2 * Linux INET6 implementation
3 * Forwarding Information Database
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.
14 * Yuji SEKIYA @USAGI: Support default route on router node;
15 * remove ip6_null_entry from the top of
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>
34 #include <net/ndisc.h>
35 #include <net/addrconf.h>
36 #include <net/lwtunnel.h>
37 #include <net/fib_notifier.h>
39 #include <net/ip6_fib.h>
40 #include <net/ip6_route.h>
42 static struct kmem_cache *fib6_node_kmem __read_mostly;
47 int (*func)(struct fib6_info *, void *arg);
53 #ifdef CONFIG_IPV6_SUBTREES
54 #define FWS_INIT FWS_S
56 #define FWS_INIT FWS_L
59 static struct fib6_info *fib6_find_prefix(struct net *net,
60 struct fib6_table *table,
61 struct fib6_node *fn);
62 static struct fib6_node *fib6_repair_tree(struct net *net,
63 struct fib6_table *table,
64 struct fib6_node *fn);
65 static int fib6_walk(struct net *net, struct fib6_walker *w);
66 static int fib6_walk_continue(struct fib6_walker *w);
69 * A routing update causes an increase of the serial number on the
70 * affected subtree. This allows for cached routes to be asynchronously
71 * tested when modifications are made to the destination cache as a
72 * result of redirects, path MTU changes, etc.
75 static void fib6_gc_timer_cb(struct timer_list *t);
77 #define FOR_WALKERS(net, w) \
78 list_for_each_entry(w, &(net)->ipv6.fib6_walkers, lh)
80 static void fib6_walker_link(struct net *net, struct fib6_walker *w)
82 write_lock_bh(&net->ipv6.fib6_walker_lock);
83 list_add(&w->lh, &net->ipv6.fib6_walkers);
84 write_unlock_bh(&net->ipv6.fib6_walker_lock);
87 static void fib6_walker_unlink(struct net *net, struct fib6_walker *w)
89 write_lock_bh(&net->ipv6.fib6_walker_lock);
91 write_unlock_bh(&net->ipv6.fib6_walker_lock);
94 static int fib6_new_sernum(struct net *net)
99 old = atomic_read(&net->ipv6.fib6_sernum);
100 new = old < INT_MAX ? old + 1 : 1;
101 } while (atomic_cmpxchg(&net->ipv6.fib6_sernum,
107 FIB6_NO_SERNUM_CHANGE = 0,
110 void fib6_update_sernum(struct net *net, struct fib6_info *f6i)
112 struct fib6_node *fn;
114 fn = rcu_dereference_protected(f6i->fib6_node,
115 lockdep_is_held(&f6i->fib6_table->tb6_lock));
117 fn->fn_sernum = fib6_new_sernum(net);
121 * Auxiliary address test functions for the radix tree.
123 * These assume a 32bit processor (although it will work on
130 #if defined(__LITTLE_ENDIAN)
131 # define BITOP_BE32_SWIZZLE (0x1F & ~7)
133 # define BITOP_BE32_SWIZZLE 0
136 static __be32 addr_bit_set(const void *token, int fn_bit)
138 const __be32 *addr = token;
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.
146 return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
150 struct fib6_info *fib6_info_alloc(gfp_t gfp_flags)
152 struct fib6_info *f6i;
154 f6i = kzalloc(sizeof(*f6i), gfp_flags);
158 f6i->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags);
159 if (!f6i->rt6i_pcpu) {
164 INIT_LIST_HEAD(&f6i->fib6_siblings);
165 atomic_inc(&f6i->fib6_ref);
170 void fib6_info_destroy_rcu(struct rcu_head *head)
172 struct fib6_info *f6i = container_of(head, struct fib6_info, rcu);
173 struct rt6_exception_bucket *bucket;
175 WARN_ON(f6i->fib6_node);
177 bucket = rcu_dereference_protected(f6i->rt6i_exception_bucket, 1);
179 f6i->rt6i_exception_bucket = NULL;
183 if (f6i->rt6i_pcpu) {
186 for_each_possible_cpu(cpu) {
187 struct rt6_info **ppcpu_rt;
188 struct rt6_info *pcpu_rt;
190 ppcpu_rt = per_cpu_ptr(f6i->rt6i_pcpu, cpu);
193 dst_dev_put(&pcpu_rt->dst);
194 dst_release(&pcpu_rt->dst);
199 free_percpu(f6i->rt6i_pcpu);
202 lwtstate_put(f6i->fib6_nh.nh_lwtstate);
204 if (f6i->fib6_nh.nh_dev)
205 dev_put(f6i->fib6_nh.nh_dev);
207 ip_fib_metrics_put(f6i->fib6_metrics);
211 EXPORT_SYMBOL_GPL(fib6_info_destroy_rcu);
213 static struct fib6_node *node_alloc(struct net *net)
215 struct fib6_node *fn;
217 fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
219 net->ipv6.rt6_stats->fib_nodes++;
224 static void node_free_immediate(struct net *net, struct fib6_node *fn)
226 kmem_cache_free(fib6_node_kmem, fn);
227 net->ipv6.rt6_stats->fib_nodes--;
230 static void node_free_rcu(struct rcu_head *head)
232 struct fib6_node *fn = container_of(head, struct fib6_node, rcu);
234 kmem_cache_free(fib6_node_kmem, fn);
237 static void node_free(struct net *net, struct fib6_node *fn)
239 call_rcu(&fn->rcu, node_free_rcu);
240 net->ipv6.rt6_stats->fib_nodes--;
243 static void fib6_free_table(struct fib6_table *table)
245 inetpeer_invalidate_tree(&table->tb6_peers);
249 static void fib6_link_table(struct net *net, struct fib6_table *tb)
254 * Initialize table lock at a single place to give lockdep a key,
255 * tables aren't visible prior to being linked to the list.
257 spin_lock_init(&tb->tb6_lock);
258 h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
261 * No protection necessary, this is the only list mutatation
262 * operation, tables never disappear once they exist.
264 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
267 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
269 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
271 struct fib6_table *table;
273 table = kzalloc(sizeof(*table), GFP_ATOMIC);
276 rcu_assign_pointer(table->tb6_root.leaf,
277 net->ipv6.fib6_null_entry);
278 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
279 inet_peer_base_init(&table->tb6_peers);
285 struct fib6_table *fib6_new_table(struct net *net, u32 id)
287 struct fib6_table *tb;
291 tb = fib6_get_table(net, id);
295 tb = fib6_alloc_table(net, id);
297 fib6_link_table(net, tb);
301 EXPORT_SYMBOL_GPL(fib6_new_table);
303 struct fib6_table *fib6_get_table(struct net *net, u32 id)
305 struct fib6_table *tb;
306 struct hlist_head *head;
311 h = id & (FIB6_TABLE_HASHSZ - 1);
313 head = &net->ipv6.fib_table_hash[h];
314 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
315 if (tb->tb6_id == id) {
324 EXPORT_SYMBOL_GPL(fib6_get_table);
326 static void __net_init fib6_tables_init(struct net *net)
328 fib6_link_table(net, net->ipv6.fib6_main_tbl);
329 fib6_link_table(net, net->ipv6.fib6_local_tbl);
333 struct fib6_table *fib6_new_table(struct net *net, u32 id)
335 return fib6_get_table(net, id);
338 struct fib6_table *fib6_get_table(struct net *net, u32 id)
340 return net->ipv6.fib6_main_tbl;
343 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
344 const struct sk_buff *skb,
345 int flags, pol_lookup_t lookup)
349 rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, skb, flags);
350 if (rt->dst.error == -EAGAIN) {
352 rt = net->ipv6.ip6_null_entry;
359 /* called with rcu lock held; no reference taken on fib6_info */
360 struct fib6_info *fib6_lookup(struct net *net, int oif, struct flowi6 *fl6,
363 return fib6_table_lookup(net, net->ipv6.fib6_main_tbl, oif, fl6, flags);
366 static void __net_init fib6_tables_init(struct net *net)
368 fib6_link_table(net, net->ipv6.fib6_main_tbl);
373 unsigned int fib6_tables_seq_read(struct net *net)
375 unsigned int h, fib_seq = 0;
378 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
379 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
380 struct fib6_table *tb;
382 hlist_for_each_entry_rcu(tb, head, tb6_hlist)
383 fib_seq += tb->fib_seq;
390 static int call_fib6_entry_notifier(struct notifier_block *nb, struct net *net,
391 enum fib_event_type event_type,
392 struct fib6_info *rt)
394 struct fib6_entry_notifier_info info = {
398 return call_fib6_notifier(nb, net, event_type, &info.info);
401 static int call_fib6_entry_notifiers(struct net *net,
402 enum fib_event_type event_type,
403 struct fib6_info *rt,
404 struct netlink_ext_ack *extack)
406 struct fib6_entry_notifier_info info = {
407 .info.extack = extack,
411 rt->fib6_table->fib_seq++;
412 return call_fib6_notifiers(net, event_type, &info.info);
415 struct fib6_dump_arg {
417 struct notifier_block *nb;
420 static void fib6_rt_dump(struct fib6_info *rt, struct fib6_dump_arg *arg)
422 if (rt == arg->net->ipv6.fib6_null_entry)
424 call_fib6_entry_notifier(arg->nb, arg->net, FIB_EVENT_ENTRY_ADD, rt);
427 static int fib6_node_dump(struct fib6_walker *w)
429 struct fib6_info *rt;
431 for_each_fib6_walker_rt(w)
432 fib6_rt_dump(rt, w->args);
437 static void fib6_table_dump(struct net *net, struct fib6_table *tb,
438 struct fib6_walker *w)
440 w->root = &tb->tb6_root;
441 spin_lock_bh(&tb->tb6_lock);
443 spin_unlock_bh(&tb->tb6_lock);
446 /* Called with rcu_read_lock() */
447 int fib6_tables_dump(struct net *net, struct notifier_block *nb)
449 struct fib6_dump_arg arg;
450 struct fib6_walker *w;
453 w = kzalloc(sizeof(*w), GFP_ATOMIC);
457 w->func = fib6_node_dump;
462 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
463 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
464 struct fib6_table *tb;
466 hlist_for_each_entry_rcu(tb, head, tb6_hlist)
467 fib6_table_dump(net, tb, w);
475 static int fib6_dump_node(struct fib6_walker *w)
478 struct fib6_info *rt;
480 for_each_fib6_walker_rt(w) {
481 res = rt6_dump_route(rt, w->args);
483 /* Frame is full, suspend walking */
488 /* Multipath routes are dumped in one route with the
489 * RTA_MULTIPATH attribute. Jump 'rt' to point to the
490 * last sibling of this route (no need to dump the
491 * sibling routes again)
493 if (rt->fib6_nsiblings)
494 rt = list_last_entry(&rt->fib6_siblings,
502 static void fib6_dump_end(struct netlink_callback *cb)
504 struct net *net = sock_net(cb->skb->sk);
505 struct fib6_walker *w = (void *)cb->args[2];
510 fib6_walker_unlink(net, w);
515 cb->done = (void *)cb->args[3];
519 static int fib6_dump_done(struct netlink_callback *cb)
522 return cb->done ? cb->done(cb) : 0;
525 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
526 struct netlink_callback *cb)
528 struct net *net = sock_net(skb->sk);
529 struct fib6_walker *w;
532 w = (void *)cb->args[2];
533 w->root = &table->tb6_root;
535 if (cb->args[4] == 0) {
539 spin_lock_bh(&table->tb6_lock);
540 res = fib6_walk(net, w);
541 spin_unlock_bh(&table->tb6_lock);
544 cb->args[5] = w->root->fn_sernum;
547 if (cb->args[5] != w->root->fn_sernum) {
548 /* Begin at the root if the tree changed */
549 cb->args[5] = w->root->fn_sernum;
556 spin_lock_bh(&table->tb6_lock);
557 res = fib6_walk_continue(w);
558 spin_unlock_bh(&table->tb6_lock);
560 fib6_walker_unlink(net, w);
568 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
570 const struct nlmsghdr *nlh = cb->nlh;
571 struct net *net = sock_net(skb->sk);
572 struct rt6_rtnl_dump_arg arg = {};
574 unsigned int e = 0, s_e;
575 struct fib6_walker *w;
576 struct fib6_table *tb;
577 struct hlist_head *head;
580 if (cb->strict_check) {
583 err = ip_valid_fib_dump_req(net, nlh, &arg.filter, cb);
586 } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
587 struct rtmsg *rtm = nlmsg_data(nlh);
589 arg.filter.flags = rtm->rtm_flags & (RTM_F_PREFIX|RTM_F_CLONED);
592 /* fib entries are never clones */
593 if (arg.filter.flags & RTM_F_CLONED)
596 w = (void *)cb->args[2];
600 * 1. hook callback destructor.
602 cb->args[3] = (long)cb->done;
603 cb->done = fib6_dump_done;
606 * 2. allocate and initialize walker.
608 w = kzalloc(sizeof(*w), GFP_ATOMIC);
611 w->func = fib6_dump_node;
612 cb->args[2] = (long)w;
620 if (arg.filter.table_id) {
621 tb = fib6_get_table(net, arg.filter.table_id);
623 NL_SET_ERR_MSG_MOD(cb->extack, "FIB table does not exist");
627 res = fib6_dump_table(tb, skb, cb);
635 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
637 head = &net->ipv6.fib_table_hash[h];
638 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
641 res = fib6_dump_table(tb, skb, cb);
653 res = res < 0 ? res : skb->len;
659 void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val)
664 if (f6i->fib6_metrics == &dst_default_metrics) {
665 struct dst_metrics *p = kzalloc(sizeof(*p), GFP_ATOMIC);
670 refcount_set(&p->refcnt, 1);
671 f6i->fib6_metrics = p;
674 f6i->fib6_metrics->metrics[metric - 1] = val;
680 * return the appropriate node for a routing tree "add" operation
681 * by either creating and inserting or by returning an existing
685 static struct fib6_node *fib6_add_1(struct net *net,
686 struct fib6_table *table,
687 struct fib6_node *root,
688 struct in6_addr *addr, int plen,
689 int offset, int allow_create,
690 int replace_required,
691 struct netlink_ext_ack *extack)
693 struct fib6_node *fn, *in, *ln;
694 struct fib6_node *pn = NULL;
699 RT6_TRACE("fib6_add_1\n");
701 /* insert node in tree */
706 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
707 lockdep_is_held(&table->tb6_lock));
708 key = (struct rt6key *)((u8 *)leaf + offset);
713 if (plen < fn->fn_bit ||
714 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
716 if (replace_required) {
717 NL_SET_ERR_MSG(extack,
718 "Can not replace route - no match found");
719 pr_warn("Can't replace route, no match found\n");
720 return ERR_PTR(-ENOENT);
722 pr_warn("NLM_F_CREATE should be set when creating new route\n");
731 if (plen == fn->fn_bit) {
732 /* clean up an intermediate node */
733 if (!(fn->fn_flags & RTN_RTINFO)) {
734 RCU_INIT_POINTER(fn->leaf, NULL);
735 fib6_info_release(leaf);
736 /* remove null_entry in the root node */
737 } else if (fn->fn_flags & RTN_TL_ROOT &&
738 rcu_access_pointer(fn->leaf) ==
739 net->ipv6.fib6_null_entry) {
740 RCU_INIT_POINTER(fn->leaf, NULL);
747 * We have more bits to go
750 /* Try to walk down on tree. */
751 dir = addr_bit_set(addr, fn->fn_bit);
754 rcu_dereference_protected(fn->right,
755 lockdep_is_held(&table->tb6_lock)) :
756 rcu_dereference_protected(fn->left,
757 lockdep_is_held(&table->tb6_lock));
761 /* We should not create new node because
762 * NLM_F_REPLACE was specified without NLM_F_CREATE
763 * I assume it is safe to require NLM_F_CREATE when
764 * REPLACE flag is used! Later we may want to remove the
765 * check for replace_required, because according
766 * to netlink specification, NLM_F_CREATE
767 * MUST be specified if new route is created.
768 * That would keep IPv6 consistent with IPv4
770 if (replace_required) {
771 NL_SET_ERR_MSG(extack,
772 "Can not replace route - no match found");
773 pr_warn("Can't replace route, no match found\n");
774 return ERR_PTR(-ENOENT);
776 pr_warn("NLM_F_CREATE should be set when creating new route\n");
779 * We walked to the bottom of tree.
780 * Create new leaf node without children.
783 ln = node_alloc(net);
786 return ERR_PTR(-ENOMEM);
788 RCU_INIT_POINTER(ln->parent, pn);
791 rcu_assign_pointer(pn->right, ln);
793 rcu_assign_pointer(pn->left, ln);
800 * split since we don't have a common prefix anymore or
801 * we have a less significant route.
802 * we've to insert an intermediate node on the list
803 * this new node will point to the one we need to create
807 pn = rcu_dereference_protected(fn->parent,
808 lockdep_is_held(&table->tb6_lock));
810 /* find 1st bit in difference between the 2 addrs.
812 See comment in __ipv6_addr_diff: bit may be an invalid value,
813 but if it is >= plen, the value is ignored in any case.
816 bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
821 * (new leaf node)[ln] (old node)[fn]
824 in = node_alloc(net);
825 ln = node_alloc(net);
829 node_free_immediate(net, in);
831 node_free_immediate(net, ln);
832 return ERR_PTR(-ENOMEM);
836 * new intermediate node.
838 * be off since that an address that chooses one of
839 * the branches would not match less specific routes
840 * in the other branch
845 RCU_INIT_POINTER(in->parent, pn);
847 atomic_inc(&rcu_dereference_protected(in->leaf,
848 lockdep_is_held(&table->tb6_lock))->fib6_ref);
850 /* update parent pointer */
852 rcu_assign_pointer(pn->right, in);
854 rcu_assign_pointer(pn->left, in);
858 RCU_INIT_POINTER(ln->parent, in);
859 rcu_assign_pointer(fn->parent, in);
861 if (addr_bit_set(addr, bit)) {
862 rcu_assign_pointer(in->right, ln);
863 rcu_assign_pointer(in->left, fn);
865 rcu_assign_pointer(in->left, ln);
866 rcu_assign_pointer(in->right, fn);
868 } else { /* plen <= bit */
871 * (new leaf node)[ln]
873 * (old node)[fn] NULL
876 ln = node_alloc(net);
879 return ERR_PTR(-ENOMEM);
883 RCU_INIT_POINTER(ln->parent, pn);
885 if (addr_bit_set(&key->addr, plen))
886 RCU_INIT_POINTER(ln->right, fn);
888 RCU_INIT_POINTER(ln->left, fn);
890 rcu_assign_pointer(fn->parent, ln);
893 rcu_assign_pointer(pn->right, ln);
895 rcu_assign_pointer(pn->left, ln);
900 static void fib6_drop_pcpu_from(struct fib6_info *f6i,
901 const struct fib6_table *table)
905 /* release the reference to this fib entry from
906 * all of its cached pcpu routes
908 for_each_possible_cpu(cpu) {
909 struct rt6_info **ppcpu_rt;
910 struct rt6_info *pcpu_rt;
912 ppcpu_rt = per_cpu_ptr(f6i->rt6i_pcpu, cpu);
915 struct fib6_info *from;
917 from = rcu_dereference_protected(pcpu_rt->from,
918 lockdep_is_held(&table->tb6_lock));
919 rcu_assign_pointer(pcpu_rt->from, NULL);
920 fib6_info_release(from);
925 static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn,
928 struct fib6_table *table = rt->fib6_table;
930 if (atomic_read(&rt->fib6_ref) != 1) {
931 /* This route is used as dummy address holder in some split
932 * nodes. It is not leaked, but it still holds other resources,
933 * which must be released in time. So, scan ascendant nodes
934 * and replace dummy references to this route with references
935 * to still alive ones.
938 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
939 lockdep_is_held(&table->tb6_lock));
940 struct fib6_info *new_leaf;
941 if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
942 new_leaf = fib6_find_prefix(net, table, fn);
943 atomic_inc(&new_leaf->fib6_ref);
945 rcu_assign_pointer(fn->leaf, new_leaf);
946 fib6_info_release(rt);
948 fn = rcu_dereference_protected(fn->parent,
949 lockdep_is_held(&table->tb6_lock));
953 fib6_drop_pcpu_from(rt, table);
958 * Insert routing information in a node.
961 static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt,
962 struct nl_info *info,
963 struct netlink_ext_ack *extack)
965 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
966 lockdep_is_held(&rt->fib6_table->tb6_lock));
967 struct fib6_info *iter = NULL;
968 struct fib6_info __rcu **ins;
969 struct fib6_info __rcu **fallback_ins = NULL;
970 int replace = (info->nlh &&
971 (info->nlh->nlmsg_flags & NLM_F_REPLACE));
972 int add = (!info->nlh ||
973 (info->nlh->nlmsg_flags & NLM_F_CREATE));
975 bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
976 u16 nlflags = NLM_F_EXCL;
979 if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
980 nlflags |= NLM_F_APPEND;
984 for (iter = leaf; iter;
985 iter = rcu_dereference_protected(iter->fib6_next,
986 lockdep_is_held(&rt->fib6_table->tb6_lock))) {
988 * Search for duplicates
991 if (iter->fib6_metric == rt->fib6_metric) {
993 * Same priority level
996 (info->nlh->nlmsg_flags & NLM_F_EXCL))
999 nlflags &= ~NLM_F_EXCL;
1001 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
1006 fallback_ins = fallback_ins ?: ins;
1010 if (rt6_duplicate_nexthop(iter, rt)) {
1011 if (rt->fib6_nsiblings)
1012 rt->fib6_nsiblings = 0;
1013 if (!(iter->fib6_flags & RTF_EXPIRES))
1015 if (!(rt->fib6_flags & RTF_EXPIRES))
1016 fib6_clean_expires(iter);
1018 fib6_set_expires(iter, rt->expires);
1021 fib6_metric_set(iter, RTAX_MTU,
1025 /* If we have the same destination and the same metric,
1026 * but not the same gateway, then the route we try to
1027 * add is sibling to this route, increment our counter
1028 * of siblings, and later we will add our route to the
1030 * Only static routes (which don't have flag
1031 * RTF_EXPIRES) are used for ECMPv6.
1033 * To avoid long list, we only had siblings if the
1034 * route have a gateway.
1037 rt6_qualify_for_ecmp(iter))
1038 rt->fib6_nsiblings++;
1041 if (iter->fib6_metric > rt->fib6_metric)
1045 ins = &iter->fib6_next;
1048 if (fallback_ins && !found) {
1049 /* No ECMP-able route found, replace first non-ECMP one */
1051 iter = rcu_dereference_protected(*ins,
1052 lockdep_is_held(&rt->fib6_table->tb6_lock));
1056 /* Reset round-robin state, if necessary */
1057 if (ins == &fn->leaf)
1060 /* Link this route to others same route. */
1061 if (rt->fib6_nsiblings) {
1062 unsigned int fib6_nsiblings;
1063 struct fib6_info *sibling, *temp_sibling;
1065 /* Find the first route that have the same metric */
1068 if (sibling->fib6_metric == rt->fib6_metric &&
1069 rt6_qualify_for_ecmp(sibling)) {
1070 list_add_tail(&rt->fib6_siblings,
1071 &sibling->fib6_siblings);
1074 sibling = rcu_dereference_protected(sibling->fib6_next,
1075 lockdep_is_held(&rt->fib6_table->tb6_lock));
1077 /* For each sibling in the list, increment the counter of
1078 * siblings. BUG() if counters does not match, list of siblings
1082 list_for_each_entry_safe(sibling, temp_sibling,
1083 &rt->fib6_siblings, fib6_siblings) {
1084 sibling->fib6_nsiblings++;
1085 BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings);
1088 BUG_ON(fib6_nsiblings != rt->fib6_nsiblings);
1089 rt6_multipath_rebalance(temp_sibling);
1097 pr_warn("NLM_F_CREATE should be set when creating new route\n");
1100 nlflags |= NLM_F_CREATE;
1102 err = call_fib6_entry_notifiers(info->nl_net,
1103 FIB_EVENT_ENTRY_ADD,
1108 rcu_assign_pointer(rt->fib6_next, iter);
1109 atomic_inc(&rt->fib6_ref);
1110 rcu_assign_pointer(rt->fib6_node, fn);
1111 rcu_assign_pointer(*ins, rt);
1112 if (!info->skip_notify)
1113 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1114 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1116 if (!(fn->fn_flags & RTN_RTINFO)) {
1117 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1118 fn->fn_flags |= RTN_RTINFO;
1127 pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1131 err = call_fib6_entry_notifiers(info->nl_net,
1132 FIB_EVENT_ENTRY_REPLACE,
1137 atomic_inc(&rt->fib6_ref);
1138 rcu_assign_pointer(rt->fib6_node, fn);
1139 rt->fib6_next = iter->fib6_next;
1140 rcu_assign_pointer(*ins, rt);
1141 if (!info->skip_notify)
1142 inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1143 if (!(fn->fn_flags & RTN_RTINFO)) {
1144 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1145 fn->fn_flags |= RTN_RTINFO;
1147 nsiblings = iter->fib6_nsiblings;
1148 iter->fib6_node = NULL;
1149 fib6_purge_rt(iter, fn, info->nl_net);
1150 if (rcu_access_pointer(fn->rr_ptr) == iter)
1152 fib6_info_release(iter);
1155 /* Replacing an ECMP route, remove all siblings */
1156 ins = &rt->fib6_next;
1157 iter = rcu_dereference_protected(*ins,
1158 lockdep_is_held(&rt->fib6_table->tb6_lock));
1160 if (iter->fib6_metric > rt->fib6_metric)
1162 if (rt6_qualify_for_ecmp(iter)) {
1163 *ins = iter->fib6_next;
1164 iter->fib6_node = NULL;
1165 fib6_purge_rt(iter, fn, info->nl_net);
1166 if (rcu_access_pointer(fn->rr_ptr) == iter)
1168 fib6_info_release(iter);
1170 info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1172 ins = &iter->fib6_next;
1174 iter = rcu_dereference_protected(*ins,
1175 lockdep_is_held(&rt->fib6_table->tb6_lock));
1177 WARN_ON(nsiblings != 0);
1184 static void fib6_start_gc(struct net *net, struct fib6_info *rt)
1186 if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1187 (rt->fib6_flags & RTF_EXPIRES))
1188 mod_timer(&net->ipv6.ip6_fib_timer,
1189 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1192 void fib6_force_start_gc(struct net *net)
1194 if (!timer_pending(&net->ipv6.ip6_fib_timer))
1195 mod_timer(&net->ipv6.ip6_fib_timer,
1196 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1199 static void __fib6_update_sernum_upto_root(struct fib6_info *rt,
1202 struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1203 lockdep_is_held(&rt->fib6_table->tb6_lock));
1205 /* paired with smp_rmb() in rt6_get_cookie_safe() */
1208 fn->fn_sernum = sernum;
1209 fn = rcu_dereference_protected(fn->parent,
1210 lockdep_is_held(&rt->fib6_table->tb6_lock));
1214 void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt)
1216 __fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1220 * Add routing information to the routing tree.
1221 * <destination addr>/<source addr>
1222 * with source addr info in sub-trees
1223 * Need to own table->tb6_lock
1226 int fib6_add(struct fib6_node *root, struct fib6_info *rt,
1227 struct nl_info *info, struct netlink_ext_ack *extack)
1229 struct fib6_table *table = rt->fib6_table;
1230 struct fib6_node *fn, *pn = NULL;
1232 int allow_create = 1;
1233 int replace_required = 0;
1234 int sernum = fib6_new_sernum(info->nl_net);
1237 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1239 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1240 replace_required = 1;
1242 if (!allow_create && !replace_required)
1243 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1245 fn = fib6_add_1(info->nl_net, table, root,
1246 &rt->fib6_dst.addr, rt->fib6_dst.plen,
1247 offsetof(struct fib6_info, fib6_dst), allow_create,
1248 replace_required, extack);
1257 #ifdef CONFIG_IPV6_SUBTREES
1258 if (rt->fib6_src.plen) {
1259 struct fib6_node *sn;
1261 if (!rcu_access_pointer(fn->subtree)) {
1262 struct fib6_node *sfn;
1274 /* Create subtree root node */
1275 sfn = node_alloc(info->nl_net);
1279 atomic_inc(&info->nl_net->ipv6.fib6_null_entry->fib6_ref);
1280 rcu_assign_pointer(sfn->leaf,
1281 info->nl_net->ipv6.fib6_null_entry);
1282 sfn->fn_flags = RTN_ROOT;
1284 /* Now add the first leaf node to new subtree */
1286 sn = fib6_add_1(info->nl_net, table, sfn,
1287 &rt->fib6_src.addr, rt->fib6_src.plen,
1288 offsetof(struct fib6_info, fib6_src),
1289 allow_create, replace_required, extack);
1292 /* If it is failed, discard just allocated
1293 root, and then (in failure) stale node
1296 node_free_immediate(info->nl_net, sfn);
1301 /* Now link new subtree to main tree */
1302 rcu_assign_pointer(sfn->parent, fn);
1303 rcu_assign_pointer(fn->subtree, sfn);
1305 sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1306 &rt->fib6_src.addr, rt->fib6_src.plen,
1307 offsetof(struct fib6_info, fib6_src),
1308 allow_create, replace_required, extack);
1316 if (!rcu_access_pointer(fn->leaf)) {
1317 if (fn->fn_flags & RTN_TL_ROOT) {
1318 /* put back null_entry for root node */
1319 rcu_assign_pointer(fn->leaf,
1320 info->nl_net->ipv6.fib6_null_entry);
1322 atomic_inc(&rt->fib6_ref);
1323 rcu_assign_pointer(fn->leaf, rt);
1330 err = fib6_add_rt2node(fn, rt, info, extack);
1332 __fib6_update_sernum_upto_root(rt, sernum);
1333 fib6_start_gc(info->nl_net, rt);
1338 #ifdef CONFIG_IPV6_SUBTREES
1340 * If fib6_add_1 has cleared the old leaf pointer in the
1341 * super-tree leaf node we have to find a new one for it.
1344 struct fib6_info *pn_leaf =
1345 rcu_dereference_protected(pn->leaf,
1346 lockdep_is_held(&table->tb6_lock));
1347 if (pn_leaf == rt) {
1349 RCU_INIT_POINTER(pn->leaf, NULL);
1350 fib6_info_release(rt);
1352 if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1353 pn_leaf = fib6_find_prefix(info->nl_net, table,
1359 info->nl_net->ipv6.fib6_null_entry;
1362 fib6_info_hold(pn_leaf);
1363 rcu_assign_pointer(pn->leaf, pn_leaf);
1372 /* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1373 * 1. fn is an intermediate node and we failed to add the new
1374 * route to it in both subtree creation failure and fib6_add_rt2node()
1376 * 2. fn is the root node in the table and we fail to add the first
1377 * default route to it.
1380 (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1381 (fn->fn_flags & RTN_TL_ROOT &&
1382 !rcu_access_pointer(fn->leaf))))
1383 fib6_repair_tree(info->nl_net, table, fn);
1388 * Routing tree lookup
1392 struct lookup_args {
1393 int offset; /* key offset on fib6_info */
1394 const struct in6_addr *addr; /* search key */
1397 static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root,
1398 struct lookup_args *args)
1400 struct fib6_node *fn;
1403 if (unlikely(args->offset == 0))
1413 struct fib6_node *next;
1415 dir = addr_bit_set(args->addr, fn->fn_bit);
1417 next = dir ? rcu_dereference(fn->right) :
1418 rcu_dereference(fn->left);
1428 struct fib6_node *subtree = FIB6_SUBTREE(fn);
1430 if (subtree || fn->fn_flags & RTN_RTINFO) {
1431 struct fib6_info *leaf = rcu_dereference(fn->leaf);
1437 key = (struct rt6key *) ((u8 *)leaf + args->offset);
1439 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1440 #ifdef CONFIG_IPV6_SUBTREES
1442 struct fib6_node *sfn;
1443 sfn = fib6_node_lookup_1(subtree,
1450 if (fn->fn_flags & RTN_RTINFO)
1455 if (fn->fn_flags & RTN_ROOT)
1458 fn = rcu_dereference(fn->parent);
1464 /* called with rcu_read_lock() held
1466 struct fib6_node *fib6_node_lookup(struct fib6_node *root,
1467 const struct in6_addr *daddr,
1468 const struct in6_addr *saddr)
1470 struct fib6_node *fn;
1471 struct lookup_args args[] = {
1473 .offset = offsetof(struct fib6_info, fib6_dst),
1476 #ifdef CONFIG_IPV6_SUBTREES
1478 .offset = offsetof(struct fib6_info, fib6_src),
1483 .offset = 0, /* sentinel */
1487 fn = fib6_node_lookup_1(root, daddr ? args : args + 1);
1488 if (!fn || fn->fn_flags & RTN_TL_ROOT)
1495 * Get node with specified destination prefix (and source prefix,
1496 * if subtrees are used)
1497 * exact_match == true means we try to find fn with exact match of
1498 * the passed in prefix addr
1499 * exact_match == false means we try to find fn with longest prefix
1500 * match of the passed in prefix addr. This is useful for finding fn
1501 * for cached route as it will be stored in the exception table under
1502 * the node with longest prefix length.
1506 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1507 const struct in6_addr *addr,
1508 int plen, int offset,
1511 struct fib6_node *fn, *prev = NULL;
1513 for (fn = root; fn ; ) {
1514 struct fib6_info *leaf = rcu_dereference(fn->leaf);
1517 /* This node is being deleted */
1519 if (plen <= fn->fn_bit)
1525 key = (struct rt6key *)((u8 *)leaf + offset);
1530 if (plen < fn->fn_bit ||
1531 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1534 if (plen == fn->fn_bit)
1541 * We have more bits to go
1543 if (addr_bit_set(addr, fn->fn_bit))
1544 fn = rcu_dereference(fn->right);
1546 fn = rcu_dereference(fn->left);
1555 struct fib6_node *fib6_locate(struct fib6_node *root,
1556 const struct in6_addr *daddr, int dst_len,
1557 const struct in6_addr *saddr, int src_len,
1560 struct fib6_node *fn;
1562 fn = fib6_locate_1(root, daddr, dst_len,
1563 offsetof(struct fib6_info, fib6_dst),
1566 #ifdef CONFIG_IPV6_SUBTREES
1568 WARN_ON(saddr == NULL);
1570 struct fib6_node *subtree = FIB6_SUBTREE(fn);
1573 fn = fib6_locate_1(subtree, saddr, src_len,
1574 offsetof(struct fib6_info, fib6_src),
1581 if (fn && fn->fn_flags & RTN_RTINFO)
1593 static struct fib6_info *fib6_find_prefix(struct net *net,
1594 struct fib6_table *table,
1595 struct fib6_node *fn)
1597 struct fib6_node *child_left, *child_right;
1599 if (fn->fn_flags & RTN_ROOT)
1600 return net->ipv6.fib6_null_entry;
1603 child_left = rcu_dereference_protected(fn->left,
1604 lockdep_is_held(&table->tb6_lock));
1605 child_right = rcu_dereference_protected(fn->right,
1606 lockdep_is_held(&table->tb6_lock));
1608 return rcu_dereference_protected(child_left->leaf,
1609 lockdep_is_held(&table->tb6_lock));
1611 return rcu_dereference_protected(child_right->leaf,
1612 lockdep_is_held(&table->tb6_lock));
1614 fn = FIB6_SUBTREE(fn);
1620 * Called to trim the tree of intermediate nodes when possible. "fn"
1621 * is the node we want to try and remove.
1622 * Need to own table->tb6_lock
1625 static struct fib6_node *fib6_repair_tree(struct net *net,
1626 struct fib6_table *table,
1627 struct fib6_node *fn)
1631 struct fib6_node *child;
1632 struct fib6_walker *w;
1635 /* Set fn->leaf to null_entry for root node. */
1636 if (fn->fn_flags & RTN_TL_ROOT) {
1637 rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry);
1642 struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1643 lockdep_is_held(&table->tb6_lock));
1644 struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1645 lockdep_is_held(&table->tb6_lock));
1646 struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1647 lockdep_is_held(&table->tb6_lock));
1648 struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1649 lockdep_is_held(&table->tb6_lock));
1650 struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1651 lockdep_is_held(&table->tb6_lock));
1652 struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1653 lockdep_is_held(&table->tb6_lock));
1654 struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1655 lockdep_is_held(&table->tb6_lock));
1656 struct fib6_info *new_fn_leaf;
1658 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1661 WARN_ON(fn->fn_flags & RTN_RTINFO);
1662 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1668 child = fn_r, children |= 1;
1670 child = fn_l, children |= 2;
1672 if (children == 3 || FIB6_SUBTREE(fn)
1673 #ifdef CONFIG_IPV6_SUBTREES
1674 /* Subtree root (i.e. fn) may have one child */
1675 || (children && fn->fn_flags & RTN_ROOT)
1678 new_fn_leaf = fib6_find_prefix(net, table, fn);
1681 WARN_ON(!new_fn_leaf);
1682 new_fn_leaf = net->ipv6.fib6_null_entry;
1685 fib6_info_hold(new_fn_leaf);
1686 rcu_assign_pointer(fn->leaf, new_fn_leaf);
1690 #ifdef CONFIG_IPV6_SUBTREES
1691 if (FIB6_SUBTREE(pn) == fn) {
1692 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1693 RCU_INIT_POINTER(pn->subtree, NULL);
1696 WARN_ON(fn->fn_flags & RTN_ROOT);
1699 rcu_assign_pointer(pn->right, child);
1700 else if (pn_l == fn)
1701 rcu_assign_pointer(pn->left, child);
1707 rcu_assign_pointer(child->parent, pn);
1709 #ifdef CONFIG_IPV6_SUBTREES
1713 read_lock(&net->ipv6.fib6_walker_lock);
1714 FOR_WALKERS(net, w) {
1716 if (w->node == fn) {
1717 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1722 if (w->node == fn) {
1725 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1726 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1728 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1729 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1734 read_unlock(&net->ipv6.fib6_walker_lock);
1737 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1740 RCU_INIT_POINTER(pn->leaf, NULL);
1741 fib6_info_release(pn_leaf);
1746 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1747 struct fib6_info __rcu **rtp, struct nl_info *info)
1749 struct fib6_walker *w;
1750 struct fib6_info *rt = rcu_dereference_protected(*rtp,
1751 lockdep_is_held(&table->tb6_lock));
1752 struct net *net = info->nl_net;
1754 RT6_TRACE("fib6_del_route\n");
1757 *rtp = rt->fib6_next;
1758 rt->fib6_node = NULL;
1759 net->ipv6.rt6_stats->fib_rt_entries--;
1760 net->ipv6.rt6_stats->fib_discarded_routes++;
1762 /* Flush all cached dst in exception table */
1763 rt6_flush_exceptions(rt);
1765 /* Reset round-robin state, if necessary */
1766 if (rcu_access_pointer(fn->rr_ptr) == rt)
1769 /* Remove this entry from other siblings */
1770 if (rt->fib6_nsiblings) {
1771 struct fib6_info *sibling, *next_sibling;
1773 list_for_each_entry_safe(sibling, next_sibling,
1774 &rt->fib6_siblings, fib6_siblings)
1775 sibling->fib6_nsiblings--;
1776 rt->fib6_nsiblings = 0;
1777 list_del_init(&rt->fib6_siblings);
1778 rt6_multipath_rebalance(next_sibling);
1781 /* Adjust walkers */
1782 read_lock(&net->ipv6.fib6_walker_lock);
1783 FOR_WALKERS(net, w) {
1784 if (w->state == FWS_C && w->leaf == rt) {
1785 RT6_TRACE("walker %p adjusted by delroute\n", w);
1786 w->leaf = rcu_dereference_protected(rt->fib6_next,
1787 lockdep_is_held(&table->tb6_lock));
1792 read_unlock(&net->ipv6.fib6_walker_lock);
1794 /* If it was last route, call fib6_repair_tree() to:
1795 * 1. For root node, put back null_entry as how the table was created.
1796 * 2. For other nodes, expunge its radix tree node.
1798 if (!rcu_access_pointer(fn->leaf)) {
1799 if (!(fn->fn_flags & RTN_TL_ROOT)) {
1800 fn->fn_flags &= ~RTN_RTINFO;
1801 net->ipv6.rt6_stats->fib_route_nodes--;
1803 fn = fib6_repair_tree(net, table, fn);
1806 fib6_purge_rt(rt, fn, net);
1808 call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, rt, NULL);
1809 if (!info->skip_notify)
1810 inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1811 fib6_info_release(rt);
1814 /* Need to own table->tb6_lock */
1815 int fib6_del(struct fib6_info *rt, struct nl_info *info)
1817 struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1818 lockdep_is_held(&rt->fib6_table->tb6_lock));
1819 struct fib6_table *table = rt->fib6_table;
1820 struct net *net = info->nl_net;
1821 struct fib6_info __rcu **rtp;
1822 struct fib6_info __rcu **rtp_next;
1824 if (!fn || rt == net->ipv6.fib6_null_entry)
1827 WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1830 * Walk the leaf entries looking for ourself
1833 for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
1834 struct fib6_info *cur = rcu_dereference_protected(*rtp,
1835 lockdep_is_held(&table->tb6_lock));
1837 fib6_del_route(table, fn, rtp, info);
1840 rtp_next = &cur->fib6_next;
1846 * Tree traversal function.
1848 * Certainly, it is not interrupt safe.
1849 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1850 * It means, that we can modify tree during walking
1851 * and use this function for garbage collection, clone pruning,
1852 * cleaning tree when a device goes down etc. etc.
1854 * It guarantees that every node will be traversed,
1855 * and that it will be traversed only once.
1857 * Callback function w->func may return:
1858 * 0 -> continue walking.
1859 * positive value -> walking is suspended (used by tree dumps,
1860 * and probably by gc, if it will be split to several slices)
1861 * negative value -> terminate walking.
1863 * The function itself returns:
1864 * 0 -> walk is complete.
1865 * >0 -> walk is incomplete (i.e. suspended)
1866 * <0 -> walk is terminated by an error.
1868 * This function is called with tb6_lock held.
1871 static int fib6_walk_continue(struct fib6_walker *w)
1873 struct fib6_node *fn, *pn, *left, *right;
1875 /* w->root should always be table->tb6_root */
1876 WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
1884 #ifdef CONFIG_IPV6_SUBTREES
1886 if (FIB6_SUBTREE(fn)) {
1887 w->node = FIB6_SUBTREE(fn);
1894 left = rcu_dereference_protected(fn->left, 1);
1897 w->state = FWS_INIT;
1903 right = rcu_dereference_protected(fn->right, 1);
1906 w->state = FWS_INIT;
1910 w->leaf = rcu_dereference_protected(fn->leaf, 1);
1913 if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1934 pn = rcu_dereference_protected(fn->parent, 1);
1935 left = rcu_dereference_protected(pn->left, 1);
1936 right = rcu_dereference_protected(pn->right, 1);
1938 #ifdef CONFIG_IPV6_SUBTREES
1939 if (FIB6_SUBTREE(pn) == fn) {
1940 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1951 w->leaf = rcu_dereference_protected(w->node->leaf, 1);
1961 static int fib6_walk(struct net *net, struct fib6_walker *w)
1965 w->state = FWS_INIT;
1968 fib6_walker_link(net, w);
1969 res = fib6_walk_continue(w);
1971 fib6_walker_unlink(net, w);
1975 static int fib6_clean_node(struct fib6_walker *w)
1978 struct fib6_info *rt;
1979 struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1980 struct nl_info info = {
1982 .skip_notify = c->skip_notify,
1985 if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1986 w->node->fn_sernum != c->sernum)
1987 w->node->fn_sernum = c->sernum;
1990 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1995 for_each_fib6_walker_rt(w) {
1996 res = c->func(rt, c->arg);
1999 res = fib6_del(rt, &info);
2002 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
2004 rcu_access_pointer(rt->fib6_node),
2010 } else if (res == -2) {
2011 if (WARN_ON(!rt->fib6_nsiblings))
2013 rt = list_last_entry(&rt->fib6_siblings,
2014 struct fib6_info, fib6_siblings);
2024 * Convenient frontend to tree walker.
2026 * func is called on each route.
2027 * It may return -2 -> skip multipath route.
2028 * -1 -> delete this route.
2029 * 0 -> continue walking
2032 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
2033 int (*func)(struct fib6_info *, void *arg),
2034 int sernum, void *arg, bool skip_notify)
2036 struct fib6_cleaner c;
2039 c.w.func = fib6_clean_node;
2046 c.skip_notify = skip_notify;
2048 fib6_walk(net, &c.w);
2051 static void __fib6_clean_all(struct net *net,
2052 int (*func)(struct fib6_info *, void *),
2053 int sernum, void *arg, bool skip_notify)
2055 struct fib6_table *table;
2056 struct hlist_head *head;
2060 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2061 head = &net->ipv6.fib_table_hash[h];
2062 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2063 spin_lock_bh(&table->tb6_lock);
2064 fib6_clean_tree(net, &table->tb6_root,
2065 func, sernum, arg, skip_notify);
2066 spin_unlock_bh(&table->tb6_lock);
2072 void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *),
2075 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, false);
2078 void fib6_clean_all_skip_notify(struct net *net,
2079 int (*func)(struct fib6_info *, void *),
2082 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, true);
2085 static void fib6_flush_trees(struct net *net)
2087 int new_sernum = fib6_new_sernum(net);
2089 __fib6_clean_all(net, NULL, new_sernum, NULL, false);
2093 * Garbage collection
2096 static int fib6_age(struct fib6_info *rt, void *arg)
2098 struct fib6_gc_args *gc_args = arg;
2099 unsigned long now = jiffies;
2102 * check addrconf expiration here.
2103 * Routes are expired even if they are in use.
2106 if (rt->fib6_flags & RTF_EXPIRES && rt->expires) {
2107 if (time_after(now, rt->expires)) {
2108 RT6_TRACE("expiring %p\n", rt);
2114 /* Also age clones in the exception table.
2115 * Note, that clones are aged out
2116 * only if they are not in use now.
2118 rt6_age_exceptions(rt, gc_args, now);
2123 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2125 struct fib6_gc_args gc_args;
2129 spin_lock_bh(&net->ipv6.fib6_gc_lock);
2130 } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2131 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2134 gc_args.timeout = expires ? (int)expires :
2135 net->ipv6.sysctl.ip6_rt_gc_interval;
2138 fib6_clean_all(net, fib6_age, &gc_args);
2140 net->ipv6.ip6_rt_last_gc = now;
2143 mod_timer(&net->ipv6.ip6_fib_timer,
2145 + net->ipv6.sysctl.ip6_rt_gc_interval));
2147 del_timer(&net->ipv6.ip6_fib_timer);
2148 spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2151 static void fib6_gc_timer_cb(struct timer_list *t)
2153 struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2155 fib6_run_gc(0, arg, true);
2158 static int __net_init fib6_net_init(struct net *net)
2160 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2163 err = fib6_notifier_init(net);
2167 spin_lock_init(&net->ipv6.fib6_gc_lock);
2168 rwlock_init(&net->ipv6.fib6_walker_lock);
2169 INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2170 timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2172 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2173 if (!net->ipv6.rt6_stats)
2176 /* Avoid false sharing : Use at least a full cache line */
2177 size = max_t(size_t, size, L1_CACHE_BYTES);
2179 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2180 if (!net->ipv6.fib_table_hash)
2183 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2185 if (!net->ipv6.fib6_main_tbl)
2186 goto out_fib_table_hash;
2188 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2189 rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2190 net->ipv6.fib6_null_entry);
2191 net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2192 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2193 inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2195 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2196 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2198 if (!net->ipv6.fib6_local_tbl)
2199 goto out_fib6_main_tbl;
2200 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2201 rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2202 net->ipv6.fib6_null_entry);
2203 net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2204 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2205 inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2207 fib6_tables_init(net);
2211 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2213 kfree(net->ipv6.fib6_main_tbl);
2216 kfree(net->ipv6.fib_table_hash);
2218 kfree(net->ipv6.rt6_stats);
2220 fib6_notifier_exit(net);
2224 static void fib6_net_exit(struct net *net)
2228 del_timer_sync(&net->ipv6.ip6_fib_timer);
2230 for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2231 struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2232 struct hlist_node *tmp;
2233 struct fib6_table *tb;
2235 hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2236 hlist_del(&tb->tb6_hlist);
2237 fib6_free_table(tb);
2241 kfree(net->ipv6.fib_table_hash);
2242 kfree(net->ipv6.rt6_stats);
2243 fib6_notifier_exit(net);
2246 static struct pernet_operations fib6_net_ops = {
2247 .init = fib6_net_init,
2248 .exit = fib6_net_exit,
2251 int __init fib6_init(void)
2255 fib6_node_kmem = kmem_cache_create("fib6_nodes",
2256 sizeof(struct fib6_node),
2257 0, SLAB_HWCACHE_ALIGN,
2259 if (!fib6_node_kmem)
2262 ret = register_pernet_subsys(&fib6_net_ops);
2264 goto out_kmem_cache_create;
2266 ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2269 goto out_unregister_subsys;
2271 __fib6_flush_trees = fib6_flush_trees;
2275 out_unregister_subsys:
2276 unregister_pernet_subsys(&fib6_net_ops);
2277 out_kmem_cache_create:
2278 kmem_cache_destroy(fib6_node_kmem);
2282 void fib6_gc_cleanup(void)
2284 unregister_pernet_subsys(&fib6_net_ops);
2285 kmem_cache_destroy(fib6_node_kmem);
2288 #ifdef CONFIG_PROC_FS
2289 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2291 struct fib6_info *rt = v;
2292 struct ipv6_route_iter *iter = seq->private;
2293 const struct net_device *dev;
2295 seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
2297 #ifdef CONFIG_IPV6_SUBTREES
2298 seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen);
2300 seq_puts(seq, "00000000000000000000000000000000 00 ");
2302 if (rt->fib6_flags & RTF_GATEWAY)
2303 seq_printf(seq, "%pi6", &rt->fib6_nh.nh_gw);
2305 seq_puts(seq, "00000000000000000000000000000000");
2307 dev = rt->fib6_nh.nh_dev;
2308 seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2309 rt->fib6_metric, atomic_read(&rt->fib6_ref), 0,
2310 rt->fib6_flags, dev ? dev->name : "");
2311 iter->w.leaf = NULL;
2315 static int ipv6_route_yield(struct fib6_walker *w)
2317 struct ipv6_route_iter *iter = w->args;
2323 iter->w.leaf = rcu_dereference_protected(
2324 iter->w.leaf->fib6_next,
2325 lockdep_is_held(&iter->tbl->tb6_lock));
2327 if (!iter->skip && iter->w.leaf)
2329 } while (iter->w.leaf);
2334 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2337 memset(&iter->w, 0, sizeof(iter->w));
2338 iter->w.func = ipv6_route_yield;
2339 iter->w.root = &iter->tbl->tb6_root;
2340 iter->w.state = FWS_INIT;
2341 iter->w.node = iter->w.root;
2342 iter->w.args = iter;
2343 iter->sernum = iter->w.root->fn_sernum;
2344 INIT_LIST_HEAD(&iter->w.lh);
2345 fib6_walker_link(net, &iter->w);
2348 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2352 struct hlist_node *node;
2355 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2356 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2362 while (!node && h < FIB6_TABLE_HASHSZ) {
2363 node = rcu_dereference_bh(
2364 hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2366 return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2369 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2371 if (iter->sernum != iter->w.root->fn_sernum) {
2372 iter->sernum = iter->w.root->fn_sernum;
2373 iter->w.state = FWS_INIT;
2374 iter->w.node = iter->w.root;
2375 WARN_ON(iter->w.skip);
2376 iter->w.skip = iter->w.count;
2380 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2383 struct fib6_info *n;
2384 struct net *net = seq_file_net(seq);
2385 struct ipv6_route_iter *iter = seq->private;
2390 n = rcu_dereference_bh(((struct fib6_info *)v)->fib6_next);
2397 ipv6_route_check_sernum(iter);
2398 spin_lock_bh(&iter->tbl->tb6_lock);
2399 r = fib6_walk_continue(&iter->w);
2400 spin_unlock_bh(&iter->tbl->tb6_lock);
2404 return iter->w.leaf;
2406 fib6_walker_unlink(net, &iter->w);
2409 fib6_walker_unlink(net, &iter->w);
2411 iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2415 ipv6_route_seq_setup_walk(iter, net);
2419 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2422 struct net *net = seq_file_net(seq);
2423 struct ipv6_route_iter *iter = seq->private;
2426 iter->tbl = ipv6_route_seq_next_table(NULL, net);
2430 ipv6_route_seq_setup_walk(iter, net);
2431 return ipv6_route_seq_next(seq, NULL, pos);
2437 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2439 struct fib6_walker *w = &iter->w;
2440 return w->node && !(w->state == FWS_U && w->node == w->root);
2443 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2446 struct net *net = seq_file_net(seq);
2447 struct ipv6_route_iter *iter = seq->private;
2449 if (ipv6_route_iter_active(iter))
2450 fib6_walker_unlink(net, &iter->w);
2452 rcu_read_unlock_bh();
2455 const struct seq_operations ipv6_route_seq_ops = {
2456 .start = ipv6_route_seq_start,
2457 .next = ipv6_route_seq_next,
2458 .stop = ipv6_route_seq_stop,
2459 .show = ipv6_route_seq_show
2461 #endif /* CONFIG_PROC_FS */