2 * This program is free software; you can redistribute it and/or
3 * modify it under the terms of the GNU General Public License
4 * as published by the Free Software Foundation; either version
5 * 2 of the License, or (at your option) any later version.
7 * Robert Olsson <robert.olsson@its.uu.se> Uppsala Universitet
8 * & Swedish University of Agricultural Sciences.
10 * Jens Laas <jens.laas@data.slu.se> Swedish University of
11 * Agricultural Sciences.
13 * Hans Liss <hans.liss@its.uu.se> Uppsala Universitet
15 * This work is based on the LPC-trie which is originally descibed in:
17 * An experimental study of compression methods for dynamic tries
18 * Stefan Nilsson and Matti Tikkanen. Algorithmica, 33(1):19-33, 2002.
19 * http://www.nada.kth.se/~snilsson/public/papers/dyntrie2/
22 * IP-address lookup using LC-tries. Stefan Nilsson and Gunnar Karlsson
23 * IEEE Journal on Selected Areas in Communications, 17(6):1083-1092, June 1999
25 * Version: $Id: fib_trie.c,v 1.3 2005/06/08 14:20:01 robert Exp $
28 * Code from fib_hash has been reused which includes the following header:
31 * INET An implementation of the TCP/IP protocol suite for the LINUX
32 * operating system. INET is implemented using the BSD Socket
33 * interface as the means of communication with the user level.
35 * IPv4 FIB: lookup engine and maintenance routines.
38 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
40 * This program is free software; you can redistribute it and/or
41 * modify it under the terms of the GNU General Public License
42 * as published by the Free Software Foundation; either version
43 * 2 of the License, or (at your option) any later version.
45 * Substantial contributions to this work comes from:
47 * David S. Miller, <davem@davemloft.net>
48 * Stephen Hemminger <shemminger@osdl.org>
49 * Paul E. McKenney <paulmck@us.ibm.com>
50 * Patrick McHardy <kaber@trash.net>
53 #define VERSION "0.408"
55 #include <asm/uaccess.h>
56 #include <asm/system.h>
57 #include <linux/bitops.h>
58 #include <linux/types.h>
59 #include <linux/kernel.h>
61 #include <linux/string.h>
62 #include <linux/socket.h>
63 #include <linux/sockios.h>
64 #include <linux/errno.h>
66 #include <linux/inet.h>
67 #include <linux/inetdevice.h>
68 #include <linux/netdevice.h>
69 #include <linux/if_arp.h>
70 #include <linux/proc_fs.h>
71 #include <linux/rcupdate.h>
72 #include <linux/skbuff.h>
73 #include <linux/netlink.h>
74 #include <linux/init.h>
75 #include <linux/list.h>
76 #include <net/net_namespace.h>
78 #include <net/protocol.h>
79 #include <net/route.h>
82 #include <net/ip_fib.h>
83 #include "fib_lookup.h"
85 #undef CONFIG_IP_FIB_TRIE_STATS
86 #define MAX_STAT_DEPTH 32
88 #define KEYLENGTH (8*sizeof(t_key))
90 typedef unsigned int t_key;
94 #define NODE_TYPE_MASK 0x1UL
95 #define NODE_TYPE(node) ((node)->parent & NODE_TYPE_MASK)
97 #define IS_TNODE(n) (!(n->parent & T_LEAF))
98 #define IS_LEAF(n) (n->parent & T_LEAF)
102 unsigned long parent;
107 unsigned long parent;
108 struct hlist_head list;
113 struct hlist_node hlist;
116 struct list_head falh;
121 unsigned long parent;
122 unsigned short pos:5; /* 2log(KEYLENGTH) bits needed */
123 unsigned short bits:5; /* 2log(KEYLENGTH) bits needed */
124 unsigned short full_children; /* KEYLENGTH bits needed */
125 unsigned short empty_children; /* KEYLENGTH bits needed */
127 struct node *child[0];
130 #ifdef CONFIG_IP_FIB_TRIE_STATS
131 struct trie_use_stats {
133 unsigned int backtrack;
134 unsigned int semantic_match_passed;
135 unsigned int semantic_match_miss;
136 unsigned int null_node_hit;
137 unsigned int resize_node_skipped;
142 unsigned int totdepth;
143 unsigned int maxdepth;
146 unsigned int nullpointers;
147 unsigned int nodesizes[MAX_STAT_DEPTH];
152 #ifdef CONFIG_IP_FIB_TRIE_STATS
153 struct trie_use_stats stats;
156 unsigned int revision;
159 static void put_child(struct trie *t, struct tnode *tn, int i, struct node *n);
160 static void tnode_put_child_reorg(struct tnode *tn, int i, struct node *n, int wasfull);
161 static struct node *resize(struct trie *t, struct tnode *tn);
162 static struct tnode *inflate(struct trie *t, struct tnode *tn);
163 static struct tnode *halve(struct trie *t, struct tnode *tn);
164 static void tnode_free(struct tnode *tn);
166 static struct kmem_cache *fn_alias_kmem __read_mostly;
168 static inline struct tnode *node_parent(struct node *node)
172 ret = (struct tnode *)(node->parent & ~NODE_TYPE_MASK);
173 return rcu_dereference(ret);
176 static inline void node_set_parent(struct node *node, struct tnode *ptr)
178 rcu_assign_pointer(node->parent,
179 (unsigned long)ptr | NODE_TYPE(node));
182 /* rcu_read_lock needs to be hold by caller from readside */
184 static inline struct node *tnode_get_child(struct tnode *tn, int i)
186 BUG_ON(i >= 1 << tn->bits);
188 return rcu_dereference(tn->child[i]);
191 static inline int tnode_child_length(const struct tnode *tn)
193 return 1 << tn->bits;
196 static inline t_key mask_pfx(t_key k, unsigned short l)
198 return (l == 0) ? 0 : k >> (KEYLENGTH-l) << (KEYLENGTH-l);
201 static inline t_key tkey_extract_bits(t_key a, int offset, int bits)
203 if (offset < KEYLENGTH)
204 return ((t_key)(a << offset)) >> (KEYLENGTH - bits);
209 static inline int tkey_equals(t_key a, t_key b)
214 static inline int tkey_sub_equals(t_key a, int offset, int bits, t_key b)
216 if (bits == 0 || offset >= KEYLENGTH)
218 bits = bits > KEYLENGTH ? KEYLENGTH : bits;
219 return ((a ^ b) << offset) >> (KEYLENGTH - bits) == 0;
222 static inline int tkey_mismatch(t_key a, int offset, t_key b)
229 while ((diff << i) >> (KEYLENGTH-1) == 0)
235 To understand this stuff, an understanding of keys and all their bits is
236 necessary. Every node in the trie has a key associated with it, but not
237 all of the bits in that key are significant.
239 Consider a node 'n' and its parent 'tp'.
241 If n is a leaf, every bit in its key is significant. Its presence is
242 necessitated by path compression, since during a tree traversal (when
243 searching for a leaf - unless we are doing an insertion) we will completely
244 ignore all skipped bits we encounter. Thus we need to verify, at the end of
245 a potentially successful search, that we have indeed been walking the
248 Note that we can never "miss" the correct key in the tree if present by
249 following the wrong path. Path compression ensures that segments of the key
250 that are the same for all keys with a given prefix are skipped, but the
251 skipped part *is* identical for each node in the subtrie below the skipped
252 bit! trie_insert() in this implementation takes care of that - note the
253 call to tkey_sub_equals() in trie_insert().
255 if n is an internal node - a 'tnode' here, the various parts of its key
256 have many different meanings.
259 _________________________________________________________________
260 | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
261 -----------------------------------------------------------------
262 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
264 _________________________________________________________________
265 | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
266 -----------------------------------------------------------------
267 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
274 First, let's just ignore the bits that come before the parent tp, that is
275 the bits from 0 to (tp->pos-1). They are *known* but at this point we do
276 not use them for anything.
278 The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
279 index into the parent's child array. That is, they will be used to find
280 'n' among tp's children.
282 The bits from (tp->pos + tp->bits) to (n->pos - 1) - "S" - are skipped bits
285 All the bits we have seen so far are significant to the node n. The rest
286 of the bits are really not needed or indeed known in n->key.
288 The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
289 n's child array, and will of course be different for each child.
292 The rest of the bits, from (n->pos + n->bits) onward, are completely unknown
297 static inline void check_tnode(const struct tnode *tn)
299 WARN_ON(tn && tn->pos+tn->bits > 32);
302 static const int halve_threshold = 25;
303 static const int inflate_threshold = 50;
304 static const int halve_threshold_root = 8;
305 static const int inflate_threshold_root = 15;
308 static void __alias_free_mem(struct rcu_head *head)
310 struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
311 kmem_cache_free(fn_alias_kmem, fa);
314 static inline void alias_free_mem_rcu(struct fib_alias *fa)
316 call_rcu(&fa->rcu, __alias_free_mem);
319 static void __leaf_free_rcu(struct rcu_head *head)
321 kfree(container_of(head, struct leaf, rcu));
324 static void __leaf_info_free_rcu(struct rcu_head *head)
326 kfree(container_of(head, struct leaf_info, rcu));
329 static inline void free_leaf_info(struct leaf_info *leaf)
331 call_rcu(&leaf->rcu, __leaf_info_free_rcu);
334 static struct tnode *tnode_alloc(unsigned int size)
338 if (size <= PAGE_SIZE)
339 return kcalloc(size, 1, GFP_KERNEL);
341 pages = alloc_pages(GFP_KERNEL|__GFP_ZERO, get_order(size));
345 return page_address(pages);
348 static void __tnode_free_rcu(struct rcu_head *head)
350 struct tnode *tn = container_of(head, struct tnode, rcu);
351 unsigned int size = sizeof(struct tnode) +
352 (1 << tn->bits) * sizeof(struct node *);
354 if (size <= PAGE_SIZE)
357 free_pages((unsigned long)tn, get_order(size));
360 static inline void tnode_free(struct tnode *tn)
363 struct leaf *l = (struct leaf *) tn;
364 call_rcu_bh(&l->rcu, __leaf_free_rcu);
366 call_rcu(&tn->rcu, __tnode_free_rcu);
369 static struct leaf *leaf_new(void)
371 struct leaf *l = kmalloc(sizeof(struct leaf), GFP_KERNEL);
374 INIT_HLIST_HEAD(&l->list);
379 static struct leaf_info *leaf_info_new(int plen)
381 struct leaf_info *li = kmalloc(sizeof(struct leaf_info), GFP_KERNEL);
384 INIT_LIST_HEAD(&li->falh);
389 static struct tnode* tnode_new(t_key key, int pos, int bits)
391 int nchildren = 1<<bits;
392 int sz = sizeof(struct tnode) + nchildren * sizeof(struct node *);
393 struct tnode *tn = tnode_alloc(sz);
397 tn->parent = T_TNODE;
401 tn->full_children = 0;
402 tn->empty_children = 1<<bits;
405 pr_debug("AT %p s=%u %u\n", tn, (unsigned int) sizeof(struct tnode),
406 (unsigned int) (sizeof(struct node) * 1<<bits));
411 * Check whether a tnode 'n' is "full", i.e. it is an internal node
412 * and no bits are skipped. See discussion in dyntree paper p. 6
415 static inline int tnode_full(const struct tnode *tn, const struct node *n)
417 if (n == NULL || IS_LEAF(n))
420 return ((struct tnode *) n)->pos == tn->pos + tn->bits;
423 static inline void put_child(struct trie *t, struct tnode *tn, int i, struct node *n)
425 tnode_put_child_reorg(tn, i, n, -1);
429 * Add a child at position i overwriting the old value.
430 * Update the value of full_children and empty_children.
433 static void tnode_put_child_reorg(struct tnode *tn, int i, struct node *n, int wasfull)
435 struct node *chi = tn->child[i];
438 BUG_ON(i >= 1<<tn->bits);
441 /* update emptyChildren */
442 if (n == NULL && chi != NULL)
443 tn->empty_children++;
444 else if (n != NULL && chi == NULL)
445 tn->empty_children--;
447 /* update fullChildren */
449 wasfull = tnode_full(tn, chi);
451 isfull = tnode_full(tn, n);
452 if (wasfull && !isfull)
454 else if (!wasfull && isfull)
458 node_set_parent(n, tn);
460 rcu_assign_pointer(tn->child[i], n);
463 static struct node *resize(struct trie *t, struct tnode *tn)
467 struct tnode *old_tn;
468 int inflate_threshold_use;
469 int halve_threshold_use;
475 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
476 tn, inflate_threshold, halve_threshold);
479 if (tn->empty_children == tnode_child_length(tn)) {
484 if (tn->empty_children == tnode_child_length(tn) - 1)
485 for (i = 0; i < tnode_child_length(tn); i++) {
492 /* compress one level */
493 node_set_parent(n, NULL);
498 * Double as long as the resulting node has a number of
499 * nonempty nodes that are above the threshold.
503 * From "Implementing a dynamic compressed trie" by Stefan Nilsson of
504 * the Helsinki University of Technology and Matti Tikkanen of Nokia
505 * Telecommunications, page 6:
506 * "A node is doubled if the ratio of non-empty children to all
507 * children in the *doubled* node is at least 'high'."
509 * 'high' in this instance is the variable 'inflate_threshold'. It
510 * is expressed as a percentage, so we multiply it with
511 * tnode_child_length() and instead of multiplying by 2 (since the
512 * child array will be doubled by inflate()) and multiplying
513 * the left-hand side by 100 (to handle the percentage thing) we
514 * multiply the left-hand side by 50.
516 * The left-hand side may look a bit weird: tnode_child_length(tn)
517 * - tn->empty_children is of course the number of non-null children
518 * in the current node. tn->full_children is the number of "full"
519 * children, that is non-null tnodes with a skip value of 0.
520 * All of those will be doubled in the resulting inflated tnode, so
521 * we just count them one extra time here.
523 * A clearer way to write this would be:
525 * to_be_doubled = tn->full_children;
526 * not_to_be_doubled = tnode_child_length(tn) - tn->empty_children -
529 * new_child_length = tnode_child_length(tn) * 2;
531 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
533 * if (new_fill_factor >= inflate_threshold)
535 * ...and so on, tho it would mess up the while () loop.
538 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
542 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
543 * inflate_threshold * new_child_length
545 * expand not_to_be_doubled and to_be_doubled, and shorten:
546 * 100 * (tnode_child_length(tn) - tn->empty_children +
547 * tn->full_children) >= inflate_threshold * new_child_length
549 * expand new_child_length:
550 * 100 * (tnode_child_length(tn) - tn->empty_children +
551 * tn->full_children) >=
552 * inflate_threshold * tnode_child_length(tn) * 2
555 * 50 * (tn->full_children + tnode_child_length(tn) -
556 * tn->empty_children) >= inflate_threshold *
557 * tnode_child_length(tn)
563 /* Keep root node larger */
566 inflate_threshold_use = inflate_threshold_root;
568 inflate_threshold_use = inflate_threshold;
572 while ((tn->full_children > 0 && max_resize-- &&
573 50 * (tn->full_children + tnode_child_length(tn) - tn->empty_children) >=
574 inflate_threshold_use * tnode_child_length(tn))) {
580 #ifdef CONFIG_IP_FIB_TRIE_STATS
581 t->stats.resize_node_skipped++;
587 if (max_resize < 0) {
589 printk(KERN_WARNING "Fix inflate_threshold_root. Now=%d size=%d bits\n",
590 inflate_threshold_root, tn->bits);
592 printk(KERN_WARNING "Fix inflate_threshold. Now=%d size=%d bits\n",
593 inflate_threshold, tn->bits);
599 * Halve as long as the number of empty children in this
600 * node is above threshold.
604 /* Keep root node larger */
607 halve_threshold_use = halve_threshold_root;
609 halve_threshold_use = halve_threshold;
613 while (tn->bits > 1 && max_resize-- &&
614 100 * (tnode_child_length(tn) - tn->empty_children) <
615 halve_threshold_use * tnode_child_length(tn)) {
621 #ifdef CONFIG_IP_FIB_TRIE_STATS
622 t->stats.resize_node_skipped++;
628 if (max_resize < 0) {
630 printk(KERN_WARNING "Fix halve_threshold_root. Now=%d size=%d bits\n",
631 halve_threshold_root, tn->bits);
633 printk(KERN_WARNING "Fix halve_threshold. Now=%d size=%d bits\n",
634 halve_threshold, tn->bits);
637 /* Only one child remains */
638 if (tn->empty_children == tnode_child_length(tn) - 1)
639 for (i = 0; i < tnode_child_length(tn); i++) {
646 /* compress one level */
648 node_set_parent(n, NULL);
653 return (struct node *) tn;
656 static struct tnode *inflate(struct trie *t, struct tnode *tn)
659 struct tnode *oldtnode = tn;
660 int olen = tnode_child_length(tn);
663 pr_debug("In inflate\n");
665 tn = tnode_new(oldtnode->key, oldtnode->pos, oldtnode->bits + 1);
668 return ERR_PTR(-ENOMEM);
671 * Preallocate and store tnodes before the actual work so we
672 * don't get into an inconsistent state if memory allocation
673 * fails. In case of failure we return the oldnode and inflate
674 * of tnode is ignored.
677 for (i = 0; i < olen; i++) {
678 struct tnode *inode = (struct tnode *) tnode_get_child(oldtnode, i);
682 inode->pos == oldtnode->pos + oldtnode->bits &&
684 struct tnode *left, *right;
685 t_key m = ~0U << (KEYLENGTH - 1) >> inode->pos;
687 left = tnode_new(inode->key&(~m), inode->pos + 1,
692 right = tnode_new(inode->key|m, inode->pos + 1,
700 put_child(t, tn, 2*i, (struct node *) left);
701 put_child(t, tn, 2*i+1, (struct node *) right);
705 for (i = 0; i < olen; i++) {
706 struct node *node = tnode_get_child(oldtnode, i);
707 struct tnode *left, *right;
714 /* A leaf or an internal node with skipped bits */
716 if (IS_LEAF(node) || ((struct tnode *) node)->pos >
717 tn->pos + tn->bits - 1) {
718 if (tkey_extract_bits(node->key, oldtnode->pos + oldtnode->bits,
720 put_child(t, tn, 2*i, node);
722 put_child(t, tn, 2*i+1, node);
726 /* An internal node with two children */
727 inode = (struct tnode *) node;
729 if (inode->bits == 1) {
730 put_child(t, tn, 2*i, inode->child[0]);
731 put_child(t, tn, 2*i+1, inode->child[1]);
737 /* An internal node with more than two children */
739 /* We will replace this node 'inode' with two new
740 * ones, 'left' and 'right', each with half of the
741 * original children. The two new nodes will have
742 * a position one bit further down the key and this
743 * means that the "significant" part of their keys
744 * (see the discussion near the top of this file)
745 * will differ by one bit, which will be "0" in
746 * left's key and "1" in right's key. Since we are
747 * moving the key position by one step, the bit that
748 * we are moving away from - the bit at position
749 * (inode->pos) - is the one that will differ between
750 * left and right. So... we synthesize that bit in the
752 * The mask 'm' below will be a single "one" bit at
753 * the position (inode->pos)
756 /* Use the old key, but set the new significant
760 left = (struct tnode *) tnode_get_child(tn, 2*i);
761 put_child(t, tn, 2*i, NULL);
765 right = (struct tnode *) tnode_get_child(tn, 2*i+1);
766 put_child(t, tn, 2*i+1, NULL);
770 size = tnode_child_length(left);
771 for (j = 0; j < size; j++) {
772 put_child(t, left, j, inode->child[j]);
773 put_child(t, right, j, inode->child[j + size]);
775 put_child(t, tn, 2*i, resize(t, left));
776 put_child(t, tn, 2*i+1, resize(t, right));
780 tnode_free(oldtnode);
784 int size = tnode_child_length(tn);
787 for (j = 0; j < size; j++)
789 tnode_free((struct tnode *)tn->child[j]);
793 return ERR_PTR(-ENOMEM);
797 static struct tnode *halve(struct trie *t, struct tnode *tn)
799 struct tnode *oldtnode = tn;
800 struct node *left, *right;
802 int olen = tnode_child_length(tn);
804 pr_debug("In halve\n");
806 tn = tnode_new(oldtnode->key, oldtnode->pos, oldtnode->bits - 1);
809 return ERR_PTR(-ENOMEM);
812 * Preallocate and store tnodes before the actual work so we
813 * don't get into an inconsistent state if memory allocation
814 * fails. In case of failure we return the oldnode and halve
815 * of tnode is ignored.
818 for (i = 0; i < olen; i += 2) {
819 left = tnode_get_child(oldtnode, i);
820 right = tnode_get_child(oldtnode, i+1);
822 /* Two nonempty children */
826 newn = tnode_new(left->key, tn->pos + tn->bits, 1);
831 put_child(t, tn, i/2, (struct node *)newn);
836 for (i = 0; i < olen; i += 2) {
837 struct tnode *newBinNode;
839 left = tnode_get_child(oldtnode, i);
840 right = tnode_get_child(oldtnode, i+1);
842 /* At least one of the children is empty */
844 if (right == NULL) /* Both are empty */
846 put_child(t, tn, i/2, right);
851 put_child(t, tn, i/2, left);
855 /* Two nonempty children */
856 newBinNode = (struct tnode *) tnode_get_child(tn, i/2);
857 put_child(t, tn, i/2, NULL);
858 put_child(t, newBinNode, 0, left);
859 put_child(t, newBinNode, 1, right);
860 put_child(t, tn, i/2, resize(t, newBinNode));
862 tnode_free(oldtnode);
866 int size = tnode_child_length(tn);
869 for (j = 0; j < size; j++)
871 tnode_free((struct tnode *)tn->child[j]);
875 return ERR_PTR(-ENOMEM);
879 /* readside must use rcu_read_lock currently dump routines
880 via get_fa_head and dump */
882 static struct leaf_info *find_leaf_info(struct leaf *l, int plen)
884 struct hlist_head *head = &l->list;
885 struct hlist_node *node;
886 struct leaf_info *li;
888 hlist_for_each_entry_rcu(li, node, head, hlist)
889 if (li->plen == plen)
895 static inline struct list_head * get_fa_head(struct leaf *l, int plen)
897 struct leaf_info *li = find_leaf_info(l, plen);
905 static void insert_leaf_info(struct hlist_head *head, struct leaf_info *new)
907 struct leaf_info *li = NULL, *last = NULL;
908 struct hlist_node *node;
910 if (hlist_empty(head)) {
911 hlist_add_head_rcu(&new->hlist, head);
913 hlist_for_each_entry(li, node, head, hlist) {
914 if (new->plen > li->plen)
920 hlist_add_after_rcu(&last->hlist, &new->hlist);
922 hlist_add_before_rcu(&new->hlist, &li->hlist);
926 /* rcu_read_lock needs to be hold by caller from readside */
929 fib_find_node(struct trie *t, u32 key)
936 n = rcu_dereference(t->trie);
938 while (n != NULL && NODE_TYPE(n) == T_TNODE) {
939 tn = (struct tnode *) n;
943 if (tkey_sub_equals(tn->key, pos, tn->pos-pos, key)) {
944 pos = tn->pos + tn->bits;
945 n = tnode_get_child(tn, tkey_extract_bits(key, tn->pos, tn->bits));
949 /* Case we have found a leaf. Compare prefixes */
951 if (n != NULL && IS_LEAF(n) && tkey_equals(key, n->key))
952 return (struct leaf *)n;
957 static struct node *trie_rebalance(struct trie *t, struct tnode *tn)
960 t_key cindex, key = tn->key;
963 while (tn != NULL && (tp = node_parent((struct node *)tn)) != NULL) {
964 cindex = tkey_extract_bits(key, tp->pos, tp->bits);
965 wasfull = tnode_full(tp, tnode_get_child(tp, cindex));
966 tn = (struct tnode *) resize (t, (struct tnode *)tn);
967 tnode_put_child_reorg((struct tnode *)tp, cindex,(struct node*)tn, wasfull);
969 tp = node_parent((struct node *) tn);
975 /* Handle last (top) tnode */
977 tn = (struct tnode*) resize(t, (struct tnode *)tn);
979 return (struct node*) tn;
982 /* only used from updater-side */
984 static struct list_head *
985 fib_insert_node(struct trie *t, int *err, u32 key, int plen)
988 struct tnode *tp = NULL, *tn = NULL;
992 struct list_head *fa_head = NULL;
993 struct leaf_info *li;
999 /* If we point to NULL, stop. Either the tree is empty and we should
1000 * just put a new leaf in if, or we have reached an empty child slot,
1001 * and we should just put our new leaf in that.
1002 * If we point to a T_TNODE, check if it matches our key. Note that
1003 * a T_TNODE might be skipping any number of bits - its 'pos' need
1004 * not be the parent's 'pos'+'bits'!
1006 * If it does match the current key, get pos/bits from it, extract
1007 * the index from our key, push the T_TNODE and walk the tree.
1009 * If it doesn't, we have to replace it with a new T_TNODE.
1011 * If we point to a T_LEAF, it might or might not have the same key
1012 * as we do. If it does, just change the value, update the T_LEAF's
1013 * value, and return it.
1014 * If it doesn't, we need to replace it with a T_TNODE.
1017 while (n != NULL && NODE_TYPE(n) == T_TNODE) {
1018 tn = (struct tnode *) n;
1022 if (tkey_sub_equals(tn->key, pos, tn->pos-pos, key)) {
1024 pos = tn->pos + tn->bits;
1025 n = tnode_get_child(tn, tkey_extract_bits(key, tn->pos, tn->bits));
1027 BUG_ON(n && node_parent(n) != tn);
1033 * n ----> NULL, LEAF or TNODE
1035 * tp is n's (parent) ----> NULL or TNODE
1038 BUG_ON(tp && IS_LEAF(tp));
1040 /* Case 1: n is a leaf. Compare prefixes */
1042 if (n != NULL && IS_LEAF(n) && tkey_equals(key, n->key)) {
1043 struct leaf *l = (struct leaf *) n;
1045 li = leaf_info_new(plen);
1052 fa_head = &li->falh;
1053 insert_leaf_info(&l->list, li);
1065 li = leaf_info_new(plen);
1068 tnode_free((struct tnode *) l);
1073 fa_head = &li->falh;
1074 insert_leaf_info(&l->list, li);
1076 if (t->trie && n == NULL) {
1077 /* Case 2: n is NULL, and will just insert a new leaf */
1079 node_set_parent((struct node *)l, tp);
1081 cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1082 put_child(t, (struct tnode *)tp, cindex, (struct node *)l);
1084 /* Case 3: n is a LEAF or a TNODE and the key doesn't match. */
1086 * Add a new tnode here
1087 * first tnode need some special handling
1091 pos = tp->pos+tp->bits;
1096 newpos = tkey_mismatch(key, pos, n->key);
1097 tn = tnode_new(n->key, newpos, 1);
1100 tn = tnode_new(key, newpos, 1); /* First tnode */
1105 tnode_free((struct tnode *) l);
1110 node_set_parent((struct node *)tn, tp);
1112 missbit = tkey_extract_bits(key, newpos, 1);
1113 put_child(t, tn, missbit, (struct node *)l);
1114 put_child(t, tn, 1-missbit, n);
1117 cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1118 put_child(t, (struct tnode *)tp, cindex, (struct node *)tn);
1120 rcu_assign_pointer(t->trie, (struct node *)tn); /* First tnode */
1125 if (tp && tp->pos + tp->bits > 32)
1126 printk(KERN_WARNING "fib_trie tp=%p pos=%d, bits=%d, key=%0x plen=%d\n",
1127 tp, tp->pos, tp->bits, key, plen);
1129 /* Rebalance the trie */
1131 rcu_assign_pointer(t->trie, trie_rebalance(t, tp));
1139 * Caller must hold RTNL.
1141 static int fn_trie_insert(struct fib_table *tb, struct fib_config *cfg)
1143 struct trie *t = (struct trie *) tb->tb_data;
1144 struct fib_alias *fa, *new_fa;
1145 struct list_head *fa_head = NULL;
1146 struct fib_info *fi;
1147 int plen = cfg->fc_dst_len;
1148 u8 tos = cfg->fc_tos;
1156 key = ntohl(cfg->fc_dst);
1158 pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1160 mask = ntohl(inet_make_mask(plen));
1167 fi = fib_create_info(cfg);
1173 l = fib_find_node(t, key);
1177 fa_head = get_fa_head(l, plen);
1178 fa = fib_find_alias(fa_head, tos, fi->fib_priority);
1181 /* Now fa, if non-NULL, points to the first fib alias
1182 * with the same keys [prefix,tos,priority], if such key already
1183 * exists or to the node before which we will insert new one.
1185 * If fa is NULL, we will need to allocate a new one and
1186 * insert to the head of f.
1188 * If f is NULL, no fib node matched the destination key
1189 * and we need to allocate a new one of those as well.
1192 if (fa && fa->fa_info->fib_priority == fi->fib_priority) {
1193 struct fib_alias *fa_orig;
1196 if (cfg->fc_nlflags & NLM_F_EXCL)
1199 if (cfg->fc_nlflags & NLM_F_REPLACE) {
1200 struct fib_info *fi_drop;
1203 if (fi->fib_treeref > 1)
1207 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1211 fi_drop = fa->fa_info;
1212 new_fa->fa_tos = fa->fa_tos;
1213 new_fa->fa_info = fi;
1214 new_fa->fa_type = cfg->fc_type;
1215 new_fa->fa_scope = cfg->fc_scope;
1216 state = fa->fa_state;
1217 new_fa->fa_state &= ~FA_S_ACCESSED;
1219 list_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1220 alias_free_mem_rcu(fa);
1222 fib_release_info(fi_drop);
1223 if (state & FA_S_ACCESSED)
1225 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1226 tb->tb_id, &cfg->fc_nlinfo, NLM_F_REPLACE);
1230 /* Error if we find a perfect match which
1231 * uses the same scope, type, and nexthop
1235 list_for_each_entry(fa, fa_orig->fa_list.prev, fa_list) {
1236 if (fa->fa_tos != tos)
1238 if (fa->fa_info->fib_priority != fi->fib_priority)
1240 if (fa->fa_type == cfg->fc_type &&
1241 fa->fa_scope == cfg->fc_scope &&
1242 fa->fa_info == fi) {
1246 if (!(cfg->fc_nlflags & NLM_F_APPEND))
1250 if (!(cfg->fc_nlflags & NLM_F_CREATE))
1254 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1258 new_fa->fa_info = fi;
1259 new_fa->fa_tos = tos;
1260 new_fa->fa_type = cfg->fc_type;
1261 new_fa->fa_scope = cfg->fc_scope;
1262 new_fa->fa_state = 0;
1264 * Insert new entry to the list.
1269 fa_head = fib_insert_node(t, &err, key, plen);
1271 goto out_free_new_fa;
1274 list_add_tail_rcu(&new_fa->fa_list,
1275 (fa ? &fa->fa_list : fa_head));
1278 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, tb->tb_id,
1279 &cfg->fc_nlinfo, 0);
1284 kmem_cache_free(fn_alias_kmem, new_fa);
1286 fib_release_info(fi);
1292 /* should be called with rcu_read_lock */
1293 static inline int check_leaf(struct trie *t, struct leaf *l,
1294 t_key key, int *plen, const struct flowi *flp,
1295 struct fib_result *res)
1299 struct leaf_info *li;
1300 struct hlist_head *hhead = &l->list;
1301 struct hlist_node *node;
1303 hlist_for_each_entry_rcu(li, node, hhead, hlist) {
1305 mask = inet_make_mask(i);
1306 if (l->key != (key & ntohl(mask)))
1309 if ((err = fib_semantic_match(&li->falh, flp, res, htonl(l->key), mask, i)) <= 0) {
1311 #ifdef CONFIG_IP_FIB_TRIE_STATS
1312 t->stats.semantic_match_passed++;
1316 #ifdef CONFIG_IP_FIB_TRIE_STATS
1317 t->stats.semantic_match_miss++;
1324 fn_trie_lookup(struct fib_table *tb, const struct flowi *flp, struct fib_result *res)
1326 struct trie *t = (struct trie *) tb->tb_data;
1331 t_key key = ntohl(flp->fl4_dst);
1334 int current_prefix_length = KEYLENGTH;
1336 t_key node_prefix, key_prefix, pref_mismatch;
1341 n = rcu_dereference(t->trie);
1345 #ifdef CONFIG_IP_FIB_TRIE_STATS
1351 if ((ret = check_leaf(t, (struct leaf *)n, key, &plen, flp, res)) <= 0)
1355 pn = (struct tnode *) n;
1363 cindex = tkey_extract_bits(mask_pfx(key, current_prefix_length),
1366 n = tnode_get_child(pn, cindex);
1369 #ifdef CONFIG_IP_FIB_TRIE_STATS
1370 t->stats.null_node_hit++;
1376 if ((ret = check_leaf(t, (struct leaf *)n, key, &plen, flp, res)) <= 0)
1384 cn = (struct tnode *)n;
1387 * It's a tnode, and we can do some extra checks here if we
1388 * like, to avoid descending into a dead-end branch.
1389 * This tnode is in the parent's child array at index
1390 * key[p_pos..p_pos+p_bits] but potentially with some bits
1391 * chopped off, so in reality the index may be just a
1392 * subprefix, padded with zero at the end.
1393 * We can also take a look at any skipped bits in this
1394 * tnode - everything up to p_pos is supposed to be ok,
1395 * and the non-chopped bits of the index (se previous
1396 * paragraph) are also guaranteed ok, but the rest is
1397 * considered unknown.
1399 * The skipped bits are key[pos+bits..cn->pos].
1402 /* If current_prefix_length < pos+bits, we are already doing
1403 * actual prefix matching, which means everything from
1404 * pos+(bits-chopped_off) onward must be zero along some
1405 * branch of this subtree - otherwise there is *no* valid
1406 * prefix present. Here we can only check the skipped
1407 * bits. Remember, since we have already indexed into the
1408 * parent's child array, we know that the bits we chopped of
1412 /* NOTA BENE: CHECKING ONLY SKIPPED BITS FOR THE NEW NODE HERE */
1414 if (current_prefix_length < pos+bits) {
1415 if (tkey_extract_bits(cn->key, current_prefix_length,
1416 cn->pos - current_prefix_length) != 0 ||
1422 * If chopped_off=0, the index is fully validated and we
1423 * only need to look at the skipped bits for this, the new,
1424 * tnode. What we actually want to do is to find out if
1425 * these skipped bits match our key perfectly, or if we will
1426 * have to count on finding a matching prefix further down,
1427 * because if we do, we would like to have some way of
1428 * verifying the existence of such a prefix at this point.
1431 /* The only thing we can do at this point is to verify that
1432 * any such matching prefix can indeed be a prefix to our
1433 * key, and if the bits in the node we are inspecting that
1434 * do not match our key are not ZERO, this cannot be true.
1435 * Thus, find out where there is a mismatch (before cn->pos)
1436 * and verify that all the mismatching bits are zero in the
1440 /* Note: We aren't very concerned about the piece of the key
1441 * that precede pn->pos+pn->bits, since these have already been
1442 * checked. The bits after cn->pos aren't checked since these are
1443 * by definition "unknown" at this point. Thus, what we want to
1444 * see is if we are about to enter the "prefix matching" state,
1445 * and in that case verify that the skipped bits that will prevail
1446 * throughout this subtree are zero, as they have to be if we are
1447 * to find a matching prefix.
1450 node_prefix = mask_pfx(cn->key, cn->pos);
1451 key_prefix = mask_pfx(key, cn->pos);
1452 pref_mismatch = key_prefix^node_prefix;
1455 /* In short: If skipped bits in this node do not match the search
1456 * key, enter the "prefix matching" state.directly.
1458 if (pref_mismatch) {
1459 while (!(pref_mismatch & (1<<(KEYLENGTH-1)))) {
1461 pref_mismatch = pref_mismatch <<1;
1463 key_prefix = tkey_extract_bits(cn->key, mp, cn->pos-mp);
1465 if (key_prefix != 0)
1468 if (current_prefix_length >= cn->pos)
1469 current_prefix_length = mp;
1472 pn = (struct tnode *)n; /* Descend */
1479 /* As zero don't change the child key (cindex) */
1480 while ((chopped_off <= pn->bits) && !(cindex & (1<<(chopped_off-1))))
1483 /* Decrease current_... with bits chopped off */
1484 if (current_prefix_length > pn->pos + pn->bits - chopped_off)
1485 current_prefix_length = pn->pos + pn->bits - chopped_off;
1488 * Either we do the actual chop off according or if we have
1489 * chopped off all bits in this tnode walk up to our parent.
1492 if (chopped_off <= pn->bits) {
1493 cindex &= ~(1 << (chopped_off-1));
1495 struct tnode *parent = node_parent((struct node *) pn);
1499 /* Get Child's index */
1500 cindex = tkey_extract_bits(pn->key, parent->pos, parent->bits);
1504 #ifdef CONFIG_IP_FIB_TRIE_STATS
1505 t->stats.backtrack++;
1517 /* only called from updater side */
1518 static int trie_leaf_remove(struct trie *t, t_key key)
1521 struct tnode *tp = NULL;
1522 struct node *n = t->trie;
1525 pr_debug("entering trie_leaf_remove(%p)\n", n);
1527 /* Note that in the case skipped bits, those bits are *not* checked!
1528 * When we finish this, we will have NULL or a T_LEAF, and the
1529 * T_LEAF may or may not match our key.
1532 while (n != NULL && IS_TNODE(n)) {
1533 struct tnode *tn = (struct tnode *) n;
1535 n = tnode_get_child(tn ,tkey_extract_bits(key, tn->pos, tn->bits));
1537 BUG_ON(n && node_parent(n) != tn);
1539 l = (struct leaf *) n;
1541 if (!n || !tkey_equals(l->key, key))
1546 * Remove the leaf and rebalance the tree
1552 tp = node_parent(n);
1553 tnode_free((struct tnode *) n);
1556 cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1557 put_child(t, (struct tnode *)tp, cindex, NULL);
1558 rcu_assign_pointer(t->trie, trie_rebalance(t, tp));
1560 rcu_assign_pointer(t->trie, NULL);
1566 * Caller must hold RTNL.
1568 static int fn_trie_delete(struct fib_table *tb, struct fib_config *cfg)
1570 struct trie *t = (struct trie *) tb->tb_data;
1572 int plen = cfg->fc_dst_len;
1573 u8 tos = cfg->fc_tos;
1574 struct fib_alias *fa, *fa_to_delete;
1575 struct list_head *fa_head;
1577 struct leaf_info *li;
1582 key = ntohl(cfg->fc_dst);
1583 mask = ntohl(inet_make_mask(plen));
1589 l = fib_find_node(t, key);
1594 fa_head = get_fa_head(l, plen);
1595 fa = fib_find_alias(fa_head, tos, 0);
1600 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1602 fa_to_delete = NULL;
1603 fa_head = fa->fa_list.prev;
1605 list_for_each_entry(fa, fa_head, fa_list) {
1606 struct fib_info *fi = fa->fa_info;
1608 if (fa->fa_tos != tos)
1611 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1612 (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1613 fa->fa_scope == cfg->fc_scope) &&
1614 (!cfg->fc_protocol ||
1615 fi->fib_protocol == cfg->fc_protocol) &&
1616 fib_nh_match(cfg, fi) == 0) {
1626 rtmsg_fib(RTM_DELROUTE, htonl(key), fa, plen, tb->tb_id,
1627 &cfg->fc_nlinfo, 0);
1629 l = fib_find_node(t, key);
1630 li = find_leaf_info(l, plen);
1632 list_del_rcu(&fa->fa_list);
1634 if (list_empty(fa_head)) {
1635 hlist_del_rcu(&li->hlist);
1639 if (hlist_empty(&l->list))
1640 trie_leaf_remove(t, key);
1642 if (fa->fa_state & FA_S_ACCESSED)
1645 fib_release_info(fa->fa_info);
1646 alias_free_mem_rcu(fa);
1650 static int trie_flush_list(struct trie *t, struct list_head *head)
1652 struct fib_alias *fa, *fa_node;
1655 list_for_each_entry_safe(fa, fa_node, head, fa_list) {
1656 struct fib_info *fi = fa->fa_info;
1658 if (fi && (fi->fib_flags & RTNH_F_DEAD)) {
1659 list_del_rcu(&fa->fa_list);
1660 fib_release_info(fa->fa_info);
1661 alias_free_mem_rcu(fa);
1668 static int trie_flush_leaf(struct trie *t, struct leaf *l)
1671 struct hlist_head *lih = &l->list;
1672 struct hlist_node *node, *tmp;
1673 struct leaf_info *li = NULL;
1675 hlist_for_each_entry_safe(li, node, tmp, lih, hlist) {
1676 found += trie_flush_list(t, &li->falh);
1678 if (list_empty(&li->falh)) {
1679 hlist_del_rcu(&li->hlist);
1686 /* rcu_read_lock needs to be hold by caller from readside */
1688 static struct leaf *nextleaf(struct trie *t, struct leaf *thisleaf)
1690 struct node *c = (struct node *) thisleaf;
1693 struct node *trie = rcu_dereference(t->trie);
1699 if (IS_LEAF(trie)) /* trie w. just a leaf */
1700 return (struct leaf *) trie;
1702 p = (struct tnode*) trie; /* Start */
1709 /* Find the next child of the parent */
1711 pos = 1 + tkey_extract_bits(c->key, p->pos, p->bits);
1715 last = 1 << p->bits;
1716 for (idx = pos; idx < last ; idx++) {
1717 c = rcu_dereference(p->child[idx]);
1722 /* Decend if tnode */
1723 while (IS_TNODE(c)) {
1724 p = (struct tnode *) c;
1727 /* Rightmost non-NULL branch */
1728 if (p && IS_TNODE(p))
1729 while (!(c = rcu_dereference(p->child[idx]))
1730 && idx < (1<<p->bits)) idx++;
1732 /* Done with this tnode? */
1733 if (idx >= (1 << p->bits) || !c)
1736 return (struct leaf *) c;
1739 /* No more children go up one step */
1740 c = (struct node *) p;
1743 return NULL; /* Ready. Root of trie */
1747 * Caller must hold RTNL.
1749 static int fn_trie_flush(struct fib_table *tb)
1751 struct trie *t = (struct trie *) tb->tb_data;
1752 struct leaf *ll = NULL, *l = NULL;
1757 for (h = 0; (l = nextleaf(t, l)) != NULL; h++) {
1758 found += trie_flush_leaf(t, l);
1760 if (ll && hlist_empty(&ll->list))
1761 trie_leaf_remove(t, ll->key);
1765 if (ll && hlist_empty(&ll->list))
1766 trie_leaf_remove(t, ll->key);
1768 pr_debug("trie_flush found=%d\n", found);
1773 fn_trie_select_default(struct fib_table *tb, const struct flowi *flp, struct fib_result *res)
1775 struct trie *t = (struct trie *) tb->tb_data;
1776 int order, last_idx;
1777 struct fib_info *fi = NULL;
1778 struct fib_info *last_resort;
1779 struct fib_alias *fa = NULL;
1780 struct list_head *fa_head;
1789 l = fib_find_node(t, 0);
1793 fa_head = get_fa_head(l, 0);
1797 if (list_empty(fa_head))
1800 list_for_each_entry_rcu(fa, fa_head, fa_list) {
1801 struct fib_info *next_fi = fa->fa_info;
1803 if (fa->fa_scope != res->scope ||
1804 fa->fa_type != RTN_UNICAST)
1807 if (next_fi->fib_priority > res->fi->fib_priority)
1809 if (!next_fi->fib_nh[0].nh_gw ||
1810 next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1812 fa->fa_state |= FA_S_ACCESSED;
1815 if (next_fi != res->fi)
1817 } else if (!fib_detect_death(fi, order, &last_resort,
1818 &last_idx, tb->tb_default)) {
1819 fib_result_assign(res, fi);
1820 tb->tb_default = order;
1826 if (order <= 0 || fi == NULL) {
1827 tb->tb_default = -1;
1831 if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1833 fib_result_assign(res, fi);
1834 tb->tb_default = order;
1838 fib_result_assign(res, last_resort);
1839 tb->tb_default = last_idx;
1844 static int fn_trie_dump_fa(t_key key, int plen, struct list_head *fah, struct fib_table *tb,
1845 struct sk_buff *skb, struct netlink_callback *cb)
1848 struct fib_alias *fa;
1850 __be32 xkey = htonl(key);
1855 /* rcu_read_lock is hold by caller */
1857 list_for_each_entry_rcu(fa, fah, fa_list) {
1862 BUG_ON(!fa->fa_info);
1864 if (fib_dump_info(skb, NETLINK_CB(cb->skb).pid,
1873 fa->fa_info, 0) < 0) {
1883 static int fn_trie_dump_plen(struct trie *t, int plen, struct fib_table *tb, struct sk_buff *skb,
1884 struct netlink_callback *cb)
1887 struct list_head *fa_head;
1888 struct leaf *l = NULL;
1892 for (h = 0; (l = nextleaf(t, l)) != NULL; h++) {
1896 memset(&cb->args[4], 0,
1897 sizeof(cb->args) - 4*sizeof(cb->args[0]));
1899 fa_head = get_fa_head(l, plen);
1904 if (list_empty(fa_head))
1907 if (fn_trie_dump_fa(l->key, plen, fa_head, tb, skb, cb)<0) {
1916 static int fn_trie_dump(struct fib_table *tb, struct sk_buff *skb, struct netlink_callback *cb)
1919 struct trie *t = (struct trie *) tb->tb_data;
1924 for (m = 0; m <= 32; m++) {
1928 memset(&cb->args[3], 0,
1929 sizeof(cb->args) - 3*sizeof(cb->args[0]));
1931 if (fn_trie_dump_plen(t, 32-m, tb, skb, cb)<0) {
1944 /* Fix more generic FIB names for init later */
1946 struct fib_table *fib_hash_init(u32 id)
1948 struct fib_table *tb;
1951 if (fn_alias_kmem == NULL)
1952 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
1953 sizeof(struct fib_alias),
1954 0, SLAB_HWCACHE_ALIGN,
1957 tb = kmalloc(sizeof(struct fib_table) + sizeof(struct trie),
1963 tb->tb_default = -1;
1964 tb->tb_lookup = fn_trie_lookup;
1965 tb->tb_insert = fn_trie_insert;
1966 tb->tb_delete = fn_trie_delete;
1967 tb->tb_flush = fn_trie_flush;
1968 tb->tb_select_default = fn_trie_select_default;
1969 tb->tb_dump = fn_trie_dump;
1971 t = (struct trie *) tb->tb_data;
1972 memset(t, 0, sizeof(*t));
1974 if (id == RT_TABLE_LOCAL)
1975 printk(KERN_INFO "IPv4 FIB: Using LC-trie version %s\n", VERSION);
1980 #ifdef CONFIG_PROC_FS
1981 /* Depth first Trie walk iterator */
1982 struct fib_trie_iter {
1983 struct seq_net_private p;
1984 struct trie *trie_local, *trie_main;
1985 struct tnode *tnode;
1991 static struct node *fib_trie_get_next(struct fib_trie_iter *iter)
1993 struct tnode *tn = iter->tnode;
1994 unsigned cindex = iter->index;
1997 /* A single entry routing table */
2001 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2002 iter->tnode, iter->index, iter->depth);
2004 while (cindex < (1<<tn->bits)) {
2005 struct node *n = tnode_get_child(tn, cindex);
2010 iter->index = cindex + 1;
2012 /* push down one level */
2013 iter->tnode = (struct tnode *) n;
2023 /* Current node exhausted, pop back up */
2024 p = node_parent((struct node *)tn);
2026 cindex = tkey_extract_bits(tn->key, p->pos, p->bits)+1;
2036 static struct node *fib_trie_get_first(struct fib_trie_iter *iter,
2044 n = rcu_dereference(t->trie);
2051 iter->tnode = (struct tnode *) n;
2066 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2069 struct fib_trie_iter iter;
2071 memset(s, 0, sizeof(*s));
2074 for (n = fib_trie_get_first(&iter, t); n;
2075 n = fib_trie_get_next(&iter)) {
2078 s->totdepth += iter.depth;
2079 if (iter.depth > s->maxdepth)
2080 s->maxdepth = iter.depth;
2082 const struct tnode *tn = (const struct tnode *) n;
2086 if (tn->bits < MAX_STAT_DEPTH)
2087 s->nodesizes[tn->bits]++;
2089 for (i = 0; i < (1<<tn->bits); i++)
2098 * This outputs /proc/net/fib_triestats
2100 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2102 unsigned i, max, pointers, bytes, avdepth;
2105 avdepth = stat->totdepth*100 / stat->leaves;
2109 seq_printf(seq, "\tAver depth: %d.%02d\n", avdepth / 100, avdepth % 100 );
2110 seq_printf(seq, "\tMax depth: %u\n", stat->maxdepth);
2112 seq_printf(seq, "\tLeaves: %u\n", stat->leaves);
2114 bytes = sizeof(struct leaf) * stat->leaves;
2115 seq_printf(seq, "\tInternal nodes: %d\n\t", stat->tnodes);
2116 bytes += sizeof(struct tnode) * stat->tnodes;
2118 max = MAX_STAT_DEPTH;
2119 while (max > 0 && stat->nodesizes[max-1] == 0)
2123 for (i = 1; i <= max; i++)
2124 if (stat->nodesizes[i] != 0) {
2125 seq_printf(seq, " %d: %d", i, stat->nodesizes[i]);
2126 pointers += (1<<i) * stat->nodesizes[i];
2128 seq_putc(seq, '\n');
2129 seq_printf(seq, "\tPointers: %d\n", pointers);
2131 bytes += sizeof(struct node *) * pointers;
2132 seq_printf(seq, "Null ptrs: %d\n", stat->nullpointers);
2133 seq_printf(seq, "Total size: %d kB\n", (bytes + 1023) / 1024);
2135 #ifdef CONFIG_IP_FIB_TRIE_STATS
2136 seq_printf(seq, "Counters:\n---------\n");
2137 seq_printf(seq,"gets = %d\n", t->stats.gets);
2138 seq_printf(seq,"backtracks = %d\n", t->stats.backtrack);
2139 seq_printf(seq,"semantic match passed = %d\n", t->stats.semantic_match_passed);
2140 seq_printf(seq,"semantic match miss = %d\n", t->stats.semantic_match_miss);
2141 seq_printf(seq,"null node hit= %d\n", t->stats.null_node_hit);
2142 seq_printf(seq,"skipped node resize = %d\n", t->stats.resize_node_skipped);
2144 memset(&(t->stats), 0, sizeof(t->stats));
2146 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2149 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2151 struct net *net = (struct net *)seq->private;
2152 struct trie *trie_local, *trie_main;
2153 struct trie_stat *stat;
2154 struct fib_table *tb;
2157 tb = fib_get_table(net, RT_TABLE_LOCAL);
2159 trie_local = (struct trie *) tb->tb_data;
2162 tb = fib_get_table(net, RT_TABLE_MAIN);
2164 trie_main = (struct trie *) tb->tb_data;
2167 stat = kmalloc(sizeof(*stat), GFP_KERNEL);
2171 seq_printf(seq, "Basic info: size of leaf: %Zd bytes, size of tnode: %Zd bytes.\n",
2172 sizeof(struct leaf), sizeof(struct tnode));
2175 seq_printf(seq, "Local:\n");
2176 trie_collect_stats(trie_local, stat);
2177 trie_show_stats(seq, stat);
2181 seq_printf(seq, "Main:\n");
2182 trie_collect_stats(trie_main, stat);
2183 trie_show_stats(seq, stat);
2190 static int fib_triestat_seq_open(struct inode *inode, struct file *file)
2195 net = get_proc_net(inode);
2198 err = single_open(file, fib_triestat_seq_show, net);
2206 static int fib_triestat_seq_release(struct inode *ino, struct file *f)
2208 struct seq_file *seq = f->private_data;
2209 put_net(seq->private);
2210 return single_release(ino, f);
2213 static const struct file_operations fib_triestat_fops = {
2214 .owner = THIS_MODULE,
2215 .open = fib_triestat_seq_open,
2217 .llseek = seq_lseek,
2218 .release = fib_triestat_seq_release,
2221 static struct node *fib_trie_get_idx(struct fib_trie_iter *iter,
2227 for (n = fib_trie_get_first(iter, iter->trie_local);
2228 n; ++idx, n = fib_trie_get_next(iter)) {
2233 for (n = fib_trie_get_first(iter, iter->trie_main);
2234 n; ++idx, n = fib_trie_get_next(iter)) {
2241 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2243 struct fib_trie_iter *iter = seq->private;
2244 struct fib_table *tb;
2246 if (!iter->trie_local) {
2247 tb = fib_get_table(iter->p.net, RT_TABLE_LOCAL);
2249 iter->trie_local = (struct trie *) tb->tb_data;
2251 if (!iter->trie_main) {
2252 tb = fib_get_table(iter->p.net, RT_TABLE_MAIN);
2254 iter->trie_main = (struct trie *) tb->tb_data;
2258 return SEQ_START_TOKEN;
2259 return fib_trie_get_idx(iter, *pos - 1);
2262 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2264 struct fib_trie_iter *iter = seq->private;
2268 if (v == SEQ_START_TOKEN)
2269 return fib_trie_get_idx(iter, 0);
2271 v = fib_trie_get_next(iter);
2276 /* continue scan in next trie */
2277 if (iter->trie == iter->trie_local)
2278 return fib_trie_get_first(iter, iter->trie_main);
2283 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2288 static void seq_indent(struct seq_file *seq, int n)
2290 while (n-- > 0) seq_puts(seq, " ");
2293 static inline const char *rtn_scope(enum rt_scope_t s)
2295 static char buf[32];
2298 case RT_SCOPE_UNIVERSE: return "universe";
2299 case RT_SCOPE_SITE: return "site";
2300 case RT_SCOPE_LINK: return "link";
2301 case RT_SCOPE_HOST: return "host";
2302 case RT_SCOPE_NOWHERE: return "nowhere";
2304 snprintf(buf, sizeof(buf), "scope=%d", s);
2309 static const char *rtn_type_names[__RTN_MAX] = {
2310 [RTN_UNSPEC] = "UNSPEC",
2311 [RTN_UNICAST] = "UNICAST",
2312 [RTN_LOCAL] = "LOCAL",
2313 [RTN_BROADCAST] = "BROADCAST",
2314 [RTN_ANYCAST] = "ANYCAST",
2315 [RTN_MULTICAST] = "MULTICAST",
2316 [RTN_BLACKHOLE] = "BLACKHOLE",
2317 [RTN_UNREACHABLE] = "UNREACHABLE",
2318 [RTN_PROHIBIT] = "PROHIBIT",
2319 [RTN_THROW] = "THROW",
2321 [RTN_XRESOLVE] = "XRESOLVE",
2324 static inline const char *rtn_type(unsigned t)
2326 static char buf[32];
2328 if (t < __RTN_MAX && rtn_type_names[t])
2329 return rtn_type_names[t];
2330 snprintf(buf, sizeof(buf), "type %d", t);
2334 /* Pretty print the trie */
2335 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2337 const struct fib_trie_iter *iter = seq->private;
2340 if (v == SEQ_START_TOKEN)
2343 if (!node_parent(n)) {
2344 if (iter->trie == iter->trie_local)
2345 seq_puts(seq, "<local>:\n");
2347 seq_puts(seq, "<main>:\n");
2351 struct tnode *tn = (struct tnode *) n;
2352 __be32 prf = htonl(mask_pfx(tn->key, tn->pos));
2354 seq_indent(seq, iter->depth-1);
2355 seq_printf(seq, " +-- %d.%d.%d.%d/%d %d %d %d\n",
2356 NIPQUAD(prf), tn->pos, tn->bits, tn->full_children,
2357 tn->empty_children);
2360 struct leaf *l = (struct leaf *) n;
2362 __be32 val = htonl(l->key);
2364 seq_indent(seq, iter->depth);
2365 seq_printf(seq, " |-- %d.%d.%d.%d\n", NIPQUAD(val));
2366 for (i = 32; i >= 0; i--) {
2367 struct leaf_info *li = find_leaf_info(l, i);
2369 struct fib_alias *fa;
2370 list_for_each_entry_rcu(fa, &li->falh, fa_list) {
2371 seq_indent(seq, iter->depth+1);
2372 seq_printf(seq, " /%d %s %s", i,
2373 rtn_scope(fa->fa_scope),
2374 rtn_type(fa->fa_type));
2376 seq_printf(seq, "tos =%d\n",
2378 seq_putc(seq, '\n');
2387 static const struct seq_operations fib_trie_seq_ops = {
2388 .start = fib_trie_seq_start,
2389 .next = fib_trie_seq_next,
2390 .stop = fib_trie_seq_stop,
2391 .show = fib_trie_seq_show,
2394 static int fib_trie_seq_open(struct inode *inode, struct file *file)
2396 return seq_open_net(inode, file, &fib_trie_seq_ops,
2397 sizeof(struct fib_trie_iter));
2400 static const struct file_operations fib_trie_fops = {
2401 .owner = THIS_MODULE,
2402 .open = fib_trie_seq_open,
2404 .llseek = seq_lseek,
2405 .release = seq_release_net,
2408 static unsigned fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2410 static unsigned type2flags[RTN_MAX + 1] = {
2411 [7] = RTF_REJECT, [8] = RTF_REJECT,
2413 unsigned flags = type2flags[type];
2415 if (fi && fi->fib_nh->nh_gw)
2416 flags |= RTF_GATEWAY;
2417 if (mask == htonl(0xFFFFFFFF))
2424 * This outputs /proc/net/route.
2425 * The format of the file is not supposed to be changed
2426 * and needs to be same as fib_hash output to avoid breaking
2429 static int fib_route_seq_show(struct seq_file *seq, void *v)
2431 const struct fib_trie_iter *iter = seq->private;
2436 if (v == SEQ_START_TOKEN) {
2437 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2438 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2443 if (iter->trie == iter->trie_local)
2448 for (i=32; i>=0; i--) {
2449 struct leaf_info *li = find_leaf_info(l, i);
2450 struct fib_alias *fa;
2451 __be32 mask, prefix;
2456 mask = inet_make_mask(li->plen);
2457 prefix = htonl(l->key);
2459 list_for_each_entry_rcu(fa, &li->falh, fa_list) {
2460 const struct fib_info *fi = fa->fa_info;
2461 unsigned flags = fib_flag_trans(fa->fa_type, mask, fi);
2463 if (fa->fa_type == RTN_BROADCAST
2464 || fa->fa_type == RTN_MULTICAST)
2468 snprintf(bf, sizeof(bf),
2469 "%s\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u",
2470 fi->fib_dev ? fi->fib_dev->name : "*",
2472 fi->fib_nh->nh_gw, flags, 0, 0,
2475 (fi->fib_advmss ? fi->fib_advmss + 40 : 0),
2479 snprintf(bf, sizeof(bf),
2480 "*\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u",
2481 prefix, 0, flags, 0, 0, 0,
2484 seq_printf(seq, "%-127s\n", bf);
2491 static const struct seq_operations fib_route_seq_ops = {
2492 .start = fib_trie_seq_start,
2493 .next = fib_trie_seq_next,
2494 .stop = fib_trie_seq_stop,
2495 .show = fib_route_seq_show,
2498 static int fib_route_seq_open(struct inode *inode, struct file *file)
2500 return seq_open_net(inode, file, &fib_route_seq_ops,
2501 sizeof(struct fib_trie_iter));
2504 static const struct file_operations fib_route_fops = {
2505 .owner = THIS_MODULE,
2506 .open = fib_route_seq_open,
2508 .llseek = seq_lseek,
2509 .release = seq_release_net,
2512 int __net_init fib_proc_init(struct net *net)
2514 if (!proc_net_fops_create(net, "fib_trie", S_IRUGO, &fib_trie_fops))
2517 if (!proc_net_fops_create(net, "fib_triestat", S_IRUGO,
2518 &fib_triestat_fops))
2521 if (!proc_net_fops_create(net, "route", S_IRUGO, &fib_route_fops))
2527 proc_net_remove(net, "fib_triestat");
2529 proc_net_remove(net, "fib_trie");
2534 void __net_exit fib_proc_exit(struct net *net)
2536 proc_net_remove(net, "fib_trie");
2537 proc_net_remove(net, "fib_triestat");
2538 proc_net_remove(net, "route");
2541 #endif /* CONFIG_PROC_FS */