2 * This program is free software; you can redistribute it and/or
3 * modify it under the terms of the GNU General Public License
4 * as published by the Free Software Foundation; either version
5 * 2 of the License, or (at your option) any later version.
7 * Robert Olsson <robert.olsson@its.uu.se> Uppsala Universitet
8 * & Swedish University of Agricultural Sciences.
10 * Jens Laas <jens.laas@data.slu.se> Swedish University of
11 * Agricultural Sciences.
13 * Hans Liss <hans.liss@its.uu.se> Uppsala Universitet
15 * This work is based on the LPC-trie which is originally described in:
17 * An experimental study of compression methods for dynamic tries
18 * Stefan Nilsson and Matti Tikkanen. Algorithmica, 33(1):19-33, 2002.
19 * http://www.csc.kth.se/~snilsson/software/dyntrie2/
22 * IP-address lookup using LC-tries. Stefan Nilsson and Gunnar Karlsson
23 * IEEE Journal on Selected Areas in Communications, 17(6):1083-1092, June 1999
26 * Code from fib_hash has been reused which includes the following header:
29 * INET An implementation of the TCP/IP protocol suite for the LINUX
30 * operating system. INET is implemented using the BSD Socket
31 * interface as the means of communication with the user level.
33 * IPv4 FIB: lookup engine and maintenance routines.
36 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
38 * This program is free software; you can redistribute it and/or
39 * modify it under the terms of the GNU General Public License
40 * as published by the Free Software Foundation; either version
41 * 2 of the License, or (at your option) any later version.
43 * Substantial contributions to this work comes from:
45 * David S. Miller, <davem@davemloft.net>
46 * Stephen Hemminger <shemminger@osdl.org>
47 * Paul E. McKenney <paulmck@us.ibm.com>
48 * Patrick McHardy <kaber@trash.net>
51 #define VERSION "0.409"
53 #include <asm/uaccess.h>
54 #include <linux/bitops.h>
55 #include <linux/types.h>
56 #include <linux/kernel.h>
58 #include <linux/string.h>
59 #include <linux/socket.h>
60 #include <linux/sockios.h>
61 #include <linux/errno.h>
63 #include <linux/inet.h>
64 #include <linux/inetdevice.h>
65 #include <linux/netdevice.h>
66 #include <linux/if_arp.h>
67 #include <linux/proc_fs.h>
68 #include <linux/rcupdate.h>
69 #include <linux/skbuff.h>
70 #include <linux/netlink.h>
71 #include <linux/init.h>
72 #include <linux/list.h>
73 #include <linux/slab.h>
74 #include <linux/export.h>
75 #include <net/net_namespace.h>
77 #include <net/protocol.h>
78 #include <net/route.h>
81 #include <net/ip_fib.h>
82 #include "fib_lookup.h"
84 #define MAX_STAT_DEPTH 32
86 #define KEYLENGTH (8*sizeof(t_key))
88 typedef unsigned int t_key
;
90 #define IS_TNODE(n) ((n)->bits)
91 #define IS_LEAF(n) (!(n)->bits)
93 #define get_index(_key, _kv) (((_key) ^ (_kv)->key) >> (_kv)->pos)
97 unsigned char bits
; /* 2log(KEYLENGTH) bits needed */
98 unsigned char pos
; /* 2log(KEYLENGTH) bits needed */
99 struct tnode __rcu
*parent
;
102 /* The fields in this struct are valid if bits > 0 (TNODE) */
104 unsigned int full_children
; /* KEYLENGTH bits needed */
105 unsigned int empty_children
; /* KEYLENGTH bits needed */
106 struct tnode __rcu
*child
[0];
108 /* This list pointer if valid if bits == 0 (LEAF) */
109 struct hlist_head list
;
114 struct hlist_node hlist
;
116 u32 mask_plen
; /* ntohl(inet_make_mask(plen)) */
117 struct list_head falh
;
121 #ifdef CONFIG_IP_FIB_TRIE_STATS
122 struct trie_use_stats
{
124 unsigned int backtrack
;
125 unsigned int semantic_match_passed
;
126 unsigned int semantic_match_miss
;
127 unsigned int null_node_hit
;
128 unsigned int resize_node_skipped
;
133 unsigned int totdepth
;
134 unsigned int maxdepth
;
137 unsigned int nullpointers
;
138 unsigned int prefixes
;
139 unsigned int nodesizes
[MAX_STAT_DEPTH
];
143 struct tnode __rcu
*trie
;
144 #ifdef CONFIG_IP_FIB_TRIE_STATS
145 struct trie_use_stats __percpu
*stats
;
149 static void tnode_put_child_reorg(struct tnode
*tn
, unsigned long i
,
150 struct tnode
*n
, int wasfull
);
151 static struct tnode
*resize(struct trie
*t
, struct tnode
*tn
);
152 /* tnodes to free after resize(); protected by RTNL */
153 static struct callback_head
*tnode_free_head
;
154 static size_t tnode_free_size
;
157 * synchronize_rcu after call_rcu for that many pages; it should be especially
158 * useful before resizing the root node with PREEMPT_NONE configs; the value was
159 * obtained experimentally, aiming to avoid visible slowdown.
161 static const int sync_pages
= 128;
163 static struct kmem_cache
*fn_alias_kmem __read_mostly
;
164 static struct kmem_cache
*trie_leaf_kmem __read_mostly
;
166 /* caller must hold RTNL */
167 #define node_parent(n) rtnl_dereference((n)->parent)
169 /* caller must hold RCU read lock or RTNL */
170 #define node_parent_rcu(n) rcu_dereference_rtnl((n)->parent)
172 /* wrapper for rcu_assign_pointer */
173 static inline void node_set_parent(struct tnode
*n
, struct tnode
*tp
)
176 rcu_assign_pointer(n
->parent
, tp
);
179 #define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER((n)->parent, p)
181 /* This provides us with the number of children in this node, in the case of a
182 * leaf this will return 0 meaning none of the children are accessible.
184 static inline unsigned long tnode_child_length(const struct tnode
*tn
)
186 return (1ul << tn
->bits
) & ~(1ul);
189 /* caller must hold RTNL */
190 static inline struct tnode
*tnode_get_child(const struct tnode
*tn
,
193 BUG_ON(i
>= tnode_child_length(tn
));
195 return rtnl_dereference(tn
->child
[i
]);
198 /* caller must hold RCU read lock or RTNL */
199 static inline struct tnode
*tnode_get_child_rcu(const struct tnode
*tn
,
202 BUG_ON(i
>= tnode_child_length(tn
));
204 return rcu_dereference_rtnl(tn
->child
[i
]);
207 /* To understand this stuff, an understanding of keys and all their bits is
208 * necessary. Every node in the trie has a key associated with it, but not
209 * all of the bits in that key are significant.
211 * Consider a node 'n' and its parent 'tp'.
213 * If n is a leaf, every bit in its key is significant. Its presence is
214 * necessitated by path compression, since during a tree traversal (when
215 * searching for a leaf - unless we are doing an insertion) we will completely
216 * ignore all skipped bits we encounter. Thus we need to verify, at the end of
217 * a potentially successful search, that we have indeed been walking the
220 * Note that we can never "miss" the correct key in the tree if present by
221 * following the wrong path. Path compression ensures that segments of the key
222 * that are the same for all keys with a given prefix are skipped, but the
223 * skipped part *is* identical for each node in the subtrie below the skipped
224 * bit! trie_insert() in this implementation takes care of that.
226 * if n is an internal node - a 'tnode' here, the various parts of its key
227 * have many different meanings.
230 * _________________________________________________________________
231 * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
232 * -----------------------------------------------------------------
233 * 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16
235 * _________________________________________________________________
236 * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
237 * -----------------------------------------------------------------
238 * 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
245 * First, let's just ignore the bits that come before the parent tp, that is
246 * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
247 * point we do not use them for anything.
249 * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
250 * index into the parent's child array. That is, they will be used to find
251 * 'n' among tp's children.
253 * The bits from (n->pos + n->bits) to (tn->pos - 1) - "S" - are skipped bits
256 * All the bits we have seen so far are significant to the node n. The rest
257 * of the bits are really not needed or indeed known in n->key.
259 * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
260 * n's child array, and will of course be different for each child.
262 * The rest of the bits, from 0 to (n->pos + n->bits), are completely unknown
266 static const int halve_threshold
= 25;
267 static const int inflate_threshold
= 50;
268 static const int halve_threshold_root
= 15;
269 static const int inflate_threshold_root
= 30;
271 static void __alias_free_mem(struct rcu_head
*head
)
273 struct fib_alias
*fa
= container_of(head
, struct fib_alias
, rcu
);
274 kmem_cache_free(fn_alias_kmem
, fa
);
277 static inline void alias_free_mem_rcu(struct fib_alias
*fa
)
279 call_rcu(&fa
->rcu
, __alias_free_mem
);
282 #define TNODE_KMALLOC_MAX \
283 ilog2((PAGE_SIZE - sizeof(struct tnode)) / sizeof(struct tnode *))
285 static void __node_free_rcu(struct rcu_head
*head
)
287 struct tnode
*n
= container_of(head
, struct tnode
, rcu
);
290 kmem_cache_free(trie_leaf_kmem
, n
);
291 else if (n
->bits
<= TNODE_KMALLOC_MAX
)
297 #define node_free(n) call_rcu(&n->rcu, __node_free_rcu)
299 static inline void free_leaf_info(struct leaf_info
*leaf
)
301 kfree_rcu(leaf
, rcu
);
304 static struct tnode
*tnode_alloc(size_t size
)
306 if (size
<= PAGE_SIZE
)
307 return kzalloc(size
, GFP_KERNEL
);
309 return vzalloc(size
);
312 static void tnode_free_safe(struct tnode
*tn
)
315 tn
->rcu
.next
= tnode_free_head
;
316 tnode_free_head
= &tn
->rcu
;
319 static void tnode_free_flush(void)
321 struct callback_head
*head
;
323 while ((head
= tnode_free_head
)) {
324 struct tnode
*tn
= container_of(head
, struct tnode
, rcu
);
326 tnode_free_head
= head
->next
;
327 tnode_free_size
+= offsetof(struct tnode
, child
[1 << tn
->bits
]);
332 if (tnode_free_size
>= PAGE_SIZE
* sync_pages
) {
338 static struct tnode
*leaf_new(t_key key
)
340 struct tnode
*l
= kmem_cache_alloc(trie_leaf_kmem
, GFP_KERNEL
);
343 /* set key and pos to reflect full key value
344 * any trailing zeros in the key should be ignored
345 * as the nodes are searched
349 /* set bits to 0 indicating we are not a tnode */
352 INIT_HLIST_HEAD(&l
->list
);
357 static struct leaf_info
*leaf_info_new(int plen
)
359 struct leaf_info
*li
= kmalloc(sizeof(struct leaf_info
), GFP_KERNEL
);
362 li
->mask_plen
= ntohl(inet_make_mask(plen
));
363 INIT_LIST_HEAD(&li
->falh
);
368 static struct tnode
*tnode_new(t_key key
, int pos
, int bits
)
370 size_t sz
= offsetof(struct tnode
, child
[1 << bits
]);
371 struct tnode
*tn
= tnode_alloc(sz
);
372 unsigned int shift
= pos
+ bits
;
374 /* verify bits and pos their msb bits clear and values are valid */
375 BUG_ON(!bits
|| (shift
> KEYLENGTH
));
381 tn
->key
= (shift
< KEYLENGTH
) ? (key
>> shift
) << shift
: 0;
382 tn
->full_children
= 0;
383 tn
->empty_children
= 1<<bits
;
386 pr_debug("AT %p s=%zu %zu\n", tn
, sizeof(struct tnode
),
387 sizeof(struct tnode
*) << bits
);
391 /* Check whether a tnode 'n' is "full", i.e. it is an internal node
392 * and no bits are skipped. See discussion in dyntree paper p. 6
394 static inline int tnode_full(const struct tnode
*tn
, const struct tnode
*n
)
396 return n
&& ((n
->pos
+ n
->bits
) == tn
->pos
) && IS_TNODE(n
);
399 static inline void put_child(struct tnode
*tn
, unsigned long i
,
402 tnode_put_child_reorg(tn
, i
, n
, -1);
406 * Add a child at position i overwriting the old value.
407 * Update the value of full_children and empty_children.
410 static void tnode_put_child_reorg(struct tnode
*tn
, unsigned long i
,
411 struct tnode
*n
, int wasfull
)
413 struct tnode
*chi
= rtnl_dereference(tn
->child
[i
]);
416 BUG_ON(i
>= tnode_child_length(tn
));
418 /* update emptyChildren */
419 if (n
== NULL
&& chi
!= NULL
)
420 tn
->empty_children
++;
421 else if (n
!= NULL
&& chi
== NULL
)
422 tn
->empty_children
--;
424 /* update fullChildren */
426 wasfull
= tnode_full(tn
, chi
);
428 isfull
= tnode_full(tn
, n
);
429 if (wasfull
&& !isfull
)
431 else if (!wasfull
&& isfull
)
434 node_set_parent(n
, tn
);
436 rcu_assign_pointer(tn
->child
[i
], n
);
439 static void put_child_root(struct tnode
*tp
, struct trie
*t
,
440 t_key key
, struct tnode
*n
)
443 put_child(tp
, get_index(key
, tp
), n
);
445 rcu_assign_pointer(t
->trie
, n
);
448 static void tnode_clean_free(struct tnode
*tn
)
450 struct tnode
*tofree
;
453 for (i
= 0; i
< tnode_child_length(tn
); i
++) {
454 tofree
= tnode_get_child(tn
, i
);
461 static struct tnode
*inflate(struct trie
*t
, struct tnode
*oldtnode
)
463 unsigned long olen
= tnode_child_length(oldtnode
);
468 pr_debug("In inflate\n");
470 tn
= tnode_new(oldtnode
->key
, oldtnode
->pos
- 1, oldtnode
->bits
+ 1);
473 return ERR_PTR(-ENOMEM
);
476 * Preallocate and store tnodes before the actual work so we
477 * don't get into an inconsistent state if memory allocation
478 * fails. In case of failure we return the oldnode and inflate
479 * of tnode is ignored.
481 for (i
= 0, m
= 1u << tn
->pos
; i
< olen
; i
++) {
482 struct tnode
*inode
= tnode_get_child(oldtnode
, i
);
484 if (tnode_full(oldtnode
, inode
) && (inode
->bits
> 1)) {
485 struct tnode
*left
, *right
;
487 left
= tnode_new(inode
->key
& ~m
, inode
->pos
,
492 right
= tnode_new(inode
->key
| m
, inode
->pos
,
500 put_child(tn
, 2*i
, left
);
501 put_child(tn
, 2*i
+1, right
);
505 for (i
= 0; i
< olen
; i
++) {
506 struct tnode
*inode
= tnode_get_child(oldtnode
, i
);
507 struct tnode
*left
, *right
;
508 unsigned long size
, j
;
514 /* A leaf or an internal node with skipped bits */
515 if (!tnode_full(oldtnode
, inode
)) {
516 put_child(tn
, get_index(inode
->key
, tn
), inode
);
520 /* An internal node with two children */
521 if (inode
->bits
== 1) {
522 put_child(tn
, 2*i
, rtnl_dereference(inode
->child
[0]));
523 put_child(tn
, 2*i
+1, rtnl_dereference(inode
->child
[1]));
525 tnode_free_safe(inode
);
529 /* An internal node with more than two children */
531 /* We will replace this node 'inode' with two new
532 * ones, 'left' and 'right', each with half of the
533 * original children. The two new nodes will have
534 * a position one bit further down the key and this
535 * means that the "significant" part of their keys
536 * (see the discussion near the top of this file)
537 * will differ by one bit, which will be "0" in
538 * left's key and "1" in right's key. Since we are
539 * moving the key position by one step, the bit that
540 * we are moving away from - the bit at position
541 * (inode->pos) - is the one that will differ between
542 * left and right. So... we synthesize that bit in the
544 * The mask 'm' below will be a single "one" bit at
545 * the position (inode->pos)
548 /* Use the old key, but set the new significant
552 left
= tnode_get_child(tn
, 2*i
);
553 put_child(tn
, 2*i
, NULL
);
557 right
= tnode_get_child(tn
, 2*i
+1);
558 put_child(tn
, 2*i
+1, NULL
);
562 size
= tnode_child_length(left
);
563 for (j
= 0; j
< size
; j
++) {
564 put_child(left
, j
, rtnl_dereference(inode
->child
[j
]));
565 put_child(right
, j
, rtnl_dereference(inode
->child
[j
+ size
]));
567 put_child(tn
, 2*i
, resize(t
, left
));
568 put_child(tn
, 2*i
+1, resize(t
, right
));
570 tnode_free_safe(inode
);
572 tnode_free_safe(oldtnode
);
575 tnode_clean_free(tn
);
576 return ERR_PTR(-ENOMEM
);
579 static struct tnode
*halve(struct trie
*t
, struct tnode
*oldtnode
)
581 unsigned long olen
= tnode_child_length(oldtnode
);
582 struct tnode
*tn
, *left
, *right
;
585 pr_debug("In halve\n");
587 tn
= tnode_new(oldtnode
->key
, oldtnode
->pos
+ 1, oldtnode
->bits
- 1);
590 return ERR_PTR(-ENOMEM
);
593 * Preallocate and store tnodes before the actual work so we
594 * don't get into an inconsistent state if memory allocation
595 * fails. In case of failure we return the oldnode and halve
596 * of tnode is ignored.
599 for (i
= 0; i
< olen
; i
+= 2) {
600 left
= tnode_get_child(oldtnode
, i
);
601 right
= tnode_get_child(oldtnode
, i
+1);
603 /* Two nonempty children */
607 newn
= tnode_new(left
->key
, oldtnode
->pos
, 1);
612 put_child(tn
, i
/2, newn
);
617 for (i
= 0; i
< olen
; i
+= 2) {
618 struct tnode
*newBinNode
;
620 left
= tnode_get_child(oldtnode
, i
);
621 right
= tnode_get_child(oldtnode
, i
+1);
623 /* At least one of the children is empty */
625 if (right
== NULL
) /* Both are empty */
627 put_child(tn
, i
/2, right
);
632 put_child(tn
, i
/2, left
);
636 /* Two nonempty children */
637 newBinNode
= tnode_get_child(tn
, i
/2);
638 put_child(tn
, i
/2, NULL
);
639 put_child(newBinNode
, 0, left
);
640 put_child(newBinNode
, 1, right
);
641 put_child(tn
, i
/2, resize(t
, newBinNode
));
643 tnode_free_safe(oldtnode
);
646 tnode_clean_free(tn
);
647 return ERR_PTR(-ENOMEM
);
651 static struct tnode
*resize(struct trie
*t
, struct tnode
*tn
)
653 struct tnode
*old_tn
, *n
= NULL
;
654 int inflate_threshold_use
;
655 int halve_threshold_use
;
661 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
662 tn
, inflate_threshold
, halve_threshold
);
665 if (tn
->empty_children
> (tnode_child_length(tn
) - 1))
669 if (tn
->empty_children
== (tnode_child_length(tn
) - 1))
672 * Double as long as the resulting node has a number of
673 * nonempty nodes that are above the threshold.
677 * From "Implementing a dynamic compressed trie" by Stefan Nilsson of
678 * the Helsinki University of Technology and Matti Tikkanen of Nokia
679 * Telecommunications, page 6:
680 * "A node is doubled if the ratio of non-empty children to all
681 * children in the *doubled* node is at least 'high'."
683 * 'high' in this instance is the variable 'inflate_threshold'. It
684 * is expressed as a percentage, so we multiply it with
685 * tnode_child_length() and instead of multiplying by 2 (since the
686 * child array will be doubled by inflate()) and multiplying
687 * the left-hand side by 100 (to handle the percentage thing) we
688 * multiply the left-hand side by 50.
690 * The left-hand side may look a bit weird: tnode_child_length(tn)
691 * - tn->empty_children is of course the number of non-null children
692 * in the current node. tn->full_children is the number of "full"
693 * children, that is non-null tnodes with a skip value of 0.
694 * All of those will be doubled in the resulting inflated tnode, so
695 * we just count them one extra time here.
697 * A clearer way to write this would be:
699 * to_be_doubled = tn->full_children;
700 * not_to_be_doubled = tnode_child_length(tn) - tn->empty_children -
703 * new_child_length = tnode_child_length(tn) * 2;
705 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
707 * if (new_fill_factor >= inflate_threshold)
709 * ...and so on, tho it would mess up the while () loop.
712 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
716 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
717 * inflate_threshold * new_child_length
719 * expand not_to_be_doubled and to_be_doubled, and shorten:
720 * 100 * (tnode_child_length(tn) - tn->empty_children +
721 * tn->full_children) >= inflate_threshold * new_child_length
723 * expand new_child_length:
724 * 100 * (tnode_child_length(tn) - tn->empty_children +
725 * tn->full_children) >=
726 * inflate_threshold * tnode_child_length(tn) * 2
729 * 50 * (tn->full_children + tnode_child_length(tn) -
730 * tn->empty_children) >= inflate_threshold *
731 * tnode_child_length(tn)
735 /* Keep root node larger */
737 if (!node_parent(tn
)) {
738 inflate_threshold_use
= inflate_threshold_root
;
739 halve_threshold_use
= halve_threshold_root
;
741 inflate_threshold_use
= inflate_threshold
;
742 halve_threshold_use
= halve_threshold
;
746 while ((tn
->full_children
> 0 && max_work
-- &&
747 50 * (tn
->full_children
+ tnode_child_length(tn
)
748 - tn
->empty_children
)
749 >= inflate_threshold_use
* tnode_child_length(tn
))) {
756 #ifdef CONFIG_IP_FIB_TRIE_STATS
757 this_cpu_inc(t
->stats
->resize_node_skipped
);
763 /* Return if at least one inflate is run */
764 if (max_work
!= MAX_WORK
)
768 * Halve as long as the number of empty children in this
769 * node is above threshold.
773 while (tn
->bits
> 1 && max_work
-- &&
774 100 * (tnode_child_length(tn
) - tn
->empty_children
) <
775 halve_threshold_use
* tnode_child_length(tn
)) {
781 #ifdef CONFIG_IP_FIB_TRIE_STATS
782 this_cpu_inc(t
->stats
->resize_node_skipped
);
789 /* Only one child remains */
790 if (tn
->empty_children
== (tnode_child_length(tn
) - 1)) {
793 for (i
= tnode_child_length(tn
); !n
&& i
;)
794 n
= tnode_get_child(tn
, --i
);
796 /* compress one level */
797 node_set_parent(n
, NULL
);
804 /* readside must use rcu_read_lock currently dump routines
805 via get_fa_head and dump */
807 static struct leaf_info
*find_leaf_info(struct tnode
*l
, int plen
)
809 struct hlist_head
*head
= &l
->list
;
810 struct leaf_info
*li
;
812 hlist_for_each_entry_rcu(li
, head
, hlist
)
813 if (li
->plen
== plen
)
819 static inline struct list_head
*get_fa_head(struct tnode
*l
, int plen
)
821 struct leaf_info
*li
= find_leaf_info(l
, plen
);
829 static void insert_leaf_info(struct hlist_head
*head
, struct leaf_info
*new)
831 struct leaf_info
*li
= NULL
, *last
= NULL
;
833 if (hlist_empty(head
)) {
834 hlist_add_head_rcu(&new->hlist
, head
);
836 hlist_for_each_entry(li
, head
, hlist
) {
837 if (new->plen
> li
->plen
)
843 hlist_add_behind_rcu(&new->hlist
, &last
->hlist
);
845 hlist_add_before_rcu(&new->hlist
, &li
->hlist
);
849 /* rcu_read_lock needs to be hold by caller from readside */
850 static struct tnode
*fib_find_node(struct trie
*t
, u32 key
)
852 struct tnode
*n
= rcu_dereference_rtnl(t
->trie
);
855 unsigned long index
= get_index(key
, n
);
857 /* This bit of code is a bit tricky but it combines multiple
858 * checks into a single check. The prefix consists of the
859 * prefix plus zeros for the bits in the cindex. The index
860 * is the difference between the key and this value. From
861 * this we can actually derive several pieces of data.
862 * if !(index >> bits)
863 * we know the value is cindex
865 * we have a mismatch in skip bits and failed
867 if (index
>> n
->bits
)
870 /* we have found a leaf. Prefixes have already been compared */
874 n
= rcu_dereference_rtnl(n
->child
[index
]);
880 static void trie_rebalance(struct trie
*t
, struct tnode
*tn
)
888 while (tn
!= NULL
&& (tp
= node_parent(tn
)) != NULL
) {
889 cindex
= get_index(key
, tp
);
890 wasfull
= tnode_full(tp
, tnode_get_child(tp
, cindex
));
893 tnode_put_child_reorg(tp
, cindex
, tn
, wasfull
);
895 tp
= node_parent(tn
);
897 rcu_assign_pointer(t
->trie
, tn
);
905 /* Handle last (top) tnode */
909 rcu_assign_pointer(t
->trie
, tn
);
913 /* only used from updater-side */
915 static struct list_head
*fib_insert_node(struct trie
*t
, u32 key
, int plen
)
917 struct list_head
*fa_head
= NULL
;
918 struct tnode
*l
, *n
, *tp
= NULL
;
919 struct leaf_info
*li
;
921 li
= leaf_info_new(plen
);
926 n
= rtnl_dereference(t
->trie
);
928 /* If we point to NULL, stop. Either the tree is empty and we should
929 * just put a new leaf in if, or we have reached an empty child slot,
930 * and we should just put our new leaf in that.
932 * If we hit a node with a key that does't match then we should stop
933 * and create a new tnode to replace that node and insert ourselves
934 * and the other node into the new tnode.
937 unsigned long index
= get_index(key
, n
);
939 /* This bit of code is a bit tricky but it combines multiple
940 * checks into a single check. The prefix consists of the
941 * prefix plus zeros for the "bits" in the prefix. The index
942 * is the difference between the key and this value. From
943 * this we can actually derive several pieces of data.
944 * if !(index >> bits)
945 * we know the value is child index
947 * we have a mismatch in skip bits and failed
949 if (index
>> n
->bits
)
952 /* we have found a leaf. Prefixes have already been compared */
954 /* Case 1: n is a leaf, and prefixes match*/
955 insert_leaf_info(&n
->list
, li
);
960 n
= rcu_dereference_rtnl(n
->child
[index
]);
969 insert_leaf_info(&l
->list
, li
);
971 /* Case 2: n is a LEAF or a TNODE and the key doesn't match.
973 * Add a new tnode here
974 * first tnode need some special handling
975 * leaves us in position for handling as case 3
980 tn
= tnode_new(key
, __fls(key
^ n
->key
), 1);
987 /* initialize routes out of node */
988 NODE_INIT_PARENT(tn
, tp
);
989 put_child(tn
, get_index(key
, tn
) ^ 1, n
);
991 /* start adding routes into the node */
992 put_child_root(tp
, t
, key
, tn
);
993 node_set_parent(n
, tn
);
995 /* parent now has a NULL spot where the leaf can go */
999 /* Case 3: n is NULL, and will just insert a new leaf */
1001 NODE_INIT_PARENT(l
, tp
);
1002 put_child(tp
, get_index(key
, tp
), l
);
1003 trie_rebalance(t
, tp
);
1005 rcu_assign_pointer(t
->trie
, l
);
1012 * Caller must hold RTNL.
1014 int fib_table_insert(struct fib_table
*tb
, struct fib_config
*cfg
)
1016 struct trie
*t
= (struct trie
*) tb
->tb_data
;
1017 struct fib_alias
*fa
, *new_fa
;
1018 struct list_head
*fa_head
= NULL
;
1019 struct fib_info
*fi
;
1020 int plen
= cfg
->fc_dst_len
;
1021 u8 tos
= cfg
->fc_tos
;
1029 key
= ntohl(cfg
->fc_dst
);
1031 pr_debug("Insert table=%u %08x/%d\n", tb
->tb_id
, key
, plen
);
1033 mask
= ntohl(inet_make_mask(plen
));
1040 fi
= fib_create_info(cfg
);
1046 l
= fib_find_node(t
, key
);
1050 fa_head
= get_fa_head(l
, plen
);
1051 fa
= fib_find_alias(fa_head
, tos
, fi
->fib_priority
);
1054 /* Now fa, if non-NULL, points to the first fib alias
1055 * with the same keys [prefix,tos,priority], if such key already
1056 * exists or to the node before which we will insert new one.
1058 * If fa is NULL, we will need to allocate a new one and
1059 * insert to the head of f.
1061 * If f is NULL, no fib node matched the destination key
1062 * and we need to allocate a new one of those as well.
1065 if (fa
&& fa
->fa_tos
== tos
&&
1066 fa
->fa_info
->fib_priority
== fi
->fib_priority
) {
1067 struct fib_alias
*fa_first
, *fa_match
;
1070 if (cfg
->fc_nlflags
& NLM_F_EXCL
)
1074 * 1. Find exact match for type, scope, fib_info to avoid
1076 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1080 fa
= list_entry(fa
->fa_list
.prev
, struct fib_alias
, fa_list
);
1081 list_for_each_entry_continue(fa
, fa_head
, fa_list
) {
1082 if (fa
->fa_tos
!= tos
)
1084 if (fa
->fa_info
->fib_priority
!= fi
->fib_priority
)
1086 if (fa
->fa_type
== cfg
->fc_type
&&
1087 fa
->fa_info
== fi
) {
1093 if (cfg
->fc_nlflags
& NLM_F_REPLACE
) {
1094 struct fib_info
*fi_drop
;
1104 new_fa
= kmem_cache_alloc(fn_alias_kmem
, GFP_KERNEL
);
1108 fi_drop
= fa
->fa_info
;
1109 new_fa
->fa_tos
= fa
->fa_tos
;
1110 new_fa
->fa_info
= fi
;
1111 new_fa
->fa_type
= cfg
->fc_type
;
1112 state
= fa
->fa_state
;
1113 new_fa
->fa_state
= state
& ~FA_S_ACCESSED
;
1115 list_replace_rcu(&fa
->fa_list
, &new_fa
->fa_list
);
1116 alias_free_mem_rcu(fa
);
1118 fib_release_info(fi_drop
);
1119 if (state
& FA_S_ACCESSED
)
1120 rt_cache_flush(cfg
->fc_nlinfo
.nl_net
);
1121 rtmsg_fib(RTM_NEWROUTE
, htonl(key
), new_fa
, plen
,
1122 tb
->tb_id
, &cfg
->fc_nlinfo
, NLM_F_REPLACE
);
1126 /* Error if we find a perfect match which
1127 * uses the same scope, type, and nexthop
1133 if (!(cfg
->fc_nlflags
& NLM_F_APPEND
))
1137 if (!(cfg
->fc_nlflags
& NLM_F_CREATE
))
1141 new_fa
= kmem_cache_alloc(fn_alias_kmem
, GFP_KERNEL
);
1145 new_fa
->fa_info
= fi
;
1146 new_fa
->fa_tos
= tos
;
1147 new_fa
->fa_type
= cfg
->fc_type
;
1148 new_fa
->fa_state
= 0;
1150 * Insert new entry to the list.
1154 fa_head
= fib_insert_node(t
, key
, plen
);
1155 if (unlikely(!fa_head
)) {
1157 goto out_free_new_fa
;
1162 tb
->tb_num_default
++;
1164 list_add_tail_rcu(&new_fa
->fa_list
,
1165 (fa
? &fa
->fa_list
: fa_head
));
1167 rt_cache_flush(cfg
->fc_nlinfo
.nl_net
);
1168 rtmsg_fib(RTM_NEWROUTE
, htonl(key
), new_fa
, plen
, tb
->tb_id
,
1169 &cfg
->fc_nlinfo
, 0);
1174 kmem_cache_free(fn_alias_kmem
, new_fa
);
1176 fib_release_info(fi
);
1181 static inline t_key
prefix_mismatch(t_key key
, struct tnode
*n
)
1183 t_key prefix
= n
->key
;
1185 return (key
^ prefix
) & (prefix
| -prefix
);
1188 /* should be called with rcu_read_lock */
1189 int fib_table_lookup(struct fib_table
*tb
, const struct flowi4
*flp
,
1190 struct fib_result
*res
, int fib_flags
)
1192 struct trie
*t
= (struct trie
*)tb
->tb_data
;
1193 #ifdef CONFIG_IP_FIB_TRIE_STATS
1194 struct trie_use_stats __percpu
*stats
= t
->stats
;
1196 const t_key key
= ntohl(flp
->daddr
);
1197 struct tnode
*n
, *pn
;
1198 struct leaf_info
*li
;
1201 n
= rcu_dereference(t
->trie
);
1205 #ifdef CONFIG_IP_FIB_TRIE_STATS
1206 this_cpu_inc(stats
->gets
);
1212 /* Step 1: Travel to the longest prefix match in the trie */
1214 unsigned long index
= get_index(key
, n
);
1216 /* This bit of code is a bit tricky but it combines multiple
1217 * checks into a single check. The prefix consists of the
1218 * prefix plus zeros for the "bits" in the prefix. The index
1219 * is the difference between the key and this value. From
1220 * this we can actually derive several pieces of data.
1221 * if !(index >> bits)
1222 * we know the value is child index
1224 * we have a mismatch in skip bits and failed
1226 if (index
>> n
->bits
)
1229 /* we have found a leaf. Prefixes have already been compared */
1233 /* only record pn and cindex if we are going to be chopping
1234 * bits later. Otherwise we are just wasting cycles.
1241 n
= rcu_dereference(n
->child
[index
]);
1246 /* Step 2: Sort out leaves and begin backtracing for longest prefix */
1248 /* record the pointer where our next node pointer is stored */
1249 struct tnode __rcu
**cptr
= n
->child
;
1251 /* This test verifies that none of the bits that differ
1252 * between the key and the prefix exist in the region of
1253 * the lsb and higher in the prefix.
1255 if (unlikely(prefix_mismatch(key
, n
)))
1258 /* exit out and process leaf */
1259 if (unlikely(IS_LEAF(n
)))
1262 /* Don't bother recording parent info. Since we are in
1263 * prefix match mode we will have to come back to wherever
1264 * we started this traversal anyway
1267 while ((n
= rcu_dereference(*cptr
)) == NULL
) {
1269 #ifdef CONFIG_IP_FIB_TRIE_STATS
1271 this_cpu_inc(stats
->null_node_hit
);
1273 /* If we are at cindex 0 there are no more bits for
1274 * us to strip at this level so we must ascend back
1275 * up one level to see if there are any more bits to
1276 * be stripped there.
1279 t_key pkey
= pn
->key
;
1281 pn
= node_parent_rcu(pn
);
1284 #ifdef CONFIG_IP_FIB_TRIE_STATS
1285 this_cpu_inc(stats
->backtrack
);
1287 /* Get Child's index */
1288 cindex
= get_index(pkey
, pn
);
1291 /* strip the least significant bit from the cindex */
1292 cindex
&= cindex
- 1;
1294 /* grab pointer for next child node */
1295 cptr
= &pn
->child
[cindex
];
1300 /* Step 3: Process the leaf, if that fails fall back to backtracing */
1301 hlist_for_each_entry_rcu(li
, &n
->list
, hlist
) {
1302 struct fib_alias
*fa
;
1304 if ((key
^ n
->key
) & li
->mask_plen
)
1307 list_for_each_entry_rcu(fa
, &li
->falh
, fa_list
) {
1308 struct fib_info
*fi
= fa
->fa_info
;
1311 if (fa
->fa_tos
&& fa
->fa_tos
!= flp
->flowi4_tos
)
1315 if (fa
->fa_info
->fib_scope
< flp
->flowi4_scope
)
1317 fib_alias_accessed(fa
);
1318 err
= fib_props
[fa
->fa_type
].error
;
1319 if (unlikely(err
< 0)) {
1320 #ifdef CONFIG_IP_FIB_TRIE_STATS
1321 this_cpu_inc(stats
->semantic_match_passed
);
1325 if (fi
->fib_flags
& RTNH_F_DEAD
)
1327 for (nhsel
= 0; nhsel
< fi
->fib_nhs
; nhsel
++) {
1328 const struct fib_nh
*nh
= &fi
->fib_nh
[nhsel
];
1330 if (nh
->nh_flags
& RTNH_F_DEAD
)
1332 if (flp
->flowi4_oif
&& flp
->flowi4_oif
!= nh
->nh_oif
)
1335 if (!(fib_flags
& FIB_LOOKUP_NOREF
))
1336 atomic_inc(&fi
->fib_clntref
);
1338 res
->prefixlen
= li
->plen
;
1339 res
->nh_sel
= nhsel
;
1340 res
->type
= fa
->fa_type
;
1341 res
->scope
= fi
->fib_scope
;
1344 res
->fa_head
= &li
->falh
;
1345 #ifdef CONFIG_IP_FIB_TRIE_STATS
1346 this_cpu_inc(stats
->semantic_match_passed
);
1352 #ifdef CONFIG_IP_FIB_TRIE_STATS
1353 this_cpu_inc(stats
->semantic_match_miss
);
1358 EXPORT_SYMBOL_GPL(fib_table_lookup
);
1361 * Remove the leaf and return parent.
1363 static void trie_leaf_remove(struct trie
*t
, struct tnode
*l
)
1365 struct tnode
*tp
= node_parent(l
);
1367 pr_debug("entering trie_leaf_remove(%p)\n", l
);
1370 put_child(tp
, get_index(l
->key
, tp
), NULL
);
1371 trie_rebalance(t
, tp
);
1373 RCU_INIT_POINTER(t
->trie
, NULL
);
1380 * Caller must hold RTNL.
1382 int fib_table_delete(struct fib_table
*tb
, struct fib_config
*cfg
)
1384 struct trie
*t
= (struct trie
*) tb
->tb_data
;
1386 int plen
= cfg
->fc_dst_len
;
1387 u8 tos
= cfg
->fc_tos
;
1388 struct fib_alias
*fa
, *fa_to_delete
;
1389 struct list_head
*fa_head
;
1391 struct leaf_info
*li
;
1396 key
= ntohl(cfg
->fc_dst
);
1397 mask
= ntohl(inet_make_mask(plen
));
1403 l
= fib_find_node(t
, key
);
1408 li
= find_leaf_info(l
, plen
);
1413 fa_head
= &li
->falh
;
1414 fa
= fib_find_alias(fa_head
, tos
, 0);
1419 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key
, plen
, tos
, t
);
1421 fa_to_delete
= NULL
;
1422 fa
= list_entry(fa
->fa_list
.prev
, struct fib_alias
, fa_list
);
1423 list_for_each_entry_continue(fa
, fa_head
, fa_list
) {
1424 struct fib_info
*fi
= fa
->fa_info
;
1426 if (fa
->fa_tos
!= tos
)
1429 if ((!cfg
->fc_type
|| fa
->fa_type
== cfg
->fc_type
) &&
1430 (cfg
->fc_scope
== RT_SCOPE_NOWHERE
||
1431 fa
->fa_info
->fib_scope
== cfg
->fc_scope
) &&
1432 (!cfg
->fc_prefsrc
||
1433 fi
->fib_prefsrc
== cfg
->fc_prefsrc
) &&
1434 (!cfg
->fc_protocol
||
1435 fi
->fib_protocol
== cfg
->fc_protocol
) &&
1436 fib_nh_match(cfg
, fi
) == 0) {
1446 rtmsg_fib(RTM_DELROUTE
, htonl(key
), fa
, plen
, tb
->tb_id
,
1447 &cfg
->fc_nlinfo
, 0);
1449 list_del_rcu(&fa
->fa_list
);
1452 tb
->tb_num_default
--;
1454 if (list_empty(fa_head
)) {
1455 hlist_del_rcu(&li
->hlist
);
1459 if (hlist_empty(&l
->list
))
1460 trie_leaf_remove(t
, l
);
1462 if (fa
->fa_state
& FA_S_ACCESSED
)
1463 rt_cache_flush(cfg
->fc_nlinfo
.nl_net
);
1465 fib_release_info(fa
->fa_info
);
1466 alias_free_mem_rcu(fa
);
1470 static int trie_flush_list(struct list_head
*head
)
1472 struct fib_alias
*fa
, *fa_node
;
1475 list_for_each_entry_safe(fa
, fa_node
, head
, fa_list
) {
1476 struct fib_info
*fi
= fa
->fa_info
;
1478 if (fi
&& (fi
->fib_flags
& RTNH_F_DEAD
)) {
1479 list_del_rcu(&fa
->fa_list
);
1480 fib_release_info(fa
->fa_info
);
1481 alias_free_mem_rcu(fa
);
1488 static int trie_flush_leaf(struct tnode
*l
)
1491 struct hlist_head
*lih
= &l
->list
;
1492 struct hlist_node
*tmp
;
1493 struct leaf_info
*li
= NULL
;
1495 hlist_for_each_entry_safe(li
, tmp
, lih
, hlist
) {
1496 found
+= trie_flush_list(&li
->falh
);
1498 if (list_empty(&li
->falh
)) {
1499 hlist_del_rcu(&li
->hlist
);
1507 * Scan for the next right leaf starting at node p->child[idx]
1508 * Since we have back pointer, no recursion necessary.
1510 static struct tnode
*leaf_walk_rcu(struct tnode
*p
, struct tnode
*c
)
1513 unsigned long idx
= c
? idx
= get_index(c
->key
, p
) + 1 : 0;
1515 while (idx
< tnode_child_length(p
)) {
1516 c
= tnode_get_child_rcu(p
, idx
++);
1523 /* Rescan start scanning in new node */
1528 /* Node empty, walk back up to parent */
1530 } while ((p
= node_parent_rcu(c
)) != NULL
);
1532 return NULL
; /* Root of trie */
1535 static struct tnode
*trie_firstleaf(struct trie
*t
)
1537 struct tnode
*n
= rcu_dereference_rtnl(t
->trie
);
1542 if (IS_LEAF(n
)) /* trie is just a leaf */
1545 return leaf_walk_rcu(n
, NULL
);
1548 static struct tnode
*trie_nextleaf(struct tnode
*l
)
1550 struct tnode
*p
= node_parent_rcu(l
);
1553 return NULL
; /* trie with just one leaf */
1555 return leaf_walk_rcu(p
, l
);
1558 static struct tnode
*trie_leafindex(struct trie
*t
, int index
)
1560 struct tnode
*l
= trie_firstleaf(t
);
1562 while (l
&& index
-- > 0)
1563 l
= trie_nextleaf(l
);
1570 * Caller must hold RTNL.
1572 int fib_table_flush(struct fib_table
*tb
)
1574 struct trie
*t
= (struct trie
*) tb
->tb_data
;
1575 struct tnode
*l
, *ll
= NULL
;
1578 for (l
= trie_firstleaf(t
); l
; l
= trie_nextleaf(l
)) {
1579 found
+= trie_flush_leaf(l
);
1581 if (ll
&& hlist_empty(&ll
->list
))
1582 trie_leaf_remove(t
, ll
);
1586 if (ll
&& hlist_empty(&ll
->list
))
1587 trie_leaf_remove(t
, ll
);
1589 pr_debug("trie_flush found=%d\n", found
);
1593 void fib_free_table(struct fib_table
*tb
)
1595 #ifdef CONFIG_IP_FIB_TRIE_STATS
1596 struct trie
*t
= (struct trie
*)tb
->tb_data
;
1598 free_percpu(t
->stats
);
1599 #endif /* CONFIG_IP_FIB_TRIE_STATS */
1603 static int fn_trie_dump_fa(t_key key
, int plen
, struct list_head
*fah
,
1604 struct fib_table
*tb
,
1605 struct sk_buff
*skb
, struct netlink_callback
*cb
)
1608 struct fib_alias
*fa
;
1609 __be32 xkey
= htonl(key
);
1614 /* rcu_read_lock is hold by caller */
1616 list_for_each_entry_rcu(fa
, fah
, fa_list
) {
1622 if (fib_dump_info(skb
, NETLINK_CB(cb
->skb
).portid
,
1630 fa
->fa_info
, NLM_F_MULTI
) < 0) {
1640 static int fn_trie_dump_leaf(struct tnode
*l
, struct fib_table
*tb
,
1641 struct sk_buff
*skb
, struct netlink_callback
*cb
)
1643 struct leaf_info
*li
;
1649 /* rcu_read_lock is hold by caller */
1650 hlist_for_each_entry_rcu(li
, &l
->list
, hlist
) {
1659 if (list_empty(&li
->falh
))
1662 if (fn_trie_dump_fa(l
->key
, li
->plen
, &li
->falh
, tb
, skb
, cb
) < 0) {
1673 int fib_table_dump(struct fib_table
*tb
, struct sk_buff
*skb
,
1674 struct netlink_callback
*cb
)
1677 struct trie
*t
= (struct trie
*) tb
->tb_data
;
1678 t_key key
= cb
->args
[2];
1679 int count
= cb
->args
[3];
1682 /* Dump starting at last key.
1683 * Note: 0.0.0.0/0 (ie default) is first key.
1686 l
= trie_firstleaf(t
);
1688 /* Normally, continue from last key, but if that is missing
1689 * fallback to using slow rescan
1691 l
= fib_find_node(t
, key
);
1693 l
= trie_leafindex(t
, count
);
1697 cb
->args
[2] = l
->key
;
1698 if (fn_trie_dump_leaf(l
, tb
, skb
, cb
) < 0) {
1699 cb
->args
[3] = count
;
1705 l
= trie_nextleaf(l
);
1706 memset(&cb
->args
[4], 0,
1707 sizeof(cb
->args
) - 4*sizeof(cb
->args
[0]));
1709 cb
->args
[3] = count
;
1715 void __init
fib_trie_init(void)
1717 fn_alias_kmem
= kmem_cache_create("ip_fib_alias",
1718 sizeof(struct fib_alias
),
1719 0, SLAB_PANIC
, NULL
);
1721 trie_leaf_kmem
= kmem_cache_create("ip_fib_trie",
1722 max(sizeof(struct tnode
),
1723 sizeof(struct leaf_info
)),
1724 0, SLAB_PANIC
, NULL
);
1728 struct fib_table
*fib_trie_table(u32 id
)
1730 struct fib_table
*tb
;
1733 tb
= kmalloc(sizeof(struct fib_table
) + sizeof(struct trie
),
1739 tb
->tb_default
= -1;
1740 tb
->tb_num_default
= 0;
1742 t
= (struct trie
*) tb
->tb_data
;
1743 RCU_INIT_POINTER(t
->trie
, NULL
);
1744 #ifdef CONFIG_IP_FIB_TRIE_STATS
1745 t
->stats
= alloc_percpu(struct trie_use_stats
);
1755 #ifdef CONFIG_PROC_FS
1756 /* Depth first Trie walk iterator */
1757 struct fib_trie_iter
{
1758 struct seq_net_private p
;
1759 struct fib_table
*tb
;
1760 struct tnode
*tnode
;
1765 static struct tnode
*fib_trie_get_next(struct fib_trie_iter
*iter
)
1767 unsigned long cindex
= iter
->index
;
1768 struct tnode
*tn
= iter
->tnode
;
1771 /* A single entry routing table */
1775 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
1776 iter
->tnode
, iter
->index
, iter
->depth
);
1778 while (cindex
< tnode_child_length(tn
)) {
1779 struct tnode
*n
= tnode_get_child_rcu(tn
, cindex
);
1784 iter
->index
= cindex
+ 1;
1786 /* push down one level */
1797 /* Current node exhausted, pop back up */
1798 p
= node_parent_rcu(tn
);
1800 cindex
= get_index(tn
->key
, p
) + 1;
1810 static struct tnode
*fib_trie_get_first(struct fib_trie_iter
*iter
,
1818 n
= rcu_dereference(t
->trie
);
1835 static void trie_collect_stats(struct trie
*t
, struct trie_stat
*s
)
1838 struct fib_trie_iter iter
;
1840 memset(s
, 0, sizeof(*s
));
1843 for (n
= fib_trie_get_first(&iter
, t
); n
; n
= fib_trie_get_next(&iter
)) {
1845 struct leaf_info
*li
;
1848 s
->totdepth
+= iter
.depth
;
1849 if (iter
.depth
> s
->maxdepth
)
1850 s
->maxdepth
= iter
.depth
;
1852 hlist_for_each_entry_rcu(li
, &n
->list
, hlist
)
1858 if (n
->bits
< MAX_STAT_DEPTH
)
1859 s
->nodesizes
[n
->bits
]++;
1861 for (i
= 0; i
< tnode_child_length(n
); i
++) {
1862 if (!rcu_access_pointer(n
->child
[i
]))
1871 * This outputs /proc/net/fib_triestats
1873 static void trie_show_stats(struct seq_file
*seq
, struct trie_stat
*stat
)
1875 unsigned int i
, max
, pointers
, bytes
, avdepth
;
1878 avdepth
= stat
->totdepth
*100 / stat
->leaves
;
1882 seq_printf(seq
, "\tAver depth: %u.%02d\n",
1883 avdepth
/ 100, avdepth
% 100);
1884 seq_printf(seq
, "\tMax depth: %u\n", stat
->maxdepth
);
1886 seq_printf(seq
, "\tLeaves: %u\n", stat
->leaves
);
1887 bytes
= sizeof(struct tnode
) * stat
->leaves
;
1889 seq_printf(seq
, "\tPrefixes: %u\n", stat
->prefixes
);
1890 bytes
+= sizeof(struct leaf_info
) * stat
->prefixes
;
1892 seq_printf(seq
, "\tInternal nodes: %u\n\t", stat
->tnodes
);
1893 bytes
+= sizeof(struct tnode
) * stat
->tnodes
;
1895 max
= MAX_STAT_DEPTH
;
1896 while (max
> 0 && stat
->nodesizes
[max
-1] == 0)
1900 for (i
= 1; i
< max
; i
++)
1901 if (stat
->nodesizes
[i
] != 0) {
1902 seq_printf(seq
, " %u: %u", i
, stat
->nodesizes
[i
]);
1903 pointers
+= (1<<i
) * stat
->nodesizes
[i
];
1905 seq_putc(seq
, '\n');
1906 seq_printf(seq
, "\tPointers: %u\n", pointers
);
1908 bytes
+= sizeof(struct tnode
*) * pointers
;
1909 seq_printf(seq
, "Null ptrs: %u\n", stat
->nullpointers
);
1910 seq_printf(seq
, "Total size: %u kB\n", (bytes
+ 1023) / 1024);
1913 #ifdef CONFIG_IP_FIB_TRIE_STATS
1914 static void trie_show_usage(struct seq_file
*seq
,
1915 const struct trie_use_stats __percpu
*stats
)
1917 struct trie_use_stats s
= { 0 };
1920 /* loop through all of the CPUs and gather up the stats */
1921 for_each_possible_cpu(cpu
) {
1922 const struct trie_use_stats
*pcpu
= per_cpu_ptr(stats
, cpu
);
1924 s
.gets
+= pcpu
->gets
;
1925 s
.backtrack
+= pcpu
->backtrack
;
1926 s
.semantic_match_passed
+= pcpu
->semantic_match_passed
;
1927 s
.semantic_match_miss
+= pcpu
->semantic_match_miss
;
1928 s
.null_node_hit
+= pcpu
->null_node_hit
;
1929 s
.resize_node_skipped
+= pcpu
->resize_node_skipped
;
1932 seq_printf(seq
, "\nCounters:\n---------\n");
1933 seq_printf(seq
, "gets = %u\n", s
.gets
);
1934 seq_printf(seq
, "backtracks = %u\n", s
.backtrack
);
1935 seq_printf(seq
, "semantic match passed = %u\n",
1936 s
.semantic_match_passed
);
1937 seq_printf(seq
, "semantic match miss = %u\n", s
.semantic_match_miss
);
1938 seq_printf(seq
, "null node hit= %u\n", s
.null_node_hit
);
1939 seq_printf(seq
, "skipped node resize = %u\n\n", s
.resize_node_skipped
);
1941 #endif /* CONFIG_IP_FIB_TRIE_STATS */
1943 static void fib_table_print(struct seq_file
*seq
, struct fib_table
*tb
)
1945 if (tb
->tb_id
== RT_TABLE_LOCAL
)
1946 seq_puts(seq
, "Local:\n");
1947 else if (tb
->tb_id
== RT_TABLE_MAIN
)
1948 seq_puts(seq
, "Main:\n");
1950 seq_printf(seq
, "Id %d:\n", tb
->tb_id
);
1954 static int fib_triestat_seq_show(struct seq_file
*seq
, void *v
)
1956 struct net
*net
= (struct net
*)seq
->private;
1960 "Basic info: size of leaf:"
1961 " %Zd bytes, size of tnode: %Zd bytes.\n",
1962 sizeof(struct tnode
), sizeof(struct tnode
));
1964 for (h
= 0; h
< FIB_TABLE_HASHSZ
; h
++) {
1965 struct hlist_head
*head
= &net
->ipv4
.fib_table_hash
[h
];
1966 struct fib_table
*tb
;
1968 hlist_for_each_entry_rcu(tb
, head
, tb_hlist
) {
1969 struct trie
*t
= (struct trie
*) tb
->tb_data
;
1970 struct trie_stat stat
;
1975 fib_table_print(seq
, tb
);
1977 trie_collect_stats(t
, &stat
);
1978 trie_show_stats(seq
, &stat
);
1979 #ifdef CONFIG_IP_FIB_TRIE_STATS
1980 trie_show_usage(seq
, t
->stats
);
1988 static int fib_triestat_seq_open(struct inode
*inode
, struct file
*file
)
1990 return single_open_net(inode
, file
, fib_triestat_seq_show
);
1993 static const struct file_operations fib_triestat_fops
= {
1994 .owner
= THIS_MODULE
,
1995 .open
= fib_triestat_seq_open
,
1997 .llseek
= seq_lseek
,
1998 .release
= single_release_net
,
2001 static struct tnode
*fib_trie_get_idx(struct seq_file
*seq
, loff_t pos
)
2003 struct fib_trie_iter
*iter
= seq
->private;
2004 struct net
*net
= seq_file_net(seq
);
2008 for (h
= 0; h
< FIB_TABLE_HASHSZ
; h
++) {
2009 struct hlist_head
*head
= &net
->ipv4
.fib_table_hash
[h
];
2010 struct fib_table
*tb
;
2012 hlist_for_each_entry_rcu(tb
, head
, tb_hlist
) {
2015 for (n
= fib_trie_get_first(iter
,
2016 (struct trie
*) tb
->tb_data
);
2017 n
; n
= fib_trie_get_next(iter
))
2028 static void *fib_trie_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2032 return fib_trie_get_idx(seq
, *pos
);
2035 static void *fib_trie_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2037 struct fib_trie_iter
*iter
= seq
->private;
2038 struct net
*net
= seq_file_net(seq
);
2039 struct fib_table
*tb
= iter
->tb
;
2040 struct hlist_node
*tb_node
;
2045 /* next node in same table */
2046 n
= fib_trie_get_next(iter
);
2050 /* walk rest of this hash chain */
2051 h
= tb
->tb_id
& (FIB_TABLE_HASHSZ
- 1);
2052 while ((tb_node
= rcu_dereference(hlist_next_rcu(&tb
->tb_hlist
)))) {
2053 tb
= hlist_entry(tb_node
, struct fib_table
, tb_hlist
);
2054 n
= fib_trie_get_first(iter
, (struct trie
*) tb
->tb_data
);
2059 /* new hash chain */
2060 while (++h
< FIB_TABLE_HASHSZ
) {
2061 struct hlist_head
*head
= &net
->ipv4
.fib_table_hash
[h
];
2062 hlist_for_each_entry_rcu(tb
, head
, tb_hlist
) {
2063 n
= fib_trie_get_first(iter
, (struct trie
*) tb
->tb_data
);
2075 static void fib_trie_seq_stop(struct seq_file
*seq
, void *v
)
2081 static void seq_indent(struct seq_file
*seq
, int n
)
2087 static inline const char *rtn_scope(char *buf
, size_t len
, enum rt_scope_t s
)
2090 case RT_SCOPE_UNIVERSE
: return "universe";
2091 case RT_SCOPE_SITE
: return "site";
2092 case RT_SCOPE_LINK
: return "link";
2093 case RT_SCOPE_HOST
: return "host";
2094 case RT_SCOPE_NOWHERE
: return "nowhere";
2096 snprintf(buf
, len
, "scope=%d", s
);
2101 static const char *const rtn_type_names
[__RTN_MAX
] = {
2102 [RTN_UNSPEC
] = "UNSPEC",
2103 [RTN_UNICAST
] = "UNICAST",
2104 [RTN_LOCAL
] = "LOCAL",
2105 [RTN_BROADCAST
] = "BROADCAST",
2106 [RTN_ANYCAST
] = "ANYCAST",
2107 [RTN_MULTICAST
] = "MULTICAST",
2108 [RTN_BLACKHOLE
] = "BLACKHOLE",
2109 [RTN_UNREACHABLE
] = "UNREACHABLE",
2110 [RTN_PROHIBIT
] = "PROHIBIT",
2111 [RTN_THROW
] = "THROW",
2113 [RTN_XRESOLVE
] = "XRESOLVE",
2116 static inline const char *rtn_type(char *buf
, size_t len
, unsigned int t
)
2118 if (t
< __RTN_MAX
&& rtn_type_names
[t
])
2119 return rtn_type_names
[t
];
2120 snprintf(buf
, len
, "type %u", t
);
2124 /* Pretty print the trie */
2125 static int fib_trie_seq_show(struct seq_file
*seq
, void *v
)
2127 const struct fib_trie_iter
*iter
= seq
->private;
2128 struct tnode
*n
= v
;
2130 if (!node_parent_rcu(n
))
2131 fib_table_print(seq
, iter
->tb
);
2134 __be32 prf
= htonl(n
->key
);
2136 seq_indent(seq
, iter
->depth
-1);
2137 seq_printf(seq
, " +-- %pI4/%zu %u %u %u\n",
2138 &prf
, KEYLENGTH
- n
->pos
- n
->bits
, n
->bits
,
2139 n
->full_children
, n
->empty_children
);
2141 struct leaf_info
*li
;
2142 __be32 val
= htonl(n
->key
);
2144 seq_indent(seq
, iter
->depth
);
2145 seq_printf(seq
, " |-- %pI4\n", &val
);
2147 hlist_for_each_entry_rcu(li
, &n
->list
, hlist
) {
2148 struct fib_alias
*fa
;
2150 list_for_each_entry_rcu(fa
, &li
->falh
, fa_list
) {
2151 char buf1
[32], buf2
[32];
2153 seq_indent(seq
, iter
->depth
+1);
2154 seq_printf(seq
, " /%d %s %s", li
->plen
,
2155 rtn_scope(buf1
, sizeof(buf1
),
2156 fa
->fa_info
->fib_scope
),
2157 rtn_type(buf2
, sizeof(buf2
),
2160 seq_printf(seq
, " tos=%d", fa
->fa_tos
);
2161 seq_putc(seq
, '\n');
2169 static const struct seq_operations fib_trie_seq_ops
= {
2170 .start
= fib_trie_seq_start
,
2171 .next
= fib_trie_seq_next
,
2172 .stop
= fib_trie_seq_stop
,
2173 .show
= fib_trie_seq_show
,
2176 static int fib_trie_seq_open(struct inode
*inode
, struct file
*file
)
2178 return seq_open_net(inode
, file
, &fib_trie_seq_ops
,
2179 sizeof(struct fib_trie_iter
));
2182 static const struct file_operations fib_trie_fops
= {
2183 .owner
= THIS_MODULE
,
2184 .open
= fib_trie_seq_open
,
2186 .llseek
= seq_lseek
,
2187 .release
= seq_release_net
,
2190 struct fib_route_iter
{
2191 struct seq_net_private p
;
2192 struct trie
*main_trie
;
2197 static struct tnode
*fib_route_get_idx(struct fib_route_iter
*iter
, loff_t pos
)
2199 struct tnode
*l
= NULL
;
2200 struct trie
*t
= iter
->main_trie
;
2202 /* use cache location of last found key */
2203 if (iter
->pos
> 0 && pos
>= iter
->pos
&& (l
= fib_find_node(t
, iter
->key
)))
2207 l
= trie_firstleaf(t
);
2210 while (l
&& pos
-- > 0) {
2212 l
= trie_nextleaf(l
);
2216 iter
->key
= pos
; /* remember it */
2218 iter
->pos
= 0; /* forget it */
2223 static void *fib_route_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2226 struct fib_route_iter
*iter
= seq
->private;
2227 struct fib_table
*tb
;
2230 tb
= fib_get_table(seq_file_net(seq
), RT_TABLE_MAIN
);
2234 iter
->main_trie
= (struct trie
*) tb
->tb_data
;
2236 return SEQ_START_TOKEN
;
2238 return fib_route_get_idx(iter
, *pos
- 1);
2241 static void *fib_route_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2243 struct fib_route_iter
*iter
= seq
->private;
2244 struct tnode
*l
= v
;
2247 if (v
== SEQ_START_TOKEN
) {
2249 l
= trie_firstleaf(iter
->main_trie
);
2252 l
= trie_nextleaf(l
);
2262 static void fib_route_seq_stop(struct seq_file
*seq
, void *v
)
2268 static unsigned int fib_flag_trans(int type
, __be32 mask
, const struct fib_info
*fi
)
2270 unsigned int flags
= 0;
2272 if (type
== RTN_UNREACHABLE
|| type
== RTN_PROHIBIT
)
2274 if (fi
&& fi
->fib_nh
->nh_gw
)
2275 flags
|= RTF_GATEWAY
;
2276 if (mask
== htonl(0xFFFFFFFF))
2283 * This outputs /proc/net/route.
2284 * The format of the file is not supposed to be changed
2285 * and needs to be same as fib_hash output to avoid breaking
2288 static int fib_route_seq_show(struct seq_file
*seq
, void *v
)
2290 struct tnode
*l
= v
;
2291 struct leaf_info
*li
;
2293 if (v
== SEQ_START_TOKEN
) {
2294 seq_printf(seq
, "%-127s\n", "Iface\tDestination\tGateway "
2295 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2300 hlist_for_each_entry_rcu(li
, &l
->list
, hlist
) {
2301 struct fib_alias
*fa
;
2302 __be32 mask
, prefix
;
2304 mask
= inet_make_mask(li
->plen
);
2305 prefix
= htonl(l
->key
);
2307 list_for_each_entry_rcu(fa
, &li
->falh
, fa_list
) {
2308 const struct fib_info
*fi
= fa
->fa_info
;
2309 unsigned int flags
= fib_flag_trans(fa
->fa_type
, mask
, fi
);
2311 if (fa
->fa_type
== RTN_BROADCAST
2312 || fa
->fa_type
== RTN_MULTICAST
)
2315 seq_setwidth(seq
, 127);
2319 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2320 "%d\t%08X\t%d\t%u\t%u",
2321 fi
->fib_dev
? fi
->fib_dev
->name
: "*",
2323 fi
->fib_nh
->nh_gw
, flags
, 0, 0,
2327 fi
->fib_advmss
+ 40 : 0),
2332 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2333 "%d\t%08X\t%d\t%u\t%u",
2334 prefix
, 0, flags
, 0, 0, 0,
2344 static const struct seq_operations fib_route_seq_ops
= {
2345 .start
= fib_route_seq_start
,
2346 .next
= fib_route_seq_next
,
2347 .stop
= fib_route_seq_stop
,
2348 .show
= fib_route_seq_show
,
2351 static int fib_route_seq_open(struct inode
*inode
, struct file
*file
)
2353 return seq_open_net(inode
, file
, &fib_route_seq_ops
,
2354 sizeof(struct fib_route_iter
));
2357 static const struct file_operations fib_route_fops
= {
2358 .owner
= THIS_MODULE
,
2359 .open
= fib_route_seq_open
,
2361 .llseek
= seq_lseek
,
2362 .release
= seq_release_net
,
2365 int __net_init
fib_proc_init(struct net
*net
)
2367 if (!proc_create("fib_trie", S_IRUGO
, net
->proc_net
, &fib_trie_fops
))
2370 if (!proc_create("fib_triestat", S_IRUGO
, net
->proc_net
,
2371 &fib_triestat_fops
))
2374 if (!proc_create("route", S_IRUGO
, net
->proc_net
, &fib_route_fops
))
2380 remove_proc_entry("fib_triestat", net
->proc_net
);
2382 remove_proc_entry("fib_trie", net
->proc_net
);
2387 void __net_exit
fib_proc_exit(struct net
*net
)
2389 remove_proc_entry("fib_trie", net
->proc_net
);
2390 remove_proc_entry("fib_triestat", net
->proc_net
);
2391 remove_proc_entry("route", net
->proc_net
);
2394 #endif /* CONFIG_PROC_FS */