fib_trie.c revision 22bd5b9b13f2931ac80949f8bfbc40e8cab05be7
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.
6 *
7 *   Robert Olsson <robert.olsson@its.uu.se> Uppsala Universitet
8 *     & Swedish University of Agricultural Sciences.
9 *
10 *   Jens Laas <jens.laas@data.slu.se> Swedish University of
11 *     Agricultural Sciences.
12 *
13 *   Hans Liss <hans.liss@its.uu.se>  Uppsala Universitet
14 *
15 * This work is based on the LPC-trie which is originally descibed in:
16 *
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/
20 *
21 *
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
24 *
25 *
26 * Code from fib_hash has been reused which includes the following header:
27 *
28 *
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.
32 *
33 *		IPv4 FIB: lookup engine and maintenance routines.
34 *
35 *
36 * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
37 *
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.
42 *
43 * Substantial contributions to this work comes from:
44 *
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>
49 */
50
51#define VERSION "0.409"
52
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"
83
84#define MAX_STAT_DEPTH 32
85
86#define KEYLENGTH (8*sizeof(t_key))
87
88typedef unsigned int t_key;
89
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)
94
95#define IS_TNODE(n) (!(n->parent & T_LEAF))
96#define IS_LEAF(n) (n->parent & T_LEAF)
97
98struct rt_trie_node {
99	unsigned long parent;
100	t_key key;
101};
102
103struct leaf {
104	unsigned long parent;
105	t_key key;
106	struct hlist_head list;
107	struct rcu_head rcu;
108};
109
110struct leaf_info {
111	struct hlist_node hlist;
112	struct rcu_head rcu;
113	int plen;
114	struct list_head falh;
115};
116
117struct 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;
128	};
129	struct rt_trie_node *child[0];
130};
131
132#ifdef CONFIG_IP_FIB_TRIE_STATS
133struct 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;
140};
141#endif
142
143struct 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];
151};
152
153struct trie {
154	struct rt_trie_node *trie;
155#ifdef CONFIG_IP_FIB_TRIE_STATS
156	struct trie_use_stats stats;
157#endif
158};
159
160static void put_child(struct trie *t, struct tnode *tn, int i, struct rt_trie_node *n);
161static void tnode_put_child_reorg(struct tnode *tn, int i, struct rt_trie_node *n,
162				  int wasfull);
163static struct rt_trie_node *resize(struct trie *t, struct tnode *tn);
164static struct tnode *inflate(struct trie *t, struct tnode *tn);
165static struct tnode *halve(struct trie *t, struct tnode *tn);
166/* tnodes to free after resize(); protected by RTNL */
167static struct tnode *tnode_free_head;
168static size_t tnode_free_size;
169
170/*
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.
174 */
175static const int sync_pages = 128;
176
177static struct kmem_cache *fn_alias_kmem __read_mostly;
178static struct kmem_cache *trie_leaf_kmem __read_mostly;
179
180static inline struct tnode *node_parent(struct rt_trie_node *node)
181{
182	return (struct tnode *)(node->parent & ~NODE_TYPE_MASK);
183}
184
185static inline struct tnode *node_parent_rcu(struct rt_trie_node *node)
186{
187	struct tnode *ret = node_parent(node);
188
189	return rcu_dereference_rtnl(ret);
190}
191
192/* Same as rcu_assign_pointer
193 * but that macro() assumes that value is a pointer.
194 */
195static inline void node_set_parent(struct rt_trie_node *node, struct tnode *ptr)
196{
197	smp_wmb();
198	node->parent = (unsigned long)ptr | NODE_TYPE(node);
199}
200
201static inline struct rt_trie_node *tnode_get_child(struct tnode *tn, unsigned int i)
202{
203	BUG_ON(i >= 1U << tn->bits);
204
205	return tn->child[i];
206}
207
208static inline struct rt_trie_node *tnode_get_child_rcu(struct tnode *tn, unsigned int i)
209{
210	struct rt_trie_node *ret = tnode_get_child(tn, i);
211
212	return rcu_dereference_rtnl(ret);
213}
214
215static inline int tnode_child_length(const struct tnode *tn)
216{
217	return 1 << tn->bits;
218}
219
220static inline t_key mask_pfx(t_key k, unsigned int l)
221{
222	return (l == 0) ? 0 : k >> (KEYLENGTH-l) << (KEYLENGTH-l);
223}
224
225static inline t_key tkey_extract_bits(t_key a, unsigned int offset, unsigned int bits)
226{
227	if (offset < KEYLENGTH)
228		return ((t_key)(a << offset)) >> (KEYLENGTH - bits);
229	else
230		return 0;
231}
232
233static inline int tkey_equals(t_key a, t_key b)
234{
235	return a == b;
236}
237
238static inline int tkey_sub_equals(t_key a, int offset, int bits, t_key b)
239{
240	if (bits == 0 || offset >= KEYLENGTH)
241		return 1;
242	bits = bits > KEYLENGTH ? KEYLENGTH : bits;
243	return ((a ^ b) << offset) >> (KEYLENGTH - bits) == 0;
244}
245
246static inline int tkey_mismatch(t_key a, int offset, t_key b)
247{
248	t_key diff = a ^ b;
249	int i = offset;
250
251	if (!diff)
252		return 0;
253	while ((diff << i) >> (KEYLENGTH-1) == 0)
254		i++;
255	return i;
256}
257
258/*
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.
262
263  Consider a node 'n' and its parent 'tp'.
264
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.
271
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().
278
279  if n is an internal node - a 'tnode' here, the various parts of its key
280  have many different meanings.
281
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
287
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
292
293  tp->pos = 7
294  tp->bits = 3
295  n->pos = 15
296  n->bits = 4
297
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.
301
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.
305
306  The bits from (tp->pos + tp->bits) to (n->pos - 1) - "S" - are skipped bits
307  for the node n.
308
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.
311
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.
314
315
316  The rest of the bits, from (n->pos + n->bits) onward, are completely unknown
317  at this point.
318
319*/
320
321static inline void check_tnode(const struct tnode *tn)
322{
323	WARN_ON(tn && tn->pos+tn->bits > 32);
324}
325
326static const int halve_threshold = 25;
327static const int inflate_threshold = 50;
328static const int halve_threshold_root = 15;
329static const int inflate_threshold_root = 30;
330
331static void __alias_free_mem(struct rcu_head *head)
332{
333	struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
334	kmem_cache_free(fn_alias_kmem, fa);
335}
336
337static inline void alias_free_mem_rcu(struct fib_alias *fa)
338{
339	call_rcu(&fa->rcu, __alias_free_mem);
340}
341
342static void __leaf_free_rcu(struct rcu_head *head)
343{
344	struct leaf *l = container_of(head, struct leaf, rcu);
345	kmem_cache_free(trie_leaf_kmem, l);
346}
347
348static inline void free_leaf(struct leaf *l)
349{
350	call_rcu_bh(&l->rcu, __leaf_free_rcu);
351}
352
353static void __leaf_info_free_rcu(struct rcu_head *head)
354{
355	kfree(container_of(head, struct leaf_info, rcu));
356}
357
358static inline void free_leaf_info(struct leaf_info *leaf)
359{
360	call_rcu(&leaf->rcu, __leaf_info_free_rcu);
361}
362
363static struct tnode *tnode_alloc(size_t size)
364{
365	if (size <= PAGE_SIZE)
366		return kzalloc(size, GFP_KERNEL);
367	else
368		return vzalloc(size);
369}
370
371static void __tnode_vfree(struct work_struct *arg)
372{
373	struct tnode *tn = container_of(arg, struct tnode, work);
374	vfree(tn);
375}
376
377static void __tnode_free_rcu(struct rcu_head *head)
378{
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);
382
383	if (size <= PAGE_SIZE)
384		kfree(tn);
385	else {
386		INIT_WORK(&tn->work, __tnode_vfree);
387		schedule_work(&tn->work);
388	}
389}
390
391static inline void tnode_free(struct tnode *tn)
392{
393	if (IS_LEAF(tn))
394		free_leaf((struct leaf *) tn);
395	else
396		call_rcu(&tn->rcu, __tnode_free_rcu);
397}
398
399static void tnode_free_safe(struct tnode *tn)
400{
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);
406}
407
408static void tnode_free_flush(void)
409{
410	struct tnode *tn;
411
412	while ((tn = tnode_free_head)) {
413		tnode_free_head = tn->tnode_free;
414		tn->tnode_free = NULL;
415		tnode_free(tn);
416	}
417
418	if (tnode_free_size >= PAGE_SIZE * sync_pages) {
419		tnode_free_size = 0;
420		synchronize_rcu();
421	}
422}
423
424static struct leaf *leaf_new(void)
425{
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);
430	}
431	return l;
432}
433
434static struct leaf_info *leaf_info_new(int plen)
435{
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);
440	}
441	return li;
442}
443
444static struct tnode *tnode_new(t_key key, int pos, int bits)
445{
446	size_t sz = sizeof(struct tnode) + (sizeof(struct rt_trie_node *) << bits);
447	struct tnode *tn = tnode_alloc(sz);
448
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;
456	}
457
458	pr_debug("AT %p s=%zu %zu\n", tn, sizeof(struct tnode),
459		 sizeof(struct rt_trie_node) << bits);
460	return tn;
461}
462
463/*
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
466 */
467
468static inline int tnode_full(const struct tnode *tn, const struct rt_trie_node *n)
469{
470	if (n == NULL || IS_LEAF(n))
471		return 0;
472
473	return ((struct tnode *) n)->pos == tn->pos + tn->bits;
474}
475
476static inline void put_child(struct trie *t, struct tnode *tn, int i,
477			     struct rt_trie_node *n)
478{
479	tnode_put_child_reorg(tn, i, n, -1);
480}
481
482 /*
483  * Add a child at position i overwriting the old value.
484  * Update the value of full_children and empty_children.
485  */
486
487static void tnode_put_child_reorg(struct tnode *tn, int i, struct rt_trie_node *n,
488				  int wasfull)
489{
490	struct rt_trie_node *chi = tn->child[i];
491	int isfull;
492
493	BUG_ON(i >= 1<<tn->bits);
494
495	/* update emptyChildren */
496	if (n == NULL && chi != NULL)
497		tn->empty_children++;
498	else if (n != NULL && chi == NULL)
499		tn->empty_children--;
500
501	/* update fullChildren */
502	if (wasfull == -1)
503		wasfull = tnode_full(tn, chi);
504
505	isfull = tnode_full(tn, n);
506	if (wasfull && !isfull)
507		tn->full_children--;
508	else if (!wasfull && isfull)
509		tn->full_children++;
510
511	if (n)
512		node_set_parent(n, tn);
513
514	rcu_assign_pointer(tn->child[i], n);
515}
516
517#define MAX_WORK 10
518static struct rt_trie_node *resize(struct trie *t, struct tnode *tn)
519{
520	int i;
521	struct tnode *old_tn;
522	int inflate_threshold_use;
523	int halve_threshold_use;
524	int max_work;
525
526	if (!tn)
527		return NULL;
528
529	pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
530		 tn, inflate_threshold, halve_threshold);
531
532	/* No children */
533	if (tn->empty_children == tnode_child_length(tn)) {
534		tnode_free_safe(tn);
535		return NULL;
536	}
537	/* One child */
538	if (tn->empty_children == tnode_child_length(tn) - 1)
539		goto one_child;
540	/*
541	 * Double as long as the resulting node has a number of
542	 * nonempty nodes that are above the threshold.
543	 */
544
545	/*
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'."
551	 *
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.
558	 *
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.
565	 *
566	 * A clearer way to write this would be:
567	 *
568	 * to_be_doubled = tn->full_children;
569	 * not_to_be_doubled = tnode_child_length(tn) - tn->empty_children -
570	 *     tn->full_children;
571	 *
572	 * new_child_length = tnode_child_length(tn) * 2;
573	 *
574	 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
575	 *      new_child_length;
576	 * if (new_fill_factor >= inflate_threshold)
577	 *
578	 * ...and so on, tho it would mess up the while () loop.
579	 *
580	 * anyway,
581	 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
582	 *      inflate_threshold
583	 *
584	 * avoid a division:
585	 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
586	 *      inflate_threshold * new_child_length
587	 *
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
591	 *
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
596	 *
597	 * shorten again:
598	 * 50 * (tn->full_children + tnode_child_length(tn) -
599	 *    tn->empty_children) >= inflate_threshold *
600	 *    tnode_child_length(tn)
601	 *
602	 */
603
604	check_tnode(tn);
605
606	/* Keep root node larger  */
607
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;
614	}
615
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))) {
621
622		old_tn = tn;
623		tn = inflate(t, tn);
624
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;
631		}
632	}
633
634	check_tnode(tn);
635
636	/* Return if at least one inflate is run */
637	if (max_work != MAX_WORK)
638		return (struct rt_trie_node *) tn;
639
640	/*
641	 * Halve as long as the number of empty children in this
642	 * node is above threshold.
643	 */
644
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)) {
649
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;
658		}
659	}
660
661
662	/* Only one child remains */
663	if (tn->empty_children == tnode_child_length(tn) - 1) {
664one_child:
665		for (i = 0; i < tnode_child_length(tn); i++) {
666			struct rt_trie_node *n;
667
668			n = tn->child[i];
669			if (!n)
670				continue;
671
672			/* compress one level */
673
674			node_set_parent(n, NULL);
675			tnode_free_safe(tn);
676			return n;
677		}
678	}
679	return (struct rt_trie_node *) tn;
680}
681
682static struct tnode *inflate(struct trie *t, struct tnode *tn)
683{
684	struct tnode *oldtnode = tn;
685	int olen = tnode_child_length(tn);
686	int i;
687
688	pr_debug("In inflate\n");
689
690	tn = tnode_new(oldtnode->key, oldtnode->pos, oldtnode->bits + 1);
691
692	if (!tn)
693		return ERR_PTR(-ENOMEM);
694
695	/*
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.
700	 */
701
702	for (i = 0; i < olen; i++) {
703		struct tnode *inode;
704
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;
712
713			left = tnode_new(inode->key&(~m), inode->pos + 1,
714					 inode->bits - 1);
715			if (!left)
716				goto nomem;
717
718			right = tnode_new(inode->key|m, inode->pos + 1,
719					  inode->bits - 1);
720
721			if (!right) {
722				tnode_free(left);
723				goto nomem;
724			}
725
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);
728		}
729	}
730
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;
736
737		/* An empty child */
738		if (node == NULL)
739			continue;
740
741		/* A leaf or an internal node with skipped bits */
742
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;
752		}
753
754		/* An internal node with two children */
755		inode = (struct tnode *) node;
756
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]);
760
761			tnode_free_safe(inode);
762			continue;
763		}
764
765		/* An internal node with more than two children */
766
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)
782		 */
783
784		/* Use the old key, but set the new significant
785		 *   bit to zero.
786		 */
787
788		left = (struct tnode *) tnode_get_child(tn, 2*i);
789		put_child(t, tn, 2*i, NULL);
790
791		BUG_ON(!left);
792
793		right = (struct tnode *) tnode_get_child(tn, 2*i+1);
794		put_child(t, tn, 2*i+1, NULL);
795
796		BUG_ON(!right);
797
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]);
802		}
803		put_child(t, tn, 2*i, resize(t, left));
804		put_child(t, tn, 2*i+1, resize(t, right));
805
806		tnode_free_safe(inode);
807	}
808	tnode_free_safe(oldtnode);
809	return tn;
810nomem:
811	{
812		int size = tnode_child_length(tn);
813		int j;
814
815		for (j = 0; j < size; j++)
816			if (tn->child[j])
817				tnode_free((struct tnode *)tn->child[j]);
818
819		tnode_free(tn);
820
821		return ERR_PTR(-ENOMEM);
822	}
823}
824
825static struct tnode *halve(struct trie *t, struct tnode *tn)
826{
827	struct tnode *oldtnode = tn;
828	struct rt_trie_node *left, *right;
829	int i;
830	int olen = tnode_child_length(tn);
831
832	pr_debug("In halve\n");
833
834	tn = tnode_new(oldtnode->key, oldtnode->pos, oldtnode->bits - 1);
835
836	if (!tn)
837		return ERR_PTR(-ENOMEM);
838
839	/*
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.
844	 */
845
846	for (i = 0; i < olen; i += 2) {
847		left = tnode_get_child(oldtnode, i);
848		right = tnode_get_child(oldtnode, i+1);
849
850		/* Two nonempty children */
851		if (left && right) {
852			struct tnode *newn;
853
854			newn = tnode_new(left->key, tn->pos + tn->bits, 1);
855
856			if (!newn)
857				goto nomem;
858
859			put_child(t, tn, i/2, (struct rt_trie_node *)newn);
860		}
861
862	}
863
864	for (i = 0; i < olen; i += 2) {
865		struct tnode *newBinNode;
866
867		left = tnode_get_child(oldtnode, i);
868		right = tnode_get_child(oldtnode, i+1);
869
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;
876		}
877
878		if (right == NULL) {
879			put_child(t, tn, i/2, left);
880			continue;
881		}
882
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));
889	}
890	tnode_free_safe(oldtnode);
891	return tn;
892nomem:
893	{
894		int size = tnode_child_length(tn);
895		int j;
896
897		for (j = 0; j < size; j++)
898			if (tn->child[j])
899				tnode_free((struct tnode *)tn->child[j]);
900
901		tnode_free(tn);
902
903		return ERR_PTR(-ENOMEM);
904	}
905}
906
907/* readside must use rcu_read_lock currently dump routines
908 via get_fa_head and dump */
909
910static struct leaf_info *find_leaf_info(struct leaf *l, int plen)
911{
912	struct hlist_head *head = &l->list;
913	struct hlist_node *node;
914	struct leaf_info *li;
915
916	hlist_for_each_entry_rcu(li, node, head, hlist)
917		if (li->plen == plen)
918			return li;
919
920	return NULL;
921}
922
923static inline struct list_head *get_fa_head(struct leaf *l, int plen)
924{
925	struct leaf_info *li = find_leaf_info(l, plen);
926
927	if (!li)
928		return NULL;
929
930	return &li->falh;
931}
932
933static void insert_leaf_info(struct hlist_head *head, struct leaf_info *new)
934{
935	struct leaf_info *li = NULL, *last = NULL;
936	struct hlist_node *node;
937
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;
944
945			last = li;
946		}
947		if (last)
948			hlist_add_after_rcu(&last->hlist, &new->hlist);
949		else
950			hlist_add_before_rcu(&new->hlist, &li->hlist);
951	}
952}
953
954/* rcu_read_lock needs to be hold by caller from readside */
955
956static struct leaf *
957fib_find_node(struct trie *t, u32 key)
958{
959	int pos;
960	struct tnode *tn;
961	struct rt_trie_node *n;
962
963	pos = 0;
964	n = rcu_dereference_rtnl(t->trie);
965
966	while (n != NULL &&  NODE_TYPE(n) == T_TNODE) {
967		tn = (struct tnode *) n;
968
969		check_tnode(tn);
970
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;
979	}
980	/* Case we have found a leaf. Compare prefixes */
981
982	if (n != NULL && IS_LEAF(n) && tkey_equals(key, n->key))
983		return (struct leaf *)n;
984
985	return NULL;
986}
987
988static void trie_rebalance(struct trie *t, struct tnode *tn)
989{
990	int wasfull;
991	t_key cindex, key;
992	struct tnode *tp;
993
994	key = tn->key;
995
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);
1000
1001		tnode_put_child_reorg((struct tnode *)tp, cindex,
1002				      (struct rt_trie_node *)tn, wasfull);
1003
1004		tp = node_parent((struct rt_trie_node *) tn);
1005		if (!tp)
1006			rcu_assign_pointer(t->trie, (struct rt_trie_node *)tn);
1007
1008		tnode_free_flush();
1009		if (!tp)
1010			break;
1011		tn = tp;
1012	}
1013
1014	/* Handle last (top) tnode */
1015	if (IS_TNODE(tn))
1016		tn = (struct tnode *)resize(t, (struct tnode *)tn);
1017
1018	rcu_assign_pointer(t->trie, (struct rt_trie_node *)tn);
1019	tnode_free_flush();
1020}
1021
1022/* only used from updater-side */
1023
1024static struct list_head *fib_insert_node(struct trie *t, u32 key, int plen)
1025{
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;
1034
1035	pos = 0;
1036	n = t->trie;
1037
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'!
1044	 *
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.
1047	 *
1048	 * If it doesn't, we have to replace it with a new T_TNODE.
1049	 *
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.
1054	 */
1055
1056	while (n != NULL &&  NODE_TYPE(n) == T_TNODE) {
1057		tn = (struct tnode *) n;
1058
1059		check_tnode(tn);
1060
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));
1068
1069			BUG_ON(n && node_parent(n) != tn);
1070		} else
1071			break;
1072	}
1073
1074	/*
1075	 * n  ----> NULL, LEAF or TNODE
1076	 *
1077	 * tp is n's (parent) ----> NULL or TNODE
1078	 */
1079
1080	BUG_ON(tp && IS_LEAF(tp));
1081
1082	/* Case 1: n is a leaf. Compare prefixes */
1083
1084	if (n != NULL && IS_LEAF(n) && tkey_equals(key, n->key)) {
1085		l = (struct leaf *) n;
1086		li = leaf_info_new(plen);
1087
1088		if (!li)
1089			return NULL;
1090
1091		fa_head = &li->falh;
1092		insert_leaf_info(&l->list, li);
1093		goto done;
1094	}
1095	l = leaf_new();
1096
1097	if (!l)
1098		return NULL;
1099
1100	l->key = key;
1101	li = leaf_info_new(plen);
1102
1103	if (!li) {
1104		free_leaf(l);
1105		return NULL;
1106	}
1107
1108	fa_head = &li->falh;
1109	insert_leaf_info(&l->list, li);
1110
1111	if (t->trie && n == NULL) {
1112		/* Case 2: n is NULL, and will just insert a new leaf */
1113
1114		node_set_parent((struct rt_trie_node *)l, tp);
1115
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. */
1120		/*
1121		 *  Add a new tnode here
1122		 *  first tnode need some special handling
1123		 */
1124
1125		if (tp)
1126			pos = tp->pos+tp->bits;
1127		else
1128			pos = 0;
1129
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 */
1136		}
1137
1138		if (!tn) {
1139			free_leaf_info(li);
1140			free_leaf(l);
1141			return NULL;
1142		}
1143
1144		node_set_parent((struct rt_trie_node *)tn, tp);
1145
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);
1149
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;
1157		}
1158	}
1159
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);
1164
1165	/* Rebalance the trie */
1166
1167	trie_rebalance(t, tp);
1168done:
1169	return fa_head;
1170}
1171
1172/*
1173 * Caller must hold RTNL.
1174 */
1175int fib_table_insert(struct fib_table *tb, struct fib_config *cfg)
1176{
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;
1186
1187	if (plen > 32)
1188		return -EINVAL;
1189
1190	key = ntohl(cfg->fc_dst);
1191
1192	pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1193
1194	mask = ntohl(inet_make_mask(plen));
1195
1196	if (key & ~mask)
1197		return -EINVAL;
1198
1199	key = key & mask;
1200
1201	fi = fib_create_info(cfg);
1202	if (IS_ERR(fi)) {
1203		err = PTR_ERR(fi);
1204		goto err;
1205	}
1206
1207	l = fib_find_node(t, key);
1208	fa = NULL;
1209
1210	if (l) {
1211		fa_head = get_fa_head(l, plen);
1212		fa = fib_find_alias(fa_head, tos, fi->fib_priority);
1213	}
1214
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.
1218	 *
1219	 * If fa is NULL, we will need to allocate a new one and
1220	 * insert to the head of f.
1221	 *
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.
1224	 */
1225
1226	if (fa && fa->fa_tos == tos &&
1227	    fa->fa_info->fib_priority == fi->fib_priority) {
1228		struct fib_alias *fa_first, *fa_match;
1229
1230		err = -EEXIST;
1231		if (cfg->fc_nlflags & NLM_F_EXCL)
1232			goto out;
1233
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
1238		 */
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;
1252			}
1253		}
1254
1255		if (cfg->fc_nlflags & NLM_F_REPLACE) {
1256			struct fib_info *fi_drop;
1257			u8 state;
1258
1259			fa = fa_first;
1260			if (fa_match) {
1261				if (fa == fa_match)
1262					err = 0;
1263				goto out;
1264			}
1265			err = -ENOBUFS;
1266			new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1267			if (new_fa == NULL)
1268				goto out;
1269
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;
1277
1278			list_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1279			alias_free_mem_rcu(fa);
1280
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);
1286
1287			goto succeeded;
1288		}
1289		/* Error if we find a perfect match which
1290		 * uses the same scope, type, and nexthop
1291		 * information.
1292		 */
1293		if (fa_match)
1294			goto out;
1295
1296		if (!(cfg->fc_nlflags & NLM_F_APPEND))
1297			fa = fa_first;
1298	}
1299	err = -ENOENT;
1300	if (!(cfg->fc_nlflags & NLM_F_CREATE))
1301		goto out;
1302
1303	err = -ENOBUFS;
1304	new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1305	if (new_fa == NULL)
1306		goto out;
1307
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;
1313	/*
1314	 * Insert new entry to the list.
1315	 */
1316
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;
1322		}
1323	}
1324
1325	list_add_tail_rcu(&new_fa->fa_list,
1326			  (fa ? &fa->fa_list : fa_head));
1327
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);
1331succeeded:
1332	return 0;
1333
1334out_free_new_fa:
1335	kmem_cache_free(fn_alias_kmem, new_fa);
1336out:
1337	fib_release_info(fi);
1338err:
1339	return err;
1340}
1341
1342/* should be called with rcu_read_lock */
1343static int check_leaf(struct fib_table *tb, struct trie *t, struct leaf *l,
1344		      t_key key,  const struct flowi4 *flp,
1345		      struct fib_result *res, int fib_flags)
1346{
1347	struct leaf_info *li;
1348	struct hlist_head *hhead = &l->list;
1349	struct hlist_node *node;
1350
1351	hlist_for_each_entry_rcu(li, node, hhead, hlist) {
1352		struct fib_alias *fa;
1353		int plen = li->plen;
1354		__be32 mask = inet_make_mask(plen);
1355
1356		if (l->key != (key & ntohl(mask)))
1357			continue;
1358
1359		list_for_each_entry_rcu(fa, &li->falh, fa_list) {
1360			struct fib_info *fi = fa->fa_info;
1361			int nhsel, err;
1362
1363			if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1364				continue;
1365			if (fa->fa_scope < flp->flowi4_scope)
1366				continue;
1367			fib_alias_accessed(fa);
1368			err = fib_props[fa->fa_type].error;
1369			if (err) {
1370#ifdef CONFIG_IP_FIB_TRIE_STATS
1371				t->stats.semantic_match_miss++;
1372#endif
1373				return 1;
1374			}
1375			if (fi->fib_flags & RTNH_F_DEAD)
1376				continue;
1377			for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
1378				const struct fib_nh *nh = &fi->fib_nh[nhsel];
1379
1380				if (nh->nh_flags & RTNH_F_DEAD)
1381					continue;
1382				if (flp->flowi4_oif && flp->flowi4_oif != nh->nh_oif)
1383					continue;
1384
1385#ifdef CONFIG_IP_FIB_TRIE_STATS
1386				t->stats.semantic_match_passed++;
1387#endif
1388				res->prefixlen = plen;
1389				res->nh_sel = nhsel;
1390				res->type = fa->fa_type;
1391				res->scope = fa->fa_scope;
1392				res->fi = fi;
1393				res->table = tb;
1394				res->fa_head = &li->falh;
1395				if (!(fib_flags & FIB_LOOKUP_NOREF))
1396					atomic_inc(&res->fi->fib_clntref);
1397				return 0;
1398			}
1399		}
1400
1401#ifdef CONFIG_IP_FIB_TRIE_STATS
1402		t->stats.semantic_match_miss++;
1403#endif
1404	}
1405
1406	return 1;
1407}
1408
1409int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1410		     struct fib_result *res, int fib_flags)
1411{
1412	struct trie *t = (struct trie *) tb->tb_data;
1413	int ret;
1414	struct rt_trie_node *n;
1415	struct tnode *pn;
1416	unsigned int pos, bits;
1417	t_key key = ntohl(flp->daddr);
1418	unsigned int chopped_off;
1419	t_key cindex = 0;
1420	unsigned int current_prefix_length = KEYLENGTH;
1421	struct tnode *cn;
1422	t_key pref_mismatch;
1423
1424	rcu_read_lock();
1425
1426	n = rcu_dereference(t->trie);
1427	if (!n)
1428		goto failed;
1429
1430#ifdef CONFIG_IP_FIB_TRIE_STATS
1431	t->stats.gets++;
1432#endif
1433
1434	/* Just a leaf? */
1435	if (IS_LEAF(n)) {
1436		ret = check_leaf(tb, t, (struct leaf *)n, key, flp, res, fib_flags);
1437		goto found;
1438	}
1439
1440	pn = (struct tnode *) n;
1441	chopped_off = 0;
1442
1443	while (pn) {
1444		pos = pn->pos;
1445		bits = pn->bits;
1446
1447		if (!chopped_off)
1448			cindex = tkey_extract_bits(mask_pfx(key, current_prefix_length),
1449						   pos, bits);
1450
1451		n = tnode_get_child_rcu(pn, cindex);
1452
1453		if (n == NULL) {
1454#ifdef CONFIG_IP_FIB_TRIE_STATS
1455			t->stats.null_node_hit++;
1456#endif
1457			goto backtrace;
1458		}
1459
1460		if (IS_LEAF(n)) {
1461			ret = check_leaf(tb, t, (struct leaf *)n, key, flp, res, fib_flags);
1462			if (ret > 0)
1463				goto backtrace;
1464			goto found;
1465		}
1466
1467		cn = (struct tnode *)n;
1468
1469		/*
1470		 * It's a tnode, and we can do some extra checks here if we
1471		 * like, to avoid descending into a dead-end branch.
1472		 * This tnode is in the parent's child array at index
1473		 * key[p_pos..p_pos+p_bits] but potentially with some bits
1474		 * chopped off, so in reality the index may be just a
1475		 * subprefix, padded with zero at the end.
1476		 * We can also take a look at any skipped bits in this
1477		 * tnode - everything up to p_pos is supposed to be ok,
1478		 * and the non-chopped bits of the index (se previous
1479		 * paragraph) are also guaranteed ok, but the rest is
1480		 * considered unknown.
1481		 *
1482		 * The skipped bits are key[pos+bits..cn->pos].
1483		 */
1484
1485		/* If current_prefix_length < pos+bits, we are already doing
1486		 * actual prefix  matching, which means everything from
1487		 * pos+(bits-chopped_off) onward must be zero along some
1488		 * branch of this subtree - otherwise there is *no* valid
1489		 * prefix present. Here we can only check the skipped
1490		 * bits. Remember, since we have already indexed into the
1491		 * parent's child array, we know that the bits we chopped of
1492		 * *are* zero.
1493		 */
1494
1495		/* NOTA BENE: Checking only skipped bits
1496		   for the new node here */
1497
1498		if (current_prefix_length < pos+bits) {
1499			if (tkey_extract_bits(cn->key, current_prefix_length,
1500						cn->pos - current_prefix_length)
1501			    || !(cn->child[0]))
1502				goto backtrace;
1503		}
1504
1505		/*
1506		 * If chopped_off=0, the index is fully validated and we
1507		 * only need to look at the skipped bits for this, the new,
1508		 * tnode. What we actually want to do is to find out if
1509		 * these skipped bits match our key perfectly, or if we will
1510		 * have to count on finding a matching prefix further down,
1511		 * because if we do, we would like to have some way of
1512		 * verifying the existence of such a prefix at this point.
1513		 */
1514
1515		/* The only thing we can do at this point is to verify that
1516		 * any such matching prefix can indeed be a prefix to our
1517		 * key, and if the bits in the node we are inspecting that
1518		 * do not match our key are not ZERO, this cannot be true.
1519		 * Thus, find out where there is a mismatch (before cn->pos)
1520		 * and verify that all the mismatching bits are zero in the
1521		 * new tnode's key.
1522		 */
1523
1524		/*
1525		 * Note: We aren't very concerned about the piece of
1526		 * the key that precede pn->pos+pn->bits, since these
1527		 * have already been checked. The bits after cn->pos
1528		 * aren't checked since these are by definition
1529		 * "unknown" at this point. Thus, what we want to see
1530		 * is if we are about to enter the "prefix matching"
1531		 * state, and in that case verify that the skipped
1532		 * bits that will prevail throughout this subtree are
1533		 * zero, as they have to be if we are to find a
1534		 * matching prefix.
1535		 */
1536
1537		pref_mismatch = mask_pfx(cn->key ^ key, cn->pos);
1538
1539		/*
1540		 * In short: If skipped bits in this node do not match
1541		 * the search key, enter the "prefix matching"
1542		 * state.directly.
1543		 */
1544		if (pref_mismatch) {
1545			int mp = KEYLENGTH - fls(pref_mismatch);
1546
1547			if (tkey_extract_bits(cn->key, mp, cn->pos - mp) != 0)
1548				goto backtrace;
1549
1550			if (current_prefix_length >= cn->pos)
1551				current_prefix_length = mp;
1552		}
1553
1554		pn = (struct tnode *)n; /* Descend */
1555		chopped_off = 0;
1556		continue;
1557
1558backtrace:
1559		chopped_off++;
1560
1561		/* As zero don't change the child key (cindex) */
1562		while ((chopped_off <= pn->bits)
1563		       && !(cindex & (1<<(chopped_off-1))))
1564			chopped_off++;
1565
1566		/* Decrease current_... with bits chopped off */
1567		if (current_prefix_length > pn->pos + pn->bits - chopped_off)
1568			current_prefix_length = pn->pos + pn->bits
1569				- chopped_off;
1570
1571		/*
1572		 * Either we do the actual chop off according or if we have
1573		 * chopped off all bits in this tnode walk up to our parent.
1574		 */
1575
1576		if (chopped_off <= pn->bits) {
1577			cindex &= ~(1 << (chopped_off-1));
1578		} else {
1579			struct tnode *parent = node_parent_rcu((struct rt_trie_node *) pn);
1580			if (!parent)
1581				goto failed;
1582
1583			/* Get Child's index */
1584			cindex = tkey_extract_bits(pn->key, parent->pos, parent->bits);
1585			pn = parent;
1586			chopped_off = 0;
1587
1588#ifdef CONFIG_IP_FIB_TRIE_STATS
1589			t->stats.backtrack++;
1590#endif
1591			goto backtrace;
1592		}
1593	}
1594failed:
1595	ret = 1;
1596found:
1597	rcu_read_unlock();
1598	return ret;
1599}
1600
1601/*
1602 * Remove the leaf and return parent.
1603 */
1604static void trie_leaf_remove(struct trie *t, struct leaf *l)
1605{
1606	struct tnode *tp = node_parent((struct rt_trie_node *) l);
1607
1608	pr_debug("entering trie_leaf_remove(%p)\n", l);
1609
1610	if (tp) {
1611		t_key cindex = tkey_extract_bits(l->key, tp->pos, tp->bits);
1612		put_child(t, (struct tnode *)tp, cindex, NULL);
1613		trie_rebalance(t, tp);
1614	} else
1615		rcu_assign_pointer(t->trie, NULL);
1616
1617	free_leaf(l);
1618}
1619
1620/*
1621 * Caller must hold RTNL.
1622 */
1623int fib_table_delete(struct fib_table *tb, struct fib_config *cfg)
1624{
1625	struct trie *t = (struct trie *) tb->tb_data;
1626	u32 key, mask;
1627	int plen = cfg->fc_dst_len;
1628	u8 tos = cfg->fc_tos;
1629	struct fib_alias *fa, *fa_to_delete;
1630	struct list_head *fa_head;
1631	struct leaf *l;
1632	struct leaf_info *li;
1633
1634	if (plen > 32)
1635		return -EINVAL;
1636
1637	key = ntohl(cfg->fc_dst);
1638	mask = ntohl(inet_make_mask(plen));
1639
1640	if (key & ~mask)
1641		return -EINVAL;
1642
1643	key = key & mask;
1644	l = fib_find_node(t, key);
1645
1646	if (!l)
1647		return -ESRCH;
1648
1649	fa_head = get_fa_head(l, plen);
1650	fa = fib_find_alias(fa_head, tos, 0);
1651
1652	if (!fa)
1653		return -ESRCH;
1654
1655	pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1656
1657	fa_to_delete = NULL;
1658	fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
1659	list_for_each_entry_continue(fa, fa_head, fa_list) {
1660		struct fib_info *fi = fa->fa_info;
1661
1662		if (fa->fa_tos != tos)
1663			break;
1664
1665		if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1666		    (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1667		     fa->fa_scope == cfg->fc_scope) &&
1668		    (!cfg->fc_protocol ||
1669		     fi->fib_protocol == cfg->fc_protocol) &&
1670		    fib_nh_match(cfg, fi) == 0) {
1671			fa_to_delete = fa;
1672			break;
1673		}
1674	}
1675
1676	if (!fa_to_delete)
1677		return -ESRCH;
1678
1679	fa = fa_to_delete;
1680	rtmsg_fib(RTM_DELROUTE, htonl(key), fa, plen, tb->tb_id,
1681		  &cfg->fc_nlinfo, 0);
1682
1683	l = fib_find_node(t, key);
1684	li = find_leaf_info(l, plen);
1685
1686	list_del_rcu(&fa->fa_list);
1687
1688	if (list_empty(fa_head)) {
1689		hlist_del_rcu(&li->hlist);
1690		free_leaf_info(li);
1691	}
1692
1693	if (hlist_empty(&l->list))
1694		trie_leaf_remove(t, l);
1695
1696	if (fa->fa_state & FA_S_ACCESSED)
1697		rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1698
1699	fib_release_info(fa->fa_info);
1700	alias_free_mem_rcu(fa);
1701	return 0;
1702}
1703
1704static int trie_flush_list(struct list_head *head)
1705{
1706	struct fib_alias *fa, *fa_node;
1707	int found = 0;
1708
1709	list_for_each_entry_safe(fa, fa_node, head, fa_list) {
1710		struct fib_info *fi = fa->fa_info;
1711
1712		if (fi && (fi->fib_flags & RTNH_F_DEAD)) {
1713			list_del_rcu(&fa->fa_list);
1714			fib_release_info(fa->fa_info);
1715			alias_free_mem_rcu(fa);
1716			found++;
1717		}
1718	}
1719	return found;
1720}
1721
1722static int trie_flush_leaf(struct leaf *l)
1723{
1724	int found = 0;
1725	struct hlist_head *lih = &l->list;
1726	struct hlist_node *node, *tmp;
1727	struct leaf_info *li = NULL;
1728
1729	hlist_for_each_entry_safe(li, node, tmp, lih, hlist) {
1730		found += trie_flush_list(&li->falh);
1731
1732		if (list_empty(&li->falh)) {
1733			hlist_del_rcu(&li->hlist);
1734			free_leaf_info(li);
1735		}
1736	}
1737	return found;
1738}
1739
1740/*
1741 * Scan for the next right leaf starting at node p->child[idx]
1742 * Since we have back pointer, no recursion necessary.
1743 */
1744static struct leaf *leaf_walk_rcu(struct tnode *p, struct rt_trie_node *c)
1745{
1746	do {
1747		t_key idx;
1748
1749		if (c)
1750			idx = tkey_extract_bits(c->key, p->pos, p->bits) + 1;
1751		else
1752			idx = 0;
1753
1754		while (idx < 1u << p->bits) {
1755			c = tnode_get_child_rcu(p, idx++);
1756			if (!c)
1757				continue;
1758
1759			if (IS_LEAF(c)) {
1760				prefetch(p->child[idx]);
1761				return (struct leaf *) c;
1762			}
1763
1764			/* Rescan start scanning in new node */
1765			p = (struct tnode *) c;
1766			idx = 0;
1767		}
1768
1769		/* Node empty, walk back up to parent */
1770		c = (struct rt_trie_node *) p;
1771	} while ((p = node_parent_rcu(c)) != NULL);
1772
1773	return NULL; /* Root of trie */
1774}
1775
1776static struct leaf *trie_firstleaf(struct trie *t)
1777{
1778	struct tnode *n = (struct tnode *)rcu_dereference_rtnl(t->trie);
1779
1780	if (!n)
1781		return NULL;
1782
1783	if (IS_LEAF(n))          /* trie is just a leaf */
1784		return (struct leaf *) n;
1785
1786	return leaf_walk_rcu(n, NULL);
1787}
1788
1789static struct leaf *trie_nextleaf(struct leaf *l)
1790{
1791	struct rt_trie_node *c = (struct rt_trie_node *) l;
1792	struct tnode *p = node_parent_rcu(c);
1793
1794	if (!p)
1795		return NULL;	/* trie with just one leaf */
1796
1797	return leaf_walk_rcu(p, c);
1798}
1799
1800static struct leaf *trie_leafindex(struct trie *t, int index)
1801{
1802	struct leaf *l = trie_firstleaf(t);
1803
1804	while (l && index-- > 0)
1805		l = trie_nextleaf(l);
1806
1807	return l;
1808}
1809
1810
1811/*
1812 * Caller must hold RTNL.
1813 */
1814int fib_table_flush(struct fib_table *tb)
1815{
1816	struct trie *t = (struct trie *) tb->tb_data;
1817	struct leaf *l, *ll = NULL;
1818	int found = 0;
1819
1820	for (l = trie_firstleaf(t); l; l = trie_nextleaf(l)) {
1821		found += trie_flush_leaf(l);
1822
1823		if (ll && hlist_empty(&ll->list))
1824			trie_leaf_remove(t, ll);
1825		ll = l;
1826	}
1827
1828	if (ll && hlist_empty(&ll->list))
1829		trie_leaf_remove(t, ll);
1830
1831	pr_debug("trie_flush found=%d\n", found);
1832	return found;
1833}
1834
1835void fib_free_table(struct fib_table *tb)
1836{
1837	kfree(tb);
1838}
1839
1840static int fn_trie_dump_fa(t_key key, int plen, struct list_head *fah,
1841			   struct fib_table *tb,
1842			   struct sk_buff *skb, struct netlink_callback *cb)
1843{
1844	int i, s_i;
1845	struct fib_alias *fa;
1846	__be32 xkey = htonl(key);
1847
1848	s_i = cb->args[5];
1849	i = 0;
1850
1851	/* rcu_read_lock is hold by caller */
1852
1853	list_for_each_entry_rcu(fa, fah, fa_list) {
1854		if (i < s_i) {
1855			i++;
1856			continue;
1857		}
1858
1859		if (fib_dump_info(skb, NETLINK_CB(cb->skb).pid,
1860				  cb->nlh->nlmsg_seq,
1861				  RTM_NEWROUTE,
1862				  tb->tb_id,
1863				  fa->fa_type,
1864				  fa->fa_scope,
1865				  xkey,
1866				  plen,
1867				  fa->fa_tos,
1868				  fa->fa_info, NLM_F_MULTI) < 0) {
1869			cb->args[5] = i;
1870			return -1;
1871		}
1872		i++;
1873	}
1874	cb->args[5] = i;
1875	return skb->len;
1876}
1877
1878static int fn_trie_dump_leaf(struct leaf *l, struct fib_table *tb,
1879			struct sk_buff *skb, struct netlink_callback *cb)
1880{
1881	struct leaf_info *li;
1882	struct hlist_node *node;
1883	int i, s_i;
1884
1885	s_i = cb->args[4];
1886	i = 0;
1887
1888	/* rcu_read_lock is hold by caller */
1889	hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
1890		if (i < s_i) {
1891			i++;
1892			continue;
1893		}
1894
1895		if (i > s_i)
1896			cb->args[5] = 0;
1897
1898		if (list_empty(&li->falh))
1899			continue;
1900
1901		if (fn_trie_dump_fa(l->key, li->plen, &li->falh, tb, skb, cb) < 0) {
1902			cb->args[4] = i;
1903			return -1;
1904		}
1905		i++;
1906	}
1907
1908	cb->args[4] = i;
1909	return skb->len;
1910}
1911
1912int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
1913		   struct netlink_callback *cb)
1914{
1915	struct leaf *l;
1916	struct trie *t = (struct trie *) tb->tb_data;
1917	t_key key = cb->args[2];
1918	int count = cb->args[3];
1919
1920	rcu_read_lock();
1921	/* Dump starting at last key.
1922	 * Note: 0.0.0.0/0 (ie default) is first key.
1923	 */
1924	if (count == 0)
1925		l = trie_firstleaf(t);
1926	else {
1927		/* Normally, continue from last key, but if that is missing
1928		 * fallback to using slow rescan
1929		 */
1930		l = fib_find_node(t, key);
1931		if (!l)
1932			l = trie_leafindex(t, count);
1933	}
1934
1935	while (l) {
1936		cb->args[2] = l->key;
1937		if (fn_trie_dump_leaf(l, tb, skb, cb) < 0) {
1938			cb->args[3] = count;
1939			rcu_read_unlock();
1940			return -1;
1941		}
1942
1943		++count;
1944		l = trie_nextleaf(l);
1945		memset(&cb->args[4], 0,
1946		       sizeof(cb->args) - 4*sizeof(cb->args[0]));
1947	}
1948	cb->args[3] = count;
1949	rcu_read_unlock();
1950
1951	return skb->len;
1952}
1953
1954void __init fib_trie_init(void)
1955{
1956	fn_alias_kmem = kmem_cache_create("ip_fib_alias",
1957					  sizeof(struct fib_alias),
1958					  0, SLAB_PANIC, NULL);
1959
1960	trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
1961					   max(sizeof(struct leaf),
1962					       sizeof(struct leaf_info)),
1963					   0, SLAB_PANIC, NULL);
1964}
1965
1966
1967struct fib_table *fib_trie_table(u32 id)
1968{
1969	struct fib_table *tb;
1970	struct trie *t;
1971
1972	tb = kmalloc(sizeof(struct fib_table) + sizeof(struct trie),
1973		     GFP_KERNEL);
1974	if (tb == NULL)
1975		return NULL;
1976
1977	tb->tb_id = id;
1978	tb->tb_default = -1;
1979
1980	t = (struct trie *) tb->tb_data;
1981	memset(t, 0, sizeof(*t));
1982
1983	if (id == RT_TABLE_LOCAL)
1984		pr_info("IPv4 FIB: Using LC-trie version %s\n", VERSION);
1985
1986	return tb;
1987}
1988
1989#ifdef CONFIG_PROC_FS
1990/* Depth first Trie walk iterator */
1991struct fib_trie_iter {
1992	struct seq_net_private p;
1993	struct fib_table *tb;
1994	struct tnode *tnode;
1995	unsigned int index;
1996	unsigned int depth;
1997};
1998
1999static struct rt_trie_node *fib_trie_get_next(struct fib_trie_iter *iter)
2000{
2001	struct tnode *tn = iter->tnode;
2002	unsigned int cindex = iter->index;
2003	struct tnode *p;
2004
2005	/* A single entry routing table */
2006	if (!tn)
2007		return NULL;
2008
2009	pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2010		 iter->tnode, iter->index, iter->depth);
2011rescan:
2012	while (cindex < (1<<tn->bits)) {
2013		struct rt_trie_node *n = tnode_get_child_rcu(tn, cindex);
2014
2015		if (n) {
2016			if (IS_LEAF(n)) {
2017				iter->tnode = tn;
2018				iter->index = cindex + 1;
2019			} else {
2020				/* push down one level */
2021				iter->tnode = (struct tnode *) n;
2022				iter->index = 0;
2023				++iter->depth;
2024			}
2025			return n;
2026		}
2027
2028		++cindex;
2029	}
2030
2031	/* Current node exhausted, pop back up */
2032	p = node_parent_rcu((struct rt_trie_node *)tn);
2033	if (p) {
2034		cindex = tkey_extract_bits(tn->key, p->pos, p->bits)+1;
2035		tn = p;
2036		--iter->depth;
2037		goto rescan;
2038	}
2039
2040	/* got root? */
2041	return NULL;
2042}
2043
2044static struct rt_trie_node *fib_trie_get_first(struct fib_trie_iter *iter,
2045				       struct trie *t)
2046{
2047	struct rt_trie_node *n;
2048
2049	if (!t)
2050		return NULL;
2051
2052	n = rcu_dereference(t->trie);
2053	if (!n)
2054		return NULL;
2055
2056	if (IS_TNODE(n)) {
2057		iter->tnode = (struct tnode *) n;
2058		iter->index = 0;
2059		iter->depth = 1;
2060	} else {
2061		iter->tnode = NULL;
2062		iter->index = 0;
2063		iter->depth = 0;
2064	}
2065
2066	return n;
2067}
2068
2069static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2070{
2071	struct rt_trie_node *n;
2072	struct fib_trie_iter iter;
2073
2074	memset(s, 0, sizeof(*s));
2075
2076	rcu_read_lock();
2077	for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2078		if (IS_LEAF(n)) {
2079			struct leaf *l = (struct leaf *)n;
2080			struct leaf_info *li;
2081			struct hlist_node *tmp;
2082
2083			s->leaves++;
2084			s->totdepth += iter.depth;
2085			if (iter.depth > s->maxdepth)
2086				s->maxdepth = iter.depth;
2087
2088			hlist_for_each_entry_rcu(li, tmp, &l->list, hlist)
2089				++s->prefixes;
2090		} else {
2091			const struct tnode *tn = (const struct tnode *) n;
2092			int i;
2093
2094			s->tnodes++;
2095			if (tn->bits < MAX_STAT_DEPTH)
2096				s->nodesizes[tn->bits]++;
2097
2098			for (i = 0; i < (1<<tn->bits); i++)
2099				if (!tn->child[i])
2100					s->nullpointers++;
2101		}
2102	}
2103	rcu_read_unlock();
2104}
2105
2106/*
2107 *	This outputs /proc/net/fib_triestats
2108 */
2109static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2110{
2111	unsigned int i, max, pointers, bytes, avdepth;
2112
2113	if (stat->leaves)
2114		avdepth = stat->totdepth*100 / stat->leaves;
2115	else
2116		avdepth = 0;
2117
2118	seq_printf(seq, "\tAver depth:     %u.%02d\n",
2119		   avdepth / 100, avdepth % 100);
2120	seq_printf(seq, "\tMax depth:      %u\n", stat->maxdepth);
2121
2122	seq_printf(seq, "\tLeaves:         %u\n", stat->leaves);
2123	bytes = sizeof(struct leaf) * stat->leaves;
2124
2125	seq_printf(seq, "\tPrefixes:       %u\n", stat->prefixes);
2126	bytes += sizeof(struct leaf_info) * stat->prefixes;
2127
2128	seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2129	bytes += sizeof(struct tnode) * stat->tnodes;
2130
2131	max = MAX_STAT_DEPTH;
2132	while (max > 0 && stat->nodesizes[max-1] == 0)
2133		max--;
2134
2135	pointers = 0;
2136	for (i = 1; i <= max; i++)
2137		if (stat->nodesizes[i] != 0) {
2138			seq_printf(seq, "  %u: %u",  i, stat->nodesizes[i]);
2139			pointers += (1<<i) * stat->nodesizes[i];
2140		}
2141	seq_putc(seq, '\n');
2142	seq_printf(seq, "\tPointers: %u\n", pointers);
2143
2144	bytes += sizeof(struct rt_trie_node *) * pointers;
2145	seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2146	seq_printf(seq, "Total size: %u  kB\n", (bytes + 1023) / 1024);
2147}
2148
2149#ifdef CONFIG_IP_FIB_TRIE_STATS
2150static void trie_show_usage(struct seq_file *seq,
2151			    const struct trie_use_stats *stats)
2152{
2153	seq_printf(seq, "\nCounters:\n---------\n");
2154	seq_printf(seq, "gets = %u\n", stats->gets);
2155	seq_printf(seq, "backtracks = %u\n", stats->backtrack);
2156	seq_printf(seq, "semantic match passed = %u\n",
2157		   stats->semantic_match_passed);
2158	seq_printf(seq, "semantic match miss = %u\n",
2159		   stats->semantic_match_miss);
2160	seq_printf(seq, "null node hit= %u\n", stats->null_node_hit);
2161	seq_printf(seq, "skipped node resize = %u\n\n",
2162		   stats->resize_node_skipped);
2163}
2164#endif /*  CONFIG_IP_FIB_TRIE_STATS */
2165
2166static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2167{
2168	if (tb->tb_id == RT_TABLE_LOCAL)
2169		seq_puts(seq, "Local:\n");
2170	else if (tb->tb_id == RT_TABLE_MAIN)
2171		seq_puts(seq, "Main:\n");
2172	else
2173		seq_printf(seq, "Id %d:\n", tb->tb_id);
2174}
2175
2176
2177static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2178{
2179	struct net *net = (struct net *)seq->private;
2180	unsigned int h;
2181
2182	seq_printf(seq,
2183		   "Basic info: size of leaf:"
2184		   " %Zd bytes, size of tnode: %Zd bytes.\n",
2185		   sizeof(struct leaf), sizeof(struct tnode));
2186
2187	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2188		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2189		struct hlist_node *node;
2190		struct fib_table *tb;
2191
2192		hlist_for_each_entry_rcu(tb, node, head, tb_hlist) {
2193			struct trie *t = (struct trie *) tb->tb_data;
2194			struct trie_stat stat;
2195
2196			if (!t)
2197				continue;
2198
2199			fib_table_print(seq, tb);
2200
2201			trie_collect_stats(t, &stat);
2202			trie_show_stats(seq, &stat);
2203#ifdef CONFIG_IP_FIB_TRIE_STATS
2204			trie_show_usage(seq, &t->stats);
2205#endif
2206		}
2207	}
2208
2209	return 0;
2210}
2211
2212static int fib_triestat_seq_open(struct inode *inode, struct file *file)
2213{
2214	return single_open_net(inode, file, fib_triestat_seq_show);
2215}
2216
2217static const struct file_operations fib_triestat_fops = {
2218	.owner	= THIS_MODULE,
2219	.open	= fib_triestat_seq_open,
2220	.read	= seq_read,
2221	.llseek	= seq_lseek,
2222	.release = single_release_net,
2223};
2224
2225static struct rt_trie_node *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2226{
2227	struct fib_trie_iter *iter = seq->private;
2228	struct net *net = seq_file_net(seq);
2229	loff_t idx = 0;
2230	unsigned int h;
2231
2232	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2233		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2234		struct hlist_node *node;
2235		struct fib_table *tb;
2236
2237		hlist_for_each_entry_rcu(tb, node, head, tb_hlist) {
2238			struct rt_trie_node *n;
2239
2240			for (n = fib_trie_get_first(iter,
2241						    (struct trie *) tb->tb_data);
2242			     n; n = fib_trie_get_next(iter))
2243				if (pos == idx++) {
2244					iter->tb = tb;
2245					return n;
2246				}
2247		}
2248	}
2249
2250	return NULL;
2251}
2252
2253static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2254	__acquires(RCU)
2255{
2256	rcu_read_lock();
2257	return fib_trie_get_idx(seq, *pos);
2258}
2259
2260static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2261{
2262	struct fib_trie_iter *iter = seq->private;
2263	struct net *net = seq_file_net(seq);
2264	struct fib_table *tb = iter->tb;
2265	struct hlist_node *tb_node;
2266	unsigned int h;
2267	struct rt_trie_node *n;
2268
2269	++*pos;
2270	/* next node in same table */
2271	n = fib_trie_get_next(iter);
2272	if (n)
2273		return n;
2274
2275	/* walk rest of this hash chain */
2276	h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2277	while ( (tb_node = rcu_dereference(tb->tb_hlist.next)) ) {
2278		tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2279		n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2280		if (n)
2281			goto found;
2282	}
2283
2284	/* new hash chain */
2285	while (++h < FIB_TABLE_HASHSZ) {
2286		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2287		hlist_for_each_entry_rcu(tb, tb_node, head, tb_hlist) {
2288			n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2289			if (n)
2290				goto found;
2291		}
2292	}
2293	return NULL;
2294
2295found:
2296	iter->tb = tb;
2297	return n;
2298}
2299
2300static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2301	__releases(RCU)
2302{
2303	rcu_read_unlock();
2304}
2305
2306static void seq_indent(struct seq_file *seq, int n)
2307{
2308	while (n-- > 0)
2309		seq_puts(seq, "   ");
2310}
2311
2312static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2313{
2314	switch (s) {
2315	case RT_SCOPE_UNIVERSE: return "universe";
2316	case RT_SCOPE_SITE:	return "site";
2317	case RT_SCOPE_LINK:	return "link";
2318	case RT_SCOPE_HOST:	return "host";
2319	case RT_SCOPE_NOWHERE:	return "nowhere";
2320	default:
2321		snprintf(buf, len, "scope=%d", s);
2322		return buf;
2323	}
2324}
2325
2326static const char *const rtn_type_names[__RTN_MAX] = {
2327	[RTN_UNSPEC] = "UNSPEC",
2328	[RTN_UNICAST] = "UNICAST",
2329	[RTN_LOCAL] = "LOCAL",
2330	[RTN_BROADCAST] = "BROADCAST",
2331	[RTN_ANYCAST] = "ANYCAST",
2332	[RTN_MULTICAST] = "MULTICAST",
2333	[RTN_BLACKHOLE] = "BLACKHOLE",
2334	[RTN_UNREACHABLE] = "UNREACHABLE",
2335	[RTN_PROHIBIT] = "PROHIBIT",
2336	[RTN_THROW] = "THROW",
2337	[RTN_NAT] = "NAT",
2338	[RTN_XRESOLVE] = "XRESOLVE",
2339};
2340
2341static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2342{
2343	if (t < __RTN_MAX && rtn_type_names[t])
2344		return rtn_type_names[t];
2345	snprintf(buf, len, "type %u", t);
2346	return buf;
2347}
2348
2349/* Pretty print the trie */
2350static int fib_trie_seq_show(struct seq_file *seq, void *v)
2351{
2352	const struct fib_trie_iter *iter = seq->private;
2353	struct rt_trie_node *n = v;
2354
2355	if (!node_parent_rcu(n))
2356		fib_table_print(seq, iter->tb);
2357
2358	if (IS_TNODE(n)) {
2359		struct tnode *tn = (struct tnode *) n;
2360		__be32 prf = htonl(mask_pfx(tn->key, tn->pos));
2361
2362		seq_indent(seq, iter->depth-1);
2363		seq_printf(seq, "  +-- %pI4/%d %d %d %d\n",
2364			   &prf, tn->pos, tn->bits, tn->full_children,
2365			   tn->empty_children);
2366
2367	} else {
2368		struct leaf *l = (struct leaf *) n;
2369		struct leaf_info *li;
2370		struct hlist_node *node;
2371		__be32 val = htonl(l->key);
2372
2373		seq_indent(seq, iter->depth);
2374		seq_printf(seq, "  |-- %pI4\n", &val);
2375
2376		hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
2377			struct fib_alias *fa;
2378
2379			list_for_each_entry_rcu(fa, &li->falh, fa_list) {
2380				char buf1[32], buf2[32];
2381
2382				seq_indent(seq, iter->depth+1);
2383				seq_printf(seq, "  /%d %s %s", li->plen,
2384					   rtn_scope(buf1, sizeof(buf1),
2385						     fa->fa_scope),
2386					   rtn_type(buf2, sizeof(buf2),
2387						    fa->fa_type));
2388				if (fa->fa_tos)
2389					seq_printf(seq, " tos=%d", fa->fa_tos);
2390				seq_putc(seq, '\n');
2391			}
2392		}
2393	}
2394
2395	return 0;
2396}
2397
2398static const struct seq_operations fib_trie_seq_ops = {
2399	.start  = fib_trie_seq_start,
2400	.next   = fib_trie_seq_next,
2401	.stop   = fib_trie_seq_stop,
2402	.show   = fib_trie_seq_show,
2403};
2404
2405static int fib_trie_seq_open(struct inode *inode, struct file *file)
2406{
2407	return seq_open_net(inode, file, &fib_trie_seq_ops,
2408			    sizeof(struct fib_trie_iter));
2409}
2410
2411static const struct file_operations fib_trie_fops = {
2412	.owner  = THIS_MODULE,
2413	.open   = fib_trie_seq_open,
2414	.read   = seq_read,
2415	.llseek = seq_lseek,
2416	.release = seq_release_net,
2417};
2418
2419struct fib_route_iter {
2420	struct seq_net_private p;
2421	struct trie *main_trie;
2422	loff_t	pos;
2423	t_key	key;
2424};
2425
2426static struct leaf *fib_route_get_idx(struct fib_route_iter *iter, loff_t pos)
2427{
2428	struct leaf *l = NULL;
2429	struct trie *t = iter->main_trie;
2430
2431	/* use cache location of last found key */
2432	if (iter->pos > 0 && pos >= iter->pos && (l = fib_find_node(t, iter->key)))
2433		pos -= iter->pos;
2434	else {
2435		iter->pos = 0;
2436		l = trie_firstleaf(t);
2437	}
2438
2439	while (l && pos-- > 0) {
2440		iter->pos++;
2441		l = trie_nextleaf(l);
2442	}
2443
2444	if (l)
2445		iter->key = pos;	/* remember it */
2446	else
2447		iter->pos = 0;		/* forget it */
2448
2449	return l;
2450}
2451
2452static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2453	__acquires(RCU)
2454{
2455	struct fib_route_iter *iter = seq->private;
2456	struct fib_table *tb;
2457
2458	rcu_read_lock();
2459	tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2460	if (!tb)
2461		return NULL;
2462
2463	iter->main_trie = (struct trie *) tb->tb_data;
2464	if (*pos == 0)
2465		return SEQ_START_TOKEN;
2466	else
2467		return fib_route_get_idx(iter, *pos - 1);
2468}
2469
2470static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2471{
2472	struct fib_route_iter *iter = seq->private;
2473	struct leaf *l = v;
2474
2475	++*pos;
2476	if (v == SEQ_START_TOKEN) {
2477		iter->pos = 0;
2478		l = trie_firstleaf(iter->main_trie);
2479	} else {
2480		iter->pos++;
2481		l = trie_nextleaf(l);
2482	}
2483
2484	if (l)
2485		iter->key = l->key;
2486	else
2487		iter->pos = 0;
2488	return l;
2489}
2490
2491static void fib_route_seq_stop(struct seq_file *seq, void *v)
2492	__releases(RCU)
2493{
2494	rcu_read_unlock();
2495}
2496
2497static unsigned int fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2498{
2499	unsigned int flags = 0;
2500
2501	if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2502		flags = RTF_REJECT;
2503	if (fi && fi->fib_nh->nh_gw)
2504		flags |= RTF_GATEWAY;
2505	if (mask == htonl(0xFFFFFFFF))
2506		flags |= RTF_HOST;
2507	flags |= RTF_UP;
2508	return flags;
2509}
2510
2511/*
2512 *	This outputs /proc/net/route.
2513 *	The format of the file is not supposed to be changed
2514 *	and needs to be same as fib_hash output to avoid breaking
2515 *	legacy utilities
2516 */
2517static int fib_route_seq_show(struct seq_file *seq, void *v)
2518{
2519	struct leaf *l = v;
2520	struct leaf_info *li;
2521	struct hlist_node *node;
2522
2523	if (v == SEQ_START_TOKEN) {
2524		seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2525			   "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2526			   "\tWindow\tIRTT");
2527		return 0;
2528	}
2529
2530	hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
2531		struct fib_alias *fa;
2532		__be32 mask, prefix;
2533
2534		mask = inet_make_mask(li->plen);
2535		prefix = htonl(l->key);
2536
2537		list_for_each_entry_rcu(fa, &li->falh, fa_list) {
2538			const struct fib_info *fi = fa->fa_info;
2539			unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2540			int len;
2541
2542			if (fa->fa_type == RTN_BROADCAST
2543			    || fa->fa_type == RTN_MULTICAST)
2544				continue;
2545
2546			if (fi)
2547				seq_printf(seq,
2548					 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2549					 "%d\t%08X\t%d\t%u\t%u%n",
2550					 fi->fib_dev ? fi->fib_dev->name : "*",
2551					 prefix,
2552					 fi->fib_nh->nh_gw, flags, 0, 0,
2553					 fi->fib_priority,
2554					 mask,
2555					 (fi->fib_advmss ?
2556					  fi->fib_advmss + 40 : 0),
2557					 fi->fib_window,
2558					 fi->fib_rtt >> 3, &len);
2559			else
2560				seq_printf(seq,
2561					 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2562					 "%d\t%08X\t%d\t%u\t%u%n",
2563					 prefix, 0, flags, 0, 0, 0,
2564					 mask, 0, 0, 0, &len);
2565
2566			seq_printf(seq, "%*s\n", 127 - len, "");
2567		}
2568	}
2569
2570	return 0;
2571}
2572
2573static const struct seq_operations fib_route_seq_ops = {
2574	.start  = fib_route_seq_start,
2575	.next   = fib_route_seq_next,
2576	.stop   = fib_route_seq_stop,
2577	.show   = fib_route_seq_show,
2578};
2579
2580static int fib_route_seq_open(struct inode *inode, struct file *file)
2581{
2582	return seq_open_net(inode, file, &fib_route_seq_ops,
2583			    sizeof(struct fib_route_iter));
2584}
2585
2586static const struct file_operations fib_route_fops = {
2587	.owner  = THIS_MODULE,
2588	.open   = fib_route_seq_open,
2589	.read   = seq_read,
2590	.llseek = seq_lseek,
2591	.release = seq_release_net,
2592};
2593
2594int __net_init fib_proc_init(struct net *net)
2595{
2596	if (!proc_net_fops_create(net, "fib_trie", S_IRUGO, &fib_trie_fops))
2597		goto out1;
2598
2599	if (!proc_net_fops_create(net, "fib_triestat", S_IRUGO,
2600				  &fib_triestat_fops))
2601		goto out2;
2602
2603	if (!proc_net_fops_create(net, "route", S_IRUGO, &fib_route_fops))
2604		goto out3;
2605
2606	return 0;
2607
2608out3:
2609	proc_net_remove(net, "fib_triestat");
2610out2:
2611	proc_net_remove(net, "fib_trie");
2612out1:
2613	return -ENOMEM;
2614}
2615
2616void __net_exit fib_proc_exit(struct net *net)
2617{
2618	proc_net_remove(net, "fib_trie");
2619	proc_net_remove(net, "fib_triestat");
2620	proc_net_remove(net, "route");
2621}
2622
2623#endif /* CONFIG_PROC_FS */
2624