fib_trie.c 61.2 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
 * An experimental study of compression methods for dynamic tries
 * Stefan Nilsson and Matti Tikkanen. Algorithmica, 33(1):19-33, 2002.
19
 * http://www.csc.kth.se/~snilsson/software/dyntrie2/
20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41
 *
 *
 * 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
 */

J
Jens Låås 已提交
51
#define VERSION "0.409"
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 <linux/slab.h>
75
#include <net/net_namespace.h>
76 77 78 79 80 81 82 83
#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
84
#define MAX_STAT_DEPTH 32
85 86 87 88 89 90 91 92

#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
93 94
#define NODE_TYPE(node) ((node)->parent & NODE_TYPE_MASK)

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

98
struct rt_trie_node {
O
Olof Johansson 已提交
99
	unsigned long parent;
100
	t_key key;
101 102 103
};

struct leaf {
O
Olof Johansson 已提交
104
	unsigned long parent;
105
	t_key key;
106
	struct hlist_head list;
R
Robert Olsson 已提交
107
	struct rcu_head rcu;
108 109 110 111
};

struct leaf_info {
	struct hlist_node hlist;
R
Robert Olsson 已提交
112
	struct rcu_head rcu;
113 114 115 116 117
	int plen;
	struct list_head falh;
};

struct tnode {
O
Olof Johansson 已提交
118
	unsigned long parent;
119
	t_key key;
120 121
	unsigned char pos;		/* 2log(KEYLENGTH) bits needed */
	unsigned char bits;		/* 2log(KEYLENGTH) bits needed */
122 123
	unsigned int full_children;	/* KEYLENGTH bits needed */
	unsigned int empty_children;	/* KEYLENGTH bits needed */
124 125 126
	union {
		struct rcu_head rcu;
		struct work_struct work;
J
Jarek Poplawski 已提交
127
		struct tnode *tnode_free;
128
	};
129
	struct rt_trie_node *child[0];
130 131 132 133 134 135 136 137 138
};

#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;
139
	unsigned int resize_node_skipped;
140 141 142 143 144 145 146 147 148
};
#endif

struct trie_stat {
	unsigned int totdepth;
	unsigned int maxdepth;
	unsigned int tnodes;
	unsigned int leaves;
	unsigned int nullpointers;
149
	unsigned int prefixes;
R
Robert Olsson 已提交
150
	unsigned int nodesizes[MAX_STAT_DEPTH];
151
};
152 153

struct trie {
154
	struct rt_trie_node *trie;
155 156 157 158 159
#ifdef CONFIG_IP_FIB_TRIE_STATS
	struct trie_use_stats stats;
#endif
};

160 161
static void put_child(struct trie *t, struct tnode *tn, int i, struct rt_trie_node *n);
static void tnode_put_child_reorg(struct tnode *tn, int i, struct rt_trie_node *n,
162
				  int wasfull);
163
static struct rt_trie_node *resize(struct trie *t, struct tnode *tn);
164 165
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
166 167
/* tnodes to free after resize(); protected by RTNL */
static struct tnode *tnode_free_head;
168 169 170 171 172 173 174 175
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;
176

177
static struct kmem_cache *fn_alias_kmem __read_mostly;
178
static struct kmem_cache *trie_leaf_kmem __read_mostly;
179

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

185
static inline struct tnode *node_parent_rcu(struct rt_trie_node *node)
186 187
{
	struct tnode *ret = node_parent(node);
S
Stephen Hemminger 已提交
188

E
Eric Dumazet 已提交
189
	return rcu_dereference_rtnl(ret);
S
Stephen Hemminger 已提交
190 191
}

192 193 194
/* Same as rcu_assign_pointer
 * but that macro() assumes that value is a pointer.
 */
195
static inline void node_set_parent(struct rt_trie_node *node, struct tnode *ptr)
S
Stephen Hemminger 已提交
196
{
197 198
	smp_wmb();
	node->parent = (unsigned long)ptr | NODE_TYPE(node);
S
Stephen Hemminger 已提交
199
}
R
Robert Olsson 已提交
200

201
static inline struct rt_trie_node *tnode_get_child(struct tnode *tn, unsigned int i)
202 203
{
	BUG_ON(i >= 1U << tn->bits);
R
Robert Olsson 已提交
204

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

208
static inline struct rt_trie_node *tnode_get_child_rcu(struct tnode *tn, unsigned int i)
209
{
210
	struct rt_trie_node *ret = tnode_get_child(tn, i);
211

E
Eric Dumazet 已提交
212
	return rcu_dereference_rtnl(ret);
213 214
}

S
Stephen Hemmigner 已提交
215
static inline int tnode_child_length(const struct tnode *tn)
216
{
O
Olof Johansson 已提交
217
	return 1 << tn->bits;
218 219
}

220
static inline t_key mask_pfx(t_key k, unsigned int l)
S
Stephen Hemminger 已提交
221 222 223 224
{
	return (l == 0) ? 0 : k >> (KEYLENGTH-l) << (KEYLENGTH-l);
}

225
static inline t_key tkey_extract_bits(t_key a, unsigned int offset, unsigned int bits)
226
{
O
Olof Johansson 已提交
227
	if (offset < KEYLENGTH)
228
		return ((t_key)(a << offset)) >> (KEYLENGTH - bits);
O
Olof Johansson 已提交
229
	else
230 231 232 233 234
		return 0;
}

static inline int tkey_equals(t_key a, t_key b)
{
235
	return a == b;
236 237 238 239
}

static inline int tkey_sub_equals(t_key a, int offset, int bits, t_key b)
{
240 241
	if (bits == 0 || offset >= KEYLENGTH)
		return 1;
O
Olof Johansson 已提交
242 243
	bits = bits > KEYLENGTH ? KEYLENGTH : bits;
	return ((a ^ b) << offset) >> (KEYLENGTH - bits) == 0;
244
}
245 246 247 248 249 250

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

251 252 253
	if (!diff)
		return 0;
	while ((diff << i) >> (KEYLENGTH-1) == 0)
254 255 256 257 258
		i++;
	return i;
}

/*
259 260
  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
261 262 263 264
  all of the bits in that key are significant.

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

265 266 267 268 269
  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
270 271
  correct key path.

272 273 274 275 276
  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
277 278
  call to tkey_sub_equals() in trie_insert().

279
  if n is an internal node - a 'tnode' here, the various parts of its key
280 281
  have many different meanings.

282
  Example:
283 284 285
  _________________________________________________________________
  | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
  -----------------------------------------------------------------
286
    0   1   2   3   4   5   6   7   8   9  10  11  12  13  14  15
287 288 289 290 291 292 293 294 295

  _________________________________________________________________
  | 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
296
  n->bits = 4
297

298 299
  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
300 301 302
  not use them for anything.

  The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
303
  index into the parent's child array. That is, they will be used to find
304 305 306 307 308
  'n' among tp's children.

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

309
  All the bits we have seen so far are significant to the node n. The rest
310 311
  of the bits are really not needed or indeed known in n->key.

312
  The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
313
  n's child array, and will of course be different for each child.
314

315

316 317 318 319 320
  The rest of the bits, from (n->pos + n->bits) onward, are completely unknown
  at this point.

*/

S
Stephen Hemminger 已提交
321
static inline void check_tnode(const struct tnode *tn)
322
{
S
Stephen Hemminger 已提交
323
	WARN_ON(tn && tn->pos+tn->bits > 32);
324 325
}

326 327
static const int halve_threshold = 25;
static const int inflate_threshold = 50;
328
static const int halve_threshold_root = 15;
J
Jens Låås 已提交
329
static const int inflate_threshold_root = 30;
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
	else
368
		return vzalloc(size);
369
}
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
	size_t size = sizeof(struct tnode) +
381
		      (sizeof(struct rt_trie_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
	tnode_free_size += sizeof(struct tnode) +
405
			   (sizeof(struct rt_trie_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 rt_trie_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

E
Eric Dumazet 已提交
458
	pr_debug("AT %p s=%zu %zu\n", tn, sizeof(struct tnode),
459
		 sizeof(struct rt_trie_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
 */

468
static inline int tnode_full(const struct tnode *tn, const struct rt_trie_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
static inline void put_child(struct trie *t, struct tnode *tn, int i,
477
			     struct rt_trie_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
static void tnode_put_child_reorg(struct tnode *tn, int i, struct rt_trie_node *n,
488
				  int wasfull)
489
{
490
	struct rt_trie_node *chi = tn->child[i];
491 492
	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
}

J
Jens Låås 已提交
517
#define MAX_WORK 10
518
static struct rt_trie_node *resize(struct trie *t, struct tnode *tn)
519 520
{
	int i;
521
	struct tnode *old_tn;
522 523
	int inflate_threshold_use;
	int halve_threshold_use;
J
Jens Låås 已提交
524
	int max_work;
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
		return NULL;
	}
	/* One child */
	if (tn->empty_children == tnode_child_length(tn) - 1)
J
Jens Låås 已提交
539
		goto one_child;
540
	/*
541 542 543 544 545
	 * Double as long as the resulting node has a number of
	 * nonempty nodes that are above the threshold.
	 */

	/*
546 547
	 * From "Implementing a dynamic compressed trie" by Stefan Nilsson of
	 * the Helsinki University of Technology and Matti Tikkanen of Nokia
548
	 * Telecommunications, page 6:
549
	 * "A node is doubled if the ratio of non-empty children to all
550 551
	 * children in the *doubled* node is at least 'high'."
	 *
552 553 554 555 556
	 * '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
557
	 * multiply the left-hand side by 50.
558 559 560 561
	 *
	 * 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"
562
	 * children, that is non-null tnodes with a skip value of 0.
563
	 * All of those will be doubled in the resulting inflated tnode, so
564
	 * we just count them one extra time here.
565
	 *
566
	 * A clearer way to write this would be:
567
	 *
568
	 * to_be_doubled = tn->full_children;
569
	 * not_to_be_doubled = tnode_child_length(tn) - tn->empty_children -
570 571 572 573
	 *     tn->full_children;
	 *
	 * new_child_length = tnode_child_length(tn) * 2;
	 *
574
	 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
575 576
	 *      new_child_length;
	 * if (new_fill_factor >= inflate_threshold)
577 578 579
	 *
	 * ...and so on, tho it would mess up the while () loop.
	 *
580 581 582
	 * anyway,
	 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
	 *      inflate_threshold
583
	 *
584 585 586
	 * avoid a division:
	 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
	 *      inflate_threshold * new_child_length
587
	 *
588
	 * expand not_to_be_doubled and to_be_doubled, and shorten:
589
	 * 100 * (tnode_child_length(tn) - tn->empty_children +
O
Olof Johansson 已提交
590
	 *    tn->full_children) >= inflate_threshold * new_child_length
591
	 *
592
	 * expand new_child_length:
593
	 * 100 * (tnode_child_length(tn) - tn->empty_children +
O
Olof Johansson 已提交
594
	 *    tn->full_children) >=
595
	 *      inflate_threshold * tnode_child_length(tn) * 2
596
	 *
597
	 * shorten again:
598
	 * 50 * (tn->full_children + tnode_child_length(tn) -
O
Olof Johansson 已提交
599
	 *    tn->empty_children) >= inflate_threshold *
600
	 *    tnode_child_length(tn)
601
	 *
602 603 604
	 */

	check_tnode(tn);
605

606 607
	/* Keep root node larger  */

608
	if (!node_parent((struct rt_trie_node *)tn)) {
J
Jens Låås 已提交
609 610
		inflate_threshold_use = inflate_threshold_root;
		halve_threshold_use = halve_threshold_root;
E
Eric Dumazet 已提交
611
	} else {
612
		inflate_threshold_use = inflate_threshold;
J
Jens Låås 已提交
613 614
		halve_threshold_use = halve_threshold;
	}
615

J
Jens Låås 已提交
616 617
	max_work = MAX_WORK;
	while ((tn->full_children > 0 &&  max_work-- &&
618 619 620
		50 * (tn->full_children + tnode_child_length(tn)
		      - tn->empty_children)
		>= inflate_threshold_use * tnode_child_length(tn))) {
621

622 623
		old_tn = tn;
		tn = inflate(t, tn);
624

625 626
		if (IS_ERR(tn)) {
			tn = old_tn;
627 628 629 630 631
#ifdef CONFIG_IP_FIB_TRIE_STATS
			t->stats.resize_node_skipped++;
#endif
			break;
		}
632 633 634 635
	}

	check_tnode(tn);

J
Jens Låås 已提交
636
	/* Return if at least one inflate is run */
E
Eric Dumazet 已提交
637
	if (max_work != MAX_WORK)
638
		return (struct rt_trie_node *) tn;
J
Jens Låås 已提交
639

640 641 642 643
	/*
	 * Halve as long as the number of empty children in this
	 * node is above threshold.
	 */
644

J
Jens Låås 已提交
645 646
	max_work = MAX_WORK;
	while (tn->bits > 1 &&  max_work-- &&
647
	       100 * (tnode_child_length(tn) - tn->empty_children) <
648
	       halve_threshold_use * tnode_child_length(tn)) {
649

650 651 652 653
		old_tn = tn;
		tn = halve(t, tn);
		if (IS_ERR(tn)) {
			tn = old_tn;
654 655 656 657 658 659
#ifdef CONFIG_IP_FIB_TRIE_STATS
			t->stats.resize_node_skipped++;
#endif
			break;
		}
	}
660

661

662
	/* Only one child remains */
J
Jens Låås 已提交
663 664
	if (tn->empty_children == tnode_child_length(tn) - 1) {
one_child:
665
		for (i = 0; i < tnode_child_length(tn); i++) {
666
			struct rt_trie_node *n;
667

O
Olof Johansson 已提交
668
			n = tn->child[i];
R
Robert Olsson 已提交
669
			if (!n)
O
Olof Johansson 已提交
670 671 672 673
				continue;

			/* compress one level */

S
Stephen Hemminger 已提交
674
			node_set_parent(n, NULL);
J
Jarek Poplawski 已提交
675
			tnode_free_safe(tn);
O
Olof Johansson 已提交
676
			return n;
677
		}
J
Jens Låås 已提交
678
	}
679
	return (struct rt_trie_node *) tn;
680 681
}

682
static struct tnode *inflate(struct trie *t, struct tnode *tn)
683 684 685 686 687
{
	struct tnode *oldtnode = tn;
	int olen = tnode_child_length(tn);
	int i;

S
Stephen Hemminger 已提交
688
	pr_debug("In inflate\n");
689 690 691

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

S
Stephen Hemminger 已提交
692
	if (!tn)
693
		return ERR_PTR(-ENOMEM);
694 695

	/*
696 697 698
	 * 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
699 700
	 * of tnode is ignored.
	 */
O
Olof Johansson 已提交
701 702

	for (i = 0; i < olen; i++) {
703
		struct tnode *inode;
704

705
		inode = (struct tnode *) tnode_get_child(oldtnode, i);
706 707 708 709 710
		if (inode &&
		    IS_TNODE(inode) &&
		    inode->pos == oldtnode->pos + oldtnode->bits &&
		    inode->bits > 1) {
			struct tnode *left, *right;
S
Stephen Hemminger 已提交
711
			t_key m = ~0U << (KEYLENGTH - 1) >> inode->pos;
712

713 714
			left = tnode_new(inode->key&(~m), inode->pos + 1,
					 inode->bits - 1);
715 716
			if (!left)
				goto nomem;
O
Olof Johansson 已提交
717

718 719 720
			right = tnode_new(inode->key|m, inode->pos + 1,
					  inode->bits - 1);

721
			if (!right) {
722 723
				tnode_free(left);
				goto nomem;
724
			}
725

726 727
			put_child(t, tn, 2*i, (struct rt_trie_node *) left);
			put_child(t, tn, 2*i+1, (struct rt_trie_node *) right);
728 729 730
		}
	}

O
Olof Johansson 已提交
731
	for (i = 0; i < olen; i++) {
732
		struct tnode *inode;
733
		struct rt_trie_node *node = tnode_get_child(oldtnode, i);
O
Olof Johansson 已提交
734 735
		struct tnode *left, *right;
		int size, j;
736

737 738 739 740 741 742
		/* An empty child */
		if (node == NULL)
			continue;

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

743
		if (IS_LEAF(node) || ((struct tnode *) node)->pos >
744
		   tn->pos + tn->bits - 1) {
745 746 747
			if (tkey_extract_bits(node->key,
					      oldtnode->pos + oldtnode->bits,
					      1) == 0)
748 749 750 751 752 753 754 755 756 757 758 759 760
				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
761
			tnode_free_safe(inode);
O
Olof Johansson 已提交
762
			continue;
763 764
		}

O
Olof Johansson 已提交
765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782
		/* 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)
		 */
783

O
Olof Johansson 已提交
784 785 786
		/* Use the old key, but set the new significant
		 *   bit to zero.
		 */
787

O
Olof Johansson 已提交
788 789
		left = (struct tnode *) tnode_get_child(tn, 2*i);
		put_child(t, tn, 2*i, NULL);
790

O
Olof Johansson 已提交
791
		BUG_ON(!left);
792

O
Olof Johansson 已提交
793 794
		right = (struct tnode *) tnode_get_child(tn, 2*i+1);
		put_child(t, tn, 2*i+1, NULL);
795

O
Olof Johansson 已提交
796
		BUG_ON(!right);
797

O
Olof Johansson 已提交
798 799 800 801
		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]);
802
		}
O
Olof Johansson 已提交
803 804 805
		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
806
		tnode_free_safe(inode);
807
	}
J
Jarek Poplawski 已提交
808
	tnode_free_safe(oldtnode);
809
	return tn;
810 811 812 813 814
nomem:
	{
		int size = tnode_child_length(tn);
		int j;

S
Stephen Hemminger 已提交
815
		for (j = 0; j < size; j++)
816 817 818 819
			if (tn->child[j])
				tnode_free((struct tnode *)tn->child[j]);

		tnode_free(tn);
S
Stephen Hemminger 已提交
820

821 822
		return ERR_PTR(-ENOMEM);
	}
823 824
}

825
static struct tnode *halve(struct trie *t, struct tnode *tn)
826 827
{
	struct tnode *oldtnode = tn;
828
	struct rt_trie_node *left, *right;
829 830 831
	int i;
	int olen = tnode_child_length(tn);

S
Stephen Hemminger 已提交
832
	pr_debug("In halve\n");
833 834

	tn = tnode_new(oldtnode->key, oldtnode->pos, oldtnode->bits - 1);
835

836 837
	if (!tn)
		return ERR_PTR(-ENOMEM);
838 839

	/*
840 841 842
	 * 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
843 844 845
	 * of tnode is ignored.
	 */

O
Olof Johansson 已提交
846
	for (i = 0; i < olen; i += 2) {
847 848
		left = tnode_get_child(oldtnode, i);
		right = tnode_get_child(oldtnode, i+1);
849

850
		/* Two nonempty children */
S
Stephen Hemminger 已提交
851
		if (left && right) {
852
			struct tnode *newn;
S
Stephen Hemminger 已提交
853

854
			newn = tnode_new(left->key, tn->pos + tn->bits, 1);
S
Stephen Hemminger 已提交
855 856

			if (!newn)
857
				goto nomem;
S
Stephen Hemminger 已提交
858

859
			put_child(t, tn, i/2, (struct rt_trie_node *)newn);
860 861 862
		}

	}
863

O
Olof Johansson 已提交
864 865 866
	for (i = 0; i < olen; i += 2) {
		struct tnode *newBinNode;

867 868
		left = tnode_get_child(oldtnode, i);
		right = tnode_get_child(oldtnode, i+1);
869

870 871 872 873 874
		/* 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
875
			continue;
S
Stephen Hemminger 已提交
876
		}
O
Olof Johansson 已提交
877 878

		if (right == NULL) {
879
			put_child(t, tn, i/2, left);
O
Olof Johansson 已提交
880 881
			continue;
		}
882

883
		/* Two nonempty children */
O
Olof Johansson 已提交
884 885 886 887 888
		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));
889
	}
J
Jarek Poplawski 已提交
890
	tnode_free_safe(oldtnode);
891
	return tn;
892 893 894 895 896
nomem:
	{
		int size = tnode_child_length(tn);
		int j;

S
Stephen Hemminger 已提交
897
		for (j = 0; j < size; j++)
898 899 900 901
			if (tn->child[j])
				tnode_free((struct tnode *)tn->child[j]);

		tnode_free(tn);
S
Stephen Hemminger 已提交
902

903 904
		return ERR_PTR(-ENOMEM);
	}
905 906
}

R
Robert Olsson 已提交
907
/* readside must use rcu_read_lock currently dump routines
R
Robert Olsson 已提交
908 909
 via get_fa_head and dump */

R
Robert Olsson 已提交
910
static struct leaf_info *find_leaf_info(struct leaf *l, int plen)
911
{
R
Robert Olsson 已提交
912
	struct hlist_head *head = &l->list;
913 914 915
	struct hlist_node *node;
	struct leaf_info *li;

R
Robert Olsson 已提交
916
	hlist_for_each_entry_rcu(li, node, head, hlist)
917
		if (li->plen == plen)
918
			return li;
O
Olof Johansson 已提交
919

920 921 922
	return NULL;
}

923
static inline struct list_head *get_fa_head(struct leaf *l, int plen)
924
{
R
Robert Olsson 已提交
925
	struct leaf_info *li = find_leaf_info(l, plen);
926

O
Olof Johansson 已提交
927 928
	if (!li)
		return NULL;
929

O
Olof Johansson 已提交
930
	return &li->falh;
931 932 933 934
}

static void insert_leaf_info(struct hlist_head *head, struct leaf_info *new)
{
935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951
	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);
	}
952 953
}

R
Robert Olsson 已提交
954 955
/* rcu_read_lock needs to be hold by caller from readside */

956 957 958 959 960
static struct leaf *
fib_find_node(struct trie *t, u32 key)
{
	int pos;
	struct tnode *tn;
961
	struct rt_trie_node *n;
962 963

	pos = 0;
E
Eric Dumazet 已提交
964
	n = rcu_dereference_rtnl(t->trie);
965 966 967

	while (n != NULL &&  NODE_TYPE(n) == T_TNODE) {
		tn = (struct tnode *) n;
O
Olof Johansson 已提交
968

969
		check_tnode(tn);
O
Olof Johansson 已提交
970

971
		if (tkey_sub_equals(tn->key, pos, tn->pos-pos, key)) {
O
Olof Johansson 已提交
972
			pos = tn->pos + tn->bits;
973 974 975 976
			n = tnode_get_child_rcu(tn,
						tkey_extract_bits(key,
								  tn->pos,
								  tn->bits));
O
Olof Johansson 已提交
977
		} else
978 979 980 981
			break;
	}
	/* Case we have found a leaf. Compare prefixes */

O
Olof Johansson 已提交
982 983 984
	if (n != NULL && IS_LEAF(n) && tkey_equals(key, n->key))
		return (struct leaf *)n;

985 986 987
	return NULL;
}

988
static void trie_rebalance(struct trie *t, struct tnode *tn)
989 990
{
	int wasfull;
R
Robert Olsson 已提交
991
	t_key cindex, key;
S
Stephen Hemminger 已提交
992
	struct tnode *tp;
993

R
Robert Olsson 已提交
994 995
	key = tn->key;

996
	while (tn != NULL && (tp = node_parent((struct rt_trie_node *)tn)) != NULL) {
997 998
		cindex = tkey_extract_bits(key, tp->pos, tp->bits);
		wasfull = tnode_full(tp, tnode_get_child(tp, cindex));
999 1000 1001
		tn = (struct tnode *) resize(t, (struct tnode *)tn);

		tnode_put_child_reorg((struct tnode *)tp, cindex,
1002
				      (struct rt_trie_node *)tn, wasfull);
O
Olof Johansson 已提交
1003

1004
		tp = node_parent((struct rt_trie_node *) tn);
1005
		if (!tp)
1006
			rcu_assign_pointer(t->trie, (struct rt_trie_node *)tn);
1007

J
Jarek Poplawski 已提交
1008
		tnode_free_flush();
S
Stephen Hemminger 已提交
1009
		if (!tp)
1010
			break;
S
Stephen Hemminger 已提交
1011
		tn = tp;
1012
	}
S
Stephen Hemminger 已提交
1013

1014
	/* Handle last (top) tnode */
1015
	if (IS_TNODE(tn))
1016
		tn = (struct tnode *)resize(t, (struct tnode *)tn);
1017

1018
	rcu_assign_pointer(t->trie, (struct rt_trie_node *)tn);
1019
	tnode_free_flush();
1020 1021
}

R
Robert Olsson 已提交
1022 1023
/* only used from updater-side */

1024
static struct list_head *fib_insert_node(struct trie *t, u32 key, int plen)
1025 1026 1027
{
	int pos, newpos;
	struct tnode *tp = NULL, *tn = NULL;
1028
	struct rt_trie_node *n;
1029 1030
	struct leaf *l;
	int missbit;
1031
	struct list_head *fa_head = NULL;
1032 1033 1034 1035
	struct leaf_info *li;
	t_key cindex;

	pos = 0;
1036
	n = t->trie;
1037

1038 1039
	/* 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,
1040
	 * and we should just put our new leaf in that.
1041 1042
	 * 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
1043 1044
	 * not be the parent's 'pos'+'bits'!
	 *
1045
	 * If it does match the current key, get pos/bits from it, extract
1046 1047 1048 1049
	 * 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.
	 *
1050 1051 1052
	 * 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.
1053 1054 1055 1056 1057
	 * 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
1058

1059
		check_tnode(tn);
O
Olof Johansson 已提交
1060

1061
		if (tkey_sub_equals(tn->key, pos, tn->pos-pos, key)) {
1062
			tp = tn;
O
Olof Johansson 已提交
1063
			pos = tn->pos + tn->bits;
1064 1065 1066 1067
			n = tnode_get_child(tn,
					    tkey_extract_bits(key,
							      tn->pos,
							      tn->bits));
1068

S
Stephen Hemminger 已提交
1069
			BUG_ON(n && node_parent(n) != tn);
O
Olof Johansson 已提交
1070
		} else
1071 1072 1073 1074 1075 1076
			break;
	}

	/*
	 * n  ----> NULL, LEAF or TNODE
	 *
1077
	 * tp is n's (parent) ----> NULL or TNODE
1078 1079
	 */

O
Olof Johansson 已提交
1080
	BUG_ON(tp && IS_LEAF(tp));
1081 1082 1083

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

1084
	if (n != NULL && IS_LEAF(n) && tkey_equals(key, n->key)) {
1085
		l = (struct leaf *) n;
1086
		li = leaf_info_new(plen);
O
Olof Johansson 已提交
1087

1088 1089
		if (!li)
			return NULL;
1090 1091 1092 1093 1094 1095 1096

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

1097 1098
	if (!l)
		return NULL;
1099 1100 1101 1102

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

1103
	if (!li) {
1104
		free_leaf(l);
1105
		return NULL;
1106
	}
1107 1108 1109 1110 1111

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

	if (t->trie && n == NULL) {
O
Olof Johansson 已提交
1112
		/* Case 2: n is NULL, and will just insert a new leaf */
1113

1114
		node_set_parent((struct rt_trie_node *)l, tp);
1115

O
Olof Johansson 已提交
1116
		cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1117
		put_child(t, (struct tnode *)tp, cindex, (struct rt_trie_node *)l);
O
Olof Johansson 已提交
1118 1119
	} else {
		/* Case 3: n is a LEAF or a TNODE and the key doesn't match. */
1120 1121
		/*
		 *  Add a new tnode here
1122 1123 1124 1125
		 *  first tnode need some special handling
		 */

		if (tp)
O
Olof Johansson 已提交
1126
			pos = tp->pos+tp->bits;
1127
		else
O
Olof Johansson 已提交
1128 1129
			pos = 0;

1130
		if (n) {
1131 1132
			newpos = tkey_mismatch(key, pos, n->key);
			tn = tnode_new(n->key, newpos, 1);
O
Olof Johansson 已提交
1133
		} else {
1134
			newpos = 0;
1135
			tn = tnode_new(key, newpos, 1); /* First tnode */
1136 1137
		}

1138
		if (!tn) {
1139
			free_leaf_info(li);
1140
			free_leaf(l);
1141
			return NULL;
O
Olof Johansson 已提交
1142 1143
		}

1144
		node_set_parent((struct rt_trie_node *)tn, tp);
1145

O
Olof Johansson 已提交
1146
		missbit = tkey_extract_bits(key, newpos, 1);
1147
		put_child(t, tn, missbit, (struct rt_trie_node *)l);
1148 1149
		put_child(t, tn, 1-missbit, n);

1150
		if (tp) {
1151
			cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1152
			put_child(t, (struct tnode *)tp, cindex,
1153
				  (struct rt_trie_node *)tn);
O
Olof Johansson 已提交
1154
		} else {
1155
			rcu_assign_pointer(t->trie, (struct rt_trie_node *)tn);
1156 1157 1158
			tp = tn;
		}
	}
O
Olof Johansson 已提交
1159 1160

	if (tp && tp->pos + tp->bits > 32)
1161 1162 1163
		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
1164

1165
	/* Rebalance the trie */
R
Robert Olsson 已提交
1166

1167
	trie_rebalance(t, tp);
1168
done:
1169 1170 1171
	return fa_head;
}

1172 1173 1174
/*
 * Caller must hold RTNL.
 */
1175
int fib_table_insert(struct fib_table *tb, struct fib_config *cfg)
1176 1177 1178
{
	struct trie *t = (struct trie *) tb->tb_data;
	struct fib_alias *fa, *new_fa;
1179
	struct list_head *fa_head = NULL;
1180
	struct fib_info *fi;
1181 1182
	int plen = cfg->fc_dst_len;
	u8 tos = cfg->fc_tos;
1183 1184 1185 1186 1187 1188 1189
	u32 key, mask;
	int err;
	struct leaf *l;

	if (plen > 32)
		return -EINVAL;

1190
	key = ntohl(cfg->fc_dst);
1191

1192
	pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1193

O
Olof Johansson 已提交
1194
	mask = ntohl(inet_make_mask(plen));
1195

1196
	if (key & ~mask)
1197 1198 1199 1200
		return -EINVAL;

	key = key & mask;

1201 1202 1203
	fi = fib_create_info(cfg);
	if (IS_ERR(fi)) {
		err = PTR_ERR(fi);
1204
		goto err;
1205
	}
1206 1207

	l = fib_find_node(t, key);
1208
	fa = NULL;
1209

1210
	if (l) {
1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225
		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.
	 */

1226 1227 1228
	if (fa && fa->fa_tos == tos &&
	    fa->fa_info->fib_priority == fi->fib_priority) {
		struct fib_alias *fa_first, *fa_match;
1229 1230

		err = -EEXIST;
1231
		if (cfg->fc_nlflags & NLM_F_EXCL)
1232 1233
			goto out;

1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254
		/* 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;
			}
		}

1255
		if (cfg->fc_nlflags & NLM_F_REPLACE) {
1256 1257 1258
			struct fib_info *fi_drop;
			u8 state;

1259 1260 1261 1262
			fa = fa_first;
			if (fa_match) {
				if (fa == fa_match)
					err = 0;
1263
				goto out;
1264
			}
R
Robert Olsson 已提交
1265
			err = -ENOBUFS;
1266
			new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
R
Robert Olsson 已提交
1267 1268
			if (new_fa == NULL)
				goto out;
1269 1270

			fi_drop = fa->fa_info;
R
Robert Olsson 已提交
1271 1272
			new_fa->fa_tos = fa->fa_tos;
			new_fa->fa_info = fi;
1273 1274
			new_fa->fa_type = cfg->fc_type;
			new_fa->fa_scope = cfg->fc_scope;
1275
			state = fa->fa_state;
1276
			new_fa->fa_state = state & ~FA_S_ACCESSED;
1277

R
Robert Olsson 已提交
1278 1279
			list_replace_rcu(&fa->fa_list, &new_fa->fa_list);
			alias_free_mem_rcu(fa);
1280 1281 1282

			fib_release_info(fi_drop);
			if (state & FA_S_ACCESSED)
1283
				rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1284 1285
			rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
				tb->tb_id, &cfg->fc_nlinfo, NLM_F_REPLACE);
1286

O
Olof Johansson 已提交
1287
			goto succeeded;
1288 1289 1290 1291 1292
		}
		/* Error if we find a perfect match which
		 * uses the same scope, type, and nexthop
		 * information.
		 */
1293 1294
		if (fa_match)
			goto out;
1295

1296
		if (!(cfg->fc_nlflags & NLM_F_APPEND))
1297
			fa = fa_first;
1298 1299
	}
	err = -ENOENT;
1300
	if (!(cfg->fc_nlflags & NLM_F_CREATE))
1301 1302 1303
		goto out;

	err = -ENOBUFS;
1304
	new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1305 1306 1307 1308 1309
	if (new_fa == NULL)
		goto out;

	new_fa->fa_info = fi;
	new_fa->fa_tos = tos;
1310 1311
	new_fa->fa_type = cfg->fc_type;
	new_fa->fa_scope = cfg->fc_scope;
1312 1313 1314 1315 1316
	new_fa->fa_state = 0;
	/*
	 * Insert new entry to the list.
	 */

1317
	if (!fa_head) {
1318 1319 1320
		fa_head = fib_insert_node(t, key, plen);
		if (unlikely(!fa_head)) {
			err = -ENOMEM;
1321
			goto out_free_new_fa;
1322
		}
1323
	}
1324

R
Robert Olsson 已提交
1325 1326
	list_add_tail_rcu(&new_fa->fa_list,
			  (fa ? &fa->fa_list : fa_head));
1327

1328
	rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1329
	rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, tb->tb_id,
1330
		  &cfg->fc_nlinfo, 0);
1331 1332
succeeded:
	return 0;
1333 1334 1335

out_free_new_fa:
	kmem_cache_free(fn_alias_kmem, new_fa);
1336 1337
out:
	fib_release_info(fi);
O
Olof Johansson 已提交
1338
err:
1339 1340 1341
	return err;
}

R
Robert Olsson 已提交
1342
/* should be called with rcu_read_lock */
1343
static int check_leaf(struct fib_table *tb, struct trie *t, struct leaf *l,
1344
		      t_key key,  const struct flowi *flp,
E
Eric Dumazet 已提交
1345
		      struct fib_result *res, int fib_flags)
1346 1347 1348 1349
{
	struct leaf_info *li;
	struct hlist_head *hhead = &l->list;
	struct hlist_node *node;
1350

R
Robert Olsson 已提交
1351
	hlist_for_each_entry_rcu(li, node, hhead, hlist) {
1352
		struct fib_alias *fa;
1353 1354 1355
		int plen = li->plen;
		__be32 mask = inet_make_mask(plen);

1356
		if (l->key != (key & ntohl(mask)))
1357 1358
			continue;

1359 1360 1361
		list_for_each_entry_rcu(fa, &li->falh, fa_list) {
			struct fib_info *fi = fa->fa_info;
			int nhsel, err;
1362

1363 1364 1365 1366 1367 1368 1369
			if (fa->fa_tos && fa->fa_tos != flp->fl4_tos)
				continue;
			if (fa->fa_scope < flp->fl4_scope)
				continue;
			fib_alias_accessed(fa);
			err = fib_props[fa->fa_type].error;
			if (err) {
1370
#ifdef CONFIG_IP_FIB_TRIE_STATS
1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381
				t->stats.semantic_match_miss++;
#endif
				return 1;
			}
			if (fi->fib_flags & RTNH_F_DEAD)
				continue;
			for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
				const struct fib_nh *nh = &fi->fib_nh[nhsel];

				if (nh->nh_flags & RTNH_F_DEAD)
					continue;
1382
				if (flp->flowi_oif && flp->flowi_oif != nh->nh_oif)
1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402
					continue;

#ifdef CONFIG_IP_FIB_TRIE_STATS
				t->stats.semantic_match_passed++;
#endif
				res->prefixlen = plen;
				res->nh_sel = nhsel;
				res->type = fa->fa_type;
				res->scope = fa->fa_scope;
				res->fi = fi;
				res->table = tb;
				res->fa_head = &li->falh;
				if (!(fib_flags & FIB_LOOKUP_NOREF))
					atomic_inc(&res->fi->fib_clntref);
				return 0;
			}
		}

#ifdef CONFIG_IP_FIB_TRIE_STATS
		t->stats.semantic_match_miss++;
1403 1404
#endif
	}
1405

1406
	return 1;
1407 1408
}

1409
int fib_table_lookup(struct fib_table *tb, const struct flowi *flp,
E
Eric Dumazet 已提交
1410
		     struct fib_result *res, int fib_flags)
1411 1412
{
	struct trie *t = (struct trie *) tb->tb_data;
1413
	int ret;
1414
	struct rt_trie_node *n;
1415
	struct tnode *pn;
1416
	unsigned int pos, bits;
O
Olof Johansson 已提交
1417
	t_key key = ntohl(flp->fl4_dst);
1418
	unsigned int chopped_off;
1419
	t_key cindex = 0;
1420
	unsigned int current_prefix_length = KEYLENGTH;
O
Olof Johansson 已提交
1421
	struct tnode *cn;
1422
	t_key pref_mismatch;
O
Olof Johansson 已提交
1423

R
Robert Olsson 已提交
1424
	rcu_read_lock();
O
Olof Johansson 已提交
1425

R
Robert Olsson 已提交
1426
	n = rcu_dereference(t->trie);
1427
	if (!n)
1428 1429 1430 1431 1432 1433 1434 1435
		goto failed;

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

	/* Just a leaf? */
	if (IS_LEAF(n)) {
1436
		ret = check_leaf(tb, t, (struct leaf *)n, key, flp, res, fib_flags);
1437
		goto found;
1438
	}
1439

1440 1441
	pn = (struct tnode *) n;
	chopped_off = 0;
1442

O
Olof Johansson 已提交
1443
	while (pn) {
1444 1445 1446
		pos = pn->pos;
		bits = pn->bits;

1447
		if (!chopped_off)
S
Stephen Hemminger 已提交
1448 1449
			cindex = tkey_extract_bits(mask_pfx(key, current_prefix_length),
						   pos, bits);
1450

1451
		n = tnode_get_child_rcu(pn, cindex);
1452 1453 1454 1455 1456 1457 1458 1459

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

O
Olof Johansson 已提交
1460
		if (IS_LEAF(n)) {
1461
			ret = check_leaf(tb, t, (struct leaf *)n, key, flp, res, fib_flags);
1462
			if (ret > 0)
O
Olof Johansson 已提交
1463
				goto backtrace;
1464
			goto found;
O
Olof Johansson 已提交
1465 1466 1467
		}

		cn = (struct tnode *)n;
1468

O
Olof Johansson 已提交
1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483
		/*
		 * 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].
		 */
1484

O
Olof Johansson 已提交
1485 1486 1487 1488 1489 1490 1491 1492 1493
		/* 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.
		 */
1494

1495 1496
		/* NOTA BENE: Checking only skipped bits
		   for the new node here */
1497

O
Olof Johansson 已提交
1498 1499
		if (current_prefix_length < pos+bits) {
			if (tkey_extract_bits(cn->key, current_prefix_length,
1500 1501
						cn->pos - current_prefix_length)
			    || !(cn->child[0]))
O
Olof Johansson 已提交
1502 1503
				goto backtrace;
		}
1504

O
Olof Johansson 已提交
1505 1506 1507 1508 1509 1510 1511 1512 1513
		/*
		 * 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.
		 */
1514

O
Olof Johansson 已提交
1515 1516 1517 1518 1519 1520 1521 1522
		/* 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.
		 */
1523

1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534
		/*
		 * 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
1535 1536
		 */

1537
		pref_mismatch = mask_pfx(cn->key ^ key, cn->pos);
O
Olof Johansson 已提交
1538

1539 1540 1541 1542
		/*
		 * 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
1543 1544
		 */
		if (pref_mismatch) {
1545
			int mp = KEYLENGTH - fls(pref_mismatch);
O
Olof Johansson 已提交
1546

1547
			if (tkey_extract_bits(cn->key, mp, cn->pos - mp) != 0)
O
Olof Johansson 已提交
1548 1549 1550 1551
				goto backtrace;

			if (current_prefix_length >= cn->pos)
				current_prefix_length = mp;
1552
		}
1553

O
Olof Johansson 已提交
1554 1555 1556 1557
		pn = (struct tnode *)n; /* Descend */
		chopped_off = 0;
		continue;

1558 1559 1560 1561
backtrace:
		chopped_off++;

		/* As zero don't change the child key (cindex) */
1562 1563
		while ((chopped_off <= pn->bits)
		       && !(cindex & (1<<(chopped_off-1))))
1564 1565 1566 1567
			chopped_off++;

		/* Decrease current_... with bits chopped off */
		if (current_prefix_length > pn->pos + pn->bits - chopped_off)
1568 1569
			current_prefix_length = pn->pos + pn->bits
				- chopped_off;
O
Olof Johansson 已提交
1570

1571
		/*
1572
		 * Either we do the actual chop off according or if we have
1573 1574 1575
		 * chopped off all bits in this tnode walk up to our parent.
		 */

O
Olof Johansson 已提交
1576
		if (chopped_off <= pn->bits) {
1577
			cindex &= ~(1 << (chopped_off-1));
O
Olof Johansson 已提交
1578
		} else {
1579
			struct tnode *parent = node_parent_rcu((struct rt_trie_node *) pn);
S
Stephen Hemminger 已提交
1580
			if (!parent)
1581
				goto failed;
O
Olof Johansson 已提交
1582

1583
			/* Get Child's index */
S
Stephen Hemminger 已提交
1584 1585
			cindex = tkey_extract_bits(pn->key, parent->pos, parent->bits);
			pn = parent;
1586 1587 1588 1589 1590 1591
			chopped_off = 0;

#ifdef CONFIG_IP_FIB_TRIE_STATS
			t->stats.backtrack++;
#endif
			goto backtrace;
1592
		}
1593 1594
	}
failed:
1595
	ret = 1;
1596
found:
R
Robert Olsson 已提交
1597
	rcu_read_unlock();
1598 1599 1600
	return ret;
}

1601 1602 1603 1604
/*
 * Remove the leaf and return parent.
 */
static void trie_leaf_remove(struct trie *t, struct leaf *l)
1605
{
1606
	struct tnode *tp = node_parent((struct rt_trie_node *) l);
1607

1608
	pr_debug("entering trie_leaf_remove(%p)\n", l);
1609

1610
	if (tp) {
1611
		t_key cindex = tkey_extract_bits(l->key, tp->pos, tp->bits);
1612
		put_child(t, (struct tnode *)tp, cindex, NULL);
1613
		trie_rebalance(t, tp);
O
Olof Johansson 已提交
1614
	} else
R
Robert Olsson 已提交
1615
		rcu_assign_pointer(t->trie, NULL);
1616

1617
	free_leaf(l);
1618 1619
}

1620 1621 1622
/*
 * Caller must hold RTNL.
 */
1623
int fib_table_delete(struct fib_table *tb, struct fib_config *cfg)
1624 1625 1626
{
	struct trie *t = (struct trie *) tb->tb_data;
	u32 key, mask;
1627 1628
	int plen = cfg->fc_dst_len;
	u8 tos = cfg->fc_tos;
1629 1630 1631
	struct fib_alias *fa, *fa_to_delete;
	struct list_head *fa_head;
	struct leaf *l;
O
Olof Johansson 已提交
1632 1633
	struct leaf_info *li;

1634
	if (plen > 32)
1635 1636
		return -EINVAL;

1637
	key = ntohl(cfg->fc_dst);
O
Olof Johansson 已提交
1638
	mask = ntohl(inet_make_mask(plen));
1639

1640
	if (key & ~mask)
1641 1642 1643 1644 1645
		return -EINVAL;

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

1646
	if (!l)
1647 1648 1649 1650 1651 1652 1653 1654
		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
1655
	pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1656 1657

	fa_to_delete = NULL;
1658 1659
	fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
	list_for_each_entry_continue(fa, fa_head, fa_list) {
1660 1661 1662 1663 1664
		struct fib_info *fi = fa->fa_info;

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

1665 1666 1667 1668 1669 1670
		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) {
1671 1672 1673 1674 1675
			fa_to_delete = fa;
			break;
		}
	}

O
Olof Johansson 已提交
1676 1677
	if (!fa_to_delete)
		return -ESRCH;
1678

O
Olof Johansson 已提交
1679
	fa = fa_to_delete;
1680
	rtmsg_fib(RTM_DELROUTE, htonl(key), fa, plen, tb->tb_id,
1681
		  &cfg->fc_nlinfo, 0);
O
Olof Johansson 已提交
1682 1683

	l = fib_find_node(t, key);
R
Robert Olsson 已提交
1684
	li = find_leaf_info(l, plen);
1685

R
Robert Olsson 已提交
1686
	list_del_rcu(&fa->fa_list);
1687

O
Olof Johansson 已提交
1688
	if (list_empty(fa_head)) {
R
Robert Olsson 已提交
1689
		hlist_del_rcu(&li->hlist);
O
Olof Johansson 已提交
1690
		free_leaf_info(li);
R
Robert Olsson 已提交
1691
	}
1692

O
Olof Johansson 已提交
1693
	if (hlist_empty(&l->list))
1694
		trie_leaf_remove(t, l);
1695

O
Olof Johansson 已提交
1696
	if (fa->fa_state & FA_S_ACCESSED)
1697
		rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1698

R
Robert Olsson 已提交
1699 1700
	fib_release_info(fa->fa_info);
	alias_free_mem_rcu(fa);
O
Olof Johansson 已提交
1701
	return 0;
1702 1703
}

1704
static int trie_flush_list(struct list_head *head)
1705 1706 1707 1708 1709 1710 1711
{
	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
1712 1713 1714 1715
		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);
1716 1717 1718 1719 1720 1721
			found++;
		}
	}
	return found;
}

1722
static int trie_flush_leaf(struct leaf *l)
1723 1724 1725 1726 1727 1728 1729
{
	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) {
1730
		found += trie_flush_list(&li->falh);
1731 1732

		if (list_empty(&li->falh)) {
R
Robert Olsson 已提交
1733
			hlist_del_rcu(&li->hlist);
1734 1735 1736 1737 1738 1739
			free_leaf_info(li);
		}
	}
	return found;
}

1740 1741 1742 1743
/*
 * Scan for the next right leaf starting at node p->child[idx]
 * Since we have back pointer, no recursion necessary.
 */
1744
static struct leaf *leaf_walk_rcu(struct tnode *p, struct rt_trie_node *c)
1745
{
1746 1747
	do {
		t_key idx;
1748 1749

		if (c)
1750
			idx = tkey_extract_bits(c->key, p->pos, p->bits) + 1;
1751
		else
1752
			idx = 0;
R
Robert Olsson 已提交
1753

1754 1755
		while (idx < 1u << p->bits) {
			c = tnode_get_child_rcu(p, idx++);
R
Robert Olsson 已提交
1756
			if (!c)
O
Olof Johansson 已提交
1757 1758
				continue;

1759 1760 1761
			if (IS_LEAF(c)) {
				prefetch(p->child[idx]);
				return (struct leaf *) c;
1762
			}
1763 1764 1765 1766

			/* Rescan start scanning in new node */
			p = (struct tnode *) c;
			idx = 0;
1767
		}
1768 1769

		/* Node empty, walk back up to parent */
1770
		c = (struct rt_trie_node *) p;
E
Eric Dumazet 已提交
1771
	} while ((p = node_parent_rcu(c)) != NULL);
1772 1773 1774 1775 1776 1777

	return NULL; /* Root of trie */
}

static struct leaf *trie_firstleaf(struct trie *t)
{
E
Eric Dumazet 已提交
1778
	struct tnode *n = (struct tnode *)rcu_dereference_rtnl(t->trie);
1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790

	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)
{
1791
	struct rt_trie_node *c = (struct rt_trie_node *) l;
1792
	struct tnode *p = node_parent_rcu(c);
1793 1794 1795 1796 1797

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

	return leaf_walk_rcu(p, c);
1798 1799
}

1800 1801 1802 1803
static struct leaf *trie_leafindex(struct trie *t, int index)
{
	struct leaf *l = trie_firstleaf(t);

S
Stephen Hemminger 已提交
1804
	while (l && index-- > 0)
1805
		l = trie_nextleaf(l);
S
Stephen Hemminger 已提交
1806

1807 1808 1809 1810
	return l;
}


1811 1812 1813
/*
 * Caller must hold RTNL.
 */
1814
int fib_table_flush(struct fib_table *tb)
1815 1816
{
	struct trie *t = (struct trie *) tb->tb_data;
1817
	struct leaf *l, *ll = NULL;
1818
	int found = 0;
1819

1820
	for (l = trie_firstleaf(t); l; l = trie_nextleaf(l)) {
1821
		found += trie_flush_leaf(l);
1822 1823

		if (ll && hlist_empty(&ll->list))
1824
			trie_leaf_remove(t, ll);
1825 1826 1827 1828
		ll = l;
	}

	if (ll && hlist_empty(&ll->list))
1829
		trie_leaf_remove(t, ll);
1830

S
Stephen Hemminger 已提交
1831
	pr_debug("trie_flush found=%d\n", found);
1832 1833 1834
	return found;
}

1835 1836 1837 1838 1839
void fib_free_table(struct fib_table *tb)
{
	kfree(tb);
}

1840 1841
static int fn_trie_dump_fa(t_key key, int plen, struct list_head *fah,
			   struct fib_table *tb,
1842 1843 1844 1845
			   struct sk_buff *skb, struct netlink_callback *cb)
{
	int i, s_i;
	struct fib_alias *fa;
A
Al Viro 已提交
1846
	__be32 xkey = htonl(key);
1847

1848
	s_i = cb->args[5];
1849 1850
	i = 0;

R
Robert Olsson 已提交
1851 1852 1853
	/* rcu_read_lock is hold by caller */

	list_for_each_entry_rcu(fa, fah, fa_list) {
1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864
		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,
1865
				  xkey,
1866 1867
				  plen,
				  fa->fa_tos,
1868
				  fa->fa_info, NLM_F_MULTI) < 0) {
1869
			cb->args[5] = i;
1870
			return -1;
O
Olof Johansson 已提交
1871
		}
1872 1873
		i++;
	}
1874
	cb->args[5] = i;
1875 1876 1877
	return skb->len;
}

1878 1879
static int fn_trie_dump_leaf(struct leaf *l, struct fib_table *tb,
			struct sk_buff *skb, struct netlink_callback *cb)
1880
{
1881 1882 1883
	struct leaf_info *li;
	struct hlist_node *node;
	int i, s_i;
1884

1885
	s_i = cb->args[4];
1886
	i = 0;
1887

1888 1889 1890 1891
	/* rcu_read_lock is hold by caller */
	hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
		if (i < s_i) {
			i++;
1892
			continue;
1893
		}
O
Olof Johansson 已提交
1894

1895
		if (i > s_i)
1896
			cb->args[5] = 0;
1897

1898
		if (list_empty(&li->falh))
1899 1900
			continue;

1901
		if (fn_trie_dump_fa(l->key, li->plen, &li->falh, tb, skb, cb) < 0) {
1902
			cb->args[4] = i;
1903 1904
			return -1;
		}
1905
		i++;
1906
	}
1907

1908
	cb->args[4] = i;
1909 1910 1911
	return skb->len;
}

1912 1913
int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
		   struct netlink_callback *cb)
1914
{
1915
	struct leaf *l;
1916
	struct trie *t = (struct trie *) tb->tb_data;
1917
	t_key key = cb->args[2];
1918
	int count = cb->args[3];
1919

R
Robert Olsson 已提交
1920
	rcu_read_lock();
1921 1922 1923
	/* Dump starting at last key.
	 * Note: 0.0.0.0/0 (ie default) is first key.
	 */
1924
	if (count == 0)
1925 1926
		l = trie_firstleaf(t);
	else {
1927 1928 1929
		/* Normally, continue from last key, but if that is missing
		 * fallback to using slow rescan
		 */
1930
		l = fib_find_node(t, key);
1931 1932
		if (!l)
			l = trie_leafindex(t, count);
1933
	}
1934

1935 1936
	while (l) {
		cb->args[2] = l->key;
1937
		if (fn_trie_dump_leaf(l, tb, skb, cb) < 0) {
1938
			cb->args[3] = count;
1939 1940
			rcu_read_unlock();
			return -1;
1941
		}
1942

1943
		++count;
1944
		l = trie_nextleaf(l);
1945 1946
		memset(&cb->args[4], 0,
		       sizeof(cb->args) - 4*sizeof(cb->args[0]));
1947
	}
1948
	cb->args[3] = count;
R
Robert Olsson 已提交
1949
	rcu_read_unlock();
1950

1951 1952 1953
	return skb->len;
}

1954
void __init fib_trie_init(void)
1955
{
1956 1957
	fn_alias_kmem = kmem_cache_create("ip_fib_alias",
					  sizeof(struct fib_alias),
1958 1959 1960 1961 1962 1963
					  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);
1964
}
1965

1966

1967
struct fib_table *fib_trie_table(u32 id)
1968 1969 1970 1971 1972 1973 1974 1975 1976 1977
{
	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;
1978
	tb->tb_default = -1;
1979 1980

	t = (struct trie *) tb->tb_data;
1981
	memset(t, 0, sizeof(*t));
1982 1983

	if (id == RT_TABLE_LOCAL)
1984
		pr_info("IPv4 FIB: Using LC-trie version %s\n", VERSION);
1985 1986 1987 1988

	return tb;
}

1989 1990 1991
#ifdef CONFIG_PROC_FS
/* Depth first Trie walk iterator */
struct fib_trie_iter {
1992
	struct seq_net_private p;
1993
	struct fib_table *tb;
1994
	struct tnode *tnode;
E
Eric Dumazet 已提交
1995 1996
	unsigned int index;
	unsigned int depth;
1997
};
1998

1999
static struct rt_trie_node *fib_trie_get_next(struct fib_trie_iter *iter)
2000
{
2001
	struct tnode *tn = iter->tnode;
E
Eric Dumazet 已提交
2002
	unsigned int cindex = iter->index;
2003
	struct tnode *p;
2004

2005 2006 2007 2008
	/* A single entry routing table */
	if (!tn)
		return NULL;

2009 2010 2011 2012
	pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
		 iter->tnode, iter->index, iter->depth);
rescan:
	while (cindex < (1<<tn->bits)) {
2013
		struct rt_trie_node *n = tnode_get_child_rcu(tn, cindex);
2014

2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026
		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;
		}
2027

2028 2029
		++cindex;
	}
O
Olof Johansson 已提交
2030

2031
	/* Current node exhausted, pop back up */
2032
	p = node_parent_rcu((struct rt_trie_node *)tn);
2033 2034 2035 2036 2037
	if (p) {
		cindex = tkey_extract_bits(tn->key, p->pos, p->bits)+1;
		tn = p;
		--iter->depth;
		goto rescan;
2038
	}
2039 2040 2041

	/* got root? */
	return NULL;
2042 2043
}

2044
static struct rt_trie_node *fib_trie_get_first(struct fib_trie_iter *iter,
2045
				       struct trie *t)
2046
{
2047
	struct rt_trie_node *n;
2048

S
Stephen Hemminger 已提交
2049
	if (!t)
2050 2051 2052
		return NULL;

	n = rcu_dereference(t->trie);
2053
	if (!n)
2054
		return NULL;
2055

2056 2057 2058 2059 2060 2061 2062 2063
	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
2064
	}
2065 2066

	return n;
2067
}
O
Olof Johansson 已提交
2068

2069 2070
static void trie_collect_stats(struct trie *t, struct trie_stat *s)
{
2071
	struct rt_trie_node *n;
2072
	struct fib_trie_iter iter;
O
Olof Johansson 已提交
2073

2074
	memset(s, 0, sizeof(*s));
O
Olof Johansson 已提交
2075

2076
	rcu_read_lock();
2077
	for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2078
		if (IS_LEAF(n)) {
2079 2080 2081 2082
			struct leaf *l = (struct leaf *)n;
			struct leaf_info *li;
			struct hlist_node *tmp;

2083 2084 2085 2086
			s->leaves++;
			s->totdepth += iter.depth;
			if (iter.depth > s->maxdepth)
				s->maxdepth = iter.depth;
2087 2088 2089

			hlist_for_each_entry_rcu(li, tmp, &l->list, hlist)
				++s->prefixes;
2090 2091 2092 2093 2094
		} else {
			const struct tnode *tn = (const struct tnode *) n;
			int i;

			s->tnodes++;
S
Stephen Hemminger 已提交
2095
			if (tn->bits < MAX_STAT_DEPTH)
R
Robert Olsson 已提交
2096 2097
				s->nodesizes[tn->bits]++;

2098 2099 2100
			for (i = 0; i < (1<<tn->bits); i++)
				if (!tn->child[i])
					s->nullpointers++;
2101 2102
		}
	}
R
Robert Olsson 已提交
2103
	rcu_read_unlock();
2104 2105
}

2106 2107 2108 2109
/*
 *	This outputs /proc/net/fib_triestats
 */
static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2110
{
E
Eric Dumazet 已提交
2111
	unsigned int i, max, pointers, bytes, avdepth;
2112

2113 2114 2115 2116
	if (stat->leaves)
		avdepth = stat->totdepth*100 / stat->leaves;
	else
		avdepth = 0;
O
Olof Johansson 已提交
2117

2118 2119
	seq_printf(seq, "\tAver depth:     %u.%02d\n",
		   avdepth / 100, avdepth % 100);
2120
	seq_printf(seq, "\tMax depth:      %u\n", stat->maxdepth);
O
Olof Johansson 已提交
2121

2122 2123
	seq_printf(seq, "\tLeaves:         %u\n", stat->leaves);
	bytes = sizeof(struct leaf) * stat->leaves;
2124 2125 2126 2127

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

2128
	seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2129
	bytes += sizeof(struct tnode) * stat->tnodes;
2130

R
Robert Olsson 已提交
2131 2132
	max = MAX_STAT_DEPTH;
	while (max > 0 && stat->nodesizes[max-1] == 0)
2133
		max--;
2134

2135 2136 2137
	pointers = 0;
	for (i = 1; i <= max; i++)
		if (stat->nodesizes[i] != 0) {
2138
			seq_printf(seq, "  %u: %u",  i, stat->nodesizes[i]);
2139 2140 2141
			pointers += (1<<i) * stat->nodesizes[i];
		}
	seq_putc(seq, '\n');
2142
	seq_printf(seq, "\tPointers: %u\n", pointers);
R
Robert Olsson 已提交
2143

2144
	bytes += sizeof(struct rt_trie_node *) * pointers;
2145 2146
	seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
	seq_printf(seq, "Total size: %u  kB\n", (bytes + 1023) / 1024);
2147
}
R
Robert Olsson 已提交
2148

2149
#ifdef CONFIG_IP_FIB_TRIE_STATS
2150 2151 2152 2153
static void trie_show_usage(struct seq_file *seq,
			    const struct trie_use_stats *stats)
{
	seq_printf(seq, "\nCounters:\n---------\n");
2154 2155 2156 2157 2158 2159 2160 2161 2162
	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);
2163
}
2164 2165
#endif /*  CONFIG_IP_FIB_TRIE_STATS */

2166
static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2167
{
2168 2169 2170 2171 2172 2173
	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);
2174
}
2175

2176

2177 2178
static int fib_triestat_seq_show(struct seq_file *seq, void *v)
{
2179
	struct net *net = (struct net *)seq->private;
2180
	unsigned int h;
2181

2182
	seq_printf(seq,
2183 2184
		   "Basic info: size of leaf:"
		   " %Zd bytes, size of tnode: %Zd bytes.\n",
2185 2186
		   sizeof(struct leaf), sizeof(struct tnode));

2187 2188 2189 2190 2191 2192 2193 2194
	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;
2195

2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207
			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
		}
	}
2208

2209
	return 0;
2210 2211
}

2212
static int fib_triestat_seq_open(struct inode *inode, struct file *file)
2213
{
2214
	return single_open_net(inode, file, fib_triestat_seq_show);
2215 2216
}

2217
static const struct file_operations fib_triestat_fops = {
2218 2219 2220 2221
	.owner	= THIS_MODULE,
	.open	= fib_triestat_seq_open,
	.read	= seq_read,
	.llseek	= seq_lseek,
2222
	.release = single_release_net,
2223 2224
};

2225
static struct rt_trie_node *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2226
{
2227 2228
	struct fib_trie_iter *iter = seq->private;
	struct net *net = seq_file_net(seq);
2229
	loff_t idx = 0;
2230
	unsigned int h;
2231

2232 2233 2234 2235
	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;
2236

2237
		hlist_for_each_entry_rcu(tb, node, head, tb_hlist) {
2238
			struct rt_trie_node *n;
2239 2240 2241 2242 2243 2244 2245 2246 2247

			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;
				}
		}
2248
	}
2249

2250 2251 2252
	return NULL;
}

2253
static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2254
	__acquires(RCU)
2255
{
2256
	rcu_read_lock();
2257
	return fib_trie_get_idx(seq, *pos);
2258 2259
}

2260
static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2261
{
2262
	struct fib_trie_iter *iter = seq->private;
2263
	struct net *net = seq_file_net(seq);
2264 2265 2266
	struct fib_table *tb = iter->tb;
	struct hlist_node *tb_node;
	unsigned int h;
2267
	struct rt_trie_node *n;
2268

2269
	++*pos;
2270 2271 2272 2273
	/* next node in same table */
	n = fib_trie_get_next(iter);
	if (n)
		return n;
2274

2275 2276 2277 2278 2279 2280 2281 2282
	/* 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;
	}
2283

2284 2285 2286 2287 2288 2289 2290 2291 2292
	/* 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;
		}
	}
2293
	return NULL;
2294 2295 2296 2297

found:
	iter->tb = tb;
	return n;
2298
}
2299

2300
static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2301
	__releases(RCU)
2302
{
2303 2304
	rcu_read_unlock();
}
O
Olof Johansson 已提交
2305

2306 2307
static void seq_indent(struct seq_file *seq, int n)
{
E
Eric Dumazet 已提交
2308 2309
	while (n-- > 0)
		seq_puts(seq, "   ");
2310
}
2311

2312
static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2313
{
S
Stephen Hemminger 已提交
2314
	switch (s) {
2315 2316 2317 2318 2319 2320
	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:
2321
		snprintf(buf, len, "scope=%d", s);
2322 2323 2324
		return buf;
	}
}
2325

2326
static const char *const rtn_type_names[__RTN_MAX] = {
2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339
	[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",
};
2340

E
Eric Dumazet 已提交
2341
static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2342 2343 2344
{
	if (t < __RTN_MAX && rtn_type_names[t])
		return rtn_type_names[t];
2345
	snprintf(buf, len, "type %u", t);
2346
	return buf;
2347 2348
}

2349 2350
/* Pretty print the trie */
static int fib_trie_seq_show(struct seq_file *seq, void *v)
2351
{
2352
	const struct fib_trie_iter *iter = seq->private;
2353
	struct rt_trie_node *n = v;
2354

2355 2356
	if (!node_parent_rcu(n))
		fib_table_print(seq, iter->tb);
2357

2358 2359
	if (IS_TNODE(n)) {
		struct tnode *tn = (struct tnode *) n;
S
Stephen Hemminger 已提交
2360
		__be32 prf = htonl(mask_pfx(tn->key, tn->pos));
O
Olof Johansson 已提交
2361

2362
		seq_indent(seq, iter->depth-1);
2363 2364
		seq_printf(seq, "  +-- %pI4/%d %d %d %d\n",
			   &prf, tn->pos, tn->bits, tn->full_children,
2365
			   tn->empty_children);
2366

2367 2368
	} else {
		struct leaf *l = (struct leaf *) n;
2369 2370
		struct leaf_info *li;
		struct hlist_node *node;
A
Al Viro 已提交
2371
		__be32 val = htonl(l->key);
2372 2373

		seq_indent(seq, iter->depth);
2374
		seq_printf(seq, "  |-- %pI4\n", &val);
2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388

		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)
2389
					seq_printf(seq, " tos=%d", fa->fa_tos);
2390
				seq_putc(seq, '\n');
2391 2392
			}
		}
2393
	}
2394

2395 2396 2397
	return 0;
}

2398
static const struct seq_operations fib_trie_seq_ops = {
2399 2400 2401 2402
	.start  = fib_trie_seq_start,
	.next   = fib_trie_seq_next,
	.stop   = fib_trie_seq_stop,
	.show   = fib_trie_seq_show,
2403 2404
};

2405
static int fib_trie_seq_open(struct inode *inode, struct file *file)
2406
{
2407 2408
	return seq_open_net(inode, file, &fib_trie_seq_ops,
			    sizeof(struct fib_trie_iter));
2409 2410
}

2411
static const struct file_operations fib_trie_fops = {
2412 2413 2414 2415
	.owner  = THIS_MODULE,
	.open   = fib_trie_seq_open,
	.read   = seq_read,
	.llseek = seq_lseek,
2416
	.release = seq_release_net,
2417 2418
};

2419 2420 2421 2422 2423 2424 2425 2426 2427 2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458
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();
2459
	tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495 2496
	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();
}

E
Eric Dumazet 已提交
2497
static unsigned int fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2498
{
E
Eric Dumazet 已提交
2499
	unsigned int flags = 0;
2500

E
Eric Dumazet 已提交
2501 2502
	if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
		flags = RTF_REJECT;
2503 2504
	if (fi && fi->fib_nh->nh_gw)
		flags |= RTF_GATEWAY;
A
Al Viro 已提交
2505
	if (mask == htonl(0xFFFFFFFF))
2506 2507 2508
		flags |= RTF_HOST;
	flags |= RTF_UP;
	return flags;
2509 2510
}

2511 2512 2513
/*
 *	This outputs /proc/net/route.
 *	The format of the file is not supposed to be changed
E
Eric Dumazet 已提交
2514
 *	and needs to be same as fib_hash output to avoid breaking
2515 2516 2517
 *	legacy utilities
 */
static int fib_route_seq_show(struct seq_file *seq, void *v)
2518
{
2519
	struct leaf *l = v;
2520 2521
	struct leaf_info *li;
	struct hlist_node *node;
2522

2523 2524 2525 2526 2527 2528
	if (v == SEQ_START_TOKEN) {
		seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
			   "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
			   "\tWindow\tIRTT");
		return 0;
	}
2529

2530
	hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
2531
		struct fib_alias *fa;
A
Al Viro 已提交
2532
		__be32 mask, prefix;
O
Olof Johansson 已提交
2533

2534 2535
		mask = inet_make_mask(li->plen);
		prefix = htonl(l->key);
2536

2537
		list_for_each_entry_rcu(fa, &li->falh, fa_list) {
2538
			const struct fib_info *fi = fa->fa_info;
E
Eric Dumazet 已提交
2539
			unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2540
			int len;
2541

2542 2543 2544
			if (fa->fa_type == RTN_BROADCAST
			    || fa->fa_type == RTN_MULTICAST)
				continue;
2545

2546
			if (fi)
2547 2548 2549
				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",
2550 2551 2552 2553 2554
					 fi->fib_dev ? fi->fib_dev->name : "*",
					 prefix,
					 fi->fib_nh->nh_gw, flags, 0, 0,
					 fi->fib_priority,
					 mask,
2555 2556
					 (fi->fib_advmss ?
					  fi->fib_advmss + 40 : 0),
2557
					 fi->fib_window,
2558
					 fi->fib_rtt >> 3, &len);
2559
			else
2560 2561 2562
				seq_printf(seq,
					 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
					 "%d\t%08X\t%d\t%u\t%u%n",
2563
					 prefix, 0, flags, 0, 0, 0,
2564
					 mask, 0, 0, 0, &len);
2565

2566
			seq_printf(seq, "%*s\n", 127 - len, "");
2567
		}
2568 2569 2570 2571 2572
	}

	return 0;
}

2573
static const struct seq_operations fib_route_seq_ops = {
2574 2575 2576
	.start  = fib_route_seq_start,
	.next   = fib_route_seq_next,
	.stop   = fib_route_seq_stop,
2577
	.show   = fib_route_seq_show,
2578 2579
};

2580
static int fib_route_seq_open(struct inode *inode, struct file *file)
2581
{
2582
	return seq_open_net(inode, file, &fib_route_seq_ops,
2583
			    sizeof(struct fib_route_iter));
2584 2585
}

2586
static const struct file_operations fib_route_fops = {
2587 2588 2589 2590
	.owner  = THIS_MODULE,
	.open   = fib_route_seq_open,
	.read   = seq_read,
	.llseek = seq_lseek,
2591
	.release = seq_release_net,
2592 2593
};

2594
int __net_init fib_proc_init(struct net *net)
2595
{
2596
	if (!proc_net_fops_create(net, "fib_trie", S_IRUGO, &fib_trie_fops))
2597 2598
		goto out1;

2599 2600
	if (!proc_net_fops_create(net, "fib_triestat", S_IRUGO,
				  &fib_triestat_fops))
2601 2602
		goto out2;

2603
	if (!proc_net_fops_create(net, "route", S_IRUGO, &fib_route_fops))
2604 2605
		goto out3;

2606
	return 0;
2607 2608

out3:
2609
	proc_net_remove(net, "fib_triestat");
2610
out2:
2611
	proc_net_remove(net, "fib_trie");
2612 2613
out1:
	return -ENOMEM;
2614 2615
}

2616
void __net_exit fib_proc_exit(struct net *net)
2617
{
2618 2619 2620
	proc_net_remove(net, "fib_trie");
	proc_net_remove(net, "fib_triestat");
	proc_net_remove(net, "route");
2621 2622 2623
}

#endif /* CONFIG_PROC_FS */