmsg_flags & NLM_F_REPLACE));
742 int add = (!info->nlh ||
743 (info->nlh->nlmsg_flags & NLM_F_CREATE));
744 int found = 0;
745 bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
746 u16 nlflags = NLM_F_EXCL;
747 int err;
749 ins = &fn->leaf;
751 for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) {
753 * Search for duplicates
756 if (iter->rt6i_metric == rt->rt6i_metric) {
758 * Same priority level
760 if (info->nlh &&
761 (info->nlh->nlmsg_flags & NLM_F_EXCL))
762 return -EEXIST;
764 nlflags &= ~NLM_F_EXCL;
765 if (replace) {
766 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
767 found++;
768 break;
770 if (rt_can_ecmp)
771 fallback_ins = fallback_ins ?: ins;
772 goto next_iter;
775 if (iter->dst.dev == rt->dst.dev &&
776 iter->rt6i_idev == rt->rt6i_idev &&
777 ipv6_addr_equal(&iter->rt6i_gateway,
778 &rt->rt6i_gateway)) {
779 if (rt->rt6i_nsiblings)
780 rt->rt6i_nsiblings = 0;
781 if (!(iter->rt6i_flags & RTF_EXPIRES))
782 return -EEXIST;
783 if (!(rt->rt6i_flags & RTF_EXPIRES))
784 rt6_clean_expires(iter);
785 else
786 rt6_set_expires(iter, rt->dst.expires);
787 iter->rt6i_pmtu = rt->rt6i_pmtu;
788 return -EEXIST;
790 /* If we have the same destination and the same metric,
791 * but not the same gateway, then the route we try to
792 * add is sibling to this route, increment our counter
793 * of siblings, and later we will add our route to the
794 * list.
795 * Only static routes (which don't have flag
796 * RTF_EXPIRES) are used for ECMPv6.
798 * To avoid long list, we only had siblings if the
799 * route have a gateway.
801 if (rt_can_ecmp &&
802 rt6_qualify_for_ecmp(iter))
803 rt->rt6i_nsiblings++;
806 if (iter->rt6i_metric > rt->rt6i_metric)
807 break;
809 next_iter:
810 ins = &iter->dst.rt6_next;
813 if (fallback_ins && !found) {
814 /* No ECMP-able route found, replace first non-ECMP one */
815 ins = fallback_ins;
816 iter = *ins;
817 found++;
820 /* Reset round-robin state, if necessary */
821 if (ins == &fn->leaf)
822 fn->rr_ptr = NULL;
824 /* Link this route to others same route. */
825 if (rt->rt6i_nsiblings) {
826 unsigned int rt6i_nsiblings;
827 struct rt6_info *sibling, *temp_sibling;
829 /* Find the first route that have the same metric */
830 sibling = fn->leaf;
831 while (sibling) {
832 if (sibling->rt6i_metric == rt->rt6i_metric &&
833 rt6_qualify_for_ecmp(sibling)) {
834 list_add_tail(&rt->rt6i_siblings,
835 &sibling->rt6i_siblings);
836 break;
838 sibling = sibling->dst.rt6_next;
840 /* For each sibling in the list, increment the counter of
841 * siblings. BUG() if counters does not match, list of siblings
842 * is broken!
844 rt6i_nsiblings = 0;
845 list_for_each_entry_safe(sibling, temp_sibling,
846 &rt->rt6i_siblings, rt6i_siblings) {
847 sibling->rt6i_nsiblings++;
848 BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings);
849 rt6i_nsiblings++;
851 BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings);
855 * insert node
857 if (!replace) {
858 if (!add)
859 pr_warn("NLM_F_CREATE should be set when creating new route\n");
861 add:
862 nlflags |= NLM_F_CREATE;
863 err = fib6_commit_metrics(&rt->dst, mxc);
864 if (err)
865 return err;
867 rt->dst.rt6_next = iter;
868 *ins = rt;
869 rt->rt6i_node = fn;
870 atomic_inc(&rt->rt6i_ref);
871 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
872 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
874 if (!(fn->fn_flags & RTN_RTINFO)) {
875 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
876 fn->fn_flags |= RTN_RTINFO;
879 } else {
880 int nsiblings;
882 if (!found) {
883 if (add)
884 goto add;
885 pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
886 return -ENOENT;
889 err = fib6_commit_metrics(&rt->dst, mxc);
890 if (err)
891 return err;
893 *ins = rt;
894 rt->rt6i_node = fn;
895 rt->dst.rt6_next = iter->dst.rt6_next;
896 atomic_inc(&rt->rt6i_ref);
897 inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
898 if (!(fn->fn_flags & RTN_RTINFO)) {
899 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
900 fn->fn_flags |= RTN_RTINFO;
902 nsiblings = iter->rt6i_nsiblings;
903 fib6_purge_rt(iter, fn, info->nl_net);
904 rt6_release(iter);
906 if (nsiblings) {
907 /* Replacing an ECMP route, remove all siblings */
908 ins = &rt->dst.rt6_next;
909 iter = *ins;
910 while (iter) {
911 if (rt6_qualify_for_ecmp(iter)) {
912 *ins = iter->dst.rt6_next;
913 fib6_purge_rt(iter, fn, info->nl_net);
914 rt6_release(iter);
915 nsiblings--;
916 } else {
917 ins = &iter->dst.rt6_next;
919 iter = *ins;
921 WARN_ON(nsiblings != 0);
925 return 0;
928 static void fib6_start_gc(struct net *net, struct rt6_info *rt)
930 if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
931 (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
932 mod_timer(&net->ipv6.ip6_fib_timer,
933 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
936 void fib6_force_start_gc(struct net *net)
938 if (!timer_pending(&net->ipv6.ip6_fib_timer))
939 mod_timer(&net->ipv6.ip6_fib_timer,
940 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
944 * Add routing information to the routing tree.
945 * <destination addr>/<source addr>
946 * with source addr info in sub-trees
949 int fib6_add(struct fib6_node *root, struct rt6_info *rt,
950 struct nl_info *info, struct mx6_config *mxc)
952 struct fib6_node *fn, *pn = NULL;
953 int err = -ENOMEM;
954 int allow_create = 1;
955 int replace_required = 0;
956 int sernum = fib6_new_sernum(info->nl_net);
958 if (WARN_ON_ONCE((rt->dst.flags & DST_NOCACHE) &&
959 !atomic_read(&rt->dst.__refcnt)))
960 return -EINVAL;
962 if (info->nlh) {
963 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
964 allow_create = 0;
965 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
966 replace_required = 1;
968 if (!allow_create && !replace_required)
969 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
971 fn = fib6_add_1(root, &rt->rt6i_dst.addr, rt->rt6i_dst.plen,
972 offsetof(struct rt6_info, rt6i_dst), allow_create,
973 replace_required, sernum);
974 if (IS_ERR(fn)) {
975 err = PTR_ERR(fn);
976 fn = NULL;
977 goto out;
980 pn = fn;
982 #ifdef CONFIG_IPV6_SUBTREES
983 if (rt->rt6i_src.plen) {
984 struct fib6_node *sn;
986 if (!fn->subtree) {
987 struct fib6_node *sfn;
990 * Create subtree.
992 * fn[main tree]
994 * sfn[subtree root]
996 * sn[new leaf node]
999 /* Create subtree root node */
1000 sfn = node_alloc();
1001 if (!sfn)
1002 goto st_failure;
1004 sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
1005 atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
1006 sfn->fn_flags = RTN_ROOT;
1007 sfn->fn_sernum = sernum;
1009 /* Now add the first leaf node to new subtree */
1011 sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
1012 rt->rt6i_src.plen,
1013 offsetof(struct rt6_info, rt6i_src),
1014 allow_create, replace_required, sernum);
1016 if (IS_ERR(sn)) {
1017 /* If it is failed, discard just allocated
1018 root, and then (in st_failure) stale node
1019 in main tree.
1021 node_free(sfn);
1022 err = PTR_ERR(sn);
1023 goto st_failure;
1026 /* Now link new subtree to main tree */
1027 sfn->parent = fn;
1028 fn->subtree = sfn;
1029 } else {
1030 sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
1031 rt->rt6i_src.plen,
1032 offsetof(struct rt6_info, rt6i_src),
1033 allow_create, replace_required, sernum);
1035 if (IS_ERR(sn)) {
1036 err = PTR_ERR(sn);
1037 goto st_failure;
1041 if (!fn->leaf) {
1042 fn->leaf = rt;
1043 atomic_inc(&rt->rt6i_ref);
1045 fn = sn;
1047 #endif
1049 err = fib6_add_rt2node(fn, rt, info, mxc);
1050 if (!err) {
1051 fib6_start_gc(info->nl_net, rt);
1052 if (!(rt->rt6i_flags & RTF_CACHE))
1053 fib6_prune_clones(info->nl_net, pn);
1054 rt->dst.flags &= ~DST_NOCACHE;
1057 out:
1058 if (err) {
1059 #ifdef CONFIG_IPV6_SUBTREES
1061 * If fib6_add_1 has cleared the old leaf pointer in the
1062 * super-tree leaf node we have to find a new one for it.
1064 if (pn != fn && pn->leaf == rt) {
1065 pn->leaf = NULL;
1066 atomic_dec(&rt->rt6i_ref);
1068 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
1069 pn->leaf = fib6_find_prefix(info->nl_net, pn);
1070 #if RT6_DEBUG >= 2
1071 if (!pn->leaf) {
1072 WARN_ON(pn->leaf == NULL);
1073 pn->leaf = info->nl_net->ipv6.ip6_null_entry;
1075 #endif
1076 atomic_inc(&pn->leaf->rt6i_ref);
1078 #endif
1079 if (!(rt->dst.flags & DST_NOCACHE))
1080 dst_free(&rt->dst);
1082 return err;
1084 #ifdef CONFIG_IPV6_SUBTREES
1085 /* Subtree creation failed, probably main tree node
1086 is orphan. If it is, shoot it.
1088 st_failure:
1089 if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
1090 fib6_repair_tree(info->nl_net, fn);
1091 if (!(rt->dst.flags & DST_NOCACHE))
1092 dst_free(&rt->dst);
1093 return err;
1094 #endif
1098 * Routing tree lookup
1102 struct lookup_args {
1103 int offset; /* key offset on rt6_info */
1104 const struct in6_addr *addr; /* search key */
1107 static struct fib6_node *fib6_lookup_1(struct fib6_node *root,
1108 struct lookup_args *args)
1110 struct fib6_node *fn;
1111 __be32 dir;
1113 if (unlikely(args->offset == 0))
1114 return NULL;
1117 * Descend on a tree
1120 fn = root;
1122 for (;;) {
1123 struct fib6_node *next;
1125 dir = addr_bit_set(args->addr, fn->fn_bit);
1127 next = dir ? fn->right : fn->left;
1129 if (next) {
1130 fn = next;
1131 continue;
1133 break;
1136 while (fn) {
1137 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
1138 struct rt6key *key;
1140 key = (struct rt6key *) ((u8 *) fn->leaf +
1141 args->offset);
1143 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1144 #ifdef CONFIG_IPV6_SUBTREES
1145 if (fn->subtree) {
1146 struct fib6_node *sfn;
1147 sfn = fib6_lookup_1(fn->subtree,
1148 args + 1);
1149 if (!sfn)
1150 goto backtrack;
1151 fn = sfn;
1153 #endif
1154 if (fn->fn_flags & RTN_RTINFO)
1155 return fn;
1158 #ifdef CONFIG_IPV6_SUBTREES
1159 backtrack:
1160 #endif
1161 if (fn->fn_flags & RTN_ROOT)
1162 break;
1164 fn = fn->parent;
1167 return NULL;
1170 struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
1171 const struct in6_addr *saddr)
1173 struct fib6_node *fn;
1174 struct lookup_args args[] = {
1176 .offset = offsetof(struct rt6_info, rt6i_dst),
1177 .addr = daddr,
1179 #ifdef CONFIG_IPV6_SUBTREES
1181 .offset = offsetof(struct rt6_info, rt6i_src),
1182 .addr = saddr,
1184 #endif
1186 .offset = 0, /* sentinel */
1190 fn = fib6_lookup_1(root, daddr ? args : args + 1);
1191 if (!fn || fn->fn_flags & RTN_TL_ROOT)
1192 fn = root;
1194 return fn;
1198 * Get node with specified destination prefix (and source prefix,
1199 * if subtrees are used)
1203 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1204 const struct in6_addr *addr,
1205 int plen, int offset)
1207 struct fib6_node *fn;
1209 for (fn = root; fn ; ) {
1210 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
1213 * Prefix match
1215 if (plen < fn->fn_bit ||
1216 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1217 return NULL;
1219 if (plen == fn->fn_bit)
1220 return fn;
1223 * We have more bits to go
1225 if (addr_bit_set(addr, fn->fn_bit))
1226 fn = fn->right;
1227 else
1228 fn = fn->left;
1230 return NULL;
1233 struct fib6_node *fib6_locate(struct fib6_node *root,
1234 const struct in6_addr *daddr, int dst_len,
1235 const struct in6_addr *saddr, int src_len)
1237 struct fib6_node *fn;
1239 fn = fib6_locate_1(root, daddr, dst_len,
1240 offsetof(struct rt6_info, rt6i_dst));
1242 #ifdef CONFIG_IPV6_SUBTREES
1243 if (src_len) {
1244 WARN_ON(saddr == NULL);
1245 if (fn && fn->subtree)
1246 fn = fib6_locate_1(fn->subtree, saddr, src_len,
1247 offsetof(struct rt6_info, rt6i_src));
1249 #endif
1251 if (fn && fn->fn_flags & RTN_RTINFO)
1252 return fn;
1254 return NULL;
1259 * Deletion
1263 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
1265 if (fn->fn_flags & RTN_ROOT)
1266 return net->ipv6.ip6_null_entry;
1268 while (fn) {
1269 if (fn->left)
1270 return fn->left->leaf;
1271 if (fn->right)
1272 return fn->right->leaf;
1274 fn = FIB6_SUBTREE(fn);
1276 return NULL;
1280 * Called to trim the tree of intermediate nodes when possible. "fn"
1281 * is the node we want to try and remove.
1284 static struct fib6_node *fib6_repair_tree(struct net *net,
1285 struct fib6_node *fn)
1287 int children;
1288 int nstate;
1289 struct fib6_node *child, *pn;
1290 struct fib6_walker *w;
1291 int iter = 0;
1293 for (;;) {
1294 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1295 iter++;
1297 WARN_ON(fn->fn_flags & RTN_RTINFO);
1298 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1299 WARN_ON(fn->leaf);
1301 children = 0;
1302 child = NULL;
1303 if (fn->right)
1304 child = fn->right, children |= 1;
1305 if (fn->left)
1306 child = fn->left, children |= 2;
1308 if (children == 3 || FIB6_SUBTREE(fn)
1309 #ifdef CONFIG_IPV6_SUBTREES
1310 /* Subtree root (i.e. fn) may have one child */
1311 || (children && fn->fn_flags & RTN_ROOT)
1312 #endif
1314 fn->leaf = fib6_find_prefix(net, fn);
1315 #if RT6_DEBUG >= 2
1316 if (!fn->leaf) {
1317 WARN_ON(!fn->leaf);
1318 fn->leaf = net->ipv6.ip6_null_entry;
1320 #endif
1321 atomic_inc(&fn->leaf->rt6i_ref);
1322 return fn->parent;
1325 pn = fn->parent;
1326 #ifdef CONFIG_IPV6_SUBTREES
1327 if (FIB6_SUBTREE(pn) == fn) {
1328 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1329 FIB6_SUBTREE(pn) = NULL;
1330 nstate = FWS_L;
1331 } else {
1332 WARN_ON(fn->fn_flags & RTN_ROOT);
1333 #endif
1334 if (pn->right == fn)
1335 pn->right = child;
1336 else if (pn->left == fn)
1337 pn->left = child;
1338 #if RT6_DEBUG >= 2
1339 else
1340 WARN_ON(1);
1341 #endif
1342 if (child)
1343 child->parent = pn;
1344 nstate = FWS_R;
1345 #ifdef CONFIG_IPV6_SUBTREES
1347 #endif
1349 read_lock(&net->ipv6.fib6_walker_lock);
1350 FOR_WALKERS(net, w) {
1351 if (!child) {
1352 if (w->root == fn) {
1353 w->root = w->node = NULL;
1354 RT6_TRACE("W %p adjusted by delroot 1\n", w);
1355 } else if (w->node == fn) {
1356 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1357 w->node = pn;
1358 w->state = nstate;
1360 } else {
1361 if (w->root == fn) {
1362 w->root = child;
1363 RT6_TRACE("W %p adjusted by delroot 2\n", w);
1365 if (w->node == fn) {
1366 w->node = child;
1367 if (children&2) {
1368 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1369 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1370 } else {
1371 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1372 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1377 read_unlock(&net->ipv6.fib6_walker_lock);
1379 node_free(fn);
1380 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1381 return pn;
1383 rt6_release(pn->leaf);
1384 pn->leaf = NULL;
1385 fn = pn;
1389 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1390 struct nl_info *info)
1392 struct fib6_walker *w;
1393 struct rt6_info *rt = *rtp;
1394 struct net *net = info->nl_net;
1396 RT6_TRACE("fib6_del_route\n");
1398 /* Unlink it */
1399 *rtp = rt->dst.rt6_next;
1400 rt->rt6i_node = NULL;
1401 net->ipv6.rt6_stats->fib_rt_entries--;
1402 net->ipv6.rt6_stats->fib_discarded_routes++;
1404 /* Reset round-robin state, if necessary */
1405 if (fn->rr_ptr == rt)
1406 fn->rr_ptr = NULL;
1408 /* Remove this entry from other siblings */
1409 if (rt->rt6i_nsiblings) {
1410 struct rt6_info *sibling, *next_sibling;
1412 list_for_each_entry_safe(sibling, next_sibling,
1413 &rt->rt6i_siblings, rt6i_siblings)
1414 sibling->rt6i_nsiblings--;
1415 rt->rt6i_nsiblings = 0;
1416 list_del_init(&rt->rt6i_siblings);
1419 /* Adjust walkers */
1420 read_lock(&net->ipv6.fib6_walker_lock);
1421 FOR_WALKERS(net, w) {
1422 if (w->state == FWS_C && w->leaf == rt) {
1423 RT6_TRACE("walker %p adjusted by delroute\n", w);
1424 w->leaf = rt->dst.rt6_next;
1425 if (!w->leaf)
1426 w->state = FWS_U;
1429 read_unlock(&net->ipv6.fib6_walker_lock);
1431 rt->dst.rt6_next = NULL;
1433 /* If it was last route, expunge its radix tree node */
1434 if (!fn->leaf) {
1435 fn->fn_flags &= ~RTN_RTINFO;
1436 net->ipv6.rt6_stats->fib_route_nodes--;
1437 fn = fib6_repair_tree(net, fn);
1440 fib6_purge_rt(rt, fn, net);
1442 inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1443 rt6_release(rt);
1446 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1448 struct net *net = info->nl_net;
1449 struct fib6_node *fn = rt->rt6i_node;
1450 struct rt6_info **rtp;
1452 #if RT6_DEBUG >= 2
1453 if (rt->dst.obsolete > 0) {
1454 WARN_ON(fn);
1455 return -ENOENT;
1457 #endif
1458 if (!fn || rt == net->ipv6.ip6_null_entry)
1459 return -ENOENT;
1461 WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1463 if (!(rt->rt6i_flags & RTF_CACHE)) {
1464 struct fib6_node *pn = fn;
1465 #ifdef CONFIG_IPV6_SUBTREES
1466 /* clones of this route might be in another subtree */
1467 if (rt->rt6i_src.plen) {
1468 while (!(pn->fn_flags & RTN_ROOT))
1469 pn = pn->parent;
1470 pn = pn->parent;
1472 #endif
1473 fib6_prune_clones(info->nl_net, pn);
1477 * Walk the leaf entries looking for ourself
1480 for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1481 if (*rtp == rt) {
1482 fib6_del_route(fn, rtp, info);
1483 return 0;
1486 return -ENOENT;
1490 * Tree traversal function.
1492 * Certainly, it is not interrupt safe.
1493 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1494 * It means, that we can modify tree during walking
1495 * and use this function for garbage collection, clone pruning,
1496 * cleaning tree when a device goes down etc. etc.
1498 * It guarantees that every node will be traversed,
1499 * and that it will be traversed only once.
1501 * Callback function w->func may return:
1502 * 0 -> continue walking.
1503 * positive value -> walking is suspended (used by tree dumps,
1504 * and probably by gc, if it will be split to several slices)
1505 * negative value -> terminate walking.
1507 * The function itself returns:
1508 * 0 -> walk is complete.
1509 * >0 -> walk is incomplete (i.e. suspended)
1510 * <0 -> walk is terminated by an error.
1513 static int fib6_walk_continue(struct fib6_walker *w)
1515 struct fib6_node *fn, *pn;
1517 for (;;) {
1518 fn = w->node;
1519 if (!fn)
1520 return 0;
1522 if (w->prune && fn != w->root &&
1523 fn->fn_flags & RTN_RTINFO && w->state < FWS_C) {
1524 w->state = FWS_C;
1525 w->leaf = fn->leaf;
1527 switch (w->state) {
1528 #ifdef CONFIG_IPV6_SUBTREES
1529 case FWS_S:
1530 if (FIB6_SUBTREE(fn)) {
1531 w->node = FIB6_SUBTREE(fn);
1532 continue;
1534 w->state = FWS_L;
1535 #endif
1536 case FWS_L:
1537 if (fn->left) {
1538 w->node = fn->left;
1539 w->state = FWS_INIT;
1540 continue;
1542 w->state = FWS_R;
1543 case FWS_R:
1544 if (fn->right) {
1545 w->node = fn->right;
1546 w->state = FWS_INIT;
1547 continue;
1549 w->state = FWS_C;
1550 w->leaf = fn->leaf;
1551 case FWS_C:
1552 if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1553 int err;
1555 if (w->skip) {
1556 w->skip--;
1557 goto skip;
1560 err = w->func(w);
1561 if (err)
1562 return err;
1564 w->count++;
1565 continue;
1567 skip:
1568 w->state = FWS_U;
1569 case FWS_U:
1570 if (fn == w->root)
1571 return 0;
1572 pn = fn->parent;
1573 w->node = pn;
1574 #ifdef CONFIG_IPV6_SUBTREES
1575 if (FIB6_SUBTREE(pn) == fn) {
1576 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1577 w->state = FWS_L;
1578 continue;
1580 #endif
1581 if (pn->left == fn) {
1582 w->state = FWS_R;
1583 continue;
1585 if (pn->right == fn) {
1586 w->state = FWS_C;
1587 w->leaf = w->node->leaf;
1588 continue;
1590 #if RT6_DEBUG >= 2
1591 WARN_ON(1);
1592 #endif
1597 static int fib6_walk(struct net *net, struct fib6_walker *w)
1599 int res;
1601 w->state = FWS_INIT;
1602 w->node = w->root;
1604 fib6_walker_link(net, w);
1605 res = fib6_walk_continue(w);
1606 if (res <= 0)
1607 fib6_walker_unlink(net, w);
1608 return res;
1611 static int fib6_clean_node(struct fib6_walker *w)
1613 int res;
1614 struct rt6_info *rt;
1615 struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1616 struct nl_info info = {
1617 .nl_net = c->net,
1620 if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1621 w->node->fn_sernum != c->sernum)
1622 w->node->fn_sernum = c->sernum;
1624 if (!c->func) {
1625 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1626 w->leaf = NULL;
1627 return 0;
1630 for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1631 res = c->func(rt, c->arg);
1632 if (res < 0) {
1633 w->leaf = rt;
1634 res = fib6_del(rt, &info);
1635 if (res) {
1636 #if RT6_DEBUG >= 2
1637 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1638 __func__, rt, rt->rt6i_node, res);
1639 #endif
1640 continue;
1642 return 0;
1644 WARN_ON(res != 0);
1646 w->leaf = rt;
1647 return 0;
1651 * Convenient frontend to tree walker.
1653 * func is called on each route.
1654 * It may return -1 -> delete this route.
1655 * 0 -> continue walking
1657 * prune==1 -> only immediate children of node (certainly,
1658 * ignoring pure split nodes) will be scanned.
1661 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1662 int (*func)(struct rt6_info *, void *arg),
1663 bool prune, int sernum, void *arg)
1665 struct fib6_cleaner c;
1667 c.w.root = root;
1668 c.w.func = fib6_clean_node;
1669 c.w.prune = prune;
1670 c.w.count = 0;
1671 c.w.skip = 0;
1672 c.func = func;
1673 c.sernum = sernum;
1674 c.arg = arg;
1675 c.net = net;
1677 fib6_walk(net, &c.w);
1680 static void __fib6_clean_all(struct net *net,
1681 int (*func)(struct rt6_info *, void *),
1682 int sernum, void *arg)
1684 struct fib6_table *table;
1685 struct hlist_head *head;
1686 unsigned int h;
1688 rcu_read_lock();
1689 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1690 head = &net->ipv6.fib_table_hash[h];
1691 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
1692 write_lock_bh(&table->tb6_lock);
1693 fib6_clean_tree(net, &table->tb6_root,
1694 func, false, sernum, arg);
1695 write_unlock_bh(&table->tb6_lock);
1698 rcu_read_unlock();
1701 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *),
1702 void *arg)
1704 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg);
1707 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1709 if (rt->rt6i_flags & RTF_CACHE) {
1710 RT6_TRACE("pruning clone %p\n", rt);
1711 return -1;
1714 return 0;
1717 static void fib6_prune_clones(struct net *net, struct fib6_node *fn)
1719 fib6_clean_tree(net, fn, fib6_prune_clone, true,
1720 FIB6_NO_SERNUM_CHANGE, NULL);
1723 static void fib6_flush_trees(struct net *net)
1725 int new_sernum = fib6_new_sernum(net);
1727 __fib6_clean_all(net, NULL, new_sernum, NULL);
1731 * Garbage collection
1734 struct fib6_gc_args
1736 int timeout;
1737 int more;
1740 static int fib6_age(struct rt6_info *rt, void *arg)
1742 struct fib6_gc_args *gc_args = arg;
1743 unsigned long now = jiffies;
1746 * check addrconf expiration here.
1747 * Routes are expired even if they are in use.
1749 * Also age clones. Note, that clones are aged out
1750 * only if they are not in use now.
1753 if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
1754 if (time_after(now, rt->dst.expires)) {
1755 RT6_TRACE("expiring %p\n", rt);
1756 return -1;
1758 gc_args->more++;
1759 } else if (rt->rt6i_flags & RTF_CACHE) {
1760 if (atomic_read(&rt->dst.__refcnt) == 0 &&
1761 time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) {
1762 RT6_TRACE("aging clone %p\n", rt);
1763 return -1;
1764 } else if (rt->rt6i_flags & RTF_GATEWAY) {
1765 struct neighbour *neigh;
1766 __u8 neigh_flags = 0;
1768 neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway);
1769 if (neigh) {
1770 neigh_flags = neigh->flags;
1771 neigh_release(neigh);
1773 if (!(neigh_flags & NTF_ROUTER)) {
1774 RT6_TRACE("purging route %p via non-router but gateway\n",
1775 rt);
1776 return -1;
1779 gc_args->more++;
1782 return 0;
1785 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
1787 struct fib6_gc_args gc_args;
1788 unsigned long now;
1790 if (force) {
1791 spin_lock_bh(&net->ipv6.fib6_gc_lock);
1792 } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
1793 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1794 return;
1796 gc_args.timeout = expires ? (int)expires :
1797 net->ipv6.sysctl.ip6_rt_gc_interval;
1799 gc_args.more = icmp6_dst_gc();
1801 fib6_clean_all(net, fib6_age, &gc_args);
1802 now = jiffies;
1803 net->ipv6.ip6_rt_last_gc = now;
1805 if (gc_args.more)
1806 mod_timer(&net->ipv6.ip6_fib_timer,
1807 round_jiffies(now
1808 + net->ipv6.sysctl.ip6_rt_gc_interval));
1809 else
1810 del_timer(&net->ipv6.ip6_fib_timer);
1811 spin_unlock_bh(&net->ipv6.fib6_gc_lock);
1814 static void fib6_gc_timer_cb(unsigned long arg)
1816 fib6_run_gc(0, (struct net *)arg, true);
1819 static int __net_init fib6_net_init(struct net *net)
1821 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1823 spin_lock_init(&net->ipv6.fib6_gc_lock);
1824 rwlock_init(&net->ipv6.fib6_walker_lock);
1825 INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
1826 setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1828 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1829 if (!net->ipv6.rt6_stats)
1830 goto out_timer;
1832 /* Avoid false sharing : Use at least a full cache line */
1833 size = max_t(size_t, size, L1_CACHE_BYTES);
1835 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1836 if (!net->ipv6.fib_table_hash)
1837 goto out_rt6_stats;
1839 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1840 GFP_KERNEL);
1841 if (!net->ipv6.fib6_main_tbl)
1842 goto out_fib_table_hash;
1844 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1845 net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1846 net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1847 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1848 inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
1850 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1851 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1852 GFP_KERNEL);
1853 if (!net->ipv6.fib6_local_tbl)
1854 goto out_fib6_main_tbl;
1855 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1856 net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1857 net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1858 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1859 inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
1860 #endif
1861 fib6_tables_init(net);
1863 return 0;
1865 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1866 out_fib6_main_tbl:
1867 kfree(net->ipv6.fib6_main_tbl);
1868 #endif
1869 out_fib_table_hash:
1870 kfree(net->ipv6.fib_table_hash);
1871 out_rt6_stats:
1872 kfree(net->ipv6.rt6_stats);
1873 out_timer:
1874 return -ENOMEM;
1877 static void fib6_net_exit(struct net *net)
1879 rt6_ifdown(net, NULL);
1880 del_timer_sync(&net->ipv6.ip6_fib_timer);
1882 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1883 inetpeer_invalidate_tree(&net->ipv6.fib6_local_tbl->tb6_peers);
1884 kfree(net->ipv6.fib6_local_tbl);
1885 #endif
1886 inetpeer_invalidate_tree(&net->ipv6.fib6_main_tbl->tb6_peers);
1887 kfree(net->ipv6.fib6_main_tbl);
1888 kfree(net->ipv6.fib_table_hash);
1889 kfree(net->ipv6.rt6_stats);
1892 static struct pernet_operations fib6_net_ops = {
1893 .init = fib6_net_init,
1894 .exit = fib6_net_exit,
1897 int __init fib6_init(void)
1899 int ret = -ENOMEM;
1901 fib6_node_kmem = kmem_cache_create("fib6_nodes",
1902 sizeof(struct fib6_node),
1903 0, SLAB_HWCACHE_ALIGN,
1904 NULL);
1905 if (!fib6_node_kmem)
1906 goto out;
1908 ret = register_pernet_subsys(&fib6_net_ops);
1909 if (ret)
1910 goto out_kmem_cache_create;
1912 ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
1913 NULL);
1914 if (ret)
1915 goto out_unregister_subsys;
1917 __fib6_flush_trees = fib6_flush_trees;
1918 out:
1919 return ret;
1921 out_unregister_subsys:
1922 unregister_pernet_subsys(&fib6_net_ops);
1923 out_kmem_cache_create:
1924 kmem_cache_destroy(fib6_node_kmem);
1925 goto out;
1928 void fib6_gc_cleanup(void)
1930 unregister_pernet_subsys(&fib6_net_ops);
1931 kmem_cache_destroy(fib6_node_kmem);
1934 #ifdef CONFIG_PROC_FS
1936 struct ipv6_route_iter {
1937 struct seq_net_private p;
1938 struct fib6_walker w;
1939 loff_t skip;
1940 struct fib6_table *tbl;
1941 int sernum;
1944 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
1946 struct rt6_info *rt = v;
1947 struct ipv6_route_iter *iter = seq->private;
1949 seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen);
1951 #ifdef CONFIG_IPV6_SUBTREES
1952 seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen);
1953 #else
1954 seq_puts(seq, "00000000000000000000000000000000 00 ");
1955 #endif
1956 if (rt->rt6i_flags & RTF_GATEWAY)
1957 seq_printf(seq, "%pi6", &rt->rt6i_gateway);
1958 else
1959 seq_puts(seq, "00000000000000000000000000000000");
1961 seq_printf(seq, " %08x %08x %08x %08x %8s\n",
1962 rt->rt6i_metric, atomic_read(&rt->dst.__refcnt),
1963 rt->dst.__use, rt->rt6i_flags,
1964 rt->dst.dev ? rt->dst.dev->name : "");
1965 iter->w.leaf = NULL;
1966 return 0;
1969 static int ipv6_route_yield(struct fib6_walker *w)
1971 struct ipv6_route_iter *iter = w->args;
1973 if (!iter->skip)
1974 return 1;
1976 do {
1977 iter->w.leaf = iter->w.leaf->dst.rt6_next;
1978 iter->skip--;
1979 if (!iter->skip && iter->w.leaf)
1980 return 1;
1981 } while (iter->w.leaf);
1983 return 0;
1986 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
1987 struct net *net)
1989 memset(&iter->w, 0, sizeof(iter->w));
1990 iter->w.func = ipv6_route_yield;
1991 iter->w.root = &iter->tbl->tb6_root;
1992 iter->w.state = FWS_INIT;
1993 iter->w.node = iter->w.root;
1994 iter->w.args = iter;
1995 iter->sernum = iter->w.root->fn_sernum;
1996 INIT_LIST_HEAD(&iter->w.lh);
1997 fib6_walker_link(net, &iter->w);
2000 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2001 struct net *net)
2003 unsigned int h;
2004 struct hlist_node *node;
2006 if (tbl) {
2007 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2008 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2009 } else {
2010 h = 0;
2011 node = NULL;
2014 while (!node && h < FIB6_TABLE_HASHSZ) {
2015 node = rcu_dereference_bh(
2016 hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2018 return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2021 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2023 if (iter->sernum != iter->w.root->fn_sernum) {
2024 iter->sernum = iter->w.root->fn_sernum;
2025 iter->w.state = FWS_INIT;
2026 iter->w.node = iter->w.root;
2027 WARN_ON(iter->w.skip);
2028 iter->w.skip = iter->w.count;
2032 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2034 int r;
2035 struct rt6_info *n;
2036 struct net *net = seq_file_net(seq);
2037 struct ipv6_route_iter *iter = seq->private;
2039 if (!v)
2040 goto iter_table;
2042 n = ((struct rt6_info *)v)->dst.rt6_next;
2043 if (n) {
2044 ++*pos;
2045 return n;
2048 iter_table:
2049 ipv6_route_check_sernum(iter);
2050 read_lock(&iter->tbl->tb6_lock);
2051 r = fib6_walk_continue(&iter->w);
2052 read_unlock(&iter->tbl->tb6_lock);
2053 if (r > 0) {
2054 if (v)
2055 ++*pos;
2056 return iter->w.leaf;
2057 } else if (r < 0) {
2058 fib6_walker_unlink(net, &iter->w);
2059 return NULL;
2061 fib6_walker_unlink(net, &iter->w);
2063 iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2064 if (!iter->tbl)
2065 return NULL;
2067 ipv6_route_seq_setup_walk(iter, net);
2068 goto iter_table;
2071 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2072 __acquires(RCU_BH)
2074 struct net *net = seq_file_net(seq);
2075 struct ipv6_route_iter *iter = seq->private;
2077 rcu_read_lock_bh();
2078 iter->tbl = ipv6_route_seq_next_table(NULL, net);
2079 iter->skip = *pos;
2081 if (iter->tbl) {
2082 ipv6_route_seq_setup_walk(iter, net);
2083 return ipv6_route_seq_next(seq, NULL, pos);
2084 } else {
2085 return NULL;
2089 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2091 struct fib6_walker *w = &iter->w;
2092 return w->node && !(w->state == FWS_U && w->node == w->root);
2095 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2096 __releases(RCU_BH)
2098 struct net *net = seq_file_net(seq);
2099 struct ipv6_route_iter *iter = seq->private;
2101 if (ipv6_route_iter_active(iter))
2102 fib6_walker_unlink(net, &iter->w);
2104 rcu_read_unlock_bh();
2107 static const struct seq_operations ipv6_route_seq_ops = {
2108 .start = ipv6_route_seq_start,
2109 .next = ipv6_route_seq_next,
2110 .stop = ipv6_route_seq_stop,
2111 .show = ipv6_route_seq_show
2114 int ipv6_route_open(struct inode *inode, struct file *file)
2116 return seq_open_net(inode, file, &ipv6_route_seq_ops,
2117 sizeof(struct ipv6_route_iter));
2120 #endif /* CONFIG_PROC_FS */