2 * This program is free software; you can redistribute it and/or
3 * modify it under the terms of the GNU General Public License
4 * as published by the Free Software Foundation; either version
5 * 2 of the License, or (at your option) any later version.
7 * Robert Olsson <robert.olsson@its.uu.se> Uppsala Universitet
8 * & Swedish University of Agricultural Sciences.
10 * Jens Laas <jens.laas@data.slu.se> Swedish University of
11 * Agricultural Sciences.
13 * Hans Liss <hans.liss@its.uu.se> Uppsala Universitet
15 * This work is based on the LPC-trie which is originally descibed in:
17 * An experimental study of compression methods for dynamic tries
18 * Stefan Nilsson and Matti Tikkanen. Algorithmica, 33(1):19-33, 2002.
19 * http://www.nada.kth.se/~snilsson/public/papers/dyntrie2/
22 * IP-address lookup using LC-tries. Stefan Nilsson and Gunnar Karlsson
23 * IEEE Journal on Selected Areas in Communications, 17(6):1083-1092, June 1999
26 * Code from fib_hash has been reused which includes the following header:
29 * INET An implementation of the TCP/IP protocol suite for the LINUX
30 * operating system. INET is implemented using the BSD Socket
31 * interface as the means of communication with the user level.
33 * IPv4 FIB: lookup engine and maintenance routines.
36 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
38 * This program is free software; you can redistribute it and/or
39 * modify it under the terms of the GNU General Public License
40 * as published by the Free Software Foundation; either version
41 * 2 of the License, or (at your option) any later version.
43 * Substantial contributions to this work comes from:
45 * David S. Miller, <davem@davemloft.net>
46 * Stephen Hemminger <shemminger@osdl.org>
47 * Paul E. McKenney <paulmck@us.ibm.com>
48 * Patrick McHardy <kaber@trash.net>
51 #define VERSION "0.408"
53 #include <asm/uaccess.h>
54 #include <asm/system.h>
55 #include <linux/bitops.h>
56 #include <linux/types.h>
57 #include <linux/kernel.h>
59 #include <linux/string.h>
60 #include <linux/socket.h>
61 #include <linux/sockios.h>
62 #include <linux/errno.h>
64 #include <linux/inet.h>
65 #include <linux/inetdevice.h>
66 #include <linux/netdevice.h>
67 #include <linux/if_arp.h>
68 #include <linux/proc_fs.h>
69 #include <linux/rcupdate.h>
70 #include <linux/skbuff.h>
71 #include <linux/netlink.h>
72 #include <linux/init.h>
73 #include <linux/list.h>
74 #include <net/net_namespace.h>
76 #include <net/protocol.h>
77 #include <net/route.h>
80 #include <net/ip_fib.h>
81 #include "fib_lookup.h"
83 #define MAX_STAT_DEPTH 32
85 #define KEYLENGTH (8*sizeof(t_key))
87 typedef unsigned int t_key;
91 #define NODE_TYPE_MASK 0x1UL
92 #define NODE_TYPE(node) ((node)->parent & NODE_TYPE_MASK)
94 #define IS_TNODE(n) (!(n->parent & T_LEAF))
95 #define IS_LEAF(n) (n->parent & T_LEAF)
103 unsigned long parent;
105 struct hlist_head list;
110 struct hlist_node hlist;
113 struct list_head falh;
117 unsigned long parent;
119 unsigned char pos; /* 2log(KEYLENGTH) bits needed */
120 unsigned char bits; /* 2log(KEYLENGTH) bits needed */
121 unsigned int full_children; /* KEYLENGTH bits needed */
122 unsigned int empty_children; /* KEYLENGTH bits needed */
125 struct work_struct work;
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 prefixes;
148 unsigned int nodesizes[MAX_STAT_DEPTH];
153 #ifdef CONFIG_IP_FIB_TRIE_STATS
154 struct trie_use_stats stats;
158 static void put_child(struct trie *t, struct tnode *tn, int i, struct node *n);
159 static void tnode_put_child_reorg(struct tnode *tn, int i, struct node *n,
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);
165 static struct kmem_cache *fn_alias_kmem __read_mostly;
166 static struct kmem_cache *trie_leaf_kmem __read_mostly;
168 static inline struct tnode *node_parent(struct node *node)
170 return (struct tnode *)(node->parent & ~NODE_TYPE_MASK);
173 static inline struct tnode *node_parent_rcu(struct node *node)
175 struct tnode *ret = node_parent(node);
177 return rcu_dereference(ret);
180 /* Same as rcu_assign_pointer
181 * but that macro() assumes that value is a pointer.
183 static inline void node_set_parent(struct node *node, struct tnode *ptr)
186 node->parent = (unsigned long)ptr | NODE_TYPE(node);
189 static inline struct node *tnode_get_child(struct tnode *tn, unsigned int i)
191 BUG_ON(i >= 1U << tn->bits);
196 static inline struct node *tnode_get_child_rcu(struct tnode *tn, unsigned int i)
198 struct node *ret = tnode_get_child(tn, i);
200 return rcu_dereference(ret);
203 static inline int tnode_child_length(const struct tnode *tn)
205 return 1 << tn->bits;
208 static inline t_key mask_pfx(t_key k, unsigned short l)
210 return (l == 0) ? 0 : k >> (KEYLENGTH-l) << (KEYLENGTH-l);
213 static inline t_key tkey_extract_bits(t_key a, int offset, int bits)
215 if (offset < KEYLENGTH)
216 return ((t_key)(a << offset)) >> (KEYLENGTH - bits);
221 static inline int tkey_equals(t_key a, t_key b)
226 static inline int tkey_sub_equals(t_key a, int offset, int bits, t_key b)
228 if (bits == 0 || offset >= KEYLENGTH)
230 bits = bits > KEYLENGTH ? KEYLENGTH : bits;
231 return ((a ^ b) << offset) >> (KEYLENGTH - bits) == 0;
234 static inline int tkey_mismatch(t_key a, int offset, t_key b)
241 while ((diff << i) >> (KEYLENGTH-1) == 0)
247 To understand this stuff, an understanding of keys and all their bits is
248 necessary. Every node in the trie has a key associated with it, but not
249 all of the bits in that key are significant.
251 Consider a node 'n' and its parent 'tp'.
253 If n is a leaf, every bit in its key is significant. Its presence is
254 necessitated by path compression, since during a tree traversal (when
255 searching for a leaf - unless we are doing an insertion) we will completely
256 ignore all skipped bits we encounter. Thus we need to verify, at the end of
257 a potentially successful search, that we have indeed been walking the
260 Note that we can never "miss" the correct key in the tree if present by
261 following the wrong path. Path compression ensures that segments of the key
262 that are the same for all keys with a given prefix are skipped, but the
263 skipped part *is* identical for each node in the subtrie below the skipped
264 bit! trie_insert() in this implementation takes care of that - note the
265 call to tkey_sub_equals() in trie_insert().
267 if n is an internal node - a 'tnode' here, the various parts of its key
268 have many different meanings.
271 _________________________________________________________________
272 | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
273 -----------------------------------------------------------------
274 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
276 _________________________________________________________________
277 | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
278 -----------------------------------------------------------------
279 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
286 First, let's just ignore the bits that come before the parent tp, that is
287 the bits from 0 to (tp->pos-1). They are *known* but at this point we do
288 not use them for anything.
290 The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
291 index into the parent's child array. That is, they will be used to find
292 'n' among tp's children.
294 The bits from (tp->pos + tp->bits) to (n->pos - 1) - "S" - are skipped bits
297 All the bits we have seen so far are significant to the node n. The rest
298 of the bits are really not needed or indeed known in n->key.
300 The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
301 n's child array, and will of course be different for each child.
304 The rest of the bits, from (n->pos + n->bits) onward, are completely unknown
309 static inline void check_tnode(const struct tnode *tn)
311 WARN_ON(tn && tn->pos+tn->bits > 32);
314 static const int halve_threshold = 25;
315 static const int inflate_threshold = 50;
316 static const int halve_threshold_root = 8;
317 static const int inflate_threshold_root = 15;
320 static void __alias_free_mem(struct rcu_head *head)
322 struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
323 kmem_cache_free(fn_alias_kmem, fa);
326 static inline void alias_free_mem_rcu(struct fib_alias *fa)
328 call_rcu(&fa->rcu, __alias_free_mem);
331 static void __leaf_free_rcu(struct rcu_head *head)
333 struct leaf *l = container_of(head, struct leaf, rcu);
334 kmem_cache_free(trie_leaf_kmem, l);
337 static inline void free_leaf(struct leaf *l)
339 call_rcu_bh(&l->rcu, __leaf_free_rcu);
342 static void __leaf_info_free_rcu(struct rcu_head *head)
344 kfree(container_of(head, struct leaf_info, rcu));
347 static inline void free_leaf_info(struct leaf_info *leaf)
349 call_rcu(&leaf->rcu, __leaf_info_free_rcu);
352 static struct tnode *tnode_alloc(size_t size)
354 if (size <= PAGE_SIZE)
355 return kzalloc(size, GFP_KERNEL);
357 return __vmalloc(size, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL);
360 static void __tnode_vfree(struct work_struct *arg)
362 struct tnode *tn = container_of(arg, struct tnode, work);
366 static void __tnode_free_rcu(struct rcu_head *head)
368 struct tnode *tn = container_of(head, struct tnode, rcu);
369 size_t size = sizeof(struct tnode) +
370 (sizeof(struct node *) << tn->bits);
372 if (size <= PAGE_SIZE)
375 INIT_WORK(&tn->work, __tnode_vfree);
376 schedule_work(&tn->work);
380 static inline void tnode_free(struct tnode *tn)
383 free_leaf((struct leaf *) tn);
385 call_rcu(&tn->rcu, __tnode_free_rcu);
388 static struct leaf *leaf_new(void)
390 struct leaf *l = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
393 INIT_HLIST_HEAD(&l->list);
398 static struct leaf_info *leaf_info_new(int plen)
400 struct leaf_info *li = kmalloc(sizeof(struct leaf_info), GFP_KERNEL);
403 INIT_LIST_HEAD(&li->falh);
408 static struct tnode *tnode_new(t_key key, int pos, int bits)
410 size_t sz = sizeof(struct tnode) + (sizeof(struct node *) << bits);
411 struct tnode *tn = tnode_alloc(sz);
414 tn->parent = T_TNODE;
418 tn->full_children = 0;
419 tn->empty_children = 1<<bits;
422 pr_debug("AT %p s=%u %lu\n", tn, (unsigned int) sizeof(struct tnode),
423 (unsigned long) (sizeof(struct node) << bits));
428 * Check whether a tnode 'n' is "full", i.e. it is an internal node
429 * and no bits are skipped. See discussion in dyntree paper p. 6
432 static inline int tnode_full(const struct tnode *tn, const struct node *n)
434 if (n == NULL || IS_LEAF(n))
437 return ((struct tnode *) n)->pos == tn->pos + tn->bits;
440 static inline void put_child(struct trie *t, struct tnode *tn, int i,
443 tnode_put_child_reorg(tn, i, n, -1);
447 * Add a child at position i overwriting the old value.
448 * Update the value of full_children and empty_children.
451 static void tnode_put_child_reorg(struct tnode *tn, int i, struct node *n,
454 struct node *chi = tn->child[i];
457 BUG_ON(i >= 1<<tn->bits);
459 /* update emptyChildren */
460 if (n == NULL && chi != NULL)
461 tn->empty_children++;
462 else if (n != NULL && chi == NULL)
463 tn->empty_children--;
465 /* update fullChildren */
467 wasfull = tnode_full(tn, chi);
469 isfull = tnode_full(tn, n);
470 if (wasfull && !isfull)
472 else if (!wasfull && isfull)
476 node_set_parent(n, tn);
478 rcu_assign_pointer(tn->child[i], n);
481 static struct node *resize(struct trie *t, struct tnode *tn)
485 struct tnode *old_tn;
486 int inflate_threshold_use;
487 int halve_threshold_use;
493 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
494 tn, inflate_threshold, halve_threshold);
497 if (tn->empty_children == tnode_child_length(tn)) {
502 if (tn->empty_children == tnode_child_length(tn) - 1)
503 for (i = 0; i < tnode_child_length(tn); i++) {
510 /* compress one level */
511 node_set_parent(n, NULL);
516 * Double as long as the resulting node has a number of
517 * nonempty nodes that are above the threshold.
521 * From "Implementing a dynamic compressed trie" by Stefan Nilsson of
522 * the Helsinki University of Technology and Matti Tikkanen of Nokia
523 * Telecommunications, page 6:
524 * "A node is doubled if the ratio of non-empty children to all
525 * children in the *doubled* node is at least 'high'."
527 * 'high' in this instance is the variable 'inflate_threshold'. It
528 * is expressed as a percentage, so we multiply it with
529 * tnode_child_length() and instead of multiplying by 2 (since the
530 * child array will be doubled by inflate()) and multiplying
531 * the left-hand side by 100 (to handle the percentage thing) we
532 * multiply the left-hand side by 50.
534 * The left-hand side may look a bit weird: tnode_child_length(tn)
535 * - tn->empty_children is of course the number of non-null children
536 * in the current node. tn->full_children is the number of "full"
537 * children, that is non-null tnodes with a skip value of 0.
538 * All of those will be doubled in the resulting inflated tnode, so
539 * we just count them one extra time here.
541 * A clearer way to write this would be:
543 * to_be_doubled = tn->full_children;
544 * not_to_be_doubled = tnode_child_length(tn) - tn->empty_children -
547 * new_child_length = tnode_child_length(tn) * 2;
549 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
551 * if (new_fill_factor >= inflate_threshold)
553 * ...and so on, tho it would mess up the while () loop.
556 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
560 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
561 * inflate_threshold * new_child_length
563 * expand not_to_be_doubled and to_be_doubled, and shorten:
564 * 100 * (tnode_child_length(tn) - tn->empty_children +
565 * tn->full_children) >= inflate_threshold * new_child_length
567 * expand new_child_length:
568 * 100 * (tnode_child_length(tn) - tn->empty_children +
569 * tn->full_children) >=
570 * inflate_threshold * tnode_child_length(tn) * 2
573 * 50 * (tn->full_children + tnode_child_length(tn) -
574 * tn->empty_children) >= inflate_threshold *
575 * tnode_child_length(tn)
581 /* Keep root node larger */
584 inflate_threshold_use = inflate_threshold_root;
586 inflate_threshold_use = inflate_threshold;
590 while ((tn->full_children > 0 && max_resize-- &&
591 50 * (tn->full_children + tnode_child_length(tn)
592 - tn->empty_children)
593 >= inflate_threshold_use * tnode_child_length(tn))) {
600 #ifdef CONFIG_IP_FIB_TRIE_STATS
601 t->stats.resize_node_skipped++;
607 if (max_resize < 0) {
609 pr_warning("Fix inflate_threshold_root."
610 " Now=%d size=%d bits\n",
611 inflate_threshold_root, tn->bits);
613 pr_warning("Fix inflate_threshold."
614 " Now=%d size=%d bits\n",
615 inflate_threshold, tn->bits);
621 * Halve as long as the number of empty children in this
622 * node is above threshold.
626 /* Keep root node larger */
629 halve_threshold_use = halve_threshold_root;
631 halve_threshold_use = halve_threshold;
635 while (tn->bits > 1 && max_resize-- &&
636 100 * (tnode_child_length(tn) - tn->empty_children) <
637 halve_threshold_use * tnode_child_length(tn)) {
643 #ifdef CONFIG_IP_FIB_TRIE_STATS
644 t->stats.resize_node_skipped++;
650 if (max_resize < 0) {
652 pr_warning("Fix halve_threshold_root."
653 " Now=%d size=%d bits\n",
654 halve_threshold_root, tn->bits);
656 pr_warning("Fix halve_threshold."
657 " Now=%d size=%d bits\n",
658 halve_threshold, tn->bits);
661 /* Only one child remains */
662 if (tn->empty_children == tnode_child_length(tn) - 1)
663 for (i = 0; i < tnode_child_length(tn); i++) {
670 /* compress one level */
672 node_set_parent(n, NULL);
677 return (struct node *) tn;
680 static struct tnode *inflate(struct trie *t, struct tnode *tn)
682 struct tnode *oldtnode = tn;
683 int olen = tnode_child_length(tn);
686 pr_debug("In inflate\n");
688 tn = tnode_new(oldtnode->key, oldtnode->pos, oldtnode->bits + 1);
691 return ERR_PTR(-ENOMEM);
694 * Preallocate and store tnodes before the actual work so we
695 * don't get into an inconsistent state if memory allocation
696 * fails. In case of failure we return the oldnode and inflate
697 * of tnode is ignored.
700 for (i = 0; i < olen; i++) {
703 inode = (struct tnode *) tnode_get_child(oldtnode, i);
706 inode->pos == oldtnode->pos + oldtnode->bits &&
708 struct tnode *left, *right;
709 t_key m = ~0U << (KEYLENGTH - 1) >> inode->pos;
711 left = tnode_new(inode->key&(~m), inode->pos + 1,
716 right = tnode_new(inode->key|m, inode->pos + 1,
724 put_child(t, tn, 2*i, (struct node *) left);
725 put_child(t, tn, 2*i+1, (struct node *) right);
729 for (i = 0; i < olen; i++) {
731 struct node *node = tnode_get_child(oldtnode, i);
732 struct tnode *left, *right;
739 /* A leaf or an internal node with skipped bits */
741 if (IS_LEAF(node) || ((struct tnode *) node)->pos >
742 tn->pos + tn->bits - 1) {
743 if (tkey_extract_bits(node->key,
744 oldtnode->pos + oldtnode->bits,
746 put_child(t, tn, 2*i, node);
748 put_child(t, tn, 2*i+1, node);
752 /* An internal node with two children */
753 inode = (struct tnode *) node;
755 if (inode->bits == 1) {
756 put_child(t, tn, 2*i, inode->child[0]);
757 put_child(t, tn, 2*i+1, inode->child[1]);
763 /* An internal node with more than two children */
765 /* We will replace this node 'inode' with two new
766 * ones, 'left' and 'right', each with half of the
767 * original children. The two new nodes will have
768 * a position one bit further down the key and this
769 * means that the "significant" part of their keys
770 * (see the discussion near the top of this file)
771 * will differ by one bit, which will be "0" in
772 * left's key and "1" in right's key. Since we are
773 * moving the key position by one step, the bit that
774 * we are moving away from - the bit at position
775 * (inode->pos) - is the one that will differ between
776 * left and right. So... we synthesize that bit in the
778 * The mask 'm' below will be a single "one" bit at
779 * the position (inode->pos)
782 /* Use the old key, but set the new significant
786 left = (struct tnode *) tnode_get_child(tn, 2*i);
787 put_child(t, tn, 2*i, NULL);
791 right = (struct tnode *) tnode_get_child(tn, 2*i+1);
792 put_child(t, tn, 2*i+1, NULL);
796 size = tnode_child_length(left);
797 for (j = 0; j < size; j++) {
798 put_child(t, left, j, inode->child[j]);
799 put_child(t, right, j, inode->child[j + size]);
801 put_child(t, tn, 2*i, resize(t, left));
802 put_child(t, tn, 2*i+1, resize(t, right));
806 tnode_free(oldtnode);
810 int size = tnode_child_length(tn);
813 for (j = 0; j < size; j++)
815 tnode_free((struct tnode *)tn->child[j]);
819 return ERR_PTR(-ENOMEM);
823 static struct tnode *halve(struct trie *t, struct tnode *tn)
825 struct tnode *oldtnode = tn;
826 struct node *left, *right;
828 int olen = tnode_child_length(tn);
830 pr_debug("In halve\n");
832 tn = tnode_new(oldtnode->key, oldtnode->pos, oldtnode->bits - 1);
835 return ERR_PTR(-ENOMEM);
838 * Preallocate and store tnodes before the actual work so we
839 * don't get into an inconsistent state if memory allocation
840 * fails. In case of failure we return the oldnode and halve
841 * of tnode is ignored.
844 for (i = 0; i < olen; i += 2) {
845 left = tnode_get_child(oldtnode, i);
846 right = tnode_get_child(oldtnode, i+1);
848 /* Two nonempty children */
852 newn = tnode_new(left->key, tn->pos + tn->bits, 1);
857 put_child(t, tn, i/2, (struct node *)newn);
862 for (i = 0; i < olen; i += 2) {
863 struct tnode *newBinNode;
865 left = tnode_get_child(oldtnode, i);
866 right = tnode_get_child(oldtnode, i+1);
868 /* At least one of the children is empty */
870 if (right == NULL) /* Both are empty */
872 put_child(t, tn, i/2, right);
877 put_child(t, tn, i/2, left);
881 /* Two nonempty children */
882 newBinNode = (struct tnode *) tnode_get_child(tn, i/2);
883 put_child(t, tn, i/2, NULL);
884 put_child(t, newBinNode, 0, left);
885 put_child(t, newBinNode, 1, right);
886 put_child(t, tn, i/2, resize(t, newBinNode));
888 tnode_free(oldtnode);
892 int size = tnode_child_length(tn);
895 for (j = 0; j < size; j++)
897 tnode_free((struct tnode *)tn->child[j]);
901 return ERR_PTR(-ENOMEM);
905 /* readside must use rcu_read_lock currently dump routines
906 via get_fa_head and dump */
908 static struct leaf_info *find_leaf_info(struct leaf *l, int plen)
910 struct hlist_head *head = &l->list;
911 struct hlist_node *node;
912 struct leaf_info *li;
914 hlist_for_each_entry_rcu(li, node, head, hlist)
915 if (li->plen == plen)
921 static inline struct list_head *get_fa_head(struct leaf *l, int plen)
923 struct leaf_info *li = find_leaf_info(l, plen);
931 static void insert_leaf_info(struct hlist_head *head, struct leaf_info *new)
933 struct leaf_info *li = NULL, *last = NULL;
934 struct hlist_node *node;
936 if (hlist_empty(head)) {
937 hlist_add_head_rcu(&new->hlist, head);
939 hlist_for_each_entry(li, node, head, hlist) {
940 if (new->plen > li->plen)
946 hlist_add_after_rcu(&last->hlist, &new->hlist);
948 hlist_add_before_rcu(&new->hlist, &li->hlist);
952 /* rcu_read_lock needs to be hold by caller from readside */
955 fib_find_node(struct trie *t, u32 key)
962 n = rcu_dereference(t->trie);
964 while (n != NULL && NODE_TYPE(n) == T_TNODE) {
965 tn = (struct tnode *) n;
969 if (tkey_sub_equals(tn->key, pos, tn->pos-pos, key)) {
970 pos = tn->pos + tn->bits;
971 n = tnode_get_child_rcu(tn,
972 tkey_extract_bits(key,
978 /* Case we have found a leaf. Compare prefixes */
980 if (n != NULL && IS_LEAF(n) && tkey_equals(key, n->key))
981 return (struct leaf *)n;
986 static struct node *trie_rebalance(struct trie *t, struct tnode *tn)
995 while (tn != NULL && (tp = node_parent((struct node *)tn)) != NULL) {
996 cindex = tkey_extract_bits(key, tp->pos, tp->bits);
997 wasfull = tnode_full(tp, tnode_get_child(tp, cindex));
998 tn = (struct tnode *) resize(t, (struct tnode *)tn);
1000 tnode_put_child_reorg((struct tnode *)tp, cindex,
1001 (struct node *)tn, wasfull);
1003 tp = node_parent((struct node *) tn);
1009 /* Handle last (top) tnode */
1011 tn = (struct tnode *)resize(t, (struct tnode *)tn);
1014 return (struct node *)tn;
1017 /* only used from updater-side */
1019 static struct list_head *fib_insert_node(struct trie *t, u32 key, int plen)
1022 struct tnode *tp = NULL, *tn = NULL;
1026 struct list_head *fa_head = NULL;
1027 struct leaf_info *li;
1033 /* If we point to NULL, stop. Either the tree is empty and we should
1034 * just put a new leaf in if, or we have reached an empty child slot,
1035 * and we should just put our new leaf in that.
1036 * If we point to a T_TNODE, check if it matches our key. Note that
1037 * a T_TNODE might be skipping any number of bits - its 'pos' need
1038 * not be the parent's 'pos'+'bits'!
1040 * If it does match the current key, get pos/bits from it, extract
1041 * the index from our key, push the T_TNODE and walk the tree.
1043 * If it doesn't, we have to replace it with a new T_TNODE.
1045 * If we point to a T_LEAF, it might or might not have the same key
1046 * as we do. If it does, just change the value, update the T_LEAF's
1047 * value, and return it.
1048 * If it doesn't, we need to replace it with a T_TNODE.
1051 while (n != NULL && NODE_TYPE(n) == T_TNODE) {
1052 tn = (struct tnode *) n;
1056 if (tkey_sub_equals(tn->key, pos, tn->pos-pos, key)) {
1058 pos = tn->pos + tn->bits;
1059 n = tnode_get_child(tn,
1060 tkey_extract_bits(key,
1064 BUG_ON(n && node_parent(n) != tn);
1070 * n ----> NULL, LEAF or TNODE
1072 * tp is n's (parent) ----> NULL or TNODE
1075 BUG_ON(tp && IS_LEAF(tp));
1077 /* Case 1: n is a leaf. Compare prefixes */
1079 if (n != NULL && IS_LEAF(n) && tkey_equals(key, n->key)) {
1080 l = (struct leaf *) n;
1081 li = leaf_info_new(plen);
1086 fa_head = &li->falh;
1087 insert_leaf_info(&l->list, li);
1096 li = leaf_info_new(plen);
1103 fa_head = &li->falh;
1104 insert_leaf_info(&l->list, li);
1106 if (t->trie && n == NULL) {
1107 /* Case 2: n is NULL, and will just insert a new leaf */
1109 node_set_parent((struct node *)l, tp);
1111 cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1112 put_child(t, (struct tnode *)tp, cindex, (struct node *)l);
1114 /* Case 3: n is a LEAF or a TNODE and the key doesn't match. */
1116 * Add a new tnode here
1117 * first tnode need some special handling
1121 pos = tp->pos+tp->bits;
1126 newpos = tkey_mismatch(key, pos, n->key);
1127 tn = tnode_new(n->key, newpos, 1);
1130 tn = tnode_new(key, newpos, 1); /* First tnode */
1139 node_set_parent((struct node *)tn, tp);
1141 missbit = tkey_extract_bits(key, newpos, 1);
1142 put_child(t, tn, missbit, (struct node *)l);
1143 put_child(t, tn, 1-missbit, n);
1146 cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1147 put_child(t, (struct tnode *)tp, cindex,
1150 rcu_assign_pointer(t->trie, (struct node *)tn);
1155 if (tp && tp->pos + tp->bits > 32)
1156 pr_warning("fib_trie"
1157 " tp=%p pos=%d, bits=%d, key=%0x plen=%d\n",
1158 tp, tp->pos, tp->bits, key, plen);
1160 /* Rebalance the trie */
1162 rcu_assign_pointer(t->trie, trie_rebalance(t, tp));
1168 * Caller must hold RTNL.
1170 static int fn_trie_insert(struct fib_table *tb, struct fib_config *cfg)
1172 struct trie *t = (struct trie *) tb->tb_data;
1173 struct fib_alias *fa, *new_fa;
1174 struct list_head *fa_head = NULL;
1175 struct fib_info *fi;
1176 int plen = cfg->fc_dst_len;
1177 u8 tos = cfg->fc_tos;
1185 key = ntohl(cfg->fc_dst);
1187 pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1189 mask = ntohl(inet_make_mask(plen));
1196 fi = fib_create_info(cfg);
1202 l = fib_find_node(t, key);
1206 fa_head = get_fa_head(l, plen);
1207 fa = fib_find_alias(fa_head, tos, fi->fib_priority);
1210 /* Now fa, if non-NULL, points to the first fib alias
1211 * with the same keys [prefix,tos,priority], if such key already
1212 * exists or to the node before which we will insert new one.
1214 * If fa is NULL, we will need to allocate a new one and
1215 * insert to the head of f.
1217 * If f is NULL, no fib node matched the destination key
1218 * and we need to allocate a new one of those as well.
1221 if (fa && fa->fa_tos == tos &&
1222 fa->fa_info->fib_priority == fi->fib_priority) {
1223 struct fib_alias *fa_first, *fa_match;
1226 if (cfg->fc_nlflags & NLM_F_EXCL)
1230 * 1. Find exact match for type, scope, fib_info to avoid
1232 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1236 fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
1237 list_for_each_entry_continue(fa, fa_head, fa_list) {
1238 if (fa->fa_tos != tos)
1240 if (fa->fa_info->fib_priority != fi->fib_priority)
1242 if (fa->fa_type == cfg->fc_type &&
1243 fa->fa_scope == cfg->fc_scope &&
1244 fa->fa_info == fi) {
1250 if (cfg->fc_nlflags & NLM_F_REPLACE) {
1251 struct fib_info *fi_drop;
1261 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1265 fi_drop = fa->fa_info;
1266 new_fa->fa_tos = fa->fa_tos;
1267 new_fa->fa_info = fi;
1268 new_fa->fa_type = cfg->fc_type;
1269 new_fa->fa_scope = cfg->fc_scope;
1270 state = fa->fa_state;
1271 new_fa->fa_state = state & ~FA_S_ACCESSED;
1273 list_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1274 alias_free_mem_rcu(fa);
1276 fib_release_info(fi_drop);
1277 if (state & FA_S_ACCESSED)
1278 rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1279 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1280 tb->tb_id, &cfg->fc_nlinfo, NLM_F_REPLACE);
1284 /* Error if we find a perfect match which
1285 * uses the same scope, type, and nexthop
1291 if (!(cfg->fc_nlflags & NLM_F_APPEND))
1295 if (!(cfg->fc_nlflags & NLM_F_CREATE))
1299 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1303 new_fa->fa_info = fi;
1304 new_fa->fa_tos = tos;
1305 new_fa->fa_type = cfg->fc_type;
1306 new_fa->fa_scope = cfg->fc_scope;
1307 new_fa->fa_state = 0;
1309 * Insert new entry to the list.
1313 fa_head = fib_insert_node(t, key, plen);
1314 if (unlikely(!fa_head)) {
1316 goto out_free_new_fa;
1320 list_add_tail_rcu(&new_fa->fa_list,
1321 (fa ? &fa->fa_list : fa_head));
1323 rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1324 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, tb->tb_id,
1325 &cfg->fc_nlinfo, 0);
1330 kmem_cache_free(fn_alias_kmem, new_fa);
1332 fib_release_info(fi);
1337 /* should be called with rcu_read_lock */
1338 static int check_leaf(struct trie *t, struct leaf *l,
1339 t_key key, const struct flowi *flp,
1340 struct fib_result *res)
1342 struct leaf_info *li;
1343 struct hlist_head *hhead = &l->list;
1344 struct hlist_node *node;
1346 hlist_for_each_entry_rcu(li, node, hhead, hlist) {
1348 int plen = li->plen;
1349 __be32 mask = inet_make_mask(plen);
1351 if (l->key != (key & ntohl(mask)))
1354 err = fib_semantic_match(&li->falh, flp, res, plen);
1356 #ifdef CONFIG_IP_FIB_TRIE_STATS
1358 t->stats.semantic_match_passed++;
1360 t->stats.semantic_match_miss++;
1369 static int fn_trie_lookup(struct fib_table *tb, const struct flowi *flp,
1370 struct fib_result *res)
1372 struct trie *t = (struct trie *) tb->tb_data;
1377 t_key key = ntohl(flp->fl4_dst);
1380 int current_prefix_length = KEYLENGTH;
1382 t_key node_prefix, key_prefix, pref_mismatch;
1387 n = rcu_dereference(t->trie);
1391 #ifdef CONFIG_IP_FIB_TRIE_STATS
1397 ret = check_leaf(t, (struct leaf *)n, key, flp, res);
1401 pn = (struct tnode *) n;
1409 cindex = tkey_extract_bits(mask_pfx(key, current_prefix_length),
1412 n = tnode_get_child(pn, cindex);
1415 #ifdef CONFIG_IP_FIB_TRIE_STATS
1416 t->stats.null_node_hit++;
1422 ret = check_leaf(t, (struct leaf *)n, key, flp, res);
1428 cn = (struct tnode *)n;
1431 * It's a tnode, and we can do some extra checks here if we
1432 * like, to avoid descending into a dead-end branch.
1433 * This tnode is in the parent's child array at index
1434 * key[p_pos..p_pos+p_bits] but potentially with some bits
1435 * chopped off, so in reality the index may be just a
1436 * subprefix, padded with zero at the end.
1437 * We can also take a look at any skipped bits in this
1438 * tnode - everything up to p_pos is supposed to be ok,
1439 * and the non-chopped bits of the index (se previous
1440 * paragraph) are also guaranteed ok, but the rest is
1441 * considered unknown.
1443 * The skipped bits are key[pos+bits..cn->pos].
1446 /* If current_prefix_length < pos+bits, we are already doing
1447 * actual prefix matching, which means everything from
1448 * pos+(bits-chopped_off) onward must be zero along some
1449 * branch of this subtree - otherwise there is *no* valid
1450 * prefix present. Here we can only check the skipped
1451 * bits. Remember, since we have already indexed into the
1452 * parent's child array, we know that the bits we chopped of
1456 /* NOTA BENE: Checking only skipped bits
1457 for the new node here */
1459 if (current_prefix_length < pos+bits) {
1460 if (tkey_extract_bits(cn->key, current_prefix_length,
1461 cn->pos - current_prefix_length)
1467 * If chopped_off=0, the index is fully validated and we
1468 * only need to look at the skipped bits for this, the new,
1469 * tnode. What we actually want to do is to find out if
1470 * these skipped bits match our key perfectly, or if we will
1471 * have to count on finding a matching prefix further down,
1472 * because if we do, we would like to have some way of
1473 * verifying the existence of such a prefix at this point.
1476 /* The only thing we can do at this point is to verify that
1477 * any such matching prefix can indeed be a prefix to our
1478 * key, and if the bits in the node we are inspecting that
1479 * do not match our key are not ZERO, this cannot be true.
1480 * Thus, find out where there is a mismatch (before cn->pos)
1481 * and verify that all the mismatching bits are zero in the
1486 * Note: We aren't very concerned about the piece of
1487 * the key that precede pn->pos+pn->bits, since these
1488 * have already been checked. The bits after cn->pos
1489 * aren't checked since these are by definition
1490 * "unknown" at this point. Thus, what we want to see
1491 * is if we are about to enter the "prefix matching"
1492 * state, and in that case verify that the skipped
1493 * bits that will prevail throughout this subtree are
1494 * zero, as they have to be if we are to find a
1498 node_prefix = mask_pfx(cn->key, cn->pos);
1499 key_prefix = mask_pfx(key, cn->pos);
1500 pref_mismatch = key_prefix^node_prefix;
1504 * In short: If skipped bits in this node do not match
1505 * the search key, enter the "prefix matching"
1508 if (pref_mismatch) {
1509 while (!(pref_mismatch & (1<<(KEYLENGTH-1)))) {
1511 pref_mismatch = pref_mismatch << 1;
1513 key_prefix = tkey_extract_bits(cn->key, mp, cn->pos-mp);
1515 if (key_prefix != 0)
1518 if (current_prefix_length >= cn->pos)
1519 current_prefix_length = mp;
1522 pn = (struct tnode *)n; /* Descend */
1529 /* As zero don't change the child key (cindex) */
1530 while ((chopped_off <= pn->bits)
1531 && !(cindex & (1<<(chopped_off-1))))
1534 /* Decrease current_... with bits chopped off */
1535 if (current_prefix_length > pn->pos + pn->bits - chopped_off)
1536 current_prefix_length = pn->pos + pn->bits
1540 * Either we do the actual chop off according or if we have
1541 * chopped off all bits in this tnode walk up to our parent.
1544 if (chopped_off <= pn->bits) {
1545 cindex &= ~(1 << (chopped_off-1));
1547 struct tnode *parent = node_parent((struct node *) pn);
1551 /* Get Child's index */
1552 cindex = tkey_extract_bits(pn->key, parent->pos, parent->bits);
1556 #ifdef CONFIG_IP_FIB_TRIE_STATS
1557 t->stats.backtrack++;
1570 * Remove the leaf and return parent.
1572 static void trie_leaf_remove(struct trie *t, struct leaf *l)
1574 struct tnode *tp = node_parent((struct node *) l);
1576 pr_debug("entering trie_leaf_remove(%p)\n", l);
1579 t_key cindex = tkey_extract_bits(l->key, tp->pos, tp->bits);
1580 put_child(t, (struct tnode *)tp, cindex, NULL);
1581 rcu_assign_pointer(t->trie, trie_rebalance(t, tp));
1583 rcu_assign_pointer(t->trie, NULL);
1589 * Caller must hold RTNL.
1591 static int fn_trie_delete(struct fib_table *tb, struct fib_config *cfg)
1593 struct trie *t = (struct trie *) tb->tb_data;
1595 int plen = cfg->fc_dst_len;
1596 u8 tos = cfg->fc_tos;
1597 struct fib_alias *fa, *fa_to_delete;
1598 struct list_head *fa_head;
1600 struct leaf_info *li;
1605 key = ntohl(cfg->fc_dst);
1606 mask = ntohl(inet_make_mask(plen));
1612 l = fib_find_node(t, key);
1617 fa_head = get_fa_head(l, plen);
1618 fa = fib_find_alias(fa_head, tos, 0);
1623 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1625 fa_to_delete = NULL;
1626 fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
1627 list_for_each_entry_continue(fa, fa_head, fa_list) {
1628 struct fib_info *fi = fa->fa_info;
1630 if (fa->fa_tos != tos)
1633 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1634 (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1635 fa->fa_scope == cfg->fc_scope) &&
1636 (!cfg->fc_protocol ||
1637 fi->fib_protocol == cfg->fc_protocol) &&
1638 fib_nh_match(cfg, fi) == 0) {
1648 rtmsg_fib(RTM_DELROUTE, htonl(key), fa, plen, tb->tb_id,
1649 &cfg->fc_nlinfo, 0);
1651 l = fib_find_node(t, key);
1652 li = find_leaf_info(l, plen);
1654 list_del_rcu(&fa->fa_list);
1656 if (list_empty(fa_head)) {
1657 hlist_del_rcu(&li->hlist);
1661 if (hlist_empty(&l->list))
1662 trie_leaf_remove(t, l);
1664 if (fa->fa_state & FA_S_ACCESSED)
1665 rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1667 fib_release_info(fa->fa_info);
1668 alias_free_mem_rcu(fa);
1672 static int trie_flush_list(struct list_head *head)
1674 struct fib_alias *fa, *fa_node;
1677 list_for_each_entry_safe(fa, fa_node, head, fa_list) {
1678 struct fib_info *fi = fa->fa_info;
1680 if (fi && (fi->fib_flags & RTNH_F_DEAD)) {
1681 list_del_rcu(&fa->fa_list);
1682 fib_release_info(fa->fa_info);
1683 alias_free_mem_rcu(fa);
1690 static int trie_flush_leaf(struct leaf *l)
1693 struct hlist_head *lih = &l->list;
1694 struct hlist_node *node, *tmp;
1695 struct leaf_info *li = NULL;
1697 hlist_for_each_entry_safe(li, node, tmp, lih, hlist) {
1698 found += trie_flush_list(&li->falh);
1700 if (list_empty(&li->falh)) {
1701 hlist_del_rcu(&li->hlist);
1709 * Scan for the next right leaf starting at node p->child[idx]
1710 * Since we have back pointer, no recursion necessary.
1712 static struct leaf *leaf_walk_rcu(struct tnode *p, struct node *c)
1718 idx = tkey_extract_bits(c->key, p->pos, p->bits) + 1;
1722 while (idx < 1u << p->bits) {
1723 c = tnode_get_child_rcu(p, idx++);
1728 prefetch(p->child[idx]);
1729 return (struct leaf *) c;
1732 /* Rescan start scanning in new node */
1733 p = (struct tnode *) c;
1737 /* Node empty, walk back up to parent */
1738 c = (struct node *) p;
1739 } while ( (p = node_parent_rcu(c)) != NULL);
1741 return NULL; /* Root of trie */
1744 static struct leaf *trie_firstleaf(struct trie *t)
1746 struct tnode *n = (struct tnode *) rcu_dereference(t->trie);
1751 if (IS_LEAF(n)) /* trie is just a leaf */
1752 return (struct leaf *) n;
1754 return leaf_walk_rcu(n, NULL);
1757 static struct leaf *trie_nextleaf(struct leaf *l)
1759 struct node *c = (struct node *) l;
1760 struct tnode *p = node_parent(c);
1763 return NULL; /* trie with just one leaf */
1765 return leaf_walk_rcu(p, c);
1768 static struct leaf *trie_leafindex(struct trie *t, int index)
1770 struct leaf *l = trie_firstleaf(t);
1772 while (l && index-- > 0)
1773 l = trie_nextleaf(l);
1780 * Caller must hold RTNL.
1782 static int fn_trie_flush(struct fib_table *tb)
1784 struct trie *t = (struct trie *) tb->tb_data;
1785 struct leaf *l, *ll = NULL;
1788 for (l = trie_firstleaf(t); l; l = trie_nextleaf(l)) {
1789 found += trie_flush_leaf(l);
1791 if (ll && hlist_empty(&ll->list))
1792 trie_leaf_remove(t, ll);
1796 if (ll && hlist_empty(&ll->list))
1797 trie_leaf_remove(t, ll);
1799 pr_debug("trie_flush found=%d\n", found);
1803 static void fn_trie_select_default(struct fib_table *tb,
1804 const struct flowi *flp,
1805 struct fib_result *res)
1807 struct trie *t = (struct trie *) tb->tb_data;
1808 int order, last_idx;
1809 struct fib_info *fi = NULL;
1810 struct fib_info *last_resort;
1811 struct fib_alias *fa = NULL;
1812 struct list_head *fa_head;
1821 l = fib_find_node(t, 0);
1825 fa_head = get_fa_head(l, 0);
1829 if (list_empty(fa_head))
1832 list_for_each_entry_rcu(fa, fa_head, fa_list) {
1833 struct fib_info *next_fi = fa->fa_info;
1835 if (fa->fa_scope != res->scope ||
1836 fa->fa_type != RTN_UNICAST)
1839 if (next_fi->fib_priority > res->fi->fib_priority)
1841 if (!next_fi->fib_nh[0].nh_gw ||
1842 next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1844 fa->fa_state |= FA_S_ACCESSED;
1847 if (next_fi != res->fi)
1849 } else if (!fib_detect_death(fi, order, &last_resort,
1850 &last_idx, tb->tb_default)) {
1851 fib_result_assign(res, fi);
1852 tb->tb_default = order;
1858 if (order <= 0 || fi == NULL) {
1859 tb->tb_default = -1;
1863 if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1865 fib_result_assign(res, fi);
1866 tb->tb_default = order;
1870 fib_result_assign(res, last_resort);
1871 tb->tb_default = last_idx;
1876 static int fn_trie_dump_fa(t_key key, int plen, struct list_head *fah,
1877 struct fib_table *tb,
1878 struct sk_buff *skb, struct netlink_callback *cb)
1881 struct fib_alias *fa;
1882 __be32 xkey = htonl(key);
1887 /* rcu_read_lock is hold by caller */
1889 list_for_each_entry_rcu(fa, fah, fa_list) {
1895 if (fib_dump_info(skb, NETLINK_CB(cb->skb).pid,
1904 fa->fa_info, NLM_F_MULTI) < 0) {
1914 static int fn_trie_dump_leaf(struct leaf *l, struct fib_table *tb,
1915 struct sk_buff *skb, struct netlink_callback *cb)
1917 struct leaf_info *li;
1918 struct hlist_node *node;
1924 /* rcu_read_lock is hold by caller */
1925 hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
1934 if (list_empty(&li->falh))
1937 if (fn_trie_dump_fa(l->key, li->plen, &li->falh, tb, skb, cb) < 0) {
1948 static int fn_trie_dump(struct fib_table *tb, struct sk_buff *skb,
1949 struct netlink_callback *cb)
1952 struct trie *t = (struct trie *) tb->tb_data;
1953 t_key key = cb->args[2];
1954 int count = cb->args[3];
1957 /* Dump starting at last key.
1958 * Note: 0.0.0.0/0 (ie default) is first key.
1961 l = trie_firstleaf(t);
1963 /* Normally, continue from last key, but if that is missing
1964 * fallback to using slow rescan
1966 l = fib_find_node(t, key);
1968 l = trie_leafindex(t, count);
1972 cb->args[2] = l->key;
1973 if (fn_trie_dump_leaf(l, tb, skb, cb) < 0) {
1974 cb->args[3] = count;
1980 l = trie_nextleaf(l);
1981 memset(&cb->args[4], 0,
1982 sizeof(cb->args) - 4*sizeof(cb->args[0]));
1984 cb->args[3] = count;
1990 void __init fib_hash_init(void)
1992 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
1993 sizeof(struct fib_alias),
1994 0, SLAB_PANIC, NULL);
1996 trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
1997 max(sizeof(struct leaf),
1998 sizeof(struct leaf_info)),
1999 0, SLAB_PANIC, NULL);
2003 /* Fix more generic FIB names for init later */
2004 struct fib_table *fib_hash_table(u32 id)
2006 struct fib_table *tb;
2009 tb = kmalloc(sizeof(struct fib_table) + sizeof(struct trie),
2015 tb->tb_default = -1;
2016 tb->tb_lookup = fn_trie_lookup;
2017 tb->tb_insert = fn_trie_insert;
2018 tb->tb_delete = fn_trie_delete;
2019 tb->tb_flush = fn_trie_flush;
2020 tb->tb_select_default = fn_trie_select_default;
2021 tb->tb_dump = fn_trie_dump;
2023 t = (struct trie *) tb->tb_data;
2024 memset(t, 0, sizeof(*t));
2026 if (id == RT_TABLE_LOCAL)
2027 pr_info("IPv4 FIB: Using LC-trie version %s\n", VERSION);
2032 #ifdef CONFIG_PROC_FS
2033 /* Depth first Trie walk iterator */
2034 struct fib_trie_iter {
2035 struct seq_net_private p;
2036 struct fib_table *tb;
2037 struct tnode *tnode;
2042 static struct node *fib_trie_get_next(struct fib_trie_iter *iter)
2044 struct tnode *tn = iter->tnode;
2045 unsigned cindex = iter->index;
2048 /* A single entry routing table */
2052 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2053 iter->tnode, iter->index, iter->depth);
2055 while (cindex < (1<<tn->bits)) {
2056 struct node *n = tnode_get_child_rcu(tn, cindex);
2061 iter->index = cindex + 1;
2063 /* push down one level */
2064 iter->tnode = (struct tnode *) n;
2074 /* Current node exhausted, pop back up */
2075 p = node_parent_rcu((struct node *)tn);
2077 cindex = tkey_extract_bits(tn->key, p->pos, p->bits)+1;
2087 static struct node *fib_trie_get_first(struct fib_trie_iter *iter,
2095 n = rcu_dereference(t->trie);
2100 iter->tnode = (struct tnode *) n;
2112 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2115 struct fib_trie_iter iter;
2117 memset(s, 0, sizeof(*s));
2120 for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2122 struct leaf *l = (struct leaf *)n;
2123 struct leaf_info *li;
2124 struct hlist_node *tmp;
2127 s->totdepth += iter.depth;
2128 if (iter.depth > s->maxdepth)
2129 s->maxdepth = iter.depth;
2131 hlist_for_each_entry_rcu(li, tmp, &l->list, hlist)
2134 const struct tnode *tn = (const struct tnode *) n;
2138 if (tn->bits < MAX_STAT_DEPTH)
2139 s->nodesizes[tn->bits]++;
2141 for (i = 0; i < (1<<tn->bits); i++)
2150 * This outputs /proc/net/fib_triestats
2152 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2154 unsigned i, max, pointers, bytes, avdepth;
2157 avdepth = stat->totdepth*100 / stat->leaves;
2161 seq_printf(seq, "\tAver depth: %u.%02d\n",
2162 avdepth / 100, avdepth % 100);
2163 seq_printf(seq, "\tMax depth: %u\n", stat->maxdepth);
2165 seq_printf(seq, "\tLeaves: %u\n", stat->leaves);
2166 bytes = sizeof(struct leaf) * stat->leaves;
2168 seq_printf(seq, "\tPrefixes: %u\n", stat->prefixes);
2169 bytes += sizeof(struct leaf_info) * stat->prefixes;
2171 seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2172 bytes += sizeof(struct tnode) * stat->tnodes;
2174 max = MAX_STAT_DEPTH;
2175 while (max > 0 && stat->nodesizes[max-1] == 0)
2179 for (i = 1; i <= max; i++)
2180 if (stat->nodesizes[i] != 0) {
2181 seq_printf(seq, " %u: %u", i, stat->nodesizes[i]);
2182 pointers += (1<<i) * stat->nodesizes[i];
2184 seq_putc(seq, '\n');
2185 seq_printf(seq, "\tPointers: %u\n", pointers);
2187 bytes += sizeof(struct node *) * pointers;
2188 seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2189 seq_printf(seq, "Total size: %u kB\n", (bytes + 1023) / 1024);
2192 #ifdef CONFIG_IP_FIB_TRIE_STATS
2193 static void trie_show_usage(struct seq_file *seq,
2194 const struct trie_use_stats *stats)
2196 seq_printf(seq, "\nCounters:\n---------\n");
2197 seq_printf(seq, "gets = %u\n", stats->gets);
2198 seq_printf(seq, "backtracks = %u\n", stats->backtrack);
2199 seq_printf(seq, "semantic match passed = %u\n",
2200 stats->semantic_match_passed);
2201 seq_printf(seq, "semantic match miss = %u\n",
2202 stats->semantic_match_miss);
2203 seq_printf(seq, "null node hit= %u\n", stats->null_node_hit);
2204 seq_printf(seq, "skipped node resize = %u\n\n",
2205 stats->resize_node_skipped);
2207 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2209 static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2211 if (tb->tb_id == RT_TABLE_LOCAL)
2212 seq_puts(seq, "Local:\n");
2213 else if (tb->tb_id == RT_TABLE_MAIN)
2214 seq_puts(seq, "Main:\n");
2216 seq_printf(seq, "Id %d:\n", tb->tb_id);
2220 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2222 struct net *net = (struct net *)seq->private;
2226 "Basic info: size of leaf:"
2227 " %Zd bytes, size of tnode: %Zd bytes.\n",
2228 sizeof(struct leaf), sizeof(struct tnode));
2230 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2231 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2232 struct hlist_node *node;
2233 struct fib_table *tb;
2235 hlist_for_each_entry_rcu(tb, node, head, tb_hlist) {
2236 struct trie *t = (struct trie *) tb->tb_data;
2237 struct trie_stat stat;
2242 fib_table_print(seq, tb);
2244 trie_collect_stats(t, &stat);
2245 trie_show_stats(seq, &stat);
2246 #ifdef CONFIG_IP_FIB_TRIE_STATS
2247 trie_show_usage(seq, &t->stats);
2255 static int fib_triestat_seq_open(struct inode *inode, struct file *file)
2257 return single_open_net(inode, file, fib_triestat_seq_show);
2260 static const struct file_operations fib_triestat_fops = {
2261 .owner = THIS_MODULE,
2262 .open = fib_triestat_seq_open,
2264 .llseek = seq_lseek,
2265 .release = single_release_net,
2268 static struct node *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2270 struct fib_trie_iter *iter = seq->private;
2271 struct net *net = seq_file_net(seq);
2275 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2276 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2277 struct hlist_node *node;
2278 struct fib_table *tb;
2280 hlist_for_each_entry_rcu(tb, node, head, tb_hlist) {
2283 for (n = fib_trie_get_first(iter,
2284 (struct trie *) tb->tb_data);
2285 n; n = fib_trie_get_next(iter))
2296 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2300 return fib_trie_get_idx(seq, *pos);
2303 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2305 struct fib_trie_iter *iter = seq->private;
2306 struct net *net = seq_file_net(seq);
2307 struct fib_table *tb = iter->tb;
2308 struct hlist_node *tb_node;
2313 /* next node in same table */
2314 n = fib_trie_get_next(iter);
2318 /* walk rest of this hash chain */
2319 h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2320 while ( (tb_node = rcu_dereference(tb->tb_hlist.next)) ) {
2321 tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2322 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2327 /* new hash chain */
2328 while (++h < FIB_TABLE_HASHSZ) {
2329 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2330 hlist_for_each_entry_rcu(tb, tb_node, head, tb_hlist) {
2331 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2343 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2349 static void seq_indent(struct seq_file *seq, int n)
2351 while (n-- > 0) seq_puts(seq, " ");
2354 static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2357 case RT_SCOPE_UNIVERSE: return "universe";
2358 case RT_SCOPE_SITE: return "site";
2359 case RT_SCOPE_LINK: return "link";
2360 case RT_SCOPE_HOST: return "host";
2361 case RT_SCOPE_NOWHERE: return "nowhere";
2363 snprintf(buf, len, "scope=%d", s);
2368 static const char *rtn_type_names[__RTN_MAX] = {
2369 [RTN_UNSPEC] = "UNSPEC",
2370 [RTN_UNICAST] = "UNICAST",
2371 [RTN_LOCAL] = "LOCAL",
2372 [RTN_BROADCAST] = "BROADCAST",
2373 [RTN_ANYCAST] = "ANYCAST",
2374 [RTN_MULTICAST] = "MULTICAST",
2375 [RTN_BLACKHOLE] = "BLACKHOLE",
2376 [RTN_UNREACHABLE] = "UNREACHABLE",
2377 [RTN_PROHIBIT] = "PROHIBIT",
2378 [RTN_THROW] = "THROW",
2380 [RTN_XRESOLVE] = "XRESOLVE",
2383 static inline const char *rtn_type(char *buf, size_t len, unsigned t)
2385 if (t < __RTN_MAX && rtn_type_names[t])
2386 return rtn_type_names[t];
2387 snprintf(buf, len, "type %u", t);
2391 /* Pretty print the trie */
2392 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2394 const struct fib_trie_iter *iter = seq->private;
2397 if (!node_parent_rcu(n))
2398 fib_table_print(seq, iter->tb);
2401 struct tnode *tn = (struct tnode *) n;
2402 __be32 prf = htonl(mask_pfx(tn->key, tn->pos));
2404 seq_indent(seq, iter->depth-1);
2405 seq_printf(seq, " +-- %pI4/%d %d %d %d\n",
2406 &prf, tn->pos, tn->bits, tn->full_children,
2407 tn->empty_children);
2410 struct leaf *l = (struct leaf *) n;
2411 struct leaf_info *li;
2412 struct hlist_node *node;
2413 __be32 val = htonl(l->key);
2415 seq_indent(seq, iter->depth);
2416 seq_printf(seq, " |-- %pI4\n", &val);
2418 hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
2419 struct fib_alias *fa;
2421 list_for_each_entry_rcu(fa, &li->falh, fa_list) {
2422 char buf1[32], buf2[32];
2424 seq_indent(seq, iter->depth+1);
2425 seq_printf(seq, " /%d %s %s", li->plen,
2426 rtn_scope(buf1, sizeof(buf1),
2428 rtn_type(buf2, sizeof(buf2),
2431 seq_printf(seq, " tos=%d", fa->fa_tos);
2432 seq_putc(seq, '\n');
2440 static const struct seq_operations fib_trie_seq_ops = {
2441 .start = fib_trie_seq_start,
2442 .next = fib_trie_seq_next,
2443 .stop = fib_trie_seq_stop,
2444 .show = fib_trie_seq_show,
2447 static int fib_trie_seq_open(struct inode *inode, struct file *file)
2449 return seq_open_net(inode, file, &fib_trie_seq_ops,
2450 sizeof(struct fib_trie_iter));
2453 static const struct file_operations fib_trie_fops = {
2454 .owner = THIS_MODULE,
2455 .open = fib_trie_seq_open,
2457 .llseek = seq_lseek,
2458 .release = seq_release_net,
2461 struct fib_route_iter {
2462 struct seq_net_private p;
2463 struct trie *main_trie;
2468 static struct leaf *fib_route_get_idx(struct fib_route_iter *iter, loff_t pos)
2470 struct leaf *l = NULL;
2471 struct trie *t = iter->main_trie;
2473 /* use cache location of last found key */
2474 if (iter->pos > 0 && pos >= iter->pos && (l = fib_find_node(t, iter->key)))
2478 l = trie_firstleaf(t);
2481 while (l && pos-- > 0) {
2483 l = trie_nextleaf(l);
2487 iter->key = pos; /* remember it */
2489 iter->pos = 0; /* forget it */
2494 static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2497 struct fib_route_iter *iter = seq->private;
2498 struct fib_table *tb;
2501 tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2505 iter->main_trie = (struct trie *) tb->tb_data;
2507 return SEQ_START_TOKEN;
2509 return fib_route_get_idx(iter, *pos - 1);
2512 static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2514 struct fib_route_iter *iter = seq->private;
2518 if (v == SEQ_START_TOKEN) {
2520 l = trie_firstleaf(iter->main_trie);
2523 l = trie_nextleaf(l);
2533 static void fib_route_seq_stop(struct seq_file *seq, void *v)
2539 static unsigned fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2541 static unsigned type2flags[RTN_MAX + 1] = {
2542 [7] = RTF_REJECT, [8] = RTF_REJECT,
2544 unsigned flags = type2flags[type];
2546 if (fi && fi->fib_nh->nh_gw)
2547 flags |= RTF_GATEWAY;
2548 if (mask == htonl(0xFFFFFFFF))
2555 * This outputs /proc/net/route.
2556 * The format of the file is not supposed to be changed
2557 * and needs to be same as fib_hash output to avoid breaking
2560 static int fib_route_seq_show(struct seq_file *seq, void *v)
2563 struct leaf_info *li;
2564 struct hlist_node *node;
2566 if (v == SEQ_START_TOKEN) {
2567 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2568 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2573 hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
2574 struct fib_alias *fa;
2575 __be32 mask, prefix;
2577 mask = inet_make_mask(li->plen);
2578 prefix = htonl(l->key);
2580 list_for_each_entry_rcu(fa, &li->falh, fa_list) {
2581 const struct fib_info *fi = fa->fa_info;
2582 unsigned flags = fib_flag_trans(fa->fa_type, mask, fi);
2585 if (fa->fa_type == RTN_BROADCAST
2586 || fa->fa_type == RTN_MULTICAST)
2591 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2592 "%d\t%08X\t%d\t%u\t%u%n",
2593 fi->fib_dev ? fi->fib_dev->name : "*",
2595 fi->fib_nh->nh_gw, flags, 0, 0,
2599 fi->fib_advmss + 40 : 0),
2601 fi->fib_rtt >> 3, &len);
2604 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2605 "%d\t%08X\t%d\t%u\t%u%n",
2606 prefix, 0, flags, 0, 0, 0,
2607 mask, 0, 0, 0, &len);
2609 seq_printf(seq, "%*s\n", 127 - len, "");
2616 static const struct seq_operations fib_route_seq_ops = {
2617 .start = fib_route_seq_start,
2618 .next = fib_route_seq_next,
2619 .stop = fib_route_seq_stop,
2620 .show = fib_route_seq_show,
2623 static int fib_route_seq_open(struct inode *inode, struct file *file)
2625 return seq_open_net(inode, file, &fib_route_seq_ops,
2626 sizeof(struct fib_route_iter));
2629 static const struct file_operations fib_route_fops = {
2630 .owner = THIS_MODULE,
2631 .open = fib_route_seq_open,
2633 .llseek = seq_lseek,
2634 .release = seq_release_net,
2637 int __net_init fib_proc_init(struct net *net)
2639 if (!proc_net_fops_create(net, "fib_trie", S_IRUGO, &fib_trie_fops))
2642 if (!proc_net_fops_create(net, "fib_triestat", S_IRUGO,
2643 &fib_triestat_fops))
2646 if (!proc_net_fops_create(net, "route", S_IRUGO, &fib_route_fops))
2652 proc_net_remove(net, "fib_triestat");
2654 proc_net_remove(net, "fib_trie");
2659 void __net_exit fib_proc_exit(struct net *net)
2661 proc_net_remove(net, "fib_trie");
2662 proc_net_remove(net, "fib_triestat");
2663 proc_net_remove(net, "route");
2666 #endif /* CONFIG_PROC_FS */