1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Linux INET6 implementation
4 * Forwarding Information Database
7 * Pedro Roque <roque@di.fc.ul.pt>
10 * Yuji SEKIYA @USAGI: Support default route on router node;
11 * remove ip6_null_entry from the top of
13 * Ville Nuorvala: Fixed routing subtrees.
16 #define pr_fmt(fmt) "IPv6: " fmt
18 #include <linux/errno.h>
19 #include <linux/types.h>
20 #include <linux/net.h>
21 #include <linux/route.h>
22 #include <linux/netdevice.h>
23 #include <linux/in6.h>
24 #include <linux/init.h>
25 #include <linux/list.h>
26 #include <linux/slab.h>
30 #include <net/ndisc.h>
31 #include <net/addrconf.h>
32 #include <net/lwtunnel.h>
33 #include <net/fib_notifier.h>
35 #include <net/ip6_fib.h>
36 #include <net/ip6_route.h>
38 static struct kmem_cache *fib6_node_kmem __read_mostly;
43 int (*func)(struct fib6_info *, void *arg);
49 #ifdef CONFIG_IPV6_SUBTREES
50 #define FWS_INIT FWS_S
52 #define FWS_INIT FWS_L
55 static struct fib6_info *fib6_find_prefix(struct net *net,
56 struct fib6_table *table,
57 struct fib6_node *fn);
58 static struct fib6_node *fib6_repair_tree(struct net *net,
59 struct fib6_table *table,
60 struct fib6_node *fn);
61 static int fib6_walk(struct net *net, struct fib6_walker *w);
62 static int fib6_walk_continue(struct fib6_walker *w);
65 * A routing update causes an increase of the serial number on the
66 * affected subtree. This allows for cached routes to be asynchronously
67 * tested when modifications are made to the destination cache as a
68 * result of redirects, path MTU changes, etc.
71 static void fib6_gc_timer_cb(struct timer_list *t);
73 #define FOR_WALKERS(net, w) \
74 list_for_each_entry(w, &(net)->ipv6.fib6_walkers, lh)
76 static void fib6_walker_link(struct net *net, struct fib6_walker *w)
78 write_lock_bh(&net->ipv6.fib6_walker_lock);
79 list_add(&w->lh, &net->ipv6.fib6_walkers);
80 write_unlock_bh(&net->ipv6.fib6_walker_lock);
83 static void fib6_walker_unlink(struct net *net, struct fib6_walker *w)
85 write_lock_bh(&net->ipv6.fib6_walker_lock);
87 write_unlock_bh(&net->ipv6.fib6_walker_lock);
90 static int fib6_new_sernum(struct net *net)
95 old = atomic_read(&net->ipv6.fib6_sernum);
96 new = old < INT_MAX ? old + 1 : 1;
97 } while (atomic_cmpxchg(&net->ipv6.fib6_sernum,
103 FIB6_NO_SERNUM_CHANGE = 0,
106 void fib6_update_sernum(struct net *net, struct fib6_info *f6i)
108 struct fib6_node *fn;
110 fn = rcu_dereference_protected(f6i->fib6_node,
111 lockdep_is_held(&f6i->fib6_table->tb6_lock));
113 fn->fn_sernum = fib6_new_sernum(net);
117 * Auxiliary address test functions for the radix tree.
119 * These assume a 32bit processor (although it will work on
126 #if defined(__LITTLE_ENDIAN)
127 # define BITOP_BE32_SWIZZLE (0x1F & ~7)
129 # define BITOP_BE32_SWIZZLE 0
132 static __be32 addr_bit_set(const void *token, int fn_bit)
134 const __be32 *addr = token;
137 * 1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
138 * is optimized version of
139 * htonl(1 << ((~fn_bit)&0x1F))
140 * See include/asm-generic/bitops/le.h.
142 return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
146 struct fib6_info *fib6_info_alloc(gfp_t gfp_flags, bool with_fib6_nh)
148 struct fib6_info *f6i;
149 size_t sz = sizeof(*f6i);
152 sz += sizeof(struct fib6_nh);
154 f6i = kzalloc(sz, gfp_flags);
158 /* fib6_siblings is a union with nh_list, so this initializes both */
159 INIT_LIST_HEAD(&f6i->fib6_siblings);
160 refcount_set(&f6i->fib6_ref, 1);
165 void fib6_info_destroy_rcu(struct rcu_head *head)
167 struct fib6_info *f6i = container_of(head, struct fib6_info, rcu);
169 WARN_ON(f6i->fib6_node);
172 nexthop_put(f6i->nh);
174 fib6_nh_release(f6i->fib6_nh);
176 ip_fib_metrics_put(f6i->fib6_metrics);
179 EXPORT_SYMBOL_GPL(fib6_info_destroy_rcu);
181 static struct fib6_node *node_alloc(struct net *net)
183 struct fib6_node *fn;
185 fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
187 net->ipv6.rt6_stats->fib_nodes++;
192 static void node_free_immediate(struct net *net, struct fib6_node *fn)
194 kmem_cache_free(fib6_node_kmem, fn);
195 net->ipv6.rt6_stats->fib_nodes--;
198 static void node_free_rcu(struct rcu_head *head)
200 struct fib6_node *fn = container_of(head, struct fib6_node, rcu);
202 kmem_cache_free(fib6_node_kmem, fn);
205 static void node_free(struct net *net, struct fib6_node *fn)
207 call_rcu(&fn->rcu, node_free_rcu);
208 net->ipv6.rt6_stats->fib_nodes--;
211 static void fib6_free_table(struct fib6_table *table)
213 inetpeer_invalidate_tree(&table->tb6_peers);
217 static void fib6_link_table(struct net *net, struct fib6_table *tb)
222 * Initialize table lock at a single place to give lockdep a key,
223 * tables aren't visible prior to being linked to the list.
225 spin_lock_init(&tb->tb6_lock);
226 h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
229 * No protection necessary, this is the only list mutatation
230 * operation, tables never disappear once they exist.
232 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
235 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
237 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
239 struct fib6_table *table;
241 table = kzalloc(sizeof(*table), GFP_ATOMIC);
244 rcu_assign_pointer(table->tb6_root.leaf,
245 net->ipv6.fib6_null_entry);
246 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
247 inet_peer_base_init(&table->tb6_peers);
253 struct fib6_table *fib6_new_table(struct net *net, u32 id)
255 struct fib6_table *tb;
259 tb = fib6_get_table(net, id);
263 tb = fib6_alloc_table(net, id);
265 fib6_link_table(net, tb);
269 EXPORT_SYMBOL_GPL(fib6_new_table);
271 struct fib6_table *fib6_get_table(struct net *net, u32 id)
273 struct fib6_table *tb;
274 struct hlist_head *head;
279 h = id & (FIB6_TABLE_HASHSZ - 1);
281 head = &net->ipv6.fib_table_hash[h];
282 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
283 if (tb->tb6_id == id) {
292 EXPORT_SYMBOL_GPL(fib6_get_table);
294 static void __net_init fib6_tables_init(struct net *net)
296 fib6_link_table(net, net->ipv6.fib6_main_tbl);
297 fib6_link_table(net, net->ipv6.fib6_local_tbl);
301 struct fib6_table *fib6_new_table(struct net *net, u32 id)
303 return fib6_get_table(net, id);
306 struct fib6_table *fib6_get_table(struct net *net, u32 id)
308 return net->ipv6.fib6_main_tbl;
311 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
312 const struct sk_buff *skb,
313 int flags, pol_lookup_t lookup)
317 rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, skb, flags);
318 if (rt->dst.error == -EAGAIN) {
319 ip6_rt_put_flags(rt, flags);
320 rt = net->ipv6.ip6_null_entry;
321 if (!(flags & RT6_LOOKUP_F_DST_NOREF))
328 /* called with rcu lock held; no reference taken on fib6_info */
329 int fib6_lookup(struct net *net, int oif, struct flowi6 *fl6,
330 struct fib6_result *res, int flags)
332 return fib6_table_lookup(net, net->ipv6.fib6_main_tbl, oif, fl6,
336 static void __net_init fib6_tables_init(struct net *net)
338 fib6_link_table(net, net->ipv6.fib6_main_tbl);
343 unsigned int fib6_tables_seq_read(struct net *net)
345 unsigned int h, fib_seq = 0;
348 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
349 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
350 struct fib6_table *tb;
352 hlist_for_each_entry_rcu(tb, head, tb6_hlist)
353 fib_seq += tb->fib_seq;
360 static int call_fib6_entry_notifier(struct notifier_block *nb,
361 enum fib_event_type event_type,
362 struct fib6_info *rt,
363 struct netlink_ext_ack *extack)
365 struct fib6_entry_notifier_info info = {
366 .info.extack = extack,
370 return call_fib6_notifier(nb, event_type, &info.info);
373 int call_fib6_entry_notifiers(struct net *net,
374 enum fib_event_type event_type,
375 struct fib6_info *rt,
376 struct netlink_ext_ack *extack)
378 struct fib6_entry_notifier_info info = {
379 .info.extack = extack,
383 rt->fib6_table->fib_seq++;
384 return call_fib6_notifiers(net, event_type, &info.info);
387 int call_fib6_multipath_entry_notifiers(struct net *net,
388 enum fib_event_type event_type,
389 struct fib6_info *rt,
390 unsigned int nsiblings,
391 struct netlink_ext_ack *extack)
393 struct fib6_entry_notifier_info info = {
394 .info.extack = extack,
396 .nsiblings = nsiblings,
399 rt->fib6_table->fib_seq++;
400 return call_fib6_notifiers(net, event_type, &info.info);
403 struct fib6_dump_arg {
405 struct notifier_block *nb;
406 struct netlink_ext_ack *extack;
409 static int fib6_rt_dump(struct fib6_info *rt, struct fib6_dump_arg *arg)
411 if (rt == arg->net->ipv6.fib6_null_entry)
413 return call_fib6_entry_notifier(arg->nb, FIB_EVENT_ENTRY_ADD,
417 static int fib6_node_dump(struct fib6_walker *w)
419 struct fib6_info *rt;
422 for_each_fib6_walker_rt(w) {
423 err = fib6_rt_dump(rt, w->args);
431 static int fib6_table_dump(struct net *net, struct fib6_table *tb,
432 struct fib6_walker *w)
436 w->root = &tb->tb6_root;
437 spin_lock_bh(&tb->tb6_lock);
438 err = fib6_walk(net, w);
439 spin_unlock_bh(&tb->tb6_lock);
443 /* Called with rcu_read_lock() */
444 int fib6_tables_dump(struct net *net, struct notifier_block *nb,
445 struct netlink_ext_ack *extack)
447 struct fib6_dump_arg arg;
448 struct fib6_walker *w;
452 w = kzalloc(sizeof(*w), GFP_ATOMIC);
456 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 err = fib6_table_dump(net, tb, w);
479 static int fib6_dump_node(struct fib6_walker *w)
482 struct fib6_info *rt;
484 for_each_fib6_walker_rt(w) {
485 res = rt6_dump_route(rt, w->args, w->skip_in_node);
487 /* Frame is full, suspend walking */
490 /* We'll restart from this node, so if some routes were
491 * already dumped, skip them next time.
493 w->skip_in_node += res;
499 /* Multipath routes are dumped in one route with the
500 * RTA_MULTIPATH attribute. Jump 'rt' to point to the
501 * last sibling of this route (no need to dump the
502 * sibling routes again)
504 if (rt->fib6_nsiblings)
505 rt = list_last_entry(&rt->fib6_siblings,
513 static void fib6_dump_end(struct netlink_callback *cb)
515 struct net *net = sock_net(cb->skb->sk);
516 struct fib6_walker *w = (void *)cb->args[2];
521 fib6_walker_unlink(net, w);
526 cb->done = (void *)cb->args[3];
530 static int fib6_dump_done(struct netlink_callback *cb)
533 return cb->done ? cb->done(cb) : 0;
536 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
537 struct netlink_callback *cb)
539 struct net *net = sock_net(skb->sk);
540 struct fib6_walker *w;
543 w = (void *)cb->args[2];
544 w->root = &table->tb6_root;
546 if (cb->args[4] == 0) {
551 spin_lock_bh(&table->tb6_lock);
552 res = fib6_walk(net, w);
553 spin_unlock_bh(&table->tb6_lock);
556 cb->args[5] = w->root->fn_sernum;
559 if (cb->args[5] != w->root->fn_sernum) {
560 /* Begin at the root if the tree changed */
561 cb->args[5] = w->root->fn_sernum;
569 spin_lock_bh(&table->tb6_lock);
570 res = fib6_walk_continue(w);
571 spin_unlock_bh(&table->tb6_lock);
573 fib6_walker_unlink(net, w);
581 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
583 struct rt6_rtnl_dump_arg arg = { .filter.dump_exceptions = true,
584 .filter.dump_routes = true };
585 const struct nlmsghdr *nlh = cb->nlh;
586 struct net *net = sock_net(skb->sk);
588 unsigned int e = 0, s_e;
589 struct fib6_walker *w;
590 struct fib6_table *tb;
591 struct hlist_head *head;
594 if (cb->strict_check) {
597 err = ip_valid_fib_dump_req(net, nlh, &arg.filter, cb);
600 } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
601 struct rtmsg *rtm = nlmsg_data(nlh);
603 if (rtm->rtm_flags & RTM_F_PREFIX)
604 arg.filter.flags = RTM_F_PREFIX;
607 w = (void *)cb->args[2];
611 * 1. hook callback destructor.
613 cb->args[3] = (long)cb->done;
614 cb->done = fib6_dump_done;
617 * 2. allocate and initialize walker.
619 w = kzalloc(sizeof(*w), GFP_ATOMIC);
622 w->func = fib6_dump_node;
623 cb->args[2] = (long)w;
631 if (arg.filter.table_id) {
632 tb = fib6_get_table(net, arg.filter.table_id);
634 if (arg.filter.dump_all_families)
637 NL_SET_ERR_MSG_MOD(cb->extack, "FIB table does not exist");
642 res = fib6_dump_table(tb, skb, cb);
653 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
655 head = &net->ipv6.fib_table_hash[h];
656 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
659 res = fib6_dump_table(tb, skb, cb);
671 res = res < 0 ? res : skb->len;
677 void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val)
682 if (f6i->fib6_metrics == &dst_default_metrics) {
683 struct dst_metrics *p = kzalloc(sizeof(*p), GFP_ATOMIC);
688 refcount_set(&p->refcnt, 1);
689 f6i->fib6_metrics = p;
692 f6i->fib6_metrics->metrics[metric - 1] = val;
698 * return the appropriate node for a routing tree "add" operation
699 * by either creating and inserting or by returning an existing
703 static struct fib6_node *fib6_add_1(struct net *net,
704 struct fib6_table *table,
705 struct fib6_node *root,
706 struct in6_addr *addr, int plen,
707 int offset, int allow_create,
708 int replace_required,
709 struct netlink_ext_ack *extack)
711 struct fib6_node *fn, *in, *ln;
712 struct fib6_node *pn = NULL;
717 RT6_TRACE("fib6_add_1\n");
719 /* insert node in tree */
724 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
725 lockdep_is_held(&table->tb6_lock));
726 key = (struct rt6key *)((u8 *)leaf + offset);
731 if (plen < fn->fn_bit ||
732 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
734 if (replace_required) {
735 NL_SET_ERR_MSG(extack,
736 "Can not replace route - no match found");
737 pr_warn("Can't replace route, no match found\n");
738 return ERR_PTR(-ENOENT);
740 pr_warn("NLM_F_CREATE should be set when creating new route\n");
749 if (plen == fn->fn_bit) {
750 /* clean up an intermediate node */
751 if (!(fn->fn_flags & RTN_RTINFO)) {
752 RCU_INIT_POINTER(fn->leaf, NULL);
753 fib6_info_release(leaf);
754 /* remove null_entry in the root node */
755 } else if (fn->fn_flags & RTN_TL_ROOT &&
756 rcu_access_pointer(fn->leaf) ==
757 net->ipv6.fib6_null_entry) {
758 RCU_INIT_POINTER(fn->leaf, NULL);
765 * We have more bits to go
768 /* Try to walk down on tree. */
769 dir = addr_bit_set(addr, fn->fn_bit);
772 rcu_dereference_protected(fn->right,
773 lockdep_is_held(&table->tb6_lock)) :
774 rcu_dereference_protected(fn->left,
775 lockdep_is_held(&table->tb6_lock));
779 /* We should not create new node because
780 * NLM_F_REPLACE was specified without NLM_F_CREATE
781 * I assume it is safe to require NLM_F_CREATE when
782 * REPLACE flag is used! Later we may want to remove the
783 * check for replace_required, because according
784 * to netlink specification, NLM_F_CREATE
785 * MUST be specified if new route is created.
786 * That would keep IPv6 consistent with IPv4
788 if (replace_required) {
789 NL_SET_ERR_MSG(extack,
790 "Can not replace route - no match found");
791 pr_warn("Can't replace route, no match found\n");
792 return ERR_PTR(-ENOENT);
794 pr_warn("NLM_F_CREATE should be set when creating new route\n");
797 * We walked to the bottom of tree.
798 * Create new leaf node without children.
801 ln = node_alloc(net);
804 return ERR_PTR(-ENOMEM);
806 RCU_INIT_POINTER(ln->parent, pn);
809 rcu_assign_pointer(pn->right, ln);
811 rcu_assign_pointer(pn->left, ln);
818 * split since we don't have a common prefix anymore or
819 * we have a less significant route.
820 * we've to insert an intermediate node on the list
821 * this new node will point to the one we need to create
825 pn = rcu_dereference_protected(fn->parent,
826 lockdep_is_held(&table->tb6_lock));
828 /* find 1st bit in difference between the 2 addrs.
830 See comment in __ipv6_addr_diff: bit may be an invalid value,
831 but if it is >= plen, the value is ignored in any case.
834 bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
839 * (new leaf node)[ln] (old node)[fn]
842 in = node_alloc(net);
843 ln = node_alloc(net);
847 node_free_immediate(net, in);
849 node_free_immediate(net, ln);
850 return ERR_PTR(-ENOMEM);
854 * new intermediate node.
856 * be off since that an address that chooses one of
857 * the branches would not match less specific routes
858 * in the other branch
863 RCU_INIT_POINTER(in->parent, pn);
865 fib6_info_hold(rcu_dereference_protected(in->leaf,
866 lockdep_is_held(&table->tb6_lock)));
868 /* update parent pointer */
870 rcu_assign_pointer(pn->right, in);
872 rcu_assign_pointer(pn->left, in);
876 RCU_INIT_POINTER(ln->parent, in);
877 rcu_assign_pointer(fn->parent, in);
879 if (addr_bit_set(addr, bit)) {
880 rcu_assign_pointer(in->right, ln);
881 rcu_assign_pointer(in->left, fn);
883 rcu_assign_pointer(in->left, ln);
884 rcu_assign_pointer(in->right, fn);
886 } else { /* plen <= bit */
889 * (new leaf node)[ln]
891 * (old node)[fn] NULL
894 ln = node_alloc(net);
897 return ERR_PTR(-ENOMEM);
901 RCU_INIT_POINTER(ln->parent, pn);
903 if (addr_bit_set(&key->addr, plen))
904 RCU_INIT_POINTER(ln->right, fn);
906 RCU_INIT_POINTER(ln->left, fn);
908 rcu_assign_pointer(fn->parent, ln);
911 rcu_assign_pointer(pn->right, ln);
913 rcu_assign_pointer(pn->left, ln);
918 static void __fib6_drop_pcpu_from(struct fib6_nh *fib6_nh,
919 const struct fib6_info *match,
920 const struct fib6_table *table)
924 if (!fib6_nh->rt6i_pcpu)
927 /* release the reference to this fib entry from
928 * all of its cached pcpu routes
930 for_each_possible_cpu(cpu) {
931 struct rt6_info **ppcpu_rt;
932 struct rt6_info *pcpu_rt;
934 ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
937 /* only dropping the 'from' reference if the cached route
938 * is using 'match'. The cached pcpu_rt->from only changes
939 * from a fib6_info to NULL (ip6_dst_destroy); it can never
940 * change from one fib6_info reference to another
942 if (pcpu_rt && rcu_access_pointer(pcpu_rt->from) == match) {
943 struct fib6_info *from;
945 from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
946 fib6_info_release(from);
951 struct fib6_nh_pcpu_arg {
952 struct fib6_info *from;
953 const struct fib6_table *table;
956 static int fib6_nh_drop_pcpu_from(struct fib6_nh *nh, void *_arg)
958 struct fib6_nh_pcpu_arg *arg = _arg;
960 __fib6_drop_pcpu_from(nh, arg->from, arg->table);
964 static void fib6_drop_pcpu_from(struct fib6_info *f6i,
965 const struct fib6_table *table)
967 /* Make sure rt6_make_pcpu_route() wont add other percpu routes
968 * while we are cleaning them here.
970 f6i->fib6_destroying = 1;
971 mb(); /* paired with the cmpxchg() in rt6_make_pcpu_route() */
974 struct fib6_nh_pcpu_arg arg = {
979 nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_drop_pcpu_from,
982 struct fib6_nh *fib6_nh;
984 fib6_nh = f6i->fib6_nh;
985 __fib6_drop_pcpu_from(fib6_nh, f6i, table);
989 static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn,
992 struct fib6_table *table = rt->fib6_table;
994 fib6_drop_pcpu_from(rt, table);
996 if (rt->nh && !list_empty(&rt->nh_list))
997 list_del_init(&rt->nh_list);
999 if (refcount_read(&rt->fib6_ref) != 1) {
1000 /* This route is used as dummy address holder in some split
1001 * nodes. It is not leaked, but it still holds other resources,
1002 * which must be released in time. So, scan ascendant nodes
1003 * and replace dummy references to this route with references
1004 * to still alive ones.
1007 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
1008 lockdep_is_held(&table->tb6_lock));
1009 struct fib6_info *new_leaf;
1010 if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
1011 new_leaf = fib6_find_prefix(net, table, fn);
1012 fib6_info_hold(new_leaf);
1014 rcu_assign_pointer(fn->leaf, new_leaf);
1015 fib6_info_release(rt);
1017 fn = rcu_dereference_protected(fn->parent,
1018 lockdep_is_held(&table->tb6_lock));
1024 * Insert routing information in a node.
1027 static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt,
1028 struct nl_info *info,
1029 struct netlink_ext_ack *extack)
1031 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
1032 lockdep_is_held(&rt->fib6_table->tb6_lock));
1033 struct fib6_info *iter = NULL;
1034 struct fib6_info __rcu **ins;
1035 struct fib6_info __rcu **fallback_ins = NULL;
1036 int replace = (info->nlh &&
1037 (info->nlh->nlmsg_flags & NLM_F_REPLACE));
1038 int add = (!info->nlh ||
1039 (info->nlh->nlmsg_flags & NLM_F_CREATE));
1041 bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
1042 u16 nlflags = NLM_F_EXCL;
1045 if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
1046 nlflags |= NLM_F_APPEND;
1050 for (iter = leaf; iter;
1051 iter = rcu_dereference_protected(iter->fib6_next,
1052 lockdep_is_held(&rt->fib6_table->tb6_lock))) {
1054 * Search for duplicates
1057 if (iter->fib6_metric == rt->fib6_metric) {
1059 * Same priority level
1062 (info->nlh->nlmsg_flags & NLM_F_EXCL))
1065 nlflags &= ~NLM_F_EXCL;
1067 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
1072 fallback_ins = fallback_ins ?: ins;
1076 if (rt6_duplicate_nexthop(iter, rt)) {
1077 if (rt->fib6_nsiblings)
1078 rt->fib6_nsiblings = 0;
1079 if (!(iter->fib6_flags & RTF_EXPIRES))
1081 if (!(rt->fib6_flags & RTF_EXPIRES))
1082 fib6_clean_expires(iter);
1084 fib6_set_expires(iter, rt->expires);
1087 fib6_metric_set(iter, RTAX_MTU,
1091 /* If we have the same destination and the same metric,
1092 * but not the same gateway, then the route we try to
1093 * add is sibling to this route, increment our counter
1094 * of siblings, and later we will add our route to the
1096 * Only static routes (which don't have flag
1097 * RTF_EXPIRES) are used for ECMPv6.
1099 * To avoid long list, we only had siblings if the
1100 * route have a gateway.
1103 rt6_qualify_for_ecmp(iter))
1104 rt->fib6_nsiblings++;
1107 if (iter->fib6_metric > rt->fib6_metric)
1111 ins = &iter->fib6_next;
1114 if (fallback_ins && !found) {
1115 /* No ECMP-able route found, replace first non-ECMP one */
1117 iter = rcu_dereference_protected(*ins,
1118 lockdep_is_held(&rt->fib6_table->tb6_lock));
1122 /* Reset round-robin state, if necessary */
1123 if (ins == &fn->leaf)
1126 /* Link this route to others same route. */
1127 if (rt->fib6_nsiblings) {
1128 unsigned int fib6_nsiblings;
1129 struct fib6_info *sibling, *temp_sibling;
1131 /* Find the first route that have the same metric */
1134 if (sibling->fib6_metric == rt->fib6_metric &&
1135 rt6_qualify_for_ecmp(sibling)) {
1136 list_add_tail(&rt->fib6_siblings,
1137 &sibling->fib6_siblings);
1140 sibling = rcu_dereference_protected(sibling->fib6_next,
1141 lockdep_is_held(&rt->fib6_table->tb6_lock));
1143 /* For each sibling in the list, increment the counter of
1144 * siblings. BUG() if counters does not match, list of siblings
1148 list_for_each_entry_safe(sibling, temp_sibling,
1149 &rt->fib6_siblings, fib6_siblings) {
1150 sibling->fib6_nsiblings++;
1151 BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings);
1154 BUG_ON(fib6_nsiblings != rt->fib6_nsiblings);
1155 rt6_multipath_rebalance(temp_sibling);
1163 pr_warn("NLM_F_CREATE should be set when creating new route\n");
1166 nlflags |= NLM_F_CREATE;
1168 if (!info->skip_notify_kernel) {
1169 err = call_fib6_entry_notifiers(info->nl_net,
1170 FIB_EVENT_ENTRY_ADD,
1173 struct fib6_info *sibling, *next_sibling;
1175 /* If the route has siblings, then it first
1176 * needs to be unlinked from them.
1178 if (!rt->fib6_nsiblings)
1181 list_for_each_entry_safe(sibling, next_sibling,
1184 sibling->fib6_nsiblings--;
1185 rt->fib6_nsiblings = 0;
1186 list_del_init(&rt->fib6_siblings);
1187 rt6_multipath_rebalance(next_sibling);
1192 rcu_assign_pointer(rt->fib6_next, iter);
1194 rcu_assign_pointer(rt->fib6_node, fn);
1195 rcu_assign_pointer(*ins, rt);
1196 if (!info->skip_notify)
1197 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1198 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1200 if (!(fn->fn_flags & RTN_RTINFO)) {
1201 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1202 fn->fn_flags |= RTN_RTINFO;
1211 pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1215 if (!info->skip_notify_kernel) {
1216 err = call_fib6_entry_notifiers(info->nl_net,
1217 FIB_EVENT_ENTRY_REPLACE,
1224 rcu_assign_pointer(rt->fib6_node, fn);
1225 rt->fib6_next = iter->fib6_next;
1226 rcu_assign_pointer(*ins, rt);
1227 if (!info->skip_notify)
1228 inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1229 if (!(fn->fn_flags & RTN_RTINFO)) {
1230 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1231 fn->fn_flags |= RTN_RTINFO;
1233 nsiblings = iter->fib6_nsiblings;
1234 iter->fib6_node = NULL;
1235 fib6_purge_rt(iter, fn, info->nl_net);
1236 if (rcu_access_pointer(fn->rr_ptr) == iter)
1238 fib6_info_release(iter);
1241 /* Replacing an ECMP route, remove all siblings */
1242 ins = &rt->fib6_next;
1243 iter = rcu_dereference_protected(*ins,
1244 lockdep_is_held(&rt->fib6_table->tb6_lock));
1246 if (iter->fib6_metric > rt->fib6_metric)
1248 if (rt6_qualify_for_ecmp(iter)) {
1249 *ins = iter->fib6_next;
1250 iter->fib6_node = NULL;
1251 fib6_purge_rt(iter, fn, info->nl_net);
1252 if (rcu_access_pointer(fn->rr_ptr) == iter)
1254 fib6_info_release(iter);
1256 info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1258 ins = &iter->fib6_next;
1260 iter = rcu_dereference_protected(*ins,
1261 lockdep_is_held(&rt->fib6_table->tb6_lock));
1263 WARN_ON(nsiblings != 0);
1270 static void fib6_start_gc(struct net *net, struct fib6_info *rt)
1272 if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1273 (rt->fib6_flags & RTF_EXPIRES))
1274 mod_timer(&net->ipv6.ip6_fib_timer,
1275 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1278 void fib6_force_start_gc(struct net *net)
1280 if (!timer_pending(&net->ipv6.ip6_fib_timer))
1281 mod_timer(&net->ipv6.ip6_fib_timer,
1282 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1285 static void __fib6_update_sernum_upto_root(struct fib6_info *rt,
1288 struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1289 lockdep_is_held(&rt->fib6_table->tb6_lock));
1291 /* paired with smp_rmb() in rt6_get_cookie_safe() */
1294 fn->fn_sernum = sernum;
1295 fn = rcu_dereference_protected(fn->parent,
1296 lockdep_is_held(&rt->fib6_table->tb6_lock));
1300 void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt)
1302 __fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1305 /* allow ipv4 to update sernum via ipv6_stub */
1306 void fib6_update_sernum_stub(struct net *net, struct fib6_info *f6i)
1308 spin_lock_bh(&f6i->fib6_table->tb6_lock);
1309 fib6_update_sernum_upto_root(net, f6i);
1310 spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1314 * Add routing information to the routing tree.
1315 * <destination addr>/<source addr>
1316 * with source addr info in sub-trees
1317 * Need to own table->tb6_lock
1320 int fib6_add(struct fib6_node *root, struct fib6_info *rt,
1321 struct nl_info *info, struct netlink_ext_ack *extack)
1323 struct fib6_table *table = rt->fib6_table;
1324 struct fib6_node *fn, *pn = NULL;
1326 int allow_create = 1;
1327 int replace_required = 0;
1328 int sernum = fib6_new_sernum(info->nl_net);
1331 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1333 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1334 replace_required = 1;
1336 if (!allow_create && !replace_required)
1337 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1339 fn = fib6_add_1(info->nl_net, table, root,
1340 &rt->fib6_dst.addr, rt->fib6_dst.plen,
1341 offsetof(struct fib6_info, fib6_dst), allow_create,
1342 replace_required, extack);
1351 #ifdef CONFIG_IPV6_SUBTREES
1352 if (rt->fib6_src.plen) {
1353 struct fib6_node *sn;
1355 if (!rcu_access_pointer(fn->subtree)) {
1356 struct fib6_node *sfn;
1368 /* Create subtree root node */
1369 sfn = node_alloc(info->nl_net);
1373 fib6_info_hold(info->nl_net->ipv6.fib6_null_entry);
1374 rcu_assign_pointer(sfn->leaf,
1375 info->nl_net->ipv6.fib6_null_entry);
1376 sfn->fn_flags = RTN_ROOT;
1378 /* Now add the first leaf node to new subtree */
1380 sn = fib6_add_1(info->nl_net, table, sfn,
1381 &rt->fib6_src.addr, rt->fib6_src.plen,
1382 offsetof(struct fib6_info, fib6_src),
1383 allow_create, replace_required, extack);
1386 /* If it is failed, discard just allocated
1387 root, and then (in failure) stale node
1390 node_free_immediate(info->nl_net, sfn);
1395 /* Now link new subtree to main tree */
1396 rcu_assign_pointer(sfn->parent, fn);
1397 rcu_assign_pointer(fn->subtree, sfn);
1399 sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1400 &rt->fib6_src.addr, rt->fib6_src.plen,
1401 offsetof(struct fib6_info, fib6_src),
1402 allow_create, replace_required, extack);
1410 if (!rcu_access_pointer(fn->leaf)) {
1411 if (fn->fn_flags & RTN_TL_ROOT) {
1412 /* put back null_entry for root node */
1413 rcu_assign_pointer(fn->leaf,
1414 info->nl_net->ipv6.fib6_null_entry);
1417 rcu_assign_pointer(fn->leaf, rt);
1424 err = fib6_add_rt2node(fn, rt, info, extack);
1427 list_add(&rt->nh_list, &rt->nh->f6i_list);
1428 __fib6_update_sernum_upto_root(rt, sernum);
1429 fib6_start_gc(info->nl_net, rt);
1434 #ifdef CONFIG_IPV6_SUBTREES
1436 * If fib6_add_1 has cleared the old leaf pointer in the
1437 * super-tree leaf node we have to find a new one for it.
1440 struct fib6_info *pn_leaf =
1441 rcu_dereference_protected(pn->leaf,
1442 lockdep_is_held(&table->tb6_lock));
1443 if (pn_leaf == rt) {
1445 RCU_INIT_POINTER(pn->leaf, NULL);
1446 fib6_info_release(rt);
1448 if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1449 pn_leaf = fib6_find_prefix(info->nl_net, table,
1455 info->nl_net->ipv6.fib6_null_entry;
1458 fib6_info_hold(pn_leaf);
1459 rcu_assign_pointer(pn->leaf, pn_leaf);
1468 /* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1469 * 1. fn is an intermediate node and we failed to add the new
1470 * route to it in both subtree creation failure and fib6_add_rt2node()
1472 * 2. fn is the root node in the table and we fail to add the first
1473 * default route to it.
1476 (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1477 (fn->fn_flags & RTN_TL_ROOT &&
1478 !rcu_access_pointer(fn->leaf))))
1479 fib6_repair_tree(info->nl_net, table, fn);
1484 * Routing tree lookup
1488 struct lookup_args {
1489 int offset; /* key offset on fib6_info */
1490 const struct in6_addr *addr; /* search key */
1493 static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root,
1494 struct lookup_args *args)
1496 struct fib6_node *fn;
1499 if (unlikely(args->offset == 0))
1509 struct fib6_node *next;
1511 dir = addr_bit_set(args->addr, fn->fn_bit);
1513 next = dir ? rcu_dereference(fn->right) :
1514 rcu_dereference(fn->left);
1524 struct fib6_node *subtree = FIB6_SUBTREE(fn);
1526 if (subtree || fn->fn_flags & RTN_RTINFO) {
1527 struct fib6_info *leaf = rcu_dereference(fn->leaf);
1533 key = (struct rt6key *) ((u8 *)leaf + args->offset);
1535 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1536 #ifdef CONFIG_IPV6_SUBTREES
1538 struct fib6_node *sfn;
1539 sfn = fib6_node_lookup_1(subtree,
1546 if (fn->fn_flags & RTN_RTINFO)
1551 if (fn->fn_flags & RTN_ROOT)
1554 fn = rcu_dereference(fn->parent);
1560 /* called with rcu_read_lock() held
1562 struct fib6_node *fib6_node_lookup(struct fib6_node *root,
1563 const struct in6_addr *daddr,
1564 const struct in6_addr *saddr)
1566 struct fib6_node *fn;
1567 struct lookup_args args[] = {
1569 .offset = offsetof(struct fib6_info, fib6_dst),
1572 #ifdef CONFIG_IPV6_SUBTREES
1574 .offset = offsetof(struct fib6_info, fib6_src),
1579 .offset = 0, /* sentinel */
1583 fn = fib6_node_lookup_1(root, daddr ? args : args + 1);
1584 if (!fn || fn->fn_flags & RTN_TL_ROOT)
1591 * Get node with specified destination prefix (and source prefix,
1592 * if subtrees are used)
1593 * exact_match == true means we try to find fn with exact match of
1594 * the passed in prefix addr
1595 * exact_match == false means we try to find fn with longest prefix
1596 * match of the passed in prefix addr. This is useful for finding fn
1597 * for cached route as it will be stored in the exception table under
1598 * the node with longest prefix length.
1602 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1603 const struct in6_addr *addr,
1604 int plen, int offset,
1607 struct fib6_node *fn, *prev = NULL;
1609 for (fn = root; fn ; ) {
1610 struct fib6_info *leaf = rcu_dereference(fn->leaf);
1613 /* This node is being deleted */
1615 if (plen <= fn->fn_bit)
1621 key = (struct rt6key *)((u8 *)leaf + offset);
1626 if (plen < fn->fn_bit ||
1627 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1630 if (plen == fn->fn_bit)
1633 if (fn->fn_flags & RTN_RTINFO)
1638 * We have more bits to go
1640 if (addr_bit_set(addr, fn->fn_bit))
1641 fn = rcu_dereference(fn->right);
1643 fn = rcu_dereference(fn->left);
1652 struct fib6_node *fib6_locate(struct fib6_node *root,
1653 const struct in6_addr *daddr, int dst_len,
1654 const struct in6_addr *saddr, int src_len,
1657 struct fib6_node *fn;
1659 fn = fib6_locate_1(root, daddr, dst_len,
1660 offsetof(struct fib6_info, fib6_dst),
1663 #ifdef CONFIG_IPV6_SUBTREES
1665 WARN_ON(saddr == NULL);
1667 struct fib6_node *subtree = FIB6_SUBTREE(fn);
1670 fn = fib6_locate_1(subtree, saddr, src_len,
1671 offsetof(struct fib6_info, fib6_src),
1678 if (fn && fn->fn_flags & RTN_RTINFO)
1690 static struct fib6_info *fib6_find_prefix(struct net *net,
1691 struct fib6_table *table,
1692 struct fib6_node *fn)
1694 struct fib6_node *child_left, *child_right;
1696 if (fn->fn_flags & RTN_ROOT)
1697 return net->ipv6.fib6_null_entry;
1700 child_left = rcu_dereference_protected(fn->left,
1701 lockdep_is_held(&table->tb6_lock));
1702 child_right = rcu_dereference_protected(fn->right,
1703 lockdep_is_held(&table->tb6_lock));
1705 return rcu_dereference_protected(child_left->leaf,
1706 lockdep_is_held(&table->tb6_lock));
1708 return rcu_dereference_protected(child_right->leaf,
1709 lockdep_is_held(&table->tb6_lock));
1711 fn = FIB6_SUBTREE(fn);
1717 * Called to trim the tree of intermediate nodes when possible. "fn"
1718 * is the node we want to try and remove.
1719 * Need to own table->tb6_lock
1722 static struct fib6_node *fib6_repair_tree(struct net *net,
1723 struct fib6_table *table,
1724 struct fib6_node *fn)
1728 struct fib6_node *child;
1729 struct fib6_walker *w;
1732 /* Set fn->leaf to null_entry for root node. */
1733 if (fn->fn_flags & RTN_TL_ROOT) {
1734 rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry);
1739 struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1740 lockdep_is_held(&table->tb6_lock));
1741 struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1742 lockdep_is_held(&table->tb6_lock));
1743 struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1744 lockdep_is_held(&table->tb6_lock));
1745 struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1746 lockdep_is_held(&table->tb6_lock));
1747 struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1748 lockdep_is_held(&table->tb6_lock));
1749 struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1750 lockdep_is_held(&table->tb6_lock));
1751 struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1752 lockdep_is_held(&table->tb6_lock));
1753 struct fib6_info *new_fn_leaf;
1755 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1758 WARN_ON(fn->fn_flags & RTN_RTINFO);
1759 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1765 child = fn_r, children |= 1;
1767 child = fn_l, children |= 2;
1769 if (children == 3 || FIB6_SUBTREE(fn)
1770 #ifdef CONFIG_IPV6_SUBTREES
1771 /* Subtree root (i.e. fn) may have one child */
1772 || (children && fn->fn_flags & RTN_ROOT)
1775 new_fn_leaf = fib6_find_prefix(net, table, fn);
1778 WARN_ON(!new_fn_leaf);
1779 new_fn_leaf = net->ipv6.fib6_null_entry;
1782 fib6_info_hold(new_fn_leaf);
1783 rcu_assign_pointer(fn->leaf, new_fn_leaf);
1787 #ifdef CONFIG_IPV6_SUBTREES
1788 if (FIB6_SUBTREE(pn) == fn) {
1789 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1790 RCU_INIT_POINTER(pn->subtree, NULL);
1793 WARN_ON(fn->fn_flags & RTN_ROOT);
1796 rcu_assign_pointer(pn->right, child);
1797 else if (pn_l == fn)
1798 rcu_assign_pointer(pn->left, child);
1804 rcu_assign_pointer(child->parent, pn);
1806 #ifdef CONFIG_IPV6_SUBTREES
1810 read_lock(&net->ipv6.fib6_walker_lock);
1811 FOR_WALKERS(net, w) {
1813 if (w->node == fn) {
1814 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1819 if (w->node == fn) {
1822 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1823 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1825 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1826 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1831 read_unlock(&net->ipv6.fib6_walker_lock);
1834 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1837 RCU_INIT_POINTER(pn->leaf, NULL);
1838 fib6_info_release(pn_leaf);
1843 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1844 struct fib6_info __rcu **rtp, struct nl_info *info)
1846 struct fib6_walker *w;
1847 struct fib6_info *rt = rcu_dereference_protected(*rtp,
1848 lockdep_is_held(&table->tb6_lock));
1849 struct net *net = info->nl_net;
1851 RT6_TRACE("fib6_del_route\n");
1854 *rtp = rt->fib6_next;
1855 rt->fib6_node = NULL;
1856 net->ipv6.rt6_stats->fib_rt_entries--;
1857 net->ipv6.rt6_stats->fib_discarded_routes++;
1859 /* Flush all cached dst in exception table */
1860 rt6_flush_exceptions(rt);
1862 /* Reset round-robin state, if necessary */
1863 if (rcu_access_pointer(fn->rr_ptr) == rt)
1866 /* Remove this entry from other siblings */
1867 if (rt->fib6_nsiblings) {
1868 struct fib6_info *sibling, *next_sibling;
1870 list_for_each_entry_safe(sibling, next_sibling,
1871 &rt->fib6_siblings, fib6_siblings)
1872 sibling->fib6_nsiblings--;
1873 rt->fib6_nsiblings = 0;
1874 list_del_init(&rt->fib6_siblings);
1875 rt6_multipath_rebalance(next_sibling);
1878 /* Adjust walkers */
1879 read_lock(&net->ipv6.fib6_walker_lock);
1880 FOR_WALKERS(net, w) {
1881 if (w->state == FWS_C && w->leaf == rt) {
1882 RT6_TRACE("walker %p adjusted by delroute\n", w);
1883 w->leaf = rcu_dereference_protected(rt->fib6_next,
1884 lockdep_is_held(&table->tb6_lock));
1889 read_unlock(&net->ipv6.fib6_walker_lock);
1891 /* If it was last route, call fib6_repair_tree() to:
1892 * 1. For root node, put back null_entry as how the table was created.
1893 * 2. For other nodes, expunge its radix tree node.
1895 if (!rcu_access_pointer(fn->leaf)) {
1896 if (!(fn->fn_flags & RTN_TL_ROOT)) {
1897 fn->fn_flags &= ~RTN_RTINFO;
1898 net->ipv6.rt6_stats->fib_route_nodes--;
1900 fn = fib6_repair_tree(net, table, fn);
1903 fib6_purge_rt(rt, fn, net);
1905 if (!info->skip_notify_kernel)
1906 call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, rt, NULL);
1907 if (!info->skip_notify)
1908 inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1910 fib6_info_release(rt);
1913 /* Need to own table->tb6_lock */
1914 int fib6_del(struct fib6_info *rt, struct nl_info *info)
1916 struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1917 lockdep_is_held(&rt->fib6_table->tb6_lock));
1918 struct fib6_table *table = rt->fib6_table;
1919 struct net *net = info->nl_net;
1920 struct fib6_info __rcu **rtp;
1921 struct fib6_info __rcu **rtp_next;
1923 if (!fn || rt == net->ipv6.fib6_null_entry)
1926 WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1929 * Walk the leaf entries looking for ourself
1932 for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
1933 struct fib6_info *cur = rcu_dereference_protected(*rtp,
1934 lockdep_is_held(&table->tb6_lock));
1936 fib6_del_route(table, fn, rtp, info);
1939 rtp_next = &cur->fib6_next;
1945 * Tree traversal function.
1947 * Certainly, it is not interrupt safe.
1948 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1949 * It means, that we can modify tree during walking
1950 * and use this function for garbage collection, clone pruning,
1951 * cleaning tree when a device goes down etc. etc.
1953 * It guarantees that every node will be traversed,
1954 * and that it will be traversed only once.
1956 * Callback function w->func may return:
1957 * 0 -> continue walking.
1958 * positive value -> walking is suspended (used by tree dumps,
1959 * and probably by gc, if it will be split to several slices)
1960 * negative value -> terminate walking.
1962 * The function itself returns:
1963 * 0 -> walk is complete.
1964 * >0 -> walk is incomplete (i.e. suspended)
1965 * <0 -> walk is terminated by an error.
1967 * This function is called with tb6_lock held.
1970 static int fib6_walk_continue(struct fib6_walker *w)
1972 struct fib6_node *fn, *pn, *left, *right;
1974 /* w->root should always be table->tb6_root */
1975 WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
1983 #ifdef CONFIG_IPV6_SUBTREES
1985 if (FIB6_SUBTREE(fn)) {
1986 w->node = FIB6_SUBTREE(fn);
1993 left = rcu_dereference_protected(fn->left, 1);
1996 w->state = FWS_INIT;
2002 right = rcu_dereference_protected(fn->right, 1);
2005 w->state = FWS_INIT;
2009 w->leaf = rcu_dereference_protected(fn->leaf, 1);
2012 if (w->leaf && fn->fn_flags & RTN_RTINFO) {
2033 pn = rcu_dereference_protected(fn->parent, 1);
2034 left = rcu_dereference_protected(pn->left, 1);
2035 right = rcu_dereference_protected(pn->right, 1);
2037 #ifdef CONFIG_IPV6_SUBTREES
2038 if (FIB6_SUBTREE(pn) == fn) {
2039 WARN_ON(!(fn->fn_flags & RTN_ROOT));
2050 w->leaf = rcu_dereference_protected(w->node->leaf, 1);
2060 static int fib6_walk(struct net *net, struct fib6_walker *w)
2064 w->state = FWS_INIT;
2067 fib6_walker_link(net, w);
2068 res = fib6_walk_continue(w);
2070 fib6_walker_unlink(net, w);
2074 static int fib6_clean_node(struct fib6_walker *w)
2077 struct fib6_info *rt;
2078 struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
2079 struct nl_info info = {
2081 .skip_notify = c->skip_notify,
2084 if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
2085 w->node->fn_sernum != c->sernum)
2086 w->node->fn_sernum = c->sernum;
2089 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
2094 for_each_fib6_walker_rt(w) {
2095 res = c->func(rt, c->arg);
2098 res = fib6_del(rt, &info);
2101 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
2103 rcu_access_pointer(rt->fib6_node),
2109 } else if (res == -2) {
2110 if (WARN_ON(!rt->fib6_nsiblings))
2112 rt = list_last_entry(&rt->fib6_siblings,
2113 struct fib6_info, fib6_siblings);
2123 * Convenient frontend to tree walker.
2125 * func is called on each route.
2126 * It may return -2 -> skip multipath route.
2127 * -1 -> delete this route.
2128 * 0 -> continue walking
2131 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
2132 int (*func)(struct fib6_info *, void *arg),
2133 int sernum, void *arg, bool skip_notify)
2135 struct fib6_cleaner c;
2138 c.w.func = fib6_clean_node;
2141 c.w.skip_in_node = 0;
2146 c.skip_notify = skip_notify;
2148 fib6_walk(net, &c.w);
2151 static void __fib6_clean_all(struct net *net,
2152 int (*func)(struct fib6_info *, void *),
2153 int sernum, void *arg, bool skip_notify)
2155 struct fib6_table *table;
2156 struct hlist_head *head;
2160 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2161 head = &net->ipv6.fib_table_hash[h];
2162 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2163 spin_lock_bh(&table->tb6_lock);
2164 fib6_clean_tree(net, &table->tb6_root,
2165 func, sernum, arg, skip_notify);
2166 spin_unlock_bh(&table->tb6_lock);
2172 void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *),
2175 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, false);
2178 void fib6_clean_all_skip_notify(struct net *net,
2179 int (*func)(struct fib6_info *, void *),
2182 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, true);
2185 static void fib6_flush_trees(struct net *net)
2187 int new_sernum = fib6_new_sernum(net);
2189 __fib6_clean_all(net, NULL, new_sernum, NULL, false);
2193 * Garbage collection
2196 static int fib6_age(struct fib6_info *rt, void *arg)
2198 struct fib6_gc_args *gc_args = arg;
2199 unsigned long now = jiffies;
2202 * check addrconf expiration here.
2203 * Routes are expired even if they are in use.
2206 if (rt->fib6_flags & RTF_EXPIRES && rt->expires) {
2207 if (time_after(now, rt->expires)) {
2208 RT6_TRACE("expiring %p\n", rt);
2214 /* Also age clones in the exception table.
2215 * Note, that clones are aged out
2216 * only if they are not in use now.
2218 rt6_age_exceptions(rt, gc_args, now);
2223 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2225 struct fib6_gc_args gc_args;
2229 spin_lock_bh(&net->ipv6.fib6_gc_lock);
2230 } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2231 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2234 gc_args.timeout = expires ? (int)expires :
2235 net->ipv6.sysctl.ip6_rt_gc_interval;
2238 fib6_clean_all(net, fib6_age, &gc_args);
2240 net->ipv6.ip6_rt_last_gc = now;
2243 mod_timer(&net->ipv6.ip6_fib_timer,
2245 + net->ipv6.sysctl.ip6_rt_gc_interval));
2247 del_timer(&net->ipv6.ip6_fib_timer);
2248 spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2251 static void fib6_gc_timer_cb(struct timer_list *t)
2253 struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2255 fib6_run_gc(0, arg, true);
2258 static int __net_init fib6_net_init(struct net *net)
2260 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2263 err = fib6_notifier_init(net);
2267 spin_lock_init(&net->ipv6.fib6_gc_lock);
2268 rwlock_init(&net->ipv6.fib6_walker_lock);
2269 INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2270 timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2272 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2273 if (!net->ipv6.rt6_stats)
2276 /* Avoid false sharing : Use at least a full cache line */
2277 size = max_t(size_t, size, L1_CACHE_BYTES);
2279 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2280 if (!net->ipv6.fib_table_hash)
2283 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2285 if (!net->ipv6.fib6_main_tbl)
2286 goto out_fib_table_hash;
2288 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2289 rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2290 net->ipv6.fib6_null_entry);
2291 net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2292 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2293 inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2295 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2296 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2298 if (!net->ipv6.fib6_local_tbl)
2299 goto out_fib6_main_tbl;
2300 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2301 rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2302 net->ipv6.fib6_null_entry);
2303 net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2304 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2305 inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2307 fib6_tables_init(net);
2311 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2313 kfree(net->ipv6.fib6_main_tbl);
2316 kfree(net->ipv6.fib_table_hash);
2318 kfree(net->ipv6.rt6_stats);
2320 fib6_notifier_exit(net);
2324 static void fib6_net_exit(struct net *net)
2328 del_timer_sync(&net->ipv6.ip6_fib_timer);
2330 for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2331 struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2332 struct hlist_node *tmp;
2333 struct fib6_table *tb;
2335 hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2336 hlist_del(&tb->tb6_hlist);
2337 fib6_free_table(tb);
2341 kfree(net->ipv6.fib_table_hash);
2342 kfree(net->ipv6.rt6_stats);
2343 fib6_notifier_exit(net);
2346 static struct pernet_operations fib6_net_ops = {
2347 .init = fib6_net_init,
2348 .exit = fib6_net_exit,
2351 int __init fib6_init(void)
2355 fib6_node_kmem = kmem_cache_create("fib6_nodes",
2356 sizeof(struct fib6_node),
2357 0, SLAB_HWCACHE_ALIGN,
2359 if (!fib6_node_kmem)
2362 ret = register_pernet_subsys(&fib6_net_ops);
2364 goto out_kmem_cache_create;
2366 ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2369 goto out_unregister_subsys;
2371 __fib6_flush_trees = fib6_flush_trees;
2375 out_unregister_subsys:
2376 unregister_pernet_subsys(&fib6_net_ops);
2377 out_kmem_cache_create:
2378 kmem_cache_destroy(fib6_node_kmem);
2382 void fib6_gc_cleanup(void)
2384 unregister_pernet_subsys(&fib6_net_ops);
2385 kmem_cache_destroy(fib6_node_kmem);
2388 #ifdef CONFIG_PROC_FS
2389 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2391 struct fib6_info *rt = v;
2392 struct ipv6_route_iter *iter = seq->private;
2393 struct fib6_nh *fib6_nh = rt->fib6_nh;
2394 unsigned int flags = rt->fib6_flags;
2395 const struct net_device *dev;
2398 fib6_nh = nexthop_fib6_nh(rt->nh);
2400 seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
2402 #ifdef CONFIG_IPV6_SUBTREES
2403 seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen);
2405 seq_puts(seq, "00000000000000000000000000000000 00 ");
2407 if (fib6_nh->fib_nh_gw_family) {
2408 flags |= RTF_GATEWAY;
2409 seq_printf(seq, "%pi6", &fib6_nh->fib_nh_gw6);
2411 seq_puts(seq, "00000000000000000000000000000000");
2414 dev = fib6_nh->fib_nh_dev;
2415 seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2416 rt->fib6_metric, refcount_read(&rt->fib6_ref), 0,
2417 flags, dev ? dev->name : "");
2418 iter->w.leaf = NULL;
2422 static int ipv6_route_yield(struct fib6_walker *w)
2424 struct ipv6_route_iter *iter = w->args;
2430 iter->w.leaf = rcu_dereference_protected(
2431 iter->w.leaf->fib6_next,
2432 lockdep_is_held(&iter->tbl->tb6_lock));
2434 if (!iter->skip && iter->w.leaf)
2436 } while (iter->w.leaf);
2441 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2444 memset(&iter->w, 0, sizeof(iter->w));
2445 iter->w.func = ipv6_route_yield;
2446 iter->w.root = &iter->tbl->tb6_root;
2447 iter->w.state = FWS_INIT;
2448 iter->w.node = iter->w.root;
2449 iter->w.args = iter;
2450 iter->sernum = iter->w.root->fn_sernum;
2451 INIT_LIST_HEAD(&iter->w.lh);
2452 fib6_walker_link(net, &iter->w);
2455 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2459 struct hlist_node *node;
2462 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2463 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2469 while (!node && h < FIB6_TABLE_HASHSZ) {
2470 node = rcu_dereference_bh(
2471 hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2473 return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2476 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2478 if (iter->sernum != iter->w.root->fn_sernum) {
2479 iter->sernum = iter->w.root->fn_sernum;
2480 iter->w.state = FWS_INIT;
2481 iter->w.node = iter->w.root;
2482 WARN_ON(iter->w.skip);
2483 iter->w.skip = iter->w.count;
2487 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2490 struct fib6_info *n;
2491 struct net *net = seq_file_net(seq);
2492 struct ipv6_route_iter *iter = seq->private;
2497 n = rcu_dereference_bh(((struct fib6_info *)v)->fib6_next);
2504 ipv6_route_check_sernum(iter);
2505 spin_lock_bh(&iter->tbl->tb6_lock);
2506 r = fib6_walk_continue(&iter->w);
2507 spin_unlock_bh(&iter->tbl->tb6_lock);
2511 return iter->w.leaf;
2513 fib6_walker_unlink(net, &iter->w);
2516 fib6_walker_unlink(net, &iter->w);
2518 iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2522 ipv6_route_seq_setup_walk(iter, net);
2526 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2529 struct net *net = seq_file_net(seq);
2530 struct ipv6_route_iter *iter = seq->private;
2533 iter->tbl = ipv6_route_seq_next_table(NULL, net);
2537 ipv6_route_seq_setup_walk(iter, net);
2538 return ipv6_route_seq_next(seq, NULL, pos);
2544 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2546 struct fib6_walker *w = &iter->w;
2547 return w->node && !(w->state == FWS_U && w->node == w->root);
2550 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2553 struct net *net = seq_file_net(seq);
2554 struct ipv6_route_iter *iter = seq->private;
2556 if (ipv6_route_iter_active(iter))
2557 fib6_walker_unlink(net, &iter->w);
2559 rcu_read_unlock_bh();
2562 const struct seq_operations ipv6_route_seq_ops = {
2563 .start = ipv6_route_seq_start,
2564 .next = ipv6_route_seq_next,
2565 .stop = ipv6_route_seq_stop,
2566 .show = ipv6_route_seq_show
2568 #endif /* CONFIG_PROC_FS */