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 = pol_lookup_func(lookup,
318 net, net->ipv6.fib6_main_tbl, fl6, skb, flags);
319 if (rt->dst.error == -EAGAIN) {
320 ip6_rt_put_flags(rt, flags);
321 rt = net->ipv6.ip6_null_entry;
322 if (!(flags & RT6_LOOKUP_F_DST_NOREF))
329 /* called with rcu lock held; no reference taken on fib6_info */
330 int fib6_lookup(struct net *net, int oif, struct flowi6 *fl6,
331 struct fib6_result *res, int flags)
333 return fib6_table_lookup(net, net->ipv6.fib6_main_tbl, oif, fl6,
337 static void __net_init fib6_tables_init(struct net *net)
339 fib6_link_table(net, net->ipv6.fib6_main_tbl);
344 unsigned int fib6_tables_seq_read(struct net *net)
346 unsigned int h, fib_seq = 0;
349 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
350 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
351 struct fib6_table *tb;
353 hlist_for_each_entry_rcu(tb, head, tb6_hlist)
354 fib_seq += tb->fib_seq;
361 static int call_fib6_entry_notifier(struct notifier_block *nb,
362 enum fib_event_type event_type,
363 struct fib6_info *rt,
364 struct netlink_ext_ack *extack)
366 struct fib6_entry_notifier_info info = {
367 .info.extack = extack,
371 return call_fib6_notifier(nb, event_type, &info.info);
374 static int call_fib6_multipath_entry_notifier(struct notifier_block *nb,
375 enum fib_event_type event_type,
376 struct fib6_info *rt,
377 unsigned int nsiblings,
378 struct netlink_ext_ack *extack)
380 struct fib6_entry_notifier_info info = {
381 .info.extack = extack,
383 .nsiblings = nsiblings,
386 return call_fib6_notifier(nb, event_type, &info.info);
389 int call_fib6_entry_notifiers(struct net *net,
390 enum fib_event_type event_type,
391 struct fib6_info *rt,
392 struct netlink_ext_ack *extack)
394 struct fib6_entry_notifier_info info = {
395 .info.extack = extack,
399 rt->fib6_table->fib_seq++;
400 return call_fib6_notifiers(net, event_type, &info.info);
403 int call_fib6_multipath_entry_notifiers(struct net *net,
404 enum fib_event_type event_type,
405 struct fib6_info *rt,
406 unsigned int nsiblings,
407 struct netlink_ext_ack *extack)
409 struct fib6_entry_notifier_info info = {
410 .info.extack = extack,
412 .nsiblings = nsiblings,
415 rt->fib6_table->fib_seq++;
416 return call_fib6_notifiers(net, event_type, &info.info);
419 int call_fib6_entry_notifiers_replace(struct net *net, struct fib6_info *rt)
421 struct fib6_entry_notifier_info info = {
423 .nsiblings = rt->fib6_nsiblings,
426 rt->fib6_table->fib_seq++;
427 return call_fib6_notifiers(net, FIB_EVENT_ENTRY_REPLACE, &info.info);
430 struct fib6_dump_arg {
432 struct notifier_block *nb;
433 struct netlink_ext_ack *extack;
436 static int fib6_rt_dump(struct fib6_info *rt, struct fib6_dump_arg *arg)
438 enum fib_event_type fib_event = FIB_EVENT_ENTRY_REPLACE;
441 if (!rt || rt == arg->net->ipv6.fib6_null_entry)
444 if (rt->fib6_nsiblings)
445 err = call_fib6_multipath_entry_notifier(arg->nb, fib_event,
450 err = call_fib6_entry_notifier(arg->nb, fib_event, rt,
456 static int fib6_node_dump(struct fib6_walker *w)
460 err = fib6_rt_dump(w->leaf, w->args);
465 static int fib6_table_dump(struct net *net, struct fib6_table *tb,
466 struct fib6_walker *w)
470 w->root = &tb->tb6_root;
471 spin_lock_bh(&tb->tb6_lock);
472 err = fib6_walk(net, w);
473 spin_unlock_bh(&tb->tb6_lock);
477 /* Called with rcu_read_lock() */
478 int fib6_tables_dump(struct net *net, struct notifier_block *nb,
479 struct netlink_ext_ack *extack)
481 struct fib6_dump_arg arg;
482 struct fib6_walker *w;
486 w = kzalloc(sizeof(*w), GFP_ATOMIC);
490 w->func = fib6_node_dump;
496 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
497 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
498 struct fib6_table *tb;
500 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
501 err = fib6_table_dump(net, tb, w);
513 static int fib6_dump_node(struct fib6_walker *w)
516 struct fib6_info *rt;
518 for_each_fib6_walker_rt(w) {
519 res = rt6_dump_route(rt, w->args, w->skip_in_node);
521 /* Frame is full, suspend walking */
524 /* We'll restart from this node, so if some routes were
525 * already dumped, skip them next time.
527 w->skip_in_node += res;
533 /* Multipath routes are dumped in one route with the
534 * RTA_MULTIPATH attribute. Jump 'rt' to point to the
535 * last sibling of this route (no need to dump the
536 * sibling routes again)
538 if (rt->fib6_nsiblings)
539 rt = list_last_entry(&rt->fib6_siblings,
547 static void fib6_dump_end(struct netlink_callback *cb)
549 struct net *net = sock_net(cb->skb->sk);
550 struct fib6_walker *w = (void *)cb->args[2];
555 fib6_walker_unlink(net, w);
560 cb->done = (void *)cb->args[3];
564 static int fib6_dump_done(struct netlink_callback *cb)
567 return cb->done ? cb->done(cb) : 0;
570 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
571 struct netlink_callback *cb)
573 struct net *net = sock_net(skb->sk);
574 struct fib6_walker *w;
577 w = (void *)cb->args[2];
578 w->root = &table->tb6_root;
580 if (cb->args[4] == 0) {
585 spin_lock_bh(&table->tb6_lock);
586 res = fib6_walk(net, w);
587 spin_unlock_bh(&table->tb6_lock);
590 cb->args[5] = w->root->fn_sernum;
593 if (cb->args[5] != w->root->fn_sernum) {
594 /* Begin at the root if the tree changed */
595 cb->args[5] = w->root->fn_sernum;
603 spin_lock_bh(&table->tb6_lock);
604 res = fib6_walk_continue(w);
605 spin_unlock_bh(&table->tb6_lock);
607 fib6_walker_unlink(net, w);
615 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
617 struct rt6_rtnl_dump_arg arg = { .filter.dump_exceptions = true,
618 .filter.dump_routes = true };
619 const struct nlmsghdr *nlh = cb->nlh;
620 struct net *net = sock_net(skb->sk);
622 unsigned int e = 0, s_e;
623 struct fib6_walker *w;
624 struct fib6_table *tb;
625 struct hlist_head *head;
628 if (cb->strict_check) {
631 err = ip_valid_fib_dump_req(net, nlh, &arg.filter, cb);
634 } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
635 struct rtmsg *rtm = nlmsg_data(nlh);
637 if (rtm->rtm_flags & RTM_F_PREFIX)
638 arg.filter.flags = RTM_F_PREFIX;
641 w = (void *)cb->args[2];
645 * 1. hook callback destructor.
647 cb->args[3] = (long)cb->done;
648 cb->done = fib6_dump_done;
651 * 2. allocate and initialize walker.
653 w = kzalloc(sizeof(*w), GFP_ATOMIC);
656 w->func = fib6_dump_node;
657 cb->args[2] = (long)w;
665 if (arg.filter.table_id) {
666 tb = fib6_get_table(net, arg.filter.table_id);
668 if (rtnl_msg_family(cb->nlh) != PF_INET6)
671 NL_SET_ERR_MSG_MOD(cb->extack, "FIB table does not exist");
676 res = fib6_dump_table(tb, skb, cb);
687 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
689 head = &net->ipv6.fib_table_hash[h];
690 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
693 res = fib6_dump_table(tb, skb, cb);
705 res = res < 0 ? res : skb->len;
711 void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val)
716 if (f6i->fib6_metrics == &dst_default_metrics) {
717 struct dst_metrics *p = kzalloc(sizeof(*p), GFP_ATOMIC);
722 refcount_set(&p->refcnt, 1);
723 f6i->fib6_metrics = p;
726 f6i->fib6_metrics->metrics[metric - 1] = val;
732 * return the appropriate node for a routing tree "add" operation
733 * by either creating and inserting or by returning an existing
737 static struct fib6_node *fib6_add_1(struct net *net,
738 struct fib6_table *table,
739 struct fib6_node *root,
740 struct in6_addr *addr, int plen,
741 int offset, int allow_create,
742 int replace_required,
743 struct netlink_ext_ack *extack)
745 struct fib6_node *fn, *in, *ln;
746 struct fib6_node *pn = NULL;
751 RT6_TRACE("fib6_add_1\n");
753 /* insert node in tree */
758 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
759 lockdep_is_held(&table->tb6_lock));
760 key = (struct rt6key *)((u8 *)leaf + offset);
765 if (plen < fn->fn_bit ||
766 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
768 if (replace_required) {
769 NL_SET_ERR_MSG(extack,
770 "Can not replace route - no match found");
771 pr_warn("Can't replace route, no match found\n");
772 return ERR_PTR(-ENOENT);
774 pr_warn("NLM_F_CREATE should be set when creating new route\n");
783 if (plen == fn->fn_bit) {
784 /* clean up an intermediate node */
785 if (!(fn->fn_flags & RTN_RTINFO)) {
786 RCU_INIT_POINTER(fn->leaf, NULL);
787 fib6_info_release(leaf);
788 /* remove null_entry in the root node */
789 } else if (fn->fn_flags & RTN_TL_ROOT &&
790 rcu_access_pointer(fn->leaf) ==
791 net->ipv6.fib6_null_entry) {
792 RCU_INIT_POINTER(fn->leaf, NULL);
799 * We have more bits to go
802 /* Try to walk down on tree. */
803 dir = addr_bit_set(addr, fn->fn_bit);
806 rcu_dereference_protected(fn->right,
807 lockdep_is_held(&table->tb6_lock)) :
808 rcu_dereference_protected(fn->left,
809 lockdep_is_held(&table->tb6_lock));
813 /* We should not create new node because
814 * NLM_F_REPLACE was specified without NLM_F_CREATE
815 * I assume it is safe to require NLM_F_CREATE when
816 * REPLACE flag is used! Later we may want to remove the
817 * check for replace_required, because according
818 * to netlink specification, NLM_F_CREATE
819 * MUST be specified if new route is created.
820 * That would keep IPv6 consistent with IPv4
822 if (replace_required) {
823 NL_SET_ERR_MSG(extack,
824 "Can not replace route - no match found");
825 pr_warn("Can't replace route, no match found\n");
826 return ERR_PTR(-ENOENT);
828 pr_warn("NLM_F_CREATE should be set when creating new route\n");
831 * We walked to the bottom of tree.
832 * Create new leaf node without children.
835 ln = node_alloc(net);
838 return ERR_PTR(-ENOMEM);
840 RCU_INIT_POINTER(ln->parent, pn);
843 rcu_assign_pointer(pn->right, ln);
845 rcu_assign_pointer(pn->left, ln);
852 * split since we don't have a common prefix anymore or
853 * we have a less significant route.
854 * we've to insert an intermediate node on the list
855 * this new node will point to the one we need to create
859 pn = rcu_dereference_protected(fn->parent,
860 lockdep_is_held(&table->tb6_lock));
862 /* find 1st bit in difference between the 2 addrs.
864 See comment in __ipv6_addr_diff: bit may be an invalid value,
865 but if it is >= plen, the value is ignored in any case.
868 bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
873 * (new leaf node)[ln] (old node)[fn]
876 in = node_alloc(net);
877 ln = node_alloc(net);
881 node_free_immediate(net, in);
883 node_free_immediate(net, ln);
884 return ERR_PTR(-ENOMEM);
888 * new intermediate node.
890 * be off since that an address that chooses one of
891 * the branches would not match less specific routes
892 * in the other branch
897 RCU_INIT_POINTER(in->parent, pn);
899 fib6_info_hold(rcu_dereference_protected(in->leaf,
900 lockdep_is_held(&table->tb6_lock)));
902 /* update parent pointer */
904 rcu_assign_pointer(pn->right, in);
906 rcu_assign_pointer(pn->left, in);
910 RCU_INIT_POINTER(ln->parent, in);
911 rcu_assign_pointer(fn->parent, in);
913 if (addr_bit_set(addr, bit)) {
914 rcu_assign_pointer(in->right, ln);
915 rcu_assign_pointer(in->left, fn);
917 rcu_assign_pointer(in->left, ln);
918 rcu_assign_pointer(in->right, fn);
920 } else { /* plen <= bit */
923 * (new leaf node)[ln]
925 * (old node)[fn] NULL
928 ln = node_alloc(net);
931 return ERR_PTR(-ENOMEM);
935 RCU_INIT_POINTER(ln->parent, pn);
937 if (addr_bit_set(&key->addr, plen))
938 RCU_INIT_POINTER(ln->right, fn);
940 RCU_INIT_POINTER(ln->left, fn);
942 rcu_assign_pointer(fn->parent, ln);
945 rcu_assign_pointer(pn->right, ln);
947 rcu_assign_pointer(pn->left, ln);
952 static void __fib6_drop_pcpu_from(struct fib6_nh *fib6_nh,
953 const struct fib6_info *match,
954 const struct fib6_table *table)
958 if (!fib6_nh->rt6i_pcpu)
961 /* release the reference to this fib entry from
962 * all of its cached pcpu routes
964 for_each_possible_cpu(cpu) {
965 struct rt6_info **ppcpu_rt;
966 struct rt6_info *pcpu_rt;
968 ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
971 /* only dropping the 'from' reference if the cached route
972 * is using 'match'. The cached pcpu_rt->from only changes
973 * from a fib6_info to NULL (ip6_dst_destroy); it can never
974 * change from one fib6_info reference to another
976 if (pcpu_rt && rcu_access_pointer(pcpu_rt->from) == match) {
977 struct fib6_info *from;
979 from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
980 fib6_info_release(from);
985 struct fib6_nh_pcpu_arg {
986 struct fib6_info *from;
987 const struct fib6_table *table;
990 static int fib6_nh_drop_pcpu_from(struct fib6_nh *nh, void *_arg)
992 struct fib6_nh_pcpu_arg *arg = _arg;
994 __fib6_drop_pcpu_from(nh, arg->from, arg->table);
998 static void fib6_drop_pcpu_from(struct fib6_info *f6i,
999 const struct fib6_table *table)
1001 /* Make sure rt6_make_pcpu_route() wont add other percpu routes
1002 * while we are cleaning them here.
1004 f6i->fib6_destroying = 1;
1005 mb(); /* paired with the cmpxchg() in rt6_make_pcpu_route() */
1008 struct fib6_nh_pcpu_arg arg = {
1013 nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_drop_pcpu_from,
1016 struct fib6_nh *fib6_nh;
1018 fib6_nh = f6i->fib6_nh;
1019 __fib6_drop_pcpu_from(fib6_nh, f6i, table);
1023 static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn,
1026 struct fib6_table *table = rt->fib6_table;
1028 fib6_drop_pcpu_from(rt, table);
1030 if (rt->nh && !list_empty(&rt->nh_list))
1031 list_del_init(&rt->nh_list);
1033 if (refcount_read(&rt->fib6_ref) != 1) {
1034 /* This route is used as dummy address holder in some split
1035 * nodes. It is not leaked, but it still holds other resources,
1036 * which must be released in time. So, scan ascendant nodes
1037 * and replace dummy references to this route with references
1038 * to still alive ones.
1041 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
1042 lockdep_is_held(&table->tb6_lock));
1043 struct fib6_info *new_leaf;
1044 if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
1045 new_leaf = fib6_find_prefix(net, table, fn);
1046 fib6_info_hold(new_leaf);
1048 rcu_assign_pointer(fn->leaf, new_leaf);
1049 fib6_info_release(rt);
1051 fn = rcu_dereference_protected(fn->parent,
1052 lockdep_is_held(&table->tb6_lock));
1058 * Insert routing information in a node.
1061 static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt,
1062 struct nl_info *info,
1063 struct netlink_ext_ack *extack)
1065 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
1066 lockdep_is_held(&rt->fib6_table->tb6_lock));
1067 struct fib6_info *iter = NULL;
1068 struct fib6_info __rcu **ins;
1069 struct fib6_info __rcu **fallback_ins = NULL;
1070 int replace = (info->nlh &&
1071 (info->nlh->nlmsg_flags & NLM_F_REPLACE));
1072 int add = (!info->nlh ||
1073 (info->nlh->nlmsg_flags & NLM_F_CREATE));
1075 bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
1076 bool notify_sibling_rt = false;
1077 u16 nlflags = NLM_F_EXCL;
1080 if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
1081 nlflags |= NLM_F_APPEND;
1085 for (iter = leaf; iter;
1086 iter = rcu_dereference_protected(iter->fib6_next,
1087 lockdep_is_held(&rt->fib6_table->tb6_lock))) {
1089 * Search for duplicates
1092 if (iter->fib6_metric == rt->fib6_metric) {
1094 * Same priority level
1097 (info->nlh->nlmsg_flags & NLM_F_EXCL))
1100 nlflags &= ~NLM_F_EXCL;
1102 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
1106 fallback_ins = fallback_ins ?: ins;
1110 if (rt6_duplicate_nexthop(iter, rt)) {
1111 if (rt->fib6_nsiblings)
1112 rt->fib6_nsiblings = 0;
1113 if (!(iter->fib6_flags & RTF_EXPIRES))
1115 if (!(rt->fib6_flags & RTF_EXPIRES))
1116 fib6_clean_expires(iter);
1118 fib6_set_expires(iter, rt->expires);
1121 fib6_metric_set(iter, RTAX_MTU,
1125 /* If we have the same destination and the same metric,
1126 * but not the same gateway, then the route we try to
1127 * add is sibling to this route, increment our counter
1128 * of siblings, and later we will add our route to the
1130 * Only static routes (which don't have flag
1131 * RTF_EXPIRES) are used for ECMPv6.
1133 * To avoid long list, we only had siblings if the
1134 * route have a gateway.
1137 rt6_qualify_for_ecmp(iter))
1138 rt->fib6_nsiblings++;
1141 if (iter->fib6_metric > rt->fib6_metric)
1145 ins = &iter->fib6_next;
1148 if (fallback_ins && !found) {
1149 /* No matching route with same ecmp-able-ness found, replace
1150 * first matching route
1153 iter = rcu_dereference_protected(*ins,
1154 lockdep_is_held(&rt->fib6_table->tb6_lock));
1158 /* Reset round-robin state, if necessary */
1159 if (ins == &fn->leaf)
1162 /* Link this route to others same route. */
1163 if (rt->fib6_nsiblings) {
1164 unsigned int fib6_nsiblings;
1165 struct fib6_info *sibling, *temp_sibling;
1167 /* Find the first route that have the same metric */
1169 notify_sibling_rt = true;
1171 if (sibling->fib6_metric == rt->fib6_metric &&
1172 rt6_qualify_for_ecmp(sibling)) {
1173 list_add_tail(&rt->fib6_siblings,
1174 &sibling->fib6_siblings);
1177 sibling = rcu_dereference_protected(sibling->fib6_next,
1178 lockdep_is_held(&rt->fib6_table->tb6_lock));
1179 notify_sibling_rt = false;
1181 /* For each sibling in the list, increment the counter of
1182 * siblings. BUG() if counters does not match, list of siblings
1186 list_for_each_entry_safe(sibling, temp_sibling,
1187 &rt->fib6_siblings, fib6_siblings) {
1188 sibling->fib6_nsiblings++;
1189 BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings);
1192 BUG_ON(fib6_nsiblings != rt->fib6_nsiblings);
1193 rt6_multipath_rebalance(temp_sibling);
1201 pr_warn("NLM_F_CREATE should be set when creating new route\n");
1204 nlflags |= NLM_F_CREATE;
1206 /* The route should only be notified if it is the first
1207 * route in the node or if it is added as a sibling
1208 * route to the first route in the node.
1210 if (!info->skip_notify_kernel &&
1211 (notify_sibling_rt || ins == &fn->leaf)) {
1212 enum fib_event_type fib_event;
1214 if (notify_sibling_rt)
1215 fib_event = FIB_EVENT_ENTRY_APPEND;
1217 fib_event = FIB_EVENT_ENTRY_REPLACE;
1218 err = call_fib6_entry_notifiers(info->nl_net,
1222 struct fib6_info *sibling, *next_sibling;
1224 /* If the route has siblings, then it first
1225 * needs to be unlinked from them.
1227 if (!rt->fib6_nsiblings)
1230 list_for_each_entry_safe(sibling, next_sibling,
1233 sibling->fib6_nsiblings--;
1234 rt->fib6_nsiblings = 0;
1235 list_del_init(&rt->fib6_siblings);
1236 rt6_multipath_rebalance(next_sibling);
1241 rcu_assign_pointer(rt->fib6_next, iter);
1243 rcu_assign_pointer(rt->fib6_node, fn);
1244 rcu_assign_pointer(*ins, rt);
1245 if (!info->skip_notify)
1246 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1247 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1249 if (!(fn->fn_flags & RTN_RTINFO)) {
1250 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1251 fn->fn_flags |= RTN_RTINFO;
1260 pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1264 if (!info->skip_notify_kernel && ins == &fn->leaf) {
1265 err = call_fib6_entry_notifiers(info->nl_net,
1266 FIB_EVENT_ENTRY_REPLACE,
1273 rcu_assign_pointer(rt->fib6_node, fn);
1274 rt->fib6_next = iter->fib6_next;
1275 rcu_assign_pointer(*ins, rt);
1276 if (!info->skip_notify)
1277 inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1278 if (!(fn->fn_flags & RTN_RTINFO)) {
1279 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1280 fn->fn_flags |= RTN_RTINFO;
1282 nsiblings = iter->fib6_nsiblings;
1283 iter->fib6_node = NULL;
1284 fib6_purge_rt(iter, fn, info->nl_net);
1285 if (rcu_access_pointer(fn->rr_ptr) == iter)
1287 fib6_info_release(iter);
1290 /* Replacing an ECMP route, remove all siblings */
1291 ins = &rt->fib6_next;
1292 iter = rcu_dereference_protected(*ins,
1293 lockdep_is_held(&rt->fib6_table->tb6_lock));
1295 if (iter->fib6_metric > rt->fib6_metric)
1297 if (rt6_qualify_for_ecmp(iter)) {
1298 *ins = iter->fib6_next;
1299 iter->fib6_node = NULL;
1300 fib6_purge_rt(iter, fn, info->nl_net);
1301 if (rcu_access_pointer(fn->rr_ptr) == iter)
1303 fib6_info_release(iter);
1305 info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1307 ins = &iter->fib6_next;
1309 iter = rcu_dereference_protected(*ins,
1310 lockdep_is_held(&rt->fib6_table->tb6_lock));
1312 WARN_ON(nsiblings != 0);
1319 static void fib6_start_gc(struct net *net, struct fib6_info *rt)
1321 if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1322 (rt->fib6_flags & RTF_EXPIRES))
1323 mod_timer(&net->ipv6.ip6_fib_timer,
1324 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1327 void fib6_force_start_gc(struct net *net)
1329 if (!timer_pending(&net->ipv6.ip6_fib_timer))
1330 mod_timer(&net->ipv6.ip6_fib_timer,
1331 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1334 static void __fib6_update_sernum_upto_root(struct fib6_info *rt,
1337 struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1338 lockdep_is_held(&rt->fib6_table->tb6_lock));
1340 /* paired with smp_rmb() in rt6_get_cookie_safe() */
1343 fn->fn_sernum = sernum;
1344 fn = rcu_dereference_protected(fn->parent,
1345 lockdep_is_held(&rt->fib6_table->tb6_lock));
1349 void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt)
1351 __fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1354 /* allow ipv4 to update sernum via ipv6_stub */
1355 void fib6_update_sernum_stub(struct net *net, struct fib6_info *f6i)
1357 spin_lock_bh(&f6i->fib6_table->tb6_lock);
1358 fib6_update_sernum_upto_root(net, f6i);
1359 spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1363 * Add routing information to the routing tree.
1364 * <destination addr>/<source addr>
1365 * with source addr info in sub-trees
1366 * Need to own table->tb6_lock
1369 int fib6_add(struct fib6_node *root, struct fib6_info *rt,
1370 struct nl_info *info, struct netlink_ext_ack *extack)
1372 struct fib6_table *table = rt->fib6_table;
1373 struct fib6_node *fn, *pn = NULL;
1375 int allow_create = 1;
1376 int replace_required = 0;
1377 int sernum = fib6_new_sernum(info->nl_net);
1380 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1382 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1383 replace_required = 1;
1385 if (!allow_create && !replace_required)
1386 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1388 fn = fib6_add_1(info->nl_net, table, root,
1389 &rt->fib6_dst.addr, rt->fib6_dst.plen,
1390 offsetof(struct fib6_info, fib6_dst), allow_create,
1391 replace_required, extack);
1400 #ifdef CONFIG_IPV6_SUBTREES
1401 if (rt->fib6_src.plen) {
1402 struct fib6_node *sn;
1404 if (!rcu_access_pointer(fn->subtree)) {
1405 struct fib6_node *sfn;
1417 /* Create subtree root node */
1418 sfn = node_alloc(info->nl_net);
1422 fib6_info_hold(info->nl_net->ipv6.fib6_null_entry);
1423 rcu_assign_pointer(sfn->leaf,
1424 info->nl_net->ipv6.fib6_null_entry);
1425 sfn->fn_flags = RTN_ROOT;
1427 /* Now add the first leaf node to new subtree */
1429 sn = fib6_add_1(info->nl_net, table, sfn,
1430 &rt->fib6_src.addr, rt->fib6_src.plen,
1431 offsetof(struct fib6_info, fib6_src),
1432 allow_create, replace_required, extack);
1435 /* If it is failed, discard just allocated
1436 root, and then (in failure) stale node
1439 node_free_immediate(info->nl_net, sfn);
1444 /* Now link new subtree to main tree */
1445 rcu_assign_pointer(sfn->parent, fn);
1446 rcu_assign_pointer(fn->subtree, sfn);
1448 sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1449 &rt->fib6_src.addr, rt->fib6_src.plen,
1450 offsetof(struct fib6_info, fib6_src),
1451 allow_create, replace_required, extack);
1459 if (!rcu_access_pointer(fn->leaf)) {
1460 if (fn->fn_flags & RTN_TL_ROOT) {
1461 /* put back null_entry for root node */
1462 rcu_assign_pointer(fn->leaf,
1463 info->nl_net->ipv6.fib6_null_entry);
1466 rcu_assign_pointer(fn->leaf, rt);
1473 err = fib6_add_rt2node(fn, rt, info, extack);
1476 list_add(&rt->nh_list, &rt->nh->f6i_list);
1477 __fib6_update_sernum_upto_root(rt, sernum);
1478 fib6_start_gc(info->nl_net, rt);
1483 #ifdef CONFIG_IPV6_SUBTREES
1485 * If fib6_add_1 has cleared the old leaf pointer in the
1486 * super-tree leaf node we have to find a new one for it.
1489 struct fib6_info *pn_leaf =
1490 rcu_dereference_protected(pn->leaf,
1491 lockdep_is_held(&table->tb6_lock));
1492 if (pn_leaf == rt) {
1494 RCU_INIT_POINTER(pn->leaf, NULL);
1495 fib6_info_release(rt);
1497 if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1498 pn_leaf = fib6_find_prefix(info->nl_net, table,
1504 info->nl_net->ipv6.fib6_null_entry;
1507 fib6_info_hold(pn_leaf);
1508 rcu_assign_pointer(pn->leaf, pn_leaf);
1513 } else if (fib6_requires_src(rt)) {
1514 fib6_routes_require_src_inc(info->nl_net);
1519 /* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1520 * 1. fn is an intermediate node and we failed to add the new
1521 * route to it in both subtree creation failure and fib6_add_rt2node()
1523 * 2. fn is the root node in the table and we fail to add the first
1524 * default route to it.
1527 (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1528 (fn->fn_flags & RTN_TL_ROOT &&
1529 !rcu_access_pointer(fn->leaf))))
1530 fib6_repair_tree(info->nl_net, table, fn);
1535 * Routing tree lookup
1539 struct lookup_args {
1540 int offset; /* key offset on fib6_info */
1541 const struct in6_addr *addr; /* search key */
1544 static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root,
1545 struct lookup_args *args)
1547 struct fib6_node *fn;
1550 if (unlikely(args->offset == 0))
1560 struct fib6_node *next;
1562 dir = addr_bit_set(args->addr, fn->fn_bit);
1564 next = dir ? rcu_dereference(fn->right) :
1565 rcu_dereference(fn->left);
1575 struct fib6_node *subtree = FIB6_SUBTREE(fn);
1577 if (subtree || fn->fn_flags & RTN_RTINFO) {
1578 struct fib6_info *leaf = rcu_dereference(fn->leaf);
1584 key = (struct rt6key *) ((u8 *)leaf + args->offset);
1586 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1587 #ifdef CONFIG_IPV6_SUBTREES
1589 struct fib6_node *sfn;
1590 sfn = fib6_node_lookup_1(subtree,
1597 if (fn->fn_flags & RTN_RTINFO)
1602 if (fn->fn_flags & RTN_ROOT)
1605 fn = rcu_dereference(fn->parent);
1611 /* called with rcu_read_lock() held
1613 struct fib6_node *fib6_node_lookup(struct fib6_node *root,
1614 const struct in6_addr *daddr,
1615 const struct in6_addr *saddr)
1617 struct fib6_node *fn;
1618 struct lookup_args args[] = {
1620 .offset = offsetof(struct fib6_info, fib6_dst),
1623 #ifdef CONFIG_IPV6_SUBTREES
1625 .offset = offsetof(struct fib6_info, fib6_src),
1630 .offset = 0, /* sentinel */
1634 fn = fib6_node_lookup_1(root, daddr ? args : args + 1);
1635 if (!fn || fn->fn_flags & RTN_TL_ROOT)
1642 * Get node with specified destination prefix (and source prefix,
1643 * if subtrees are used)
1644 * exact_match == true means we try to find fn with exact match of
1645 * the passed in prefix addr
1646 * exact_match == false means we try to find fn with longest prefix
1647 * match of the passed in prefix addr. This is useful for finding fn
1648 * for cached route as it will be stored in the exception table under
1649 * the node with longest prefix length.
1653 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1654 const struct in6_addr *addr,
1655 int plen, int offset,
1658 struct fib6_node *fn, *prev = NULL;
1660 for (fn = root; fn ; ) {
1661 struct fib6_info *leaf = rcu_dereference(fn->leaf);
1664 /* This node is being deleted */
1666 if (plen <= fn->fn_bit)
1672 key = (struct rt6key *)((u8 *)leaf + offset);
1677 if (plen < fn->fn_bit ||
1678 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1681 if (plen == fn->fn_bit)
1684 if (fn->fn_flags & RTN_RTINFO)
1689 * We have more bits to go
1691 if (addr_bit_set(addr, fn->fn_bit))
1692 fn = rcu_dereference(fn->right);
1694 fn = rcu_dereference(fn->left);
1703 struct fib6_node *fib6_locate(struct fib6_node *root,
1704 const struct in6_addr *daddr, int dst_len,
1705 const struct in6_addr *saddr, int src_len,
1708 struct fib6_node *fn;
1710 fn = fib6_locate_1(root, daddr, dst_len,
1711 offsetof(struct fib6_info, fib6_dst),
1714 #ifdef CONFIG_IPV6_SUBTREES
1716 WARN_ON(saddr == NULL);
1718 struct fib6_node *subtree = FIB6_SUBTREE(fn);
1721 fn = fib6_locate_1(subtree, saddr, src_len,
1722 offsetof(struct fib6_info, fib6_src),
1729 if (fn && fn->fn_flags & RTN_RTINFO)
1741 static struct fib6_info *fib6_find_prefix(struct net *net,
1742 struct fib6_table *table,
1743 struct fib6_node *fn)
1745 struct fib6_node *child_left, *child_right;
1747 if (fn->fn_flags & RTN_ROOT)
1748 return net->ipv6.fib6_null_entry;
1751 child_left = rcu_dereference_protected(fn->left,
1752 lockdep_is_held(&table->tb6_lock));
1753 child_right = rcu_dereference_protected(fn->right,
1754 lockdep_is_held(&table->tb6_lock));
1756 return rcu_dereference_protected(child_left->leaf,
1757 lockdep_is_held(&table->tb6_lock));
1759 return rcu_dereference_protected(child_right->leaf,
1760 lockdep_is_held(&table->tb6_lock));
1762 fn = FIB6_SUBTREE(fn);
1768 * Called to trim the tree of intermediate nodes when possible. "fn"
1769 * is the node we want to try and remove.
1770 * Need to own table->tb6_lock
1773 static struct fib6_node *fib6_repair_tree(struct net *net,
1774 struct fib6_table *table,
1775 struct fib6_node *fn)
1779 struct fib6_node *child;
1780 struct fib6_walker *w;
1783 /* Set fn->leaf to null_entry for root node. */
1784 if (fn->fn_flags & RTN_TL_ROOT) {
1785 rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry);
1790 struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1791 lockdep_is_held(&table->tb6_lock));
1792 struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1793 lockdep_is_held(&table->tb6_lock));
1794 struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1795 lockdep_is_held(&table->tb6_lock));
1796 struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1797 lockdep_is_held(&table->tb6_lock));
1798 struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1799 lockdep_is_held(&table->tb6_lock));
1800 struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1801 lockdep_is_held(&table->tb6_lock));
1802 struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1803 lockdep_is_held(&table->tb6_lock));
1804 struct fib6_info *new_fn_leaf;
1806 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1809 WARN_ON(fn->fn_flags & RTN_RTINFO);
1810 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1816 child = fn_r, children |= 1;
1818 child = fn_l, children |= 2;
1820 if (children == 3 || FIB6_SUBTREE(fn)
1821 #ifdef CONFIG_IPV6_SUBTREES
1822 /* Subtree root (i.e. fn) may have one child */
1823 || (children && fn->fn_flags & RTN_ROOT)
1826 new_fn_leaf = fib6_find_prefix(net, table, fn);
1829 WARN_ON(!new_fn_leaf);
1830 new_fn_leaf = net->ipv6.fib6_null_entry;
1833 fib6_info_hold(new_fn_leaf);
1834 rcu_assign_pointer(fn->leaf, new_fn_leaf);
1838 #ifdef CONFIG_IPV6_SUBTREES
1839 if (FIB6_SUBTREE(pn) == fn) {
1840 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1841 RCU_INIT_POINTER(pn->subtree, NULL);
1844 WARN_ON(fn->fn_flags & RTN_ROOT);
1847 rcu_assign_pointer(pn->right, child);
1848 else if (pn_l == fn)
1849 rcu_assign_pointer(pn->left, child);
1855 rcu_assign_pointer(child->parent, pn);
1857 #ifdef CONFIG_IPV6_SUBTREES
1861 read_lock(&net->ipv6.fib6_walker_lock);
1862 FOR_WALKERS(net, w) {
1864 if (w->node == fn) {
1865 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1870 if (w->node == fn) {
1873 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1874 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1876 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1877 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1882 read_unlock(&net->ipv6.fib6_walker_lock);
1885 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1888 RCU_INIT_POINTER(pn->leaf, NULL);
1889 fib6_info_release(pn_leaf);
1894 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1895 struct fib6_info __rcu **rtp, struct nl_info *info)
1897 struct fib6_info *leaf, *replace_rt = NULL;
1898 struct fib6_walker *w;
1899 struct fib6_info *rt = rcu_dereference_protected(*rtp,
1900 lockdep_is_held(&table->tb6_lock));
1901 struct net *net = info->nl_net;
1902 bool notify_del = false;
1904 RT6_TRACE("fib6_del_route\n");
1906 /* If the deleted route is the first in the node and it is not part of
1907 * a multipath route, then we need to replace it with the next route
1908 * in the node, if exists.
1910 leaf = rcu_dereference_protected(fn->leaf,
1911 lockdep_is_held(&table->tb6_lock));
1912 if (leaf == rt && !rt->fib6_nsiblings) {
1913 if (rcu_access_pointer(rt->fib6_next))
1914 replace_rt = rcu_dereference_protected(rt->fib6_next,
1915 lockdep_is_held(&table->tb6_lock));
1921 *rtp = rt->fib6_next;
1922 rt->fib6_node = NULL;
1923 net->ipv6.rt6_stats->fib_rt_entries--;
1924 net->ipv6.rt6_stats->fib_discarded_routes++;
1926 /* Flush all cached dst in exception table */
1927 rt6_flush_exceptions(rt);
1929 /* Reset round-robin state, if necessary */
1930 if (rcu_access_pointer(fn->rr_ptr) == rt)
1933 /* Remove this entry from other siblings */
1934 if (rt->fib6_nsiblings) {
1935 struct fib6_info *sibling, *next_sibling;
1937 /* The route is deleted from a multipath route. If this
1938 * multipath route is the first route in the node, then we need
1939 * to emit a delete notification. Otherwise, we need to skip
1942 if (rt->fib6_metric == leaf->fib6_metric &&
1943 rt6_qualify_for_ecmp(leaf))
1945 list_for_each_entry_safe(sibling, next_sibling,
1946 &rt->fib6_siblings, fib6_siblings)
1947 sibling->fib6_nsiblings--;
1948 rt->fib6_nsiblings = 0;
1949 list_del_init(&rt->fib6_siblings);
1950 rt6_multipath_rebalance(next_sibling);
1953 /* Adjust walkers */
1954 read_lock(&net->ipv6.fib6_walker_lock);
1955 FOR_WALKERS(net, w) {
1956 if (w->state == FWS_C && w->leaf == rt) {
1957 RT6_TRACE("walker %p adjusted by delroute\n", w);
1958 w->leaf = rcu_dereference_protected(rt->fib6_next,
1959 lockdep_is_held(&table->tb6_lock));
1964 read_unlock(&net->ipv6.fib6_walker_lock);
1966 /* If it was last route, call fib6_repair_tree() to:
1967 * 1. For root node, put back null_entry as how the table was created.
1968 * 2. For other nodes, expunge its radix tree node.
1970 if (!rcu_access_pointer(fn->leaf)) {
1971 if (!(fn->fn_flags & RTN_TL_ROOT)) {
1972 fn->fn_flags &= ~RTN_RTINFO;
1973 net->ipv6.rt6_stats->fib_route_nodes--;
1975 fn = fib6_repair_tree(net, table, fn);
1978 fib6_purge_rt(rt, fn, net);
1980 if (!info->skip_notify_kernel) {
1982 call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL,
1984 else if (replace_rt)
1985 call_fib6_entry_notifiers_replace(net, replace_rt);
1987 if (!info->skip_notify)
1988 inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1990 fib6_info_release(rt);
1993 /* Need to own table->tb6_lock */
1994 int fib6_del(struct fib6_info *rt, struct nl_info *info)
1996 struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1997 lockdep_is_held(&rt->fib6_table->tb6_lock));
1998 struct fib6_table *table = rt->fib6_table;
1999 struct net *net = info->nl_net;
2000 struct fib6_info __rcu **rtp;
2001 struct fib6_info __rcu **rtp_next;
2003 if (!fn || rt == net->ipv6.fib6_null_entry)
2006 WARN_ON(!(fn->fn_flags & RTN_RTINFO));
2009 * Walk the leaf entries looking for ourself
2012 for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
2013 struct fib6_info *cur = rcu_dereference_protected(*rtp,
2014 lockdep_is_held(&table->tb6_lock));
2016 if (fib6_requires_src(cur))
2017 fib6_routes_require_src_dec(info->nl_net);
2018 fib6_del_route(table, fn, rtp, info);
2021 rtp_next = &cur->fib6_next;
2027 * Tree traversal function.
2029 * Certainly, it is not interrupt safe.
2030 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
2031 * It means, that we can modify tree during walking
2032 * and use this function for garbage collection, clone pruning,
2033 * cleaning tree when a device goes down etc. etc.
2035 * It guarantees that every node will be traversed,
2036 * and that it will be traversed only once.
2038 * Callback function w->func may return:
2039 * 0 -> continue walking.
2040 * positive value -> walking is suspended (used by tree dumps,
2041 * and probably by gc, if it will be split to several slices)
2042 * negative value -> terminate walking.
2044 * The function itself returns:
2045 * 0 -> walk is complete.
2046 * >0 -> walk is incomplete (i.e. suspended)
2047 * <0 -> walk is terminated by an error.
2049 * This function is called with tb6_lock held.
2052 static int fib6_walk_continue(struct fib6_walker *w)
2054 struct fib6_node *fn, *pn, *left, *right;
2056 /* w->root should always be table->tb6_root */
2057 WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
2065 #ifdef CONFIG_IPV6_SUBTREES
2067 if (FIB6_SUBTREE(fn)) {
2068 w->node = FIB6_SUBTREE(fn);
2075 left = rcu_dereference_protected(fn->left, 1);
2078 w->state = FWS_INIT;
2084 right = rcu_dereference_protected(fn->right, 1);
2087 w->state = FWS_INIT;
2091 w->leaf = rcu_dereference_protected(fn->leaf, 1);
2094 if (w->leaf && fn->fn_flags & RTN_RTINFO) {
2115 pn = rcu_dereference_protected(fn->parent, 1);
2116 left = rcu_dereference_protected(pn->left, 1);
2117 right = rcu_dereference_protected(pn->right, 1);
2119 #ifdef CONFIG_IPV6_SUBTREES
2120 if (FIB6_SUBTREE(pn) == fn) {
2121 WARN_ON(!(fn->fn_flags & RTN_ROOT));
2132 w->leaf = rcu_dereference_protected(w->node->leaf, 1);
2142 static int fib6_walk(struct net *net, struct fib6_walker *w)
2146 w->state = FWS_INIT;
2149 fib6_walker_link(net, w);
2150 res = fib6_walk_continue(w);
2152 fib6_walker_unlink(net, w);
2156 static int fib6_clean_node(struct fib6_walker *w)
2159 struct fib6_info *rt;
2160 struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
2161 struct nl_info info = {
2163 .skip_notify = c->skip_notify,
2166 if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
2167 w->node->fn_sernum != c->sernum)
2168 w->node->fn_sernum = c->sernum;
2171 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
2176 for_each_fib6_walker_rt(w) {
2177 res = c->func(rt, c->arg);
2180 res = fib6_del(rt, &info);
2183 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
2185 rcu_access_pointer(rt->fib6_node),
2191 } else if (res == -2) {
2192 if (WARN_ON(!rt->fib6_nsiblings))
2194 rt = list_last_entry(&rt->fib6_siblings,
2195 struct fib6_info, fib6_siblings);
2205 * Convenient frontend to tree walker.
2207 * func is called on each route.
2208 * It may return -2 -> skip multipath route.
2209 * -1 -> delete this route.
2210 * 0 -> continue walking
2213 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
2214 int (*func)(struct fib6_info *, void *arg),
2215 int sernum, void *arg, bool skip_notify)
2217 struct fib6_cleaner c;
2220 c.w.func = fib6_clean_node;
2223 c.w.skip_in_node = 0;
2228 c.skip_notify = skip_notify;
2230 fib6_walk(net, &c.w);
2233 static void __fib6_clean_all(struct net *net,
2234 int (*func)(struct fib6_info *, void *),
2235 int sernum, void *arg, bool skip_notify)
2237 struct fib6_table *table;
2238 struct hlist_head *head;
2242 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2243 head = &net->ipv6.fib_table_hash[h];
2244 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2245 spin_lock_bh(&table->tb6_lock);
2246 fib6_clean_tree(net, &table->tb6_root,
2247 func, sernum, arg, skip_notify);
2248 spin_unlock_bh(&table->tb6_lock);
2254 void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *),
2257 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, false);
2260 void fib6_clean_all_skip_notify(struct net *net,
2261 int (*func)(struct fib6_info *, void *),
2264 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, true);
2267 static void fib6_flush_trees(struct net *net)
2269 int new_sernum = fib6_new_sernum(net);
2271 __fib6_clean_all(net, NULL, new_sernum, NULL, false);
2275 * Garbage collection
2278 static int fib6_age(struct fib6_info *rt, void *arg)
2280 struct fib6_gc_args *gc_args = arg;
2281 unsigned long now = jiffies;
2284 * check addrconf expiration here.
2285 * Routes are expired even if they are in use.
2288 if (rt->fib6_flags & RTF_EXPIRES && rt->expires) {
2289 if (time_after(now, rt->expires)) {
2290 RT6_TRACE("expiring %p\n", rt);
2296 /* Also age clones in the exception table.
2297 * Note, that clones are aged out
2298 * only if they are not in use now.
2300 rt6_age_exceptions(rt, gc_args, now);
2305 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2307 struct fib6_gc_args gc_args;
2311 spin_lock_bh(&net->ipv6.fib6_gc_lock);
2312 } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2313 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2316 gc_args.timeout = expires ? (int)expires :
2317 net->ipv6.sysctl.ip6_rt_gc_interval;
2320 fib6_clean_all(net, fib6_age, &gc_args);
2322 net->ipv6.ip6_rt_last_gc = now;
2325 mod_timer(&net->ipv6.ip6_fib_timer,
2327 + net->ipv6.sysctl.ip6_rt_gc_interval));
2329 del_timer(&net->ipv6.ip6_fib_timer);
2330 spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2333 static void fib6_gc_timer_cb(struct timer_list *t)
2335 struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2337 fib6_run_gc(0, arg, true);
2340 static int __net_init fib6_net_init(struct net *net)
2342 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2345 err = fib6_notifier_init(net);
2349 spin_lock_init(&net->ipv6.fib6_gc_lock);
2350 rwlock_init(&net->ipv6.fib6_walker_lock);
2351 INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2352 timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2354 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2355 if (!net->ipv6.rt6_stats)
2358 /* Avoid false sharing : Use at least a full cache line */
2359 size = max_t(size_t, size, L1_CACHE_BYTES);
2361 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2362 if (!net->ipv6.fib_table_hash)
2365 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2367 if (!net->ipv6.fib6_main_tbl)
2368 goto out_fib_table_hash;
2370 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2371 rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2372 net->ipv6.fib6_null_entry);
2373 net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2374 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2375 inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2377 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2378 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2380 if (!net->ipv6.fib6_local_tbl)
2381 goto out_fib6_main_tbl;
2382 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2383 rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2384 net->ipv6.fib6_null_entry);
2385 net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2386 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2387 inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2389 fib6_tables_init(net);
2393 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2395 kfree(net->ipv6.fib6_main_tbl);
2398 kfree(net->ipv6.fib_table_hash);
2400 kfree(net->ipv6.rt6_stats);
2402 fib6_notifier_exit(net);
2406 static void fib6_net_exit(struct net *net)
2410 del_timer_sync(&net->ipv6.ip6_fib_timer);
2412 for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2413 struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2414 struct hlist_node *tmp;
2415 struct fib6_table *tb;
2417 hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2418 hlist_del(&tb->tb6_hlist);
2419 fib6_free_table(tb);
2423 kfree(net->ipv6.fib_table_hash);
2424 kfree(net->ipv6.rt6_stats);
2425 fib6_notifier_exit(net);
2428 static struct pernet_operations fib6_net_ops = {
2429 .init = fib6_net_init,
2430 .exit = fib6_net_exit,
2433 int __init fib6_init(void)
2437 fib6_node_kmem = kmem_cache_create("fib6_nodes",
2438 sizeof(struct fib6_node),
2439 0, SLAB_HWCACHE_ALIGN,
2441 if (!fib6_node_kmem)
2444 ret = register_pernet_subsys(&fib6_net_ops);
2446 goto out_kmem_cache_create;
2448 ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2451 goto out_unregister_subsys;
2453 __fib6_flush_trees = fib6_flush_trees;
2457 out_unregister_subsys:
2458 unregister_pernet_subsys(&fib6_net_ops);
2459 out_kmem_cache_create:
2460 kmem_cache_destroy(fib6_node_kmem);
2464 void fib6_gc_cleanup(void)
2466 unregister_pernet_subsys(&fib6_net_ops);
2467 kmem_cache_destroy(fib6_node_kmem);
2470 #ifdef CONFIG_PROC_FS
2471 static int ipv6_route_native_seq_show(struct seq_file *seq, void *v)
2473 struct fib6_info *rt = v;
2474 struct ipv6_route_iter *iter = seq->private;
2475 struct fib6_nh *fib6_nh = rt->fib6_nh;
2476 unsigned int flags = rt->fib6_flags;
2477 const struct net_device *dev;
2480 fib6_nh = nexthop_fib6_nh(rt->nh);
2482 seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
2484 #ifdef CONFIG_IPV6_SUBTREES
2485 seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen);
2487 seq_puts(seq, "00000000000000000000000000000000 00 ");
2489 if (fib6_nh->fib_nh_gw_family) {
2490 flags |= RTF_GATEWAY;
2491 seq_printf(seq, "%pi6", &fib6_nh->fib_nh_gw6);
2493 seq_puts(seq, "00000000000000000000000000000000");
2496 dev = fib6_nh->fib_nh_dev;
2497 seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2498 rt->fib6_metric, refcount_read(&rt->fib6_ref), 0,
2499 flags, dev ? dev->name : "");
2500 iter->w.leaf = NULL;
2504 static int ipv6_route_yield(struct fib6_walker *w)
2506 struct ipv6_route_iter *iter = w->args;
2512 iter->w.leaf = rcu_dereference_protected(
2513 iter->w.leaf->fib6_next,
2514 lockdep_is_held(&iter->tbl->tb6_lock));
2516 if (!iter->skip && iter->w.leaf)
2518 } while (iter->w.leaf);
2523 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2526 memset(&iter->w, 0, sizeof(iter->w));
2527 iter->w.func = ipv6_route_yield;
2528 iter->w.root = &iter->tbl->tb6_root;
2529 iter->w.state = FWS_INIT;
2530 iter->w.node = iter->w.root;
2531 iter->w.args = iter;
2532 iter->sernum = iter->w.root->fn_sernum;
2533 INIT_LIST_HEAD(&iter->w.lh);
2534 fib6_walker_link(net, &iter->w);
2537 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2541 struct hlist_node *node;
2544 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2545 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2551 while (!node && h < FIB6_TABLE_HASHSZ) {
2552 node = rcu_dereference_bh(
2553 hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2555 return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2558 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2560 if (iter->sernum != iter->w.root->fn_sernum) {
2561 iter->sernum = iter->w.root->fn_sernum;
2562 iter->w.state = FWS_INIT;
2563 iter->w.node = iter->w.root;
2564 WARN_ON(iter->w.skip);
2565 iter->w.skip = iter->w.count;
2569 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2572 struct fib6_info *n;
2573 struct net *net = seq_file_net(seq);
2574 struct ipv6_route_iter *iter = seq->private;
2580 n = rcu_dereference_bh(((struct fib6_info *)v)->fib6_next);
2585 ipv6_route_check_sernum(iter);
2586 spin_lock_bh(&iter->tbl->tb6_lock);
2587 r = fib6_walk_continue(&iter->w);
2588 spin_unlock_bh(&iter->tbl->tb6_lock);
2590 return iter->w.leaf;
2592 fib6_walker_unlink(net, &iter->w);
2595 fib6_walker_unlink(net, &iter->w);
2597 iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2601 ipv6_route_seq_setup_walk(iter, net);
2605 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2608 struct net *net = seq_file_net(seq);
2609 struct ipv6_route_iter *iter = seq->private;
2612 iter->tbl = ipv6_route_seq_next_table(NULL, net);
2616 ipv6_route_seq_setup_walk(iter, net);
2617 return ipv6_route_seq_next(seq, NULL, pos);
2623 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2625 struct fib6_walker *w = &iter->w;
2626 return w->node && !(w->state == FWS_U && w->node == w->root);
2629 static void ipv6_route_native_seq_stop(struct seq_file *seq, void *v)
2632 struct net *net = seq_file_net(seq);
2633 struct ipv6_route_iter *iter = seq->private;
2635 if (ipv6_route_iter_active(iter))
2636 fib6_walker_unlink(net, &iter->w);
2638 rcu_read_unlock_bh();
2641 #if IS_BUILTIN(CONFIG_IPV6) && defined(CONFIG_BPF_SYSCALL)
2642 static int ipv6_route_prog_seq_show(struct bpf_prog *prog,
2643 struct bpf_iter_meta *meta,
2646 struct bpf_iter__ipv6_route ctx;
2650 return bpf_iter_run_prog(prog, &ctx);
2653 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2655 struct ipv6_route_iter *iter = seq->private;
2656 struct bpf_iter_meta meta;
2657 struct bpf_prog *prog;
2661 prog = bpf_iter_get_info(&meta, false);
2663 return ipv6_route_native_seq_show(seq, v);
2665 ret = ipv6_route_prog_seq_show(prog, &meta, v);
2666 iter->w.leaf = NULL;
2671 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2673 struct bpf_iter_meta meta;
2674 struct bpf_prog *prog;
2678 prog = bpf_iter_get_info(&meta, true);
2680 (void)ipv6_route_prog_seq_show(prog, &meta, v);
2683 ipv6_route_native_seq_stop(seq, v);
2686 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2688 return ipv6_route_native_seq_show(seq, v);
2691 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2693 ipv6_route_native_seq_stop(seq, v);
2697 const struct seq_operations ipv6_route_seq_ops = {
2698 .start = ipv6_route_seq_start,
2699 .next = ipv6_route_seq_next,
2700 .stop = ipv6_route_seq_stop,
2701 .show = ipv6_route_seq_show
2703 #endif /* CONFIG_PROC_FS */