2 * This program is free software; you can redistribute it and/or
3 * modify it under the terms of the GNU General Public License
4 * as published by the Free Software Foundation; either version
5 * 2 of the License, or (at your option) any later version.
7 * Robert Olsson <robert.olsson@its.uu.se> Uppsala Universitet
8 * & Swedish University of Agricultural Sciences.
10 * Jens Laas <jens.laas@data.slu.se> Swedish University of
11 * Agricultural Sciences.
13 * Hans Liss <hans.liss@its.uu.se> Uppsala Universitet
15 * This work is based on the LPC-trie which is originally described in:
17 * An experimental study of compression methods for dynamic tries
18 * Stefan Nilsson and Matti Tikkanen. Algorithmica, 33(1):19-33, 2002.
19 * http://www.csc.kth.se/~snilsson/software/dyntrie2/
22 * IP-address lookup using LC-tries. Stefan Nilsson and Gunnar Karlsson
23 * IEEE Journal on Selected Areas in Communications, 17(6):1083-1092, June 1999
26 * Code from fib_hash has been reused which includes the following header:
29 * INET An implementation of the TCP/IP protocol suite for the LINUX
30 * operating system. INET is implemented using the BSD Socket
31 * interface as the means of communication with the user level.
33 * IPv4 FIB: lookup engine and maintenance routines.
36 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
38 * This program is free software; you can redistribute it and/or
39 * modify it under the terms of the GNU General Public License
40 * as published by the Free Software Foundation; either version
41 * 2 of the License, or (at your option) any later version.
43 * Substantial contributions to this work comes from:
45 * David S. Miller, <davem@davemloft.net>
46 * Stephen Hemminger <shemminger@osdl.org>
47 * Paul E. McKenney <paulmck@us.ibm.com>
48 * Patrick McHardy <kaber@trash.net>
51 #define VERSION "0.409"
53 #include <asm/uaccess.h>
54 #include <linux/bitops.h>
55 #include <linux/types.h>
56 #include <linux/kernel.h>
58 #include <linux/string.h>
59 #include <linux/socket.h>
60 #include <linux/sockios.h>
61 #include <linux/errno.h>
63 #include <linux/inet.h>
64 #include <linux/inetdevice.h>
65 #include <linux/netdevice.h>
66 #include <linux/if_arp.h>
67 #include <linux/proc_fs.h>
68 #include <linux/rcupdate.h>
69 #include <linux/skbuff.h>
70 #include <linux/netlink.h>
71 #include <linux/init.h>
72 #include <linux/list.h>
73 #include <linux/slab.h>
74 #include <linux/export.h>
75 #include <net/net_namespace.h>
77 #include <net/protocol.h>
78 #include <net/route.h>
81 #include <net/ip_fib.h>
82 #include "fib_lookup.h"
84 #define MAX_STAT_DEPTH 32
86 #define KEYLENGTH (8*sizeof(t_key))
88 typedef unsigned int t_key
;
92 #define NODE_TYPE_MASK 0x1UL
93 #define NODE_TYPE(node) ((node)->parent & NODE_TYPE_MASK)
95 #define IS_TNODE(n) (!(n->parent & T_LEAF))
96 #define IS_LEAF(n) (n->parent & T_LEAF)
104 unsigned long parent
;
106 struct hlist_head list
;
111 struct hlist_node hlist
;
113 u32 mask_plen
; /* ntohl(inet_make_mask(plen)) */
114 struct list_head falh
;
119 unsigned long parent
;
121 unsigned char pos
; /* 2log(KEYLENGTH) bits needed */
122 unsigned char bits
; /* 2log(KEYLENGTH) bits needed */
123 unsigned int full_children
; /* KEYLENGTH bits needed */
124 unsigned int empty_children
; /* KEYLENGTH bits needed */
127 struct tnode
*tnode_free
;
129 struct rt_trie_node __rcu
*child
[0];
132 #ifdef CONFIG_IP_FIB_TRIE_STATS
133 struct trie_use_stats
{
135 unsigned int backtrack
;
136 unsigned int semantic_match_passed
;
137 unsigned int semantic_match_miss
;
138 unsigned int null_node_hit
;
139 unsigned int resize_node_skipped
;
144 unsigned int totdepth
;
145 unsigned int maxdepth
;
148 unsigned int nullpointers
;
149 unsigned int prefixes
;
150 unsigned int nodesizes
[MAX_STAT_DEPTH
];
154 struct rt_trie_node __rcu
*trie
;
155 #ifdef CONFIG_IP_FIB_TRIE_STATS
156 struct trie_use_stats stats
;
160 static void tnode_put_child_reorg(struct tnode
*tn
, int i
, struct rt_trie_node
*n
,
162 static struct rt_trie_node
*resize(struct trie
*t
, struct tnode
*tn
);
163 static struct tnode
*inflate(struct trie
*t
, struct tnode
*tn
);
164 static struct tnode
*halve(struct trie
*t
, struct tnode
*tn
);
165 /* tnodes to free after resize(); protected by RTNL */
166 static struct tnode
*tnode_free_head
;
167 static size_t tnode_free_size
;
170 * synchronize_rcu after call_rcu for that many pages; it should be especially
171 * useful before resizing the root node with PREEMPT_NONE configs; the value was
172 * obtained experimentally, aiming to avoid visible slowdown.
174 static const int sync_pages
= 128;
176 static struct kmem_cache
*fn_alias_kmem __read_mostly
;
177 static struct kmem_cache
*trie_leaf_kmem __read_mostly
;
180 * caller must hold RTNL
182 static inline struct tnode
*node_parent(const struct rt_trie_node
*node
)
184 unsigned long parent
;
186 parent
= rcu_dereference_index_check(node
->parent
, lockdep_rtnl_is_held());
188 return (struct tnode
*)(parent
& ~NODE_TYPE_MASK
);
192 * caller must hold RCU read lock or RTNL
194 static inline struct tnode
*node_parent_rcu(const struct rt_trie_node
*node
)
196 unsigned long parent
;
198 parent
= rcu_dereference_index_check(node
->parent
, rcu_read_lock_held() ||
199 lockdep_rtnl_is_held());
201 return (struct tnode
*)(parent
& ~NODE_TYPE_MASK
);
204 /* Same as rcu_assign_pointer
205 * but that macro() assumes that value is a pointer.
207 static inline void node_set_parent(struct rt_trie_node
*node
, struct tnode
*ptr
)
210 node
->parent
= (unsigned long)ptr
| NODE_TYPE(node
);
214 * caller must hold RTNL
216 static inline struct rt_trie_node
*tnode_get_child(const struct tnode
*tn
, unsigned int i
)
218 BUG_ON(i
>= 1U << tn
->bits
);
220 return rtnl_dereference(tn
->child
[i
]);
224 * caller must hold RCU read lock or RTNL
226 static inline struct rt_trie_node
*tnode_get_child_rcu(const struct tnode
*tn
, unsigned int i
)
228 BUG_ON(i
>= 1U << tn
->bits
);
230 return rcu_dereference_rtnl(tn
->child
[i
]);
233 static inline int tnode_child_length(const struct tnode
*tn
)
235 return 1 << tn
->bits
;
238 static inline t_key
mask_pfx(t_key k
, unsigned int l
)
240 return (l
== 0) ? 0 : k
>> (KEYLENGTH
-l
) << (KEYLENGTH
-l
);
243 static inline t_key
tkey_extract_bits(t_key a
, unsigned int offset
, unsigned int bits
)
245 if (offset
< KEYLENGTH
)
246 return ((t_key
)(a
<< offset
)) >> (KEYLENGTH
- bits
);
251 static inline int tkey_equals(t_key a
, t_key b
)
256 static inline int tkey_sub_equals(t_key a
, int offset
, int bits
, t_key b
)
258 if (bits
== 0 || offset
>= KEYLENGTH
)
260 bits
= bits
> KEYLENGTH
? KEYLENGTH
: bits
;
261 return ((a
^ b
) << offset
) >> (KEYLENGTH
- bits
) == 0;
264 static inline int tkey_mismatch(t_key a
, int offset
, t_key b
)
271 while ((diff
<< i
) >> (KEYLENGTH
-1) == 0)
277 To understand this stuff, an understanding of keys and all their bits is
278 necessary. Every node in the trie has a key associated with it, but not
279 all of the bits in that key are significant.
281 Consider a node 'n' and its parent 'tp'.
283 If n is a leaf, every bit in its key is significant. Its presence is
284 necessitated by path compression, since during a tree traversal (when
285 searching for a leaf - unless we are doing an insertion) we will completely
286 ignore all skipped bits we encounter. Thus we need to verify, at the end of
287 a potentially successful search, that we have indeed been walking the
290 Note that we can never "miss" the correct key in the tree if present by
291 following the wrong path. Path compression ensures that segments of the key
292 that are the same for all keys with a given prefix are skipped, but the
293 skipped part *is* identical for each node in the subtrie below the skipped
294 bit! trie_insert() in this implementation takes care of that - note the
295 call to tkey_sub_equals() in trie_insert().
297 if n is an internal node - a 'tnode' here, the various parts of its key
298 have many different meanings.
301 _________________________________________________________________
302 | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
303 -----------------------------------------------------------------
304 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
306 _________________________________________________________________
307 | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
308 -----------------------------------------------------------------
309 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
316 First, let's just ignore the bits that come before the parent tp, that is
317 the bits from 0 to (tp->pos-1). They are *known* but at this point we do
318 not use them for anything.
320 The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
321 index into the parent's child array. That is, they will be used to find
322 'n' among tp's children.
324 The bits from (tp->pos + tp->bits) to (n->pos - 1) - "S" - are skipped bits
327 All the bits we have seen so far are significant to the node n. The rest
328 of the bits are really not needed or indeed known in n->key.
330 The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
331 n's child array, and will of course be different for each child.
334 The rest of the bits, from (n->pos + n->bits) onward, are completely unknown
339 static inline void check_tnode(const struct tnode
*tn
)
341 WARN_ON(tn
&& tn
->pos
+tn
->bits
> 32);
344 static const int halve_threshold
= 25;
345 static const int inflate_threshold
= 50;
346 static const int halve_threshold_root
= 15;
347 static const int inflate_threshold_root
= 30;
349 static void __alias_free_mem(struct rcu_head
*head
)
351 struct fib_alias
*fa
= container_of(head
, struct fib_alias
, rcu
);
352 kmem_cache_free(fn_alias_kmem
, fa
);
355 static inline void alias_free_mem_rcu(struct fib_alias
*fa
)
357 call_rcu(&fa
->rcu
, __alias_free_mem
);
360 static void __leaf_free_rcu(struct rcu_head
*head
)
362 struct leaf
*l
= container_of(head
, struct leaf
, rcu
);
363 kmem_cache_free(trie_leaf_kmem
, l
);
366 static inline void free_leaf(struct leaf
*l
)
368 call_rcu(&l
->rcu
, __leaf_free_rcu
);
371 static inline void free_leaf_info(struct leaf_info
*leaf
)
373 kfree_rcu(leaf
, rcu
);
376 static struct tnode
*tnode_alloc(size_t size
)
378 if (size
<= PAGE_SIZE
)
379 return kzalloc(size
, GFP_KERNEL
);
381 return vzalloc(size
);
384 static void __tnode_free_rcu(struct rcu_head
*head
)
386 struct tnode
*tn
= container_of(head
, struct tnode
, rcu
);
387 size_t size
= sizeof(struct tnode
) +
388 (sizeof(struct rt_trie_node
*) << tn
->bits
);
390 if (size
<= PAGE_SIZE
)
396 static inline void tnode_free(struct tnode
*tn
)
399 free_leaf((struct leaf
*) tn
);
401 call_rcu(&tn
->rcu
, __tnode_free_rcu
);
404 static void tnode_free_safe(struct tnode
*tn
)
407 tn
->tnode_free
= tnode_free_head
;
408 tnode_free_head
= tn
;
409 tnode_free_size
+= sizeof(struct tnode
) +
410 (sizeof(struct rt_trie_node
*) << tn
->bits
);
413 static void tnode_free_flush(void)
417 while ((tn
= tnode_free_head
)) {
418 tnode_free_head
= tn
->tnode_free
;
419 tn
->tnode_free
= NULL
;
423 if (tnode_free_size
>= PAGE_SIZE
* sync_pages
) {
429 static struct leaf
*leaf_new(void)
431 struct leaf
*l
= kmem_cache_alloc(trie_leaf_kmem
, GFP_KERNEL
);
434 INIT_HLIST_HEAD(&l
->list
);
439 static struct leaf_info
*leaf_info_new(int plen
)
441 struct leaf_info
*li
= kmalloc(sizeof(struct leaf_info
), GFP_KERNEL
);
444 li
->mask_plen
= ntohl(inet_make_mask(plen
));
445 INIT_LIST_HEAD(&li
->falh
);
450 static struct tnode
*tnode_new(t_key key
, int pos
, int bits
)
452 size_t sz
= sizeof(struct tnode
) + (sizeof(struct rt_trie_node
*) << bits
);
453 struct tnode
*tn
= tnode_alloc(sz
);
456 tn
->parent
= T_TNODE
;
460 tn
->full_children
= 0;
461 tn
->empty_children
= 1<<bits
;
464 pr_debug("AT %p s=%zu %zu\n", tn
, sizeof(struct tnode
),
465 sizeof(struct rt_trie_node
*) << bits
);
470 * Check whether a tnode 'n' is "full", i.e. it is an internal node
471 * and no bits are skipped. See discussion in dyntree paper p. 6
474 static inline int tnode_full(const struct tnode
*tn
, const struct rt_trie_node
*n
)
476 if (n
== NULL
|| IS_LEAF(n
))
479 return ((struct tnode
*) n
)->pos
== tn
->pos
+ tn
->bits
;
482 static inline void put_child(struct tnode
*tn
, int i
,
483 struct rt_trie_node
*n
)
485 tnode_put_child_reorg(tn
, i
, n
, -1);
489 * Add a child at position i overwriting the old value.
490 * Update the value of full_children and empty_children.
493 static void tnode_put_child_reorg(struct tnode
*tn
, int i
, struct rt_trie_node
*n
,
496 struct rt_trie_node
*chi
= rtnl_dereference(tn
->child
[i
]);
499 BUG_ON(i
>= 1<<tn
->bits
);
501 /* update emptyChildren */
502 if (n
== NULL
&& chi
!= NULL
)
503 tn
->empty_children
++;
504 else if (n
!= NULL
&& chi
== NULL
)
505 tn
->empty_children
--;
507 /* update fullChildren */
509 wasfull
= tnode_full(tn
, chi
);
511 isfull
= tnode_full(tn
, n
);
512 if (wasfull
&& !isfull
)
514 else if (!wasfull
&& isfull
)
518 node_set_parent(n
, tn
);
520 rcu_assign_pointer(tn
->child
[i
], n
);
524 static struct rt_trie_node
*resize(struct trie
*t
, struct tnode
*tn
)
527 struct tnode
*old_tn
;
528 int inflate_threshold_use
;
529 int halve_threshold_use
;
535 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
536 tn
, inflate_threshold
, halve_threshold
);
539 if (tn
->empty_children
== tnode_child_length(tn
)) {
544 if (tn
->empty_children
== tnode_child_length(tn
) - 1)
547 * Double as long as the resulting node has a number of
548 * nonempty nodes that are above the threshold.
552 * From "Implementing a dynamic compressed trie" by Stefan Nilsson of
553 * the Helsinki University of Technology and Matti Tikkanen of Nokia
554 * Telecommunications, page 6:
555 * "A node is doubled if the ratio of non-empty children to all
556 * children in the *doubled* node is at least 'high'."
558 * 'high' in this instance is the variable 'inflate_threshold'. It
559 * is expressed as a percentage, so we multiply it with
560 * tnode_child_length() and instead of multiplying by 2 (since the
561 * child array will be doubled by inflate()) and multiplying
562 * the left-hand side by 100 (to handle the percentage thing) we
563 * multiply the left-hand side by 50.
565 * The left-hand side may look a bit weird: tnode_child_length(tn)
566 * - tn->empty_children is of course the number of non-null children
567 * in the current node. tn->full_children is the number of "full"
568 * children, that is non-null tnodes with a skip value of 0.
569 * All of those will be doubled in the resulting inflated tnode, so
570 * we just count them one extra time here.
572 * A clearer way to write this would be:
574 * to_be_doubled = tn->full_children;
575 * not_to_be_doubled = tnode_child_length(tn) - tn->empty_children -
578 * new_child_length = tnode_child_length(tn) * 2;
580 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
582 * if (new_fill_factor >= inflate_threshold)
584 * ...and so on, tho it would mess up the while () loop.
587 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
591 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
592 * inflate_threshold * new_child_length
594 * expand not_to_be_doubled and to_be_doubled, and shorten:
595 * 100 * (tnode_child_length(tn) - tn->empty_children +
596 * tn->full_children) >= inflate_threshold * new_child_length
598 * expand new_child_length:
599 * 100 * (tnode_child_length(tn) - tn->empty_children +
600 * tn->full_children) >=
601 * inflate_threshold * tnode_child_length(tn) * 2
604 * 50 * (tn->full_children + tnode_child_length(tn) -
605 * tn->empty_children) >= inflate_threshold *
606 * tnode_child_length(tn)
612 /* Keep root node larger */
614 if (!node_parent((struct rt_trie_node
*)tn
)) {
615 inflate_threshold_use
= inflate_threshold_root
;
616 halve_threshold_use
= halve_threshold_root
;
618 inflate_threshold_use
= inflate_threshold
;
619 halve_threshold_use
= halve_threshold
;
623 while ((tn
->full_children
> 0 && max_work
-- &&
624 50 * (tn
->full_children
+ tnode_child_length(tn
)
625 - tn
->empty_children
)
626 >= inflate_threshold_use
* tnode_child_length(tn
))) {
633 #ifdef CONFIG_IP_FIB_TRIE_STATS
634 t
->stats
.resize_node_skipped
++;
642 /* Return if at least one inflate is run */
643 if (max_work
!= MAX_WORK
)
644 return (struct rt_trie_node
*) tn
;
647 * Halve as long as the number of empty children in this
648 * node is above threshold.
652 while (tn
->bits
> 1 && max_work
-- &&
653 100 * (tnode_child_length(tn
) - tn
->empty_children
) <
654 halve_threshold_use
* tnode_child_length(tn
)) {
660 #ifdef CONFIG_IP_FIB_TRIE_STATS
661 t
->stats
.resize_node_skipped
++;
668 /* Only one child remains */
669 if (tn
->empty_children
== tnode_child_length(tn
) - 1) {
671 for (i
= 0; i
< tnode_child_length(tn
); i
++) {
672 struct rt_trie_node
*n
;
674 n
= rtnl_dereference(tn
->child
[i
]);
678 /* compress one level */
680 node_set_parent(n
, NULL
);
685 return (struct rt_trie_node
*) tn
;
689 static void tnode_clean_free(struct tnode
*tn
)
692 struct tnode
*tofree
;
694 for (i
= 0; i
< tnode_child_length(tn
); i
++) {
695 tofree
= (struct tnode
*)rtnl_dereference(tn
->child
[i
]);
702 static struct tnode
*inflate(struct trie
*t
, struct tnode
*tn
)
704 struct tnode
*oldtnode
= tn
;
705 int olen
= tnode_child_length(tn
);
708 pr_debug("In inflate\n");
710 tn
= tnode_new(oldtnode
->key
, oldtnode
->pos
, oldtnode
->bits
+ 1);
713 return ERR_PTR(-ENOMEM
);
716 * Preallocate and store tnodes before the actual work so we
717 * don't get into an inconsistent state if memory allocation
718 * fails. In case of failure we return the oldnode and inflate
719 * of tnode is ignored.
722 for (i
= 0; i
< olen
; i
++) {
725 inode
= (struct tnode
*) tnode_get_child(oldtnode
, i
);
728 inode
->pos
== oldtnode
->pos
+ oldtnode
->bits
&&
730 struct tnode
*left
, *right
;
731 t_key m
= ~0U << (KEYLENGTH
- 1) >> inode
->pos
;
733 left
= tnode_new(inode
->key
&(~m
), inode
->pos
+ 1,
738 right
= tnode_new(inode
->key
|m
, inode
->pos
+ 1,
746 put_child(tn
, 2*i
, (struct rt_trie_node
*) left
);
747 put_child(tn
, 2*i
+1, (struct rt_trie_node
*) right
);
751 for (i
= 0; i
< olen
; i
++) {
753 struct rt_trie_node
*node
= tnode_get_child(oldtnode
, i
);
754 struct tnode
*left
, *right
;
761 /* A leaf or an internal node with skipped bits */
763 if (IS_LEAF(node
) || ((struct tnode
*) node
)->pos
>
764 tn
->pos
+ tn
->bits
- 1) {
765 if (tkey_extract_bits(node
->key
,
766 oldtnode
->pos
+ oldtnode
->bits
,
768 put_child(tn
, 2*i
, node
);
770 put_child(tn
, 2*i
+1, node
);
774 /* An internal node with two children */
775 inode
= (struct tnode
*) node
;
777 if (inode
->bits
== 1) {
778 put_child(tn
, 2*i
, rtnl_dereference(inode
->child
[0]));
779 put_child(tn
, 2*i
+1, rtnl_dereference(inode
->child
[1]));
781 tnode_free_safe(inode
);
785 /* An internal node with more than two children */
787 /* We will replace this node 'inode' with two new
788 * ones, 'left' and 'right', each with half of the
789 * original children. The two new nodes will have
790 * a position one bit further down the key and this
791 * means that the "significant" part of their keys
792 * (see the discussion near the top of this file)
793 * will differ by one bit, which will be "0" in
794 * left's key and "1" in right's key. Since we are
795 * moving the key position by one step, the bit that
796 * we are moving away from - the bit at position
797 * (inode->pos) - is the one that will differ between
798 * left and right. So... we synthesize that bit in the
800 * The mask 'm' below will be a single "one" bit at
801 * the position (inode->pos)
804 /* Use the old key, but set the new significant
808 left
= (struct tnode
*) tnode_get_child(tn
, 2*i
);
809 put_child(tn
, 2*i
, NULL
);
813 right
= (struct tnode
*) tnode_get_child(tn
, 2*i
+1);
814 put_child(tn
, 2*i
+1, NULL
);
818 size
= tnode_child_length(left
);
819 for (j
= 0; j
< size
; j
++) {
820 put_child(left
, j
, rtnl_dereference(inode
->child
[j
]));
821 put_child(right
, j
, rtnl_dereference(inode
->child
[j
+ size
]));
823 put_child(tn
, 2*i
, resize(t
, left
));
824 put_child(tn
, 2*i
+1, resize(t
, right
));
826 tnode_free_safe(inode
);
828 tnode_free_safe(oldtnode
);
831 tnode_clean_free(tn
);
832 return ERR_PTR(-ENOMEM
);
835 static struct tnode
*halve(struct trie
*t
, struct tnode
*tn
)
837 struct tnode
*oldtnode
= tn
;
838 struct rt_trie_node
*left
, *right
;
840 int olen
= tnode_child_length(tn
);
842 pr_debug("In halve\n");
844 tn
= tnode_new(oldtnode
->key
, oldtnode
->pos
, oldtnode
->bits
- 1);
847 return ERR_PTR(-ENOMEM
);
850 * Preallocate and store tnodes before the actual work so we
851 * don't get into an inconsistent state if memory allocation
852 * fails. In case of failure we return the oldnode and halve
853 * of tnode is ignored.
856 for (i
= 0; i
< olen
; i
+= 2) {
857 left
= tnode_get_child(oldtnode
, i
);
858 right
= tnode_get_child(oldtnode
, i
+1);
860 /* Two nonempty children */
864 newn
= tnode_new(left
->key
, tn
->pos
+ tn
->bits
, 1);
869 put_child(tn
, i
/2, (struct rt_trie_node
*)newn
);
874 for (i
= 0; i
< olen
; i
+= 2) {
875 struct tnode
*newBinNode
;
877 left
= tnode_get_child(oldtnode
, i
);
878 right
= tnode_get_child(oldtnode
, i
+1);
880 /* At least one of the children is empty */
882 if (right
== NULL
) /* Both are empty */
884 put_child(tn
, i
/2, right
);
889 put_child(tn
, i
/2, left
);
893 /* Two nonempty children */
894 newBinNode
= (struct tnode
*) tnode_get_child(tn
, i
/2);
895 put_child(tn
, i
/2, NULL
);
896 put_child(newBinNode
, 0, left
);
897 put_child(newBinNode
, 1, right
);
898 put_child(tn
, i
/2, resize(t
, newBinNode
));
900 tnode_free_safe(oldtnode
);
903 tnode_clean_free(tn
);
904 return ERR_PTR(-ENOMEM
);
907 /* readside must use rcu_read_lock currently dump routines
908 via get_fa_head and dump */
910 static struct leaf_info
*find_leaf_info(struct leaf
*l
, int plen
)
912 struct hlist_head
*head
= &l
->list
;
913 struct leaf_info
*li
;
915 hlist_for_each_entry_rcu(li
, head
, hlist
)
916 if (li
->plen
== plen
)
922 static inline struct list_head
*get_fa_head(struct leaf
*l
, int plen
)
924 struct leaf_info
*li
= find_leaf_info(l
, plen
);
932 static void insert_leaf_info(struct hlist_head
*head
, struct leaf_info
*new)
934 struct leaf_info
*li
= NULL
, *last
= NULL
;
936 if (hlist_empty(head
)) {
937 hlist_add_head_rcu(&new->hlist
, head
);
939 hlist_for_each_entry(li
, head
, hlist
) {
940 if (new->plen
> li
->plen
)
946 hlist_add_after_rcu(&last
->hlist
, &new->hlist
);
948 hlist_add_before_rcu(&new->hlist
, &li
->hlist
);
952 /* rcu_read_lock needs to be hold by caller from readside */
955 fib_find_node(struct trie
*t
, u32 key
)
959 struct rt_trie_node
*n
;
962 n
= rcu_dereference_rtnl(t
->trie
);
964 while (n
!= NULL
&& NODE_TYPE(n
) == T_TNODE
) {
965 tn
= (struct tnode
*) n
;
969 if (tkey_sub_equals(tn
->key
, pos
, tn
->pos
-pos
, key
)) {
970 pos
= tn
->pos
+ tn
->bits
;
971 n
= tnode_get_child_rcu(tn
,
972 tkey_extract_bits(key
,
978 /* Case we have found a leaf. Compare prefixes */
980 if (n
!= NULL
&& IS_LEAF(n
) && tkey_equals(key
, n
->key
))
981 return (struct leaf
*)n
;
986 static void trie_rebalance(struct trie
*t
, struct tnode
*tn
)
994 while (tn
!= NULL
&& (tp
= node_parent((struct rt_trie_node
*)tn
)) != NULL
) {
995 cindex
= tkey_extract_bits(key
, tp
->pos
, tp
->bits
);
996 wasfull
= tnode_full(tp
, tnode_get_child(tp
, cindex
));
997 tn
= (struct tnode
*)resize(t
, tn
);
999 tnode_put_child_reorg(tp
, cindex
,
1000 (struct rt_trie_node
*)tn
, wasfull
);
1002 tp
= node_parent((struct rt_trie_node
*) tn
);
1004 rcu_assign_pointer(t
->trie
, (struct rt_trie_node
*)tn
);
1012 /* Handle last (top) tnode */
1014 tn
= (struct tnode
*)resize(t
, tn
);
1016 rcu_assign_pointer(t
->trie
, (struct rt_trie_node
*)tn
);
1020 /* only used from updater-side */
1022 static struct list_head
*fib_insert_node(struct trie
*t
, u32 key
, int plen
)
1025 struct tnode
*tp
= NULL
, *tn
= NULL
;
1026 struct rt_trie_node
*n
;
1029 struct list_head
*fa_head
= NULL
;
1030 struct leaf_info
*li
;
1034 n
= rtnl_dereference(t
->trie
);
1036 /* If we point to NULL, stop. Either the tree is empty and we should
1037 * just put a new leaf in if, or we have reached an empty child slot,
1038 * and we should just put our new leaf in that.
1039 * If we point to a T_TNODE, check if it matches our key. Note that
1040 * a T_TNODE might be skipping any number of bits - its 'pos' need
1041 * not be the parent's 'pos'+'bits'!
1043 * If it does match the current key, get pos/bits from it, extract
1044 * the index from our key, push the T_TNODE and walk the tree.
1046 * If it doesn't, we have to replace it with a new T_TNODE.
1048 * If we point to a T_LEAF, it might or might not have the same key
1049 * as we do. If it does, just change the value, update the T_LEAF's
1050 * value, and return it.
1051 * If it doesn't, we need to replace it with a T_TNODE.
1054 while (n
!= NULL
&& NODE_TYPE(n
) == T_TNODE
) {
1055 tn
= (struct tnode
*) n
;
1059 if (tkey_sub_equals(tn
->key
, pos
, tn
->pos
-pos
, key
)) {
1061 pos
= tn
->pos
+ tn
->bits
;
1062 n
= tnode_get_child(tn
,
1063 tkey_extract_bits(key
,
1067 BUG_ON(n
&& node_parent(n
) != tn
);
1073 * n ----> NULL, LEAF or TNODE
1075 * tp is n's (parent) ----> NULL or TNODE
1078 BUG_ON(tp
&& IS_LEAF(tp
));
1080 /* Case 1: n is a leaf. Compare prefixes */
1082 if (n
!= NULL
&& IS_LEAF(n
) && tkey_equals(key
, n
->key
)) {
1083 l
= (struct leaf
*) n
;
1084 li
= leaf_info_new(plen
);
1089 fa_head
= &li
->falh
;
1090 insert_leaf_info(&l
->list
, li
);
1099 li
= leaf_info_new(plen
);
1106 fa_head
= &li
->falh
;
1107 insert_leaf_info(&l
->list
, li
);
1109 if (t
->trie
&& n
== NULL
) {
1110 /* Case 2: n is NULL, and will just insert a new leaf */
1112 node_set_parent((struct rt_trie_node
*)l
, tp
);
1114 cindex
= tkey_extract_bits(key
, tp
->pos
, tp
->bits
);
1115 put_child(tp
, cindex
, (struct rt_trie_node
*)l
);
1117 /* Case 3: n is a LEAF or a TNODE and the key doesn't match. */
1119 * Add a new tnode here
1120 * first tnode need some special handling
1124 pos
= tp
->pos
+tp
->bits
;
1129 newpos
= tkey_mismatch(key
, pos
, n
->key
);
1130 tn
= tnode_new(n
->key
, newpos
, 1);
1133 tn
= tnode_new(key
, newpos
, 1); /* First tnode */
1142 node_set_parent((struct rt_trie_node
*)tn
, tp
);
1144 missbit
= tkey_extract_bits(key
, newpos
, 1);
1145 put_child(tn
, missbit
, (struct rt_trie_node
*)l
);
1146 put_child(tn
, 1-missbit
, n
);
1149 cindex
= tkey_extract_bits(key
, tp
->pos
, tp
->bits
);
1150 put_child(tp
, cindex
, (struct rt_trie_node
*)tn
);
1152 rcu_assign_pointer(t
->trie
, (struct rt_trie_node
*)tn
);
1157 if (tp
&& tp
->pos
+ tp
->bits
> 32)
1158 pr_warn("fib_trie tp=%p pos=%d, bits=%d, key=%0x plen=%d\n",
1159 tp
, tp
->pos
, tp
->bits
, key
, plen
);
1161 /* Rebalance the trie */
1163 trie_rebalance(t
, tp
);
1169 * Caller must hold RTNL.
1171 int fib_table_insert(struct fib_table
*tb
, struct fib_config
*cfg
)
1173 struct trie
*t
= (struct trie
*) tb
->tb_data
;
1174 struct fib_alias
*fa
, *new_fa
;
1175 struct list_head
*fa_head
= NULL
;
1176 struct fib_info
*fi
;
1177 int plen
= cfg
->fc_dst_len
;
1178 u8 tos
= cfg
->fc_tos
;
1186 key
= ntohl(cfg
->fc_dst
);
1188 pr_debug("Insert table=%u %08x/%d\n", tb
->tb_id
, key
, plen
);
1190 mask
= ntohl(inet_make_mask(plen
));
1197 fi
= fib_create_info(cfg
);
1203 l
= fib_find_node(t
, key
);
1207 fa_head
= get_fa_head(l
, plen
);
1208 fa
= fib_find_alias(fa_head
, tos
, fi
->fib_priority
);
1211 /* Now fa, if non-NULL, points to the first fib alias
1212 * with the same keys [prefix,tos,priority], if such key already
1213 * exists or to the node before which we will insert new one.
1215 * If fa is NULL, we will need to allocate a new one and
1216 * insert to the head of f.
1218 * If f is NULL, no fib node matched the destination key
1219 * and we need to allocate a new one of those as well.
1222 if (fa
&& fa
->fa_tos
== tos
&&
1223 fa
->fa_info
->fib_priority
== fi
->fib_priority
) {
1224 struct fib_alias
*fa_first
, *fa_match
;
1227 if (cfg
->fc_nlflags
& NLM_F_EXCL
)
1231 * 1. Find exact match for type, scope, fib_info to avoid
1233 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1237 fa
= list_entry(fa
->fa_list
.prev
, struct fib_alias
, fa_list
);
1238 list_for_each_entry_continue(fa
, fa_head
, fa_list
) {
1239 if (fa
->fa_tos
!= tos
)
1241 if (fa
->fa_info
->fib_priority
!= fi
->fib_priority
)
1243 if (fa
->fa_type
== cfg
->fc_type
&&
1244 fa
->fa_info
== fi
) {
1250 if (cfg
->fc_nlflags
& NLM_F_REPLACE
) {
1251 struct fib_info
*fi_drop
;
1261 new_fa
= kmem_cache_alloc(fn_alias_kmem
, GFP_KERNEL
);
1265 fi_drop
= fa
->fa_info
;
1266 new_fa
->fa_tos
= fa
->fa_tos
;
1267 new_fa
->fa_info
= fi
;
1268 new_fa
->fa_type
= cfg
->fc_type
;
1269 state
= fa
->fa_state
;
1270 new_fa
->fa_state
= state
& ~FA_S_ACCESSED
;
1272 list_replace_rcu(&fa
->fa_list
, &new_fa
->fa_list
);
1273 alias_free_mem_rcu(fa
);
1275 fib_release_info(fi_drop
);
1276 if (state
& FA_S_ACCESSED
)
1277 rt_cache_flush(cfg
->fc_nlinfo
.nl_net
);
1278 rtmsg_fib(RTM_NEWROUTE
, htonl(key
), new_fa
, plen
,
1279 tb
->tb_id
, &cfg
->fc_nlinfo
, NLM_F_REPLACE
);
1283 /* Error if we find a perfect match which
1284 * uses the same scope, type, and nexthop
1290 if (!(cfg
->fc_nlflags
& NLM_F_APPEND
))
1294 if (!(cfg
->fc_nlflags
& NLM_F_CREATE
))
1298 new_fa
= kmem_cache_alloc(fn_alias_kmem
, GFP_KERNEL
);
1302 new_fa
->fa_info
= fi
;
1303 new_fa
->fa_tos
= tos
;
1304 new_fa
->fa_type
= cfg
->fc_type
;
1305 new_fa
->fa_state
= 0;
1307 * Insert new entry to the list.
1311 fa_head
= fib_insert_node(t
, key
, plen
);
1312 if (unlikely(!fa_head
)) {
1314 goto out_free_new_fa
;
1319 tb
->tb_num_default
++;
1321 list_add_tail_rcu(&new_fa
->fa_list
,
1322 (fa
? &fa
->fa_list
: fa_head
));
1324 rt_cache_flush(cfg
->fc_nlinfo
.nl_net
);
1325 rtmsg_fib(RTM_NEWROUTE
, htonl(key
), new_fa
, plen
, tb
->tb_id
,
1326 &cfg
->fc_nlinfo
, 0);
1331 kmem_cache_free(fn_alias_kmem
, new_fa
);
1333 fib_release_info(fi
);
1338 /* should be called with rcu_read_lock */
1339 static int check_leaf(struct fib_table
*tb
, struct trie
*t
, struct leaf
*l
,
1340 t_key key
, const struct flowi4
*flp
,
1341 struct fib_result
*res
, int fib_flags
)
1343 struct leaf_info
*li
;
1344 struct hlist_head
*hhead
= &l
->list
;
1346 hlist_for_each_entry_rcu(li
, hhead
, hlist
) {
1347 struct fib_alias
*fa
;
1349 if (l
->key
!= (key
& li
->mask_plen
))
1352 list_for_each_entry_rcu(fa
, &li
->falh
, fa_list
) {
1353 struct fib_info
*fi
= fa
->fa_info
;
1356 if (fa
->fa_tos
&& fa
->fa_tos
!= flp
->flowi4_tos
)
1360 if (fa
->fa_info
->fib_scope
< flp
->flowi4_scope
)
1362 fib_alias_accessed(fa
);
1363 err
= fib_props
[fa
->fa_type
].error
;
1365 #ifdef CONFIG_IP_FIB_TRIE_STATS
1366 t
->stats
.semantic_match_passed
++;
1370 if (fi
->fib_flags
& RTNH_F_DEAD
)
1372 for (nhsel
= 0; nhsel
< fi
->fib_nhs
; nhsel
++) {
1373 const struct fib_nh
*nh
= &fi
->fib_nh
[nhsel
];
1375 if (nh
->nh_flags
& RTNH_F_DEAD
)
1377 if (flp
->flowi4_oif
&& flp
->flowi4_oif
!= nh
->nh_oif
)
1380 #ifdef CONFIG_IP_FIB_TRIE_STATS
1381 t
->stats
.semantic_match_passed
++;
1383 res
->prefixlen
= li
->plen
;
1384 res
->nh_sel
= nhsel
;
1385 res
->type
= fa
->fa_type
;
1386 res
->scope
= fa
->fa_info
->fib_scope
;
1389 res
->fa_head
= &li
->falh
;
1390 if (!(fib_flags
& FIB_LOOKUP_NOREF
))
1391 atomic_inc(&fi
->fib_clntref
);
1396 #ifdef CONFIG_IP_FIB_TRIE_STATS
1397 t
->stats
.semantic_match_miss
++;
1404 int fib_table_lookup(struct fib_table
*tb
, const struct flowi4
*flp
,
1405 struct fib_result
*res
, int fib_flags
)
1407 struct trie
*t
= (struct trie
*) tb
->tb_data
;
1409 struct rt_trie_node
*n
;
1411 unsigned int pos
, bits
;
1412 t_key key
= ntohl(flp
->daddr
);
1413 unsigned int chopped_off
;
1415 unsigned int current_prefix_length
= KEYLENGTH
;
1417 t_key pref_mismatch
;
1421 n
= rcu_dereference(t
->trie
);
1425 #ifdef CONFIG_IP_FIB_TRIE_STATS
1431 ret
= check_leaf(tb
, t
, (struct leaf
*)n
, key
, flp
, res
, fib_flags
);
1435 pn
= (struct tnode
*) n
;
1443 cindex
= tkey_extract_bits(mask_pfx(key
, current_prefix_length
),
1446 n
= tnode_get_child_rcu(pn
, cindex
);
1449 #ifdef CONFIG_IP_FIB_TRIE_STATS
1450 t
->stats
.null_node_hit
++;
1456 ret
= check_leaf(tb
, t
, (struct leaf
*)n
, key
, flp
, res
, fib_flags
);
1462 cn
= (struct tnode
*)n
;
1465 * It's a tnode, and we can do some extra checks here if we
1466 * like, to avoid descending into a dead-end branch.
1467 * This tnode is in the parent's child array at index
1468 * key[p_pos..p_pos+p_bits] but potentially with some bits
1469 * chopped off, so in reality the index may be just a
1470 * subprefix, padded with zero at the end.
1471 * We can also take a look at any skipped bits in this
1472 * tnode - everything up to p_pos is supposed to be ok,
1473 * and the non-chopped bits of the index (se previous
1474 * paragraph) are also guaranteed ok, but the rest is
1475 * considered unknown.
1477 * The skipped bits are key[pos+bits..cn->pos].
1480 /* If current_prefix_length < pos+bits, we are already doing
1481 * actual prefix matching, which means everything from
1482 * pos+(bits-chopped_off) onward must be zero along some
1483 * branch of this subtree - otherwise there is *no* valid
1484 * prefix present. Here we can only check the skipped
1485 * bits. Remember, since we have already indexed into the
1486 * parent's child array, we know that the bits we chopped of
1490 /* NOTA BENE: Checking only skipped bits
1491 for the new node here */
1493 if (current_prefix_length
< pos
+bits
) {
1494 if (tkey_extract_bits(cn
->key
, current_prefix_length
,
1495 cn
->pos
- current_prefix_length
)
1501 * If chopped_off=0, the index is fully validated and we
1502 * only need to look at the skipped bits for this, the new,
1503 * tnode. What we actually want to do is to find out if
1504 * these skipped bits match our key perfectly, or if we will
1505 * have to count on finding a matching prefix further down,
1506 * because if we do, we would like to have some way of
1507 * verifying the existence of such a prefix at this point.
1510 /* The only thing we can do at this point is to verify that
1511 * any such matching prefix can indeed be a prefix to our
1512 * key, and if the bits in the node we are inspecting that
1513 * do not match our key are not ZERO, this cannot be true.
1514 * Thus, find out where there is a mismatch (before cn->pos)
1515 * and verify that all the mismatching bits are zero in the
1520 * Note: We aren't very concerned about the piece of
1521 * the key that precede pn->pos+pn->bits, since these
1522 * have already been checked. The bits after cn->pos
1523 * aren't checked since these are by definition
1524 * "unknown" at this point. Thus, what we want to see
1525 * is if we are about to enter the "prefix matching"
1526 * state, and in that case verify that the skipped
1527 * bits that will prevail throughout this subtree are
1528 * zero, as they have to be if we are to find a
1532 pref_mismatch
= mask_pfx(cn
->key
^ key
, cn
->pos
);
1535 * In short: If skipped bits in this node do not match
1536 * the search key, enter the "prefix matching"
1539 if (pref_mismatch
) {
1540 /* fls(x) = __fls(x) + 1 */
1541 int mp
= KEYLENGTH
- __fls(pref_mismatch
) - 1;
1543 if (tkey_extract_bits(cn
->key
, mp
, cn
->pos
- mp
) != 0)
1546 if (current_prefix_length
>= cn
->pos
)
1547 current_prefix_length
= mp
;
1550 pn
= (struct tnode
*)n
; /* Descend */
1557 /* As zero don't change the child key (cindex) */
1558 while ((chopped_off
<= pn
->bits
)
1559 && !(cindex
& (1<<(chopped_off
-1))))
1562 /* Decrease current_... with bits chopped off */
1563 if (current_prefix_length
> pn
->pos
+ pn
->bits
- chopped_off
)
1564 current_prefix_length
= pn
->pos
+ pn
->bits
1568 * Either we do the actual chop off according or if we have
1569 * chopped off all bits in this tnode walk up to our parent.
1572 if (chopped_off
<= pn
->bits
) {
1573 cindex
&= ~(1 << (chopped_off
-1));
1575 struct tnode
*parent
= node_parent_rcu((struct rt_trie_node
*) pn
);
1579 /* Get Child's index */
1580 cindex
= tkey_extract_bits(pn
->key
, parent
->pos
, parent
->bits
);
1584 #ifdef CONFIG_IP_FIB_TRIE_STATS
1585 t
->stats
.backtrack
++;
1596 EXPORT_SYMBOL_GPL(fib_table_lookup
);
1599 * Remove the leaf and return parent.
1601 static void trie_leaf_remove(struct trie
*t
, struct leaf
*l
)
1603 struct tnode
*tp
= node_parent((struct rt_trie_node
*) l
);
1605 pr_debug("entering trie_leaf_remove(%p)\n", l
);
1608 t_key cindex
= tkey_extract_bits(l
->key
, tp
->pos
, tp
->bits
);
1609 put_child(tp
, cindex
, NULL
);
1610 trie_rebalance(t
, tp
);
1612 RCU_INIT_POINTER(t
->trie
, NULL
);
1618 * Caller must hold RTNL.
1620 int fib_table_delete(struct fib_table
*tb
, struct fib_config
*cfg
)
1622 struct trie
*t
= (struct trie
*) tb
->tb_data
;
1624 int plen
= cfg
->fc_dst_len
;
1625 u8 tos
= cfg
->fc_tos
;
1626 struct fib_alias
*fa
, *fa_to_delete
;
1627 struct list_head
*fa_head
;
1629 struct leaf_info
*li
;
1634 key
= ntohl(cfg
->fc_dst
);
1635 mask
= ntohl(inet_make_mask(plen
));
1641 l
= fib_find_node(t
, key
);
1646 li
= find_leaf_info(l
, plen
);
1651 fa_head
= &li
->falh
;
1652 fa
= fib_find_alias(fa_head
, tos
, 0);
1657 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key
, plen
, tos
, t
);
1659 fa_to_delete
= NULL
;
1660 fa
= list_entry(fa
->fa_list
.prev
, struct fib_alias
, fa_list
);
1661 list_for_each_entry_continue(fa
, fa_head
, fa_list
) {
1662 struct fib_info
*fi
= fa
->fa_info
;
1664 if (fa
->fa_tos
!= tos
)
1667 if ((!cfg
->fc_type
|| fa
->fa_type
== cfg
->fc_type
) &&
1668 (cfg
->fc_scope
== RT_SCOPE_NOWHERE
||
1669 fa
->fa_info
->fib_scope
== cfg
->fc_scope
) &&
1670 (!cfg
->fc_prefsrc
||
1671 fi
->fib_prefsrc
== cfg
->fc_prefsrc
) &&
1672 (!cfg
->fc_protocol
||
1673 fi
->fib_protocol
== cfg
->fc_protocol
) &&
1674 fib_nh_match(cfg
, fi
) == 0) {
1684 rtmsg_fib(RTM_DELROUTE
, htonl(key
), fa
, plen
, tb
->tb_id
,
1685 &cfg
->fc_nlinfo
, 0);
1687 list_del_rcu(&fa
->fa_list
);
1690 tb
->tb_num_default
--;
1692 if (list_empty(fa_head
)) {
1693 hlist_del_rcu(&li
->hlist
);
1697 if (hlist_empty(&l
->list
))
1698 trie_leaf_remove(t
, l
);
1700 if (fa
->fa_state
& FA_S_ACCESSED
)
1701 rt_cache_flush(cfg
->fc_nlinfo
.nl_net
);
1703 fib_release_info(fa
->fa_info
);
1704 alias_free_mem_rcu(fa
);
1708 static int trie_flush_list(struct list_head
*head
)
1710 struct fib_alias
*fa
, *fa_node
;
1713 list_for_each_entry_safe(fa
, fa_node
, head
, fa_list
) {
1714 struct fib_info
*fi
= fa
->fa_info
;
1716 if (fi
&& (fi
->fib_flags
& RTNH_F_DEAD
)) {
1717 list_del_rcu(&fa
->fa_list
);
1718 fib_release_info(fa
->fa_info
);
1719 alias_free_mem_rcu(fa
);
1726 static int trie_flush_leaf(struct leaf
*l
)
1729 struct hlist_head
*lih
= &l
->list
;
1730 struct hlist_node
*tmp
;
1731 struct leaf_info
*li
= NULL
;
1733 hlist_for_each_entry_safe(li
, tmp
, lih
, hlist
) {
1734 found
+= trie_flush_list(&li
->falh
);
1736 if (list_empty(&li
->falh
)) {
1737 hlist_del_rcu(&li
->hlist
);
1745 * Scan for the next right leaf starting at node p->child[idx]
1746 * Since we have back pointer, no recursion necessary.
1748 static struct leaf
*leaf_walk_rcu(struct tnode
*p
, struct rt_trie_node
*c
)
1754 idx
= tkey_extract_bits(c
->key
, p
->pos
, p
->bits
) + 1;
1758 while (idx
< 1u << p
->bits
) {
1759 c
= tnode_get_child_rcu(p
, idx
++);
1764 return (struct leaf
*) c
;
1766 /* Rescan start scanning in new node */
1767 p
= (struct tnode
*) c
;
1771 /* Node empty, walk back up to parent */
1772 c
= (struct rt_trie_node
*) p
;
1773 } while ((p
= node_parent_rcu(c
)) != NULL
);
1775 return NULL
; /* Root of trie */
1778 static struct leaf
*trie_firstleaf(struct trie
*t
)
1780 struct tnode
*n
= (struct tnode
*)rcu_dereference_rtnl(t
->trie
);
1785 if (IS_LEAF(n
)) /* trie is just a leaf */
1786 return (struct leaf
*) n
;
1788 return leaf_walk_rcu(n
, NULL
);
1791 static struct leaf
*trie_nextleaf(struct leaf
*l
)
1793 struct rt_trie_node
*c
= (struct rt_trie_node
*) l
;
1794 struct tnode
*p
= node_parent_rcu(c
);
1797 return NULL
; /* trie with just one leaf */
1799 return leaf_walk_rcu(p
, c
);
1802 static struct leaf
*trie_leafindex(struct trie
*t
, int index
)
1804 struct leaf
*l
= trie_firstleaf(t
);
1806 while (l
&& index
-- > 0)
1807 l
= trie_nextleaf(l
);
1814 * Caller must hold RTNL.
1816 int fib_table_flush(struct fib_table
*tb
)
1818 struct trie
*t
= (struct trie
*) tb
->tb_data
;
1819 struct leaf
*l
, *ll
= NULL
;
1822 for (l
= trie_firstleaf(t
); l
; l
= trie_nextleaf(l
)) {
1823 found
+= trie_flush_leaf(l
);
1825 if (ll
&& hlist_empty(&ll
->list
))
1826 trie_leaf_remove(t
, ll
);
1830 if (ll
&& hlist_empty(&ll
->list
))
1831 trie_leaf_remove(t
, ll
);
1833 pr_debug("trie_flush found=%d\n", found
);
1837 void fib_free_table(struct fib_table
*tb
)
1842 static int fn_trie_dump_fa(t_key key
, int plen
, struct list_head
*fah
,
1843 struct fib_table
*tb
,
1844 struct sk_buff
*skb
, struct netlink_callback
*cb
)
1847 struct fib_alias
*fa
;
1848 __be32 xkey
= htonl(key
);
1853 /* rcu_read_lock is hold by caller */
1855 list_for_each_entry_rcu(fa
, fah
, fa_list
) {
1861 if (fib_dump_info(skb
, NETLINK_CB(cb
->skb
).portid
,
1869 fa
->fa_info
, NLM_F_MULTI
) < 0) {
1879 static int fn_trie_dump_leaf(struct leaf
*l
, struct fib_table
*tb
,
1880 struct sk_buff
*skb
, struct netlink_callback
*cb
)
1882 struct leaf_info
*li
;
1888 /* rcu_read_lock is hold by caller */
1889 hlist_for_each_entry_rcu(li
, &l
->list
, hlist
) {
1898 if (list_empty(&li
->falh
))
1901 if (fn_trie_dump_fa(l
->key
, li
->plen
, &li
->falh
, tb
, skb
, cb
) < 0) {
1912 int fib_table_dump(struct fib_table
*tb
, struct sk_buff
*skb
,
1913 struct netlink_callback
*cb
)
1916 struct trie
*t
= (struct trie
*) tb
->tb_data
;
1917 t_key key
= cb
->args
[2];
1918 int count
= cb
->args
[3];
1921 /* Dump starting at last key.
1922 * Note: 0.0.0.0/0 (ie default) is first key.
1925 l
= trie_firstleaf(t
);
1927 /* Normally, continue from last key, but if that is missing
1928 * fallback to using slow rescan
1930 l
= fib_find_node(t
, key
);
1932 l
= trie_leafindex(t
, count
);
1936 cb
->args
[2] = l
->key
;
1937 if (fn_trie_dump_leaf(l
, tb
, skb
, cb
) < 0) {
1938 cb
->args
[3] = count
;
1944 l
= trie_nextleaf(l
);
1945 memset(&cb
->args
[4], 0,
1946 sizeof(cb
->args
) - 4*sizeof(cb
->args
[0]));
1948 cb
->args
[3] = count
;
1954 void __init
fib_trie_init(void)
1956 fn_alias_kmem
= kmem_cache_create("ip_fib_alias",
1957 sizeof(struct fib_alias
),
1958 0, SLAB_PANIC
, NULL
);
1960 trie_leaf_kmem
= kmem_cache_create("ip_fib_trie",
1961 max(sizeof(struct leaf
),
1962 sizeof(struct leaf_info
)),
1963 0, SLAB_PANIC
, NULL
);
1967 struct fib_table
*fib_trie_table(u32 id
)
1969 struct fib_table
*tb
;
1972 tb
= kmalloc(sizeof(struct fib_table
) + sizeof(struct trie
),
1978 tb
->tb_default
= -1;
1979 tb
->tb_num_default
= 0;
1981 t
= (struct trie
*) tb
->tb_data
;
1982 memset(t
, 0, sizeof(*t
));
1987 #ifdef CONFIG_PROC_FS
1988 /* Depth first Trie walk iterator */
1989 struct fib_trie_iter
{
1990 struct seq_net_private p
;
1991 struct fib_table
*tb
;
1992 struct tnode
*tnode
;
1997 static struct rt_trie_node
*fib_trie_get_next(struct fib_trie_iter
*iter
)
1999 struct tnode
*tn
= iter
->tnode
;
2000 unsigned int cindex
= iter
->index
;
2003 /* A single entry routing table */
2007 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2008 iter
->tnode
, iter
->index
, iter
->depth
);
2010 while (cindex
< (1<<tn
->bits
)) {
2011 struct rt_trie_node
*n
= tnode_get_child_rcu(tn
, cindex
);
2016 iter
->index
= cindex
+ 1;
2018 /* push down one level */
2019 iter
->tnode
= (struct tnode
*) n
;
2029 /* Current node exhausted, pop back up */
2030 p
= node_parent_rcu((struct rt_trie_node
*)tn
);
2032 cindex
= tkey_extract_bits(tn
->key
, p
->pos
, p
->bits
)+1;
2042 static struct rt_trie_node
*fib_trie_get_first(struct fib_trie_iter
*iter
,
2045 struct rt_trie_node
*n
;
2050 n
= rcu_dereference(t
->trie
);
2055 iter
->tnode
= (struct tnode
*) n
;
2067 static void trie_collect_stats(struct trie
*t
, struct trie_stat
*s
)
2069 struct rt_trie_node
*n
;
2070 struct fib_trie_iter iter
;
2072 memset(s
, 0, sizeof(*s
));
2075 for (n
= fib_trie_get_first(&iter
, t
); n
; n
= fib_trie_get_next(&iter
)) {
2077 struct leaf
*l
= (struct leaf
*)n
;
2078 struct leaf_info
*li
;
2081 s
->totdepth
+= iter
.depth
;
2082 if (iter
.depth
> s
->maxdepth
)
2083 s
->maxdepth
= iter
.depth
;
2085 hlist_for_each_entry_rcu(li
, &l
->list
, hlist
)
2088 const struct tnode
*tn
= (const struct tnode
*) n
;
2092 if (tn
->bits
< MAX_STAT_DEPTH
)
2093 s
->nodesizes
[tn
->bits
]++;
2095 for (i
= 0; i
< (1<<tn
->bits
); i
++)
2104 * This outputs /proc/net/fib_triestats
2106 static void trie_show_stats(struct seq_file
*seq
, struct trie_stat
*stat
)
2108 unsigned int i
, max
, pointers
, bytes
, avdepth
;
2111 avdepth
= stat
->totdepth
*100 / stat
->leaves
;
2115 seq_printf(seq
, "\tAver depth: %u.%02d\n",
2116 avdepth
/ 100, avdepth
% 100);
2117 seq_printf(seq
, "\tMax depth: %u\n", stat
->maxdepth
);
2119 seq_printf(seq
, "\tLeaves: %u\n", stat
->leaves
);
2120 bytes
= sizeof(struct leaf
) * stat
->leaves
;
2122 seq_printf(seq
, "\tPrefixes: %u\n", stat
->prefixes
);
2123 bytes
+= sizeof(struct leaf_info
) * stat
->prefixes
;
2125 seq_printf(seq
, "\tInternal nodes: %u\n\t", stat
->tnodes
);
2126 bytes
+= sizeof(struct tnode
) * stat
->tnodes
;
2128 max
= MAX_STAT_DEPTH
;
2129 while (max
> 0 && stat
->nodesizes
[max
-1] == 0)
2133 for (i
= 1; i
< max
; i
++)
2134 if (stat
->nodesizes
[i
] != 0) {
2135 seq_printf(seq
, " %u: %u", i
, stat
->nodesizes
[i
]);
2136 pointers
+= (1<<i
) * stat
->nodesizes
[i
];
2138 seq_putc(seq
, '\n');
2139 seq_printf(seq
, "\tPointers: %u\n", pointers
);
2141 bytes
+= sizeof(struct rt_trie_node
*) * pointers
;
2142 seq_printf(seq
, "Null ptrs: %u\n", stat
->nullpointers
);
2143 seq_printf(seq
, "Total size: %u kB\n", (bytes
+ 1023) / 1024);
2146 #ifdef CONFIG_IP_FIB_TRIE_STATS
2147 static void trie_show_usage(struct seq_file
*seq
,
2148 const struct trie_use_stats
*stats
)
2150 seq_printf(seq
, "\nCounters:\n---------\n");
2151 seq_printf(seq
, "gets = %u\n", stats
->gets
);
2152 seq_printf(seq
, "backtracks = %u\n", stats
->backtrack
);
2153 seq_printf(seq
, "semantic match passed = %u\n",
2154 stats
->semantic_match_passed
);
2155 seq_printf(seq
, "semantic match miss = %u\n",
2156 stats
->semantic_match_miss
);
2157 seq_printf(seq
, "null node hit= %u\n", stats
->null_node_hit
);
2158 seq_printf(seq
, "skipped node resize = %u\n\n",
2159 stats
->resize_node_skipped
);
2161 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2163 static void fib_table_print(struct seq_file
*seq
, struct fib_table
*tb
)
2165 if (tb
->tb_id
== RT_TABLE_LOCAL
)
2166 seq_puts(seq
, "Local:\n");
2167 else if (tb
->tb_id
== RT_TABLE_MAIN
)
2168 seq_puts(seq
, "Main:\n");
2170 seq_printf(seq
, "Id %d:\n", tb
->tb_id
);
2174 static int fib_triestat_seq_show(struct seq_file
*seq
, void *v
)
2176 struct net
*net
= (struct net
*)seq
->private;
2180 "Basic info: size of leaf:"
2181 " %Zd bytes, size of tnode: %Zd bytes.\n",
2182 sizeof(struct leaf
), sizeof(struct tnode
));
2184 for (h
= 0; h
< FIB_TABLE_HASHSZ
; h
++) {
2185 struct hlist_head
*head
= &net
->ipv4
.fib_table_hash
[h
];
2186 struct fib_table
*tb
;
2188 hlist_for_each_entry_rcu(tb
, head
, tb_hlist
) {
2189 struct trie
*t
= (struct trie
*) tb
->tb_data
;
2190 struct trie_stat stat
;
2195 fib_table_print(seq
, tb
);
2197 trie_collect_stats(t
, &stat
);
2198 trie_show_stats(seq
, &stat
);
2199 #ifdef CONFIG_IP_FIB_TRIE_STATS
2200 trie_show_usage(seq
, &t
->stats
);
2208 static int fib_triestat_seq_open(struct inode
*inode
, struct file
*file
)
2210 return single_open_net(inode
, file
, fib_triestat_seq_show
);
2213 static const struct file_operations fib_triestat_fops
= {
2214 .owner
= THIS_MODULE
,
2215 .open
= fib_triestat_seq_open
,
2217 .llseek
= seq_lseek
,
2218 .release
= single_release_net
,
2221 static struct rt_trie_node
*fib_trie_get_idx(struct seq_file
*seq
, loff_t pos
)
2223 struct fib_trie_iter
*iter
= seq
->private;
2224 struct net
*net
= seq_file_net(seq
);
2228 for (h
= 0; h
< FIB_TABLE_HASHSZ
; h
++) {
2229 struct hlist_head
*head
= &net
->ipv4
.fib_table_hash
[h
];
2230 struct fib_table
*tb
;
2232 hlist_for_each_entry_rcu(tb
, head
, tb_hlist
) {
2233 struct rt_trie_node
*n
;
2235 for (n
= fib_trie_get_first(iter
,
2236 (struct trie
*) tb
->tb_data
);
2237 n
; n
= fib_trie_get_next(iter
))
2248 static void *fib_trie_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2252 return fib_trie_get_idx(seq
, *pos
);
2255 static void *fib_trie_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2257 struct fib_trie_iter
*iter
= seq
->private;
2258 struct net
*net
= seq_file_net(seq
);
2259 struct fib_table
*tb
= iter
->tb
;
2260 struct hlist_node
*tb_node
;
2262 struct rt_trie_node
*n
;
2265 /* next node in same table */
2266 n
= fib_trie_get_next(iter
);
2270 /* walk rest of this hash chain */
2271 h
= tb
->tb_id
& (FIB_TABLE_HASHSZ
- 1);
2272 while ((tb_node
= rcu_dereference(hlist_next_rcu(&tb
->tb_hlist
)))) {
2273 tb
= hlist_entry(tb_node
, struct fib_table
, tb_hlist
);
2274 n
= fib_trie_get_first(iter
, (struct trie
*) tb
->tb_data
);
2279 /* new hash chain */
2280 while (++h
< FIB_TABLE_HASHSZ
) {
2281 struct hlist_head
*head
= &net
->ipv4
.fib_table_hash
[h
];
2282 hlist_for_each_entry_rcu(tb
, head
, tb_hlist
) {
2283 n
= fib_trie_get_first(iter
, (struct trie
*) tb
->tb_data
);
2295 static void fib_trie_seq_stop(struct seq_file
*seq
, void *v
)
2301 static void seq_indent(struct seq_file
*seq
, int n
)
2307 static inline const char *rtn_scope(char *buf
, size_t len
, enum rt_scope_t s
)
2310 case RT_SCOPE_UNIVERSE
: return "universe";
2311 case RT_SCOPE_SITE
: return "site";
2312 case RT_SCOPE_LINK
: return "link";
2313 case RT_SCOPE_HOST
: return "host";
2314 case RT_SCOPE_NOWHERE
: return "nowhere";
2316 snprintf(buf
, len
, "scope=%d", s
);
2321 static const char *const rtn_type_names
[__RTN_MAX
] = {
2322 [RTN_UNSPEC
] = "UNSPEC",
2323 [RTN_UNICAST
] = "UNICAST",
2324 [RTN_LOCAL
] = "LOCAL",
2325 [RTN_BROADCAST
] = "BROADCAST",
2326 [RTN_ANYCAST
] = "ANYCAST",
2327 [RTN_MULTICAST
] = "MULTICAST",
2328 [RTN_BLACKHOLE
] = "BLACKHOLE",
2329 [RTN_UNREACHABLE
] = "UNREACHABLE",
2330 [RTN_PROHIBIT
] = "PROHIBIT",
2331 [RTN_THROW
] = "THROW",
2333 [RTN_XRESOLVE
] = "XRESOLVE",
2336 static inline const char *rtn_type(char *buf
, size_t len
, unsigned int t
)
2338 if (t
< __RTN_MAX
&& rtn_type_names
[t
])
2339 return rtn_type_names
[t
];
2340 snprintf(buf
, len
, "type %u", t
);
2344 /* Pretty print the trie */
2345 static int fib_trie_seq_show(struct seq_file
*seq
, void *v
)
2347 const struct fib_trie_iter
*iter
= seq
->private;
2348 struct rt_trie_node
*n
= v
;
2350 if (!node_parent_rcu(n
))
2351 fib_table_print(seq
, iter
->tb
);
2354 struct tnode
*tn
= (struct tnode
*) n
;
2355 __be32 prf
= htonl(mask_pfx(tn
->key
, tn
->pos
));
2357 seq_indent(seq
, iter
->depth
-1);
2358 seq_printf(seq
, " +-- %pI4/%d %d %d %d\n",
2359 &prf
, tn
->pos
, tn
->bits
, tn
->full_children
,
2360 tn
->empty_children
);
2363 struct leaf
*l
= (struct leaf
*) n
;
2364 struct leaf_info
*li
;
2365 __be32 val
= htonl(l
->key
);
2367 seq_indent(seq
, iter
->depth
);
2368 seq_printf(seq
, " |-- %pI4\n", &val
);
2370 hlist_for_each_entry_rcu(li
, &l
->list
, hlist
) {
2371 struct fib_alias
*fa
;
2373 list_for_each_entry_rcu(fa
, &li
->falh
, fa_list
) {
2374 char buf1
[32], buf2
[32];
2376 seq_indent(seq
, iter
->depth
+1);
2377 seq_printf(seq
, " /%d %s %s", li
->plen
,
2378 rtn_scope(buf1
, sizeof(buf1
),
2379 fa
->fa_info
->fib_scope
),
2380 rtn_type(buf2
, sizeof(buf2
),
2383 seq_printf(seq
, " tos=%d", fa
->fa_tos
);
2384 seq_putc(seq
, '\n');
2392 static const struct seq_operations fib_trie_seq_ops
= {
2393 .start
= fib_trie_seq_start
,
2394 .next
= fib_trie_seq_next
,
2395 .stop
= fib_trie_seq_stop
,
2396 .show
= fib_trie_seq_show
,
2399 static int fib_trie_seq_open(struct inode
*inode
, struct file
*file
)
2401 return seq_open_net(inode
, file
, &fib_trie_seq_ops
,
2402 sizeof(struct fib_trie_iter
));
2405 static const struct file_operations fib_trie_fops
= {
2406 .owner
= THIS_MODULE
,
2407 .open
= fib_trie_seq_open
,
2409 .llseek
= seq_lseek
,
2410 .release
= seq_release_net
,
2413 struct fib_route_iter
{
2414 struct seq_net_private p
;
2415 struct trie
*main_trie
;
2420 static struct leaf
*fib_route_get_idx(struct fib_route_iter
*iter
, loff_t pos
)
2422 struct leaf
*l
= NULL
;
2423 struct trie
*t
= iter
->main_trie
;
2425 /* use cache location of last found key */
2426 if (iter
->pos
> 0 && pos
>= iter
->pos
&& (l
= fib_find_node(t
, iter
->key
)))
2430 l
= trie_firstleaf(t
);
2433 while (l
&& pos
-- > 0) {
2435 l
= trie_nextleaf(l
);
2439 iter
->key
= pos
; /* remember it */
2441 iter
->pos
= 0; /* forget it */
2446 static void *fib_route_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2449 struct fib_route_iter
*iter
= seq
->private;
2450 struct fib_table
*tb
;
2453 tb
= fib_get_table(seq_file_net(seq
), RT_TABLE_MAIN
);
2457 iter
->main_trie
= (struct trie
*) tb
->tb_data
;
2459 return SEQ_START_TOKEN
;
2461 return fib_route_get_idx(iter
, *pos
- 1);
2464 static void *fib_route_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2466 struct fib_route_iter
*iter
= seq
->private;
2470 if (v
== SEQ_START_TOKEN
) {
2472 l
= trie_firstleaf(iter
->main_trie
);
2475 l
= trie_nextleaf(l
);
2485 static void fib_route_seq_stop(struct seq_file
*seq
, void *v
)
2491 static unsigned int fib_flag_trans(int type
, __be32 mask
, const struct fib_info
*fi
)
2493 unsigned int flags
= 0;
2495 if (type
== RTN_UNREACHABLE
|| type
== RTN_PROHIBIT
)
2497 if (fi
&& fi
->fib_nh
->nh_gw
)
2498 flags
|= RTF_GATEWAY
;
2499 if (mask
== htonl(0xFFFFFFFF))
2506 * This outputs /proc/net/route.
2507 * The format of the file is not supposed to be changed
2508 * and needs to be same as fib_hash output to avoid breaking
2511 static int fib_route_seq_show(struct seq_file
*seq
, void *v
)
2514 struct leaf_info
*li
;
2516 if (v
== SEQ_START_TOKEN
) {
2517 seq_printf(seq
, "%-127s\n", "Iface\tDestination\tGateway "
2518 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2523 hlist_for_each_entry_rcu(li
, &l
->list
, hlist
) {
2524 struct fib_alias
*fa
;
2525 __be32 mask
, prefix
;
2527 mask
= inet_make_mask(li
->plen
);
2528 prefix
= htonl(l
->key
);
2530 list_for_each_entry_rcu(fa
, &li
->falh
, fa_list
) {
2531 const struct fib_info
*fi
= fa
->fa_info
;
2532 unsigned int flags
= fib_flag_trans(fa
->fa_type
, mask
, fi
);
2535 if (fa
->fa_type
== RTN_BROADCAST
2536 || fa
->fa_type
== RTN_MULTICAST
)
2541 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2542 "%d\t%08X\t%d\t%u\t%u%n",
2543 fi
->fib_dev
? fi
->fib_dev
->name
: "*",
2545 fi
->fib_nh
->nh_gw
, flags
, 0, 0,
2549 fi
->fib_advmss
+ 40 : 0),
2551 fi
->fib_rtt
>> 3, &len
);
2554 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2555 "%d\t%08X\t%d\t%u\t%u%n",
2556 prefix
, 0, flags
, 0, 0, 0,
2557 mask
, 0, 0, 0, &len
);
2559 seq_printf(seq
, "%*s\n", 127 - len
, "");
2566 static const struct seq_operations fib_route_seq_ops
= {
2567 .start
= fib_route_seq_start
,
2568 .next
= fib_route_seq_next
,
2569 .stop
= fib_route_seq_stop
,
2570 .show
= fib_route_seq_show
,
2573 static int fib_route_seq_open(struct inode
*inode
, struct file
*file
)
2575 return seq_open_net(inode
, file
, &fib_route_seq_ops
,
2576 sizeof(struct fib_route_iter
));
2579 static const struct file_operations fib_route_fops
= {
2580 .owner
= THIS_MODULE
,
2581 .open
= fib_route_seq_open
,
2583 .llseek
= seq_lseek
,
2584 .release
= seq_release_net
,
2587 int __net_init
fib_proc_init(struct net
*net
)
2589 if (!proc_create("fib_trie", S_IRUGO
, net
->proc_net
, &fib_trie_fops
))
2592 if (!proc_create("fib_triestat", S_IRUGO
, net
->proc_net
,
2593 &fib_triestat_fops
))
2596 if (!proc_create("route", S_IRUGO
, net
->proc_net
, &fib_route_fops
))
2602 remove_proc_entry("fib_triestat", net
->proc_net
);
2604 remove_proc_entry("fib_trie", net
->proc_net
);
2609 void __net_exit
fib_proc_exit(struct net
*net
)
2611 remove_proc_entry("fib_trie", net
->proc_net
);
2612 remove_proc_entry("fib_triestat", net
->proc_net
);
2613 remove_proc_entry("route", net
->proc_net
);
2616 #endif /* CONFIG_PROC_FS */