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/bpf.h>
19 #include <linux/errno.h>
20 #include <linux/types.h>
21 #include <linux/net.h>
22 #include <linux/route.h>
23 #include <linux/netdevice.h>
24 #include <linux/in6.h>
25 #include <linux/init.h>
26 #include <linux/list.h>
27 #include <linux/slab.h>
31 #include <net/ndisc.h>
32 #include <net/addrconf.h>
33 #include <net/lwtunnel.h>
34 #include <net/fib_notifier.h>
36 #include <net/ip_fib.h>
37 #include <net/ip6_fib.h>
38 #include <net/ip6_route.h>
40 static struct kmem_cache *fib6_node_kmem __read_mostly;
45 int (*func)(struct fib6_info *, void *arg);
51 #ifdef CONFIG_IPV6_SUBTREES
52 #define FWS_INIT FWS_S
54 #define FWS_INIT FWS_L
57 static struct fib6_info *fib6_find_prefix(struct net *net,
58 struct fib6_table *table,
59 struct fib6_node *fn);
60 static struct fib6_node *fib6_repair_tree(struct net *net,
61 struct fib6_table *table,
62 struct fib6_node *fn);
63 static int fib6_walk(struct net *net, struct fib6_walker *w);
64 static int fib6_walk_continue(struct fib6_walker *w);
67 * A routing update causes an increase of the serial number on the
68 * affected subtree. This allows for cached routes to be asynchronously
69 * tested when modifications are made to the destination cache as a
70 * result of redirects, path MTU changes, etc.
73 static void fib6_gc_timer_cb(struct timer_list *t);
75 #define FOR_WALKERS(net, w) \
76 list_for_each_entry(w, &(net)->ipv6.fib6_walkers, lh)
78 static void fib6_walker_link(struct net *net, struct fib6_walker *w)
80 write_lock_bh(&net->ipv6.fib6_walker_lock);
81 list_add(&w->lh, &net->ipv6.fib6_walkers);
82 write_unlock_bh(&net->ipv6.fib6_walker_lock);
85 static void fib6_walker_unlink(struct net *net, struct fib6_walker *w)
87 write_lock_bh(&net->ipv6.fib6_walker_lock);
89 write_unlock_bh(&net->ipv6.fib6_walker_lock);
92 static int fib6_new_sernum(struct net *net)
94 int new, old = atomic_read(&net->ipv6.fib6_sernum);
97 new = old < INT_MAX ? old + 1 : 1;
98 } while (!atomic_try_cmpxchg(&net->ipv6.fib6_sernum, &old, new));
104 FIB6_NO_SERNUM_CHANGE = 0,
107 void fib6_update_sernum(struct net *net, struct fib6_info *f6i)
109 struct fib6_node *fn;
111 fn = rcu_dereference_protected(f6i->fib6_node,
112 lockdep_is_held(&f6i->fib6_table->tb6_lock));
114 WRITE_ONCE(fn->fn_sernum, fib6_new_sernum(net));
118 * Auxiliary address test functions for the radix tree.
120 * These assume a 32bit processor (although it will work on
127 #if defined(__LITTLE_ENDIAN)
128 # define BITOP_BE32_SWIZZLE (0x1F & ~7)
130 # define BITOP_BE32_SWIZZLE 0
133 static __be32 addr_bit_set(const void *token, int fn_bit)
135 const __be32 *addr = token;
138 * 1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
139 * is optimized version of
140 * htonl(1 << ((~fn_bit)&0x1F))
141 * See include/asm-generic/bitops/le.h.
143 return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
147 struct fib6_info *fib6_info_alloc(gfp_t gfp_flags, bool with_fib6_nh)
149 struct fib6_info *f6i;
150 size_t sz = sizeof(*f6i);
153 sz += sizeof(struct fib6_nh);
155 f6i = kzalloc(sz, gfp_flags);
159 /* fib6_siblings is a union with nh_list, so this initializes both */
160 INIT_LIST_HEAD(&f6i->fib6_siblings);
161 refcount_set(&f6i->fib6_ref, 1);
166 void fib6_info_destroy_rcu(struct rcu_head *head)
168 struct fib6_info *f6i = container_of(head, struct fib6_info, rcu);
170 WARN_ON(f6i->fib6_node);
173 nexthop_put(f6i->nh);
175 fib6_nh_release(f6i->fib6_nh);
177 ip_fib_metrics_put(f6i->fib6_metrics);
180 EXPORT_SYMBOL_GPL(fib6_info_destroy_rcu);
182 static struct fib6_node *node_alloc(struct net *net)
184 struct fib6_node *fn;
186 fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
188 net->ipv6.rt6_stats->fib_nodes++;
193 static void node_free_immediate(struct net *net, struct fib6_node *fn)
195 kmem_cache_free(fib6_node_kmem, fn);
196 net->ipv6.rt6_stats->fib_nodes--;
199 static void node_free_rcu(struct rcu_head *head)
201 struct fib6_node *fn = container_of(head, struct fib6_node, rcu);
203 kmem_cache_free(fib6_node_kmem, fn);
206 static void node_free(struct net *net, struct fib6_node *fn)
208 call_rcu(&fn->rcu, node_free_rcu);
209 net->ipv6.rt6_stats->fib_nodes--;
212 static void fib6_free_table(struct fib6_table *table)
214 inetpeer_invalidate_tree(&table->tb6_peers);
218 static void fib6_link_table(struct net *net, struct fib6_table *tb)
223 * Initialize table lock at a single place to give lockdep a key,
224 * tables aren't visible prior to being linked to the list.
226 spin_lock_init(&tb->tb6_lock);
227 h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
230 * No protection necessary, this is the only list mutatation
231 * operation, tables never disappear once they exist.
233 hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
236 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
238 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
240 struct fib6_table *table;
242 table = kzalloc(sizeof(*table), GFP_ATOMIC);
245 rcu_assign_pointer(table->tb6_root.leaf,
246 net->ipv6.fib6_null_entry);
247 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
248 inet_peer_base_init(&table->tb6_peers);
254 struct fib6_table *fib6_new_table(struct net *net, u32 id)
256 struct fib6_table *tb;
260 tb = fib6_get_table(net, id);
264 tb = fib6_alloc_table(net, id);
266 fib6_link_table(net, tb);
270 EXPORT_SYMBOL_GPL(fib6_new_table);
272 struct fib6_table *fib6_get_table(struct net *net, u32 id)
274 struct fib6_table *tb;
275 struct hlist_head *head;
280 h = id & (FIB6_TABLE_HASHSZ - 1);
282 head = &net->ipv6.fib_table_hash[h];
283 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
284 if (tb->tb6_id == id) {
293 EXPORT_SYMBOL_GPL(fib6_get_table);
295 static void __net_init fib6_tables_init(struct net *net)
297 fib6_link_table(net, net->ipv6.fib6_main_tbl);
298 fib6_link_table(net, net->ipv6.fib6_local_tbl);
302 struct fib6_table *fib6_new_table(struct net *net, u32 id)
304 return fib6_get_table(net, id);
307 struct fib6_table *fib6_get_table(struct net *net, u32 id)
309 return net->ipv6.fib6_main_tbl;
312 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
313 const struct sk_buff *skb,
314 int flags, pol_lookup_t lookup)
318 rt = pol_lookup_func(lookup,
319 net, net->ipv6.fib6_main_tbl, fl6, skb, flags);
320 if (rt->dst.error == -EAGAIN) {
321 ip6_rt_put_flags(rt, flags);
322 rt = net->ipv6.ip6_null_entry;
323 if (!(flags & RT6_LOOKUP_F_DST_NOREF))
330 /* called with rcu lock held; no reference taken on fib6_info */
331 int fib6_lookup(struct net *net, int oif, struct flowi6 *fl6,
332 struct fib6_result *res, int flags)
334 return fib6_table_lookup(net, net->ipv6.fib6_main_tbl, oif, fl6,
338 static void __net_init fib6_tables_init(struct net *net)
340 fib6_link_table(net, net->ipv6.fib6_main_tbl);
345 unsigned int fib6_tables_seq_read(struct net *net)
347 unsigned int h, fib_seq = 0;
350 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
351 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
352 struct fib6_table *tb;
354 hlist_for_each_entry_rcu(tb, head, tb6_hlist)
355 fib_seq += tb->fib_seq;
362 static int call_fib6_entry_notifier(struct notifier_block *nb,
363 enum fib_event_type event_type,
364 struct fib6_info *rt,
365 struct netlink_ext_ack *extack)
367 struct fib6_entry_notifier_info info = {
368 .info.extack = extack,
372 return call_fib6_notifier(nb, event_type, &info.info);
375 static int call_fib6_multipath_entry_notifier(struct notifier_block *nb,
376 enum fib_event_type event_type,
377 struct fib6_info *rt,
378 unsigned int nsiblings,
379 struct netlink_ext_ack *extack)
381 struct fib6_entry_notifier_info info = {
382 .info.extack = extack,
384 .nsiblings = nsiblings,
387 return call_fib6_notifier(nb, event_type, &info.info);
390 int call_fib6_entry_notifiers(struct net *net,
391 enum fib_event_type event_type,
392 struct fib6_info *rt,
393 struct netlink_ext_ack *extack)
395 struct fib6_entry_notifier_info info = {
396 .info.extack = extack,
400 rt->fib6_table->fib_seq++;
401 return call_fib6_notifiers(net, event_type, &info.info);
404 int call_fib6_multipath_entry_notifiers(struct net *net,
405 enum fib_event_type event_type,
406 struct fib6_info *rt,
407 unsigned int nsiblings,
408 struct netlink_ext_ack *extack)
410 struct fib6_entry_notifier_info info = {
411 .info.extack = extack,
413 .nsiblings = nsiblings,
416 rt->fib6_table->fib_seq++;
417 return call_fib6_notifiers(net, event_type, &info.info);
420 int call_fib6_entry_notifiers_replace(struct net *net, struct fib6_info *rt)
422 struct fib6_entry_notifier_info info = {
424 .nsiblings = rt->fib6_nsiblings,
427 rt->fib6_table->fib_seq++;
428 return call_fib6_notifiers(net, FIB_EVENT_ENTRY_REPLACE, &info.info);
431 struct fib6_dump_arg {
433 struct notifier_block *nb;
434 struct netlink_ext_ack *extack;
437 static int fib6_rt_dump(struct fib6_info *rt, struct fib6_dump_arg *arg)
439 enum fib_event_type fib_event = FIB_EVENT_ENTRY_REPLACE;
442 if (!rt || rt == arg->net->ipv6.fib6_null_entry)
445 if (rt->fib6_nsiblings)
446 err = call_fib6_multipath_entry_notifier(arg->nb, fib_event,
451 err = call_fib6_entry_notifier(arg->nb, fib_event, rt,
457 static int fib6_node_dump(struct fib6_walker *w)
461 err = fib6_rt_dump(w->leaf, w->args);
466 static int fib6_table_dump(struct net *net, struct fib6_table *tb,
467 struct fib6_walker *w)
471 w->root = &tb->tb6_root;
472 spin_lock_bh(&tb->tb6_lock);
473 err = fib6_walk(net, w);
474 spin_unlock_bh(&tb->tb6_lock);
478 /* Called with rcu_read_lock() */
479 int fib6_tables_dump(struct net *net, struct notifier_block *nb,
480 struct netlink_ext_ack *extack)
482 struct fib6_dump_arg arg;
483 struct fib6_walker *w;
487 w = kzalloc(sizeof(*w), GFP_ATOMIC);
491 w->func = fib6_node_dump;
497 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
498 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
499 struct fib6_table *tb;
501 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
502 err = fib6_table_dump(net, tb, w);
511 /* The tree traversal function should never return a positive value. */
512 return err > 0 ? -EINVAL : err;
515 static int fib6_dump_node(struct fib6_walker *w)
518 struct fib6_info *rt;
520 for_each_fib6_walker_rt(w) {
521 res = rt6_dump_route(rt, w->args, w->skip_in_node);
523 /* Frame is full, suspend walking */
526 /* We'll restart from this node, so if some routes were
527 * already dumped, skip them next time.
529 w->skip_in_node += res;
535 /* Multipath routes are dumped in one route with the
536 * RTA_MULTIPATH attribute. Jump 'rt' to point to the
537 * last sibling of this route (no need to dump the
538 * sibling routes again)
540 if (rt->fib6_nsiblings)
541 rt = list_last_entry(&rt->fib6_siblings,
549 static void fib6_dump_end(struct netlink_callback *cb)
551 struct net *net = sock_net(cb->skb->sk);
552 struct fib6_walker *w = (void *)cb->args[2];
557 fib6_walker_unlink(net, w);
562 cb->done = (void *)cb->args[3];
566 static int fib6_dump_done(struct netlink_callback *cb)
569 return cb->done ? cb->done(cb) : 0;
572 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
573 struct netlink_callback *cb)
575 struct net *net = sock_net(skb->sk);
576 struct fib6_walker *w;
579 w = (void *)cb->args[2];
580 w->root = &table->tb6_root;
582 if (cb->args[4] == 0) {
587 spin_lock_bh(&table->tb6_lock);
588 res = fib6_walk(net, w);
589 spin_unlock_bh(&table->tb6_lock);
592 cb->args[5] = READ_ONCE(w->root->fn_sernum);
595 int sernum = READ_ONCE(w->root->fn_sernum);
596 if (cb->args[5] != sernum) {
597 /* Begin at the root if the tree changed */
598 cb->args[5] = sernum;
606 spin_lock_bh(&table->tb6_lock);
607 res = fib6_walk_continue(w);
608 spin_unlock_bh(&table->tb6_lock);
610 fib6_walker_unlink(net, w);
618 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
620 struct rt6_rtnl_dump_arg arg = { .filter.dump_exceptions = true,
621 .filter.dump_routes = true };
622 const struct nlmsghdr *nlh = cb->nlh;
623 struct net *net = sock_net(skb->sk);
625 unsigned int e = 0, s_e;
626 struct fib6_walker *w;
627 struct fib6_table *tb;
628 struct hlist_head *head;
631 if (cb->strict_check) {
634 err = ip_valid_fib_dump_req(net, nlh, &arg.filter, cb);
637 } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
638 struct rtmsg *rtm = nlmsg_data(nlh);
640 if (rtm->rtm_flags & RTM_F_PREFIX)
641 arg.filter.flags = RTM_F_PREFIX;
644 w = (void *)cb->args[2];
648 * 1. allocate and initialize walker.
650 w = kzalloc(sizeof(*w), GFP_ATOMIC);
653 w->func = fib6_dump_node;
654 cb->args[2] = (long)w;
656 /* 2. hook callback destructor.
658 cb->args[3] = (long)cb->done;
659 cb->done = fib6_dump_done;
668 if (arg.filter.table_id) {
669 tb = fib6_get_table(net, arg.filter.table_id);
671 if (rtnl_msg_family(cb->nlh) != PF_INET6)
674 NL_SET_ERR_MSG_MOD(cb->extack, "FIB table does not exist");
679 res = fib6_dump_table(tb, skb, cb);
690 for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
692 head = &net->ipv6.fib_table_hash[h];
693 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
696 res = fib6_dump_table(tb, skb, cb);
708 res = res < 0 ? res : skb->len;
714 void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val)
719 if (f6i->fib6_metrics == &dst_default_metrics) {
720 struct dst_metrics *p = kzalloc(sizeof(*p), GFP_ATOMIC);
725 refcount_set(&p->refcnt, 1);
726 f6i->fib6_metrics = p;
729 f6i->fib6_metrics->metrics[metric - 1] = val;
735 * return the appropriate node for a routing tree "add" operation
736 * by either creating and inserting or by returning an existing
740 static struct fib6_node *fib6_add_1(struct net *net,
741 struct fib6_table *table,
742 struct fib6_node *root,
743 struct in6_addr *addr, int plen,
744 int offset, int allow_create,
745 int replace_required,
746 struct netlink_ext_ack *extack)
748 struct fib6_node *fn, *in, *ln;
749 struct fib6_node *pn = NULL;
754 RT6_TRACE("fib6_add_1\n");
756 /* insert node in tree */
761 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
762 lockdep_is_held(&table->tb6_lock));
763 key = (struct rt6key *)((u8 *)leaf + offset);
768 if (plen < fn->fn_bit ||
769 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
771 if (replace_required) {
772 NL_SET_ERR_MSG(extack,
773 "Can not replace route - no match found");
774 pr_warn("Can't replace route, no match found\n");
775 return ERR_PTR(-ENOENT);
777 pr_warn("NLM_F_CREATE should be set when creating new route\n");
786 if (plen == fn->fn_bit) {
787 /* clean up an intermediate node */
788 if (!(fn->fn_flags & RTN_RTINFO)) {
789 RCU_INIT_POINTER(fn->leaf, NULL);
790 fib6_info_release(leaf);
791 /* remove null_entry in the root node */
792 } else if (fn->fn_flags & RTN_TL_ROOT &&
793 rcu_access_pointer(fn->leaf) ==
794 net->ipv6.fib6_null_entry) {
795 RCU_INIT_POINTER(fn->leaf, NULL);
802 * We have more bits to go
805 /* Try to walk down on tree. */
806 dir = addr_bit_set(addr, fn->fn_bit);
809 rcu_dereference_protected(fn->right,
810 lockdep_is_held(&table->tb6_lock)) :
811 rcu_dereference_protected(fn->left,
812 lockdep_is_held(&table->tb6_lock));
816 /* We should not create new node because
817 * NLM_F_REPLACE was specified without NLM_F_CREATE
818 * I assume it is safe to require NLM_F_CREATE when
819 * REPLACE flag is used! Later we may want to remove the
820 * check for replace_required, because according
821 * to netlink specification, NLM_F_CREATE
822 * MUST be specified if new route is created.
823 * That would keep IPv6 consistent with IPv4
825 if (replace_required) {
826 NL_SET_ERR_MSG(extack,
827 "Can not replace route - no match found");
828 pr_warn("Can't replace route, no match found\n");
829 return ERR_PTR(-ENOENT);
831 pr_warn("NLM_F_CREATE should be set when creating new route\n");
834 * We walked to the bottom of tree.
835 * Create new leaf node without children.
838 ln = node_alloc(net);
841 return ERR_PTR(-ENOMEM);
843 RCU_INIT_POINTER(ln->parent, pn);
846 rcu_assign_pointer(pn->right, ln);
848 rcu_assign_pointer(pn->left, ln);
855 * split since we don't have a common prefix anymore or
856 * we have a less significant route.
857 * we've to insert an intermediate node on the list
858 * this new node will point to the one we need to create
862 pn = rcu_dereference_protected(fn->parent,
863 lockdep_is_held(&table->tb6_lock));
865 /* find 1st bit in difference between the 2 addrs.
867 See comment in __ipv6_addr_diff: bit may be an invalid value,
868 but if it is >= plen, the value is ignored in any case.
871 bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
876 * (new leaf node)[ln] (old node)[fn]
879 in = node_alloc(net);
880 ln = node_alloc(net);
884 node_free_immediate(net, in);
886 node_free_immediate(net, ln);
887 return ERR_PTR(-ENOMEM);
891 * new intermediate node.
893 * be off since that an address that chooses one of
894 * the branches would not match less specific routes
895 * in the other branch
900 RCU_INIT_POINTER(in->parent, pn);
902 fib6_info_hold(rcu_dereference_protected(in->leaf,
903 lockdep_is_held(&table->tb6_lock)));
905 /* update parent pointer */
907 rcu_assign_pointer(pn->right, in);
909 rcu_assign_pointer(pn->left, in);
913 RCU_INIT_POINTER(ln->parent, in);
914 rcu_assign_pointer(fn->parent, in);
916 if (addr_bit_set(addr, bit)) {
917 rcu_assign_pointer(in->right, ln);
918 rcu_assign_pointer(in->left, fn);
920 rcu_assign_pointer(in->left, ln);
921 rcu_assign_pointer(in->right, fn);
923 } else { /* plen <= bit */
926 * (new leaf node)[ln]
928 * (old node)[fn] NULL
931 ln = node_alloc(net);
934 return ERR_PTR(-ENOMEM);
938 RCU_INIT_POINTER(ln->parent, pn);
940 if (addr_bit_set(&key->addr, plen))
941 RCU_INIT_POINTER(ln->right, fn);
943 RCU_INIT_POINTER(ln->left, fn);
945 rcu_assign_pointer(fn->parent, ln);
948 rcu_assign_pointer(pn->right, ln);
950 rcu_assign_pointer(pn->left, ln);
955 static void __fib6_drop_pcpu_from(struct fib6_nh *fib6_nh,
956 const struct fib6_info *match,
957 const struct fib6_table *table)
961 if (!fib6_nh->rt6i_pcpu)
964 /* release the reference to this fib entry from
965 * all of its cached pcpu routes
967 for_each_possible_cpu(cpu) {
968 struct rt6_info **ppcpu_rt;
969 struct rt6_info *pcpu_rt;
971 ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
974 /* only dropping the 'from' reference if the cached route
975 * is using 'match'. The cached pcpu_rt->from only changes
976 * from a fib6_info to NULL (ip6_dst_destroy); it can never
977 * change from one fib6_info reference to another
979 if (pcpu_rt && rcu_access_pointer(pcpu_rt->from) == match) {
980 struct fib6_info *from;
982 from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
983 fib6_info_release(from);
988 struct fib6_nh_pcpu_arg {
989 struct fib6_info *from;
990 const struct fib6_table *table;
993 static int fib6_nh_drop_pcpu_from(struct fib6_nh *nh, void *_arg)
995 struct fib6_nh_pcpu_arg *arg = _arg;
997 __fib6_drop_pcpu_from(nh, arg->from, arg->table);
1001 static void fib6_drop_pcpu_from(struct fib6_info *f6i,
1002 const struct fib6_table *table)
1004 /* Make sure rt6_make_pcpu_route() wont add other percpu routes
1005 * while we are cleaning them here.
1007 f6i->fib6_destroying = 1;
1008 mb(); /* paired with the cmpxchg() in rt6_make_pcpu_route() */
1011 struct fib6_nh_pcpu_arg arg = {
1016 nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_drop_pcpu_from,
1019 struct fib6_nh *fib6_nh;
1021 fib6_nh = f6i->fib6_nh;
1022 __fib6_drop_pcpu_from(fib6_nh, f6i, table);
1026 static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn,
1029 struct fib6_table *table = rt->fib6_table;
1031 /* Flush all cached dst in exception table */
1032 rt6_flush_exceptions(rt);
1033 fib6_drop_pcpu_from(rt, table);
1035 if (rt->nh && !list_empty(&rt->nh_list))
1036 list_del_init(&rt->nh_list);
1038 if (refcount_read(&rt->fib6_ref) != 1) {
1039 /* This route is used as dummy address holder in some split
1040 * nodes. It is not leaked, but it still holds other resources,
1041 * which must be released in time. So, scan ascendant nodes
1042 * and replace dummy references to this route with references
1043 * to still alive ones.
1046 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
1047 lockdep_is_held(&table->tb6_lock));
1048 struct fib6_info *new_leaf;
1049 if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
1050 new_leaf = fib6_find_prefix(net, table, fn);
1051 fib6_info_hold(new_leaf);
1053 rcu_assign_pointer(fn->leaf, new_leaf);
1054 fib6_info_release(rt);
1056 fn = rcu_dereference_protected(fn->parent,
1057 lockdep_is_held(&table->tb6_lock));
1063 * Insert routing information in a node.
1066 static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt,
1067 struct nl_info *info,
1068 struct netlink_ext_ack *extack)
1070 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
1071 lockdep_is_held(&rt->fib6_table->tb6_lock));
1072 struct fib6_info *iter = NULL;
1073 struct fib6_info __rcu **ins;
1074 struct fib6_info __rcu **fallback_ins = NULL;
1075 int replace = (info->nlh &&
1076 (info->nlh->nlmsg_flags & NLM_F_REPLACE));
1077 int add = (!info->nlh ||
1078 (info->nlh->nlmsg_flags & NLM_F_CREATE));
1080 bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
1081 bool notify_sibling_rt = false;
1082 u16 nlflags = NLM_F_EXCL;
1085 if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
1086 nlflags |= NLM_F_APPEND;
1090 for (iter = leaf; iter;
1091 iter = rcu_dereference_protected(iter->fib6_next,
1092 lockdep_is_held(&rt->fib6_table->tb6_lock))) {
1094 * Search for duplicates
1097 if (iter->fib6_metric == rt->fib6_metric) {
1099 * Same priority level
1102 (info->nlh->nlmsg_flags & NLM_F_EXCL))
1105 nlflags &= ~NLM_F_EXCL;
1107 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
1111 fallback_ins = fallback_ins ?: ins;
1115 if (rt6_duplicate_nexthop(iter, rt)) {
1116 if (rt->fib6_nsiblings)
1117 rt->fib6_nsiblings = 0;
1118 if (!(iter->fib6_flags & RTF_EXPIRES))
1120 if (!(rt->fib6_flags & RTF_EXPIRES))
1121 fib6_clean_expires(iter);
1123 fib6_set_expires(iter, rt->expires);
1126 fib6_metric_set(iter, RTAX_MTU,
1130 /* If we have the same destination and the same metric,
1131 * but not the same gateway, then the route we try to
1132 * add is sibling to this route, increment our counter
1133 * of siblings, and later we will add our route to the
1135 * Only static routes (which don't have flag
1136 * RTF_EXPIRES) are used for ECMPv6.
1138 * To avoid long list, we only had siblings if the
1139 * route have a gateway.
1142 rt6_qualify_for_ecmp(iter))
1143 rt->fib6_nsiblings++;
1146 if (iter->fib6_metric > rt->fib6_metric)
1150 ins = &iter->fib6_next;
1153 if (fallback_ins && !found) {
1154 /* No matching route with same ecmp-able-ness found, replace
1155 * first matching route
1158 iter = rcu_dereference_protected(*ins,
1159 lockdep_is_held(&rt->fib6_table->tb6_lock));
1163 /* Reset round-robin state, if necessary */
1164 if (ins == &fn->leaf)
1167 /* Link this route to others same route. */
1168 if (rt->fib6_nsiblings) {
1169 unsigned int fib6_nsiblings;
1170 struct fib6_info *sibling, *temp_sibling;
1172 /* Find the first route that have the same metric */
1174 notify_sibling_rt = true;
1176 if (sibling->fib6_metric == rt->fib6_metric &&
1177 rt6_qualify_for_ecmp(sibling)) {
1178 list_add_tail(&rt->fib6_siblings,
1179 &sibling->fib6_siblings);
1182 sibling = rcu_dereference_protected(sibling->fib6_next,
1183 lockdep_is_held(&rt->fib6_table->tb6_lock));
1184 notify_sibling_rt = false;
1186 /* For each sibling in the list, increment the counter of
1187 * siblings. BUG() if counters does not match, list of siblings
1191 list_for_each_entry_safe(sibling, temp_sibling,
1192 &rt->fib6_siblings, fib6_siblings) {
1193 sibling->fib6_nsiblings++;
1194 BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings);
1197 BUG_ON(fib6_nsiblings != rt->fib6_nsiblings);
1198 rt6_multipath_rebalance(temp_sibling);
1206 pr_warn("NLM_F_CREATE should be set when creating new route\n");
1209 nlflags |= NLM_F_CREATE;
1211 /* The route should only be notified if it is the first
1212 * route in the node or if it is added as a sibling
1213 * route to the first route in the node.
1215 if (!info->skip_notify_kernel &&
1216 (notify_sibling_rt || ins == &fn->leaf)) {
1217 enum fib_event_type fib_event;
1219 if (notify_sibling_rt)
1220 fib_event = FIB_EVENT_ENTRY_APPEND;
1222 fib_event = FIB_EVENT_ENTRY_REPLACE;
1223 err = call_fib6_entry_notifiers(info->nl_net,
1227 struct fib6_info *sibling, *next_sibling;
1229 /* If the route has siblings, then it first
1230 * needs to be unlinked from them.
1232 if (!rt->fib6_nsiblings)
1235 list_for_each_entry_safe(sibling, next_sibling,
1238 sibling->fib6_nsiblings--;
1239 rt->fib6_nsiblings = 0;
1240 list_del_init(&rt->fib6_siblings);
1241 rt6_multipath_rebalance(next_sibling);
1246 rcu_assign_pointer(rt->fib6_next, iter);
1248 rcu_assign_pointer(rt->fib6_node, fn);
1249 rcu_assign_pointer(*ins, rt);
1250 if (!info->skip_notify)
1251 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1252 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1254 if (!(fn->fn_flags & RTN_RTINFO)) {
1255 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1256 fn->fn_flags |= RTN_RTINFO;
1265 pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1269 if (!info->skip_notify_kernel && ins == &fn->leaf) {
1270 err = call_fib6_entry_notifiers(info->nl_net,
1271 FIB_EVENT_ENTRY_REPLACE,
1278 rcu_assign_pointer(rt->fib6_node, fn);
1279 rt->fib6_next = iter->fib6_next;
1280 rcu_assign_pointer(*ins, rt);
1281 if (!info->skip_notify)
1282 inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1283 if (!(fn->fn_flags & RTN_RTINFO)) {
1284 info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1285 fn->fn_flags |= RTN_RTINFO;
1287 nsiblings = iter->fib6_nsiblings;
1288 iter->fib6_node = NULL;
1289 fib6_purge_rt(iter, fn, info->nl_net);
1290 if (rcu_access_pointer(fn->rr_ptr) == iter)
1292 fib6_info_release(iter);
1295 /* Replacing an ECMP route, remove all siblings */
1296 ins = &rt->fib6_next;
1297 iter = rcu_dereference_protected(*ins,
1298 lockdep_is_held(&rt->fib6_table->tb6_lock));
1300 if (iter->fib6_metric > rt->fib6_metric)
1302 if (rt6_qualify_for_ecmp(iter)) {
1303 *ins = iter->fib6_next;
1304 iter->fib6_node = NULL;
1305 fib6_purge_rt(iter, fn, info->nl_net);
1306 if (rcu_access_pointer(fn->rr_ptr) == iter)
1308 fib6_info_release(iter);
1310 info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1312 ins = &iter->fib6_next;
1314 iter = rcu_dereference_protected(*ins,
1315 lockdep_is_held(&rt->fib6_table->tb6_lock));
1317 WARN_ON(nsiblings != 0);
1324 static void fib6_start_gc(struct net *net, struct fib6_info *rt)
1326 if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1327 (rt->fib6_flags & RTF_EXPIRES))
1328 mod_timer(&net->ipv6.ip6_fib_timer,
1329 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1332 void fib6_force_start_gc(struct net *net)
1334 if (!timer_pending(&net->ipv6.ip6_fib_timer))
1335 mod_timer(&net->ipv6.ip6_fib_timer,
1336 jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1339 static void __fib6_update_sernum_upto_root(struct fib6_info *rt,
1342 struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1343 lockdep_is_held(&rt->fib6_table->tb6_lock));
1345 /* paired with smp_rmb() in fib6_get_cookie_safe() */
1348 WRITE_ONCE(fn->fn_sernum, sernum);
1349 fn = rcu_dereference_protected(fn->parent,
1350 lockdep_is_held(&rt->fib6_table->tb6_lock));
1354 void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt)
1356 __fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1359 /* allow ipv4 to update sernum via ipv6_stub */
1360 void fib6_update_sernum_stub(struct net *net, struct fib6_info *f6i)
1362 spin_lock_bh(&f6i->fib6_table->tb6_lock);
1363 fib6_update_sernum_upto_root(net, f6i);
1364 spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1368 * Add routing information to the routing tree.
1369 * <destination addr>/<source addr>
1370 * with source addr info in sub-trees
1371 * Need to own table->tb6_lock
1374 int fib6_add(struct fib6_node *root, struct fib6_info *rt,
1375 struct nl_info *info, struct netlink_ext_ack *extack)
1377 struct fib6_table *table = rt->fib6_table;
1378 struct fib6_node *fn;
1379 #ifdef CONFIG_IPV6_SUBTREES
1380 struct fib6_node *pn = NULL;
1383 int allow_create = 1;
1384 int replace_required = 0;
1387 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1389 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1390 replace_required = 1;
1392 if (!allow_create && !replace_required)
1393 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1395 fn = fib6_add_1(info->nl_net, table, root,
1396 &rt->fib6_dst.addr, rt->fib6_dst.plen,
1397 offsetof(struct fib6_info, fib6_dst), allow_create,
1398 replace_required, extack);
1405 #ifdef CONFIG_IPV6_SUBTREES
1408 if (rt->fib6_src.plen) {
1409 struct fib6_node *sn;
1411 if (!rcu_access_pointer(fn->subtree)) {
1412 struct fib6_node *sfn;
1424 /* Create subtree root node */
1425 sfn = node_alloc(info->nl_net);
1429 fib6_info_hold(info->nl_net->ipv6.fib6_null_entry);
1430 rcu_assign_pointer(sfn->leaf,
1431 info->nl_net->ipv6.fib6_null_entry);
1432 sfn->fn_flags = RTN_ROOT;
1434 /* Now add the first leaf node to new subtree */
1436 sn = fib6_add_1(info->nl_net, table, sfn,
1437 &rt->fib6_src.addr, rt->fib6_src.plen,
1438 offsetof(struct fib6_info, fib6_src),
1439 allow_create, replace_required, extack);
1442 /* If it is failed, discard just allocated
1443 root, and then (in failure) stale node
1446 node_free_immediate(info->nl_net, sfn);
1451 /* Now link new subtree to main tree */
1452 rcu_assign_pointer(sfn->parent, fn);
1453 rcu_assign_pointer(fn->subtree, sfn);
1455 sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1456 &rt->fib6_src.addr, rt->fib6_src.plen,
1457 offsetof(struct fib6_info, fib6_src),
1458 allow_create, replace_required, extack);
1466 if (!rcu_access_pointer(fn->leaf)) {
1467 if (fn->fn_flags & RTN_TL_ROOT) {
1468 /* put back null_entry for root node */
1469 rcu_assign_pointer(fn->leaf,
1470 info->nl_net->ipv6.fib6_null_entry);
1473 rcu_assign_pointer(fn->leaf, rt);
1480 err = fib6_add_rt2node(fn, rt, info, extack);
1483 list_add(&rt->nh_list, &rt->nh->f6i_list);
1484 __fib6_update_sernum_upto_root(rt, fib6_new_sernum(info->nl_net));
1485 fib6_start_gc(info->nl_net, rt);
1490 #ifdef CONFIG_IPV6_SUBTREES
1492 * If fib6_add_1 has cleared the old leaf pointer in the
1493 * super-tree leaf node we have to find a new one for it.
1496 struct fib6_info *pn_leaf =
1497 rcu_dereference_protected(pn->leaf,
1498 lockdep_is_held(&table->tb6_lock));
1499 if (pn_leaf == rt) {
1501 RCU_INIT_POINTER(pn->leaf, NULL);
1502 fib6_info_release(rt);
1504 if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1505 pn_leaf = fib6_find_prefix(info->nl_net, table,
1509 info->nl_net->ipv6.fib6_null_entry;
1510 fib6_info_hold(pn_leaf);
1511 rcu_assign_pointer(pn->leaf, pn_leaf);
1516 } else if (fib6_requires_src(rt)) {
1517 fib6_routes_require_src_inc(info->nl_net);
1522 /* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1523 * 1. fn is an intermediate node and we failed to add the new
1524 * route to it in both subtree creation failure and fib6_add_rt2node()
1526 * 2. fn is the root node in the table and we fail to add the first
1527 * default route to it.
1530 (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1531 (fn->fn_flags & RTN_TL_ROOT &&
1532 !rcu_access_pointer(fn->leaf))))
1533 fib6_repair_tree(info->nl_net, table, fn);
1538 * Routing tree lookup
1542 struct lookup_args {
1543 int offset; /* key offset on fib6_info */
1544 const struct in6_addr *addr; /* search key */
1547 static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root,
1548 struct lookup_args *args)
1550 struct fib6_node *fn;
1553 if (unlikely(args->offset == 0))
1563 struct fib6_node *next;
1565 dir = addr_bit_set(args->addr, fn->fn_bit);
1567 next = dir ? rcu_dereference(fn->right) :
1568 rcu_dereference(fn->left);
1578 struct fib6_node *subtree = FIB6_SUBTREE(fn);
1580 if (subtree || fn->fn_flags & RTN_RTINFO) {
1581 struct fib6_info *leaf = rcu_dereference(fn->leaf);
1587 key = (struct rt6key *) ((u8 *)leaf + args->offset);
1589 if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1590 #ifdef CONFIG_IPV6_SUBTREES
1592 struct fib6_node *sfn;
1593 sfn = fib6_node_lookup_1(subtree,
1600 if (fn->fn_flags & RTN_RTINFO)
1605 if (fn->fn_flags & RTN_ROOT)
1608 fn = rcu_dereference(fn->parent);
1614 /* called with rcu_read_lock() held
1616 struct fib6_node *fib6_node_lookup(struct fib6_node *root,
1617 const struct in6_addr *daddr,
1618 const struct in6_addr *saddr)
1620 struct fib6_node *fn;
1621 struct lookup_args args[] = {
1623 .offset = offsetof(struct fib6_info, fib6_dst),
1626 #ifdef CONFIG_IPV6_SUBTREES
1628 .offset = offsetof(struct fib6_info, fib6_src),
1633 .offset = 0, /* sentinel */
1637 fn = fib6_node_lookup_1(root, daddr ? args : args + 1);
1638 if (!fn || fn->fn_flags & RTN_TL_ROOT)
1645 * Get node with specified destination prefix (and source prefix,
1646 * if subtrees are used)
1647 * exact_match == true means we try to find fn with exact match of
1648 * the passed in prefix addr
1649 * exact_match == false means we try to find fn with longest prefix
1650 * match of the passed in prefix addr. This is useful for finding fn
1651 * for cached route as it will be stored in the exception table under
1652 * the node with longest prefix length.
1656 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1657 const struct in6_addr *addr,
1658 int plen, int offset,
1661 struct fib6_node *fn, *prev = NULL;
1663 for (fn = root; fn ; ) {
1664 struct fib6_info *leaf = rcu_dereference(fn->leaf);
1667 /* This node is being deleted */
1669 if (plen <= fn->fn_bit)
1675 key = (struct rt6key *)((u8 *)leaf + offset);
1680 if (plen < fn->fn_bit ||
1681 !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1684 if (plen == fn->fn_bit)
1687 if (fn->fn_flags & RTN_RTINFO)
1692 * We have more bits to go
1694 if (addr_bit_set(addr, fn->fn_bit))
1695 fn = rcu_dereference(fn->right);
1697 fn = rcu_dereference(fn->left);
1706 struct fib6_node *fib6_locate(struct fib6_node *root,
1707 const struct in6_addr *daddr, int dst_len,
1708 const struct in6_addr *saddr, int src_len,
1711 struct fib6_node *fn;
1713 fn = fib6_locate_1(root, daddr, dst_len,
1714 offsetof(struct fib6_info, fib6_dst),
1717 #ifdef CONFIG_IPV6_SUBTREES
1719 WARN_ON(saddr == NULL);
1721 struct fib6_node *subtree = FIB6_SUBTREE(fn);
1724 fn = fib6_locate_1(subtree, saddr, src_len,
1725 offsetof(struct fib6_info, fib6_src),
1732 if (fn && fn->fn_flags & RTN_RTINFO)
1744 static struct fib6_info *fib6_find_prefix(struct net *net,
1745 struct fib6_table *table,
1746 struct fib6_node *fn)
1748 struct fib6_node *child_left, *child_right;
1750 if (fn->fn_flags & RTN_ROOT)
1751 return net->ipv6.fib6_null_entry;
1754 child_left = rcu_dereference_protected(fn->left,
1755 lockdep_is_held(&table->tb6_lock));
1756 child_right = rcu_dereference_protected(fn->right,
1757 lockdep_is_held(&table->tb6_lock));
1759 return rcu_dereference_protected(child_left->leaf,
1760 lockdep_is_held(&table->tb6_lock));
1762 return rcu_dereference_protected(child_right->leaf,
1763 lockdep_is_held(&table->tb6_lock));
1765 fn = FIB6_SUBTREE(fn);
1771 * Called to trim the tree of intermediate nodes when possible. "fn"
1772 * is the node we want to try and remove.
1773 * Need to own table->tb6_lock
1776 static struct fib6_node *fib6_repair_tree(struct net *net,
1777 struct fib6_table *table,
1778 struct fib6_node *fn)
1782 struct fib6_node *child;
1783 struct fib6_walker *w;
1786 /* Set fn->leaf to null_entry for root node. */
1787 if (fn->fn_flags & RTN_TL_ROOT) {
1788 rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry);
1793 struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1794 lockdep_is_held(&table->tb6_lock));
1795 struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1796 lockdep_is_held(&table->tb6_lock));
1797 struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1798 lockdep_is_held(&table->tb6_lock));
1799 struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1800 lockdep_is_held(&table->tb6_lock));
1801 struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1802 lockdep_is_held(&table->tb6_lock));
1803 struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1804 lockdep_is_held(&table->tb6_lock));
1805 struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1806 lockdep_is_held(&table->tb6_lock));
1807 struct fib6_info *new_fn_leaf;
1809 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1812 WARN_ON(fn->fn_flags & RTN_RTINFO);
1813 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1827 if (children == 3 || FIB6_SUBTREE(fn)
1828 #ifdef CONFIG_IPV6_SUBTREES
1829 /* Subtree root (i.e. fn) may have one child */
1830 || (children && fn->fn_flags & RTN_ROOT)
1833 new_fn_leaf = fib6_find_prefix(net, table, fn);
1836 WARN_ON(!new_fn_leaf);
1837 new_fn_leaf = net->ipv6.fib6_null_entry;
1840 fib6_info_hold(new_fn_leaf);
1841 rcu_assign_pointer(fn->leaf, new_fn_leaf);
1845 #ifdef CONFIG_IPV6_SUBTREES
1846 if (FIB6_SUBTREE(pn) == fn) {
1847 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1848 RCU_INIT_POINTER(pn->subtree, NULL);
1851 WARN_ON(fn->fn_flags & RTN_ROOT);
1854 rcu_assign_pointer(pn->right, child);
1855 else if (pn_l == fn)
1856 rcu_assign_pointer(pn->left, child);
1862 rcu_assign_pointer(child->parent, pn);
1864 #ifdef CONFIG_IPV6_SUBTREES
1868 read_lock(&net->ipv6.fib6_walker_lock);
1869 FOR_WALKERS(net, w) {
1871 if (w->node == fn) {
1872 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1877 if (w->node == fn) {
1880 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1881 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1883 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1884 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1889 read_unlock(&net->ipv6.fib6_walker_lock);
1892 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1895 RCU_INIT_POINTER(pn->leaf, NULL);
1896 fib6_info_release(pn_leaf);
1901 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1902 struct fib6_info __rcu **rtp, struct nl_info *info)
1904 struct fib6_info *leaf, *replace_rt = NULL;
1905 struct fib6_walker *w;
1906 struct fib6_info *rt = rcu_dereference_protected(*rtp,
1907 lockdep_is_held(&table->tb6_lock));
1908 struct net *net = info->nl_net;
1909 bool notify_del = false;
1911 RT6_TRACE("fib6_del_route\n");
1913 /* If the deleted route is the first in the node and it is not part of
1914 * a multipath route, then we need to replace it with the next route
1915 * in the node, if exists.
1917 leaf = rcu_dereference_protected(fn->leaf,
1918 lockdep_is_held(&table->tb6_lock));
1919 if (leaf == rt && !rt->fib6_nsiblings) {
1920 if (rcu_access_pointer(rt->fib6_next))
1921 replace_rt = rcu_dereference_protected(rt->fib6_next,
1922 lockdep_is_held(&table->tb6_lock));
1928 *rtp = rt->fib6_next;
1929 rt->fib6_node = NULL;
1930 net->ipv6.rt6_stats->fib_rt_entries--;
1931 net->ipv6.rt6_stats->fib_discarded_routes++;
1933 /* Reset round-robin state, if necessary */
1934 if (rcu_access_pointer(fn->rr_ptr) == rt)
1937 /* Remove this entry from other siblings */
1938 if (rt->fib6_nsiblings) {
1939 struct fib6_info *sibling, *next_sibling;
1941 /* The route is deleted from a multipath route. If this
1942 * multipath route is the first route in the node, then we need
1943 * to emit a delete notification. Otherwise, we need to skip
1946 if (rt->fib6_metric == leaf->fib6_metric &&
1947 rt6_qualify_for_ecmp(leaf))
1949 list_for_each_entry_safe(sibling, next_sibling,
1950 &rt->fib6_siblings, fib6_siblings)
1951 sibling->fib6_nsiblings--;
1952 rt->fib6_nsiblings = 0;
1953 list_del_init(&rt->fib6_siblings);
1954 rt6_multipath_rebalance(next_sibling);
1957 /* Adjust walkers */
1958 read_lock(&net->ipv6.fib6_walker_lock);
1959 FOR_WALKERS(net, w) {
1960 if (w->state == FWS_C && w->leaf == rt) {
1961 RT6_TRACE("walker %p adjusted by delroute\n", w);
1962 w->leaf = rcu_dereference_protected(rt->fib6_next,
1963 lockdep_is_held(&table->tb6_lock));
1968 read_unlock(&net->ipv6.fib6_walker_lock);
1970 /* If it was last route, call fib6_repair_tree() to:
1971 * 1. For root node, put back null_entry as how the table was created.
1972 * 2. For other nodes, expunge its radix tree node.
1974 if (!rcu_access_pointer(fn->leaf)) {
1975 if (!(fn->fn_flags & RTN_TL_ROOT)) {
1976 fn->fn_flags &= ~RTN_RTINFO;
1977 net->ipv6.rt6_stats->fib_route_nodes--;
1979 fn = fib6_repair_tree(net, table, fn);
1982 fib6_purge_rt(rt, fn, net);
1984 if (!info->skip_notify_kernel) {
1986 call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL,
1988 else if (replace_rt)
1989 call_fib6_entry_notifiers_replace(net, replace_rt);
1991 if (!info->skip_notify)
1992 inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1994 fib6_info_release(rt);
1997 /* Need to own table->tb6_lock */
1998 int fib6_del(struct fib6_info *rt, struct nl_info *info)
2000 struct net *net = info->nl_net;
2001 struct fib6_info __rcu **rtp;
2002 struct fib6_info __rcu **rtp_next;
2003 struct fib6_table *table;
2004 struct fib6_node *fn;
2006 if (rt == net->ipv6.fib6_null_entry)
2009 table = rt->fib6_table;
2010 fn = rcu_dereference_protected(rt->fib6_node,
2011 lockdep_is_held(&table->tb6_lock));
2015 WARN_ON(!(fn->fn_flags & RTN_RTINFO));
2018 * Walk the leaf entries looking for ourself
2021 for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
2022 struct fib6_info *cur = rcu_dereference_protected(*rtp,
2023 lockdep_is_held(&table->tb6_lock));
2025 if (fib6_requires_src(cur))
2026 fib6_routes_require_src_dec(info->nl_net);
2027 fib6_del_route(table, fn, rtp, info);
2030 rtp_next = &cur->fib6_next;
2036 * Tree traversal function.
2038 * Certainly, it is not interrupt safe.
2039 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
2040 * It means, that we can modify tree during walking
2041 * and use this function for garbage collection, clone pruning,
2042 * cleaning tree when a device goes down etc. etc.
2044 * It guarantees that every node will be traversed,
2045 * and that it will be traversed only once.
2047 * Callback function w->func may return:
2048 * 0 -> continue walking.
2049 * positive value -> walking is suspended (used by tree dumps,
2050 * and probably by gc, if it will be split to several slices)
2051 * negative value -> terminate walking.
2053 * The function itself returns:
2054 * 0 -> walk is complete.
2055 * >0 -> walk is incomplete (i.e. suspended)
2056 * <0 -> walk is terminated by an error.
2058 * This function is called with tb6_lock held.
2061 static int fib6_walk_continue(struct fib6_walker *w)
2063 struct fib6_node *fn, *pn, *left, *right;
2065 /* w->root should always be table->tb6_root */
2066 WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
2074 #ifdef CONFIG_IPV6_SUBTREES
2076 if (FIB6_SUBTREE(fn)) {
2077 w->node = FIB6_SUBTREE(fn);
2084 left = rcu_dereference_protected(fn->left, 1);
2087 w->state = FWS_INIT;
2093 right = rcu_dereference_protected(fn->right, 1);
2096 w->state = FWS_INIT;
2100 w->leaf = rcu_dereference_protected(fn->leaf, 1);
2103 if (w->leaf && fn->fn_flags & RTN_RTINFO) {
2124 pn = rcu_dereference_protected(fn->parent, 1);
2125 left = rcu_dereference_protected(pn->left, 1);
2126 right = rcu_dereference_protected(pn->right, 1);
2128 #ifdef CONFIG_IPV6_SUBTREES
2129 if (FIB6_SUBTREE(pn) == fn) {
2130 WARN_ON(!(fn->fn_flags & RTN_ROOT));
2141 w->leaf = rcu_dereference_protected(w->node->leaf, 1);
2151 static int fib6_walk(struct net *net, struct fib6_walker *w)
2155 w->state = FWS_INIT;
2158 fib6_walker_link(net, w);
2159 res = fib6_walk_continue(w);
2161 fib6_walker_unlink(net, w);
2165 static int fib6_clean_node(struct fib6_walker *w)
2168 struct fib6_info *rt;
2169 struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
2170 struct nl_info info = {
2172 .skip_notify = c->skip_notify,
2175 if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
2176 READ_ONCE(w->node->fn_sernum) != c->sernum)
2177 WRITE_ONCE(w->node->fn_sernum, c->sernum);
2180 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
2185 for_each_fib6_walker_rt(w) {
2186 res = c->func(rt, c->arg);
2189 res = fib6_del(rt, &info);
2192 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
2194 rcu_access_pointer(rt->fib6_node),
2200 } else if (res == -2) {
2201 if (WARN_ON(!rt->fib6_nsiblings))
2203 rt = list_last_entry(&rt->fib6_siblings,
2204 struct fib6_info, fib6_siblings);
2214 * Convenient frontend to tree walker.
2216 * func is called on each route.
2217 * It may return -2 -> skip multipath route.
2218 * -1 -> delete this route.
2219 * 0 -> continue walking
2222 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
2223 int (*func)(struct fib6_info *, void *arg),
2224 int sernum, void *arg, bool skip_notify)
2226 struct fib6_cleaner c;
2229 c.w.func = fib6_clean_node;
2232 c.w.skip_in_node = 0;
2237 c.skip_notify = skip_notify;
2239 fib6_walk(net, &c.w);
2242 static void __fib6_clean_all(struct net *net,
2243 int (*func)(struct fib6_info *, void *),
2244 int sernum, void *arg, bool skip_notify)
2246 struct fib6_table *table;
2247 struct hlist_head *head;
2251 for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2252 head = &net->ipv6.fib_table_hash[h];
2253 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2254 spin_lock_bh(&table->tb6_lock);
2255 fib6_clean_tree(net, &table->tb6_root,
2256 func, sernum, arg, skip_notify);
2257 spin_unlock_bh(&table->tb6_lock);
2263 void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *),
2266 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, false);
2269 void fib6_clean_all_skip_notify(struct net *net,
2270 int (*func)(struct fib6_info *, void *),
2273 __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, true);
2276 static void fib6_flush_trees(struct net *net)
2278 int new_sernum = fib6_new_sernum(net);
2280 __fib6_clean_all(net, NULL, new_sernum, NULL, false);
2284 * Garbage collection
2287 static int fib6_age(struct fib6_info *rt, void *arg)
2289 struct fib6_gc_args *gc_args = arg;
2290 unsigned long now = jiffies;
2293 * check addrconf expiration here.
2294 * Routes are expired even if they are in use.
2297 if (rt->fib6_flags & RTF_EXPIRES && rt->expires) {
2298 if (time_after(now, rt->expires)) {
2299 RT6_TRACE("expiring %p\n", rt);
2305 /* Also age clones in the exception table.
2306 * Note, that clones are aged out
2307 * only if they are not in use now.
2309 rt6_age_exceptions(rt, gc_args, now);
2314 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2316 struct fib6_gc_args gc_args;
2320 spin_lock_bh(&net->ipv6.fib6_gc_lock);
2321 } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2322 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2325 gc_args.timeout = expires ? (int)expires :
2326 net->ipv6.sysctl.ip6_rt_gc_interval;
2329 fib6_clean_all(net, fib6_age, &gc_args);
2331 net->ipv6.ip6_rt_last_gc = now;
2334 mod_timer(&net->ipv6.ip6_fib_timer,
2336 + net->ipv6.sysctl.ip6_rt_gc_interval));
2338 del_timer(&net->ipv6.ip6_fib_timer);
2339 spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2342 static void fib6_gc_timer_cb(struct timer_list *t)
2344 struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2346 fib6_run_gc(0, arg, true);
2349 static int __net_init fib6_net_init(struct net *net)
2351 size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2354 err = fib6_notifier_init(net);
2358 /* Default to 3-tuple */
2359 net->ipv6.sysctl.multipath_hash_fields =
2360 FIB_MULTIPATH_HASH_FIELD_DEFAULT_MASK;
2362 spin_lock_init(&net->ipv6.fib6_gc_lock);
2363 rwlock_init(&net->ipv6.fib6_walker_lock);
2364 INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2365 timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2367 net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2368 if (!net->ipv6.rt6_stats)
2371 /* Avoid false sharing : Use at least a full cache line */
2372 size = max_t(size_t, size, L1_CACHE_BYTES);
2374 net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2375 if (!net->ipv6.fib_table_hash)
2378 net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2380 if (!net->ipv6.fib6_main_tbl)
2381 goto out_fib_table_hash;
2383 net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2384 rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2385 net->ipv6.fib6_null_entry);
2386 net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2387 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2388 inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2390 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2391 net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2393 if (!net->ipv6.fib6_local_tbl)
2394 goto out_fib6_main_tbl;
2395 net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2396 rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2397 net->ipv6.fib6_null_entry);
2398 net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2399 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2400 inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2402 fib6_tables_init(net);
2406 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2408 kfree(net->ipv6.fib6_main_tbl);
2411 kfree(net->ipv6.fib_table_hash);
2413 kfree(net->ipv6.rt6_stats);
2415 fib6_notifier_exit(net);
2419 static void fib6_net_exit(struct net *net)
2423 del_timer_sync(&net->ipv6.ip6_fib_timer);
2425 for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2426 struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2427 struct hlist_node *tmp;
2428 struct fib6_table *tb;
2430 hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2431 hlist_del(&tb->tb6_hlist);
2432 fib6_free_table(tb);
2436 kfree(net->ipv6.fib_table_hash);
2437 kfree(net->ipv6.rt6_stats);
2438 fib6_notifier_exit(net);
2441 static struct pernet_operations fib6_net_ops = {
2442 .init = fib6_net_init,
2443 .exit = fib6_net_exit,
2446 int __init fib6_init(void)
2450 fib6_node_kmem = kmem_cache_create("fib6_nodes",
2451 sizeof(struct fib6_node), 0,
2452 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT,
2454 if (!fib6_node_kmem)
2457 ret = register_pernet_subsys(&fib6_net_ops);
2459 goto out_kmem_cache_create;
2461 ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2464 goto out_unregister_subsys;
2466 __fib6_flush_trees = fib6_flush_trees;
2470 out_unregister_subsys:
2471 unregister_pernet_subsys(&fib6_net_ops);
2472 out_kmem_cache_create:
2473 kmem_cache_destroy(fib6_node_kmem);
2477 void fib6_gc_cleanup(void)
2479 unregister_pernet_subsys(&fib6_net_ops);
2480 kmem_cache_destroy(fib6_node_kmem);
2483 #ifdef CONFIG_PROC_FS
2484 static int ipv6_route_native_seq_show(struct seq_file *seq, void *v)
2486 struct fib6_info *rt = v;
2487 struct ipv6_route_iter *iter = seq->private;
2488 struct fib6_nh *fib6_nh = rt->fib6_nh;
2489 unsigned int flags = rt->fib6_flags;
2490 const struct net_device *dev;
2493 fib6_nh = nexthop_fib6_nh(rt->nh);
2495 seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
2497 #ifdef CONFIG_IPV6_SUBTREES
2498 seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen);
2500 seq_puts(seq, "00000000000000000000000000000000 00 ");
2502 if (fib6_nh->fib_nh_gw_family) {
2503 flags |= RTF_GATEWAY;
2504 seq_printf(seq, "%pi6", &fib6_nh->fib_nh_gw6);
2506 seq_puts(seq, "00000000000000000000000000000000");
2509 dev = fib6_nh->fib_nh_dev;
2510 seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2511 rt->fib6_metric, refcount_read(&rt->fib6_ref), 0,
2512 flags, dev ? dev->name : "");
2513 iter->w.leaf = NULL;
2517 static int ipv6_route_yield(struct fib6_walker *w)
2519 struct ipv6_route_iter *iter = w->args;
2525 iter->w.leaf = rcu_dereference_protected(
2526 iter->w.leaf->fib6_next,
2527 lockdep_is_held(&iter->tbl->tb6_lock));
2529 if (!iter->skip && iter->w.leaf)
2531 } while (iter->w.leaf);
2536 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2539 memset(&iter->w, 0, sizeof(iter->w));
2540 iter->w.func = ipv6_route_yield;
2541 iter->w.root = &iter->tbl->tb6_root;
2542 iter->w.state = FWS_INIT;
2543 iter->w.node = iter->w.root;
2544 iter->w.args = iter;
2545 iter->sernum = READ_ONCE(iter->w.root->fn_sernum);
2546 INIT_LIST_HEAD(&iter->w.lh);
2547 fib6_walker_link(net, &iter->w);
2550 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2554 struct hlist_node *node;
2557 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2558 node = rcu_dereference(hlist_next_rcu(&tbl->tb6_hlist));
2564 while (!node && h < FIB6_TABLE_HASHSZ) {
2565 node = rcu_dereference(
2566 hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2568 return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2571 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2573 int sernum = READ_ONCE(iter->w.root->fn_sernum);
2575 if (iter->sernum != sernum) {
2576 iter->sernum = sernum;
2577 iter->w.state = FWS_INIT;
2578 iter->w.node = iter->w.root;
2579 WARN_ON(iter->w.skip);
2580 iter->w.skip = iter->w.count;
2584 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2587 struct fib6_info *n;
2588 struct net *net = seq_file_net(seq);
2589 struct ipv6_route_iter *iter = seq->private;
2595 n = rcu_dereference(((struct fib6_info *)v)->fib6_next);
2600 ipv6_route_check_sernum(iter);
2601 spin_lock_bh(&iter->tbl->tb6_lock);
2602 r = fib6_walk_continue(&iter->w);
2603 spin_unlock_bh(&iter->tbl->tb6_lock);
2605 return iter->w.leaf;
2607 fib6_walker_unlink(net, &iter->w);
2610 fib6_walker_unlink(net, &iter->w);
2612 iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2616 ipv6_route_seq_setup_walk(iter, net);
2620 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2623 struct net *net = seq_file_net(seq);
2624 struct ipv6_route_iter *iter = seq->private;
2627 iter->tbl = ipv6_route_seq_next_table(NULL, net);
2633 ipv6_route_seq_setup_walk(iter, net);
2634 return ipv6_route_seq_next(seq, NULL, &p);
2640 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2642 struct fib6_walker *w = &iter->w;
2643 return w->node && !(w->state == FWS_U && w->node == w->root);
2646 static void ipv6_route_native_seq_stop(struct seq_file *seq, void *v)
2649 struct net *net = seq_file_net(seq);
2650 struct ipv6_route_iter *iter = seq->private;
2652 if (ipv6_route_iter_active(iter))
2653 fib6_walker_unlink(net, &iter->w);
2658 #if IS_BUILTIN(CONFIG_IPV6) && defined(CONFIG_BPF_SYSCALL)
2659 static int ipv6_route_prog_seq_show(struct bpf_prog *prog,
2660 struct bpf_iter_meta *meta,
2663 struct bpf_iter__ipv6_route ctx;
2667 return bpf_iter_run_prog(prog, &ctx);
2670 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2672 struct ipv6_route_iter *iter = seq->private;
2673 struct bpf_iter_meta meta;
2674 struct bpf_prog *prog;
2678 prog = bpf_iter_get_info(&meta, false);
2680 return ipv6_route_native_seq_show(seq, v);
2682 ret = ipv6_route_prog_seq_show(prog, &meta, v);
2683 iter->w.leaf = NULL;
2688 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2690 struct bpf_iter_meta meta;
2691 struct bpf_prog *prog;
2695 prog = bpf_iter_get_info(&meta, true);
2697 (void)ipv6_route_prog_seq_show(prog, &meta, v);
2700 ipv6_route_native_seq_stop(seq, v);
2703 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2705 return ipv6_route_native_seq_show(seq, v);
2708 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2710 ipv6_route_native_seq_stop(seq, v);
2714 const struct seq_operations ipv6_route_seq_ops = {
2715 .start = ipv6_route_seq_start,
2716 .next = ipv6_route_seq_next,
2717 .stop = ipv6_route_seq_stop,
2718 .show = ipv6_route_seq_show
2720 #endif /* CONFIG_PROC_FS */