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)
989 t_key cindex, key = tn->key;
992 while (tn != NULL && (tp = node_parent((struct node *)tn)) != NULL) {
993 cindex = tkey_extract_bits(key, tp->pos, tp->bits);
994 wasfull = tnode_full(tp, tnode_get_child(tp, cindex));
995 tn = (struct tnode *) resize(t, (struct tnode *)tn);
997 tnode_put_child_reorg((struct tnode *)tp, cindex,
998 (struct node *)tn, wasfull);
1000 tp = node_parent((struct node *) tn);
1006 /* Handle last (top) tnode */
1008 tn = (struct tnode *)resize(t, (struct tnode *)tn);
1010 return (struct node *)tn;
1013 /* only used from updater-side */
1015 static struct list_head *fib_insert_node(struct trie *t, u32 key, int plen)
1018 struct tnode *tp = NULL, *tn = NULL;
1022 struct list_head *fa_head = NULL;
1023 struct leaf_info *li;
1029 /* If we point to NULL, stop. Either the tree is empty and we should
1030 * just put a new leaf in if, or we have reached an empty child slot,
1031 * and we should just put our new leaf in that.
1032 * If we point to a T_TNODE, check if it matches our key. Note that
1033 * a T_TNODE might be skipping any number of bits - its 'pos' need
1034 * not be the parent's 'pos'+'bits'!
1036 * If it does match the current key, get pos/bits from it, extract
1037 * the index from our key, push the T_TNODE and walk the tree.
1039 * If it doesn't, we have to replace it with a new T_TNODE.
1041 * If we point to a T_LEAF, it might or might not have the same key
1042 * as we do. If it does, just change the value, update the T_LEAF's
1043 * value, and return it.
1044 * If it doesn't, we need to replace it with a T_TNODE.
1047 while (n != NULL && NODE_TYPE(n) == T_TNODE) {
1048 tn = (struct tnode *) n;
1052 if (tkey_sub_equals(tn->key, pos, tn->pos-pos, key)) {
1054 pos = tn->pos + tn->bits;
1055 n = tnode_get_child(tn,
1056 tkey_extract_bits(key,
1060 BUG_ON(n && node_parent(n) != tn);
1066 * n ----> NULL, LEAF or TNODE
1068 * tp is n's (parent) ----> NULL or TNODE
1071 BUG_ON(tp && IS_LEAF(tp));
1073 /* Case 1: n is a leaf. Compare prefixes */
1075 if (n != NULL && IS_LEAF(n) && tkey_equals(key, n->key)) {
1076 l = (struct leaf *) n;
1077 li = leaf_info_new(plen);
1082 fa_head = &li->falh;
1083 insert_leaf_info(&l->list, li);
1092 li = leaf_info_new(plen);
1099 fa_head = &li->falh;
1100 insert_leaf_info(&l->list, li);
1102 if (t->trie && n == NULL) {
1103 /* Case 2: n is NULL, and will just insert a new leaf */
1105 node_set_parent((struct node *)l, tp);
1107 cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1108 put_child(t, (struct tnode *)tp, cindex, (struct node *)l);
1110 /* Case 3: n is a LEAF or a TNODE and the key doesn't match. */
1112 * Add a new tnode here
1113 * first tnode need some special handling
1117 pos = tp->pos+tp->bits;
1122 newpos = tkey_mismatch(key, pos, n->key);
1123 tn = tnode_new(n->key, newpos, 1);
1126 tn = tnode_new(key, newpos, 1); /* First tnode */
1135 node_set_parent((struct node *)tn, tp);
1137 missbit = tkey_extract_bits(key, newpos, 1);
1138 put_child(t, tn, missbit, (struct node *)l);
1139 put_child(t, tn, 1-missbit, n);
1142 cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1143 put_child(t, (struct tnode *)tp, cindex,
1146 rcu_assign_pointer(t->trie, (struct node *)tn);
1151 if (tp && tp->pos + tp->bits > 32)
1152 pr_warning("fib_trie"
1153 " tp=%p pos=%d, bits=%d, key=%0x plen=%d\n",
1154 tp, tp->pos, tp->bits, key, plen);
1156 /* Rebalance the trie */
1158 rcu_assign_pointer(t->trie, trie_rebalance(t, tp));
1164 * Caller must hold RTNL.
1166 static int fn_trie_insert(struct fib_table *tb, struct fib_config *cfg)
1168 struct trie *t = (struct trie *) tb->tb_data;
1169 struct fib_alias *fa, *new_fa;
1170 struct list_head *fa_head = NULL;
1171 struct fib_info *fi;
1172 int plen = cfg->fc_dst_len;
1173 u8 tos = cfg->fc_tos;
1181 key = ntohl(cfg->fc_dst);
1183 pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1185 mask = ntohl(inet_make_mask(plen));
1192 fi = fib_create_info(cfg);
1198 l = fib_find_node(t, key);
1202 fa_head = get_fa_head(l, plen);
1203 fa = fib_find_alias(fa_head, tos, fi->fib_priority);
1206 /* Now fa, if non-NULL, points to the first fib alias
1207 * with the same keys [prefix,tos,priority], if such key already
1208 * exists or to the node before which we will insert new one.
1210 * If fa is NULL, we will need to allocate a new one and
1211 * insert to the head of f.
1213 * If f is NULL, no fib node matched the destination key
1214 * and we need to allocate a new one of those as well.
1217 if (fa && fa->fa_tos == tos &&
1218 fa->fa_info->fib_priority == fi->fib_priority) {
1219 struct fib_alias *fa_first, *fa_match;
1222 if (cfg->fc_nlflags & NLM_F_EXCL)
1226 * 1. Find exact match for type, scope, fib_info to avoid
1228 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1232 fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
1233 list_for_each_entry_continue(fa, fa_head, fa_list) {
1234 if (fa->fa_tos != tos)
1236 if (fa->fa_info->fib_priority != fi->fib_priority)
1238 if (fa->fa_type == cfg->fc_type &&
1239 fa->fa_scope == cfg->fc_scope &&
1240 fa->fa_info == fi) {
1246 if (cfg->fc_nlflags & NLM_F_REPLACE) {
1247 struct fib_info *fi_drop;
1257 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1261 fi_drop = fa->fa_info;
1262 new_fa->fa_tos = fa->fa_tos;
1263 new_fa->fa_info = fi;
1264 new_fa->fa_type = cfg->fc_type;
1265 new_fa->fa_scope = cfg->fc_scope;
1266 state = fa->fa_state;
1267 new_fa->fa_state = state & ~FA_S_ACCESSED;
1269 list_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1270 alias_free_mem_rcu(fa);
1272 fib_release_info(fi_drop);
1273 if (state & FA_S_ACCESSED)
1274 rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1275 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1276 tb->tb_id, &cfg->fc_nlinfo, NLM_F_REPLACE);
1280 /* Error if we find a perfect match which
1281 * uses the same scope, type, and nexthop
1287 if (!(cfg->fc_nlflags & NLM_F_APPEND))
1291 if (!(cfg->fc_nlflags & NLM_F_CREATE))
1295 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1299 new_fa->fa_info = fi;
1300 new_fa->fa_tos = tos;
1301 new_fa->fa_type = cfg->fc_type;
1302 new_fa->fa_scope = cfg->fc_scope;
1303 new_fa->fa_state = 0;
1305 * Insert new entry to the list.
1309 fa_head = fib_insert_node(t, key, plen);
1310 if (unlikely(!fa_head)) {
1312 goto out_free_new_fa;
1316 list_add_tail_rcu(&new_fa->fa_list,
1317 (fa ? &fa->fa_list : fa_head));
1319 rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1320 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, tb->tb_id,
1321 &cfg->fc_nlinfo, 0);
1326 kmem_cache_free(fn_alias_kmem, new_fa);
1328 fib_release_info(fi);
1333 /* should be called with rcu_read_lock */
1334 static int check_leaf(struct trie *t, struct leaf *l,
1335 t_key key, const struct flowi *flp,
1336 struct fib_result *res)
1338 struct leaf_info *li;
1339 struct hlist_head *hhead = &l->list;
1340 struct hlist_node *node;
1342 hlist_for_each_entry_rcu(li, node, hhead, hlist) {
1344 int plen = li->plen;
1345 __be32 mask = inet_make_mask(plen);
1347 if (l->key != (key & ntohl(mask)))
1350 err = fib_semantic_match(&li->falh, flp, res,
1351 htonl(l->key), mask, plen);
1353 #ifdef CONFIG_IP_FIB_TRIE_STATS
1355 t->stats.semantic_match_passed++;
1357 t->stats.semantic_match_miss++;
1366 static int fn_trie_lookup(struct fib_table *tb, const struct flowi *flp,
1367 struct fib_result *res)
1369 struct trie *t = (struct trie *) tb->tb_data;
1374 t_key key = ntohl(flp->fl4_dst);
1377 int current_prefix_length = KEYLENGTH;
1379 t_key node_prefix, key_prefix, pref_mismatch;
1384 n = rcu_dereference(t->trie);
1388 #ifdef CONFIG_IP_FIB_TRIE_STATS
1394 ret = check_leaf(t, (struct leaf *)n, key, flp, res);
1398 pn = (struct tnode *) n;
1406 cindex = tkey_extract_bits(mask_pfx(key, current_prefix_length),
1409 n = tnode_get_child(pn, cindex);
1412 #ifdef CONFIG_IP_FIB_TRIE_STATS
1413 t->stats.null_node_hit++;
1419 ret = check_leaf(t, (struct leaf *)n, key, flp, res);
1425 cn = (struct tnode *)n;
1428 * It's a tnode, and we can do some extra checks here if we
1429 * like, to avoid descending into a dead-end branch.
1430 * This tnode is in the parent's child array at index
1431 * key[p_pos..p_pos+p_bits] but potentially with some bits
1432 * chopped off, so in reality the index may be just a
1433 * subprefix, padded with zero at the end.
1434 * We can also take a look at any skipped bits in this
1435 * tnode - everything up to p_pos is supposed to be ok,
1436 * and the non-chopped bits of the index (se previous
1437 * paragraph) are also guaranteed ok, but the rest is
1438 * considered unknown.
1440 * The skipped bits are key[pos+bits..cn->pos].
1443 /* If current_prefix_length < pos+bits, we are already doing
1444 * actual prefix matching, which means everything from
1445 * pos+(bits-chopped_off) onward must be zero along some
1446 * branch of this subtree - otherwise there is *no* valid
1447 * prefix present. Here we can only check the skipped
1448 * bits. Remember, since we have already indexed into the
1449 * parent's child array, we know that the bits we chopped of
1453 /* NOTA BENE: Checking only skipped bits
1454 for the new node here */
1456 if (current_prefix_length < pos+bits) {
1457 if (tkey_extract_bits(cn->key, current_prefix_length,
1458 cn->pos - current_prefix_length)
1464 * If chopped_off=0, the index is fully validated and we
1465 * only need to look at the skipped bits for this, the new,
1466 * tnode. What we actually want to do is to find out if
1467 * these skipped bits match our key perfectly, or if we will
1468 * have to count on finding a matching prefix further down,
1469 * because if we do, we would like to have some way of
1470 * verifying the existence of such a prefix at this point.
1473 /* The only thing we can do at this point is to verify that
1474 * any such matching prefix can indeed be a prefix to our
1475 * key, and if the bits in the node we are inspecting that
1476 * do not match our key are not ZERO, this cannot be true.
1477 * Thus, find out where there is a mismatch (before cn->pos)
1478 * and verify that all the mismatching bits are zero in the
1483 * Note: We aren't very concerned about the piece of
1484 * the key that precede pn->pos+pn->bits, since these
1485 * have already been checked. The bits after cn->pos
1486 * aren't checked since these are by definition
1487 * "unknown" at this point. Thus, what we want to see
1488 * is if we are about to enter the "prefix matching"
1489 * state, and in that case verify that the skipped
1490 * bits that will prevail throughout this subtree are
1491 * zero, as they have to be if we are to find a
1495 node_prefix = mask_pfx(cn->key, cn->pos);
1496 key_prefix = mask_pfx(key, cn->pos);
1497 pref_mismatch = key_prefix^node_prefix;
1501 * In short: If skipped bits in this node do not match
1502 * the search key, enter the "prefix matching"
1505 if (pref_mismatch) {
1506 while (!(pref_mismatch & (1<<(KEYLENGTH-1)))) {
1508 pref_mismatch = pref_mismatch << 1;
1510 key_prefix = tkey_extract_bits(cn->key, mp, cn->pos-mp);
1512 if (key_prefix != 0)
1515 if (current_prefix_length >= cn->pos)
1516 current_prefix_length = mp;
1519 pn = (struct tnode *)n; /* Descend */
1526 /* As zero don't change the child key (cindex) */
1527 while ((chopped_off <= pn->bits)
1528 && !(cindex & (1<<(chopped_off-1))))
1531 /* Decrease current_... with bits chopped off */
1532 if (current_prefix_length > pn->pos + pn->bits - chopped_off)
1533 current_prefix_length = pn->pos + pn->bits
1537 * Either we do the actual chop off according or if we have
1538 * chopped off all bits in this tnode walk up to our parent.
1541 if (chopped_off <= pn->bits) {
1542 cindex &= ~(1 << (chopped_off-1));
1544 struct tnode *parent = node_parent((struct node *) pn);
1548 /* Get Child's index */
1549 cindex = tkey_extract_bits(pn->key, parent->pos, parent->bits);
1553 #ifdef CONFIG_IP_FIB_TRIE_STATS
1554 t->stats.backtrack++;
1567 * Remove the leaf and return parent.
1569 static void trie_leaf_remove(struct trie *t, struct leaf *l)
1571 struct tnode *tp = node_parent((struct node *) l);
1573 pr_debug("entering trie_leaf_remove(%p)\n", l);
1576 t_key cindex = tkey_extract_bits(l->key, tp->pos, tp->bits);
1577 put_child(t, (struct tnode *)tp, cindex, NULL);
1578 rcu_assign_pointer(t->trie, trie_rebalance(t, tp));
1580 rcu_assign_pointer(t->trie, NULL);
1586 * Caller must hold RTNL.
1588 static int fn_trie_delete(struct fib_table *tb, struct fib_config *cfg)
1590 struct trie *t = (struct trie *) tb->tb_data;
1592 int plen = cfg->fc_dst_len;
1593 u8 tos = cfg->fc_tos;
1594 struct fib_alias *fa, *fa_to_delete;
1595 struct list_head *fa_head;
1597 struct leaf_info *li;
1602 key = ntohl(cfg->fc_dst);
1603 mask = ntohl(inet_make_mask(plen));
1609 l = fib_find_node(t, key);
1614 fa_head = get_fa_head(l, plen);
1615 fa = fib_find_alias(fa_head, tos, 0);
1620 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1622 fa_to_delete = NULL;
1623 fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
1624 list_for_each_entry_continue(fa, fa_head, fa_list) {
1625 struct fib_info *fi = fa->fa_info;
1627 if (fa->fa_tos != tos)
1630 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1631 (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1632 fa->fa_scope == cfg->fc_scope) &&
1633 (!cfg->fc_protocol ||
1634 fi->fib_protocol == cfg->fc_protocol) &&
1635 fib_nh_match(cfg, fi) == 0) {
1645 rtmsg_fib(RTM_DELROUTE, htonl(key), fa, plen, tb->tb_id,
1646 &cfg->fc_nlinfo, 0);
1648 l = fib_find_node(t, key);
1649 li = find_leaf_info(l, plen);
1651 list_del_rcu(&fa->fa_list);
1653 if (list_empty(fa_head)) {
1654 hlist_del_rcu(&li->hlist);
1658 if (hlist_empty(&l->list))
1659 trie_leaf_remove(t, l);
1661 if (fa->fa_state & FA_S_ACCESSED)
1662 rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1664 fib_release_info(fa->fa_info);
1665 alias_free_mem_rcu(fa);
1669 static int trie_flush_list(struct list_head *head)
1671 struct fib_alias *fa, *fa_node;
1674 list_for_each_entry_safe(fa, fa_node, head, fa_list) {
1675 struct fib_info *fi = fa->fa_info;
1677 if (fi && (fi->fib_flags & RTNH_F_DEAD)) {
1678 list_del_rcu(&fa->fa_list);
1679 fib_release_info(fa->fa_info);
1680 alias_free_mem_rcu(fa);
1687 static int trie_flush_leaf(struct leaf *l)
1690 struct hlist_head *lih = &l->list;
1691 struct hlist_node *node, *tmp;
1692 struct leaf_info *li = NULL;
1694 hlist_for_each_entry_safe(li, node, tmp, lih, hlist) {
1695 found += trie_flush_list(&li->falh);
1697 if (list_empty(&li->falh)) {
1698 hlist_del_rcu(&li->hlist);
1706 * Scan for the next right leaf starting at node p->child[idx]
1707 * Since we have back pointer, no recursion necessary.
1709 static struct leaf *leaf_walk_rcu(struct tnode *p, struct node *c)
1715 idx = tkey_extract_bits(c->key, p->pos, p->bits) + 1;
1719 while (idx < 1u << p->bits) {
1720 c = tnode_get_child_rcu(p, idx++);
1725 prefetch(p->child[idx]);
1726 return (struct leaf *) c;
1729 /* Rescan start scanning in new node */
1730 p = (struct tnode *) c;
1734 /* Node empty, walk back up to parent */
1735 c = (struct node *) p;
1736 } while ( (p = node_parent_rcu(c)) != NULL);
1738 return NULL; /* Root of trie */
1741 static struct leaf *trie_firstleaf(struct trie *t)
1743 struct tnode *n = (struct tnode *) rcu_dereference(t->trie);
1748 if (IS_LEAF(n)) /* trie is just a leaf */
1749 return (struct leaf *) n;
1751 return leaf_walk_rcu(n, NULL);
1754 static struct leaf *trie_nextleaf(struct leaf *l)
1756 struct node *c = (struct node *) l;
1757 struct tnode *p = node_parent(c);
1760 return NULL; /* trie with just one leaf */
1762 return leaf_walk_rcu(p, c);
1765 static struct leaf *trie_leafindex(struct trie *t, int index)
1767 struct leaf *l = trie_firstleaf(t);
1769 while (l && index-- > 0)
1770 l = trie_nextleaf(l);
1777 * Caller must hold RTNL.
1779 static int fn_trie_flush(struct fib_table *tb)
1781 struct trie *t = (struct trie *) tb->tb_data;
1782 struct leaf *l, *ll = NULL;
1785 for (l = trie_firstleaf(t); l; l = trie_nextleaf(l)) {
1786 found += trie_flush_leaf(l);
1788 if (ll && hlist_empty(&ll->list))
1789 trie_leaf_remove(t, ll);
1793 if (ll && hlist_empty(&ll->list))
1794 trie_leaf_remove(t, ll);
1796 pr_debug("trie_flush found=%d\n", found);
1800 static void fn_trie_select_default(struct fib_table *tb,
1801 const struct flowi *flp,
1802 struct fib_result *res)
1804 struct trie *t = (struct trie *) tb->tb_data;
1805 int order, last_idx;
1806 struct fib_info *fi = NULL;
1807 struct fib_info *last_resort;
1808 struct fib_alias *fa = NULL;
1809 struct list_head *fa_head;
1818 l = fib_find_node(t, 0);
1822 fa_head = get_fa_head(l, 0);
1826 if (list_empty(fa_head))
1829 list_for_each_entry_rcu(fa, fa_head, fa_list) {
1830 struct fib_info *next_fi = fa->fa_info;
1832 if (fa->fa_scope != res->scope ||
1833 fa->fa_type != RTN_UNICAST)
1836 if (next_fi->fib_priority > res->fi->fib_priority)
1838 if (!next_fi->fib_nh[0].nh_gw ||
1839 next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1841 fa->fa_state |= FA_S_ACCESSED;
1844 if (next_fi != res->fi)
1846 } else if (!fib_detect_death(fi, order, &last_resort,
1847 &last_idx, tb->tb_default)) {
1848 fib_result_assign(res, fi);
1849 tb->tb_default = order;
1855 if (order <= 0 || fi == NULL) {
1856 tb->tb_default = -1;
1860 if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1862 fib_result_assign(res, fi);
1863 tb->tb_default = order;
1867 fib_result_assign(res, last_resort);
1868 tb->tb_default = last_idx;
1873 static int fn_trie_dump_fa(t_key key, int plen, struct list_head *fah,
1874 struct fib_table *tb,
1875 struct sk_buff *skb, struct netlink_callback *cb)
1878 struct fib_alias *fa;
1879 __be32 xkey = htonl(key);
1884 /* rcu_read_lock is hold by caller */
1886 list_for_each_entry_rcu(fa, fah, fa_list) {
1892 if (fib_dump_info(skb, NETLINK_CB(cb->skb).pid,
1901 fa->fa_info, NLM_F_MULTI) < 0) {
1911 static int fn_trie_dump_leaf(struct leaf *l, struct fib_table *tb,
1912 struct sk_buff *skb, struct netlink_callback *cb)
1914 struct leaf_info *li;
1915 struct hlist_node *node;
1921 /* rcu_read_lock is hold by caller */
1922 hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
1931 if (list_empty(&li->falh))
1934 if (fn_trie_dump_fa(l->key, li->plen, &li->falh, tb, skb, cb) < 0) {
1945 static int fn_trie_dump(struct fib_table *tb, struct sk_buff *skb,
1946 struct netlink_callback *cb)
1949 struct trie *t = (struct trie *) tb->tb_data;
1950 t_key key = cb->args[2];
1951 int count = cb->args[3];
1954 /* Dump starting at last key.
1955 * Note: 0.0.0.0/0 (ie default) is first key.
1958 l = trie_firstleaf(t);
1960 /* Normally, continue from last key, but if that is missing
1961 * fallback to using slow rescan
1963 l = fib_find_node(t, key);
1965 l = trie_leafindex(t, count);
1969 cb->args[2] = l->key;
1970 if (fn_trie_dump_leaf(l, tb, skb, cb) < 0) {
1971 cb->args[3] = count;
1977 l = trie_nextleaf(l);
1978 memset(&cb->args[4], 0,
1979 sizeof(cb->args) - 4*sizeof(cb->args[0]));
1981 cb->args[3] = count;
1987 void __init fib_hash_init(void)
1989 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
1990 sizeof(struct fib_alias),
1991 0, SLAB_PANIC, NULL);
1993 trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
1994 max(sizeof(struct leaf),
1995 sizeof(struct leaf_info)),
1996 0, SLAB_PANIC, NULL);
2000 /* Fix more generic FIB names for init later */
2001 struct fib_table *fib_hash_table(u32 id)
2003 struct fib_table *tb;
2006 tb = kmalloc(sizeof(struct fib_table) + sizeof(struct trie),
2012 tb->tb_default = -1;
2013 tb->tb_lookup = fn_trie_lookup;
2014 tb->tb_insert = fn_trie_insert;
2015 tb->tb_delete = fn_trie_delete;
2016 tb->tb_flush = fn_trie_flush;
2017 tb->tb_select_default = fn_trie_select_default;
2018 tb->tb_dump = fn_trie_dump;
2020 t = (struct trie *) tb->tb_data;
2021 memset(t, 0, sizeof(*t));
2023 if (id == RT_TABLE_LOCAL)
2024 pr_info("IPv4 FIB: Using LC-trie version %s\n", VERSION);
2029 #ifdef CONFIG_PROC_FS
2030 /* Depth first Trie walk iterator */
2031 struct fib_trie_iter {
2032 struct seq_net_private p;
2033 struct fib_table *tb;
2034 struct tnode *tnode;
2039 static struct node *fib_trie_get_next(struct fib_trie_iter *iter)
2041 struct tnode *tn = iter->tnode;
2042 unsigned cindex = iter->index;
2045 /* A single entry routing table */
2049 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2050 iter->tnode, iter->index, iter->depth);
2052 while (cindex < (1<<tn->bits)) {
2053 struct node *n = tnode_get_child_rcu(tn, cindex);
2058 iter->index = cindex + 1;
2060 /* push down one level */
2061 iter->tnode = (struct tnode *) n;
2071 /* Current node exhausted, pop back up */
2072 p = node_parent_rcu((struct node *)tn);
2074 cindex = tkey_extract_bits(tn->key, p->pos, p->bits)+1;
2084 static struct node *fib_trie_get_first(struct fib_trie_iter *iter,
2092 n = rcu_dereference(t->trie);
2097 iter->tnode = (struct tnode *) n;
2109 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2112 struct fib_trie_iter iter;
2114 memset(s, 0, sizeof(*s));
2117 for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2119 struct leaf *l = (struct leaf *)n;
2120 struct leaf_info *li;
2121 struct hlist_node *tmp;
2124 s->totdepth += iter.depth;
2125 if (iter.depth > s->maxdepth)
2126 s->maxdepth = iter.depth;
2128 hlist_for_each_entry_rcu(li, tmp, &l->list, hlist)
2131 const struct tnode *tn = (const struct tnode *) n;
2135 if (tn->bits < MAX_STAT_DEPTH)
2136 s->nodesizes[tn->bits]++;
2138 for (i = 0; i < (1<<tn->bits); i++)
2147 * This outputs /proc/net/fib_triestats
2149 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2151 unsigned i, max, pointers, bytes, avdepth;
2154 avdepth = stat->totdepth*100 / stat->leaves;
2158 seq_printf(seq, "\tAver depth: %u.%02d\n",
2159 avdepth / 100, avdepth % 100);
2160 seq_printf(seq, "\tMax depth: %u\n", stat->maxdepth);
2162 seq_printf(seq, "\tLeaves: %u\n", stat->leaves);
2163 bytes = sizeof(struct leaf) * stat->leaves;
2165 seq_printf(seq, "\tPrefixes: %u\n", stat->prefixes);
2166 bytes += sizeof(struct leaf_info) * stat->prefixes;
2168 seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2169 bytes += sizeof(struct tnode) * stat->tnodes;
2171 max = MAX_STAT_DEPTH;
2172 while (max > 0 && stat->nodesizes[max-1] == 0)
2176 for (i = 1; i <= max; i++)
2177 if (stat->nodesizes[i] != 0) {
2178 seq_printf(seq, " %u: %u", i, stat->nodesizes[i]);
2179 pointers += (1<<i) * stat->nodesizes[i];
2181 seq_putc(seq, '\n');
2182 seq_printf(seq, "\tPointers: %u\n", pointers);
2184 bytes += sizeof(struct node *) * pointers;
2185 seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2186 seq_printf(seq, "Total size: %u kB\n", (bytes + 1023) / 1024);
2189 #ifdef CONFIG_IP_FIB_TRIE_STATS
2190 static void trie_show_usage(struct seq_file *seq,
2191 const struct trie_use_stats *stats)
2193 seq_printf(seq, "\nCounters:\n---------\n");
2194 seq_printf(seq, "gets = %u\n", stats->gets);
2195 seq_printf(seq, "backtracks = %u\n", stats->backtrack);
2196 seq_printf(seq, "semantic match passed = %u\n",
2197 stats->semantic_match_passed);
2198 seq_printf(seq, "semantic match miss = %u\n",
2199 stats->semantic_match_miss);
2200 seq_printf(seq, "null node hit= %u\n", stats->null_node_hit);
2201 seq_printf(seq, "skipped node resize = %u\n\n",
2202 stats->resize_node_skipped);
2204 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2206 static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2208 if (tb->tb_id == RT_TABLE_LOCAL)
2209 seq_puts(seq, "Local:\n");
2210 else if (tb->tb_id == RT_TABLE_MAIN)
2211 seq_puts(seq, "Main:\n");
2213 seq_printf(seq, "Id %d:\n", tb->tb_id);
2217 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2219 struct net *net = (struct net *)seq->private;
2223 "Basic info: size of leaf:"
2224 " %Zd bytes, size of tnode: %Zd bytes.\n",
2225 sizeof(struct leaf), sizeof(struct tnode));
2227 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2228 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2229 struct hlist_node *node;
2230 struct fib_table *tb;
2232 hlist_for_each_entry_rcu(tb, node, head, tb_hlist) {
2233 struct trie *t = (struct trie *) tb->tb_data;
2234 struct trie_stat stat;
2239 fib_table_print(seq, tb);
2241 trie_collect_stats(t, &stat);
2242 trie_show_stats(seq, &stat);
2243 #ifdef CONFIG_IP_FIB_TRIE_STATS
2244 trie_show_usage(seq, &t->stats);
2252 static int fib_triestat_seq_open(struct inode *inode, struct file *file)
2254 return single_open_net(inode, file, fib_triestat_seq_show);
2257 static const struct file_operations fib_triestat_fops = {
2258 .owner = THIS_MODULE,
2259 .open = fib_triestat_seq_open,
2261 .llseek = seq_lseek,
2262 .release = single_release_net,
2265 static struct node *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2267 struct fib_trie_iter *iter = seq->private;
2268 struct net *net = seq_file_net(seq);
2272 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2273 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2274 struct hlist_node *node;
2275 struct fib_table *tb;
2277 hlist_for_each_entry_rcu(tb, node, head, tb_hlist) {
2280 for (n = fib_trie_get_first(iter,
2281 (struct trie *) tb->tb_data);
2282 n; n = fib_trie_get_next(iter))
2293 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2297 return fib_trie_get_idx(seq, *pos);
2300 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2302 struct fib_trie_iter *iter = seq->private;
2303 struct net *net = seq_file_net(seq);
2304 struct fib_table *tb = iter->tb;
2305 struct hlist_node *tb_node;
2310 /* next node in same table */
2311 n = fib_trie_get_next(iter);
2315 /* walk rest of this hash chain */
2316 h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2317 while ( (tb_node = rcu_dereference(tb->tb_hlist.next)) ) {
2318 tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2319 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2324 /* new hash chain */
2325 while (++h < FIB_TABLE_HASHSZ) {
2326 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2327 hlist_for_each_entry_rcu(tb, tb_node, head, tb_hlist) {
2328 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2340 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2346 static void seq_indent(struct seq_file *seq, int n)
2348 while (n-- > 0) seq_puts(seq, " ");
2351 static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2354 case RT_SCOPE_UNIVERSE: return "universe";
2355 case RT_SCOPE_SITE: return "site";
2356 case RT_SCOPE_LINK: return "link";
2357 case RT_SCOPE_HOST: return "host";
2358 case RT_SCOPE_NOWHERE: return "nowhere";
2360 snprintf(buf, len, "scope=%d", s);
2365 static const char *rtn_type_names[__RTN_MAX] = {
2366 [RTN_UNSPEC] = "UNSPEC",
2367 [RTN_UNICAST] = "UNICAST",
2368 [RTN_LOCAL] = "LOCAL",
2369 [RTN_BROADCAST] = "BROADCAST",
2370 [RTN_ANYCAST] = "ANYCAST",
2371 [RTN_MULTICAST] = "MULTICAST",
2372 [RTN_BLACKHOLE] = "BLACKHOLE",
2373 [RTN_UNREACHABLE] = "UNREACHABLE",
2374 [RTN_PROHIBIT] = "PROHIBIT",
2375 [RTN_THROW] = "THROW",
2377 [RTN_XRESOLVE] = "XRESOLVE",
2380 static inline const char *rtn_type(char *buf, size_t len, unsigned t)
2382 if (t < __RTN_MAX && rtn_type_names[t])
2383 return rtn_type_names[t];
2384 snprintf(buf, len, "type %u", t);
2388 /* Pretty print the trie */
2389 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2391 const struct fib_trie_iter *iter = seq->private;
2394 if (!node_parent_rcu(n))
2395 fib_table_print(seq, iter->tb);
2398 struct tnode *tn = (struct tnode *) n;
2399 __be32 prf = htonl(mask_pfx(tn->key, tn->pos));
2401 seq_indent(seq, iter->depth-1);
2402 seq_printf(seq, " +-- " NIPQUAD_FMT "/%d %d %d %d\n",
2403 NIPQUAD(prf), tn->pos, tn->bits, tn->full_children,
2404 tn->empty_children);
2407 struct leaf *l = (struct leaf *) n;
2408 struct leaf_info *li;
2409 struct hlist_node *node;
2410 __be32 val = htonl(l->key);
2412 seq_indent(seq, iter->depth);
2413 seq_printf(seq, " |-- " NIPQUAD_FMT "\n", NIPQUAD(val));
2415 hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
2416 struct fib_alias *fa;
2418 list_for_each_entry_rcu(fa, &li->falh, fa_list) {
2419 char buf1[32], buf2[32];
2421 seq_indent(seq, iter->depth+1);
2422 seq_printf(seq, " /%d %s %s", li->plen,
2423 rtn_scope(buf1, sizeof(buf1),
2425 rtn_type(buf2, sizeof(buf2),
2428 seq_printf(seq, " tos=%d", fa->fa_tos);
2429 seq_putc(seq, '\n');
2437 static const struct seq_operations fib_trie_seq_ops = {
2438 .start = fib_trie_seq_start,
2439 .next = fib_trie_seq_next,
2440 .stop = fib_trie_seq_stop,
2441 .show = fib_trie_seq_show,
2444 static int fib_trie_seq_open(struct inode *inode, struct file *file)
2446 return seq_open_net(inode, file, &fib_trie_seq_ops,
2447 sizeof(struct fib_trie_iter));
2450 static const struct file_operations fib_trie_fops = {
2451 .owner = THIS_MODULE,
2452 .open = fib_trie_seq_open,
2454 .llseek = seq_lseek,
2455 .release = seq_release_net,
2458 struct fib_route_iter {
2459 struct seq_net_private p;
2460 struct trie *main_trie;
2465 static struct leaf *fib_route_get_idx(struct fib_route_iter *iter, loff_t pos)
2467 struct leaf *l = NULL;
2468 struct trie *t = iter->main_trie;
2470 /* use cache location of last found key */
2471 if (iter->pos > 0 && pos >= iter->pos && (l = fib_find_node(t, iter->key)))
2475 l = trie_firstleaf(t);
2478 while (l && pos-- > 0) {
2480 l = trie_nextleaf(l);
2484 iter->key = pos; /* remember it */
2486 iter->pos = 0; /* forget it */
2491 static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2494 struct fib_route_iter *iter = seq->private;
2495 struct fib_table *tb;
2498 tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2502 iter->main_trie = (struct trie *) tb->tb_data;
2504 return SEQ_START_TOKEN;
2506 return fib_route_get_idx(iter, *pos - 1);
2509 static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2511 struct fib_route_iter *iter = seq->private;
2515 if (v == SEQ_START_TOKEN) {
2517 l = trie_firstleaf(iter->main_trie);
2520 l = trie_nextleaf(l);
2530 static void fib_route_seq_stop(struct seq_file *seq, void *v)
2536 static unsigned fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2538 static unsigned type2flags[RTN_MAX + 1] = {
2539 [7] = RTF_REJECT, [8] = RTF_REJECT,
2541 unsigned flags = type2flags[type];
2543 if (fi && fi->fib_nh->nh_gw)
2544 flags |= RTF_GATEWAY;
2545 if (mask == htonl(0xFFFFFFFF))
2552 * This outputs /proc/net/route.
2553 * The format of the file is not supposed to be changed
2554 * and needs to be same as fib_hash output to avoid breaking
2557 static int fib_route_seq_show(struct seq_file *seq, void *v)
2560 struct leaf_info *li;
2561 struct hlist_node *node;
2563 if (v == SEQ_START_TOKEN) {
2564 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2565 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2570 hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
2571 struct fib_alias *fa;
2572 __be32 mask, prefix;
2574 mask = inet_make_mask(li->plen);
2575 prefix = htonl(l->key);
2577 list_for_each_entry_rcu(fa, &li->falh, fa_list) {
2578 const struct fib_info *fi = fa->fa_info;
2579 unsigned flags = fib_flag_trans(fa->fa_type, mask, fi);
2582 if (fa->fa_type == RTN_BROADCAST
2583 || fa->fa_type == RTN_MULTICAST)
2588 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2589 "%d\t%08X\t%d\t%u\t%u%n",
2590 fi->fib_dev ? fi->fib_dev->name : "*",
2592 fi->fib_nh->nh_gw, flags, 0, 0,
2596 fi->fib_advmss + 40 : 0),
2598 fi->fib_rtt >> 3, &len);
2601 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2602 "%d\t%08X\t%d\t%u\t%u%n",
2603 prefix, 0, flags, 0, 0, 0,
2604 mask, 0, 0, 0, &len);
2606 seq_printf(seq, "%*s\n", 127 - len, "");
2613 static const struct seq_operations fib_route_seq_ops = {
2614 .start = fib_route_seq_start,
2615 .next = fib_route_seq_next,
2616 .stop = fib_route_seq_stop,
2617 .show = fib_route_seq_show,
2620 static int fib_route_seq_open(struct inode *inode, struct file *file)
2622 return seq_open_net(inode, file, &fib_route_seq_ops,
2623 sizeof(struct fib_route_iter));
2626 static const struct file_operations fib_route_fops = {
2627 .owner = THIS_MODULE,
2628 .open = fib_route_seq_open,
2630 .llseek = seq_lseek,
2631 .release = seq_release_net,
2634 int __net_init fib_proc_init(struct net *net)
2636 if (!proc_net_fops_create(net, "fib_trie", S_IRUGO, &fib_trie_fops))
2639 if (!proc_net_fops_create(net, "fib_triestat", S_IRUGO,
2640 &fib_triestat_fops))
2643 if (!proc_net_fops_create(net, "route", S_IRUGO, &fib_route_fops))
2649 proc_net_remove(net, "fib_triestat");
2651 proc_net_remove(net, "fib_trie");
2656 void __net_exit fib_proc_exit(struct net *net)
2658 proc_net_remove(net, "fib_trie");
2659 proc_net_remove(net, "fib_triestat");
2660 proc_net_remove(net, "route");
2663 #endif /* CONFIG_PROC_FS */