net: reintroduce missing rcu_assign_pointer() calls
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / net / ipv4 / fib_trie.c
blob3ce23f9d936395135394d292ac4f3ddc84d43645
1 /*
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 <asm/system.h>
55 #include <linux/bitops.h>
56 #include <linux/types.h>
57 #include <linux/kernel.h>
58 #include <linux/mm.h>
59 #include <linux/string.h>
60 #include <linux/socket.h>
61 #include <linux/sockios.h>
62 #include <linux/errno.h>
63 #include <linux/in.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 <linux/slab.h>
75 #include <linux/prefetch.h>
76 #include <linux/export.h>
77 #include <net/net_namespace.h>
78 #include <net/ip.h>
79 #include <net/protocol.h>
80 #include <net/route.h>
81 #include <net/tcp.h>
82 #include <net/sock.h>
83 #include <net/ip_fib.h>
84 #include "fib_lookup.h"
86 #define MAX_STAT_DEPTH 32
88 #define KEYLENGTH (8*sizeof(t_key))
90 typedef unsigned int t_key;
92 #define T_TNODE 0
93 #define T_LEAF 1
94 #define NODE_TYPE_MASK 0x1UL
95 #define NODE_TYPE(node) ((node)->parent & NODE_TYPE_MASK)
97 #define IS_TNODE(n) (!(n->parent & T_LEAF))
98 #define IS_LEAF(n) (n->parent & T_LEAF)
100 struct rt_trie_node {
101 unsigned long parent;
102 t_key key;
105 struct leaf {
106 unsigned long parent;
107 t_key key;
108 struct hlist_head list;
109 struct rcu_head rcu;
112 struct leaf_info {
113 struct hlist_node hlist;
114 int plen;
115 u32 mask_plen; /* ntohl(inet_make_mask(plen)) */
116 struct list_head falh;
117 struct rcu_head rcu;
120 struct tnode {
121 unsigned long parent;
122 t_key key;
123 unsigned char pos; /* 2log(KEYLENGTH) bits needed */
124 unsigned char bits; /* 2log(KEYLENGTH) bits needed */
125 unsigned int full_children; /* KEYLENGTH bits needed */
126 unsigned int empty_children; /* KEYLENGTH bits needed */
127 union {
128 struct rcu_head rcu;
129 struct work_struct work;
130 struct tnode *tnode_free;
132 struct rt_trie_node __rcu *child[0];
135 #ifdef CONFIG_IP_FIB_TRIE_STATS
136 struct trie_use_stats {
137 unsigned int gets;
138 unsigned int backtrack;
139 unsigned int semantic_match_passed;
140 unsigned int semantic_match_miss;
141 unsigned int null_node_hit;
142 unsigned int resize_node_skipped;
144 #endif
146 struct trie_stat {
147 unsigned int totdepth;
148 unsigned int maxdepth;
149 unsigned int tnodes;
150 unsigned int leaves;
151 unsigned int nullpointers;
152 unsigned int prefixes;
153 unsigned int nodesizes[MAX_STAT_DEPTH];
156 struct trie {
157 struct rt_trie_node __rcu *trie;
158 #ifdef CONFIG_IP_FIB_TRIE_STATS
159 struct trie_use_stats stats;
160 #endif
163 static void put_child(struct trie *t, struct tnode *tn, int i, struct rt_trie_node *n);
164 static void tnode_put_child_reorg(struct tnode *tn, int i, struct rt_trie_node *n,
165 int wasfull);
166 static struct rt_trie_node *resize(struct trie *t, struct tnode *tn);
167 static struct tnode *inflate(struct trie *t, struct tnode *tn);
168 static struct tnode *halve(struct trie *t, struct tnode *tn);
169 /* tnodes to free after resize(); protected by RTNL */
170 static struct tnode *tnode_free_head;
171 static size_t tnode_free_size;
174 * synchronize_rcu after call_rcu for that many pages; it should be especially
175 * useful before resizing the root node with PREEMPT_NONE configs; the value was
176 * obtained experimentally, aiming to avoid visible slowdown.
178 static const int sync_pages = 128;
180 static struct kmem_cache *fn_alias_kmem __read_mostly;
181 static struct kmem_cache *trie_leaf_kmem __read_mostly;
184 * caller must hold RTNL
186 static inline struct tnode *node_parent(const struct rt_trie_node *node)
188 unsigned long parent;
190 parent = rcu_dereference_index_check(node->parent, lockdep_rtnl_is_held());
192 return (struct tnode *)(parent & ~NODE_TYPE_MASK);
196 * caller must hold RCU read lock or RTNL
198 static inline struct tnode *node_parent_rcu(const struct rt_trie_node *node)
200 unsigned long parent;
202 parent = rcu_dereference_index_check(node->parent, rcu_read_lock_held() ||
203 lockdep_rtnl_is_held());
205 return (struct tnode *)(parent & ~NODE_TYPE_MASK);
208 /* Same as rcu_assign_pointer
209 * but that macro() assumes that value is a pointer.
211 static inline void node_set_parent(struct rt_trie_node *node, struct tnode *ptr)
213 smp_wmb();
214 node->parent = (unsigned long)ptr | NODE_TYPE(node);
218 * caller must hold RTNL
220 static inline struct rt_trie_node *tnode_get_child(const struct tnode *tn, unsigned int i)
222 BUG_ON(i >= 1U << tn->bits);
224 return rtnl_dereference(tn->child[i]);
228 * caller must hold RCU read lock or RTNL
230 static inline struct rt_trie_node *tnode_get_child_rcu(const struct tnode *tn, unsigned int i)
232 BUG_ON(i >= 1U << tn->bits);
234 return rcu_dereference_rtnl(tn->child[i]);
237 static inline int tnode_child_length(const struct tnode *tn)
239 return 1 << tn->bits;
242 static inline t_key mask_pfx(t_key k, unsigned int l)
244 return (l == 0) ? 0 : k >> (KEYLENGTH-l) << (KEYLENGTH-l);
247 static inline t_key tkey_extract_bits(t_key a, unsigned int offset, unsigned int bits)
249 if (offset < KEYLENGTH)
250 return ((t_key)(a << offset)) >> (KEYLENGTH - bits);
251 else
252 return 0;
255 static inline int tkey_equals(t_key a, t_key b)
257 return a == b;
260 static inline int tkey_sub_equals(t_key a, int offset, int bits, t_key b)
262 if (bits == 0 || offset >= KEYLENGTH)
263 return 1;
264 bits = bits > KEYLENGTH ? KEYLENGTH : bits;
265 return ((a ^ b) << offset) >> (KEYLENGTH - bits) == 0;
268 static inline int tkey_mismatch(t_key a, int offset, t_key b)
270 t_key diff = a ^ b;
271 int i = offset;
273 if (!diff)
274 return 0;
275 while ((diff << i) >> (KEYLENGTH-1) == 0)
276 i++;
277 return i;
281 To understand this stuff, an understanding of keys and all their bits is
282 necessary. Every node in the trie has a key associated with it, but not
283 all of the bits in that key are significant.
285 Consider a node 'n' and its parent 'tp'.
287 If n is a leaf, every bit in its key is significant. Its presence is
288 necessitated by path compression, since during a tree traversal (when
289 searching for a leaf - unless we are doing an insertion) we will completely
290 ignore all skipped bits we encounter. Thus we need to verify, at the end of
291 a potentially successful search, that we have indeed been walking the
292 correct key path.
294 Note that we can never "miss" the correct key in the tree if present by
295 following the wrong path. Path compression ensures that segments of the key
296 that are the same for all keys with a given prefix are skipped, but the
297 skipped part *is* identical for each node in the subtrie below the skipped
298 bit! trie_insert() in this implementation takes care of that - note the
299 call to tkey_sub_equals() in trie_insert().
301 if n is an internal node - a 'tnode' here, the various parts of its key
302 have many different meanings.
304 Example:
305 _________________________________________________________________
306 | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
307 -----------------------------------------------------------------
308 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
310 _________________________________________________________________
311 | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
312 -----------------------------------------------------------------
313 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
315 tp->pos = 7
316 tp->bits = 3
317 n->pos = 15
318 n->bits = 4
320 First, let's just ignore the bits that come before the parent tp, that is
321 the bits from 0 to (tp->pos-1). They are *known* but at this point we do
322 not use them for anything.
324 The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
325 index into the parent's child array. That is, they will be used to find
326 'n' among tp's children.
328 The bits from (tp->pos + tp->bits) to (n->pos - 1) - "S" - are skipped bits
329 for the node n.
331 All the bits we have seen so far are significant to the node n. The rest
332 of the bits are really not needed or indeed known in n->key.
334 The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
335 n's child array, and will of course be different for each child.
338 The rest of the bits, from (n->pos + n->bits) onward, are completely unknown
339 at this point.
343 static inline void check_tnode(const struct tnode *tn)
345 WARN_ON(tn && tn->pos+tn->bits > 32);
348 static const int halve_threshold = 25;
349 static const int inflate_threshold = 50;
350 static const int halve_threshold_root = 15;
351 static const int inflate_threshold_root = 30;
353 static void __alias_free_mem(struct rcu_head *head)
355 struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
356 kmem_cache_free(fn_alias_kmem, fa);
359 static inline void alias_free_mem_rcu(struct fib_alias *fa)
361 call_rcu(&fa->rcu, __alias_free_mem);
364 static void __leaf_free_rcu(struct rcu_head *head)
366 struct leaf *l = container_of(head, struct leaf, rcu);
367 kmem_cache_free(trie_leaf_kmem, l);
370 static inline void free_leaf(struct leaf *l)
372 call_rcu_bh(&l->rcu, __leaf_free_rcu);
375 static inline void free_leaf_info(struct leaf_info *leaf)
377 kfree_rcu(leaf, rcu);
380 static struct tnode *tnode_alloc(size_t size)
382 if (size <= PAGE_SIZE)
383 return kzalloc(size, GFP_KERNEL);
384 else
385 return vzalloc(size);
388 static void __tnode_vfree(struct work_struct *arg)
390 struct tnode *tn = container_of(arg, struct tnode, work);
391 vfree(tn);
394 static void __tnode_free_rcu(struct rcu_head *head)
396 struct tnode *tn = container_of(head, struct tnode, rcu);
397 size_t size = sizeof(struct tnode) +
398 (sizeof(struct rt_trie_node *) << tn->bits);
400 if (size <= PAGE_SIZE)
401 kfree(tn);
402 else {
403 INIT_WORK(&tn->work, __tnode_vfree);
404 schedule_work(&tn->work);
408 static inline void tnode_free(struct tnode *tn)
410 if (IS_LEAF(tn))
411 free_leaf((struct leaf *) tn);
412 else
413 call_rcu(&tn->rcu, __tnode_free_rcu);
416 static void tnode_free_safe(struct tnode *tn)
418 BUG_ON(IS_LEAF(tn));
419 tn->tnode_free = tnode_free_head;
420 tnode_free_head = tn;
421 tnode_free_size += sizeof(struct tnode) +
422 (sizeof(struct rt_trie_node *) << tn->bits);
425 static void tnode_free_flush(void)
427 struct tnode *tn;
429 while ((tn = tnode_free_head)) {
430 tnode_free_head = tn->tnode_free;
431 tn->tnode_free = NULL;
432 tnode_free(tn);
435 if (tnode_free_size >= PAGE_SIZE * sync_pages) {
436 tnode_free_size = 0;
437 synchronize_rcu();
441 static struct leaf *leaf_new(void)
443 struct leaf *l = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
444 if (l) {
445 l->parent = T_LEAF;
446 INIT_HLIST_HEAD(&l->list);
448 return l;
451 static struct leaf_info *leaf_info_new(int plen)
453 struct leaf_info *li = kmalloc(sizeof(struct leaf_info), GFP_KERNEL);
454 if (li) {
455 li->plen = plen;
456 li->mask_plen = ntohl(inet_make_mask(plen));
457 INIT_LIST_HEAD(&li->falh);
459 return li;
462 static struct tnode *tnode_new(t_key key, int pos, int bits)
464 size_t sz = sizeof(struct tnode) + (sizeof(struct rt_trie_node *) << bits);
465 struct tnode *tn = tnode_alloc(sz);
467 if (tn) {
468 tn->parent = T_TNODE;
469 tn->pos = pos;
470 tn->bits = bits;
471 tn->key = key;
472 tn->full_children = 0;
473 tn->empty_children = 1<<bits;
476 pr_debug("AT %p s=%zu %zu\n", tn, sizeof(struct tnode),
477 sizeof(struct rt_trie_node) << bits);
478 return tn;
482 * Check whether a tnode 'n' is "full", i.e. it is an internal node
483 * and no bits are skipped. See discussion in dyntree paper p. 6
486 static inline int tnode_full(const struct tnode *tn, const struct rt_trie_node *n)
488 if (n == NULL || IS_LEAF(n))
489 return 0;
491 return ((struct tnode *) n)->pos == tn->pos + tn->bits;
494 static inline void put_child(struct trie *t, struct tnode *tn, int i,
495 struct rt_trie_node *n)
497 tnode_put_child_reorg(tn, i, n, -1);
501 * Add a child at position i overwriting the old value.
502 * Update the value of full_children and empty_children.
505 static void tnode_put_child_reorg(struct tnode *tn, int i, struct rt_trie_node *n,
506 int wasfull)
508 struct rt_trie_node *chi = rtnl_dereference(tn->child[i]);
509 int isfull;
511 BUG_ON(i >= 1<<tn->bits);
513 /* update emptyChildren */
514 if (n == NULL && chi != NULL)
515 tn->empty_children++;
516 else if (n != NULL && chi == NULL)
517 tn->empty_children--;
519 /* update fullChildren */
520 if (wasfull == -1)
521 wasfull = tnode_full(tn, chi);
523 isfull = tnode_full(tn, n);
524 if (wasfull && !isfull)
525 tn->full_children--;
526 else if (!wasfull && isfull)
527 tn->full_children++;
529 if (n)
530 node_set_parent(n, tn);
532 rcu_assign_pointer(tn->child[i], n);
535 #define MAX_WORK 10
536 static struct rt_trie_node *resize(struct trie *t, struct tnode *tn)
538 int i;
539 struct tnode *old_tn;
540 int inflate_threshold_use;
541 int halve_threshold_use;
542 int max_work;
544 if (!tn)
545 return NULL;
547 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
548 tn, inflate_threshold, halve_threshold);
550 /* No children */
551 if (tn->empty_children == tnode_child_length(tn)) {
552 tnode_free_safe(tn);
553 return NULL;
555 /* One child */
556 if (tn->empty_children == tnode_child_length(tn) - 1)
557 goto one_child;
559 * Double as long as the resulting node has a number of
560 * nonempty nodes that are above the threshold.
564 * From "Implementing a dynamic compressed trie" by Stefan Nilsson of
565 * the Helsinki University of Technology and Matti Tikkanen of Nokia
566 * Telecommunications, page 6:
567 * "A node is doubled if the ratio of non-empty children to all
568 * children in the *doubled* node is at least 'high'."
570 * 'high' in this instance is the variable 'inflate_threshold'. It
571 * is expressed as a percentage, so we multiply it with
572 * tnode_child_length() and instead of multiplying by 2 (since the
573 * child array will be doubled by inflate()) and multiplying
574 * the left-hand side by 100 (to handle the percentage thing) we
575 * multiply the left-hand side by 50.
577 * The left-hand side may look a bit weird: tnode_child_length(tn)
578 * - tn->empty_children is of course the number of non-null children
579 * in the current node. tn->full_children is the number of "full"
580 * children, that is non-null tnodes with a skip value of 0.
581 * All of those will be doubled in the resulting inflated tnode, so
582 * we just count them one extra time here.
584 * A clearer way to write this would be:
586 * to_be_doubled = tn->full_children;
587 * not_to_be_doubled = tnode_child_length(tn) - tn->empty_children -
588 * tn->full_children;
590 * new_child_length = tnode_child_length(tn) * 2;
592 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
593 * new_child_length;
594 * if (new_fill_factor >= inflate_threshold)
596 * ...and so on, tho it would mess up the while () loop.
598 * anyway,
599 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
600 * inflate_threshold
602 * avoid a division:
603 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
604 * inflate_threshold * new_child_length
606 * expand not_to_be_doubled and to_be_doubled, and shorten:
607 * 100 * (tnode_child_length(tn) - tn->empty_children +
608 * tn->full_children) >= inflate_threshold * new_child_length
610 * expand new_child_length:
611 * 100 * (tnode_child_length(tn) - tn->empty_children +
612 * tn->full_children) >=
613 * inflate_threshold * tnode_child_length(tn) * 2
615 * shorten again:
616 * 50 * (tn->full_children + tnode_child_length(tn) -
617 * tn->empty_children) >= inflate_threshold *
618 * tnode_child_length(tn)
622 check_tnode(tn);
624 /* Keep root node larger */
626 if (!node_parent((struct rt_trie_node *)tn)) {
627 inflate_threshold_use = inflate_threshold_root;
628 halve_threshold_use = halve_threshold_root;
629 } else {
630 inflate_threshold_use = inflate_threshold;
631 halve_threshold_use = halve_threshold;
634 max_work = MAX_WORK;
635 while ((tn->full_children > 0 && max_work-- &&
636 50 * (tn->full_children + tnode_child_length(tn)
637 - tn->empty_children)
638 >= inflate_threshold_use * tnode_child_length(tn))) {
640 old_tn = tn;
641 tn = inflate(t, tn);
643 if (IS_ERR(tn)) {
644 tn = old_tn;
645 #ifdef CONFIG_IP_FIB_TRIE_STATS
646 t->stats.resize_node_skipped++;
647 #endif
648 break;
652 check_tnode(tn);
654 /* Return if at least one inflate is run */
655 if (max_work != MAX_WORK)
656 return (struct rt_trie_node *) tn;
659 * Halve as long as the number of empty children in this
660 * node is above threshold.
663 max_work = MAX_WORK;
664 while (tn->bits > 1 && max_work-- &&
665 100 * (tnode_child_length(tn) - tn->empty_children) <
666 halve_threshold_use * tnode_child_length(tn)) {
668 old_tn = tn;
669 tn = halve(t, tn);
670 if (IS_ERR(tn)) {
671 tn = old_tn;
672 #ifdef CONFIG_IP_FIB_TRIE_STATS
673 t->stats.resize_node_skipped++;
674 #endif
675 break;
680 /* Only one child remains */
681 if (tn->empty_children == tnode_child_length(tn) - 1) {
682 one_child:
683 for (i = 0; i < tnode_child_length(tn); i++) {
684 struct rt_trie_node *n;
686 n = rtnl_dereference(tn->child[i]);
687 if (!n)
688 continue;
690 /* compress one level */
692 node_set_parent(n, NULL);
693 tnode_free_safe(tn);
694 return n;
697 return (struct rt_trie_node *) tn;
701 static void tnode_clean_free(struct tnode *tn)
703 int i;
704 struct tnode *tofree;
706 for (i = 0; i < tnode_child_length(tn); i++) {
707 tofree = (struct tnode *)rtnl_dereference(tn->child[i]);
708 if (tofree)
709 tnode_free(tofree);
711 tnode_free(tn);
714 static struct tnode *inflate(struct trie *t, struct tnode *tn)
716 struct tnode *oldtnode = tn;
717 int olen = tnode_child_length(tn);
718 int i;
720 pr_debug("In inflate\n");
722 tn = tnode_new(oldtnode->key, oldtnode->pos, oldtnode->bits + 1);
724 if (!tn)
725 return ERR_PTR(-ENOMEM);
728 * Preallocate and store tnodes before the actual work so we
729 * don't get into an inconsistent state if memory allocation
730 * fails. In case of failure we return the oldnode and inflate
731 * of tnode is ignored.
734 for (i = 0; i < olen; i++) {
735 struct tnode *inode;
737 inode = (struct tnode *) tnode_get_child(oldtnode, i);
738 if (inode &&
739 IS_TNODE(inode) &&
740 inode->pos == oldtnode->pos + oldtnode->bits &&
741 inode->bits > 1) {
742 struct tnode *left, *right;
743 t_key m = ~0U << (KEYLENGTH - 1) >> inode->pos;
745 left = tnode_new(inode->key&(~m), inode->pos + 1,
746 inode->bits - 1);
747 if (!left)
748 goto nomem;
750 right = tnode_new(inode->key|m, inode->pos + 1,
751 inode->bits - 1);
753 if (!right) {
754 tnode_free(left);
755 goto nomem;
758 put_child(t, tn, 2*i, (struct rt_trie_node *) left);
759 put_child(t, tn, 2*i+1, (struct rt_trie_node *) right);
763 for (i = 0; i < olen; i++) {
764 struct tnode *inode;
765 struct rt_trie_node *node = tnode_get_child(oldtnode, i);
766 struct tnode *left, *right;
767 int size, j;
769 /* An empty child */
770 if (node == NULL)
771 continue;
773 /* A leaf or an internal node with skipped bits */
775 if (IS_LEAF(node) || ((struct tnode *) node)->pos >
776 tn->pos + tn->bits - 1) {
777 if (tkey_extract_bits(node->key,
778 oldtnode->pos + oldtnode->bits,
779 1) == 0)
780 put_child(t, tn, 2*i, node);
781 else
782 put_child(t, tn, 2*i+1, node);
783 continue;
786 /* An internal node with two children */
787 inode = (struct tnode *) node;
789 if (inode->bits == 1) {
790 put_child(t, tn, 2*i, rtnl_dereference(inode->child[0]));
791 put_child(t, tn, 2*i+1, rtnl_dereference(inode->child[1]));
793 tnode_free_safe(inode);
794 continue;
797 /* An internal node with more than two children */
799 /* We will replace this node 'inode' with two new
800 * ones, 'left' and 'right', each with half of the
801 * original children. The two new nodes will have
802 * a position one bit further down the key and this
803 * means that the "significant" part of their keys
804 * (see the discussion near the top of this file)
805 * will differ by one bit, which will be "0" in
806 * left's key and "1" in right's key. Since we are
807 * moving the key position by one step, the bit that
808 * we are moving away from - the bit at position
809 * (inode->pos) - is the one that will differ between
810 * left and right. So... we synthesize that bit in the
811 * two new keys.
812 * The mask 'm' below will be a single "one" bit at
813 * the position (inode->pos)
816 /* Use the old key, but set the new significant
817 * bit to zero.
820 left = (struct tnode *) tnode_get_child(tn, 2*i);
821 put_child(t, tn, 2*i, NULL);
823 BUG_ON(!left);
825 right = (struct tnode *) tnode_get_child(tn, 2*i+1);
826 put_child(t, tn, 2*i+1, NULL);
828 BUG_ON(!right);
830 size = tnode_child_length(left);
831 for (j = 0; j < size; j++) {
832 put_child(t, left, j, rtnl_dereference(inode->child[j]));
833 put_child(t, right, j, rtnl_dereference(inode->child[j + size]));
835 put_child(t, tn, 2*i, resize(t, left));
836 put_child(t, tn, 2*i+1, resize(t, right));
838 tnode_free_safe(inode);
840 tnode_free_safe(oldtnode);
841 return tn;
842 nomem:
843 tnode_clean_free(tn);
844 return ERR_PTR(-ENOMEM);
847 static struct tnode *halve(struct trie *t, struct tnode *tn)
849 struct tnode *oldtnode = tn;
850 struct rt_trie_node *left, *right;
851 int i;
852 int olen = tnode_child_length(tn);
854 pr_debug("In halve\n");
856 tn = tnode_new(oldtnode->key, oldtnode->pos, oldtnode->bits - 1);
858 if (!tn)
859 return ERR_PTR(-ENOMEM);
862 * Preallocate and store tnodes before the actual work so we
863 * don't get into an inconsistent state if memory allocation
864 * fails. In case of failure we return the oldnode and halve
865 * of tnode is ignored.
868 for (i = 0; i < olen; i += 2) {
869 left = tnode_get_child(oldtnode, i);
870 right = tnode_get_child(oldtnode, i+1);
872 /* Two nonempty children */
873 if (left && right) {
874 struct tnode *newn;
876 newn = tnode_new(left->key, tn->pos + tn->bits, 1);
878 if (!newn)
879 goto nomem;
881 put_child(t, tn, i/2, (struct rt_trie_node *)newn);
886 for (i = 0; i < olen; i += 2) {
887 struct tnode *newBinNode;
889 left = tnode_get_child(oldtnode, i);
890 right = tnode_get_child(oldtnode, i+1);
892 /* At least one of the children is empty */
893 if (left == NULL) {
894 if (right == NULL) /* Both are empty */
895 continue;
896 put_child(t, tn, i/2, right);
897 continue;
900 if (right == NULL) {
901 put_child(t, tn, i/2, left);
902 continue;
905 /* Two nonempty children */
906 newBinNode = (struct tnode *) tnode_get_child(tn, i/2);
907 put_child(t, tn, i/2, NULL);
908 put_child(t, newBinNode, 0, left);
909 put_child(t, newBinNode, 1, right);
910 put_child(t, tn, i/2, resize(t, newBinNode));
912 tnode_free_safe(oldtnode);
913 return tn;
914 nomem:
915 tnode_clean_free(tn);
916 return ERR_PTR(-ENOMEM);
919 /* readside must use rcu_read_lock currently dump routines
920 via get_fa_head and dump */
922 static struct leaf_info *find_leaf_info(struct leaf *l, int plen)
924 struct hlist_head *head = &l->list;
925 struct hlist_node *node;
926 struct leaf_info *li;
928 hlist_for_each_entry_rcu(li, node, head, hlist)
929 if (li->plen == plen)
930 return li;
932 return NULL;
935 static inline struct list_head *get_fa_head(struct leaf *l, int plen)
937 struct leaf_info *li = find_leaf_info(l, plen);
939 if (!li)
940 return NULL;
942 return &li->falh;
945 static void insert_leaf_info(struct hlist_head *head, struct leaf_info *new)
947 struct leaf_info *li = NULL, *last = NULL;
948 struct hlist_node *node;
950 if (hlist_empty(head)) {
951 hlist_add_head_rcu(&new->hlist, head);
952 } else {
953 hlist_for_each_entry(li, node, head, hlist) {
954 if (new->plen > li->plen)
955 break;
957 last = li;
959 if (last)
960 hlist_add_after_rcu(&last->hlist, &new->hlist);
961 else
962 hlist_add_before_rcu(&new->hlist, &li->hlist);
966 /* rcu_read_lock needs to be hold by caller from readside */
968 static struct leaf *
969 fib_find_node(struct trie *t, u32 key)
971 int pos;
972 struct tnode *tn;
973 struct rt_trie_node *n;
975 pos = 0;
976 n = rcu_dereference_rtnl(t->trie);
978 while (n != NULL && NODE_TYPE(n) == T_TNODE) {
979 tn = (struct tnode *) n;
981 check_tnode(tn);
983 if (tkey_sub_equals(tn->key, pos, tn->pos-pos, key)) {
984 pos = tn->pos + tn->bits;
985 n = tnode_get_child_rcu(tn,
986 tkey_extract_bits(key,
987 tn->pos,
988 tn->bits));
989 } else
990 break;
992 /* Case we have found a leaf. Compare prefixes */
994 if (n != NULL && IS_LEAF(n) && tkey_equals(key, n->key))
995 return (struct leaf *)n;
997 return NULL;
1000 static void trie_rebalance(struct trie *t, struct tnode *tn)
1002 int wasfull;
1003 t_key cindex, key;
1004 struct tnode *tp;
1006 key = tn->key;
1008 while (tn != NULL && (tp = node_parent((struct rt_trie_node *)tn)) != NULL) {
1009 cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1010 wasfull = tnode_full(tp, tnode_get_child(tp, cindex));
1011 tn = (struct tnode *) resize(t, (struct tnode *)tn);
1013 tnode_put_child_reorg((struct tnode *)tp, cindex,
1014 (struct rt_trie_node *)tn, wasfull);
1016 tp = node_parent((struct rt_trie_node *) tn);
1017 if (!tp)
1018 rcu_assign_pointer(t->trie, (struct rt_trie_node *)tn);
1020 tnode_free_flush();
1021 if (!tp)
1022 break;
1023 tn = tp;
1026 /* Handle last (top) tnode */
1027 if (IS_TNODE(tn))
1028 tn = (struct tnode *)resize(t, (struct tnode *)tn);
1030 rcu_assign_pointer(t->trie, (struct rt_trie_node *)tn);
1031 tnode_free_flush();
1034 /* only used from updater-side */
1036 static struct list_head *fib_insert_node(struct trie *t, u32 key, int plen)
1038 int pos, newpos;
1039 struct tnode *tp = NULL, *tn = NULL;
1040 struct rt_trie_node *n;
1041 struct leaf *l;
1042 int missbit;
1043 struct list_head *fa_head = NULL;
1044 struct leaf_info *li;
1045 t_key cindex;
1047 pos = 0;
1048 n = rtnl_dereference(t->trie);
1050 /* If we point to NULL, stop. Either the tree is empty and we should
1051 * just put a new leaf in if, or we have reached an empty child slot,
1052 * and we should just put our new leaf in that.
1053 * If we point to a T_TNODE, check if it matches our key. Note that
1054 * a T_TNODE might be skipping any number of bits - its 'pos' need
1055 * not be the parent's 'pos'+'bits'!
1057 * If it does match the current key, get pos/bits from it, extract
1058 * the index from our key, push the T_TNODE and walk the tree.
1060 * If it doesn't, we have to replace it with a new T_TNODE.
1062 * If we point to a T_LEAF, it might or might not have the same key
1063 * as we do. If it does, just change the value, update the T_LEAF's
1064 * value, and return it.
1065 * If it doesn't, we need to replace it with a T_TNODE.
1068 while (n != NULL && NODE_TYPE(n) == T_TNODE) {
1069 tn = (struct tnode *) n;
1071 check_tnode(tn);
1073 if (tkey_sub_equals(tn->key, pos, tn->pos-pos, key)) {
1074 tp = tn;
1075 pos = tn->pos + tn->bits;
1076 n = tnode_get_child(tn,
1077 tkey_extract_bits(key,
1078 tn->pos,
1079 tn->bits));
1081 BUG_ON(n && node_parent(n) != tn);
1082 } else
1083 break;
1087 * n ----> NULL, LEAF or TNODE
1089 * tp is n's (parent) ----> NULL or TNODE
1092 BUG_ON(tp && IS_LEAF(tp));
1094 /* Case 1: n is a leaf. Compare prefixes */
1096 if (n != NULL && IS_LEAF(n) && tkey_equals(key, n->key)) {
1097 l = (struct leaf *) n;
1098 li = leaf_info_new(plen);
1100 if (!li)
1101 return NULL;
1103 fa_head = &li->falh;
1104 insert_leaf_info(&l->list, li);
1105 goto done;
1107 l = leaf_new();
1109 if (!l)
1110 return NULL;
1112 l->key = key;
1113 li = leaf_info_new(plen);
1115 if (!li) {
1116 free_leaf(l);
1117 return NULL;
1120 fa_head = &li->falh;
1121 insert_leaf_info(&l->list, li);
1123 if (t->trie && n == NULL) {
1124 /* Case 2: n is NULL, and will just insert a new leaf */
1126 node_set_parent((struct rt_trie_node *)l, tp);
1128 cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1129 put_child(t, (struct tnode *)tp, cindex, (struct rt_trie_node *)l);
1130 } else {
1131 /* Case 3: n is a LEAF or a TNODE and the key doesn't match. */
1133 * Add a new tnode here
1134 * first tnode need some special handling
1137 if (tp)
1138 pos = tp->pos+tp->bits;
1139 else
1140 pos = 0;
1142 if (n) {
1143 newpos = tkey_mismatch(key, pos, n->key);
1144 tn = tnode_new(n->key, newpos, 1);
1145 } else {
1146 newpos = 0;
1147 tn = tnode_new(key, newpos, 1); /* First tnode */
1150 if (!tn) {
1151 free_leaf_info(li);
1152 free_leaf(l);
1153 return NULL;
1156 node_set_parent((struct rt_trie_node *)tn, tp);
1158 missbit = tkey_extract_bits(key, newpos, 1);
1159 put_child(t, tn, missbit, (struct rt_trie_node *)l);
1160 put_child(t, tn, 1-missbit, n);
1162 if (tp) {
1163 cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1164 put_child(t, (struct tnode *)tp, cindex,
1165 (struct rt_trie_node *)tn);
1166 } else {
1167 rcu_assign_pointer(t->trie, (struct rt_trie_node *)tn);
1168 tp = tn;
1172 if (tp && tp->pos + tp->bits > 32)
1173 pr_warning("fib_trie"
1174 " tp=%p pos=%d, bits=%d, key=%0x plen=%d\n",
1175 tp, tp->pos, tp->bits, key, plen);
1177 /* Rebalance the trie */
1179 trie_rebalance(t, tp);
1180 done:
1181 return fa_head;
1185 * Caller must hold RTNL.
1187 int fib_table_insert(struct fib_table *tb, struct fib_config *cfg)
1189 struct trie *t = (struct trie *) tb->tb_data;
1190 struct fib_alias *fa, *new_fa;
1191 struct list_head *fa_head = NULL;
1192 struct fib_info *fi;
1193 int plen = cfg->fc_dst_len;
1194 u8 tos = cfg->fc_tos;
1195 u32 key, mask;
1196 int err;
1197 struct leaf *l;
1199 if (plen > 32)
1200 return -EINVAL;
1202 key = ntohl(cfg->fc_dst);
1204 pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1206 mask = ntohl(inet_make_mask(plen));
1208 if (key & ~mask)
1209 return -EINVAL;
1211 key = key & mask;
1213 fi = fib_create_info(cfg);
1214 if (IS_ERR(fi)) {
1215 err = PTR_ERR(fi);
1216 goto err;
1219 l = fib_find_node(t, key);
1220 fa = NULL;
1222 if (l) {
1223 fa_head = get_fa_head(l, plen);
1224 fa = fib_find_alias(fa_head, tos, fi->fib_priority);
1227 /* Now fa, if non-NULL, points to the first fib alias
1228 * with the same keys [prefix,tos,priority], if such key already
1229 * exists or to the node before which we will insert new one.
1231 * If fa is NULL, we will need to allocate a new one and
1232 * insert to the head of f.
1234 * If f is NULL, no fib node matched the destination key
1235 * and we need to allocate a new one of those as well.
1238 if (fa && fa->fa_tos == tos &&
1239 fa->fa_info->fib_priority == fi->fib_priority) {
1240 struct fib_alias *fa_first, *fa_match;
1242 err = -EEXIST;
1243 if (cfg->fc_nlflags & NLM_F_EXCL)
1244 goto out;
1246 /* We have 2 goals:
1247 * 1. Find exact match for type, scope, fib_info to avoid
1248 * duplicate routes
1249 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1251 fa_match = NULL;
1252 fa_first = fa;
1253 fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
1254 list_for_each_entry_continue(fa, fa_head, fa_list) {
1255 if (fa->fa_tos != tos)
1256 break;
1257 if (fa->fa_info->fib_priority != fi->fib_priority)
1258 break;
1259 if (fa->fa_type == cfg->fc_type &&
1260 fa->fa_info == fi) {
1261 fa_match = fa;
1262 break;
1266 if (cfg->fc_nlflags & NLM_F_REPLACE) {
1267 struct fib_info *fi_drop;
1268 u8 state;
1270 fa = fa_first;
1271 if (fa_match) {
1272 if (fa == fa_match)
1273 err = 0;
1274 goto out;
1276 err = -ENOBUFS;
1277 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1278 if (new_fa == NULL)
1279 goto out;
1281 fi_drop = fa->fa_info;
1282 new_fa->fa_tos = fa->fa_tos;
1283 new_fa->fa_info = fi;
1284 new_fa->fa_type = cfg->fc_type;
1285 state = fa->fa_state;
1286 new_fa->fa_state = state & ~FA_S_ACCESSED;
1288 list_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1289 alias_free_mem_rcu(fa);
1291 fib_release_info(fi_drop);
1292 if (state & FA_S_ACCESSED)
1293 rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1294 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1295 tb->tb_id, &cfg->fc_nlinfo, NLM_F_REPLACE);
1297 goto succeeded;
1299 /* Error if we find a perfect match which
1300 * uses the same scope, type, and nexthop
1301 * information.
1303 if (fa_match)
1304 goto out;
1306 if (!(cfg->fc_nlflags & NLM_F_APPEND))
1307 fa = fa_first;
1309 err = -ENOENT;
1310 if (!(cfg->fc_nlflags & NLM_F_CREATE))
1311 goto out;
1313 err = -ENOBUFS;
1314 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1315 if (new_fa == NULL)
1316 goto out;
1318 new_fa->fa_info = fi;
1319 new_fa->fa_tos = tos;
1320 new_fa->fa_type = cfg->fc_type;
1321 new_fa->fa_state = 0;
1323 * Insert new entry to the list.
1326 if (!fa_head) {
1327 fa_head = fib_insert_node(t, key, plen);
1328 if (unlikely(!fa_head)) {
1329 err = -ENOMEM;
1330 goto out_free_new_fa;
1334 if (!plen)
1335 tb->tb_num_default++;
1337 list_add_tail_rcu(&new_fa->fa_list,
1338 (fa ? &fa->fa_list : fa_head));
1340 rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1341 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, tb->tb_id,
1342 &cfg->fc_nlinfo, 0);
1343 succeeded:
1344 return 0;
1346 out_free_new_fa:
1347 kmem_cache_free(fn_alias_kmem, new_fa);
1348 out:
1349 fib_release_info(fi);
1350 err:
1351 return err;
1354 /* should be called with rcu_read_lock */
1355 static int check_leaf(struct fib_table *tb, struct trie *t, struct leaf *l,
1356 t_key key, const struct flowi4 *flp,
1357 struct fib_result *res, int fib_flags)
1359 struct leaf_info *li;
1360 struct hlist_head *hhead = &l->list;
1361 struct hlist_node *node;
1363 hlist_for_each_entry_rcu(li, node, hhead, hlist) {
1364 struct fib_alias *fa;
1366 if (l->key != (key & li->mask_plen))
1367 continue;
1369 list_for_each_entry_rcu(fa, &li->falh, fa_list) {
1370 struct fib_info *fi = fa->fa_info;
1371 int nhsel, err;
1373 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1374 continue;
1375 if (fa->fa_info->fib_scope < flp->flowi4_scope)
1376 continue;
1377 fib_alias_accessed(fa);
1378 err = fib_props[fa->fa_type].error;
1379 if (err) {
1380 #ifdef CONFIG_IP_FIB_TRIE_STATS
1381 t->stats.semantic_match_passed++;
1382 #endif
1383 return err;
1385 if (fi->fib_flags & RTNH_F_DEAD)
1386 continue;
1387 for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
1388 const struct fib_nh *nh = &fi->fib_nh[nhsel];
1390 if (nh->nh_flags & RTNH_F_DEAD)
1391 continue;
1392 if (flp->flowi4_oif && flp->flowi4_oif != nh->nh_oif)
1393 continue;
1395 #ifdef CONFIG_IP_FIB_TRIE_STATS
1396 t->stats.semantic_match_passed++;
1397 #endif
1398 res->prefixlen = li->plen;
1399 res->nh_sel = nhsel;
1400 res->type = fa->fa_type;
1401 res->scope = fa->fa_info->fib_scope;
1402 res->fi = fi;
1403 res->table = tb;
1404 res->fa_head = &li->falh;
1405 if (!(fib_flags & FIB_LOOKUP_NOREF))
1406 atomic_inc(&fi->fib_clntref);
1407 return 0;
1411 #ifdef CONFIG_IP_FIB_TRIE_STATS
1412 t->stats.semantic_match_miss++;
1413 #endif
1416 return 1;
1419 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1420 struct fib_result *res, int fib_flags)
1422 struct trie *t = (struct trie *) tb->tb_data;
1423 int ret;
1424 struct rt_trie_node *n;
1425 struct tnode *pn;
1426 unsigned int pos, bits;
1427 t_key key = ntohl(flp->daddr);
1428 unsigned int chopped_off;
1429 t_key cindex = 0;
1430 unsigned int current_prefix_length = KEYLENGTH;
1431 struct tnode *cn;
1432 t_key pref_mismatch;
1434 rcu_read_lock();
1436 n = rcu_dereference(t->trie);
1437 if (!n)
1438 goto failed;
1440 #ifdef CONFIG_IP_FIB_TRIE_STATS
1441 t->stats.gets++;
1442 #endif
1444 /* Just a leaf? */
1445 if (IS_LEAF(n)) {
1446 ret = check_leaf(tb, t, (struct leaf *)n, key, flp, res, fib_flags);
1447 goto found;
1450 pn = (struct tnode *) n;
1451 chopped_off = 0;
1453 while (pn) {
1454 pos = pn->pos;
1455 bits = pn->bits;
1457 if (!chopped_off)
1458 cindex = tkey_extract_bits(mask_pfx(key, current_prefix_length),
1459 pos, bits);
1461 n = tnode_get_child_rcu(pn, cindex);
1463 if (n == NULL) {
1464 #ifdef CONFIG_IP_FIB_TRIE_STATS
1465 t->stats.null_node_hit++;
1466 #endif
1467 goto backtrace;
1470 if (IS_LEAF(n)) {
1471 ret = check_leaf(tb, t, (struct leaf *)n, key, flp, res, fib_flags);
1472 if (ret > 0)
1473 goto backtrace;
1474 goto found;
1477 cn = (struct tnode *)n;
1480 * It's a tnode, and we can do some extra checks here if we
1481 * like, to avoid descending into a dead-end branch.
1482 * This tnode is in the parent's child array at index
1483 * key[p_pos..p_pos+p_bits] but potentially with some bits
1484 * chopped off, so in reality the index may be just a
1485 * subprefix, padded with zero at the end.
1486 * We can also take a look at any skipped bits in this
1487 * tnode - everything up to p_pos is supposed to be ok,
1488 * and the non-chopped bits of the index (se previous
1489 * paragraph) are also guaranteed ok, but the rest is
1490 * considered unknown.
1492 * The skipped bits are key[pos+bits..cn->pos].
1495 /* If current_prefix_length < pos+bits, we are already doing
1496 * actual prefix matching, which means everything from
1497 * pos+(bits-chopped_off) onward must be zero along some
1498 * branch of this subtree - otherwise there is *no* valid
1499 * prefix present. Here we can only check the skipped
1500 * bits. Remember, since we have already indexed into the
1501 * parent's child array, we know that the bits we chopped of
1502 * *are* zero.
1505 /* NOTA BENE: Checking only skipped bits
1506 for the new node here */
1508 if (current_prefix_length < pos+bits) {
1509 if (tkey_extract_bits(cn->key, current_prefix_length,
1510 cn->pos - current_prefix_length)
1511 || !(cn->child[0]))
1512 goto backtrace;
1516 * If chopped_off=0, the index is fully validated and we
1517 * only need to look at the skipped bits for this, the new,
1518 * tnode. What we actually want to do is to find out if
1519 * these skipped bits match our key perfectly, or if we will
1520 * have to count on finding a matching prefix further down,
1521 * because if we do, we would like to have some way of
1522 * verifying the existence of such a prefix at this point.
1525 /* The only thing we can do at this point is to verify that
1526 * any such matching prefix can indeed be a prefix to our
1527 * key, and if the bits in the node we are inspecting that
1528 * do not match our key are not ZERO, this cannot be true.
1529 * Thus, find out where there is a mismatch (before cn->pos)
1530 * and verify that all the mismatching bits are zero in the
1531 * new tnode's key.
1535 * Note: We aren't very concerned about the piece of
1536 * the key that precede pn->pos+pn->bits, since these
1537 * have already been checked. The bits after cn->pos
1538 * aren't checked since these are by definition
1539 * "unknown" at this point. Thus, what we want to see
1540 * is if we are about to enter the "prefix matching"
1541 * state, and in that case verify that the skipped
1542 * bits that will prevail throughout this subtree are
1543 * zero, as they have to be if we are to find a
1544 * matching prefix.
1547 pref_mismatch = mask_pfx(cn->key ^ key, cn->pos);
1550 * In short: If skipped bits in this node do not match
1551 * the search key, enter the "prefix matching"
1552 * state.directly.
1554 if (pref_mismatch) {
1555 int mp = KEYLENGTH - fls(pref_mismatch);
1557 if (tkey_extract_bits(cn->key, mp, cn->pos - mp) != 0)
1558 goto backtrace;
1560 if (current_prefix_length >= cn->pos)
1561 current_prefix_length = mp;
1564 pn = (struct tnode *)n; /* Descend */
1565 chopped_off = 0;
1566 continue;
1568 backtrace:
1569 chopped_off++;
1571 /* As zero don't change the child key (cindex) */
1572 while ((chopped_off <= pn->bits)
1573 && !(cindex & (1<<(chopped_off-1))))
1574 chopped_off++;
1576 /* Decrease current_... with bits chopped off */
1577 if (current_prefix_length > pn->pos + pn->bits - chopped_off)
1578 current_prefix_length = pn->pos + pn->bits
1579 - chopped_off;
1582 * Either we do the actual chop off according or if we have
1583 * chopped off all bits in this tnode walk up to our parent.
1586 if (chopped_off <= pn->bits) {
1587 cindex &= ~(1 << (chopped_off-1));
1588 } else {
1589 struct tnode *parent = node_parent_rcu((struct rt_trie_node *) pn);
1590 if (!parent)
1591 goto failed;
1593 /* Get Child's index */
1594 cindex = tkey_extract_bits(pn->key, parent->pos, parent->bits);
1595 pn = parent;
1596 chopped_off = 0;
1598 #ifdef CONFIG_IP_FIB_TRIE_STATS
1599 t->stats.backtrack++;
1600 #endif
1601 goto backtrace;
1604 failed:
1605 ret = 1;
1606 found:
1607 rcu_read_unlock();
1608 return ret;
1612 * Remove the leaf and return parent.
1614 static void trie_leaf_remove(struct trie *t, struct leaf *l)
1616 struct tnode *tp = node_parent((struct rt_trie_node *) l);
1618 pr_debug("entering trie_leaf_remove(%p)\n", l);
1620 if (tp) {
1621 t_key cindex = tkey_extract_bits(l->key, tp->pos, tp->bits);
1622 put_child(t, (struct tnode *)tp, cindex, NULL);
1623 trie_rebalance(t, tp);
1624 } else
1625 RCU_INIT_POINTER(t->trie, NULL);
1627 free_leaf(l);
1631 * Caller must hold RTNL.
1633 int fib_table_delete(struct fib_table *tb, struct fib_config *cfg)
1635 struct trie *t = (struct trie *) tb->tb_data;
1636 u32 key, mask;
1637 int plen = cfg->fc_dst_len;
1638 u8 tos = cfg->fc_tos;
1639 struct fib_alias *fa, *fa_to_delete;
1640 struct list_head *fa_head;
1641 struct leaf *l;
1642 struct leaf_info *li;
1644 if (plen > 32)
1645 return -EINVAL;
1647 key = ntohl(cfg->fc_dst);
1648 mask = ntohl(inet_make_mask(plen));
1650 if (key & ~mask)
1651 return -EINVAL;
1653 key = key & mask;
1654 l = fib_find_node(t, key);
1656 if (!l)
1657 return -ESRCH;
1659 fa_head = get_fa_head(l, plen);
1660 fa = fib_find_alias(fa_head, tos, 0);
1662 if (!fa)
1663 return -ESRCH;
1665 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1667 fa_to_delete = NULL;
1668 fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
1669 list_for_each_entry_continue(fa, fa_head, fa_list) {
1670 struct fib_info *fi = fa->fa_info;
1672 if (fa->fa_tos != tos)
1673 break;
1675 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1676 (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1677 fa->fa_info->fib_scope == cfg->fc_scope) &&
1678 (!cfg->fc_prefsrc ||
1679 fi->fib_prefsrc == cfg->fc_prefsrc) &&
1680 (!cfg->fc_protocol ||
1681 fi->fib_protocol == cfg->fc_protocol) &&
1682 fib_nh_match(cfg, fi) == 0) {
1683 fa_to_delete = fa;
1684 break;
1688 if (!fa_to_delete)
1689 return -ESRCH;
1691 fa = fa_to_delete;
1692 rtmsg_fib(RTM_DELROUTE, htonl(key), fa, plen, tb->tb_id,
1693 &cfg->fc_nlinfo, 0);
1695 l = fib_find_node(t, key);
1696 li = find_leaf_info(l, plen);
1698 list_del_rcu(&fa->fa_list);
1700 if (!plen)
1701 tb->tb_num_default--;
1703 if (list_empty(fa_head)) {
1704 hlist_del_rcu(&li->hlist);
1705 free_leaf_info(li);
1708 if (hlist_empty(&l->list))
1709 trie_leaf_remove(t, l);
1711 if (fa->fa_state & FA_S_ACCESSED)
1712 rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1714 fib_release_info(fa->fa_info);
1715 alias_free_mem_rcu(fa);
1716 return 0;
1719 static int trie_flush_list(struct list_head *head)
1721 struct fib_alias *fa, *fa_node;
1722 int found = 0;
1724 list_for_each_entry_safe(fa, fa_node, head, fa_list) {
1725 struct fib_info *fi = fa->fa_info;
1727 if (fi && (fi->fib_flags & RTNH_F_DEAD)) {
1728 list_del_rcu(&fa->fa_list);
1729 fib_release_info(fa->fa_info);
1730 alias_free_mem_rcu(fa);
1731 found++;
1734 return found;
1737 static int trie_flush_leaf(struct leaf *l)
1739 int found = 0;
1740 struct hlist_head *lih = &l->list;
1741 struct hlist_node *node, *tmp;
1742 struct leaf_info *li = NULL;
1744 hlist_for_each_entry_safe(li, node, tmp, lih, hlist) {
1745 found += trie_flush_list(&li->falh);
1747 if (list_empty(&li->falh)) {
1748 hlist_del_rcu(&li->hlist);
1749 free_leaf_info(li);
1752 return found;
1756 * Scan for the next right leaf starting at node p->child[idx]
1757 * Since we have back pointer, no recursion necessary.
1759 static struct leaf *leaf_walk_rcu(struct tnode *p, struct rt_trie_node *c)
1761 do {
1762 t_key idx;
1764 if (c)
1765 idx = tkey_extract_bits(c->key, p->pos, p->bits) + 1;
1766 else
1767 idx = 0;
1769 while (idx < 1u << p->bits) {
1770 c = tnode_get_child_rcu(p, idx++);
1771 if (!c)
1772 continue;
1774 if (IS_LEAF(c)) {
1775 prefetch(rcu_dereference_rtnl(p->child[idx]));
1776 return (struct leaf *) c;
1779 /* Rescan start scanning in new node */
1780 p = (struct tnode *) c;
1781 idx = 0;
1784 /* Node empty, walk back up to parent */
1785 c = (struct rt_trie_node *) p;
1786 } while ((p = node_parent_rcu(c)) != NULL);
1788 return NULL; /* Root of trie */
1791 static struct leaf *trie_firstleaf(struct trie *t)
1793 struct tnode *n = (struct tnode *)rcu_dereference_rtnl(t->trie);
1795 if (!n)
1796 return NULL;
1798 if (IS_LEAF(n)) /* trie is just a leaf */
1799 return (struct leaf *) n;
1801 return leaf_walk_rcu(n, NULL);
1804 static struct leaf *trie_nextleaf(struct leaf *l)
1806 struct rt_trie_node *c = (struct rt_trie_node *) l;
1807 struct tnode *p = node_parent_rcu(c);
1809 if (!p)
1810 return NULL; /* trie with just one leaf */
1812 return leaf_walk_rcu(p, c);
1815 static struct leaf *trie_leafindex(struct trie *t, int index)
1817 struct leaf *l = trie_firstleaf(t);
1819 while (l && index-- > 0)
1820 l = trie_nextleaf(l);
1822 return l;
1827 * Caller must hold RTNL.
1829 int fib_table_flush(struct fib_table *tb)
1831 struct trie *t = (struct trie *) tb->tb_data;
1832 struct leaf *l, *ll = NULL;
1833 int found = 0;
1835 for (l = trie_firstleaf(t); l; l = trie_nextleaf(l)) {
1836 found += trie_flush_leaf(l);
1838 if (ll && hlist_empty(&ll->list))
1839 trie_leaf_remove(t, ll);
1840 ll = l;
1843 if (ll && hlist_empty(&ll->list))
1844 trie_leaf_remove(t, ll);
1846 pr_debug("trie_flush found=%d\n", found);
1847 return found;
1850 void fib_free_table(struct fib_table *tb)
1852 kfree(tb);
1855 static int fn_trie_dump_fa(t_key key, int plen, struct list_head *fah,
1856 struct fib_table *tb,
1857 struct sk_buff *skb, struct netlink_callback *cb)
1859 int i, s_i;
1860 struct fib_alias *fa;
1861 __be32 xkey = htonl(key);
1863 s_i = cb->args[5];
1864 i = 0;
1866 /* rcu_read_lock is hold by caller */
1868 list_for_each_entry_rcu(fa, fah, fa_list) {
1869 if (i < s_i) {
1870 i++;
1871 continue;
1874 if (fib_dump_info(skb, NETLINK_CB(cb->skb).pid,
1875 cb->nlh->nlmsg_seq,
1876 RTM_NEWROUTE,
1877 tb->tb_id,
1878 fa->fa_type,
1879 xkey,
1880 plen,
1881 fa->fa_tos,
1882 fa->fa_info, NLM_F_MULTI) < 0) {
1883 cb->args[5] = i;
1884 return -1;
1886 i++;
1888 cb->args[5] = i;
1889 return skb->len;
1892 static int fn_trie_dump_leaf(struct leaf *l, struct fib_table *tb,
1893 struct sk_buff *skb, struct netlink_callback *cb)
1895 struct leaf_info *li;
1896 struct hlist_node *node;
1897 int i, s_i;
1899 s_i = cb->args[4];
1900 i = 0;
1902 /* rcu_read_lock is hold by caller */
1903 hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
1904 if (i < s_i) {
1905 i++;
1906 continue;
1909 if (i > s_i)
1910 cb->args[5] = 0;
1912 if (list_empty(&li->falh))
1913 continue;
1915 if (fn_trie_dump_fa(l->key, li->plen, &li->falh, tb, skb, cb) < 0) {
1916 cb->args[4] = i;
1917 return -1;
1919 i++;
1922 cb->args[4] = i;
1923 return skb->len;
1926 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
1927 struct netlink_callback *cb)
1929 struct leaf *l;
1930 struct trie *t = (struct trie *) tb->tb_data;
1931 t_key key = cb->args[2];
1932 int count = cb->args[3];
1934 rcu_read_lock();
1935 /* Dump starting at last key.
1936 * Note: 0.0.0.0/0 (ie default) is first key.
1938 if (count == 0)
1939 l = trie_firstleaf(t);
1940 else {
1941 /* Normally, continue from last key, but if that is missing
1942 * fallback to using slow rescan
1944 l = fib_find_node(t, key);
1945 if (!l)
1946 l = trie_leafindex(t, count);
1949 while (l) {
1950 cb->args[2] = l->key;
1951 if (fn_trie_dump_leaf(l, tb, skb, cb) < 0) {
1952 cb->args[3] = count;
1953 rcu_read_unlock();
1954 return -1;
1957 ++count;
1958 l = trie_nextleaf(l);
1959 memset(&cb->args[4], 0,
1960 sizeof(cb->args) - 4*sizeof(cb->args[0]));
1962 cb->args[3] = count;
1963 rcu_read_unlock();
1965 return skb->len;
1968 void __init fib_trie_init(void)
1970 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
1971 sizeof(struct fib_alias),
1972 0, SLAB_PANIC, NULL);
1974 trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
1975 max(sizeof(struct leaf),
1976 sizeof(struct leaf_info)),
1977 0, SLAB_PANIC, NULL);
1981 struct fib_table *fib_trie_table(u32 id)
1983 struct fib_table *tb;
1984 struct trie *t;
1986 tb = kmalloc(sizeof(struct fib_table) + sizeof(struct trie),
1987 GFP_KERNEL);
1988 if (tb == NULL)
1989 return NULL;
1991 tb->tb_id = id;
1992 tb->tb_default = -1;
1993 tb->tb_num_default = 0;
1995 t = (struct trie *) tb->tb_data;
1996 memset(t, 0, sizeof(*t));
1998 return tb;
2001 #ifdef CONFIG_PROC_FS
2002 /* Depth first Trie walk iterator */
2003 struct fib_trie_iter {
2004 struct seq_net_private p;
2005 struct fib_table *tb;
2006 struct tnode *tnode;
2007 unsigned int index;
2008 unsigned int depth;
2011 static struct rt_trie_node *fib_trie_get_next(struct fib_trie_iter *iter)
2013 struct tnode *tn = iter->tnode;
2014 unsigned int cindex = iter->index;
2015 struct tnode *p;
2017 /* A single entry routing table */
2018 if (!tn)
2019 return NULL;
2021 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2022 iter->tnode, iter->index, iter->depth);
2023 rescan:
2024 while (cindex < (1<<tn->bits)) {
2025 struct rt_trie_node *n = tnode_get_child_rcu(tn, cindex);
2027 if (n) {
2028 if (IS_LEAF(n)) {
2029 iter->tnode = tn;
2030 iter->index = cindex + 1;
2031 } else {
2032 /* push down one level */
2033 iter->tnode = (struct tnode *) n;
2034 iter->index = 0;
2035 ++iter->depth;
2037 return n;
2040 ++cindex;
2043 /* Current node exhausted, pop back up */
2044 p = node_parent_rcu((struct rt_trie_node *)tn);
2045 if (p) {
2046 cindex = tkey_extract_bits(tn->key, p->pos, p->bits)+1;
2047 tn = p;
2048 --iter->depth;
2049 goto rescan;
2052 /* got root? */
2053 return NULL;
2056 static struct rt_trie_node *fib_trie_get_first(struct fib_trie_iter *iter,
2057 struct trie *t)
2059 struct rt_trie_node *n;
2061 if (!t)
2062 return NULL;
2064 n = rcu_dereference(t->trie);
2065 if (!n)
2066 return NULL;
2068 if (IS_TNODE(n)) {
2069 iter->tnode = (struct tnode *) n;
2070 iter->index = 0;
2071 iter->depth = 1;
2072 } else {
2073 iter->tnode = NULL;
2074 iter->index = 0;
2075 iter->depth = 0;
2078 return n;
2081 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2083 struct rt_trie_node *n;
2084 struct fib_trie_iter iter;
2086 memset(s, 0, sizeof(*s));
2088 rcu_read_lock();
2089 for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2090 if (IS_LEAF(n)) {
2091 struct leaf *l = (struct leaf *)n;
2092 struct leaf_info *li;
2093 struct hlist_node *tmp;
2095 s->leaves++;
2096 s->totdepth += iter.depth;
2097 if (iter.depth > s->maxdepth)
2098 s->maxdepth = iter.depth;
2100 hlist_for_each_entry_rcu(li, tmp, &l->list, hlist)
2101 ++s->prefixes;
2102 } else {
2103 const struct tnode *tn = (const struct tnode *) n;
2104 int i;
2106 s->tnodes++;
2107 if (tn->bits < MAX_STAT_DEPTH)
2108 s->nodesizes[tn->bits]++;
2110 for (i = 0; i < (1<<tn->bits); i++)
2111 if (!tn->child[i])
2112 s->nullpointers++;
2115 rcu_read_unlock();
2119 * This outputs /proc/net/fib_triestats
2121 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2123 unsigned int i, max, pointers, bytes, avdepth;
2125 if (stat->leaves)
2126 avdepth = stat->totdepth*100 / stat->leaves;
2127 else
2128 avdepth = 0;
2130 seq_printf(seq, "\tAver depth: %u.%02d\n",
2131 avdepth / 100, avdepth % 100);
2132 seq_printf(seq, "\tMax depth: %u\n", stat->maxdepth);
2134 seq_printf(seq, "\tLeaves: %u\n", stat->leaves);
2135 bytes = sizeof(struct leaf) * stat->leaves;
2137 seq_printf(seq, "\tPrefixes: %u\n", stat->prefixes);
2138 bytes += sizeof(struct leaf_info) * stat->prefixes;
2140 seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2141 bytes += sizeof(struct tnode) * stat->tnodes;
2143 max = MAX_STAT_DEPTH;
2144 while (max > 0 && stat->nodesizes[max-1] == 0)
2145 max--;
2147 pointers = 0;
2148 for (i = 1; i <= max; i++)
2149 if (stat->nodesizes[i] != 0) {
2150 seq_printf(seq, " %u: %u", i, stat->nodesizes[i]);
2151 pointers += (1<<i) * stat->nodesizes[i];
2153 seq_putc(seq, '\n');
2154 seq_printf(seq, "\tPointers: %u\n", pointers);
2156 bytes += sizeof(struct rt_trie_node *) * pointers;
2157 seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2158 seq_printf(seq, "Total size: %u kB\n", (bytes + 1023) / 1024);
2161 #ifdef CONFIG_IP_FIB_TRIE_STATS
2162 static void trie_show_usage(struct seq_file *seq,
2163 const struct trie_use_stats *stats)
2165 seq_printf(seq, "\nCounters:\n---------\n");
2166 seq_printf(seq, "gets = %u\n", stats->gets);
2167 seq_printf(seq, "backtracks = %u\n", stats->backtrack);
2168 seq_printf(seq, "semantic match passed = %u\n",
2169 stats->semantic_match_passed);
2170 seq_printf(seq, "semantic match miss = %u\n",
2171 stats->semantic_match_miss);
2172 seq_printf(seq, "null node hit= %u\n", stats->null_node_hit);
2173 seq_printf(seq, "skipped node resize = %u\n\n",
2174 stats->resize_node_skipped);
2176 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2178 static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2180 if (tb->tb_id == RT_TABLE_LOCAL)
2181 seq_puts(seq, "Local:\n");
2182 else if (tb->tb_id == RT_TABLE_MAIN)
2183 seq_puts(seq, "Main:\n");
2184 else
2185 seq_printf(seq, "Id %d:\n", tb->tb_id);
2189 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2191 struct net *net = (struct net *)seq->private;
2192 unsigned int h;
2194 seq_printf(seq,
2195 "Basic info: size of leaf:"
2196 " %Zd bytes, size of tnode: %Zd bytes.\n",
2197 sizeof(struct leaf), sizeof(struct tnode));
2199 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2200 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2201 struct hlist_node *node;
2202 struct fib_table *tb;
2204 hlist_for_each_entry_rcu(tb, node, head, tb_hlist) {
2205 struct trie *t = (struct trie *) tb->tb_data;
2206 struct trie_stat stat;
2208 if (!t)
2209 continue;
2211 fib_table_print(seq, tb);
2213 trie_collect_stats(t, &stat);
2214 trie_show_stats(seq, &stat);
2215 #ifdef CONFIG_IP_FIB_TRIE_STATS
2216 trie_show_usage(seq, &t->stats);
2217 #endif
2221 return 0;
2224 static int fib_triestat_seq_open(struct inode *inode, struct file *file)
2226 return single_open_net(inode, file, fib_triestat_seq_show);
2229 static const struct file_operations fib_triestat_fops = {
2230 .owner = THIS_MODULE,
2231 .open = fib_triestat_seq_open,
2232 .read = seq_read,
2233 .llseek = seq_lseek,
2234 .release = single_release_net,
2237 static struct rt_trie_node *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2239 struct fib_trie_iter *iter = seq->private;
2240 struct net *net = seq_file_net(seq);
2241 loff_t idx = 0;
2242 unsigned int h;
2244 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2245 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2246 struct hlist_node *node;
2247 struct fib_table *tb;
2249 hlist_for_each_entry_rcu(tb, node, head, tb_hlist) {
2250 struct rt_trie_node *n;
2252 for (n = fib_trie_get_first(iter,
2253 (struct trie *) tb->tb_data);
2254 n; n = fib_trie_get_next(iter))
2255 if (pos == idx++) {
2256 iter->tb = tb;
2257 return n;
2262 return NULL;
2265 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2266 __acquires(RCU)
2268 rcu_read_lock();
2269 return fib_trie_get_idx(seq, *pos);
2272 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2274 struct fib_trie_iter *iter = seq->private;
2275 struct net *net = seq_file_net(seq);
2276 struct fib_table *tb = iter->tb;
2277 struct hlist_node *tb_node;
2278 unsigned int h;
2279 struct rt_trie_node *n;
2281 ++*pos;
2282 /* next node in same table */
2283 n = fib_trie_get_next(iter);
2284 if (n)
2285 return n;
2287 /* walk rest of this hash chain */
2288 h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2289 while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2290 tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2291 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2292 if (n)
2293 goto found;
2296 /* new hash chain */
2297 while (++h < FIB_TABLE_HASHSZ) {
2298 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2299 hlist_for_each_entry_rcu(tb, tb_node, head, tb_hlist) {
2300 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2301 if (n)
2302 goto found;
2305 return NULL;
2307 found:
2308 iter->tb = tb;
2309 return n;
2312 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2313 __releases(RCU)
2315 rcu_read_unlock();
2318 static void seq_indent(struct seq_file *seq, int n)
2320 while (n-- > 0)
2321 seq_puts(seq, " ");
2324 static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2326 switch (s) {
2327 case RT_SCOPE_UNIVERSE: return "universe";
2328 case RT_SCOPE_SITE: return "site";
2329 case RT_SCOPE_LINK: return "link";
2330 case RT_SCOPE_HOST: return "host";
2331 case RT_SCOPE_NOWHERE: return "nowhere";
2332 default:
2333 snprintf(buf, len, "scope=%d", s);
2334 return buf;
2338 static const char *const rtn_type_names[__RTN_MAX] = {
2339 [RTN_UNSPEC] = "UNSPEC",
2340 [RTN_UNICAST] = "UNICAST",
2341 [RTN_LOCAL] = "LOCAL",
2342 [RTN_BROADCAST] = "BROADCAST",
2343 [RTN_ANYCAST] = "ANYCAST",
2344 [RTN_MULTICAST] = "MULTICAST",
2345 [RTN_BLACKHOLE] = "BLACKHOLE",
2346 [RTN_UNREACHABLE] = "UNREACHABLE",
2347 [RTN_PROHIBIT] = "PROHIBIT",
2348 [RTN_THROW] = "THROW",
2349 [RTN_NAT] = "NAT",
2350 [RTN_XRESOLVE] = "XRESOLVE",
2353 static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2355 if (t < __RTN_MAX && rtn_type_names[t])
2356 return rtn_type_names[t];
2357 snprintf(buf, len, "type %u", t);
2358 return buf;
2361 /* Pretty print the trie */
2362 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2364 const struct fib_trie_iter *iter = seq->private;
2365 struct rt_trie_node *n = v;
2367 if (!node_parent_rcu(n))
2368 fib_table_print(seq, iter->tb);
2370 if (IS_TNODE(n)) {
2371 struct tnode *tn = (struct tnode *) n;
2372 __be32 prf = htonl(mask_pfx(tn->key, tn->pos));
2374 seq_indent(seq, iter->depth-1);
2375 seq_printf(seq, " +-- %pI4/%d %d %d %d\n",
2376 &prf, tn->pos, tn->bits, tn->full_children,
2377 tn->empty_children);
2379 } else {
2380 struct leaf *l = (struct leaf *) n;
2381 struct leaf_info *li;
2382 struct hlist_node *node;
2383 __be32 val = htonl(l->key);
2385 seq_indent(seq, iter->depth);
2386 seq_printf(seq, " |-- %pI4\n", &val);
2388 hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
2389 struct fib_alias *fa;
2391 list_for_each_entry_rcu(fa, &li->falh, fa_list) {
2392 char buf1[32], buf2[32];
2394 seq_indent(seq, iter->depth+1);
2395 seq_printf(seq, " /%d %s %s", li->plen,
2396 rtn_scope(buf1, sizeof(buf1),
2397 fa->fa_info->fib_scope),
2398 rtn_type(buf2, sizeof(buf2),
2399 fa->fa_type));
2400 if (fa->fa_tos)
2401 seq_printf(seq, " tos=%d", fa->fa_tos);
2402 seq_putc(seq, '\n');
2407 return 0;
2410 static const struct seq_operations fib_trie_seq_ops = {
2411 .start = fib_trie_seq_start,
2412 .next = fib_trie_seq_next,
2413 .stop = fib_trie_seq_stop,
2414 .show = fib_trie_seq_show,
2417 static int fib_trie_seq_open(struct inode *inode, struct file *file)
2419 return seq_open_net(inode, file, &fib_trie_seq_ops,
2420 sizeof(struct fib_trie_iter));
2423 static const struct file_operations fib_trie_fops = {
2424 .owner = THIS_MODULE,
2425 .open = fib_trie_seq_open,
2426 .read = seq_read,
2427 .llseek = seq_lseek,
2428 .release = seq_release_net,
2431 struct fib_route_iter {
2432 struct seq_net_private p;
2433 struct trie *main_trie;
2434 loff_t pos;
2435 t_key key;
2438 static struct leaf *fib_route_get_idx(struct fib_route_iter *iter, loff_t pos)
2440 struct leaf *l = NULL;
2441 struct trie *t = iter->main_trie;
2443 /* use cache location of last found key */
2444 if (iter->pos > 0 && pos >= iter->pos && (l = fib_find_node(t, iter->key)))
2445 pos -= iter->pos;
2446 else {
2447 iter->pos = 0;
2448 l = trie_firstleaf(t);
2451 while (l && pos-- > 0) {
2452 iter->pos++;
2453 l = trie_nextleaf(l);
2456 if (l)
2457 iter->key = pos; /* remember it */
2458 else
2459 iter->pos = 0; /* forget it */
2461 return l;
2464 static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2465 __acquires(RCU)
2467 struct fib_route_iter *iter = seq->private;
2468 struct fib_table *tb;
2470 rcu_read_lock();
2471 tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2472 if (!tb)
2473 return NULL;
2475 iter->main_trie = (struct trie *) tb->tb_data;
2476 if (*pos == 0)
2477 return SEQ_START_TOKEN;
2478 else
2479 return fib_route_get_idx(iter, *pos - 1);
2482 static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2484 struct fib_route_iter *iter = seq->private;
2485 struct leaf *l = v;
2487 ++*pos;
2488 if (v == SEQ_START_TOKEN) {
2489 iter->pos = 0;
2490 l = trie_firstleaf(iter->main_trie);
2491 } else {
2492 iter->pos++;
2493 l = trie_nextleaf(l);
2496 if (l)
2497 iter->key = l->key;
2498 else
2499 iter->pos = 0;
2500 return l;
2503 static void fib_route_seq_stop(struct seq_file *seq, void *v)
2504 __releases(RCU)
2506 rcu_read_unlock();
2509 static unsigned int fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2511 unsigned int flags = 0;
2513 if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2514 flags = RTF_REJECT;
2515 if (fi && fi->fib_nh->nh_gw)
2516 flags |= RTF_GATEWAY;
2517 if (mask == htonl(0xFFFFFFFF))
2518 flags |= RTF_HOST;
2519 flags |= RTF_UP;
2520 return flags;
2524 * This outputs /proc/net/route.
2525 * The format of the file is not supposed to be changed
2526 * and needs to be same as fib_hash output to avoid breaking
2527 * legacy utilities
2529 static int fib_route_seq_show(struct seq_file *seq, void *v)
2531 struct leaf *l = v;
2532 struct leaf_info *li;
2533 struct hlist_node *node;
2535 if (v == SEQ_START_TOKEN) {
2536 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2537 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2538 "\tWindow\tIRTT");
2539 return 0;
2542 hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
2543 struct fib_alias *fa;
2544 __be32 mask, prefix;
2546 mask = inet_make_mask(li->plen);
2547 prefix = htonl(l->key);
2549 list_for_each_entry_rcu(fa, &li->falh, fa_list) {
2550 const struct fib_info *fi = fa->fa_info;
2551 unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2552 int len;
2554 if (fa->fa_type == RTN_BROADCAST
2555 || fa->fa_type == RTN_MULTICAST)
2556 continue;
2558 if (fi)
2559 seq_printf(seq,
2560 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2561 "%d\t%08X\t%d\t%u\t%u%n",
2562 fi->fib_dev ? fi->fib_dev->name : "*",
2563 prefix,
2564 fi->fib_nh->nh_gw, flags, 0, 0,
2565 fi->fib_priority,
2566 mask,
2567 (fi->fib_advmss ?
2568 fi->fib_advmss + 40 : 0),
2569 fi->fib_window,
2570 fi->fib_rtt >> 3, &len);
2571 else
2572 seq_printf(seq,
2573 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2574 "%d\t%08X\t%d\t%u\t%u%n",
2575 prefix, 0, flags, 0, 0, 0,
2576 mask, 0, 0, 0, &len);
2578 seq_printf(seq, "%*s\n", 127 - len, "");
2582 return 0;
2585 static const struct seq_operations fib_route_seq_ops = {
2586 .start = fib_route_seq_start,
2587 .next = fib_route_seq_next,
2588 .stop = fib_route_seq_stop,
2589 .show = fib_route_seq_show,
2592 static int fib_route_seq_open(struct inode *inode, struct file *file)
2594 return seq_open_net(inode, file, &fib_route_seq_ops,
2595 sizeof(struct fib_route_iter));
2598 static const struct file_operations fib_route_fops = {
2599 .owner = THIS_MODULE,
2600 .open = fib_route_seq_open,
2601 .read = seq_read,
2602 .llseek = seq_lseek,
2603 .release = seq_release_net,
2606 int __net_init fib_proc_init(struct net *net)
2608 if (!proc_net_fops_create(net, "fib_trie", S_IRUGO, &fib_trie_fops))
2609 goto out1;
2611 if (!proc_net_fops_create(net, "fib_triestat", S_IRUGO,
2612 &fib_triestat_fops))
2613 goto out2;
2615 if (!proc_net_fops_create(net, "route", S_IRUGO, &fib_route_fops))
2616 goto out3;
2618 return 0;
2620 out3:
2621 proc_net_remove(net, "fib_triestat");
2622 out2:
2623 proc_net_remove(net, "fib_trie");
2624 out1:
2625 return -ENOMEM;
2628 void __net_exit fib_proc_exit(struct net *net)
2630 proc_net_remove(net, "fib_trie");
2631 proc_net_remove(net, "fib_triestat");
2632 proc_net_remove(net, "route");
2635 #endif /* CONFIG_PROC_FS */