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 WRITE_ONCE(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] = READ_ONCE(w->root->fn_sernum);
593 int sernum = READ_ONCE(w->root->fn_sernum);
594 if (cb->args[5] != sernum) {
595 /* Begin at the root if the tree changed */
596 cb->args[5] = sernum;
604 spin_lock_bh(&table->tb6_lock);
605 res = fib6_walk_continue(w);
606 spin_unlock_bh(&table->tb6_lock);
608 fib6_walker_unlink(net, w);
616 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
618 struct rt6_rtnl_dump_arg arg = { .filter.dump_exceptions = true,
619 .filter.dump_routes = true };
620 const struct nlmsghdr *nlh = cb->nlh;
621 struct net *net = sock_net(skb->sk);
623 unsigned int e = 0, s_e;
624 struct fib6_walker *w;
625 struct fib6_table *tb;
626 struct hlist_head *head;
629 if (cb->strict_check) {
632 err = ip_valid_fib_dump_req(net, nlh, &arg.filter, cb);
635 } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
636 struct rtmsg *rtm = nlmsg_data(nlh);
638 if (rtm->rtm_flags & RTM_F_PREFIX)
639 arg.filter.flags = RTM_F_PREFIX;
642 w = (void *)cb->args[2];
646 * 1. allocate and initialize walker.
648 w = kzalloc(sizeof(*w), GFP_ATOMIC);
651 w->func = fib6_dump_node;
652 cb->args[2] = (long)w;
654 /* 2. hook callback destructor.
656 cb->args[3] = (long)cb->done;
657 cb->done = fib6_dump_done;
666 if (arg.filter.table_id) {
667 tb = fib6_get_table(net, arg.filter.table_id);
669 if (rtnl_msg_family(cb->nlh) != PF_INET6)
672 NL_SET_ERR_MSG_MOD(cb->extack, "FIB table does not exist");
677 res = fib6_dump_table(tb, skb, cb);
688 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
690 head = &net->ipv6.fib_table_hash[h];
691 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
694 res = fib6_dump_table(tb, skb, cb);
706 res = res < 0 ? res : skb->len;
712 void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val)
717 if (f6i->fib6_metrics == &dst_default_metrics) {
718 struct dst_metrics *p = kzalloc(sizeof(*p), GFP_ATOMIC);
723 refcount_set(&p->refcnt, 1);
724 f6i->fib6_metrics = p;
727 f6i->fib6_metrics->metrics[metric - 1] = val;
733 * return the appropriate node for a routing tree "add" operation
734 * by either creating and inserting or by returning an existing
738 static struct fib6_node *fib6_add_1(struct net *net,
739 struct fib6_table *table,
740 struct fib6_node *root,
741 struct in6_addr *addr, int plen,
742 int offset, int allow_create,
743 int replace_required,
744 struct netlink_ext_ack *extack)
746 struct fib6_node *fn, *in, *ln;
747 struct fib6_node *pn = NULL;
752 RT6_TRACE("fib6_add_1\n");
754 /* insert node in tree */
759 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
760 lockdep_is_held(&table->tb6_lock));
761 key = (struct rt6key *)((u8 *)leaf + offset);
766 if (plen < fn->fn_bit ||
767 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
769 if (replace_required) {
770 NL_SET_ERR_MSG(extack,
771 "Can not replace route - no match found");
772 pr_warn("Can't replace route, no match found\n");
773 return ERR_PTR(-ENOENT);
775 pr_warn("NLM_F_CREATE should be set when creating new route\n");
784 if (plen == fn->fn_bit) {
785 /* clean up an intermediate node */
786 if (!(fn->fn_flags & RTN_RTINFO)) {
787 RCU_INIT_POINTER(fn->leaf, NULL);
788 fib6_info_release(leaf);
789 /* remove null_entry in the root node */
790 } else if (fn->fn_flags & RTN_TL_ROOT &&
791 rcu_access_pointer(fn->leaf) ==
792 net->ipv6.fib6_null_entry) {
793 RCU_INIT_POINTER(fn->leaf, NULL);
800 * We have more bits to go
803 /* Try to walk down on tree. */
804 dir = addr_bit_set(addr, fn->fn_bit);
807 rcu_dereference_protected(fn->right,
808 lockdep_is_held(&table->tb6_lock)) :
809 rcu_dereference_protected(fn->left,
810 lockdep_is_held(&table->tb6_lock));
814 /* We should not create new node because
815 * NLM_F_REPLACE was specified without NLM_F_CREATE
816 * I assume it is safe to require NLM_F_CREATE when
817 * REPLACE flag is used! Later we may want to remove the
818 * check for replace_required, because according
819 * to netlink specification, NLM_F_CREATE
820 * MUST be specified if new route is created.
821 * That would keep IPv6 consistent with IPv4
823 if (replace_required) {
824 NL_SET_ERR_MSG(extack,
825 "Can not replace route - no match found");
826 pr_warn("Can't replace route, no match found\n");
827 return ERR_PTR(-ENOENT);
829 pr_warn("NLM_F_CREATE should be set when creating new route\n");
832 * We walked to the bottom of tree.
833 * Create new leaf node without children.
836 ln = node_alloc(net);
839 return ERR_PTR(-ENOMEM);
841 RCU_INIT_POINTER(ln->parent, pn);
844 rcu_assign_pointer(pn->right, ln);
846 rcu_assign_pointer(pn->left, ln);
853 * split since we don't have a common prefix anymore or
854 * we have a less significant route.
855 * we've to insert an intermediate node on the list
856 * this new node will point to the one we need to create
860 pn = rcu_dereference_protected(fn->parent,
861 lockdep_is_held(&table->tb6_lock));
863 /* find 1st bit in difference between the 2 addrs.
865 See comment in __ipv6_addr_diff: bit may be an invalid value,
866 but if it is >= plen, the value is ignored in any case.
869 bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
874 * (new leaf node)[ln] (old node)[fn]
877 in = node_alloc(net);
878 ln = node_alloc(net);
882 node_free_immediate(net, in);
884 node_free_immediate(net, ln);
885 return ERR_PTR(-ENOMEM);
889 * new intermediate node.
891 * be off since that an address that chooses one of
892 * the branches would not match less specific routes
893 * in the other branch
898 RCU_INIT_POINTER(in->parent, pn);
900 fib6_info_hold(rcu_dereference_protected(in->leaf,
901 lockdep_is_held(&table->tb6_lock)));
903 /* update parent pointer */
905 rcu_assign_pointer(pn->right, in);
907 rcu_assign_pointer(pn->left, in);
911 RCU_INIT_POINTER(ln->parent, in);
912 rcu_assign_pointer(fn->parent, in);
914 if (addr_bit_set(addr, bit)) {
915 rcu_assign_pointer(in->right, ln);
916 rcu_assign_pointer(in->left, fn);
918 rcu_assign_pointer(in->left, ln);
919 rcu_assign_pointer(in->right, fn);
921 } else { /* plen <= bit */
924 * (new leaf node)[ln]
926 * (old node)[fn] NULL
929 ln = node_alloc(net);
932 return ERR_PTR(-ENOMEM);
936 RCU_INIT_POINTER(ln->parent, pn);
938 if (addr_bit_set(&key->addr, plen))
939 RCU_INIT_POINTER(ln->right, fn);
941 RCU_INIT_POINTER(ln->left, fn);
943 rcu_assign_pointer(fn->parent, ln);
946 rcu_assign_pointer(pn->right, ln);
948 rcu_assign_pointer(pn->left, ln);
953 static void __fib6_drop_pcpu_from(struct fib6_nh *fib6_nh,
954 const struct fib6_info *match,
955 const struct fib6_table *table)
959 if (!fib6_nh->rt6i_pcpu)
962 /* release the reference to this fib entry from
963 * all of its cached pcpu routes
965 for_each_possible_cpu(cpu) {
966 struct rt6_info **ppcpu_rt;
967 struct rt6_info *pcpu_rt;
969 ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
972 /* only dropping the 'from' reference if the cached route
973 * is using 'match'. The cached pcpu_rt->from only changes
974 * from a fib6_info to NULL (ip6_dst_destroy); it can never
975 * change from one fib6_info reference to another
977 if (pcpu_rt && rcu_access_pointer(pcpu_rt->from) == match) {
978 struct fib6_info *from;
980 from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
981 fib6_info_release(from);
986 struct fib6_nh_pcpu_arg {
987 struct fib6_info *from;
988 const struct fib6_table *table;
991 static int fib6_nh_drop_pcpu_from(struct fib6_nh *nh, void *_arg)
993 struct fib6_nh_pcpu_arg *arg = _arg;
995 __fib6_drop_pcpu_from(nh, arg->from, arg->table);
999 static void fib6_drop_pcpu_from(struct fib6_info *f6i,
1000 const struct fib6_table *table)
1002 /* Make sure rt6_make_pcpu_route() wont add other percpu routes
1003 * while we are cleaning them here.
1005 f6i->fib6_destroying = 1;
1006 mb(); /* paired with the cmpxchg() in rt6_make_pcpu_route() */
1009 struct fib6_nh_pcpu_arg arg = {
1014 nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_drop_pcpu_from,
1017 struct fib6_nh *fib6_nh;
1019 fib6_nh = f6i->fib6_nh;
1020 __fib6_drop_pcpu_from(fib6_nh, f6i, table);
1024 static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn,
1027 struct fib6_table *table = rt->fib6_table;
1029 /* Flush all cached dst in exception table */
1030 rt6_flush_exceptions(rt);
1031 fib6_drop_pcpu_from(rt, table);
1033 if (rt->nh && !list_empty(&rt->nh_list))
1034 list_del_init(&rt->nh_list);
1036 if (refcount_read(&rt->fib6_ref) != 1) {
1037 /* This route is used as dummy address holder in some split
1038 * nodes. It is not leaked, but it still holds other resources,
1039 * which must be released in time. So, scan ascendant nodes
1040 * and replace dummy references to this route with references
1041 * to still alive ones.
1044 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
1045 lockdep_is_held(&table->tb6_lock));
1046 struct fib6_info *new_leaf;
1047 if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
1048 new_leaf = fib6_find_prefix(net, table, fn);
1049 fib6_info_hold(new_leaf);
1051 rcu_assign_pointer(fn->leaf, new_leaf);
1052 fib6_info_release(rt);
1054 fn = rcu_dereference_protected(fn->parent,
1055 lockdep_is_held(&table->tb6_lock));
1061 * Insert routing information in a node.
1064 static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt,
1065 struct nl_info *info,
1066 struct netlink_ext_ack *extack)
1068 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
1069 lockdep_is_held(&rt->fib6_table->tb6_lock));
1070 struct fib6_info *iter = NULL;
1071 struct fib6_info __rcu **ins;
1072 struct fib6_info __rcu **fallback_ins = NULL;
1073 int replace = (info->nlh &&
1074 (info->nlh->nlmsg_flags & NLM_F_REPLACE));
1075 int add = (!info->nlh ||
1076 (info->nlh->nlmsg_flags & NLM_F_CREATE));
1078 bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
1079 bool notify_sibling_rt = false;
1080 u16 nlflags = NLM_F_EXCL;
1083 if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
1084 nlflags |= NLM_F_APPEND;
1088 for (iter = leaf; iter;
1089 iter = rcu_dereference_protected(iter->fib6_next,
1090 lockdep_is_held(&rt->fib6_table->tb6_lock))) {
1092 * Search for duplicates
1095 if (iter->fib6_metric == rt->fib6_metric) {
1097 * Same priority level
1100 (info->nlh->nlmsg_flags & NLM_F_EXCL))
1103 nlflags &= ~NLM_F_EXCL;
1105 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
1109 fallback_ins = fallback_ins ?: ins;
1113 if (rt6_duplicate_nexthop(iter, rt)) {
1114 if (rt->fib6_nsiblings)
1115 rt->fib6_nsiblings = 0;
1116 if (!(iter->fib6_flags & RTF_EXPIRES))
1118 if (!(rt->fib6_flags & RTF_EXPIRES))
1119 fib6_clean_expires(iter);
1121 fib6_set_expires(iter, rt->expires);
1124 fib6_metric_set(iter, RTAX_MTU,
1128 /* If we have the same destination and the same metric,
1129 * but not the same gateway, then the route we try to
1130 * add is sibling to this route, increment our counter
1131 * of siblings, and later we will add our route to the
1133 * Only static routes (which don't have flag
1134 * RTF_EXPIRES) are used for ECMPv6.
1136 * To avoid long list, we only had siblings if the
1137 * route have a gateway.
1140 rt6_qualify_for_ecmp(iter))
1141 rt->fib6_nsiblings++;
1144 if (iter->fib6_metric > rt->fib6_metric)
1148 ins = &iter->fib6_next;
1151 if (fallback_ins && !found) {
1152 /* No matching route with same ecmp-able-ness found, replace
1153 * first matching route
1156 iter = rcu_dereference_protected(*ins,
1157 lockdep_is_held(&rt->fib6_table->tb6_lock));
1161 /* Reset round-robin state, if necessary */
1162 if (ins == &fn->leaf)
1165 /* Link this route to others same route. */
1166 if (rt->fib6_nsiblings) {
1167 unsigned int fib6_nsiblings;
1168 struct fib6_info *sibling, *temp_sibling;
1170 /* Find the first route that have the same metric */
1172 notify_sibling_rt = true;
1174 if (sibling->fib6_metric == rt->fib6_metric &&
1175 rt6_qualify_for_ecmp(sibling)) {
1176 list_add_tail(&rt->fib6_siblings,
1177 &sibling->fib6_siblings);
1180 sibling = rcu_dereference_protected(sibling->fib6_next,
1181 lockdep_is_held(&rt->fib6_table->tb6_lock));
1182 notify_sibling_rt = false;
1184 /* For each sibling in the list, increment the counter of
1185 * siblings. BUG() if counters does not match, list of siblings
1189 list_for_each_entry_safe(sibling, temp_sibling,
1190 &rt->fib6_siblings, fib6_siblings) {
1191 sibling->fib6_nsiblings++;
1192 BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings);
1195 BUG_ON(fib6_nsiblings != rt->fib6_nsiblings);
1196 rt6_multipath_rebalance(temp_sibling);
1204 pr_warn("NLM_F_CREATE should be set when creating new route\n");
1207 nlflags |= NLM_F_CREATE;
1209 /* The route should only be notified if it is the first
1210 * route in the node or if it is added as a sibling
1211 * route to the first route in the node.
1213 if (!info->skip_notify_kernel &&
1214 (notify_sibling_rt || ins == &fn->leaf)) {
1215 enum fib_event_type fib_event;
1217 if (notify_sibling_rt)
1218 fib_event = FIB_EVENT_ENTRY_APPEND;
1220 fib_event = FIB_EVENT_ENTRY_REPLACE;
1221 err = call_fib6_entry_notifiers(info->nl_net,
1225 struct fib6_info *sibling, *next_sibling;
1227 /* If the route has siblings, then it first
1228 * needs to be unlinked from them.
1230 if (!rt->fib6_nsiblings)
1233 list_for_each_entry_safe(sibling, next_sibling,
1236 sibling->fib6_nsiblings--;
1237 rt->fib6_nsiblings = 0;
1238 list_del_init(&rt->fib6_siblings);
1239 rt6_multipath_rebalance(next_sibling);
1244 rcu_assign_pointer(rt->fib6_next, iter);
1246 rcu_assign_pointer(rt->fib6_node, fn);
1247 rcu_assign_pointer(*ins, rt);
1248 if (!info->skip_notify)
1249 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1250 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1252 if (!(fn->fn_flags & RTN_RTINFO)) {
1253 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1254 fn->fn_flags |= RTN_RTINFO;
1263 pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1267 if (!info->skip_notify_kernel && ins == &fn->leaf) {
1268 err = call_fib6_entry_notifiers(info->nl_net,
1269 FIB_EVENT_ENTRY_REPLACE,
1276 rcu_assign_pointer(rt->fib6_node, fn);
1277 rt->fib6_next = iter->fib6_next;
1278 rcu_assign_pointer(*ins, rt);
1279 if (!info->skip_notify)
1280 inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1281 if (!(fn->fn_flags & RTN_RTINFO)) {
1282 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1283 fn->fn_flags |= RTN_RTINFO;
1285 nsiblings = iter->fib6_nsiblings;
1286 iter->fib6_node = NULL;
1287 fib6_purge_rt(iter, fn, info->nl_net);
1288 if (rcu_access_pointer(fn->rr_ptr) == iter)
1290 fib6_info_release(iter);
1293 /* Replacing an ECMP route, remove all siblings */
1294 ins = &rt->fib6_next;
1295 iter = rcu_dereference_protected(*ins,
1296 lockdep_is_held(&rt->fib6_table->tb6_lock));
1298 if (iter->fib6_metric > rt->fib6_metric)
1300 if (rt6_qualify_for_ecmp(iter)) {
1301 *ins = iter->fib6_next;
1302 iter->fib6_node = NULL;
1303 fib6_purge_rt(iter, fn, info->nl_net);
1304 if (rcu_access_pointer(fn->rr_ptr) == iter)
1306 fib6_info_release(iter);
1308 info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1310 ins = &iter->fib6_next;
1312 iter = rcu_dereference_protected(*ins,
1313 lockdep_is_held(&rt->fib6_table->tb6_lock));
1315 WARN_ON(nsiblings != 0);
1322 static void fib6_start_gc(struct net *net, struct fib6_info *rt)
1324 if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1325 (rt->fib6_flags & RTF_EXPIRES))
1326 mod_timer(&net->ipv6.ip6_fib_timer,
1327 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1330 void fib6_force_start_gc(struct net *net)
1332 if (!timer_pending(&net->ipv6.ip6_fib_timer))
1333 mod_timer(&net->ipv6.ip6_fib_timer,
1334 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1337 static void __fib6_update_sernum_upto_root(struct fib6_info *rt,
1340 struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1341 lockdep_is_held(&rt->fib6_table->tb6_lock));
1343 /* paired with smp_rmb() in rt6_get_cookie_safe() */
1346 WRITE_ONCE(fn->fn_sernum, sernum);
1347 fn = rcu_dereference_protected(fn->parent,
1348 lockdep_is_held(&rt->fib6_table->tb6_lock));
1352 void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt)
1354 __fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1357 /* allow ipv4 to update sernum via ipv6_stub */
1358 void fib6_update_sernum_stub(struct net *net, struct fib6_info *f6i)
1360 spin_lock_bh(&f6i->fib6_table->tb6_lock);
1361 fib6_update_sernum_upto_root(net, f6i);
1362 spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1366 * Add routing information to the routing tree.
1367 * <destination addr>/<source addr>
1368 * with source addr info in sub-trees
1369 * Need to own table->tb6_lock
1372 int fib6_add(struct fib6_node *root, struct fib6_info *rt,
1373 struct nl_info *info, struct netlink_ext_ack *extack)
1375 struct fib6_table *table = rt->fib6_table;
1376 struct fib6_node *fn;
1377 #ifdef CONFIG_IPV6_SUBTREES
1378 struct fib6_node *pn = NULL;
1381 int allow_create = 1;
1382 int replace_required = 0;
1385 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1387 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1388 replace_required = 1;
1390 if (!allow_create && !replace_required)
1391 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1393 fn = fib6_add_1(info->nl_net, table, root,
1394 &rt->fib6_dst.addr, rt->fib6_dst.plen,
1395 offsetof(struct fib6_info, fib6_dst), allow_create,
1396 replace_required, extack);
1403 #ifdef CONFIG_IPV6_SUBTREES
1406 if (rt->fib6_src.plen) {
1407 struct fib6_node *sn;
1409 if (!rcu_access_pointer(fn->subtree)) {
1410 struct fib6_node *sfn;
1422 /* Create subtree root node */
1423 sfn = node_alloc(info->nl_net);
1427 fib6_info_hold(info->nl_net->ipv6.fib6_null_entry);
1428 rcu_assign_pointer(sfn->leaf,
1429 info->nl_net->ipv6.fib6_null_entry);
1430 sfn->fn_flags = RTN_ROOT;
1432 /* Now add the first leaf node to new subtree */
1434 sn = fib6_add_1(info->nl_net, table, sfn,
1435 &rt->fib6_src.addr, rt->fib6_src.plen,
1436 offsetof(struct fib6_info, fib6_src),
1437 allow_create, replace_required, extack);
1440 /* If it is failed, discard just allocated
1441 root, and then (in failure) stale node
1444 node_free_immediate(info->nl_net, sfn);
1449 /* Now link new subtree to main tree */
1450 rcu_assign_pointer(sfn->parent, fn);
1451 rcu_assign_pointer(fn->subtree, sfn);
1453 sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1454 &rt->fib6_src.addr, rt->fib6_src.plen,
1455 offsetof(struct fib6_info, fib6_src),
1456 allow_create, replace_required, extack);
1464 if (!rcu_access_pointer(fn->leaf)) {
1465 if (fn->fn_flags & RTN_TL_ROOT) {
1466 /* put back null_entry for root node */
1467 rcu_assign_pointer(fn->leaf,
1468 info->nl_net->ipv6.fib6_null_entry);
1471 rcu_assign_pointer(fn->leaf, rt);
1478 err = fib6_add_rt2node(fn, rt, info, extack);
1481 list_add(&rt->nh_list, &rt->nh->f6i_list);
1482 __fib6_update_sernum_upto_root(rt, fib6_new_sernum(info->nl_net));
1483 fib6_start_gc(info->nl_net, rt);
1488 #ifdef CONFIG_IPV6_SUBTREES
1490 * If fib6_add_1 has cleared the old leaf pointer in the
1491 * super-tree leaf node we have to find a new one for it.
1494 struct fib6_info *pn_leaf =
1495 rcu_dereference_protected(pn->leaf,
1496 lockdep_is_held(&table->tb6_lock));
1497 if (pn_leaf == rt) {
1499 RCU_INIT_POINTER(pn->leaf, NULL);
1500 fib6_info_release(rt);
1502 if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1503 pn_leaf = fib6_find_prefix(info->nl_net, table,
1507 info->nl_net->ipv6.fib6_null_entry;
1508 fib6_info_hold(pn_leaf);
1509 rcu_assign_pointer(pn->leaf, pn_leaf);
1514 } else if (fib6_requires_src(rt)) {
1515 fib6_routes_require_src_inc(info->nl_net);
1520 /* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1521 * 1. fn is an intermediate node and we failed to add the new
1522 * route to it in both subtree creation failure and fib6_add_rt2node()
1524 * 2. fn is the root node in the table and we fail to add the first
1525 * default route to it.
1528 (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1529 (fn->fn_flags & RTN_TL_ROOT &&
1530 !rcu_access_pointer(fn->leaf))))
1531 fib6_repair_tree(info->nl_net, table, fn);
1536 * Routing tree lookup
1540 struct lookup_args {
1541 int offset; /* key offset on fib6_info */
1542 const struct in6_addr *addr; /* search key */
1545 static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root,
1546 struct lookup_args *args)
1548 struct fib6_node *fn;
1551 if (unlikely(args->offset == 0))
1561 struct fib6_node *next;
1563 dir = addr_bit_set(args->addr, fn->fn_bit);
1565 next = dir ? rcu_dereference(fn->right) :
1566 rcu_dereference(fn->left);
1576 struct fib6_node *subtree = FIB6_SUBTREE(fn);
1578 if (subtree || fn->fn_flags & RTN_RTINFO) {
1579 struct fib6_info *leaf = rcu_dereference(fn->leaf);
1585 key = (struct rt6key *) ((u8 *)leaf + args->offset);
1587 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1588 #ifdef CONFIG_IPV6_SUBTREES
1590 struct fib6_node *sfn;
1591 sfn = fib6_node_lookup_1(subtree,
1598 if (fn->fn_flags & RTN_RTINFO)
1603 if (fn->fn_flags & RTN_ROOT)
1606 fn = rcu_dereference(fn->parent);
1612 /* called with rcu_read_lock() held
1614 struct fib6_node *fib6_node_lookup(struct fib6_node *root,
1615 const struct in6_addr *daddr,
1616 const struct in6_addr *saddr)
1618 struct fib6_node *fn;
1619 struct lookup_args args[] = {
1621 .offset = offsetof(struct fib6_info, fib6_dst),
1624 #ifdef CONFIG_IPV6_SUBTREES
1626 .offset = offsetof(struct fib6_info, fib6_src),
1631 .offset = 0, /* sentinel */
1635 fn = fib6_node_lookup_1(root, daddr ? args : args + 1);
1636 if (!fn || fn->fn_flags & RTN_TL_ROOT)
1643 * Get node with specified destination prefix (and source prefix,
1644 * if subtrees are used)
1645 * exact_match == true means we try to find fn with exact match of
1646 * the passed in prefix addr
1647 * exact_match == false means we try to find fn with longest prefix
1648 * match of the passed in prefix addr. This is useful for finding fn
1649 * for cached route as it will be stored in the exception table under
1650 * the node with longest prefix length.
1654 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1655 const struct in6_addr *addr,
1656 int plen, int offset,
1659 struct fib6_node *fn, *prev = NULL;
1661 for (fn = root; fn ; ) {
1662 struct fib6_info *leaf = rcu_dereference(fn->leaf);
1665 /* This node is being deleted */
1667 if (plen <= fn->fn_bit)
1673 key = (struct rt6key *)((u8 *)leaf + offset);
1678 if (plen < fn->fn_bit ||
1679 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1682 if (plen == fn->fn_bit)
1685 if (fn->fn_flags & RTN_RTINFO)
1690 * We have more bits to go
1692 if (addr_bit_set(addr, fn->fn_bit))
1693 fn = rcu_dereference(fn->right);
1695 fn = rcu_dereference(fn->left);
1704 struct fib6_node *fib6_locate(struct fib6_node *root,
1705 const struct in6_addr *daddr, int dst_len,
1706 const struct in6_addr *saddr, int src_len,
1709 struct fib6_node *fn;
1711 fn = fib6_locate_1(root, daddr, dst_len,
1712 offsetof(struct fib6_info, fib6_dst),
1715 #ifdef CONFIG_IPV6_SUBTREES
1717 WARN_ON(saddr == NULL);
1719 struct fib6_node *subtree = FIB6_SUBTREE(fn);
1722 fn = fib6_locate_1(subtree, saddr, src_len,
1723 offsetof(struct fib6_info, fib6_src),
1730 if (fn && fn->fn_flags & RTN_RTINFO)
1742 static struct fib6_info *fib6_find_prefix(struct net *net,
1743 struct fib6_table *table,
1744 struct fib6_node *fn)
1746 struct fib6_node *child_left, *child_right;
1748 if (fn->fn_flags & RTN_ROOT)
1749 return net->ipv6.fib6_null_entry;
1752 child_left = rcu_dereference_protected(fn->left,
1753 lockdep_is_held(&table->tb6_lock));
1754 child_right = rcu_dereference_protected(fn->right,
1755 lockdep_is_held(&table->tb6_lock));
1757 return rcu_dereference_protected(child_left->leaf,
1758 lockdep_is_held(&table->tb6_lock));
1760 return rcu_dereference_protected(child_right->leaf,
1761 lockdep_is_held(&table->tb6_lock));
1763 fn = FIB6_SUBTREE(fn);
1769 * Called to trim the tree of intermediate nodes when possible. "fn"
1770 * is the node we want to try and remove.
1771 * Need to own table->tb6_lock
1774 static struct fib6_node *fib6_repair_tree(struct net *net,
1775 struct fib6_table *table,
1776 struct fib6_node *fn)
1780 struct fib6_node *child;
1781 struct fib6_walker *w;
1784 /* Set fn->leaf to null_entry for root node. */
1785 if (fn->fn_flags & RTN_TL_ROOT) {
1786 rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry);
1791 struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1792 lockdep_is_held(&table->tb6_lock));
1793 struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1794 lockdep_is_held(&table->tb6_lock));
1795 struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1796 lockdep_is_held(&table->tb6_lock));
1797 struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1798 lockdep_is_held(&table->tb6_lock));
1799 struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1800 lockdep_is_held(&table->tb6_lock));
1801 struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1802 lockdep_is_held(&table->tb6_lock));
1803 struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1804 lockdep_is_held(&table->tb6_lock));
1805 struct fib6_info *new_fn_leaf;
1807 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1810 WARN_ON(fn->fn_flags & RTN_RTINFO);
1811 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1825 if (children == 3 || FIB6_SUBTREE(fn)
1826 #ifdef CONFIG_IPV6_SUBTREES
1827 /* Subtree root (i.e. fn) may have one child */
1828 || (children && fn->fn_flags & RTN_ROOT)
1831 new_fn_leaf = fib6_find_prefix(net, table, fn);
1834 WARN_ON(!new_fn_leaf);
1835 new_fn_leaf = net->ipv6.fib6_null_entry;
1838 fib6_info_hold(new_fn_leaf);
1839 rcu_assign_pointer(fn->leaf, new_fn_leaf);
1843 #ifdef CONFIG_IPV6_SUBTREES
1844 if (FIB6_SUBTREE(pn) == fn) {
1845 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1846 RCU_INIT_POINTER(pn->subtree, NULL);
1849 WARN_ON(fn->fn_flags & RTN_ROOT);
1852 rcu_assign_pointer(pn->right, child);
1853 else if (pn_l == fn)
1854 rcu_assign_pointer(pn->left, child);
1860 rcu_assign_pointer(child->parent, pn);
1862 #ifdef CONFIG_IPV6_SUBTREES
1866 read_lock(&net->ipv6.fib6_walker_lock);
1867 FOR_WALKERS(net, w) {
1869 if (w->node == fn) {
1870 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1875 if (w->node == fn) {
1878 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1879 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1881 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1882 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1887 read_unlock(&net->ipv6.fib6_walker_lock);
1890 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1893 RCU_INIT_POINTER(pn->leaf, NULL);
1894 fib6_info_release(pn_leaf);
1899 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1900 struct fib6_info __rcu **rtp, struct nl_info *info)
1902 struct fib6_info *leaf, *replace_rt = NULL;
1903 struct fib6_walker *w;
1904 struct fib6_info *rt = rcu_dereference_protected(*rtp,
1905 lockdep_is_held(&table->tb6_lock));
1906 struct net *net = info->nl_net;
1907 bool notify_del = false;
1909 RT6_TRACE("fib6_del_route\n");
1911 /* If the deleted route is the first in the node and it is not part of
1912 * a multipath route, then we need to replace it with the next route
1913 * in the node, if exists.
1915 leaf = rcu_dereference_protected(fn->leaf,
1916 lockdep_is_held(&table->tb6_lock));
1917 if (leaf == rt && !rt->fib6_nsiblings) {
1918 if (rcu_access_pointer(rt->fib6_next))
1919 replace_rt = rcu_dereference_protected(rt->fib6_next,
1920 lockdep_is_held(&table->tb6_lock));
1926 *rtp = rt->fib6_next;
1927 rt->fib6_node = NULL;
1928 net->ipv6.rt6_stats->fib_rt_entries--;
1929 net->ipv6.rt6_stats->fib_discarded_routes++;
1931 /* Reset round-robin state, if necessary */
1932 if (rcu_access_pointer(fn->rr_ptr) == rt)
1935 /* Remove this entry from other siblings */
1936 if (rt->fib6_nsiblings) {
1937 struct fib6_info *sibling, *next_sibling;
1939 /* The route is deleted from a multipath route. If this
1940 * multipath route is the first route in the node, then we need
1941 * to emit a delete notification. Otherwise, we need to skip
1944 if (rt->fib6_metric == leaf->fib6_metric &&
1945 rt6_qualify_for_ecmp(leaf))
1947 list_for_each_entry_safe(sibling, next_sibling,
1948 &rt->fib6_siblings, fib6_siblings)
1949 sibling->fib6_nsiblings--;
1950 rt->fib6_nsiblings = 0;
1951 list_del_init(&rt->fib6_siblings);
1952 rt6_multipath_rebalance(next_sibling);
1955 /* Adjust walkers */
1956 read_lock(&net->ipv6.fib6_walker_lock);
1957 FOR_WALKERS(net, w) {
1958 if (w->state == FWS_C && w->leaf == rt) {
1959 RT6_TRACE("walker %p adjusted by delroute\n", w);
1960 w->leaf = rcu_dereference_protected(rt->fib6_next,
1961 lockdep_is_held(&table->tb6_lock));
1966 read_unlock(&net->ipv6.fib6_walker_lock);
1968 /* If it was last route, call fib6_repair_tree() to:
1969 * 1. For root node, put back null_entry as how the table was created.
1970 * 2. For other nodes, expunge its radix tree node.
1972 if (!rcu_access_pointer(fn->leaf)) {
1973 if (!(fn->fn_flags & RTN_TL_ROOT)) {
1974 fn->fn_flags &= ~RTN_RTINFO;
1975 net->ipv6.rt6_stats->fib_route_nodes--;
1977 fn = fib6_repair_tree(net, table, fn);
1980 fib6_purge_rt(rt, fn, net);
1982 if (!info->skip_notify_kernel) {
1984 call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL,
1986 else if (replace_rt)
1987 call_fib6_entry_notifiers_replace(net, replace_rt);
1989 if (!info->skip_notify)
1990 inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1992 fib6_info_release(rt);
1995 /* Need to own table->tb6_lock */
1996 int fib6_del(struct fib6_info *rt, struct nl_info *info)
1998 struct net *net = info->nl_net;
1999 struct fib6_info __rcu **rtp;
2000 struct fib6_info __rcu **rtp_next;
2001 struct fib6_table *table;
2002 struct fib6_node *fn;
2004 if (rt == net->ipv6.fib6_null_entry)
2007 table = rt->fib6_table;
2008 fn = rcu_dereference_protected(rt->fib6_node,
2009 lockdep_is_held(&table->tb6_lock));
2013 WARN_ON(!(fn->fn_flags & RTN_RTINFO));
2016 * Walk the leaf entries looking for ourself
2019 for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
2020 struct fib6_info *cur = rcu_dereference_protected(*rtp,
2021 lockdep_is_held(&table->tb6_lock));
2023 if (fib6_requires_src(cur))
2024 fib6_routes_require_src_dec(info->nl_net);
2025 fib6_del_route(table, fn, rtp, info);
2028 rtp_next = &cur->fib6_next;
2034 * Tree traversal function.
2036 * Certainly, it is not interrupt safe.
2037 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
2038 * It means, that we can modify tree during walking
2039 * and use this function for garbage collection, clone pruning,
2040 * cleaning tree when a device goes down etc. etc.
2042 * It guarantees that every node will be traversed,
2043 * and that it will be traversed only once.
2045 * Callback function w->func may return:
2046 * 0 -> continue walking.
2047 * positive value -> walking is suspended (used by tree dumps,
2048 * and probably by gc, if it will be split to several slices)
2049 * negative value -> terminate walking.
2051 * The function itself returns:
2052 * 0 -> walk is complete.
2053 * >0 -> walk is incomplete (i.e. suspended)
2054 * <0 -> walk is terminated by an error.
2056 * This function is called with tb6_lock held.
2059 static int fib6_walk_continue(struct fib6_walker *w)
2061 struct fib6_node *fn, *pn, *left, *right;
2063 /* w->root should always be table->tb6_root */
2064 WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
2072 #ifdef CONFIG_IPV6_SUBTREES
2074 if (FIB6_SUBTREE(fn)) {
2075 w->node = FIB6_SUBTREE(fn);
2082 left = rcu_dereference_protected(fn->left, 1);
2085 w->state = FWS_INIT;
2091 right = rcu_dereference_protected(fn->right, 1);
2094 w->state = FWS_INIT;
2098 w->leaf = rcu_dereference_protected(fn->leaf, 1);
2101 if (w->leaf && fn->fn_flags & RTN_RTINFO) {
2122 pn = rcu_dereference_protected(fn->parent, 1);
2123 left = rcu_dereference_protected(pn->left, 1);
2124 right = rcu_dereference_protected(pn->right, 1);
2126 #ifdef CONFIG_IPV6_SUBTREES
2127 if (FIB6_SUBTREE(pn) == fn) {
2128 WARN_ON(!(fn->fn_flags & RTN_ROOT));
2139 w->leaf = rcu_dereference_protected(w->node->leaf, 1);
2149 static int fib6_walk(struct net *net, struct fib6_walker *w)
2153 w->state = FWS_INIT;
2156 fib6_walker_link(net, w);
2157 res = fib6_walk_continue(w);
2159 fib6_walker_unlink(net, w);
2163 static int fib6_clean_node(struct fib6_walker *w)
2166 struct fib6_info *rt;
2167 struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
2168 struct nl_info info = {
2170 .skip_notify = c->skip_notify,
2173 if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
2174 READ_ONCE(w->node->fn_sernum) != c->sernum)
2175 WRITE_ONCE(w->node->fn_sernum, c->sernum);
2178 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
2183 for_each_fib6_walker_rt(w) {
2184 res = c->func(rt, c->arg);
2187 res = fib6_del(rt, &info);
2190 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
2192 rcu_access_pointer(rt->fib6_node),
2198 } else if (res == -2) {
2199 if (WARN_ON(!rt->fib6_nsiblings))
2201 rt = list_last_entry(&rt->fib6_siblings,
2202 struct fib6_info, fib6_siblings);
2212 * Convenient frontend to tree walker.
2214 * func is called on each route.
2215 * It may return -2 -> skip multipath route.
2216 * -1 -> delete this route.
2217 * 0 -> continue walking
2220 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
2221 int (*func)(struct fib6_info *, void *arg),
2222 int sernum, void *arg, bool skip_notify)
2224 struct fib6_cleaner c;
2227 c.w.func = fib6_clean_node;
2230 c.w.skip_in_node = 0;
2235 c.skip_notify = skip_notify;
2237 fib6_walk(net, &c.w);
2240 static void __fib6_clean_all(struct net *net,
2241 int (*func)(struct fib6_info *, void *),
2242 int sernum, void *arg, bool skip_notify)
2244 struct fib6_table *table;
2245 struct hlist_head *head;
2249 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2250 head = &net->ipv6.fib_table_hash[h];
2251 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2252 spin_lock_bh(&table->tb6_lock);
2253 fib6_clean_tree(net, &table->tb6_root,
2254 func, sernum, arg, skip_notify);
2255 spin_unlock_bh(&table->tb6_lock);
2261 void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *),
2264 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, false);
2267 void fib6_clean_all_skip_notify(struct net *net,
2268 int (*func)(struct fib6_info *, void *),
2271 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, true);
2274 static void fib6_flush_trees(struct net *net)
2276 int new_sernum = fib6_new_sernum(net);
2278 __fib6_clean_all(net, NULL, new_sernum, NULL, false);
2282 * Garbage collection
2285 static int fib6_age(struct fib6_info *rt, void *arg)
2287 struct fib6_gc_args *gc_args = arg;
2288 unsigned long now = jiffies;
2291 * check addrconf expiration here.
2292 * Routes are expired even if they are in use.
2295 if (rt->fib6_flags & RTF_EXPIRES && rt->expires) {
2296 if (time_after(now, rt->expires)) {
2297 RT6_TRACE("expiring %p\n", rt);
2303 /* Also age clones in the exception table.
2304 * Note, that clones are aged out
2305 * only if they are not in use now.
2307 rt6_age_exceptions(rt, gc_args, now);
2312 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2314 struct fib6_gc_args gc_args;
2318 spin_lock_bh(&net->ipv6.fib6_gc_lock);
2319 } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2320 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2323 gc_args.timeout = expires ? (int)expires :
2324 net->ipv6.sysctl.ip6_rt_gc_interval;
2327 fib6_clean_all(net, fib6_age, &gc_args);
2329 net->ipv6.ip6_rt_last_gc = now;
2332 mod_timer(&net->ipv6.ip6_fib_timer,
2334 + net->ipv6.sysctl.ip6_rt_gc_interval));
2336 del_timer(&net->ipv6.ip6_fib_timer);
2337 spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2340 static void fib6_gc_timer_cb(struct timer_list *t)
2342 struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2344 fib6_run_gc(0, arg, true);
2347 static int __net_init fib6_net_init(struct net *net)
2349 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2352 err = fib6_notifier_init(net);
2356 spin_lock_init(&net->ipv6.fib6_gc_lock);
2357 rwlock_init(&net->ipv6.fib6_walker_lock);
2358 INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2359 timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2361 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2362 if (!net->ipv6.rt6_stats)
2365 /* Avoid false sharing : Use at least a full cache line */
2366 size = max_t(size_t, size, L1_CACHE_BYTES);
2368 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2369 if (!net->ipv6.fib_table_hash)
2372 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2374 if (!net->ipv6.fib6_main_tbl)
2375 goto out_fib_table_hash;
2377 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2378 rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2379 net->ipv6.fib6_null_entry);
2380 net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2381 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2382 inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2384 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2385 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2387 if (!net->ipv6.fib6_local_tbl)
2388 goto out_fib6_main_tbl;
2389 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2390 rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2391 net->ipv6.fib6_null_entry);
2392 net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2393 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2394 inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2396 fib6_tables_init(net);
2400 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2402 kfree(net->ipv6.fib6_main_tbl);
2405 kfree(net->ipv6.fib_table_hash);
2407 kfree(net->ipv6.rt6_stats);
2409 fib6_notifier_exit(net);
2413 static void fib6_net_exit(struct net *net)
2417 del_timer_sync(&net->ipv6.ip6_fib_timer);
2419 for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2420 struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2421 struct hlist_node *tmp;
2422 struct fib6_table *tb;
2424 hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2425 hlist_del(&tb->tb6_hlist);
2426 fib6_free_table(tb);
2430 kfree(net->ipv6.fib_table_hash);
2431 kfree(net->ipv6.rt6_stats);
2432 fib6_notifier_exit(net);
2435 static struct pernet_operations fib6_net_ops = {
2436 .init = fib6_net_init,
2437 .exit = fib6_net_exit,
2440 int __init fib6_init(void)
2444 fib6_node_kmem = kmem_cache_create("fib6_nodes",
2445 sizeof(struct fib6_node),
2446 0, SLAB_HWCACHE_ALIGN,
2448 if (!fib6_node_kmem)
2451 ret = register_pernet_subsys(&fib6_net_ops);
2453 goto out_kmem_cache_create;
2455 ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2458 goto out_unregister_subsys;
2460 __fib6_flush_trees = fib6_flush_trees;
2464 out_unregister_subsys:
2465 unregister_pernet_subsys(&fib6_net_ops);
2466 out_kmem_cache_create:
2467 kmem_cache_destroy(fib6_node_kmem);
2471 void fib6_gc_cleanup(void)
2473 unregister_pernet_subsys(&fib6_net_ops);
2474 kmem_cache_destroy(fib6_node_kmem);
2477 #ifdef CONFIG_PROC_FS
2478 static int ipv6_route_native_seq_show(struct seq_file *seq, void *v)
2480 struct fib6_info *rt = v;
2481 struct ipv6_route_iter *iter = seq->private;
2482 struct fib6_nh *fib6_nh = rt->fib6_nh;
2483 unsigned int flags = rt->fib6_flags;
2484 const struct net_device *dev;
2487 fib6_nh = nexthop_fib6_nh_bh(rt->nh);
2489 seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
2491 #ifdef CONFIG_IPV6_SUBTREES
2492 seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen);
2494 seq_puts(seq, "00000000000000000000000000000000 00 ");
2496 if (fib6_nh->fib_nh_gw_family) {
2497 flags |= RTF_GATEWAY;
2498 seq_printf(seq, "%pi6", &fib6_nh->fib_nh_gw6);
2500 seq_puts(seq, "00000000000000000000000000000000");
2503 dev = fib6_nh->fib_nh_dev;
2504 seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2505 rt->fib6_metric, refcount_read(&rt->fib6_ref), 0,
2506 flags, dev ? dev->name : "");
2507 iter->w.leaf = NULL;
2511 static int ipv6_route_yield(struct fib6_walker *w)
2513 struct ipv6_route_iter *iter = w->args;
2519 iter->w.leaf = rcu_dereference_protected(
2520 iter->w.leaf->fib6_next,
2521 lockdep_is_held(&iter->tbl->tb6_lock));
2523 if (!iter->skip && iter->w.leaf)
2525 } while (iter->w.leaf);
2530 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2533 memset(&iter->w, 0, sizeof(iter->w));
2534 iter->w.func = ipv6_route_yield;
2535 iter->w.root = &iter->tbl->tb6_root;
2536 iter->w.state = FWS_INIT;
2537 iter->w.node = iter->w.root;
2538 iter->w.args = iter;
2539 iter->sernum = READ_ONCE(iter->w.root->fn_sernum);
2540 INIT_LIST_HEAD(&iter->w.lh);
2541 fib6_walker_link(net, &iter->w);
2544 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2548 struct hlist_node *node;
2551 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2552 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2558 while (!node && h < FIB6_TABLE_HASHSZ) {
2559 node = rcu_dereference_bh(
2560 hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2562 return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2565 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2567 int sernum = READ_ONCE(iter->w.root->fn_sernum);
2569 if (iter->sernum != sernum) {
2570 iter->sernum = sernum;
2571 iter->w.state = FWS_INIT;
2572 iter->w.node = iter->w.root;
2573 WARN_ON(iter->w.skip);
2574 iter->w.skip = iter->w.count;
2578 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2581 struct fib6_info *n;
2582 struct net *net = seq_file_net(seq);
2583 struct ipv6_route_iter *iter = seq->private;
2589 n = rcu_dereference_bh(((struct fib6_info *)v)->fib6_next);
2594 ipv6_route_check_sernum(iter);
2595 spin_lock_bh(&iter->tbl->tb6_lock);
2596 r = fib6_walk_continue(&iter->w);
2597 spin_unlock_bh(&iter->tbl->tb6_lock);
2599 return iter->w.leaf;
2601 fib6_walker_unlink(net, &iter->w);
2604 fib6_walker_unlink(net, &iter->w);
2606 iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2610 ipv6_route_seq_setup_walk(iter, net);
2614 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2617 struct net *net = seq_file_net(seq);
2618 struct ipv6_route_iter *iter = seq->private;
2621 iter->tbl = ipv6_route_seq_next_table(NULL, net);
2627 ipv6_route_seq_setup_walk(iter, net);
2628 return ipv6_route_seq_next(seq, NULL, &p);
2634 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2636 struct fib6_walker *w = &iter->w;
2637 return w->node && !(w->state == FWS_U && w->node == w->root);
2640 static void ipv6_route_native_seq_stop(struct seq_file *seq, void *v)
2643 struct net *net = seq_file_net(seq);
2644 struct ipv6_route_iter *iter = seq->private;
2646 if (ipv6_route_iter_active(iter))
2647 fib6_walker_unlink(net, &iter->w);
2649 rcu_read_unlock_bh();
2652 #if IS_BUILTIN(CONFIG_IPV6) && defined(CONFIG_BPF_SYSCALL)
2653 static int ipv6_route_prog_seq_show(struct bpf_prog *prog,
2654 struct bpf_iter_meta *meta,
2657 struct bpf_iter__ipv6_route ctx;
2661 return bpf_iter_run_prog(prog, &ctx);
2664 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2666 struct ipv6_route_iter *iter = seq->private;
2667 struct bpf_iter_meta meta;
2668 struct bpf_prog *prog;
2672 prog = bpf_iter_get_info(&meta, false);
2674 return ipv6_route_native_seq_show(seq, v);
2676 ret = ipv6_route_prog_seq_show(prog, &meta, v);
2677 iter->w.leaf = NULL;
2682 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2684 struct bpf_iter_meta meta;
2685 struct bpf_prog *prog;
2689 prog = bpf_iter_get_info(&meta, true);
2691 (void)ipv6_route_prog_seq_show(prog, &meta, v);
2694 ipv6_route_native_seq_stop(seq, v);
2697 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2699 return ipv6_route_native_seq_show(seq, v);
2702 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2704 ipv6_route_native_seq_stop(seq, v);
2708 const struct seq_operations ipv6_route_seq_ops = {
2709 .start = ipv6_route_seq_start,
2710 .next = ipv6_route_seq_next,
2711 .stop = ipv6_route_seq_stop,
2712 .show = ipv6_route_seq_show
2714 #endif /* CONFIG_PROC_FS */