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/prefetch.h>
75 #include <linux/export.h>
76 #include <net/net_namespace.h>
78 #include <net/protocol.h>
79 #include <net/route.h>
82 #include <net/ip_fib.h>
83 #include "fib_lookup.h"
85 #define MAX_STAT_DEPTH 32
87 #define KEYLENGTH (8*sizeof(t_key))
89 typedef unsigned int t_key
;
93 #define NODE_TYPE_MASK 0x1UL
94 #define NODE_TYPE(node) ((node)->parent & NODE_TYPE_MASK)
96 #define IS_TNODE(n) (!(n->parent & T_LEAF))
97 #define IS_LEAF(n) (n->parent & T_LEAF)
100 unsigned long parent
;
105 unsigned long parent
;
107 struct hlist_head list
;
112 struct hlist_node hlist
;
114 u32 mask_plen
; /* ntohl(inet_make_mask(plen)) */
115 struct list_head falh
;
120 unsigned long parent
;
122 unsigned char pos
; /* 2log(KEYLENGTH) bits needed */
123 unsigned char bits
; /* 2log(KEYLENGTH) bits needed */
124 unsigned int full_children
; /* KEYLENGTH bits needed */
125 unsigned int empty_children
; /* KEYLENGTH bits needed */
128 struct tnode
*tnode_free
;
130 struct rt_trie_node __rcu
*child
[0];
133 #ifdef CONFIG_IP_FIB_TRIE_STATS
134 struct trie_use_stats
{
136 unsigned int backtrack
;
137 unsigned int semantic_match_passed
;
138 unsigned int semantic_match_miss
;
139 unsigned int null_node_hit
;
140 unsigned int resize_node_skipped
;
145 unsigned int totdepth
;
146 unsigned int maxdepth
;
149 unsigned int nullpointers
;
150 unsigned int prefixes
;
151 unsigned int nodesizes
[MAX_STAT_DEPTH
];
155 struct rt_trie_node __rcu
*trie
;
156 #ifdef CONFIG_IP_FIB_TRIE_STATS
157 struct trie_use_stats stats
;
161 static void tnode_put_child_reorg(struct tnode
*tn
, int i
, struct rt_trie_node
*n
,
163 static struct rt_trie_node
*resize(struct trie
*t
, struct tnode
*tn
);
164 static struct tnode
*inflate(struct trie
*t
, struct tnode
*tn
);
165 static struct tnode
*halve(struct trie
*t
, struct tnode
*tn
);
166 /* tnodes to free after resize(); protected by RTNL */
167 static struct tnode
*tnode_free_head
;
168 static size_t tnode_free_size
;
171 * synchronize_rcu after call_rcu for that many pages; it should be especially
172 * useful before resizing the root node with PREEMPT_NONE configs; the value was
173 * obtained experimentally, aiming to avoid visible slowdown.
175 static const int sync_pages
= 128;
177 static struct kmem_cache
*fn_alias_kmem __read_mostly
;
178 static struct kmem_cache
*trie_leaf_kmem __read_mostly
;
181 * caller must hold RTNL
183 static inline struct tnode
*node_parent(const struct rt_trie_node
*node
)
185 unsigned long parent
;
187 parent
= rcu_dereference_index_check(node
->parent
, lockdep_rtnl_is_held());
189 return (struct tnode
*)(parent
& ~NODE_TYPE_MASK
);
193 * caller must hold RCU read lock or RTNL
195 static inline struct tnode
*node_parent_rcu(const struct rt_trie_node
*node
)
197 unsigned long parent
;
199 parent
= rcu_dereference_index_check(node
->parent
, rcu_read_lock_held() ||
200 lockdep_rtnl_is_held());
202 return (struct tnode
*)(parent
& ~NODE_TYPE_MASK
);
205 /* Same as rcu_assign_pointer
206 * but that macro() assumes that value is a pointer.
208 static inline void node_set_parent(struct rt_trie_node
*node
, struct tnode
*ptr
)
211 node
->parent
= (unsigned long)ptr
| NODE_TYPE(node
);
215 * caller must hold RTNL
217 static inline struct rt_trie_node
*tnode_get_child(const struct tnode
*tn
, unsigned int i
)
219 BUG_ON(i
>= 1U << tn
->bits
);
221 return rtnl_dereference(tn
->child
[i
]);
225 * caller must hold RCU read lock or RTNL
227 static inline struct rt_trie_node
*tnode_get_child_rcu(const struct tnode
*tn
, unsigned int i
)
229 BUG_ON(i
>= 1U << tn
->bits
);
231 return rcu_dereference_rtnl(tn
->child
[i
]);
234 static inline int tnode_child_length(const struct tnode
*tn
)
236 return 1 << tn
->bits
;
239 static inline t_key
mask_pfx(t_key k
, unsigned int l
)
241 return (l
== 0) ? 0 : k
>> (KEYLENGTH
-l
) << (KEYLENGTH
-l
);
244 static inline t_key
tkey_extract_bits(t_key a
, unsigned int offset
, unsigned int bits
)
246 if (offset
< KEYLENGTH
)
247 return ((t_key
)(a
<< offset
)) >> (KEYLENGTH
- bits
);
252 static inline int tkey_equals(t_key a
, t_key b
)
257 static inline int tkey_sub_equals(t_key a
, int offset
, int bits
, t_key b
)
259 if (bits
== 0 || offset
>= KEYLENGTH
)
261 bits
= bits
> KEYLENGTH
? KEYLENGTH
: bits
;
262 return ((a
^ b
) << offset
) >> (KEYLENGTH
- bits
) == 0;
265 static inline int tkey_mismatch(t_key a
, int offset
, t_key b
)
272 while ((diff
<< i
) >> (KEYLENGTH
-1) == 0)
278 To understand this stuff, an understanding of keys and all their bits is
279 necessary. Every node in the trie has a key associated with it, but not
280 all of the bits in that key are significant.
282 Consider a node 'n' and its parent 'tp'.
284 If n is a leaf, every bit in its key is significant. Its presence is
285 necessitated by path compression, since during a tree traversal (when
286 searching for a leaf - unless we are doing an insertion) we will completely
287 ignore all skipped bits we encounter. Thus we need to verify, at the end of
288 a potentially successful search, that we have indeed been walking the
291 Note that we can never "miss" the correct key in the tree if present by
292 following the wrong path. Path compression ensures that segments of the key
293 that are the same for all keys with a given prefix are skipped, but the
294 skipped part *is* identical for each node in the subtrie below the skipped
295 bit! trie_insert() in this implementation takes care of that - note the
296 call to tkey_sub_equals() in trie_insert().
298 if n is an internal node - a 'tnode' here, the various parts of its key
299 have many different meanings.
302 _________________________________________________________________
303 | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
304 -----------------------------------------------------------------
305 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
307 _________________________________________________________________
308 | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
309 -----------------------------------------------------------------
310 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
317 First, let's just ignore the bits that come before the parent tp, that is
318 the bits from 0 to (tp->pos-1). They are *known* but at this point we do
319 not use them for anything.
321 The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
322 index into the parent's child array. That is, they will be used to find
323 'n' among tp's children.
325 The bits from (tp->pos + tp->bits) to (n->pos - 1) - "S" - are skipped bits
328 All the bits we have seen so far are significant to the node n. The rest
329 of the bits are really not needed or indeed known in n->key.
331 The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
332 n's child array, and will of course be different for each child.
335 The rest of the bits, from (n->pos + n->bits) onward, are completely unknown
340 static inline void check_tnode(const struct tnode
*tn
)
342 WARN_ON(tn
&& tn
->pos
+tn
->bits
> 32);
345 static const int halve_threshold
= 25;
346 static const int inflate_threshold
= 50;
347 static const int halve_threshold_root
= 15;
348 static const int inflate_threshold_root
= 30;
350 static void __alias_free_mem(struct rcu_head
*head
)
352 struct fib_alias
*fa
= container_of(head
, struct fib_alias
, rcu
);
353 kmem_cache_free(fn_alias_kmem
, fa
);
356 static inline void alias_free_mem_rcu(struct fib_alias
*fa
)
358 call_rcu(&fa
->rcu
, __alias_free_mem
);
361 static void __leaf_free_rcu(struct rcu_head
*head
)
363 struct leaf
*l
= container_of(head
, struct leaf
, rcu
);
364 kmem_cache_free(trie_leaf_kmem
, l
);
367 static inline void free_leaf(struct leaf
*l
)
369 call_rcu(&l
->rcu
, __leaf_free_rcu
);
372 static inline void free_leaf_info(struct leaf_info
*leaf
)
374 kfree_rcu(leaf
, rcu
);
377 static struct tnode
*tnode_alloc(size_t size
)
379 if (size
<= PAGE_SIZE
)
380 return kzalloc(size
, GFP_KERNEL
);
382 return vzalloc(size
);
385 static void __tnode_free_rcu(struct rcu_head
*head
)
387 struct tnode
*tn
= container_of(head
, struct tnode
, rcu
);
388 size_t size
= sizeof(struct tnode
) +
389 (sizeof(struct rt_trie_node
*) << tn
->bits
);
391 if (size
<= PAGE_SIZE
)
397 static inline void tnode_free(struct tnode
*tn
)
400 free_leaf((struct leaf
*) tn
);
402 call_rcu(&tn
->rcu
, __tnode_free_rcu
);
405 static void tnode_free_safe(struct tnode
*tn
)
408 tn
->tnode_free
= tnode_free_head
;
409 tnode_free_head
= tn
;
410 tnode_free_size
+= sizeof(struct tnode
) +
411 (sizeof(struct rt_trie_node
*) << tn
->bits
);
414 static void tnode_free_flush(void)
418 while ((tn
= tnode_free_head
)) {
419 tnode_free_head
= tn
->tnode_free
;
420 tn
->tnode_free
= NULL
;
424 if (tnode_free_size
>= PAGE_SIZE
* sync_pages
) {
430 static struct leaf
*leaf_new(void)
432 struct leaf
*l
= kmem_cache_alloc(trie_leaf_kmem
, GFP_KERNEL
);
435 INIT_HLIST_HEAD(&l
->list
);
440 static struct leaf_info
*leaf_info_new(int plen
)
442 struct leaf_info
*li
= kmalloc(sizeof(struct leaf_info
), GFP_KERNEL
);
445 li
->mask_plen
= ntohl(inet_make_mask(plen
));
446 INIT_LIST_HEAD(&li
->falh
);
451 static struct tnode
*tnode_new(t_key key
, int pos
, int bits
)
453 size_t sz
= sizeof(struct tnode
) + (sizeof(struct rt_trie_node
*) << bits
);
454 struct tnode
*tn
= tnode_alloc(sz
);
457 tn
->parent
= T_TNODE
;
461 tn
->full_children
= 0;
462 tn
->empty_children
= 1<<bits
;
465 pr_debug("AT %p s=%zu %zu\n", tn
, sizeof(struct tnode
),
466 sizeof(struct rt_trie_node
*) << bits
);
471 * Check whether a tnode 'n' is "full", i.e. it is an internal node
472 * and no bits are skipped. See discussion in dyntree paper p. 6
475 static inline int tnode_full(const struct tnode
*tn
, const struct rt_trie_node
*n
)
477 if (n
== NULL
|| IS_LEAF(n
))
480 return ((struct tnode
*) n
)->pos
== tn
->pos
+ tn
->bits
;
483 static inline void put_child(struct tnode
*tn
, int i
,
484 struct rt_trie_node
*n
)
486 tnode_put_child_reorg(tn
, i
, n
, -1);
490 * Add a child at position i overwriting the old value.
491 * Update the value of full_children and empty_children.
494 static void tnode_put_child_reorg(struct tnode
*tn
, int i
, struct rt_trie_node
*n
,
497 struct rt_trie_node
*chi
= rtnl_dereference(tn
->child
[i
]);
500 BUG_ON(i
>= 1<<tn
->bits
);
502 /* update emptyChildren */
503 if (n
== NULL
&& chi
!= NULL
)
504 tn
->empty_children
++;
505 else if (n
!= NULL
&& chi
== NULL
)
506 tn
->empty_children
--;
508 /* update fullChildren */
510 wasfull
= tnode_full(tn
, chi
);
512 isfull
= tnode_full(tn
, n
);
513 if (wasfull
&& !isfull
)
515 else if (!wasfull
&& isfull
)
519 node_set_parent(n
, tn
);
521 rcu_assign_pointer(tn
->child
[i
], n
);
525 static struct rt_trie_node
*resize(struct trie
*t
, struct tnode
*tn
)
528 struct tnode
*old_tn
;
529 int inflate_threshold_use
;
530 int halve_threshold_use
;
536 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
537 tn
, inflate_threshold
, halve_threshold
);
540 if (tn
->empty_children
== tnode_child_length(tn
)) {
545 if (tn
->empty_children
== tnode_child_length(tn
) - 1)
548 * Double as long as the resulting node has a number of
549 * nonempty nodes that are above the threshold.
553 * From "Implementing a dynamic compressed trie" by Stefan Nilsson of
554 * the Helsinki University of Technology and Matti Tikkanen of Nokia
555 * Telecommunications, page 6:
556 * "A node is doubled if the ratio of non-empty children to all
557 * children in the *doubled* node is at least 'high'."
559 * 'high' in this instance is the variable 'inflate_threshold'. It
560 * is expressed as a percentage, so we multiply it with
561 * tnode_child_length() and instead of multiplying by 2 (since the
562 * child array will be doubled by inflate()) and multiplying
563 * the left-hand side by 100 (to handle the percentage thing) we
564 * multiply the left-hand side by 50.
566 * The left-hand side may look a bit weird: tnode_child_length(tn)
567 * - tn->empty_children is of course the number of non-null children
568 * in the current node. tn->full_children is the number of "full"
569 * children, that is non-null tnodes with a skip value of 0.
570 * All of those will be doubled in the resulting inflated tnode, so
571 * we just count them one extra time here.
573 * A clearer way to write this would be:
575 * to_be_doubled = tn->full_children;
576 * not_to_be_doubled = tnode_child_length(tn) - tn->empty_children -
579 * new_child_length = tnode_child_length(tn) * 2;
581 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
583 * if (new_fill_factor >= inflate_threshold)
585 * ...and so on, tho it would mess up the while () loop.
588 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
592 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
593 * inflate_threshold * new_child_length
595 * expand not_to_be_doubled and to_be_doubled, and shorten:
596 * 100 * (tnode_child_length(tn) - tn->empty_children +
597 * tn->full_children) >= inflate_threshold * new_child_length
599 * expand new_child_length:
600 * 100 * (tnode_child_length(tn) - tn->empty_children +
601 * tn->full_children) >=
602 * inflate_threshold * tnode_child_length(tn) * 2
605 * 50 * (tn->full_children + tnode_child_length(tn) -
606 * tn->empty_children) >= inflate_threshold *
607 * tnode_child_length(tn)
613 /* Keep root node larger */
615 if (!node_parent((struct rt_trie_node
*)tn
)) {
616 inflate_threshold_use
= inflate_threshold_root
;
617 halve_threshold_use
= halve_threshold_root
;
619 inflate_threshold_use
= inflate_threshold
;
620 halve_threshold_use
= halve_threshold
;
624 while ((tn
->full_children
> 0 && max_work
-- &&
625 50 * (tn
->full_children
+ tnode_child_length(tn
)
626 - tn
->empty_children
)
627 >= inflate_threshold_use
* tnode_child_length(tn
))) {
634 #ifdef CONFIG_IP_FIB_TRIE_STATS
635 t
->stats
.resize_node_skipped
++;
643 /* Return if at least one inflate is run */
644 if (max_work
!= MAX_WORK
)
645 return (struct rt_trie_node
*) tn
;
648 * Halve as long as the number of empty children in this
649 * node is above threshold.
653 while (tn
->bits
> 1 && max_work
-- &&
654 100 * (tnode_child_length(tn
) - tn
->empty_children
) <
655 halve_threshold_use
* tnode_child_length(tn
)) {
661 #ifdef CONFIG_IP_FIB_TRIE_STATS
662 t
->stats
.resize_node_skipped
++;
669 /* Only one child remains */
670 if (tn
->empty_children
== tnode_child_length(tn
) - 1) {
672 for (i
= 0; i
< tnode_child_length(tn
); i
++) {
673 struct rt_trie_node
*n
;
675 n
= rtnl_dereference(tn
->child
[i
]);
679 /* compress one level */
681 node_set_parent(n
, NULL
);
686 return (struct rt_trie_node
*) tn
;
690 static void tnode_clean_free(struct tnode
*tn
)
693 struct tnode
*tofree
;
695 for (i
= 0; i
< tnode_child_length(tn
); i
++) {
696 tofree
= (struct tnode
*)rtnl_dereference(tn
->child
[i
]);
703 static struct tnode
*inflate(struct trie
*t
, struct tnode
*tn
)
705 struct tnode
*oldtnode
= tn
;
706 int olen
= tnode_child_length(tn
);
709 pr_debug("In inflate\n");
711 tn
= tnode_new(oldtnode
->key
, oldtnode
->pos
, oldtnode
->bits
+ 1);
714 return ERR_PTR(-ENOMEM
);
717 * Preallocate and store tnodes before the actual work so we
718 * don't get into an inconsistent state if memory allocation
719 * fails. In case of failure we return the oldnode and inflate
720 * of tnode is ignored.
723 for (i
= 0; i
< olen
; i
++) {
726 inode
= (struct tnode
*) tnode_get_child(oldtnode
, i
);
729 inode
->pos
== oldtnode
->pos
+ oldtnode
->bits
&&
731 struct tnode
*left
, *right
;
732 t_key m
= ~0U << (KEYLENGTH
- 1) >> inode
->pos
;
734 left
= tnode_new(inode
->key
&(~m
), inode
->pos
+ 1,
739 right
= tnode_new(inode
->key
|m
, inode
->pos
+ 1,
747 put_child(tn
, 2*i
, (struct rt_trie_node
*) left
);
748 put_child(tn
, 2*i
+1, (struct rt_trie_node
*) right
);
752 for (i
= 0; i
< olen
; i
++) {
754 struct rt_trie_node
*node
= tnode_get_child(oldtnode
, i
);
755 struct tnode
*left
, *right
;
762 /* A leaf or an internal node with skipped bits */
764 if (IS_LEAF(node
) || ((struct tnode
*) node
)->pos
>
765 tn
->pos
+ tn
->bits
- 1) {
766 if (tkey_extract_bits(node
->key
,
767 oldtnode
->pos
+ oldtnode
->bits
,
769 put_child(tn
, 2*i
, node
);
771 put_child(tn
, 2*i
+1, node
);
775 /* An internal node with two children */
776 inode
= (struct tnode
*) node
;
778 if (inode
->bits
== 1) {
779 put_child(tn
, 2*i
, rtnl_dereference(inode
->child
[0]));
780 put_child(tn
, 2*i
+1, rtnl_dereference(inode
->child
[1]));
782 tnode_free_safe(inode
);
786 /* An internal node with more than two children */
788 /* We will replace this node 'inode' with two new
789 * ones, 'left' and 'right', each with half of the
790 * original children. The two new nodes will have
791 * a position one bit further down the key and this
792 * means that the "significant" part of their keys
793 * (see the discussion near the top of this file)
794 * will differ by one bit, which will be "0" in
795 * left's key and "1" in right's key. Since we are
796 * moving the key position by one step, the bit that
797 * we are moving away from - the bit at position
798 * (inode->pos) - is the one that will differ between
799 * left and right. So... we synthesize that bit in the
801 * The mask 'm' below will be a single "one" bit at
802 * the position (inode->pos)
805 /* Use the old key, but set the new significant
809 left
= (struct tnode
*) tnode_get_child(tn
, 2*i
);
810 put_child(tn
, 2*i
, NULL
);
814 right
= (struct tnode
*) tnode_get_child(tn
, 2*i
+1);
815 put_child(tn
, 2*i
+1, NULL
);
819 size
= tnode_child_length(left
);
820 for (j
= 0; j
< size
; j
++) {
821 put_child(left
, j
, rtnl_dereference(inode
->child
[j
]));
822 put_child(right
, j
, rtnl_dereference(inode
->child
[j
+ size
]));
824 put_child(tn
, 2*i
, resize(t
, left
));
825 put_child(tn
, 2*i
+1, resize(t
, right
));
827 tnode_free_safe(inode
);
829 tnode_free_safe(oldtnode
);
832 tnode_clean_free(tn
);
833 return ERR_PTR(-ENOMEM
);
836 static struct tnode
*halve(struct trie
*t
, struct tnode
*tn
)
838 struct tnode
*oldtnode
= tn
;
839 struct rt_trie_node
*left
, *right
;
841 int olen
= tnode_child_length(tn
);
843 pr_debug("In halve\n");
845 tn
= tnode_new(oldtnode
->key
, oldtnode
->pos
, oldtnode
->bits
- 1);
848 return ERR_PTR(-ENOMEM
);
851 * Preallocate and store tnodes before the actual work so we
852 * don't get into an inconsistent state if memory allocation
853 * fails. In case of failure we return the oldnode and halve
854 * of tnode is ignored.
857 for (i
= 0; i
< olen
; i
+= 2) {
858 left
= tnode_get_child(oldtnode
, i
);
859 right
= tnode_get_child(oldtnode
, i
+1);
861 /* Two nonempty children */
865 newn
= tnode_new(left
->key
, tn
->pos
+ tn
->bits
, 1);
870 put_child(tn
, i
/2, (struct rt_trie_node
*)newn
);
875 for (i
= 0; i
< olen
; i
+= 2) {
876 struct tnode
*newBinNode
;
878 left
= tnode_get_child(oldtnode
, i
);
879 right
= tnode_get_child(oldtnode
, i
+1);
881 /* At least one of the children is empty */
883 if (right
== NULL
) /* Both are empty */
885 put_child(tn
, i
/2, right
);
890 put_child(tn
, i
/2, left
);
894 /* Two nonempty children */
895 newBinNode
= (struct tnode
*) tnode_get_child(tn
, i
/2);
896 put_child(tn
, i
/2, NULL
);
897 put_child(newBinNode
, 0, left
);
898 put_child(newBinNode
, 1, right
);
899 put_child(tn
, i
/2, resize(t
, newBinNode
));
901 tnode_free_safe(oldtnode
);
904 tnode_clean_free(tn
);
905 return ERR_PTR(-ENOMEM
);
908 /* readside must use rcu_read_lock currently dump routines
909 via get_fa_head and dump */
911 static struct leaf_info
*find_leaf_info(struct leaf
*l
, int plen
)
913 struct hlist_head
*head
= &l
->list
;
914 struct leaf_info
*li
;
916 hlist_for_each_entry_rcu(li
, head
, hlist
)
917 if (li
->plen
== plen
)
923 static inline struct list_head
*get_fa_head(struct leaf
*l
, int plen
)
925 struct leaf_info
*li
= find_leaf_info(l
, plen
);
933 static void insert_leaf_info(struct hlist_head
*head
, struct leaf_info
*new)
935 struct leaf_info
*li
= NULL
, *last
= NULL
;
937 if (hlist_empty(head
)) {
938 hlist_add_head_rcu(&new->hlist
, head
);
940 hlist_for_each_entry(li
, head
, hlist
) {
941 if (new->plen
> li
->plen
)
947 hlist_add_after_rcu(&last
->hlist
, &new->hlist
);
949 hlist_add_before_rcu(&new->hlist
, &li
->hlist
);
953 /* rcu_read_lock needs to be hold by caller from readside */
956 fib_find_node(struct trie
*t
, u32 key
)
960 struct rt_trie_node
*n
;
963 n
= rcu_dereference_rtnl(t
->trie
);
965 while (n
!= NULL
&& NODE_TYPE(n
) == T_TNODE
) {
966 tn
= (struct tnode
*) n
;
970 if (tkey_sub_equals(tn
->key
, pos
, tn
->pos
-pos
, key
)) {
971 pos
= tn
->pos
+ tn
->bits
;
972 n
= tnode_get_child_rcu(tn
,
973 tkey_extract_bits(key
,
979 /* Case we have found a leaf. Compare prefixes */
981 if (n
!= NULL
&& IS_LEAF(n
) && tkey_equals(key
, n
->key
))
982 return (struct leaf
*)n
;
987 static void trie_rebalance(struct trie
*t
, struct tnode
*tn
)
995 while (tn
!= NULL
&& (tp
= node_parent((struct rt_trie_node
*)tn
)) != NULL
) {
996 cindex
= tkey_extract_bits(key
, tp
->pos
, tp
->bits
);
997 wasfull
= tnode_full(tp
, tnode_get_child(tp
, cindex
));
998 tn
= (struct tnode
*)resize(t
, tn
);
1000 tnode_put_child_reorg(tp
, cindex
,
1001 (struct rt_trie_node
*)tn
, wasfull
);
1003 tp
= node_parent((struct rt_trie_node
*) tn
);
1005 rcu_assign_pointer(t
->trie
, (struct rt_trie_node
*)tn
);
1013 /* Handle last (top) tnode */
1015 tn
= (struct tnode
*)resize(t
, tn
);
1017 rcu_assign_pointer(t
->trie
, (struct rt_trie_node
*)tn
);
1021 /* only used from updater-side */
1023 static struct list_head
*fib_insert_node(struct trie
*t
, u32 key
, int plen
)
1026 struct tnode
*tp
= NULL
, *tn
= NULL
;
1027 struct rt_trie_node
*n
;
1030 struct list_head
*fa_head
= NULL
;
1031 struct leaf_info
*li
;
1035 n
= rtnl_dereference(t
->trie
);
1037 /* If we point to NULL, stop. Either the tree is empty and we should
1038 * just put a new leaf in if, or we have reached an empty child slot,
1039 * and we should just put our new leaf in that.
1040 * If we point to a T_TNODE, check if it matches our key. Note that
1041 * a T_TNODE might be skipping any number of bits - its 'pos' need
1042 * not be the parent's 'pos'+'bits'!
1044 * If it does match the current key, get pos/bits from it, extract
1045 * the index from our key, push the T_TNODE and walk the tree.
1047 * If it doesn't, we have to replace it with a new T_TNODE.
1049 * If we point to a T_LEAF, it might or might not have the same key
1050 * as we do. If it does, just change the value, update the T_LEAF's
1051 * value, and return it.
1052 * If it doesn't, we need to replace it with a T_TNODE.
1055 while (n
!= NULL
&& NODE_TYPE(n
) == T_TNODE
) {
1056 tn
= (struct tnode
*) n
;
1060 if (tkey_sub_equals(tn
->key
, pos
, tn
->pos
-pos
, key
)) {
1062 pos
= tn
->pos
+ tn
->bits
;
1063 n
= tnode_get_child(tn
,
1064 tkey_extract_bits(key
,
1068 BUG_ON(n
&& node_parent(n
) != tn
);
1074 * n ----> NULL, LEAF or TNODE
1076 * tp is n's (parent) ----> NULL or TNODE
1079 BUG_ON(tp
&& IS_LEAF(tp
));
1081 /* Case 1: n is a leaf. Compare prefixes */
1083 if (n
!= NULL
&& IS_LEAF(n
) && tkey_equals(key
, n
->key
)) {
1084 l
= (struct leaf
*) n
;
1085 li
= leaf_info_new(plen
);
1090 fa_head
= &li
->falh
;
1091 insert_leaf_info(&l
->list
, li
);
1100 li
= leaf_info_new(plen
);
1107 fa_head
= &li
->falh
;
1108 insert_leaf_info(&l
->list
, li
);
1110 if (t
->trie
&& n
== NULL
) {
1111 /* Case 2: n is NULL, and will just insert a new leaf */
1113 node_set_parent((struct rt_trie_node
*)l
, tp
);
1115 cindex
= tkey_extract_bits(key
, tp
->pos
, tp
->bits
);
1116 put_child(tp
, cindex
, (struct rt_trie_node
*)l
);
1118 /* Case 3: n is a LEAF or a TNODE and the key doesn't match. */
1120 * Add a new tnode here
1121 * first tnode need some special handling
1125 pos
= tp
->pos
+tp
->bits
;
1130 newpos
= tkey_mismatch(key
, pos
, n
->key
);
1131 tn
= tnode_new(n
->key
, newpos
, 1);
1134 tn
= tnode_new(key
, newpos
, 1); /* First tnode */
1143 node_set_parent((struct rt_trie_node
*)tn
, tp
);
1145 missbit
= tkey_extract_bits(key
, newpos
, 1);
1146 put_child(tn
, missbit
, (struct rt_trie_node
*)l
);
1147 put_child(tn
, 1-missbit
, n
);
1150 cindex
= tkey_extract_bits(key
, tp
->pos
, tp
->bits
);
1151 put_child(tp
, cindex
, (struct rt_trie_node
*)tn
);
1153 rcu_assign_pointer(t
->trie
, (struct rt_trie_node
*)tn
);
1158 if (tp
&& tp
->pos
+ tp
->bits
> 32)
1159 pr_warn("fib_trie tp=%p pos=%d, bits=%d, key=%0x plen=%d\n",
1160 tp
, tp
->pos
, tp
->bits
, key
, plen
);
1162 /* Rebalance the trie */
1164 trie_rebalance(t
, tp
);
1170 * Caller must hold RTNL.
1172 int fib_table_insert(struct fib_table
*tb
, struct fib_config
*cfg
)
1174 struct trie
*t
= (struct trie
*) tb
->tb_data
;
1175 struct fib_alias
*fa
, *new_fa
;
1176 struct list_head
*fa_head
= NULL
;
1177 struct fib_info
*fi
;
1178 int plen
= cfg
->fc_dst_len
;
1179 u8 tos
= cfg
->fc_tos
;
1187 key
= ntohl(cfg
->fc_dst
);
1189 pr_debug("Insert table=%u %08x/%d\n", tb
->tb_id
, key
, plen
);
1191 mask
= ntohl(inet_make_mask(plen
));
1198 fi
= fib_create_info(cfg
);
1204 l
= fib_find_node(t
, key
);
1208 fa_head
= get_fa_head(l
, plen
);
1209 fa
= fib_find_alias(fa_head
, tos
, fi
->fib_priority
);
1212 /* Now fa, if non-NULL, points to the first fib alias
1213 * with the same keys [prefix,tos,priority], if such key already
1214 * exists or to the node before which we will insert new one.
1216 * If fa is NULL, we will need to allocate a new one and
1217 * insert to the head of f.
1219 * If f is NULL, no fib node matched the destination key
1220 * and we need to allocate a new one of those as well.
1223 if (fa
&& fa
->fa_tos
== tos
&&
1224 fa
->fa_info
->fib_priority
== fi
->fib_priority
) {
1225 struct fib_alias
*fa_first
, *fa_match
;
1228 if (cfg
->fc_nlflags
& NLM_F_EXCL
)
1232 * 1. Find exact match for type, scope, fib_info to avoid
1234 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1238 fa
= list_entry(fa
->fa_list
.prev
, struct fib_alias
, fa_list
);
1239 list_for_each_entry_continue(fa
, fa_head
, fa_list
) {
1240 if (fa
->fa_tos
!= tos
)
1242 if (fa
->fa_info
->fib_priority
!= fi
->fib_priority
)
1244 if (fa
->fa_type
== cfg
->fc_type
&&
1245 fa
->fa_info
== fi
) {
1251 if (cfg
->fc_nlflags
& NLM_F_REPLACE
) {
1252 struct fib_info
*fi_drop
;
1262 new_fa
= kmem_cache_alloc(fn_alias_kmem
, GFP_KERNEL
);
1266 fi_drop
= fa
->fa_info
;
1267 new_fa
->fa_tos
= fa
->fa_tos
;
1268 new_fa
->fa_info
= fi
;
1269 new_fa
->fa_type
= cfg
->fc_type
;
1270 state
= fa
->fa_state
;
1271 new_fa
->fa_state
= state
& ~FA_S_ACCESSED
;
1273 list_replace_rcu(&fa
->fa_list
, &new_fa
->fa_list
);
1274 alias_free_mem_rcu(fa
);
1276 fib_release_info(fi_drop
);
1277 if (state
& FA_S_ACCESSED
)
1278 rt_cache_flush(cfg
->fc_nlinfo
.nl_net
);
1279 rtmsg_fib(RTM_NEWROUTE
, htonl(key
), new_fa
, plen
,
1280 tb
->tb_id
, &cfg
->fc_nlinfo
, NLM_F_REPLACE
);
1284 /* Error if we find a perfect match which
1285 * uses the same scope, type, and nexthop
1291 if (!(cfg
->fc_nlflags
& NLM_F_APPEND
))
1295 if (!(cfg
->fc_nlflags
& NLM_F_CREATE
))
1299 new_fa
= kmem_cache_alloc(fn_alias_kmem
, GFP_KERNEL
);
1303 new_fa
->fa_info
= fi
;
1304 new_fa
->fa_tos
= tos
;
1305 new_fa
->fa_type
= cfg
->fc_type
;
1306 new_fa
->fa_state
= 0;
1308 * Insert new entry to the list.
1312 fa_head
= fib_insert_node(t
, key
, plen
);
1313 if (unlikely(!fa_head
)) {
1315 goto out_free_new_fa
;
1320 tb
->tb_num_default
++;
1322 list_add_tail_rcu(&new_fa
->fa_list
,
1323 (fa
? &fa
->fa_list
: fa_head
));
1325 rt_cache_flush(cfg
->fc_nlinfo
.nl_net
);
1326 rtmsg_fib(RTM_NEWROUTE
, htonl(key
), new_fa
, plen
, tb
->tb_id
,
1327 &cfg
->fc_nlinfo
, 0);
1332 kmem_cache_free(fn_alias_kmem
, new_fa
);
1334 fib_release_info(fi
);
1339 /* should be called with rcu_read_lock */
1340 static int check_leaf(struct fib_table
*tb
, struct trie
*t
, struct leaf
*l
,
1341 t_key key
, const struct flowi4
*flp
,
1342 struct fib_result
*res
, int fib_flags
)
1344 struct leaf_info
*li
;
1345 struct hlist_head
*hhead
= &l
->list
;
1347 hlist_for_each_entry_rcu(li
, hhead
, hlist
) {
1348 struct fib_alias
*fa
;
1350 if (l
->key
!= (key
& li
->mask_plen
))
1353 list_for_each_entry_rcu(fa
, &li
->falh
, fa_list
) {
1354 struct fib_info
*fi
= fa
->fa_info
;
1357 if (fa
->fa_tos
&& fa
->fa_tos
!= flp
->flowi4_tos
)
1361 if (fa
->fa_info
->fib_scope
< flp
->flowi4_scope
)
1363 fib_alias_accessed(fa
);
1364 err
= fib_props
[fa
->fa_type
].error
;
1366 #ifdef CONFIG_IP_FIB_TRIE_STATS
1367 t
->stats
.semantic_match_passed
++;
1371 if (fi
->fib_flags
& RTNH_F_DEAD
)
1373 for (nhsel
= 0; nhsel
< fi
->fib_nhs
; nhsel
++) {
1374 const struct fib_nh
*nh
= &fi
->fib_nh
[nhsel
];
1376 if (nh
->nh_flags
& RTNH_F_DEAD
)
1378 if (flp
->flowi4_oif
&& flp
->flowi4_oif
!= nh
->nh_oif
)
1381 #ifdef CONFIG_IP_FIB_TRIE_STATS
1382 t
->stats
.semantic_match_passed
++;
1384 res
->prefixlen
= li
->plen
;
1385 res
->nh_sel
= nhsel
;
1386 res
->type
= fa
->fa_type
;
1387 res
->scope
= fa
->fa_info
->fib_scope
;
1390 res
->fa_head
= &li
->falh
;
1391 if (!(fib_flags
& FIB_LOOKUP_NOREF
))
1392 atomic_inc(&fi
->fib_clntref
);
1397 #ifdef CONFIG_IP_FIB_TRIE_STATS
1398 t
->stats
.semantic_match_miss
++;
1405 int fib_table_lookup(struct fib_table
*tb
, const struct flowi4
*flp
,
1406 struct fib_result
*res
, int fib_flags
)
1408 struct trie
*t
= (struct trie
*) tb
->tb_data
;
1410 struct rt_trie_node
*n
;
1412 unsigned int pos
, bits
;
1413 t_key key
= ntohl(flp
->daddr
);
1414 unsigned int chopped_off
;
1416 unsigned int current_prefix_length
= KEYLENGTH
;
1418 t_key pref_mismatch
;
1422 n
= rcu_dereference(t
->trie
);
1426 #ifdef CONFIG_IP_FIB_TRIE_STATS
1432 ret
= check_leaf(tb
, t
, (struct leaf
*)n
, key
, flp
, res
, fib_flags
);
1436 pn
= (struct tnode
*) n
;
1444 cindex
= tkey_extract_bits(mask_pfx(key
, current_prefix_length
),
1447 n
= tnode_get_child_rcu(pn
, cindex
);
1450 #ifdef CONFIG_IP_FIB_TRIE_STATS
1451 t
->stats
.null_node_hit
++;
1457 ret
= check_leaf(tb
, t
, (struct leaf
*)n
, key
, flp
, res
, fib_flags
);
1463 cn
= (struct tnode
*)n
;
1466 * It's a tnode, and we can do some extra checks here if we
1467 * like, to avoid descending into a dead-end branch.
1468 * This tnode is in the parent's child array at index
1469 * key[p_pos..p_pos+p_bits] but potentially with some bits
1470 * chopped off, so in reality the index may be just a
1471 * subprefix, padded with zero at the end.
1472 * We can also take a look at any skipped bits in this
1473 * tnode - everything up to p_pos is supposed to be ok,
1474 * and the non-chopped bits of the index (se previous
1475 * paragraph) are also guaranteed ok, but the rest is
1476 * considered unknown.
1478 * The skipped bits are key[pos+bits..cn->pos].
1481 /* If current_prefix_length < pos+bits, we are already doing
1482 * actual prefix matching, which means everything from
1483 * pos+(bits-chopped_off) onward must be zero along some
1484 * branch of this subtree - otherwise there is *no* valid
1485 * prefix present. Here we can only check the skipped
1486 * bits. Remember, since we have already indexed into the
1487 * parent's child array, we know that the bits we chopped of
1491 /* NOTA BENE: Checking only skipped bits
1492 for the new node here */
1494 if (current_prefix_length
< pos
+bits
) {
1495 if (tkey_extract_bits(cn
->key
, current_prefix_length
,
1496 cn
->pos
- current_prefix_length
)
1502 * If chopped_off=0, the index is fully validated and we
1503 * only need to look at the skipped bits for this, the new,
1504 * tnode. What we actually want to do is to find out if
1505 * these skipped bits match our key perfectly, or if we will
1506 * have to count on finding a matching prefix further down,
1507 * because if we do, we would like to have some way of
1508 * verifying the existence of such a prefix at this point.
1511 /* The only thing we can do at this point is to verify that
1512 * any such matching prefix can indeed be a prefix to our
1513 * key, and if the bits in the node we are inspecting that
1514 * do not match our key are not ZERO, this cannot be true.
1515 * Thus, find out where there is a mismatch (before cn->pos)
1516 * and verify that all the mismatching bits are zero in the
1521 * Note: We aren't very concerned about the piece of
1522 * the key that precede pn->pos+pn->bits, since these
1523 * have already been checked. The bits after cn->pos
1524 * aren't checked since these are by definition
1525 * "unknown" at this point. Thus, what we want to see
1526 * is if we are about to enter the "prefix matching"
1527 * state, and in that case verify that the skipped
1528 * bits that will prevail throughout this subtree are
1529 * zero, as they have to be if we are to find a
1533 pref_mismatch
= mask_pfx(cn
->key
^ key
, cn
->pos
);
1536 * In short: If skipped bits in this node do not match
1537 * the search key, enter the "prefix matching"
1540 if (pref_mismatch
) {
1541 /* fls(x) = __fls(x) + 1 */
1542 int mp
= KEYLENGTH
- __fls(pref_mismatch
) - 1;
1544 if (tkey_extract_bits(cn
->key
, mp
, cn
->pos
- mp
) != 0)
1547 if (current_prefix_length
>= cn
->pos
)
1548 current_prefix_length
= mp
;
1551 pn
= (struct tnode
*)n
; /* Descend */
1558 /* As zero don't change the child key (cindex) */
1559 while ((chopped_off
<= pn
->bits
)
1560 && !(cindex
& (1<<(chopped_off
-1))))
1563 /* Decrease current_... with bits chopped off */
1564 if (current_prefix_length
> pn
->pos
+ pn
->bits
- chopped_off
)
1565 current_prefix_length
= pn
->pos
+ pn
->bits
1569 * Either we do the actual chop off according or if we have
1570 * chopped off all bits in this tnode walk up to our parent.
1573 if (chopped_off
<= pn
->bits
) {
1574 cindex
&= ~(1 << (chopped_off
-1));
1576 struct tnode
*parent
= node_parent_rcu((struct rt_trie_node
*) pn
);
1580 /* Get Child's index */
1581 cindex
= tkey_extract_bits(pn
->key
, parent
->pos
, parent
->bits
);
1585 #ifdef CONFIG_IP_FIB_TRIE_STATS
1586 t
->stats
.backtrack
++;
1597 EXPORT_SYMBOL_GPL(fib_table_lookup
);
1600 * Remove the leaf and return parent.
1602 static void trie_leaf_remove(struct trie
*t
, struct leaf
*l
)
1604 struct tnode
*tp
= node_parent((struct rt_trie_node
*) l
);
1606 pr_debug("entering trie_leaf_remove(%p)\n", l
);
1609 t_key cindex
= tkey_extract_bits(l
->key
, tp
->pos
, tp
->bits
);
1610 put_child(tp
, cindex
, NULL
);
1611 trie_rebalance(t
, tp
);
1613 RCU_INIT_POINTER(t
->trie
, NULL
);
1619 * Caller must hold RTNL.
1621 int fib_table_delete(struct fib_table
*tb
, struct fib_config
*cfg
)
1623 struct trie
*t
= (struct trie
*) tb
->tb_data
;
1625 int plen
= cfg
->fc_dst_len
;
1626 u8 tos
= cfg
->fc_tos
;
1627 struct fib_alias
*fa
, *fa_to_delete
;
1628 struct list_head
*fa_head
;
1630 struct leaf_info
*li
;
1635 key
= ntohl(cfg
->fc_dst
);
1636 mask
= ntohl(inet_make_mask(plen
));
1642 l
= fib_find_node(t
, key
);
1647 li
= find_leaf_info(l
, plen
);
1652 fa_head
= &li
->falh
;
1653 fa
= fib_find_alias(fa_head
, tos
, 0);
1658 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key
, plen
, tos
, t
);
1660 fa_to_delete
= NULL
;
1661 fa
= list_entry(fa
->fa_list
.prev
, struct fib_alias
, fa_list
);
1662 list_for_each_entry_continue(fa
, fa_head
, fa_list
) {
1663 struct fib_info
*fi
= fa
->fa_info
;
1665 if (fa
->fa_tos
!= tos
)
1668 if ((!cfg
->fc_type
|| fa
->fa_type
== cfg
->fc_type
) &&
1669 (cfg
->fc_scope
== RT_SCOPE_NOWHERE
||
1670 fa
->fa_info
->fib_scope
== cfg
->fc_scope
) &&
1671 (!cfg
->fc_prefsrc
||
1672 fi
->fib_prefsrc
== cfg
->fc_prefsrc
) &&
1673 (!cfg
->fc_protocol
||
1674 fi
->fib_protocol
== cfg
->fc_protocol
) &&
1675 fib_nh_match(cfg
, fi
) == 0) {
1685 rtmsg_fib(RTM_DELROUTE
, htonl(key
), fa
, plen
, tb
->tb_id
,
1686 &cfg
->fc_nlinfo
, 0);
1688 list_del_rcu(&fa
->fa_list
);
1691 tb
->tb_num_default
--;
1693 if (list_empty(fa_head
)) {
1694 hlist_del_rcu(&li
->hlist
);
1698 if (hlist_empty(&l
->list
))
1699 trie_leaf_remove(t
, l
);
1701 if (fa
->fa_state
& FA_S_ACCESSED
)
1702 rt_cache_flush(cfg
->fc_nlinfo
.nl_net
);
1704 fib_release_info(fa
->fa_info
);
1705 alias_free_mem_rcu(fa
);
1709 static int trie_flush_list(struct list_head
*head
)
1711 struct fib_alias
*fa
, *fa_node
;
1714 list_for_each_entry_safe(fa
, fa_node
, head
, fa_list
) {
1715 struct fib_info
*fi
= fa
->fa_info
;
1717 if (fi
&& (fi
->fib_flags
& RTNH_F_DEAD
)) {
1718 list_del_rcu(&fa
->fa_list
);
1719 fib_release_info(fa
->fa_info
);
1720 alias_free_mem_rcu(fa
);
1727 static int trie_flush_leaf(struct leaf
*l
)
1730 struct hlist_head
*lih
= &l
->list
;
1731 struct hlist_node
*tmp
;
1732 struct leaf_info
*li
= NULL
;
1734 hlist_for_each_entry_safe(li
, tmp
, lih
, hlist
) {
1735 found
+= trie_flush_list(&li
->falh
);
1737 if (list_empty(&li
->falh
)) {
1738 hlist_del_rcu(&li
->hlist
);
1746 * Scan for the next right leaf starting at node p->child[idx]
1747 * Since we have back pointer, no recursion necessary.
1749 static struct leaf
*leaf_walk_rcu(struct tnode
*p
, struct rt_trie_node
*c
)
1755 idx
= tkey_extract_bits(c
->key
, p
->pos
, p
->bits
) + 1;
1759 while (idx
< 1u << p
->bits
) {
1760 c
= tnode_get_child_rcu(p
, idx
++);
1765 prefetch(rcu_dereference_rtnl(p
->child
[idx
]));
1766 return (struct leaf
*) c
;
1769 /* Rescan start scanning in new node */
1770 p
= (struct tnode
*) c
;
1774 /* Node empty, walk back up to parent */
1775 c
= (struct rt_trie_node
*) p
;
1776 } while ((p
= node_parent_rcu(c
)) != NULL
);
1778 return NULL
; /* Root of trie */
1781 static struct leaf
*trie_firstleaf(struct trie
*t
)
1783 struct tnode
*n
= (struct tnode
*)rcu_dereference_rtnl(t
->trie
);
1788 if (IS_LEAF(n
)) /* trie is just a leaf */
1789 return (struct leaf
*) n
;
1791 return leaf_walk_rcu(n
, NULL
);
1794 static struct leaf
*trie_nextleaf(struct leaf
*l
)
1796 struct rt_trie_node
*c
= (struct rt_trie_node
*) l
;
1797 struct tnode
*p
= node_parent_rcu(c
);
1800 return NULL
; /* trie with just one leaf */
1802 return leaf_walk_rcu(p
, c
);
1805 static struct leaf
*trie_leafindex(struct trie
*t
, int index
)
1807 struct leaf
*l
= trie_firstleaf(t
);
1809 while (l
&& index
-- > 0)
1810 l
= trie_nextleaf(l
);
1817 * Caller must hold RTNL.
1819 int fib_table_flush(struct fib_table
*tb
)
1821 struct trie
*t
= (struct trie
*) tb
->tb_data
;
1822 struct leaf
*l
, *ll
= NULL
;
1825 for (l
= trie_firstleaf(t
); l
; l
= trie_nextleaf(l
)) {
1826 found
+= trie_flush_leaf(l
);
1828 if (ll
&& hlist_empty(&ll
->list
))
1829 trie_leaf_remove(t
, ll
);
1833 if (ll
&& hlist_empty(&ll
->list
))
1834 trie_leaf_remove(t
, ll
);
1836 pr_debug("trie_flush found=%d\n", found
);
1840 void fib_free_table(struct fib_table
*tb
)
1845 static int fn_trie_dump_fa(t_key key
, int plen
, struct list_head
*fah
,
1846 struct fib_table
*tb
,
1847 struct sk_buff
*skb
, struct netlink_callback
*cb
)
1850 struct fib_alias
*fa
;
1851 __be32 xkey
= htonl(key
);
1856 /* rcu_read_lock is hold by caller */
1858 list_for_each_entry_rcu(fa
, fah
, fa_list
) {
1864 if (fib_dump_info(skb
, NETLINK_CB(cb
->skb
).portid
,
1872 fa
->fa_info
, NLM_F_MULTI
) < 0) {
1882 static int fn_trie_dump_leaf(struct leaf
*l
, struct fib_table
*tb
,
1883 struct sk_buff
*skb
, struct netlink_callback
*cb
)
1885 struct leaf_info
*li
;
1891 /* rcu_read_lock is hold by caller */
1892 hlist_for_each_entry_rcu(li
, &l
->list
, hlist
) {
1901 if (list_empty(&li
->falh
))
1904 if (fn_trie_dump_fa(l
->key
, li
->plen
, &li
->falh
, tb
, skb
, cb
) < 0) {
1915 int fib_table_dump(struct fib_table
*tb
, struct sk_buff
*skb
,
1916 struct netlink_callback
*cb
)
1919 struct trie
*t
= (struct trie
*) tb
->tb_data
;
1920 t_key key
= cb
->args
[2];
1921 int count
= cb
->args
[3];
1924 /* Dump starting at last key.
1925 * Note: 0.0.0.0/0 (ie default) is first key.
1928 l
= trie_firstleaf(t
);
1930 /* Normally, continue from last key, but if that is missing
1931 * fallback to using slow rescan
1933 l
= fib_find_node(t
, key
);
1935 l
= trie_leafindex(t
, count
);
1939 cb
->args
[2] = l
->key
;
1940 if (fn_trie_dump_leaf(l
, tb
, skb
, cb
) < 0) {
1941 cb
->args
[3] = count
;
1947 l
= trie_nextleaf(l
);
1948 memset(&cb
->args
[4], 0,
1949 sizeof(cb
->args
) - 4*sizeof(cb
->args
[0]));
1951 cb
->args
[3] = count
;
1957 void __init
fib_trie_init(void)
1959 fn_alias_kmem
= kmem_cache_create("ip_fib_alias",
1960 sizeof(struct fib_alias
),
1961 0, SLAB_PANIC
, NULL
);
1963 trie_leaf_kmem
= kmem_cache_create("ip_fib_trie",
1964 max(sizeof(struct leaf
),
1965 sizeof(struct leaf_info
)),
1966 0, SLAB_PANIC
, NULL
);
1970 struct fib_table
*fib_trie_table(u32 id
)
1972 struct fib_table
*tb
;
1975 tb
= kmalloc(sizeof(struct fib_table
) + sizeof(struct trie
),
1981 tb
->tb_default
= -1;
1982 tb
->tb_num_default
= 0;
1984 t
= (struct trie
*) tb
->tb_data
;
1985 memset(t
, 0, sizeof(*t
));
1990 #ifdef CONFIG_PROC_FS
1991 /* Depth first Trie walk iterator */
1992 struct fib_trie_iter
{
1993 struct seq_net_private p
;
1994 struct fib_table
*tb
;
1995 struct tnode
*tnode
;
2000 static struct rt_trie_node
*fib_trie_get_next(struct fib_trie_iter
*iter
)
2002 struct tnode
*tn
= iter
->tnode
;
2003 unsigned int cindex
= iter
->index
;
2006 /* A single entry routing table */
2010 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2011 iter
->tnode
, iter
->index
, iter
->depth
);
2013 while (cindex
< (1<<tn
->bits
)) {
2014 struct rt_trie_node
*n
= tnode_get_child_rcu(tn
, cindex
);
2019 iter
->index
= cindex
+ 1;
2021 /* push down one level */
2022 iter
->tnode
= (struct tnode
*) n
;
2032 /* Current node exhausted, pop back up */
2033 p
= node_parent_rcu((struct rt_trie_node
*)tn
);
2035 cindex
= tkey_extract_bits(tn
->key
, p
->pos
, p
->bits
)+1;
2045 static struct rt_trie_node
*fib_trie_get_first(struct fib_trie_iter
*iter
,
2048 struct rt_trie_node
*n
;
2053 n
= rcu_dereference(t
->trie
);
2058 iter
->tnode
= (struct tnode
*) n
;
2070 static void trie_collect_stats(struct trie
*t
, struct trie_stat
*s
)
2072 struct rt_trie_node
*n
;
2073 struct fib_trie_iter iter
;
2075 memset(s
, 0, sizeof(*s
));
2078 for (n
= fib_trie_get_first(&iter
, t
); n
; n
= fib_trie_get_next(&iter
)) {
2080 struct leaf
*l
= (struct leaf
*)n
;
2081 struct leaf_info
*li
;
2084 s
->totdepth
+= iter
.depth
;
2085 if (iter
.depth
> s
->maxdepth
)
2086 s
->maxdepth
= iter
.depth
;
2088 hlist_for_each_entry_rcu(li
, &l
->list
, hlist
)
2091 const struct tnode
*tn
= (const struct tnode
*) n
;
2095 if (tn
->bits
< MAX_STAT_DEPTH
)
2096 s
->nodesizes
[tn
->bits
]++;
2098 for (i
= 0; i
< (1<<tn
->bits
); i
++)
2107 * This outputs /proc/net/fib_triestats
2109 static void trie_show_stats(struct seq_file
*seq
, struct trie_stat
*stat
)
2111 unsigned int i
, max
, pointers
, bytes
, avdepth
;
2114 avdepth
= stat
->totdepth
*100 / stat
->leaves
;
2118 seq_printf(seq
, "\tAver depth: %u.%02d\n",
2119 avdepth
/ 100, avdepth
% 100);
2120 seq_printf(seq
, "\tMax depth: %u\n", stat
->maxdepth
);
2122 seq_printf(seq
, "\tLeaves: %u\n", stat
->leaves
);
2123 bytes
= sizeof(struct leaf
) * stat
->leaves
;
2125 seq_printf(seq
, "\tPrefixes: %u\n", stat
->prefixes
);
2126 bytes
+= sizeof(struct leaf_info
) * stat
->prefixes
;
2128 seq_printf(seq
, "\tInternal nodes: %u\n\t", stat
->tnodes
);
2129 bytes
+= sizeof(struct tnode
) * stat
->tnodes
;
2131 max
= MAX_STAT_DEPTH
;
2132 while (max
> 0 && stat
->nodesizes
[max
-1] == 0)
2136 for (i
= 1; i
<= max
; i
++)
2137 if (stat
->nodesizes
[i
] != 0) {
2138 seq_printf(seq
, " %u: %u", i
, stat
->nodesizes
[i
]);
2139 pointers
+= (1<<i
) * stat
->nodesizes
[i
];
2141 seq_putc(seq
, '\n');
2142 seq_printf(seq
, "\tPointers: %u\n", pointers
);
2144 bytes
+= sizeof(struct rt_trie_node
*) * pointers
;
2145 seq_printf(seq
, "Null ptrs: %u\n", stat
->nullpointers
);
2146 seq_printf(seq
, "Total size: %u kB\n", (bytes
+ 1023) / 1024);
2149 #ifdef CONFIG_IP_FIB_TRIE_STATS
2150 static void trie_show_usage(struct seq_file
*seq
,
2151 const struct trie_use_stats
*stats
)
2153 seq_printf(seq
, "\nCounters:\n---------\n");
2154 seq_printf(seq
, "gets = %u\n", stats
->gets
);
2155 seq_printf(seq
, "backtracks = %u\n", stats
->backtrack
);
2156 seq_printf(seq
, "semantic match passed = %u\n",
2157 stats
->semantic_match_passed
);
2158 seq_printf(seq
, "semantic match miss = %u\n",
2159 stats
->semantic_match_miss
);
2160 seq_printf(seq
, "null node hit= %u\n", stats
->null_node_hit
);
2161 seq_printf(seq
, "skipped node resize = %u\n\n",
2162 stats
->resize_node_skipped
);
2164 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2166 static void fib_table_print(struct seq_file
*seq
, struct fib_table
*tb
)
2168 if (tb
->tb_id
== RT_TABLE_LOCAL
)
2169 seq_puts(seq
, "Local:\n");
2170 else if (tb
->tb_id
== RT_TABLE_MAIN
)
2171 seq_puts(seq
, "Main:\n");
2173 seq_printf(seq
, "Id %d:\n", tb
->tb_id
);
2177 static int fib_triestat_seq_show(struct seq_file
*seq
, void *v
)
2179 struct net
*net
= (struct net
*)seq
->private;
2183 "Basic info: size of leaf:"
2184 " %Zd bytes, size of tnode: %Zd bytes.\n",
2185 sizeof(struct leaf
), sizeof(struct tnode
));
2187 for (h
= 0; h
< FIB_TABLE_HASHSZ
; h
++) {
2188 struct hlist_head
*head
= &net
->ipv4
.fib_table_hash
[h
];
2189 struct fib_table
*tb
;
2191 hlist_for_each_entry_rcu(tb
, head
, tb_hlist
) {
2192 struct trie
*t
= (struct trie
*) tb
->tb_data
;
2193 struct trie_stat stat
;
2198 fib_table_print(seq
, tb
);
2200 trie_collect_stats(t
, &stat
);
2201 trie_show_stats(seq
, &stat
);
2202 #ifdef CONFIG_IP_FIB_TRIE_STATS
2203 trie_show_usage(seq
, &t
->stats
);
2211 static int fib_triestat_seq_open(struct inode
*inode
, struct file
*file
)
2213 return single_open_net(inode
, file
, fib_triestat_seq_show
);
2216 static const struct file_operations fib_triestat_fops
= {
2217 .owner
= THIS_MODULE
,
2218 .open
= fib_triestat_seq_open
,
2220 .llseek
= seq_lseek
,
2221 .release
= single_release_net
,
2224 static struct rt_trie_node
*fib_trie_get_idx(struct seq_file
*seq
, loff_t pos
)
2226 struct fib_trie_iter
*iter
= seq
->private;
2227 struct net
*net
= seq_file_net(seq
);
2231 for (h
= 0; h
< FIB_TABLE_HASHSZ
; h
++) {
2232 struct hlist_head
*head
= &net
->ipv4
.fib_table_hash
[h
];
2233 struct fib_table
*tb
;
2235 hlist_for_each_entry_rcu(tb
, head
, tb_hlist
) {
2236 struct rt_trie_node
*n
;
2238 for (n
= fib_trie_get_first(iter
,
2239 (struct trie
*) tb
->tb_data
);
2240 n
; n
= fib_trie_get_next(iter
))
2251 static void *fib_trie_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2255 return fib_trie_get_idx(seq
, *pos
);
2258 static void *fib_trie_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2260 struct fib_trie_iter
*iter
= seq
->private;
2261 struct net
*net
= seq_file_net(seq
);
2262 struct fib_table
*tb
= iter
->tb
;
2263 struct hlist_node
*tb_node
;
2265 struct rt_trie_node
*n
;
2268 /* next node in same table */
2269 n
= fib_trie_get_next(iter
);
2273 /* walk rest of this hash chain */
2274 h
= tb
->tb_id
& (FIB_TABLE_HASHSZ
- 1);
2275 while ((tb_node
= rcu_dereference(hlist_next_rcu(&tb
->tb_hlist
)))) {
2276 tb
= hlist_entry(tb_node
, struct fib_table
, tb_hlist
);
2277 n
= fib_trie_get_first(iter
, (struct trie
*) tb
->tb_data
);
2282 /* new hash chain */
2283 while (++h
< FIB_TABLE_HASHSZ
) {
2284 struct hlist_head
*head
= &net
->ipv4
.fib_table_hash
[h
];
2285 hlist_for_each_entry_rcu(tb
, head
, tb_hlist
) {
2286 n
= fib_trie_get_first(iter
, (struct trie
*) tb
->tb_data
);
2298 static void fib_trie_seq_stop(struct seq_file
*seq
, void *v
)
2304 static void seq_indent(struct seq_file
*seq
, int n
)
2310 static inline const char *rtn_scope(char *buf
, size_t len
, enum rt_scope_t s
)
2313 case RT_SCOPE_UNIVERSE
: return "universe";
2314 case RT_SCOPE_SITE
: return "site";
2315 case RT_SCOPE_LINK
: return "link";
2316 case RT_SCOPE_HOST
: return "host";
2317 case RT_SCOPE_NOWHERE
: return "nowhere";
2319 snprintf(buf
, len
, "scope=%d", s
);
2324 static const char *const rtn_type_names
[__RTN_MAX
] = {
2325 [RTN_UNSPEC
] = "UNSPEC",
2326 [RTN_UNICAST
] = "UNICAST",
2327 [RTN_LOCAL
] = "LOCAL",
2328 [RTN_BROADCAST
] = "BROADCAST",
2329 [RTN_ANYCAST
] = "ANYCAST",
2330 [RTN_MULTICAST
] = "MULTICAST",
2331 [RTN_BLACKHOLE
] = "BLACKHOLE",
2332 [RTN_UNREACHABLE
] = "UNREACHABLE",
2333 [RTN_PROHIBIT
] = "PROHIBIT",
2334 [RTN_THROW
] = "THROW",
2336 [RTN_XRESOLVE
] = "XRESOLVE",
2339 static inline const char *rtn_type(char *buf
, size_t len
, unsigned int t
)
2341 if (t
< __RTN_MAX
&& rtn_type_names
[t
])
2342 return rtn_type_names
[t
];
2343 snprintf(buf
, len
, "type %u", t
);
2347 /* Pretty print the trie */
2348 static int fib_trie_seq_show(struct seq_file
*seq
, void *v
)
2350 const struct fib_trie_iter
*iter
= seq
->private;
2351 struct rt_trie_node
*n
= v
;
2353 if (!node_parent_rcu(n
))
2354 fib_table_print(seq
, iter
->tb
);
2357 struct tnode
*tn
= (struct tnode
*) n
;
2358 __be32 prf
= htonl(mask_pfx(tn
->key
, tn
->pos
));
2360 seq_indent(seq
, iter
->depth
-1);
2361 seq_printf(seq
, " +-- %pI4/%d %d %d %d\n",
2362 &prf
, tn
->pos
, tn
->bits
, tn
->full_children
,
2363 tn
->empty_children
);
2366 struct leaf
*l
= (struct leaf
*) n
;
2367 struct leaf_info
*li
;
2368 __be32 val
= htonl(l
->key
);
2370 seq_indent(seq
, iter
->depth
);
2371 seq_printf(seq
, " |-- %pI4\n", &val
);
2373 hlist_for_each_entry_rcu(li
, &l
->list
, hlist
) {
2374 struct fib_alias
*fa
;
2376 list_for_each_entry_rcu(fa
, &li
->falh
, fa_list
) {
2377 char buf1
[32], buf2
[32];
2379 seq_indent(seq
, iter
->depth
+1);
2380 seq_printf(seq
, " /%d %s %s", li
->plen
,
2381 rtn_scope(buf1
, sizeof(buf1
),
2382 fa
->fa_info
->fib_scope
),
2383 rtn_type(buf2
, sizeof(buf2
),
2386 seq_printf(seq
, " tos=%d", fa
->fa_tos
);
2387 seq_putc(seq
, '\n');
2395 static const struct seq_operations fib_trie_seq_ops
= {
2396 .start
= fib_trie_seq_start
,
2397 .next
= fib_trie_seq_next
,
2398 .stop
= fib_trie_seq_stop
,
2399 .show
= fib_trie_seq_show
,
2402 static int fib_trie_seq_open(struct inode
*inode
, struct file
*file
)
2404 return seq_open_net(inode
, file
, &fib_trie_seq_ops
,
2405 sizeof(struct fib_trie_iter
));
2408 static const struct file_operations fib_trie_fops
= {
2409 .owner
= THIS_MODULE
,
2410 .open
= fib_trie_seq_open
,
2412 .llseek
= seq_lseek
,
2413 .release
= seq_release_net
,
2416 struct fib_route_iter
{
2417 struct seq_net_private p
;
2418 struct trie
*main_trie
;
2423 static struct leaf
*fib_route_get_idx(struct fib_route_iter
*iter
, loff_t pos
)
2425 struct leaf
*l
= NULL
;
2426 struct trie
*t
= iter
->main_trie
;
2428 /* use cache location of last found key */
2429 if (iter
->pos
> 0 && pos
>= iter
->pos
&& (l
= fib_find_node(t
, iter
->key
)))
2433 l
= trie_firstleaf(t
);
2436 while (l
&& pos
-- > 0) {
2438 l
= trie_nextleaf(l
);
2442 iter
->key
= pos
; /* remember it */
2444 iter
->pos
= 0; /* forget it */
2449 static void *fib_route_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2452 struct fib_route_iter
*iter
= seq
->private;
2453 struct fib_table
*tb
;
2456 tb
= fib_get_table(seq_file_net(seq
), RT_TABLE_MAIN
);
2460 iter
->main_trie
= (struct trie
*) tb
->tb_data
;
2462 return SEQ_START_TOKEN
;
2464 return fib_route_get_idx(iter
, *pos
- 1);
2467 static void *fib_route_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2469 struct fib_route_iter
*iter
= seq
->private;
2473 if (v
== SEQ_START_TOKEN
) {
2475 l
= trie_firstleaf(iter
->main_trie
);
2478 l
= trie_nextleaf(l
);
2488 static void fib_route_seq_stop(struct seq_file
*seq
, void *v
)
2494 static unsigned int fib_flag_trans(int type
, __be32 mask
, const struct fib_info
*fi
)
2496 unsigned int flags
= 0;
2498 if (type
== RTN_UNREACHABLE
|| type
== RTN_PROHIBIT
)
2500 if (fi
&& fi
->fib_nh
->nh_gw
)
2501 flags
|= RTF_GATEWAY
;
2502 if (mask
== htonl(0xFFFFFFFF))
2509 * This outputs /proc/net/route.
2510 * The format of the file is not supposed to be changed
2511 * and needs to be same as fib_hash output to avoid breaking
2514 static int fib_route_seq_show(struct seq_file
*seq
, void *v
)
2517 struct leaf_info
*li
;
2519 if (v
== SEQ_START_TOKEN
) {
2520 seq_printf(seq
, "%-127s\n", "Iface\tDestination\tGateway "
2521 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2526 hlist_for_each_entry_rcu(li
, &l
->list
, hlist
) {
2527 struct fib_alias
*fa
;
2528 __be32 mask
, prefix
;
2530 mask
= inet_make_mask(li
->plen
);
2531 prefix
= htonl(l
->key
);
2533 list_for_each_entry_rcu(fa
, &li
->falh
, fa_list
) {
2534 const struct fib_info
*fi
= fa
->fa_info
;
2535 unsigned int flags
= fib_flag_trans(fa
->fa_type
, mask
, fi
);
2538 if (fa
->fa_type
== RTN_BROADCAST
2539 || fa
->fa_type
== RTN_MULTICAST
)
2544 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2545 "%d\t%08X\t%d\t%u\t%u%n",
2546 fi
->fib_dev
? fi
->fib_dev
->name
: "*",
2548 fi
->fib_nh
->nh_gw
, flags
, 0, 0,
2552 fi
->fib_advmss
+ 40 : 0),
2554 fi
->fib_rtt
>> 3, &len
);
2557 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2558 "%d\t%08X\t%d\t%u\t%u%n",
2559 prefix
, 0, flags
, 0, 0, 0,
2560 mask
, 0, 0, 0, &len
);
2562 seq_printf(seq
, "%*s\n", 127 - len
, "");
2569 static const struct seq_operations fib_route_seq_ops
= {
2570 .start
= fib_route_seq_start
,
2571 .next
= fib_route_seq_next
,
2572 .stop
= fib_route_seq_stop
,
2573 .show
= fib_route_seq_show
,
2576 static int fib_route_seq_open(struct inode
*inode
, struct file
*file
)
2578 return seq_open_net(inode
, file
, &fib_route_seq_ops
,
2579 sizeof(struct fib_route_iter
));
2582 static const struct file_operations fib_route_fops
= {
2583 .owner
= THIS_MODULE
,
2584 .open
= fib_route_seq_open
,
2586 .llseek
= seq_lseek
,
2587 .release
= seq_release_net
,
2590 int __net_init
fib_proc_init(struct net
*net
)
2592 if (!proc_create("fib_trie", S_IRUGO
, net
->proc_net
, &fib_trie_fops
))
2595 if (!proc_create("fib_triestat", S_IRUGO
, net
->proc_net
,
2596 &fib_triestat_fops
))
2599 if (!proc_create("route", S_IRUGO
, net
->proc_net
, &fib_route_fops
))
2605 remove_proc_entry("fib_triestat", net
->proc_net
);
2607 remove_proc_entry("fib_trie", net
->proc_net
);
2612 void __net_exit
fib_proc_exit(struct net
*net
)
2614 remove_proc_entry("fib_trie", net
->proc_net
);
2615 remove_proc_entry("fib_triestat", net
->proc_net
);
2616 remove_proc_entry("route", net
->proc_net
);
2619 #endif /* CONFIG_PROC_FS */