2 * Linux INET6 implementation
3 * Forwarding Information Database
6 * Pedro Roque <roque@di.fc.ul.pt>
8 * $Id: ip6_fib.c,v 1.25 2001/10/31 21:55:55 davem Exp $
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or (at your option) any later version.
18 * Yuji SEKIYA @USAGI: Support default route on router node;
19 * remove ip6_null_entry from the top of
21 * Ville Nuorvala: Fixed routing subtrees.
23 #include <linux/errno.h>
24 #include <linux/types.h>
25 #include <linux/net.h>
26 #include <linux/route.h>
27 #include <linux/netdevice.h>
28 #include <linux/in6.h>
29 #include <linux/init.h>
30 #include <linux/list.h>
33 #include <linux/proc_fs.h>
37 #include <net/ndisc.h>
38 #include <net/addrconf.h>
40 #include <net/ip6_fib.h>
41 #include <net/ip6_route.h>
46 #define RT6_TRACE(x...) printk(KERN_DEBUG x)
48 #define RT6_TRACE(x...) do { ; } while (0)
51 struct rt6_statistics rt6_stats
;
53 static struct kmem_cache
* fib6_node_kmem __read_mostly
;
57 #ifdef CONFIG_IPV6_SUBTREES
68 struct fib6_walker_t w
;
69 int (*func
)(struct rt6_info
*, void *arg
);
73 static DEFINE_RWLOCK(fib6_walker_lock
);
75 #ifdef CONFIG_IPV6_SUBTREES
76 #define FWS_INIT FWS_S
78 #define FWS_INIT FWS_L
81 static void fib6_prune_clones(struct fib6_node
*fn
, struct rt6_info
*rt
);
82 static struct rt6_info
* fib6_find_prefix(struct fib6_node
*fn
);
83 static struct fib6_node
* fib6_repair_tree(struct fib6_node
*fn
);
84 static int fib6_walk(struct fib6_walker_t
*w
);
85 static int fib6_walk_continue(struct fib6_walker_t
*w
);
88 * A routing update causes an increase of the serial number on the
89 * affected subtree. This allows for cached routes to be asynchronously
90 * tested when modifications are made to the destination cache as a
91 * result of redirects, path MTU changes, etc.
94 static __u32 rt_sernum
;
96 static DEFINE_TIMER(ip6_fib_timer
, fib6_run_gc
, 0, 0);
98 static struct fib6_walker_t fib6_walker_list
= {
99 .prev
= &fib6_walker_list
,
100 .next
= &fib6_walker_list
,
103 #define FOR_WALKERS(w) for ((w)=fib6_walker_list.next; (w) != &fib6_walker_list; (w)=(w)->next)
105 static inline void fib6_walker_link(struct fib6_walker_t
*w
)
107 write_lock_bh(&fib6_walker_lock
);
108 w
->next
= fib6_walker_list
.next
;
109 w
->prev
= &fib6_walker_list
;
112 write_unlock_bh(&fib6_walker_lock
);
115 static inline void fib6_walker_unlink(struct fib6_walker_t
*w
)
117 write_lock_bh(&fib6_walker_lock
);
118 w
->next
->prev
= w
->prev
;
119 w
->prev
->next
= w
->next
;
120 w
->prev
= w
->next
= w
;
121 write_unlock_bh(&fib6_walker_lock
);
123 static __inline__ u32
fib6_new_sernum(void)
132 * Auxiliary address test functions for the radix tree.
134 * These assume a 32bit processor (although it will work on
142 static __inline__ __be32
addr_bit_set(void *token
, int fn_bit
)
144 __be32
*addr
= token
;
146 return htonl(1 << ((~fn_bit
)&0x1F)) & addr
[fn_bit
>>5];
149 static __inline__
struct fib6_node
* node_alloc(void)
151 struct fib6_node
*fn
;
153 if ((fn
= kmem_cache_alloc(fib6_node_kmem
, GFP_ATOMIC
)) != NULL
)
154 memset(fn
, 0, sizeof(struct fib6_node
));
159 static __inline__
void node_free(struct fib6_node
* fn
)
161 kmem_cache_free(fib6_node_kmem
, fn
);
164 static __inline__
void rt6_release(struct rt6_info
*rt
)
166 if (atomic_dec_and_test(&rt
->rt6i_ref
))
167 dst_free(&rt
->u
.dst
);
170 static struct fib6_table fib6_main_tbl
= {
171 .tb6_id
= RT6_TABLE_MAIN
,
173 .leaf
= &ip6_null_entry
,
174 .fn_flags
= RTN_ROOT
| RTN_TL_ROOT
| RTN_RTINFO
,
178 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
179 #define FIB_TABLE_HASHSZ 256
181 #define FIB_TABLE_HASHSZ 1
183 static struct hlist_head fib_table_hash
[FIB_TABLE_HASHSZ
];
185 static void fib6_link_table(struct fib6_table
*tb
)
190 * Initialize table lock at a single place to give lockdep a key,
191 * tables aren't visible prior to being linked to the list.
193 rwlock_init(&tb
->tb6_lock
);
195 h
= tb
->tb6_id
& (FIB_TABLE_HASHSZ
- 1);
198 * No protection necessary, this is the only list mutatation
199 * operation, tables never disappear once they exist.
201 hlist_add_head_rcu(&tb
->tb6_hlist
, &fib_table_hash
[h
]);
204 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
205 static struct fib6_table fib6_local_tbl
= {
206 .tb6_id
= RT6_TABLE_LOCAL
,
208 .leaf
= &ip6_null_entry
,
209 .fn_flags
= RTN_ROOT
| RTN_TL_ROOT
| RTN_RTINFO
,
213 static struct fib6_table
*fib6_alloc_table(u32 id
)
215 struct fib6_table
*table
;
217 table
= kzalloc(sizeof(*table
), GFP_ATOMIC
);
220 table
->tb6_root
.leaf
= &ip6_null_entry
;
221 table
->tb6_root
.fn_flags
= RTN_ROOT
| RTN_TL_ROOT
| RTN_RTINFO
;
227 struct fib6_table
*fib6_new_table(u32 id
)
229 struct fib6_table
*tb
;
233 tb
= fib6_get_table(id
);
237 tb
= fib6_alloc_table(id
);
244 struct fib6_table
*fib6_get_table(u32 id
)
246 struct fib6_table
*tb
;
247 struct hlist_node
*node
;
252 h
= id
& (FIB_TABLE_HASHSZ
- 1);
254 hlist_for_each_entry_rcu(tb
, node
, &fib_table_hash
[h
], tb6_hlist
) {
255 if (tb
->tb6_id
== id
) {
265 static void __init
fib6_tables_init(void)
267 fib6_link_table(&fib6_main_tbl
);
268 fib6_link_table(&fib6_local_tbl
);
273 struct fib6_table
*fib6_new_table(u32 id
)
275 return fib6_get_table(id
);
278 struct fib6_table
*fib6_get_table(u32 id
)
280 return &fib6_main_tbl
;
283 struct dst_entry
*fib6_rule_lookup(struct flowi
*fl
, int flags
,
286 return (struct dst_entry
*) lookup(&fib6_main_tbl
, fl
, flags
);
289 static void __init
fib6_tables_init(void)
291 fib6_link_table(&fib6_main_tbl
);
296 static int fib6_dump_node(struct fib6_walker_t
*w
)
301 for (rt
= w
->leaf
; rt
; rt
= rt
->u
.next
) {
302 res
= rt6_dump_route(rt
, w
->args
);
304 /* Frame is full, suspend walking */
314 static void fib6_dump_end(struct netlink_callback
*cb
)
316 struct fib6_walker_t
*w
= (void*)cb
->args
[2];
322 cb
->done
= (void*)cb
->args
[3];
326 static int fib6_dump_done(struct netlink_callback
*cb
)
329 return cb
->done
? cb
->done(cb
) : 0;
332 static int fib6_dump_table(struct fib6_table
*table
, struct sk_buff
*skb
,
333 struct netlink_callback
*cb
)
335 struct fib6_walker_t
*w
;
338 w
= (void *)cb
->args
[2];
339 w
->root
= &table
->tb6_root
;
341 if (cb
->args
[4] == 0) {
342 read_lock_bh(&table
->tb6_lock
);
344 read_unlock_bh(&table
->tb6_lock
);
348 read_lock_bh(&table
->tb6_lock
);
349 res
= fib6_walk_continue(w
);
350 read_unlock_bh(&table
->tb6_lock
);
353 fib6_walker_unlink(w
);
356 fib6_walker_unlink(w
);
363 int inet6_dump_fib(struct sk_buff
*skb
, struct netlink_callback
*cb
)
366 unsigned int e
= 0, s_e
;
367 struct rt6_rtnl_dump_arg arg
;
368 struct fib6_walker_t
*w
;
369 struct fib6_table
*tb
;
370 struct hlist_node
*node
;
376 w
= (void *)cb
->args
[2];
380 * 1. hook callback destructor.
382 cb
->args
[3] = (long)cb
->done
;
383 cb
->done
= fib6_dump_done
;
386 * 2. allocate and initialize walker.
388 w
= kzalloc(sizeof(*w
), GFP_ATOMIC
);
391 w
->func
= fib6_dump_node
;
392 cb
->args
[2] = (long)w
;
399 for (h
= s_h
; h
< FIB_TABLE_HASHSZ
; h
++, s_e
= 0) {
401 hlist_for_each_entry(tb
, node
, &fib_table_hash
[h
], tb6_hlist
) {
404 res
= fib6_dump_table(tb
, skb
, cb
);
415 res
= res
< 0 ? res
: skb
->len
;
424 * return the appropriate node for a routing tree "add" operation
425 * by either creating and inserting or by returning an existing
429 static struct fib6_node
* fib6_add_1(struct fib6_node
*root
, void *addr
,
430 int addrlen
, int plen
,
433 struct fib6_node
*fn
, *in
, *ln
;
434 struct fib6_node
*pn
= NULL
;
438 __u32 sernum
= fib6_new_sernum();
440 RT6_TRACE("fib6_add_1\n");
442 /* insert node in tree */
447 key
= (struct rt6key
*)((u8
*)fn
->leaf
+ offset
);
452 if (plen
< fn
->fn_bit
||
453 !ipv6_prefix_equal(&key
->addr
, addr
, fn
->fn_bit
))
460 if (plen
== fn
->fn_bit
) {
461 /* clean up an intermediate node */
462 if ((fn
->fn_flags
& RTN_RTINFO
) == 0) {
463 rt6_release(fn
->leaf
);
467 fn
->fn_sernum
= sernum
;
473 * We have more bits to go
476 /* Try to walk down on tree. */
477 fn
->fn_sernum
= sernum
;
478 dir
= addr_bit_set(addr
, fn
->fn_bit
);
480 fn
= dir
? fn
->right
: fn
->left
;
484 * We walked to the bottom of tree.
485 * Create new leaf node without children.
495 ln
->fn_sernum
= sernum
;
507 * split since we don't have a common prefix anymore or
508 * we have a less significant route.
509 * we've to insert an intermediate node on the list
510 * this new node will point to the one we need to create
516 /* find 1st bit in difference between the 2 addrs.
518 See comment in __ipv6_addr_diff: bit may be an invalid value,
519 but if it is >= plen, the value is ignored in any case.
522 bit
= __ipv6_addr_diff(addr
, &key
->addr
, addrlen
);
527 * (new leaf node)[ln] (old node)[fn]
533 if (in
== NULL
|| ln
== NULL
) {
542 * new intermediate node.
544 * be off since that an address that chooses one of
545 * the branches would not match less specific routes
546 * in the other branch
553 atomic_inc(&in
->leaf
->rt6i_ref
);
555 in
->fn_sernum
= sernum
;
557 /* update parent pointer */
568 ln
->fn_sernum
= sernum
;
570 if (addr_bit_set(addr
, bit
)) {
577 } else { /* plen <= bit */
580 * (new leaf node)[ln]
582 * (old node)[fn] NULL
594 ln
->fn_sernum
= sernum
;
601 if (addr_bit_set(&key
->addr
, plen
))
612 * Insert routing information in a node.
615 static int fib6_add_rt2node(struct fib6_node
*fn
, struct rt6_info
*rt
,
616 struct nl_info
*info
)
618 struct rt6_info
*iter
= NULL
;
619 struct rt6_info
**ins
;
623 for (iter
= fn
->leaf
; iter
; iter
=iter
->u
.next
) {
625 * Search for duplicates
628 if (iter
->rt6i_metric
== rt
->rt6i_metric
) {
630 * Same priority level
633 if (iter
->rt6i_dev
== rt
->rt6i_dev
&&
634 iter
->rt6i_idev
== rt
->rt6i_idev
&&
635 ipv6_addr_equal(&iter
->rt6i_gateway
,
636 &rt
->rt6i_gateway
)) {
637 if (!(iter
->rt6i_flags
&RTF_EXPIRES
))
639 iter
->rt6i_expires
= rt
->rt6i_expires
;
640 if (!(rt
->rt6i_flags
&RTF_EXPIRES
)) {
641 iter
->rt6i_flags
&= ~RTF_EXPIRES
;
642 iter
->rt6i_expires
= 0;
648 if (iter
->rt6i_metric
> rt
->rt6i_metric
)
654 /* Reset round-robin state, if necessary */
655 if (ins
== &fn
->leaf
)
666 atomic_inc(&rt
->rt6i_ref
);
667 inet6_rt_notify(RTM_NEWROUTE
, rt
, info
);
668 rt6_stats
.fib_rt_entries
++;
670 if ((fn
->fn_flags
& RTN_RTINFO
) == 0) {
671 rt6_stats
.fib_route_nodes
++;
672 fn
->fn_flags
|= RTN_RTINFO
;
678 static __inline__
void fib6_start_gc(struct rt6_info
*rt
)
680 if (ip6_fib_timer
.expires
== 0 &&
681 (rt
->rt6i_flags
& (RTF_EXPIRES
|RTF_CACHE
)))
682 mod_timer(&ip6_fib_timer
, jiffies
+ ip6_rt_gc_interval
);
685 void fib6_force_start_gc(void)
687 if (ip6_fib_timer
.expires
== 0)
688 mod_timer(&ip6_fib_timer
, jiffies
+ ip6_rt_gc_interval
);
692 * Add routing information to the routing tree.
693 * <destination addr>/<source addr>
694 * with source addr info in sub-trees
697 int fib6_add(struct fib6_node
*root
, struct rt6_info
*rt
, struct nl_info
*info
)
699 struct fib6_node
*fn
, *pn
= NULL
;
702 fn
= fib6_add_1(root
, &rt
->rt6i_dst
.addr
, sizeof(struct in6_addr
),
703 rt
->rt6i_dst
.plen
, offsetof(struct rt6_info
, rt6i_dst
));
710 #ifdef CONFIG_IPV6_SUBTREES
711 if (rt
->rt6i_src
.plen
) {
712 struct fib6_node
*sn
;
714 if (fn
->subtree
== NULL
) {
715 struct fib6_node
*sfn
;
727 /* Create subtree root node */
732 sfn
->leaf
= &ip6_null_entry
;
733 atomic_inc(&ip6_null_entry
.rt6i_ref
);
734 sfn
->fn_flags
= RTN_ROOT
;
735 sfn
->fn_sernum
= fib6_new_sernum();
737 /* Now add the first leaf node to new subtree */
739 sn
= fib6_add_1(sfn
, &rt
->rt6i_src
.addr
,
740 sizeof(struct in6_addr
), rt
->rt6i_src
.plen
,
741 offsetof(struct rt6_info
, rt6i_src
));
744 /* If it is failed, discard just allocated
745 root, and then (in st_failure) stale node
752 /* Now link new subtree to main tree */
756 sn
= fib6_add_1(fn
->subtree
, &rt
->rt6i_src
.addr
,
757 sizeof(struct in6_addr
), rt
->rt6i_src
.plen
,
758 offsetof(struct rt6_info
, rt6i_src
));
764 if (fn
->leaf
== NULL
) {
766 atomic_inc(&rt
->rt6i_ref
);
772 err
= fib6_add_rt2node(fn
, rt
, info
);
776 if (!(rt
->rt6i_flags
&RTF_CACHE
))
777 fib6_prune_clones(pn
, rt
);
782 #ifdef CONFIG_IPV6_SUBTREES
784 * If fib6_add_1 has cleared the old leaf pointer in the
785 * super-tree leaf node we have to find a new one for it.
787 if (pn
!= fn
&& !pn
->leaf
&& !(pn
->fn_flags
& RTN_RTINFO
)) {
788 pn
->leaf
= fib6_find_prefix(pn
);
791 BUG_TRAP(pn
->leaf
!= NULL
);
792 pn
->leaf
= &ip6_null_entry
;
795 atomic_inc(&pn
->leaf
->rt6i_ref
);
798 dst_free(&rt
->u
.dst
);
802 #ifdef CONFIG_IPV6_SUBTREES
803 /* Subtree creation failed, probably main tree node
804 is orphan. If it is, shoot it.
807 if (fn
&& !(fn
->fn_flags
& (RTN_RTINFO
|RTN_ROOT
)))
808 fib6_repair_tree(fn
);
809 dst_free(&rt
->u
.dst
);
815 * Routing tree lookup
820 int offset
; /* key offset on rt6_info */
821 struct in6_addr
*addr
; /* search key */
824 static struct fib6_node
* fib6_lookup_1(struct fib6_node
*root
,
825 struct lookup_args
*args
)
827 struct fib6_node
*fn
;
830 if (unlikely(args
->offset
== 0))
840 struct fib6_node
*next
;
842 dir
= addr_bit_set(args
->addr
, fn
->fn_bit
);
844 next
= dir
? fn
->right
: fn
->left
;
855 if (FIB6_SUBTREE(fn
) || fn
->fn_flags
& RTN_RTINFO
) {
858 key
= (struct rt6key
*) ((u8
*) fn
->leaf
+
861 if (ipv6_prefix_equal(&key
->addr
, args
->addr
, key
->plen
)) {
862 #ifdef CONFIG_IPV6_SUBTREES
864 fn
= fib6_lookup_1(fn
->subtree
, args
+ 1);
866 if (!fn
|| fn
->fn_flags
& RTN_RTINFO
)
871 if (fn
->fn_flags
& RTN_ROOT
)
880 struct fib6_node
* fib6_lookup(struct fib6_node
*root
, struct in6_addr
*daddr
,
881 struct in6_addr
*saddr
)
883 struct fib6_node
*fn
;
884 struct lookup_args args
[] = {
886 .offset
= offsetof(struct rt6_info
, rt6i_dst
),
889 #ifdef CONFIG_IPV6_SUBTREES
891 .offset
= offsetof(struct rt6_info
, rt6i_src
),
896 .offset
= 0, /* sentinel */
900 fn
= fib6_lookup_1(root
, daddr
? args
: args
+ 1);
902 if (fn
== NULL
|| fn
->fn_flags
& RTN_TL_ROOT
)
909 * Get node with specified destination prefix (and source prefix,
910 * if subtrees are used)
914 static struct fib6_node
* fib6_locate_1(struct fib6_node
*root
,
915 struct in6_addr
*addr
,
916 int plen
, int offset
)
918 struct fib6_node
*fn
;
920 for (fn
= root
; fn
; ) {
921 struct rt6key
*key
= (struct rt6key
*)((u8
*)fn
->leaf
+ offset
);
926 if (plen
< fn
->fn_bit
||
927 !ipv6_prefix_equal(&key
->addr
, addr
, fn
->fn_bit
))
930 if (plen
== fn
->fn_bit
)
934 * We have more bits to go
936 if (addr_bit_set(addr
, fn
->fn_bit
))
944 struct fib6_node
* fib6_locate(struct fib6_node
*root
,
945 struct in6_addr
*daddr
, int dst_len
,
946 struct in6_addr
*saddr
, int src_len
)
948 struct fib6_node
*fn
;
950 fn
= fib6_locate_1(root
, daddr
, dst_len
,
951 offsetof(struct rt6_info
, rt6i_dst
));
953 #ifdef CONFIG_IPV6_SUBTREES
955 BUG_TRAP(saddr
!=NULL
);
956 if (fn
&& fn
->subtree
)
957 fn
= fib6_locate_1(fn
->subtree
, saddr
, src_len
,
958 offsetof(struct rt6_info
, rt6i_src
));
962 if (fn
&& fn
->fn_flags
&RTN_RTINFO
)
974 static struct rt6_info
* fib6_find_prefix(struct fib6_node
*fn
)
976 if (fn
->fn_flags
&RTN_ROOT
)
977 return &ip6_null_entry
;
981 return fn
->left
->leaf
;
984 return fn
->right
->leaf
;
986 fn
= FIB6_SUBTREE(fn
);
992 * Called to trim the tree of intermediate nodes when possible. "fn"
993 * is the node we want to try and remove.
996 static struct fib6_node
* fib6_repair_tree(struct fib6_node
*fn
)
1000 struct fib6_node
*child
, *pn
;
1001 struct fib6_walker_t
*w
;
1005 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn
->fn_bit
, iter
);
1008 BUG_TRAP(!(fn
->fn_flags
&RTN_RTINFO
));
1009 BUG_TRAP(!(fn
->fn_flags
&RTN_TL_ROOT
));
1010 BUG_TRAP(fn
->leaf
==NULL
);
1014 if (fn
->right
) child
= fn
->right
, children
|= 1;
1015 if (fn
->left
) child
= fn
->left
, children
|= 2;
1017 if (children
== 3 || FIB6_SUBTREE(fn
)
1018 #ifdef CONFIG_IPV6_SUBTREES
1019 /* Subtree root (i.e. fn) may have one child */
1020 || (children
&& fn
->fn_flags
&RTN_ROOT
)
1023 fn
->leaf
= fib6_find_prefix(fn
);
1025 if (fn
->leaf
==NULL
) {
1027 fn
->leaf
= &ip6_null_entry
;
1030 atomic_inc(&fn
->leaf
->rt6i_ref
);
1035 #ifdef CONFIG_IPV6_SUBTREES
1036 if (FIB6_SUBTREE(pn
) == fn
) {
1037 BUG_TRAP(fn
->fn_flags
&RTN_ROOT
);
1038 FIB6_SUBTREE(pn
) = NULL
;
1041 BUG_TRAP(!(fn
->fn_flags
&RTN_ROOT
));
1043 if (pn
->right
== fn
) pn
->right
= child
;
1044 else if (pn
->left
== fn
) pn
->left
= child
;
1051 #ifdef CONFIG_IPV6_SUBTREES
1055 read_lock(&fib6_walker_lock
);
1057 if (child
== NULL
) {
1058 if (w
->root
== fn
) {
1059 w
->root
= w
->node
= NULL
;
1060 RT6_TRACE("W %p adjusted by delroot 1\n", w
);
1061 } else if (w
->node
== fn
) {
1062 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w
, w
->state
, nstate
);
1067 if (w
->root
== fn
) {
1069 RT6_TRACE("W %p adjusted by delroot 2\n", w
);
1071 if (w
->node
== fn
) {
1074 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w
, w
->state
);
1075 w
->state
= w
->state
>=FWS_R
? FWS_U
: FWS_INIT
;
1077 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w
, w
->state
);
1078 w
->state
= w
->state
>=FWS_C
? FWS_U
: FWS_INIT
;
1083 read_unlock(&fib6_walker_lock
);
1086 if (pn
->fn_flags
&RTN_RTINFO
|| FIB6_SUBTREE(pn
))
1089 rt6_release(pn
->leaf
);
1095 static void fib6_del_route(struct fib6_node
*fn
, struct rt6_info
**rtp
,
1096 struct nl_info
*info
)
1098 struct fib6_walker_t
*w
;
1099 struct rt6_info
*rt
= *rtp
;
1101 RT6_TRACE("fib6_del_route\n");
1105 rt
->rt6i_node
= NULL
;
1106 rt6_stats
.fib_rt_entries
--;
1107 rt6_stats
.fib_discarded_routes
++;
1109 /* Reset round-robin state, if necessary */
1110 if (fn
->rr_ptr
== rt
)
1113 /* Adjust walkers */
1114 read_lock(&fib6_walker_lock
);
1116 if (w
->state
== FWS_C
&& w
->leaf
== rt
) {
1117 RT6_TRACE("walker %p adjusted by delroute\n", w
);
1118 w
->leaf
= rt
->u
.next
;
1119 if (w
->leaf
== NULL
)
1123 read_unlock(&fib6_walker_lock
);
1127 if (fn
->leaf
== NULL
&& fn
->fn_flags
&RTN_TL_ROOT
)
1128 fn
->leaf
= &ip6_null_entry
;
1130 /* If it was last route, expunge its radix tree node */
1131 if (fn
->leaf
== NULL
) {
1132 fn
->fn_flags
&= ~RTN_RTINFO
;
1133 rt6_stats
.fib_route_nodes
--;
1134 fn
= fib6_repair_tree(fn
);
1137 if (atomic_read(&rt
->rt6i_ref
) != 1) {
1138 /* This route is used as dummy address holder in some split
1139 * nodes. It is not leaked, but it still holds other resources,
1140 * which must be released in time. So, scan ascendant nodes
1141 * and replace dummy references to this route with references
1142 * to still alive ones.
1145 if (!(fn
->fn_flags
&RTN_RTINFO
) && fn
->leaf
== rt
) {
1146 fn
->leaf
= fib6_find_prefix(fn
);
1147 atomic_inc(&fn
->leaf
->rt6i_ref
);
1152 /* No more references are possible at this point. */
1153 if (atomic_read(&rt
->rt6i_ref
) != 1) BUG();
1156 inet6_rt_notify(RTM_DELROUTE
, rt
, info
);
1160 int fib6_del(struct rt6_info
*rt
, struct nl_info
*info
)
1162 struct fib6_node
*fn
= rt
->rt6i_node
;
1163 struct rt6_info
**rtp
;
1166 if (rt
->u
.dst
.obsolete
>0) {
1171 if (fn
== NULL
|| rt
== &ip6_null_entry
)
1174 BUG_TRAP(fn
->fn_flags
&RTN_RTINFO
);
1176 if (!(rt
->rt6i_flags
&RTF_CACHE
)) {
1177 struct fib6_node
*pn
= fn
;
1178 #ifdef CONFIG_IPV6_SUBTREES
1179 /* clones of this route might be in another subtree */
1180 if (rt
->rt6i_src
.plen
) {
1181 while (!(pn
->fn_flags
&RTN_ROOT
))
1186 fib6_prune_clones(pn
, rt
);
1190 * Walk the leaf entries looking for ourself
1193 for (rtp
= &fn
->leaf
; *rtp
; rtp
= &(*rtp
)->u
.next
) {
1195 fib6_del_route(fn
, rtp
, info
);
1203 * Tree traversal function.
1205 * Certainly, it is not interrupt safe.
1206 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1207 * It means, that we can modify tree during walking
1208 * and use this function for garbage collection, clone pruning,
1209 * cleaning tree when a device goes down etc. etc.
1211 * It guarantees that every node will be traversed,
1212 * and that it will be traversed only once.
1214 * Callback function w->func may return:
1215 * 0 -> continue walking.
1216 * positive value -> walking is suspended (used by tree dumps,
1217 * and probably by gc, if it will be split to several slices)
1218 * negative value -> terminate walking.
1220 * The function itself returns:
1221 * 0 -> walk is complete.
1222 * >0 -> walk is incomplete (i.e. suspended)
1223 * <0 -> walk is terminated by an error.
1226 static int fib6_walk_continue(struct fib6_walker_t
*w
)
1228 struct fib6_node
*fn
, *pn
;
1235 if (w
->prune
&& fn
!= w
->root
&&
1236 fn
->fn_flags
&RTN_RTINFO
&& w
->state
< FWS_C
) {
1241 #ifdef CONFIG_IPV6_SUBTREES
1243 if (FIB6_SUBTREE(fn
)) {
1244 w
->node
= FIB6_SUBTREE(fn
);
1252 w
->state
= FWS_INIT
;
1258 w
->node
= fn
->right
;
1259 w
->state
= FWS_INIT
;
1265 if (w
->leaf
&& fn
->fn_flags
&RTN_RTINFO
) {
1266 int err
= w
->func(w
);
1277 #ifdef CONFIG_IPV6_SUBTREES
1278 if (FIB6_SUBTREE(pn
) == fn
) {
1279 BUG_TRAP(fn
->fn_flags
&RTN_ROOT
);
1284 if (pn
->left
== fn
) {
1288 if (pn
->right
== fn
) {
1290 w
->leaf
= w
->node
->leaf
;
1300 static int fib6_walk(struct fib6_walker_t
*w
)
1304 w
->state
= FWS_INIT
;
1307 fib6_walker_link(w
);
1308 res
= fib6_walk_continue(w
);
1310 fib6_walker_unlink(w
);
1314 static int fib6_clean_node(struct fib6_walker_t
*w
)
1317 struct rt6_info
*rt
;
1318 struct fib6_cleaner_t
*c
= (struct fib6_cleaner_t
*)w
;
1320 for (rt
= w
->leaf
; rt
; rt
= rt
->u
.next
) {
1321 res
= c
->func(rt
, c
->arg
);
1324 res
= fib6_del(rt
, NULL
);
1327 printk(KERN_DEBUG
"fib6_clean_node: del failed: rt=%p@%p err=%d\n", rt
, rt
->rt6i_node
, res
);
1340 * Convenient frontend to tree walker.
1342 * func is called on each route.
1343 * It may return -1 -> delete this route.
1344 * 0 -> continue walking
1346 * prune==1 -> only immediate children of node (certainly,
1347 * ignoring pure split nodes) will be scanned.
1350 static void fib6_clean_tree(struct fib6_node
*root
,
1351 int (*func
)(struct rt6_info
*, void *arg
),
1352 int prune
, void *arg
)
1354 struct fib6_cleaner_t c
;
1357 c
.w
.func
= fib6_clean_node
;
1365 void fib6_clean_all(int (*func
)(struct rt6_info
*, void *arg
),
1366 int prune
, void *arg
)
1368 struct fib6_table
*table
;
1369 struct hlist_node
*node
;
1373 for (h
= 0; h
< FIB_TABLE_HASHSZ
; h
++) {
1374 hlist_for_each_entry_rcu(table
, node
, &fib_table_hash
[h
],
1376 write_lock_bh(&table
->tb6_lock
);
1377 fib6_clean_tree(&table
->tb6_root
, func
, prune
, arg
);
1378 write_unlock_bh(&table
->tb6_lock
);
1384 static int fib6_prune_clone(struct rt6_info
*rt
, void *arg
)
1386 if (rt
->rt6i_flags
& RTF_CACHE
) {
1387 RT6_TRACE("pruning clone %p\n", rt
);
1394 static void fib6_prune_clones(struct fib6_node
*fn
, struct rt6_info
*rt
)
1396 fib6_clean_tree(fn
, fib6_prune_clone
, 1, rt
);
1400 * Garbage collection
1403 static struct fib6_gc_args
1409 static int fib6_age(struct rt6_info
*rt
, void *arg
)
1411 unsigned long now
= jiffies
;
1414 * check addrconf expiration here.
1415 * Routes are expired even if they are in use.
1417 * Also age clones. Note, that clones are aged out
1418 * only if they are not in use now.
1421 if (rt
->rt6i_flags
&RTF_EXPIRES
&& rt
->rt6i_expires
) {
1422 if (time_after(now
, rt
->rt6i_expires
)) {
1423 RT6_TRACE("expiring %p\n", rt
);
1427 } else if (rt
->rt6i_flags
& RTF_CACHE
) {
1428 if (atomic_read(&rt
->u
.dst
.__refcnt
) == 0 &&
1429 time_after_eq(now
, rt
->u
.dst
.lastuse
+ gc_args
.timeout
)) {
1430 RT6_TRACE("aging clone %p\n", rt
);
1432 } else if ((rt
->rt6i_flags
& RTF_GATEWAY
) &&
1433 (!(rt
->rt6i_nexthop
->flags
& NTF_ROUTER
))) {
1434 RT6_TRACE("purging route %p via non-router but gateway\n",
1444 static DEFINE_SPINLOCK(fib6_gc_lock
);
1446 void fib6_run_gc(unsigned long dummy
)
1448 if (dummy
!= ~0UL) {
1449 spin_lock_bh(&fib6_gc_lock
);
1450 gc_args
.timeout
= dummy
? (int)dummy
: ip6_rt_gc_interval
;
1453 if (!spin_trylock(&fib6_gc_lock
)) {
1454 mod_timer(&ip6_fib_timer
, jiffies
+ HZ
);
1458 gc_args
.timeout
= ip6_rt_gc_interval
;
1462 ndisc_dst_gc(&gc_args
.more
);
1463 fib6_clean_all(fib6_age
, 0, NULL
);
1466 mod_timer(&ip6_fib_timer
, jiffies
+ ip6_rt_gc_interval
);
1468 del_timer(&ip6_fib_timer
);
1469 ip6_fib_timer
.expires
= 0;
1471 spin_unlock_bh(&fib6_gc_lock
);
1474 void __init
fib6_init(void)
1476 fib6_node_kmem
= kmem_cache_create("fib6_nodes",
1477 sizeof(struct fib6_node
),
1478 0, SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
1484 void fib6_gc_cleanup(void)
1486 del_timer(&ip6_fib_timer
);
1487 kmem_cache_destroy(fib6_node_kmem
);