xref: /openbmc/linux/net/ipv4/fib_trie.c (revision 615c36f5)
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 <asm/system.h>
55 #include <linux/bitops.h>
56 #include <linux/types.h>
57 #include <linux/kernel.h>
58 #include <linux/mm.h>
59 #include <linux/string.h>
60 #include <linux/socket.h>
61 #include <linux/sockios.h>
62 #include <linux/errno.h>
63 #include <linux/in.h>
64 #include <linux/inet.h>
65 #include <linux/inetdevice.h>
66 #include <linux/netdevice.h>
67 #include <linux/if_arp.h>
68 #include <linux/proc_fs.h>
69 #include <linux/rcupdate.h>
70 #include <linux/skbuff.h>
71 #include <linux/netlink.h>
72 #include <linux/init.h>
73 #include <linux/list.h>
74 #include <linux/slab.h>
75 #include <linux/prefetch.h>
76 #include <linux/export.h>
77 #include <net/net_namespace.h>
78 #include <net/ip.h>
79 #include <net/protocol.h>
80 #include <net/route.h>
81 #include <net/tcp.h>
82 #include <net/sock.h>
83 #include <net/ip_fib.h>
84 #include "fib_lookup.h"
85 
86 #define MAX_STAT_DEPTH 32
87 
88 #define KEYLENGTH (8*sizeof(t_key))
89 
90 typedef unsigned int t_key;
91 
92 #define T_TNODE 0
93 #define T_LEAF  1
94 #define NODE_TYPE_MASK	0x1UL
95 #define NODE_TYPE(node) ((node)->parent & NODE_TYPE_MASK)
96 
97 #define IS_TNODE(n) (!(n->parent & T_LEAF))
98 #define IS_LEAF(n) (n->parent & T_LEAF)
99 
100 struct rt_trie_node {
101 	unsigned long parent;
102 	t_key key;
103 };
104 
105 struct leaf {
106 	unsigned long parent;
107 	t_key key;
108 	struct hlist_head list;
109 	struct rcu_head rcu;
110 };
111 
112 struct leaf_info {
113 	struct hlist_node hlist;
114 	int plen;
115 	u32 mask_plen; /* ntohl(inet_make_mask(plen)) */
116 	struct list_head falh;
117 	struct rcu_head rcu;
118 };
119 
120 struct tnode {
121 	unsigned long parent;
122 	t_key key;
123 	unsigned char pos;		/* 2log(KEYLENGTH) bits needed */
124 	unsigned char bits;		/* 2log(KEYLENGTH) bits needed */
125 	unsigned int full_children;	/* KEYLENGTH bits needed */
126 	unsigned int empty_children;	/* KEYLENGTH bits needed */
127 	union {
128 		struct rcu_head rcu;
129 		struct work_struct work;
130 		struct tnode *tnode_free;
131 	};
132 	struct rt_trie_node __rcu *child[0];
133 };
134 
135 #ifdef CONFIG_IP_FIB_TRIE_STATS
136 struct trie_use_stats {
137 	unsigned int gets;
138 	unsigned int backtrack;
139 	unsigned int semantic_match_passed;
140 	unsigned int semantic_match_miss;
141 	unsigned int null_node_hit;
142 	unsigned int resize_node_skipped;
143 };
144 #endif
145 
146 struct trie_stat {
147 	unsigned int totdepth;
148 	unsigned int maxdepth;
149 	unsigned int tnodes;
150 	unsigned int leaves;
151 	unsigned int nullpointers;
152 	unsigned int prefixes;
153 	unsigned int nodesizes[MAX_STAT_DEPTH];
154 };
155 
156 struct trie {
157 	struct rt_trie_node __rcu *trie;
158 #ifdef CONFIG_IP_FIB_TRIE_STATS
159 	struct trie_use_stats stats;
160 #endif
161 };
162 
163 static void put_child(struct trie *t, struct tnode *tn, int i, struct rt_trie_node *n);
164 static void tnode_put_child_reorg(struct tnode *tn, int i, struct rt_trie_node *n,
165 				  int wasfull);
166 static struct rt_trie_node *resize(struct trie *t, struct tnode *tn);
167 static struct tnode *inflate(struct trie *t, struct tnode *tn);
168 static struct tnode *halve(struct trie *t, struct tnode *tn);
169 /* tnodes to free after resize(); protected by RTNL */
170 static struct tnode *tnode_free_head;
171 static size_t tnode_free_size;
172 
173 /*
174  * synchronize_rcu after call_rcu for that many pages; it should be especially
175  * useful before resizing the root node with PREEMPT_NONE configs; the value was
176  * obtained experimentally, aiming to avoid visible slowdown.
177  */
178 static const int sync_pages = 128;
179 
180 static struct kmem_cache *fn_alias_kmem __read_mostly;
181 static struct kmem_cache *trie_leaf_kmem __read_mostly;
182 
183 /*
184  * caller must hold RTNL
185  */
186 static inline struct tnode *node_parent(const struct rt_trie_node *node)
187 {
188 	unsigned long parent;
189 
190 	parent = rcu_dereference_index_check(node->parent, lockdep_rtnl_is_held());
191 
192 	return (struct tnode *)(parent & ~NODE_TYPE_MASK);
193 }
194 
195 /*
196  * caller must hold RCU read lock or RTNL
197  */
198 static inline struct tnode *node_parent_rcu(const struct rt_trie_node *node)
199 {
200 	unsigned long parent;
201 
202 	parent = rcu_dereference_index_check(node->parent, rcu_read_lock_held() ||
203 							   lockdep_rtnl_is_held());
204 
205 	return (struct tnode *)(parent & ~NODE_TYPE_MASK);
206 }
207 
208 /* Same as RCU_INIT_POINTER
209  * but that macro() assumes that value is a pointer.
210  */
211 static inline void node_set_parent(struct rt_trie_node *node, struct tnode *ptr)
212 {
213 	smp_wmb();
214 	node->parent = (unsigned long)ptr | NODE_TYPE(node);
215 }
216 
217 /*
218  * caller must hold RTNL
219  */
220 static inline struct rt_trie_node *tnode_get_child(const struct tnode *tn, unsigned int i)
221 {
222 	BUG_ON(i >= 1U << tn->bits);
223 
224 	return rtnl_dereference(tn->child[i]);
225 }
226 
227 /*
228  * caller must hold RCU read lock or RTNL
229  */
230 static inline struct rt_trie_node *tnode_get_child_rcu(const struct tnode *tn, unsigned int i)
231 {
232 	BUG_ON(i >= 1U << tn->bits);
233 
234 	return rcu_dereference_rtnl(tn->child[i]);
235 }
236 
237 static inline int tnode_child_length(const struct tnode *tn)
238 {
239 	return 1 << tn->bits;
240 }
241 
242 static inline t_key mask_pfx(t_key k, unsigned int l)
243 {
244 	return (l == 0) ? 0 : k >> (KEYLENGTH-l) << (KEYLENGTH-l);
245 }
246 
247 static inline t_key tkey_extract_bits(t_key a, unsigned int offset, unsigned int bits)
248 {
249 	if (offset < KEYLENGTH)
250 		return ((t_key)(a << offset)) >> (KEYLENGTH - bits);
251 	else
252 		return 0;
253 }
254 
255 static inline int tkey_equals(t_key a, t_key b)
256 {
257 	return a == b;
258 }
259 
260 static inline int tkey_sub_equals(t_key a, int offset, int bits, t_key b)
261 {
262 	if (bits == 0 || offset >= KEYLENGTH)
263 		return 1;
264 	bits = bits > KEYLENGTH ? KEYLENGTH : bits;
265 	return ((a ^ b) << offset) >> (KEYLENGTH - bits) == 0;
266 }
267 
268 static inline int tkey_mismatch(t_key a, int offset, t_key b)
269 {
270 	t_key diff = a ^ b;
271 	int i = offset;
272 
273 	if (!diff)
274 		return 0;
275 	while ((diff << i) >> (KEYLENGTH-1) == 0)
276 		i++;
277 	return i;
278 }
279 
280 /*
281   To understand this stuff, an understanding of keys and all their bits is
282   necessary. Every node in the trie has a key associated with it, but not
283   all of the bits in that key are significant.
284 
285   Consider a node 'n' and its parent 'tp'.
286 
287   If n is a leaf, every bit in its key is significant. Its presence is
288   necessitated by path compression, since during a tree traversal (when
289   searching for a leaf - unless we are doing an insertion) we will completely
290   ignore all skipped bits we encounter. Thus we need to verify, at the end of
291   a potentially successful search, that we have indeed been walking the
292   correct key path.
293 
294   Note that we can never "miss" the correct key in the tree if present by
295   following the wrong path. Path compression ensures that segments of the key
296   that are the same for all keys with a given prefix are skipped, but the
297   skipped part *is* identical for each node in the subtrie below the skipped
298   bit! trie_insert() in this implementation takes care of that - note the
299   call to tkey_sub_equals() in trie_insert().
300 
301   if n is an internal node - a 'tnode' here, the various parts of its key
302   have many different meanings.
303 
304   Example:
305   _________________________________________________________________
306   | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
307   -----------------------------------------------------------------
308     0   1   2   3   4   5   6   7   8   9  10  11  12  13  14  15
309 
310   _________________________________________________________________
311   | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
312   -----------------------------------------------------------------
313    16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31
314 
315   tp->pos = 7
316   tp->bits = 3
317   n->pos = 15
318   n->bits = 4
319 
320   First, let's just ignore the bits that come before the parent tp, that is
321   the bits from 0 to (tp->pos-1). They are *known* but at this point we do
322   not use them for anything.
323 
324   The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
325   index into the parent's child array. That is, they will be used to find
326   'n' among tp's children.
327 
328   The bits from (tp->pos + tp->bits) to (n->pos - 1) - "S" - are skipped bits
329   for the node n.
330 
331   All the bits we have seen so far are significant to the node n. The rest
332   of the bits are really not needed or indeed known in n->key.
333 
334   The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
335   n's child array, and will of course be different for each child.
336 
337 
338   The rest of the bits, from (n->pos + n->bits) onward, are completely unknown
339   at this point.
340 
341 */
342 
343 static inline void check_tnode(const struct tnode *tn)
344 {
345 	WARN_ON(tn && tn->pos+tn->bits > 32);
346 }
347 
348 static const int halve_threshold = 25;
349 static const int inflate_threshold = 50;
350 static const int halve_threshold_root = 15;
351 static const int inflate_threshold_root = 30;
352 
353 static void __alias_free_mem(struct rcu_head *head)
354 {
355 	struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
356 	kmem_cache_free(fn_alias_kmem, fa);
357 }
358 
359 static inline void alias_free_mem_rcu(struct fib_alias *fa)
360 {
361 	call_rcu(&fa->rcu, __alias_free_mem);
362 }
363 
364 static void __leaf_free_rcu(struct rcu_head *head)
365 {
366 	struct leaf *l = container_of(head, struct leaf, rcu);
367 	kmem_cache_free(trie_leaf_kmem, l);
368 }
369 
370 static inline void free_leaf(struct leaf *l)
371 {
372 	call_rcu_bh(&l->rcu, __leaf_free_rcu);
373 }
374 
375 static inline void free_leaf_info(struct leaf_info *leaf)
376 {
377 	kfree_rcu(leaf, rcu);
378 }
379 
380 static struct tnode *tnode_alloc(size_t size)
381 {
382 	if (size <= PAGE_SIZE)
383 		return kzalloc(size, GFP_KERNEL);
384 	else
385 		return vzalloc(size);
386 }
387 
388 static void __tnode_vfree(struct work_struct *arg)
389 {
390 	struct tnode *tn = container_of(arg, struct tnode, work);
391 	vfree(tn);
392 }
393 
394 static void __tnode_free_rcu(struct rcu_head *head)
395 {
396 	struct tnode *tn = container_of(head, struct tnode, rcu);
397 	size_t size = sizeof(struct tnode) +
398 		      (sizeof(struct rt_trie_node *) << tn->bits);
399 
400 	if (size <= PAGE_SIZE)
401 		kfree(tn);
402 	else {
403 		INIT_WORK(&tn->work, __tnode_vfree);
404 		schedule_work(&tn->work);
405 	}
406 }
407 
408 static inline void tnode_free(struct tnode *tn)
409 {
410 	if (IS_LEAF(tn))
411 		free_leaf((struct leaf *) tn);
412 	else
413 		call_rcu(&tn->rcu, __tnode_free_rcu);
414 }
415 
416 static void tnode_free_safe(struct tnode *tn)
417 {
418 	BUG_ON(IS_LEAF(tn));
419 	tn->tnode_free = tnode_free_head;
420 	tnode_free_head = tn;
421 	tnode_free_size += sizeof(struct tnode) +
422 			   (sizeof(struct rt_trie_node *) << tn->bits);
423 }
424 
425 static void tnode_free_flush(void)
426 {
427 	struct tnode *tn;
428 
429 	while ((tn = tnode_free_head)) {
430 		tnode_free_head = tn->tnode_free;
431 		tn->tnode_free = NULL;
432 		tnode_free(tn);
433 	}
434 
435 	if (tnode_free_size >= PAGE_SIZE * sync_pages) {
436 		tnode_free_size = 0;
437 		synchronize_rcu();
438 	}
439 }
440 
441 static struct leaf *leaf_new(void)
442 {
443 	struct leaf *l = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
444 	if (l) {
445 		l->parent = T_LEAF;
446 		INIT_HLIST_HEAD(&l->list);
447 	}
448 	return l;
449 }
450 
451 static struct leaf_info *leaf_info_new(int plen)
452 {
453 	struct leaf_info *li = kmalloc(sizeof(struct leaf_info),  GFP_KERNEL);
454 	if (li) {
455 		li->plen = plen;
456 		li->mask_plen = ntohl(inet_make_mask(plen));
457 		INIT_LIST_HEAD(&li->falh);
458 	}
459 	return li;
460 }
461 
462 static struct tnode *tnode_new(t_key key, int pos, int bits)
463 {
464 	size_t sz = sizeof(struct tnode) + (sizeof(struct rt_trie_node *) << bits);
465 	struct tnode *tn = tnode_alloc(sz);
466 
467 	if (tn) {
468 		tn->parent = T_TNODE;
469 		tn->pos = pos;
470 		tn->bits = bits;
471 		tn->key = key;
472 		tn->full_children = 0;
473 		tn->empty_children = 1<<bits;
474 	}
475 
476 	pr_debug("AT %p s=%zu %zu\n", tn, sizeof(struct tnode),
477 		 sizeof(struct rt_trie_node) << bits);
478 	return tn;
479 }
480 
481 /*
482  * Check whether a tnode 'n' is "full", i.e. it is an internal node
483  * and no bits are skipped. See discussion in dyntree paper p. 6
484  */
485 
486 static inline int tnode_full(const struct tnode *tn, const struct rt_trie_node *n)
487 {
488 	if (n == NULL || IS_LEAF(n))
489 		return 0;
490 
491 	return ((struct tnode *) n)->pos == tn->pos + tn->bits;
492 }
493 
494 static inline void put_child(struct trie *t, struct tnode *tn, int i,
495 			     struct rt_trie_node *n)
496 {
497 	tnode_put_child_reorg(tn, i, n, -1);
498 }
499 
500  /*
501   * Add a child at position i overwriting the old value.
502   * Update the value of full_children and empty_children.
503   */
504 
505 static void tnode_put_child_reorg(struct tnode *tn, int i, struct rt_trie_node *n,
506 				  int wasfull)
507 {
508 	struct rt_trie_node *chi = rtnl_dereference(tn->child[i]);
509 	int isfull;
510 
511 	BUG_ON(i >= 1<<tn->bits);
512 
513 	/* update emptyChildren */
514 	if (n == NULL && chi != NULL)
515 		tn->empty_children++;
516 	else if (n != NULL && chi == NULL)
517 		tn->empty_children--;
518 
519 	/* update fullChildren */
520 	if (wasfull == -1)
521 		wasfull = tnode_full(tn, chi);
522 
523 	isfull = tnode_full(tn, n);
524 	if (wasfull && !isfull)
525 		tn->full_children--;
526 	else if (!wasfull && isfull)
527 		tn->full_children++;
528 
529 	if (n)
530 		node_set_parent(n, tn);
531 
532 	RCU_INIT_POINTER(tn->child[i], n);
533 }
534 
535 #define MAX_WORK 10
536 static struct rt_trie_node *resize(struct trie *t, struct tnode *tn)
537 {
538 	int i;
539 	struct tnode *old_tn;
540 	int inflate_threshold_use;
541 	int halve_threshold_use;
542 	int max_work;
543 
544 	if (!tn)
545 		return NULL;
546 
547 	pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
548 		 tn, inflate_threshold, halve_threshold);
549 
550 	/* No children */
551 	if (tn->empty_children == tnode_child_length(tn)) {
552 		tnode_free_safe(tn);
553 		return NULL;
554 	}
555 	/* One child */
556 	if (tn->empty_children == tnode_child_length(tn) - 1)
557 		goto one_child;
558 	/*
559 	 * Double as long as the resulting node has a number of
560 	 * nonempty nodes that are above the threshold.
561 	 */
562 
563 	/*
564 	 * From "Implementing a dynamic compressed trie" by Stefan Nilsson of
565 	 * the Helsinki University of Technology and Matti Tikkanen of Nokia
566 	 * Telecommunications, page 6:
567 	 * "A node is doubled if the ratio of non-empty children to all
568 	 * children in the *doubled* node is at least 'high'."
569 	 *
570 	 * 'high' in this instance is the variable 'inflate_threshold'. It
571 	 * is expressed as a percentage, so we multiply it with
572 	 * tnode_child_length() and instead of multiplying by 2 (since the
573 	 * child array will be doubled by inflate()) and multiplying
574 	 * the left-hand side by 100 (to handle the percentage thing) we
575 	 * multiply the left-hand side by 50.
576 	 *
577 	 * The left-hand side may look a bit weird: tnode_child_length(tn)
578 	 * - tn->empty_children is of course the number of non-null children
579 	 * in the current node. tn->full_children is the number of "full"
580 	 * children, that is non-null tnodes with a skip value of 0.
581 	 * All of those will be doubled in the resulting inflated tnode, so
582 	 * we just count them one extra time here.
583 	 *
584 	 * A clearer way to write this would be:
585 	 *
586 	 * to_be_doubled = tn->full_children;
587 	 * not_to_be_doubled = tnode_child_length(tn) - tn->empty_children -
588 	 *     tn->full_children;
589 	 *
590 	 * new_child_length = tnode_child_length(tn) * 2;
591 	 *
592 	 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
593 	 *      new_child_length;
594 	 * if (new_fill_factor >= inflate_threshold)
595 	 *
596 	 * ...and so on, tho it would mess up the while () loop.
597 	 *
598 	 * anyway,
599 	 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
600 	 *      inflate_threshold
601 	 *
602 	 * avoid a division:
603 	 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
604 	 *      inflate_threshold * new_child_length
605 	 *
606 	 * expand not_to_be_doubled and to_be_doubled, and shorten:
607 	 * 100 * (tnode_child_length(tn) - tn->empty_children +
608 	 *    tn->full_children) >= inflate_threshold * new_child_length
609 	 *
610 	 * expand new_child_length:
611 	 * 100 * (tnode_child_length(tn) - tn->empty_children +
612 	 *    tn->full_children) >=
613 	 *      inflate_threshold * tnode_child_length(tn) * 2
614 	 *
615 	 * shorten again:
616 	 * 50 * (tn->full_children + tnode_child_length(tn) -
617 	 *    tn->empty_children) >= inflate_threshold *
618 	 *    tnode_child_length(tn)
619 	 *
620 	 */
621 
622 	check_tnode(tn);
623 
624 	/* Keep root node larger  */
625 
626 	if (!node_parent((struct rt_trie_node *)tn)) {
627 		inflate_threshold_use = inflate_threshold_root;
628 		halve_threshold_use = halve_threshold_root;
629 	} else {
630 		inflate_threshold_use = inflate_threshold;
631 		halve_threshold_use = halve_threshold;
632 	}
633 
634 	max_work = MAX_WORK;
635 	while ((tn->full_children > 0 &&  max_work-- &&
636 		50 * (tn->full_children + tnode_child_length(tn)
637 		      - tn->empty_children)
638 		>= inflate_threshold_use * tnode_child_length(tn))) {
639 
640 		old_tn = tn;
641 		tn = inflate(t, tn);
642 
643 		if (IS_ERR(tn)) {
644 			tn = old_tn;
645 #ifdef CONFIG_IP_FIB_TRIE_STATS
646 			t->stats.resize_node_skipped++;
647 #endif
648 			break;
649 		}
650 	}
651 
652 	check_tnode(tn);
653 
654 	/* Return if at least one inflate is run */
655 	if (max_work != MAX_WORK)
656 		return (struct rt_trie_node *) tn;
657 
658 	/*
659 	 * Halve as long as the number of empty children in this
660 	 * node is above threshold.
661 	 */
662 
663 	max_work = MAX_WORK;
664 	while (tn->bits > 1 &&  max_work-- &&
665 	       100 * (tnode_child_length(tn) - tn->empty_children) <
666 	       halve_threshold_use * tnode_child_length(tn)) {
667 
668 		old_tn = tn;
669 		tn = halve(t, tn);
670 		if (IS_ERR(tn)) {
671 			tn = old_tn;
672 #ifdef CONFIG_IP_FIB_TRIE_STATS
673 			t->stats.resize_node_skipped++;
674 #endif
675 			break;
676 		}
677 	}
678 
679 
680 	/* Only one child remains */
681 	if (tn->empty_children == tnode_child_length(tn) - 1) {
682 one_child:
683 		for (i = 0; i < tnode_child_length(tn); i++) {
684 			struct rt_trie_node *n;
685 
686 			n = rtnl_dereference(tn->child[i]);
687 			if (!n)
688 				continue;
689 
690 			/* compress one level */
691 
692 			node_set_parent(n, NULL);
693 			tnode_free_safe(tn);
694 			return n;
695 		}
696 	}
697 	return (struct rt_trie_node *) tn;
698 }
699 
700 
701 static void tnode_clean_free(struct tnode *tn)
702 {
703 	int i;
704 	struct tnode *tofree;
705 
706 	for (i = 0; i < tnode_child_length(tn); i++) {
707 		tofree = (struct tnode *)rtnl_dereference(tn->child[i]);
708 		if (tofree)
709 			tnode_free(tofree);
710 	}
711 	tnode_free(tn);
712 }
713 
714 static struct tnode *inflate(struct trie *t, struct tnode *tn)
715 {
716 	struct tnode *oldtnode = tn;
717 	int olen = tnode_child_length(tn);
718 	int i;
719 
720 	pr_debug("In inflate\n");
721 
722 	tn = tnode_new(oldtnode->key, oldtnode->pos, oldtnode->bits + 1);
723 
724 	if (!tn)
725 		return ERR_PTR(-ENOMEM);
726 
727 	/*
728 	 * Preallocate and store tnodes before the actual work so we
729 	 * don't get into an inconsistent state if memory allocation
730 	 * fails. In case of failure we return the oldnode and  inflate
731 	 * of tnode is ignored.
732 	 */
733 
734 	for (i = 0; i < olen; i++) {
735 		struct tnode *inode;
736 
737 		inode = (struct tnode *) tnode_get_child(oldtnode, i);
738 		if (inode &&
739 		    IS_TNODE(inode) &&
740 		    inode->pos == oldtnode->pos + oldtnode->bits &&
741 		    inode->bits > 1) {
742 			struct tnode *left, *right;
743 			t_key m = ~0U << (KEYLENGTH - 1) >> inode->pos;
744 
745 			left = tnode_new(inode->key&(~m), inode->pos + 1,
746 					 inode->bits - 1);
747 			if (!left)
748 				goto nomem;
749 
750 			right = tnode_new(inode->key|m, inode->pos + 1,
751 					  inode->bits - 1);
752 
753 			if (!right) {
754 				tnode_free(left);
755 				goto nomem;
756 			}
757 
758 			put_child(t, tn, 2*i, (struct rt_trie_node *) left);
759 			put_child(t, tn, 2*i+1, (struct rt_trie_node *) right);
760 		}
761 	}
762 
763 	for (i = 0; i < olen; i++) {
764 		struct tnode *inode;
765 		struct rt_trie_node *node = tnode_get_child(oldtnode, i);
766 		struct tnode *left, *right;
767 		int size, j;
768 
769 		/* An empty child */
770 		if (node == NULL)
771 			continue;
772 
773 		/* A leaf or an internal node with skipped bits */
774 
775 		if (IS_LEAF(node) || ((struct tnode *) node)->pos >
776 		   tn->pos + tn->bits - 1) {
777 			if (tkey_extract_bits(node->key,
778 					      oldtnode->pos + oldtnode->bits,
779 					      1) == 0)
780 				put_child(t, tn, 2*i, node);
781 			else
782 				put_child(t, tn, 2*i+1, node);
783 			continue;
784 		}
785 
786 		/* An internal node with two children */
787 		inode = (struct tnode *) node;
788 
789 		if (inode->bits == 1) {
790 			put_child(t, tn, 2*i, rtnl_dereference(inode->child[0]));
791 			put_child(t, tn, 2*i+1, rtnl_dereference(inode->child[1]));
792 
793 			tnode_free_safe(inode);
794 			continue;
795 		}
796 
797 		/* An internal node with more than two children */
798 
799 		/* We will replace this node 'inode' with two new
800 		 * ones, 'left' and 'right', each with half of the
801 		 * original children. The two new nodes will have
802 		 * a position one bit further down the key and this
803 		 * means that the "significant" part of their keys
804 		 * (see the discussion near the top of this file)
805 		 * will differ by one bit, which will be "0" in
806 		 * left's key and "1" in right's key. Since we are
807 		 * moving the key position by one step, the bit that
808 		 * we are moving away from - the bit at position
809 		 * (inode->pos) - is the one that will differ between
810 		 * left and right. So... we synthesize that bit in the
811 		 * two  new keys.
812 		 * The mask 'm' below will be a single "one" bit at
813 		 * the position (inode->pos)
814 		 */
815 
816 		/* Use the old key, but set the new significant
817 		 *   bit to zero.
818 		 */
819 
820 		left = (struct tnode *) tnode_get_child(tn, 2*i);
821 		put_child(t, tn, 2*i, NULL);
822 
823 		BUG_ON(!left);
824 
825 		right = (struct tnode *) tnode_get_child(tn, 2*i+1);
826 		put_child(t, tn, 2*i+1, NULL);
827 
828 		BUG_ON(!right);
829 
830 		size = tnode_child_length(left);
831 		for (j = 0; j < size; j++) {
832 			put_child(t, left, j, rtnl_dereference(inode->child[j]));
833 			put_child(t, right, j, rtnl_dereference(inode->child[j + size]));
834 		}
835 		put_child(t, tn, 2*i, resize(t, left));
836 		put_child(t, tn, 2*i+1, resize(t, right));
837 
838 		tnode_free_safe(inode);
839 	}
840 	tnode_free_safe(oldtnode);
841 	return tn;
842 nomem:
843 	tnode_clean_free(tn);
844 	return ERR_PTR(-ENOMEM);
845 }
846 
847 static struct tnode *halve(struct trie *t, struct tnode *tn)
848 {
849 	struct tnode *oldtnode = tn;
850 	struct rt_trie_node *left, *right;
851 	int i;
852 	int olen = tnode_child_length(tn);
853 
854 	pr_debug("In halve\n");
855 
856 	tn = tnode_new(oldtnode->key, oldtnode->pos, oldtnode->bits - 1);
857 
858 	if (!tn)
859 		return ERR_PTR(-ENOMEM);
860 
861 	/*
862 	 * Preallocate and store tnodes before the actual work so we
863 	 * don't get into an inconsistent state if memory allocation
864 	 * fails. In case of failure we return the oldnode and halve
865 	 * of tnode is ignored.
866 	 */
867 
868 	for (i = 0; i < olen; i += 2) {
869 		left = tnode_get_child(oldtnode, i);
870 		right = tnode_get_child(oldtnode, i+1);
871 
872 		/* Two nonempty children */
873 		if (left && right) {
874 			struct tnode *newn;
875 
876 			newn = tnode_new(left->key, tn->pos + tn->bits, 1);
877 
878 			if (!newn)
879 				goto nomem;
880 
881 			put_child(t, tn, i/2, (struct rt_trie_node *)newn);
882 		}
883 
884 	}
885 
886 	for (i = 0; i < olen; i += 2) {
887 		struct tnode *newBinNode;
888 
889 		left = tnode_get_child(oldtnode, i);
890 		right = tnode_get_child(oldtnode, i+1);
891 
892 		/* At least one of the children is empty */
893 		if (left == NULL) {
894 			if (right == NULL)    /* Both are empty */
895 				continue;
896 			put_child(t, tn, i/2, right);
897 			continue;
898 		}
899 
900 		if (right == NULL) {
901 			put_child(t, tn, i/2, left);
902 			continue;
903 		}
904 
905 		/* Two nonempty children */
906 		newBinNode = (struct tnode *) tnode_get_child(tn, i/2);
907 		put_child(t, tn, i/2, NULL);
908 		put_child(t, newBinNode, 0, left);
909 		put_child(t, newBinNode, 1, right);
910 		put_child(t, tn, i/2, resize(t, newBinNode));
911 	}
912 	tnode_free_safe(oldtnode);
913 	return tn;
914 nomem:
915 	tnode_clean_free(tn);
916 	return ERR_PTR(-ENOMEM);
917 }
918 
919 /* readside must use rcu_read_lock currently dump routines
920  via get_fa_head and dump */
921 
922 static struct leaf_info *find_leaf_info(struct leaf *l, int plen)
923 {
924 	struct hlist_head *head = &l->list;
925 	struct hlist_node *node;
926 	struct leaf_info *li;
927 
928 	hlist_for_each_entry_rcu(li, node, head, hlist)
929 		if (li->plen == plen)
930 			return li;
931 
932 	return NULL;
933 }
934 
935 static inline struct list_head *get_fa_head(struct leaf *l, int plen)
936 {
937 	struct leaf_info *li = find_leaf_info(l, plen);
938 
939 	if (!li)
940 		return NULL;
941 
942 	return &li->falh;
943 }
944 
945 static void insert_leaf_info(struct hlist_head *head, struct leaf_info *new)
946 {
947 	struct leaf_info *li = NULL, *last = NULL;
948 	struct hlist_node *node;
949 
950 	if (hlist_empty(head)) {
951 		hlist_add_head_rcu(&new->hlist, head);
952 	} else {
953 		hlist_for_each_entry(li, node, head, hlist) {
954 			if (new->plen > li->plen)
955 				break;
956 
957 			last = li;
958 		}
959 		if (last)
960 			hlist_add_after_rcu(&last->hlist, &new->hlist);
961 		else
962 			hlist_add_before_rcu(&new->hlist, &li->hlist);
963 	}
964 }
965 
966 /* rcu_read_lock needs to be hold by caller from readside */
967 
968 static struct leaf *
969 fib_find_node(struct trie *t, u32 key)
970 {
971 	int pos;
972 	struct tnode *tn;
973 	struct rt_trie_node *n;
974 
975 	pos = 0;
976 	n = rcu_dereference_rtnl(t->trie);
977 
978 	while (n != NULL &&  NODE_TYPE(n) == T_TNODE) {
979 		tn = (struct tnode *) n;
980 
981 		check_tnode(tn);
982 
983 		if (tkey_sub_equals(tn->key, pos, tn->pos-pos, key)) {
984 			pos = tn->pos + tn->bits;
985 			n = tnode_get_child_rcu(tn,
986 						tkey_extract_bits(key,
987 								  tn->pos,
988 								  tn->bits));
989 		} else
990 			break;
991 	}
992 	/* Case we have found a leaf. Compare prefixes */
993 
994 	if (n != NULL && IS_LEAF(n) && tkey_equals(key, n->key))
995 		return (struct leaf *)n;
996 
997 	return NULL;
998 }
999 
1000 static void trie_rebalance(struct trie *t, struct tnode *tn)
1001 {
1002 	int wasfull;
1003 	t_key cindex, key;
1004 	struct tnode *tp;
1005 
1006 	key = tn->key;
1007 
1008 	while (tn != NULL && (tp = node_parent((struct rt_trie_node *)tn)) != NULL) {
1009 		cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1010 		wasfull = tnode_full(tp, tnode_get_child(tp, cindex));
1011 		tn = (struct tnode *) resize(t, (struct tnode *)tn);
1012 
1013 		tnode_put_child_reorg((struct tnode *)tp, cindex,
1014 				      (struct rt_trie_node *)tn, wasfull);
1015 
1016 		tp = node_parent((struct rt_trie_node *) tn);
1017 		if (!tp)
1018 			RCU_INIT_POINTER(t->trie, (struct rt_trie_node *)tn);
1019 
1020 		tnode_free_flush();
1021 		if (!tp)
1022 			break;
1023 		tn = tp;
1024 	}
1025 
1026 	/* Handle last (top) tnode */
1027 	if (IS_TNODE(tn))
1028 		tn = (struct tnode *)resize(t, (struct tnode *)tn);
1029 
1030 	RCU_INIT_POINTER(t->trie, (struct rt_trie_node *)tn);
1031 	tnode_free_flush();
1032 }
1033 
1034 /* only used from updater-side */
1035 
1036 static struct list_head *fib_insert_node(struct trie *t, u32 key, int plen)
1037 {
1038 	int pos, newpos;
1039 	struct tnode *tp = NULL, *tn = NULL;
1040 	struct rt_trie_node *n;
1041 	struct leaf *l;
1042 	int missbit;
1043 	struct list_head *fa_head = NULL;
1044 	struct leaf_info *li;
1045 	t_key cindex;
1046 
1047 	pos = 0;
1048 	n = rtnl_dereference(t->trie);
1049 
1050 	/* If we point to NULL, stop. Either the tree is empty and we should
1051 	 * just put a new leaf in if, or we have reached an empty child slot,
1052 	 * and we should just put our new leaf in that.
1053 	 * If we point to a T_TNODE, check if it matches our key. Note that
1054 	 * a T_TNODE might be skipping any number of bits - its 'pos' need
1055 	 * not be the parent's 'pos'+'bits'!
1056 	 *
1057 	 * If it does match the current key, get pos/bits from it, extract
1058 	 * the index from our key, push the T_TNODE and walk the tree.
1059 	 *
1060 	 * If it doesn't, we have to replace it with a new T_TNODE.
1061 	 *
1062 	 * If we point to a T_LEAF, it might or might not have the same key
1063 	 * as we do. If it does, just change the value, update the T_LEAF's
1064 	 * value, and return it.
1065 	 * If it doesn't, we need to replace it with a T_TNODE.
1066 	 */
1067 
1068 	while (n != NULL &&  NODE_TYPE(n) == T_TNODE) {
1069 		tn = (struct tnode *) n;
1070 
1071 		check_tnode(tn);
1072 
1073 		if (tkey_sub_equals(tn->key, pos, tn->pos-pos, key)) {
1074 			tp = tn;
1075 			pos = tn->pos + tn->bits;
1076 			n = tnode_get_child(tn,
1077 					    tkey_extract_bits(key,
1078 							      tn->pos,
1079 							      tn->bits));
1080 
1081 			BUG_ON(n && node_parent(n) != tn);
1082 		} else
1083 			break;
1084 	}
1085 
1086 	/*
1087 	 * n  ----> NULL, LEAF or TNODE
1088 	 *
1089 	 * tp is n's (parent) ----> NULL or TNODE
1090 	 */
1091 
1092 	BUG_ON(tp && IS_LEAF(tp));
1093 
1094 	/* Case 1: n is a leaf. Compare prefixes */
1095 
1096 	if (n != NULL && IS_LEAF(n) && tkey_equals(key, n->key)) {
1097 		l = (struct leaf *) n;
1098 		li = leaf_info_new(plen);
1099 
1100 		if (!li)
1101 			return NULL;
1102 
1103 		fa_head = &li->falh;
1104 		insert_leaf_info(&l->list, li);
1105 		goto done;
1106 	}
1107 	l = leaf_new();
1108 
1109 	if (!l)
1110 		return NULL;
1111 
1112 	l->key = key;
1113 	li = leaf_info_new(plen);
1114 
1115 	if (!li) {
1116 		free_leaf(l);
1117 		return NULL;
1118 	}
1119 
1120 	fa_head = &li->falh;
1121 	insert_leaf_info(&l->list, li);
1122 
1123 	if (t->trie && n == NULL) {
1124 		/* Case 2: n is NULL, and will just insert a new leaf */
1125 
1126 		node_set_parent((struct rt_trie_node *)l, tp);
1127 
1128 		cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1129 		put_child(t, (struct tnode *)tp, cindex, (struct rt_trie_node *)l);
1130 	} else {
1131 		/* Case 3: n is a LEAF or a TNODE and the key doesn't match. */
1132 		/*
1133 		 *  Add a new tnode here
1134 		 *  first tnode need some special handling
1135 		 */
1136 
1137 		if (tp)
1138 			pos = tp->pos+tp->bits;
1139 		else
1140 			pos = 0;
1141 
1142 		if (n) {
1143 			newpos = tkey_mismatch(key, pos, n->key);
1144 			tn = tnode_new(n->key, newpos, 1);
1145 		} else {
1146 			newpos = 0;
1147 			tn = tnode_new(key, newpos, 1); /* First tnode */
1148 		}
1149 
1150 		if (!tn) {
1151 			free_leaf_info(li);
1152 			free_leaf(l);
1153 			return NULL;
1154 		}
1155 
1156 		node_set_parent((struct rt_trie_node *)tn, tp);
1157 
1158 		missbit = tkey_extract_bits(key, newpos, 1);
1159 		put_child(t, tn, missbit, (struct rt_trie_node *)l);
1160 		put_child(t, tn, 1-missbit, n);
1161 
1162 		if (tp) {
1163 			cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1164 			put_child(t, (struct tnode *)tp, cindex,
1165 				  (struct rt_trie_node *)tn);
1166 		} else {
1167 			RCU_INIT_POINTER(t->trie, (struct rt_trie_node *)tn);
1168 			tp = tn;
1169 		}
1170 	}
1171 
1172 	if (tp && tp->pos + tp->bits > 32)
1173 		pr_warning("fib_trie"
1174 			   " tp=%p pos=%d, bits=%d, key=%0x plen=%d\n",
1175 			   tp, tp->pos, tp->bits, key, plen);
1176 
1177 	/* Rebalance the trie */
1178 
1179 	trie_rebalance(t, tp);
1180 done:
1181 	return fa_head;
1182 }
1183 
1184 /*
1185  * Caller must hold RTNL.
1186  */
1187 int fib_table_insert(struct fib_table *tb, struct fib_config *cfg)
1188 {
1189 	struct trie *t = (struct trie *) tb->tb_data;
1190 	struct fib_alias *fa, *new_fa;
1191 	struct list_head *fa_head = NULL;
1192 	struct fib_info *fi;
1193 	int plen = cfg->fc_dst_len;
1194 	u8 tos = cfg->fc_tos;
1195 	u32 key, mask;
1196 	int err;
1197 	struct leaf *l;
1198 
1199 	if (plen > 32)
1200 		return -EINVAL;
1201 
1202 	key = ntohl(cfg->fc_dst);
1203 
1204 	pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1205 
1206 	mask = ntohl(inet_make_mask(plen));
1207 
1208 	if (key & ~mask)
1209 		return -EINVAL;
1210 
1211 	key = key & mask;
1212 
1213 	fi = fib_create_info(cfg);
1214 	if (IS_ERR(fi)) {
1215 		err = PTR_ERR(fi);
1216 		goto err;
1217 	}
1218 
1219 	l = fib_find_node(t, key);
1220 	fa = NULL;
1221 
1222 	if (l) {
1223 		fa_head = get_fa_head(l, plen);
1224 		fa = fib_find_alias(fa_head, tos, fi->fib_priority);
1225 	}
1226 
1227 	/* Now fa, if non-NULL, points to the first fib alias
1228 	 * with the same keys [prefix,tos,priority], if such key already
1229 	 * exists or to the node before which we will insert new one.
1230 	 *
1231 	 * If fa is NULL, we will need to allocate a new one and
1232 	 * insert to the head of f.
1233 	 *
1234 	 * If f is NULL, no fib node matched the destination key
1235 	 * and we need to allocate a new one of those as well.
1236 	 */
1237 
1238 	if (fa && fa->fa_tos == tos &&
1239 	    fa->fa_info->fib_priority == fi->fib_priority) {
1240 		struct fib_alias *fa_first, *fa_match;
1241 
1242 		err = -EEXIST;
1243 		if (cfg->fc_nlflags & NLM_F_EXCL)
1244 			goto out;
1245 
1246 		/* We have 2 goals:
1247 		 * 1. Find exact match for type, scope, fib_info to avoid
1248 		 * duplicate routes
1249 		 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1250 		 */
1251 		fa_match = NULL;
1252 		fa_first = fa;
1253 		fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
1254 		list_for_each_entry_continue(fa, fa_head, fa_list) {
1255 			if (fa->fa_tos != tos)
1256 				break;
1257 			if (fa->fa_info->fib_priority != fi->fib_priority)
1258 				break;
1259 			if (fa->fa_type == cfg->fc_type &&
1260 			    fa->fa_info == fi) {
1261 				fa_match = fa;
1262 				break;
1263 			}
1264 		}
1265 
1266 		if (cfg->fc_nlflags & NLM_F_REPLACE) {
1267 			struct fib_info *fi_drop;
1268 			u8 state;
1269 
1270 			fa = fa_first;
1271 			if (fa_match) {
1272 				if (fa == fa_match)
1273 					err = 0;
1274 				goto out;
1275 			}
1276 			err = -ENOBUFS;
1277 			new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1278 			if (new_fa == NULL)
1279 				goto out;
1280 
1281 			fi_drop = fa->fa_info;
1282 			new_fa->fa_tos = fa->fa_tos;
1283 			new_fa->fa_info = fi;
1284 			new_fa->fa_type = cfg->fc_type;
1285 			state = fa->fa_state;
1286 			new_fa->fa_state = state & ~FA_S_ACCESSED;
1287 
1288 			list_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1289 			alias_free_mem_rcu(fa);
1290 
1291 			fib_release_info(fi_drop);
1292 			if (state & FA_S_ACCESSED)
1293 				rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1294 			rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1295 				tb->tb_id, &cfg->fc_nlinfo, NLM_F_REPLACE);
1296 
1297 			goto succeeded;
1298 		}
1299 		/* Error if we find a perfect match which
1300 		 * uses the same scope, type, and nexthop
1301 		 * information.
1302 		 */
1303 		if (fa_match)
1304 			goto out;
1305 
1306 		if (!(cfg->fc_nlflags & NLM_F_APPEND))
1307 			fa = fa_first;
1308 	}
1309 	err = -ENOENT;
1310 	if (!(cfg->fc_nlflags & NLM_F_CREATE))
1311 		goto out;
1312 
1313 	err = -ENOBUFS;
1314 	new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1315 	if (new_fa == NULL)
1316 		goto out;
1317 
1318 	new_fa->fa_info = fi;
1319 	new_fa->fa_tos = tos;
1320 	new_fa->fa_type = cfg->fc_type;
1321 	new_fa->fa_state = 0;
1322 	/*
1323 	 * Insert new entry to the list.
1324 	 */
1325 
1326 	if (!fa_head) {
1327 		fa_head = fib_insert_node(t, key, plen);
1328 		if (unlikely(!fa_head)) {
1329 			err = -ENOMEM;
1330 			goto out_free_new_fa;
1331 		}
1332 	}
1333 
1334 	if (!plen)
1335 		tb->tb_num_default++;
1336 
1337 	list_add_tail_rcu(&new_fa->fa_list,
1338 			  (fa ? &fa->fa_list : fa_head));
1339 
1340 	rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1341 	rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, tb->tb_id,
1342 		  &cfg->fc_nlinfo, 0);
1343 succeeded:
1344 	return 0;
1345 
1346 out_free_new_fa:
1347 	kmem_cache_free(fn_alias_kmem, new_fa);
1348 out:
1349 	fib_release_info(fi);
1350 err:
1351 	return err;
1352 }
1353 
1354 /* should be called with rcu_read_lock */
1355 static int check_leaf(struct fib_table *tb, struct trie *t, struct leaf *l,
1356 		      t_key key,  const struct flowi4 *flp,
1357 		      struct fib_result *res, int fib_flags)
1358 {
1359 	struct leaf_info *li;
1360 	struct hlist_head *hhead = &l->list;
1361 	struct hlist_node *node;
1362 
1363 	hlist_for_each_entry_rcu(li, node, hhead, hlist) {
1364 		struct fib_alias *fa;
1365 
1366 		if (l->key != (key & li->mask_plen))
1367 			continue;
1368 
1369 		list_for_each_entry_rcu(fa, &li->falh, fa_list) {
1370 			struct fib_info *fi = fa->fa_info;
1371 			int nhsel, err;
1372 
1373 			if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1374 				continue;
1375 			if (fa->fa_info->fib_scope < flp->flowi4_scope)
1376 				continue;
1377 			fib_alias_accessed(fa);
1378 			err = fib_props[fa->fa_type].error;
1379 			if (err) {
1380 #ifdef CONFIG_IP_FIB_TRIE_STATS
1381 				t->stats.semantic_match_passed++;
1382 #endif
1383 				return err;
1384 			}
1385 			if (fi->fib_flags & RTNH_F_DEAD)
1386 				continue;
1387 			for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
1388 				const struct fib_nh *nh = &fi->fib_nh[nhsel];
1389 
1390 				if (nh->nh_flags & RTNH_F_DEAD)
1391 					continue;
1392 				if (flp->flowi4_oif && flp->flowi4_oif != nh->nh_oif)
1393 					continue;
1394 
1395 #ifdef CONFIG_IP_FIB_TRIE_STATS
1396 				t->stats.semantic_match_passed++;
1397 #endif
1398 				res->prefixlen = li->plen;
1399 				res->nh_sel = nhsel;
1400 				res->type = fa->fa_type;
1401 				res->scope = fa->fa_info->fib_scope;
1402 				res->fi = fi;
1403 				res->table = tb;
1404 				res->fa_head = &li->falh;
1405 				if (!(fib_flags & FIB_LOOKUP_NOREF))
1406 					atomic_inc(&fi->fib_clntref);
1407 				return 0;
1408 			}
1409 		}
1410 
1411 #ifdef CONFIG_IP_FIB_TRIE_STATS
1412 		t->stats.semantic_match_miss++;
1413 #endif
1414 	}
1415 
1416 	return 1;
1417 }
1418 
1419 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1420 		     struct fib_result *res, int fib_flags)
1421 {
1422 	struct trie *t = (struct trie *) tb->tb_data;
1423 	int ret;
1424 	struct rt_trie_node *n;
1425 	struct tnode *pn;
1426 	unsigned int pos, bits;
1427 	t_key key = ntohl(flp->daddr);
1428 	unsigned int chopped_off;
1429 	t_key cindex = 0;
1430 	unsigned int current_prefix_length = KEYLENGTH;
1431 	struct tnode *cn;
1432 	t_key pref_mismatch;
1433 
1434 	rcu_read_lock();
1435 
1436 	n = rcu_dereference(t->trie);
1437 	if (!n)
1438 		goto failed;
1439 
1440 #ifdef CONFIG_IP_FIB_TRIE_STATS
1441 	t->stats.gets++;
1442 #endif
1443 
1444 	/* Just a leaf? */
1445 	if (IS_LEAF(n)) {
1446 		ret = check_leaf(tb, t, (struct leaf *)n, key, flp, res, fib_flags);
1447 		goto found;
1448 	}
1449 
1450 	pn = (struct tnode *) n;
1451 	chopped_off = 0;
1452 
1453 	while (pn) {
1454 		pos = pn->pos;
1455 		bits = pn->bits;
1456 
1457 		if (!chopped_off)
1458 			cindex = tkey_extract_bits(mask_pfx(key, current_prefix_length),
1459 						   pos, bits);
1460 
1461 		n = tnode_get_child_rcu(pn, cindex);
1462 
1463 		if (n == NULL) {
1464 #ifdef CONFIG_IP_FIB_TRIE_STATS
1465 			t->stats.null_node_hit++;
1466 #endif
1467 			goto backtrace;
1468 		}
1469 
1470 		if (IS_LEAF(n)) {
1471 			ret = check_leaf(tb, t, (struct leaf *)n, key, flp, res, fib_flags);
1472 			if (ret > 0)
1473 				goto backtrace;
1474 			goto found;
1475 		}
1476 
1477 		cn = (struct tnode *)n;
1478 
1479 		/*
1480 		 * It's a tnode, and we can do some extra checks here if we
1481 		 * like, to avoid descending into a dead-end branch.
1482 		 * This tnode is in the parent's child array at index
1483 		 * key[p_pos..p_pos+p_bits] but potentially with some bits
1484 		 * chopped off, so in reality the index may be just a
1485 		 * subprefix, padded with zero at the end.
1486 		 * We can also take a look at any skipped bits in this
1487 		 * tnode - everything up to p_pos is supposed to be ok,
1488 		 * and the non-chopped bits of the index (se previous
1489 		 * paragraph) are also guaranteed ok, but the rest is
1490 		 * considered unknown.
1491 		 *
1492 		 * The skipped bits are key[pos+bits..cn->pos].
1493 		 */
1494 
1495 		/* If current_prefix_length < pos+bits, we are already doing
1496 		 * actual prefix  matching, which means everything from
1497 		 * pos+(bits-chopped_off) onward must be zero along some
1498 		 * branch of this subtree - otherwise there is *no* valid
1499 		 * prefix present. Here we can only check the skipped
1500 		 * bits. Remember, since we have already indexed into the
1501 		 * parent's child array, we know that the bits we chopped of
1502 		 * *are* zero.
1503 		 */
1504 
1505 		/* NOTA BENE: Checking only skipped bits
1506 		   for the new node here */
1507 
1508 		if (current_prefix_length < pos+bits) {
1509 			if (tkey_extract_bits(cn->key, current_prefix_length,
1510 						cn->pos - current_prefix_length)
1511 			    || !(cn->child[0]))
1512 				goto backtrace;
1513 		}
1514 
1515 		/*
1516 		 * If chopped_off=0, the index is fully validated and we
1517 		 * only need to look at the skipped bits for this, the new,
1518 		 * tnode. What we actually want to do is to find out if
1519 		 * these skipped bits match our key perfectly, or if we will
1520 		 * have to count on finding a matching prefix further down,
1521 		 * because if we do, we would like to have some way of
1522 		 * verifying the existence of such a prefix at this point.
1523 		 */
1524 
1525 		/* The only thing we can do at this point is to verify that
1526 		 * any such matching prefix can indeed be a prefix to our
1527 		 * key, and if the bits in the node we are inspecting that
1528 		 * do not match our key are not ZERO, this cannot be true.
1529 		 * Thus, find out where there is a mismatch (before cn->pos)
1530 		 * and verify that all the mismatching bits are zero in the
1531 		 * new tnode's key.
1532 		 */
1533 
1534 		/*
1535 		 * Note: We aren't very concerned about the piece of
1536 		 * the key that precede pn->pos+pn->bits, since these
1537 		 * have already been checked. The bits after cn->pos
1538 		 * aren't checked since these are by definition
1539 		 * "unknown" at this point. Thus, what we want to see
1540 		 * is if we are about to enter the "prefix matching"
1541 		 * state, and in that case verify that the skipped
1542 		 * bits that will prevail throughout this subtree are
1543 		 * zero, as they have to be if we are to find a
1544 		 * matching prefix.
1545 		 */
1546 
1547 		pref_mismatch = mask_pfx(cn->key ^ key, cn->pos);
1548 
1549 		/*
1550 		 * In short: If skipped bits in this node do not match
1551 		 * the search key, enter the "prefix matching"
1552 		 * state.directly.
1553 		 */
1554 		if (pref_mismatch) {
1555 			int mp = KEYLENGTH - fls(pref_mismatch);
1556 
1557 			if (tkey_extract_bits(cn->key, mp, cn->pos - mp) != 0)
1558 				goto backtrace;
1559 
1560 			if (current_prefix_length >= cn->pos)
1561 				current_prefix_length = mp;
1562 		}
1563 
1564 		pn = (struct tnode *)n; /* Descend */
1565 		chopped_off = 0;
1566 		continue;
1567 
1568 backtrace:
1569 		chopped_off++;
1570 
1571 		/* As zero don't change the child key (cindex) */
1572 		while ((chopped_off <= pn->bits)
1573 		       && !(cindex & (1<<(chopped_off-1))))
1574 			chopped_off++;
1575 
1576 		/* Decrease current_... with bits chopped off */
1577 		if (current_prefix_length > pn->pos + pn->bits - chopped_off)
1578 			current_prefix_length = pn->pos + pn->bits
1579 				- chopped_off;
1580 
1581 		/*
1582 		 * Either we do the actual chop off according or if we have
1583 		 * chopped off all bits in this tnode walk up to our parent.
1584 		 */
1585 
1586 		if (chopped_off <= pn->bits) {
1587 			cindex &= ~(1 << (chopped_off-1));
1588 		} else {
1589 			struct tnode *parent = node_parent_rcu((struct rt_trie_node *) pn);
1590 			if (!parent)
1591 				goto failed;
1592 
1593 			/* Get Child's index */
1594 			cindex = tkey_extract_bits(pn->key, parent->pos, parent->bits);
1595 			pn = parent;
1596 			chopped_off = 0;
1597 
1598 #ifdef CONFIG_IP_FIB_TRIE_STATS
1599 			t->stats.backtrack++;
1600 #endif
1601 			goto backtrace;
1602 		}
1603 	}
1604 failed:
1605 	ret = 1;
1606 found:
1607 	rcu_read_unlock();
1608 	return ret;
1609 }
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(t, (struct tnode *)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 	fa_head = get_fa_head(l, plen);
1660 	fa = fib_find_alias(fa_head, tos, 0);
1661 
1662 	if (!fa)
1663 		return -ESRCH;
1664 
1665 	pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1666 
1667 	fa_to_delete = NULL;
1668 	fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
1669 	list_for_each_entry_continue(fa, fa_head, fa_list) {
1670 		struct fib_info *fi = fa->fa_info;
1671 
1672 		if (fa->fa_tos != tos)
1673 			break;
1674 
1675 		if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1676 		    (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1677 		     fa->fa_info->fib_scope == cfg->fc_scope) &&
1678 		    (!cfg->fc_prefsrc ||
1679 		     fi->fib_prefsrc == cfg->fc_prefsrc) &&
1680 		    (!cfg->fc_protocol ||
1681 		     fi->fib_protocol == cfg->fc_protocol) &&
1682 		    fib_nh_match(cfg, fi) == 0) {
1683 			fa_to_delete = fa;
1684 			break;
1685 		}
1686 	}
1687 
1688 	if (!fa_to_delete)
1689 		return -ESRCH;
1690 
1691 	fa = fa_to_delete;
1692 	rtmsg_fib(RTM_DELROUTE, htonl(key), fa, plen, tb->tb_id,
1693 		  &cfg->fc_nlinfo, 0);
1694 
1695 	l = fib_find_node(t, key);
1696 	li = find_leaf_info(l, plen);
1697 
1698 	list_del_rcu(&fa->fa_list);
1699 
1700 	if (!plen)
1701 		tb->tb_num_default--;
1702 
1703 	if (list_empty(fa_head)) {
1704 		hlist_del_rcu(&li->hlist);
1705 		free_leaf_info(li);
1706 	}
1707 
1708 	if (hlist_empty(&l->list))
1709 		trie_leaf_remove(t, l);
1710 
1711 	if (fa->fa_state & FA_S_ACCESSED)
1712 		rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1713 
1714 	fib_release_info(fa->fa_info);
1715 	alias_free_mem_rcu(fa);
1716 	return 0;
1717 }
1718 
1719 static int trie_flush_list(struct list_head *head)
1720 {
1721 	struct fib_alias *fa, *fa_node;
1722 	int found = 0;
1723 
1724 	list_for_each_entry_safe(fa, fa_node, head, fa_list) {
1725 		struct fib_info *fi = fa->fa_info;
1726 
1727 		if (fi && (fi->fib_flags & RTNH_F_DEAD)) {
1728 			list_del_rcu(&fa->fa_list);
1729 			fib_release_info(fa->fa_info);
1730 			alias_free_mem_rcu(fa);
1731 			found++;
1732 		}
1733 	}
1734 	return found;
1735 }
1736 
1737 static int trie_flush_leaf(struct leaf *l)
1738 {
1739 	int found = 0;
1740 	struct hlist_head *lih = &l->list;
1741 	struct hlist_node *node, *tmp;
1742 	struct leaf_info *li = NULL;
1743 
1744 	hlist_for_each_entry_safe(li, node, tmp, lih, hlist) {
1745 		found += trie_flush_list(&li->falh);
1746 
1747 		if (list_empty(&li->falh)) {
1748 			hlist_del_rcu(&li->hlist);
1749 			free_leaf_info(li);
1750 		}
1751 	}
1752 	return found;
1753 }
1754 
1755 /*
1756  * Scan for the next right leaf starting at node p->child[idx]
1757  * Since we have back pointer, no recursion necessary.
1758  */
1759 static struct leaf *leaf_walk_rcu(struct tnode *p, struct rt_trie_node *c)
1760 {
1761 	do {
1762 		t_key idx;
1763 
1764 		if (c)
1765 			idx = tkey_extract_bits(c->key, p->pos, p->bits) + 1;
1766 		else
1767 			idx = 0;
1768 
1769 		while (idx < 1u << p->bits) {
1770 			c = tnode_get_child_rcu(p, idx++);
1771 			if (!c)
1772 				continue;
1773 
1774 			if (IS_LEAF(c)) {
1775 				prefetch(rcu_dereference_rtnl(p->child[idx]));
1776 				return (struct leaf *) c;
1777 			}
1778 
1779 			/* Rescan start scanning in new node */
1780 			p = (struct tnode *) c;
1781 			idx = 0;
1782 		}
1783 
1784 		/* Node empty, walk back up to parent */
1785 		c = (struct rt_trie_node *) p;
1786 	} while ((p = node_parent_rcu(c)) != NULL);
1787 
1788 	return NULL; /* Root of trie */
1789 }
1790 
1791 static struct leaf *trie_firstleaf(struct trie *t)
1792 {
1793 	struct tnode *n = (struct tnode *)rcu_dereference_rtnl(t->trie);
1794 
1795 	if (!n)
1796 		return NULL;
1797 
1798 	if (IS_LEAF(n))          /* trie is just a leaf */
1799 		return (struct leaf *) n;
1800 
1801 	return leaf_walk_rcu(n, NULL);
1802 }
1803 
1804 static struct leaf *trie_nextleaf(struct leaf *l)
1805 {
1806 	struct rt_trie_node *c = (struct rt_trie_node *) l;
1807 	struct tnode *p = node_parent_rcu(c);
1808 
1809 	if (!p)
1810 		return NULL;	/* trie with just one leaf */
1811 
1812 	return leaf_walk_rcu(p, c);
1813 }
1814 
1815 static struct leaf *trie_leafindex(struct trie *t, int index)
1816 {
1817 	struct leaf *l = trie_firstleaf(t);
1818 
1819 	while (l && index-- > 0)
1820 		l = trie_nextleaf(l);
1821 
1822 	return l;
1823 }
1824 
1825 
1826 /*
1827  * Caller must hold RTNL.
1828  */
1829 int fib_table_flush(struct fib_table *tb)
1830 {
1831 	struct trie *t = (struct trie *) tb->tb_data;
1832 	struct leaf *l, *ll = NULL;
1833 	int found = 0;
1834 
1835 	for (l = trie_firstleaf(t); l; l = trie_nextleaf(l)) {
1836 		found += trie_flush_leaf(l);
1837 
1838 		if (ll && hlist_empty(&ll->list))
1839 			trie_leaf_remove(t, ll);
1840 		ll = l;
1841 	}
1842 
1843 	if (ll && hlist_empty(&ll->list))
1844 		trie_leaf_remove(t, ll);
1845 
1846 	pr_debug("trie_flush found=%d\n", found);
1847 	return found;
1848 }
1849 
1850 void fib_free_table(struct fib_table *tb)
1851 {
1852 	kfree(tb);
1853 }
1854 
1855 static int fn_trie_dump_fa(t_key key, int plen, struct list_head *fah,
1856 			   struct fib_table *tb,
1857 			   struct sk_buff *skb, struct netlink_callback *cb)
1858 {
1859 	int i, s_i;
1860 	struct fib_alias *fa;
1861 	__be32 xkey = htonl(key);
1862 
1863 	s_i = cb->args[5];
1864 	i = 0;
1865 
1866 	/* rcu_read_lock is hold by caller */
1867 
1868 	list_for_each_entry_rcu(fa, fah, fa_list) {
1869 		if (i < s_i) {
1870 			i++;
1871 			continue;
1872 		}
1873 
1874 		if (fib_dump_info(skb, NETLINK_CB(cb->skb).pid,
1875 				  cb->nlh->nlmsg_seq,
1876 				  RTM_NEWROUTE,
1877 				  tb->tb_id,
1878 				  fa->fa_type,
1879 				  xkey,
1880 				  plen,
1881 				  fa->fa_tos,
1882 				  fa->fa_info, NLM_F_MULTI) < 0) {
1883 			cb->args[5] = i;
1884 			return -1;
1885 		}
1886 		i++;
1887 	}
1888 	cb->args[5] = i;
1889 	return skb->len;
1890 }
1891 
1892 static int fn_trie_dump_leaf(struct leaf *l, struct fib_table *tb,
1893 			struct sk_buff *skb, struct netlink_callback *cb)
1894 {
1895 	struct leaf_info *li;
1896 	struct hlist_node *node;
1897 	int i, s_i;
1898 
1899 	s_i = cb->args[4];
1900 	i = 0;
1901 
1902 	/* rcu_read_lock is hold by caller */
1903 	hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
1904 		if (i < s_i) {
1905 			i++;
1906 			continue;
1907 		}
1908 
1909 		if (i > s_i)
1910 			cb->args[5] = 0;
1911 
1912 		if (list_empty(&li->falh))
1913 			continue;
1914 
1915 		if (fn_trie_dump_fa(l->key, li->plen, &li->falh, tb, skb, cb) < 0) {
1916 			cb->args[4] = i;
1917 			return -1;
1918 		}
1919 		i++;
1920 	}
1921 
1922 	cb->args[4] = i;
1923 	return skb->len;
1924 }
1925 
1926 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
1927 		   struct netlink_callback *cb)
1928 {
1929 	struct leaf *l;
1930 	struct trie *t = (struct trie *) tb->tb_data;
1931 	t_key key = cb->args[2];
1932 	int count = cb->args[3];
1933 
1934 	rcu_read_lock();
1935 	/* Dump starting at last key.
1936 	 * Note: 0.0.0.0/0 (ie default) is first key.
1937 	 */
1938 	if (count == 0)
1939 		l = trie_firstleaf(t);
1940 	else {
1941 		/* Normally, continue from last key, but if that is missing
1942 		 * fallback to using slow rescan
1943 		 */
1944 		l = fib_find_node(t, key);
1945 		if (!l)
1946 			l = trie_leafindex(t, count);
1947 	}
1948 
1949 	while (l) {
1950 		cb->args[2] = l->key;
1951 		if (fn_trie_dump_leaf(l, tb, skb, cb) < 0) {
1952 			cb->args[3] = count;
1953 			rcu_read_unlock();
1954 			return -1;
1955 		}
1956 
1957 		++count;
1958 		l = trie_nextleaf(l);
1959 		memset(&cb->args[4], 0,
1960 		       sizeof(cb->args) - 4*sizeof(cb->args[0]));
1961 	}
1962 	cb->args[3] = count;
1963 	rcu_read_unlock();
1964 
1965 	return skb->len;
1966 }
1967 
1968 void __init fib_trie_init(void)
1969 {
1970 	fn_alias_kmem = kmem_cache_create("ip_fib_alias",
1971 					  sizeof(struct fib_alias),
1972 					  0, SLAB_PANIC, NULL);
1973 
1974 	trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
1975 					   max(sizeof(struct leaf),
1976 					       sizeof(struct leaf_info)),
1977 					   0, SLAB_PANIC, NULL);
1978 }
1979 
1980 
1981 struct fib_table *fib_trie_table(u32 id)
1982 {
1983 	struct fib_table *tb;
1984 	struct trie *t;
1985 
1986 	tb = kmalloc(sizeof(struct fib_table) + sizeof(struct trie),
1987 		     GFP_KERNEL);
1988 	if (tb == NULL)
1989 		return NULL;
1990 
1991 	tb->tb_id = id;
1992 	tb->tb_default = -1;
1993 	tb->tb_num_default = 0;
1994 
1995 	t = (struct trie *) tb->tb_data;
1996 	memset(t, 0, sizeof(*t));
1997 
1998 	return tb;
1999 }
2000 
2001 #ifdef CONFIG_PROC_FS
2002 /* Depth first Trie walk iterator */
2003 struct fib_trie_iter {
2004 	struct seq_net_private p;
2005 	struct fib_table *tb;
2006 	struct tnode *tnode;
2007 	unsigned int index;
2008 	unsigned int depth;
2009 };
2010 
2011 static struct rt_trie_node *fib_trie_get_next(struct fib_trie_iter *iter)
2012 {
2013 	struct tnode *tn = iter->tnode;
2014 	unsigned int cindex = iter->index;
2015 	struct tnode *p;
2016 
2017 	/* A single entry routing table */
2018 	if (!tn)
2019 		return NULL;
2020 
2021 	pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2022 		 iter->tnode, iter->index, iter->depth);
2023 rescan:
2024 	while (cindex < (1<<tn->bits)) {
2025 		struct rt_trie_node *n = tnode_get_child_rcu(tn, cindex);
2026 
2027 		if (n) {
2028 			if (IS_LEAF(n)) {
2029 				iter->tnode = tn;
2030 				iter->index = cindex + 1;
2031 			} else {
2032 				/* push down one level */
2033 				iter->tnode = (struct tnode *) n;
2034 				iter->index = 0;
2035 				++iter->depth;
2036 			}
2037 			return n;
2038 		}
2039 
2040 		++cindex;
2041 	}
2042 
2043 	/* Current node exhausted, pop back up */
2044 	p = node_parent_rcu((struct rt_trie_node *)tn);
2045 	if (p) {
2046 		cindex = tkey_extract_bits(tn->key, p->pos, p->bits)+1;
2047 		tn = p;
2048 		--iter->depth;
2049 		goto rescan;
2050 	}
2051 
2052 	/* got root? */
2053 	return NULL;
2054 }
2055 
2056 static struct rt_trie_node *fib_trie_get_first(struct fib_trie_iter *iter,
2057 				       struct trie *t)
2058 {
2059 	struct rt_trie_node *n;
2060 
2061 	if (!t)
2062 		return NULL;
2063 
2064 	n = rcu_dereference(t->trie);
2065 	if (!n)
2066 		return NULL;
2067 
2068 	if (IS_TNODE(n)) {
2069 		iter->tnode = (struct tnode *) n;
2070 		iter->index = 0;
2071 		iter->depth = 1;
2072 	} else {
2073 		iter->tnode = NULL;
2074 		iter->index = 0;
2075 		iter->depth = 0;
2076 	}
2077 
2078 	return n;
2079 }
2080 
2081 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2082 {
2083 	struct rt_trie_node *n;
2084 	struct fib_trie_iter iter;
2085 
2086 	memset(s, 0, sizeof(*s));
2087 
2088 	rcu_read_lock();
2089 	for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2090 		if (IS_LEAF(n)) {
2091 			struct leaf *l = (struct leaf *)n;
2092 			struct leaf_info *li;
2093 			struct hlist_node *tmp;
2094 
2095 			s->leaves++;
2096 			s->totdepth += iter.depth;
2097 			if (iter.depth > s->maxdepth)
2098 				s->maxdepth = iter.depth;
2099 
2100 			hlist_for_each_entry_rcu(li, tmp, &l->list, hlist)
2101 				++s->prefixes;
2102 		} else {
2103 			const struct tnode *tn = (const struct tnode *) n;
2104 			int i;
2105 
2106 			s->tnodes++;
2107 			if (tn->bits < MAX_STAT_DEPTH)
2108 				s->nodesizes[tn->bits]++;
2109 
2110 			for (i = 0; i < (1<<tn->bits); i++)
2111 				if (!tn->child[i])
2112 					s->nullpointers++;
2113 		}
2114 	}
2115 	rcu_read_unlock();
2116 }
2117 
2118 /*
2119  *	This outputs /proc/net/fib_triestats
2120  */
2121 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2122 {
2123 	unsigned int i, max, pointers, bytes, avdepth;
2124 
2125 	if (stat->leaves)
2126 		avdepth = stat->totdepth*100 / stat->leaves;
2127 	else
2128 		avdepth = 0;
2129 
2130 	seq_printf(seq, "\tAver depth:     %u.%02d\n",
2131 		   avdepth / 100, avdepth % 100);
2132 	seq_printf(seq, "\tMax depth:      %u\n", stat->maxdepth);
2133 
2134 	seq_printf(seq, "\tLeaves:         %u\n", stat->leaves);
2135 	bytes = sizeof(struct leaf) * stat->leaves;
2136 
2137 	seq_printf(seq, "\tPrefixes:       %u\n", stat->prefixes);
2138 	bytes += sizeof(struct leaf_info) * stat->prefixes;
2139 
2140 	seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2141 	bytes += sizeof(struct tnode) * stat->tnodes;
2142 
2143 	max = MAX_STAT_DEPTH;
2144 	while (max > 0 && stat->nodesizes[max-1] == 0)
2145 		max--;
2146 
2147 	pointers = 0;
2148 	for (i = 1; i <= max; i++)
2149 		if (stat->nodesizes[i] != 0) {
2150 			seq_printf(seq, "  %u: %u",  i, stat->nodesizes[i]);
2151 			pointers += (1<<i) * stat->nodesizes[i];
2152 		}
2153 	seq_putc(seq, '\n');
2154 	seq_printf(seq, "\tPointers: %u\n", pointers);
2155 
2156 	bytes += sizeof(struct rt_trie_node *) * pointers;
2157 	seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2158 	seq_printf(seq, "Total size: %u  kB\n", (bytes + 1023) / 1024);
2159 }
2160 
2161 #ifdef CONFIG_IP_FIB_TRIE_STATS
2162 static void trie_show_usage(struct seq_file *seq,
2163 			    const struct trie_use_stats *stats)
2164 {
2165 	seq_printf(seq, "\nCounters:\n---------\n");
2166 	seq_printf(seq, "gets = %u\n", stats->gets);
2167 	seq_printf(seq, "backtracks = %u\n", stats->backtrack);
2168 	seq_printf(seq, "semantic match passed = %u\n",
2169 		   stats->semantic_match_passed);
2170 	seq_printf(seq, "semantic match miss = %u\n",
2171 		   stats->semantic_match_miss);
2172 	seq_printf(seq, "null node hit= %u\n", stats->null_node_hit);
2173 	seq_printf(seq, "skipped node resize = %u\n\n",
2174 		   stats->resize_node_skipped);
2175 }
2176 #endif /*  CONFIG_IP_FIB_TRIE_STATS */
2177 
2178 static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2179 {
2180 	if (tb->tb_id == RT_TABLE_LOCAL)
2181 		seq_puts(seq, "Local:\n");
2182 	else if (tb->tb_id == RT_TABLE_MAIN)
2183 		seq_puts(seq, "Main:\n");
2184 	else
2185 		seq_printf(seq, "Id %d:\n", tb->tb_id);
2186 }
2187 
2188 
2189 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2190 {
2191 	struct net *net = (struct net *)seq->private;
2192 	unsigned int h;
2193 
2194 	seq_printf(seq,
2195 		   "Basic info: size of leaf:"
2196 		   " %Zd bytes, size of tnode: %Zd bytes.\n",
2197 		   sizeof(struct leaf), sizeof(struct tnode));
2198 
2199 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2200 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2201 		struct hlist_node *node;
2202 		struct fib_table *tb;
2203 
2204 		hlist_for_each_entry_rcu(tb, node, head, tb_hlist) {
2205 			struct trie *t = (struct trie *) tb->tb_data;
2206 			struct trie_stat stat;
2207 
2208 			if (!t)
2209 				continue;
2210 
2211 			fib_table_print(seq, tb);
2212 
2213 			trie_collect_stats(t, &stat);
2214 			trie_show_stats(seq, &stat);
2215 #ifdef CONFIG_IP_FIB_TRIE_STATS
2216 			trie_show_usage(seq, &t->stats);
2217 #endif
2218 		}
2219 	}
2220 
2221 	return 0;
2222 }
2223 
2224 static int fib_triestat_seq_open(struct inode *inode, struct file *file)
2225 {
2226 	return single_open_net(inode, file, fib_triestat_seq_show);
2227 }
2228 
2229 static const struct file_operations fib_triestat_fops = {
2230 	.owner	= THIS_MODULE,
2231 	.open	= fib_triestat_seq_open,
2232 	.read	= seq_read,
2233 	.llseek	= seq_lseek,
2234 	.release = single_release_net,
2235 };
2236 
2237 static struct rt_trie_node *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2238 {
2239 	struct fib_trie_iter *iter = seq->private;
2240 	struct net *net = seq_file_net(seq);
2241 	loff_t idx = 0;
2242 	unsigned int h;
2243 
2244 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2245 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2246 		struct hlist_node *node;
2247 		struct fib_table *tb;
2248 
2249 		hlist_for_each_entry_rcu(tb, node, head, tb_hlist) {
2250 			struct rt_trie_node *n;
2251 
2252 			for (n = fib_trie_get_first(iter,
2253 						    (struct trie *) tb->tb_data);
2254 			     n; n = fib_trie_get_next(iter))
2255 				if (pos == idx++) {
2256 					iter->tb = tb;
2257 					return n;
2258 				}
2259 		}
2260 	}
2261 
2262 	return NULL;
2263 }
2264 
2265 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2266 	__acquires(RCU)
2267 {
2268 	rcu_read_lock();
2269 	return fib_trie_get_idx(seq, *pos);
2270 }
2271 
2272 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2273 {
2274 	struct fib_trie_iter *iter = seq->private;
2275 	struct net *net = seq_file_net(seq);
2276 	struct fib_table *tb = iter->tb;
2277 	struct hlist_node *tb_node;
2278 	unsigned int h;
2279 	struct rt_trie_node *n;
2280 
2281 	++*pos;
2282 	/* next node in same table */
2283 	n = fib_trie_get_next(iter);
2284 	if (n)
2285 		return n;
2286 
2287 	/* walk rest of this hash chain */
2288 	h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2289 	while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2290 		tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2291 		n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2292 		if (n)
2293 			goto found;
2294 	}
2295 
2296 	/* new hash chain */
2297 	while (++h < FIB_TABLE_HASHSZ) {
2298 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2299 		hlist_for_each_entry_rcu(tb, tb_node, head, tb_hlist) {
2300 			n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2301 			if (n)
2302 				goto found;
2303 		}
2304 	}
2305 	return NULL;
2306 
2307 found:
2308 	iter->tb = tb;
2309 	return n;
2310 }
2311 
2312 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2313 	__releases(RCU)
2314 {
2315 	rcu_read_unlock();
2316 }
2317 
2318 static void seq_indent(struct seq_file *seq, int n)
2319 {
2320 	while (n-- > 0)
2321 		seq_puts(seq, "   ");
2322 }
2323 
2324 static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2325 {
2326 	switch (s) {
2327 	case RT_SCOPE_UNIVERSE: return "universe";
2328 	case RT_SCOPE_SITE:	return "site";
2329 	case RT_SCOPE_LINK:	return "link";
2330 	case RT_SCOPE_HOST:	return "host";
2331 	case RT_SCOPE_NOWHERE:	return "nowhere";
2332 	default:
2333 		snprintf(buf, len, "scope=%d", s);
2334 		return buf;
2335 	}
2336 }
2337 
2338 static const char *const rtn_type_names[__RTN_MAX] = {
2339 	[RTN_UNSPEC] = "UNSPEC",
2340 	[RTN_UNICAST] = "UNICAST",
2341 	[RTN_LOCAL] = "LOCAL",
2342 	[RTN_BROADCAST] = "BROADCAST",
2343 	[RTN_ANYCAST] = "ANYCAST",
2344 	[RTN_MULTICAST] = "MULTICAST",
2345 	[RTN_BLACKHOLE] = "BLACKHOLE",
2346 	[RTN_UNREACHABLE] = "UNREACHABLE",
2347 	[RTN_PROHIBIT] = "PROHIBIT",
2348 	[RTN_THROW] = "THROW",
2349 	[RTN_NAT] = "NAT",
2350 	[RTN_XRESOLVE] = "XRESOLVE",
2351 };
2352 
2353 static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2354 {
2355 	if (t < __RTN_MAX && rtn_type_names[t])
2356 		return rtn_type_names[t];
2357 	snprintf(buf, len, "type %u", t);
2358 	return buf;
2359 }
2360 
2361 /* Pretty print the trie */
2362 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2363 {
2364 	const struct fib_trie_iter *iter = seq->private;
2365 	struct rt_trie_node *n = v;
2366 
2367 	if (!node_parent_rcu(n))
2368 		fib_table_print(seq, iter->tb);
2369 
2370 	if (IS_TNODE(n)) {
2371 		struct tnode *tn = (struct tnode *) n;
2372 		__be32 prf = htonl(mask_pfx(tn->key, tn->pos));
2373 
2374 		seq_indent(seq, iter->depth-1);
2375 		seq_printf(seq, "  +-- %pI4/%d %d %d %d\n",
2376 			   &prf, tn->pos, tn->bits, tn->full_children,
2377 			   tn->empty_children);
2378 
2379 	} else {
2380 		struct leaf *l = (struct leaf *) n;
2381 		struct leaf_info *li;
2382 		struct hlist_node *node;
2383 		__be32 val = htonl(l->key);
2384 
2385 		seq_indent(seq, iter->depth);
2386 		seq_printf(seq, "  |-- %pI4\n", &val);
2387 
2388 		hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
2389 			struct fib_alias *fa;
2390 
2391 			list_for_each_entry_rcu(fa, &li->falh, fa_list) {
2392 				char buf1[32], buf2[32];
2393 
2394 				seq_indent(seq, iter->depth+1);
2395 				seq_printf(seq, "  /%d %s %s", li->plen,
2396 					   rtn_scope(buf1, sizeof(buf1),
2397 						     fa->fa_info->fib_scope),
2398 					   rtn_type(buf2, sizeof(buf2),
2399 						    fa->fa_type));
2400 				if (fa->fa_tos)
2401 					seq_printf(seq, " tos=%d", fa->fa_tos);
2402 				seq_putc(seq, '\n');
2403 			}
2404 		}
2405 	}
2406 
2407 	return 0;
2408 }
2409 
2410 static const struct seq_operations fib_trie_seq_ops = {
2411 	.start  = fib_trie_seq_start,
2412 	.next   = fib_trie_seq_next,
2413 	.stop   = fib_trie_seq_stop,
2414 	.show   = fib_trie_seq_show,
2415 };
2416 
2417 static int fib_trie_seq_open(struct inode *inode, struct file *file)
2418 {
2419 	return seq_open_net(inode, file, &fib_trie_seq_ops,
2420 			    sizeof(struct fib_trie_iter));
2421 }
2422 
2423 static const struct file_operations fib_trie_fops = {
2424 	.owner  = THIS_MODULE,
2425 	.open   = fib_trie_seq_open,
2426 	.read   = seq_read,
2427 	.llseek = seq_lseek,
2428 	.release = seq_release_net,
2429 };
2430 
2431 struct fib_route_iter {
2432 	struct seq_net_private p;
2433 	struct trie *main_trie;
2434 	loff_t	pos;
2435 	t_key	key;
2436 };
2437 
2438 static struct leaf *fib_route_get_idx(struct fib_route_iter *iter, loff_t pos)
2439 {
2440 	struct leaf *l = NULL;
2441 	struct trie *t = iter->main_trie;
2442 
2443 	/* use cache location of last found key */
2444 	if (iter->pos > 0 && pos >= iter->pos && (l = fib_find_node(t, iter->key)))
2445 		pos -= iter->pos;
2446 	else {
2447 		iter->pos = 0;
2448 		l = trie_firstleaf(t);
2449 	}
2450 
2451 	while (l && pos-- > 0) {
2452 		iter->pos++;
2453 		l = trie_nextleaf(l);
2454 	}
2455 
2456 	if (l)
2457 		iter->key = pos;	/* remember it */
2458 	else
2459 		iter->pos = 0;		/* forget it */
2460 
2461 	return l;
2462 }
2463 
2464 static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2465 	__acquires(RCU)
2466 {
2467 	struct fib_route_iter *iter = seq->private;
2468 	struct fib_table *tb;
2469 
2470 	rcu_read_lock();
2471 	tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2472 	if (!tb)
2473 		return NULL;
2474 
2475 	iter->main_trie = (struct trie *) tb->tb_data;
2476 	if (*pos == 0)
2477 		return SEQ_START_TOKEN;
2478 	else
2479 		return fib_route_get_idx(iter, *pos - 1);
2480 }
2481 
2482 static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2483 {
2484 	struct fib_route_iter *iter = seq->private;
2485 	struct leaf *l = v;
2486 
2487 	++*pos;
2488 	if (v == SEQ_START_TOKEN) {
2489 		iter->pos = 0;
2490 		l = trie_firstleaf(iter->main_trie);
2491 	} else {
2492 		iter->pos++;
2493 		l = trie_nextleaf(l);
2494 	}
2495 
2496 	if (l)
2497 		iter->key = l->key;
2498 	else
2499 		iter->pos = 0;
2500 	return l;
2501 }
2502 
2503 static void fib_route_seq_stop(struct seq_file *seq, void *v)
2504 	__releases(RCU)
2505 {
2506 	rcu_read_unlock();
2507 }
2508 
2509 static unsigned int fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2510 {
2511 	unsigned int flags = 0;
2512 
2513 	if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2514 		flags = RTF_REJECT;
2515 	if (fi && fi->fib_nh->nh_gw)
2516 		flags |= RTF_GATEWAY;
2517 	if (mask == htonl(0xFFFFFFFF))
2518 		flags |= RTF_HOST;
2519 	flags |= RTF_UP;
2520 	return flags;
2521 }
2522 
2523 /*
2524  *	This outputs /proc/net/route.
2525  *	The format of the file is not supposed to be changed
2526  *	and needs to be same as fib_hash output to avoid breaking
2527  *	legacy utilities
2528  */
2529 static int fib_route_seq_show(struct seq_file *seq, void *v)
2530 {
2531 	struct leaf *l = v;
2532 	struct leaf_info *li;
2533 	struct hlist_node *node;
2534 
2535 	if (v == SEQ_START_TOKEN) {
2536 		seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2537 			   "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2538 			   "\tWindow\tIRTT");
2539 		return 0;
2540 	}
2541 
2542 	hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
2543 		struct fib_alias *fa;
2544 		__be32 mask, prefix;
2545 
2546 		mask = inet_make_mask(li->plen);
2547 		prefix = htonl(l->key);
2548 
2549 		list_for_each_entry_rcu(fa, &li->falh, fa_list) {
2550 			const struct fib_info *fi = fa->fa_info;
2551 			unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2552 			int len;
2553 
2554 			if (fa->fa_type == RTN_BROADCAST
2555 			    || fa->fa_type == RTN_MULTICAST)
2556 				continue;
2557 
2558 			if (fi)
2559 				seq_printf(seq,
2560 					 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2561 					 "%d\t%08X\t%d\t%u\t%u%n",
2562 					 fi->fib_dev ? fi->fib_dev->name : "*",
2563 					 prefix,
2564 					 fi->fib_nh->nh_gw, flags, 0, 0,
2565 					 fi->fib_priority,
2566 					 mask,
2567 					 (fi->fib_advmss ?
2568 					  fi->fib_advmss + 40 : 0),
2569 					 fi->fib_window,
2570 					 fi->fib_rtt >> 3, &len);
2571 			else
2572 				seq_printf(seq,
2573 					 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2574 					 "%d\t%08X\t%d\t%u\t%u%n",
2575 					 prefix, 0, flags, 0, 0, 0,
2576 					 mask, 0, 0, 0, &len);
2577 
2578 			seq_printf(seq, "%*s\n", 127 - len, "");
2579 		}
2580 	}
2581 
2582 	return 0;
2583 }
2584 
2585 static const struct seq_operations fib_route_seq_ops = {
2586 	.start  = fib_route_seq_start,
2587 	.next   = fib_route_seq_next,
2588 	.stop   = fib_route_seq_stop,
2589 	.show   = fib_route_seq_show,
2590 };
2591 
2592 static int fib_route_seq_open(struct inode *inode, struct file *file)
2593 {
2594 	return seq_open_net(inode, file, &fib_route_seq_ops,
2595 			    sizeof(struct fib_route_iter));
2596 }
2597 
2598 static const struct file_operations fib_route_fops = {
2599 	.owner  = THIS_MODULE,
2600 	.open   = fib_route_seq_open,
2601 	.read   = seq_read,
2602 	.llseek = seq_lseek,
2603 	.release = seq_release_net,
2604 };
2605 
2606 int __net_init fib_proc_init(struct net *net)
2607 {
2608 	if (!proc_net_fops_create(net, "fib_trie", S_IRUGO, &fib_trie_fops))
2609 		goto out1;
2610 
2611 	if (!proc_net_fops_create(net, "fib_triestat", S_IRUGO,
2612 				  &fib_triestat_fops))
2613 		goto out2;
2614 
2615 	if (!proc_net_fops_create(net, "route", S_IRUGO, &fib_route_fops))
2616 		goto out3;
2617 
2618 	return 0;
2619 
2620 out3:
2621 	proc_net_remove(net, "fib_triestat");
2622 out2:
2623 	proc_net_remove(net, "fib_trie");
2624 out1:
2625 	return -ENOMEM;
2626 }
2627 
2628 void __net_exit fib_proc_exit(struct net *net)
2629 {
2630 	proc_net_remove(net, "fib_trie");
2631 	proc_net_remove(net, "fib_triestat");
2632 	proc_net_remove(net, "route");
2633 }
2634 
2635 #endif /* CONFIG_PROC_FS */
2636