xref: /openbmc/linux/net/ipv4/fib_trie.c (revision 75f25bd3)
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 <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 put_child(struct trie *t, struct tnode *tn, int i, struct rt_trie_node *n);
163 static void tnode_put_child_reorg(struct tnode *tn, int i, struct rt_trie_node *n,
164 				  int wasfull);
165 static struct rt_trie_node *resize(struct trie *t, struct tnode *tn);
166 static struct tnode *inflate(struct trie *t, struct tnode *tn);
167 static struct tnode *halve(struct trie *t, struct tnode *tn);
168 /* tnodes to free after resize(); protected by RTNL */
169 static struct tnode *tnode_free_head;
170 static size_t tnode_free_size;
171 
172 /*
173  * synchronize_rcu after call_rcu for that many pages; it should be especially
174  * useful before resizing the root node with PREEMPT_NONE configs; the value was
175  * obtained experimentally, aiming to avoid visible slowdown.
176  */
177 static const int sync_pages = 128;
178 
179 static struct kmem_cache *fn_alias_kmem __read_mostly;
180 static struct kmem_cache *trie_leaf_kmem __read_mostly;
181 
182 /*
183  * caller must hold RTNL
184  */
185 static inline struct tnode *node_parent(const struct rt_trie_node *node)
186 {
187 	unsigned long parent;
188 
189 	parent = rcu_dereference_index_check(node->parent, lockdep_rtnl_is_held());
190 
191 	return (struct tnode *)(parent & ~NODE_TYPE_MASK);
192 }
193 
194 /*
195  * caller must hold RCU read lock or RTNL
196  */
197 static inline struct tnode *node_parent_rcu(const struct rt_trie_node *node)
198 {
199 	unsigned long parent;
200 
201 	parent = rcu_dereference_index_check(node->parent, rcu_read_lock_held() ||
202 							   lockdep_rtnl_is_held());
203 
204 	return (struct tnode *)(parent & ~NODE_TYPE_MASK);
205 }
206 
207 /* Same as rcu_assign_pointer
208  * but that macro() assumes that value is a pointer.
209  */
210 static inline void node_set_parent(struct rt_trie_node *node, struct tnode *ptr)
211 {
212 	smp_wmb();
213 	node->parent = (unsigned long)ptr | NODE_TYPE(node);
214 }
215 
216 /*
217  * caller must hold RTNL
218  */
219 static inline struct rt_trie_node *tnode_get_child(const struct tnode *tn, unsigned int i)
220 {
221 	BUG_ON(i >= 1U << tn->bits);
222 
223 	return rtnl_dereference(tn->child[i]);
224 }
225 
226 /*
227  * caller must hold RCU read lock or RTNL
228  */
229 static inline struct rt_trie_node *tnode_get_child_rcu(const struct tnode *tn, unsigned int i)
230 {
231 	BUG_ON(i >= 1U << tn->bits);
232 
233 	return rcu_dereference_rtnl(tn->child[i]);
234 }
235 
236 static inline int tnode_child_length(const struct tnode *tn)
237 {
238 	return 1 << tn->bits;
239 }
240 
241 static inline t_key mask_pfx(t_key k, unsigned int l)
242 {
243 	return (l == 0) ? 0 : k >> (KEYLENGTH-l) << (KEYLENGTH-l);
244 }
245 
246 static inline t_key tkey_extract_bits(t_key a, unsigned int offset, unsigned int bits)
247 {
248 	if (offset < KEYLENGTH)
249 		return ((t_key)(a << offset)) >> (KEYLENGTH - bits);
250 	else
251 		return 0;
252 }
253 
254 static inline int tkey_equals(t_key a, t_key b)
255 {
256 	return a == b;
257 }
258 
259 static inline int tkey_sub_equals(t_key a, int offset, int bits, t_key b)
260 {
261 	if (bits == 0 || offset >= KEYLENGTH)
262 		return 1;
263 	bits = bits > KEYLENGTH ? KEYLENGTH : bits;
264 	return ((a ^ b) << offset) >> (KEYLENGTH - bits) == 0;
265 }
266 
267 static inline int tkey_mismatch(t_key a, int offset, t_key b)
268 {
269 	t_key diff = a ^ b;
270 	int i = offset;
271 
272 	if (!diff)
273 		return 0;
274 	while ((diff << i) >> (KEYLENGTH-1) == 0)
275 		i++;
276 	return i;
277 }
278 
279 /*
280   To understand this stuff, an understanding of keys and all their bits is
281   necessary. Every node in the trie has a key associated with it, but not
282   all of the bits in that key are significant.
283 
284   Consider a node 'n' and its parent 'tp'.
285 
286   If n is a leaf, every bit in its key is significant. Its presence is
287   necessitated by path compression, since during a tree traversal (when
288   searching for a leaf - unless we are doing an insertion) we will completely
289   ignore all skipped bits we encounter. Thus we need to verify, at the end of
290   a potentially successful search, that we have indeed been walking the
291   correct key path.
292 
293   Note that we can never "miss" the correct key in the tree if present by
294   following the wrong path. Path compression ensures that segments of the key
295   that are the same for all keys with a given prefix are skipped, but the
296   skipped part *is* identical for each node in the subtrie below the skipped
297   bit! trie_insert() in this implementation takes care of that - note the
298   call to tkey_sub_equals() in trie_insert().
299 
300   if n is an internal node - a 'tnode' here, the various parts of its key
301   have many different meanings.
302 
303   Example:
304   _________________________________________________________________
305   | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
306   -----------------------------------------------------------------
307     0   1   2   3   4   5   6   7   8   9  10  11  12  13  14  15
308 
309   _________________________________________________________________
310   | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
311   -----------------------------------------------------------------
312    16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31
313 
314   tp->pos = 7
315   tp->bits = 3
316   n->pos = 15
317   n->bits = 4
318 
319   First, let's just ignore the bits that come before the parent tp, that is
320   the bits from 0 to (tp->pos-1). They are *known* but at this point we do
321   not use them for anything.
322 
323   The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
324   index into the parent's child array. That is, they will be used to find
325   'n' among tp's children.
326 
327   The bits from (tp->pos + tp->bits) to (n->pos - 1) - "S" - are skipped bits
328   for the node n.
329 
330   All the bits we have seen so far are significant to the node n. The rest
331   of the bits are really not needed or indeed known in n->key.
332 
333   The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
334   n's child array, and will of course be different for each child.
335 
336 
337   The rest of the bits, from (n->pos + n->bits) onward, are completely unknown
338   at this point.
339 
340 */
341 
342 static inline void check_tnode(const struct tnode *tn)
343 {
344 	WARN_ON(tn && tn->pos+tn->bits > 32);
345 }
346 
347 static const int halve_threshold = 25;
348 static const int inflate_threshold = 50;
349 static const int halve_threshold_root = 15;
350 static const int inflate_threshold_root = 30;
351 
352 static void __alias_free_mem(struct rcu_head *head)
353 {
354 	struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
355 	kmem_cache_free(fn_alias_kmem, fa);
356 }
357 
358 static inline void alias_free_mem_rcu(struct fib_alias *fa)
359 {
360 	call_rcu(&fa->rcu, __alias_free_mem);
361 }
362 
363 static void __leaf_free_rcu(struct rcu_head *head)
364 {
365 	struct leaf *l = container_of(head, struct leaf, rcu);
366 	kmem_cache_free(trie_leaf_kmem, l);
367 }
368 
369 static inline void free_leaf(struct leaf *l)
370 {
371 	call_rcu_bh(&l->rcu, __leaf_free_rcu);
372 }
373 
374 static inline void free_leaf_info(struct leaf_info *leaf)
375 {
376 	kfree_rcu(leaf, rcu);
377 }
378 
379 static struct tnode *tnode_alloc(size_t size)
380 {
381 	if (size <= PAGE_SIZE)
382 		return kzalloc(size, GFP_KERNEL);
383 	else
384 		return vzalloc(size);
385 }
386 
387 static void __tnode_vfree(struct work_struct *arg)
388 {
389 	struct tnode *tn = container_of(arg, struct tnode, work);
390 	vfree(tn);
391 }
392 
393 static void __tnode_free_rcu(struct rcu_head *head)
394 {
395 	struct tnode *tn = container_of(head, struct tnode, rcu);
396 	size_t size = sizeof(struct tnode) +
397 		      (sizeof(struct rt_trie_node *) << tn->bits);
398 
399 	if (size <= PAGE_SIZE)
400 		kfree(tn);
401 	else {
402 		INIT_WORK(&tn->work, __tnode_vfree);
403 		schedule_work(&tn->work);
404 	}
405 }
406 
407 static inline void tnode_free(struct tnode *tn)
408 {
409 	if (IS_LEAF(tn))
410 		free_leaf((struct leaf *) tn);
411 	else
412 		call_rcu(&tn->rcu, __tnode_free_rcu);
413 }
414 
415 static void tnode_free_safe(struct tnode *tn)
416 {
417 	BUG_ON(IS_LEAF(tn));
418 	tn->tnode_free = tnode_free_head;
419 	tnode_free_head = tn;
420 	tnode_free_size += sizeof(struct tnode) +
421 			   (sizeof(struct rt_trie_node *) << tn->bits);
422 }
423 
424 static void tnode_free_flush(void)
425 {
426 	struct tnode *tn;
427 
428 	while ((tn = tnode_free_head)) {
429 		tnode_free_head = tn->tnode_free;
430 		tn->tnode_free = NULL;
431 		tnode_free(tn);
432 	}
433 
434 	if (tnode_free_size >= PAGE_SIZE * sync_pages) {
435 		tnode_free_size = 0;
436 		synchronize_rcu();
437 	}
438 }
439 
440 static struct leaf *leaf_new(void)
441 {
442 	struct leaf *l = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
443 	if (l) {
444 		l->parent = T_LEAF;
445 		INIT_HLIST_HEAD(&l->list);
446 	}
447 	return l;
448 }
449 
450 static struct leaf_info *leaf_info_new(int plen)
451 {
452 	struct leaf_info *li = kmalloc(sizeof(struct leaf_info),  GFP_KERNEL);
453 	if (li) {
454 		li->plen = plen;
455 		li->mask_plen = ntohl(inet_make_mask(plen));
456 		INIT_LIST_HEAD(&li->falh);
457 	}
458 	return li;
459 }
460 
461 static struct tnode *tnode_new(t_key key, int pos, int bits)
462 {
463 	size_t sz = sizeof(struct tnode) + (sizeof(struct rt_trie_node *) << bits);
464 	struct tnode *tn = tnode_alloc(sz);
465 
466 	if (tn) {
467 		tn->parent = T_TNODE;
468 		tn->pos = pos;
469 		tn->bits = bits;
470 		tn->key = key;
471 		tn->full_children = 0;
472 		tn->empty_children = 1<<bits;
473 	}
474 
475 	pr_debug("AT %p s=%zu %zu\n", tn, sizeof(struct tnode),
476 		 sizeof(struct rt_trie_node) << bits);
477 	return tn;
478 }
479 
480 /*
481  * Check whether a tnode 'n' is "full", i.e. it is an internal node
482  * and no bits are skipped. See discussion in dyntree paper p. 6
483  */
484 
485 static inline int tnode_full(const struct tnode *tn, const struct rt_trie_node *n)
486 {
487 	if (n == NULL || IS_LEAF(n))
488 		return 0;
489 
490 	return ((struct tnode *) n)->pos == tn->pos + tn->bits;
491 }
492 
493 static inline void put_child(struct trie *t, struct tnode *tn, int i,
494 			     struct rt_trie_node *n)
495 {
496 	tnode_put_child_reorg(tn, i, n, -1);
497 }
498 
499  /*
500   * Add a child at position i overwriting the old value.
501   * Update the value of full_children and empty_children.
502   */
503 
504 static void tnode_put_child_reorg(struct tnode *tn, int i, struct rt_trie_node *n,
505 				  int wasfull)
506 {
507 	struct rt_trie_node *chi = rtnl_dereference(tn->child[i]);
508 	int isfull;
509 
510 	BUG_ON(i >= 1<<tn->bits);
511 
512 	/* update emptyChildren */
513 	if (n == NULL && chi != NULL)
514 		tn->empty_children++;
515 	else if (n != NULL && chi == NULL)
516 		tn->empty_children--;
517 
518 	/* update fullChildren */
519 	if (wasfull == -1)
520 		wasfull = tnode_full(tn, chi);
521 
522 	isfull = tnode_full(tn, n);
523 	if (wasfull && !isfull)
524 		tn->full_children--;
525 	else if (!wasfull && isfull)
526 		tn->full_children++;
527 
528 	if (n)
529 		node_set_parent(n, tn);
530 
531 	rcu_assign_pointer(tn->child[i], n);
532 }
533 
534 #define MAX_WORK 10
535 static struct rt_trie_node *resize(struct trie *t, struct tnode *tn)
536 {
537 	int i;
538 	struct tnode *old_tn;
539 	int inflate_threshold_use;
540 	int halve_threshold_use;
541 	int max_work;
542 
543 	if (!tn)
544 		return NULL;
545 
546 	pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
547 		 tn, inflate_threshold, halve_threshold);
548 
549 	/* No children */
550 	if (tn->empty_children == tnode_child_length(tn)) {
551 		tnode_free_safe(tn);
552 		return NULL;
553 	}
554 	/* One child */
555 	if (tn->empty_children == tnode_child_length(tn) - 1)
556 		goto one_child;
557 	/*
558 	 * Double as long as the resulting node has a number of
559 	 * nonempty nodes that are above the threshold.
560 	 */
561 
562 	/*
563 	 * From "Implementing a dynamic compressed trie" by Stefan Nilsson of
564 	 * the Helsinki University of Technology and Matti Tikkanen of Nokia
565 	 * Telecommunications, page 6:
566 	 * "A node is doubled if the ratio of non-empty children to all
567 	 * children in the *doubled* node is at least 'high'."
568 	 *
569 	 * 'high' in this instance is the variable 'inflate_threshold'. It
570 	 * is expressed as a percentage, so we multiply it with
571 	 * tnode_child_length() and instead of multiplying by 2 (since the
572 	 * child array will be doubled by inflate()) and multiplying
573 	 * the left-hand side by 100 (to handle the percentage thing) we
574 	 * multiply the left-hand side by 50.
575 	 *
576 	 * The left-hand side may look a bit weird: tnode_child_length(tn)
577 	 * - tn->empty_children is of course the number of non-null children
578 	 * in the current node. tn->full_children is the number of "full"
579 	 * children, that is non-null tnodes with a skip value of 0.
580 	 * All of those will be doubled in the resulting inflated tnode, so
581 	 * we just count them one extra time here.
582 	 *
583 	 * A clearer way to write this would be:
584 	 *
585 	 * to_be_doubled = tn->full_children;
586 	 * not_to_be_doubled = tnode_child_length(tn) - tn->empty_children -
587 	 *     tn->full_children;
588 	 *
589 	 * new_child_length = tnode_child_length(tn) * 2;
590 	 *
591 	 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
592 	 *      new_child_length;
593 	 * if (new_fill_factor >= inflate_threshold)
594 	 *
595 	 * ...and so on, tho it would mess up the while () loop.
596 	 *
597 	 * anyway,
598 	 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
599 	 *      inflate_threshold
600 	 *
601 	 * avoid a division:
602 	 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
603 	 *      inflate_threshold * new_child_length
604 	 *
605 	 * expand not_to_be_doubled and to_be_doubled, and shorten:
606 	 * 100 * (tnode_child_length(tn) - tn->empty_children +
607 	 *    tn->full_children) >= inflate_threshold * new_child_length
608 	 *
609 	 * expand new_child_length:
610 	 * 100 * (tnode_child_length(tn) - tn->empty_children +
611 	 *    tn->full_children) >=
612 	 *      inflate_threshold * tnode_child_length(tn) * 2
613 	 *
614 	 * shorten again:
615 	 * 50 * (tn->full_children + tnode_child_length(tn) -
616 	 *    tn->empty_children) >= inflate_threshold *
617 	 *    tnode_child_length(tn)
618 	 *
619 	 */
620 
621 	check_tnode(tn);
622 
623 	/* Keep root node larger  */
624 
625 	if (!node_parent((struct rt_trie_node *)tn)) {
626 		inflate_threshold_use = inflate_threshold_root;
627 		halve_threshold_use = halve_threshold_root;
628 	} else {
629 		inflate_threshold_use = inflate_threshold;
630 		halve_threshold_use = halve_threshold;
631 	}
632 
633 	max_work = MAX_WORK;
634 	while ((tn->full_children > 0 &&  max_work-- &&
635 		50 * (tn->full_children + tnode_child_length(tn)
636 		      - tn->empty_children)
637 		>= inflate_threshold_use * tnode_child_length(tn))) {
638 
639 		old_tn = tn;
640 		tn = inflate(t, tn);
641 
642 		if (IS_ERR(tn)) {
643 			tn = old_tn;
644 #ifdef CONFIG_IP_FIB_TRIE_STATS
645 			t->stats.resize_node_skipped++;
646 #endif
647 			break;
648 		}
649 	}
650 
651 	check_tnode(tn);
652 
653 	/* Return if at least one inflate is run */
654 	if (max_work != MAX_WORK)
655 		return (struct rt_trie_node *) tn;
656 
657 	/*
658 	 * Halve as long as the number of empty children in this
659 	 * node is above threshold.
660 	 */
661 
662 	max_work = MAX_WORK;
663 	while (tn->bits > 1 &&  max_work-- &&
664 	       100 * (tnode_child_length(tn) - tn->empty_children) <
665 	       halve_threshold_use * tnode_child_length(tn)) {
666 
667 		old_tn = tn;
668 		tn = halve(t, tn);
669 		if (IS_ERR(tn)) {
670 			tn = old_tn;
671 #ifdef CONFIG_IP_FIB_TRIE_STATS
672 			t->stats.resize_node_skipped++;
673 #endif
674 			break;
675 		}
676 	}
677 
678 
679 	/* Only one child remains */
680 	if (tn->empty_children == tnode_child_length(tn) - 1) {
681 one_child:
682 		for (i = 0; i < tnode_child_length(tn); i++) {
683 			struct rt_trie_node *n;
684 
685 			n = rtnl_dereference(tn->child[i]);
686 			if (!n)
687 				continue;
688 
689 			/* compress one level */
690 
691 			node_set_parent(n, NULL);
692 			tnode_free_safe(tn);
693 			return n;
694 		}
695 	}
696 	return (struct rt_trie_node *) tn;
697 }
698 
699 
700 static void tnode_clean_free(struct tnode *tn)
701 {
702 	int i;
703 	struct tnode *tofree;
704 
705 	for (i = 0; i < tnode_child_length(tn); i++) {
706 		tofree = (struct tnode *)rtnl_dereference(tn->child[i]);
707 		if (tofree)
708 			tnode_free(tofree);
709 	}
710 	tnode_free(tn);
711 }
712 
713 static struct tnode *inflate(struct trie *t, struct tnode *tn)
714 {
715 	struct tnode *oldtnode = tn;
716 	int olen = tnode_child_length(tn);
717 	int i;
718 
719 	pr_debug("In inflate\n");
720 
721 	tn = tnode_new(oldtnode->key, oldtnode->pos, oldtnode->bits + 1);
722 
723 	if (!tn)
724 		return ERR_PTR(-ENOMEM);
725 
726 	/*
727 	 * Preallocate and store tnodes before the actual work so we
728 	 * don't get into an inconsistent state if memory allocation
729 	 * fails. In case of failure we return the oldnode and  inflate
730 	 * of tnode is ignored.
731 	 */
732 
733 	for (i = 0; i < olen; i++) {
734 		struct tnode *inode;
735 
736 		inode = (struct tnode *) tnode_get_child(oldtnode, i);
737 		if (inode &&
738 		    IS_TNODE(inode) &&
739 		    inode->pos == oldtnode->pos + oldtnode->bits &&
740 		    inode->bits > 1) {
741 			struct tnode *left, *right;
742 			t_key m = ~0U << (KEYLENGTH - 1) >> inode->pos;
743 
744 			left = tnode_new(inode->key&(~m), inode->pos + 1,
745 					 inode->bits - 1);
746 			if (!left)
747 				goto nomem;
748 
749 			right = tnode_new(inode->key|m, inode->pos + 1,
750 					  inode->bits - 1);
751 
752 			if (!right) {
753 				tnode_free(left);
754 				goto nomem;
755 			}
756 
757 			put_child(t, tn, 2*i, (struct rt_trie_node *) left);
758 			put_child(t, tn, 2*i+1, (struct rt_trie_node *) right);
759 		}
760 	}
761 
762 	for (i = 0; i < olen; i++) {
763 		struct tnode *inode;
764 		struct rt_trie_node *node = tnode_get_child(oldtnode, i);
765 		struct tnode *left, *right;
766 		int size, j;
767 
768 		/* An empty child */
769 		if (node == NULL)
770 			continue;
771 
772 		/* A leaf or an internal node with skipped bits */
773 
774 		if (IS_LEAF(node) || ((struct tnode *) node)->pos >
775 		   tn->pos + tn->bits - 1) {
776 			if (tkey_extract_bits(node->key,
777 					      oldtnode->pos + oldtnode->bits,
778 					      1) == 0)
779 				put_child(t, tn, 2*i, node);
780 			else
781 				put_child(t, tn, 2*i+1, node);
782 			continue;
783 		}
784 
785 		/* An internal node with two children */
786 		inode = (struct tnode *) node;
787 
788 		if (inode->bits == 1) {
789 			put_child(t, tn, 2*i, rtnl_dereference(inode->child[0]));
790 			put_child(t, tn, 2*i+1, rtnl_dereference(inode->child[1]));
791 
792 			tnode_free_safe(inode);
793 			continue;
794 		}
795 
796 		/* An internal node with more than two children */
797 
798 		/* We will replace this node 'inode' with two new
799 		 * ones, 'left' and 'right', each with half of the
800 		 * original children. The two new nodes will have
801 		 * a position one bit further down the key and this
802 		 * means that the "significant" part of their keys
803 		 * (see the discussion near the top of this file)
804 		 * will differ by one bit, which will be "0" in
805 		 * left's key and "1" in right's key. Since we are
806 		 * moving the key position by one step, the bit that
807 		 * we are moving away from - the bit at position
808 		 * (inode->pos) - is the one that will differ between
809 		 * left and right. So... we synthesize that bit in the
810 		 * two  new keys.
811 		 * The mask 'm' below will be a single "one" bit at
812 		 * the position (inode->pos)
813 		 */
814 
815 		/* Use the old key, but set the new significant
816 		 *   bit to zero.
817 		 */
818 
819 		left = (struct tnode *) tnode_get_child(tn, 2*i);
820 		put_child(t, tn, 2*i, NULL);
821 
822 		BUG_ON(!left);
823 
824 		right = (struct tnode *) tnode_get_child(tn, 2*i+1);
825 		put_child(t, tn, 2*i+1, NULL);
826 
827 		BUG_ON(!right);
828 
829 		size = tnode_child_length(left);
830 		for (j = 0; j < size; j++) {
831 			put_child(t, left, j, rtnl_dereference(inode->child[j]));
832 			put_child(t, right, j, rtnl_dereference(inode->child[j + size]));
833 		}
834 		put_child(t, tn, 2*i, resize(t, left));
835 		put_child(t, tn, 2*i+1, resize(t, right));
836 
837 		tnode_free_safe(inode);
838 	}
839 	tnode_free_safe(oldtnode);
840 	return tn;
841 nomem:
842 	tnode_clean_free(tn);
843 	return ERR_PTR(-ENOMEM);
844 }
845 
846 static struct tnode *halve(struct trie *t, struct tnode *tn)
847 {
848 	struct tnode *oldtnode = tn;
849 	struct rt_trie_node *left, *right;
850 	int i;
851 	int olen = tnode_child_length(tn);
852 
853 	pr_debug("In halve\n");
854 
855 	tn = tnode_new(oldtnode->key, oldtnode->pos, oldtnode->bits - 1);
856 
857 	if (!tn)
858 		return ERR_PTR(-ENOMEM);
859 
860 	/*
861 	 * Preallocate and store tnodes before the actual work so we
862 	 * don't get into an inconsistent state if memory allocation
863 	 * fails. In case of failure we return the oldnode and halve
864 	 * of tnode is ignored.
865 	 */
866 
867 	for (i = 0; i < olen; i += 2) {
868 		left = tnode_get_child(oldtnode, i);
869 		right = tnode_get_child(oldtnode, i+1);
870 
871 		/* Two nonempty children */
872 		if (left && right) {
873 			struct tnode *newn;
874 
875 			newn = tnode_new(left->key, tn->pos + tn->bits, 1);
876 
877 			if (!newn)
878 				goto nomem;
879 
880 			put_child(t, tn, i/2, (struct rt_trie_node *)newn);
881 		}
882 
883 	}
884 
885 	for (i = 0; i < olen; i += 2) {
886 		struct tnode *newBinNode;
887 
888 		left = tnode_get_child(oldtnode, i);
889 		right = tnode_get_child(oldtnode, i+1);
890 
891 		/* At least one of the children is empty */
892 		if (left == NULL) {
893 			if (right == NULL)    /* Both are empty */
894 				continue;
895 			put_child(t, tn, i/2, right);
896 			continue;
897 		}
898 
899 		if (right == NULL) {
900 			put_child(t, tn, i/2, left);
901 			continue;
902 		}
903 
904 		/* Two nonempty children */
905 		newBinNode = (struct tnode *) tnode_get_child(tn, i/2);
906 		put_child(t, tn, i/2, NULL);
907 		put_child(t, newBinNode, 0, left);
908 		put_child(t, newBinNode, 1, right);
909 		put_child(t, tn, i/2, resize(t, newBinNode));
910 	}
911 	tnode_free_safe(oldtnode);
912 	return tn;
913 nomem:
914 	tnode_clean_free(tn);
915 	return ERR_PTR(-ENOMEM);
916 }
917 
918 /* readside must use rcu_read_lock currently dump routines
919  via get_fa_head and dump */
920 
921 static struct leaf_info *find_leaf_info(struct leaf *l, int plen)
922 {
923 	struct hlist_head *head = &l->list;
924 	struct hlist_node *node;
925 	struct leaf_info *li;
926 
927 	hlist_for_each_entry_rcu(li, node, head, hlist)
928 		if (li->plen == plen)
929 			return li;
930 
931 	return NULL;
932 }
933 
934 static inline struct list_head *get_fa_head(struct leaf *l, int plen)
935 {
936 	struct leaf_info *li = find_leaf_info(l, plen);
937 
938 	if (!li)
939 		return NULL;
940 
941 	return &li->falh;
942 }
943 
944 static void insert_leaf_info(struct hlist_head *head, struct leaf_info *new)
945 {
946 	struct leaf_info *li = NULL, *last = NULL;
947 	struct hlist_node *node;
948 
949 	if (hlist_empty(head)) {
950 		hlist_add_head_rcu(&new->hlist, head);
951 	} else {
952 		hlist_for_each_entry(li, node, head, hlist) {
953 			if (new->plen > li->plen)
954 				break;
955 
956 			last = li;
957 		}
958 		if (last)
959 			hlist_add_after_rcu(&last->hlist, &new->hlist);
960 		else
961 			hlist_add_before_rcu(&new->hlist, &li->hlist);
962 	}
963 }
964 
965 /* rcu_read_lock needs to be hold by caller from readside */
966 
967 static struct leaf *
968 fib_find_node(struct trie *t, u32 key)
969 {
970 	int pos;
971 	struct tnode *tn;
972 	struct rt_trie_node *n;
973 
974 	pos = 0;
975 	n = rcu_dereference_rtnl(t->trie);
976 
977 	while (n != NULL &&  NODE_TYPE(n) == T_TNODE) {
978 		tn = (struct tnode *) n;
979 
980 		check_tnode(tn);
981 
982 		if (tkey_sub_equals(tn->key, pos, tn->pos-pos, key)) {
983 			pos = tn->pos + tn->bits;
984 			n = tnode_get_child_rcu(tn,
985 						tkey_extract_bits(key,
986 								  tn->pos,
987 								  tn->bits));
988 		} else
989 			break;
990 	}
991 	/* Case we have found a leaf. Compare prefixes */
992 
993 	if (n != NULL && IS_LEAF(n) && tkey_equals(key, n->key))
994 		return (struct leaf *)n;
995 
996 	return NULL;
997 }
998 
999 static void trie_rebalance(struct trie *t, struct tnode *tn)
1000 {
1001 	int wasfull;
1002 	t_key cindex, key;
1003 	struct tnode *tp;
1004 
1005 	key = tn->key;
1006 
1007 	while (tn != NULL && (tp = node_parent((struct rt_trie_node *)tn)) != NULL) {
1008 		cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1009 		wasfull = tnode_full(tp, tnode_get_child(tp, cindex));
1010 		tn = (struct tnode *) resize(t, (struct tnode *)tn);
1011 
1012 		tnode_put_child_reorg((struct tnode *)tp, cindex,
1013 				      (struct rt_trie_node *)tn, wasfull);
1014 
1015 		tp = node_parent((struct rt_trie_node *) tn);
1016 		if (!tp)
1017 			rcu_assign_pointer(t->trie, (struct rt_trie_node *)tn);
1018 
1019 		tnode_free_flush();
1020 		if (!tp)
1021 			break;
1022 		tn = tp;
1023 	}
1024 
1025 	/* Handle last (top) tnode */
1026 	if (IS_TNODE(tn))
1027 		tn = (struct tnode *)resize(t, (struct tnode *)tn);
1028 
1029 	rcu_assign_pointer(t->trie, (struct rt_trie_node *)tn);
1030 	tnode_free_flush();
1031 }
1032 
1033 /* only used from updater-side */
1034 
1035 static struct list_head *fib_insert_node(struct trie *t, u32 key, int plen)
1036 {
1037 	int pos, newpos;
1038 	struct tnode *tp = NULL, *tn = NULL;
1039 	struct rt_trie_node *n;
1040 	struct leaf *l;
1041 	int missbit;
1042 	struct list_head *fa_head = NULL;
1043 	struct leaf_info *li;
1044 	t_key cindex;
1045 
1046 	pos = 0;
1047 	n = rtnl_dereference(t->trie);
1048 
1049 	/* If we point to NULL, stop. Either the tree is empty and we should
1050 	 * just put a new leaf in if, or we have reached an empty child slot,
1051 	 * and we should just put our new leaf in that.
1052 	 * If we point to a T_TNODE, check if it matches our key. Note that
1053 	 * a T_TNODE might be skipping any number of bits - its 'pos' need
1054 	 * not be the parent's 'pos'+'bits'!
1055 	 *
1056 	 * If it does match the current key, get pos/bits from it, extract
1057 	 * the index from our key, push the T_TNODE and walk the tree.
1058 	 *
1059 	 * If it doesn't, we have to replace it with a new T_TNODE.
1060 	 *
1061 	 * If we point to a T_LEAF, it might or might not have the same key
1062 	 * as we do. If it does, just change the value, update the T_LEAF's
1063 	 * value, and return it.
1064 	 * If it doesn't, we need to replace it with a T_TNODE.
1065 	 */
1066 
1067 	while (n != NULL &&  NODE_TYPE(n) == T_TNODE) {
1068 		tn = (struct tnode *) n;
1069 
1070 		check_tnode(tn);
1071 
1072 		if (tkey_sub_equals(tn->key, pos, tn->pos-pos, key)) {
1073 			tp = tn;
1074 			pos = tn->pos + tn->bits;
1075 			n = tnode_get_child(tn,
1076 					    tkey_extract_bits(key,
1077 							      tn->pos,
1078 							      tn->bits));
1079 
1080 			BUG_ON(n && node_parent(n) != tn);
1081 		} else
1082 			break;
1083 	}
1084 
1085 	/*
1086 	 * n  ----> NULL, LEAF or TNODE
1087 	 *
1088 	 * tp is n's (parent) ----> NULL or TNODE
1089 	 */
1090 
1091 	BUG_ON(tp && IS_LEAF(tp));
1092 
1093 	/* Case 1: n is a leaf. Compare prefixes */
1094 
1095 	if (n != NULL && IS_LEAF(n) && tkey_equals(key, n->key)) {
1096 		l = (struct leaf *) n;
1097 		li = leaf_info_new(plen);
1098 
1099 		if (!li)
1100 			return NULL;
1101 
1102 		fa_head = &li->falh;
1103 		insert_leaf_info(&l->list, li);
1104 		goto done;
1105 	}
1106 	l = leaf_new();
1107 
1108 	if (!l)
1109 		return NULL;
1110 
1111 	l->key = key;
1112 	li = leaf_info_new(plen);
1113 
1114 	if (!li) {
1115 		free_leaf(l);
1116 		return NULL;
1117 	}
1118 
1119 	fa_head = &li->falh;
1120 	insert_leaf_info(&l->list, li);
1121 
1122 	if (t->trie && n == NULL) {
1123 		/* Case 2: n is NULL, and will just insert a new leaf */
1124 
1125 		node_set_parent((struct rt_trie_node *)l, tp);
1126 
1127 		cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1128 		put_child(t, (struct tnode *)tp, cindex, (struct rt_trie_node *)l);
1129 	} else {
1130 		/* Case 3: n is a LEAF or a TNODE and the key doesn't match. */
1131 		/*
1132 		 *  Add a new tnode here
1133 		 *  first tnode need some special handling
1134 		 */
1135 
1136 		if (tp)
1137 			pos = tp->pos+tp->bits;
1138 		else
1139 			pos = 0;
1140 
1141 		if (n) {
1142 			newpos = tkey_mismatch(key, pos, n->key);
1143 			tn = tnode_new(n->key, newpos, 1);
1144 		} else {
1145 			newpos = 0;
1146 			tn = tnode_new(key, newpos, 1); /* First tnode */
1147 		}
1148 
1149 		if (!tn) {
1150 			free_leaf_info(li);
1151 			free_leaf(l);
1152 			return NULL;
1153 		}
1154 
1155 		node_set_parent((struct rt_trie_node *)tn, tp);
1156 
1157 		missbit = tkey_extract_bits(key, newpos, 1);
1158 		put_child(t, tn, missbit, (struct rt_trie_node *)l);
1159 		put_child(t, tn, 1-missbit, n);
1160 
1161 		if (tp) {
1162 			cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1163 			put_child(t, (struct tnode *)tp, cindex,
1164 				  (struct rt_trie_node *)tn);
1165 		} else {
1166 			rcu_assign_pointer(t->trie, (struct rt_trie_node *)tn);
1167 			tp = tn;
1168 		}
1169 	}
1170 
1171 	if (tp && tp->pos + tp->bits > 32)
1172 		pr_warning("fib_trie"
1173 			   " tp=%p pos=%d, bits=%d, key=%0x plen=%d\n",
1174 			   tp, tp->pos, tp->bits, key, plen);
1175 
1176 	/* Rebalance the trie */
1177 
1178 	trie_rebalance(t, tp);
1179 done:
1180 	return fa_head;
1181 }
1182 
1183 /*
1184  * Caller must hold RTNL.
1185  */
1186 int fib_table_insert(struct fib_table *tb, struct fib_config *cfg)
1187 {
1188 	struct trie *t = (struct trie *) tb->tb_data;
1189 	struct fib_alias *fa, *new_fa;
1190 	struct list_head *fa_head = NULL;
1191 	struct fib_info *fi;
1192 	int plen = cfg->fc_dst_len;
1193 	u8 tos = cfg->fc_tos;
1194 	u32 key, mask;
1195 	int err;
1196 	struct leaf *l;
1197 
1198 	if (plen > 32)
1199 		return -EINVAL;
1200 
1201 	key = ntohl(cfg->fc_dst);
1202 
1203 	pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1204 
1205 	mask = ntohl(inet_make_mask(plen));
1206 
1207 	if (key & ~mask)
1208 		return -EINVAL;
1209 
1210 	key = key & mask;
1211 
1212 	fi = fib_create_info(cfg);
1213 	if (IS_ERR(fi)) {
1214 		err = PTR_ERR(fi);
1215 		goto err;
1216 	}
1217 
1218 	l = fib_find_node(t, key);
1219 	fa = NULL;
1220 
1221 	if (l) {
1222 		fa_head = get_fa_head(l, plen);
1223 		fa = fib_find_alias(fa_head, tos, fi->fib_priority);
1224 	}
1225 
1226 	/* Now fa, if non-NULL, points to the first fib alias
1227 	 * with the same keys [prefix,tos,priority], if such key already
1228 	 * exists or to the node before which we will insert new one.
1229 	 *
1230 	 * If fa is NULL, we will need to allocate a new one and
1231 	 * insert to the head of f.
1232 	 *
1233 	 * If f is NULL, no fib node matched the destination key
1234 	 * and we need to allocate a new one of those as well.
1235 	 */
1236 
1237 	if (fa && fa->fa_tos == tos &&
1238 	    fa->fa_info->fib_priority == fi->fib_priority) {
1239 		struct fib_alias *fa_first, *fa_match;
1240 
1241 		err = -EEXIST;
1242 		if (cfg->fc_nlflags & NLM_F_EXCL)
1243 			goto out;
1244 
1245 		/* We have 2 goals:
1246 		 * 1. Find exact match for type, scope, fib_info to avoid
1247 		 * duplicate routes
1248 		 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1249 		 */
1250 		fa_match = NULL;
1251 		fa_first = fa;
1252 		fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
1253 		list_for_each_entry_continue(fa, fa_head, fa_list) {
1254 			if (fa->fa_tos != tos)
1255 				break;
1256 			if (fa->fa_info->fib_priority != fi->fib_priority)
1257 				break;
1258 			if (fa->fa_type == cfg->fc_type &&
1259 			    fa->fa_info == fi) {
1260 				fa_match = fa;
1261 				break;
1262 			}
1263 		}
1264 
1265 		if (cfg->fc_nlflags & NLM_F_REPLACE) {
1266 			struct fib_info *fi_drop;
1267 			u8 state;
1268 
1269 			fa = fa_first;
1270 			if (fa_match) {
1271 				if (fa == fa_match)
1272 					err = 0;
1273 				goto out;
1274 			}
1275 			err = -ENOBUFS;
1276 			new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1277 			if (new_fa == NULL)
1278 				goto out;
1279 
1280 			fi_drop = fa->fa_info;
1281 			new_fa->fa_tos = fa->fa_tos;
1282 			new_fa->fa_info = fi;
1283 			new_fa->fa_type = cfg->fc_type;
1284 			state = fa->fa_state;
1285 			new_fa->fa_state = state & ~FA_S_ACCESSED;
1286 
1287 			list_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1288 			alias_free_mem_rcu(fa);
1289 
1290 			fib_release_info(fi_drop);
1291 			if (state & FA_S_ACCESSED)
1292 				rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1293 			rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1294 				tb->tb_id, &cfg->fc_nlinfo, NLM_F_REPLACE);
1295 
1296 			goto succeeded;
1297 		}
1298 		/* Error if we find a perfect match which
1299 		 * uses the same scope, type, and nexthop
1300 		 * information.
1301 		 */
1302 		if (fa_match)
1303 			goto out;
1304 
1305 		if (!(cfg->fc_nlflags & NLM_F_APPEND))
1306 			fa = fa_first;
1307 	}
1308 	err = -ENOENT;
1309 	if (!(cfg->fc_nlflags & NLM_F_CREATE))
1310 		goto out;
1311 
1312 	err = -ENOBUFS;
1313 	new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1314 	if (new_fa == NULL)
1315 		goto out;
1316 
1317 	new_fa->fa_info = fi;
1318 	new_fa->fa_tos = tos;
1319 	new_fa->fa_type = cfg->fc_type;
1320 	new_fa->fa_state = 0;
1321 	/*
1322 	 * Insert new entry to the list.
1323 	 */
1324 
1325 	if (!fa_head) {
1326 		fa_head = fib_insert_node(t, key, plen);
1327 		if (unlikely(!fa_head)) {
1328 			err = -ENOMEM;
1329 			goto out_free_new_fa;
1330 		}
1331 	}
1332 
1333 	if (!plen)
1334 		tb->tb_num_default++;
1335 
1336 	list_add_tail_rcu(&new_fa->fa_list,
1337 			  (fa ? &fa->fa_list : fa_head));
1338 
1339 	rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1340 	rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, tb->tb_id,
1341 		  &cfg->fc_nlinfo, 0);
1342 succeeded:
1343 	return 0;
1344 
1345 out_free_new_fa:
1346 	kmem_cache_free(fn_alias_kmem, new_fa);
1347 out:
1348 	fib_release_info(fi);
1349 err:
1350 	return err;
1351 }
1352 
1353 /* should be called with rcu_read_lock */
1354 static int check_leaf(struct fib_table *tb, struct trie *t, struct leaf *l,
1355 		      t_key key,  const struct flowi4 *flp,
1356 		      struct fib_result *res, int fib_flags)
1357 {
1358 	struct leaf_info *li;
1359 	struct hlist_head *hhead = &l->list;
1360 	struct hlist_node *node;
1361 
1362 	hlist_for_each_entry_rcu(li, node, hhead, hlist) {
1363 		struct fib_alias *fa;
1364 
1365 		if (l->key != (key & li->mask_plen))
1366 			continue;
1367 
1368 		list_for_each_entry_rcu(fa, &li->falh, fa_list) {
1369 			struct fib_info *fi = fa->fa_info;
1370 			int nhsel, err;
1371 
1372 			if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1373 				continue;
1374 			if (fa->fa_info->fib_scope < flp->flowi4_scope)
1375 				continue;
1376 			fib_alias_accessed(fa);
1377 			err = fib_props[fa->fa_type].error;
1378 			if (err) {
1379 #ifdef CONFIG_IP_FIB_TRIE_STATS
1380 				t->stats.semantic_match_passed++;
1381 #endif
1382 				return err;
1383 			}
1384 			if (fi->fib_flags & RTNH_F_DEAD)
1385 				continue;
1386 			for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
1387 				const struct fib_nh *nh = &fi->fib_nh[nhsel];
1388 
1389 				if (nh->nh_flags & RTNH_F_DEAD)
1390 					continue;
1391 				if (flp->flowi4_oif && flp->flowi4_oif != nh->nh_oif)
1392 					continue;
1393 
1394 #ifdef CONFIG_IP_FIB_TRIE_STATS
1395 				t->stats.semantic_match_passed++;
1396 #endif
1397 				res->prefixlen = li->plen;
1398 				res->nh_sel = nhsel;
1399 				res->type = fa->fa_type;
1400 				res->scope = fa->fa_info->fib_scope;
1401 				res->fi = fi;
1402 				res->table = tb;
1403 				res->fa_head = &li->falh;
1404 				if (!(fib_flags & FIB_LOOKUP_NOREF))
1405 					atomic_inc(&fi->fib_clntref);
1406 				return 0;
1407 			}
1408 		}
1409 
1410 #ifdef CONFIG_IP_FIB_TRIE_STATS
1411 		t->stats.semantic_match_miss++;
1412 #endif
1413 	}
1414 
1415 	return 1;
1416 }
1417 
1418 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1419 		     struct fib_result *res, int fib_flags)
1420 {
1421 	struct trie *t = (struct trie *) tb->tb_data;
1422 	int ret;
1423 	struct rt_trie_node *n;
1424 	struct tnode *pn;
1425 	unsigned int pos, bits;
1426 	t_key key = ntohl(flp->daddr);
1427 	unsigned int chopped_off;
1428 	t_key cindex = 0;
1429 	unsigned int current_prefix_length = KEYLENGTH;
1430 	struct tnode *cn;
1431 	t_key pref_mismatch;
1432 
1433 	rcu_read_lock();
1434 
1435 	n = rcu_dereference(t->trie);
1436 	if (!n)
1437 		goto failed;
1438 
1439 #ifdef CONFIG_IP_FIB_TRIE_STATS
1440 	t->stats.gets++;
1441 #endif
1442 
1443 	/* Just a leaf? */
1444 	if (IS_LEAF(n)) {
1445 		ret = check_leaf(tb, t, (struct leaf *)n, key, flp, res, fib_flags);
1446 		goto found;
1447 	}
1448 
1449 	pn = (struct tnode *) n;
1450 	chopped_off = 0;
1451 
1452 	while (pn) {
1453 		pos = pn->pos;
1454 		bits = pn->bits;
1455 
1456 		if (!chopped_off)
1457 			cindex = tkey_extract_bits(mask_pfx(key, current_prefix_length),
1458 						   pos, bits);
1459 
1460 		n = tnode_get_child_rcu(pn, cindex);
1461 
1462 		if (n == NULL) {
1463 #ifdef CONFIG_IP_FIB_TRIE_STATS
1464 			t->stats.null_node_hit++;
1465 #endif
1466 			goto backtrace;
1467 		}
1468 
1469 		if (IS_LEAF(n)) {
1470 			ret = check_leaf(tb, t, (struct leaf *)n, key, flp, res, fib_flags);
1471 			if (ret > 0)
1472 				goto backtrace;
1473 			goto found;
1474 		}
1475 
1476 		cn = (struct tnode *)n;
1477 
1478 		/*
1479 		 * It's a tnode, and we can do some extra checks here if we
1480 		 * like, to avoid descending into a dead-end branch.
1481 		 * This tnode is in the parent's child array at index
1482 		 * key[p_pos..p_pos+p_bits] but potentially with some bits
1483 		 * chopped off, so in reality the index may be just a
1484 		 * subprefix, padded with zero at the end.
1485 		 * We can also take a look at any skipped bits in this
1486 		 * tnode - everything up to p_pos is supposed to be ok,
1487 		 * and the non-chopped bits of the index (se previous
1488 		 * paragraph) are also guaranteed ok, but the rest is
1489 		 * considered unknown.
1490 		 *
1491 		 * The skipped bits are key[pos+bits..cn->pos].
1492 		 */
1493 
1494 		/* If current_prefix_length < pos+bits, we are already doing
1495 		 * actual prefix  matching, which means everything from
1496 		 * pos+(bits-chopped_off) onward must be zero along some
1497 		 * branch of this subtree - otherwise there is *no* valid
1498 		 * prefix present. Here we can only check the skipped
1499 		 * bits. Remember, since we have already indexed into the
1500 		 * parent's child array, we know that the bits we chopped of
1501 		 * *are* zero.
1502 		 */
1503 
1504 		/* NOTA BENE: Checking only skipped bits
1505 		   for the new node here */
1506 
1507 		if (current_prefix_length < pos+bits) {
1508 			if (tkey_extract_bits(cn->key, current_prefix_length,
1509 						cn->pos - current_prefix_length)
1510 			    || !(cn->child[0]))
1511 				goto backtrace;
1512 		}
1513 
1514 		/*
1515 		 * If chopped_off=0, the index is fully validated and we
1516 		 * only need to look at the skipped bits for this, the new,
1517 		 * tnode. What we actually want to do is to find out if
1518 		 * these skipped bits match our key perfectly, or if we will
1519 		 * have to count on finding a matching prefix further down,
1520 		 * because if we do, we would like to have some way of
1521 		 * verifying the existence of such a prefix at this point.
1522 		 */
1523 
1524 		/* The only thing we can do at this point is to verify that
1525 		 * any such matching prefix can indeed be a prefix to our
1526 		 * key, and if the bits in the node we are inspecting that
1527 		 * do not match our key are not ZERO, this cannot be true.
1528 		 * Thus, find out where there is a mismatch (before cn->pos)
1529 		 * and verify that all the mismatching bits are zero in the
1530 		 * new tnode's key.
1531 		 */
1532 
1533 		/*
1534 		 * Note: We aren't very concerned about the piece of
1535 		 * the key that precede pn->pos+pn->bits, since these
1536 		 * have already been checked. The bits after cn->pos
1537 		 * aren't checked since these are by definition
1538 		 * "unknown" at this point. Thus, what we want to see
1539 		 * is if we are about to enter the "prefix matching"
1540 		 * state, and in that case verify that the skipped
1541 		 * bits that will prevail throughout this subtree are
1542 		 * zero, as they have to be if we are to find a
1543 		 * matching prefix.
1544 		 */
1545 
1546 		pref_mismatch = mask_pfx(cn->key ^ key, cn->pos);
1547 
1548 		/*
1549 		 * In short: If skipped bits in this node do not match
1550 		 * the search key, enter the "prefix matching"
1551 		 * state.directly.
1552 		 */
1553 		if (pref_mismatch) {
1554 			int mp = KEYLENGTH - fls(pref_mismatch);
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 
1610 /*
1611  * Remove the leaf and return parent.
1612  */
1613 static void trie_leaf_remove(struct trie *t, struct leaf *l)
1614 {
1615 	struct tnode *tp = node_parent((struct rt_trie_node *) l);
1616 
1617 	pr_debug("entering trie_leaf_remove(%p)\n", l);
1618 
1619 	if (tp) {
1620 		t_key cindex = tkey_extract_bits(l->key, tp->pos, tp->bits);
1621 		put_child(t, (struct tnode *)tp, cindex, NULL);
1622 		trie_rebalance(t, tp);
1623 	} else
1624 		rcu_assign_pointer(t->trie, NULL);
1625 
1626 	free_leaf(l);
1627 }
1628 
1629 /*
1630  * Caller must hold RTNL.
1631  */
1632 int fib_table_delete(struct fib_table *tb, struct fib_config *cfg)
1633 {
1634 	struct trie *t = (struct trie *) tb->tb_data;
1635 	u32 key, mask;
1636 	int plen = cfg->fc_dst_len;
1637 	u8 tos = cfg->fc_tos;
1638 	struct fib_alias *fa, *fa_to_delete;
1639 	struct list_head *fa_head;
1640 	struct leaf *l;
1641 	struct leaf_info *li;
1642 
1643 	if (plen > 32)
1644 		return -EINVAL;
1645 
1646 	key = ntohl(cfg->fc_dst);
1647 	mask = ntohl(inet_make_mask(plen));
1648 
1649 	if (key & ~mask)
1650 		return -EINVAL;
1651 
1652 	key = key & mask;
1653 	l = fib_find_node(t, key);
1654 
1655 	if (!l)
1656 		return -ESRCH;
1657 
1658 	fa_head = get_fa_head(l, plen);
1659 	fa = fib_find_alias(fa_head, tos, 0);
1660 
1661 	if (!fa)
1662 		return -ESRCH;
1663 
1664 	pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1665 
1666 	fa_to_delete = NULL;
1667 	fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
1668 	list_for_each_entry_continue(fa, fa_head, fa_list) {
1669 		struct fib_info *fi = fa->fa_info;
1670 
1671 		if (fa->fa_tos != tos)
1672 			break;
1673 
1674 		if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1675 		    (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1676 		     fa->fa_info->fib_scope == cfg->fc_scope) &&
1677 		    (!cfg->fc_prefsrc ||
1678 		     fi->fib_prefsrc == cfg->fc_prefsrc) &&
1679 		    (!cfg->fc_protocol ||
1680 		     fi->fib_protocol == cfg->fc_protocol) &&
1681 		    fib_nh_match(cfg, fi) == 0) {
1682 			fa_to_delete = fa;
1683 			break;
1684 		}
1685 	}
1686 
1687 	if (!fa_to_delete)
1688 		return -ESRCH;
1689 
1690 	fa = fa_to_delete;
1691 	rtmsg_fib(RTM_DELROUTE, htonl(key), fa, plen, tb->tb_id,
1692 		  &cfg->fc_nlinfo, 0);
1693 
1694 	l = fib_find_node(t, key);
1695 	li = find_leaf_info(l, plen);
1696 
1697 	list_del_rcu(&fa->fa_list);
1698 
1699 	if (!plen)
1700 		tb->tb_num_default--;
1701 
1702 	if (list_empty(fa_head)) {
1703 		hlist_del_rcu(&li->hlist);
1704 		free_leaf_info(li);
1705 	}
1706 
1707 	if (hlist_empty(&l->list))
1708 		trie_leaf_remove(t, l);
1709 
1710 	if (fa->fa_state & FA_S_ACCESSED)
1711 		rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1712 
1713 	fib_release_info(fa->fa_info);
1714 	alias_free_mem_rcu(fa);
1715 	return 0;
1716 }
1717 
1718 static int trie_flush_list(struct list_head *head)
1719 {
1720 	struct fib_alias *fa, *fa_node;
1721 	int found = 0;
1722 
1723 	list_for_each_entry_safe(fa, fa_node, head, fa_list) {
1724 		struct fib_info *fi = fa->fa_info;
1725 
1726 		if (fi && (fi->fib_flags & RTNH_F_DEAD)) {
1727 			list_del_rcu(&fa->fa_list);
1728 			fib_release_info(fa->fa_info);
1729 			alias_free_mem_rcu(fa);
1730 			found++;
1731 		}
1732 	}
1733 	return found;
1734 }
1735 
1736 static int trie_flush_leaf(struct leaf *l)
1737 {
1738 	int found = 0;
1739 	struct hlist_head *lih = &l->list;
1740 	struct hlist_node *node, *tmp;
1741 	struct leaf_info *li = NULL;
1742 
1743 	hlist_for_each_entry_safe(li, node, tmp, lih, hlist) {
1744 		found += trie_flush_list(&li->falh);
1745 
1746 		if (list_empty(&li->falh)) {
1747 			hlist_del_rcu(&li->hlist);
1748 			free_leaf_info(li);
1749 		}
1750 	}
1751 	return found;
1752 }
1753 
1754 /*
1755  * Scan for the next right leaf starting at node p->child[idx]
1756  * Since we have back pointer, no recursion necessary.
1757  */
1758 static struct leaf *leaf_walk_rcu(struct tnode *p, struct rt_trie_node *c)
1759 {
1760 	do {
1761 		t_key idx;
1762 
1763 		if (c)
1764 			idx = tkey_extract_bits(c->key, p->pos, p->bits) + 1;
1765 		else
1766 			idx = 0;
1767 
1768 		while (idx < 1u << p->bits) {
1769 			c = tnode_get_child_rcu(p, idx++);
1770 			if (!c)
1771 				continue;
1772 
1773 			if (IS_LEAF(c)) {
1774 				prefetch(rcu_dereference_rtnl(p->child[idx]));
1775 				return (struct leaf *) c;
1776 			}
1777 
1778 			/* Rescan start scanning in new node */
1779 			p = (struct tnode *) c;
1780 			idx = 0;
1781 		}
1782 
1783 		/* Node empty, walk back up to parent */
1784 		c = (struct rt_trie_node *) p;
1785 	} while ((p = node_parent_rcu(c)) != NULL);
1786 
1787 	return NULL; /* Root of trie */
1788 }
1789 
1790 static struct leaf *trie_firstleaf(struct trie *t)
1791 {
1792 	struct tnode *n = (struct tnode *)rcu_dereference_rtnl(t->trie);
1793 
1794 	if (!n)
1795 		return NULL;
1796 
1797 	if (IS_LEAF(n))          /* trie is just a leaf */
1798 		return (struct leaf *) n;
1799 
1800 	return leaf_walk_rcu(n, NULL);
1801 }
1802 
1803 static struct leaf *trie_nextleaf(struct leaf *l)
1804 {
1805 	struct rt_trie_node *c = (struct rt_trie_node *) l;
1806 	struct tnode *p = node_parent_rcu(c);
1807 
1808 	if (!p)
1809 		return NULL;	/* trie with just one leaf */
1810 
1811 	return leaf_walk_rcu(p, c);
1812 }
1813 
1814 static struct leaf *trie_leafindex(struct trie *t, int index)
1815 {
1816 	struct leaf *l = trie_firstleaf(t);
1817 
1818 	while (l && index-- > 0)
1819 		l = trie_nextleaf(l);
1820 
1821 	return l;
1822 }
1823 
1824 
1825 /*
1826  * Caller must hold RTNL.
1827  */
1828 int fib_table_flush(struct fib_table *tb)
1829 {
1830 	struct trie *t = (struct trie *) tb->tb_data;
1831 	struct leaf *l, *ll = NULL;
1832 	int found = 0;
1833 
1834 	for (l = trie_firstleaf(t); l; l = trie_nextleaf(l)) {
1835 		found += trie_flush_leaf(l);
1836 
1837 		if (ll && hlist_empty(&ll->list))
1838 			trie_leaf_remove(t, ll);
1839 		ll = l;
1840 	}
1841 
1842 	if (ll && hlist_empty(&ll->list))
1843 		trie_leaf_remove(t, ll);
1844 
1845 	pr_debug("trie_flush found=%d\n", found);
1846 	return found;
1847 }
1848 
1849 void fib_free_table(struct fib_table *tb)
1850 {
1851 	kfree(tb);
1852 }
1853 
1854 static int fn_trie_dump_fa(t_key key, int plen, struct list_head *fah,
1855 			   struct fib_table *tb,
1856 			   struct sk_buff *skb, struct netlink_callback *cb)
1857 {
1858 	int i, s_i;
1859 	struct fib_alias *fa;
1860 	__be32 xkey = htonl(key);
1861 
1862 	s_i = cb->args[5];
1863 	i = 0;
1864 
1865 	/* rcu_read_lock is hold by caller */
1866 
1867 	list_for_each_entry_rcu(fa, fah, fa_list) {
1868 		if (i < s_i) {
1869 			i++;
1870 			continue;
1871 		}
1872 
1873 		if (fib_dump_info(skb, NETLINK_CB(cb->skb).pid,
1874 				  cb->nlh->nlmsg_seq,
1875 				  RTM_NEWROUTE,
1876 				  tb->tb_id,
1877 				  fa->fa_type,
1878 				  xkey,
1879 				  plen,
1880 				  fa->fa_tos,
1881 				  fa->fa_info, NLM_F_MULTI) < 0) {
1882 			cb->args[5] = i;
1883 			return -1;
1884 		}
1885 		i++;
1886 	}
1887 	cb->args[5] = i;
1888 	return skb->len;
1889 }
1890 
1891 static int fn_trie_dump_leaf(struct leaf *l, struct fib_table *tb,
1892 			struct sk_buff *skb, struct netlink_callback *cb)
1893 {
1894 	struct leaf_info *li;
1895 	struct hlist_node *node;
1896 	int i, s_i;
1897 
1898 	s_i = cb->args[4];
1899 	i = 0;
1900 
1901 	/* rcu_read_lock is hold by caller */
1902 	hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
1903 		if (i < s_i) {
1904 			i++;
1905 			continue;
1906 		}
1907 
1908 		if (i > s_i)
1909 			cb->args[5] = 0;
1910 
1911 		if (list_empty(&li->falh))
1912 			continue;
1913 
1914 		if (fn_trie_dump_fa(l->key, li->plen, &li->falh, tb, skb, cb) < 0) {
1915 			cb->args[4] = i;
1916 			return -1;
1917 		}
1918 		i++;
1919 	}
1920 
1921 	cb->args[4] = i;
1922 	return skb->len;
1923 }
1924 
1925 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
1926 		   struct netlink_callback *cb)
1927 {
1928 	struct leaf *l;
1929 	struct trie *t = (struct trie *) tb->tb_data;
1930 	t_key key = cb->args[2];
1931 	int count = cb->args[3];
1932 
1933 	rcu_read_lock();
1934 	/* Dump starting at last key.
1935 	 * Note: 0.0.0.0/0 (ie default) is first key.
1936 	 */
1937 	if (count == 0)
1938 		l = trie_firstleaf(t);
1939 	else {
1940 		/* Normally, continue from last key, but if that is missing
1941 		 * fallback to using slow rescan
1942 		 */
1943 		l = fib_find_node(t, key);
1944 		if (!l)
1945 			l = trie_leafindex(t, count);
1946 	}
1947 
1948 	while (l) {
1949 		cb->args[2] = l->key;
1950 		if (fn_trie_dump_leaf(l, tb, skb, cb) < 0) {
1951 			cb->args[3] = count;
1952 			rcu_read_unlock();
1953 			return -1;
1954 		}
1955 
1956 		++count;
1957 		l = trie_nextleaf(l);
1958 		memset(&cb->args[4], 0,
1959 		       sizeof(cb->args) - 4*sizeof(cb->args[0]));
1960 	}
1961 	cb->args[3] = count;
1962 	rcu_read_unlock();
1963 
1964 	return skb->len;
1965 }
1966 
1967 void __init fib_trie_init(void)
1968 {
1969 	fn_alias_kmem = kmem_cache_create("ip_fib_alias",
1970 					  sizeof(struct fib_alias),
1971 					  0, SLAB_PANIC, NULL);
1972 
1973 	trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
1974 					   max(sizeof(struct leaf),
1975 					       sizeof(struct leaf_info)),
1976 					   0, SLAB_PANIC, NULL);
1977 }
1978 
1979 
1980 struct fib_table *fib_trie_table(u32 id)
1981 {
1982 	struct fib_table *tb;
1983 	struct trie *t;
1984 
1985 	tb = kmalloc(sizeof(struct fib_table) + sizeof(struct trie),
1986 		     GFP_KERNEL);
1987 	if (tb == NULL)
1988 		return NULL;
1989 
1990 	tb->tb_id = id;
1991 	tb->tb_default = -1;
1992 	tb->tb_num_default = 0;
1993 
1994 	t = (struct trie *) tb->tb_data;
1995 	memset(t, 0, sizeof(*t));
1996 
1997 	return tb;
1998 }
1999 
2000 #ifdef CONFIG_PROC_FS
2001 /* Depth first Trie walk iterator */
2002 struct fib_trie_iter {
2003 	struct seq_net_private p;
2004 	struct fib_table *tb;
2005 	struct tnode *tnode;
2006 	unsigned int index;
2007 	unsigned int depth;
2008 };
2009 
2010 static struct rt_trie_node *fib_trie_get_next(struct fib_trie_iter *iter)
2011 {
2012 	struct tnode *tn = iter->tnode;
2013 	unsigned int cindex = iter->index;
2014 	struct tnode *p;
2015 
2016 	/* A single entry routing table */
2017 	if (!tn)
2018 		return NULL;
2019 
2020 	pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2021 		 iter->tnode, iter->index, iter->depth);
2022 rescan:
2023 	while (cindex < (1<<tn->bits)) {
2024 		struct rt_trie_node *n = tnode_get_child_rcu(tn, cindex);
2025 
2026 		if (n) {
2027 			if (IS_LEAF(n)) {
2028 				iter->tnode = tn;
2029 				iter->index = cindex + 1;
2030 			} else {
2031 				/* push down one level */
2032 				iter->tnode = (struct tnode *) n;
2033 				iter->index = 0;
2034 				++iter->depth;
2035 			}
2036 			return n;
2037 		}
2038 
2039 		++cindex;
2040 	}
2041 
2042 	/* Current node exhausted, pop back up */
2043 	p = node_parent_rcu((struct rt_trie_node *)tn);
2044 	if (p) {
2045 		cindex = tkey_extract_bits(tn->key, p->pos, p->bits)+1;
2046 		tn = p;
2047 		--iter->depth;
2048 		goto rescan;
2049 	}
2050 
2051 	/* got root? */
2052 	return NULL;
2053 }
2054 
2055 static struct rt_trie_node *fib_trie_get_first(struct fib_trie_iter *iter,
2056 				       struct trie *t)
2057 {
2058 	struct rt_trie_node *n;
2059 
2060 	if (!t)
2061 		return NULL;
2062 
2063 	n = rcu_dereference(t->trie);
2064 	if (!n)
2065 		return NULL;
2066 
2067 	if (IS_TNODE(n)) {
2068 		iter->tnode = (struct tnode *) n;
2069 		iter->index = 0;
2070 		iter->depth = 1;
2071 	} else {
2072 		iter->tnode = NULL;
2073 		iter->index = 0;
2074 		iter->depth = 0;
2075 	}
2076 
2077 	return n;
2078 }
2079 
2080 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2081 {
2082 	struct rt_trie_node *n;
2083 	struct fib_trie_iter iter;
2084 
2085 	memset(s, 0, sizeof(*s));
2086 
2087 	rcu_read_lock();
2088 	for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2089 		if (IS_LEAF(n)) {
2090 			struct leaf *l = (struct leaf *)n;
2091 			struct leaf_info *li;
2092 			struct hlist_node *tmp;
2093 
2094 			s->leaves++;
2095 			s->totdepth += iter.depth;
2096 			if (iter.depth > s->maxdepth)
2097 				s->maxdepth = iter.depth;
2098 
2099 			hlist_for_each_entry_rcu(li, tmp, &l->list, hlist)
2100 				++s->prefixes;
2101 		} else {
2102 			const struct tnode *tn = (const struct tnode *) n;
2103 			int i;
2104 
2105 			s->tnodes++;
2106 			if (tn->bits < MAX_STAT_DEPTH)
2107 				s->nodesizes[tn->bits]++;
2108 
2109 			for (i = 0; i < (1<<tn->bits); i++)
2110 				if (!tn->child[i])
2111 					s->nullpointers++;
2112 		}
2113 	}
2114 	rcu_read_unlock();
2115 }
2116 
2117 /*
2118  *	This outputs /proc/net/fib_triestats
2119  */
2120 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2121 {
2122 	unsigned int i, max, pointers, bytes, avdepth;
2123 
2124 	if (stat->leaves)
2125 		avdepth = stat->totdepth*100 / stat->leaves;
2126 	else
2127 		avdepth = 0;
2128 
2129 	seq_printf(seq, "\tAver depth:     %u.%02d\n",
2130 		   avdepth / 100, avdepth % 100);
2131 	seq_printf(seq, "\tMax depth:      %u\n", stat->maxdepth);
2132 
2133 	seq_printf(seq, "\tLeaves:         %u\n", stat->leaves);
2134 	bytes = sizeof(struct leaf) * stat->leaves;
2135 
2136 	seq_printf(seq, "\tPrefixes:       %u\n", stat->prefixes);
2137 	bytes += sizeof(struct leaf_info) * stat->prefixes;
2138 
2139 	seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2140 	bytes += sizeof(struct tnode) * stat->tnodes;
2141 
2142 	max = MAX_STAT_DEPTH;
2143 	while (max > 0 && stat->nodesizes[max-1] == 0)
2144 		max--;
2145 
2146 	pointers = 0;
2147 	for (i = 1; i <= max; i++)
2148 		if (stat->nodesizes[i] != 0) {
2149 			seq_printf(seq, "  %u: %u",  i, stat->nodesizes[i]);
2150 			pointers += (1<<i) * stat->nodesizes[i];
2151 		}
2152 	seq_putc(seq, '\n');
2153 	seq_printf(seq, "\tPointers: %u\n", pointers);
2154 
2155 	bytes += sizeof(struct rt_trie_node *) * pointers;
2156 	seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2157 	seq_printf(seq, "Total size: %u  kB\n", (bytes + 1023) / 1024);
2158 }
2159 
2160 #ifdef CONFIG_IP_FIB_TRIE_STATS
2161 static void trie_show_usage(struct seq_file *seq,
2162 			    const struct trie_use_stats *stats)
2163 {
2164 	seq_printf(seq, "\nCounters:\n---------\n");
2165 	seq_printf(seq, "gets = %u\n", stats->gets);
2166 	seq_printf(seq, "backtracks = %u\n", stats->backtrack);
2167 	seq_printf(seq, "semantic match passed = %u\n",
2168 		   stats->semantic_match_passed);
2169 	seq_printf(seq, "semantic match miss = %u\n",
2170 		   stats->semantic_match_miss);
2171 	seq_printf(seq, "null node hit= %u\n", stats->null_node_hit);
2172 	seq_printf(seq, "skipped node resize = %u\n\n",
2173 		   stats->resize_node_skipped);
2174 }
2175 #endif /*  CONFIG_IP_FIB_TRIE_STATS */
2176 
2177 static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2178 {
2179 	if (tb->tb_id == RT_TABLE_LOCAL)
2180 		seq_puts(seq, "Local:\n");
2181 	else if (tb->tb_id == RT_TABLE_MAIN)
2182 		seq_puts(seq, "Main:\n");
2183 	else
2184 		seq_printf(seq, "Id %d:\n", tb->tb_id);
2185 }
2186 
2187 
2188 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2189 {
2190 	struct net *net = (struct net *)seq->private;
2191 	unsigned int h;
2192 
2193 	seq_printf(seq,
2194 		   "Basic info: size of leaf:"
2195 		   " %Zd bytes, size of tnode: %Zd bytes.\n",
2196 		   sizeof(struct leaf), sizeof(struct tnode));
2197 
2198 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2199 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2200 		struct hlist_node *node;
2201 		struct fib_table *tb;
2202 
2203 		hlist_for_each_entry_rcu(tb, node, head, tb_hlist) {
2204 			struct trie *t = (struct trie *) tb->tb_data;
2205 			struct trie_stat stat;
2206 
2207 			if (!t)
2208 				continue;
2209 
2210 			fib_table_print(seq, tb);
2211 
2212 			trie_collect_stats(t, &stat);
2213 			trie_show_stats(seq, &stat);
2214 #ifdef CONFIG_IP_FIB_TRIE_STATS
2215 			trie_show_usage(seq, &t->stats);
2216 #endif
2217 		}
2218 	}
2219 
2220 	return 0;
2221 }
2222 
2223 static int fib_triestat_seq_open(struct inode *inode, struct file *file)
2224 {
2225 	return single_open_net(inode, file, fib_triestat_seq_show);
2226 }
2227 
2228 static const struct file_operations fib_triestat_fops = {
2229 	.owner	= THIS_MODULE,
2230 	.open	= fib_triestat_seq_open,
2231 	.read	= seq_read,
2232 	.llseek	= seq_lseek,
2233 	.release = single_release_net,
2234 };
2235 
2236 static struct rt_trie_node *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2237 {
2238 	struct fib_trie_iter *iter = seq->private;
2239 	struct net *net = seq_file_net(seq);
2240 	loff_t idx = 0;
2241 	unsigned int h;
2242 
2243 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2244 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2245 		struct hlist_node *node;
2246 		struct fib_table *tb;
2247 
2248 		hlist_for_each_entry_rcu(tb, node, head, tb_hlist) {
2249 			struct rt_trie_node *n;
2250 
2251 			for (n = fib_trie_get_first(iter,
2252 						    (struct trie *) tb->tb_data);
2253 			     n; n = fib_trie_get_next(iter))
2254 				if (pos == idx++) {
2255 					iter->tb = tb;
2256 					return n;
2257 				}
2258 		}
2259 	}
2260 
2261 	return NULL;
2262 }
2263 
2264 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2265 	__acquires(RCU)
2266 {
2267 	rcu_read_lock();
2268 	return fib_trie_get_idx(seq, *pos);
2269 }
2270 
2271 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2272 {
2273 	struct fib_trie_iter *iter = seq->private;
2274 	struct net *net = seq_file_net(seq);
2275 	struct fib_table *tb = iter->tb;
2276 	struct hlist_node *tb_node;
2277 	unsigned int h;
2278 	struct rt_trie_node *n;
2279 
2280 	++*pos;
2281 	/* next node in same table */
2282 	n = fib_trie_get_next(iter);
2283 	if (n)
2284 		return n;
2285 
2286 	/* walk rest of this hash chain */
2287 	h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2288 	while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2289 		tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2290 		n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2291 		if (n)
2292 			goto found;
2293 	}
2294 
2295 	/* new hash chain */
2296 	while (++h < FIB_TABLE_HASHSZ) {
2297 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2298 		hlist_for_each_entry_rcu(tb, tb_node, head, tb_hlist) {
2299 			n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2300 			if (n)
2301 				goto found;
2302 		}
2303 	}
2304 	return NULL;
2305 
2306 found:
2307 	iter->tb = tb;
2308 	return n;
2309 }
2310 
2311 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2312 	__releases(RCU)
2313 {
2314 	rcu_read_unlock();
2315 }
2316 
2317 static void seq_indent(struct seq_file *seq, int n)
2318 {
2319 	while (n-- > 0)
2320 		seq_puts(seq, "   ");
2321 }
2322 
2323 static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2324 {
2325 	switch (s) {
2326 	case RT_SCOPE_UNIVERSE: return "universe";
2327 	case RT_SCOPE_SITE:	return "site";
2328 	case RT_SCOPE_LINK:	return "link";
2329 	case RT_SCOPE_HOST:	return "host";
2330 	case RT_SCOPE_NOWHERE:	return "nowhere";
2331 	default:
2332 		snprintf(buf, len, "scope=%d", s);
2333 		return buf;
2334 	}
2335 }
2336 
2337 static const char *const rtn_type_names[__RTN_MAX] = {
2338 	[RTN_UNSPEC] = "UNSPEC",
2339 	[RTN_UNICAST] = "UNICAST",
2340 	[RTN_LOCAL] = "LOCAL",
2341 	[RTN_BROADCAST] = "BROADCAST",
2342 	[RTN_ANYCAST] = "ANYCAST",
2343 	[RTN_MULTICAST] = "MULTICAST",
2344 	[RTN_BLACKHOLE] = "BLACKHOLE",
2345 	[RTN_UNREACHABLE] = "UNREACHABLE",
2346 	[RTN_PROHIBIT] = "PROHIBIT",
2347 	[RTN_THROW] = "THROW",
2348 	[RTN_NAT] = "NAT",
2349 	[RTN_XRESOLVE] = "XRESOLVE",
2350 };
2351 
2352 static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2353 {
2354 	if (t < __RTN_MAX && rtn_type_names[t])
2355 		return rtn_type_names[t];
2356 	snprintf(buf, len, "type %u", t);
2357 	return buf;
2358 }
2359 
2360 /* Pretty print the trie */
2361 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2362 {
2363 	const struct fib_trie_iter *iter = seq->private;
2364 	struct rt_trie_node *n = v;
2365 
2366 	if (!node_parent_rcu(n))
2367 		fib_table_print(seq, iter->tb);
2368 
2369 	if (IS_TNODE(n)) {
2370 		struct tnode *tn = (struct tnode *) n;
2371 		__be32 prf = htonl(mask_pfx(tn->key, tn->pos));
2372 
2373 		seq_indent(seq, iter->depth-1);
2374 		seq_printf(seq, "  +-- %pI4/%d %d %d %d\n",
2375 			   &prf, tn->pos, tn->bits, tn->full_children,
2376 			   tn->empty_children);
2377 
2378 	} else {
2379 		struct leaf *l = (struct leaf *) n;
2380 		struct leaf_info *li;
2381 		struct hlist_node *node;
2382 		__be32 val = htonl(l->key);
2383 
2384 		seq_indent(seq, iter->depth);
2385 		seq_printf(seq, "  |-- %pI4\n", &val);
2386 
2387 		hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
2388 			struct fib_alias *fa;
2389 
2390 			list_for_each_entry_rcu(fa, &li->falh, fa_list) {
2391 				char buf1[32], buf2[32];
2392 
2393 				seq_indent(seq, iter->depth+1);
2394 				seq_printf(seq, "  /%d %s %s", li->plen,
2395 					   rtn_scope(buf1, sizeof(buf1),
2396 						     fa->fa_info->fib_scope),
2397 					   rtn_type(buf2, sizeof(buf2),
2398 						    fa->fa_type));
2399 				if (fa->fa_tos)
2400 					seq_printf(seq, " tos=%d", fa->fa_tos);
2401 				seq_putc(seq, '\n');
2402 			}
2403 		}
2404 	}
2405 
2406 	return 0;
2407 }
2408 
2409 static const struct seq_operations fib_trie_seq_ops = {
2410 	.start  = fib_trie_seq_start,
2411 	.next   = fib_trie_seq_next,
2412 	.stop   = fib_trie_seq_stop,
2413 	.show   = fib_trie_seq_show,
2414 };
2415 
2416 static int fib_trie_seq_open(struct inode *inode, struct file *file)
2417 {
2418 	return seq_open_net(inode, file, &fib_trie_seq_ops,
2419 			    sizeof(struct fib_trie_iter));
2420 }
2421 
2422 static const struct file_operations fib_trie_fops = {
2423 	.owner  = THIS_MODULE,
2424 	.open   = fib_trie_seq_open,
2425 	.read   = seq_read,
2426 	.llseek = seq_lseek,
2427 	.release = seq_release_net,
2428 };
2429 
2430 struct fib_route_iter {
2431 	struct seq_net_private p;
2432 	struct trie *main_trie;
2433 	loff_t	pos;
2434 	t_key	key;
2435 };
2436 
2437 static struct leaf *fib_route_get_idx(struct fib_route_iter *iter, loff_t pos)
2438 {
2439 	struct leaf *l = NULL;
2440 	struct trie *t = iter->main_trie;
2441 
2442 	/* use cache location of last found key */
2443 	if (iter->pos > 0 && pos >= iter->pos && (l = fib_find_node(t, iter->key)))
2444 		pos -= iter->pos;
2445 	else {
2446 		iter->pos = 0;
2447 		l = trie_firstleaf(t);
2448 	}
2449 
2450 	while (l && pos-- > 0) {
2451 		iter->pos++;
2452 		l = trie_nextleaf(l);
2453 	}
2454 
2455 	if (l)
2456 		iter->key = pos;	/* remember it */
2457 	else
2458 		iter->pos = 0;		/* forget it */
2459 
2460 	return l;
2461 }
2462 
2463 static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2464 	__acquires(RCU)
2465 {
2466 	struct fib_route_iter *iter = seq->private;
2467 	struct fib_table *tb;
2468 
2469 	rcu_read_lock();
2470 	tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2471 	if (!tb)
2472 		return NULL;
2473 
2474 	iter->main_trie = (struct trie *) tb->tb_data;
2475 	if (*pos == 0)
2476 		return SEQ_START_TOKEN;
2477 	else
2478 		return fib_route_get_idx(iter, *pos - 1);
2479 }
2480 
2481 static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2482 {
2483 	struct fib_route_iter *iter = seq->private;
2484 	struct leaf *l = v;
2485 
2486 	++*pos;
2487 	if (v == SEQ_START_TOKEN) {
2488 		iter->pos = 0;
2489 		l = trie_firstleaf(iter->main_trie);
2490 	} else {
2491 		iter->pos++;
2492 		l = trie_nextleaf(l);
2493 	}
2494 
2495 	if (l)
2496 		iter->key = l->key;
2497 	else
2498 		iter->pos = 0;
2499 	return l;
2500 }
2501 
2502 static void fib_route_seq_stop(struct seq_file *seq, void *v)
2503 	__releases(RCU)
2504 {
2505 	rcu_read_unlock();
2506 }
2507 
2508 static unsigned int fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2509 {
2510 	unsigned int flags = 0;
2511 
2512 	if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2513 		flags = RTF_REJECT;
2514 	if (fi && fi->fib_nh->nh_gw)
2515 		flags |= RTF_GATEWAY;
2516 	if (mask == htonl(0xFFFFFFFF))
2517 		flags |= RTF_HOST;
2518 	flags |= RTF_UP;
2519 	return flags;
2520 }
2521 
2522 /*
2523  *	This outputs /proc/net/route.
2524  *	The format of the file is not supposed to be changed
2525  *	and needs to be same as fib_hash output to avoid breaking
2526  *	legacy utilities
2527  */
2528 static int fib_route_seq_show(struct seq_file *seq, void *v)
2529 {
2530 	struct leaf *l = v;
2531 	struct leaf_info *li;
2532 	struct hlist_node *node;
2533 
2534 	if (v == SEQ_START_TOKEN) {
2535 		seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2536 			   "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2537 			   "\tWindow\tIRTT");
2538 		return 0;
2539 	}
2540 
2541 	hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
2542 		struct fib_alias *fa;
2543 		__be32 mask, prefix;
2544 
2545 		mask = inet_make_mask(li->plen);
2546 		prefix = htonl(l->key);
2547 
2548 		list_for_each_entry_rcu(fa, &li->falh, fa_list) {
2549 			const struct fib_info *fi = fa->fa_info;
2550 			unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2551 			int len;
2552 
2553 			if (fa->fa_type == RTN_BROADCAST
2554 			    || fa->fa_type == RTN_MULTICAST)
2555 				continue;
2556 
2557 			if (fi)
2558 				seq_printf(seq,
2559 					 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2560 					 "%d\t%08X\t%d\t%u\t%u%n",
2561 					 fi->fib_dev ? fi->fib_dev->name : "*",
2562 					 prefix,
2563 					 fi->fib_nh->nh_gw, flags, 0, 0,
2564 					 fi->fib_priority,
2565 					 mask,
2566 					 (fi->fib_advmss ?
2567 					  fi->fib_advmss + 40 : 0),
2568 					 fi->fib_window,
2569 					 fi->fib_rtt >> 3, &len);
2570 			else
2571 				seq_printf(seq,
2572 					 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2573 					 "%d\t%08X\t%d\t%u\t%u%n",
2574 					 prefix, 0, flags, 0, 0, 0,
2575 					 mask, 0, 0, 0, &len);
2576 
2577 			seq_printf(seq, "%*s\n", 127 - len, "");
2578 		}
2579 	}
2580 
2581 	return 0;
2582 }
2583 
2584 static const struct seq_operations fib_route_seq_ops = {
2585 	.start  = fib_route_seq_start,
2586 	.next   = fib_route_seq_next,
2587 	.stop   = fib_route_seq_stop,
2588 	.show   = fib_route_seq_show,
2589 };
2590 
2591 static int fib_route_seq_open(struct inode *inode, struct file *file)
2592 {
2593 	return seq_open_net(inode, file, &fib_route_seq_ops,
2594 			    sizeof(struct fib_route_iter));
2595 }
2596 
2597 static const struct file_operations fib_route_fops = {
2598 	.owner  = THIS_MODULE,
2599 	.open   = fib_route_seq_open,
2600 	.read   = seq_read,
2601 	.llseek = seq_lseek,
2602 	.release = seq_release_net,
2603 };
2604 
2605 int __net_init fib_proc_init(struct net *net)
2606 {
2607 	if (!proc_net_fops_create(net, "fib_trie", S_IRUGO, &fib_trie_fops))
2608 		goto out1;
2609 
2610 	if (!proc_net_fops_create(net, "fib_triestat", S_IRUGO,
2611 				  &fib_triestat_fops))
2612 		goto out2;
2613 
2614 	if (!proc_net_fops_create(net, "route", S_IRUGO, &fib_route_fops))
2615 		goto out3;
2616 
2617 	return 0;
2618 
2619 out3:
2620 	proc_net_remove(net, "fib_triestat");
2621 out2:
2622 	proc_net_remove(net, "fib_trie");
2623 out1:
2624 	return -ENOMEM;
2625 }
2626 
2627 void __net_exit fib_proc_exit(struct net *net)
2628 {
2629 	proc_net_remove(net, "fib_trie");
2630 	proc_net_remove(net, "fib_triestat");
2631 	proc_net_remove(net, "route");
2632 }
2633 
2634 #endif /* CONFIG_PROC_FS */
2635