ipv4: Kill can_sleep arg to ip_route_output_flow()
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / net / ipv4 / fib_trie.c
blobedf3b0997e0195bee3c098e5a6d005833d249850
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 descibed in:
17 * An experimental study of compression methods for dynamic tries
18 * Stefan Nilsson and Matti Tikkanen. Algorithmica, 33(1):19-33, 2002.
19 * http://www.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 <net/net_namespace.h>
76 #include <net/ip.h>
77 #include <net/protocol.h>
78 #include <net/route.h>
79 #include <net/tcp.h>
80 #include <net/sock.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))
88 typedef unsigned int t_key;
90 #define T_TNODE 0
91 #define T_LEAF 1
92 #define NODE_TYPE_MASK 0x1UL
93 #define NODE_TYPE(node) ((node)->parent & NODE_TYPE_MASK)
95 #define IS_TNODE(n) (!(n->parent & T_LEAF))
96 #define IS_LEAF(n) (n->parent & T_LEAF)
98 struct rt_trie_node {
99 unsigned long parent;
100 t_key key;
103 struct leaf {
104 unsigned long parent;
105 t_key key;
106 struct hlist_head list;
107 struct rcu_head rcu;
110 struct leaf_info {
111 struct hlist_node hlist;
112 struct rcu_head rcu;
113 int plen;
114 struct list_head falh;
117 struct tnode {
118 unsigned long parent;
119 t_key key;
120 unsigned char pos; /* 2log(KEYLENGTH) bits needed */
121 unsigned char bits; /* 2log(KEYLENGTH) bits needed */
122 unsigned int full_children; /* KEYLENGTH bits needed */
123 unsigned int empty_children; /* KEYLENGTH bits needed */
124 union {
125 struct rcu_head rcu;
126 struct work_struct work;
127 struct tnode *tnode_free;
129 struct rt_trie_node *child[0];
132 #ifdef CONFIG_IP_FIB_TRIE_STATS
133 struct trie_use_stats {
134 unsigned int gets;
135 unsigned int backtrack;
136 unsigned int semantic_match_passed;
137 unsigned int semantic_match_miss;
138 unsigned int null_node_hit;
139 unsigned int resize_node_skipped;
141 #endif
143 struct trie_stat {
144 unsigned int totdepth;
145 unsigned int maxdepth;
146 unsigned int tnodes;
147 unsigned int leaves;
148 unsigned int nullpointers;
149 unsigned int prefixes;
150 unsigned int nodesizes[MAX_STAT_DEPTH];
153 struct trie {
154 struct rt_trie_node *trie;
155 #ifdef CONFIG_IP_FIB_TRIE_STATS
156 struct trie_use_stats stats;
157 #endif
160 static void put_child(struct trie *t, struct tnode *tn, int i, struct rt_trie_node *n);
161 static void tnode_put_child_reorg(struct tnode *tn, int i, struct rt_trie_node *n,
162 int wasfull);
163 static struct rt_trie_node *resize(struct trie *t, struct tnode *tn);
164 static struct tnode *inflate(struct trie *t, struct tnode *tn);
165 static struct tnode *halve(struct trie *t, struct tnode *tn);
166 /* tnodes to free after resize(); protected by RTNL */
167 static struct tnode *tnode_free_head;
168 static size_t tnode_free_size;
171 * synchronize_rcu after call_rcu for that many pages; it should be especially
172 * useful before resizing the root node with PREEMPT_NONE configs; the value was
173 * obtained experimentally, aiming to avoid visible slowdown.
175 static const int sync_pages = 128;
177 static struct kmem_cache *fn_alias_kmem __read_mostly;
178 static struct kmem_cache *trie_leaf_kmem __read_mostly;
180 static inline struct tnode *node_parent(struct rt_trie_node *node)
182 return (struct tnode *)(node->parent & ~NODE_TYPE_MASK);
185 static inline struct tnode *node_parent_rcu(struct rt_trie_node *node)
187 struct tnode *ret = node_parent(node);
189 return rcu_dereference_rtnl(ret);
192 /* Same as rcu_assign_pointer
193 * but that macro() assumes that value is a pointer.
195 static inline void node_set_parent(struct rt_trie_node *node, struct tnode *ptr)
197 smp_wmb();
198 node->parent = (unsigned long)ptr | NODE_TYPE(node);
201 static inline struct rt_trie_node *tnode_get_child(struct tnode *tn, unsigned int i)
203 BUG_ON(i >= 1U << tn->bits);
205 return tn->child[i];
208 static inline struct rt_trie_node *tnode_get_child_rcu(struct tnode *tn, unsigned int i)
210 struct rt_trie_node *ret = tnode_get_child(tn, i);
212 return rcu_dereference_rtnl(ret);
215 static inline int tnode_child_length(const struct tnode *tn)
217 return 1 << tn->bits;
220 static inline t_key mask_pfx(t_key k, unsigned int l)
222 return (l == 0) ? 0 : k >> (KEYLENGTH-l) << (KEYLENGTH-l);
225 static inline t_key tkey_extract_bits(t_key a, unsigned int offset, unsigned int bits)
227 if (offset < KEYLENGTH)
228 return ((t_key)(a << offset)) >> (KEYLENGTH - bits);
229 else
230 return 0;
233 static inline int tkey_equals(t_key a, t_key b)
235 return a == b;
238 static inline int tkey_sub_equals(t_key a, int offset, int bits, t_key b)
240 if (bits == 0 || offset >= KEYLENGTH)
241 return 1;
242 bits = bits > KEYLENGTH ? KEYLENGTH : bits;
243 return ((a ^ b) << offset) >> (KEYLENGTH - bits) == 0;
246 static inline int tkey_mismatch(t_key a, int offset, t_key b)
248 t_key diff = a ^ b;
249 int i = offset;
251 if (!diff)
252 return 0;
253 while ((diff << i) >> (KEYLENGTH-1) == 0)
254 i++;
255 return i;
259 To understand this stuff, an understanding of keys and all their bits is
260 necessary. Every node in the trie has a key associated with it, but not
261 all of the bits in that key are significant.
263 Consider a node 'n' and its parent 'tp'.
265 If n is a leaf, every bit in its key is significant. Its presence is
266 necessitated by path compression, since during a tree traversal (when
267 searching for a leaf - unless we are doing an insertion) we will completely
268 ignore all skipped bits we encounter. Thus we need to verify, at the end of
269 a potentially successful search, that we have indeed been walking the
270 correct key path.
272 Note that we can never "miss" the correct key in the tree if present by
273 following the wrong path. Path compression ensures that segments of the key
274 that are the same for all keys with a given prefix are skipped, but the
275 skipped part *is* identical for each node in the subtrie below the skipped
276 bit! trie_insert() in this implementation takes care of that - note the
277 call to tkey_sub_equals() in trie_insert().
279 if n is an internal node - a 'tnode' here, the various parts of its key
280 have many different meanings.
282 Example:
283 _________________________________________________________________
284 | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
285 -----------------------------------------------------------------
286 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
288 _________________________________________________________________
289 | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
290 -----------------------------------------------------------------
291 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
293 tp->pos = 7
294 tp->bits = 3
295 n->pos = 15
296 n->bits = 4
298 First, let's just ignore the bits that come before the parent tp, that is
299 the bits from 0 to (tp->pos-1). They are *known* but at this point we do
300 not use them for anything.
302 The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
303 index into the parent's child array. That is, they will be used to find
304 'n' among tp's children.
306 The bits from (tp->pos + tp->bits) to (n->pos - 1) - "S" - are skipped bits
307 for the node n.
309 All the bits we have seen so far are significant to the node n. The rest
310 of the bits are really not needed or indeed known in n->key.
312 The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
313 n's child array, and will of course be different for each child.
316 The rest of the bits, from (n->pos + n->bits) onward, are completely unknown
317 at this point.
321 static inline void check_tnode(const struct tnode *tn)
323 WARN_ON(tn && tn->pos+tn->bits > 32);
326 static const int halve_threshold = 25;
327 static const int inflate_threshold = 50;
328 static const int halve_threshold_root = 15;
329 static const int inflate_threshold_root = 30;
331 static void __alias_free_mem(struct rcu_head *head)
333 struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
334 kmem_cache_free(fn_alias_kmem, fa);
337 static inline void alias_free_mem_rcu(struct fib_alias *fa)
339 call_rcu(&fa->rcu, __alias_free_mem);
342 static void __leaf_free_rcu(struct rcu_head *head)
344 struct leaf *l = container_of(head, struct leaf, rcu);
345 kmem_cache_free(trie_leaf_kmem, l);
348 static inline void free_leaf(struct leaf *l)
350 call_rcu_bh(&l->rcu, __leaf_free_rcu);
353 static void __leaf_info_free_rcu(struct rcu_head *head)
355 kfree(container_of(head, struct leaf_info, rcu));
358 static inline void free_leaf_info(struct leaf_info *leaf)
360 call_rcu(&leaf->rcu, __leaf_info_free_rcu);
363 static struct tnode *tnode_alloc(size_t size)
365 if (size <= PAGE_SIZE)
366 return kzalloc(size, GFP_KERNEL);
367 else
368 return vzalloc(size);
371 static void __tnode_vfree(struct work_struct *arg)
373 struct tnode *tn = container_of(arg, struct tnode, work);
374 vfree(tn);
377 static void __tnode_free_rcu(struct rcu_head *head)
379 struct tnode *tn = container_of(head, struct tnode, rcu);
380 size_t size = sizeof(struct tnode) +
381 (sizeof(struct rt_trie_node *) << tn->bits);
383 if (size <= PAGE_SIZE)
384 kfree(tn);
385 else {
386 INIT_WORK(&tn->work, __tnode_vfree);
387 schedule_work(&tn->work);
391 static inline void tnode_free(struct tnode *tn)
393 if (IS_LEAF(tn))
394 free_leaf((struct leaf *) tn);
395 else
396 call_rcu(&tn->rcu, __tnode_free_rcu);
399 static void tnode_free_safe(struct tnode *tn)
401 BUG_ON(IS_LEAF(tn));
402 tn->tnode_free = tnode_free_head;
403 tnode_free_head = tn;
404 tnode_free_size += sizeof(struct tnode) +
405 (sizeof(struct rt_trie_node *) << tn->bits);
408 static void tnode_free_flush(void)
410 struct tnode *tn;
412 while ((tn = tnode_free_head)) {
413 tnode_free_head = tn->tnode_free;
414 tn->tnode_free = NULL;
415 tnode_free(tn);
418 if (tnode_free_size >= PAGE_SIZE * sync_pages) {
419 tnode_free_size = 0;
420 synchronize_rcu();
424 static struct leaf *leaf_new(void)
426 struct leaf *l = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
427 if (l) {
428 l->parent = T_LEAF;
429 INIT_HLIST_HEAD(&l->list);
431 return l;
434 static struct leaf_info *leaf_info_new(int plen)
436 struct leaf_info *li = kmalloc(sizeof(struct leaf_info), GFP_KERNEL);
437 if (li) {
438 li->plen = plen;
439 INIT_LIST_HEAD(&li->falh);
441 return li;
444 static struct tnode *tnode_new(t_key key, int pos, int bits)
446 size_t sz = sizeof(struct tnode) + (sizeof(struct rt_trie_node *) << bits);
447 struct tnode *tn = tnode_alloc(sz);
449 if (tn) {
450 tn->parent = T_TNODE;
451 tn->pos = pos;
452 tn->bits = bits;
453 tn->key = key;
454 tn->full_children = 0;
455 tn->empty_children = 1<<bits;
458 pr_debug("AT %p s=%zu %zu\n", tn, sizeof(struct tnode),
459 sizeof(struct rt_trie_node) << bits);
460 return tn;
464 * Check whether a tnode 'n' is "full", i.e. it is an internal node
465 * and no bits are skipped. See discussion in dyntree paper p. 6
468 static inline int tnode_full(const struct tnode *tn, const struct rt_trie_node *n)
470 if (n == NULL || IS_LEAF(n))
471 return 0;
473 return ((struct tnode *) n)->pos == tn->pos + tn->bits;
476 static inline void put_child(struct trie *t, struct tnode *tn, int i,
477 struct rt_trie_node *n)
479 tnode_put_child_reorg(tn, i, n, -1);
483 * Add a child at position i overwriting the old value.
484 * Update the value of full_children and empty_children.
487 static void tnode_put_child_reorg(struct tnode *tn, int i, struct rt_trie_node *n,
488 int wasfull)
490 struct rt_trie_node *chi = tn->child[i];
491 int isfull;
493 BUG_ON(i >= 1<<tn->bits);
495 /* update emptyChildren */
496 if (n == NULL && chi != NULL)
497 tn->empty_children++;
498 else if (n != NULL && chi == NULL)
499 tn->empty_children--;
501 /* update fullChildren */
502 if (wasfull == -1)
503 wasfull = tnode_full(tn, chi);
505 isfull = tnode_full(tn, n);
506 if (wasfull && !isfull)
507 tn->full_children--;
508 else if (!wasfull && isfull)
509 tn->full_children++;
511 if (n)
512 node_set_parent(n, tn);
514 rcu_assign_pointer(tn->child[i], n);
517 #define MAX_WORK 10
518 static struct rt_trie_node *resize(struct trie *t, struct tnode *tn)
520 int i;
521 struct tnode *old_tn;
522 int inflate_threshold_use;
523 int halve_threshold_use;
524 int max_work;
526 if (!tn)
527 return NULL;
529 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
530 tn, inflate_threshold, halve_threshold);
532 /* No children */
533 if (tn->empty_children == tnode_child_length(tn)) {
534 tnode_free_safe(tn);
535 return NULL;
537 /* One child */
538 if (tn->empty_children == tnode_child_length(tn) - 1)
539 goto one_child;
541 * Double as long as the resulting node has a number of
542 * nonempty nodes that are above the threshold.
546 * From "Implementing a dynamic compressed trie" by Stefan Nilsson of
547 * the Helsinki University of Technology and Matti Tikkanen of Nokia
548 * Telecommunications, page 6:
549 * "A node is doubled if the ratio of non-empty children to all
550 * children in the *doubled* node is at least 'high'."
552 * 'high' in this instance is the variable 'inflate_threshold'. It
553 * is expressed as a percentage, so we multiply it with
554 * tnode_child_length() and instead of multiplying by 2 (since the
555 * child array will be doubled by inflate()) and multiplying
556 * the left-hand side by 100 (to handle the percentage thing) we
557 * multiply the left-hand side by 50.
559 * The left-hand side may look a bit weird: tnode_child_length(tn)
560 * - tn->empty_children is of course the number of non-null children
561 * in the current node. tn->full_children is the number of "full"
562 * children, that is non-null tnodes with a skip value of 0.
563 * All of those will be doubled in the resulting inflated tnode, so
564 * we just count them one extra time here.
566 * A clearer way to write this would be:
568 * to_be_doubled = tn->full_children;
569 * not_to_be_doubled = tnode_child_length(tn) - tn->empty_children -
570 * tn->full_children;
572 * new_child_length = tnode_child_length(tn) * 2;
574 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
575 * new_child_length;
576 * if (new_fill_factor >= inflate_threshold)
578 * ...and so on, tho it would mess up the while () loop.
580 * anyway,
581 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
582 * inflate_threshold
584 * avoid a division:
585 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
586 * inflate_threshold * new_child_length
588 * expand not_to_be_doubled and to_be_doubled, and shorten:
589 * 100 * (tnode_child_length(tn) - tn->empty_children +
590 * tn->full_children) >= inflate_threshold * new_child_length
592 * expand new_child_length:
593 * 100 * (tnode_child_length(tn) - tn->empty_children +
594 * tn->full_children) >=
595 * inflate_threshold * tnode_child_length(tn) * 2
597 * shorten again:
598 * 50 * (tn->full_children + tnode_child_length(tn) -
599 * tn->empty_children) >= inflate_threshold *
600 * tnode_child_length(tn)
604 check_tnode(tn);
606 /* Keep root node larger */
608 if (!node_parent((struct rt_trie_node *)tn)) {
609 inflate_threshold_use = inflate_threshold_root;
610 halve_threshold_use = halve_threshold_root;
611 } else {
612 inflate_threshold_use = inflate_threshold;
613 halve_threshold_use = halve_threshold;
616 max_work = MAX_WORK;
617 while ((tn->full_children > 0 && max_work-- &&
618 50 * (tn->full_children + tnode_child_length(tn)
619 - tn->empty_children)
620 >= inflate_threshold_use * tnode_child_length(tn))) {
622 old_tn = tn;
623 tn = inflate(t, tn);
625 if (IS_ERR(tn)) {
626 tn = old_tn;
627 #ifdef CONFIG_IP_FIB_TRIE_STATS
628 t->stats.resize_node_skipped++;
629 #endif
630 break;
634 check_tnode(tn);
636 /* Return if at least one inflate is run */
637 if (max_work != MAX_WORK)
638 return (struct rt_trie_node *) tn;
641 * Halve as long as the number of empty children in this
642 * node is above threshold.
645 max_work = MAX_WORK;
646 while (tn->bits > 1 && max_work-- &&
647 100 * (tnode_child_length(tn) - tn->empty_children) <
648 halve_threshold_use * tnode_child_length(tn)) {
650 old_tn = tn;
651 tn = halve(t, tn);
652 if (IS_ERR(tn)) {
653 tn = old_tn;
654 #ifdef CONFIG_IP_FIB_TRIE_STATS
655 t->stats.resize_node_skipped++;
656 #endif
657 break;
662 /* Only one child remains */
663 if (tn->empty_children == tnode_child_length(tn) - 1) {
664 one_child:
665 for (i = 0; i < tnode_child_length(tn); i++) {
666 struct rt_trie_node *n;
668 n = tn->child[i];
669 if (!n)
670 continue;
672 /* compress one level */
674 node_set_parent(n, NULL);
675 tnode_free_safe(tn);
676 return n;
679 return (struct rt_trie_node *) tn;
682 static struct tnode *inflate(struct trie *t, struct tnode *tn)
684 struct tnode *oldtnode = tn;
685 int olen = tnode_child_length(tn);
686 int i;
688 pr_debug("In inflate\n");
690 tn = tnode_new(oldtnode->key, oldtnode->pos, oldtnode->bits + 1);
692 if (!tn)
693 return ERR_PTR(-ENOMEM);
696 * Preallocate and store tnodes before the actual work so we
697 * don't get into an inconsistent state if memory allocation
698 * fails. In case of failure we return the oldnode and inflate
699 * of tnode is ignored.
702 for (i = 0; i < olen; i++) {
703 struct tnode *inode;
705 inode = (struct tnode *) tnode_get_child(oldtnode, i);
706 if (inode &&
707 IS_TNODE(inode) &&
708 inode->pos == oldtnode->pos + oldtnode->bits &&
709 inode->bits > 1) {
710 struct tnode *left, *right;
711 t_key m = ~0U << (KEYLENGTH - 1) >> inode->pos;
713 left = tnode_new(inode->key&(~m), inode->pos + 1,
714 inode->bits - 1);
715 if (!left)
716 goto nomem;
718 right = tnode_new(inode->key|m, inode->pos + 1,
719 inode->bits - 1);
721 if (!right) {
722 tnode_free(left);
723 goto nomem;
726 put_child(t, tn, 2*i, (struct rt_trie_node *) left);
727 put_child(t, tn, 2*i+1, (struct rt_trie_node *) right);
731 for (i = 0; i < olen; i++) {
732 struct tnode *inode;
733 struct rt_trie_node *node = tnode_get_child(oldtnode, i);
734 struct tnode *left, *right;
735 int size, j;
737 /* An empty child */
738 if (node == NULL)
739 continue;
741 /* A leaf or an internal node with skipped bits */
743 if (IS_LEAF(node) || ((struct tnode *) node)->pos >
744 tn->pos + tn->bits - 1) {
745 if (tkey_extract_bits(node->key,
746 oldtnode->pos + oldtnode->bits,
747 1) == 0)
748 put_child(t, tn, 2*i, node);
749 else
750 put_child(t, tn, 2*i+1, node);
751 continue;
754 /* An internal node with two children */
755 inode = (struct tnode *) node;
757 if (inode->bits == 1) {
758 put_child(t, tn, 2*i, inode->child[0]);
759 put_child(t, tn, 2*i+1, inode->child[1]);
761 tnode_free_safe(inode);
762 continue;
765 /* An internal node with more than two children */
767 /* We will replace this node 'inode' with two new
768 * ones, 'left' and 'right', each with half of the
769 * original children. The two new nodes will have
770 * a position one bit further down the key and this
771 * means that the "significant" part of their keys
772 * (see the discussion near the top of this file)
773 * will differ by one bit, which will be "0" in
774 * left's key and "1" in right's key. Since we are
775 * moving the key position by one step, the bit that
776 * we are moving away from - the bit at position
777 * (inode->pos) - is the one that will differ between
778 * left and right. So... we synthesize that bit in the
779 * two new keys.
780 * The mask 'm' below will be a single "one" bit at
781 * the position (inode->pos)
784 /* Use the old key, but set the new significant
785 * bit to zero.
788 left = (struct tnode *) tnode_get_child(tn, 2*i);
789 put_child(t, tn, 2*i, NULL);
791 BUG_ON(!left);
793 right = (struct tnode *) tnode_get_child(tn, 2*i+1);
794 put_child(t, tn, 2*i+1, NULL);
796 BUG_ON(!right);
798 size = tnode_child_length(left);
799 for (j = 0; j < size; j++) {
800 put_child(t, left, j, inode->child[j]);
801 put_child(t, right, j, inode->child[j + size]);
803 put_child(t, tn, 2*i, resize(t, left));
804 put_child(t, tn, 2*i+1, resize(t, right));
806 tnode_free_safe(inode);
808 tnode_free_safe(oldtnode);
809 return tn;
810 nomem:
812 int size = tnode_child_length(tn);
813 int j;
815 for (j = 0; j < size; j++)
816 if (tn->child[j])
817 tnode_free((struct tnode *)tn->child[j]);
819 tnode_free(tn);
821 return ERR_PTR(-ENOMEM);
825 static struct tnode *halve(struct trie *t, struct tnode *tn)
827 struct tnode *oldtnode = tn;
828 struct rt_trie_node *left, *right;
829 int i;
830 int olen = tnode_child_length(tn);
832 pr_debug("In halve\n");
834 tn = tnode_new(oldtnode->key, oldtnode->pos, oldtnode->bits - 1);
836 if (!tn)
837 return ERR_PTR(-ENOMEM);
840 * Preallocate and store tnodes before the actual work so we
841 * don't get into an inconsistent state if memory allocation
842 * fails. In case of failure we return the oldnode and halve
843 * of tnode is ignored.
846 for (i = 0; i < olen; i += 2) {
847 left = tnode_get_child(oldtnode, i);
848 right = tnode_get_child(oldtnode, i+1);
850 /* Two nonempty children */
851 if (left && right) {
852 struct tnode *newn;
854 newn = tnode_new(left->key, tn->pos + tn->bits, 1);
856 if (!newn)
857 goto nomem;
859 put_child(t, tn, i/2, (struct rt_trie_node *)newn);
864 for (i = 0; i < olen; i += 2) {
865 struct tnode *newBinNode;
867 left = tnode_get_child(oldtnode, i);
868 right = tnode_get_child(oldtnode, i+1);
870 /* At least one of the children is empty */
871 if (left == NULL) {
872 if (right == NULL) /* Both are empty */
873 continue;
874 put_child(t, tn, i/2, right);
875 continue;
878 if (right == NULL) {
879 put_child(t, tn, i/2, left);
880 continue;
883 /* Two nonempty children */
884 newBinNode = (struct tnode *) tnode_get_child(tn, i/2);
885 put_child(t, tn, i/2, NULL);
886 put_child(t, newBinNode, 0, left);
887 put_child(t, newBinNode, 1, right);
888 put_child(t, tn, i/2, resize(t, newBinNode));
890 tnode_free_safe(oldtnode);
891 return tn;
892 nomem:
894 int size = tnode_child_length(tn);
895 int j;
897 for (j = 0; j < size; j++)
898 if (tn->child[j])
899 tnode_free((struct tnode *)tn->child[j]);
901 tnode_free(tn);
903 return ERR_PTR(-ENOMEM);
907 /* readside must use rcu_read_lock currently dump routines
908 via get_fa_head and dump */
910 static struct leaf_info *find_leaf_info(struct leaf *l, int plen)
912 struct hlist_head *head = &l->list;
913 struct hlist_node *node;
914 struct leaf_info *li;
916 hlist_for_each_entry_rcu(li, node, head, hlist)
917 if (li->plen == plen)
918 return li;
920 return NULL;
923 static inline struct list_head *get_fa_head(struct leaf *l, int plen)
925 struct leaf_info *li = find_leaf_info(l, plen);
927 if (!li)
928 return NULL;
930 return &li->falh;
933 static void insert_leaf_info(struct hlist_head *head, struct leaf_info *new)
935 struct leaf_info *li = NULL, *last = NULL;
936 struct hlist_node *node;
938 if (hlist_empty(head)) {
939 hlist_add_head_rcu(&new->hlist, head);
940 } else {
941 hlist_for_each_entry(li, node, head, hlist) {
942 if (new->plen > li->plen)
943 break;
945 last = li;
947 if (last)
948 hlist_add_after_rcu(&last->hlist, &new->hlist);
949 else
950 hlist_add_before_rcu(&new->hlist, &li->hlist);
954 /* rcu_read_lock needs to be hold by caller from readside */
956 static struct leaf *
957 fib_find_node(struct trie *t, u32 key)
959 int pos;
960 struct tnode *tn;
961 struct rt_trie_node *n;
963 pos = 0;
964 n = rcu_dereference_rtnl(t->trie);
966 while (n != NULL && NODE_TYPE(n) == T_TNODE) {
967 tn = (struct tnode *) n;
969 check_tnode(tn);
971 if (tkey_sub_equals(tn->key, pos, tn->pos-pos, key)) {
972 pos = tn->pos + tn->bits;
973 n = tnode_get_child_rcu(tn,
974 tkey_extract_bits(key,
975 tn->pos,
976 tn->bits));
977 } else
978 break;
980 /* Case we have found a leaf. Compare prefixes */
982 if (n != NULL && IS_LEAF(n) && tkey_equals(key, n->key))
983 return (struct leaf *)n;
985 return NULL;
988 static void trie_rebalance(struct trie *t, struct tnode *tn)
990 int wasfull;
991 t_key cindex, key;
992 struct tnode *tp;
994 key = tn->key;
996 while (tn != NULL && (tp = node_parent((struct rt_trie_node *)tn)) != NULL) {
997 cindex = tkey_extract_bits(key, tp->pos, tp->bits);
998 wasfull = tnode_full(tp, tnode_get_child(tp, cindex));
999 tn = (struct tnode *) resize(t, (struct tnode *)tn);
1001 tnode_put_child_reorg((struct tnode *)tp, cindex,
1002 (struct rt_trie_node *)tn, wasfull);
1004 tp = node_parent((struct rt_trie_node *) tn);
1005 if (!tp)
1006 rcu_assign_pointer(t->trie, (struct rt_trie_node *)tn);
1008 tnode_free_flush();
1009 if (!tp)
1010 break;
1011 tn = tp;
1014 /* Handle last (top) tnode */
1015 if (IS_TNODE(tn))
1016 tn = (struct tnode *)resize(t, (struct tnode *)tn);
1018 rcu_assign_pointer(t->trie, (struct rt_trie_node *)tn);
1019 tnode_free_flush();
1022 /* only used from updater-side */
1024 static struct list_head *fib_insert_node(struct trie *t, u32 key, int plen)
1026 int pos, newpos;
1027 struct tnode *tp = NULL, *tn = NULL;
1028 struct rt_trie_node *n;
1029 struct leaf *l;
1030 int missbit;
1031 struct list_head *fa_head = NULL;
1032 struct leaf_info *li;
1033 t_key cindex;
1035 pos = 0;
1036 n = t->trie;
1038 /* If we point to NULL, stop. Either the tree is empty and we should
1039 * just put a new leaf in if, or we have reached an empty child slot,
1040 * and we should just put our new leaf in that.
1041 * If we point to a T_TNODE, check if it matches our key. Note that
1042 * a T_TNODE might be skipping any number of bits - its 'pos' need
1043 * not be the parent's 'pos'+'bits'!
1045 * If it does match the current key, get pos/bits from it, extract
1046 * the index from our key, push the T_TNODE and walk the tree.
1048 * If it doesn't, we have to replace it with a new T_TNODE.
1050 * If we point to a T_LEAF, it might or might not have the same key
1051 * as we do. If it does, just change the value, update the T_LEAF's
1052 * value, and return it.
1053 * If it doesn't, we need to replace it with a T_TNODE.
1056 while (n != NULL && NODE_TYPE(n) == T_TNODE) {
1057 tn = (struct tnode *) n;
1059 check_tnode(tn);
1061 if (tkey_sub_equals(tn->key, pos, tn->pos-pos, key)) {
1062 tp = tn;
1063 pos = tn->pos + tn->bits;
1064 n = tnode_get_child(tn,
1065 tkey_extract_bits(key,
1066 tn->pos,
1067 tn->bits));
1069 BUG_ON(n && node_parent(n) != tn);
1070 } else
1071 break;
1075 * n ----> NULL, LEAF or TNODE
1077 * tp is n's (parent) ----> NULL or TNODE
1080 BUG_ON(tp && IS_LEAF(tp));
1082 /* Case 1: n is a leaf. Compare prefixes */
1084 if (n != NULL && IS_LEAF(n) && tkey_equals(key, n->key)) {
1085 l = (struct leaf *) n;
1086 li = leaf_info_new(plen);
1088 if (!li)
1089 return NULL;
1091 fa_head = &li->falh;
1092 insert_leaf_info(&l->list, li);
1093 goto done;
1095 l = leaf_new();
1097 if (!l)
1098 return NULL;
1100 l->key = key;
1101 li = leaf_info_new(plen);
1103 if (!li) {
1104 free_leaf(l);
1105 return NULL;
1108 fa_head = &li->falh;
1109 insert_leaf_info(&l->list, li);
1111 if (t->trie && n == NULL) {
1112 /* Case 2: n is NULL, and will just insert a new leaf */
1114 node_set_parent((struct rt_trie_node *)l, tp);
1116 cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1117 put_child(t, (struct tnode *)tp, cindex, (struct rt_trie_node *)l);
1118 } else {
1119 /* Case 3: n is a LEAF or a TNODE and the key doesn't match. */
1121 * Add a new tnode here
1122 * first tnode need some special handling
1125 if (tp)
1126 pos = tp->pos+tp->bits;
1127 else
1128 pos = 0;
1130 if (n) {
1131 newpos = tkey_mismatch(key, pos, n->key);
1132 tn = tnode_new(n->key, newpos, 1);
1133 } else {
1134 newpos = 0;
1135 tn = tnode_new(key, newpos, 1); /* First tnode */
1138 if (!tn) {
1139 free_leaf_info(li);
1140 free_leaf(l);
1141 return NULL;
1144 node_set_parent((struct rt_trie_node *)tn, tp);
1146 missbit = tkey_extract_bits(key, newpos, 1);
1147 put_child(t, tn, missbit, (struct rt_trie_node *)l);
1148 put_child(t, tn, 1-missbit, n);
1150 if (tp) {
1151 cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1152 put_child(t, (struct tnode *)tp, cindex,
1153 (struct rt_trie_node *)tn);
1154 } else {
1155 rcu_assign_pointer(t->trie, (struct rt_trie_node *)tn);
1156 tp = tn;
1160 if (tp && tp->pos + tp->bits > 32)
1161 pr_warning("fib_trie"
1162 " tp=%p pos=%d, bits=%d, key=%0x plen=%d\n",
1163 tp, tp->pos, tp->bits, key, plen);
1165 /* Rebalance the trie */
1167 trie_rebalance(t, tp);
1168 done:
1169 return fa_head;
1173 * Caller must hold RTNL.
1175 int fib_table_insert(struct fib_table *tb, struct fib_config *cfg)
1177 struct trie *t = (struct trie *) tb->tb_data;
1178 struct fib_alias *fa, *new_fa;
1179 struct list_head *fa_head = NULL;
1180 struct fib_info *fi;
1181 int plen = cfg->fc_dst_len;
1182 u8 tos = cfg->fc_tos;
1183 u32 key, mask;
1184 int err;
1185 struct leaf *l;
1187 if (plen > 32)
1188 return -EINVAL;
1190 key = ntohl(cfg->fc_dst);
1192 pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1194 mask = ntohl(inet_make_mask(plen));
1196 if (key & ~mask)
1197 return -EINVAL;
1199 key = key & mask;
1201 fi = fib_create_info(cfg);
1202 if (IS_ERR(fi)) {
1203 err = PTR_ERR(fi);
1204 goto err;
1207 l = fib_find_node(t, key);
1208 fa = NULL;
1210 if (l) {
1211 fa_head = get_fa_head(l, plen);
1212 fa = fib_find_alias(fa_head, tos, fi->fib_priority);
1215 /* Now fa, if non-NULL, points to the first fib alias
1216 * with the same keys [prefix,tos,priority], if such key already
1217 * exists or to the node before which we will insert new one.
1219 * If fa is NULL, we will need to allocate a new one and
1220 * insert to the head of f.
1222 * If f is NULL, no fib node matched the destination key
1223 * and we need to allocate a new one of those as well.
1226 if (fa && fa->fa_tos == tos &&
1227 fa->fa_info->fib_priority == fi->fib_priority) {
1228 struct fib_alias *fa_first, *fa_match;
1230 err = -EEXIST;
1231 if (cfg->fc_nlflags & NLM_F_EXCL)
1232 goto out;
1234 /* We have 2 goals:
1235 * 1. Find exact match for type, scope, fib_info to avoid
1236 * duplicate routes
1237 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1239 fa_match = NULL;
1240 fa_first = fa;
1241 fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
1242 list_for_each_entry_continue(fa, fa_head, fa_list) {
1243 if (fa->fa_tos != tos)
1244 break;
1245 if (fa->fa_info->fib_priority != fi->fib_priority)
1246 break;
1247 if (fa->fa_type == cfg->fc_type &&
1248 fa->fa_scope == cfg->fc_scope &&
1249 fa->fa_info == fi) {
1250 fa_match = fa;
1251 break;
1255 if (cfg->fc_nlflags & NLM_F_REPLACE) {
1256 struct fib_info *fi_drop;
1257 u8 state;
1259 fa = fa_first;
1260 if (fa_match) {
1261 if (fa == fa_match)
1262 err = 0;
1263 goto out;
1265 err = -ENOBUFS;
1266 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1267 if (new_fa == NULL)
1268 goto out;
1270 fi_drop = fa->fa_info;
1271 new_fa->fa_tos = fa->fa_tos;
1272 new_fa->fa_info = fi;
1273 new_fa->fa_type = cfg->fc_type;
1274 new_fa->fa_scope = cfg->fc_scope;
1275 state = fa->fa_state;
1276 new_fa->fa_state = state & ~FA_S_ACCESSED;
1278 list_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1279 alias_free_mem_rcu(fa);
1281 fib_release_info(fi_drop);
1282 if (state & FA_S_ACCESSED)
1283 rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1284 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1285 tb->tb_id, &cfg->fc_nlinfo, NLM_F_REPLACE);
1287 goto succeeded;
1289 /* Error if we find a perfect match which
1290 * uses the same scope, type, and nexthop
1291 * information.
1293 if (fa_match)
1294 goto out;
1296 if (!(cfg->fc_nlflags & NLM_F_APPEND))
1297 fa = fa_first;
1299 err = -ENOENT;
1300 if (!(cfg->fc_nlflags & NLM_F_CREATE))
1301 goto out;
1303 err = -ENOBUFS;
1304 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1305 if (new_fa == NULL)
1306 goto out;
1308 new_fa->fa_info = fi;
1309 new_fa->fa_tos = tos;
1310 new_fa->fa_type = cfg->fc_type;
1311 new_fa->fa_scope = cfg->fc_scope;
1312 new_fa->fa_state = 0;
1314 * Insert new entry to the list.
1317 if (!fa_head) {
1318 fa_head = fib_insert_node(t, key, plen);
1319 if (unlikely(!fa_head)) {
1320 err = -ENOMEM;
1321 goto out_free_new_fa;
1325 list_add_tail_rcu(&new_fa->fa_list,
1326 (fa ? &fa->fa_list : fa_head));
1328 rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1329 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, tb->tb_id,
1330 &cfg->fc_nlinfo, 0);
1331 succeeded:
1332 return 0;
1334 out_free_new_fa:
1335 kmem_cache_free(fn_alias_kmem, new_fa);
1336 out:
1337 fib_release_info(fi);
1338 err:
1339 return err;
1342 /* should be called with rcu_read_lock */
1343 static int check_leaf(struct fib_table *tb, struct trie *t, struct leaf *l,
1344 t_key key, const struct flowi *flp,
1345 struct fib_result *res, int fib_flags)
1347 struct leaf_info *li;
1348 struct hlist_head *hhead = &l->list;
1349 struct hlist_node *node;
1351 hlist_for_each_entry_rcu(li, node, hhead, hlist) {
1352 int err;
1353 int plen = li->plen;
1354 __be32 mask = inet_make_mask(plen);
1356 if (l->key != (key & ntohl(mask)))
1357 continue;
1359 err = fib_semantic_match(tb, &li->falh, flp, res, plen, fib_flags);
1361 #ifdef CONFIG_IP_FIB_TRIE_STATS
1362 if (err <= 0)
1363 t->stats.semantic_match_passed++;
1364 else
1365 t->stats.semantic_match_miss++;
1366 #endif
1367 if (err <= 0)
1368 return err;
1371 return 1;
1374 int fib_table_lookup(struct fib_table *tb, const struct flowi *flp,
1375 struct fib_result *res, int fib_flags)
1377 struct trie *t = (struct trie *) tb->tb_data;
1378 int ret;
1379 struct rt_trie_node *n;
1380 struct tnode *pn;
1381 unsigned int pos, bits;
1382 t_key key = ntohl(flp->fl4_dst);
1383 unsigned int chopped_off;
1384 t_key cindex = 0;
1385 unsigned int current_prefix_length = KEYLENGTH;
1386 struct tnode *cn;
1387 t_key pref_mismatch;
1389 rcu_read_lock();
1391 n = rcu_dereference(t->trie);
1392 if (!n)
1393 goto failed;
1395 #ifdef CONFIG_IP_FIB_TRIE_STATS
1396 t->stats.gets++;
1397 #endif
1399 /* Just a leaf? */
1400 if (IS_LEAF(n)) {
1401 ret = check_leaf(tb, t, (struct leaf *)n, key, flp, res, fib_flags);
1402 goto found;
1405 pn = (struct tnode *) n;
1406 chopped_off = 0;
1408 while (pn) {
1409 pos = pn->pos;
1410 bits = pn->bits;
1412 if (!chopped_off)
1413 cindex = tkey_extract_bits(mask_pfx(key, current_prefix_length),
1414 pos, bits);
1416 n = tnode_get_child_rcu(pn, cindex);
1418 if (n == NULL) {
1419 #ifdef CONFIG_IP_FIB_TRIE_STATS
1420 t->stats.null_node_hit++;
1421 #endif
1422 goto backtrace;
1425 if (IS_LEAF(n)) {
1426 ret = check_leaf(tb, t, (struct leaf *)n, key, flp, res, fib_flags);
1427 if (ret > 0)
1428 goto backtrace;
1429 goto found;
1432 cn = (struct tnode *)n;
1435 * It's a tnode, and we can do some extra checks here if we
1436 * like, to avoid descending into a dead-end branch.
1437 * This tnode is in the parent's child array at index
1438 * key[p_pos..p_pos+p_bits] but potentially with some bits
1439 * chopped off, so in reality the index may be just a
1440 * subprefix, padded with zero at the end.
1441 * We can also take a look at any skipped bits in this
1442 * tnode - everything up to p_pos is supposed to be ok,
1443 * and the non-chopped bits of the index (se previous
1444 * paragraph) are also guaranteed ok, but the rest is
1445 * considered unknown.
1447 * The skipped bits are key[pos+bits..cn->pos].
1450 /* If current_prefix_length < pos+bits, we are already doing
1451 * actual prefix matching, which means everything from
1452 * pos+(bits-chopped_off) onward must be zero along some
1453 * branch of this subtree - otherwise there is *no* valid
1454 * prefix present. Here we can only check the skipped
1455 * bits. Remember, since we have already indexed into the
1456 * parent's child array, we know that the bits we chopped of
1457 * *are* zero.
1460 /* NOTA BENE: Checking only skipped bits
1461 for the new node here */
1463 if (current_prefix_length < pos+bits) {
1464 if (tkey_extract_bits(cn->key, current_prefix_length,
1465 cn->pos - current_prefix_length)
1466 || !(cn->child[0]))
1467 goto backtrace;
1471 * If chopped_off=0, the index is fully validated and we
1472 * only need to look at the skipped bits for this, the new,
1473 * tnode. What we actually want to do is to find out if
1474 * these skipped bits match our key perfectly, or if we will
1475 * have to count on finding a matching prefix further down,
1476 * because if we do, we would like to have some way of
1477 * verifying the existence of such a prefix at this point.
1480 /* The only thing we can do at this point is to verify that
1481 * any such matching prefix can indeed be a prefix to our
1482 * key, and if the bits in the node we are inspecting that
1483 * do not match our key are not ZERO, this cannot be true.
1484 * Thus, find out where there is a mismatch (before cn->pos)
1485 * and verify that all the mismatching bits are zero in the
1486 * new tnode's key.
1490 * Note: We aren't very concerned about the piece of
1491 * the key that precede pn->pos+pn->bits, since these
1492 * have already been checked. The bits after cn->pos
1493 * aren't checked since these are by definition
1494 * "unknown" at this point. Thus, what we want to see
1495 * is if we are about to enter the "prefix matching"
1496 * state, and in that case verify that the skipped
1497 * bits that will prevail throughout this subtree are
1498 * zero, as they have to be if we are to find a
1499 * matching prefix.
1502 pref_mismatch = mask_pfx(cn->key ^ key, cn->pos);
1505 * In short: If skipped bits in this node do not match
1506 * the search key, enter the "prefix matching"
1507 * state.directly.
1509 if (pref_mismatch) {
1510 int mp = KEYLENGTH - fls(pref_mismatch);
1512 if (tkey_extract_bits(cn->key, mp, cn->pos - mp) != 0)
1513 goto backtrace;
1515 if (current_prefix_length >= cn->pos)
1516 current_prefix_length = mp;
1519 pn = (struct tnode *)n; /* Descend */
1520 chopped_off = 0;
1521 continue;
1523 backtrace:
1524 chopped_off++;
1526 /* As zero don't change the child key (cindex) */
1527 while ((chopped_off <= pn->bits)
1528 && !(cindex & (1<<(chopped_off-1))))
1529 chopped_off++;
1531 /* Decrease current_... with bits chopped off */
1532 if (current_prefix_length > pn->pos + pn->bits - chopped_off)
1533 current_prefix_length = pn->pos + pn->bits
1534 - chopped_off;
1537 * Either we do the actual chop off according or if we have
1538 * chopped off all bits in this tnode walk up to our parent.
1541 if (chopped_off <= pn->bits) {
1542 cindex &= ~(1 << (chopped_off-1));
1543 } else {
1544 struct tnode *parent = node_parent_rcu((struct rt_trie_node *) pn);
1545 if (!parent)
1546 goto failed;
1548 /* Get Child's index */
1549 cindex = tkey_extract_bits(pn->key, parent->pos, parent->bits);
1550 pn = parent;
1551 chopped_off = 0;
1553 #ifdef CONFIG_IP_FIB_TRIE_STATS
1554 t->stats.backtrack++;
1555 #endif
1556 goto backtrace;
1559 failed:
1560 ret = 1;
1561 found:
1562 rcu_read_unlock();
1563 return ret;
1567 * Remove the leaf and return parent.
1569 static void trie_leaf_remove(struct trie *t, struct leaf *l)
1571 struct tnode *tp = node_parent((struct rt_trie_node *) l);
1573 pr_debug("entering trie_leaf_remove(%p)\n", l);
1575 if (tp) {
1576 t_key cindex = tkey_extract_bits(l->key, tp->pos, tp->bits);
1577 put_child(t, (struct tnode *)tp, cindex, NULL);
1578 trie_rebalance(t, tp);
1579 } else
1580 rcu_assign_pointer(t->trie, NULL);
1582 free_leaf(l);
1586 * Caller must hold RTNL.
1588 int fib_table_delete(struct fib_table *tb, struct fib_config *cfg)
1590 struct trie *t = (struct trie *) tb->tb_data;
1591 u32 key, mask;
1592 int plen = cfg->fc_dst_len;
1593 u8 tos = cfg->fc_tos;
1594 struct fib_alias *fa, *fa_to_delete;
1595 struct list_head *fa_head;
1596 struct leaf *l;
1597 struct leaf_info *li;
1599 if (plen > 32)
1600 return -EINVAL;
1602 key = ntohl(cfg->fc_dst);
1603 mask = ntohl(inet_make_mask(plen));
1605 if (key & ~mask)
1606 return -EINVAL;
1608 key = key & mask;
1609 l = fib_find_node(t, key);
1611 if (!l)
1612 return -ESRCH;
1614 fa_head = get_fa_head(l, plen);
1615 fa = fib_find_alias(fa_head, tos, 0);
1617 if (!fa)
1618 return -ESRCH;
1620 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1622 fa_to_delete = NULL;
1623 fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
1624 list_for_each_entry_continue(fa, fa_head, fa_list) {
1625 struct fib_info *fi = fa->fa_info;
1627 if (fa->fa_tos != tos)
1628 break;
1630 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1631 (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1632 fa->fa_scope == cfg->fc_scope) &&
1633 (!cfg->fc_protocol ||
1634 fi->fib_protocol == cfg->fc_protocol) &&
1635 fib_nh_match(cfg, fi) == 0) {
1636 fa_to_delete = fa;
1637 break;
1641 if (!fa_to_delete)
1642 return -ESRCH;
1644 fa = fa_to_delete;
1645 rtmsg_fib(RTM_DELROUTE, htonl(key), fa, plen, tb->tb_id,
1646 &cfg->fc_nlinfo, 0);
1648 l = fib_find_node(t, key);
1649 li = find_leaf_info(l, plen);
1651 list_del_rcu(&fa->fa_list);
1653 if (list_empty(fa_head)) {
1654 hlist_del_rcu(&li->hlist);
1655 free_leaf_info(li);
1658 if (hlist_empty(&l->list))
1659 trie_leaf_remove(t, l);
1661 if (fa->fa_state & FA_S_ACCESSED)
1662 rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1664 fib_release_info(fa->fa_info);
1665 alias_free_mem_rcu(fa);
1666 return 0;
1669 static int trie_flush_list(struct list_head *head)
1671 struct fib_alias *fa, *fa_node;
1672 int found = 0;
1674 list_for_each_entry_safe(fa, fa_node, head, fa_list) {
1675 struct fib_info *fi = fa->fa_info;
1677 if (fi && (fi->fib_flags & RTNH_F_DEAD)) {
1678 list_del_rcu(&fa->fa_list);
1679 fib_release_info(fa->fa_info);
1680 alias_free_mem_rcu(fa);
1681 found++;
1684 return found;
1687 static int trie_flush_leaf(struct leaf *l)
1689 int found = 0;
1690 struct hlist_head *lih = &l->list;
1691 struct hlist_node *node, *tmp;
1692 struct leaf_info *li = NULL;
1694 hlist_for_each_entry_safe(li, node, tmp, lih, hlist) {
1695 found += trie_flush_list(&li->falh);
1697 if (list_empty(&li->falh)) {
1698 hlist_del_rcu(&li->hlist);
1699 free_leaf_info(li);
1702 return found;
1706 * Scan for the next right leaf starting at node p->child[idx]
1707 * Since we have back pointer, no recursion necessary.
1709 static struct leaf *leaf_walk_rcu(struct tnode *p, struct rt_trie_node *c)
1711 do {
1712 t_key idx;
1714 if (c)
1715 idx = tkey_extract_bits(c->key, p->pos, p->bits) + 1;
1716 else
1717 idx = 0;
1719 while (idx < 1u << p->bits) {
1720 c = tnode_get_child_rcu(p, idx++);
1721 if (!c)
1722 continue;
1724 if (IS_LEAF(c)) {
1725 prefetch(p->child[idx]);
1726 return (struct leaf *) c;
1729 /* Rescan start scanning in new node */
1730 p = (struct tnode *) c;
1731 idx = 0;
1734 /* Node empty, walk back up to parent */
1735 c = (struct rt_trie_node *) p;
1736 } while ((p = node_parent_rcu(c)) != NULL);
1738 return NULL; /* Root of trie */
1741 static struct leaf *trie_firstleaf(struct trie *t)
1743 struct tnode *n = (struct tnode *)rcu_dereference_rtnl(t->trie);
1745 if (!n)
1746 return NULL;
1748 if (IS_LEAF(n)) /* trie is just a leaf */
1749 return (struct leaf *) n;
1751 return leaf_walk_rcu(n, NULL);
1754 static struct leaf *trie_nextleaf(struct leaf *l)
1756 struct rt_trie_node *c = (struct rt_trie_node *) l;
1757 struct tnode *p = node_parent_rcu(c);
1759 if (!p)
1760 return NULL; /* trie with just one leaf */
1762 return leaf_walk_rcu(p, c);
1765 static struct leaf *trie_leafindex(struct trie *t, int index)
1767 struct leaf *l = trie_firstleaf(t);
1769 while (l && index-- > 0)
1770 l = trie_nextleaf(l);
1772 return l;
1777 * Caller must hold RTNL.
1779 int fib_table_flush(struct fib_table *tb)
1781 struct trie *t = (struct trie *) tb->tb_data;
1782 struct leaf *l, *ll = NULL;
1783 int found = 0;
1785 for (l = trie_firstleaf(t); l; l = trie_nextleaf(l)) {
1786 found += trie_flush_leaf(l);
1788 if (ll && hlist_empty(&ll->list))
1789 trie_leaf_remove(t, ll);
1790 ll = l;
1793 if (ll && hlist_empty(&ll->list))
1794 trie_leaf_remove(t, ll);
1796 pr_debug("trie_flush found=%d\n", found);
1797 return found;
1800 void fib_free_table(struct fib_table *tb)
1802 kfree(tb);
1805 static int fn_trie_dump_fa(t_key key, int plen, struct list_head *fah,
1806 struct fib_table *tb,
1807 struct sk_buff *skb, struct netlink_callback *cb)
1809 int i, s_i;
1810 struct fib_alias *fa;
1811 __be32 xkey = htonl(key);
1813 s_i = cb->args[5];
1814 i = 0;
1816 /* rcu_read_lock is hold by caller */
1818 list_for_each_entry_rcu(fa, fah, fa_list) {
1819 if (i < s_i) {
1820 i++;
1821 continue;
1824 if (fib_dump_info(skb, NETLINK_CB(cb->skb).pid,
1825 cb->nlh->nlmsg_seq,
1826 RTM_NEWROUTE,
1827 tb->tb_id,
1828 fa->fa_type,
1829 fa->fa_scope,
1830 xkey,
1831 plen,
1832 fa->fa_tos,
1833 fa->fa_info, NLM_F_MULTI) < 0) {
1834 cb->args[5] = i;
1835 return -1;
1837 i++;
1839 cb->args[5] = i;
1840 return skb->len;
1843 static int fn_trie_dump_leaf(struct leaf *l, struct fib_table *tb,
1844 struct sk_buff *skb, struct netlink_callback *cb)
1846 struct leaf_info *li;
1847 struct hlist_node *node;
1848 int i, s_i;
1850 s_i = cb->args[4];
1851 i = 0;
1853 /* rcu_read_lock is hold by caller */
1854 hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
1855 if (i < s_i) {
1856 i++;
1857 continue;
1860 if (i > s_i)
1861 cb->args[5] = 0;
1863 if (list_empty(&li->falh))
1864 continue;
1866 if (fn_trie_dump_fa(l->key, li->plen, &li->falh, tb, skb, cb) < 0) {
1867 cb->args[4] = i;
1868 return -1;
1870 i++;
1873 cb->args[4] = i;
1874 return skb->len;
1877 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
1878 struct netlink_callback *cb)
1880 struct leaf *l;
1881 struct trie *t = (struct trie *) tb->tb_data;
1882 t_key key = cb->args[2];
1883 int count = cb->args[3];
1885 rcu_read_lock();
1886 /* Dump starting at last key.
1887 * Note: 0.0.0.0/0 (ie default) is first key.
1889 if (count == 0)
1890 l = trie_firstleaf(t);
1891 else {
1892 /* Normally, continue from last key, but if that is missing
1893 * fallback to using slow rescan
1895 l = fib_find_node(t, key);
1896 if (!l)
1897 l = trie_leafindex(t, count);
1900 while (l) {
1901 cb->args[2] = l->key;
1902 if (fn_trie_dump_leaf(l, tb, skb, cb) < 0) {
1903 cb->args[3] = count;
1904 rcu_read_unlock();
1905 return -1;
1908 ++count;
1909 l = trie_nextleaf(l);
1910 memset(&cb->args[4], 0,
1911 sizeof(cb->args) - 4*sizeof(cb->args[0]));
1913 cb->args[3] = count;
1914 rcu_read_unlock();
1916 return skb->len;
1919 void __init fib_trie_init(void)
1921 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
1922 sizeof(struct fib_alias),
1923 0, SLAB_PANIC, NULL);
1925 trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
1926 max(sizeof(struct leaf),
1927 sizeof(struct leaf_info)),
1928 0, SLAB_PANIC, NULL);
1932 struct fib_table *fib_trie_table(u32 id)
1934 struct fib_table *tb;
1935 struct trie *t;
1937 tb = kmalloc(sizeof(struct fib_table) + sizeof(struct trie),
1938 GFP_KERNEL);
1939 if (tb == NULL)
1940 return NULL;
1942 tb->tb_id = id;
1943 tb->tb_default = -1;
1945 t = (struct trie *) tb->tb_data;
1946 memset(t, 0, sizeof(*t));
1948 if (id == RT_TABLE_LOCAL)
1949 pr_info("IPv4 FIB: Using LC-trie version %s\n", VERSION);
1951 return tb;
1954 #ifdef CONFIG_PROC_FS
1955 /* Depth first Trie walk iterator */
1956 struct fib_trie_iter {
1957 struct seq_net_private p;
1958 struct fib_table *tb;
1959 struct tnode *tnode;
1960 unsigned int index;
1961 unsigned int depth;
1964 static struct rt_trie_node *fib_trie_get_next(struct fib_trie_iter *iter)
1966 struct tnode *tn = iter->tnode;
1967 unsigned int cindex = iter->index;
1968 struct tnode *p;
1970 /* A single entry routing table */
1971 if (!tn)
1972 return NULL;
1974 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
1975 iter->tnode, iter->index, iter->depth);
1976 rescan:
1977 while (cindex < (1<<tn->bits)) {
1978 struct rt_trie_node *n = tnode_get_child_rcu(tn, cindex);
1980 if (n) {
1981 if (IS_LEAF(n)) {
1982 iter->tnode = tn;
1983 iter->index = cindex + 1;
1984 } else {
1985 /* push down one level */
1986 iter->tnode = (struct tnode *) n;
1987 iter->index = 0;
1988 ++iter->depth;
1990 return n;
1993 ++cindex;
1996 /* Current node exhausted, pop back up */
1997 p = node_parent_rcu((struct rt_trie_node *)tn);
1998 if (p) {
1999 cindex = tkey_extract_bits(tn->key, p->pos, p->bits)+1;
2000 tn = p;
2001 --iter->depth;
2002 goto rescan;
2005 /* got root? */
2006 return NULL;
2009 static struct rt_trie_node *fib_trie_get_first(struct fib_trie_iter *iter,
2010 struct trie *t)
2012 struct rt_trie_node *n;
2014 if (!t)
2015 return NULL;
2017 n = rcu_dereference(t->trie);
2018 if (!n)
2019 return NULL;
2021 if (IS_TNODE(n)) {
2022 iter->tnode = (struct tnode *) n;
2023 iter->index = 0;
2024 iter->depth = 1;
2025 } else {
2026 iter->tnode = NULL;
2027 iter->index = 0;
2028 iter->depth = 0;
2031 return n;
2034 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2036 struct rt_trie_node *n;
2037 struct fib_trie_iter iter;
2039 memset(s, 0, sizeof(*s));
2041 rcu_read_lock();
2042 for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2043 if (IS_LEAF(n)) {
2044 struct leaf *l = (struct leaf *)n;
2045 struct leaf_info *li;
2046 struct hlist_node *tmp;
2048 s->leaves++;
2049 s->totdepth += iter.depth;
2050 if (iter.depth > s->maxdepth)
2051 s->maxdepth = iter.depth;
2053 hlist_for_each_entry_rcu(li, tmp, &l->list, hlist)
2054 ++s->prefixes;
2055 } else {
2056 const struct tnode *tn = (const struct tnode *) n;
2057 int i;
2059 s->tnodes++;
2060 if (tn->bits < MAX_STAT_DEPTH)
2061 s->nodesizes[tn->bits]++;
2063 for (i = 0; i < (1<<tn->bits); i++)
2064 if (!tn->child[i])
2065 s->nullpointers++;
2068 rcu_read_unlock();
2072 * This outputs /proc/net/fib_triestats
2074 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2076 unsigned int i, max, pointers, bytes, avdepth;
2078 if (stat->leaves)
2079 avdepth = stat->totdepth*100 / stat->leaves;
2080 else
2081 avdepth = 0;
2083 seq_printf(seq, "\tAver depth: %u.%02d\n",
2084 avdepth / 100, avdepth % 100);
2085 seq_printf(seq, "\tMax depth: %u\n", stat->maxdepth);
2087 seq_printf(seq, "\tLeaves: %u\n", stat->leaves);
2088 bytes = sizeof(struct leaf) * stat->leaves;
2090 seq_printf(seq, "\tPrefixes: %u\n", stat->prefixes);
2091 bytes += sizeof(struct leaf_info) * stat->prefixes;
2093 seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2094 bytes += sizeof(struct tnode) * stat->tnodes;
2096 max = MAX_STAT_DEPTH;
2097 while (max > 0 && stat->nodesizes[max-1] == 0)
2098 max--;
2100 pointers = 0;
2101 for (i = 1; i <= max; i++)
2102 if (stat->nodesizes[i] != 0) {
2103 seq_printf(seq, " %u: %u", i, stat->nodesizes[i]);
2104 pointers += (1<<i) * stat->nodesizes[i];
2106 seq_putc(seq, '\n');
2107 seq_printf(seq, "\tPointers: %u\n", pointers);
2109 bytes += sizeof(struct rt_trie_node *) * pointers;
2110 seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2111 seq_printf(seq, "Total size: %u kB\n", (bytes + 1023) / 1024);
2114 #ifdef CONFIG_IP_FIB_TRIE_STATS
2115 static void trie_show_usage(struct seq_file *seq,
2116 const struct trie_use_stats *stats)
2118 seq_printf(seq, "\nCounters:\n---------\n");
2119 seq_printf(seq, "gets = %u\n", stats->gets);
2120 seq_printf(seq, "backtracks = %u\n", stats->backtrack);
2121 seq_printf(seq, "semantic match passed = %u\n",
2122 stats->semantic_match_passed);
2123 seq_printf(seq, "semantic match miss = %u\n",
2124 stats->semantic_match_miss);
2125 seq_printf(seq, "null node hit= %u\n", stats->null_node_hit);
2126 seq_printf(seq, "skipped node resize = %u\n\n",
2127 stats->resize_node_skipped);
2129 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2131 static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2133 if (tb->tb_id == RT_TABLE_LOCAL)
2134 seq_puts(seq, "Local:\n");
2135 else if (tb->tb_id == RT_TABLE_MAIN)
2136 seq_puts(seq, "Main:\n");
2137 else
2138 seq_printf(seq, "Id %d:\n", tb->tb_id);
2142 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2144 struct net *net = (struct net *)seq->private;
2145 unsigned int h;
2147 seq_printf(seq,
2148 "Basic info: size of leaf:"
2149 " %Zd bytes, size of tnode: %Zd bytes.\n",
2150 sizeof(struct leaf), sizeof(struct tnode));
2152 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2153 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2154 struct hlist_node *node;
2155 struct fib_table *tb;
2157 hlist_for_each_entry_rcu(tb, node, head, tb_hlist) {
2158 struct trie *t = (struct trie *) tb->tb_data;
2159 struct trie_stat stat;
2161 if (!t)
2162 continue;
2164 fib_table_print(seq, tb);
2166 trie_collect_stats(t, &stat);
2167 trie_show_stats(seq, &stat);
2168 #ifdef CONFIG_IP_FIB_TRIE_STATS
2169 trie_show_usage(seq, &t->stats);
2170 #endif
2174 return 0;
2177 static int fib_triestat_seq_open(struct inode *inode, struct file *file)
2179 return single_open_net(inode, file, fib_triestat_seq_show);
2182 static const struct file_operations fib_triestat_fops = {
2183 .owner = THIS_MODULE,
2184 .open = fib_triestat_seq_open,
2185 .read = seq_read,
2186 .llseek = seq_lseek,
2187 .release = single_release_net,
2190 static struct rt_trie_node *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2192 struct fib_trie_iter *iter = seq->private;
2193 struct net *net = seq_file_net(seq);
2194 loff_t idx = 0;
2195 unsigned int h;
2197 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2198 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2199 struct hlist_node *node;
2200 struct fib_table *tb;
2202 hlist_for_each_entry_rcu(tb, node, head, tb_hlist) {
2203 struct rt_trie_node *n;
2205 for (n = fib_trie_get_first(iter,
2206 (struct trie *) tb->tb_data);
2207 n; n = fib_trie_get_next(iter))
2208 if (pos == idx++) {
2209 iter->tb = tb;
2210 return n;
2215 return NULL;
2218 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2219 __acquires(RCU)
2221 rcu_read_lock();
2222 return fib_trie_get_idx(seq, *pos);
2225 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2227 struct fib_trie_iter *iter = seq->private;
2228 struct net *net = seq_file_net(seq);
2229 struct fib_table *tb = iter->tb;
2230 struct hlist_node *tb_node;
2231 unsigned int h;
2232 struct rt_trie_node *n;
2234 ++*pos;
2235 /* next node in same table */
2236 n = fib_trie_get_next(iter);
2237 if (n)
2238 return n;
2240 /* walk rest of this hash chain */
2241 h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2242 while ( (tb_node = rcu_dereference(tb->tb_hlist.next)) ) {
2243 tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2244 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2245 if (n)
2246 goto found;
2249 /* new hash chain */
2250 while (++h < FIB_TABLE_HASHSZ) {
2251 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2252 hlist_for_each_entry_rcu(tb, tb_node, head, tb_hlist) {
2253 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2254 if (n)
2255 goto found;
2258 return NULL;
2260 found:
2261 iter->tb = tb;
2262 return n;
2265 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2266 __releases(RCU)
2268 rcu_read_unlock();
2271 static void seq_indent(struct seq_file *seq, int n)
2273 while (n-- > 0)
2274 seq_puts(seq, " ");
2277 static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2279 switch (s) {
2280 case RT_SCOPE_UNIVERSE: return "universe";
2281 case RT_SCOPE_SITE: return "site";
2282 case RT_SCOPE_LINK: return "link";
2283 case RT_SCOPE_HOST: return "host";
2284 case RT_SCOPE_NOWHERE: return "nowhere";
2285 default:
2286 snprintf(buf, len, "scope=%d", s);
2287 return buf;
2291 static const char *const rtn_type_names[__RTN_MAX] = {
2292 [RTN_UNSPEC] = "UNSPEC",
2293 [RTN_UNICAST] = "UNICAST",
2294 [RTN_LOCAL] = "LOCAL",
2295 [RTN_BROADCAST] = "BROADCAST",
2296 [RTN_ANYCAST] = "ANYCAST",
2297 [RTN_MULTICAST] = "MULTICAST",
2298 [RTN_BLACKHOLE] = "BLACKHOLE",
2299 [RTN_UNREACHABLE] = "UNREACHABLE",
2300 [RTN_PROHIBIT] = "PROHIBIT",
2301 [RTN_THROW] = "THROW",
2302 [RTN_NAT] = "NAT",
2303 [RTN_XRESOLVE] = "XRESOLVE",
2306 static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2308 if (t < __RTN_MAX && rtn_type_names[t])
2309 return rtn_type_names[t];
2310 snprintf(buf, len, "type %u", t);
2311 return buf;
2314 /* Pretty print the trie */
2315 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2317 const struct fib_trie_iter *iter = seq->private;
2318 struct rt_trie_node *n = v;
2320 if (!node_parent_rcu(n))
2321 fib_table_print(seq, iter->tb);
2323 if (IS_TNODE(n)) {
2324 struct tnode *tn = (struct tnode *) n;
2325 __be32 prf = htonl(mask_pfx(tn->key, tn->pos));
2327 seq_indent(seq, iter->depth-1);
2328 seq_printf(seq, " +-- %pI4/%d %d %d %d\n",
2329 &prf, tn->pos, tn->bits, tn->full_children,
2330 tn->empty_children);
2332 } else {
2333 struct leaf *l = (struct leaf *) n;
2334 struct leaf_info *li;
2335 struct hlist_node *node;
2336 __be32 val = htonl(l->key);
2338 seq_indent(seq, iter->depth);
2339 seq_printf(seq, " |-- %pI4\n", &val);
2341 hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
2342 struct fib_alias *fa;
2344 list_for_each_entry_rcu(fa, &li->falh, fa_list) {
2345 char buf1[32], buf2[32];
2347 seq_indent(seq, iter->depth+1);
2348 seq_printf(seq, " /%d %s %s", li->plen,
2349 rtn_scope(buf1, sizeof(buf1),
2350 fa->fa_scope),
2351 rtn_type(buf2, sizeof(buf2),
2352 fa->fa_type));
2353 if (fa->fa_tos)
2354 seq_printf(seq, " tos=%d", fa->fa_tos);
2355 seq_putc(seq, '\n');
2360 return 0;
2363 static const struct seq_operations fib_trie_seq_ops = {
2364 .start = fib_trie_seq_start,
2365 .next = fib_trie_seq_next,
2366 .stop = fib_trie_seq_stop,
2367 .show = fib_trie_seq_show,
2370 static int fib_trie_seq_open(struct inode *inode, struct file *file)
2372 return seq_open_net(inode, file, &fib_trie_seq_ops,
2373 sizeof(struct fib_trie_iter));
2376 static const struct file_operations fib_trie_fops = {
2377 .owner = THIS_MODULE,
2378 .open = fib_trie_seq_open,
2379 .read = seq_read,
2380 .llseek = seq_lseek,
2381 .release = seq_release_net,
2384 struct fib_route_iter {
2385 struct seq_net_private p;
2386 struct trie *main_trie;
2387 loff_t pos;
2388 t_key key;
2391 static struct leaf *fib_route_get_idx(struct fib_route_iter *iter, loff_t pos)
2393 struct leaf *l = NULL;
2394 struct trie *t = iter->main_trie;
2396 /* use cache location of last found key */
2397 if (iter->pos > 0 && pos >= iter->pos && (l = fib_find_node(t, iter->key)))
2398 pos -= iter->pos;
2399 else {
2400 iter->pos = 0;
2401 l = trie_firstleaf(t);
2404 while (l && pos-- > 0) {
2405 iter->pos++;
2406 l = trie_nextleaf(l);
2409 if (l)
2410 iter->key = pos; /* remember it */
2411 else
2412 iter->pos = 0; /* forget it */
2414 return l;
2417 static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2418 __acquires(RCU)
2420 struct fib_route_iter *iter = seq->private;
2421 struct fib_table *tb;
2423 rcu_read_lock();
2424 tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2425 if (!tb)
2426 return NULL;
2428 iter->main_trie = (struct trie *) tb->tb_data;
2429 if (*pos == 0)
2430 return SEQ_START_TOKEN;
2431 else
2432 return fib_route_get_idx(iter, *pos - 1);
2435 static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2437 struct fib_route_iter *iter = seq->private;
2438 struct leaf *l = v;
2440 ++*pos;
2441 if (v == SEQ_START_TOKEN) {
2442 iter->pos = 0;
2443 l = trie_firstleaf(iter->main_trie);
2444 } else {
2445 iter->pos++;
2446 l = trie_nextleaf(l);
2449 if (l)
2450 iter->key = l->key;
2451 else
2452 iter->pos = 0;
2453 return l;
2456 static void fib_route_seq_stop(struct seq_file *seq, void *v)
2457 __releases(RCU)
2459 rcu_read_unlock();
2462 static unsigned int fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2464 unsigned int flags = 0;
2466 if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2467 flags = RTF_REJECT;
2468 if (fi && fi->fib_nh->nh_gw)
2469 flags |= RTF_GATEWAY;
2470 if (mask == htonl(0xFFFFFFFF))
2471 flags |= RTF_HOST;
2472 flags |= RTF_UP;
2473 return flags;
2477 * This outputs /proc/net/route.
2478 * The format of the file is not supposed to be changed
2479 * and needs to be same as fib_hash output to avoid breaking
2480 * legacy utilities
2482 static int fib_route_seq_show(struct seq_file *seq, void *v)
2484 struct leaf *l = v;
2485 struct leaf_info *li;
2486 struct hlist_node *node;
2488 if (v == SEQ_START_TOKEN) {
2489 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2490 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2491 "\tWindow\tIRTT");
2492 return 0;
2495 hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
2496 struct fib_alias *fa;
2497 __be32 mask, prefix;
2499 mask = inet_make_mask(li->plen);
2500 prefix = htonl(l->key);
2502 list_for_each_entry_rcu(fa, &li->falh, fa_list) {
2503 const struct fib_info *fi = fa->fa_info;
2504 unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2505 int len;
2507 if (fa->fa_type == RTN_BROADCAST
2508 || fa->fa_type == RTN_MULTICAST)
2509 continue;
2511 if (fi)
2512 seq_printf(seq,
2513 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2514 "%d\t%08X\t%d\t%u\t%u%n",
2515 fi->fib_dev ? fi->fib_dev->name : "*",
2516 prefix,
2517 fi->fib_nh->nh_gw, flags, 0, 0,
2518 fi->fib_priority,
2519 mask,
2520 (fi->fib_advmss ?
2521 fi->fib_advmss + 40 : 0),
2522 fi->fib_window,
2523 fi->fib_rtt >> 3, &len);
2524 else
2525 seq_printf(seq,
2526 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2527 "%d\t%08X\t%d\t%u\t%u%n",
2528 prefix, 0, flags, 0, 0, 0,
2529 mask, 0, 0, 0, &len);
2531 seq_printf(seq, "%*s\n", 127 - len, "");
2535 return 0;
2538 static const struct seq_operations fib_route_seq_ops = {
2539 .start = fib_route_seq_start,
2540 .next = fib_route_seq_next,
2541 .stop = fib_route_seq_stop,
2542 .show = fib_route_seq_show,
2545 static int fib_route_seq_open(struct inode *inode, struct file *file)
2547 return seq_open_net(inode, file, &fib_route_seq_ops,
2548 sizeof(struct fib_route_iter));
2551 static const struct file_operations fib_route_fops = {
2552 .owner = THIS_MODULE,
2553 .open = fib_route_seq_open,
2554 .read = seq_read,
2555 .llseek = seq_lseek,
2556 .release = seq_release_net,
2559 int __net_init fib_proc_init(struct net *net)
2561 if (!proc_net_fops_create(net, "fib_trie", S_IRUGO, &fib_trie_fops))
2562 goto out1;
2564 if (!proc_net_fops_create(net, "fib_triestat", S_IRUGO,
2565 &fib_triestat_fops))
2566 goto out2;
2568 if (!proc_net_fops_create(net, "route", S_IRUGO, &fib_route_fops))
2569 goto out3;
2571 return 0;
2573 out3:
2574 proc_net_remove(net, "fib_triestat");
2575 out2:
2576 proc_net_remove(net, "fib_trie");
2577 out1:
2578 return -ENOMEM;
2581 void __net_exit fib_proc_exit(struct net *net)
2583 proc_net_remove(net, "fib_trie");
2584 proc_net_remove(net, "fib_triestat");
2585 proc_net_remove(net, "route");
2588 #endif /* CONFIG_PROC_FS */