2 * This program is free software; you can redistribute it and/or
3 * modify it under the terms of the GNU General Public License
4 * as published by the Free Software Foundation; either version
5 * 2 of the License, or (at your option) any later version.
7 * Robert Olsson <robert.olsson@its.uu.se> Uppsala Universitet
8 * & Swedish University of Agricultural Sciences.
10 * Jens Laas <jens.laas@data.slu.se> Swedish University of
11 * Agricultural Sciences.
13 * Hans Liss <hans.liss@its.uu.se> Uppsala Universitet
15 * This work is based on the LPC-trie which is originally described in:
17 * An experimental study of compression methods for dynamic tries
18 * Stefan Nilsson and Matti Tikkanen. Algorithmica, 33(1):19-33, 2002.
19 * http://www.csc.kth.se/~snilsson/software/dyntrie2/
22 * IP-address lookup using LC-tries. Stefan Nilsson and Gunnar Karlsson
23 * IEEE Journal on Selected Areas in Communications, 17(6):1083-1092, June 1999
26 * Code from fib_hash has been reused which includes the following header:
29 * INET An implementation of the TCP/IP protocol suite for the LINUX
30 * operating system. INET is implemented using the BSD Socket
31 * interface as the means of communication with the user level.
33 * IPv4 FIB: lookup engine and maintenance routines.
36 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
38 * This program is free software; you can redistribute it and/or
39 * modify it under the terms of the GNU General Public License
40 * as published by the Free Software Foundation; either version
41 * 2 of the License, or (at your option) any later version.
43 * Substantial contributions to this work comes from:
45 * David S. Miller, <davem@davemloft.net>
46 * Stephen Hemminger <shemminger@osdl.org>
47 * Paul E. McKenney <paulmck@us.ibm.com>
48 * Patrick McHardy <kaber@trash.net>
51 #define VERSION "0.409"
53 #include <asm/uaccess.h>
54 #include <linux/bitops.h>
55 #include <linux/types.h>
56 #include <linux/kernel.h>
58 #include <linux/string.h>
59 #include <linux/socket.h>
60 #include <linux/sockios.h>
61 #include <linux/errno.h>
63 #include <linux/inet.h>
64 #include <linux/inetdevice.h>
65 #include <linux/netdevice.h>
66 #include <linux/if_arp.h>
67 #include <linux/proc_fs.h>
68 #include <linux/rcupdate.h>
69 #include <linux/skbuff.h>
70 #include <linux/netlink.h>
71 #include <linux/init.h>
72 #include <linux/list.h>
73 #include <linux/slab.h>
74 #include <linux/export.h>
75 #include <net/net_namespace.h>
77 #include <net/protocol.h>
78 #include <net/route.h>
81 #include <net/ip_fib.h>
82 #include "fib_lookup.h"
84 #define MAX_STAT_DEPTH 32
86 #define KEYLENGTH (8*sizeof(t_key))
87 #define KEY_MAX ((t_key)~0)
89 typedef unsigned int t_key;
91 #define IS_TNODE(n) ((n)->bits)
92 #define IS_LEAF(n) (!(n)->bits)
94 #define get_index(_key, _kv) (((_key) ^ (_kv)->key) >> (_kv)->pos)
98 unsigned char bits; /* 2log(KEYLENGTH) bits needed */
99 unsigned char pos; /* 2log(KEYLENGTH) bits needed */
101 struct tnode __rcu *parent;
104 /* The fields in this struct are valid if bits > 0 (TNODE) */
106 t_key empty_children; /* KEYLENGTH bits needed */
107 t_key full_children; /* KEYLENGTH bits needed */
108 struct tnode __rcu *child[0];
110 /* This list pointer if valid if bits == 0 (LEAF) */
111 struct hlist_head leaf;
115 #ifdef CONFIG_IP_FIB_TRIE_STATS
116 struct trie_use_stats {
118 unsigned int backtrack;
119 unsigned int semantic_match_passed;
120 unsigned int semantic_match_miss;
121 unsigned int null_node_hit;
122 unsigned int resize_node_skipped;
127 unsigned int totdepth;
128 unsigned int maxdepth;
131 unsigned int nullpointers;
132 unsigned int prefixes;
133 unsigned int nodesizes[MAX_STAT_DEPTH];
137 struct tnode __rcu *trie;
138 #ifdef CONFIG_IP_FIB_TRIE_STATS
139 struct trie_use_stats __percpu *stats;
143 static void resize(struct trie *t, struct tnode *tn);
144 static size_t tnode_free_size;
147 * synchronize_rcu after call_rcu for that many pages; it should be especially
148 * useful before resizing the root node with PREEMPT_NONE configs; the value was
149 * obtained experimentally, aiming to avoid visible slowdown.
151 static const int sync_pages = 128;
153 static struct kmem_cache *fn_alias_kmem __read_mostly;
154 static struct kmem_cache *trie_leaf_kmem __read_mostly;
156 /* caller must hold RTNL */
157 #define node_parent(n) rtnl_dereference((n)->parent)
159 /* caller must hold RCU read lock or RTNL */
160 #define node_parent_rcu(n) rcu_dereference_rtnl((n)->parent)
162 /* wrapper for rcu_assign_pointer */
163 static inline void node_set_parent(struct tnode *n, struct tnode *tp)
166 rcu_assign_pointer(n->parent, tp);
169 #define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER((n)->parent, p)
171 /* This provides us with the number of children in this node, in the case of a
172 * leaf this will return 0 meaning none of the children are accessible.
174 static inline unsigned long tnode_child_length(const struct tnode *tn)
176 return (1ul << tn->bits) & ~(1ul);
179 /* caller must hold RTNL */
180 static inline struct tnode *tnode_get_child(const struct tnode *tn,
183 return rtnl_dereference(tn->child[i]);
186 /* caller must hold RCU read lock or RTNL */
187 static inline struct tnode *tnode_get_child_rcu(const struct tnode *tn,
190 return rcu_dereference_rtnl(tn->child[i]);
193 /* To understand this stuff, an understanding of keys and all their bits is
194 * necessary. Every node in the trie has a key associated with it, but not
195 * all of the bits in that key are significant.
197 * Consider a node 'n' and its parent 'tp'.
199 * If n is a leaf, every bit in its key is significant. Its presence is
200 * necessitated by path compression, since during a tree traversal (when
201 * searching for a leaf - unless we are doing an insertion) we will completely
202 * ignore all skipped bits we encounter. Thus we need to verify, at the end of
203 * a potentially successful search, that we have indeed been walking the
206 * Note that we can never "miss" the correct key in the tree if present by
207 * following the wrong path. Path compression ensures that segments of the key
208 * that are the same for all keys with a given prefix are skipped, but the
209 * skipped part *is* identical for each node in the subtrie below the skipped
210 * bit! trie_insert() in this implementation takes care of that.
212 * if n is an internal node - a 'tnode' here, the various parts of its key
213 * have many different meanings.
216 * _________________________________________________________________
217 * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
218 * -----------------------------------------------------------------
219 * 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16
221 * _________________________________________________________________
222 * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
223 * -----------------------------------------------------------------
224 * 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
231 * First, let's just ignore the bits that come before the parent tp, that is
232 * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
233 * point we do not use them for anything.
235 * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
236 * index into the parent's child array. That is, they will be used to find
237 * 'n' among tp's children.
239 * The bits from (n->pos + n->bits) to (tn->pos - 1) - "S" - are skipped bits
242 * All the bits we have seen so far are significant to the node n. The rest
243 * of the bits are really not needed or indeed known in n->key.
245 * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
246 * n's child array, and will of course be different for each child.
248 * The rest of the bits, from 0 to (n->pos + n->bits), are completely unknown
252 static const int halve_threshold = 25;
253 static const int inflate_threshold = 50;
254 static const int halve_threshold_root = 15;
255 static const int inflate_threshold_root = 30;
257 static void __alias_free_mem(struct rcu_head *head)
259 struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
260 kmem_cache_free(fn_alias_kmem, fa);
263 static inline void alias_free_mem_rcu(struct fib_alias *fa)
265 call_rcu(&fa->rcu, __alias_free_mem);
268 #define TNODE_KMALLOC_MAX \
269 ilog2((PAGE_SIZE - sizeof(struct tnode)) / sizeof(struct tnode *))
271 static void __node_free_rcu(struct rcu_head *head)
273 struct tnode *n = container_of(head, struct tnode, rcu);
276 kmem_cache_free(trie_leaf_kmem, n);
277 else if (n->bits <= TNODE_KMALLOC_MAX)
283 #define node_free(n) call_rcu(&n->rcu, __node_free_rcu)
285 static struct tnode *tnode_alloc(size_t size)
287 if (size <= PAGE_SIZE)
288 return kzalloc(size, GFP_KERNEL);
290 return vzalloc(size);
293 static inline void empty_child_inc(struct tnode *n)
295 ++n->empty_children ? : ++n->full_children;
298 static inline void empty_child_dec(struct tnode *n)
300 n->empty_children-- ? : n->full_children--;
303 static struct tnode *leaf_new(t_key key)
305 struct tnode *l = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
308 /* set key and pos to reflect full key value
309 * any trailing zeros in the key should be ignored
310 * as the nodes are searched
315 /* set bits to 0 indicating we are not a tnode */
318 INIT_HLIST_HEAD(&l->leaf);
323 static struct tnode *tnode_new(t_key key, int pos, int bits)
325 size_t sz = offsetof(struct tnode, child[1ul << bits]);
326 struct tnode *tn = tnode_alloc(sz);
327 unsigned int shift = pos + bits;
329 /* verify bits and pos their msb bits clear and values are valid */
330 BUG_ON(!bits || (shift > KEYLENGTH));
337 tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
338 if (bits == KEYLENGTH)
339 tn->full_children = 1;
341 tn->empty_children = 1ul << bits;
344 pr_debug("AT %p s=%zu %zu\n", tn, sizeof(struct tnode),
345 sizeof(struct tnode *) << bits);
349 /* Check whether a tnode 'n' is "full", i.e. it is an internal node
350 * and no bits are skipped. See discussion in dyntree paper p. 6
352 static inline int tnode_full(const struct tnode *tn, const struct tnode *n)
354 return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
357 /* Add a child at position i overwriting the old value.
358 * Update the value of full_children and empty_children.
360 static void put_child(struct tnode *tn, unsigned long i, struct tnode *n)
362 struct tnode *chi = tnode_get_child(tn, i);
365 BUG_ON(i >= tnode_child_length(tn));
367 /* update emptyChildren, overflow into fullChildren */
368 if (n == NULL && chi != NULL)
370 if (n != NULL && chi == NULL)
373 /* update fullChildren */
374 wasfull = tnode_full(tn, chi);
375 isfull = tnode_full(tn, n);
377 if (wasfull && !isfull)
379 else if (!wasfull && isfull)
382 if (n && (tn->slen < n->slen))
385 rcu_assign_pointer(tn->child[i], n);
388 static void update_children(struct tnode *tn)
392 /* update all of the child parent pointers */
393 for (i = tnode_child_length(tn); i;) {
394 struct tnode *inode = tnode_get_child(tn, --i);
399 /* Either update the children of a tnode that
400 * already belongs to us or update the child
401 * to point to ourselves.
403 if (node_parent(inode) == tn)
404 update_children(inode);
406 node_set_parent(inode, tn);
410 static inline void put_child_root(struct tnode *tp, struct trie *t,
411 t_key key, struct tnode *n)
414 put_child(tp, get_index(key, tp), n);
416 rcu_assign_pointer(t->trie, n);
419 static inline void tnode_free_init(struct tnode *tn)
424 static inline void tnode_free_append(struct tnode *tn, struct tnode *n)
426 n->rcu.next = tn->rcu.next;
427 tn->rcu.next = &n->rcu;
430 static void tnode_free(struct tnode *tn)
432 struct callback_head *head = &tn->rcu;
436 tnode_free_size += offsetof(struct tnode, child[1 << tn->bits]);
439 tn = container_of(head, struct tnode, rcu);
442 if (tnode_free_size >= PAGE_SIZE * sync_pages) {
448 static void replace(struct trie *t, struct tnode *oldtnode, struct tnode *tn)
450 struct tnode *tp = node_parent(oldtnode);
453 /* setup the parent pointer out of and back into this node */
454 NODE_INIT_PARENT(tn, tp);
455 put_child_root(tp, t, tn->key, tn);
457 /* update all of the child parent pointers */
460 /* all pointers should be clean so we are done */
461 tnode_free(oldtnode);
463 /* resize children now that oldtnode is freed */
464 for (i = tnode_child_length(tn); i;) {
465 struct tnode *inode = tnode_get_child(tn, --i);
467 /* resize child node */
468 if (tnode_full(tn, inode))
473 static int inflate(struct trie *t, struct tnode *oldtnode)
479 pr_debug("In inflate\n");
481 tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
485 /* prepare oldtnode to be freed */
486 tnode_free_init(oldtnode);
488 /* Assemble all of the pointers in our cluster, in this case that
489 * represents all of the pointers out of our allocated nodes that
490 * point to existing tnodes and the links between our allocated
493 for (i = tnode_child_length(oldtnode), m = 1u << tn->pos; i;) {
494 struct tnode *inode = tnode_get_child(oldtnode, --i);
495 struct tnode *node0, *node1;
502 /* A leaf or an internal node with skipped bits */
503 if (!tnode_full(oldtnode, inode)) {
504 put_child(tn, get_index(inode->key, tn), inode);
508 /* drop the node in the old tnode free list */
509 tnode_free_append(oldtnode, inode);
511 /* An internal node with two children */
512 if (inode->bits == 1) {
513 put_child(tn, 2 * i + 1, tnode_get_child(inode, 1));
514 put_child(tn, 2 * i, tnode_get_child(inode, 0));
518 /* We will replace this node 'inode' with two new
519 * ones, 'node0' and 'node1', each with half of the
520 * original children. The two new nodes will have
521 * a position one bit further down the key and this
522 * means that the "significant" part of their keys
523 * (see the discussion near the top of this file)
524 * will differ by one bit, which will be "0" in
525 * node0's key and "1" in node1's key. Since we are
526 * moving the key position by one step, the bit that
527 * we are moving away from - the bit at position
528 * (tn->pos) - is the one that will differ between
529 * node0 and node1. So... we synthesize that bit in the
532 node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
535 node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
537 tnode_free_append(tn, node1);
540 tnode_free_append(tn, node0);
542 /* populate child pointers in new nodes */
543 for (k = tnode_child_length(inode), j = k / 2; j;) {
544 put_child(node1, --j, tnode_get_child(inode, --k));
545 put_child(node0, j, tnode_get_child(inode, j));
546 put_child(node1, --j, tnode_get_child(inode, --k));
547 put_child(node0, j, tnode_get_child(inode, j));
550 /* link new nodes to parent */
551 NODE_INIT_PARENT(node1, tn);
552 NODE_INIT_PARENT(node0, tn);
554 /* link parent to nodes */
555 put_child(tn, 2 * i + 1, node1);
556 put_child(tn, 2 * i, node0);
559 /* setup the parent pointers into and out of this node */
560 replace(t, oldtnode, tn);
564 /* all pointers should be clean so we are done */
569 static int halve(struct trie *t, struct tnode *oldtnode)
574 pr_debug("In halve\n");
576 tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
580 /* prepare oldtnode to be freed */
581 tnode_free_init(oldtnode);
583 /* Assemble all of the pointers in our cluster, in this case that
584 * represents all of the pointers out of our allocated nodes that
585 * point to existing tnodes and the links between our allocated
588 for (i = tnode_child_length(oldtnode); i;) {
589 struct tnode *node1 = tnode_get_child(oldtnode, --i);
590 struct tnode *node0 = tnode_get_child(oldtnode, --i);
593 /* At least one of the children is empty */
594 if (!node1 || !node0) {
595 put_child(tn, i / 2, node1 ? : node0);
599 /* Two nonempty children */
600 inode = tnode_new(node0->key, oldtnode->pos, 1);
605 tnode_free_append(tn, inode);
607 /* initialize pointers out of node */
608 put_child(inode, 1, node1);
609 put_child(inode, 0, node0);
610 NODE_INIT_PARENT(inode, tn);
612 /* link parent to node */
613 put_child(tn, i / 2, inode);
616 /* setup the parent pointers into and out of this node */
617 replace(t, oldtnode, tn);
622 static void collapse(struct trie *t, struct tnode *oldtnode)
624 struct tnode *n, *tp;
627 /* scan the tnode looking for that one child that might still exist */
628 for (n = NULL, i = tnode_child_length(oldtnode); !n && i;)
629 n = tnode_get_child(oldtnode, --i);
631 /* compress one level */
632 tp = node_parent(oldtnode);
633 put_child_root(tp, t, oldtnode->key, n);
634 node_set_parent(n, tp);
640 static unsigned char update_suffix(struct tnode *tn)
642 unsigned char slen = tn->pos;
643 unsigned long stride, i;
645 /* search though the list of children looking for nodes that might
646 * have a suffix greater than the one we currently have. This is
647 * why we start with a stride of 2 since a stride of 1 would
648 * represent the nodes with suffix length equal to tn->pos
650 for (i = 0, stride = 0x2ul ; i < tnode_child_length(tn); i += stride) {
651 struct tnode *n = tnode_get_child(tn, i);
653 if (!n || (n->slen <= slen))
656 /* update stride and slen based on new value */
657 stride <<= (n->slen - slen);
661 /* if slen covers all but the last bit we can stop here
662 * there will be nothing longer than that since only node
663 * 0 and 1 << (bits - 1) could have that as their suffix
666 if ((slen + 1) >= (tn->pos + tn->bits))
675 /* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
676 * the Helsinki University of Technology and Matti Tikkanen of Nokia
677 * Telecommunications, page 6:
678 * "A node is doubled if the ratio of non-empty children to all
679 * children in the *doubled* node is at least 'high'."
681 * 'high' in this instance is the variable 'inflate_threshold'. It
682 * is expressed as a percentage, so we multiply it with
683 * tnode_child_length() and instead of multiplying by 2 (since the
684 * child array will be doubled by inflate()) and multiplying
685 * the left-hand side by 100 (to handle the percentage thing) we
686 * multiply the left-hand side by 50.
688 * The left-hand side may look a bit weird: tnode_child_length(tn)
689 * - tn->empty_children is of course the number of non-null children
690 * in the current node. tn->full_children is the number of "full"
691 * children, that is non-null tnodes with a skip value of 0.
692 * All of those will be doubled in the resulting inflated tnode, so
693 * we just count them one extra time here.
695 * A clearer way to write this would be:
697 * to_be_doubled = tn->full_children;
698 * not_to_be_doubled = tnode_child_length(tn) - tn->empty_children -
701 * new_child_length = tnode_child_length(tn) * 2;
703 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
705 * if (new_fill_factor >= inflate_threshold)
707 * ...and so on, tho it would mess up the while () loop.
710 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
714 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
715 * inflate_threshold * new_child_length
717 * expand not_to_be_doubled and to_be_doubled, and shorten:
718 * 100 * (tnode_child_length(tn) - tn->empty_children +
719 * tn->full_children) >= inflate_threshold * new_child_length
721 * expand new_child_length:
722 * 100 * (tnode_child_length(tn) - tn->empty_children +
723 * tn->full_children) >=
724 * inflate_threshold * tnode_child_length(tn) * 2
727 * 50 * (tn->full_children + tnode_child_length(tn) -
728 * tn->empty_children) >= inflate_threshold *
729 * tnode_child_length(tn)
732 static bool should_inflate(const struct tnode *tp, const struct tnode *tn)
734 unsigned long used = tnode_child_length(tn);
735 unsigned long threshold = used;
737 /* Keep root node larger */
738 threshold *= tp ? inflate_threshold : inflate_threshold_root;
739 used -= tn->empty_children;
740 used += tn->full_children;
742 /* if bits == KEYLENGTH then pos = 0, and will fail below */
744 return (used > 1) && tn->pos && ((50 * used) >= threshold);
747 static bool should_halve(const struct tnode *tp, const struct tnode *tn)
749 unsigned long used = tnode_child_length(tn);
750 unsigned long threshold = used;
752 /* Keep root node larger */
753 threshold *= tp ? halve_threshold : halve_threshold_root;
754 used -= tn->empty_children;
756 /* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
758 return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
761 static bool should_collapse(const struct tnode *tn)
763 unsigned long used = tnode_child_length(tn);
765 used -= tn->empty_children;
767 /* account for bits == KEYLENGTH case */
768 if ((tn->bits == KEYLENGTH) && tn->full_children)
771 /* One child or none, time to drop us from the trie */
776 static void resize(struct trie *t, struct tnode *tn)
778 struct tnode *tp = node_parent(tn);
779 struct tnode __rcu **cptr;
780 int max_work = MAX_WORK;
782 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
783 tn, inflate_threshold, halve_threshold);
785 /* track the tnode via the pointer from the parent instead of
786 * doing it ourselves. This way we can let RCU fully do its
787 * thing without us interfering
789 cptr = tp ? &tp->child[get_index(tn->key, tp)] : &t->trie;
790 BUG_ON(tn != rtnl_dereference(*cptr));
792 /* Double as long as the resulting node has a number of
793 * nonempty nodes that are above the threshold.
795 while (should_inflate(tp, tn) && max_work) {
796 if (inflate(t, tn)) {
797 #ifdef CONFIG_IP_FIB_TRIE_STATS
798 this_cpu_inc(t->stats->resize_node_skipped);
804 tn = rtnl_dereference(*cptr);
807 /* Return if at least one inflate is run */
808 if (max_work != MAX_WORK)
811 /* Halve as long as the number of empty children in this
812 * node is above threshold.
814 while (should_halve(tp, tn) && max_work) {
816 #ifdef CONFIG_IP_FIB_TRIE_STATS
817 this_cpu_inc(t->stats->resize_node_skipped);
823 tn = rtnl_dereference(*cptr);
826 /* Only one child remains */
827 if (should_collapse(tn)) {
832 /* Return if at least one deflate was run */
833 if (max_work != MAX_WORK)
836 /* push the suffix length to the parent node */
837 if (tn->slen > tn->pos) {
838 unsigned char slen = update_suffix(tn);
840 if (tp && (slen > tp->slen))
845 static void leaf_pull_suffix(struct tnode *l)
847 struct tnode *tp = node_parent(l);
849 while (tp && (tp->slen > tp->pos) && (tp->slen > l->slen)) {
850 if (update_suffix(tp) > l->slen)
852 tp = node_parent(tp);
856 static void leaf_push_suffix(struct tnode *l)
858 struct tnode *tn = node_parent(l);
860 /* if this is a new leaf then tn will be NULL and we can sort
861 * out parent suffix lengths as a part of trie_rebalance
863 while (tn && (tn->slen < l->slen)) {
865 tn = node_parent(tn);
869 static void fib_remove_alias(struct tnode *l, struct fib_alias *old)
871 /* record the location of the previous list_info entry */
872 struct hlist_node **pprev = old->fa_list.pprev;
873 struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
875 /* remove the fib_alias from the list */
876 hlist_del_rcu(&old->fa_list);
878 /* only access fa if it is pointing at the last valid hlist_node */
879 if (hlist_empty(&l->leaf) || (*pprev))
882 /* update the trie with the latest suffix length */
883 l->slen = fa->fa_slen;
887 static void fib_insert_alias(struct tnode *l, struct fib_alias *fa,
888 struct fib_alias *new)
891 hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
893 struct fib_alias *last;
895 hlist_for_each_entry(last, &l->leaf, fa_list) {
896 if (new->fa_slen < last->fa_slen)
902 hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
904 hlist_add_head_rcu(&new->fa_list, &l->leaf);
907 /* if we added to the tail node then we need to update slen */
908 if (l->slen < new->fa_slen) {
909 l->slen = new->fa_slen;
914 /* rcu_read_lock needs to be hold by caller from readside */
915 static struct tnode *fib_find_node(struct trie *t, u32 key)
917 struct tnode *n = rcu_dereference_rtnl(t->trie);
920 unsigned long index = get_index(key, n);
922 /* This bit of code is a bit tricky but it combines multiple
923 * checks into a single check. The prefix consists of the
924 * prefix plus zeros for the bits in the cindex. The index
925 * is the difference between the key and this value. From
926 * this we can actually derive several pieces of data.
927 * if (index & (~0ul << bits))
928 * we have a mismatch in skip bits and failed
930 * we know the value is cindex
932 if (index & (~0ul << n->bits))
935 /* we have found a leaf. Prefixes have already been compared */
939 n = tnode_get_child_rcu(n, index);
945 /* Return the first fib alias matching TOS with
946 * priority less than or equal to PRIO.
948 static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
951 struct fib_alias *fa;
956 hlist_for_each_entry(fa, fah, fa_list) {
957 if (fa->fa_slen < slen)
959 if (fa->fa_slen != slen)
961 if (fa->fa_tos > tos)
963 if (fa->fa_info->fib_priority >= prio || fa->fa_tos < tos)
970 static void trie_rebalance(struct trie *t, struct tnode *tn)
974 while ((tp = node_parent(tn)) != NULL) {
979 /* Handle last (top) tnode */
984 /* only used from updater-side */
986 static struct tnode *fib_insert_node(struct trie *t, u32 key, int plen)
988 struct tnode *l, *n, *tp = NULL;
990 n = rtnl_dereference(t->trie);
992 /* If we point to NULL, stop. Either the tree is empty and we should
993 * just put a new leaf in if, or we have reached an empty child slot,
994 * and we should just put our new leaf in that.
996 * If we hit a node with a key that does't match then we should stop
997 * and create a new tnode to replace that node and insert ourselves
998 * and the other node into the new tnode.
1001 unsigned long index = get_index(key, n);
1003 /* This bit of code is a bit tricky but it combines multiple
1004 * checks into a single check. The prefix consists of the
1005 * prefix plus zeros for the "bits" in the prefix. The index
1006 * is the difference between the key and this value. From
1007 * this we can actually derive several pieces of data.
1008 * if !(index >> bits)
1009 * we know the value is child index
1011 * we have a mismatch in skip bits and failed
1013 if (index >> n->bits)
1016 /* we have found a leaf. Prefixes have already been compared */
1018 /* Case 1: n is a leaf, and prefixes match*/
1023 n = tnode_get_child_rcu(n, index);
1030 /* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1032 * Add a new tnode here
1033 * first tnode need some special handling
1034 * leaves us in position for handling as case 3
1039 tn = tnode_new(key, __fls(key ^ n->key), 1);
1045 /* initialize routes out of node */
1046 NODE_INIT_PARENT(tn, tp);
1047 put_child(tn, get_index(key, tn) ^ 1, n);
1049 /* start adding routes into the node */
1050 put_child_root(tp, t, key, tn);
1051 node_set_parent(n, tn);
1053 /* parent now has a NULL spot where the leaf can go */
1057 /* Case 3: n is NULL, and will just insert a new leaf */
1059 NODE_INIT_PARENT(l, tp);
1060 put_child(tp, get_index(key, tp), l);
1061 trie_rebalance(t, tp);
1063 rcu_assign_pointer(t->trie, l);
1070 * Caller must hold RTNL.
1072 int fib_table_insert(struct fib_table *tb, struct fib_config *cfg)
1074 struct trie *t = (struct trie *) tb->tb_data;
1075 struct fib_alias *fa, *new_fa;
1076 struct fib_info *fi;
1077 u8 plen = cfg->fc_dst_len;
1078 u8 slen = KEYLENGTH - plen;
1079 u8 tos = cfg->fc_tos;
1084 if (plen > KEYLENGTH)
1087 key = ntohl(cfg->fc_dst);
1089 pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1091 mask = ntohl(inet_make_mask(plen));
1096 fi = fib_create_info(cfg);
1102 l = fib_find_node(t, key);
1103 fa = l ? fib_find_alias(&l->leaf, slen, tos, fi->fib_priority) : NULL;
1105 /* Now fa, if non-NULL, points to the first fib alias
1106 * with the same keys [prefix,tos,priority], if such key already
1107 * exists or to the node before which we will insert new one.
1109 * If fa is NULL, we will need to allocate a new one and
1110 * insert to the tail of the section matching the suffix length
1114 if (fa && fa->fa_tos == tos &&
1115 fa->fa_info->fib_priority == fi->fib_priority) {
1116 struct fib_alias *fa_first, *fa_match;
1119 if (cfg->fc_nlflags & NLM_F_EXCL)
1123 * 1. Find exact match for type, scope, fib_info to avoid
1125 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1129 hlist_for_each_entry_from(fa, fa_list) {
1130 if ((fa->fa_slen != slen) || (fa->fa_tos != tos))
1132 if (fa->fa_info->fib_priority != fi->fib_priority)
1134 if (fa->fa_type == cfg->fc_type &&
1135 fa->fa_info == fi) {
1141 if (cfg->fc_nlflags & NLM_F_REPLACE) {
1142 struct fib_info *fi_drop;
1152 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1156 fi_drop = fa->fa_info;
1157 new_fa->fa_tos = fa->fa_tos;
1158 new_fa->fa_info = fi;
1159 new_fa->fa_type = cfg->fc_type;
1160 state = fa->fa_state;
1161 new_fa->fa_state = state & ~FA_S_ACCESSED;
1162 new_fa->fa_slen = fa->fa_slen;
1164 hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1165 alias_free_mem_rcu(fa);
1167 fib_release_info(fi_drop);
1168 if (state & FA_S_ACCESSED)
1169 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1170 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1171 tb->tb_id, &cfg->fc_nlinfo, NLM_F_REPLACE);
1175 /* Error if we find a perfect match which
1176 * uses the same scope, type, and nexthop
1182 if (!(cfg->fc_nlflags & NLM_F_APPEND))
1186 if (!(cfg->fc_nlflags & NLM_F_CREATE))
1190 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1194 new_fa->fa_info = fi;
1195 new_fa->fa_tos = tos;
1196 new_fa->fa_type = cfg->fc_type;
1197 new_fa->fa_state = 0;
1198 new_fa->fa_slen = slen;
1200 /* Insert new entry to the list. */
1202 l = fib_insert_node(t, key, plen);
1205 goto out_free_new_fa;
1210 tb->tb_num_default++;
1212 fib_insert_alias(l, fa, new_fa);
1214 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1215 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, tb->tb_id,
1216 &cfg->fc_nlinfo, 0);
1221 kmem_cache_free(fn_alias_kmem, new_fa);
1223 fib_release_info(fi);
1228 static inline t_key prefix_mismatch(t_key key, struct tnode *n)
1230 t_key prefix = n->key;
1232 return (key ^ prefix) & (prefix | -prefix);
1235 /* should be called with rcu_read_lock */
1236 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1237 struct fib_result *res, int fib_flags)
1239 struct trie *t = (struct trie *)tb->tb_data;
1240 #ifdef CONFIG_IP_FIB_TRIE_STATS
1241 struct trie_use_stats __percpu *stats = t->stats;
1243 const t_key key = ntohl(flp->daddr);
1244 struct tnode *n, *pn;
1245 struct fib_alias *fa;
1248 n = rcu_dereference(t->trie);
1252 #ifdef CONFIG_IP_FIB_TRIE_STATS
1253 this_cpu_inc(stats->gets);
1259 /* Step 1: Travel to the longest prefix match in the trie */
1261 unsigned long index = get_index(key, n);
1263 /* This bit of code is a bit tricky but it combines multiple
1264 * checks into a single check. The prefix consists of the
1265 * prefix plus zeros for the "bits" in the prefix. The index
1266 * is the difference between the key and this value. From
1267 * this we can actually derive several pieces of data.
1268 * if (index & (~0ul << bits))
1269 * we have a mismatch in skip bits and failed
1271 * we know the value is cindex
1273 if (index & (~0ul << n->bits))
1276 /* we have found a leaf. Prefixes have already been compared */
1280 /* only record pn and cindex if we are going to be chopping
1281 * bits later. Otherwise we are just wasting cycles.
1283 if (n->slen > n->pos) {
1288 n = tnode_get_child_rcu(n, index);
1293 /* Step 2: Sort out leaves and begin backtracing for longest prefix */
1295 /* record the pointer where our next node pointer is stored */
1296 struct tnode __rcu **cptr = n->child;
1298 /* This test verifies that none of the bits that differ
1299 * between the key and the prefix exist in the region of
1300 * the lsb and higher in the prefix.
1302 if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1305 /* exit out and process leaf */
1306 if (unlikely(IS_LEAF(n)))
1309 /* Don't bother recording parent info. Since we are in
1310 * prefix match mode we will have to come back to wherever
1311 * we started this traversal anyway
1314 while ((n = rcu_dereference(*cptr)) == NULL) {
1316 #ifdef CONFIG_IP_FIB_TRIE_STATS
1318 this_cpu_inc(stats->null_node_hit);
1320 /* If we are at cindex 0 there are no more bits for
1321 * us to strip at this level so we must ascend back
1322 * up one level to see if there are any more bits to
1323 * be stripped there.
1326 t_key pkey = pn->key;
1328 pn = node_parent_rcu(pn);
1331 #ifdef CONFIG_IP_FIB_TRIE_STATS
1332 this_cpu_inc(stats->backtrack);
1334 /* Get Child's index */
1335 cindex = get_index(pkey, pn);
1338 /* strip the least significant bit from the cindex */
1339 cindex &= cindex - 1;
1341 /* grab pointer for next child node */
1342 cptr = &pn->child[cindex];
1347 /* Step 3: Process the leaf, if that fails fall back to backtracing */
1348 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1349 struct fib_info *fi = fa->fa_info;
1352 if (((key ^ n->key) >= (1ul << fa->fa_slen)) &&
1353 ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen != KEYLENGTH)))
1355 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1359 if (fa->fa_info->fib_scope < flp->flowi4_scope)
1361 fib_alias_accessed(fa);
1362 err = fib_props[fa->fa_type].error;
1363 if (unlikely(err < 0)) {
1364 #ifdef CONFIG_IP_FIB_TRIE_STATS
1365 this_cpu_inc(stats->semantic_match_passed);
1369 if (fi->fib_flags & RTNH_F_DEAD)
1371 for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
1372 const struct fib_nh *nh = &fi->fib_nh[nhsel];
1374 if (nh->nh_flags & RTNH_F_DEAD)
1376 if (flp->flowi4_oif && flp->flowi4_oif != nh->nh_oif)
1379 if (!(fib_flags & FIB_LOOKUP_NOREF))
1380 atomic_inc(&fi->fib_clntref);
1382 res->prefixlen = KEYLENGTH - fa->fa_slen;
1383 res->nh_sel = nhsel;
1384 res->type = fa->fa_type;
1385 res->scope = fi->fib_scope;
1388 res->fa_head = &n->leaf;
1389 #ifdef CONFIG_IP_FIB_TRIE_STATS
1390 this_cpu_inc(stats->semantic_match_passed);
1395 #ifdef CONFIG_IP_FIB_TRIE_STATS
1396 this_cpu_inc(stats->semantic_match_miss);
1400 EXPORT_SYMBOL_GPL(fib_table_lookup);
1403 * Remove the leaf and return parent.
1405 static void trie_leaf_remove(struct trie *t, struct tnode *l)
1407 struct tnode *tp = node_parent(l);
1409 pr_debug("entering trie_leaf_remove(%p)\n", l);
1412 put_child(tp, get_index(l->key, tp), NULL);
1413 trie_rebalance(t, tp);
1415 RCU_INIT_POINTER(t->trie, NULL);
1422 * Caller must hold RTNL.
1424 int fib_table_delete(struct fib_table *tb, struct fib_config *cfg)
1426 struct trie *t = (struct trie *) tb->tb_data;
1427 struct fib_alias *fa, *fa_to_delete;
1428 u8 plen = cfg->fc_dst_len;
1429 u8 tos = cfg->fc_tos;
1430 u8 slen = KEYLENGTH - plen;
1434 if (plen > KEYLENGTH)
1437 key = ntohl(cfg->fc_dst);
1438 mask = ntohl(inet_make_mask(plen));
1443 l = fib_find_node(t, key);
1447 fa = fib_find_alias(&l->leaf, slen, tos, 0);
1452 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1454 fa_to_delete = NULL;
1455 hlist_for_each_entry_from(fa, fa_list) {
1456 struct fib_info *fi = fa->fa_info;
1458 if ((fa->fa_slen != slen) || (fa->fa_tos != tos))
1461 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1462 (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1463 fa->fa_info->fib_scope == cfg->fc_scope) &&
1464 (!cfg->fc_prefsrc ||
1465 fi->fib_prefsrc == cfg->fc_prefsrc) &&
1466 (!cfg->fc_protocol ||
1467 fi->fib_protocol == cfg->fc_protocol) &&
1468 fib_nh_match(cfg, fi) == 0) {
1478 rtmsg_fib(RTM_DELROUTE, htonl(key), fa, plen, tb->tb_id,
1479 &cfg->fc_nlinfo, 0);
1481 fib_remove_alias(l, fa);
1484 tb->tb_num_default--;
1486 if (hlist_empty(&l->leaf))
1487 trie_leaf_remove(t, l);
1489 if (fa->fa_state & FA_S_ACCESSED)
1490 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1492 fib_release_info(fa->fa_info);
1493 alias_free_mem_rcu(fa);
1497 static int trie_flush_leaf(struct tnode *l)
1499 struct hlist_node *tmp;
1500 unsigned char slen = 0;
1501 struct fib_alias *fa;
1504 hlist_for_each_entry_safe(fa, tmp, &l->leaf, fa_list) {
1505 struct fib_info *fi = fa->fa_info;
1507 if (fi && (fi->fib_flags & RTNH_F_DEAD)) {
1508 hlist_del_rcu(&fa->fa_list);
1509 fib_release_info(fa->fa_info);
1510 alias_free_mem_rcu(fa);
1524 /* Scan for the next right leaf starting at node p->child[idx]
1525 * Since we have back pointer, no recursion necessary.
1527 static struct tnode *leaf_walk_rcu(struct tnode *p, struct tnode *c)
1530 unsigned long idx = c ? idx = get_index(c->key, p) + 1 : 0;
1532 while (idx < tnode_child_length(p)) {
1533 c = tnode_get_child_rcu(p, idx++);
1540 /* Rescan start scanning in new node */
1545 /* Node empty, walk back up to parent */
1547 } while ((p = node_parent_rcu(c)) != NULL);
1549 return NULL; /* Root of trie */
1552 static struct tnode *trie_firstleaf(struct trie *t)
1554 struct tnode *n = rcu_dereference_rtnl(t->trie);
1559 if (IS_LEAF(n)) /* trie is just a leaf */
1562 return leaf_walk_rcu(n, NULL);
1565 static struct tnode *trie_nextleaf(struct tnode *l)
1567 struct tnode *p = node_parent_rcu(l);
1570 return NULL; /* trie with just one leaf */
1572 return leaf_walk_rcu(p, l);
1575 static struct tnode *trie_leafindex(struct trie *t, int index)
1577 struct tnode *l = trie_firstleaf(t);
1579 while (l && index-- > 0)
1580 l = trie_nextleaf(l);
1587 * Caller must hold RTNL.
1589 int fib_table_flush(struct fib_table *tb)
1591 struct trie *t = (struct trie *) tb->tb_data;
1592 struct tnode *l, *ll = NULL;
1595 for (l = trie_firstleaf(t); l; l = trie_nextleaf(l)) {
1596 found += trie_flush_leaf(l);
1599 if (hlist_empty(&ll->leaf))
1600 trie_leaf_remove(t, ll);
1602 leaf_pull_suffix(ll);
1609 if (hlist_empty(&ll->leaf))
1610 trie_leaf_remove(t, ll);
1612 leaf_pull_suffix(ll);
1615 pr_debug("trie_flush found=%d\n", found);
1619 void fib_free_table(struct fib_table *tb)
1621 #ifdef CONFIG_IP_FIB_TRIE_STATS
1622 struct trie *t = (struct trie *)tb->tb_data;
1624 free_percpu(t->stats);
1625 #endif /* CONFIG_IP_FIB_TRIE_STATS */
1629 static int fn_trie_dump_leaf(struct tnode *l, struct fib_table *tb,
1630 struct sk_buff *skb, struct netlink_callback *cb)
1632 __be32 xkey = htonl(l->key);
1633 struct fib_alias *fa;
1639 /* rcu_read_lock is hold by caller */
1640 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1646 if (fib_dump_info(skb, NETLINK_CB(cb->skb).portid,
1652 KEYLENGTH - fa->fa_slen,
1654 fa->fa_info, NLM_F_MULTI) < 0) {
1665 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
1666 struct netlink_callback *cb)
1669 struct trie *t = (struct trie *) tb->tb_data;
1670 t_key key = cb->args[2];
1671 int count = cb->args[3];
1674 /* Dump starting at last key.
1675 * Note: 0.0.0.0/0 (ie default) is first key.
1678 l = trie_firstleaf(t);
1680 /* Normally, continue from last key, but if that is missing
1681 * fallback to using slow rescan
1683 l = fib_find_node(t, key);
1685 l = trie_leafindex(t, count);
1689 cb->args[2] = l->key;
1690 if (fn_trie_dump_leaf(l, tb, skb, cb) < 0) {
1691 cb->args[3] = count;
1697 l = trie_nextleaf(l);
1698 memset(&cb->args[4], 0,
1699 sizeof(cb->args) - 4*sizeof(cb->args[0]));
1701 cb->args[3] = count;
1707 void __init fib_trie_init(void)
1709 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
1710 sizeof(struct fib_alias),
1711 0, SLAB_PANIC, NULL);
1713 trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
1714 sizeof(struct tnode),
1715 0, SLAB_PANIC, NULL);
1719 struct fib_table *fib_trie_table(u32 id)
1721 struct fib_table *tb;
1724 tb = kmalloc(sizeof(struct fib_table) + sizeof(struct trie),
1730 tb->tb_default = -1;
1731 tb->tb_num_default = 0;
1733 t = (struct trie *) tb->tb_data;
1734 RCU_INIT_POINTER(t->trie, NULL);
1735 #ifdef CONFIG_IP_FIB_TRIE_STATS
1736 t->stats = alloc_percpu(struct trie_use_stats);
1746 #ifdef CONFIG_PROC_FS
1747 /* Depth first Trie walk iterator */
1748 struct fib_trie_iter {
1749 struct seq_net_private p;
1750 struct fib_table *tb;
1751 struct tnode *tnode;
1756 static struct tnode *fib_trie_get_next(struct fib_trie_iter *iter)
1758 unsigned long cindex = iter->index;
1759 struct tnode *tn = iter->tnode;
1762 /* A single entry routing table */
1766 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
1767 iter->tnode, iter->index, iter->depth);
1769 while (cindex < tnode_child_length(tn)) {
1770 struct tnode *n = tnode_get_child_rcu(tn, cindex);
1775 iter->index = cindex + 1;
1777 /* push down one level */
1788 /* Current node exhausted, pop back up */
1789 p = node_parent_rcu(tn);
1791 cindex = get_index(tn->key, p) + 1;
1801 static struct tnode *fib_trie_get_first(struct fib_trie_iter *iter,
1809 n = rcu_dereference(t->trie);
1826 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
1829 struct fib_trie_iter iter;
1831 memset(s, 0, sizeof(*s));
1834 for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
1836 struct fib_alias *fa;
1839 s->totdepth += iter.depth;
1840 if (iter.depth > s->maxdepth)
1841 s->maxdepth = iter.depth;
1843 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
1847 if (n->bits < MAX_STAT_DEPTH)
1848 s->nodesizes[n->bits]++;
1849 s->nullpointers += n->empty_children;
1856 * This outputs /proc/net/fib_triestats
1858 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
1860 unsigned int i, max, pointers, bytes, avdepth;
1863 avdepth = stat->totdepth*100 / stat->leaves;
1867 seq_printf(seq, "\tAver depth: %u.%02d\n",
1868 avdepth / 100, avdepth % 100);
1869 seq_printf(seq, "\tMax depth: %u\n", stat->maxdepth);
1871 seq_printf(seq, "\tLeaves: %u\n", stat->leaves);
1872 bytes = sizeof(struct tnode) * stat->leaves;
1874 seq_printf(seq, "\tPrefixes: %u\n", stat->prefixes);
1875 bytes += sizeof(struct fib_alias) * stat->prefixes;
1877 seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
1878 bytes += sizeof(struct tnode) * stat->tnodes;
1880 max = MAX_STAT_DEPTH;
1881 while (max > 0 && stat->nodesizes[max-1] == 0)
1885 for (i = 1; i < max; i++)
1886 if (stat->nodesizes[i] != 0) {
1887 seq_printf(seq, " %u: %u", i, stat->nodesizes[i]);
1888 pointers += (1<<i) * stat->nodesizes[i];
1890 seq_putc(seq, '\n');
1891 seq_printf(seq, "\tPointers: %u\n", pointers);
1893 bytes += sizeof(struct tnode *) * pointers;
1894 seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
1895 seq_printf(seq, "Total size: %u kB\n", (bytes + 1023) / 1024);
1898 #ifdef CONFIG_IP_FIB_TRIE_STATS
1899 static void trie_show_usage(struct seq_file *seq,
1900 const struct trie_use_stats __percpu *stats)
1902 struct trie_use_stats s = { 0 };
1905 /* loop through all of the CPUs and gather up the stats */
1906 for_each_possible_cpu(cpu) {
1907 const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
1909 s.gets += pcpu->gets;
1910 s.backtrack += pcpu->backtrack;
1911 s.semantic_match_passed += pcpu->semantic_match_passed;
1912 s.semantic_match_miss += pcpu->semantic_match_miss;
1913 s.null_node_hit += pcpu->null_node_hit;
1914 s.resize_node_skipped += pcpu->resize_node_skipped;
1917 seq_printf(seq, "\nCounters:\n---------\n");
1918 seq_printf(seq, "gets = %u\n", s.gets);
1919 seq_printf(seq, "backtracks = %u\n", s.backtrack);
1920 seq_printf(seq, "semantic match passed = %u\n",
1921 s.semantic_match_passed);
1922 seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
1923 seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
1924 seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
1926 #endif /* CONFIG_IP_FIB_TRIE_STATS */
1928 static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
1930 if (tb->tb_id == RT_TABLE_LOCAL)
1931 seq_puts(seq, "Local:\n");
1932 else if (tb->tb_id == RT_TABLE_MAIN)
1933 seq_puts(seq, "Main:\n");
1935 seq_printf(seq, "Id %d:\n", tb->tb_id);
1939 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
1941 struct net *net = (struct net *)seq->private;
1945 "Basic info: size of leaf:"
1946 " %Zd bytes, size of tnode: %Zd bytes.\n",
1947 sizeof(struct tnode), sizeof(struct tnode));
1949 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
1950 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
1951 struct fib_table *tb;
1953 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
1954 struct trie *t = (struct trie *) tb->tb_data;
1955 struct trie_stat stat;
1960 fib_table_print(seq, tb);
1962 trie_collect_stats(t, &stat);
1963 trie_show_stats(seq, &stat);
1964 #ifdef CONFIG_IP_FIB_TRIE_STATS
1965 trie_show_usage(seq, t->stats);
1973 static int fib_triestat_seq_open(struct inode *inode, struct file *file)
1975 return single_open_net(inode, file, fib_triestat_seq_show);
1978 static const struct file_operations fib_triestat_fops = {
1979 .owner = THIS_MODULE,
1980 .open = fib_triestat_seq_open,
1982 .llseek = seq_lseek,
1983 .release = single_release_net,
1986 static struct tnode *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
1988 struct fib_trie_iter *iter = seq->private;
1989 struct net *net = seq_file_net(seq);
1993 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
1994 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
1995 struct fib_table *tb;
1997 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2000 for (n = fib_trie_get_first(iter,
2001 (struct trie *) tb->tb_data);
2002 n; n = fib_trie_get_next(iter))
2013 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2017 return fib_trie_get_idx(seq, *pos);
2020 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2022 struct fib_trie_iter *iter = seq->private;
2023 struct net *net = seq_file_net(seq);
2024 struct fib_table *tb = iter->tb;
2025 struct hlist_node *tb_node;
2030 /* next node in same table */
2031 n = fib_trie_get_next(iter);
2035 /* walk rest of this hash chain */
2036 h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2037 while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2038 tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2039 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2044 /* new hash chain */
2045 while (++h < FIB_TABLE_HASHSZ) {
2046 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2047 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2048 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2060 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2066 static void seq_indent(struct seq_file *seq, int n)
2072 static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2075 case RT_SCOPE_UNIVERSE: return "universe";
2076 case RT_SCOPE_SITE: return "site";
2077 case RT_SCOPE_LINK: return "link";
2078 case RT_SCOPE_HOST: return "host";
2079 case RT_SCOPE_NOWHERE: return "nowhere";
2081 snprintf(buf, len, "scope=%d", s);
2086 static const char *const rtn_type_names[__RTN_MAX] = {
2087 [RTN_UNSPEC] = "UNSPEC",
2088 [RTN_UNICAST] = "UNICAST",
2089 [RTN_LOCAL] = "LOCAL",
2090 [RTN_BROADCAST] = "BROADCAST",
2091 [RTN_ANYCAST] = "ANYCAST",
2092 [RTN_MULTICAST] = "MULTICAST",
2093 [RTN_BLACKHOLE] = "BLACKHOLE",
2094 [RTN_UNREACHABLE] = "UNREACHABLE",
2095 [RTN_PROHIBIT] = "PROHIBIT",
2096 [RTN_THROW] = "THROW",
2098 [RTN_XRESOLVE] = "XRESOLVE",
2101 static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2103 if (t < __RTN_MAX && rtn_type_names[t])
2104 return rtn_type_names[t];
2105 snprintf(buf, len, "type %u", t);
2109 /* Pretty print the trie */
2110 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2112 const struct fib_trie_iter *iter = seq->private;
2113 struct tnode *n = v;
2115 if (!node_parent_rcu(n))
2116 fib_table_print(seq, iter->tb);
2119 __be32 prf = htonl(n->key);
2121 seq_indent(seq, iter->depth-1);
2122 seq_printf(seq, " +-- %pI4/%zu %u %u %u\n",
2123 &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2124 n->full_children, n->empty_children);
2126 __be32 val = htonl(n->key);
2127 struct fib_alias *fa;
2129 seq_indent(seq, iter->depth);
2130 seq_printf(seq, " |-- %pI4\n", &val);
2132 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2133 char buf1[32], buf2[32];
2135 seq_indent(seq, iter->depth + 1);
2136 seq_printf(seq, " /%zu %s %s",
2137 KEYLENGTH - fa->fa_slen,
2138 rtn_scope(buf1, sizeof(buf1),
2139 fa->fa_info->fib_scope),
2140 rtn_type(buf2, sizeof(buf2),
2143 seq_printf(seq, " tos=%d", fa->fa_tos);
2144 seq_putc(seq, '\n');
2151 static const struct seq_operations fib_trie_seq_ops = {
2152 .start = fib_trie_seq_start,
2153 .next = fib_trie_seq_next,
2154 .stop = fib_trie_seq_stop,
2155 .show = fib_trie_seq_show,
2158 static int fib_trie_seq_open(struct inode *inode, struct file *file)
2160 return seq_open_net(inode, file, &fib_trie_seq_ops,
2161 sizeof(struct fib_trie_iter));
2164 static const struct file_operations fib_trie_fops = {
2165 .owner = THIS_MODULE,
2166 .open = fib_trie_seq_open,
2168 .llseek = seq_lseek,
2169 .release = seq_release_net,
2172 struct fib_route_iter {
2173 struct seq_net_private p;
2174 struct trie *main_trie;
2179 static struct tnode *fib_route_get_idx(struct fib_route_iter *iter, loff_t pos)
2181 struct tnode *l = NULL;
2182 struct trie *t = iter->main_trie;
2184 /* use cache location of last found key */
2185 if (iter->pos > 0 && pos >= iter->pos && (l = fib_find_node(t, iter->key)))
2189 l = trie_firstleaf(t);
2192 while (l && pos-- > 0) {
2194 l = trie_nextleaf(l);
2198 iter->key = pos; /* remember it */
2200 iter->pos = 0; /* forget it */
2205 static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2208 struct fib_route_iter *iter = seq->private;
2209 struct fib_table *tb;
2212 tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2216 iter->main_trie = (struct trie *) tb->tb_data;
2218 return SEQ_START_TOKEN;
2220 return fib_route_get_idx(iter, *pos - 1);
2223 static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2225 struct fib_route_iter *iter = seq->private;
2226 struct tnode *l = v;
2229 if (v == SEQ_START_TOKEN) {
2231 l = trie_firstleaf(iter->main_trie);
2234 l = trie_nextleaf(l);
2244 static void fib_route_seq_stop(struct seq_file *seq, void *v)
2250 static unsigned int fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2252 unsigned int flags = 0;
2254 if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2256 if (fi && fi->fib_nh->nh_gw)
2257 flags |= RTF_GATEWAY;
2258 if (mask == htonl(0xFFFFFFFF))
2265 * This outputs /proc/net/route.
2266 * The format of the file is not supposed to be changed
2267 * and needs to be same as fib_hash output to avoid breaking
2270 static int fib_route_seq_show(struct seq_file *seq, void *v)
2272 struct fib_alias *fa;
2273 struct tnode *l = v;
2276 if (v == SEQ_START_TOKEN) {
2277 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2278 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2283 prefix = htonl(l->key);
2285 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2286 const struct fib_info *fi = fa->fa_info;
2287 __be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2288 unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2290 if ((fa->fa_type == RTN_BROADCAST) ||
2291 (fa->fa_type == RTN_MULTICAST))
2294 seq_setwidth(seq, 127);
2298 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2299 "%d\t%08X\t%d\t%u\t%u",
2300 fi->fib_dev ? fi->fib_dev->name : "*",
2302 fi->fib_nh->nh_gw, flags, 0, 0,
2306 fi->fib_advmss + 40 : 0),
2311 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2312 "%d\t%08X\t%d\t%u\t%u",
2313 prefix, 0, flags, 0, 0, 0,
2322 static const struct seq_operations fib_route_seq_ops = {
2323 .start = fib_route_seq_start,
2324 .next = fib_route_seq_next,
2325 .stop = fib_route_seq_stop,
2326 .show = fib_route_seq_show,
2329 static int fib_route_seq_open(struct inode *inode, struct file *file)
2331 return seq_open_net(inode, file, &fib_route_seq_ops,
2332 sizeof(struct fib_route_iter));
2335 static const struct file_operations fib_route_fops = {
2336 .owner = THIS_MODULE,
2337 .open = fib_route_seq_open,
2339 .llseek = seq_lseek,
2340 .release = seq_release_net,
2343 int __net_init fib_proc_init(struct net *net)
2345 if (!proc_create("fib_trie", S_IRUGO, net->proc_net, &fib_trie_fops))
2348 if (!proc_create("fib_triestat", S_IRUGO, net->proc_net,
2349 &fib_triestat_fops))
2352 if (!proc_create("route", S_IRUGO, net->proc_net, &fib_route_fops))
2358 remove_proc_entry("fib_triestat", net->proc_net);
2360 remove_proc_entry("fib_trie", net->proc_net);
2365 void __net_exit fib_proc_exit(struct net *net)
2367 remove_proc_entry("fib_trie", net->proc_net);
2368 remove_proc_entry("fib_triestat", net->proc_net);
2369 remove_proc_entry("route", net->proc_net);
2372 #endif /* CONFIG_PROC_FS */