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.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 <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 <linux/slab.h>
75 #include <net/net_namespace.h>
77 #include <net/protocol.h>
78 #include <net/route.h>
81 #include <net/ip_fib.h>
82 #include "fib_lookup.h"
84 #define MAX_STAT_DEPTH 32
86 #define KEYLENGTH (8*sizeof(t_key))
88 typedef unsigned int t_key
;
92 #define NODE_TYPE_MASK 0x1UL
93 #define NODE_TYPE(node) ((node)->parent & NODE_TYPE_MASK)
95 #define IS_TNODE(n) (!(n->parent & T_LEAF))
96 #define IS_LEAF(n) (n->parent & T_LEAF)
104 unsigned long parent
;
106 struct hlist_head list
;
111 struct hlist_node hlist
;
114 struct list_head falh
;
118 unsigned long parent
;
120 unsigned char pos
; /* 2log(KEYLENGTH) bits needed */
121 unsigned char bits
; /* 2log(KEYLENGTH) bits needed */
122 unsigned int full_children
; /* KEYLENGTH bits needed */
123 unsigned int empty_children
; /* KEYLENGTH bits needed */
126 struct work_struct work
;
127 struct tnode
*tnode_free
;
129 struct node
*child
[0];
132 #ifdef CONFIG_IP_FIB_TRIE_STATS
133 struct trie_use_stats
{
135 unsigned int backtrack
;
136 unsigned int semantic_match_passed
;
137 unsigned int semantic_match_miss
;
138 unsigned int null_node_hit
;
139 unsigned int resize_node_skipped
;
144 unsigned int totdepth
;
145 unsigned int maxdepth
;
148 unsigned int nullpointers
;
149 unsigned int prefixes
;
150 unsigned int nodesizes
[MAX_STAT_DEPTH
];
155 #ifdef CONFIG_IP_FIB_TRIE_STATS
156 struct trie_use_stats stats
;
160 static void put_child(struct trie
*t
, struct tnode
*tn
, int i
, struct node
*n
);
161 static void tnode_put_child_reorg(struct tnode
*tn
, int i
, struct node
*n
,
163 static struct 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
;
180 static inline struct tnode
*node_parent(struct node
*node
)
182 return (struct tnode
*)(node
->parent
& ~NODE_TYPE_MASK
);
185 static inline struct tnode
*node_parent_rcu(struct node
*node
)
187 struct tnode
*ret
= node_parent(node
);
189 return rcu_dereference_rtnl(ret
);
192 /* Same as rcu_assign_pointer
193 * but that macro() assumes that value is a pointer.
195 static inline void node_set_parent(struct node
*node
, struct tnode
*ptr
)
198 node
->parent
= (unsigned long)ptr
| NODE_TYPE(node
);
201 static inline struct node
*tnode_get_child(struct tnode
*tn
, unsigned int i
)
203 BUG_ON(i
>= 1U << tn
->bits
);
208 static inline struct node
*tnode_get_child_rcu(struct tnode
*tn
, unsigned int i
)
210 struct node
*ret
= tnode_get_child(tn
, i
);
212 return rcu_dereference_rtnl(ret
);
215 static inline int tnode_child_length(const struct tnode
*tn
)
217 return 1 << tn
->bits
;
220 static inline t_key
mask_pfx(t_key k
, unsigned short l
)
222 return (l
== 0) ? 0 : k
>> (KEYLENGTH
-l
) << (KEYLENGTH
-l
);
225 static inline t_key
tkey_extract_bits(t_key a
, int offset
, int bits
)
227 if (offset
< KEYLENGTH
)
228 return ((t_key
)(a
<< offset
)) >> (KEYLENGTH
- bits
);
233 static inline int tkey_equals(t_key a
, t_key b
)
238 static inline int tkey_sub_equals(t_key a
, int offset
, int bits
, t_key b
)
240 if (bits
== 0 || offset
>= KEYLENGTH
)
242 bits
= bits
> KEYLENGTH
? KEYLENGTH
: bits
;
243 return ((a
^ b
) << offset
) >> (KEYLENGTH
- bits
) == 0;
246 static inline int tkey_mismatch(t_key a
, int offset
, t_key b
)
253 while ((diff
<< i
) >> (KEYLENGTH
-1) == 0)
259 To understand this stuff, an understanding of keys and all their bits is
260 necessary. Every node in the trie has a key associated with it, but not
261 all of the bits in that key are significant.
263 Consider a node 'n' and its parent 'tp'.
265 If n is a leaf, every bit in its key is significant. Its presence is
266 necessitated by path compression, since during a tree traversal (when
267 searching for a leaf - unless we are doing an insertion) we will completely
268 ignore all skipped bits we encounter. Thus we need to verify, at the end of
269 a potentially successful search, that we have indeed been walking the
272 Note that we can never "miss" the correct key in the tree if present by
273 following the wrong path. Path compression ensures that segments of the key
274 that are the same for all keys with a given prefix are skipped, but the
275 skipped part *is* identical for each node in the subtrie below the skipped
276 bit! trie_insert() in this implementation takes care of that - note the
277 call to tkey_sub_equals() in trie_insert().
279 if n is an internal node - a 'tnode' here, the various parts of its key
280 have many different meanings.
283 _________________________________________________________________
284 | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
285 -----------------------------------------------------------------
286 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
288 _________________________________________________________________
289 | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
290 -----------------------------------------------------------------
291 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
298 First, let's just ignore the bits that come before the parent tp, that is
299 the bits from 0 to (tp->pos-1). They are *known* but at this point we do
300 not use them for anything.
302 The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
303 index into the parent's child array. That is, they will be used to find
304 'n' among tp's children.
306 The bits from (tp->pos + tp->bits) to (n->pos - 1) - "S" - are skipped bits
309 All the bits we have seen so far are significant to the node n. The rest
310 of the bits are really not needed or indeed known in n->key.
312 The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
313 n's child array, and will of course be different for each child.
316 The rest of the bits, from (n->pos + n->bits) onward, are completely unknown
321 static inline void check_tnode(const struct tnode
*tn
)
323 WARN_ON(tn
&& tn
->pos
+tn
->bits
> 32);
326 static const int halve_threshold
= 25;
327 static const int inflate_threshold
= 50;
328 static const int halve_threshold_root
= 15;
329 static const int inflate_threshold_root
= 30;
331 static void __alias_free_mem(struct rcu_head
*head
)
333 struct fib_alias
*fa
= container_of(head
, struct fib_alias
, rcu
);
334 kmem_cache_free(fn_alias_kmem
, fa
);
337 static inline void alias_free_mem_rcu(struct fib_alias
*fa
)
339 call_rcu(&fa
->rcu
, __alias_free_mem
);
342 static void __leaf_free_rcu(struct rcu_head
*head
)
344 struct leaf
*l
= container_of(head
, struct leaf
, rcu
);
345 kmem_cache_free(trie_leaf_kmem
, l
);
348 static inline void free_leaf(struct leaf
*l
)
350 call_rcu_bh(&l
->rcu
, __leaf_free_rcu
);
353 static void __leaf_info_free_rcu(struct rcu_head
*head
)
355 kfree(container_of(head
, struct leaf_info
, rcu
));
358 static inline void free_leaf_info(struct leaf_info
*leaf
)
360 call_rcu(&leaf
->rcu
, __leaf_info_free_rcu
);
363 static struct tnode
*tnode_alloc(size_t size
)
365 if (size
<= PAGE_SIZE
)
366 return kzalloc(size
, GFP_KERNEL
);
368 return __vmalloc(size
, GFP_KERNEL
| __GFP_ZERO
, PAGE_KERNEL
);
371 static void __tnode_vfree(struct work_struct
*arg
)
373 struct tnode
*tn
= container_of(arg
, struct tnode
, work
);
377 static void __tnode_free_rcu(struct rcu_head
*head
)
379 struct tnode
*tn
= container_of(head
, struct tnode
, rcu
);
380 size_t size
= sizeof(struct tnode
) +
381 (sizeof(struct node
*) << tn
->bits
);
383 if (size
<= PAGE_SIZE
)
386 INIT_WORK(&tn
->work
, __tnode_vfree
);
387 schedule_work(&tn
->work
);
391 static inline void tnode_free(struct tnode
*tn
)
394 free_leaf((struct leaf
*) tn
);
396 call_rcu(&tn
->rcu
, __tnode_free_rcu
);
399 static void tnode_free_safe(struct tnode
*tn
)
402 tn
->tnode_free
= tnode_free_head
;
403 tnode_free_head
= tn
;
404 tnode_free_size
+= sizeof(struct tnode
) +
405 (sizeof(struct node
*) << tn
->bits
);
408 static void tnode_free_flush(void)
412 while ((tn
= tnode_free_head
)) {
413 tnode_free_head
= tn
->tnode_free
;
414 tn
->tnode_free
= NULL
;
418 if (tnode_free_size
>= PAGE_SIZE
* sync_pages
) {
424 static struct leaf
*leaf_new(void)
426 struct leaf
*l
= kmem_cache_alloc(trie_leaf_kmem
, GFP_KERNEL
);
429 INIT_HLIST_HEAD(&l
->list
);
434 static struct leaf_info
*leaf_info_new(int plen
)
436 struct leaf_info
*li
= kmalloc(sizeof(struct leaf_info
), GFP_KERNEL
);
439 INIT_LIST_HEAD(&li
->falh
);
444 static struct tnode
*tnode_new(t_key key
, int pos
, int bits
)
446 size_t sz
= sizeof(struct tnode
) + (sizeof(struct node
*) << bits
);
447 struct tnode
*tn
= tnode_alloc(sz
);
450 tn
->parent
= T_TNODE
;
454 tn
->full_children
= 0;
455 tn
->empty_children
= 1<<bits
;
458 pr_debug("AT %p s=%zu %zu\n", tn
, sizeof(struct tnode
),
459 sizeof(struct node
) << bits
);
464 * Check whether a tnode 'n' is "full", i.e. it is an internal node
465 * and no bits are skipped. See discussion in dyntree paper p. 6
468 static inline int tnode_full(const struct tnode
*tn
, const struct node
*n
)
470 if (n
== NULL
|| IS_LEAF(n
))
473 return ((struct tnode
*) n
)->pos
== tn
->pos
+ tn
->bits
;
476 static inline void put_child(struct trie
*t
, struct tnode
*tn
, int i
,
479 tnode_put_child_reorg(tn
, i
, n
, -1);
483 * Add a child at position i overwriting the old value.
484 * Update the value of full_children and empty_children.
487 static void tnode_put_child_reorg(struct tnode
*tn
, int i
, struct node
*n
,
490 struct node
*chi
= tn
->child
[i
];
493 BUG_ON(i
>= 1<<tn
->bits
);
495 /* update emptyChildren */
496 if (n
== NULL
&& chi
!= NULL
)
497 tn
->empty_children
++;
498 else if (n
!= NULL
&& chi
== NULL
)
499 tn
->empty_children
--;
501 /* update fullChildren */
503 wasfull
= tnode_full(tn
, chi
);
505 isfull
= tnode_full(tn
, n
);
506 if (wasfull
&& !isfull
)
508 else if (!wasfull
&& isfull
)
512 node_set_parent(n
, tn
);
514 rcu_assign_pointer(tn
->child
[i
], n
);
518 static struct node
*resize(struct trie
*t
, struct tnode
*tn
)
521 struct tnode
*old_tn
;
522 int inflate_threshold_use
;
523 int halve_threshold_use
;
529 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
530 tn
, inflate_threshold
, halve_threshold
);
533 if (tn
->empty_children
== tnode_child_length(tn
)) {
538 if (tn
->empty_children
== tnode_child_length(tn
) - 1)
541 * Double as long as the resulting node has a number of
542 * nonempty nodes that are above the threshold.
546 * From "Implementing a dynamic compressed trie" by Stefan Nilsson of
547 * the Helsinki University of Technology and Matti Tikkanen of Nokia
548 * Telecommunications, page 6:
549 * "A node is doubled if the ratio of non-empty children to all
550 * children in the *doubled* node is at least 'high'."
552 * 'high' in this instance is the variable 'inflate_threshold'. It
553 * is expressed as a percentage, so we multiply it with
554 * tnode_child_length() and instead of multiplying by 2 (since the
555 * child array will be doubled by inflate()) and multiplying
556 * the left-hand side by 100 (to handle the percentage thing) we
557 * multiply the left-hand side by 50.
559 * The left-hand side may look a bit weird: tnode_child_length(tn)
560 * - tn->empty_children is of course the number of non-null children
561 * in the current node. tn->full_children is the number of "full"
562 * children, that is non-null tnodes with a skip value of 0.
563 * All of those will be doubled in the resulting inflated tnode, so
564 * we just count them one extra time here.
566 * A clearer way to write this would be:
568 * to_be_doubled = tn->full_children;
569 * not_to_be_doubled = tnode_child_length(tn) - tn->empty_children -
572 * new_child_length = tnode_child_length(tn) * 2;
574 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
576 * if (new_fill_factor >= inflate_threshold)
578 * ...and so on, tho it would mess up the while () loop.
581 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
585 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
586 * inflate_threshold * new_child_length
588 * expand not_to_be_doubled and to_be_doubled, and shorten:
589 * 100 * (tnode_child_length(tn) - tn->empty_children +
590 * tn->full_children) >= inflate_threshold * new_child_length
592 * expand new_child_length:
593 * 100 * (tnode_child_length(tn) - tn->empty_children +
594 * tn->full_children) >=
595 * inflate_threshold * tnode_child_length(tn) * 2
598 * 50 * (tn->full_children + tnode_child_length(tn) -
599 * tn->empty_children) >= inflate_threshold *
600 * tnode_child_length(tn)
606 /* Keep root node larger */
608 if (!node_parent((struct node
*)tn
)) {
609 inflate_threshold_use
= inflate_threshold_root
;
610 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_rtnl(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
);
1022 /* only used from updater-side */
1024 static struct list_head
*fib_insert_node(struct trie
*t
, u32 key
, int plen
)
1027 struct tnode
*tp
= NULL
, *tn
= NULL
;
1031 struct list_head
*fa_head
= NULL
;
1032 struct leaf_info
*li
;
1038 /* If we point to NULL, stop. Either the tree is empty and we should
1039 * just put a new leaf in if, or we have reached an empty child slot,
1040 * and we should just put our new leaf in that.
1041 * If we point to a T_TNODE, check if it matches our key. Note that
1042 * a T_TNODE might be skipping any number of bits - its 'pos' need
1043 * not be the parent's 'pos'+'bits'!
1045 * If it does match the current key, get pos/bits from it, extract
1046 * the index from our key, push the T_TNODE and walk the tree.
1048 * If it doesn't, we have to replace it with a new T_TNODE.
1050 * If we point to a T_LEAF, it might or might not have the same key
1051 * as we do. If it does, just change the value, update the T_LEAF's
1052 * value, and return it.
1053 * If it doesn't, we need to replace it with a T_TNODE.
1056 while (n
!= NULL
&& NODE_TYPE(n
) == T_TNODE
) {
1057 tn
= (struct tnode
*) n
;
1061 if (tkey_sub_equals(tn
->key
, pos
, tn
->pos
-pos
, key
)) {
1063 pos
= tn
->pos
+ tn
->bits
;
1064 n
= tnode_get_child(tn
,
1065 tkey_extract_bits(key
,
1069 BUG_ON(n
&& node_parent(n
) != tn
);
1075 * n ----> NULL, LEAF or TNODE
1077 * tp is n's (parent) ----> NULL or TNODE
1080 BUG_ON(tp
&& IS_LEAF(tp
));
1082 /* Case 1: n is a leaf. Compare prefixes */
1084 if (n
!= NULL
&& IS_LEAF(n
) && tkey_equals(key
, n
->key
)) {
1085 l
= (struct leaf
*) n
;
1086 li
= leaf_info_new(plen
);
1091 fa_head
= &li
->falh
;
1092 insert_leaf_info(&l
->list
, li
);
1101 li
= leaf_info_new(plen
);
1108 fa_head
= &li
->falh
;
1109 insert_leaf_info(&l
->list
, li
);
1111 if (t
->trie
&& n
== NULL
) {
1112 /* Case 2: n is NULL, and will just insert a new leaf */
1114 node_set_parent((struct node
*)l
, tp
);
1116 cindex
= tkey_extract_bits(key
, tp
->pos
, tp
->bits
);
1117 put_child(t
, (struct tnode
*)tp
, cindex
, (struct node
*)l
);
1119 /* Case 3: n is a LEAF or a TNODE and the key doesn't match. */
1121 * Add a new tnode here
1122 * first tnode need some special handling
1126 pos
= tp
->pos
+tp
->bits
;
1131 newpos
= tkey_mismatch(key
, pos
, n
->key
);
1132 tn
= tnode_new(n
->key
, newpos
, 1);
1135 tn
= tnode_new(key
, newpos
, 1); /* First tnode */
1144 node_set_parent((struct node
*)tn
, tp
);
1146 missbit
= tkey_extract_bits(key
, newpos
, 1);
1147 put_child(t
, tn
, missbit
, (struct node
*)l
);
1148 put_child(t
, tn
, 1-missbit
, n
);
1151 cindex
= tkey_extract_bits(key
, tp
->pos
, tp
->bits
);
1152 put_child(t
, (struct tnode
*)tp
, cindex
,
1155 rcu_assign_pointer(t
->trie
, (struct node
*)tn
);
1160 if (tp
&& tp
->pos
+ tp
->bits
> 32)
1161 pr_warning("fib_trie"
1162 " tp=%p pos=%d, bits=%d, key=%0x plen=%d\n",
1163 tp
, tp
->pos
, tp
->bits
, key
, plen
);
1165 /* Rebalance the trie */
1167 trie_rebalance(t
, tp
);
1173 * Caller must hold RTNL.
1175 int fib_table_insert(struct fib_table
*tb
, struct fib_config
*cfg
)
1177 struct trie
*t
= (struct trie
*) tb
->tb_data
;
1178 struct fib_alias
*fa
, *new_fa
;
1179 struct list_head
*fa_head
= NULL
;
1180 struct fib_info
*fi
;
1181 int plen
= cfg
->fc_dst_len
;
1182 u8 tos
= cfg
->fc_tos
;
1190 key
= ntohl(cfg
->fc_dst
);
1192 pr_debug("Insert table=%u %08x/%d\n", tb
->tb_id
, key
, plen
);
1194 mask
= ntohl(inet_make_mask(plen
));
1201 fi
= fib_create_info(cfg
);
1207 l
= fib_find_node(t
, key
);
1211 fa_head
= get_fa_head(l
, plen
);
1212 fa
= fib_find_alias(fa_head
, tos
, fi
->fib_priority
);
1215 /* Now fa, if non-NULL, points to the first fib alias
1216 * with the same keys [prefix,tos,priority], if such key already
1217 * exists or to the node before which we will insert new one.
1219 * If fa is NULL, we will need to allocate a new one and
1220 * insert to the head of f.
1222 * If f is NULL, no fib node matched the destination key
1223 * and we need to allocate a new one of those as well.
1226 if (fa
&& fa
->fa_tos
== tos
&&
1227 fa
->fa_info
->fib_priority
== fi
->fib_priority
) {
1228 struct fib_alias
*fa_first
, *fa_match
;
1231 if (cfg
->fc_nlflags
& NLM_F_EXCL
)
1235 * 1. Find exact match for type, scope, fib_info to avoid
1237 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1241 fa
= list_entry(fa
->fa_list
.prev
, struct fib_alias
, fa_list
);
1242 list_for_each_entry_continue(fa
, fa_head
, fa_list
) {
1243 if (fa
->fa_tos
!= tos
)
1245 if (fa
->fa_info
->fib_priority
!= fi
->fib_priority
)
1247 if (fa
->fa_type
== cfg
->fc_type
&&
1248 fa
->fa_scope
== cfg
->fc_scope
&&
1249 fa
->fa_info
== fi
) {
1255 if (cfg
->fc_nlflags
& NLM_F_REPLACE
) {
1256 struct fib_info
*fi_drop
;
1266 new_fa
= kmem_cache_alloc(fn_alias_kmem
, GFP_KERNEL
);
1270 fi_drop
= fa
->fa_info
;
1271 new_fa
->fa_tos
= fa
->fa_tos
;
1272 new_fa
->fa_info
= fi
;
1273 new_fa
->fa_type
= cfg
->fc_type
;
1274 new_fa
->fa_scope
= cfg
->fc_scope
;
1275 state
= fa
->fa_state
;
1276 new_fa
->fa_state
= state
& ~FA_S_ACCESSED
;
1278 list_replace_rcu(&fa
->fa_list
, &new_fa
->fa_list
);
1279 alias_free_mem_rcu(fa
);
1281 fib_release_info(fi_drop
);
1282 if (state
& FA_S_ACCESSED
)
1283 rt_cache_flush(cfg
->fc_nlinfo
.nl_net
, -1);
1284 rtmsg_fib(RTM_NEWROUTE
, htonl(key
), new_fa
, plen
,
1285 tb
->tb_id
, &cfg
->fc_nlinfo
, NLM_F_REPLACE
);
1289 /* Error if we find a perfect match which
1290 * uses the same scope, type, and nexthop
1296 if (!(cfg
->fc_nlflags
& NLM_F_APPEND
))
1300 if (!(cfg
->fc_nlflags
& NLM_F_CREATE
))
1304 new_fa
= kmem_cache_alloc(fn_alias_kmem
, GFP_KERNEL
);
1308 new_fa
->fa_info
= fi
;
1309 new_fa
->fa_tos
= tos
;
1310 new_fa
->fa_type
= cfg
->fc_type
;
1311 new_fa
->fa_scope
= cfg
->fc_scope
;
1312 new_fa
->fa_state
= 0;
1314 * Insert new entry to the list.
1318 fa_head
= fib_insert_node(t
, key
, plen
);
1319 if (unlikely(!fa_head
)) {
1321 goto out_free_new_fa
;
1325 list_add_tail_rcu(&new_fa
->fa_list
,
1326 (fa
? &fa
->fa_list
: fa_head
));
1328 rt_cache_flush(cfg
->fc_nlinfo
.nl_net
, -1);
1329 rtmsg_fib(RTM_NEWROUTE
, htonl(key
), new_fa
, plen
, tb
->tb_id
,
1330 &cfg
->fc_nlinfo
, 0);
1335 kmem_cache_free(fn_alias_kmem
, new_fa
);
1337 fib_release_info(fi
);
1342 /* should be called with rcu_read_lock */
1343 static int check_leaf(struct trie
*t
, struct leaf
*l
,
1344 t_key key
, const struct flowi
*flp
,
1345 struct fib_result
*res
, int fib_flags
)
1347 struct leaf_info
*li
;
1348 struct hlist_head
*hhead
= &l
->list
;
1349 struct hlist_node
*node
;
1351 hlist_for_each_entry_rcu(li
, node
, hhead
, hlist
) {
1353 int plen
= li
->plen
;
1354 __be32 mask
= inet_make_mask(plen
);
1356 if (l
->key
!= (key
& ntohl(mask
)))
1359 err
= fib_semantic_match(&li
->falh
, flp
, res
, plen
, fib_flags
);
1361 #ifdef CONFIG_IP_FIB_TRIE_STATS
1363 t
->stats
.semantic_match_passed
++;
1365 t
->stats
.semantic_match_miss
++;
1374 int fib_table_lookup(struct fib_table
*tb
, const struct flowi
*flp
,
1375 struct fib_result
*res
, int fib_flags
)
1377 struct trie
*t
= (struct trie
*) tb
->tb_data
;
1382 t_key key
= ntohl(flp
->fl4_dst
);
1385 int current_prefix_length
= KEYLENGTH
;
1387 t_key pref_mismatch
;
1391 n
= rcu_dereference(t
->trie
);
1395 #ifdef CONFIG_IP_FIB_TRIE_STATS
1401 ret
= check_leaf(t
, (struct leaf
*)n
, key
, flp
, res
, fib_flags
);
1405 pn
= (struct tnode
*) n
;
1413 cindex
= tkey_extract_bits(mask_pfx(key
, current_prefix_length
),
1416 n
= tnode_get_child_rcu(pn
, cindex
);
1419 #ifdef CONFIG_IP_FIB_TRIE_STATS
1420 t
->stats
.null_node_hit
++;
1426 ret
= check_leaf(t
, (struct leaf
*)n
, key
, flp
, res
, fib_flags
);
1432 cn
= (struct tnode
*)n
;
1435 * It's a tnode, and we can do some extra checks here if we
1436 * like, to avoid descending into a dead-end branch.
1437 * This tnode is in the parent's child array at index
1438 * key[p_pos..p_pos+p_bits] but potentially with some bits
1439 * chopped off, so in reality the index may be just a
1440 * subprefix, padded with zero at the end.
1441 * We can also take a look at any skipped bits in this
1442 * tnode - everything up to p_pos is supposed to be ok,
1443 * and the non-chopped bits of the index (se previous
1444 * paragraph) are also guaranteed ok, but the rest is
1445 * considered unknown.
1447 * The skipped bits are key[pos+bits..cn->pos].
1450 /* If current_prefix_length < pos+bits, we are already doing
1451 * actual prefix matching, which means everything from
1452 * pos+(bits-chopped_off) onward must be zero along some
1453 * branch of this subtree - otherwise there is *no* valid
1454 * prefix present. Here we can only check the skipped
1455 * bits. Remember, since we have already indexed into the
1456 * parent's child array, we know that the bits we chopped of
1460 /* NOTA BENE: Checking only skipped bits
1461 for the new node here */
1463 if (current_prefix_length
< pos
+bits
) {
1464 if (tkey_extract_bits(cn
->key
, current_prefix_length
,
1465 cn
->pos
- current_prefix_length
)
1471 * If chopped_off=0, the index is fully validated and we
1472 * only need to look at the skipped bits for this, the new,
1473 * tnode. What we actually want to do is to find out if
1474 * these skipped bits match our key perfectly, or if we will
1475 * have to count on finding a matching prefix further down,
1476 * because if we do, we would like to have some way of
1477 * verifying the existence of such a prefix at this point.
1480 /* The only thing we can do at this point is to verify that
1481 * any such matching prefix can indeed be a prefix to our
1482 * key, and if the bits in the node we are inspecting that
1483 * do not match our key are not ZERO, this cannot be true.
1484 * Thus, find out where there is a mismatch (before cn->pos)
1485 * and verify that all the mismatching bits are zero in the
1490 * Note: We aren't very concerned about the piece of
1491 * the key that precede pn->pos+pn->bits, since these
1492 * have already been checked. The bits after cn->pos
1493 * aren't checked since these are by definition
1494 * "unknown" at this point. Thus, what we want to see
1495 * is if we are about to enter the "prefix matching"
1496 * state, and in that case verify that the skipped
1497 * bits that will prevail throughout this subtree are
1498 * zero, as they have to be if we are to find a
1502 pref_mismatch
= mask_pfx(cn
->key
^ key
, cn
->pos
);
1505 * In short: If skipped bits in this node do not match
1506 * the search key, enter the "prefix matching"
1509 if (pref_mismatch
) {
1510 int mp
= KEYLENGTH
- fls(pref_mismatch
);
1512 if (tkey_extract_bits(cn
->key
, mp
, cn
->pos
- mp
) != 0)
1515 if (current_prefix_length
>= cn
->pos
)
1516 current_prefix_length
= mp
;
1519 pn
= (struct tnode
*)n
; /* Descend */
1526 /* As zero don't change the child key (cindex) */
1527 while ((chopped_off
<= pn
->bits
)
1528 && !(cindex
& (1<<(chopped_off
-1))))
1531 /* Decrease current_... with bits chopped off */
1532 if (current_prefix_length
> pn
->pos
+ pn
->bits
- chopped_off
)
1533 current_prefix_length
= pn
->pos
+ pn
->bits
1537 * Either we do the actual chop off according or if we have
1538 * chopped off all bits in this tnode walk up to our parent.
1541 if (chopped_off
<= pn
->bits
) {
1542 cindex
&= ~(1 << (chopped_off
-1));
1544 struct tnode
*parent
= node_parent_rcu((struct node
*) pn
);
1548 /* Get Child's index */
1549 cindex
= tkey_extract_bits(pn
->key
, parent
->pos
, parent
->bits
);
1553 #ifdef CONFIG_IP_FIB_TRIE_STATS
1554 t
->stats
.backtrack
++;
1567 * Remove the leaf and return parent.
1569 static void trie_leaf_remove(struct trie
*t
, struct leaf
*l
)
1571 struct tnode
*tp
= node_parent((struct node
*) l
);
1573 pr_debug("entering trie_leaf_remove(%p)\n", l
);
1576 t_key cindex
= tkey_extract_bits(l
->key
, tp
->pos
, tp
->bits
);
1577 put_child(t
, (struct tnode
*)tp
, cindex
, NULL
);
1578 trie_rebalance(t
, tp
);
1580 rcu_assign_pointer(t
->trie
, NULL
);
1586 * Caller must hold RTNL.
1588 int fib_table_delete(struct fib_table
*tb
, struct fib_config
*cfg
)
1590 struct trie
*t
= (struct trie
*) tb
->tb_data
;
1592 int plen
= cfg
->fc_dst_len
;
1593 u8 tos
= cfg
->fc_tos
;
1594 struct fib_alias
*fa
, *fa_to_delete
;
1595 struct list_head
*fa_head
;
1597 struct leaf_info
*li
;
1602 key
= ntohl(cfg
->fc_dst
);
1603 mask
= ntohl(inet_make_mask(plen
));
1609 l
= fib_find_node(t
, key
);
1614 fa_head
= get_fa_head(l
, plen
);
1615 fa
= fib_find_alias(fa_head
, tos
, 0);
1620 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key
, plen
, tos
, t
);
1622 fa_to_delete
= NULL
;
1623 fa
= list_entry(fa
->fa_list
.prev
, struct fib_alias
, fa_list
);
1624 list_for_each_entry_continue(fa
, fa_head
, fa_list
) {
1625 struct fib_info
*fi
= fa
->fa_info
;
1627 if (fa
->fa_tos
!= tos
)
1630 if ((!cfg
->fc_type
|| fa
->fa_type
== cfg
->fc_type
) &&
1631 (cfg
->fc_scope
== RT_SCOPE_NOWHERE
||
1632 fa
->fa_scope
== cfg
->fc_scope
) &&
1633 (!cfg
->fc_protocol
||
1634 fi
->fib_protocol
== cfg
->fc_protocol
) &&
1635 fib_nh_match(cfg
, fi
) == 0) {
1645 rtmsg_fib(RTM_DELROUTE
, htonl(key
), fa
, plen
, tb
->tb_id
,
1646 &cfg
->fc_nlinfo
, 0);
1648 l
= fib_find_node(t
, key
);
1649 li
= find_leaf_info(l
, plen
);
1651 list_del_rcu(&fa
->fa_list
);
1653 if (list_empty(fa_head
)) {
1654 hlist_del_rcu(&li
->hlist
);
1658 if (hlist_empty(&l
->list
))
1659 trie_leaf_remove(t
, l
);
1661 if (fa
->fa_state
& FA_S_ACCESSED
)
1662 rt_cache_flush(cfg
->fc_nlinfo
.nl_net
, -1);
1664 fib_release_info(fa
->fa_info
);
1665 alias_free_mem_rcu(fa
);
1669 static int trie_flush_list(struct list_head
*head
)
1671 struct fib_alias
*fa
, *fa_node
;
1674 list_for_each_entry_safe(fa
, fa_node
, head
, fa_list
) {
1675 struct fib_info
*fi
= fa
->fa_info
;
1677 if (fi
&& (fi
->fib_flags
& RTNH_F_DEAD
)) {
1678 list_del_rcu(&fa
->fa_list
);
1679 fib_release_info(fa
->fa_info
);
1680 alias_free_mem_rcu(fa
);
1687 static int trie_flush_leaf(struct leaf
*l
)
1690 struct hlist_head
*lih
= &l
->list
;
1691 struct hlist_node
*node
, *tmp
;
1692 struct leaf_info
*li
= NULL
;
1694 hlist_for_each_entry_safe(li
, node
, tmp
, lih
, hlist
) {
1695 found
+= trie_flush_list(&li
->falh
);
1697 if (list_empty(&li
->falh
)) {
1698 hlist_del_rcu(&li
->hlist
);
1706 * Scan for the next right leaf starting at node p->child[idx]
1707 * Since we have back pointer, no recursion necessary.
1709 static struct leaf
*leaf_walk_rcu(struct tnode
*p
, struct node
*c
)
1715 idx
= tkey_extract_bits(c
->key
, p
->pos
, p
->bits
) + 1;
1719 while (idx
< 1u << p
->bits
) {
1720 c
= tnode_get_child_rcu(p
, idx
++);
1725 prefetch(p
->child
[idx
]);
1726 return (struct leaf
*) c
;
1729 /* Rescan start scanning in new node */
1730 p
= (struct tnode
*) c
;
1734 /* Node empty, walk back up to parent */
1735 c
= (struct node
*) p
;
1736 } while ((p
= node_parent_rcu(c
)) != NULL
);
1738 return NULL
; /* Root of trie */
1741 static struct leaf
*trie_firstleaf(struct trie
*t
)
1743 struct tnode
*n
= (struct tnode
*)rcu_dereference_rtnl(t
->trie
);
1748 if (IS_LEAF(n
)) /* trie is just a leaf */
1749 return (struct leaf
*) n
;
1751 return leaf_walk_rcu(n
, NULL
);
1754 static struct leaf
*trie_nextleaf(struct leaf
*l
)
1756 struct node
*c
= (struct node
*) l
;
1757 struct tnode
*p
= node_parent_rcu(c
);
1760 return NULL
; /* trie with just one leaf */
1762 return leaf_walk_rcu(p
, c
);
1765 static struct leaf
*trie_leafindex(struct trie
*t
, int index
)
1767 struct leaf
*l
= trie_firstleaf(t
);
1769 while (l
&& index
-- > 0)
1770 l
= trie_nextleaf(l
);
1777 * Caller must hold RTNL.
1779 int fib_table_flush(struct fib_table
*tb
)
1781 struct trie
*t
= (struct trie
*) tb
->tb_data
;
1782 struct leaf
*l
, *ll
= NULL
;
1785 for (l
= trie_firstleaf(t
); l
; l
= trie_nextleaf(l
)) {
1786 found
+= trie_flush_leaf(l
);
1788 if (ll
&& hlist_empty(&ll
->list
))
1789 trie_leaf_remove(t
, ll
);
1793 if (ll
&& hlist_empty(&ll
->list
))
1794 trie_leaf_remove(t
, ll
);
1796 pr_debug("trie_flush found=%d\n", found
);
1800 void fib_free_table(struct fib_table
*tb
)
1805 void fib_table_select_default(struct fib_table
*tb
,
1806 const struct flowi
*flp
,
1807 struct fib_result
*res
)
1809 struct trie
*t
= (struct trie
*) tb
->tb_data
;
1810 int order
, last_idx
;
1811 struct fib_info
*fi
= NULL
;
1812 struct fib_info
*last_resort
;
1813 struct fib_alias
*fa
= NULL
;
1814 struct list_head
*fa_head
;
1823 l
= fib_find_node(t
, 0);
1827 fa_head
= get_fa_head(l
, 0);
1831 if (list_empty(fa_head
))
1834 list_for_each_entry_rcu(fa
, fa_head
, fa_list
) {
1835 struct fib_info
*next_fi
= fa
->fa_info
;
1837 if (fa
->fa_scope
!= res
->scope
||
1838 fa
->fa_type
!= RTN_UNICAST
)
1841 if (next_fi
->fib_priority
> res
->fi
->fib_priority
)
1843 if (!next_fi
->fib_nh
[0].nh_gw
||
1844 next_fi
->fib_nh
[0].nh_scope
!= RT_SCOPE_LINK
)
1847 fib_alias_accessed(fa
);
1850 if (next_fi
!= res
->fi
)
1852 } else if (!fib_detect_death(fi
, order
, &last_resort
,
1853 &last_idx
, tb
->tb_default
)) {
1854 fib_result_assign(res
, fi
);
1855 tb
->tb_default
= order
;
1861 if (order
<= 0 || fi
== NULL
) {
1862 tb
->tb_default
= -1;
1866 if (!fib_detect_death(fi
, order
, &last_resort
, &last_idx
,
1868 fib_result_assign(res
, fi
);
1869 tb
->tb_default
= order
;
1873 fib_result_assign(res
, last_resort
);
1874 tb
->tb_default
= last_idx
;
1879 static int fn_trie_dump_fa(t_key key
, int plen
, struct list_head
*fah
,
1880 struct fib_table
*tb
,
1881 struct sk_buff
*skb
, struct netlink_callback
*cb
)
1884 struct fib_alias
*fa
;
1885 __be32 xkey
= htonl(key
);
1890 /* rcu_read_lock is hold by caller */
1892 list_for_each_entry_rcu(fa
, fah
, fa_list
) {
1898 if (fib_dump_info(skb
, NETLINK_CB(cb
->skb
).pid
,
1907 fa
->fa_info
, NLM_F_MULTI
) < 0) {
1917 static int fn_trie_dump_leaf(struct leaf
*l
, struct fib_table
*tb
,
1918 struct sk_buff
*skb
, struct netlink_callback
*cb
)
1920 struct leaf_info
*li
;
1921 struct hlist_node
*node
;
1927 /* rcu_read_lock is hold by caller */
1928 hlist_for_each_entry_rcu(li
, node
, &l
->list
, hlist
) {
1937 if (list_empty(&li
->falh
))
1940 if (fn_trie_dump_fa(l
->key
, li
->plen
, &li
->falh
, tb
, skb
, cb
) < 0) {
1951 int fib_table_dump(struct fib_table
*tb
, struct sk_buff
*skb
,
1952 struct netlink_callback
*cb
)
1955 struct trie
*t
= (struct trie
*) tb
->tb_data
;
1956 t_key key
= cb
->args
[2];
1957 int count
= cb
->args
[3];
1960 /* Dump starting at last key.
1961 * Note: 0.0.0.0/0 (ie default) is first key.
1964 l
= trie_firstleaf(t
);
1966 /* Normally, continue from last key, but if that is missing
1967 * fallback to using slow rescan
1969 l
= fib_find_node(t
, key
);
1971 l
= trie_leafindex(t
, count
);
1975 cb
->args
[2] = l
->key
;
1976 if (fn_trie_dump_leaf(l
, tb
, skb
, cb
) < 0) {
1977 cb
->args
[3] = count
;
1983 l
= trie_nextleaf(l
);
1984 memset(&cb
->args
[4], 0,
1985 sizeof(cb
->args
) - 4*sizeof(cb
->args
[0]));
1987 cb
->args
[3] = count
;
1993 void __init
fib_hash_init(void)
1995 fn_alias_kmem
= kmem_cache_create("ip_fib_alias",
1996 sizeof(struct fib_alias
),
1997 0, SLAB_PANIC
, NULL
);
1999 trie_leaf_kmem
= kmem_cache_create("ip_fib_trie",
2000 max(sizeof(struct leaf
),
2001 sizeof(struct leaf_info
)),
2002 0, SLAB_PANIC
, NULL
);
2006 /* Fix more generic FIB names for init later */
2007 struct fib_table
*fib_hash_table(u32 id
)
2009 struct fib_table
*tb
;
2012 tb
= kmalloc(sizeof(struct fib_table
) + sizeof(struct trie
),
2018 tb
->tb_default
= -1;
2020 t
= (struct trie
*) tb
->tb_data
;
2021 memset(t
, 0, sizeof(*t
));
2023 if (id
== RT_TABLE_LOCAL
)
2024 pr_info("IPv4 FIB: Using LC-trie version %s\n", VERSION
);
2029 #ifdef CONFIG_PROC_FS
2030 /* Depth first Trie walk iterator */
2031 struct fib_trie_iter
{
2032 struct seq_net_private p
;
2033 struct fib_table
*tb
;
2034 struct tnode
*tnode
;
2039 static struct node
*fib_trie_get_next(struct fib_trie_iter
*iter
)
2041 struct tnode
*tn
= iter
->tnode
;
2042 unsigned int cindex
= iter
->index
;
2045 /* A single entry routing table */
2049 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2050 iter
->tnode
, iter
->index
, iter
->depth
);
2052 while (cindex
< (1<<tn
->bits
)) {
2053 struct node
*n
= tnode_get_child_rcu(tn
, cindex
);
2058 iter
->index
= cindex
+ 1;
2060 /* push down one level */
2061 iter
->tnode
= (struct tnode
*) n
;
2071 /* Current node exhausted, pop back up */
2072 p
= node_parent_rcu((struct node
*)tn
);
2074 cindex
= tkey_extract_bits(tn
->key
, p
->pos
, p
->bits
)+1;
2084 static struct node
*fib_trie_get_first(struct fib_trie_iter
*iter
,
2092 n
= rcu_dereference(t
->trie
);
2097 iter
->tnode
= (struct tnode
*) n
;
2109 static void trie_collect_stats(struct trie
*t
, struct trie_stat
*s
)
2112 struct fib_trie_iter iter
;
2114 memset(s
, 0, sizeof(*s
));
2117 for (n
= fib_trie_get_first(&iter
, t
); n
; n
= fib_trie_get_next(&iter
)) {
2119 struct leaf
*l
= (struct leaf
*)n
;
2120 struct leaf_info
*li
;
2121 struct hlist_node
*tmp
;
2124 s
->totdepth
+= iter
.depth
;
2125 if (iter
.depth
> s
->maxdepth
)
2126 s
->maxdepth
= iter
.depth
;
2128 hlist_for_each_entry_rcu(li
, tmp
, &l
->list
, hlist
)
2131 const struct tnode
*tn
= (const struct tnode
*) n
;
2135 if (tn
->bits
< MAX_STAT_DEPTH
)
2136 s
->nodesizes
[tn
->bits
]++;
2138 for (i
= 0; i
< (1<<tn
->bits
); i
++)
2147 * This outputs /proc/net/fib_triestats
2149 static void trie_show_stats(struct seq_file
*seq
, struct trie_stat
*stat
)
2151 unsigned int i
, max
, pointers
, bytes
, avdepth
;
2154 avdepth
= stat
->totdepth
*100 / stat
->leaves
;
2158 seq_printf(seq
, "\tAver depth: %u.%02d\n",
2159 avdepth
/ 100, avdepth
% 100);
2160 seq_printf(seq
, "\tMax depth: %u\n", stat
->maxdepth
);
2162 seq_printf(seq
, "\tLeaves: %u\n", stat
->leaves
);
2163 bytes
= sizeof(struct leaf
) * stat
->leaves
;
2165 seq_printf(seq
, "\tPrefixes: %u\n", stat
->prefixes
);
2166 bytes
+= sizeof(struct leaf_info
) * stat
->prefixes
;
2168 seq_printf(seq
, "\tInternal nodes: %u\n\t", stat
->tnodes
);
2169 bytes
+= sizeof(struct tnode
) * stat
->tnodes
;
2171 max
= MAX_STAT_DEPTH
;
2172 while (max
> 0 && stat
->nodesizes
[max
-1] == 0)
2176 for (i
= 1; i
<= max
; i
++)
2177 if (stat
->nodesizes
[i
] != 0) {
2178 seq_printf(seq
, " %u: %u", i
, stat
->nodesizes
[i
]);
2179 pointers
+= (1<<i
) * stat
->nodesizes
[i
];
2181 seq_putc(seq
, '\n');
2182 seq_printf(seq
, "\tPointers: %u\n", pointers
);
2184 bytes
+= sizeof(struct node
*) * pointers
;
2185 seq_printf(seq
, "Null ptrs: %u\n", stat
->nullpointers
);
2186 seq_printf(seq
, "Total size: %u kB\n", (bytes
+ 1023) / 1024);
2189 #ifdef CONFIG_IP_FIB_TRIE_STATS
2190 static void trie_show_usage(struct seq_file
*seq
,
2191 const struct trie_use_stats
*stats
)
2193 seq_printf(seq
, "\nCounters:\n---------\n");
2194 seq_printf(seq
, "gets = %u\n", stats
->gets
);
2195 seq_printf(seq
, "backtracks = %u\n", stats
->backtrack
);
2196 seq_printf(seq
, "semantic match passed = %u\n",
2197 stats
->semantic_match_passed
);
2198 seq_printf(seq
, "semantic match miss = %u\n",
2199 stats
->semantic_match_miss
);
2200 seq_printf(seq
, "null node hit= %u\n", stats
->null_node_hit
);
2201 seq_printf(seq
, "skipped node resize = %u\n\n",
2202 stats
->resize_node_skipped
);
2204 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2206 static void fib_table_print(struct seq_file
*seq
, struct fib_table
*tb
)
2208 if (tb
->tb_id
== RT_TABLE_LOCAL
)
2209 seq_puts(seq
, "Local:\n");
2210 else if (tb
->tb_id
== RT_TABLE_MAIN
)
2211 seq_puts(seq
, "Main:\n");
2213 seq_printf(seq
, "Id %d:\n", tb
->tb_id
);
2217 static int fib_triestat_seq_show(struct seq_file
*seq
, void *v
)
2219 struct net
*net
= (struct net
*)seq
->private;
2223 "Basic info: size of leaf:"
2224 " %Zd bytes, size of tnode: %Zd bytes.\n",
2225 sizeof(struct leaf
), sizeof(struct tnode
));
2227 for (h
= 0; h
< FIB_TABLE_HASHSZ
; h
++) {
2228 struct hlist_head
*head
= &net
->ipv4
.fib_table_hash
[h
];
2229 struct hlist_node
*node
;
2230 struct fib_table
*tb
;
2232 hlist_for_each_entry_rcu(tb
, node
, head
, tb_hlist
) {
2233 struct trie
*t
= (struct trie
*) tb
->tb_data
;
2234 struct trie_stat stat
;
2239 fib_table_print(seq
, tb
);
2241 trie_collect_stats(t
, &stat
);
2242 trie_show_stats(seq
, &stat
);
2243 #ifdef CONFIG_IP_FIB_TRIE_STATS
2244 trie_show_usage(seq
, &t
->stats
);
2252 static int fib_triestat_seq_open(struct inode
*inode
, struct file
*file
)
2254 return single_open_net(inode
, file
, fib_triestat_seq_show
);
2257 static const struct file_operations fib_triestat_fops
= {
2258 .owner
= THIS_MODULE
,
2259 .open
= fib_triestat_seq_open
,
2261 .llseek
= seq_lseek
,
2262 .release
= single_release_net
,
2265 static struct node
*fib_trie_get_idx(struct seq_file
*seq
, loff_t pos
)
2267 struct fib_trie_iter
*iter
= seq
->private;
2268 struct net
*net
= seq_file_net(seq
);
2272 for (h
= 0; h
< FIB_TABLE_HASHSZ
; h
++) {
2273 struct hlist_head
*head
= &net
->ipv4
.fib_table_hash
[h
];
2274 struct hlist_node
*node
;
2275 struct fib_table
*tb
;
2277 hlist_for_each_entry_rcu(tb
, node
, head
, tb_hlist
) {
2280 for (n
= fib_trie_get_first(iter
,
2281 (struct trie
*) tb
->tb_data
);
2282 n
; n
= fib_trie_get_next(iter
))
2293 static void *fib_trie_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2297 return fib_trie_get_idx(seq
, *pos
);
2300 static void *fib_trie_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2302 struct fib_trie_iter
*iter
= seq
->private;
2303 struct net
*net
= seq_file_net(seq
);
2304 struct fib_table
*tb
= iter
->tb
;
2305 struct hlist_node
*tb_node
;
2310 /* next node in same table */
2311 n
= fib_trie_get_next(iter
);
2315 /* walk rest of this hash chain */
2316 h
= tb
->tb_id
& (FIB_TABLE_HASHSZ
- 1);
2317 while ( (tb_node
= rcu_dereference(tb
->tb_hlist
.next
)) ) {
2318 tb
= hlist_entry(tb_node
, struct fib_table
, tb_hlist
);
2319 n
= fib_trie_get_first(iter
, (struct trie
*) tb
->tb_data
);
2324 /* new hash chain */
2325 while (++h
< FIB_TABLE_HASHSZ
) {
2326 struct hlist_head
*head
= &net
->ipv4
.fib_table_hash
[h
];
2327 hlist_for_each_entry_rcu(tb
, tb_node
, head
, tb_hlist
) {
2328 n
= fib_trie_get_first(iter
, (struct trie
*) tb
->tb_data
);
2340 static void fib_trie_seq_stop(struct seq_file
*seq
, void *v
)
2346 static void seq_indent(struct seq_file
*seq
, int n
)
2352 static inline const char *rtn_scope(char *buf
, size_t len
, enum rt_scope_t s
)
2355 case RT_SCOPE_UNIVERSE
: return "universe";
2356 case RT_SCOPE_SITE
: return "site";
2357 case RT_SCOPE_LINK
: return "link";
2358 case RT_SCOPE_HOST
: return "host";
2359 case RT_SCOPE_NOWHERE
: return "nowhere";
2361 snprintf(buf
, len
, "scope=%d", s
);
2366 static const char *const rtn_type_names
[__RTN_MAX
] = {
2367 [RTN_UNSPEC
] = "UNSPEC",
2368 [RTN_UNICAST
] = "UNICAST",
2369 [RTN_LOCAL
] = "LOCAL",
2370 [RTN_BROADCAST
] = "BROADCAST",
2371 [RTN_ANYCAST
] = "ANYCAST",
2372 [RTN_MULTICAST
] = "MULTICAST",
2373 [RTN_BLACKHOLE
] = "BLACKHOLE",
2374 [RTN_UNREACHABLE
] = "UNREACHABLE",
2375 [RTN_PROHIBIT
] = "PROHIBIT",
2376 [RTN_THROW
] = "THROW",
2378 [RTN_XRESOLVE
] = "XRESOLVE",
2381 static inline const char *rtn_type(char *buf
, size_t len
, unsigned int t
)
2383 if (t
< __RTN_MAX
&& rtn_type_names
[t
])
2384 return rtn_type_names
[t
];
2385 snprintf(buf
, len
, "type %u", t
);
2389 /* Pretty print the trie */
2390 static int fib_trie_seq_show(struct seq_file
*seq
, void *v
)
2392 const struct fib_trie_iter
*iter
= seq
->private;
2395 if (!node_parent_rcu(n
))
2396 fib_table_print(seq
, iter
->tb
);
2399 struct tnode
*tn
= (struct tnode
*) n
;
2400 __be32 prf
= htonl(mask_pfx(tn
->key
, tn
->pos
));
2402 seq_indent(seq
, iter
->depth
-1);
2403 seq_printf(seq
, " +-- %pI4/%d %d %d %d\n",
2404 &prf
, tn
->pos
, tn
->bits
, tn
->full_children
,
2405 tn
->empty_children
);
2408 struct leaf
*l
= (struct leaf
*) n
;
2409 struct leaf_info
*li
;
2410 struct hlist_node
*node
;
2411 __be32 val
= htonl(l
->key
);
2413 seq_indent(seq
, iter
->depth
);
2414 seq_printf(seq
, " |-- %pI4\n", &val
);
2416 hlist_for_each_entry_rcu(li
, node
, &l
->list
, hlist
) {
2417 struct fib_alias
*fa
;
2419 list_for_each_entry_rcu(fa
, &li
->falh
, fa_list
) {
2420 char buf1
[32], buf2
[32];
2422 seq_indent(seq
, iter
->depth
+1);
2423 seq_printf(seq
, " /%d %s %s", li
->plen
,
2424 rtn_scope(buf1
, sizeof(buf1
),
2426 rtn_type(buf2
, sizeof(buf2
),
2429 seq_printf(seq
, " tos=%d", fa
->fa_tos
);
2430 seq_putc(seq
, '\n');
2438 static const struct seq_operations fib_trie_seq_ops
= {
2439 .start
= fib_trie_seq_start
,
2440 .next
= fib_trie_seq_next
,
2441 .stop
= fib_trie_seq_stop
,
2442 .show
= fib_trie_seq_show
,
2445 static int fib_trie_seq_open(struct inode
*inode
, struct file
*file
)
2447 return seq_open_net(inode
, file
, &fib_trie_seq_ops
,
2448 sizeof(struct fib_trie_iter
));
2451 static const struct file_operations fib_trie_fops
= {
2452 .owner
= THIS_MODULE
,
2453 .open
= fib_trie_seq_open
,
2455 .llseek
= seq_lseek
,
2456 .release
= seq_release_net
,
2459 struct fib_route_iter
{
2460 struct seq_net_private p
;
2461 struct trie
*main_trie
;
2466 static struct leaf
*fib_route_get_idx(struct fib_route_iter
*iter
, loff_t pos
)
2468 struct leaf
*l
= NULL
;
2469 struct trie
*t
= iter
->main_trie
;
2471 /* use cache location of last found key */
2472 if (iter
->pos
> 0 && pos
>= iter
->pos
&& (l
= fib_find_node(t
, iter
->key
)))
2476 l
= trie_firstleaf(t
);
2479 while (l
&& pos
-- > 0) {
2481 l
= trie_nextleaf(l
);
2485 iter
->key
= pos
; /* remember it */
2487 iter
->pos
= 0; /* forget it */
2492 static void *fib_route_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2495 struct fib_route_iter
*iter
= seq
->private;
2496 struct fib_table
*tb
;
2499 tb
= fib_get_table(seq_file_net(seq
), RT_TABLE_MAIN
);
2503 iter
->main_trie
= (struct trie
*) tb
->tb_data
;
2505 return SEQ_START_TOKEN
;
2507 return fib_route_get_idx(iter
, *pos
- 1);
2510 static void *fib_route_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2512 struct fib_route_iter
*iter
= seq
->private;
2516 if (v
== SEQ_START_TOKEN
) {
2518 l
= trie_firstleaf(iter
->main_trie
);
2521 l
= trie_nextleaf(l
);
2531 static void fib_route_seq_stop(struct seq_file
*seq
, void *v
)
2537 static unsigned int fib_flag_trans(int type
, __be32 mask
, const struct fib_info
*fi
)
2539 unsigned int flags
= 0;
2541 if (type
== RTN_UNREACHABLE
|| type
== RTN_PROHIBIT
)
2543 if (fi
&& fi
->fib_nh
->nh_gw
)
2544 flags
|= RTF_GATEWAY
;
2545 if (mask
== htonl(0xFFFFFFFF))
2552 * This outputs /proc/net/route.
2553 * The format of the file is not supposed to be changed
2554 * and needs to be same as fib_hash output to avoid breaking
2557 static int fib_route_seq_show(struct seq_file
*seq
, void *v
)
2560 struct leaf_info
*li
;
2561 struct hlist_node
*node
;
2563 if (v
== SEQ_START_TOKEN
) {
2564 seq_printf(seq
, "%-127s\n", "Iface\tDestination\tGateway "
2565 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2570 hlist_for_each_entry_rcu(li
, node
, &l
->list
, hlist
) {
2571 struct fib_alias
*fa
;
2572 __be32 mask
, prefix
;
2574 mask
= inet_make_mask(li
->plen
);
2575 prefix
= htonl(l
->key
);
2577 list_for_each_entry_rcu(fa
, &li
->falh
, fa_list
) {
2578 const struct fib_info
*fi
= fa
->fa_info
;
2579 unsigned int flags
= fib_flag_trans(fa
->fa_type
, mask
, fi
);
2582 if (fa
->fa_type
== RTN_BROADCAST
2583 || fa
->fa_type
== RTN_MULTICAST
)
2588 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2589 "%d\t%08X\t%d\t%u\t%u%n",
2590 fi
->fib_dev
? fi
->fib_dev
->name
: "*",
2592 fi
->fib_nh
->nh_gw
, flags
, 0, 0,
2596 fi
->fib_advmss
+ 40 : 0),
2598 fi
->fib_rtt
>> 3, &len
);
2601 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2602 "%d\t%08X\t%d\t%u\t%u%n",
2603 prefix
, 0, flags
, 0, 0, 0,
2604 mask
, 0, 0, 0, &len
);
2606 seq_printf(seq
, "%*s\n", 127 - len
, "");
2613 static const struct seq_operations fib_route_seq_ops
= {
2614 .start
= fib_route_seq_start
,
2615 .next
= fib_route_seq_next
,
2616 .stop
= fib_route_seq_stop
,
2617 .show
= fib_route_seq_show
,
2620 static int fib_route_seq_open(struct inode
*inode
, struct file
*file
)
2622 return seq_open_net(inode
, file
, &fib_route_seq_ops
,
2623 sizeof(struct fib_route_iter
));
2626 static const struct file_operations fib_route_fops
= {
2627 .owner
= THIS_MODULE
,
2628 .open
= fib_route_seq_open
,
2630 .llseek
= seq_lseek
,
2631 .release
= seq_release_net
,
2634 int __net_init
fib_proc_init(struct net
*net
)
2636 if (!proc_net_fops_create(net
, "fib_trie", S_IRUGO
, &fib_trie_fops
))
2639 if (!proc_net_fops_create(net
, "fib_triestat", S_IRUGO
,
2640 &fib_triestat_fops
))
2643 if (!proc_net_fops_create(net
, "route", S_IRUGO
, &fib_route_fops
))
2649 proc_net_remove(net
, "fib_triestat");
2651 proc_net_remove(net
, "fib_trie");
2656 void __net_exit
fib_proc_exit(struct net
*net
)
2658 proc_net_remove(net
, "fib_trie");
2659 proc_net_remove(net
, "fib_triestat");
2660 proc_net_remove(net
, "route");
2663 #endif /* CONFIG_PROC_FS */