xref: /openbmc/linux/net/ipv4/fib_trie.c (revision 0d456bad)
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 described 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 <linux/bitops.h>
55 #include <linux/types.h>
56 #include <linux/kernel.h>
57 #include <linux/mm.h>
58 #include <linux/string.h>
59 #include <linux/socket.h>
60 #include <linux/sockios.h>
61 #include <linux/errno.h>
62 #include <linux/in.h>
63 #include <linux/inet.h>
64 #include <linux/inetdevice.h>
65 #include <linux/netdevice.h>
66 #include <linux/if_arp.h>
67 #include <linux/proc_fs.h>
68 #include <linux/rcupdate.h>
69 #include <linux/skbuff.h>
70 #include <linux/netlink.h>
71 #include <linux/init.h>
72 #include <linux/list.h>
73 #include <linux/slab.h>
74 #include <linux/prefetch.h>
75 #include <linux/export.h>
76 #include <net/net_namespace.h>
77 #include <net/ip.h>
78 #include <net/protocol.h>
79 #include <net/route.h>
80 #include <net/tcp.h>
81 #include <net/sock.h>
82 #include <net/ip_fib.h>
83 #include "fib_lookup.h"
84 
85 #define MAX_STAT_DEPTH 32
86 
87 #define KEYLENGTH (8*sizeof(t_key))
88 
89 typedef unsigned int t_key;
90 
91 #define T_TNODE 0
92 #define T_LEAF  1
93 #define NODE_TYPE_MASK	0x1UL
94 #define NODE_TYPE(node) ((node)->parent & NODE_TYPE_MASK)
95 
96 #define IS_TNODE(n) (!(n->parent & T_LEAF))
97 #define IS_LEAF(n) (n->parent & T_LEAF)
98 
99 struct rt_trie_node {
100 	unsigned long parent;
101 	t_key key;
102 };
103 
104 struct leaf {
105 	unsigned long parent;
106 	t_key key;
107 	struct hlist_head list;
108 	struct rcu_head rcu;
109 };
110 
111 struct leaf_info {
112 	struct hlist_node hlist;
113 	int plen;
114 	u32 mask_plen; /* ntohl(inet_make_mask(plen)) */
115 	struct list_head falh;
116 	struct rcu_head rcu;
117 };
118 
119 struct tnode {
120 	unsigned long parent;
121 	t_key key;
122 	unsigned char pos;		/* 2log(KEYLENGTH) bits needed */
123 	unsigned char bits;		/* 2log(KEYLENGTH) bits needed */
124 	unsigned int full_children;	/* KEYLENGTH bits needed */
125 	unsigned int empty_children;	/* KEYLENGTH bits needed */
126 	union {
127 		struct rcu_head rcu;
128 		struct work_struct work;
129 		struct tnode *tnode_free;
130 	};
131 	struct rt_trie_node __rcu *child[0];
132 };
133 
134 #ifdef CONFIG_IP_FIB_TRIE_STATS
135 struct trie_use_stats {
136 	unsigned int gets;
137 	unsigned int backtrack;
138 	unsigned int semantic_match_passed;
139 	unsigned int semantic_match_miss;
140 	unsigned int null_node_hit;
141 	unsigned int resize_node_skipped;
142 };
143 #endif
144 
145 struct trie_stat {
146 	unsigned int totdepth;
147 	unsigned int maxdepth;
148 	unsigned int tnodes;
149 	unsigned int leaves;
150 	unsigned int nullpointers;
151 	unsigned int prefixes;
152 	unsigned int nodesizes[MAX_STAT_DEPTH];
153 };
154 
155 struct trie {
156 	struct rt_trie_node __rcu *trie;
157 #ifdef CONFIG_IP_FIB_TRIE_STATS
158 	struct trie_use_stats stats;
159 #endif
160 };
161 
162 static void tnode_put_child_reorg(struct tnode *tn, int i, struct rt_trie_node *n,
163 				  int wasfull);
164 static struct rt_trie_node *resize(struct trie *t, struct tnode *tn);
165 static struct tnode *inflate(struct trie *t, struct tnode *tn);
166 static struct tnode *halve(struct trie *t, struct tnode *tn);
167 /* tnodes to free after resize(); protected by RTNL */
168 static struct tnode *tnode_free_head;
169 static size_t tnode_free_size;
170 
171 /*
172  * synchronize_rcu after call_rcu for that many pages; it should be especially
173  * useful before resizing the root node with PREEMPT_NONE configs; the value was
174  * obtained experimentally, aiming to avoid visible slowdown.
175  */
176 static const int sync_pages = 128;
177 
178 static struct kmem_cache *fn_alias_kmem __read_mostly;
179 static struct kmem_cache *trie_leaf_kmem __read_mostly;
180 
181 /*
182  * caller must hold RTNL
183  */
184 static inline struct tnode *node_parent(const struct rt_trie_node *node)
185 {
186 	unsigned long parent;
187 
188 	parent = rcu_dereference_index_check(node->parent, lockdep_rtnl_is_held());
189 
190 	return (struct tnode *)(parent & ~NODE_TYPE_MASK);
191 }
192 
193 /*
194  * caller must hold RCU read lock or RTNL
195  */
196 static inline struct tnode *node_parent_rcu(const struct rt_trie_node *node)
197 {
198 	unsigned long parent;
199 
200 	parent = rcu_dereference_index_check(node->parent, rcu_read_lock_held() ||
201 							   lockdep_rtnl_is_held());
202 
203 	return (struct tnode *)(parent & ~NODE_TYPE_MASK);
204 }
205 
206 /* Same as rcu_assign_pointer
207  * but that macro() assumes that value is a pointer.
208  */
209 static inline void node_set_parent(struct rt_trie_node *node, struct tnode *ptr)
210 {
211 	smp_wmb();
212 	node->parent = (unsigned long)ptr | NODE_TYPE(node);
213 }
214 
215 /*
216  * caller must hold RTNL
217  */
218 static inline struct rt_trie_node *tnode_get_child(const struct tnode *tn, unsigned int i)
219 {
220 	BUG_ON(i >= 1U << tn->bits);
221 
222 	return rtnl_dereference(tn->child[i]);
223 }
224 
225 /*
226  * caller must hold RCU read lock or RTNL
227  */
228 static inline struct rt_trie_node *tnode_get_child_rcu(const struct tnode *tn, unsigned int i)
229 {
230 	BUG_ON(i >= 1U << tn->bits);
231 
232 	return rcu_dereference_rtnl(tn->child[i]);
233 }
234 
235 static inline int tnode_child_length(const struct tnode *tn)
236 {
237 	return 1 << tn->bits;
238 }
239 
240 static inline t_key mask_pfx(t_key k, unsigned int l)
241 {
242 	return (l == 0) ? 0 : k >> (KEYLENGTH-l) << (KEYLENGTH-l);
243 }
244 
245 static inline t_key tkey_extract_bits(t_key a, unsigned int offset, unsigned int bits)
246 {
247 	if (offset < KEYLENGTH)
248 		return ((t_key)(a << offset)) >> (KEYLENGTH - bits);
249 	else
250 		return 0;
251 }
252 
253 static inline int tkey_equals(t_key a, t_key b)
254 {
255 	return a == b;
256 }
257 
258 static inline int tkey_sub_equals(t_key a, int offset, int bits, t_key b)
259 {
260 	if (bits == 0 || offset >= KEYLENGTH)
261 		return 1;
262 	bits = bits > KEYLENGTH ? KEYLENGTH : bits;
263 	return ((a ^ b) << offset) >> (KEYLENGTH - bits) == 0;
264 }
265 
266 static inline int tkey_mismatch(t_key a, int offset, t_key b)
267 {
268 	t_key diff = a ^ b;
269 	int i = offset;
270 
271 	if (!diff)
272 		return 0;
273 	while ((diff << i) >> (KEYLENGTH-1) == 0)
274 		i++;
275 	return i;
276 }
277 
278 /*
279   To understand this stuff, an understanding of keys and all their bits is
280   necessary. Every node in the trie has a key associated with it, but not
281   all of the bits in that key are significant.
282 
283   Consider a node 'n' and its parent 'tp'.
284 
285   If n is a leaf, every bit in its key is significant. Its presence is
286   necessitated by path compression, since during a tree traversal (when
287   searching for a leaf - unless we are doing an insertion) we will completely
288   ignore all skipped bits we encounter. Thus we need to verify, at the end of
289   a potentially successful search, that we have indeed been walking the
290   correct key path.
291 
292   Note that we can never "miss" the correct key in the tree if present by
293   following the wrong path. Path compression ensures that segments of the key
294   that are the same for all keys with a given prefix are skipped, but the
295   skipped part *is* identical for each node in the subtrie below the skipped
296   bit! trie_insert() in this implementation takes care of that - note the
297   call to tkey_sub_equals() in trie_insert().
298 
299   if n is an internal node - a 'tnode' here, the various parts of its key
300   have many different meanings.
301 
302   Example:
303   _________________________________________________________________
304   | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
305   -----------------------------------------------------------------
306     0   1   2   3   4   5   6   7   8   9  10  11  12  13  14  15
307 
308   _________________________________________________________________
309   | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
310   -----------------------------------------------------------------
311    16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31
312 
313   tp->pos = 7
314   tp->bits = 3
315   n->pos = 15
316   n->bits = 4
317 
318   First, let's just ignore the bits that come before the parent tp, that is
319   the bits from 0 to (tp->pos-1). They are *known* but at this point we do
320   not use them for anything.
321 
322   The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
323   index into the parent's child array. That is, they will be used to find
324   'n' among tp's children.
325 
326   The bits from (tp->pos + tp->bits) to (n->pos - 1) - "S" - are skipped bits
327   for the node n.
328 
329   All the bits we have seen so far are significant to the node n. The rest
330   of the bits are really not needed or indeed known in n->key.
331 
332   The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
333   n's child array, and will of course be different for each child.
334 
335 
336   The rest of the bits, from (n->pos + n->bits) onward, are completely unknown
337   at this point.
338 
339 */
340 
341 static inline void check_tnode(const struct tnode *tn)
342 {
343 	WARN_ON(tn && tn->pos+tn->bits > 32);
344 }
345 
346 static const int halve_threshold = 25;
347 static const int inflate_threshold = 50;
348 static const int halve_threshold_root = 15;
349 static const int inflate_threshold_root = 30;
350 
351 static void __alias_free_mem(struct rcu_head *head)
352 {
353 	struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
354 	kmem_cache_free(fn_alias_kmem, fa);
355 }
356 
357 static inline void alias_free_mem_rcu(struct fib_alias *fa)
358 {
359 	call_rcu(&fa->rcu, __alias_free_mem);
360 }
361 
362 static void __leaf_free_rcu(struct rcu_head *head)
363 {
364 	struct leaf *l = container_of(head, struct leaf, rcu);
365 	kmem_cache_free(trie_leaf_kmem, l);
366 }
367 
368 static inline void free_leaf(struct leaf *l)
369 {
370 	call_rcu(&l->rcu, __leaf_free_rcu);
371 }
372 
373 static inline void free_leaf_info(struct leaf_info *leaf)
374 {
375 	kfree_rcu(leaf, rcu);
376 }
377 
378 static struct tnode *tnode_alloc(size_t size)
379 {
380 	if (size <= PAGE_SIZE)
381 		return kzalloc(size, GFP_KERNEL);
382 	else
383 		return vzalloc(size);
384 }
385 
386 static void __tnode_vfree(struct work_struct *arg)
387 {
388 	struct tnode *tn = container_of(arg, struct tnode, work);
389 	vfree(tn);
390 }
391 
392 static void __tnode_free_rcu(struct rcu_head *head)
393 {
394 	struct tnode *tn = container_of(head, struct tnode, rcu);
395 	size_t size = sizeof(struct tnode) +
396 		      (sizeof(struct rt_trie_node *) << tn->bits);
397 
398 	if (size <= PAGE_SIZE)
399 		kfree(tn);
400 	else {
401 		INIT_WORK(&tn->work, __tnode_vfree);
402 		schedule_work(&tn->work);
403 	}
404 }
405 
406 static inline void tnode_free(struct tnode *tn)
407 {
408 	if (IS_LEAF(tn))
409 		free_leaf((struct leaf *) tn);
410 	else
411 		call_rcu(&tn->rcu, __tnode_free_rcu);
412 }
413 
414 static void tnode_free_safe(struct tnode *tn)
415 {
416 	BUG_ON(IS_LEAF(tn));
417 	tn->tnode_free = tnode_free_head;
418 	tnode_free_head = tn;
419 	tnode_free_size += sizeof(struct tnode) +
420 			   (sizeof(struct rt_trie_node *) << tn->bits);
421 }
422 
423 static void tnode_free_flush(void)
424 {
425 	struct tnode *tn;
426 
427 	while ((tn = tnode_free_head)) {
428 		tnode_free_head = tn->tnode_free;
429 		tn->tnode_free = NULL;
430 		tnode_free(tn);
431 	}
432 
433 	if (tnode_free_size >= PAGE_SIZE * sync_pages) {
434 		tnode_free_size = 0;
435 		synchronize_rcu();
436 	}
437 }
438 
439 static struct leaf *leaf_new(void)
440 {
441 	struct leaf *l = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
442 	if (l) {
443 		l->parent = T_LEAF;
444 		INIT_HLIST_HEAD(&l->list);
445 	}
446 	return l;
447 }
448 
449 static struct leaf_info *leaf_info_new(int plen)
450 {
451 	struct leaf_info *li = kmalloc(sizeof(struct leaf_info),  GFP_KERNEL);
452 	if (li) {
453 		li->plen = plen;
454 		li->mask_plen = ntohl(inet_make_mask(plen));
455 		INIT_LIST_HEAD(&li->falh);
456 	}
457 	return li;
458 }
459 
460 static struct tnode *tnode_new(t_key key, int pos, int bits)
461 {
462 	size_t sz = sizeof(struct tnode) + (sizeof(struct rt_trie_node *) << bits);
463 	struct tnode *tn = tnode_alloc(sz);
464 
465 	if (tn) {
466 		tn->parent = T_TNODE;
467 		tn->pos = pos;
468 		tn->bits = bits;
469 		tn->key = key;
470 		tn->full_children = 0;
471 		tn->empty_children = 1<<bits;
472 	}
473 
474 	pr_debug("AT %p s=%zu %zu\n", tn, sizeof(struct tnode),
475 		 sizeof(struct rt_trie_node *) << bits);
476 	return tn;
477 }
478 
479 /*
480  * Check whether a tnode 'n' is "full", i.e. it is an internal node
481  * and no bits are skipped. See discussion in dyntree paper p. 6
482  */
483 
484 static inline int tnode_full(const struct tnode *tn, const struct rt_trie_node *n)
485 {
486 	if (n == NULL || IS_LEAF(n))
487 		return 0;
488 
489 	return ((struct tnode *) n)->pos == tn->pos + tn->bits;
490 }
491 
492 static inline void put_child(struct tnode *tn, int i,
493 			     struct rt_trie_node *n)
494 {
495 	tnode_put_child_reorg(tn, i, n, -1);
496 }
497 
498  /*
499   * Add a child at position i overwriting the old value.
500   * Update the value of full_children and empty_children.
501   */
502 
503 static void tnode_put_child_reorg(struct tnode *tn, int i, struct rt_trie_node *n,
504 				  int wasfull)
505 {
506 	struct rt_trie_node *chi = rtnl_dereference(tn->child[i]);
507 	int isfull;
508 
509 	BUG_ON(i >= 1<<tn->bits);
510 
511 	/* update emptyChildren */
512 	if (n == NULL && chi != NULL)
513 		tn->empty_children++;
514 	else if (n != NULL && chi == NULL)
515 		tn->empty_children--;
516 
517 	/* update fullChildren */
518 	if (wasfull == -1)
519 		wasfull = tnode_full(tn, chi);
520 
521 	isfull = tnode_full(tn, n);
522 	if (wasfull && !isfull)
523 		tn->full_children--;
524 	else if (!wasfull && isfull)
525 		tn->full_children++;
526 
527 	if (n)
528 		node_set_parent(n, tn);
529 
530 	rcu_assign_pointer(tn->child[i], n);
531 }
532 
533 #define MAX_WORK 10
534 static struct rt_trie_node *resize(struct trie *t, struct tnode *tn)
535 {
536 	int i;
537 	struct tnode *old_tn;
538 	int inflate_threshold_use;
539 	int halve_threshold_use;
540 	int max_work;
541 
542 	if (!tn)
543 		return NULL;
544 
545 	pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
546 		 tn, inflate_threshold, halve_threshold);
547 
548 	/* No children */
549 	if (tn->empty_children == tnode_child_length(tn)) {
550 		tnode_free_safe(tn);
551 		return NULL;
552 	}
553 	/* One child */
554 	if (tn->empty_children == tnode_child_length(tn) - 1)
555 		goto one_child;
556 	/*
557 	 * Double as long as the resulting node has a number of
558 	 * nonempty nodes that are above the threshold.
559 	 */
560 
561 	/*
562 	 * From "Implementing a dynamic compressed trie" by Stefan Nilsson of
563 	 * the Helsinki University of Technology and Matti Tikkanen of Nokia
564 	 * Telecommunications, page 6:
565 	 * "A node is doubled if the ratio of non-empty children to all
566 	 * children in the *doubled* node is at least 'high'."
567 	 *
568 	 * 'high' in this instance is the variable 'inflate_threshold'. It
569 	 * is expressed as a percentage, so we multiply it with
570 	 * tnode_child_length() and instead of multiplying by 2 (since the
571 	 * child array will be doubled by inflate()) and multiplying
572 	 * the left-hand side by 100 (to handle the percentage thing) we
573 	 * multiply the left-hand side by 50.
574 	 *
575 	 * The left-hand side may look a bit weird: tnode_child_length(tn)
576 	 * - tn->empty_children is of course the number of non-null children
577 	 * in the current node. tn->full_children is the number of "full"
578 	 * children, that is non-null tnodes with a skip value of 0.
579 	 * All of those will be doubled in the resulting inflated tnode, so
580 	 * we just count them one extra time here.
581 	 *
582 	 * A clearer way to write this would be:
583 	 *
584 	 * to_be_doubled = tn->full_children;
585 	 * not_to_be_doubled = tnode_child_length(tn) - tn->empty_children -
586 	 *     tn->full_children;
587 	 *
588 	 * new_child_length = tnode_child_length(tn) * 2;
589 	 *
590 	 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
591 	 *      new_child_length;
592 	 * if (new_fill_factor >= inflate_threshold)
593 	 *
594 	 * ...and so on, tho it would mess up the while () loop.
595 	 *
596 	 * anyway,
597 	 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
598 	 *      inflate_threshold
599 	 *
600 	 * avoid a division:
601 	 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
602 	 *      inflate_threshold * new_child_length
603 	 *
604 	 * expand not_to_be_doubled and to_be_doubled, and shorten:
605 	 * 100 * (tnode_child_length(tn) - tn->empty_children +
606 	 *    tn->full_children) >= inflate_threshold * new_child_length
607 	 *
608 	 * expand new_child_length:
609 	 * 100 * (tnode_child_length(tn) - tn->empty_children +
610 	 *    tn->full_children) >=
611 	 *      inflate_threshold * tnode_child_length(tn) * 2
612 	 *
613 	 * shorten again:
614 	 * 50 * (tn->full_children + tnode_child_length(tn) -
615 	 *    tn->empty_children) >= inflate_threshold *
616 	 *    tnode_child_length(tn)
617 	 *
618 	 */
619 
620 	check_tnode(tn);
621 
622 	/* Keep root node larger  */
623 
624 	if (!node_parent((struct rt_trie_node *)tn)) {
625 		inflate_threshold_use = inflate_threshold_root;
626 		halve_threshold_use = halve_threshold_root;
627 	} else {
628 		inflate_threshold_use = inflate_threshold;
629 		halve_threshold_use = halve_threshold;
630 	}
631 
632 	max_work = MAX_WORK;
633 	while ((tn->full_children > 0 &&  max_work-- &&
634 		50 * (tn->full_children + tnode_child_length(tn)
635 		      - tn->empty_children)
636 		>= inflate_threshold_use * tnode_child_length(tn))) {
637 
638 		old_tn = tn;
639 		tn = inflate(t, tn);
640 
641 		if (IS_ERR(tn)) {
642 			tn = old_tn;
643 #ifdef CONFIG_IP_FIB_TRIE_STATS
644 			t->stats.resize_node_skipped++;
645 #endif
646 			break;
647 		}
648 	}
649 
650 	check_tnode(tn);
651 
652 	/* Return if at least one inflate is run */
653 	if (max_work != MAX_WORK)
654 		return (struct rt_trie_node *) tn;
655 
656 	/*
657 	 * Halve as long as the number of empty children in this
658 	 * node is above threshold.
659 	 */
660 
661 	max_work = MAX_WORK;
662 	while (tn->bits > 1 &&  max_work-- &&
663 	       100 * (tnode_child_length(tn) - tn->empty_children) <
664 	       halve_threshold_use * tnode_child_length(tn)) {
665 
666 		old_tn = tn;
667 		tn = halve(t, tn);
668 		if (IS_ERR(tn)) {
669 			tn = old_tn;
670 #ifdef CONFIG_IP_FIB_TRIE_STATS
671 			t->stats.resize_node_skipped++;
672 #endif
673 			break;
674 		}
675 	}
676 
677 
678 	/* Only one child remains */
679 	if (tn->empty_children == tnode_child_length(tn) - 1) {
680 one_child:
681 		for (i = 0; i < tnode_child_length(tn); i++) {
682 			struct rt_trie_node *n;
683 
684 			n = rtnl_dereference(tn->child[i]);
685 			if (!n)
686 				continue;
687 
688 			/* compress one level */
689 
690 			node_set_parent(n, NULL);
691 			tnode_free_safe(tn);
692 			return n;
693 		}
694 	}
695 	return (struct rt_trie_node *) tn;
696 }
697 
698 
699 static void tnode_clean_free(struct tnode *tn)
700 {
701 	int i;
702 	struct tnode *tofree;
703 
704 	for (i = 0; i < tnode_child_length(tn); i++) {
705 		tofree = (struct tnode *)rtnl_dereference(tn->child[i]);
706 		if (tofree)
707 			tnode_free(tofree);
708 	}
709 	tnode_free(tn);
710 }
711 
712 static struct tnode *inflate(struct trie *t, struct tnode *tn)
713 {
714 	struct tnode *oldtnode = tn;
715 	int olen = tnode_child_length(tn);
716 	int i;
717 
718 	pr_debug("In inflate\n");
719 
720 	tn = tnode_new(oldtnode->key, oldtnode->pos, oldtnode->bits + 1);
721 
722 	if (!tn)
723 		return ERR_PTR(-ENOMEM);
724 
725 	/*
726 	 * Preallocate and store tnodes before the actual work so we
727 	 * don't get into an inconsistent state if memory allocation
728 	 * fails. In case of failure we return the oldnode and  inflate
729 	 * of tnode is ignored.
730 	 */
731 
732 	for (i = 0; i < olen; i++) {
733 		struct tnode *inode;
734 
735 		inode = (struct tnode *) tnode_get_child(oldtnode, i);
736 		if (inode &&
737 		    IS_TNODE(inode) &&
738 		    inode->pos == oldtnode->pos + oldtnode->bits &&
739 		    inode->bits > 1) {
740 			struct tnode *left, *right;
741 			t_key m = ~0U << (KEYLENGTH - 1) >> inode->pos;
742 
743 			left = tnode_new(inode->key&(~m), inode->pos + 1,
744 					 inode->bits - 1);
745 			if (!left)
746 				goto nomem;
747 
748 			right = tnode_new(inode->key|m, inode->pos + 1,
749 					  inode->bits - 1);
750 
751 			if (!right) {
752 				tnode_free(left);
753 				goto nomem;
754 			}
755 
756 			put_child(tn, 2*i, (struct rt_trie_node *) left);
757 			put_child(tn, 2*i+1, (struct rt_trie_node *) right);
758 		}
759 	}
760 
761 	for (i = 0; i < olen; i++) {
762 		struct tnode *inode;
763 		struct rt_trie_node *node = tnode_get_child(oldtnode, i);
764 		struct tnode *left, *right;
765 		int size, j;
766 
767 		/* An empty child */
768 		if (node == NULL)
769 			continue;
770 
771 		/* A leaf or an internal node with skipped bits */
772 
773 		if (IS_LEAF(node) || ((struct tnode *) node)->pos >
774 		   tn->pos + tn->bits - 1) {
775 			if (tkey_extract_bits(node->key,
776 					      oldtnode->pos + oldtnode->bits,
777 					      1) == 0)
778 				put_child(tn, 2*i, node);
779 			else
780 				put_child(tn, 2*i+1, node);
781 			continue;
782 		}
783 
784 		/* An internal node with two children */
785 		inode = (struct tnode *) node;
786 
787 		if (inode->bits == 1) {
788 			put_child(tn, 2*i, rtnl_dereference(inode->child[0]));
789 			put_child(tn, 2*i+1, rtnl_dereference(inode->child[1]));
790 
791 			tnode_free_safe(inode);
792 			continue;
793 		}
794 
795 		/* An internal node with more than two children */
796 
797 		/* We will replace this node 'inode' with two new
798 		 * ones, 'left' and 'right', each with half of the
799 		 * original children. The two new nodes will have
800 		 * a position one bit further down the key and this
801 		 * means that the "significant" part of their keys
802 		 * (see the discussion near the top of this file)
803 		 * will differ by one bit, which will be "0" in
804 		 * left's key and "1" in right's key. Since we are
805 		 * moving the key position by one step, the bit that
806 		 * we are moving away from - the bit at position
807 		 * (inode->pos) - is the one that will differ between
808 		 * left and right. So... we synthesize that bit in the
809 		 * two  new keys.
810 		 * The mask 'm' below will be a single "one" bit at
811 		 * the position (inode->pos)
812 		 */
813 
814 		/* Use the old key, but set the new significant
815 		 *   bit to zero.
816 		 */
817 
818 		left = (struct tnode *) tnode_get_child(tn, 2*i);
819 		put_child(tn, 2*i, NULL);
820 
821 		BUG_ON(!left);
822 
823 		right = (struct tnode *) tnode_get_child(tn, 2*i+1);
824 		put_child(tn, 2*i+1, NULL);
825 
826 		BUG_ON(!right);
827 
828 		size = tnode_child_length(left);
829 		for (j = 0; j < size; j++) {
830 			put_child(left, j, rtnl_dereference(inode->child[j]));
831 			put_child(right, j, rtnl_dereference(inode->child[j + size]));
832 		}
833 		put_child(tn, 2*i, resize(t, left));
834 		put_child(tn, 2*i+1, resize(t, right));
835 
836 		tnode_free_safe(inode);
837 	}
838 	tnode_free_safe(oldtnode);
839 	return tn;
840 nomem:
841 	tnode_clean_free(tn);
842 	return ERR_PTR(-ENOMEM);
843 }
844 
845 static struct tnode *halve(struct trie *t, struct tnode *tn)
846 {
847 	struct tnode *oldtnode = tn;
848 	struct rt_trie_node *left, *right;
849 	int i;
850 	int olen = tnode_child_length(tn);
851 
852 	pr_debug("In halve\n");
853 
854 	tn = tnode_new(oldtnode->key, oldtnode->pos, oldtnode->bits - 1);
855 
856 	if (!tn)
857 		return ERR_PTR(-ENOMEM);
858 
859 	/*
860 	 * Preallocate and store tnodes before the actual work so we
861 	 * don't get into an inconsistent state if memory allocation
862 	 * fails. In case of failure we return the oldnode and halve
863 	 * of tnode is ignored.
864 	 */
865 
866 	for (i = 0; i < olen; i += 2) {
867 		left = tnode_get_child(oldtnode, i);
868 		right = tnode_get_child(oldtnode, i+1);
869 
870 		/* Two nonempty children */
871 		if (left && right) {
872 			struct tnode *newn;
873 
874 			newn = tnode_new(left->key, tn->pos + tn->bits, 1);
875 
876 			if (!newn)
877 				goto nomem;
878 
879 			put_child(tn, i/2, (struct rt_trie_node *)newn);
880 		}
881 
882 	}
883 
884 	for (i = 0; i < olen; i += 2) {
885 		struct tnode *newBinNode;
886 
887 		left = tnode_get_child(oldtnode, i);
888 		right = tnode_get_child(oldtnode, i+1);
889 
890 		/* At least one of the children is empty */
891 		if (left == NULL) {
892 			if (right == NULL)    /* Both are empty */
893 				continue;
894 			put_child(tn, i/2, right);
895 			continue;
896 		}
897 
898 		if (right == NULL) {
899 			put_child(tn, i/2, left);
900 			continue;
901 		}
902 
903 		/* Two nonempty children */
904 		newBinNode = (struct tnode *) tnode_get_child(tn, i/2);
905 		put_child(tn, i/2, NULL);
906 		put_child(newBinNode, 0, left);
907 		put_child(newBinNode, 1, right);
908 		put_child(tn, i/2, resize(t, newBinNode));
909 	}
910 	tnode_free_safe(oldtnode);
911 	return tn;
912 nomem:
913 	tnode_clean_free(tn);
914 	return ERR_PTR(-ENOMEM);
915 }
916 
917 /* readside must use rcu_read_lock currently dump routines
918  via get_fa_head and dump */
919 
920 static struct leaf_info *find_leaf_info(struct leaf *l, int plen)
921 {
922 	struct hlist_head *head = &l->list;
923 	struct hlist_node *node;
924 	struct leaf_info *li;
925 
926 	hlist_for_each_entry_rcu(li, node, head, hlist)
927 		if (li->plen == plen)
928 			return li;
929 
930 	return NULL;
931 }
932 
933 static inline struct list_head *get_fa_head(struct leaf *l, int plen)
934 {
935 	struct leaf_info *li = find_leaf_info(l, plen);
936 
937 	if (!li)
938 		return NULL;
939 
940 	return &li->falh;
941 }
942 
943 static void insert_leaf_info(struct hlist_head *head, struct leaf_info *new)
944 {
945 	struct leaf_info *li = NULL, *last = NULL;
946 	struct hlist_node *node;
947 
948 	if (hlist_empty(head)) {
949 		hlist_add_head_rcu(&new->hlist, head);
950 	} else {
951 		hlist_for_each_entry(li, node, head, hlist) {
952 			if (new->plen > li->plen)
953 				break;
954 
955 			last = li;
956 		}
957 		if (last)
958 			hlist_add_after_rcu(&last->hlist, &new->hlist);
959 		else
960 			hlist_add_before_rcu(&new->hlist, &li->hlist);
961 	}
962 }
963 
964 /* rcu_read_lock needs to be hold by caller from readside */
965 
966 static struct leaf *
967 fib_find_node(struct trie *t, u32 key)
968 {
969 	int pos;
970 	struct tnode *tn;
971 	struct rt_trie_node *n;
972 
973 	pos = 0;
974 	n = rcu_dereference_rtnl(t->trie);
975 
976 	while (n != NULL &&  NODE_TYPE(n) == T_TNODE) {
977 		tn = (struct tnode *) n;
978 
979 		check_tnode(tn);
980 
981 		if (tkey_sub_equals(tn->key, pos, tn->pos-pos, key)) {
982 			pos = tn->pos + tn->bits;
983 			n = tnode_get_child_rcu(tn,
984 						tkey_extract_bits(key,
985 								  tn->pos,
986 								  tn->bits));
987 		} else
988 			break;
989 	}
990 	/* Case we have found a leaf. Compare prefixes */
991 
992 	if (n != NULL && IS_LEAF(n) && tkey_equals(key, n->key))
993 		return (struct leaf *)n;
994 
995 	return NULL;
996 }
997 
998 static void trie_rebalance(struct trie *t, struct tnode *tn)
999 {
1000 	int wasfull;
1001 	t_key cindex, key;
1002 	struct tnode *tp;
1003 
1004 	key = tn->key;
1005 
1006 	while (tn != NULL && (tp = node_parent((struct rt_trie_node *)tn)) != NULL) {
1007 		cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1008 		wasfull = tnode_full(tp, tnode_get_child(tp, cindex));
1009 		tn = (struct tnode *)resize(t, tn);
1010 
1011 		tnode_put_child_reorg(tp, cindex,
1012 				      (struct rt_trie_node *)tn, wasfull);
1013 
1014 		tp = node_parent((struct rt_trie_node *) tn);
1015 		if (!tp)
1016 			rcu_assign_pointer(t->trie, (struct rt_trie_node *)tn);
1017 
1018 		tnode_free_flush();
1019 		if (!tp)
1020 			break;
1021 		tn = tp;
1022 	}
1023 
1024 	/* Handle last (top) tnode */
1025 	if (IS_TNODE(tn))
1026 		tn = (struct tnode *)resize(t, tn);
1027 
1028 	rcu_assign_pointer(t->trie, (struct rt_trie_node *)tn);
1029 	tnode_free_flush();
1030 }
1031 
1032 /* only used from updater-side */
1033 
1034 static struct list_head *fib_insert_node(struct trie *t, u32 key, int plen)
1035 {
1036 	int pos, newpos;
1037 	struct tnode *tp = NULL, *tn = NULL;
1038 	struct rt_trie_node *n;
1039 	struct leaf *l;
1040 	int missbit;
1041 	struct list_head *fa_head = NULL;
1042 	struct leaf_info *li;
1043 	t_key cindex;
1044 
1045 	pos = 0;
1046 	n = rtnl_dereference(t->trie);
1047 
1048 	/* If we point to NULL, stop. Either the tree is empty and we should
1049 	 * just put a new leaf in if, or we have reached an empty child slot,
1050 	 * and we should just put our new leaf in that.
1051 	 * If we point to a T_TNODE, check if it matches our key. Note that
1052 	 * a T_TNODE might be skipping any number of bits - its 'pos' need
1053 	 * not be the parent's 'pos'+'bits'!
1054 	 *
1055 	 * If it does match the current key, get pos/bits from it, extract
1056 	 * the index from our key, push the T_TNODE and walk the tree.
1057 	 *
1058 	 * If it doesn't, we have to replace it with a new T_TNODE.
1059 	 *
1060 	 * If we point to a T_LEAF, it might or might not have the same key
1061 	 * as we do. If it does, just change the value, update the T_LEAF's
1062 	 * value, and return it.
1063 	 * If it doesn't, we need to replace it with a T_TNODE.
1064 	 */
1065 
1066 	while (n != NULL &&  NODE_TYPE(n) == T_TNODE) {
1067 		tn = (struct tnode *) n;
1068 
1069 		check_tnode(tn);
1070 
1071 		if (tkey_sub_equals(tn->key, pos, tn->pos-pos, key)) {
1072 			tp = tn;
1073 			pos = tn->pos + tn->bits;
1074 			n = tnode_get_child(tn,
1075 					    tkey_extract_bits(key,
1076 							      tn->pos,
1077 							      tn->bits));
1078 
1079 			BUG_ON(n && node_parent(n) != tn);
1080 		} else
1081 			break;
1082 	}
1083 
1084 	/*
1085 	 * n  ----> NULL, LEAF or TNODE
1086 	 *
1087 	 * tp is n's (parent) ----> NULL or TNODE
1088 	 */
1089 
1090 	BUG_ON(tp && IS_LEAF(tp));
1091 
1092 	/* Case 1: n is a leaf. Compare prefixes */
1093 
1094 	if (n != NULL && IS_LEAF(n) && tkey_equals(key, n->key)) {
1095 		l = (struct leaf *) n;
1096 		li = leaf_info_new(plen);
1097 
1098 		if (!li)
1099 			return NULL;
1100 
1101 		fa_head = &li->falh;
1102 		insert_leaf_info(&l->list, li);
1103 		goto done;
1104 	}
1105 	l = leaf_new();
1106 
1107 	if (!l)
1108 		return NULL;
1109 
1110 	l->key = key;
1111 	li = leaf_info_new(plen);
1112 
1113 	if (!li) {
1114 		free_leaf(l);
1115 		return NULL;
1116 	}
1117 
1118 	fa_head = &li->falh;
1119 	insert_leaf_info(&l->list, li);
1120 
1121 	if (t->trie && n == NULL) {
1122 		/* Case 2: n is NULL, and will just insert a new leaf */
1123 
1124 		node_set_parent((struct rt_trie_node *)l, tp);
1125 
1126 		cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1127 		put_child(tp, cindex, (struct rt_trie_node *)l);
1128 	} else {
1129 		/* Case 3: n is a LEAF or a TNODE and the key doesn't match. */
1130 		/*
1131 		 *  Add a new tnode here
1132 		 *  first tnode need some special handling
1133 		 */
1134 
1135 		if (tp)
1136 			pos = tp->pos+tp->bits;
1137 		else
1138 			pos = 0;
1139 
1140 		if (n) {
1141 			newpos = tkey_mismatch(key, pos, n->key);
1142 			tn = tnode_new(n->key, newpos, 1);
1143 		} else {
1144 			newpos = 0;
1145 			tn = tnode_new(key, newpos, 1); /* First tnode */
1146 		}
1147 
1148 		if (!tn) {
1149 			free_leaf_info(li);
1150 			free_leaf(l);
1151 			return NULL;
1152 		}
1153 
1154 		node_set_parent((struct rt_trie_node *)tn, tp);
1155 
1156 		missbit = tkey_extract_bits(key, newpos, 1);
1157 		put_child(tn, missbit, (struct rt_trie_node *)l);
1158 		put_child(tn, 1-missbit, n);
1159 
1160 		if (tp) {
1161 			cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1162 			put_child(tp, cindex, (struct rt_trie_node *)tn);
1163 		} else {
1164 			rcu_assign_pointer(t->trie, (struct rt_trie_node *)tn);
1165 			tp = tn;
1166 		}
1167 	}
1168 
1169 	if (tp && tp->pos + tp->bits > 32)
1170 		pr_warn("fib_trie tp=%p pos=%d, bits=%d, key=%0x plen=%d\n",
1171 			tp, tp->pos, tp->bits, key, plen);
1172 
1173 	/* Rebalance the trie */
1174 
1175 	trie_rebalance(t, tp);
1176 done:
1177 	return fa_head;
1178 }
1179 
1180 /*
1181  * Caller must hold RTNL.
1182  */
1183 int fib_table_insert(struct fib_table *tb, struct fib_config *cfg)
1184 {
1185 	struct trie *t = (struct trie *) tb->tb_data;
1186 	struct fib_alias *fa, *new_fa;
1187 	struct list_head *fa_head = NULL;
1188 	struct fib_info *fi;
1189 	int plen = cfg->fc_dst_len;
1190 	u8 tos = cfg->fc_tos;
1191 	u32 key, mask;
1192 	int err;
1193 	struct leaf *l;
1194 
1195 	if (plen > 32)
1196 		return -EINVAL;
1197 
1198 	key = ntohl(cfg->fc_dst);
1199 
1200 	pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1201 
1202 	mask = ntohl(inet_make_mask(plen));
1203 
1204 	if (key & ~mask)
1205 		return -EINVAL;
1206 
1207 	key = key & mask;
1208 
1209 	fi = fib_create_info(cfg);
1210 	if (IS_ERR(fi)) {
1211 		err = PTR_ERR(fi);
1212 		goto err;
1213 	}
1214 
1215 	l = fib_find_node(t, key);
1216 	fa = NULL;
1217 
1218 	if (l) {
1219 		fa_head = get_fa_head(l, plen);
1220 		fa = fib_find_alias(fa_head, tos, fi->fib_priority);
1221 	}
1222 
1223 	/* Now fa, if non-NULL, points to the first fib alias
1224 	 * with the same keys [prefix,tos,priority], if such key already
1225 	 * exists or to the node before which we will insert new one.
1226 	 *
1227 	 * If fa is NULL, we will need to allocate a new one and
1228 	 * insert to the head of f.
1229 	 *
1230 	 * If f is NULL, no fib node matched the destination key
1231 	 * and we need to allocate a new one of those as well.
1232 	 */
1233 
1234 	if (fa && fa->fa_tos == tos &&
1235 	    fa->fa_info->fib_priority == fi->fib_priority) {
1236 		struct fib_alias *fa_first, *fa_match;
1237 
1238 		err = -EEXIST;
1239 		if (cfg->fc_nlflags & NLM_F_EXCL)
1240 			goto out;
1241 
1242 		/* We have 2 goals:
1243 		 * 1. Find exact match for type, scope, fib_info to avoid
1244 		 * duplicate routes
1245 		 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1246 		 */
1247 		fa_match = NULL;
1248 		fa_first = fa;
1249 		fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
1250 		list_for_each_entry_continue(fa, fa_head, fa_list) {
1251 			if (fa->fa_tos != tos)
1252 				break;
1253 			if (fa->fa_info->fib_priority != fi->fib_priority)
1254 				break;
1255 			if (fa->fa_type == cfg->fc_type &&
1256 			    fa->fa_info == fi) {
1257 				fa_match = fa;
1258 				break;
1259 			}
1260 		}
1261 
1262 		if (cfg->fc_nlflags & NLM_F_REPLACE) {
1263 			struct fib_info *fi_drop;
1264 			u8 state;
1265 
1266 			fa = fa_first;
1267 			if (fa_match) {
1268 				if (fa == fa_match)
1269 					err = 0;
1270 				goto out;
1271 			}
1272 			err = -ENOBUFS;
1273 			new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1274 			if (new_fa == NULL)
1275 				goto out;
1276 
1277 			fi_drop = fa->fa_info;
1278 			new_fa->fa_tos = fa->fa_tos;
1279 			new_fa->fa_info = fi;
1280 			new_fa->fa_type = cfg->fc_type;
1281 			state = fa->fa_state;
1282 			new_fa->fa_state = state & ~FA_S_ACCESSED;
1283 
1284 			list_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1285 			alias_free_mem_rcu(fa);
1286 
1287 			fib_release_info(fi_drop);
1288 			if (state & FA_S_ACCESSED)
1289 				rt_cache_flush(cfg->fc_nlinfo.nl_net);
1290 			rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1291 				tb->tb_id, &cfg->fc_nlinfo, NLM_F_REPLACE);
1292 
1293 			goto succeeded;
1294 		}
1295 		/* Error if we find a perfect match which
1296 		 * uses the same scope, type, and nexthop
1297 		 * information.
1298 		 */
1299 		if (fa_match)
1300 			goto out;
1301 
1302 		if (!(cfg->fc_nlflags & NLM_F_APPEND))
1303 			fa = fa_first;
1304 	}
1305 	err = -ENOENT;
1306 	if (!(cfg->fc_nlflags & NLM_F_CREATE))
1307 		goto out;
1308 
1309 	err = -ENOBUFS;
1310 	new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1311 	if (new_fa == NULL)
1312 		goto out;
1313 
1314 	new_fa->fa_info = fi;
1315 	new_fa->fa_tos = tos;
1316 	new_fa->fa_type = cfg->fc_type;
1317 	new_fa->fa_state = 0;
1318 	/*
1319 	 * Insert new entry to the list.
1320 	 */
1321 
1322 	if (!fa_head) {
1323 		fa_head = fib_insert_node(t, key, plen);
1324 		if (unlikely(!fa_head)) {
1325 			err = -ENOMEM;
1326 			goto out_free_new_fa;
1327 		}
1328 	}
1329 
1330 	if (!plen)
1331 		tb->tb_num_default++;
1332 
1333 	list_add_tail_rcu(&new_fa->fa_list,
1334 			  (fa ? &fa->fa_list : fa_head));
1335 
1336 	rt_cache_flush(cfg->fc_nlinfo.nl_net);
1337 	rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, tb->tb_id,
1338 		  &cfg->fc_nlinfo, 0);
1339 succeeded:
1340 	return 0;
1341 
1342 out_free_new_fa:
1343 	kmem_cache_free(fn_alias_kmem, new_fa);
1344 out:
1345 	fib_release_info(fi);
1346 err:
1347 	return err;
1348 }
1349 
1350 /* should be called with rcu_read_lock */
1351 static int check_leaf(struct fib_table *tb, struct trie *t, struct leaf *l,
1352 		      t_key key,  const struct flowi4 *flp,
1353 		      struct fib_result *res, int fib_flags)
1354 {
1355 	struct leaf_info *li;
1356 	struct hlist_head *hhead = &l->list;
1357 	struct hlist_node *node;
1358 
1359 	hlist_for_each_entry_rcu(li, node, hhead, hlist) {
1360 		struct fib_alias *fa;
1361 
1362 		if (l->key != (key & li->mask_plen))
1363 			continue;
1364 
1365 		list_for_each_entry_rcu(fa, &li->falh, fa_list) {
1366 			struct fib_info *fi = fa->fa_info;
1367 			int nhsel, err;
1368 
1369 			if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1370 				continue;
1371 			if (fi->fib_dead)
1372 				continue;
1373 			if (fa->fa_info->fib_scope < flp->flowi4_scope)
1374 				continue;
1375 			fib_alias_accessed(fa);
1376 			err = fib_props[fa->fa_type].error;
1377 			if (err) {
1378 #ifdef CONFIG_IP_FIB_TRIE_STATS
1379 				t->stats.semantic_match_passed++;
1380 #endif
1381 				return err;
1382 			}
1383 			if (fi->fib_flags & RTNH_F_DEAD)
1384 				continue;
1385 			for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
1386 				const struct fib_nh *nh = &fi->fib_nh[nhsel];
1387 
1388 				if (nh->nh_flags & RTNH_F_DEAD)
1389 					continue;
1390 				if (flp->flowi4_oif && flp->flowi4_oif != nh->nh_oif)
1391 					continue;
1392 
1393 #ifdef CONFIG_IP_FIB_TRIE_STATS
1394 				t->stats.semantic_match_passed++;
1395 #endif
1396 				res->prefixlen = li->plen;
1397 				res->nh_sel = nhsel;
1398 				res->type = fa->fa_type;
1399 				res->scope = fa->fa_info->fib_scope;
1400 				res->fi = fi;
1401 				res->table = tb;
1402 				res->fa_head = &li->falh;
1403 				if (!(fib_flags & FIB_LOOKUP_NOREF))
1404 					atomic_inc(&fi->fib_clntref);
1405 				return 0;
1406 			}
1407 		}
1408 
1409 #ifdef CONFIG_IP_FIB_TRIE_STATS
1410 		t->stats.semantic_match_miss++;
1411 #endif
1412 	}
1413 
1414 	return 1;
1415 }
1416 
1417 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1418 		     struct fib_result *res, int fib_flags)
1419 {
1420 	struct trie *t = (struct trie *) tb->tb_data;
1421 	int ret;
1422 	struct rt_trie_node *n;
1423 	struct tnode *pn;
1424 	unsigned int pos, bits;
1425 	t_key key = ntohl(flp->daddr);
1426 	unsigned int chopped_off;
1427 	t_key cindex = 0;
1428 	unsigned int current_prefix_length = KEYLENGTH;
1429 	struct tnode *cn;
1430 	t_key pref_mismatch;
1431 
1432 	rcu_read_lock();
1433 
1434 	n = rcu_dereference(t->trie);
1435 	if (!n)
1436 		goto failed;
1437 
1438 #ifdef CONFIG_IP_FIB_TRIE_STATS
1439 	t->stats.gets++;
1440 #endif
1441 
1442 	/* Just a leaf? */
1443 	if (IS_LEAF(n)) {
1444 		ret = check_leaf(tb, t, (struct leaf *)n, key, flp, res, fib_flags);
1445 		goto found;
1446 	}
1447 
1448 	pn = (struct tnode *) n;
1449 	chopped_off = 0;
1450 
1451 	while (pn) {
1452 		pos = pn->pos;
1453 		bits = pn->bits;
1454 
1455 		if (!chopped_off)
1456 			cindex = tkey_extract_bits(mask_pfx(key, current_prefix_length),
1457 						   pos, bits);
1458 
1459 		n = tnode_get_child_rcu(pn, cindex);
1460 
1461 		if (n == NULL) {
1462 #ifdef CONFIG_IP_FIB_TRIE_STATS
1463 			t->stats.null_node_hit++;
1464 #endif
1465 			goto backtrace;
1466 		}
1467 
1468 		if (IS_LEAF(n)) {
1469 			ret = check_leaf(tb, t, (struct leaf *)n, key, flp, res, fib_flags);
1470 			if (ret > 0)
1471 				goto backtrace;
1472 			goto found;
1473 		}
1474 
1475 		cn = (struct tnode *)n;
1476 
1477 		/*
1478 		 * It's a tnode, and we can do some extra checks here if we
1479 		 * like, to avoid descending into a dead-end branch.
1480 		 * This tnode is in the parent's child array at index
1481 		 * key[p_pos..p_pos+p_bits] but potentially with some bits
1482 		 * chopped off, so in reality the index may be just a
1483 		 * subprefix, padded with zero at the end.
1484 		 * We can also take a look at any skipped bits in this
1485 		 * tnode - everything up to p_pos is supposed to be ok,
1486 		 * and the non-chopped bits of the index (se previous
1487 		 * paragraph) are also guaranteed ok, but the rest is
1488 		 * considered unknown.
1489 		 *
1490 		 * The skipped bits are key[pos+bits..cn->pos].
1491 		 */
1492 
1493 		/* If current_prefix_length < pos+bits, we are already doing
1494 		 * actual prefix  matching, which means everything from
1495 		 * pos+(bits-chopped_off) onward must be zero along some
1496 		 * branch of this subtree - otherwise there is *no* valid
1497 		 * prefix present. Here we can only check the skipped
1498 		 * bits. Remember, since we have already indexed into the
1499 		 * parent's child array, we know that the bits we chopped of
1500 		 * *are* zero.
1501 		 */
1502 
1503 		/* NOTA BENE: Checking only skipped bits
1504 		   for the new node here */
1505 
1506 		if (current_prefix_length < pos+bits) {
1507 			if (tkey_extract_bits(cn->key, current_prefix_length,
1508 						cn->pos - current_prefix_length)
1509 			    || !(cn->child[0]))
1510 				goto backtrace;
1511 		}
1512 
1513 		/*
1514 		 * If chopped_off=0, the index is fully validated and we
1515 		 * only need to look at the skipped bits for this, the new,
1516 		 * tnode. What we actually want to do is to find out if
1517 		 * these skipped bits match our key perfectly, or if we will
1518 		 * have to count on finding a matching prefix further down,
1519 		 * because if we do, we would like to have some way of
1520 		 * verifying the existence of such a prefix at this point.
1521 		 */
1522 
1523 		/* The only thing we can do at this point is to verify that
1524 		 * any such matching prefix can indeed be a prefix to our
1525 		 * key, and if the bits in the node we are inspecting that
1526 		 * do not match our key are not ZERO, this cannot be true.
1527 		 * Thus, find out where there is a mismatch (before cn->pos)
1528 		 * and verify that all the mismatching bits are zero in the
1529 		 * new tnode's key.
1530 		 */
1531 
1532 		/*
1533 		 * Note: We aren't very concerned about the piece of
1534 		 * the key that precede pn->pos+pn->bits, since these
1535 		 * have already been checked. The bits after cn->pos
1536 		 * aren't checked since these are by definition
1537 		 * "unknown" at this point. Thus, what we want to see
1538 		 * is if we are about to enter the "prefix matching"
1539 		 * state, and in that case verify that the skipped
1540 		 * bits that will prevail throughout this subtree are
1541 		 * zero, as they have to be if we are to find a
1542 		 * matching prefix.
1543 		 */
1544 
1545 		pref_mismatch = mask_pfx(cn->key ^ key, cn->pos);
1546 
1547 		/*
1548 		 * In short: If skipped bits in this node do not match
1549 		 * the search key, enter the "prefix matching"
1550 		 * state.directly.
1551 		 */
1552 		if (pref_mismatch) {
1553 			/* fls(x) = __fls(x) + 1 */
1554 			int mp = KEYLENGTH - __fls(pref_mismatch) - 1;
1555 
1556 			if (tkey_extract_bits(cn->key, mp, cn->pos - mp) != 0)
1557 				goto backtrace;
1558 
1559 			if (current_prefix_length >= cn->pos)
1560 				current_prefix_length = mp;
1561 		}
1562 
1563 		pn = (struct tnode *)n; /* Descend */
1564 		chopped_off = 0;
1565 		continue;
1566 
1567 backtrace:
1568 		chopped_off++;
1569 
1570 		/* As zero don't change the child key (cindex) */
1571 		while ((chopped_off <= pn->bits)
1572 		       && !(cindex & (1<<(chopped_off-1))))
1573 			chopped_off++;
1574 
1575 		/* Decrease current_... with bits chopped off */
1576 		if (current_prefix_length > pn->pos + pn->bits - chopped_off)
1577 			current_prefix_length = pn->pos + pn->bits
1578 				- chopped_off;
1579 
1580 		/*
1581 		 * Either we do the actual chop off according or if we have
1582 		 * chopped off all bits in this tnode walk up to our parent.
1583 		 */
1584 
1585 		if (chopped_off <= pn->bits) {
1586 			cindex &= ~(1 << (chopped_off-1));
1587 		} else {
1588 			struct tnode *parent = node_parent_rcu((struct rt_trie_node *) pn);
1589 			if (!parent)
1590 				goto failed;
1591 
1592 			/* Get Child's index */
1593 			cindex = tkey_extract_bits(pn->key, parent->pos, parent->bits);
1594 			pn = parent;
1595 			chopped_off = 0;
1596 
1597 #ifdef CONFIG_IP_FIB_TRIE_STATS
1598 			t->stats.backtrack++;
1599 #endif
1600 			goto backtrace;
1601 		}
1602 	}
1603 failed:
1604 	ret = 1;
1605 found:
1606 	rcu_read_unlock();
1607 	return ret;
1608 }
1609 EXPORT_SYMBOL_GPL(fib_table_lookup);
1610 
1611 /*
1612  * Remove the leaf and return parent.
1613  */
1614 static void trie_leaf_remove(struct trie *t, struct leaf *l)
1615 {
1616 	struct tnode *tp = node_parent((struct rt_trie_node *) l);
1617 
1618 	pr_debug("entering trie_leaf_remove(%p)\n", l);
1619 
1620 	if (tp) {
1621 		t_key cindex = tkey_extract_bits(l->key, tp->pos, tp->bits);
1622 		put_child(tp, cindex, NULL);
1623 		trie_rebalance(t, tp);
1624 	} else
1625 		RCU_INIT_POINTER(t->trie, NULL);
1626 
1627 	free_leaf(l);
1628 }
1629 
1630 /*
1631  * Caller must hold RTNL.
1632  */
1633 int fib_table_delete(struct fib_table *tb, struct fib_config *cfg)
1634 {
1635 	struct trie *t = (struct trie *) tb->tb_data;
1636 	u32 key, mask;
1637 	int plen = cfg->fc_dst_len;
1638 	u8 tos = cfg->fc_tos;
1639 	struct fib_alias *fa, *fa_to_delete;
1640 	struct list_head *fa_head;
1641 	struct leaf *l;
1642 	struct leaf_info *li;
1643 
1644 	if (plen > 32)
1645 		return -EINVAL;
1646 
1647 	key = ntohl(cfg->fc_dst);
1648 	mask = ntohl(inet_make_mask(plen));
1649 
1650 	if (key & ~mask)
1651 		return -EINVAL;
1652 
1653 	key = key & mask;
1654 	l = fib_find_node(t, key);
1655 
1656 	if (!l)
1657 		return -ESRCH;
1658 
1659 	li = find_leaf_info(l, plen);
1660 
1661 	if (!li)
1662 		return -ESRCH;
1663 
1664 	fa_head = &li->falh;
1665 	fa = fib_find_alias(fa_head, tos, 0);
1666 
1667 	if (!fa)
1668 		return -ESRCH;
1669 
1670 	pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1671 
1672 	fa_to_delete = NULL;
1673 	fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
1674 	list_for_each_entry_continue(fa, fa_head, fa_list) {
1675 		struct fib_info *fi = fa->fa_info;
1676 
1677 		if (fa->fa_tos != tos)
1678 			break;
1679 
1680 		if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1681 		    (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1682 		     fa->fa_info->fib_scope == cfg->fc_scope) &&
1683 		    (!cfg->fc_prefsrc ||
1684 		     fi->fib_prefsrc == cfg->fc_prefsrc) &&
1685 		    (!cfg->fc_protocol ||
1686 		     fi->fib_protocol == cfg->fc_protocol) &&
1687 		    fib_nh_match(cfg, fi) == 0) {
1688 			fa_to_delete = fa;
1689 			break;
1690 		}
1691 	}
1692 
1693 	if (!fa_to_delete)
1694 		return -ESRCH;
1695 
1696 	fa = fa_to_delete;
1697 	rtmsg_fib(RTM_DELROUTE, htonl(key), fa, plen, tb->tb_id,
1698 		  &cfg->fc_nlinfo, 0);
1699 
1700 	list_del_rcu(&fa->fa_list);
1701 
1702 	if (!plen)
1703 		tb->tb_num_default--;
1704 
1705 	if (list_empty(fa_head)) {
1706 		hlist_del_rcu(&li->hlist);
1707 		free_leaf_info(li);
1708 	}
1709 
1710 	if (hlist_empty(&l->list))
1711 		trie_leaf_remove(t, l);
1712 
1713 	if (fa->fa_state & FA_S_ACCESSED)
1714 		rt_cache_flush(cfg->fc_nlinfo.nl_net);
1715 
1716 	fib_release_info(fa->fa_info);
1717 	alias_free_mem_rcu(fa);
1718 	return 0;
1719 }
1720 
1721 static int trie_flush_list(struct list_head *head)
1722 {
1723 	struct fib_alias *fa, *fa_node;
1724 	int found = 0;
1725 
1726 	list_for_each_entry_safe(fa, fa_node, head, fa_list) {
1727 		struct fib_info *fi = fa->fa_info;
1728 
1729 		if (fi && (fi->fib_flags & RTNH_F_DEAD)) {
1730 			list_del_rcu(&fa->fa_list);
1731 			fib_release_info(fa->fa_info);
1732 			alias_free_mem_rcu(fa);
1733 			found++;
1734 		}
1735 	}
1736 	return found;
1737 }
1738 
1739 static int trie_flush_leaf(struct leaf *l)
1740 {
1741 	int found = 0;
1742 	struct hlist_head *lih = &l->list;
1743 	struct hlist_node *node, *tmp;
1744 	struct leaf_info *li = NULL;
1745 
1746 	hlist_for_each_entry_safe(li, node, tmp, lih, hlist) {
1747 		found += trie_flush_list(&li->falh);
1748 
1749 		if (list_empty(&li->falh)) {
1750 			hlist_del_rcu(&li->hlist);
1751 			free_leaf_info(li);
1752 		}
1753 	}
1754 	return found;
1755 }
1756 
1757 /*
1758  * Scan for the next right leaf starting at node p->child[idx]
1759  * Since we have back pointer, no recursion necessary.
1760  */
1761 static struct leaf *leaf_walk_rcu(struct tnode *p, struct rt_trie_node *c)
1762 {
1763 	do {
1764 		t_key idx;
1765 
1766 		if (c)
1767 			idx = tkey_extract_bits(c->key, p->pos, p->bits) + 1;
1768 		else
1769 			idx = 0;
1770 
1771 		while (idx < 1u << p->bits) {
1772 			c = tnode_get_child_rcu(p, idx++);
1773 			if (!c)
1774 				continue;
1775 
1776 			if (IS_LEAF(c)) {
1777 				prefetch(rcu_dereference_rtnl(p->child[idx]));
1778 				return (struct leaf *) c;
1779 			}
1780 
1781 			/* Rescan start scanning in new node */
1782 			p = (struct tnode *) c;
1783 			idx = 0;
1784 		}
1785 
1786 		/* Node empty, walk back up to parent */
1787 		c = (struct rt_trie_node *) p;
1788 	} while ((p = node_parent_rcu(c)) != NULL);
1789 
1790 	return NULL; /* Root of trie */
1791 }
1792 
1793 static struct leaf *trie_firstleaf(struct trie *t)
1794 {
1795 	struct tnode *n = (struct tnode *)rcu_dereference_rtnl(t->trie);
1796 
1797 	if (!n)
1798 		return NULL;
1799 
1800 	if (IS_LEAF(n))          /* trie is just a leaf */
1801 		return (struct leaf *) n;
1802 
1803 	return leaf_walk_rcu(n, NULL);
1804 }
1805 
1806 static struct leaf *trie_nextleaf(struct leaf *l)
1807 {
1808 	struct rt_trie_node *c = (struct rt_trie_node *) l;
1809 	struct tnode *p = node_parent_rcu(c);
1810 
1811 	if (!p)
1812 		return NULL;	/* trie with just one leaf */
1813 
1814 	return leaf_walk_rcu(p, c);
1815 }
1816 
1817 static struct leaf *trie_leafindex(struct trie *t, int index)
1818 {
1819 	struct leaf *l = trie_firstleaf(t);
1820 
1821 	while (l && index-- > 0)
1822 		l = trie_nextleaf(l);
1823 
1824 	return l;
1825 }
1826 
1827 
1828 /*
1829  * Caller must hold RTNL.
1830  */
1831 int fib_table_flush(struct fib_table *tb)
1832 {
1833 	struct trie *t = (struct trie *) tb->tb_data;
1834 	struct leaf *l, *ll = NULL;
1835 	int found = 0;
1836 
1837 	for (l = trie_firstleaf(t); l; l = trie_nextleaf(l)) {
1838 		found += trie_flush_leaf(l);
1839 
1840 		if (ll && hlist_empty(&ll->list))
1841 			trie_leaf_remove(t, ll);
1842 		ll = l;
1843 	}
1844 
1845 	if (ll && hlist_empty(&ll->list))
1846 		trie_leaf_remove(t, ll);
1847 
1848 	pr_debug("trie_flush found=%d\n", found);
1849 	return found;
1850 }
1851 
1852 void fib_free_table(struct fib_table *tb)
1853 {
1854 	kfree(tb);
1855 }
1856 
1857 static int fn_trie_dump_fa(t_key key, int plen, struct list_head *fah,
1858 			   struct fib_table *tb,
1859 			   struct sk_buff *skb, struct netlink_callback *cb)
1860 {
1861 	int i, s_i;
1862 	struct fib_alias *fa;
1863 	__be32 xkey = htonl(key);
1864 
1865 	s_i = cb->args[5];
1866 	i = 0;
1867 
1868 	/* rcu_read_lock is hold by caller */
1869 
1870 	list_for_each_entry_rcu(fa, fah, fa_list) {
1871 		if (i < s_i) {
1872 			i++;
1873 			continue;
1874 		}
1875 
1876 		if (fib_dump_info(skb, NETLINK_CB(cb->skb).portid,
1877 				  cb->nlh->nlmsg_seq,
1878 				  RTM_NEWROUTE,
1879 				  tb->tb_id,
1880 				  fa->fa_type,
1881 				  xkey,
1882 				  plen,
1883 				  fa->fa_tos,
1884 				  fa->fa_info, NLM_F_MULTI) < 0) {
1885 			cb->args[5] = i;
1886 			return -1;
1887 		}
1888 		i++;
1889 	}
1890 	cb->args[5] = i;
1891 	return skb->len;
1892 }
1893 
1894 static int fn_trie_dump_leaf(struct leaf *l, struct fib_table *tb,
1895 			struct sk_buff *skb, struct netlink_callback *cb)
1896 {
1897 	struct leaf_info *li;
1898 	struct hlist_node *node;
1899 	int i, s_i;
1900 
1901 	s_i = cb->args[4];
1902 	i = 0;
1903 
1904 	/* rcu_read_lock is hold by caller */
1905 	hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
1906 		if (i < s_i) {
1907 			i++;
1908 			continue;
1909 		}
1910 
1911 		if (i > s_i)
1912 			cb->args[5] = 0;
1913 
1914 		if (list_empty(&li->falh))
1915 			continue;
1916 
1917 		if (fn_trie_dump_fa(l->key, li->plen, &li->falh, tb, skb, cb) < 0) {
1918 			cb->args[4] = i;
1919 			return -1;
1920 		}
1921 		i++;
1922 	}
1923 
1924 	cb->args[4] = i;
1925 	return skb->len;
1926 }
1927 
1928 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
1929 		   struct netlink_callback *cb)
1930 {
1931 	struct leaf *l;
1932 	struct trie *t = (struct trie *) tb->tb_data;
1933 	t_key key = cb->args[2];
1934 	int count = cb->args[3];
1935 
1936 	rcu_read_lock();
1937 	/* Dump starting at last key.
1938 	 * Note: 0.0.0.0/0 (ie default) is first key.
1939 	 */
1940 	if (count == 0)
1941 		l = trie_firstleaf(t);
1942 	else {
1943 		/* Normally, continue from last key, but if that is missing
1944 		 * fallback to using slow rescan
1945 		 */
1946 		l = fib_find_node(t, key);
1947 		if (!l)
1948 			l = trie_leafindex(t, count);
1949 	}
1950 
1951 	while (l) {
1952 		cb->args[2] = l->key;
1953 		if (fn_trie_dump_leaf(l, tb, skb, cb) < 0) {
1954 			cb->args[3] = count;
1955 			rcu_read_unlock();
1956 			return -1;
1957 		}
1958 
1959 		++count;
1960 		l = trie_nextleaf(l);
1961 		memset(&cb->args[4], 0,
1962 		       sizeof(cb->args) - 4*sizeof(cb->args[0]));
1963 	}
1964 	cb->args[3] = count;
1965 	rcu_read_unlock();
1966 
1967 	return skb->len;
1968 }
1969 
1970 void __init fib_trie_init(void)
1971 {
1972 	fn_alias_kmem = kmem_cache_create("ip_fib_alias",
1973 					  sizeof(struct fib_alias),
1974 					  0, SLAB_PANIC, NULL);
1975 
1976 	trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
1977 					   max(sizeof(struct leaf),
1978 					       sizeof(struct leaf_info)),
1979 					   0, SLAB_PANIC, NULL);
1980 }
1981 
1982 
1983 struct fib_table *fib_trie_table(u32 id)
1984 {
1985 	struct fib_table *tb;
1986 	struct trie *t;
1987 
1988 	tb = kmalloc(sizeof(struct fib_table) + sizeof(struct trie),
1989 		     GFP_KERNEL);
1990 	if (tb == NULL)
1991 		return NULL;
1992 
1993 	tb->tb_id = id;
1994 	tb->tb_default = -1;
1995 	tb->tb_num_default = 0;
1996 
1997 	t = (struct trie *) tb->tb_data;
1998 	memset(t, 0, sizeof(*t));
1999 
2000 	return tb;
2001 }
2002 
2003 #ifdef CONFIG_PROC_FS
2004 /* Depth first Trie walk iterator */
2005 struct fib_trie_iter {
2006 	struct seq_net_private p;
2007 	struct fib_table *tb;
2008 	struct tnode *tnode;
2009 	unsigned int index;
2010 	unsigned int depth;
2011 };
2012 
2013 static struct rt_trie_node *fib_trie_get_next(struct fib_trie_iter *iter)
2014 {
2015 	struct tnode *tn = iter->tnode;
2016 	unsigned int cindex = iter->index;
2017 	struct tnode *p;
2018 
2019 	/* A single entry routing table */
2020 	if (!tn)
2021 		return NULL;
2022 
2023 	pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2024 		 iter->tnode, iter->index, iter->depth);
2025 rescan:
2026 	while (cindex < (1<<tn->bits)) {
2027 		struct rt_trie_node *n = tnode_get_child_rcu(tn, cindex);
2028 
2029 		if (n) {
2030 			if (IS_LEAF(n)) {
2031 				iter->tnode = tn;
2032 				iter->index = cindex + 1;
2033 			} else {
2034 				/* push down one level */
2035 				iter->tnode = (struct tnode *) n;
2036 				iter->index = 0;
2037 				++iter->depth;
2038 			}
2039 			return n;
2040 		}
2041 
2042 		++cindex;
2043 	}
2044 
2045 	/* Current node exhausted, pop back up */
2046 	p = node_parent_rcu((struct rt_trie_node *)tn);
2047 	if (p) {
2048 		cindex = tkey_extract_bits(tn->key, p->pos, p->bits)+1;
2049 		tn = p;
2050 		--iter->depth;
2051 		goto rescan;
2052 	}
2053 
2054 	/* got root? */
2055 	return NULL;
2056 }
2057 
2058 static struct rt_trie_node *fib_trie_get_first(struct fib_trie_iter *iter,
2059 				       struct trie *t)
2060 {
2061 	struct rt_trie_node *n;
2062 
2063 	if (!t)
2064 		return NULL;
2065 
2066 	n = rcu_dereference(t->trie);
2067 	if (!n)
2068 		return NULL;
2069 
2070 	if (IS_TNODE(n)) {
2071 		iter->tnode = (struct tnode *) n;
2072 		iter->index = 0;
2073 		iter->depth = 1;
2074 	} else {
2075 		iter->tnode = NULL;
2076 		iter->index = 0;
2077 		iter->depth = 0;
2078 	}
2079 
2080 	return n;
2081 }
2082 
2083 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2084 {
2085 	struct rt_trie_node *n;
2086 	struct fib_trie_iter iter;
2087 
2088 	memset(s, 0, sizeof(*s));
2089 
2090 	rcu_read_lock();
2091 	for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2092 		if (IS_LEAF(n)) {
2093 			struct leaf *l = (struct leaf *)n;
2094 			struct leaf_info *li;
2095 			struct hlist_node *tmp;
2096 
2097 			s->leaves++;
2098 			s->totdepth += iter.depth;
2099 			if (iter.depth > s->maxdepth)
2100 				s->maxdepth = iter.depth;
2101 
2102 			hlist_for_each_entry_rcu(li, tmp, &l->list, hlist)
2103 				++s->prefixes;
2104 		} else {
2105 			const struct tnode *tn = (const struct tnode *) n;
2106 			int i;
2107 
2108 			s->tnodes++;
2109 			if (tn->bits < MAX_STAT_DEPTH)
2110 				s->nodesizes[tn->bits]++;
2111 
2112 			for (i = 0; i < (1<<tn->bits); i++)
2113 				if (!tn->child[i])
2114 					s->nullpointers++;
2115 		}
2116 	}
2117 	rcu_read_unlock();
2118 }
2119 
2120 /*
2121  *	This outputs /proc/net/fib_triestats
2122  */
2123 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2124 {
2125 	unsigned int i, max, pointers, bytes, avdepth;
2126 
2127 	if (stat->leaves)
2128 		avdepth = stat->totdepth*100 / stat->leaves;
2129 	else
2130 		avdepth = 0;
2131 
2132 	seq_printf(seq, "\tAver depth:     %u.%02d\n",
2133 		   avdepth / 100, avdepth % 100);
2134 	seq_printf(seq, "\tMax depth:      %u\n", stat->maxdepth);
2135 
2136 	seq_printf(seq, "\tLeaves:         %u\n", stat->leaves);
2137 	bytes = sizeof(struct leaf) * stat->leaves;
2138 
2139 	seq_printf(seq, "\tPrefixes:       %u\n", stat->prefixes);
2140 	bytes += sizeof(struct leaf_info) * stat->prefixes;
2141 
2142 	seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2143 	bytes += sizeof(struct tnode) * stat->tnodes;
2144 
2145 	max = MAX_STAT_DEPTH;
2146 	while (max > 0 && stat->nodesizes[max-1] == 0)
2147 		max--;
2148 
2149 	pointers = 0;
2150 	for (i = 1; i <= max; i++)
2151 		if (stat->nodesizes[i] != 0) {
2152 			seq_printf(seq, "  %u: %u",  i, stat->nodesizes[i]);
2153 			pointers += (1<<i) * stat->nodesizes[i];
2154 		}
2155 	seq_putc(seq, '\n');
2156 	seq_printf(seq, "\tPointers: %u\n", pointers);
2157 
2158 	bytes += sizeof(struct rt_trie_node *) * pointers;
2159 	seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2160 	seq_printf(seq, "Total size: %u  kB\n", (bytes + 1023) / 1024);
2161 }
2162 
2163 #ifdef CONFIG_IP_FIB_TRIE_STATS
2164 static void trie_show_usage(struct seq_file *seq,
2165 			    const struct trie_use_stats *stats)
2166 {
2167 	seq_printf(seq, "\nCounters:\n---------\n");
2168 	seq_printf(seq, "gets = %u\n", stats->gets);
2169 	seq_printf(seq, "backtracks = %u\n", stats->backtrack);
2170 	seq_printf(seq, "semantic match passed = %u\n",
2171 		   stats->semantic_match_passed);
2172 	seq_printf(seq, "semantic match miss = %u\n",
2173 		   stats->semantic_match_miss);
2174 	seq_printf(seq, "null node hit= %u\n", stats->null_node_hit);
2175 	seq_printf(seq, "skipped node resize = %u\n\n",
2176 		   stats->resize_node_skipped);
2177 }
2178 #endif /*  CONFIG_IP_FIB_TRIE_STATS */
2179 
2180 static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2181 {
2182 	if (tb->tb_id == RT_TABLE_LOCAL)
2183 		seq_puts(seq, "Local:\n");
2184 	else if (tb->tb_id == RT_TABLE_MAIN)
2185 		seq_puts(seq, "Main:\n");
2186 	else
2187 		seq_printf(seq, "Id %d:\n", tb->tb_id);
2188 }
2189 
2190 
2191 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2192 {
2193 	struct net *net = (struct net *)seq->private;
2194 	unsigned int h;
2195 
2196 	seq_printf(seq,
2197 		   "Basic info: size of leaf:"
2198 		   " %Zd bytes, size of tnode: %Zd bytes.\n",
2199 		   sizeof(struct leaf), sizeof(struct tnode));
2200 
2201 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2202 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2203 		struct hlist_node *node;
2204 		struct fib_table *tb;
2205 
2206 		hlist_for_each_entry_rcu(tb, node, head, tb_hlist) {
2207 			struct trie *t = (struct trie *) tb->tb_data;
2208 			struct trie_stat stat;
2209 
2210 			if (!t)
2211 				continue;
2212 
2213 			fib_table_print(seq, tb);
2214 
2215 			trie_collect_stats(t, &stat);
2216 			trie_show_stats(seq, &stat);
2217 #ifdef CONFIG_IP_FIB_TRIE_STATS
2218 			trie_show_usage(seq, &t->stats);
2219 #endif
2220 		}
2221 	}
2222 
2223 	return 0;
2224 }
2225 
2226 static int fib_triestat_seq_open(struct inode *inode, struct file *file)
2227 {
2228 	return single_open_net(inode, file, fib_triestat_seq_show);
2229 }
2230 
2231 static const struct file_operations fib_triestat_fops = {
2232 	.owner	= THIS_MODULE,
2233 	.open	= fib_triestat_seq_open,
2234 	.read	= seq_read,
2235 	.llseek	= seq_lseek,
2236 	.release = single_release_net,
2237 };
2238 
2239 static struct rt_trie_node *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2240 {
2241 	struct fib_trie_iter *iter = seq->private;
2242 	struct net *net = seq_file_net(seq);
2243 	loff_t idx = 0;
2244 	unsigned int h;
2245 
2246 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2247 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2248 		struct hlist_node *node;
2249 		struct fib_table *tb;
2250 
2251 		hlist_for_each_entry_rcu(tb, node, head, tb_hlist) {
2252 			struct rt_trie_node *n;
2253 
2254 			for (n = fib_trie_get_first(iter,
2255 						    (struct trie *) tb->tb_data);
2256 			     n; n = fib_trie_get_next(iter))
2257 				if (pos == idx++) {
2258 					iter->tb = tb;
2259 					return n;
2260 				}
2261 		}
2262 	}
2263 
2264 	return NULL;
2265 }
2266 
2267 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2268 	__acquires(RCU)
2269 {
2270 	rcu_read_lock();
2271 	return fib_trie_get_idx(seq, *pos);
2272 }
2273 
2274 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2275 {
2276 	struct fib_trie_iter *iter = seq->private;
2277 	struct net *net = seq_file_net(seq);
2278 	struct fib_table *tb = iter->tb;
2279 	struct hlist_node *tb_node;
2280 	unsigned int h;
2281 	struct rt_trie_node *n;
2282 
2283 	++*pos;
2284 	/* next node in same table */
2285 	n = fib_trie_get_next(iter);
2286 	if (n)
2287 		return n;
2288 
2289 	/* walk rest of this hash chain */
2290 	h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2291 	while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2292 		tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2293 		n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2294 		if (n)
2295 			goto found;
2296 	}
2297 
2298 	/* new hash chain */
2299 	while (++h < FIB_TABLE_HASHSZ) {
2300 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2301 		hlist_for_each_entry_rcu(tb, tb_node, head, tb_hlist) {
2302 			n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2303 			if (n)
2304 				goto found;
2305 		}
2306 	}
2307 	return NULL;
2308 
2309 found:
2310 	iter->tb = tb;
2311 	return n;
2312 }
2313 
2314 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2315 	__releases(RCU)
2316 {
2317 	rcu_read_unlock();
2318 }
2319 
2320 static void seq_indent(struct seq_file *seq, int n)
2321 {
2322 	while (n-- > 0)
2323 		seq_puts(seq, "   ");
2324 }
2325 
2326 static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2327 {
2328 	switch (s) {
2329 	case RT_SCOPE_UNIVERSE: return "universe";
2330 	case RT_SCOPE_SITE:	return "site";
2331 	case RT_SCOPE_LINK:	return "link";
2332 	case RT_SCOPE_HOST:	return "host";
2333 	case RT_SCOPE_NOWHERE:	return "nowhere";
2334 	default:
2335 		snprintf(buf, len, "scope=%d", s);
2336 		return buf;
2337 	}
2338 }
2339 
2340 static const char *const rtn_type_names[__RTN_MAX] = {
2341 	[RTN_UNSPEC] = "UNSPEC",
2342 	[RTN_UNICAST] = "UNICAST",
2343 	[RTN_LOCAL] = "LOCAL",
2344 	[RTN_BROADCAST] = "BROADCAST",
2345 	[RTN_ANYCAST] = "ANYCAST",
2346 	[RTN_MULTICAST] = "MULTICAST",
2347 	[RTN_BLACKHOLE] = "BLACKHOLE",
2348 	[RTN_UNREACHABLE] = "UNREACHABLE",
2349 	[RTN_PROHIBIT] = "PROHIBIT",
2350 	[RTN_THROW] = "THROW",
2351 	[RTN_NAT] = "NAT",
2352 	[RTN_XRESOLVE] = "XRESOLVE",
2353 };
2354 
2355 static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2356 {
2357 	if (t < __RTN_MAX && rtn_type_names[t])
2358 		return rtn_type_names[t];
2359 	snprintf(buf, len, "type %u", t);
2360 	return buf;
2361 }
2362 
2363 /* Pretty print the trie */
2364 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2365 {
2366 	const struct fib_trie_iter *iter = seq->private;
2367 	struct rt_trie_node *n = v;
2368 
2369 	if (!node_parent_rcu(n))
2370 		fib_table_print(seq, iter->tb);
2371 
2372 	if (IS_TNODE(n)) {
2373 		struct tnode *tn = (struct tnode *) n;
2374 		__be32 prf = htonl(mask_pfx(tn->key, tn->pos));
2375 
2376 		seq_indent(seq, iter->depth-1);
2377 		seq_printf(seq, "  +-- %pI4/%d %d %d %d\n",
2378 			   &prf, tn->pos, tn->bits, tn->full_children,
2379 			   tn->empty_children);
2380 
2381 	} else {
2382 		struct leaf *l = (struct leaf *) n;
2383 		struct leaf_info *li;
2384 		struct hlist_node *node;
2385 		__be32 val = htonl(l->key);
2386 
2387 		seq_indent(seq, iter->depth);
2388 		seq_printf(seq, "  |-- %pI4\n", &val);
2389 
2390 		hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
2391 			struct fib_alias *fa;
2392 
2393 			list_for_each_entry_rcu(fa, &li->falh, fa_list) {
2394 				char buf1[32], buf2[32];
2395 
2396 				seq_indent(seq, iter->depth+1);
2397 				seq_printf(seq, "  /%d %s %s", li->plen,
2398 					   rtn_scope(buf1, sizeof(buf1),
2399 						     fa->fa_info->fib_scope),
2400 					   rtn_type(buf2, sizeof(buf2),
2401 						    fa->fa_type));
2402 				if (fa->fa_tos)
2403 					seq_printf(seq, " tos=%d", fa->fa_tos);
2404 				seq_putc(seq, '\n');
2405 			}
2406 		}
2407 	}
2408 
2409 	return 0;
2410 }
2411 
2412 static const struct seq_operations fib_trie_seq_ops = {
2413 	.start  = fib_trie_seq_start,
2414 	.next   = fib_trie_seq_next,
2415 	.stop   = fib_trie_seq_stop,
2416 	.show   = fib_trie_seq_show,
2417 };
2418 
2419 static int fib_trie_seq_open(struct inode *inode, struct file *file)
2420 {
2421 	return seq_open_net(inode, file, &fib_trie_seq_ops,
2422 			    sizeof(struct fib_trie_iter));
2423 }
2424 
2425 static const struct file_operations fib_trie_fops = {
2426 	.owner  = THIS_MODULE,
2427 	.open   = fib_trie_seq_open,
2428 	.read   = seq_read,
2429 	.llseek = seq_lseek,
2430 	.release = seq_release_net,
2431 };
2432 
2433 struct fib_route_iter {
2434 	struct seq_net_private p;
2435 	struct trie *main_trie;
2436 	loff_t	pos;
2437 	t_key	key;
2438 };
2439 
2440 static struct leaf *fib_route_get_idx(struct fib_route_iter *iter, loff_t pos)
2441 {
2442 	struct leaf *l = NULL;
2443 	struct trie *t = iter->main_trie;
2444 
2445 	/* use cache location of last found key */
2446 	if (iter->pos > 0 && pos >= iter->pos && (l = fib_find_node(t, iter->key)))
2447 		pos -= iter->pos;
2448 	else {
2449 		iter->pos = 0;
2450 		l = trie_firstleaf(t);
2451 	}
2452 
2453 	while (l && pos-- > 0) {
2454 		iter->pos++;
2455 		l = trie_nextleaf(l);
2456 	}
2457 
2458 	if (l)
2459 		iter->key = pos;	/* remember it */
2460 	else
2461 		iter->pos = 0;		/* forget it */
2462 
2463 	return l;
2464 }
2465 
2466 static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2467 	__acquires(RCU)
2468 {
2469 	struct fib_route_iter *iter = seq->private;
2470 	struct fib_table *tb;
2471 
2472 	rcu_read_lock();
2473 	tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2474 	if (!tb)
2475 		return NULL;
2476 
2477 	iter->main_trie = (struct trie *) tb->tb_data;
2478 	if (*pos == 0)
2479 		return SEQ_START_TOKEN;
2480 	else
2481 		return fib_route_get_idx(iter, *pos - 1);
2482 }
2483 
2484 static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2485 {
2486 	struct fib_route_iter *iter = seq->private;
2487 	struct leaf *l = v;
2488 
2489 	++*pos;
2490 	if (v == SEQ_START_TOKEN) {
2491 		iter->pos = 0;
2492 		l = trie_firstleaf(iter->main_trie);
2493 	} else {
2494 		iter->pos++;
2495 		l = trie_nextleaf(l);
2496 	}
2497 
2498 	if (l)
2499 		iter->key = l->key;
2500 	else
2501 		iter->pos = 0;
2502 	return l;
2503 }
2504 
2505 static void fib_route_seq_stop(struct seq_file *seq, void *v)
2506 	__releases(RCU)
2507 {
2508 	rcu_read_unlock();
2509 }
2510 
2511 static unsigned int fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2512 {
2513 	unsigned int flags = 0;
2514 
2515 	if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2516 		flags = RTF_REJECT;
2517 	if (fi && fi->fib_nh->nh_gw)
2518 		flags |= RTF_GATEWAY;
2519 	if (mask == htonl(0xFFFFFFFF))
2520 		flags |= RTF_HOST;
2521 	flags |= RTF_UP;
2522 	return flags;
2523 }
2524 
2525 /*
2526  *	This outputs /proc/net/route.
2527  *	The format of the file is not supposed to be changed
2528  *	and needs to be same as fib_hash output to avoid breaking
2529  *	legacy utilities
2530  */
2531 static int fib_route_seq_show(struct seq_file *seq, void *v)
2532 {
2533 	struct leaf *l = v;
2534 	struct leaf_info *li;
2535 	struct hlist_node *node;
2536 
2537 	if (v == SEQ_START_TOKEN) {
2538 		seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2539 			   "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2540 			   "\tWindow\tIRTT");
2541 		return 0;
2542 	}
2543 
2544 	hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
2545 		struct fib_alias *fa;
2546 		__be32 mask, prefix;
2547 
2548 		mask = inet_make_mask(li->plen);
2549 		prefix = htonl(l->key);
2550 
2551 		list_for_each_entry_rcu(fa, &li->falh, fa_list) {
2552 			const struct fib_info *fi = fa->fa_info;
2553 			unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2554 			int len;
2555 
2556 			if (fa->fa_type == RTN_BROADCAST
2557 			    || fa->fa_type == RTN_MULTICAST)
2558 				continue;
2559 
2560 			if (fi)
2561 				seq_printf(seq,
2562 					 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2563 					 "%d\t%08X\t%d\t%u\t%u%n",
2564 					 fi->fib_dev ? fi->fib_dev->name : "*",
2565 					 prefix,
2566 					 fi->fib_nh->nh_gw, flags, 0, 0,
2567 					 fi->fib_priority,
2568 					 mask,
2569 					 (fi->fib_advmss ?
2570 					  fi->fib_advmss + 40 : 0),
2571 					 fi->fib_window,
2572 					 fi->fib_rtt >> 3, &len);
2573 			else
2574 				seq_printf(seq,
2575 					 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2576 					 "%d\t%08X\t%d\t%u\t%u%n",
2577 					 prefix, 0, flags, 0, 0, 0,
2578 					 mask, 0, 0, 0, &len);
2579 
2580 			seq_printf(seq, "%*s\n", 127 - len, "");
2581 		}
2582 	}
2583 
2584 	return 0;
2585 }
2586 
2587 static const struct seq_operations fib_route_seq_ops = {
2588 	.start  = fib_route_seq_start,
2589 	.next   = fib_route_seq_next,
2590 	.stop   = fib_route_seq_stop,
2591 	.show   = fib_route_seq_show,
2592 };
2593 
2594 static int fib_route_seq_open(struct inode *inode, struct file *file)
2595 {
2596 	return seq_open_net(inode, file, &fib_route_seq_ops,
2597 			    sizeof(struct fib_route_iter));
2598 }
2599 
2600 static const struct file_operations fib_route_fops = {
2601 	.owner  = THIS_MODULE,
2602 	.open   = fib_route_seq_open,
2603 	.read   = seq_read,
2604 	.llseek = seq_lseek,
2605 	.release = seq_release_net,
2606 };
2607 
2608 int __net_init fib_proc_init(struct net *net)
2609 {
2610 	if (!proc_net_fops_create(net, "fib_trie", S_IRUGO, &fib_trie_fops))
2611 		goto out1;
2612 
2613 	if (!proc_net_fops_create(net, "fib_triestat", S_IRUGO,
2614 				  &fib_triestat_fops))
2615 		goto out2;
2616 
2617 	if (!proc_net_fops_create(net, "route", S_IRUGO, &fib_route_fops))
2618 		goto out3;
2619 
2620 	return 0;
2621 
2622 out3:
2623 	proc_net_remove(net, "fib_triestat");
2624 out2:
2625 	proc_net_remove(net, "fib_trie");
2626 out1:
2627 	return -ENOMEM;
2628 }
2629 
2630 void __net_exit fib_proc_exit(struct net *net)
2631 {
2632 	proc_net_remove(net, "fib_trie");
2633 	proc_net_remove(net, "fib_triestat");
2634 	proc_net_remove(net, "route");
2635 }
2636 
2637 #endif /* CONFIG_PROC_FS */
2638