fib_trie.c 62.2 KB
Newer Older
1 2 3 4 5 6 7 8 9
/*
 *   This program is free software; you can redistribute it and/or
 *   modify it under the terms of the GNU General Public License
 *   as published by the Free Software Foundation; either version
 *   2 of the License, or (at your option) any later version.
 *
 *   Robert Olsson <robert.olsson@its.uu.se> Uppsala Universitet
 *     & Swedish University of Agricultural Sciences.
 *
10
 *   Jens Laas <jens.laas@data.slu.se> Swedish University of
11
 *     Agricultural Sciences.
12
 *
13 14 15
 *   Hans Liss <hans.liss@its.uu.se>  Uppsala Universitet
 *
 * This work is based on the LPC-trie which is originally descibed in:
16
 *
17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41
 * An experimental study of compression methods for dynamic tries
 * Stefan Nilsson and Matti Tikkanen. Algorithmica, 33(1):19-33, 2002.
 * http://www.nada.kth.se/~snilsson/public/papers/dyntrie2/
 *
 *
 * IP-address lookup using LC-tries. Stefan Nilsson and Gunnar Karlsson
 * IEEE Journal on Selected Areas in Communications, 17(6):1083-1092, June 1999
 *
 *
 * Code from fib_hash has been reused which includes the following header:
 *
 *
 * INET		An implementation of the TCP/IP protocol suite for the LINUX
 *		operating system.  INET is implemented using the  BSD Socket
 *		interface as the means of communication with the user level.
 *
 *		IPv4 FIB: lookup engine and maintenance routines.
 *
 *
 * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
 *
 *		This program is free software; you can redistribute it and/or
 *		modify it under the terms of the GNU General Public License
 *		as published by the Free Software Foundation; either version
 *		2 of the License, or (at your option) any later version.
R
Robert Olsson 已提交
42 43 44 45 46 47 48
 *
 * Substantial contributions to this work comes from:
 *
 *		David S. Miller, <davem@davemloft.net>
 *		Stephen Hemminger <shemminger@osdl.org>
 *		Paul E. McKenney <paulmck@us.ibm.com>
 *		Patrick McHardy <kaber@trash.net>
49 50
 */

R
Robert Olsson 已提交
51
#define VERSION "0.408"
52 53 54

#include <asm/uaccess.h>
#include <asm/system.h>
J
Jiri Slaby 已提交
55
#include <linux/bitops.h>
56 57 58 59 60 61 62 63 64
#include <linux/types.h>
#include <linux/kernel.h>
#include <linux/mm.h>
#include <linux/string.h>
#include <linux/socket.h>
#include <linux/sockios.h>
#include <linux/errno.h>
#include <linux/in.h>
#include <linux/inet.h>
S
Stephen Hemminger 已提交
65
#include <linux/inetdevice.h>
66 67 68
#include <linux/netdevice.h>
#include <linux/if_arp.h>
#include <linux/proc_fs.h>
R
Robert Olsson 已提交
69
#include <linux/rcupdate.h>
70 71 72 73
#include <linux/skbuff.h>
#include <linux/netlink.h>
#include <linux/init.h>
#include <linux/list.h>
74
#include <net/net_namespace.h>
75 76 77 78 79 80 81 82
#include <net/ip.h>
#include <net/protocol.h>
#include <net/route.h>
#include <net/tcp.h>
#include <net/sock.h>
#include <net/ip_fib.h>
#include "fib_lookup.h"

R
Robert Olsson 已提交
83
#define MAX_STAT_DEPTH 32
84 85 86 87 88 89 90 91

#define KEYLENGTH (8*sizeof(t_key))

typedef unsigned int t_key;

#define T_TNODE 0
#define T_LEAF  1
#define NODE_TYPE_MASK	0x1UL
R
Robert Olsson 已提交
92 93
#define NODE_TYPE(node) ((node)->parent & NODE_TYPE_MASK)

O
Olof Johansson 已提交
94 95
#define IS_TNODE(n) (!(n->parent & T_LEAF))
#define IS_LEAF(n) (n->parent & T_LEAF)
96 97

struct node {
O
Olof Johansson 已提交
98
	unsigned long parent;
99
	t_key key;
100 101 102
};

struct leaf {
O
Olof Johansson 已提交
103
	unsigned long parent;
104
	t_key key;
105
	struct hlist_head list;
R
Robert Olsson 已提交
106
	struct rcu_head rcu;
107 108 109 110
};

struct leaf_info {
	struct hlist_node hlist;
R
Robert Olsson 已提交
111
	struct rcu_head rcu;
112 113 114 115 116
	int plen;
	struct list_head falh;
};

struct tnode {
O
Olof Johansson 已提交
117
	unsigned long parent;
118
	t_key key;
119 120
	unsigned char pos;		/* 2log(KEYLENGTH) bits needed */
	unsigned char bits;		/* 2log(KEYLENGTH) bits needed */
121 122
	unsigned int full_children;	/* KEYLENGTH bits needed */
	unsigned int empty_children;	/* KEYLENGTH bits needed */
123 124 125
	union {
		struct rcu_head rcu;
		struct work_struct work;
J
Jarek Poplawski 已提交
126
		struct tnode *tnode_free;
127
	};
O
Olof Johansson 已提交
128
	struct node *child[0];
129 130 131 132 133 134 135 136 137
};

#ifdef CONFIG_IP_FIB_TRIE_STATS
struct trie_use_stats {
	unsigned int gets;
	unsigned int backtrack;
	unsigned int semantic_match_passed;
	unsigned int semantic_match_miss;
	unsigned int null_node_hit;
138
	unsigned int resize_node_skipped;
139 140 141 142 143 144 145 146 147
};
#endif

struct trie_stat {
	unsigned int totdepth;
	unsigned int maxdepth;
	unsigned int tnodes;
	unsigned int leaves;
	unsigned int nullpointers;
148
	unsigned int prefixes;
R
Robert Olsson 已提交
149
	unsigned int nodesizes[MAX_STAT_DEPTH];
150
};
151 152

struct trie {
O
Olof Johansson 已提交
153
	struct node *trie;
154 155 156 157 158 159
#ifdef CONFIG_IP_FIB_TRIE_STATS
	struct trie_use_stats stats;
#endif
};

static void put_child(struct trie *t, struct tnode *tn, int i, struct node *n);
160 161
static void tnode_put_child_reorg(struct tnode *tn, int i, struct node *n,
				  int wasfull);
162
static struct node *resize(struct trie *t, struct tnode *tn);
163 164
static struct tnode *inflate(struct trie *t, struct tnode *tn);
static struct tnode *halve(struct trie *t, struct tnode *tn);
J
Jarek Poplawski 已提交
165 166
/* tnodes to free after resize(); protected by RTNL */
static struct tnode *tnode_free_head;
167 168 169 170 171 172 173 174
static size_t tnode_free_size;

/*
 * synchronize_rcu after call_rcu for that many pages; it should be especially
 * useful before resizing the root node with PREEMPT_NONE configs; the value was
 * obtained experimentally, aiming to avoid visible slowdown.
 */
static const int sync_pages = 128;
175

176
static struct kmem_cache *fn_alias_kmem __read_mostly;
177
static struct kmem_cache *trie_leaf_kmem __read_mostly;
178

S
Stephen Hemminger 已提交
179 180
static inline struct tnode *node_parent(struct node *node)
{
181 182 183 184 185 186
	return (struct tnode *)(node->parent & ~NODE_TYPE_MASK);
}

static inline struct tnode *node_parent_rcu(struct node *node)
{
	struct tnode *ret = node_parent(node);
S
Stephen Hemminger 已提交
187 188 189 190

	return rcu_dereference(ret);
}

191 192 193
/* Same as rcu_assign_pointer
 * but that macro() assumes that value is a pointer.
 */
S
Stephen Hemminger 已提交
194 195
static inline void node_set_parent(struct node *node, struct tnode *ptr)
{
196 197
	smp_wmb();
	node->parent = (unsigned long)ptr | NODE_TYPE(node);
S
Stephen Hemminger 已提交
198
}
R
Robert Olsson 已提交
199

200 201 202
static inline struct node *tnode_get_child(struct tnode *tn, unsigned int i)
{
	BUG_ON(i >= 1U << tn->bits);
R
Robert Olsson 已提交
203

204 205 206 207
	return tn->child[i];
}

static inline struct node *tnode_get_child_rcu(struct tnode *tn, unsigned int i)
208
{
209
	struct node *ret = tnode_get_child(tn, i);
210

211
	return rcu_dereference(ret);
212 213
}

S
Stephen Hemmigner 已提交
214
static inline int tnode_child_length(const struct tnode *tn)
215
{
O
Olof Johansson 已提交
216
	return 1 << tn->bits;
217 218
}

S
Stephen Hemminger 已提交
219 220 221 222 223
static inline t_key mask_pfx(t_key k, unsigned short l)
{
	return (l == 0) ? 0 : k >> (KEYLENGTH-l) << (KEYLENGTH-l);
}

224 225
static inline t_key tkey_extract_bits(t_key a, int offset, int bits)
{
O
Olof Johansson 已提交
226
	if (offset < KEYLENGTH)
227
		return ((t_key)(a << offset)) >> (KEYLENGTH - bits);
O
Olof Johansson 已提交
228
	else
229 230 231 232 233
		return 0;
}

static inline int tkey_equals(t_key a, t_key b)
{
234
	return a == b;
235 236 237 238
}

static inline int tkey_sub_equals(t_key a, int offset, int bits, t_key b)
{
239 240
	if (bits == 0 || offset >= KEYLENGTH)
		return 1;
O
Olof Johansson 已提交
241 242
	bits = bits > KEYLENGTH ? KEYLENGTH : bits;
	return ((a ^ b) << offset) >> (KEYLENGTH - bits) == 0;
243
}
244 245 246 247 248 249

static inline int tkey_mismatch(t_key a, int offset, t_key b)
{
	t_key diff = a ^ b;
	int i = offset;

250 251 252
	if (!diff)
		return 0;
	while ((diff << i) >> (KEYLENGTH-1) == 0)
253 254 255 256 257
		i++;
	return i;
}

/*
258 259
  To understand this stuff, an understanding of keys and all their bits is
  necessary. Every node in the trie has a key associated with it, but not
260 261 262 263
  all of the bits in that key are significant.

  Consider a node 'n' and its parent 'tp'.

264 265 266 267 268
  If n is a leaf, every bit in its key is significant. Its presence is
  necessitated by path compression, since during a tree traversal (when
  searching for a leaf - unless we are doing an insertion) we will completely
  ignore all skipped bits we encounter. Thus we need to verify, at the end of
  a potentially successful search, that we have indeed been walking the
269 270
  correct key path.

271 272 273 274 275
  Note that we can never "miss" the correct key in the tree if present by
  following the wrong path. Path compression ensures that segments of the key
  that are the same for all keys with a given prefix are skipped, but the
  skipped part *is* identical for each node in the subtrie below the skipped
  bit! trie_insert() in this implementation takes care of that - note the
276 277
  call to tkey_sub_equals() in trie_insert().

278
  if n is an internal node - a 'tnode' here, the various parts of its key
279 280
  have many different meanings.

281
  Example:
282 283 284
  _________________________________________________________________
  | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
  -----------------------------------------------------------------
285
    0   1   2   3   4   5   6   7   8   9  10  11  12  13  14  15
286 287 288 289 290 291 292 293 294

  _________________________________________________________________
  | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
  -----------------------------------------------------------------
   16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31

  tp->pos = 7
  tp->bits = 3
  n->pos = 15
O
Olof Johansson 已提交
295
  n->bits = 4
296

297 298
  First, let's just ignore the bits that come before the parent tp, that is
  the bits from 0 to (tp->pos-1). They are *known* but at this point we do
299 300 301
  not use them for anything.

  The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
302
  index into the parent's child array. That is, they will be used to find
303 304 305 306 307
  'n' among tp's children.

  The bits from (tp->pos + tp->bits) to (n->pos - 1) - "S" - are skipped bits
  for the node n.

308
  All the bits we have seen so far are significant to the node n. The rest
309 310
  of the bits are really not needed or indeed known in n->key.

311
  The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
312
  n's child array, and will of course be different for each child.
313

314

315 316 317 318 319
  The rest of the bits, from (n->pos + n->bits) onward, are completely unknown
  at this point.

*/

S
Stephen Hemminger 已提交
320
static inline void check_tnode(const struct tnode *tn)
321
{
S
Stephen Hemminger 已提交
322
	WARN_ON(tn && tn->pos+tn->bits > 32);
323 324
}

325 326
static const int halve_threshold = 25;
static const int inflate_threshold = 50;
327 328
static const int halve_threshold_root = 15;
static const int inflate_threshold_root = 25;
329

R
Robert Olsson 已提交
330 331

static void __alias_free_mem(struct rcu_head *head)
332
{
R
Robert Olsson 已提交
333 334
	struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
	kmem_cache_free(fn_alias_kmem, fa);
335 336
}

R
Robert Olsson 已提交
337
static inline void alias_free_mem_rcu(struct fib_alias *fa)
338
{
R
Robert Olsson 已提交
339 340
	call_rcu(&fa->rcu, __alias_free_mem);
}
O
Olof Johansson 已提交
341

R
Robert Olsson 已提交
342 343
static void __leaf_free_rcu(struct rcu_head *head)
{
344 345
	struct leaf *l = container_of(head, struct leaf, rcu);
	kmem_cache_free(trie_leaf_kmem, l);
R
Robert Olsson 已提交
346
}
O
Olof Johansson 已提交
347

348 349 350 351 352
static inline void free_leaf(struct leaf *l)
{
	call_rcu_bh(&l->rcu, __leaf_free_rcu);
}

R
Robert Olsson 已提交
353
static void __leaf_info_free_rcu(struct rcu_head *head)
354
{
R
Robert Olsson 已提交
355
	kfree(container_of(head, struct leaf_info, rcu));
356 357
}

R
Robert Olsson 已提交
358
static inline void free_leaf_info(struct leaf_info *leaf)
359
{
R
Robert Olsson 已提交
360
	call_rcu(&leaf->rcu, __leaf_info_free_rcu);
361 362
}

363
static struct tnode *tnode_alloc(size_t size)
364
{
R
Robert Olsson 已提交
365
	if (size <= PAGE_SIZE)
366
		return kzalloc(size, GFP_KERNEL);
367 368 369
	else
		return __vmalloc(size, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL);
}
R
Robert Olsson 已提交
370

371 372 373 374
static void __tnode_vfree(struct work_struct *arg)
{
	struct tnode *tn = container_of(arg, struct tnode, work);
	vfree(tn);
375 376
}

R
Robert Olsson 已提交
377
static void __tnode_free_rcu(struct rcu_head *head)
378
{
R
Robert Olsson 已提交
379
	struct tnode *tn = container_of(head, struct tnode, rcu);
380 381
	size_t size = sizeof(struct tnode) +
		      (sizeof(struct node *) << tn->bits);
382 383 384

	if (size <= PAGE_SIZE)
		kfree(tn);
385 386 387 388
	else {
		INIT_WORK(&tn->work, __tnode_vfree);
		schedule_work(&tn->work);
	}
389 390
}

R
Robert Olsson 已提交
391 392
static inline void tnode_free(struct tnode *tn)
{
393 394 395
	if (IS_LEAF(tn))
		free_leaf((struct leaf *) tn);
	else
R
Robert Olsson 已提交
396
		call_rcu(&tn->rcu, __tnode_free_rcu);
R
Robert Olsson 已提交
397 398
}

J
Jarek Poplawski 已提交
399 400 401
static void tnode_free_safe(struct tnode *tn)
{
	BUG_ON(IS_LEAF(tn));
402 403
	tn->tnode_free = tnode_free_head;
	tnode_free_head = tn;
404 405
	tnode_free_size += sizeof(struct tnode) +
			   (sizeof(struct node *) << tn->bits);
J
Jarek Poplawski 已提交
406 407 408 409 410 411 412 413 414 415 416
}

static void tnode_free_flush(void)
{
	struct tnode *tn;

	while ((tn = tnode_free_head)) {
		tnode_free_head = tn->tnode_free;
		tn->tnode_free = NULL;
		tnode_free(tn);
	}
417 418 419 420 421

	if (tnode_free_size >= PAGE_SIZE * sync_pages) {
		tnode_free_size = 0;
		synchronize_rcu();
	}
J
Jarek Poplawski 已提交
422 423
}

R
Robert Olsson 已提交
424 425
static struct leaf *leaf_new(void)
{
426
	struct leaf *l = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
R
Robert Olsson 已提交
427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443
	if (l) {
		l->parent = T_LEAF;
		INIT_HLIST_HEAD(&l->list);
	}
	return l;
}

static struct leaf_info *leaf_info_new(int plen)
{
	struct leaf_info *li = kmalloc(sizeof(struct leaf_info),  GFP_KERNEL);
	if (li) {
		li->plen = plen;
		INIT_LIST_HEAD(&li->falh);
	}
	return li;
}

444
static struct tnode *tnode_new(t_key key, int pos, int bits)
445
{
446
	size_t sz = sizeof(struct tnode) + (sizeof(struct node *) << bits);
447
	struct tnode *tn = tnode_alloc(sz);
448

O
Olof Johansson 已提交
449
	if (tn) {
R
Robert Olsson 已提交
450
		tn->parent = T_TNODE;
451 452 453 454 455 456
		tn->pos = pos;
		tn->bits = bits;
		tn->key = key;
		tn->full_children = 0;
		tn->empty_children = 1<<bits;
	}
457

458 459
	pr_debug("AT %p s=%u %lu\n", tn, (unsigned int) sizeof(struct tnode),
		 (unsigned long) (sizeof(struct node) << bits));
460 461 462 463 464 465 466 467
	return tn;
}

/*
 * Check whether a tnode 'n' is "full", i.e. it is an internal node
 * and no bits are skipped. See discussion in dyntree paper p. 6
 */

S
Stephen Hemmigner 已提交
468
static inline int tnode_full(const struct tnode *tn, const struct node *n)
469
{
470
	if (n == NULL || IS_LEAF(n))
471 472 473 474 475
		return 0;

	return ((struct tnode *) n)->pos == tn->pos + tn->bits;
}

476 477
static inline void put_child(struct trie *t, struct tnode *tn, int i,
			     struct node *n)
478 479 480 481
{
	tnode_put_child_reorg(tn, i, n, -1);
}

482
 /*
483 484 485 486
  * Add a child at position i overwriting the old value.
  * Update the value of full_children and empty_children.
  */

487 488
static void tnode_put_child_reorg(struct tnode *tn, int i, struct node *n,
				  int wasfull)
489
{
R
Robert Olsson 已提交
490
	struct node *chi = tn->child[i];
491 492
	int isfull;

S
Stephen Hemminger 已提交
493 494
	BUG_ON(i >= 1<<tn->bits);

495 496 497 498 499
	/* update emptyChildren */
	if (n == NULL && chi != NULL)
		tn->empty_children++;
	else if (n != NULL && chi == NULL)
		tn->empty_children--;
500

501
	/* update fullChildren */
O
Olof Johansson 已提交
502
	if (wasfull == -1)
503 504 505
		wasfull = tnode_full(tn, chi);

	isfull = tnode_full(tn, n);
506
	if (wasfull && !isfull)
507
		tn->full_children--;
508
	else if (!wasfull && isfull)
509
		tn->full_children++;
O
Olof Johansson 已提交
510

511
	if (n)
S
Stephen Hemminger 已提交
512
		node_set_parent(n, tn);
513

R
Robert Olsson 已提交
514
	rcu_assign_pointer(tn->child[i], n);
515 516
}

517
static struct node *resize(struct trie *t, struct tnode *tn)
518 519
{
	int i;
520
	int err = 0;
521
	struct tnode *old_tn;
522 523
	int inflate_threshold_use;
	int halve_threshold_use;
R
Robert Olsson 已提交
524
	int max_resize;
525

526
	if (!tn)
527 528
		return NULL;

S
Stephen Hemminger 已提交
529 530
	pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
		 tn, inflate_threshold, halve_threshold);
531 532 533

	/* No children */
	if (tn->empty_children == tnode_child_length(tn)) {
J
Jarek Poplawski 已提交
534
		tnode_free_safe(tn);
535 536 537 538 539
		return NULL;
	}
	/* One child */
	if (tn->empty_children == tnode_child_length(tn) - 1)
		for (i = 0; i < tnode_child_length(tn); i++) {
O
Olof Johansson 已提交
540
			struct node *n;
541

O
Olof Johansson 已提交
542
			n = tn->child[i];
R
Robert Olsson 已提交
543
			if (!n)
O
Olof Johansson 已提交
544 545 546
				continue;

			/* compress one level */
S
Stephen Hemminger 已提交
547
			node_set_parent(n, NULL);
J
Jarek Poplawski 已提交
548
			tnode_free_safe(tn);
O
Olof Johansson 已提交
549
			return n;
550
		}
551
	/*
552 553 554 555 556
	 * Double as long as the resulting node has a number of
	 * nonempty nodes that are above the threshold.
	 */

	/*
557 558
	 * From "Implementing a dynamic compressed trie" by Stefan Nilsson of
	 * the Helsinki University of Technology and Matti Tikkanen of Nokia
559
	 * Telecommunications, page 6:
560
	 * "A node is doubled if the ratio of non-empty children to all
561 562
	 * children in the *doubled* node is at least 'high'."
	 *
563 564 565 566 567
	 * 'high' in this instance is the variable 'inflate_threshold'. It
	 * is expressed as a percentage, so we multiply it with
	 * tnode_child_length() and instead of multiplying by 2 (since the
	 * child array will be doubled by inflate()) and multiplying
	 * the left-hand side by 100 (to handle the percentage thing) we
568
	 * multiply the left-hand side by 50.
569 570 571 572
	 *
	 * The left-hand side may look a bit weird: tnode_child_length(tn)
	 * - tn->empty_children is of course the number of non-null children
	 * in the current node. tn->full_children is the number of "full"
573
	 * children, that is non-null tnodes with a skip value of 0.
574
	 * All of those will be doubled in the resulting inflated tnode, so
575
	 * we just count them one extra time here.
576
	 *
577
	 * A clearer way to write this would be:
578
	 *
579
	 * to_be_doubled = tn->full_children;
580
	 * not_to_be_doubled = tnode_child_length(tn) - tn->empty_children -
581 582 583 584
	 *     tn->full_children;
	 *
	 * new_child_length = tnode_child_length(tn) * 2;
	 *
585
	 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
586 587
	 *      new_child_length;
	 * if (new_fill_factor >= inflate_threshold)
588 589 590
	 *
	 * ...and so on, tho it would mess up the while () loop.
	 *
591 592 593
	 * anyway,
	 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
	 *      inflate_threshold
594
	 *
595 596 597
	 * avoid a division:
	 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
	 *      inflate_threshold * new_child_length
598
	 *
599
	 * expand not_to_be_doubled and to_be_doubled, and shorten:
600
	 * 100 * (tnode_child_length(tn) - tn->empty_children +
O
Olof Johansson 已提交
601
	 *    tn->full_children) >= inflate_threshold * new_child_length
602
	 *
603
	 * expand new_child_length:
604
	 * 100 * (tnode_child_length(tn) - tn->empty_children +
O
Olof Johansson 已提交
605
	 *    tn->full_children) >=
606
	 *      inflate_threshold * tnode_child_length(tn) * 2
607
	 *
608
	 * shorten again:
609
	 * 50 * (tn->full_children + tnode_child_length(tn) -
O
Olof Johansson 已提交
610
	 *    tn->empty_children) >= inflate_threshold *
611
	 *    tnode_child_length(tn)
612
	 *
613 614 615
	 */

	check_tnode(tn);
616

617 618
	/* Keep root node larger  */

S
Stephen Hemminger 已提交
619
	if (!tn->parent)
620
		inflate_threshold_use = inflate_threshold_root;
621
	else
622 623
		inflate_threshold_use = inflate_threshold;

624
	err = 0;
R
Robert Olsson 已提交
625 626
	max_resize = 10;
	while ((tn->full_children > 0 &&  max_resize-- &&
627 628 629
		50 * (tn->full_children + tnode_child_length(tn)
		      - tn->empty_children)
		>= inflate_threshold_use * tnode_child_length(tn))) {
630

631 632
		old_tn = tn;
		tn = inflate(t, tn);
633

634 635
		if (IS_ERR(tn)) {
			tn = old_tn;
636 637 638 639 640
#ifdef CONFIG_IP_FIB_TRIE_STATS
			t->stats.resize_node_skipped++;
#endif
			break;
		}
641 642
	}

R
Robert Olsson 已提交
643 644
	if (max_resize < 0) {
		if (!tn->parent)
645 646 647
			pr_warning("Fix inflate_threshold_root."
				   " Now=%d size=%d bits\n",
				   inflate_threshold_root, tn->bits);
R
Robert Olsson 已提交
648
		else
649 650 651
			pr_warning("Fix inflate_threshold."
				   " Now=%d size=%d bits\n",
				   inflate_threshold, tn->bits);
R
Robert Olsson 已提交
652 653
	}

654 655 656 657 658 659
	check_tnode(tn);

	/*
	 * Halve as long as the number of empty children in this
	 * node is above threshold.
	 */
660

661 662 663

	/* Keep root node larger  */

S
Stephen Hemminger 已提交
664
	if (!tn->parent)
665
		halve_threshold_use = halve_threshold_root;
666
	else
667 668
		halve_threshold_use = halve_threshold;

669
	err = 0;
R
Robert Olsson 已提交
670 671
	max_resize = 10;
	while (tn->bits > 1 &&  max_resize-- &&
672
	       100 * (tnode_child_length(tn) - tn->empty_children) <
673
	       halve_threshold_use * tnode_child_length(tn)) {
674

675 676 677 678
		old_tn = tn;
		tn = halve(t, tn);
		if (IS_ERR(tn)) {
			tn = old_tn;
679 680 681 682 683 684
#ifdef CONFIG_IP_FIB_TRIE_STATS
			t->stats.resize_node_skipped++;
#endif
			break;
		}
	}
685

R
Robert Olsson 已提交
686 687
	if (max_resize < 0) {
		if (!tn->parent)
688 689 690
			pr_warning("Fix halve_threshold_root."
				   " Now=%d size=%d bits\n",
				   halve_threshold_root, tn->bits);
R
Robert Olsson 已提交
691
		else
692 693 694
			pr_warning("Fix halve_threshold."
				   " Now=%d size=%d bits\n",
				   halve_threshold, tn->bits);
R
Robert Olsson 已提交
695
	}
696

697 698 699
	/* Only one child remains */
	if (tn->empty_children == tnode_child_length(tn) - 1)
		for (i = 0; i < tnode_child_length(tn); i++) {
O
Olof Johansson 已提交
700
			struct node *n;
701

O
Olof Johansson 已提交
702
			n = tn->child[i];
R
Robert Olsson 已提交
703
			if (!n)
O
Olof Johansson 已提交
704 705 706 707
				continue;

			/* compress one level */

S
Stephen Hemminger 已提交
708
			node_set_parent(n, NULL);
J
Jarek Poplawski 已提交
709
			tnode_free_safe(tn);
O
Olof Johansson 已提交
710
			return n;
711 712 713 714 715
		}

	return (struct node *) tn;
}

716
static struct tnode *inflate(struct trie *t, struct tnode *tn)
717 718 719 720 721
{
	struct tnode *oldtnode = tn;
	int olen = tnode_child_length(tn);
	int i;

S
Stephen Hemminger 已提交
722
	pr_debug("In inflate\n");
723 724 725

	tn = tnode_new(oldtnode->key, oldtnode->pos, oldtnode->bits + 1);

S
Stephen Hemminger 已提交
726
	if (!tn)
727
		return ERR_PTR(-ENOMEM);
728 729

	/*
730 731 732
	 * Preallocate and store tnodes before the actual work so we
	 * don't get into an inconsistent state if memory allocation
	 * fails. In case of failure we return the oldnode and  inflate
733 734
	 * of tnode is ignored.
	 */
O
Olof Johansson 已提交
735 736

	for (i = 0; i < olen; i++) {
737
		struct tnode *inode;
738

739
		inode = (struct tnode *) tnode_get_child(oldtnode, i);
740 741 742 743 744
		if (inode &&
		    IS_TNODE(inode) &&
		    inode->pos == oldtnode->pos + oldtnode->bits &&
		    inode->bits > 1) {
			struct tnode *left, *right;
S
Stephen Hemminger 已提交
745
			t_key m = ~0U << (KEYLENGTH - 1) >> inode->pos;
746

747 748
			left = tnode_new(inode->key&(~m), inode->pos + 1,
					 inode->bits - 1);
749 750
			if (!left)
				goto nomem;
O
Olof Johansson 已提交
751

752 753 754
			right = tnode_new(inode->key|m, inode->pos + 1,
					  inode->bits - 1);

755
			if (!right) {
756 757
				tnode_free(left);
				goto nomem;
758
			}
759 760 761 762 763 764

			put_child(t, tn, 2*i, (struct node *) left);
			put_child(t, tn, 2*i+1, (struct node *) right);
		}
	}

O
Olof Johansson 已提交
765
	for (i = 0; i < olen; i++) {
766
		struct tnode *inode;
767
		struct node *node = tnode_get_child(oldtnode, i);
O
Olof Johansson 已提交
768 769
		struct tnode *left, *right;
		int size, j;
770

771 772 773 774 775 776
		/* An empty child */
		if (node == NULL)
			continue;

		/* A leaf or an internal node with skipped bits */

777
		if (IS_LEAF(node) || ((struct tnode *) node)->pos >
778
		   tn->pos + tn->bits - 1) {
779 780 781
			if (tkey_extract_bits(node->key,
					      oldtnode->pos + oldtnode->bits,
					      1) == 0)
782 783 784 785 786 787 788 789 790 791 792 793 794
				put_child(t, tn, 2*i, node);
			else
				put_child(t, tn, 2*i+1, node);
			continue;
		}

		/* An internal node with two children */
		inode = (struct tnode *) node;

		if (inode->bits == 1) {
			put_child(t, tn, 2*i, inode->child[0]);
			put_child(t, tn, 2*i+1, inode->child[1]);

J
Jarek Poplawski 已提交
795
			tnode_free_safe(inode);
O
Olof Johansson 已提交
796
			continue;
797 798
		}

O
Olof Johansson 已提交
799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816
		/* An internal node with more than two children */

		/* We will replace this node 'inode' with two new
		 * ones, 'left' and 'right', each with half of the
		 * original children. The two new nodes will have
		 * a position one bit further down the key and this
		 * means that the "significant" part of their keys
		 * (see the discussion near the top of this file)
		 * will differ by one bit, which will be "0" in
		 * left's key and "1" in right's key. Since we are
		 * moving the key position by one step, the bit that
		 * we are moving away from - the bit at position
		 * (inode->pos) - is the one that will differ between
		 * left and right. So... we synthesize that bit in the
		 * two  new keys.
		 * The mask 'm' below will be a single "one" bit at
		 * the position (inode->pos)
		 */
817

O
Olof Johansson 已提交
818 819 820
		/* Use the old key, but set the new significant
		 *   bit to zero.
		 */
821

O
Olof Johansson 已提交
822 823
		left = (struct tnode *) tnode_get_child(tn, 2*i);
		put_child(t, tn, 2*i, NULL);
824

O
Olof Johansson 已提交
825
		BUG_ON(!left);
826

O
Olof Johansson 已提交
827 828
		right = (struct tnode *) tnode_get_child(tn, 2*i+1);
		put_child(t, tn, 2*i+1, NULL);
829

O
Olof Johansson 已提交
830
		BUG_ON(!right);
831

O
Olof Johansson 已提交
832 833 834 835
		size = tnode_child_length(left);
		for (j = 0; j < size; j++) {
			put_child(t, left, j, inode->child[j]);
			put_child(t, right, j, inode->child[j + size]);
836
		}
O
Olof Johansson 已提交
837 838 839
		put_child(t, tn, 2*i, resize(t, left));
		put_child(t, tn, 2*i+1, resize(t, right));

J
Jarek Poplawski 已提交
840
		tnode_free_safe(inode);
841
	}
J
Jarek Poplawski 已提交
842
	tnode_free_safe(oldtnode);
843
	return tn;
844 845 846 847 848
nomem:
	{
		int size = tnode_child_length(tn);
		int j;

S
Stephen Hemminger 已提交
849
		for (j = 0; j < size; j++)
850 851 852 853
			if (tn->child[j])
				tnode_free((struct tnode *)tn->child[j]);

		tnode_free(tn);
S
Stephen Hemminger 已提交
854

855 856
		return ERR_PTR(-ENOMEM);
	}
857 858
}

859
static struct tnode *halve(struct trie *t, struct tnode *tn)
860 861 862 863 864 865
{
	struct tnode *oldtnode = tn;
	struct node *left, *right;
	int i;
	int olen = tnode_child_length(tn);

S
Stephen Hemminger 已提交
866
	pr_debug("In halve\n");
867 868

	tn = tnode_new(oldtnode->key, oldtnode->pos, oldtnode->bits - 1);
869

870 871
	if (!tn)
		return ERR_PTR(-ENOMEM);
872 873

	/*
874 875 876
	 * Preallocate and store tnodes before the actual work so we
	 * don't get into an inconsistent state if memory allocation
	 * fails. In case of failure we return the oldnode and halve
877 878 879
	 * of tnode is ignored.
	 */

O
Olof Johansson 已提交
880
	for (i = 0; i < olen; i += 2) {
881 882
		left = tnode_get_child(oldtnode, i);
		right = tnode_get_child(oldtnode, i+1);
883

884
		/* Two nonempty children */
S
Stephen Hemminger 已提交
885
		if (left && right) {
886
			struct tnode *newn;
S
Stephen Hemminger 已提交
887

888
			newn = tnode_new(left->key, tn->pos + tn->bits, 1);
S
Stephen Hemminger 已提交
889 890

			if (!newn)
891
				goto nomem;
S
Stephen Hemminger 已提交
892

893
			put_child(t, tn, i/2, (struct node *)newn);
894 895 896
		}

	}
897

O
Olof Johansson 已提交
898 899 900
	for (i = 0; i < olen; i += 2) {
		struct tnode *newBinNode;

901 902
		left = tnode_get_child(oldtnode, i);
		right = tnode_get_child(oldtnode, i+1);
903

904 905 906 907 908
		/* At least one of the children is empty */
		if (left == NULL) {
			if (right == NULL)    /* Both are empty */
				continue;
			put_child(t, tn, i/2, right);
O
Olof Johansson 已提交
909
			continue;
S
Stephen Hemminger 已提交
910
		}
O
Olof Johansson 已提交
911 912

		if (right == NULL) {
913
			put_child(t, tn, i/2, left);
O
Olof Johansson 已提交
914 915
			continue;
		}
916

917
		/* Two nonempty children */
O
Olof Johansson 已提交
918 919 920 921 922
		newBinNode = (struct tnode *) tnode_get_child(tn, i/2);
		put_child(t, tn, i/2, NULL);
		put_child(t, newBinNode, 0, left);
		put_child(t, newBinNode, 1, right);
		put_child(t, tn, i/2, resize(t, newBinNode));
923
	}
J
Jarek Poplawski 已提交
924
	tnode_free_safe(oldtnode);
925
	return tn;
926 927 928 929 930
nomem:
	{
		int size = tnode_child_length(tn);
		int j;

S
Stephen Hemminger 已提交
931
		for (j = 0; j < size; j++)
932 933 934 935
			if (tn->child[j])
				tnode_free((struct tnode *)tn->child[j]);

		tnode_free(tn);
S
Stephen Hemminger 已提交
936

937 938
		return ERR_PTR(-ENOMEM);
	}
939 940
}

R
Robert Olsson 已提交
941
/* readside must use rcu_read_lock currently dump routines
R
Robert Olsson 已提交
942 943
 via get_fa_head and dump */

R
Robert Olsson 已提交
944
static struct leaf_info *find_leaf_info(struct leaf *l, int plen)
945
{
R
Robert Olsson 已提交
946
	struct hlist_head *head = &l->list;
947 948 949
	struct hlist_node *node;
	struct leaf_info *li;

R
Robert Olsson 已提交
950
	hlist_for_each_entry_rcu(li, node, head, hlist)
951
		if (li->plen == plen)
952
			return li;
O
Olof Johansson 已提交
953

954 955 956
	return NULL;
}

957
static inline struct list_head *get_fa_head(struct leaf *l, int plen)
958
{
R
Robert Olsson 已提交
959
	struct leaf_info *li = find_leaf_info(l, plen);
960

O
Olof Johansson 已提交
961 962
	if (!li)
		return NULL;
963

O
Olof Johansson 已提交
964
	return &li->falh;
965 966 967 968
}

static void insert_leaf_info(struct hlist_head *head, struct leaf_info *new)
{
969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985
	struct leaf_info *li = NULL, *last = NULL;
	struct hlist_node *node;

	if (hlist_empty(head)) {
		hlist_add_head_rcu(&new->hlist, head);
	} else {
		hlist_for_each_entry(li, node, head, hlist) {
			if (new->plen > li->plen)
				break;

			last = li;
		}
		if (last)
			hlist_add_after_rcu(&last->hlist, &new->hlist);
		else
			hlist_add_before_rcu(&new->hlist, &li->hlist);
	}
986 987
}

R
Robert Olsson 已提交
988 989
/* rcu_read_lock needs to be hold by caller from readside */

990 991 992 993 994 995 996 997
static struct leaf *
fib_find_node(struct trie *t, u32 key)
{
	int pos;
	struct tnode *tn;
	struct node *n;

	pos = 0;
R
Robert Olsson 已提交
998
	n = rcu_dereference(t->trie);
999 1000 1001

	while (n != NULL &&  NODE_TYPE(n) == T_TNODE) {
		tn = (struct tnode *) n;
O
Olof Johansson 已提交
1002

1003
		check_tnode(tn);
O
Olof Johansson 已提交
1004

1005
		if (tkey_sub_equals(tn->key, pos, tn->pos-pos, key)) {
O
Olof Johansson 已提交
1006
			pos = tn->pos + tn->bits;
1007 1008 1009 1010
			n = tnode_get_child_rcu(tn,
						tkey_extract_bits(key,
								  tn->pos,
								  tn->bits));
O
Olof Johansson 已提交
1011
		} else
1012 1013 1014 1015
			break;
	}
	/* Case we have found a leaf. Compare prefixes */

O
Olof Johansson 已提交
1016 1017 1018
	if (n != NULL && IS_LEAF(n) && tkey_equals(key, n->key))
		return (struct leaf *)n;

1019 1020 1021
	return NULL;
}

1022
static void trie_rebalance(struct trie *t, struct tnode *tn)
1023 1024
{
	int wasfull;
R
Robert Olsson 已提交
1025
	t_key cindex, key;
S
Stephen Hemminger 已提交
1026
	struct tnode *tp;
1027

R
Robert Olsson 已提交
1028 1029
	key = tn->key;

S
Stephen Hemminger 已提交
1030
	while (tn != NULL && (tp = node_parent((struct node *)tn)) != NULL) {
1031 1032
		cindex = tkey_extract_bits(key, tp->pos, tp->bits);
		wasfull = tnode_full(tp, tnode_get_child(tp, cindex));
1033 1034 1035 1036
		tn = (struct tnode *) resize(t, (struct tnode *)tn);

		tnode_put_child_reorg((struct tnode *)tp, cindex,
				      (struct node *)tn, wasfull);
O
Olof Johansson 已提交
1037

S
Stephen Hemminger 已提交
1038
		tp = node_parent((struct node *) tn);
1039 1040 1041
		if (!tp)
			rcu_assign_pointer(t->trie, (struct node *)tn);

J
Jarek Poplawski 已提交
1042
		tnode_free_flush();
S
Stephen Hemminger 已提交
1043
		if (!tp)
1044
			break;
S
Stephen Hemminger 已提交
1045
		tn = tp;
1046
	}
S
Stephen Hemminger 已提交
1047

1048
	/* Handle last (top) tnode */
1049
	if (IS_TNODE(tn))
1050
		tn = (struct tnode *)resize(t, (struct tnode *)tn);
1051

1052 1053 1054 1055
	rcu_assign_pointer(t->trie, (struct node *)tn);
	tnode_free_flush();

	return;
1056 1057
}

R
Robert Olsson 已提交
1058 1059
/* only used from updater-side */

1060
static struct list_head *fib_insert_node(struct trie *t, u32 key, int plen)
1061 1062 1063 1064 1065 1066
{
	int pos, newpos;
	struct tnode *tp = NULL, *tn = NULL;
	struct node *n;
	struct leaf *l;
	int missbit;
1067
	struct list_head *fa_head = NULL;
1068 1069 1070 1071
	struct leaf_info *li;
	t_key cindex;

	pos = 0;
1072
	n = t->trie;
1073

1074 1075
	/* If we point to NULL, stop. Either the tree is empty and we should
	 * just put a new leaf in if, or we have reached an empty child slot,
1076
	 * and we should just put our new leaf in that.
1077 1078
	 * If we point to a T_TNODE, check if it matches our key. Note that
	 * a T_TNODE might be skipping any number of bits - its 'pos' need
1079 1080
	 * not be the parent's 'pos'+'bits'!
	 *
1081
	 * If it does match the current key, get pos/bits from it, extract
1082 1083 1084 1085
	 * the index from our key, push the T_TNODE and walk the tree.
	 *
	 * If it doesn't, we have to replace it with a new T_TNODE.
	 *
1086 1087 1088
	 * If we point to a T_LEAF, it might or might not have the same key
	 * as we do. If it does, just change the value, update the T_LEAF's
	 * value, and return it.
1089 1090 1091 1092 1093
	 * If it doesn't, we need to replace it with a T_TNODE.
	 */

	while (n != NULL &&  NODE_TYPE(n) == T_TNODE) {
		tn = (struct tnode *) n;
O
Olof Johansson 已提交
1094

1095
		check_tnode(tn);
O
Olof Johansson 已提交
1096

1097
		if (tkey_sub_equals(tn->key, pos, tn->pos-pos, key)) {
1098
			tp = tn;
O
Olof Johansson 已提交
1099
			pos = tn->pos + tn->bits;
1100 1101 1102 1103
			n = tnode_get_child(tn,
					    tkey_extract_bits(key,
							      tn->pos,
							      tn->bits));
1104

S
Stephen Hemminger 已提交
1105
			BUG_ON(n && node_parent(n) != tn);
O
Olof Johansson 已提交
1106
		} else
1107 1108 1109 1110 1111 1112
			break;
	}

	/*
	 * n  ----> NULL, LEAF or TNODE
	 *
1113
	 * tp is n's (parent) ----> NULL or TNODE
1114 1115
	 */

O
Olof Johansson 已提交
1116
	BUG_ON(tp && IS_LEAF(tp));
1117 1118 1119

	/* Case 1: n is a leaf. Compare prefixes */

1120
	if (n != NULL && IS_LEAF(n) && tkey_equals(key, n->key)) {
1121
		l = (struct leaf *) n;
1122
		li = leaf_info_new(plen);
O
Olof Johansson 已提交
1123

1124 1125
		if (!li)
			return NULL;
1126 1127 1128 1129 1130 1131 1132

		fa_head = &li->falh;
		insert_leaf_info(&l->list, li);
		goto done;
	}
	l = leaf_new();

1133 1134
	if (!l)
		return NULL;
1135 1136 1137 1138

	l->key = key;
	li = leaf_info_new(plen);

1139
	if (!li) {
1140
		free_leaf(l);
1141
		return NULL;
1142
	}
1143 1144 1145 1146 1147

	fa_head = &li->falh;
	insert_leaf_info(&l->list, li);

	if (t->trie && n == NULL) {
O
Olof Johansson 已提交
1148
		/* Case 2: n is NULL, and will just insert a new leaf */
1149

S
Stephen Hemminger 已提交
1150
		node_set_parent((struct node *)l, tp);
1151

O
Olof Johansson 已提交
1152 1153 1154 1155
		cindex = tkey_extract_bits(key, tp->pos, tp->bits);
		put_child(t, (struct tnode *)tp, cindex, (struct node *)l);
	} else {
		/* Case 3: n is a LEAF or a TNODE and the key doesn't match. */
1156 1157
		/*
		 *  Add a new tnode here
1158 1159 1160 1161
		 *  first tnode need some special handling
		 */

		if (tp)
O
Olof Johansson 已提交
1162
			pos = tp->pos+tp->bits;
1163
		else
O
Olof Johansson 已提交
1164 1165
			pos = 0;

1166
		if (n) {
1167 1168
			newpos = tkey_mismatch(key, pos, n->key);
			tn = tnode_new(n->key, newpos, 1);
O
Olof Johansson 已提交
1169
		} else {
1170
			newpos = 0;
1171
			tn = tnode_new(key, newpos, 1); /* First tnode */
1172 1173
		}

1174
		if (!tn) {
1175
			free_leaf_info(li);
1176
			free_leaf(l);
1177
			return NULL;
O
Olof Johansson 已提交
1178 1179
		}

S
Stephen Hemminger 已提交
1180
		node_set_parent((struct node *)tn, tp);
1181

O
Olof Johansson 已提交
1182
		missbit = tkey_extract_bits(key, newpos, 1);
1183 1184 1185
		put_child(t, tn, missbit, (struct node *)l);
		put_child(t, tn, 1-missbit, n);

1186
		if (tp) {
1187
			cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1188 1189
			put_child(t, (struct tnode *)tp, cindex,
				  (struct node *)tn);
O
Olof Johansson 已提交
1190
		} else {
1191
			rcu_assign_pointer(t->trie, (struct node *)tn);
1192 1193 1194
			tp = tn;
		}
	}
O
Olof Johansson 已提交
1195 1196

	if (tp && tp->pos + tp->bits > 32)
1197 1198 1199
		pr_warning("fib_trie"
			   " tp=%p pos=%d, bits=%d, key=%0x plen=%d\n",
			   tp, tp->pos, tp->bits, key, plen);
O
Olof Johansson 已提交
1200

1201
	/* Rebalance the trie */
R
Robert Olsson 已提交
1202

1203
	trie_rebalance(t, tp);
1204
done:
1205 1206 1207
	return fa_head;
}

1208 1209 1210
/*
 * Caller must hold RTNL.
 */
1211
static int fn_trie_insert(struct fib_table *tb, struct fib_config *cfg)
1212 1213 1214
{
	struct trie *t = (struct trie *) tb->tb_data;
	struct fib_alias *fa, *new_fa;
1215
	struct list_head *fa_head = NULL;
1216
	struct fib_info *fi;
1217 1218
	int plen = cfg->fc_dst_len;
	u8 tos = cfg->fc_tos;
1219 1220 1221 1222 1223 1224 1225
	u32 key, mask;
	int err;
	struct leaf *l;

	if (plen > 32)
		return -EINVAL;

1226
	key = ntohl(cfg->fc_dst);
1227

1228
	pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1229

O
Olof Johansson 已提交
1230
	mask = ntohl(inet_make_mask(plen));
1231

1232
	if (key & ~mask)
1233 1234 1235 1236
		return -EINVAL;

	key = key & mask;

1237 1238 1239
	fi = fib_create_info(cfg);
	if (IS_ERR(fi)) {
		err = PTR_ERR(fi);
1240
		goto err;
1241
	}
1242 1243

	l = fib_find_node(t, key);
1244
	fa = NULL;
1245

1246
	if (l) {
1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261
		fa_head = get_fa_head(l, plen);
		fa = fib_find_alias(fa_head, tos, fi->fib_priority);
	}

	/* Now fa, if non-NULL, points to the first fib alias
	 * with the same keys [prefix,tos,priority], if such key already
	 * exists or to the node before which we will insert new one.
	 *
	 * If fa is NULL, we will need to allocate a new one and
	 * insert to the head of f.
	 *
	 * If f is NULL, no fib node matched the destination key
	 * and we need to allocate a new one of those as well.
	 */

1262 1263 1264
	if (fa && fa->fa_tos == tos &&
	    fa->fa_info->fib_priority == fi->fib_priority) {
		struct fib_alias *fa_first, *fa_match;
1265 1266

		err = -EEXIST;
1267
		if (cfg->fc_nlflags & NLM_F_EXCL)
1268 1269
			goto out;

1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290
		/* We have 2 goals:
		 * 1. Find exact match for type, scope, fib_info to avoid
		 * duplicate routes
		 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
		 */
		fa_match = NULL;
		fa_first = fa;
		fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
		list_for_each_entry_continue(fa, fa_head, fa_list) {
			if (fa->fa_tos != tos)
				break;
			if (fa->fa_info->fib_priority != fi->fib_priority)
				break;
			if (fa->fa_type == cfg->fc_type &&
			    fa->fa_scope == cfg->fc_scope &&
			    fa->fa_info == fi) {
				fa_match = fa;
				break;
			}
		}

1291
		if (cfg->fc_nlflags & NLM_F_REPLACE) {
1292 1293 1294
			struct fib_info *fi_drop;
			u8 state;

1295 1296 1297 1298
			fa = fa_first;
			if (fa_match) {
				if (fa == fa_match)
					err = 0;
1299
				goto out;
1300
			}
R
Robert Olsson 已提交
1301
			err = -ENOBUFS;
1302
			new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
R
Robert Olsson 已提交
1303 1304
			if (new_fa == NULL)
				goto out;
1305 1306

			fi_drop = fa->fa_info;
R
Robert Olsson 已提交
1307 1308
			new_fa->fa_tos = fa->fa_tos;
			new_fa->fa_info = fi;
1309 1310
			new_fa->fa_type = cfg->fc_type;
			new_fa->fa_scope = cfg->fc_scope;
1311
			state = fa->fa_state;
1312
			new_fa->fa_state = state & ~FA_S_ACCESSED;
1313

R
Robert Olsson 已提交
1314 1315
			list_replace_rcu(&fa->fa_list, &new_fa->fa_list);
			alias_free_mem_rcu(fa);
1316 1317 1318

			fib_release_info(fi_drop);
			if (state & FA_S_ACCESSED)
1319
				rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1320 1321
			rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
				tb->tb_id, &cfg->fc_nlinfo, NLM_F_REPLACE);
1322

O
Olof Johansson 已提交
1323
			goto succeeded;
1324 1325 1326 1327 1328
		}
		/* Error if we find a perfect match which
		 * uses the same scope, type, and nexthop
		 * information.
		 */
1329 1330
		if (fa_match)
			goto out;
1331

1332
		if (!(cfg->fc_nlflags & NLM_F_APPEND))
1333
			fa = fa_first;
1334 1335
	}
	err = -ENOENT;
1336
	if (!(cfg->fc_nlflags & NLM_F_CREATE))
1337 1338 1339
		goto out;

	err = -ENOBUFS;
1340
	new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1341 1342 1343 1344 1345
	if (new_fa == NULL)
		goto out;

	new_fa->fa_info = fi;
	new_fa->fa_tos = tos;
1346 1347
	new_fa->fa_type = cfg->fc_type;
	new_fa->fa_scope = cfg->fc_scope;
1348 1349 1350 1351 1352
	new_fa->fa_state = 0;
	/*
	 * Insert new entry to the list.
	 */

1353
	if (!fa_head) {
1354 1355 1356
		fa_head = fib_insert_node(t, key, plen);
		if (unlikely(!fa_head)) {
			err = -ENOMEM;
1357
			goto out_free_new_fa;
1358
		}
1359
	}
1360

R
Robert Olsson 已提交
1361 1362
	list_add_tail_rcu(&new_fa->fa_list,
			  (fa ? &fa->fa_list : fa_head));
1363

1364
	rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1365
	rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, tb->tb_id,
1366
		  &cfg->fc_nlinfo, 0);
1367 1368
succeeded:
	return 0;
1369 1370 1371

out_free_new_fa:
	kmem_cache_free(fn_alias_kmem, new_fa);
1372 1373
out:
	fib_release_info(fi);
O
Olof Johansson 已提交
1374
err:
1375 1376 1377
	return err;
}

R
Robert Olsson 已提交
1378
/* should be called with rcu_read_lock */
1379 1380 1381
static int check_leaf(struct trie *t, struct leaf *l,
		      t_key key,  const struct flowi *flp,
		      struct fib_result *res)
1382 1383 1384 1385
{
	struct leaf_info *li;
	struct hlist_head *hhead = &l->list;
	struct hlist_node *node;
1386

R
Robert Olsson 已提交
1387
	hlist_for_each_entry_rcu(li, node, hhead, hlist) {
1388 1389 1390 1391
		int err;
		int plen = li->plen;
		__be32 mask = inet_make_mask(plen);

1392
		if (l->key != (key & ntohl(mask)))
1393 1394
			continue;

1395
		err = fib_semantic_match(&li->falh, flp, res, plen);
1396

1397
#ifdef CONFIG_IP_FIB_TRIE_STATS
1398
		if (err <= 0)
1399
			t->stats.semantic_match_passed++;
1400 1401
		else
			t->stats.semantic_match_miss++;
1402
#endif
1403
		if (err <= 0)
1404
			return err;
1405
	}
1406

1407
	return 1;
1408 1409
}

1410 1411
static int fn_trie_lookup(struct fib_table *tb, const struct flowi *flp,
			  struct fib_result *res)
1412 1413
{
	struct trie *t = (struct trie *) tb->tb_data;
1414
	int ret;
1415 1416 1417
	struct node *n;
	struct tnode *pn;
	int pos, bits;
O
Olof Johansson 已提交
1418
	t_key key = ntohl(flp->fl4_dst);
1419 1420 1421
	int chopped_off;
	t_key cindex = 0;
	int current_prefix_length = KEYLENGTH;
O
Olof Johansson 已提交
1422 1423 1424 1425
	struct tnode *cn;
	t_key node_prefix, key_prefix, pref_mismatch;
	int mp;

R
Robert Olsson 已提交
1426
	rcu_read_lock();
O
Olof Johansson 已提交
1427

R
Robert Olsson 已提交
1428
	n = rcu_dereference(t->trie);
1429
	if (!n)
1430 1431 1432 1433 1434 1435 1436 1437
		goto failed;

#ifdef CONFIG_IP_FIB_TRIE_STATS
	t->stats.gets++;
#endif

	/* Just a leaf? */
	if (IS_LEAF(n)) {
1438
		ret = check_leaf(t, (struct leaf *)n, key, flp, res);
1439
		goto found;
1440
	}
1441

1442 1443
	pn = (struct tnode *) n;
	chopped_off = 0;
1444

O
Olof Johansson 已提交
1445
	while (pn) {
1446 1447 1448
		pos = pn->pos;
		bits = pn->bits;

1449
		if (!chopped_off)
S
Stephen Hemminger 已提交
1450 1451
			cindex = tkey_extract_bits(mask_pfx(key, current_prefix_length),
						   pos, bits);
1452 1453 1454 1455 1456 1457 1458 1459 1460 1461

		n = tnode_get_child(pn, cindex);

		if (n == NULL) {
#ifdef CONFIG_IP_FIB_TRIE_STATS
			t->stats.null_node_hit++;
#endif
			goto backtrace;
		}

O
Olof Johansson 已提交
1462
		if (IS_LEAF(n)) {
1463 1464
			ret = check_leaf(t, (struct leaf *)n, key, flp, res);
			if (ret > 0)
O
Olof Johansson 已提交
1465
				goto backtrace;
1466
			goto found;
O
Olof Johansson 已提交
1467 1468 1469
		}

		cn = (struct tnode *)n;
1470

O
Olof Johansson 已提交
1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485
		/*
		 * It's a tnode, and we can do some extra checks here if we
		 * like, to avoid descending into a dead-end branch.
		 * This tnode is in the parent's child array at index
		 * key[p_pos..p_pos+p_bits] but potentially with some bits
		 * chopped off, so in reality the index may be just a
		 * subprefix, padded with zero at the end.
		 * We can also take a look at any skipped bits in this
		 * tnode - everything up to p_pos is supposed to be ok,
		 * and the non-chopped bits of the index (se previous
		 * paragraph) are also guaranteed ok, but the rest is
		 * considered unknown.
		 *
		 * The skipped bits are key[pos+bits..cn->pos].
		 */
1486

O
Olof Johansson 已提交
1487 1488 1489 1490 1491 1492 1493 1494 1495
		/* If current_prefix_length < pos+bits, we are already doing
		 * actual prefix  matching, which means everything from
		 * pos+(bits-chopped_off) onward must be zero along some
		 * branch of this subtree - otherwise there is *no* valid
		 * prefix present. Here we can only check the skipped
		 * bits. Remember, since we have already indexed into the
		 * parent's child array, we know that the bits we chopped of
		 * *are* zero.
		 */
1496

1497 1498
		/* NOTA BENE: Checking only skipped bits
		   for the new node here */
1499

O
Olof Johansson 已提交
1500 1501
		if (current_prefix_length < pos+bits) {
			if (tkey_extract_bits(cn->key, current_prefix_length,
1502 1503
						cn->pos - current_prefix_length)
			    || !(cn->child[0]))
O
Olof Johansson 已提交
1504 1505
				goto backtrace;
		}
1506

O
Olof Johansson 已提交
1507 1508 1509 1510 1511 1512 1513 1514 1515
		/*
		 * If chopped_off=0, the index is fully validated and we
		 * only need to look at the skipped bits for this, the new,
		 * tnode. What we actually want to do is to find out if
		 * these skipped bits match our key perfectly, or if we will
		 * have to count on finding a matching prefix further down,
		 * because if we do, we would like to have some way of
		 * verifying the existence of such a prefix at this point.
		 */
1516

O
Olof Johansson 已提交
1517 1518 1519 1520 1521 1522 1523 1524
		/* The only thing we can do at this point is to verify that
		 * any such matching prefix can indeed be a prefix to our
		 * key, and if the bits in the node we are inspecting that
		 * do not match our key are not ZERO, this cannot be true.
		 * Thus, find out where there is a mismatch (before cn->pos)
		 * and verify that all the mismatching bits are zero in the
		 * new tnode's key.
		 */
1525

1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536
		/*
		 * Note: We aren't very concerned about the piece of
		 * the key that precede pn->pos+pn->bits, since these
		 * have already been checked. The bits after cn->pos
		 * aren't checked since these are by definition
		 * "unknown" at this point. Thus, what we want to see
		 * is if we are about to enter the "prefix matching"
		 * state, and in that case verify that the skipped
		 * bits that will prevail throughout this subtree are
		 * zero, as they have to be if we are to find a
		 * matching prefix.
O
Olof Johansson 已提交
1537 1538
		 */

S
Stephen Hemminger 已提交
1539 1540
		node_prefix = mask_pfx(cn->key, cn->pos);
		key_prefix = mask_pfx(key, cn->pos);
O
Olof Johansson 已提交
1541 1542 1543
		pref_mismatch = key_prefix^node_prefix;
		mp = 0;

1544 1545 1546 1547
		/*
		 * In short: If skipped bits in this node do not match
		 * the search key, enter the "prefix matching"
		 * state.directly.
O
Olof Johansson 已提交
1548 1549 1550 1551
		 */
		if (pref_mismatch) {
			while (!(pref_mismatch & (1<<(KEYLENGTH-1)))) {
				mp++;
1552
				pref_mismatch = pref_mismatch << 1;
O
Olof Johansson 已提交
1553 1554 1555 1556 1557 1558 1559 1560
			}
			key_prefix = tkey_extract_bits(cn->key, mp, cn->pos-mp);

			if (key_prefix != 0)
				goto backtrace;

			if (current_prefix_length >= cn->pos)
				current_prefix_length = mp;
1561
		}
1562

O
Olof Johansson 已提交
1563 1564 1565 1566
		pn = (struct tnode *)n; /* Descend */
		chopped_off = 0;
		continue;

1567 1568 1569 1570
backtrace:
		chopped_off++;

		/* As zero don't change the child key (cindex) */
1571 1572
		while ((chopped_off <= pn->bits)
		       && !(cindex & (1<<(chopped_off-1))))
1573 1574 1575 1576
			chopped_off++;

		/* Decrease current_... with bits chopped off */
		if (current_prefix_length > pn->pos + pn->bits - chopped_off)
1577 1578
			current_prefix_length = pn->pos + pn->bits
				- chopped_off;
O
Olof Johansson 已提交
1579

1580
		/*
1581
		 * Either we do the actual chop off according or if we have
1582 1583 1584
		 * chopped off all bits in this tnode walk up to our parent.
		 */

O
Olof Johansson 已提交
1585
		if (chopped_off <= pn->bits) {
1586
			cindex &= ~(1 << (chopped_off-1));
O
Olof Johansson 已提交
1587
		} else {
S
Stephen Hemminger 已提交
1588 1589
			struct tnode *parent = node_parent((struct node *) pn);
			if (!parent)
1590
				goto failed;
O
Olof Johansson 已提交
1591

1592
			/* Get Child's index */
S
Stephen Hemminger 已提交
1593 1594
			cindex = tkey_extract_bits(pn->key, parent->pos, parent->bits);
			pn = parent;
1595 1596 1597 1598 1599 1600
			chopped_off = 0;

#ifdef CONFIG_IP_FIB_TRIE_STATS
			t->stats.backtrack++;
#endif
			goto backtrace;
1601
		}
1602 1603
	}
failed:
1604
	ret = 1;
1605
found:
R
Robert Olsson 已提交
1606
	rcu_read_unlock();
1607 1608 1609
	return ret;
}

1610 1611 1612 1613
/*
 * Remove the leaf and return parent.
 */
static void trie_leaf_remove(struct trie *t, struct leaf *l)
1614
{
1615
	struct tnode *tp = node_parent((struct 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);
O
Olof Johansson 已提交
1623
	} else
R
Robert Olsson 已提交
1624
		rcu_assign_pointer(t->trie, NULL);
1625

1626
	free_leaf(l);
1627 1628
}

1629 1630 1631
/*
 * Caller must hold RTNL.
 */
1632
static int fn_trie_delete(struct fib_table *tb, struct fib_config *cfg)
1633 1634 1635
{
	struct trie *t = (struct trie *) tb->tb_data;
	u32 key, mask;
1636 1637
	int plen = cfg->fc_dst_len;
	u8 tos = cfg->fc_tos;
1638 1639 1640
	struct fib_alias *fa, *fa_to_delete;
	struct list_head *fa_head;
	struct leaf *l;
O
Olof Johansson 已提交
1641 1642
	struct leaf_info *li;

1643
	if (plen > 32)
1644 1645
		return -EINVAL;

1646
	key = ntohl(cfg->fc_dst);
O
Olof Johansson 已提交
1647
	mask = ntohl(inet_make_mask(plen));
1648

1649
	if (key & ~mask)
1650 1651 1652 1653 1654
		return -EINVAL;

	key = key & mask;
	l = fib_find_node(t, key);

1655
	if (!l)
1656 1657 1658 1659 1660 1661 1662 1663
		return -ESRCH;

	fa_head = get_fa_head(l, plen);
	fa = fib_find_alias(fa_head, tos, 0);

	if (!fa)
		return -ESRCH;

S
Stephen Hemminger 已提交
1664
	pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1665 1666

	fa_to_delete = NULL;
1667 1668
	fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
	list_for_each_entry_continue(fa, fa_head, fa_list) {
1669 1670 1671 1672 1673
		struct fib_info *fi = fa->fa_info;

		if (fa->fa_tos != tos)
			break;

1674 1675 1676 1677 1678 1679
		if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
		    (cfg->fc_scope == RT_SCOPE_NOWHERE ||
		     fa->fa_scope == cfg->fc_scope) &&
		    (!cfg->fc_protocol ||
		     fi->fib_protocol == cfg->fc_protocol) &&
		    fib_nh_match(cfg, fi) == 0) {
1680 1681 1682 1683 1684
			fa_to_delete = fa;
			break;
		}
	}

O
Olof Johansson 已提交
1685 1686
	if (!fa_to_delete)
		return -ESRCH;
1687

O
Olof Johansson 已提交
1688
	fa = fa_to_delete;
1689
	rtmsg_fib(RTM_DELROUTE, htonl(key), fa, plen, tb->tb_id,
1690
		  &cfg->fc_nlinfo, 0);
O
Olof Johansson 已提交
1691 1692

	l = fib_find_node(t, key);
R
Robert Olsson 已提交
1693
	li = find_leaf_info(l, plen);
1694

R
Robert Olsson 已提交
1695
	list_del_rcu(&fa->fa_list);
1696

O
Olof Johansson 已提交
1697
	if (list_empty(fa_head)) {
R
Robert Olsson 已提交
1698
		hlist_del_rcu(&li->hlist);
O
Olof Johansson 已提交
1699
		free_leaf_info(li);
R
Robert Olsson 已提交
1700
	}
1701

O
Olof Johansson 已提交
1702
	if (hlist_empty(&l->list))
1703
		trie_leaf_remove(t, l);
1704

O
Olof Johansson 已提交
1705
	if (fa->fa_state & FA_S_ACCESSED)
1706
		rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1707

R
Robert Olsson 已提交
1708 1709
	fib_release_info(fa->fa_info);
	alias_free_mem_rcu(fa);
O
Olof Johansson 已提交
1710
	return 0;
1711 1712
}

1713
static int trie_flush_list(struct list_head *head)
1714 1715 1716 1717 1718 1719 1720
{
	struct fib_alias *fa, *fa_node;
	int found = 0;

	list_for_each_entry_safe(fa, fa_node, head, fa_list) {
		struct fib_info *fi = fa->fa_info;

R
Robert Olsson 已提交
1721 1722 1723 1724
		if (fi && (fi->fib_flags & RTNH_F_DEAD)) {
			list_del_rcu(&fa->fa_list);
			fib_release_info(fa->fa_info);
			alias_free_mem_rcu(fa);
1725 1726 1727 1728 1729 1730
			found++;
		}
	}
	return found;
}

1731
static int trie_flush_leaf(struct leaf *l)
1732 1733 1734 1735 1736 1737 1738
{
	int found = 0;
	struct hlist_head *lih = &l->list;
	struct hlist_node *node, *tmp;
	struct leaf_info *li = NULL;

	hlist_for_each_entry_safe(li, node, tmp, lih, hlist) {
1739
		found += trie_flush_list(&li->falh);
1740 1741

		if (list_empty(&li->falh)) {
R
Robert Olsson 已提交
1742
			hlist_del_rcu(&li->hlist);
1743 1744 1745 1746 1747 1748
			free_leaf_info(li);
		}
	}
	return found;
}

1749 1750 1751 1752 1753
/*
 * Scan for the next right leaf starting at node p->child[idx]
 * Since we have back pointer, no recursion necessary.
 */
static struct leaf *leaf_walk_rcu(struct tnode *p, struct node *c)
1754
{
1755 1756
	do {
		t_key idx;
1757 1758

		if (c)
1759
			idx = tkey_extract_bits(c->key, p->pos, p->bits) + 1;
1760
		else
1761
			idx = 0;
R
Robert Olsson 已提交
1762

1763 1764
		while (idx < 1u << p->bits) {
			c = tnode_get_child_rcu(p, idx++);
R
Robert Olsson 已提交
1765
			if (!c)
O
Olof Johansson 已提交
1766 1767
				continue;

1768 1769 1770
			if (IS_LEAF(c)) {
				prefetch(p->child[idx]);
				return (struct leaf *) c;
1771
			}
1772 1773 1774 1775

			/* Rescan start scanning in new node */
			p = (struct tnode *) c;
			idx = 0;
1776
		}
1777 1778

		/* Node empty, walk back up to parent */
O
Olof Johansson 已提交
1779
		c = (struct node *) p;
1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806
	} while ( (p = node_parent_rcu(c)) != NULL);

	return NULL; /* Root of trie */
}

static struct leaf *trie_firstleaf(struct trie *t)
{
	struct tnode *n = (struct tnode *) rcu_dereference(t->trie);

	if (!n)
		return NULL;

	if (IS_LEAF(n))          /* trie is just a leaf */
		return (struct leaf *) n;

	return leaf_walk_rcu(n, NULL);
}

static struct leaf *trie_nextleaf(struct leaf *l)
{
	struct node *c = (struct node *) l;
	struct tnode *p = node_parent(c);

	if (!p)
		return NULL;	/* trie with just one leaf */

	return leaf_walk_rcu(p, c);
1807 1808
}

1809 1810 1811 1812
static struct leaf *trie_leafindex(struct trie *t, int index)
{
	struct leaf *l = trie_firstleaf(t);

S
Stephen Hemminger 已提交
1813
	while (l && index-- > 0)
1814
		l = trie_nextleaf(l);
S
Stephen Hemminger 已提交
1815

1816 1817 1818 1819
	return l;
}


1820 1821 1822
/*
 * Caller must hold RTNL.
 */
1823 1824 1825
static int fn_trie_flush(struct fib_table *tb)
{
	struct trie *t = (struct trie *) tb->tb_data;
1826
	struct leaf *l, *ll = NULL;
1827
	int found = 0;
1828

1829
	for (l = trie_firstleaf(t); l; l = trie_nextleaf(l)) {
1830
		found += trie_flush_leaf(l);
1831 1832

		if (ll && hlist_empty(&ll->list))
1833
			trie_leaf_remove(t, ll);
1834 1835 1836 1837
		ll = l;
	}

	if (ll && hlist_empty(&ll->list))
1838
		trie_leaf_remove(t, ll);
1839

S
Stephen Hemminger 已提交
1840
	pr_debug("trie_flush found=%d\n", found);
1841 1842 1843
	return found;
}

1844 1845 1846
static void fn_trie_select_default(struct fib_table *tb,
				   const struct flowi *flp,
				   struct fib_result *res)
1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859
{
	struct trie *t = (struct trie *) tb->tb_data;
	int order, last_idx;
	struct fib_info *fi = NULL;
	struct fib_info *last_resort;
	struct fib_alias *fa = NULL;
	struct list_head *fa_head;
	struct leaf *l;

	last_idx = -1;
	last_resort = NULL;
	order = -1;

R
Robert Olsson 已提交
1860
	rcu_read_lock();
1861

1862
	l = fib_find_node(t, 0);
1863
	if (!l)
1864 1865 1866
		goto out;

	fa_head = get_fa_head(l, 0);
1867
	if (!fa_head)
1868 1869
		goto out;

1870
	if (list_empty(fa_head))
1871 1872
		goto out;

R
Robert Olsson 已提交
1873
	list_for_each_entry_rcu(fa, fa_head, fa_list) {
1874
		struct fib_info *next_fi = fa->fa_info;
O
Olof Johansson 已提交
1875

1876 1877 1878
		if (fa->fa_scope != res->scope ||
		    fa->fa_type != RTN_UNICAST)
			continue;
O
Olof Johansson 已提交
1879

1880 1881 1882 1883 1884 1885
		if (next_fi->fib_priority > res->fi->fib_priority)
			break;
		if (!next_fi->fib_nh[0].nh_gw ||
		    next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
			continue;
		fa->fa_state |= FA_S_ACCESSED;
O
Olof Johansson 已提交
1886

1887 1888 1889 1890
		if (fi == NULL) {
			if (next_fi != res->fi)
				break;
		} else if (!fib_detect_death(fi, order, &last_resort,
1891
					     &last_idx, tb->tb_default)) {
1892
			fib_result_assign(res, fi);
1893
			tb->tb_default = order;
1894 1895 1896 1897 1898 1899
			goto out;
		}
		fi = next_fi;
		order++;
	}
	if (order <= 0 || fi == NULL) {
1900
		tb->tb_default = -1;
1901 1902 1903
		goto out;
	}

1904 1905
	if (!fib_detect_death(fi, order, &last_resort, &last_idx,
				tb->tb_default)) {
1906
		fib_result_assign(res, fi);
1907
		tb->tb_default = order;
1908 1909
		goto out;
	}
1910 1911
	if (last_idx >= 0)
		fib_result_assign(res, last_resort);
1912 1913
	tb->tb_default = last_idx;
out:
R
Robert Olsson 已提交
1914
	rcu_read_unlock();
1915 1916
}

1917 1918
static int fn_trie_dump_fa(t_key key, int plen, struct list_head *fah,
			   struct fib_table *tb,
1919 1920 1921 1922
			   struct sk_buff *skb, struct netlink_callback *cb)
{
	int i, s_i;
	struct fib_alias *fa;
A
Al Viro 已提交
1923
	__be32 xkey = htonl(key);
1924

1925
	s_i = cb->args[5];
1926 1927
	i = 0;

R
Robert Olsson 已提交
1928 1929 1930
	/* rcu_read_lock is hold by caller */

	list_for_each_entry_rcu(fa, fah, fa_list) {
1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941
		if (i < s_i) {
			i++;
			continue;
		}

		if (fib_dump_info(skb, NETLINK_CB(cb->skb).pid,
				  cb->nlh->nlmsg_seq,
				  RTM_NEWROUTE,
				  tb->tb_id,
				  fa->fa_type,
				  fa->fa_scope,
1942
				  xkey,
1943 1944
				  plen,
				  fa->fa_tos,
1945
				  fa->fa_info, NLM_F_MULTI) < 0) {
1946
			cb->args[5] = i;
1947
			return -1;
O
Olof Johansson 已提交
1948
		}
1949 1950
		i++;
	}
1951
	cb->args[5] = i;
1952 1953 1954
	return skb->len;
}

1955 1956
static int fn_trie_dump_leaf(struct leaf *l, struct fib_table *tb,
			struct sk_buff *skb, struct netlink_callback *cb)
1957
{
1958 1959 1960
	struct leaf_info *li;
	struct hlist_node *node;
	int i, s_i;
1961

1962
	s_i = cb->args[4];
1963
	i = 0;
1964

1965 1966 1967 1968
	/* rcu_read_lock is hold by caller */
	hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
		if (i < s_i) {
			i++;
1969
			continue;
1970
		}
O
Olof Johansson 已提交
1971

1972
		if (i > s_i)
1973
			cb->args[5] = 0;
1974

1975
		if (list_empty(&li->falh))
1976 1977
			continue;

1978
		if (fn_trie_dump_fa(l->key, li->plen, &li->falh, tb, skb, cb) < 0) {
1979
			cb->args[4] = i;
1980 1981
			return -1;
		}
1982
		i++;
1983
	}
1984

1985
	cb->args[4] = i;
1986 1987 1988
	return skb->len;
}

1989 1990
static int fn_trie_dump(struct fib_table *tb, struct sk_buff *skb,
			struct netlink_callback *cb)
1991
{
1992
	struct leaf *l;
1993
	struct trie *t = (struct trie *) tb->tb_data;
1994
	t_key key = cb->args[2];
1995
	int count = cb->args[3];
1996

R
Robert Olsson 已提交
1997
	rcu_read_lock();
1998 1999 2000
	/* Dump starting at last key.
	 * Note: 0.0.0.0/0 (ie default) is first key.
	 */
2001
	if (count == 0)
2002 2003
		l = trie_firstleaf(t);
	else {
2004 2005 2006
		/* Normally, continue from last key, but if that is missing
		 * fallback to using slow rescan
		 */
2007
		l = fib_find_node(t, key);
2008 2009
		if (!l)
			l = trie_leafindex(t, count);
2010
	}
2011

2012 2013
	while (l) {
		cb->args[2] = l->key;
2014
		if (fn_trie_dump_leaf(l, tb, skb, cb) < 0) {
2015
			cb->args[3] = count;
2016 2017
			rcu_read_unlock();
			return -1;
2018
		}
2019

2020
		++count;
2021
		l = trie_nextleaf(l);
2022 2023
		memset(&cb->args[4], 0,
		       sizeof(cb->args) - 4*sizeof(cb->args[0]));
2024
	}
2025
	cb->args[3] = count;
R
Robert Olsson 已提交
2026
	rcu_read_unlock();
2027

2028 2029 2030
	return skb->len;
}

2031 2032
void __init fib_hash_init(void)
{
2033 2034
	fn_alias_kmem = kmem_cache_create("ip_fib_alias",
					  sizeof(struct fib_alias),
2035 2036 2037 2038 2039 2040
					  0, SLAB_PANIC, NULL);

	trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
					   max(sizeof(struct leaf),
					       sizeof(struct leaf_info)),
					   0, SLAB_PANIC, NULL);
2041
}
2042

2043 2044 2045

/* Fix more generic FIB names for init later */
struct fib_table *fib_hash_table(u32 id)
2046 2047 2048 2049 2050 2051 2052 2053 2054 2055
{
	struct fib_table *tb;
	struct trie *t;

	tb = kmalloc(sizeof(struct fib_table) + sizeof(struct trie),
		     GFP_KERNEL);
	if (tb == NULL)
		return NULL;

	tb->tb_id = id;
2056
	tb->tb_default = -1;
2057 2058 2059 2060 2061 2062 2063 2064
	tb->tb_lookup = fn_trie_lookup;
	tb->tb_insert = fn_trie_insert;
	tb->tb_delete = fn_trie_delete;
	tb->tb_flush = fn_trie_flush;
	tb->tb_select_default = fn_trie_select_default;
	tb->tb_dump = fn_trie_dump;

	t = (struct trie *) tb->tb_data;
2065
	memset(t, 0, sizeof(*t));
2066 2067

	if (id == RT_TABLE_LOCAL)
2068
		pr_info("IPv4 FIB: Using LC-trie version %s\n", VERSION);
2069 2070 2071 2072

	return tb;
}

2073 2074 2075
#ifdef CONFIG_PROC_FS
/* Depth first Trie walk iterator */
struct fib_trie_iter {
2076
	struct seq_net_private p;
2077
	struct fib_table *tb;
2078 2079 2080 2081
	struct tnode *tnode;
	unsigned index;
	unsigned depth;
};
2082

2083
static struct node *fib_trie_get_next(struct fib_trie_iter *iter)
2084
{
2085 2086 2087
	struct tnode *tn = iter->tnode;
	unsigned cindex = iter->index;
	struct tnode *p;
2088

2089 2090 2091 2092
	/* A single entry routing table */
	if (!tn)
		return NULL;

2093 2094 2095 2096
	pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
		 iter->tnode, iter->index, iter->depth);
rescan:
	while (cindex < (1<<tn->bits)) {
2097
		struct node *n = tnode_get_child_rcu(tn, cindex);
2098

2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110
		if (n) {
			if (IS_LEAF(n)) {
				iter->tnode = tn;
				iter->index = cindex + 1;
			} else {
				/* push down one level */
				iter->tnode = (struct tnode *) n;
				iter->index = 0;
				++iter->depth;
			}
			return n;
		}
2111

2112 2113
		++cindex;
	}
O
Olof Johansson 已提交
2114

2115
	/* Current node exhausted, pop back up */
2116
	p = node_parent_rcu((struct node *)tn);
2117 2118 2119 2120 2121
	if (p) {
		cindex = tkey_extract_bits(tn->key, p->pos, p->bits)+1;
		tn = p;
		--iter->depth;
		goto rescan;
2122
	}
2123 2124 2125

	/* got root? */
	return NULL;
2126 2127
}

2128 2129
static struct node *fib_trie_get_first(struct fib_trie_iter *iter,
				       struct trie *t)
2130
{
2131
	struct node *n;
2132

S
Stephen Hemminger 已提交
2133
	if (!t)
2134 2135 2136
		return NULL;

	n = rcu_dereference(t->trie);
2137
	if (!n)
2138
		return NULL;
2139

2140 2141 2142 2143 2144 2145 2146 2147
	if (IS_TNODE(n)) {
		iter->tnode = (struct tnode *) n;
		iter->index = 0;
		iter->depth = 1;
	} else {
		iter->tnode = NULL;
		iter->index = 0;
		iter->depth = 0;
O
Olof Johansson 已提交
2148
	}
2149 2150

	return n;
2151
}
O
Olof Johansson 已提交
2152

2153 2154 2155 2156
static void trie_collect_stats(struct trie *t, struct trie_stat *s)
{
	struct node *n;
	struct fib_trie_iter iter;
O
Olof Johansson 已提交
2157

2158
	memset(s, 0, sizeof(*s));
O
Olof Johansson 已提交
2159

2160
	rcu_read_lock();
2161
	for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2162
		if (IS_LEAF(n)) {
2163 2164 2165 2166
			struct leaf *l = (struct leaf *)n;
			struct leaf_info *li;
			struct hlist_node *tmp;

2167 2168 2169 2170
			s->leaves++;
			s->totdepth += iter.depth;
			if (iter.depth > s->maxdepth)
				s->maxdepth = iter.depth;
2171 2172 2173

			hlist_for_each_entry_rcu(li, tmp, &l->list, hlist)
				++s->prefixes;
2174 2175 2176 2177 2178
		} else {
			const struct tnode *tn = (const struct tnode *) n;
			int i;

			s->tnodes++;
S
Stephen Hemminger 已提交
2179
			if (tn->bits < MAX_STAT_DEPTH)
R
Robert Olsson 已提交
2180 2181
				s->nodesizes[tn->bits]++;

2182 2183 2184
			for (i = 0; i < (1<<tn->bits); i++)
				if (!tn->child[i])
					s->nullpointers++;
2185 2186
		}
	}
R
Robert Olsson 已提交
2187
	rcu_read_unlock();
2188 2189
}

2190 2191 2192 2193
/*
 *	This outputs /proc/net/fib_triestats
 */
static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2194
{
2195
	unsigned i, max, pointers, bytes, avdepth;
2196

2197 2198 2199 2200
	if (stat->leaves)
		avdepth = stat->totdepth*100 / stat->leaves;
	else
		avdepth = 0;
O
Olof Johansson 已提交
2201

2202 2203
	seq_printf(seq, "\tAver depth:     %u.%02d\n",
		   avdepth / 100, avdepth % 100);
2204
	seq_printf(seq, "\tMax depth:      %u\n", stat->maxdepth);
O
Olof Johansson 已提交
2205

2206 2207
	seq_printf(seq, "\tLeaves:         %u\n", stat->leaves);
	bytes = sizeof(struct leaf) * stat->leaves;
2208 2209 2210 2211

	seq_printf(seq, "\tPrefixes:       %u\n", stat->prefixes);
	bytes += sizeof(struct leaf_info) * stat->prefixes;

2212
	seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2213
	bytes += sizeof(struct tnode) * stat->tnodes;
2214

R
Robert Olsson 已提交
2215 2216
	max = MAX_STAT_DEPTH;
	while (max > 0 && stat->nodesizes[max-1] == 0)
2217
		max--;
2218

2219 2220 2221
	pointers = 0;
	for (i = 1; i <= max; i++)
		if (stat->nodesizes[i] != 0) {
2222
			seq_printf(seq, "  %u: %u",  i, stat->nodesizes[i]);
2223 2224 2225
			pointers += (1<<i) * stat->nodesizes[i];
		}
	seq_putc(seq, '\n');
2226
	seq_printf(seq, "\tPointers: %u\n", pointers);
R
Robert Olsson 已提交
2227

2228
	bytes += sizeof(struct node *) * pointers;
2229 2230
	seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
	seq_printf(seq, "Total size: %u  kB\n", (bytes + 1023) / 1024);
2231
}
R
Robert Olsson 已提交
2232

2233
#ifdef CONFIG_IP_FIB_TRIE_STATS
2234 2235 2236 2237
static void trie_show_usage(struct seq_file *seq,
			    const struct trie_use_stats *stats)
{
	seq_printf(seq, "\nCounters:\n---------\n");
2238 2239 2240 2241 2242 2243 2244 2245 2246
	seq_printf(seq, "gets = %u\n", stats->gets);
	seq_printf(seq, "backtracks = %u\n", stats->backtrack);
	seq_printf(seq, "semantic match passed = %u\n",
		   stats->semantic_match_passed);
	seq_printf(seq, "semantic match miss = %u\n",
		   stats->semantic_match_miss);
	seq_printf(seq, "null node hit= %u\n", stats->null_node_hit);
	seq_printf(seq, "skipped node resize = %u\n\n",
		   stats->resize_node_skipped);
2247
}
2248 2249
#endif /*  CONFIG_IP_FIB_TRIE_STATS */

2250
static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2251
{
2252 2253 2254 2255 2256 2257
	if (tb->tb_id == RT_TABLE_LOCAL)
		seq_puts(seq, "Local:\n");
	else if (tb->tb_id == RT_TABLE_MAIN)
		seq_puts(seq, "Main:\n");
	else
		seq_printf(seq, "Id %d:\n", tb->tb_id);
2258
}
2259

2260

2261 2262
static int fib_triestat_seq_show(struct seq_file *seq, void *v)
{
2263
	struct net *net = (struct net *)seq->private;
2264
	unsigned int h;
2265

2266
	seq_printf(seq,
2267 2268
		   "Basic info: size of leaf:"
		   " %Zd bytes, size of tnode: %Zd bytes.\n",
2269 2270
		   sizeof(struct leaf), sizeof(struct tnode));

2271 2272 2273 2274 2275 2276 2277 2278
	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
		struct hlist_node *node;
		struct fib_table *tb;

		hlist_for_each_entry_rcu(tb, node, head, tb_hlist) {
			struct trie *t = (struct trie *) tb->tb_data;
			struct trie_stat stat;
2279

2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291
			if (!t)
				continue;

			fib_table_print(seq, tb);

			trie_collect_stats(t, &stat);
			trie_show_stats(seq, &stat);
#ifdef CONFIG_IP_FIB_TRIE_STATS
			trie_show_usage(seq, &t->stats);
#endif
		}
	}
2292

2293
	return 0;
2294 2295
}

2296
static int fib_triestat_seq_open(struct inode *inode, struct file *file)
2297
{
2298
	return single_open_net(inode, file, fib_triestat_seq_show);
2299 2300
}

2301
static const struct file_operations fib_triestat_fops = {
2302 2303 2304 2305
	.owner	= THIS_MODULE,
	.open	= fib_triestat_seq_open,
	.read	= seq_read,
	.llseek	= seq_lseek,
2306
	.release = single_release_net,
2307 2308
};

2309
static struct node *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2310
{
2311 2312
	struct fib_trie_iter *iter = seq->private;
	struct net *net = seq_file_net(seq);
2313
	loff_t idx = 0;
2314
	unsigned int h;
2315

2316 2317 2318 2319
	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
		struct hlist_node *node;
		struct fib_table *tb;
2320

2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331
		hlist_for_each_entry_rcu(tb, node, head, tb_hlist) {
			struct node *n;

			for (n = fib_trie_get_first(iter,
						    (struct trie *) tb->tb_data);
			     n; n = fib_trie_get_next(iter))
				if (pos == idx++) {
					iter->tb = tb;
					return n;
				}
		}
2332
	}
2333

2334 2335 2336
	return NULL;
}

2337
static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2338
	__acquires(RCU)
2339
{
2340
	rcu_read_lock();
2341
	return fib_trie_get_idx(seq, *pos);
2342 2343
}

2344
static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2345
{
2346
	struct fib_trie_iter *iter = seq->private;
2347
	struct net *net = seq_file_net(seq);
2348 2349 2350 2351
	struct fib_table *tb = iter->tb;
	struct hlist_node *tb_node;
	unsigned int h;
	struct node *n;
2352

2353
	++*pos;
2354 2355 2356 2357
	/* next node in same table */
	n = fib_trie_get_next(iter);
	if (n)
		return n;
2358

2359 2360 2361 2362 2363 2364 2365 2366
	/* walk rest of this hash chain */
	h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
	while ( (tb_node = rcu_dereference(tb->tb_hlist.next)) ) {
		tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
		n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
		if (n)
			goto found;
	}
2367

2368 2369 2370 2371 2372 2373 2374 2375 2376
	/* new hash chain */
	while (++h < FIB_TABLE_HASHSZ) {
		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
		hlist_for_each_entry_rcu(tb, tb_node, head, tb_hlist) {
			n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
			if (n)
				goto found;
		}
	}
2377
	return NULL;
2378 2379 2380 2381

found:
	iter->tb = tb;
	return n;
2382
}
2383

2384
static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2385
	__releases(RCU)
2386
{
2387 2388
	rcu_read_unlock();
}
O
Olof Johansson 已提交
2389

2390 2391 2392 2393
static void seq_indent(struct seq_file *seq, int n)
{
	while (n-- > 0) seq_puts(seq, "   ");
}
2394

2395
static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2396
{
S
Stephen Hemminger 已提交
2397
	switch (s) {
2398 2399 2400 2401 2402 2403
	case RT_SCOPE_UNIVERSE: return "universe";
	case RT_SCOPE_SITE:	return "site";
	case RT_SCOPE_LINK:	return "link";
	case RT_SCOPE_HOST:	return "host";
	case RT_SCOPE_NOWHERE:	return "nowhere";
	default:
2404
		snprintf(buf, len, "scope=%d", s);
2405 2406 2407
		return buf;
	}
}
2408

2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422
static const char *rtn_type_names[__RTN_MAX] = {
	[RTN_UNSPEC] = "UNSPEC",
	[RTN_UNICAST] = "UNICAST",
	[RTN_LOCAL] = "LOCAL",
	[RTN_BROADCAST] = "BROADCAST",
	[RTN_ANYCAST] = "ANYCAST",
	[RTN_MULTICAST] = "MULTICAST",
	[RTN_BLACKHOLE] = "BLACKHOLE",
	[RTN_UNREACHABLE] = "UNREACHABLE",
	[RTN_PROHIBIT] = "PROHIBIT",
	[RTN_THROW] = "THROW",
	[RTN_NAT] = "NAT",
	[RTN_XRESOLVE] = "XRESOLVE",
};
2423

2424
static inline const char *rtn_type(char *buf, size_t len, unsigned t)
2425 2426 2427
{
	if (t < __RTN_MAX && rtn_type_names[t])
		return rtn_type_names[t];
2428
	snprintf(buf, len, "type %u", t);
2429
	return buf;
2430 2431
}

2432 2433
/* Pretty print the trie */
static int fib_trie_seq_show(struct seq_file *seq, void *v)
2434
{
2435 2436
	const struct fib_trie_iter *iter = seq->private;
	struct node *n = v;
2437

2438 2439
	if (!node_parent_rcu(n))
		fib_table_print(seq, iter->tb);
2440

2441 2442
	if (IS_TNODE(n)) {
		struct tnode *tn = (struct tnode *) n;
S
Stephen Hemminger 已提交
2443
		__be32 prf = htonl(mask_pfx(tn->key, tn->pos));
O
Olof Johansson 已提交
2444

2445
		seq_indent(seq, iter->depth-1);
2446 2447
		seq_printf(seq, "  +-- %pI4/%d %d %d %d\n",
			   &prf, tn->pos, tn->bits, tn->full_children,
2448
			   tn->empty_children);
2449

2450 2451
	} else {
		struct leaf *l = (struct leaf *) n;
2452 2453
		struct leaf_info *li;
		struct hlist_node *node;
A
Al Viro 已提交
2454
		__be32 val = htonl(l->key);
2455 2456

		seq_indent(seq, iter->depth);
2457
		seq_printf(seq, "  |-- %pI4\n", &val);
2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471

		hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
			struct fib_alias *fa;

			list_for_each_entry_rcu(fa, &li->falh, fa_list) {
				char buf1[32], buf2[32];

				seq_indent(seq, iter->depth+1);
				seq_printf(seq, "  /%d %s %s", li->plen,
					   rtn_scope(buf1, sizeof(buf1),
						     fa->fa_scope),
					   rtn_type(buf2, sizeof(buf2),
						    fa->fa_type));
				if (fa->fa_tos)
2472
					seq_printf(seq, " tos=%d", fa->fa_tos);
2473
				seq_putc(seq, '\n');
2474 2475
			}
		}
2476
	}
2477

2478 2479 2480
	return 0;
}

2481
static const struct seq_operations fib_trie_seq_ops = {
2482 2483 2484 2485
	.start  = fib_trie_seq_start,
	.next   = fib_trie_seq_next,
	.stop   = fib_trie_seq_stop,
	.show   = fib_trie_seq_show,
2486 2487
};

2488
static int fib_trie_seq_open(struct inode *inode, struct file *file)
2489
{
2490 2491
	return seq_open_net(inode, file, &fib_trie_seq_ops,
			    sizeof(struct fib_trie_iter));
2492 2493
}

2494
static const struct file_operations fib_trie_fops = {
2495 2496 2497 2498
	.owner  = THIS_MODULE,
	.open   = fib_trie_seq_open,
	.read   = seq_read,
	.llseek = seq_lseek,
2499
	.release = seq_release_net,
2500 2501
};

2502 2503 2504 2505 2506 2507 2508 2509 2510 2511 2512 2513 2514 2515 2516 2517 2518 2519 2520 2521 2522 2523 2524 2525 2526 2527 2528 2529 2530 2531 2532 2533 2534 2535 2536 2537 2538 2539 2540 2541
struct fib_route_iter {
	struct seq_net_private p;
	struct trie *main_trie;
	loff_t	pos;
	t_key	key;
};

static struct leaf *fib_route_get_idx(struct fib_route_iter *iter, loff_t pos)
{
	struct leaf *l = NULL;
	struct trie *t = iter->main_trie;

	/* use cache location of last found key */
	if (iter->pos > 0 && pos >= iter->pos && (l = fib_find_node(t, iter->key)))
		pos -= iter->pos;
	else {
		iter->pos = 0;
		l = trie_firstleaf(t);
	}

	while (l && pos-- > 0) {
		iter->pos++;
		l = trie_nextleaf(l);
	}

	if (l)
		iter->key = pos;	/* remember it */
	else
		iter->pos = 0;		/* forget it */

	return l;
}

static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
	__acquires(RCU)
{
	struct fib_route_iter *iter = seq->private;
	struct fib_table *tb;

	rcu_read_lock();
2542
	tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579
	if (!tb)
		return NULL;

	iter->main_trie = (struct trie *) tb->tb_data;
	if (*pos == 0)
		return SEQ_START_TOKEN;
	else
		return fib_route_get_idx(iter, *pos - 1);
}

static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
{
	struct fib_route_iter *iter = seq->private;
	struct leaf *l = v;

	++*pos;
	if (v == SEQ_START_TOKEN) {
		iter->pos = 0;
		l = trie_firstleaf(iter->main_trie);
	} else {
		iter->pos++;
		l = trie_nextleaf(l);
	}

	if (l)
		iter->key = l->key;
	else
		iter->pos = 0;
	return l;
}

static void fib_route_seq_stop(struct seq_file *seq, void *v)
	__releases(RCU)
{
	rcu_read_unlock();
}

A
Al Viro 已提交
2580
static unsigned fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2581
{
2582 2583 2584 2585
	static unsigned type2flags[RTN_MAX + 1] = {
		[7] = RTF_REJECT, [8] = RTF_REJECT,
	};
	unsigned flags = type2flags[type];
2586

2587 2588
	if (fi && fi->fib_nh->nh_gw)
		flags |= RTF_GATEWAY;
A
Al Viro 已提交
2589
	if (mask == htonl(0xFFFFFFFF))
2590 2591 2592
		flags |= RTF_HOST;
	flags |= RTF_UP;
	return flags;
2593 2594
}

2595 2596 2597 2598 2599 2600 2601
/*
 *	This outputs /proc/net/route.
 *	The format of the file is not supposed to be changed
 * 	and needs to be same as fib_hash output to avoid breaking
 *	legacy utilities
 */
static int fib_route_seq_show(struct seq_file *seq, void *v)
2602
{
2603
	struct leaf *l = v;
2604 2605
	struct leaf_info *li;
	struct hlist_node *node;
2606

2607 2608 2609 2610 2611 2612
	if (v == SEQ_START_TOKEN) {
		seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
			   "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
			   "\tWindow\tIRTT");
		return 0;
	}
2613

2614
	hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
2615
		struct fib_alias *fa;
A
Al Viro 已提交
2616
		__be32 mask, prefix;
O
Olof Johansson 已提交
2617

2618 2619
		mask = inet_make_mask(li->plen);
		prefix = htonl(l->key);
2620

2621
		list_for_each_entry_rcu(fa, &li->falh, fa_list) {
2622
			const struct fib_info *fi = fa->fa_info;
2623
			unsigned flags = fib_flag_trans(fa->fa_type, mask, fi);
2624
			int len;
2625

2626 2627 2628
			if (fa->fa_type == RTN_BROADCAST
			    || fa->fa_type == RTN_MULTICAST)
				continue;
2629

2630
			if (fi)
2631 2632 2633
				seq_printf(seq,
					 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
					 "%d\t%08X\t%d\t%u\t%u%n",
2634 2635 2636 2637 2638
					 fi->fib_dev ? fi->fib_dev->name : "*",
					 prefix,
					 fi->fib_nh->nh_gw, flags, 0, 0,
					 fi->fib_priority,
					 mask,
2639 2640
					 (fi->fib_advmss ?
					  fi->fib_advmss + 40 : 0),
2641
					 fi->fib_window,
2642
					 fi->fib_rtt >> 3, &len);
2643
			else
2644 2645 2646
				seq_printf(seq,
					 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
					 "%d\t%08X\t%d\t%u\t%u%n",
2647
					 prefix, 0, flags, 0, 0, 0,
2648
					 mask, 0, 0, 0, &len);
2649

2650
			seq_printf(seq, "%*s\n", 127 - len, "");
2651
		}
2652 2653 2654 2655 2656
	}

	return 0;
}

2657
static const struct seq_operations fib_route_seq_ops = {
2658 2659 2660
	.start  = fib_route_seq_start,
	.next   = fib_route_seq_next,
	.stop   = fib_route_seq_stop,
2661
	.show   = fib_route_seq_show,
2662 2663
};

2664
static int fib_route_seq_open(struct inode *inode, struct file *file)
2665
{
2666
	return seq_open_net(inode, file, &fib_route_seq_ops,
2667
			    sizeof(struct fib_route_iter));
2668 2669
}

2670
static const struct file_operations fib_route_fops = {
2671 2672 2673 2674
	.owner  = THIS_MODULE,
	.open   = fib_route_seq_open,
	.read   = seq_read,
	.llseek = seq_lseek,
2675
	.release = seq_release_net,
2676 2677
};

2678
int __net_init fib_proc_init(struct net *net)
2679
{
2680
	if (!proc_net_fops_create(net, "fib_trie", S_IRUGO, &fib_trie_fops))
2681 2682
		goto out1;

2683 2684
	if (!proc_net_fops_create(net, "fib_triestat", S_IRUGO,
				  &fib_triestat_fops))
2685 2686
		goto out2;

2687
	if (!proc_net_fops_create(net, "route", S_IRUGO, &fib_route_fops))
2688 2689
		goto out3;

2690
	return 0;
2691 2692

out3:
2693
	proc_net_remove(net, "fib_triestat");
2694
out2:
2695
	proc_net_remove(net, "fib_trie");
2696 2697
out1:
	return -ENOMEM;
2698 2699
}

2700
void __net_exit fib_proc_exit(struct net *net)
2701
{
2702 2703 2704
	proc_net_remove(net, "fib_trie");
	proc_net_remove(net, "fib_triestat");
	proc_net_remove(net, "route");
2705 2706 2707
}

#endif /* CONFIG_PROC_FS */