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.
16 * Yuji SEKIYA @USAGI: Support default route on router node;
17 * remove ip6_null_entry from the top of
19 * Ville Nuorvala: Fixed routing subtrees.
22 #define pr_fmt(fmt) "IPv6: " fmt
24 #include <linux/errno.h>
25 #include <linux/types.h>
26 #include <linux/net.h>
27 #include <linux/route.h>
28 #include <linux/netdevice.h>
29 #include <linux/in6.h>
30 #include <linux/init.h>
31 #include <linux/list.h>
32 #include <linux/slab.h>
35 #include <net/ndisc.h>
36 #include <net/addrconf.h>
38 #include <net/ip6_fib.h>
39 #include <net/ip6_route.h>
44 #define RT6_TRACE(x...) pr_debug(x)
46 #define RT6_TRACE(x...) do { ; } while (0)
49 static struct kmem_cache
* fib6_node_kmem __read_mostly
;
53 #ifdef CONFIG_IPV6_SUBTREES
64 struct fib6_walker_t w
;
66 int (*func
)(struct rt6_info
*, void *arg
);
70 static DEFINE_RWLOCK(fib6_walker_lock
);
72 #ifdef CONFIG_IPV6_SUBTREES
73 #define FWS_INIT FWS_S
75 #define FWS_INIT FWS_L
78 static void fib6_prune_clones(struct net
*net
, struct fib6_node
*fn
,
80 static struct rt6_info
*fib6_find_prefix(struct net
*net
, struct fib6_node
*fn
);
81 static struct fib6_node
*fib6_repair_tree(struct net
*net
, struct fib6_node
*fn
);
82 static int fib6_walk(struct fib6_walker_t
*w
);
83 static int fib6_walk_continue(struct fib6_walker_t
*w
);
86 * A routing update causes an increase of the serial number on the
87 * affected subtree. This allows for cached routes to be asynchronously
88 * tested when modifications are made to the destination cache as a
89 * result of redirects, path MTU changes, etc.
92 static __u32 rt_sernum
;
94 static void fib6_gc_timer_cb(unsigned long arg
);
96 static LIST_HEAD(fib6_walkers
);
97 #define FOR_WALKERS(w) list_for_each_entry(w, &fib6_walkers, lh)
99 static inline void fib6_walker_link(struct fib6_walker_t
*w
)
101 write_lock_bh(&fib6_walker_lock
);
102 list_add(&w
->lh
, &fib6_walkers
);
103 write_unlock_bh(&fib6_walker_lock
);
106 static inline void fib6_walker_unlink(struct fib6_walker_t
*w
)
108 write_lock_bh(&fib6_walker_lock
);
110 write_unlock_bh(&fib6_walker_lock
);
112 static __inline__ u32
fib6_new_sernum(void)
121 * Auxiliary address test functions for the radix tree.
123 * These assume a 32bit processor (although it will work on
130 #if defined(__LITTLE_ENDIAN)
131 # define BITOP_BE32_SWIZZLE (0x1F & ~7)
133 # define BITOP_BE32_SWIZZLE 0
136 static __inline__ __be32
addr_bit_set(const void *token
, int fn_bit
)
138 const __be32
*addr
= token
;
141 * 1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
142 * is optimized version of
143 * htonl(1 << ((~fn_bit)&0x1F))
144 * See include/asm-generic/bitops/le.h.
146 return (__force __be32
)(1 << ((~fn_bit
^ BITOP_BE32_SWIZZLE
) & 0x1f)) &
150 static __inline__
struct fib6_node
* node_alloc(void)
152 struct fib6_node
*fn
;
154 fn
= kmem_cache_zalloc(fib6_node_kmem
, GFP_ATOMIC
);
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
))
170 static void fib6_link_table(struct net
*net
, struct fib6_table
*tb
)
175 * Initialize table lock at a single place to give lockdep a key,
176 * tables aren't visible prior to being linked to the list.
178 rwlock_init(&tb
->tb6_lock
);
180 h
= tb
->tb6_id
& (FIB6_TABLE_HASHSZ
- 1);
183 * No protection necessary, this is the only list mutatation
184 * operation, tables never disappear once they exist.
186 hlist_add_head_rcu(&tb
->tb6_hlist
, &net
->ipv6
.fib_table_hash
[h
]);
189 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
191 static struct fib6_table
*fib6_alloc_table(struct net
*net
, u32 id
)
193 struct fib6_table
*table
;
195 table
= kzalloc(sizeof(*table
), GFP_ATOMIC
);
198 table
->tb6_root
.leaf
= net
->ipv6
.ip6_null_entry
;
199 table
->tb6_root
.fn_flags
= RTN_ROOT
| RTN_TL_ROOT
| RTN_RTINFO
;
200 inet_peer_base_init(&table
->tb6_peers
);
206 struct fib6_table
*fib6_new_table(struct net
*net
, u32 id
)
208 struct fib6_table
*tb
;
212 tb
= fib6_get_table(net
, id
);
216 tb
= fib6_alloc_table(net
, id
);
218 fib6_link_table(net
, tb
);
223 struct fib6_table
*fib6_get_table(struct net
*net
, u32 id
)
225 struct fib6_table
*tb
;
226 struct hlist_head
*head
;
231 h
= id
& (FIB6_TABLE_HASHSZ
- 1);
233 head
= &net
->ipv6
.fib_table_hash
[h
];
234 hlist_for_each_entry_rcu(tb
, head
, tb6_hlist
) {
235 if (tb
->tb6_id
== id
) {
245 static void __net_init
fib6_tables_init(struct net
*net
)
247 fib6_link_table(net
, net
->ipv6
.fib6_main_tbl
);
248 fib6_link_table(net
, net
->ipv6
.fib6_local_tbl
);
252 struct fib6_table
*fib6_new_table(struct net
*net
, u32 id
)
254 return fib6_get_table(net
, id
);
257 struct fib6_table
*fib6_get_table(struct net
*net
, u32 id
)
259 return net
->ipv6
.fib6_main_tbl
;
262 struct dst_entry
*fib6_rule_lookup(struct net
*net
, struct flowi6
*fl6
,
263 int flags
, pol_lookup_t lookup
)
265 return (struct dst_entry
*) lookup(net
, net
->ipv6
.fib6_main_tbl
, fl6
, flags
);
268 static void __net_init
fib6_tables_init(struct net
*net
)
270 fib6_link_table(net
, net
->ipv6
.fib6_main_tbl
);
275 static int fib6_dump_node(struct fib6_walker_t
*w
)
280 for (rt
= w
->leaf
; rt
; rt
= rt
->dst
.rt6_next
) {
281 res
= rt6_dump_route(rt
, w
->args
);
283 /* Frame is full, suspend walking */
293 static void fib6_dump_end(struct netlink_callback
*cb
)
295 struct fib6_walker_t
*w
= (void*)cb
->args
[2];
300 fib6_walker_unlink(w
);
305 cb
->done
= (void*)cb
->args
[3];
309 static int fib6_dump_done(struct netlink_callback
*cb
)
312 return cb
->done
? cb
->done(cb
) : 0;
315 static int fib6_dump_table(struct fib6_table
*table
, struct sk_buff
*skb
,
316 struct netlink_callback
*cb
)
318 struct fib6_walker_t
*w
;
321 w
= (void *)cb
->args
[2];
322 w
->root
= &table
->tb6_root
;
324 if (cb
->args
[4] == 0) {
328 read_lock_bh(&table
->tb6_lock
);
330 read_unlock_bh(&table
->tb6_lock
);
333 cb
->args
[5] = w
->root
->fn_sernum
;
336 if (cb
->args
[5] != w
->root
->fn_sernum
) {
337 /* Begin at the root if the tree changed */
338 cb
->args
[5] = w
->root
->fn_sernum
;
345 read_lock_bh(&table
->tb6_lock
);
346 res
= fib6_walk_continue(w
);
347 read_unlock_bh(&table
->tb6_lock
);
349 fib6_walker_unlink(w
);
357 static int inet6_dump_fib(struct sk_buff
*skb
, struct netlink_callback
*cb
)
359 struct net
*net
= sock_net(skb
->sk
);
361 unsigned int e
= 0, s_e
;
362 struct rt6_rtnl_dump_arg arg
;
363 struct fib6_walker_t
*w
;
364 struct fib6_table
*tb
;
365 struct hlist_head
*head
;
371 w
= (void *)cb
->args
[2];
375 * 1. hook callback destructor.
377 cb
->args
[3] = (long)cb
->done
;
378 cb
->done
= fib6_dump_done
;
381 * 2. allocate and initialize walker.
383 w
= kzalloc(sizeof(*w
), GFP_ATOMIC
);
386 w
->func
= fib6_dump_node
;
387 cb
->args
[2] = (long)w
;
396 for (h
= s_h
; h
< FIB6_TABLE_HASHSZ
; h
++, s_e
= 0) {
398 head
= &net
->ipv6
.fib_table_hash
[h
];
399 hlist_for_each_entry_rcu(tb
, head
, tb6_hlist
) {
402 res
= fib6_dump_table(tb
, skb
, cb
);
414 res
= res
< 0 ? res
: skb
->len
;
423 * return the appropriate node for a routing tree "add" operation
424 * by either creating and inserting or by returning an existing
428 static struct fib6_node
* fib6_add_1(struct fib6_node
*root
, void *addr
,
429 int addrlen
, int plen
,
430 int offset
, int allow_create
,
431 int replace_required
)
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
)) {
455 if (replace_required
) {
456 pr_warn("Can't replace route, no match found\n");
457 return ERR_PTR(-ENOENT
);
459 pr_warn("NLM_F_CREATE should be set when creating new route\n");
468 if (plen
== fn
->fn_bit
) {
469 /* clean up an intermediate node */
470 if (!(fn
->fn_flags
& RTN_RTINFO
)) {
471 rt6_release(fn
->leaf
);
475 fn
->fn_sernum
= sernum
;
481 * We have more bits to go
484 /* Try to walk down on tree. */
485 fn
->fn_sernum
= sernum
;
486 dir
= addr_bit_set(addr
, fn
->fn_bit
);
488 fn
= dir
? fn
->right
: fn
->left
;
492 /* We should not create new node because
493 * NLM_F_REPLACE was specified without NLM_F_CREATE
494 * I assume it is safe to require NLM_F_CREATE when
495 * REPLACE flag is used! Later we may want to remove the
496 * check for replace_required, because according
497 * to netlink specification, NLM_F_CREATE
498 * MUST be specified if new route is created.
499 * That would keep IPv6 consistent with IPv4
501 if (replace_required
) {
502 pr_warn("Can't replace route, no match found\n");
503 return ERR_PTR(-ENOENT
);
505 pr_warn("NLM_F_CREATE should be set when creating new route\n");
508 * We walked to the bottom of tree.
509 * Create new leaf node without children.
515 return ERR_PTR(-ENOMEM
);
519 ln
->fn_sernum
= sernum
;
531 * split since we don't have a common prefix anymore or
532 * we have a less significant route.
533 * we've to insert an intermediate node on the list
534 * this new node will point to the one we need to create
540 /* find 1st bit in difference between the 2 addrs.
542 See comment in __ipv6_addr_diff: bit may be an invalid value,
543 but if it is >= plen, the value is ignored in any case.
546 bit
= __ipv6_addr_diff(addr
, &key
->addr
, addrlen
);
551 * (new leaf node)[ln] (old node)[fn]
562 return ERR_PTR(-ENOMEM
);
566 * new intermediate node.
568 * be off since that an address that chooses one of
569 * the branches would not match less specific routes
570 * in the other branch
577 atomic_inc(&in
->leaf
->rt6i_ref
);
579 in
->fn_sernum
= sernum
;
581 /* update parent pointer */
592 ln
->fn_sernum
= sernum
;
594 if (addr_bit_set(addr
, bit
)) {
601 } else { /* plen <= bit */
604 * (new leaf node)[ln]
606 * (old node)[fn] NULL
612 return ERR_PTR(-ENOMEM
);
618 ln
->fn_sernum
= sernum
;
625 if (addr_bit_set(&key
->addr
, plen
))
635 static inline bool rt6_qualify_for_ecmp(struct rt6_info
*rt
)
637 return (rt
->rt6i_flags
& (RTF_GATEWAY
|RTF_ADDRCONF
|RTF_DYNAMIC
)) ==
641 static void fib6_purge_rt(struct rt6_info
*rt
, struct fib6_node
*fn
,
644 if (atomic_read(&rt
->rt6i_ref
) != 1) {
645 /* This route is used as dummy address holder in some split
646 * nodes. It is not leaked, but it still holds other resources,
647 * which must be released in time. So, scan ascendant nodes
648 * and replace dummy references to this route with references
649 * to still alive ones.
652 if (!(fn
->fn_flags
& RTN_RTINFO
) && fn
->leaf
== rt
) {
653 fn
->leaf
= fib6_find_prefix(net
, fn
);
654 atomic_inc(&fn
->leaf
->rt6i_ref
);
659 /* No more references are possible at this point. */
660 BUG_ON(atomic_read(&rt
->rt6i_ref
) != 1);
665 * Insert routing information in a node.
668 static int fib6_add_rt2node(struct fib6_node
*fn
, struct rt6_info
*rt
,
669 struct nl_info
*info
)
671 struct rt6_info
*iter
= NULL
;
672 struct rt6_info
**ins
;
673 int replace
= (info
->nlh
&&
674 (info
->nlh
->nlmsg_flags
& NLM_F_REPLACE
));
675 int add
= (!info
->nlh
||
676 (info
->nlh
->nlmsg_flags
& NLM_F_CREATE
));
678 bool rt_can_ecmp
= rt6_qualify_for_ecmp(rt
);
682 for (iter
= fn
->leaf
; iter
; iter
= iter
->dst
.rt6_next
) {
684 * Search for duplicates
687 if (iter
->rt6i_metric
== rt
->rt6i_metric
) {
689 * Same priority level
692 (info
->nlh
->nlmsg_flags
& NLM_F_EXCL
))
699 if (iter
->dst
.dev
== rt
->dst
.dev
&&
700 iter
->rt6i_idev
== rt
->rt6i_idev
&&
701 ipv6_addr_equal(&iter
->rt6i_gateway
,
702 &rt
->rt6i_gateway
)) {
703 if (rt
->rt6i_nsiblings
)
704 rt
->rt6i_nsiblings
= 0;
705 if (!(iter
->rt6i_flags
& RTF_EXPIRES
))
707 if (!(rt
->rt6i_flags
& RTF_EXPIRES
))
708 rt6_clean_expires(iter
);
710 rt6_set_expires(iter
, rt
->dst
.expires
);
713 /* If we have the same destination and the same metric,
714 * but not the same gateway, then the route we try to
715 * add is sibling to this route, increment our counter
716 * of siblings, and later we will add our route to the
718 * Only static routes (which don't have flag
719 * RTF_EXPIRES) are used for ECMPv6.
721 * To avoid long list, we only had siblings if the
722 * route have a gateway.
725 rt6_qualify_for_ecmp(iter
))
726 rt
->rt6i_nsiblings
++;
729 if (iter
->rt6i_metric
> rt
->rt6i_metric
)
732 ins
= &iter
->dst
.rt6_next
;
735 /* Reset round-robin state, if necessary */
736 if (ins
== &fn
->leaf
)
739 /* Link this route to others same route. */
740 if (rt
->rt6i_nsiblings
) {
741 unsigned int rt6i_nsiblings
;
742 struct rt6_info
*sibling
, *temp_sibling
;
744 /* Find the first route that have the same metric */
747 if (sibling
->rt6i_metric
== rt
->rt6i_metric
&&
748 rt6_qualify_for_ecmp(sibling
)) {
749 list_add_tail(&rt
->rt6i_siblings
,
750 &sibling
->rt6i_siblings
);
753 sibling
= sibling
->dst
.rt6_next
;
755 /* For each sibling in the list, increment the counter of
756 * siblings. BUG() if counters does not match, list of siblings
760 list_for_each_entry_safe(sibling
, temp_sibling
,
761 &rt
->rt6i_siblings
, rt6i_siblings
) {
762 sibling
->rt6i_nsiblings
++;
763 BUG_ON(sibling
->rt6i_nsiblings
!= rt
->rt6i_nsiblings
);
766 BUG_ON(rt6i_nsiblings
!= rt
->rt6i_nsiblings
);
774 pr_warn("NLM_F_CREATE should be set when creating new route\n");
777 rt
->dst
.rt6_next
= iter
;
780 atomic_inc(&rt
->rt6i_ref
);
781 inet6_rt_notify(RTM_NEWROUTE
, rt
, info
);
782 info
->nl_net
->ipv6
.rt6_stats
->fib_rt_entries
++;
784 if (!(fn
->fn_flags
& RTN_RTINFO
)) {
785 info
->nl_net
->ipv6
.rt6_stats
->fib_route_nodes
++;
786 fn
->fn_flags
|= RTN_RTINFO
;
793 pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
798 rt
->dst
.rt6_next
= iter
->dst
.rt6_next
;
799 atomic_inc(&rt
->rt6i_ref
);
800 inet6_rt_notify(RTM_NEWROUTE
, rt
, info
);
801 if (!(fn
->fn_flags
& RTN_RTINFO
)) {
802 info
->nl_net
->ipv6
.rt6_stats
->fib_route_nodes
++;
803 fn
->fn_flags
|= RTN_RTINFO
;
805 fib6_purge_rt(iter
, fn
, info
->nl_net
);
812 static __inline__
void fib6_start_gc(struct net
*net
, struct rt6_info
*rt
)
814 if (!timer_pending(&net
->ipv6
.ip6_fib_timer
) &&
815 (rt
->rt6i_flags
& (RTF_EXPIRES
| RTF_CACHE
)))
816 mod_timer(&net
->ipv6
.ip6_fib_timer
,
817 jiffies
+ net
->ipv6
.sysctl
.ip6_rt_gc_interval
);
820 void fib6_force_start_gc(struct net
*net
)
822 if (!timer_pending(&net
->ipv6
.ip6_fib_timer
))
823 mod_timer(&net
->ipv6
.ip6_fib_timer
,
824 jiffies
+ net
->ipv6
.sysctl
.ip6_rt_gc_interval
);
828 * Add routing information to the routing tree.
829 * <destination addr>/<source addr>
830 * with source addr info in sub-trees
833 int fib6_add(struct fib6_node
*root
, struct rt6_info
*rt
, struct nl_info
*info
)
835 struct fib6_node
*fn
, *pn
= NULL
;
837 int allow_create
= 1;
838 int replace_required
= 0;
841 if (!(info
->nlh
->nlmsg_flags
& NLM_F_CREATE
))
843 if (info
->nlh
->nlmsg_flags
& NLM_F_REPLACE
)
844 replace_required
= 1;
846 if (!allow_create
&& !replace_required
)
847 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
849 fn
= fib6_add_1(root
, &rt
->rt6i_dst
.addr
, sizeof(struct in6_addr
),
850 rt
->rt6i_dst
.plen
, offsetof(struct rt6_info
, rt6i_dst
),
851 allow_create
, replace_required
);
860 #ifdef CONFIG_IPV6_SUBTREES
861 if (rt
->rt6i_src
.plen
) {
862 struct fib6_node
*sn
;
865 struct fib6_node
*sfn
;
877 /* Create subtree root node */
882 sfn
->leaf
= info
->nl_net
->ipv6
.ip6_null_entry
;
883 atomic_inc(&info
->nl_net
->ipv6
.ip6_null_entry
->rt6i_ref
);
884 sfn
->fn_flags
= RTN_ROOT
;
885 sfn
->fn_sernum
= fib6_new_sernum();
887 /* Now add the first leaf node to new subtree */
889 sn
= fib6_add_1(sfn
, &rt
->rt6i_src
.addr
,
890 sizeof(struct in6_addr
), rt
->rt6i_src
.plen
,
891 offsetof(struct rt6_info
, rt6i_src
),
892 allow_create
, replace_required
);
895 /* If it is failed, discard just allocated
896 root, and then (in st_failure) stale node
904 /* Now link new subtree to main tree */
908 sn
= fib6_add_1(fn
->subtree
, &rt
->rt6i_src
.addr
,
909 sizeof(struct in6_addr
), rt
->rt6i_src
.plen
,
910 offsetof(struct rt6_info
, rt6i_src
),
911 allow_create
, replace_required
);
921 atomic_inc(&rt
->rt6i_ref
);
927 err
= fib6_add_rt2node(fn
, rt
, info
);
929 fib6_start_gc(info
->nl_net
, rt
);
930 if (!(rt
->rt6i_flags
& RTF_CACHE
))
931 fib6_prune_clones(info
->nl_net
, pn
, rt
);
936 #ifdef CONFIG_IPV6_SUBTREES
938 * If fib6_add_1 has cleared the old leaf pointer in the
939 * super-tree leaf node we have to find a new one for it.
941 if (pn
!= fn
&& pn
->leaf
== rt
) {
943 atomic_dec(&rt
->rt6i_ref
);
945 if (pn
!= fn
&& !pn
->leaf
&& !(pn
->fn_flags
& RTN_RTINFO
)) {
946 pn
->leaf
= fib6_find_prefix(info
->nl_net
, pn
);
949 WARN_ON(pn
->leaf
== NULL
);
950 pn
->leaf
= info
->nl_net
->ipv6
.ip6_null_entry
;
953 atomic_inc(&pn
->leaf
->rt6i_ref
);
960 #ifdef CONFIG_IPV6_SUBTREES
961 /* Subtree creation failed, probably main tree node
962 is orphan. If it is, shoot it.
965 if (fn
&& !(fn
->fn_flags
& (RTN_RTINFO
|RTN_ROOT
)))
966 fib6_repair_tree(info
->nl_net
, fn
);
973 * Routing tree lookup
978 int offset
; /* key offset on rt6_info */
979 const struct in6_addr
*addr
; /* search key */
982 static struct fib6_node
* fib6_lookup_1(struct fib6_node
*root
,
983 struct lookup_args
*args
)
985 struct fib6_node
*fn
;
988 if (unlikely(args
->offset
== 0))
998 struct fib6_node
*next
;
1000 dir
= addr_bit_set(args
->addr
, fn
->fn_bit
);
1002 next
= dir
? fn
->right
: fn
->left
;
1012 if (FIB6_SUBTREE(fn
) || fn
->fn_flags
& RTN_RTINFO
) {
1015 key
= (struct rt6key
*) ((u8
*) fn
->leaf
+
1018 if (ipv6_prefix_equal(&key
->addr
, args
->addr
, key
->plen
)) {
1019 #ifdef CONFIG_IPV6_SUBTREES
1021 struct fib6_node
*sfn
;
1022 sfn
= fib6_lookup_1(fn
->subtree
,
1029 if (fn
->fn_flags
& RTN_RTINFO
)
1033 #ifdef CONFIG_IPV6_SUBTREES
1036 if (fn
->fn_flags
& RTN_ROOT
)
1045 struct fib6_node
* fib6_lookup(struct fib6_node
*root
, const struct in6_addr
*daddr
,
1046 const struct in6_addr
*saddr
)
1048 struct fib6_node
*fn
;
1049 struct lookup_args args
[] = {
1051 .offset
= offsetof(struct rt6_info
, rt6i_dst
),
1054 #ifdef CONFIG_IPV6_SUBTREES
1056 .offset
= offsetof(struct rt6_info
, rt6i_src
),
1061 .offset
= 0, /* sentinel */
1065 fn
= fib6_lookup_1(root
, daddr
? args
: args
+ 1);
1066 if (!fn
|| fn
->fn_flags
& RTN_TL_ROOT
)
1073 * Get node with specified destination prefix (and source prefix,
1074 * if subtrees are used)
1078 static struct fib6_node
* fib6_locate_1(struct fib6_node
*root
,
1079 const struct in6_addr
*addr
,
1080 int plen
, int offset
)
1082 struct fib6_node
*fn
;
1084 for (fn
= root
; fn
; ) {
1085 struct rt6key
*key
= (struct rt6key
*)((u8
*)fn
->leaf
+ offset
);
1090 if (plen
< fn
->fn_bit
||
1091 !ipv6_prefix_equal(&key
->addr
, addr
, fn
->fn_bit
))
1094 if (plen
== fn
->fn_bit
)
1098 * We have more bits to go
1100 if (addr_bit_set(addr
, fn
->fn_bit
))
1108 struct fib6_node
* fib6_locate(struct fib6_node
*root
,
1109 const struct in6_addr
*daddr
, int dst_len
,
1110 const struct in6_addr
*saddr
, int src_len
)
1112 struct fib6_node
*fn
;
1114 fn
= fib6_locate_1(root
, daddr
, dst_len
,
1115 offsetof(struct rt6_info
, rt6i_dst
));
1117 #ifdef CONFIG_IPV6_SUBTREES
1119 WARN_ON(saddr
== NULL
);
1120 if (fn
&& fn
->subtree
)
1121 fn
= fib6_locate_1(fn
->subtree
, saddr
, src_len
,
1122 offsetof(struct rt6_info
, rt6i_src
));
1126 if (fn
&& fn
->fn_flags
& RTN_RTINFO
)
1138 static struct rt6_info
*fib6_find_prefix(struct net
*net
, struct fib6_node
*fn
)
1140 if (fn
->fn_flags
& RTN_ROOT
)
1141 return net
->ipv6
.ip6_null_entry
;
1145 return fn
->left
->leaf
;
1147 return fn
->right
->leaf
;
1149 fn
= FIB6_SUBTREE(fn
);
1155 * Called to trim the tree of intermediate nodes when possible. "fn"
1156 * is the node we want to try and remove.
1159 static struct fib6_node
*fib6_repair_tree(struct net
*net
,
1160 struct fib6_node
*fn
)
1164 struct fib6_node
*child
, *pn
;
1165 struct fib6_walker_t
*w
;
1169 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn
->fn_bit
, iter
);
1172 WARN_ON(fn
->fn_flags
& RTN_RTINFO
);
1173 WARN_ON(fn
->fn_flags
& RTN_TL_ROOT
);
1174 WARN_ON(fn
->leaf
!= NULL
);
1178 if (fn
->right
) child
= fn
->right
, children
|= 1;
1179 if (fn
->left
) child
= fn
->left
, children
|= 2;
1181 if (children
== 3 || FIB6_SUBTREE(fn
)
1182 #ifdef CONFIG_IPV6_SUBTREES
1183 /* Subtree root (i.e. fn) may have one child */
1184 || (children
&& fn
->fn_flags
& RTN_ROOT
)
1187 fn
->leaf
= fib6_find_prefix(net
, fn
);
1191 fn
->leaf
= net
->ipv6
.ip6_null_entry
;
1194 atomic_inc(&fn
->leaf
->rt6i_ref
);
1199 #ifdef CONFIG_IPV6_SUBTREES
1200 if (FIB6_SUBTREE(pn
) == fn
) {
1201 WARN_ON(!(fn
->fn_flags
& RTN_ROOT
));
1202 FIB6_SUBTREE(pn
) = NULL
;
1205 WARN_ON(fn
->fn_flags
& RTN_ROOT
);
1207 if (pn
->right
== fn
) pn
->right
= child
;
1208 else if (pn
->left
== fn
) pn
->left
= child
;
1216 #ifdef CONFIG_IPV6_SUBTREES
1220 read_lock(&fib6_walker_lock
);
1223 if (w
->root
== fn
) {
1224 w
->root
= w
->node
= NULL
;
1225 RT6_TRACE("W %p adjusted by delroot 1\n", w
);
1226 } else if (w
->node
== fn
) {
1227 RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w
, w
->state
, nstate
);
1232 if (w
->root
== fn
) {
1234 RT6_TRACE("W %p adjusted by delroot 2\n", w
);
1236 if (w
->node
== fn
) {
1239 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w
, w
->state
);
1240 w
->state
= w
->state
>=FWS_R
? FWS_U
: FWS_INIT
;
1242 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w
, w
->state
);
1243 w
->state
= w
->state
>=FWS_C
? FWS_U
: FWS_INIT
;
1248 read_unlock(&fib6_walker_lock
);
1251 if (pn
->fn_flags
& RTN_RTINFO
|| FIB6_SUBTREE(pn
))
1254 rt6_release(pn
->leaf
);
1260 static void fib6_del_route(struct fib6_node
*fn
, struct rt6_info
**rtp
,
1261 struct nl_info
*info
)
1263 struct fib6_walker_t
*w
;
1264 struct rt6_info
*rt
= *rtp
;
1265 struct net
*net
= info
->nl_net
;
1267 RT6_TRACE("fib6_del_route\n");
1270 *rtp
= rt
->dst
.rt6_next
;
1271 rt
->rt6i_node
= NULL
;
1272 net
->ipv6
.rt6_stats
->fib_rt_entries
--;
1273 net
->ipv6
.rt6_stats
->fib_discarded_routes
++;
1275 /* Reset round-robin state, if necessary */
1276 if (fn
->rr_ptr
== rt
)
1279 /* Remove this entry from other siblings */
1280 if (rt
->rt6i_nsiblings
) {
1281 struct rt6_info
*sibling
, *next_sibling
;
1283 list_for_each_entry_safe(sibling
, next_sibling
,
1284 &rt
->rt6i_siblings
, rt6i_siblings
)
1285 sibling
->rt6i_nsiblings
--;
1286 rt
->rt6i_nsiblings
= 0;
1287 list_del_init(&rt
->rt6i_siblings
);
1290 /* Adjust walkers */
1291 read_lock(&fib6_walker_lock
);
1293 if (w
->state
== FWS_C
&& w
->leaf
== rt
) {
1294 RT6_TRACE("walker %p adjusted by delroute\n", w
);
1295 w
->leaf
= rt
->dst
.rt6_next
;
1300 read_unlock(&fib6_walker_lock
);
1302 rt
->dst
.rt6_next
= NULL
;
1304 /* If it was last route, expunge its radix tree node */
1306 fn
->fn_flags
&= ~RTN_RTINFO
;
1307 net
->ipv6
.rt6_stats
->fib_route_nodes
--;
1308 fn
= fib6_repair_tree(net
, fn
);
1311 fib6_purge_rt(rt
, fn
, net
);
1313 inet6_rt_notify(RTM_DELROUTE
, rt
, info
);
1317 int fib6_del(struct rt6_info
*rt
, struct nl_info
*info
)
1319 struct net
*net
= info
->nl_net
;
1320 struct fib6_node
*fn
= rt
->rt6i_node
;
1321 struct rt6_info
**rtp
;
1324 if (rt
->dst
.obsolete
>0) {
1325 WARN_ON(fn
!= NULL
);
1329 if (!fn
|| rt
== net
->ipv6
.ip6_null_entry
)
1332 WARN_ON(!(fn
->fn_flags
& RTN_RTINFO
));
1334 if (!(rt
->rt6i_flags
& RTF_CACHE
)) {
1335 struct fib6_node
*pn
= fn
;
1336 #ifdef CONFIG_IPV6_SUBTREES
1337 /* clones of this route might be in another subtree */
1338 if (rt
->rt6i_src
.plen
) {
1339 while (!(pn
->fn_flags
& RTN_ROOT
))
1344 fib6_prune_clones(info
->nl_net
, pn
, rt
);
1348 * Walk the leaf entries looking for ourself
1351 for (rtp
= &fn
->leaf
; *rtp
; rtp
= &(*rtp
)->dst
.rt6_next
) {
1353 fib6_del_route(fn
, rtp
, info
);
1361 * Tree traversal function.
1363 * Certainly, it is not interrupt safe.
1364 * However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1365 * It means, that we can modify tree during walking
1366 * and use this function for garbage collection, clone pruning,
1367 * cleaning tree when a device goes down etc. etc.
1369 * It guarantees that every node will be traversed,
1370 * and that it will be traversed only once.
1372 * Callback function w->func may return:
1373 * 0 -> continue walking.
1374 * positive value -> walking is suspended (used by tree dumps,
1375 * and probably by gc, if it will be split to several slices)
1376 * negative value -> terminate walking.
1378 * The function itself returns:
1379 * 0 -> walk is complete.
1380 * >0 -> walk is incomplete (i.e. suspended)
1381 * <0 -> walk is terminated by an error.
1384 static int fib6_walk_continue(struct fib6_walker_t
*w
)
1386 struct fib6_node
*fn
, *pn
;
1393 if (w
->prune
&& fn
!= w
->root
&&
1394 fn
->fn_flags
& RTN_RTINFO
&& w
->state
< FWS_C
) {
1399 #ifdef CONFIG_IPV6_SUBTREES
1401 if (FIB6_SUBTREE(fn
)) {
1402 w
->node
= FIB6_SUBTREE(fn
);
1410 w
->state
= FWS_INIT
;
1416 w
->node
= fn
->right
;
1417 w
->state
= FWS_INIT
;
1423 if (w
->leaf
&& fn
->fn_flags
& RTN_RTINFO
) {
1445 #ifdef CONFIG_IPV6_SUBTREES
1446 if (FIB6_SUBTREE(pn
) == fn
) {
1447 WARN_ON(!(fn
->fn_flags
& RTN_ROOT
));
1452 if (pn
->left
== fn
) {
1456 if (pn
->right
== fn
) {
1458 w
->leaf
= w
->node
->leaf
;
1468 static int fib6_walk(struct fib6_walker_t
*w
)
1472 w
->state
= FWS_INIT
;
1475 fib6_walker_link(w
);
1476 res
= fib6_walk_continue(w
);
1478 fib6_walker_unlink(w
);
1482 static int fib6_clean_node(struct fib6_walker_t
*w
)
1485 struct rt6_info
*rt
;
1486 struct fib6_cleaner_t
*c
= container_of(w
, struct fib6_cleaner_t
, w
);
1487 struct nl_info info
= {
1491 for (rt
= w
->leaf
; rt
; rt
= rt
->dst
.rt6_next
) {
1492 res
= c
->func(rt
, c
->arg
);
1495 res
= fib6_del(rt
, &info
);
1498 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1499 __func__
, rt
, rt
->rt6i_node
, res
);
1512 * Convenient frontend to tree walker.
1514 * func is called on each route.
1515 * It may return -1 -> delete this route.
1516 * 0 -> continue walking
1518 * prune==1 -> only immediate children of node (certainly,
1519 * ignoring pure split nodes) will be scanned.
1522 static void fib6_clean_tree(struct net
*net
, struct fib6_node
*root
,
1523 int (*func
)(struct rt6_info
*, void *arg
),
1524 int prune
, void *arg
)
1526 struct fib6_cleaner_t c
;
1529 c
.w
.func
= fib6_clean_node
;
1540 void fib6_clean_all_ro(struct net
*net
, int (*func
)(struct rt6_info
*, void *arg
),
1541 int prune
, void *arg
)
1543 struct fib6_table
*table
;
1544 struct hlist_head
*head
;
1548 for (h
= 0; h
< FIB6_TABLE_HASHSZ
; h
++) {
1549 head
= &net
->ipv6
.fib_table_hash
[h
];
1550 hlist_for_each_entry_rcu(table
, head
, tb6_hlist
) {
1551 read_lock_bh(&table
->tb6_lock
);
1552 fib6_clean_tree(net
, &table
->tb6_root
,
1554 read_unlock_bh(&table
->tb6_lock
);
1559 void fib6_clean_all(struct net
*net
, int (*func
)(struct rt6_info
*, void *arg
),
1560 int prune
, void *arg
)
1562 struct fib6_table
*table
;
1563 struct hlist_head
*head
;
1567 for (h
= 0; h
< FIB6_TABLE_HASHSZ
; h
++) {
1568 head
= &net
->ipv6
.fib_table_hash
[h
];
1569 hlist_for_each_entry_rcu(table
, head
, tb6_hlist
) {
1570 write_lock_bh(&table
->tb6_lock
);
1571 fib6_clean_tree(net
, &table
->tb6_root
,
1573 write_unlock_bh(&table
->tb6_lock
);
1579 static int fib6_prune_clone(struct rt6_info
*rt
, void *arg
)
1581 if (rt
->rt6i_flags
& RTF_CACHE
) {
1582 RT6_TRACE("pruning clone %p\n", rt
);
1589 static void fib6_prune_clones(struct net
*net
, struct fib6_node
*fn
,
1590 struct rt6_info
*rt
)
1592 fib6_clean_tree(net
, fn
, fib6_prune_clone
, 1, rt
);
1596 * Garbage collection
1599 static struct fib6_gc_args
1605 static int fib6_age(struct rt6_info
*rt
, void *arg
)
1607 unsigned long now
= jiffies
;
1610 * check addrconf expiration here.
1611 * Routes are expired even if they are in use.
1613 * Also age clones. Note, that clones are aged out
1614 * only if they are not in use now.
1617 if (rt
->rt6i_flags
& RTF_EXPIRES
&& rt
->dst
.expires
) {
1618 if (time_after(now
, rt
->dst
.expires
)) {
1619 RT6_TRACE("expiring %p\n", rt
);
1623 } else if (rt
->rt6i_flags
& RTF_CACHE
) {
1624 if (atomic_read(&rt
->dst
.__refcnt
) == 0 &&
1625 time_after_eq(now
, rt
->dst
.lastuse
+ gc_args
.timeout
)) {
1626 RT6_TRACE("aging clone %p\n", rt
);
1628 } else if (rt
->rt6i_flags
& RTF_GATEWAY
) {
1629 struct neighbour
*neigh
;
1630 __u8 neigh_flags
= 0;
1632 neigh
= dst_neigh_lookup(&rt
->dst
, &rt
->rt6i_gateway
);
1634 neigh_flags
= neigh
->flags
;
1635 neigh_release(neigh
);
1637 if (!(neigh_flags
& NTF_ROUTER
)) {
1638 RT6_TRACE("purging route %p via non-router but gateway\n",
1649 static DEFINE_SPINLOCK(fib6_gc_lock
);
1651 void fib6_run_gc(unsigned long expires
, struct net
*net
)
1653 if (expires
!= ~0UL) {
1654 spin_lock_bh(&fib6_gc_lock
);
1655 gc_args
.timeout
= expires
? (int)expires
:
1656 net
->ipv6
.sysctl
.ip6_rt_gc_interval
;
1658 if (!spin_trylock_bh(&fib6_gc_lock
)) {
1659 mod_timer(&net
->ipv6
.ip6_fib_timer
, jiffies
+ HZ
);
1662 gc_args
.timeout
= net
->ipv6
.sysctl
.ip6_rt_gc_interval
;
1665 gc_args
.more
= icmp6_dst_gc();
1667 fib6_clean_all(net
, fib6_age
, 0, NULL
);
1670 mod_timer(&net
->ipv6
.ip6_fib_timer
,
1671 round_jiffies(jiffies
1672 + net
->ipv6
.sysctl
.ip6_rt_gc_interval
));
1674 del_timer(&net
->ipv6
.ip6_fib_timer
);
1675 spin_unlock_bh(&fib6_gc_lock
);
1678 static void fib6_gc_timer_cb(unsigned long arg
)
1680 fib6_run_gc(0, (struct net
*)arg
);
1683 static int __net_init
fib6_net_init(struct net
*net
)
1685 size_t size
= sizeof(struct hlist_head
) * FIB6_TABLE_HASHSZ
;
1687 setup_timer(&net
->ipv6
.ip6_fib_timer
, fib6_gc_timer_cb
, (unsigned long)net
);
1689 net
->ipv6
.rt6_stats
= kzalloc(sizeof(*net
->ipv6
.rt6_stats
), GFP_KERNEL
);
1690 if (!net
->ipv6
.rt6_stats
)
1693 /* Avoid false sharing : Use at least a full cache line */
1694 size
= max_t(size_t, size
, L1_CACHE_BYTES
);
1696 net
->ipv6
.fib_table_hash
= kzalloc(size
, GFP_KERNEL
);
1697 if (!net
->ipv6
.fib_table_hash
)
1700 net
->ipv6
.fib6_main_tbl
= kzalloc(sizeof(*net
->ipv6
.fib6_main_tbl
),
1702 if (!net
->ipv6
.fib6_main_tbl
)
1703 goto out_fib_table_hash
;
1705 net
->ipv6
.fib6_main_tbl
->tb6_id
= RT6_TABLE_MAIN
;
1706 net
->ipv6
.fib6_main_tbl
->tb6_root
.leaf
= net
->ipv6
.ip6_null_entry
;
1707 net
->ipv6
.fib6_main_tbl
->tb6_root
.fn_flags
=
1708 RTN_ROOT
| RTN_TL_ROOT
| RTN_RTINFO
;
1709 inet_peer_base_init(&net
->ipv6
.fib6_main_tbl
->tb6_peers
);
1711 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1712 net
->ipv6
.fib6_local_tbl
= kzalloc(sizeof(*net
->ipv6
.fib6_local_tbl
),
1714 if (!net
->ipv6
.fib6_local_tbl
)
1715 goto out_fib6_main_tbl
;
1716 net
->ipv6
.fib6_local_tbl
->tb6_id
= RT6_TABLE_LOCAL
;
1717 net
->ipv6
.fib6_local_tbl
->tb6_root
.leaf
= net
->ipv6
.ip6_null_entry
;
1718 net
->ipv6
.fib6_local_tbl
->tb6_root
.fn_flags
=
1719 RTN_ROOT
| RTN_TL_ROOT
| RTN_RTINFO
;
1720 inet_peer_base_init(&net
->ipv6
.fib6_local_tbl
->tb6_peers
);
1722 fib6_tables_init(net
);
1726 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1728 kfree(net
->ipv6
.fib6_main_tbl
);
1731 kfree(net
->ipv6
.fib_table_hash
);
1733 kfree(net
->ipv6
.rt6_stats
);
1738 static void fib6_net_exit(struct net
*net
)
1740 rt6_ifdown(net
, NULL
);
1741 del_timer_sync(&net
->ipv6
.ip6_fib_timer
);
1743 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1744 inetpeer_invalidate_tree(&net
->ipv6
.fib6_local_tbl
->tb6_peers
);
1745 kfree(net
->ipv6
.fib6_local_tbl
);
1747 inetpeer_invalidate_tree(&net
->ipv6
.fib6_main_tbl
->tb6_peers
);
1748 kfree(net
->ipv6
.fib6_main_tbl
);
1749 kfree(net
->ipv6
.fib_table_hash
);
1750 kfree(net
->ipv6
.rt6_stats
);
1753 static struct pernet_operations fib6_net_ops
= {
1754 .init
= fib6_net_init
,
1755 .exit
= fib6_net_exit
,
1758 int __init
fib6_init(void)
1762 fib6_node_kmem
= kmem_cache_create("fib6_nodes",
1763 sizeof(struct fib6_node
),
1764 0, SLAB_HWCACHE_ALIGN
,
1766 if (!fib6_node_kmem
)
1769 ret
= register_pernet_subsys(&fib6_net_ops
);
1771 goto out_kmem_cache_create
;
1773 ret
= __rtnl_register(PF_INET6
, RTM_GETROUTE
, NULL
, inet6_dump_fib
,
1776 goto out_unregister_subsys
;
1780 out_unregister_subsys
:
1781 unregister_pernet_subsys(&fib6_net_ops
);
1782 out_kmem_cache_create
:
1783 kmem_cache_destroy(fib6_node_kmem
);
1787 void fib6_gc_cleanup(void)
1789 unregister_pernet_subsys(&fib6_net_ops
);
1790 kmem_cache_destroy(fib6_node_kmem
);