2 * Linux INET6 implementation
3 * Forwarding Information Database
6 * Pedro Roque <roque@di.fc.ul.pt>
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
14 * Yuji SEKIYA @USAGI: Support default route on router node;
15 * remove ip6_null_entry from the top of
17 * Ville Nuorvala: Fixed routing subtrees.
20 #define pr_fmt(fmt) "IPv6: " fmt
22 #include <linux/errno.h>
23 #include <linux/types.h>
24 #include <linux/net.h>
25 #include <linux/route.h>
26 #include <linux/netdevice.h>
27 #include <linux/in6.h>
28 #include <linux/init.h>
29 #include <linux/list.h>
30 #include <linux/slab.h>
33 #include <net/ndisc.h>
34 #include <net/addrconf.h>
35 #include <net/lwtunnel.h>
36 #include <net/fib_notifier.h>
38 #include <net/ip6_fib.h>
39 #include <net/ip6_route.h>
41 static struct kmem_cache
*fib6_node_kmem __read_mostly
;
46 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
)
97 old
= atomic_read(&net
->ipv6
.fib6_sernum
);
98 new = old
< INT_MAX
? old
+ 1 : 1;
99 } while (atomic_cmpxchg(&net
->ipv6
.fib6_sernum
,
105 FIB6_NO_SERNUM_CHANGE
= 0,
108 void fib6_update_sernum(struct net
*net
, struct fib6_info
*f6i
)
110 struct fib6_node
*fn
;
112 fn
= rcu_dereference_protected(f6i
->fib6_node
,
113 lockdep_is_held(&f6i
->fib6_table
->tb6_lock
));
115 fn
->fn_sernum
= fib6_new_sernum(net
);
119 * Auxiliary address test functions for the radix tree.
121 * These assume a 32bit processor (although it will work on
128 #if defined(__LITTLE_ENDIAN)
129 # define BITOP_BE32_SWIZZLE (0x1F & ~7)
131 # define BITOP_BE32_SWIZZLE 0
134 static __be32
addr_bit_set(const void *token
, int fn_bit
)
136 const __be32
*addr
= token
;
139 * 1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
140 * is optimized version of
141 * htonl(1 << ((~fn_bit)&0x1F))
142 * See include/asm-generic/bitops/le.h.
144 return (__force __be32
)(1 << ((~fn_bit
^ BITOP_BE32_SWIZZLE
) & 0x1f)) &
148 struct fib6_info
*fib6_info_alloc(gfp_t gfp_flags
)
150 struct fib6_info
*f6i
;
152 f6i
= kzalloc(sizeof(*f6i
), gfp_flags
);
156 f6i
->rt6i_pcpu
= alloc_percpu_gfp(struct rt6_info
*, gfp_flags
);
157 if (!f6i
->rt6i_pcpu
) {
162 INIT_LIST_HEAD(&f6i
->fib6_siblings
);
163 f6i
->fib6_metrics
= (struct dst_metrics
*)&dst_default_metrics
;
165 atomic_inc(&f6i
->fib6_ref
);
170 void fib6_info_destroy_rcu(struct rcu_head
*head
)
172 struct fib6_info
*f6i
= container_of(head
, struct fib6_info
, rcu
);
173 struct rt6_exception_bucket
*bucket
;
174 struct dst_metrics
*m
;
176 WARN_ON(f6i
->fib6_node
);
178 bucket
= rcu_dereference_protected(f6i
->rt6i_exception_bucket
, 1);
180 f6i
->rt6i_exception_bucket
= NULL
;
184 if (f6i
->rt6i_pcpu
) {
187 for_each_possible_cpu(cpu
) {
188 struct rt6_info
**ppcpu_rt
;
189 struct rt6_info
*pcpu_rt
;
191 ppcpu_rt
= per_cpu_ptr(f6i
->rt6i_pcpu
, cpu
);
194 dst_dev_put(&pcpu_rt
->dst
);
195 dst_release(&pcpu_rt
->dst
);
201 lwtstate_put(f6i
->fib6_nh
.nh_lwtstate
);
203 if (f6i
->fib6_nh
.nh_dev
)
204 dev_put(f6i
->fib6_nh
.nh_dev
);
206 m
= f6i
->fib6_metrics
;
207 if (m
!= &dst_default_metrics
&& refcount_dec_and_test(&m
->refcnt
))
212 EXPORT_SYMBOL_GPL(fib6_info_destroy_rcu
);
214 static struct fib6_node
*node_alloc(struct net
*net
)
216 struct fib6_node
*fn
;
218 fn
= kmem_cache_zalloc(fib6_node_kmem
, GFP_ATOMIC
);
220 net
->ipv6
.rt6_stats
->fib_nodes
++;
225 static void node_free_immediate(struct net
*net
, struct fib6_node
*fn
)
227 kmem_cache_free(fib6_node_kmem
, fn
);
228 net
->ipv6
.rt6_stats
->fib_nodes
--;
231 static void node_free_rcu(struct rcu_head
*head
)
233 struct fib6_node
*fn
= container_of(head
, struct fib6_node
, rcu
);
235 kmem_cache_free(fib6_node_kmem
, fn
);
238 static void node_free(struct net
*net
, struct fib6_node
*fn
)
240 call_rcu(&fn
->rcu
, node_free_rcu
);
241 net
->ipv6
.rt6_stats
->fib_nodes
--;
244 static void fib6_free_table(struct fib6_table
*table
)
246 inetpeer_invalidate_tree(&table
->tb6_peers
);
250 static void fib6_link_table(struct net
*net
, struct fib6_table
*tb
)
255 * Initialize table lock at a single place to give lockdep a key,
256 * tables aren't visible prior to being linked to the list.
258 spin_lock_init(&tb
->tb6_lock
);
259 h
= tb
->tb6_id
& (FIB6_TABLE_HASHSZ
- 1);
262 * No protection necessary, this is the only list mutatation
263 * operation, tables never disappear once they exist.
265 hlist_add_head_rcu(&tb
->tb6_hlist
, &net
->ipv6
.fib_table_hash
[h
]);
268 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
270 static struct fib6_table
*fib6_alloc_table(struct net
*net
, u32 id
)
272 struct fib6_table
*table
;
274 table
= kzalloc(sizeof(*table
), GFP_ATOMIC
);
277 rcu_assign_pointer(table
->tb6_root
.leaf
,
278 net
->ipv6
.fib6_null_entry
);
279 table
->tb6_root
.fn_flags
= RTN_ROOT
| RTN_TL_ROOT
| RTN_RTINFO
;
280 inet_peer_base_init(&table
->tb6_peers
);
286 struct fib6_table
*fib6_new_table(struct net
*net
, u32 id
)
288 struct fib6_table
*tb
;
292 tb
= fib6_get_table(net
, id
);
296 tb
= fib6_alloc_table(net
, id
);
298 fib6_link_table(net
, tb
);
302 EXPORT_SYMBOL_GPL(fib6_new_table
);
304 struct fib6_table
*fib6_get_table(struct net
*net
, u32 id
)
306 struct fib6_table
*tb
;
307 struct hlist_head
*head
;
312 h
= id
& (FIB6_TABLE_HASHSZ
- 1);
314 head
= &net
->ipv6
.fib_table_hash
[h
];
315 hlist_for_each_entry_rcu(tb
, head
, tb6_hlist
) {
316 if (tb
->tb6_id
== id
) {
325 EXPORT_SYMBOL_GPL(fib6_get_table
);
327 static void __net_init
fib6_tables_init(struct net
*net
)
329 fib6_link_table(net
, net
->ipv6
.fib6_main_tbl
);
330 fib6_link_table(net
, net
->ipv6
.fib6_local_tbl
);
334 struct fib6_table
*fib6_new_table(struct net
*net
, u32 id
)
336 return fib6_get_table(net
, id
);
339 struct fib6_table
*fib6_get_table(struct net
*net
, u32 id
)
341 return net
->ipv6
.fib6_main_tbl
;
344 struct dst_entry
*fib6_rule_lookup(struct net
*net
, struct flowi6
*fl6
,
345 const struct sk_buff
*skb
,
346 int flags
, pol_lookup_t lookup
)
350 rt
= lookup(net
, net
->ipv6
.fib6_main_tbl
, fl6
, skb
, flags
);
351 if (rt
->dst
.error
== -EAGAIN
) {
353 rt
= net
->ipv6
.ip6_null_entry
;
360 /* called with rcu lock held; no reference taken on fib6_info */
361 struct fib6_info
*fib6_lookup(struct net
*net
, int oif
, struct flowi6
*fl6
,
364 return fib6_table_lookup(net
, net
->ipv6
.fib6_main_tbl
, oif
, fl6
, flags
);
367 static void __net_init
fib6_tables_init(struct net
*net
)
369 fib6_link_table(net
, net
->ipv6
.fib6_main_tbl
);
374 unsigned int fib6_tables_seq_read(struct net
*net
)
376 unsigned int h
, fib_seq
= 0;
379 for (h
= 0; h
< FIB6_TABLE_HASHSZ
; h
++) {
380 struct hlist_head
*head
= &net
->ipv6
.fib_table_hash
[h
];
381 struct fib6_table
*tb
;
383 hlist_for_each_entry_rcu(tb
, head
, tb6_hlist
)
384 fib_seq
+= tb
->fib_seq
;
391 static int call_fib6_entry_notifier(struct notifier_block
*nb
, struct net
*net
,
392 enum fib_event_type event_type
,
393 struct fib6_info
*rt
)
395 struct fib6_entry_notifier_info info
= {
399 return call_fib6_notifier(nb
, net
, event_type
, &info
.info
);
402 static int call_fib6_entry_notifiers(struct net
*net
,
403 enum fib_event_type event_type
,
404 struct fib6_info
*rt
,
405 struct netlink_ext_ack
*extack
)
407 struct fib6_entry_notifier_info info
= {
408 .info
.extack
= extack
,
412 rt
->fib6_table
->fib_seq
++;
413 return call_fib6_notifiers(net
, event_type
, &info
.info
);
416 struct fib6_dump_arg
{
418 struct notifier_block
*nb
;
421 static void fib6_rt_dump(struct fib6_info
*rt
, struct fib6_dump_arg
*arg
)
423 if (rt
== arg
->net
->ipv6
.fib6_null_entry
)
425 call_fib6_entry_notifier(arg
->nb
, arg
->net
, FIB_EVENT_ENTRY_ADD
, rt
);
428 static int fib6_node_dump(struct fib6_walker
*w
)
430 struct fib6_info
*rt
;
432 for_each_fib6_walker_rt(w
)
433 fib6_rt_dump(rt
, w
->args
);
438 static void fib6_table_dump(struct net
*net
, struct fib6_table
*tb
,
439 struct fib6_walker
*w
)
441 w
->root
= &tb
->tb6_root
;
442 spin_lock_bh(&tb
->tb6_lock
);
444 spin_unlock_bh(&tb
->tb6_lock
);
447 /* Called with rcu_read_lock() */
448 int fib6_tables_dump(struct net
*net
, struct notifier_block
*nb
)
450 struct fib6_dump_arg arg
;
451 struct fib6_walker
*w
;
454 w
= kzalloc(sizeof(*w
), GFP_ATOMIC
);
458 w
->func
= fib6_node_dump
;
463 for (h
= 0; h
< FIB6_TABLE_HASHSZ
; h
++) {
464 struct hlist_head
*head
= &net
->ipv6
.fib_table_hash
[h
];
465 struct fib6_table
*tb
;
467 hlist_for_each_entry_rcu(tb
, head
, tb6_hlist
)
468 fib6_table_dump(net
, tb
, w
);
476 static int fib6_dump_node(struct fib6_walker
*w
)
479 struct fib6_info
*rt
;
481 for_each_fib6_walker_rt(w
) {
482 res
= rt6_dump_route(rt
, w
->args
);
484 /* Frame is full, suspend walking */
489 /* Multipath routes are dumped in one route with the
490 * RTA_MULTIPATH attribute. Jump 'rt' to point to the
491 * last sibling of this route (no need to dump the
492 * sibling routes again)
494 if (rt
->fib6_nsiblings
)
495 rt
= list_last_entry(&rt
->fib6_siblings
,
503 static void fib6_dump_end(struct netlink_callback
*cb
)
505 struct net
*net
= sock_net(cb
->skb
->sk
);
506 struct fib6_walker
*w
= (void *)cb
->args
[2];
511 fib6_walker_unlink(net
, w
);
516 cb
->done
= (void *)cb
->args
[3];
520 static int fib6_dump_done(struct netlink_callback
*cb
)
523 return cb
->done
? cb
->done(cb
) : 0;
526 static int fib6_dump_table(struct fib6_table
*table
, struct sk_buff
*skb
,
527 struct netlink_callback
*cb
)
529 struct net
*net
= sock_net(skb
->sk
);
530 struct fib6_walker
*w
;
533 w
= (void *)cb
->args
[2];
534 w
->root
= &table
->tb6_root
;
536 if (cb
->args
[4] == 0) {
540 spin_lock_bh(&table
->tb6_lock
);
541 res
= fib6_walk(net
, w
);
542 spin_unlock_bh(&table
->tb6_lock
);
545 cb
->args
[5] = w
->root
->fn_sernum
;
548 if (cb
->args
[5] != w
->root
->fn_sernum
) {
549 /* Begin at the root if the tree changed */
550 cb
->args
[5] = w
->root
->fn_sernum
;
557 spin_lock_bh(&table
->tb6_lock
);
558 res
= fib6_walk_continue(w
);
559 spin_unlock_bh(&table
->tb6_lock
);
561 fib6_walker_unlink(net
, w
);
569 static int inet6_dump_fib(struct sk_buff
*skb
, struct netlink_callback
*cb
)
571 struct net
*net
= sock_net(skb
->sk
);
573 unsigned int e
= 0, s_e
;
574 struct rt6_rtnl_dump_arg arg
;
575 struct fib6_walker
*w
;
576 struct fib6_table
*tb
;
577 struct hlist_head
*head
;
583 w
= (void *)cb
->args
[2];
587 * 1. hook callback destructor.
589 cb
->args
[3] = (long)cb
->done
;
590 cb
->done
= fib6_dump_done
;
593 * 2. allocate and initialize walker.
595 w
= kzalloc(sizeof(*w
), GFP_ATOMIC
);
598 w
->func
= fib6_dump_node
;
599 cb
->args
[2] = (long)w
;
608 for (h
= s_h
; h
< FIB6_TABLE_HASHSZ
; h
++, s_e
= 0) {
610 head
= &net
->ipv6
.fib_table_hash
[h
];
611 hlist_for_each_entry_rcu(tb
, head
, tb6_hlist
) {
614 res
= fib6_dump_table(tb
, skb
, cb
);
626 res
= res
< 0 ? res
: skb
->len
;
632 void fib6_metric_set(struct fib6_info
*f6i
, int metric
, u32 val
)
637 if (f6i
->fib6_metrics
== &dst_default_metrics
) {
638 struct dst_metrics
*p
= kzalloc(sizeof(*p
), GFP_ATOMIC
);
643 refcount_set(&p
->refcnt
, 1);
644 f6i
->fib6_metrics
= p
;
647 f6i
->fib6_metrics
->metrics
[metric
- 1] = val
;
653 * return the appropriate node for a routing tree "add" operation
654 * by either creating and inserting or by returning an existing
658 static struct fib6_node
*fib6_add_1(struct net
*net
,
659 struct fib6_table
*table
,
660 struct fib6_node
*root
,
661 struct in6_addr
*addr
, int plen
,
662 int offset
, int allow_create
,
663 int replace_required
,
664 struct netlink_ext_ack
*extack
)
666 struct fib6_node
*fn
, *in
, *ln
;
667 struct fib6_node
*pn
= NULL
;
672 RT6_TRACE("fib6_add_1\n");
674 /* insert node in tree */
679 struct fib6_info
*leaf
= rcu_dereference_protected(fn
->leaf
,
680 lockdep_is_held(&table
->tb6_lock
));
681 key
= (struct rt6key
*)((u8
*)leaf
+ offset
);
686 if (plen
< fn
->fn_bit
||
687 !ipv6_prefix_equal(&key
->addr
, addr
, fn
->fn_bit
)) {
689 if (replace_required
) {
690 NL_SET_ERR_MSG(extack
,
691 "Can not replace route - no match found");
692 pr_warn("Can't replace route, no match found\n");
693 return ERR_PTR(-ENOENT
);
695 pr_warn("NLM_F_CREATE should be set when creating new route\n");
704 if (plen
== fn
->fn_bit
) {
705 /* clean up an intermediate node */
706 if (!(fn
->fn_flags
& RTN_RTINFO
)) {
707 RCU_INIT_POINTER(fn
->leaf
, NULL
);
708 fib6_info_release(leaf
);
709 /* remove null_entry in the root node */
710 } else if (fn
->fn_flags
& RTN_TL_ROOT
&&
711 rcu_access_pointer(fn
->leaf
) ==
712 net
->ipv6
.fib6_null_entry
) {
713 RCU_INIT_POINTER(fn
->leaf
, NULL
);
720 * We have more bits to go
723 /* Try to walk down on tree. */
724 dir
= addr_bit_set(addr
, fn
->fn_bit
);
727 rcu_dereference_protected(fn
->right
,
728 lockdep_is_held(&table
->tb6_lock
)) :
729 rcu_dereference_protected(fn
->left
,
730 lockdep_is_held(&table
->tb6_lock
));
734 /* We should not create new node because
735 * NLM_F_REPLACE was specified without NLM_F_CREATE
736 * I assume it is safe to require NLM_F_CREATE when
737 * REPLACE flag is used! Later we may want to remove the
738 * check for replace_required, because according
739 * to netlink specification, NLM_F_CREATE
740 * MUST be specified if new route is created.
741 * That would keep IPv6 consistent with IPv4
743 if (replace_required
) {
744 NL_SET_ERR_MSG(extack
,
745 "Can not replace route - no match found");
746 pr_warn("Can't replace route, no match found\n");
747 return ERR_PTR(-ENOENT
);
749 pr_warn("NLM_F_CREATE should be set when creating new route\n");
752 * We walked to the bottom of tree.
753 * Create new leaf node without children.
756 ln
= node_alloc(net
);
759 return ERR_PTR(-ENOMEM
);
761 RCU_INIT_POINTER(ln
->parent
, pn
);
764 rcu_assign_pointer(pn
->right
, ln
);
766 rcu_assign_pointer(pn
->left
, ln
);
773 * split since we don't have a common prefix anymore or
774 * we have a less significant route.
775 * we've to insert an intermediate node on the list
776 * this new node will point to the one we need to create
780 pn
= rcu_dereference_protected(fn
->parent
,
781 lockdep_is_held(&table
->tb6_lock
));
783 /* find 1st bit in difference between the 2 addrs.
785 See comment in __ipv6_addr_diff: bit may be an invalid value,
786 but if it is >= plen, the value is ignored in any case.
789 bit
= __ipv6_addr_diff(addr
, &key
->addr
, sizeof(*addr
));
794 * (new leaf node)[ln] (old node)[fn]
797 in
= node_alloc(net
);
798 ln
= node_alloc(net
);
802 node_free_immediate(net
, in
);
804 node_free_immediate(net
, ln
);
805 return ERR_PTR(-ENOMEM
);
809 * new intermediate node.
811 * be off since that an address that chooses one of
812 * the branches would not match less specific routes
813 * in the other branch
818 RCU_INIT_POINTER(in
->parent
, pn
);
820 atomic_inc(&rcu_dereference_protected(in
->leaf
,
821 lockdep_is_held(&table
->tb6_lock
))->fib6_ref
);
823 /* update parent pointer */
825 rcu_assign_pointer(pn
->right
, in
);
827 rcu_assign_pointer(pn
->left
, in
);
831 RCU_INIT_POINTER(ln
->parent
, in
);
832 rcu_assign_pointer(fn
->parent
, in
);
834 if (addr_bit_set(addr
, bit
)) {
835 rcu_assign_pointer(in
->right
, ln
);
836 rcu_assign_pointer(in
->left
, fn
);
838 rcu_assign_pointer(in
->left
, ln
);
839 rcu_assign_pointer(in
->right
, fn
);
841 } else { /* plen <= bit */
844 * (new leaf node)[ln]
846 * (old node)[fn] NULL
849 ln
= node_alloc(net
);
852 return ERR_PTR(-ENOMEM
);
856 RCU_INIT_POINTER(ln
->parent
, pn
);
858 if (addr_bit_set(&key
->addr
, plen
))
859 RCU_INIT_POINTER(ln
->right
, fn
);
861 RCU_INIT_POINTER(ln
->left
, fn
);
863 rcu_assign_pointer(fn
->parent
, ln
);
866 rcu_assign_pointer(pn
->right
, ln
);
868 rcu_assign_pointer(pn
->left
, ln
);
873 static void fib6_drop_pcpu_from(struct fib6_info
*f6i
,
874 const struct fib6_table
*table
)
878 /* release the reference to this fib entry from
879 * all of its cached pcpu routes
881 for_each_possible_cpu(cpu
) {
882 struct rt6_info
**ppcpu_rt
;
883 struct rt6_info
*pcpu_rt
;
885 ppcpu_rt
= per_cpu_ptr(f6i
->rt6i_pcpu
, cpu
);
888 struct fib6_info
*from
;
890 from
= rcu_dereference_protected(pcpu_rt
->from
,
891 lockdep_is_held(&table
->tb6_lock
));
892 rcu_assign_pointer(pcpu_rt
->from
, NULL
);
893 fib6_info_release(from
);
898 static void fib6_purge_rt(struct fib6_info
*rt
, struct fib6_node
*fn
,
901 struct fib6_table
*table
= rt
->fib6_table
;
903 if (atomic_read(&rt
->fib6_ref
) != 1) {
904 /* This route is used as dummy address holder in some split
905 * nodes. It is not leaked, but it still holds other resources,
906 * which must be released in time. So, scan ascendant nodes
907 * and replace dummy references to this route with references
908 * to still alive ones.
911 struct fib6_info
*leaf
= rcu_dereference_protected(fn
->leaf
,
912 lockdep_is_held(&table
->tb6_lock
));
913 struct fib6_info
*new_leaf
;
914 if (!(fn
->fn_flags
& RTN_RTINFO
) && leaf
== rt
) {
915 new_leaf
= fib6_find_prefix(net
, table
, fn
);
916 atomic_inc(&new_leaf
->fib6_ref
);
918 rcu_assign_pointer(fn
->leaf
, new_leaf
);
919 fib6_info_release(rt
);
921 fn
= rcu_dereference_protected(fn
->parent
,
922 lockdep_is_held(&table
->tb6_lock
));
926 fib6_drop_pcpu_from(rt
, table
);
931 * Insert routing information in a node.
934 static int fib6_add_rt2node(struct fib6_node
*fn
, struct fib6_info
*rt
,
935 struct nl_info
*info
,
936 struct netlink_ext_ack
*extack
)
938 struct fib6_info
*leaf
= rcu_dereference_protected(fn
->leaf
,
939 lockdep_is_held(&rt
->fib6_table
->tb6_lock
));
940 struct fib6_info
*iter
= NULL
;
941 struct fib6_info __rcu
**ins
;
942 struct fib6_info __rcu
**fallback_ins
= NULL
;
943 int replace
= (info
->nlh
&&
944 (info
->nlh
->nlmsg_flags
& NLM_F_REPLACE
));
945 int add
= (!info
->nlh
||
946 (info
->nlh
->nlmsg_flags
& NLM_F_CREATE
));
948 bool rt_can_ecmp
= rt6_qualify_for_ecmp(rt
);
949 u16 nlflags
= NLM_F_EXCL
;
952 if (info
->nlh
&& (info
->nlh
->nlmsg_flags
& NLM_F_APPEND
))
953 nlflags
|= NLM_F_APPEND
;
957 for (iter
= leaf
; iter
;
958 iter
= rcu_dereference_protected(iter
->fib6_next
,
959 lockdep_is_held(&rt
->fib6_table
->tb6_lock
))) {
961 * Search for duplicates
964 if (iter
->fib6_metric
== rt
->fib6_metric
) {
966 * Same priority level
969 (info
->nlh
->nlmsg_flags
& NLM_F_EXCL
))
972 nlflags
&= ~NLM_F_EXCL
;
974 if (rt_can_ecmp
== rt6_qualify_for_ecmp(iter
)) {
979 fallback_ins
= fallback_ins
?: ins
;
983 if (rt6_duplicate_nexthop(iter
, rt
)) {
984 if (rt
->fib6_nsiblings
)
985 rt
->fib6_nsiblings
= 0;
986 if (!(iter
->fib6_flags
& RTF_EXPIRES
))
988 if (!(rt
->fib6_flags
& RTF_EXPIRES
))
989 fib6_clean_expires(iter
);
991 fib6_set_expires(iter
, rt
->expires
);
994 fib6_metric_set(iter
, RTAX_MTU
,
998 /* If we have the same destination and the same metric,
999 * but not the same gateway, then the route we try to
1000 * add is sibling to this route, increment our counter
1001 * of siblings, and later we will add our route to the
1003 * Only static routes (which don't have flag
1004 * RTF_EXPIRES) are used for ECMPv6.
1006 * To avoid long list, we only had siblings if the
1007 * route have a gateway.
1010 rt6_qualify_for_ecmp(iter
))
1011 rt
->fib6_nsiblings
++;
1014 if (iter
->fib6_metric
> rt
->fib6_metric
)
1018 ins
= &iter
->fib6_next
;
1021 if (fallback_ins
&& !found
) {
1022 /* No ECMP-able route found, replace first non-ECMP one */
1024 iter
= rcu_dereference_protected(*ins
,
1025 lockdep_is_held(&rt
->fib6_table
->tb6_lock
));
1029 /* Reset round-robin state, if necessary */
1030 if (ins
== &fn
->leaf
)
1033 /* Link this route to others same route. */
1034 if (rt
->fib6_nsiblings
) {
1035 unsigned int fib6_nsiblings
;
1036 struct fib6_info
*sibling
, *temp_sibling
;
1038 /* Find the first route that have the same metric */
1041 if (sibling
->fib6_metric
== rt
->fib6_metric
&&
1042 rt6_qualify_for_ecmp(sibling
)) {
1043 list_add_tail(&rt
->fib6_siblings
,
1044 &sibling
->fib6_siblings
);
1047 sibling
= rcu_dereference_protected(sibling
->fib6_next
,
1048 lockdep_is_held(&rt
->fib6_table
->tb6_lock
));
1050 /* For each sibling in the list, increment the counter of
1051 * siblings. BUG() if counters does not match, list of siblings
1055 list_for_each_entry_safe(sibling
, temp_sibling
,
1056 &rt
->fib6_siblings
, fib6_siblings
) {
1057 sibling
->fib6_nsiblings
++;
1058 BUG_ON(sibling
->fib6_nsiblings
!= rt
->fib6_nsiblings
);
1061 BUG_ON(fib6_nsiblings
!= rt
->fib6_nsiblings
);
1062 rt6_multipath_rebalance(temp_sibling
);
1070 pr_warn("NLM_F_CREATE should be set when creating new route\n");
1073 nlflags
|= NLM_F_CREATE
;
1075 err
= call_fib6_entry_notifiers(info
->nl_net
,
1076 FIB_EVENT_ENTRY_ADD
,
1081 rcu_assign_pointer(rt
->fib6_next
, iter
);
1082 atomic_inc(&rt
->fib6_ref
);
1083 rcu_assign_pointer(rt
->fib6_node
, fn
);
1084 rcu_assign_pointer(*ins
, rt
);
1085 if (!info
->skip_notify
)
1086 inet6_rt_notify(RTM_NEWROUTE
, rt
, info
, nlflags
);
1087 info
->nl_net
->ipv6
.rt6_stats
->fib_rt_entries
++;
1089 if (!(fn
->fn_flags
& RTN_RTINFO
)) {
1090 info
->nl_net
->ipv6
.rt6_stats
->fib_route_nodes
++;
1091 fn
->fn_flags
|= RTN_RTINFO
;
1100 pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1104 err
= call_fib6_entry_notifiers(info
->nl_net
,
1105 FIB_EVENT_ENTRY_REPLACE
,
1110 atomic_inc(&rt
->fib6_ref
);
1111 rcu_assign_pointer(rt
->fib6_node
, fn
);
1112 rt
->fib6_next
= iter
->fib6_next
;
1113 rcu_assign_pointer(*ins
, rt
);
1114 if (!info
->skip_notify
)
1115 inet6_rt_notify(RTM_NEWROUTE
, rt
, info
, NLM_F_REPLACE
);
1116 if (!(fn
->fn_flags
& RTN_RTINFO
)) {
1117 info
->nl_net
->ipv6
.rt6_stats
->fib_route_nodes
++;
1118 fn
->fn_flags
|= RTN_RTINFO
;
1120 nsiblings
= iter
->fib6_nsiblings
;
1121 iter
->fib6_node
= NULL
;
1122 fib6_purge_rt(iter
, fn
, info
->nl_net
);
1123 if (rcu_access_pointer(fn
->rr_ptr
) == iter
)
1125 fib6_info_release(iter
);
1128 /* Replacing an ECMP route, remove all siblings */
1129 ins
= &rt
->fib6_next
;
1130 iter
= rcu_dereference_protected(*ins
,
1131 lockdep_is_held(&rt
->fib6_table
->tb6_lock
));
1133 if (iter
->fib6_metric
> rt
->fib6_metric
)
1135 if (rt6_qualify_for_ecmp(iter
)) {
1136 *ins
= iter
->fib6_next
;
1137 iter
->fib6_node
= NULL
;
1138 fib6_purge_rt(iter
, fn
, info
->nl_net
);
1139 if (rcu_access_pointer(fn
->rr_ptr
) == iter
)
1141 fib6_info_release(iter
);
1143 info
->nl_net
->ipv6
.rt6_stats
->fib_rt_entries
--;
1145 ins
= &iter
->fib6_next
;
1147 iter
= rcu_dereference_protected(*ins
,
1148 lockdep_is_held(&rt
->fib6_table
->tb6_lock
));
1150 WARN_ON(nsiblings
!= 0);
1157 static void fib6_start_gc(struct net
*net
, struct fib6_info
*rt
)
1159 if (!timer_pending(&net
->ipv6
.ip6_fib_timer
) &&
1160 (rt
->fib6_flags
& RTF_EXPIRES
))
1161 mod_timer(&net
->ipv6
.ip6_fib_timer
,
1162 jiffies
+ net
->ipv6
.sysctl
.ip6_rt_gc_interval
);
1165 void fib6_force_start_gc(struct net
*net
)
1167 if (!timer_pending(&net
->ipv6
.ip6_fib_timer
))
1168 mod_timer(&net
->ipv6
.ip6_fib_timer
,
1169 jiffies
+ net
->ipv6
.sysctl
.ip6_rt_gc_interval
);
1172 static void __fib6_update_sernum_upto_root(struct fib6_info
*rt
,
1175 struct fib6_node
*fn
= rcu_dereference_protected(rt
->fib6_node
,
1176 lockdep_is_held(&rt
->fib6_table
->tb6_lock
));
1178 /* paired with smp_rmb() in rt6_get_cookie_safe() */
1181 fn
->fn_sernum
= sernum
;
1182 fn
= rcu_dereference_protected(fn
->parent
,
1183 lockdep_is_held(&rt
->fib6_table
->tb6_lock
));
1187 void fib6_update_sernum_upto_root(struct net
*net
, struct fib6_info
*rt
)
1189 __fib6_update_sernum_upto_root(rt
, fib6_new_sernum(net
));
1193 * Add routing information to the routing tree.
1194 * <destination addr>/<source addr>
1195 * with source addr info in sub-trees
1196 * Need to own table->tb6_lock
1199 int fib6_add(struct fib6_node
*root
, struct fib6_info
*rt
,
1200 struct nl_info
*info
, struct netlink_ext_ack
*extack
)
1202 struct fib6_table
*table
= rt
->fib6_table
;
1203 struct fib6_node
*fn
, *pn
= NULL
;
1205 int allow_create
= 1;
1206 int replace_required
= 0;
1207 int sernum
= fib6_new_sernum(info
->nl_net
);
1210 if (!(info
->nlh
->nlmsg_flags
& NLM_F_CREATE
))
1212 if (info
->nlh
->nlmsg_flags
& NLM_F_REPLACE
)
1213 replace_required
= 1;
1215 if (!allow_create
&& !replace_required
)
1216 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1218 fn
= fib6_add_1(info
->nl_net
, table
, root
,
1219 &rt
->fib6_dst
.addr
, rt
->fib6_dst
.plen
,
1220 offsetof(struct fib6_info
, fib6_dst
), allow_create
,
1221 replace_required
, extack
);
1230 #ifdef CONFIG_IPV6_SUBTREES
1231 if (rt
->fib6_src
.plen
) {
1232 struct fib6_node
*sn
;
1234 if (!rcu_access_pointer(fn
->subtree
)) {
1235 struct fib6_node
*sfn
;
1247 /* Create subtree root node */
1248 sfn
= node_alloc(info
->nl_net
);
1252 atomic_inc(&info
->nl_net
->ipv6
.fib6_null_entry
->fib6_ref
);
1253 rcu_assign_pointer(sfn
->leaf
,
1254 info
->nl_net
->ipv6
.fib6_null_entry
);
1255 sfn
->fn_flags
= RTN_ROOT
;
1257 /* Now add the first leaf node to new subtree */
1259 sn
= fib6_add_1(info
->nl_net
, table
, sfn
,
1260 &rt
->fib6_src
.addr
, rt
->fib6_src
.plen
,
1261 offsetof(struct fib6_info
, fib6_src
),
1262 allow_create
, replace_required
, extack
);
1265 /* If it is failed, discard just allocated
1266 root, and then (in failure) stale node
1269 node_free_immediate(info
->nl_net
, sfn
);
1274 /* Now link new subtree to main tree */
1275 rcu_assign_pointer(sfn
->parent
, fn
);
1276 rcu_assign_pointer(fn
->subtree
, sfn
);
1278 sn
= fib6_add_1(info
->nl_net
, table
, FIB6_SUBTREE(fn
),
1279 &rt
->fib6_src
.addr
, rt
->fib6_src
.plen
,
1280 offsetof(struct fib6_info
, fib6_src
),
1281 allow_create
, replace_required
, extack
);
1289 if (!rcu_access_pointer(fn
->leaf
)) {
1290 if (fn
->fn_flags
& RTN_TL_ROOT
) {
1291 /* put back null_entry for root node */
1292 rcu_assign_pointer(fn
->leaf
,
1293 info
->nl_net
->ipv6
.fib6_null_entry
);
1295 atomic_inc(&rt
->fib6_ref
);
1296 rcu_assign_pointer(fn
->leaf
, rt
);
1303 err
= fib6_add_rt2node(fn
, rt
, info
, extack
);
1305 __fib6_update_sernum_upto_root(rt
, sernum
);
1306 fib6_start_gc(info
->nl_net
, rt
);
1311 #ifdef CONFIG_IPV6_SUBTREES
1313 * If fib6_add_1 has cleared the old leaf pointer in the
1314 * super-tree leaf node we have to find a new one for it.
1317 struct fib6_info
*pn_leaf
=
1318 rcu_dereference_protected(pn
->leaf
,
1319 lockdep_is_held(&table
->tb6_lock
));
1320 if (pn_leaf
== rt
) {
1322 RCU_INIT_POINTER(pn
->leaf
, NULL
);
1323 fib6_info_release(rt
);
1325 if (!pn_leaf
&& !(pn
->fn_flags
& RTN_RTINFO
)) {
1326 pn_leaf
= fib6_find_prefix(info
->nl_net
, table
,
1332 info
->nl_net
->ipv6
.fib6_null_entry
;
1335 fib6_info_hold(pn_leaf
);
1336 rcu_assign_pointer(pn
->leaf
, pn_leaf
);
1345 /* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1346 * 1. fn is an intermediate node and we failed to add the new
1347 * route to it in both subtree creation failure and fib6_add_rt2node()
1349 * 2. fn is the root node in the table and we fail to add the first
1350 * default route to it.
1353 (!(fn
->fn_flags
& (RTN_RTINFO
|RTN_ROOT
)) ||
1354 (fn
->fn_flags
& RTN_TL_ROOT
&&
1355 !rcu_access_pointer(fn
->leaf
))))
1356 fib6_repair_tree(info
->nl_net
, table
, fn
);
1361 * Routing tree lookup
1365 struct lookup_args
{
1366 int offset
; /* key offset on fib6_info */
1367 const struct in6_addr
*addr
; /* search key */
1370 static struct fib6_node
*fib6_node_lookup_1(struct fib6_node
*root
,
1371 struct lookup_args
*args
)
1373 struct fib6_node
*fn
;
1376 if (unlikely(args
->offset
== 0))
1386 struct fib6_node
*next
;
1388 dir
= addr_bit_set(args
->addr
, fn
->fn_bit
);
1390 next
= dir
? rcu_dereference(fn
->right
) :
1391 rcu_dereference(fn
->left
);
1401 struct fib6_node
*subtree
= FIB6_SUBTREE(fn
);
1403 if (subtree
|| fn
->fn_flags
& RTN_RTINFO
) {
1404 struct fib6_info
*leaf
= rcu_dereference(fn
->leaf
);
1410 key
= (struct rt6key
*) ((u8
*)leaf
+ args
->offset
);
1412 if (ipv6_prefix_equal(&key
->addr
, args
->addr
, key
->plen
)) {
1413 #ifdef CONFIG_IPV6_SUBTREES
1415 struct fib6_node
*sfn
;
1416 sfn
= fib6_node_lookup_1(subtree
,
1423 if (fn
->fn_flags
& RTN_RTINFO
)
1428 if (fn
->fn_flags
& RTN_ROOT
)
1431 fn
= rcu_dereference(fn
->parent
);
1437 /* called with rcu_read_lock() held
1439 struct fib6_node
*fib6_node_lookup(struct fib6_node
*root
,
1440 const struct in6_addr
*daddr
,
1441 const struct in6_addr
*saddr
)
1443 struct fib6_node
*fn
;
1444 struct lookup_args args
[] = {
1446 .offset
= offsetof(struct fib6_info
, fib6_dst
),
1449 #ifdef CONFIG_IPV6_SUBTREES
1451 .offset
= offsetof(struct fib6_info
, fib6_src
),
1456 .offset
= 0, /* sentinel */
1460 fn
= fib6_node_lookup_1(root
, daddr
? args
: args
+ 1);
1461 if (!fn
|| fn
->fn_flags
& RTN_TL_ROOT
)
1468 * Get node with specified destination prefix (and source prefix,
1469 * if subtrees are used)
1470 * exact_match == true means we try to find fn with exact match of
1471 * the passed in prefix addr
1472 * exact_match == false means we try to find fn with longest prefix
1473 * match of the passed in prefix addr. This is useful for finding fn
1474 * for cached route as it will be stored in the exception table under
1475 * the node with longest prefix length.
1479 static struct fib6_node
*fib6_locate_1(struct fib6_node
*root
,
1480 const struct in6_addr
*addr
,
1481 int plen
, int offset
,
1484 struct fib6_node
*fn
, *prev
= NULL
;
1486 for (fn
= root
; fn
; ) {
1487 struct fib6_info
*leaf
= rcu_dereference(fn
->leaf
);
1490 /* This node is being deleted */
1492 if (plen
<= fn
->fn_bit
)
1498 key
= (struct rt6key
*)((u8
*)leaf
+ offset
);
1503 if (plen
< fn
->fn_bit
||
1504 !ipv6_prefix_equal(&key
->addr
, addr
, fn
->fn_bit
))
1507 if (plen
== fn
->fn_bit
)
1514 * We have more bits to go
1516 if (addr_bit_set(addr
, fn
->fn_bit
))
1517 fn
= rcu_dereference(fn
->right
);
1519 fn
= rcu_dereference(fn
->left
);
1528 struct fib6_node
*fib6_locate(struct fib6_node
*root
,
1529 const struct in6_addr
*daddr
, int dst_len
,
1530 const struct in6_addr
*saddr
, int src_len
,
1533 struct fib6_node
*fn
;
1535 fn
= fib6_locate_1(root
, daddr
, dst_len
,
1536 offsetof(struct fib6_info
, fib6_dst
),
1539 #ifdef CONFIG_IPV6_SUBTREES
1541 WARN_ON(saddr
== NULL
);
1543 struct fib6_node
*subtree
= FIB6_SUBTREE(fn
);
1546 fn
= fib6_locate_1(subtree
, saddr
, src_len
,
1547 offsetof(struct fib6_info
, fib6_src
),
1554 if (fn
&& fn
->fn_flags
& RTN_RTINFO
)
1566 static struct fib6_info
*fib6_find_prefix(struct net
*net
,
1567 struct fib6_table
*table
,
1568 struct fib6_node
*fn
)
1570 struct fib6_node
*child_left
, *child_right
;
1572 if (fn
->fn_flags
& RTN_ROOT
)
1573 return net
->ipv6
.fib6_null_entry
;
1576 child_left
= rcu_dereference_protected(fn
->left
,
1577 lockdep_is_held(&table
->tb6_lock
));
1578 child_right
= rcu_dereference_protected(fn
->right
,
1579 lockdep_is_held(&table
->tb6_lock
));
1581 return rcu_dereference_protected(child_left
->leaf
,
1582 lockdep_is_held(&table
->tb6_lock
));
1584 return rcu_dereference_protected(child_right
->leaf
,
1585 lockdep_is_held(&table
->tb6_lock
));
1587 fn
= FIB6_SUBTREE(fn
);
1593 * Called to trim the tree of intermediate nodes when possible. "fn"
1594 * is the node we want to try and remove.
1595 * Need to own table->tb6_lock
1598 static struct fib6_node
*fib6_repair_tree(struct net
*net
,
1599 struct fib6_table
*table
,
1600 struct fib6_node
*fn
)
1604 struct fib6_node
*child
;
1605 struct fib6_walker
*w
;
1608 /* Set fn->leaf to null_entry for root node. */
1609 if (fn
->fn_flags
& RTN_TL_ROOT
) {
1610 rcu_assign_pointer(fn
->leaf
, net
->ipv6
.fib6_null_entry
);
1615 struct fib6_node
*fn_r
= rcu_dereference_protected(fn
->right
,
1616 lockdep_is_held(&table
->tb6_lock
));
1617 struct fib6_node
*fn_l
= rcu_dereference_protected(fn
->left
,
1618 lockdep_is_held(&table
->tb6_lock
));
1619 struct fib6_node
*pn
= rcu_dereference_protected(fn
->parent
,
1620 lockdep_is_held(&table
->tb6_lock
));
1621 struct fib6_node
*pn_r
= rcu_dereference_protected(pn
->right
,
1622 lockdep_is_held(&table
->tb6_lock
));
1623 struct fib6_node
*pn_l
= rcu_dereference_protected(pn
->left
,
1624 lockdep_is_held(&table
->tb6_lock
));
1625 struct fib6_info
*fn_leaf
= rcu_dereference_protected(fn
->leaf
,
1626 lockdep_is_held(&table
->tb6_lock
));
1627 struct fib6_info
*pn_leaf
= rcu_dereference_protected(pn
->leaf
,
1628 lockdep_is_held(&table
->tb6_lock
));
1629 struct fib6_info
*new_fn_leaf
;
1631 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn
->fn_bit
, iter
);
1634 WARN_ON(fn
->fn_flags
& RTN_RTINFO
);
1635 WARN_ON(fn
->fn_flags
& RTN_TL_ROOT
);
1641 child
= fn_r
, children
|= 1;
1643 child
= fn_l
, children
|= 2;
1645 if (children
== 3 || FIB6_SUBTREE(fn
)
1646 #ifdef CONFIG_IPV6_SUBTREES
1647 /* Subtree root (i.e. fn) may have one child */
1648 || (children
&& fn
->fn_flags
& RTN_ROOT
)
1651 new_fn_leaf
= fib6_find_prefix(net
, table
, fn
);
1654 WARN_ON(!new_fn_leaf
);
1655 new_fn_leaf
= net
->ipv6
.fib6_null_entry
;
1658 fib6_info_hold(new_fn_leaf
);
1659 rcu_assign_pointer(fn
->leaf
, new_fn_leaf
);
1663 #ifdef CONFIG_IPV6_SUBTREES
1664 if (FIB6_SUBTREE(pn
) == fn
) {
1665 WARN_ON(!(fn
->fn_flags
& RTN_ROOT
));
1666 RCU_INIT_POINTER(pn
->subtree
, NULL
);
1669 WARN_ON(fn
->fn_flags
& RTN_ROOT
);
1672 rcu_assign_pointer(pn
->right
, child
);
1673 else if (pn_l
== fn
)
1674 rcu_assign_pointer(pn
->left
, child
);
1680 rcu_assign_pointer(child
->parent
, pn
);
1682 #ifdef CONFIG_IPV6_SUBTREES
1686 read_lock(&net
->ipv6
.fib6_walker_lock
);
1687 FOR_WALKERS(net
, w
) {
1689 if (w
->node
== fn
) {
1690 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w
, w
->state
, nstate
);
1695 if (w
->node
== fn
) {
1698 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w
, w
->state
);
1699 w
->state
= w
->state
>= FWS_R
? FWS_U
: FWS_INIT
;
1701 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w
, w
->state
);
1702 w
->state
= w
->state
>= FWS_C
? FWS_U
: FWS_INIT
;
1707 read_unlock(&net
->ipv6
.fib6_walker_lock
);
1710 if (pn
->fn_flags
& RTN_RTINFO
|| FIB6_SUBTREE(pn
))
1713 RCU_INIT_POINTER(pn
->leaf
, NULL
);
1714 fib6_info_release(pn_leaf
);
1719 static void fib6_del_route(struct fib6_table
*table
, struct fib6_node
*fn
,
1720 struct fib6_info __rcu
**rtp
, struct nl_info
*info
)
1722 struct fib6_walker
*w
;
1723 struct fib6_info
*rt
= rcu_dereference_protected(*rtp
,
1724 lockdep_is_held(&table
->tb6_lock
));
1725 struct net
*net
= info
->nl_net
;
1727 RT6_TRACE("fib6_del_route\n");
1730 *rtp
= rt
->fib6_next
;
1731 rt
->fib6_node
= NULL
;
1732 net
->ipv6
.rt6_stats
->fib_rt_entries
--;
1733 net
->ipv6
.rt6_stats
->fib_discarded_routes
++;
1735 /* Flush all cached dst in exception table */
1736 rt6_flush_exceptions(rt
);
1738 /* Reset round-robin state, if necessary */
1739 if (rcu_access_pointer(fn
->rr_ptr
) == rt
)
1742 /* Remove this entry from other siblings */
1743 if (rt
->fib6_nsiblings
) {
1744 struct fib6_info
*sibling
, *next_sibling
;
1746 list_for_each_entry_safe(sibling
, next_sibling
,
1747 &rt
->fib6_siblings
, fib6_siblings
)
1748 sibling
->fib6_nsiblings
--;
1749 rt
->fib6_nsiblings
= 0;
1750 list_del_init(&rt
->fib6_siblings
);
1751 rt6_multipath_rebalance(next_sibling
);
1754 /* Adjust walkers */
1755 read_lock(&net
->ipv6
.fib6_walker_lock
);
1756 FOR_WALKERS(net
, w
) {
1757 if (w
->state
== FWS_C
&& w
->leaf
== rt
) {
1758 RT6_TRACE("walker %p adjusted by delroute\n", w
);
1759 w
->leaf
= rcu_dereference_protected(rt
->fib6_next
,
1760 lockdep_is_held(&table
->tb6_lock
));
1765 read_unlock(&net
->ipv6
.fib6_walker_lock
);
1767 /* If it was last route, call fib6_repair_tree() to:
1768 * 1. For root node, put back null_entry as how the table was created.
1769 * 2. For other nodes, expunge its radix tree node.
1771 if (!rcu_access_pointer(fn
->leaf
)) {
1772 if (!(fn
->fn_flags
& RTN_TL_ROOT
)) {
1773 fn
->fn_flags
&= ~RTN_RTINFO
;
1774 net
->ipv6
.rt6_stats
->fib_route_nodes
--;
1776 fn
= fib6_repair_tree(net
, table
, fn
);
1779 fib6_purge_rt(rt
, fn
, net
);
1781 call_fib6_entry_notifiers(net
, FIB_EVENT_ENTRY_DEL
, rt
, NULL
);
1782 if (!info
->skip_notify
)
1783 inet6_rt_notify(RTM_DELROUTE
, rt
, info
, 0);
1784 fib6_info_release(rt
);
1787 /* Need to own table->tb6_lock */
1788 int fib6_del(struct fib6_info
*rt
, struct nl_info
*info
)
1790 struct fib6_node
*fn
= rcu_dereference_protected(rt
->fib6_node
,
1791 lockdep_is_held(&rt
->fib6_table
->tb6_lock
));
1792 struct fib6_table
*table
= rt
->fib6_table
;
1793 struct net
*net
= info
->nl_net
;
1794 struct fib6_info __rcu
**rtp
;
1795 struct fib6_info __rcu
**rtp_next
;
1797 if (!fn
|| rt
== net
->ipv6
.fib6_null_entry
)
1800 WARN_ON(!(fn
->fn_flags
& RTN_RTINFO
));
1803 * Walk the leaf entries looking for ourself
1806 for (rtp
= &fn
->leaf
; *rtp
; rtp
= rtp_next
) {
1807 struct fib6_info
*cur
= rcu_dereference_protected(*rtp
,
1808 lockdep_is_held(&table
->tb6_lock
));
1810 fib6_del_route(table
, fn
, rtp
, info
);
1813 rtp_next
= &cur
->fib6_next
;
1819 * Tree traversal function.
1821 * Certainly, it is not interrupt safe.
1822 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1823 * It means, that we can modify tree during walking
1824 * and use this function for garbage collection, clone pruning,
1825 * cleaning tree when a device goes down etc. etc.
1827 * It guarantees that every node will be traversed,
1828 * and that it will be traversed only once.
1830 * Callback function w->func may return:
1831 * 0 -> continue walking.
1832 * positive value -> walking is suspended (used by tree dumps,
1833 * and probably by gc, if it will be split to several slices)
1834 * negative value -> terminate walking.
1836 * The function itself returns:
1837 * 0 -> walk is complete.
1838 * >0 -> walk is incomplete (i.e. suspended)
1839 * <0 -> walk is terminated by an error.
1841 * This function is called with tb6_lock held.
1844 static int fib6_walk_continue(struct fib6_walker
*w
)
1846 struct fib6_node
*fn
, *pn
, *left
, *right
;
1848 /* w->root should always be table->tb6_root */
1849 WARN_ON_ONCE(!(w
->root
->fn_flags
& RTN_TL_ROOT
));
1857 #ifdef CONFIG_IPV6_SUBTREES
1859 if (FIB6_SUBTREE(fn
)) {
1860 w
->node
= FIB6_SUBTREE(fn
);
1867 left
= rcu_dereference_protected(fn
->left
, 1);
1870 w
->state
= FWS_INIT
;
1876 right
= rcu_dereference_protected(fn
->right
, 1);
1879 w
->state
= FWS_INIT
;
1883 w
->leaf
= rcu_dereference_protected(fn
->leaf
, 1);
1886 if (w
->leaf
&& fn
->fn_flags
& RTN_RTINFO
) {
1907 pn
= rcu_dereference_protected(fn
->parent
, 1);
1908 left
= rcu_dereference_protected(pn
->left
, 1);
1909 right
= rcu_dereference_protected(pn
->right
, 1);
1911 #ifdef CONFIG_IPV6_SUBTREES
1912 if (FIB6_SUBTREE(pn
) == fn
) {
1913 WARN_ON(!(fn
->fn_flags
& RTN_ROOT
));
1924 w
->leaf
= rcu_dereference_protected(w
->node
->leaf
, 1);
1934 static int fib6_walk(struct net
*net
, struct fib6_walker
*w
)
1938 w
->state
= FWS_INIT
;
1941 fib6_walker_link(net
, w
);
1942 res
= fib6_walk_continue(w
);
1944 fib6_walker_unlink(net
, w
);
1948 static int fib6_clean_node(struct fib6_walker
*w
)
1951 struct fib6_info
*rt
;
1952 struct fib6_cleaner
*c
= container_of(w
, struct fib6_cleaner
, w
);
1953 struct nl_info info
= {
1957 if (c
->sernum
!= FIB6_NO_SERNUM_CHANGE
&&
1958 w
->node
->fn_sernum
!= c
->sernum
)
1959 w
->node
->fn_sernum
= c
->sernum
;
1962 WARN_ON_ONCE(c
->sernum
== FIB6_NO_SERNUM_CHANGE
);
1967 for_each_fib6_walker_rt(w
) {
1968 res
= c
->func(rt
, c
->arg
);
1971 res
= fib6_del(rt
, &info
);
1974 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1976 rcu_access_pointer(rt
->fib6_node
),
1982 } else if (res
== -2) {
1983 if (WARN_ON(!rt
->fib6_nsiblings
))
1985 rt
= list_last_entry(&rt
->fib6_siblings
,
1986 struct fib6_info
, fib6_siblings
);
1996 * Convenient frontend to tree walker.
1998 * func is called on each route.
1999 * It may return -2 -> skip multipath route.
2000 * -1 -> delete this route.
2001 * 0 -> continue walking
2004 static void fib6_clean_tree(struct net
*net
, struct fib6_node
*root
,
2005 int (*func
)(struct fib6_info
*, void *arg
),
2006 int sernum
, void *arg
)
2008 struct fib6_cleaner c
;
2011 c
.w
.func
= fib6_clean_node
;
2019 fib6_walk(net
, &c
.w
);
2022 static void __fib6_clean_all(struct net
*net
,
2023 int (*func
)(struct fib6_info
*, void *),
2024 int sernum
, void *arg
)
2026 struct fib6_table
*table
;
2027 struct hlist_head
*head
;
2031 for (h
= 0; h
< FIB6_TABLE_HASHSZ
; h
++) {
2032 head
= &net
->ipv6
.fib_table_hash
[h
];
2033 hlist_for_each_entry_rcu(table
, head
, tb6_hlist
) {
2034 spin_lock_bh(&table
->tb6_lock
);
2035 fib6_clean_tree(net
, &table
->tb6_root
,
2037 spin_unlock_bh(&table
->tb6_lock
);
2043 void fib6_clean_all(struct net
*net
, int (*func
)(struct fib6_info
*, void *),
2046 __fib6_clean_all(net
, func
, FIB6_NO_SERNUM_CHANGE
, arg
);
2049 static void fib6_flush_trees(struct net
*net
)
2051 int new_sernum
= fib6_new_sernum(net
);
2053 __fib6_clean_all(net
, NULL
, new_sernum
, NULL
);
2057 * Garbage collection
2060 static int fib6_age(struct fib6_info
*rt
, void *arg
)
2062 struct fib6_gc_args
*gc_args
= arg
;
2063 unsigned long now
= jiffies
;
2066 * check addrconf expiration here.
2067 * Routes are expired even if they are in use.
2070 if (rt
->fib6_flags
& RTF_EXPIRES
&& rt
->expires
) {
2071 if (time_after(now
, rt
->expires
)) {
2072 RT6_TRACE("expiring %p\n", rt
);
2078 /* Also age clones in the exception table.
2079 * Note, that clones are aged out
2080 * only if they are not in use now.
2082 rt6_age_exceptions(rt
, gc_args
, now
);
2087 void fib6_run_gc(unsigned long expires
, struct net
*net
, bool force
)
2089 struct fib6_gc_args gc_args
;
2093 spin_lock_bh(&net
->ipv6
.fib6_gc_lock
);
2094 } else if (!spin_trylock_bh(&net
->ipv6
.fib6_gc_lock
)) {
2095 mod_timer(&net
->ipv6
.ip6_fib_timer
, jiffies
+ HZ
);
2098 gc_args
.timeout
= expires
? (int)expires
:
2099 net
->ipv6
.sysctl
.ip6_rt_gc_interval
;
2102 fib6_clean_all(net
, fib6_age
, &gc_args
);
2104 net
->ipv6
.ip6_rt_last_gc
= now
;
2107 mod_timer(&net
->ipv6
.ip6_fib_timer
,
2109 + net
->ipv6
.sysctl
.ip6_rt_gc_interval
));
2111 del_timer(&net
->ipv6
.ip6_fib_timer
);
2112 spin_unlock_bh(&net
->ipv6
.fib6_gc_lock
);
2115 static void fib6_gc_timer_cb(struct timer_list
*t
)
2117 struct net
*arg
= from_timer(arg
, t
, ipv6
.ip6_fib_timer
);
2119 fib6_run_gc(0, arg
, true);
2122 static int __net_init
fib6_net_init(struct net
*net
)
2124 size_t size
= sizeof(struct hlist_head
) * FIB6_TABLE_HASHSZ
;
2127 err
= fib6_notifier_init(net
);
2131 spin_lock_init(&net
->ipv6
.fib6_gc_lock
);
2132 rwlock_init(&net
->ipv6
.fib6_walker_lock
);
2133 INIT_LIST_HEAD(&net
->ipv6
.fib6_walkers
);
2134 timer_setup(&net
->ipv6
.ip6_fib_timer
, fib6_gc_timer_cb
, 0);
2136 net
->ipv6
.rt6_stats
= kzalloc(sizeof(*net
->ipv6
.rt6_stats
), GFP_KERNEL
);
2137 if (!net
->ipv6
.rt6_stats
)
2140 /* Avoid false sharing : Use at least a full cache line */
2141 size
= max_t(size_t, size
, L1_CACHE_BYTES
);
2143 net
->ipv6
.fib_table_hash
= kzalloc(size
, GFP_KERNEL
);
2144 if (!net
->ipv6
.fib_table_hash
)
2147 net
->ipv6
.fib6_main_tbl
= kzalloc(sizeof(*net
->ipv6
.fib6_main_tbl
),
2149 if (!net
->ipv6
.fib6_main_tbl
)
2150 goto out_fib_table_hash
;
2152 net
->ipv6
.fib6_main_tbl
->tb6_id
= RT6_TABLE_MAIN
;
2153 rcu_assign_pointer(net
->ipv6
.fib6_main_tbl
->tb6_root
.leaf
,
2154 net
->ipv6
.fib6_null_entry
);
2155 net
->ipv6
.fib6_main_tbl
->tb6_root
.fn_flags
=
2156 RTN_ROOT
| RTN_TL_ROOT
| RTN_RTINFO
;
2157 inet_peer_base_init(&net
->ipv6
.fib6_main_tbl
->tb6_peers
);
2159 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2160 net
->ipv6
.fib6_local_tbl
= kzalloc(sizeof(*net
->ipv6
.fib6_local_tbl
),
2162 if (!net
->ipv6
.fib6_local_tbl
)
2163 goto out_fib6_main_tbl
;
2164 net
->ipv6
.fib6_local_tbl
->tb6_id
= RT6_TABLE_LOCAL
;
2165 rcu_assign_pointer(net
->ipv6
.fib6_local_tbl
->tb6_root
.leaf
,
2166 net
->ipv6
.fib6_null_entry
);
2167 net
->ipv6
.fib6_local_tbl
->tb6_root
.fn_flags
=
2168 RTN_ROOT
| RTN_TL_ROOT
| RTN_RTINFO
;
2169 inet_peer_base_init(&net
->ipv6
.fib6_local_tbl
->tb6_peers
);
2171 fib6_tables_init(net
);
2175 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2177 kfree(net
->ipv6
.fib6_main_tbl
);
2180 kfree(net
->ipv6
.fib_table_hash
);
2182 kfree(net
->ipv6
.rt6_stats
);
2184 fib6_notifier_exit(net
);
2188 static void fib6_net_exit(struct net
*net
)
2192 del_timer_sync(&net
->ipv6
.ip6_fib_timer
);
2194 for (i
= 0; i
< FIB6_TABLE_HASHSZ
; i
++) {
2195 struct hlist_head
*head
= &net
->ipv6
.fib_table_hash
[i
];
2196 struct hlist_node
*tmp
;
2197 struct fib6_table
*tb
;
2199 hlist_for_each_entry_safe(tb
, tmp
, head
, tb6_hlist
) {
2200 hlist_del(&tb
->tb6_hlist
);
2201 fib6_free_table(tb
);
2205 kfree(net
->ipv6
.fib_table_hash
);
2206 kfree(net
->ipv6
.rt6_stats
);
2207 fib6_notifier_exit(net
);
2210 static struct pernet_operations fib6_net_ops
= {
2211 .init
= fib6_net_init
,
2212 .exit
= fib6_net_exit
,
2215 int __init
fib6_init(void)
2219 fib6_node_kmem
= kmem_cache_create("fib6_nodes",
2220 sizeof(struct fib6_node
),
2221 0, SLAB_HWCACHE_ALIGN
,
2223 if (!fib6_node_kmem
)
2226 ret
= register_pernet_subsys(&fib6_net_ops
);
2228 goto out_kmem_cache_create
;
2230 ret
= rtnl_register_module(THIS_MODULE
, PF_INET6
, RTM_GETROUTE
, NULL
,
2233 goto out_unregister_subsys
;
2235 __fib6_flush_trees
= fib6_flush_trees
;
2239 out_unregister_subsys
:
2240 unregister_pernet_subsys(&fib6_net_ops
);
2241 out_kmem_cache_create
:
2242 kmem_cache_destroy(fib6_node_kmem
);
2246 void fib6_gc_cleanup(void)
2248 unregister_pernet_subsys(&fib6_net_ops
);
2249 kmem_cache_destroy(fib6_node_kmem
);
2252 #ifdef CONFIG_PROC_FS
2253 static int ipv6_route_seq_show(struct seq_file
*seq
, void *v
)
2255 struct fib6_info
*rt
= v
;
2256 struct ipv6_route_iter
*iter
= seq
->private;
2257 const struct net_device
*dev
;
2259 seq_printf(seq
, "%pi6 %02x ", &rt
->fib6_dst
.addr
, rt
->fib6_dst
.plen
);
2261 #ifdef CONFIG_IPV6_SUBTREES
2262 seq_printf(seq
, "%pi6 %02x ", &rt
->fib6_src
.addr
, rt
->fib6_src
.plen
);
2264 seq_puts(seq
, "00000000000000000000000000000000 00 ");
2266 if (rt
->fib6_flags
& RTF_GATEWAY
)
2267 seq_printf(seq
, "%pi6", &rt
->fib6_nh
.nh_gw
);
2269 seq_puts(seq
, "00000000000000000000000000000000");
2271 dev
= rt
->fib6_nh
.nh_dev
;
2272 seq_printf(seq
, " %08x %08x %08x %08x %8s\n",
2273 rt
->fib6_metric
, atomic_read(&rt
->fib6_ref
), 0,
2274 rt
->fib6_flags
, dev
? dev
->name
: "");
2275 iter
->w
.leaf
= NULL
;
2279 static int ipv6_route_yield(struct fib6_walker
*w
)
2281 struct ipv6_route_iter
*iter
= w
->args
;
2287 iter
->w
.leaf
= rcu_dereference_protected(
2288 iter
->w
.leaf
->fib6_next
,
2289 lockdep_is_held(&iter
->tbl
->tb6_lock
));
2291 if (!iter
->skip
&& iter
->w
.leaf
)
2293 } while (iter
->w
.leaf
);
2298 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter
*iter
,
2301 memset(&iter
->w
, 0, sizeof(iter
->w
));
2302 iter
->w
.func
= ipv6_route_yield
;
2303 iter
->w
.root
= &iter
->tbl
->tb6_root
;
2304 iter
->w
.state
= FWS_INIT
;
2305 iter
->w
.node
= iter
->w
.root
;
2306 iter
->w
.args
= iter
;
2307 iter
->sernum
= iter
->w
.root
->fn_sernum
;
2308 INIT_LIST_HEAD(&iter
->w
.lh
);
2309 fib6_walker_link(net
, &iter
->w
);
2312 static struct fib6_table
*ipv6_route_seq_next_table(struct fib6_table
*tbl
,
2316 struct hlist_node
*node
;
2319 h
= (tbl
->tb6_id
& (FIB6_TABLE_HASHSZ
- 1)) + 1;
2320 node
= rcu_dereference_bh(hlist_next_rcu(&tbl
->tb6_hlist
));
2326 while (!node
&& h
< FIB6_TABLE_HASHSZ
) {
2327 node
= rcu_dereference_bh(
2328 hlist_first_rcu(&net
->ipv6
.fib_table_hash
[h
++]));
2330 return hlist_entry_safe(node
, struct fib6_table
, tb6_hlist
);
2333 static void ipv6_route_check_sernum(struct ipv6_route_iter
*iter
)
2335 if (iter
->sernum
!= iter
->w
.root
->fn_sernum
) {
2336 iter
->sernum
= iter
->w
.root
->fn_sernum
;
2337 iter
->w
.state
= FWS_INIT
;
2338 iter
->w
.node
= iter
->w
.root
;
2339 WARN_ON(iter
->w
.skip
);
2340 iter
->w
.skip
= iter
->w
.count
;
2344 static void *ipv6_route_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2347 struct fib6_info
*n
;
2348 struct net
*net
= seq_file_net(seq
);
2349 struct ipv6_route_iter
*iter
= seq
->private;
2354 n
= rcu_dereference_bh(((struct fib6_info
*)v
)->fib6_next
);
2361 ipv6_route_check_sernum(iter
);
2362 spin_lock_bh(&iter
->tbl
->tb6_lock
);
2363 r
= fib6_walk_continue(&iter
->w
);
2364 spin_unlock_bh(&iter
->tbl
->tb6_lock
);
2368 return iter
->w
.leaf
;
2370 fib6_walker_unlink(net
, &iter
->w
);
2373 fib6_walker_unlink(net
, &iter
->w
);
2375 iter
->tbl
= ipv6_route_seq_next_table(iter
->tbl
, net
);
2379 ipv6_route_seq_setup_walk(iter
, net
);
2383 static void *ipv6_route_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2386 struct net
*net
= seq_file_net(seq
);
2387 struct ipv6_route_iter
*iter
= seq
->private;
2390 iter
->tbl
= ipv6_route_seq_next_table(NULL
, net
);
2394 ipv6_route_seq_setup_walk(iter
, net
);
2395 return ipv6_route_seq_next(seq
, NULL
, pos
);
2401 static bool ipv6_route_iter_active(struct ipv6_route_iter
*iter
)
2403 struct fib6_walker
*w
= &iter
->w
;
2404 return w
->node
&& !(w
->state
== FWS_U
&& w
->node
== w
->root
);
2407 static void ipv6_route_seq_stop(struct seq_file
*seq
, void *v
)
2410 struct net
*net
= seq_file_net(seq
);
2411 struct ipv6_route_iter
*iter
= seq
->private;
2413 if (ipv6_route_iter_active(iter
))
2414 fib6_walker_unlink(net
, &iter
->w
);
2416 rcu_read_unlock_bh();
2419 const struct seq_operations ipv6_route_seq_ops
= {
2420 .start
= ipv6_route_seq_start
,
2421 .next
= ipv6_route_seq_next
,
2422 .stop
= ipv6_route_seq_stop
,
2423 .show
= ipv6_route_seq_show
2425 #endif /* CONFIG_PROC_FS */