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 descibed 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.nada.kth.se/~snilsson/public/papers/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 <asm/system.h>
55 #include <linux/bitops.h>
56 #include <linux/types.h>
57 #include <linux/kernel.h>
59 #include <linux/string.h>
60 #include <linux/socket.h>
61 #include <linux/sockios.h>
62 #include <linux/errno.h>
64 #include <linux/inet.h>
65 #include <linux/inetdevice.h>
66 #include <linux/netdevice.h>
67 #include <linux/if_arp.h>
68 #include <linux/proc_fs.h>
69 #include <linux/rcupdate.h>
70 #include <linux/skbuff.h>
71 #include <linux/netlink.h>
72 #include <linux/init.h>
73 #include <linux/list.h>
74 #include <net/net_namespace.h>
76 #include <net/protocol.h>
77 #include <net/route.h>
80 #include <net/ip_fib.h>
81 #include "fib_lookup.h"
83 #define MAX_STAT_DEPTH 32
85 #define KEYLENGTH (8*sizeof(t_key))
87 typedef unsigned int t_key
;
91 #define NODE_TYPE_MASK 0x1UL
92 #define NODE_TYPE(node) ((node)->parent & NODE_TYPE_MASK)
94 #define IS_TNODE(n) (!(n->parent & T_LEAF))
95 #define IS_LEAF(n) (n->parent & T_LEAF)
103 unsigned long parent
;
105 struct hlist_head list
;
110 struct hlist_node hlist
;
113 struct list_head falh
;
117 unsigned long parent
;
119 unsigned char pos
; /* 2log(KEYLENGTH) bits needed */
120 unsigned char bits
; /* 2log(KEYLENGTH) bits needed */
121 unsigned int full_children
; /* KEYLENGTH bits needed */
122 unsigned int empty_children
; /* KEYLENGTH bits needed */
125 struct work_struct work
;
126 struct tnode
*tnode_free
;
128 struct node
*child
[0];
131 #ifdef CONFIG_IP_FIB_TRIE_STATS
132 struct trie_use_stats
{
134 unsigned int backtrack
;
135 unsigned int semantic_match_passed
;
136 unsigned int semantic_match_miss
;
137 unsigned int null_node_hit
;
138 unsigned int resize_node_skipped
;
143 unsigned int totdepth
;
144 unsigned int maxdepth
;
147 unsigned int nullpointers
;
148 unsigned int prefixes
;
149 unsigned int nodesizes
[MAX_STAT_DEPTH
];
154 #ifdef CONFIG_IP_FIB_TRIE_STATS
155 struct trie_use_stats stats
;
159 static void put_child(struct trie
*t
, struct tnode
*tn
, int i
, struct node
*n
);
160 static void tnode_put_child_reorg(struct tnode
*tn
, int i
, struct node
*n
,
162 static struct 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
;
179 static inline struct tnode
*node_parent(struct node
*node
)
181 return (struct tnode
*)(node
->parent
& ~NODE_TYPE_MASK
);
184 static inline struct tnode
*node_parent_rcu(struct node
*node
)
186 struct tnode
*ret
= node_parent(node
);
188 return rcu_dereference(ret
);
191 /* Same as rcu_assign_pointer
192 * but that macro() assumes that value is a pointer.
194 static inline void node_set_parent(struct node
*node
, struct tnode
*ptr
)
197 node
->parent
= (unsigned long)ptr
| NODE_TYPE(node
);
200 static inline struct node
*tnode_get_child(struct tnode
*tn
, unsigned int i
)
202 BUG_ON(i
>= 1U << tn
->bits
);
207 static inline struct node
*tnode_get_child_rcu(struct tnode
*tn
, unsigned int i
)
209 struct node
*ret
= tnode_get_child(tn
, i
);
211 return rcu_dereference(ret
);
214 static inline int tnode_child_length(const struct tnode
*tn
)
216 return 1 << tn
->bits
;
219 static inline t_key
mask_pfx(t_key k
, unsigned short l
)
221 return (l
== 0) ? 0 : k
>> (KEYLENGTH
-l
) << (KEYLENGTH
-l
);
224 static inline t_key
tkey_extract_bits(t_key a
, int offset
, int bits
)
226 if (offset
< KEYLENGTH
)
227 return ((t_key
)(a
<< offset
)) >> (KEYLENGTH
- bits
);
232 static inline int tkey_equals(t_key a
, t_key b
)
237 static inline int tkey_sub_equals(t_key a
, int offset
, int bits
, t_key b
)
239 if (bits
== 0 || offset
>= KEYLENGTH
)
241 bits
= bits
> KEYLENGTH
? KEYLENGTH
: bits
;
242 return ((a
^ b
) << offset
) >> (KEYLENGTH
- bits
) == 0;
245 static inline int tkey_mismatch(t_key a
, int offset
, t_key b
)
252 while ((diff
<< i
) >> (KEYLENGTH
-1) == 0)
258 To understand this stuff, an understanding of keys and all their bits is
259 necessary. Every node in the trie has a key associated with it, but not
260 all of the bits in that key are significant.
262 Consider a node 'n' and its parent 'tp'.
264 If n is a leaf, every bit in its key is significant. Its presence is
265 necessitated by path compression, since during a tree traversal (when
266 searching for a leaf - unless we are doing an insertion) we will completely
267 ignore all skipped bits we encounter. Thus we need to verify, at the end of
268 a potentially successful search, that we have indeed been walking the
271 Note that we can never "miss" the correct key in the tree if present by
272 following the wrong path. Path compression ensures that segments of the key
273 that are the same for all keys with a given prefix are skipped, but the
274 skipped part *is* identical for each node in the subtrie below the skipped
275 bit! trie_insert() in this implementation takes care of that - note the
276 call to tkey_sub_equals() in trie_insert().
278 if n is an internal node - a 'tnode' here, the various parts of its key
279 have many different meanings.
282 _________________________________________________________________
283 | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
284 -----------------------------------------------------------------
285 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
287 _________________________________________________________________
288 | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
289 -----------------------------------------------------------------
290 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
297 First, let's just ignore the bits that come before the parent tp, that is
298 the bits from 0 to (tp->pos-1). They are *known* but at this point we do
299 not use them for anything.
301 The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
302 index into the parent's child array. That is, they will be used to find
303 'n' among tp's children.
305 The bits from (tp->pos + tp->bits) to (n->pos - 1) - "S" - are skipped bits
308 All the bits we have seen so far are significant to the node n. The rest
309 of the bits are really not needed or indeed known in n->key.
311 The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
312 n's child array, and will of course be different for each child.
315 The rest of the bits, from (n->pos + n->bits) onward, are completely unknown
320 static inline void check_tnode(const struct tnode
*tn
)
322 WARN_ON(tn
&& tn
->pos
+tn
->bits
> 32);
325 static const int halve_threshold
= 25;
326 static const int inflate_threshold
= 50;
327 static const int halve_threshold_root
= 15;
328 static const int inflate_threshold_root
= 30;
330 static void __alias_free_mem(struct rcu_head
*head
)
332 struct fib_alias
*fa
= container_of(head
, struct fib_alias
, rcu
);
333 kmem_cache_free(fn_alias_kmem
, fa
);
336 static inline void alias_free_mem_rcu(struct fib_alias
*fa
)
338 call_rcu(&fa
->rcu
, __alias_free_mem
);
341 static void __leaf_free_rcu(struct rcu_head
*head
)
343 struct leaf
*l
= container_of(head
, struct leaf
, rcu
);
344 kmem_cache_free(trie_leaf_kmem
, l
);
347 static inline void free_leaf(struct leaf
*l
)
349 call_rcu_bh(&l
->rcu
, __leaf_free_rcu
);
352 static void __leaf_info_free_rcu(struct rcu_head
*head
)
354 kfree(container_of(head
, struct leaf_info
, rcu
));
357 static inline void free_leaf_info(struct leaf_info
*leaf
)
359 call_rcu(&leaf
->rcu
, __leaf_info_free_rcu
);
362 static struct tnode
*tnode_alloc(size_t size
)
364 if (size
<= PAGE_SIZE
)
365 return kzalloc(size
, GFP_KERNEL
);
367 return __vmalloc(size
, GFP_KERNEL
| __GFP_ZERO
, PAGE_KERNEL
);
370 static void __tnode_vfree(struct work_struct
*arg
)
372 struct tnode
*tn
= container_of(arg
, struct tnode
, work
);
376 static void __tnode_free_rcu(struct rcu_head
*head
)
378 struct tnode
*tn
= container_of(head
, struct tnode
, rcu
);
379 size_t size
= sizeof(struct tnode
) +
380 (sizeof(struct node
*) << tn
->bits
);
382 if (size
<= PAGE_SIZE
)
385 INIT_WORK(&tn
->work
, __tnode_vfree
);
386 schedule_work(&tn
->work
);
390 static inline void tnode_free(struct tnode
*tn
)
393 free_leaf((struct leaf
*) tn
);
395 call_rcu(&tn
->rcu
, __tnode_free_rcu
);
398 static void tnode_free_safe(struct tnode
*tn
)
401 tn
->tnode_free
= tnode_free_head
;
402 tnode_free_head
= tn
;
403 tnode_free_size
+= sizeof(struct tnode
) +
404 (sizeof(struct node
*) << tn
->bits
);
407 static void tnode_free_flush(void)
411 while ((tn
= tnode_free_head
)) {
412 tnode_free_head
= tn
->tnode_free
;
413 tn
->tnode_free
= NULL
;
417 if (tnode_free_size
>= PAGE_SIZE
* sync_pages
) {
423 static struct leaf
*leaf_new(void)
425 struct leaf
*l
= kmem_cache_alloc(trie_leaf_kmem
, GFP_KERNEL
);
428 INIT_HLIST_HEAD(&l
->list
);
433 static struct leaf_info
*leaf_info_new(int plen
)
435 struct leaf_info
*li
= kmalloc(sizeof(struct leaf_info
), GFP_KERNEL
);
438 INIT_LIST_HEAD(&li
->falh
);
443 static struct tnode
*tnode_new(t_key key
, int pos
, int bits
)
445 size_t sz
= sizeof(struct tnode
) + (sizeof(struct node
*) << bits
);
446 struct tnode
*tn
= tnode_alloc(sz
);
449 tn
->parent
= T_TNODE
;
453 tn
->full_children
= 0;
454 tn
->empty_children
= 1<<bits
;
457 pr_debug("AT %p s=%u %lu\n", tn
, (unsigned int) sizeof(struct tnode
),
458 (unsigned long) (sizeof(struct node
) << bits
));
463 * Check whether a tnode 'n' is "full", i.e. it is an internal node
464 * and no bits are skipped. See discussion in dyntree paper p. 6
467 static inline int tnode_full(const struct tnode
*tn
, const struct node
*n
)
469 if (n
== NULL
|| IS_LEAF(n
))
472 return ((struct tnode
*) n
)->pos
== tn
->pos
+ tn
->bits
;
475 static inline void put_child(struct trie
*t
, struct tnode
*tn
, int i
,
478 tnode_put_child_reorg(tn
, i
, n
, -1);
482 * Add a child at position i overwriting the old value.
483 * Update the value of full_children and empty_children.
486 static void tnode_put_child_reorg(struct tnode
*tn
, int i
, struct node
*n
,
489 struct node
*chi
= tn
->child
[i
];
492 BUG_ON(i
>= 1<<tn
->bits
);
494 /* update emptyChildren */
495 if (n
== NULL
&& chi
!= NULL
)
496 tn
->empty_children
++;
497 else if (n
!= NULL
&& chi
== NULL
)
498 tn
->empty_children
--;
500 /* update fullChildren */
502 wasfull
= tnode_full(tn
, chi
);
504 isfull
= tnode_full(tn
, n
);
505 if (wasfull
&& !isfull
)
507 else if (!wasfull
&& isfull
)
511 node_set_parent(n
, tn
);
513 rcu_assign_pointer(tn
->child
[i
], n
);
517 static struct node
*resize(struct trie
*t
, struct tnode
*tn
)
520 struct tnode
*old_tn
;
521 int inflate_threshold_use
;
522 int halve_threshold_use
;
528 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
529 tn
, inflate_threshold
, halve_threshold
);
532 if (tn
->empty_children
== tnode_child_length(tn
)) {
537 if (tn
->empty_children
== tnode_child_length(tn
) - 1)
540 * Double as long as the resulting node has a number of
541 * nonempty nodes that are above the threshold.
545 * From "Implementing a dynamic compressed trie" by Stefan Nilsson of
546 * the Helsinki University of Technology and Matti Tikkanen of Nokia
547 * Telecommunications, page 6:
548 * "A node is doubled if the ratio of non-empty children to all
549 * children in the *doubled* node is at least 'high'."
551 * 'high' in this instance is the variable 'inflate_threshold'. It
552 * is expressed as a percentage, so we multiply it with
553 * tnode_child_length() and instead of multiplying by 2 (since the
554 * child array will be doubled by inflate()) and multiplying
555 * the left-hand side by 100 (to handle the percentage thing) we
556 * multiply the left-hand side by 50.
558 * The left-hand side may look a bit weird: tnode_child_length(tn)
559 * - tn->empty_children is of course the number of non-null children
560 * in the current node. tn->full_children is the number of "full"
561 * children, that is non-null tnodes with a skip value of 0.
562 * All of those will be doubled in the resulting inflated tnode, so
563 * we just count them one extra time here.
565 * A clearer way to write this would be:
567 * to_be_doubled = tn->full_children;
568 * not_to_be_doubled = tnode_child_length(tn) - tn->empty_children -
571 * new_child_length = tnode_child_length(tn) * 2;
573 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
575 * if (new_fill_factor >= inflate_threshold)
577 * ...and so on, tho it would mess up the while () loop.
580 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
584 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
585 * inflate_threshold * new_child_length
587 * expand not_to_be_doubled and to_be_doubled, and shorten:
588 * 100 * (tnode_child_length(tn) - tn->empty_children +
589 * tn->full_children) >= inflate_threshold * new_child_length
591 * expand new_child_length:
592 * 100 * (tnode_child_length(tn) - tn->empty_children +
593 * tn->full_children) >=
594 * inflate_threshold * tnode_child_length(tn) * 2
597 * 50 * (tn->full_children + tnode_child_length(tn) -
598 * tn->empty_children) >= inflate_threshold *
599 * tnode_child_length(tn)
605 /* Keep root node larger */
607 if (!node_parent((struct node
*) tn
)) {
608 inflate_threshold_use
= inflate_threshold_root
;
609 halve_threshold_use
= halve_threshold_root
;
612 inflate_threshold_use
= inflate_threshold
;
613 halve_threshold_use
= halve_threshold
;
617 while ((tn
->full_children
> 0 && max_work
-- &&
618 50 * (tn
->full_children
+ tnode_child_length(tn
)
619 - tn
->empty_children
)
620 >= inflate_threshold_use
* tnode_child_length(tn
))) {
627 #ifdef CONFIG_IP_FIB_TRIE_STATS
628 t
->stats
.resize_node_skipped
++;
636 /* Return if at least one inflate is run */
637 if( max_work
!= MAX_WORK
)
638 return (struct node
*) tn
;
641 * Halve as long as the number of empty children in this
642 * node is above threshold.
646 while (tn
->bits
> 1 && max_work
-- &&
647 100 * (tnode_child_length(tn
) - tn
->empty_children
) <
648 halve_threshold_use
* tnode_child_length(tn
)) {
654 #ifdef CONFIG_IP_FIB_TRIE_STATS
655 t
->stats
.resize_node_skipped
++;
662 /* Only one child remains */
663 if (tn
->empty_children
== tnode_child_length(tn
) - 1) {
665 for (i
= 0; i
< tnode_child_length(tn
); i
++) {
672 /* compress one level */
674 node_set_parent(n
, NULL
);
679 return (struct node
*) tn
;
682 static struct tnode
*inflate(struct trie
*t
, struct tnode
*tn
)
684 struct tnode
*oldtnode
= tn
;
685 int olen
= tnode_child_length(tn
);
688 pr_debug("In inflate\n");
690 tn
= tnode_new(oldtnode
->key
, oldtnode
->pos
, oldtnode
->bits
+ 1);
693 return ERR_PTR(-ENOMEM
);
696 * Preallocate and store tnodes before the actual work so we
697 * don't get into an inconsistent state if memory allocation
698 * fails. In case of failure we return the oldnode and inflate
699 * of tnode is ignored.
702 for (i
= 0; i
< olen
; i
++) {
705 inode
= (struct tnode
*) tnode_get_child(oldtnode
, i
);
708 inode
->pos
== oldtnode
->pos
+ oldtnode
->bits
&&
710 struct tnode
*left
, *right
;
711 t_key m
= ~0U << (KEYLENGTH
- 1) >> inode
->pos
;
713 left
= tnode_new(inode
->key
&(~m
), inode
->pos
+ 1,
718 right
= tnode_new(inode
->key
|m
, inode
->pos
+ 1,
726 put_child(t
, tn
, 2*i
, (struct node
*) left
);
727 put_child(t
, tn
, 2*i
+1, (struct node
*) right
);
731 for (i
= 0; i
< olen
; i
++) {
733 struct node
*node
= tnode_get_child(oldtnode
, i
);
734 struct tnode
*left
, *right
;
741 /* A leaf or an internal node with skipped bits */
743 if (IS_LEAF(node
) || ((struct tnode
*) node
)->pos
>
744 tn
->pos
+ tn
->bits
- 1) {
745 if (tkey_extract_bits(node
->key
,
746 oldtnode
->pos
+ oldtnode
->bits
,
748 put_child(t
, tn
, 2*i
, node
);
750 put_child(t
, tn
, 2*i
+1, node
);
754 /* An internal node with two children */
755 inode
= (struct tnode
*) node
;
757 if (inode
->bits
== 1) {
758 put_child(t
, tn
, 2*i
, inode
->child
[0]);
759 put_child(t
, tn
, 2*i
+1, inode
->child
[1]);
761 tnode_free_safe(inode
);
765 /* An internal node with more than two children */
767 /* We will replace this node 'inode' with two new
768 * ones, 'left' and 'right', each with half of the
769 * original children. The two new nodes will have
770 * a position one bit further down the key and this
771 * means that the "significant" part of their keys
772 * (see the discussion near the top of this file)
773 * will differ by one bit, which will be "0" in
774 * left's key and "1" in right's key. Since we are
775 * moving the key position by one step, the bit that
776 * we are moving away from - the bit at position
777 * (inode->pos) - is the one that will differ between
778 * left and right. So... we synthesize that bit in the
780 * The mask 'm' below will be a single "one" bit at
781 * the position (inode->pos)
784 /* Use the old key, but set the new significant
788 left
= (struct tnode
*) tnode_get_child(tn
, 2*i
);
789 put_child(t
, tn
, 2*i
, NULL
);
793 right
= (struct tnode
*) tnode_get_child(tn
, 2*i
+1);
794 put_child(t
, tn
, 2*i
+1, NULL
);
798 size
= tnode_child_length(left
);
799 for (j
= 0; j
< size
; j
++) {
800 put_child(t
, left
, j
, inode
->child
[j
]);
801 put_child(t
, right
, j
, inode
->child
[j
+ size
]);
803 put_child(t
, tn
, 2*i
, resize(t
, left
));
804 put_child(t
, tn
, 2*i
+1, resize(t
, right
));
806 tnode_free_safe(inode
);
808 tnode_free_safe(oldtnode
);
812 int size
= tnode_child_length(tn
);
815 for (j
= 0; j
< size
; j
++)
817 tnode_free((struct tnode
*)tn
->child
[j
]);
821 return ERR_PTR(-ENOMEM
);
825 static struct tnode
*halve(struct trie
*t
, struct tnode
*tn
)
827 struct tnode
*oldtnode
= tn
;
828 struct node
*left
, *right
;
830 int olen
= tnode_child_length(tn
);
832 pr_debug("In halve\n");
834 tn
= tnode_new(oldtnode
->key
, oldtnode
->pos
, oldtnode
->bits
- 1);
837 return ERR_PTR(-ENOMEM
);
840 * Preallocate and store tnodes before the actual work so we
841 * don't get into an inconsistent state if memory allocation
842 * fails. In case of failure we return the oldnode and halve
843 * of tnode is ignored.
846 for (i
= 0; i
< olen
; i
+= 2) {
847 left
= tnode_get_child(oldtnode
, i
);
848 right
= tnode_get_child(oldtnode
, i
+1);
850 /* Two nonempty children */
854 newn
= tnode_new(left
->key
, tn
->pos
+ tn
->bits
, 1);
859 put_child(t
, tn
, i
/2, (struct node
*)newn
);
864 for (i
= 0; i
< olen
; i
+= 2) {
865 struct tnode
*newBinNode
;
867 left
= tnode_get_child(oldtnode
, i
);
868 right
= tnode_get_child(oldtnode
, i
+1);
870 /* At least one of the children is empty */
872 if (right
== NULL
) /* Both are empty */
874 put_child(t
, tn
, i
/2, right
);
879 put_child(t
, tn
, i
/2, left
);
883 /* Two nonempty children */
884 newBinNode
= (struct tnode
*) tnode_get_child(tn
, i
/2);
885 put_child(t
, tn
, i
/2, NULL
);
886 put_child(t
, newBinNode
, 0, left
);
887 put_child(t
, newBinNode
, 1, right
);
888 put_child(t
, tn
, i
/2, resize(t
, newBinNode
));
890 tnode_free_safe(oldtnode
);
894 int size
= tnode_child_length(tn
);
897 for (j
= 0; j
< size
; j
++)
899 tnode_free((struct tnode
*)tn
->child
[j
]);
903 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 hlist_node
*node
;
914 struct leaf_info
*li
;
916 hlist_for_each_entry_rcu(li
, node
, 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
;
936 struct hlist_node
*node
;
938 if (hlist_empty(head
)) {
939 hlist_add_head_rcu(&new->hlist
, head
);
941 hlist_for_each_entry(li
, node
, head
, hlist
) {
942 if (new->plen
> li
->plen
)
948 hlist_add_after_rcu(&last
->hlist
, &new->hlist
);
950 hlist_add_before_rcu(&new->hlist
, &li
->hlist
);
954 /* rcu_read_lock needs to be hold by caller from readside */
957 fib_find_node(struct trie
*t
, u32 key
)
964 n
= rcu_dereference(t
->trie
);
966 while (n
!= NULL
&& NODE_TYPE(n
) == T_TNODE
) {
967 tn
= (struct tnode
*) n
;
971 if (tkey_sub_equals(tn
->key
, pos
, tn
->pos
-pos
, key
)) {
972 pos
= tn
->pos
+ tn
->bits
;
973 n
= tnode_get_child_rcu(tn
,
974 tkey_extract_bits(key
,
980 /* Case we have found a leaf. Compare prefixes */
982 if (n
!= NULL
&& IS_LEAF(n
) && tkey_equals(key
, n
->key
))
983 return (struct leaf
*)n
;
988 static void trie_rebalance(struct trie
*t
, struct tnode
*tn
)
996 while (tn
!= NULL
&& (tp
= node_parent((struct node
*)tn
)) != NULL
) {
997 cindex
= tkey_extract_bits(key
, tp
->pos
, tp
->bits
);
998 wasfull
= tnode_full(tp
, tnode_get_child(tp
, cindex
));
999 tn
= (struct tnode
*) resize(t
, (struct tnode
*)tn
);
1001 tnode_put_child_reorg((struct tnode
*)tp
, cindex
,
1002 (struct node
*)tn
, wasfull
);
1004 tp
= node_parent((struct node
*) tn
);
1006 rcu_assign_pointer(t
->trie
, (struct node
*)tn
);
1014 /* Handle last (top) tnode */
1016 tn
= (struct tnode
*)resize(t
, (struct tnode
*)tn
);
1018 rcu_assign_pointer(t
->trie
, (struct node
*)tn
);
1024 /* only used from updater-side */
1026 static struct list_head
*fib_insert_node(struct trie
*t
, u32 key
, int plen
)
1029 struct tnode
*tp
= NULL
, *tn
= NULL
;
1033 struct list_head
*fa_head
= NULL
;
1034 struct leaf_info
*li
;
1040 /* If we point to NULL, stop. Either the tree is empty and we should
1041 * just put a new leaf in if, or we have reached an empty child slot,
1042 * and we should just put our new leaf in that.
1043 * If we point to a T_TNODE, check if it matches our key. Note that
1044 * a T_TNODE might be skipping any number of bits - its 'pos' need
1045 * not be the parent's 'pos'+'bits'!
1047 * If it does match the current key, get pos/bits from it, extract
1048 * the index from our key, push the T_TNODE and walk the tree.
1050 * If it doesn't, we have to replace it with a new T_TNODE.
1052 * If we point to a T_LEAF, it might or might not have the same key
1053 * as we do. If it does, just change the value, update the T_LEAF's
1054 * value, and return it.
1055 * If it doesn't, we need to replace it with a T_TNODE.
1058 while (n
!= NULL
&& NODE_TYPE(n
) == T_TNODE
) {
1059 tn
= (struct tnode
*) n
;
1063 if (tkey_sub_equals(tn
->key
, pos
, tn
->pos
-pos
, key
)) {
1065 pos
= tn
->pos
+ tn
->bits
;
1066 n
= tnode_get_child(tn
,
1067 tkey_extract_bits(key
,
1071 BUG_ON(n
&& node_parent(n
) != tn
);
1077 * n ----> NULL, LEAF or TNODE
1079 * tp is n's (parent) ----> NULL or TNODE
1082 BUG_ON(tp
&& IS_LEAF(tp
));
1084 /* Case 1: n is a leaf. Compare prefixes */
1086 if (n
!= NULL
&& IS_LEAF(n
) && tkey_equals(key
, n
->key
)) {
1087 l
= (struct leaf
*) n
;
1088 li
= leaf_info_new(plen
);
1093 fa_head
= &li
->falh
;
1094 insert_leaf_info(&l
->list
, li
);
1103 li
= leaf_info_new(plen
);
1110 fa_head
= &li
->falh
;
1111 insert_leaf_info(&l
->list
, li
);
1113 if (t
->trie
&& n
== NULL
) {
1114 /* Case 2: n is NULL, and will just insert a new leaf */
1116 node_set_parent((struct node
*)l
, tp
);
1118 cindex
= tkey_extract_bits(key
, tp
->pos
, tp
->bits
);
1119 put_child(t
, (struct tnode
*)tp
, cindex
, (struct node
*)l
);
1121 /* Case 3: n is a LEAF or a TNODE and the key doesn't match. */
1123 * Add a new tnode here
1124 * first tnode need some special handling
1128 pos
= tp
->pos
+tp
->bits
;
1133 newpos
= tkey_mismatch(key
, pos
, n
->key
);
1134 tn
= tnode_new(n
->key
, newpos
, 1);
1137 tn
= tnode_new(key
, newpos
, 1); /* First tnode */
1146 node_set_parent((struct node
*)tn
, tp
);
1148 missbit
= tkey_extract_bits(key
, newpos
, 1);
1149 put_child(t
, tn
, missbit
, (struct node
*)l
);
1150 put_child(t
, tn
, 1-missbit
, n
);
1153 cindex
= tkey_extract_bits(key
, tp
->pos
, tp
->bits
);
1154 put_child(t
, (struct tnode
*)tp
, cindex
,
1157 rcu_assign_pointer(t
->trie
, (struct node
*)tn
);
1162 if (tp
&& tp
->pos
+ tp
->bits
> 32)
1163 pr_warning("fib_trie"
1164 " tp=%p pos=%d, bits=%d, key=%0x plen=%d\n",
1165 tp
, tp
->pos
, tp
->bits
, key
, plen
);
1167 /* Rebalance the trie */
1169 trie_rebalance(t
, tp
);
1175 * Caller must hold RTNL.
1177 static int fn_trie_insert(struct fib_table
*tb
, struct fib_config
*cfg
)
1179 struct trie
*t
= (struct trie
*) tb
->tb_data
;
1180 struct fib_alias
*fa
, *new_fa
;
1181 struct list_head
*fa_head
= NULL
;
1182 struct fib_info
*fi
;
1183 int plen
= cfg
->fc_dst_len
;
1184 u8 tos
= cfg
->fc_tos
;
1192 key
= ntohl(cfg
->fc_dst
);
1194 pr_debug("Insert table=%u %08x/%d\n", tb
->tb_id
, key
, plen
);
1196 mask
= ntohl(inet_make_mask(plen
));
1203 fi
= fib_create_info(cfg
);
1209 l
= fib_find_node(t
, key
);
1213 fa_head
= get_fa_head(l
, plen
);
1214 fa
= fib_find_alias(fa_head
, tos
, fi
->fib_priority
);
1217 /* Now fa, if non-NULL, points to the first fib alias
1218 * with the same keys [prefix,tos,priority], if such key already
1219 * exists or to the node before which we will insert new one.
1221 * If fa is NULL, we will need to allocate a new one and
1222 * insert to the head of f.
1224 * If f is NULL, no fib node matched the destination key
1225 * and we need to allocate a new one of those as well.
1228 if (fa
&& fa
->fa_tos
== tos
&&
1229 fa
->fa_info
->fib_priority
== fi
->fib_priority
) {
1230 struct fib_alias
*fa_first
, *fa_match
;
1233 if (cfg
->fc_nlflags
& NLM_F_EXCL
)
1237 * 1. Find exact match for type, scope, fib_info to avoid
1239 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1243 fa
= list_entry(fa
->fa_list
.prev
, struct fib_alias
, fa_list
);
1244 list_for_each_entry_continue(fa
, fa_head
, fa_list
) {
1245 if (fa
->fa_tos
!= tos
)
1247 if (fa
->fa_info
->fib_priority
!= fi
->fib_priority
)
1249 if (fa
->fa_type
== cfg
->fc_type
&&
1250 fa
->fa_scope
== cfg
->fc_scope
&&
1251 fa
->fa_info
== fi
) {
1257 if (cfg
->fc_nlflags
& NLM_F_REPLACE
) {
1258 struct fib_info
*fi_drop
;
1268 new_fa
= kmem_cache_alloc(fn_alias_kmem
, GFP_KERNEL
);
1272 fi_drop
= fa
->fa_info
;
1273 new_fa
->fa_tos
= fa
->fa_tos
;
1274 new_fa
->fa_info
= fi
;
1275 new_fa
->fa_type
= cfg
->fc_type
;
1276 new_fa
->fa_scope
= cfg
->fc_scope
;
1277 state
= fa
->fa_state
;
1278 new_fa
->fa_state
= state
& ~FA_S_ACCESSED
;
1280 list_replace_rcu(&fa
->fa_list
, &new_fa
->fa_list
);
1281 alias_free_mem_rcu(fa
);
1283 fib_release_info(fi_drop
);
1284 if (state
& FA_S_ACCESSED
)
1285 rt_cache_flush(cfg
->fc_nlinfo
.nl_net
, -1);
1286 rtmsg_fib(RTM_NEWROUTE
, htonl(key
), new_fa
, plen
,
1287 tb
->tb_id
, &cfg
->fc_nlinfo
, NLM_F_REPLACE
);
1291 /* Error if we find a perfect match which
1292 * uses the same scope, type, and nexthop
1298 if (!(cfg
->fc_nlflags
& NLM_F_APPEND
))
1302 if (!(cfg
->fc_nlflags
& NLM_F_CREATE
))
1306 new_fa
= kmem_cache_alloc(fn_alias_kmem
, GFP_KERNEL
);
1310 new_fa
->fa_info
= fi
;
1311 new_fa
->fa_tos
= tos
;
1312 new_fa
->fa_type
= cfg
->fc_type
;
1313 new_fa
->fa_scope
= cfg
->fc_scope
;
1314 new_fa
->fa_state
= 0;
1316 * Insert new entry to the list.
1320 fa_head
= fib_insert_node(t
, key
, plen
);
1321 if (unlikely(!fa_head
)) {
1323 goto out_free_new_fa
;
1327 list_add_tail_rcu(&new_fa
->fa_list
,
1328 (fa
? &fa
->fa_list
: fa_head
));
1330 rt_cache_flush(cfg
->fc_nlinfo
.nl_net
, -1);
1331 rtmsg_fib(RTM_NEWROUTE
, htonl(key
), new_fa
, plen
, tb
->tb_id
,
1332 &cfg
->fc_nlinfo
, 0);
1337 kmem_cache_free(fn_alias_kmem
, new_fa
);
1339 fib_release_info(fi
);
1344 /* should be called with rcu_read_lock */
1345 static int check_leaf(struct trie
*t
, struct leaf
*l
,
1346 t_key key
, const struct flowi
*flp
,
1347 struct fib_result
*res
)
1349 struct leaf_info
*li
;
1350 struct hlist_head
*hhead
= &l
->list
;
1351 struct hlist_node
*node
;
1353 hlist_for_each_entry_rcu(li
, node
, hhead
, hlist
) {
1355 int plen
= li
->plen
;
1356 __be32 mask
= inet_make_mask(plen
);
1358 if (l
->key
!= (key
& ntohl(mask
)))
1361 err
= fib_semantic_match(&li
->falh
, flp
, res
, plen
);
1363 #ifdef CONFIG_IP_FIB_TRIE_STATS
1365 t
->stats
.semantic_match_passed
++;
1367 t
->stats
.semantic_match_miss
++;
1376 static int fn_trie_lookup(struct fib_table
*tb
, const struct flowi
*flp
,
1377 struct fib_result
*res
)
1379 struct trie
*t
= (struct trie
*) tb
->tb_data
;
1384 t_key key
= ntohl(flp
->fl4_dst
);
1387 int current_prefix_length
= KEYLENGTH
;
1389 t_key node_prefix
, key_prefix
, pref_mismatch
;
1394 n
= rcu_dereference(t
->trie
);
1398 #ifdef CONFIG_IP_FIB_TRIE_STATS
1404 ret
= check_leaf(t
, (struct leaf
*)n
, key
, flp
, res
);
1408 pn
= (struct tnode
*) n
;
1416 cindex
= tkey_extract_bits(mask_pfx(key
, current_prefix_length
),
1419 n
= tnode_get_child_rcu(pn
, cindex
);
1422 #ifdef CONFIG_IP_FIB_TRIE_STATS
1423 t
->stats
.null_node_hit
++;
1429 ret
= check_leaf(t
, (struct leaf
*)n
, key
, flp
, res
);
1435 cn
= (struct tnode
*)n
;
1438 * It's a tnode, and we can do some extra checks here if we
1439 * like, to avoid descending into a dead-end branch.
1440 * This tnode is in the parent's child array at index
1441 * key[p_pos..p_pos+p_bits] but potentially with some bits
1442 * chopped off, so in reality the index may be just a
1443 * subprefix, padded with zero at the end.
1444 * We can also take a look at any skipped bits in this
1445 * tnode - everything up to p_pos is supposed to be ok,
1446 * and the non-chopped bits of the index (se previous
1447 * paragraph) are also guaranteed ok, but the rest is
1448 * considered unknown.
1450 * The skipped bits are key[pos+bits..cn->pos].
1453 /* If current_prefix_length < pos+bits, we are already doing
1454 * actual prefix matching, which means everything from
1455 * pos+(bits-chopped_off) onward must be zero along some
1456 * branch of this subtree - otherwise there is *no* valid
1457 * prefix present. Here we can only check the skipped
1458 * bits. Remember, since we have already indexed into the
1459 * parent's child array, we know that the bits we chopped of
1463 /* NOTA BENE: Checking only skipped bits
1464 for the new node here */
1466 if (current_prefix_length
< pos
+bits
) {
1467 if (tkey_extract_bits(cn
->key
, current_prefix_length
,
1468 cn
->pos
- current_prefix_length
)
1474 * If chopped_off=0, the index is fully validated and we
1475 * only need to look at the skipped bits for this, the new,
1476 * tnode. What we actually want to do is to find out if
1477 * these skipped bits match our key perfectly, or if we will
1478 * have to count on finding a matching prefix further down,
1479 * because if we do, we would like to have some way of
1480 * verifying the existence of such a prefix at this point.
1483 /* The only thing we can do at this point is to verify that
1484 * any such matching prefix can indeed be a prefix to our
1485 * key, and if the bits in the node we are inspecting that
1486 * do not match our key are not ZERO, this cannot be true.
1487 * Thus, find out where there is a mismatch (before cn->pos)
1488 * and verify that all the mismatching bits are zero in the
1493 * Note: We aren't very concerned about the piece of
1494 * the key that precede pn->pos+pn->bits, since these
1495 * have already been checked. The bits after cn->pos
1496 * aren't checked since these are by definition
1497 * "unknown" at this point. Thus, what we want to see
1498 * is if we are about to enter the "prefix matching"
1499 * state, and in that case verify that the skipped
1500 * bits that will prevail throughout this subtree are
1501 * zero, as they have to be if we are to find a
1505 node_prefix
= mask_pfx(cn
->key
, cn
->pos
);
1506 key_prefix
= mask_pfx(key
, cn
->pos
);
1507 pref_mismatch
= key_prefix
^node_prefix
;
1511 * In short: If skipped bits in this node do not match
1512 * the search key, enter the "prefix matching"
1515 if (pref_mismatch
) {
1516 while (!(pref_mismatch
& (1<<(KEYLENGTH
-1)))) {
1518 pref_mismatch
= pref_mismatch
<< 1;
1520 key_prefix
= tkey_extract_bits(cn
->key
, mp
, cn
->pos
-mp
);
1522 if (key_prefix
!= 0)
1525 if (current_prefix_length
>= cn
->pos
)
1526 current_prefix_length
= mp
;
1529 pn
= (struct tnode
*)n
; /* Descend */
1536 /* As zero don't change the child key (cindex) */
1537 while ((chopped_off
<= pn
->bits
)
1538 && !(cindex
& (1<<(chopped_off
-1))))
1541 /* Decrease current_... with bits chopped off */
1542 if (current_prefix_length
> pn
->pos
+ pn
->bits
- chopped_off
)
1543 current_prefix_length
= pn
->pos
+ pn
->bits
1547 * Either we do the actual chop off according or if we have
1548 * chopped off all bits in this tnode walk up to our parent.
1551 if (chopped_off
<= pn
->bits
) {
1552 cindex
&= ~(1 << (chopped_off
-1));
1554 struct tnode
*parent
= node_parent_rcu((struct node
*) pn
);
1558 /* Get Child's index */
1559 cindex
= tkey_extract_bits(pn
->key
, parent
->pos
, parent
->bits
);
1563 #ifdef CONFIG_IP_FIB_TRIE_STATS
1564 t
->stats
.backtrack
++;
1577 * Remove the leaf and return parent.
1579 static void trie_leaf_remove(struct trie
*t
, struct leaf
*l
)
1581 struct tnode
*tp
= node_parent((struct node
*) l
);
1583 pr_debug("entering trie_leaf_remove(%p)\n", l
);
1586 t_key cindex
= tkey_extract_bits(l
->key
, tp
->pos
, tp
->bits
);
1587 put_child(t
, (struct tnode
*)tp
, cindex
, NULL
);
1588 trie_rebalance(t
, tp
);
1590 rcu_assign_pointer(t
->trie
, NULL
);
1596 * Caller must hold RTNL.
1598 static int fn_trie_delete(struct fib_table
*tb
, struct fib_config
*cfg
)
1600 struct trie
*t
= (struct trie
*) tb
->tb_data
;
1602 int plen
= cfg
->fc_dst_len
;
1603 u8 tos
= cfg
->fc_tos
;
1604 struct fib_alias
*fa
, *fa_to_delete
;
1605 struct list_head
*fa_head
;
1607 struct leaf_info
*li
;
1612 key
= ntohl(cfg
->fc_dst
);
1613 mask
= ntohl(inet_make_mask(plen
));
1619 l
= fib_find_node(t
, key
);
1624 fa_head
= get_fa_head(l
, plen
);
1625 fa
= fib_find_alias(fa_head
, tos
, 0);
1630 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key
, plen
, tos
, t
);
1632 fa_to_delete
= NULL
;
1633 fa
= list_entry(fa
->fa_list
.prev
, struct fib_alias
, fa_list
);
1634 list_for_each_entry_continue(fa
, fa_head
, fa_list
) {
1635 struct fib_info
*fi
= fa
->fa_info
;
1637 if (fa
->fa_tos
!= tos
)
1640 if ((!cfg
->fc_type
|| fa
->fa_type
== cfg
->fc_type
) &&
1641 (cfg
->fc_scope
== RT_SCOPE_NOWHERE
||
1642 fa
->fa_scope
== cfg
->fc_scope
) &&
1643 (!cfg
->fc_protocol
||
1644 fi
->fib_protocol
== cfg
->fc_protocol
) &&
1645 fib_nh_match(cfg
, fi
) == 0) {
1655 rtmsg_fib(RTM_DELROUTE
, htonl(key
), fa
, plen
, tb
->tb_id
,
1656 &cfg
->fc_nlinfo
, 0);
1658 l
= fib_find_node(t
, key
);
1659 li
= find_leaf_info(l
, plen
);
1661 list_del_rcu(&fa
->fa_list
);
1663 if (list_empty(fa_head
)) {
1664 hlist_del_rcu(&li
->hlist
);
1668 if (hlist_empty(&l
->list
))
1669 trie_leaf_remove(t
, l
);
1671 if (fa
->fa_state
& FA_S_ACCESSED
)
1672 rt_cache_flush(cfg
->fc_nlinfo
.nl_net
, -1);
1674 fib_release_info(fa
->fa_info
);
1675 alias_free_mem_rcu(fa
);
1679 static int trie_flush_list(struct list_head
*head
)
1681 struct fib_alias
*fa
, *fa_node
;
1684 list_for_each_entry_safe(fa
, fa_node
, head
, fa_list
) {
1685 struct fib_info
*fi
= fa
->fa_info
;
1687 if (fi
&& (fi
->fib_flags
& RTNH_F_DEAD
)) {
1688 list_del_rcu(&fa
->fa_list
);
1689 fib_release_info(fa
->fa_info
);
1690 alias_free_mem_rcu(fa
);
1697 static int trie_flush_leaf(struct leaf
*l
)
1700 struct hlist_head
*lih
= &l
->list
;
1701 struct hlist_node
*node
, *tmp
;
1702 struct leaf_info
*li
= NULL
;
1704 hlist_for_each_entry_safe(li
, node
, tmp
, lih
, hlist
) {
1705 found
+= trie_flush_list(&li
->falh
);
1707 if (list_empty(&li
->falh
)) {
1708 hlist_del_rcu(&li
->hlist
);
1716 * Scan for the next right leaf starting at node p->child[idx]
1717 * Since we have back pointer, no recursion necessary.
1719 static struct leaf
*leaf_walk_rcu(struct tnode
*p
, struct node
*c
)
1725 idx
= tkey_extract_bits(c
->key
, p
->pos
, p
->bits
) + 1;
1729 while (idx
< 1u << p
->bits
) {
1730 c
= tnode_get_child_rcu(p
, idx
++);
1735 prefetch(p
->child
[idx
]);
1736 return (struct leaf
*) c
;
1739 /* Rescan start scanning in new node */
1740 p
= (struct tnode
*) c
;
1744 /* Node empty, walk back up to parent */
1745 c
= (struct node
*) p
;
1746 } while ( (p
= node_parent_rcu(c
)) != NULL
);
1748 return NULL
; /* Root of trie */
1751 static struct leaf
*trie_firstleaf(struct trie
*t
)
1753 struct tnode
*n
= (struct tnode
*) rcu_dereference(t
->trie
);
1758 if (IS_LEAF(n
)) /* trie is just a leaf */
1759 return (struct leaf
*) n
;
1761 return leaf_walk_rcu(n
, NULL
);
1764 static struct leaf
*trie_nextleaf(struct leaf
*l
)
1766 struct node
*c
= (struct node
*) l
;
1767 struct tnode
*p
= node_parent_rcu(c
);
1770 return NULL
; /* trie with just one leaf */
1772 return leaf_walk_rcu(p
, c
);
1775 static struct leaf
*trie_leafindex(struct trie
*t
, int index
)
1777 struct leaf
*l
= trie_firstleaf(t
);
1779 while (l
&& index
-- > 0)
1780 l
= trie_nextleaf(l
);
1787 * Caller must hold RTNL.
1789 static int fn_trie_flush(struct fib_table
*tb
)
1791 struct trie
*t
= (struct trie
*) tb
->tb_data
;
1792 struct leaf
*l
, *ll
= NULL
;
1795 for (l
= trie_firstleaf(t
); l
; l
= trie_nextleaf(l
)) {
1796 found
+= trie_flush_leaf(l
);
1798 if (ll
&& hlist_empty(&ll
->list
))
1799 trie_leaf_remove(t
, ll
);
1803 if (ll
&& hlist_empty(&ll
->list
))
1804 trie_leaf_remove(t
, ll
);
1806 pr_debug("trie_flush found=%d\n", found
);
1810 static void fn_trie_select_default(struct fib_table
*tb
,
1811 const struct flowi
*flp
,
1812 struct fib_result
*res
)
1814 struct trie
*t
= (struct trie
*) tb
->tb_data
;
1815 int order
, last_idx
;
1816 struct fib_info
*fi
= NULL
;
1817 struct fib_info
*last_resort
;
1818 struct fib_alias
*fa
= NULL
;
1819 struct list_head
*fa_head
;
1828 l
= fib_find_node(t
, 0);
1832 fa_head
= get_fa_head(l
, 0);
1836 if (list_empty(fa_head
))
1839 list_for_each_entry_rcu(fa
, fa_head
, fa_list
) {
1840 struct fib_info
*next_fi
= fa
->fa_info
;
1842 if (fa
->fa_scope
!= res
->scope
||
1843 fa
->fa_type
!= RTN_UNICAST
)
1846 if (next_fi
->fib_priority
> res
->fi
->fib_priority
)
1848 if (!next_fi
->fib_nh
[0].nh_gw
||
1849 next_fi
->fib_nh
[0].nh_scope
!= RT_SCOPE_LINK
)
1851 fa
->fa_state
|= FA_S_ACCESSED
;
1854 if (next_fi
!= res
->fi
)
1856 } else if (!fib_detect_death(fi
, order
, &last_resort
,
1857 &last_idx
, tb
->tb_default
)) {
1858 fib_result_assign(res
, fi
);
1859 tb
->tb_default
= order
;
1865 if (order
<= 0 || fi
== NULL
) {
1866 tb
->tb_default
= -1;
1870 if (!fib_detect_death(fi
, order
, &last_resort
, &last_idx
,
1872 fib_result_assign(res
, fi
);
1873 tb
->tb_default
= order
;
1877 fib_result_assign(res
, last_resort
);
1878 tb
->tb_default
= last_idx
;
1883 static int fn_trie_dump_fa(t_key key
, int plen
, struct list_head
*fah
,
1884 struct fib_table
*tb
,
1885 struct sk_buff
*skb
, struct netlink_callback
*cb
)
1888 struct fib_alias
*fa
;
1889 __be32 xkey
= htonl(key
);
1894 /* rcu_read_lock is hold by caller */
1896 list_for_each_entry_rcu(fa
, fah
, fa_list
) {
1902 if (fib_dump_info(skb
, NETLINK_CB(cb
->skb
).pid
,
1911 fa
->fa_info
, NLM_F_MULTI
) < 0) {
1921 static int fn_trie_dump_leaf(struct leaf
*l
, struct fib_table
*tb
,
1922 struct sk_buff
*skb
, struct netlink_callback
*cb
)
1924 struct leaf_info
*li
;
1925 struct hlist_node
*node
;
1931 /* rcu_read_lock is hold by caller */
1932 hlist_for_each_entry_rcu(li
, node
, &l
->list
, hlist
) {
1941 if (list_empty(&li
->falh
))
1944 if (fn_trie_dump_fa(l
->key
, li
->plen
, &li
->falh
, tb
, skb
, cb
) < 0) {
1955 static int fn_trie_dump(struct fib_table
*tb
, struct sk_buff
*skb
,
1956 struct netlink_callback
*cb
)
1959 struct trie
*t
= (struct trie
*) tb
->tb_data
;
1960 t_key key
= cb
->args
[2];
1961 int count
= cb
->args
[3];
1964 /* Dump starting at last key.
1965 * Note: 0.0.0.0/0 (ie default) is first key.
1968 l
= trie_firstleaf(t
);
1970 /* Normally, continue from last key, but if that is missing
1971 * fallback to using slow rescan
1973 l
= fib_find_node(t
, key
);
1975 l
= trie_leafindex(t
, count
);
1979 cb
->args
[2] = l
->key
;
1980 if (fn_trie_dump_leaf(l
, tb
, skb
, cb
) < 0) {
1981 cb
->args
[3] = count
;
1987 l
= trie_nextleaf(l
);
1988 memset(&cb
->args
[4], 0,
1989 sizeof(cb
->args
) - 4*sizeof(cb
->args
[0]));
1991 cb
->args
[3] = count
;
1997 void __init
fib_hash_init(void)
1999 fn_alias_kmem
= kmem_cache_create("ip_fib_alias",
2000 sizeof(struct fib_alias
),
2001 0, SLAB_PANIC
, NULL
);
2003 trie_leaf_kmem
= kmem_cache_create("ip_fib_trie",
2004 max(sizeof(struct leaf
),
2005 sizeof(struct leaf_info
)),
2006 0, SLAB_PANIC
, NULL
);
2010 /* Fix more generic FIB names for init later */
2011 struct fib_table
*fib_hash_table(u32 id
)
2013 struct fib_table
*tb
;
2016 tb
= kmalloc(sizeof(struct fib_table
) + sizeof(struct trie
),
2022 tb
->tb_default
= -1;
2023 tb
->tb_lookup
= fn_trie_lookup
;
2024 tb
->tb_insert
= fn_trie_insert
;
2025 tb
->tb_delete
= fn_trie_delete
;
2026 tb
->tb_flush
= fn_trie_flush
;
2027 tb
->tb_select_default
= fn_trie_select_default
;
2028 tb
->tb_dump
= fn_trie_dump
;
2030 t
= (struct trie
*) tb
->tb_data
;
2031 memset(t
, 0, sizeof(*t
));
2033 if (id
== RT_TABLE_LOCAL
)
2034 pr_info("IPv4 FIB: Using LC-trie version %s\n", VERSION
);
2039 #ifdef CONFIG_PROC_FS
2040 /* Depth first Trie walk iterator */
2041 struct fib_trie_iter
{
2042 struct seq_net_private p
;
2043 struct fib_table
*tb
;
2044 struct tnode
*tnode
;
2049 static struct node
*fib_trie_get_next(struct fib_trie_iter
*iter
)
2051 struct tnode
*tn
= iter
->tnode
;
2052 unsigned cindex
= iter
->index
;
2055 /* A single entry routing table */
2059 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2060 iter
->tnode
, iter
->index
, iter
->depth
);
2062 while (cindex
< (1<<tn
->bits
)) {
2063 struct node
*n
= tnode_get_child_rcu(tn
, cindex
);
2068 iter
->index
= cindex
+ 1;
2070 /* push down one level */
2071 iter
->tnode
= (struct tnode
*) n
;
2081 /* Current node exhausted, pop back up */
2082 p
= node_parent_rcu((struct node
*)tn
);
2084 cindex
= tkey_extract_bits(tn
->key
, p
->pos
, p
->bits
)+1;
2094 static struct node
*fib_trie_get_first(struct fib_trie_iter
*iter
,
2102 n
= rcu_dereference(t
->trie
);
2107 iter
->tnode
= (struct tnode
*) n
;
2119 static void trie_collect_stats(struct trie
*t
, struct trie_stat
*s
)
2122 struct fib_trie_iter iter
;
2124 memset(s
, 0, sizeof(*s
));
2127 for (n
= fib_trie_get_first(&iter
, t
); n
; n
= fib_trie_get_next(&iter
)) {
2129 struct leaf
*l
= (struct leaf
*)n
;
2130 struct leaf_info
*li
;
2131 struct hlist_node
*tmp
;
2134 s
->totdepth
+= iter
.depth
;
2135 if (iter
.depth
> s
->maxdepth
)
2136 s
->maxdepth
= iter
.depth
;
2138 hlist_for_each_entry_rcu(li
, tmp
, &l
->list
, hlist
)
2141 const struct tnode
*tn
= (const struct tnode
*) n
;
2145 if (tn
->bits
< MAX_STAT_DEPTH
)
2146 s
->nodesizes
[tn
->bits
]++;
2148 for (i
= 0; i
< (1<<tn
->bits
); i
++)
2157 * This outputs /proc/net/fib_triestats
2159 static void trie_show_stats(struct seq_file
*seq
, struct trie_stat
*stat
)
2161 unsigned i
, max
, pointers
, bytes
, avdepth
;
2164 avdepth
= stat
->totdepth
*100 / stat
->leaves
;
2168 seq_printf(seq
, "\tAver depth: %u.%02d\n",
2169 avdepth
/ 100, avdepth
% 100);
2170 seq_printf(seq
, "\tMax depth: %u\n", stat
->maxdepth
);
2172 seq_printf(seq
, "\tLeaves: %u\n", stat
->leaves
);
2173 bytes
= sizeof(struct leaf
) * stat
->leaves
;
2175 seq_printf(seq
, "\tPrefixes: %u\n", stat
->prefixes
);
2176 bytes
+= sizeof(struct leaf_info
) * stat
->prefixes
;
2178 seq_printf(seq
, "\tInternal nodes: %u\n\t", stat
->tnodes
);
2179 bytes
+= sizeof(struct tnode
) * stat
->tnodes
;
2181 max
= MAX_STAT_DEPTH
;
2182 while (max
> 0 && stat
->nodesizes
[max
-1] == 0)
2186 for (i
= 1; i
<= max
; i
++)
2187 if (stat
->nodesizes
[i
] != 0) {
2188 seq_printf(seq
, " %u: %u", i
, stat
->nodesizes
[i
]);
2189 pointers
+= (1<<i
) * stat
->nodesizes
[i
];
2191 seq_putc(seq
, '\n');
2192 seq_printf(seq
, "\tPointers: %u\n", pointers
);
2194 bytes
+= sizeof(struct node
*) * pointers
;
2195 seq_printf(seq
, "Null ptrs: %u\n", stat
->nullpointers
);
2196 seq_printf(seq
, "Total size: %u kB\n", (bytes
+ 1023) / 1024);
2199 #ifdef CONFIG_IP_FIB_TRIE_STATS
2200 static void trie_show_usage(struct seq_file
*seq
,
2201 const struct trie_use_stats
*stats
)
2203 seq_printf(seq
, "\nCounters:\n---------\n");
2204 seq_printf(seq
, "gets = %u\n", stats
->gets
);
2205 seq_printf(seq
, "backtracks = %u\n", stats
->backtrack
);
2206 seq_printf(seq
, "semantic match passed = %u\n",
2207 stats
->semantic_match_passed
);
2208 seq_printf(seq
, "semantic match miss = %u\n",
2209 stats
->semantic_match_miss
);
2210 seq_printf(seq
, "null node hit= %u\n", stats
->null_node_hit
);
2211 seq_printf(seq
, "skipped node resize = %u\n\n",
2212 stats
->resize_node_skipped
);
2214 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2216 static void fib_table_print(struct seq_file
*seq
, struct fib_table
*tb
)
2218 if (tb
->tb_id
== RT_TABLE_LOCAL
)
2219 seq_puts(seq
, "Local:\n");
2220 else if (tb
->tb_id
== RT_TABLE_MAIN
)
2221 seq_puts(seq
, "Main:\n");
2223 seq_printf(seq
, "Id %d:\n", tb
->tb_id
);
2227 static int fib_triestat_seq_show(struct seq_file
*seq
, void *v
)
2229 struct net
*net
= (struct net
*)seq
->private;
2233 "Basic info: size of leaf:"
2234 " %Zd bytes, size of tnode: %Zd bytes.\n",
2235 sizeof(struct leaf
), sizeof(struct tnode
));
2237 for (h
= 0; h
< FIB_TABLE_HASHSZ
; h
++) {
2238 struct hlist_head
*head
= &net
->ipv4
.fib_table_hash
[h
];
2239 struct hlist_node
*node
;
2240 struct fib_table
*tb
;
2242 hlist_for_each_entry_rcu(tb
, node
, head
, tb_hlist
) {
2243 struct trie
*t
= (struct trie
*) tb
->tb_data
;
2244 struct trie_stat stat
;
2249 fib_table_print(seq
, tb
);
2251 trie_collect_stats(t
, &stat
);
2252 trie_show_stats(seq
, &stat
);
2253 #ifdef CONFIG_IP_FIB_TRIE_STATS
2254 trie_show_usage(seq
, &t
->stats
);
2262 static int fib_triestat_seq_open(struct inode
*inode
, struct file
*file
)
2264 return single_open_net(inode
, file
, fib_triestat_seq_show
);
2267 static const struct file_operations fib_triestat_fops
= {
2268 .owner
= THIS_MODULE
,
2269 .open
= fib_triestat_seq_open
,
2271 .llseek
= seq_lseek
,
2272 .release
= single_release_net
,
2275 static struct node
*fib_trie_get_idx(struct seq_file
*seq
, loff_t pos
)
2277 struct fib_trie_iter
*iter
= seq
->private;
2278 struct net
*net
= seq_file_net(seq
);
2282 for (h
= 0; h
< FIB_TABLE_HASHSZ
; h
++) {
2283 struct hlist_head
*head
= &net
->ipv4
.fib_table_hash
[h
];
2284 struct hlist_node
*node
;
2285 struct fib_table
*tb
;
2287 hlist_for_each_entry_rcu(tb
, node
, head
, tb_hlist
) {
2290 for (n
= fib_trie_get_first(iter
,
2291 (struct trie
*) tb
->tb_data
);
2292 n
; n
= fib_trie_get_next(iter
))
2303 static void *fib_trie_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2307 return fib_trie_get_idx(seq
, *pos
);
2310 static void *fib_trie_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2312 struct fib_trie_iter
*iter
= seq
->private;
2313 struct net
*net
= seq_file_net(seq
);
2314 struct fib_table
*tb
= iter
->tb
;
2315 struct hlist_node
*tb_node
;
2320 /* next node in same table */
2321 n
= fib_trie_get_next(iter
);
2325 /* walk rest of this hash chain */
2326 h
= tb
->tb_id
& (FIB_TABLE_HASHSZ
- 1);
2327 while ( (tb_node
= rcu_dereference(tb
->tb_hlist
.next
)) ) {
2328 tb
= hlist_entry(tb_node
, struct fib_table
, tb_hlist
);
2329 n
= fib_trie_get_first(iter
, (struct trie
*) tb
->tb_data
);
2334 /* new hash chain */
2335 while (++h
< FIB_TABLE_HASHSZ
) {
2336 struct hlist_head
*head
= &net
->ipv4
.fib_table_hash
[h
];
2337 hlist_for_each_entry_rcu(tb
, tb_node
, head
, tb_hlist
) {
2338 n
= fib_trie_get_first(iter
, (struct trie
*) tb
->tb_data
);
2350 static void fib_trie_seq_stop(struct seq_file
*seq
, void *v
)
2356 static void seq_indent(struct seq_file
*seq
, int n
)
2358 while (n
-- > 0) seq_puts(seq
, " ");
2361 static inline const char *rtn_scope(char *buf
, size_t len
, enum rt_scope_t s
)
2364 case RT_SCOPE_UNIVERSE
: return "universe";
2365 case RT_SCOPE_SITE
: return "site";
2366 case RT_SCOPE_LINK
: return "link";
2367 case RT_SCOPE_HOST
: return "host";
2368 case RT_SCOPE_NOWHERE
: return "nowhere";
2370 snprintf(buf
, len
, "scope=%d", s
);
2375 static const char *const rtn_type_names
[__RTN_MAX
] = {
2376 [RTN_UNSPEC
] = "UNSPEC",
2377 [RTN_UNICAST
] = "UNICAST",
2378 [RTN_LOCAL
] = "LOCAL",
2379 [RTN_BROADCAST
] = "BROADCAST",
2380 [RTN_ANYCAST
] = "ANYCAST",
2381 [RTN_MULTICAST
] = "MULTICAST",
2382 [RTN_BLACKHOLE
] = "BLACKHOLE",
2383 [RTN_UNREACHABLE
] = "UNREACHABLE",
2384 [RTN_PROHIBIT
] = "PROHIBIT",
2385 [RTN_THROW
] = "THROW",
2387 [RTN_XRESOLVE
] = "XRESOLVE",
2390 static inline const char *rtn_type(char *buf
, size_t len
, unsigned t
)
2392 if (t
< __RTN_MAX
&& rtn_type_names
[t
])
2393 return rtn_type_names
[t
];
2394 snprintf(buf
, len
, "type %u", t
);
2398 /* Pretty print the trie */
2399 static int fib_trie_seq_show(struct seq_file
*seq
, void *v
)
2401 const struct fib_trie_iter
*iter
= seq
->private;
2404 if (!node_parent_rcu(n
))
2405 fib_table_print(seq
, iter
->tb
);
2408 struct tnode
*tn
= (struct tnode
*) n
;
2409 __be32 prf
= htonl(mask_pfx(tn
->key
, tn
->pos
));
2411 seq_indent(seq
, iter
->depth
-1);
2412 seq_printf(seq
, " +-- %pI4/%d %d %d %d\n",
2413 &prf
, tn
->pos
, tn
->bits
, tn
->full_children
,
2414 tn
->empty_children
);
2417 struct leaf
*l
= (struct leaf
*) n
;
2418 struct leaf_info
*li
;
2419 struct hlist_node
*node
;
2420 __be32 val
= htonl(l
->key
);
2422 seq_indent(seq
, iter
->depth
);
2423 seq_printf(seq
, " |-- %pI4\n", &val
);
2425 hlist_for_each_entry_rcu(li
, node
, &l
->list
, hlist
) {
2426 struct fib_alias
*fa
;
2428 list_for_each_entry_rcu(fa
, &li
->falh
, fa_list
) {
2429 char buf1
[32], buf2
[32];
2431 seq_indent(seq
, iter
->depth
+1);
2432 seq_printf(seq
, " /%d %s %s", li
->plen
,
2433 rtn_scope(buf1
, sizeof(buf1
),
2435 rtn_type(buf2
, sizeof(buf2
),
2438 seq_printf(seq
, " tos=%d", fa
->fa_tos
);
2439 seq_putc(seq
, '\n');
2447 static const struct seq_operations fib_trie_seq_ops
= {
2448 .start
= fib_trie_seq_start
,
2449 .next
= fib_trie_seq_next
,
2450 .stop
= fib_trie_seq_stop
,
2451 .show
= fib_trie_seq_show
,
2454 static int fib_trie_seq_open(struct inode
*inode
, struct file
*file
)
2456 return seq_open_net(inode
, file
, &fib_trie_seq_ops
,
2457 sizeof(struct fib_trie_iter
));
2460 static const struct file_operations fib_trie_fops
= {
2461 .owner
= THIS_MODULE
,
2462 .open
= fib_trie_seq_open
,
2464 .llseek
= seq_lseek
,
2465 .release
= seq_release_net
,
2468 struct fib_route_iter
{
2469 struct seq_net_private p
;
2470 struct trie
*main_trie
;
2475 static struct leaf
*fib_route_get_idx(struct fib_route_iter
*iter
, loff_t pos
)
2477 struct leaf
*l
= NULL
;
2478 struct trie
*t
= iter
->main_trie
;
2480 /* use cache location of last found key */
2481 if (iter
->pos
> 0 && pos
>= iter
->pos
&& (l
= fib_find_node(t
, iter
->key
)))
2485 l
= trie_firstleaf(t
);
2488 while (l
&& pos
-- > 0) {
2490 l
= trie_nextleaf(l
);
2494 iter
->key
= pos
; /* remember it */
2496 iter
->pos
= 0; /* forget it */
2501 static void *fib_route_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2504 struct fib_route_iter
*iter
= seq
->private;
2505 struct fib_table
*tb
;
2508 tb
= fib_get_table(seq_file_net(seq
), RT_TABLE_MAIN
);
2512 iter
->main_trie
= (struct trie
*) tb
->tb_data
;
2514 return SEQ_START_TOKEN
;
2516 return fib_route_get_idx(iter
, *pos
- 1);
2519 static void *fib_route_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2521 struct fib_route_iter
*iter
= seq
->private;
2525 if (v
== SEQ_START_TOKEN
) {
2527 l
= trie_firstleaf(iter
->main_trie
);
2530 l
= trie_nextleaf(l
);
2540 static void fib_route_seq_stop(struct seq_file
*seq
, void *v
)
2546 static unsigned fib_flag_trans(int type
, __be32 mask
, const struct fib_info
*fi
)
2548 static unsigned type2flags
[RTN_MAX
+ 1] = {
2549 [7] = RTF_REJECT
, [8] = RTF_REJECT
,
2551 unsigned flags
= type2flags
[type
];
2553 if (fi
&& fi
->fib_nh
->nh_gw
)
2554 flags
|= RTF_GATEWAY
;
2555 if (mask
== htonl(0xFFFFFFFF))
2562 * This outputs /proc/net/route.
2563 * The format of the file is not supposed to be changed
2564 * and needs to be same as fib_hash output to avoid breaking
2567 static int fib_route_seq_show(struct seq_file
*seq
, void *v
)
2570 struct leaf_info
*li
;
2571 struct hlist_node
*node
;
2573 if (v
== SEQ_START_TOKEN
) {
2574 seq_printf(seq
, "%-127s\n", "Iface\tDestination\tGateway "
2575 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2580 hlist_for_each_entry_rcu(li
, node
, &l
->list
, hlist
) {
2581 struct fib_alias
*fa
;
2582 __be32 mask
, prefix
;
2584 mask
= inet_make_mask(li
->plen
);
2585 prefix
= htonl(l
->key
);
2587 list_for_each_entry_rcu(fa
, &li
->falh
, fa_list
) {
2588 const struct fib_info
*fi
= fa
->fa_info
;
2589 unsigned flags
= fib_flag_trans(fa
->fa_type
, mask
, fi
);
2592 if (fa
->fa_type
== RTN_BROADCAST
2593 || fa
->fa_type
== RTN_MULTICAST
)
2598 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2599 "%d\t%08X\t%d\t%u\t%u%n",
2600 fi
->fib_dev
? fi
->fib_dev
->name
: "*",
2602 fi
->fib_nh
->nh_gw
, flags
, 0, 0,
2606 fi
->fib_advmss
+ 40 : 0),
2608 fi
->fib_rtt
>> 3, &len
);
2611 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2612 "%d\t%08X\t%d\t%u\t%u%n",
2613 prefix
, 0, flags
, 0, 0, 0,
2614 mask
, 0, 0, 0, &len
);
2616 seq_printf(seq
, "%*s\n", 127 - len
, "");
2623 static const struct seq_operations fib_route_seq_ops
= {
2624 .start
= fib_route_seq_start
,
2625 .next
= fib_route_seq_next
,
2626 .stop
= fib_route_seq_stop
,
2627 .show
= fib_route_seq_show
,
2630 static int fib_route_seq_open(struct inode
*inode
, struct file
*file
)
2632 return seq_open_net(inode
, file
, &fib_route_seq_ops
,
2633 sizeof(struct fib_route_iter
));
2636 static const struct file_operations fib_route_fops
= {
2637 .owner
= THIS_MODULE
,
2638 .open
= fib_route_seq_open
,
2640 .llseek
= seq_lseek
,
2641 .release
= seq_release_net
,
2644 int __net_init
fib_proc_init(struct net
*net
)
2646 if (!proc_net_fops_create(net
, "fib_trie", S_IRUGO
, &fib_trie_fops
))
2649 if (!proc_net_fops_create(net
, "fib_triestat", S_IRUGO
,
2650 &fib_triestat_fops
))
2653 if (!proc_net_fops_create(net
, "route", S_IRUGO
, &fib_route_fops
))
2659 proc_net_remove(net
, "fib_triestat");
2661 proc_net_remove(net
, "fib_trie");
2666 void __net_exit
fib_proc_exit(struct net
*net
)
2668 proc_net_remove(net
, "fib_trie");
2669 proc_net_remove(net
, "fib_triestat");
2670 proc_net_remove(net
, "route");
2673 #endif /* CONFIG_PROC_FS */