Linux Audio

Check our new training course

Loading...
v4.17
   1/*
   2 *   This program is free software; you can redistribute it and/or
   3 *   modify it under the terms of the GNU General Public License
   4 *   as published by the Free Software Foundation; either version
   5 *   2 of the License, or (at your option) any later version.
   6 *
   7 *   Robert Olsson <robert.olsson@its.uu.se> Uppsala Universitet
   8 *     & Swedish University of Agricultural Sciences.
   9 *
  10 *   Jens Laas <jens.laas@data.slu.se> Swedish University of
  11 *     Agricultural Sciences.
  12 *
  13 *   Hans Liss <hans.liss@its.uu.se>  Uppsala Universitet
  14 *
  15 * This work is based on the LPC-trie which is originally described in:
  16 *
  17 * An experimental study of compression methods for dynamic tries
  18 * Stefan Nilsson and Matti Tikkanen. Algorithmica, 33(1):19-33, 2002.
  19 * http://www.csc.kth.se/~snilsson/software/dyntrie2/
  20 *
  21 *
  22 * IP-address lookup using LC-tries. Stefan Nilsson and Gunnar Karlsson
  23 * IEEE Journal on Selected Areas in Communications, 17(6):1083-1092, June 1999
  24 *
  25 *
  26 * Code from fib_hash has been reused which includes the following header:
  27 *
  28 *
  29 * INET		An implementation of the TCP/IP protocol suite for the LINUX
  30 *		operating system.  INET is implemented using the  BSD Socket
  31 *		interface as the means of communication with the user level.
  32 *
  33 *		IPv4 FIB: lookup engine and maintenance routines.
  34 *
  35 *
  36 * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  37 *
  38 *		This program is free software; you can redistribute it and/or
  39 *		modify it under the terms of the GNU General Public License
  40 *		as published by the Free Software Foundation; either version
  41 *		2 of the License, or (at your option) any later version.
  42 *
  43 * Substantial contributions to this work comes from:
  44 *
  45 *		David S. Miller, <davem@davemloft.net>
  46 *		Stephen Hemminger <shemminger@osdl.org>
  47 *		Paul E. McKenney <paulmck@us.ibm.com>
  48 *		Patrick McHardy <kaber@trash.net>
  49 */
  50
  51#define VERSION "0.409"
  52
  53#include <linux/cache.h>
  54#include <linux/uaccess.h>
  55#include <linux/bitops.h>
  56#include <linux/types.h>
  57#include <linux/kernel.h>
  58#include <linux/mm.h>
  59#include <linux/string.h>
  60#include <linux/socket.h>
  61#include <linux/sockios.h>
  62#include <linux/errno.h>
  63#include <linux/in.h>
  64#include <linux/inet.h>
  65#include <linux/inetdevice.h>
  66#include <linux/netdevice.h>
  67#include <linux/if_arp.h>
  68#include <linux/proc_fs.h>
  69#include <linux/rcupdate.h>
  70#include <linux/skbuff.h>
  71#include <linux/netlink.h>
  72#include <linux/init.h>
  73#include <linux/list.h>
  74#include <linux/slab.h>
 
  75#include <linux/export.h>
  76#include <linux/vmalloc.h>
  77#include <linux/notifier.h>
  78#include <net/net_namespace.h>
  79#include <net/ip.h>
  80#include <net/protocol.h>
  81#include <net/route.h>
  82#include <net/tcp.h>
  83#include <net/sock.h>
  84#include <net/ip_fib.h>
  85#include <net/fib_notifier.h>
  86#include <trace/events/fib.h>
  87#include "fib_lookup.h"
  88
  89static int call_fib_entry_notifier(struct notifier_block *nb, struct net *net,
  90				   enum fib_event_type event_type, u32 dst,
  91				   int dst_len, struct fib_alias *fa)
  92{
  93	struct fib_entry_notifier_info info = {
  94		.dst = dst,
  95		.dst_len = dst_len,
  96		.fi = fa->fa_info,
  97		.tos = fa->fa_tos,
  98		.type = fa->fa_type,
  99		.tb_id = fa->tb_id,
 100	};
 101	return call_fib4_notifier(nb, net, event_type, &info.info);
 102}
 103
 104static int call_fib_entry_notifiers(struct net *net,
 105				    enum fib_event_type event_type, u32 dst,
 106				    int dst_len, struct fib_alias *fa,
 107				    struct netlink_ext_ack *extack)
 108{
 109	struct fib_entry_notifier_info info = {
 110		.info.extack = extack,
 111		.dst = dst,
 112		.dst_len = dst_len,
 113		.fi = fa->fa_info,
 114		.tos = fa->fa_tos,
 115		.type = fa->fa_type,
 116		.tb_id = fa->tb_id,
 117	};
 118	return call_fib4_notifiers(net, event_type, &info.info);
 119}
 120
 121#define MAX_STAT_DEPTH 32
 122
 123#define KEYLENGTH	(8*sizeof(t_key))
 124#define KEY_MAX		((t_key)~0)
 125
 126typedef unsigned int t_key;
 127
 128#define IS_TRIE(n)	((n)->pos >= KEYLENGTH)
 129#define IS_TNODE(n)	((n)->bits)
 130#define IS_LEAF(n)	(!(n)->bits)
 
 131
 132struct key_vector {
 
 
 
 
 133	t_key key;
 134	unsigned char pos;		/* 2log(KEYLENGTH) bits needed */
 135	unsigned char bits;		/* 2log(KEYLENGTH) bits needed */
 136	unsigned char slen;
 137	union {
 138		/* This list pointer if valid if (pos | bits) == 0 (LEAF) */
 139		struct hlist_head leaf;
 140		/* This array is valid if (pos | bits) > 0 (TNODE) */
 141		struct key_vector __rcu *tnode[0];
 142	};
 143};
 144
 145struct tnode {
 
 
 
 
 
 
 
 
 
 
 
 146	struct rcu_head rcu;
 147	t_key empty_children;		/* KEYLENGTH bits needed */
 148	t_key full_children;		/* KEYLENGTH bits needed */
 149	struct key_vector __rcu *parent;
 150	struct key_vector kv[1];
 151#define tn_bits kv[0].bits
 152};
 153
 154#define TNODE_SIZE(n)	offsetof(struct tnode, kv[0].tnode[n])
 155#define LEAF_SIZE	TNODE_SIZE(1)
 
 
 
 
 
 
 
 
 
 
 
 
 156
 157#ifdef CONFIG_IP_FIB_TRIE_STATS
 158struct trie_use_stats {
 159	unsigned int gets;
 160	unsigned int backtrack;
 161	unsigned int semantic_match_passed;
 162	unsigned int semantic_match_miss;
 163	unsigned int null_node_hit;
 164	unsigned int resize_node_skipped;
 165};
 166#endif
 167
 168struct trie_stat {
 169	unsigned int totdepth;
 170	unsigned int maxdepth;
 171	unsigned int tnodes;
 172	unsigned int leaves;
 173	unsigned int nullpointers;
 174	unsigned int prefixes;
 175	unsigned int nodesizes[MAX_STAT_DEPTH];
 176};
 177
 178struct trie {
 179	struct key_vector kv[1];
 180#ifdef CONFIG_IP_FIB_TRIE_STATS
 181	struct trie_use_stats __percpu *stats;
 182#endif
 183};
 184
 185static struct key_vector *resize(struct trie *t, struct key_vector *tn);
 
 
 
 
 
 
 
 186static size_t tnode_free_size;
 187
 188/*
 189 * synchronize_rcu after call_rcu for that many pages; it should be especially
 190 * useful before resizing the root node with PREEMPT_NONE configs; the value was
 191 * obtained experimentally, aiming to avoid visible slowdown.
 192 */
 193static const int sync_pages = 128;
 194
 195static struct kmem_cache *fn_alias_kmem __ro_after_init;
 196static struct kmem_cache *trie_leaf_kmem __ro_after_init;
 197
 198static inline struct tnode *tn_info(struct key_vector *kv)
 
 
 
 199{
 200	return container_of(kv, struct tnode, kv[0]);
 
 
 
 
 201}
 202
 203/* caller must hold RTNL */
 204#define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
 205#define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
 
 
 
 
 
 
 206
 207/* caller must hold RCU read lock or RTNL */
 208#define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
 209#define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
 210
 211/* wrapper for rcu_assign_pointer */
 212static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
 
 
 213{
 214	if (n)
 215		rcu_assign_pointer(tn_info(n)->parent, tp);
 216}
 217
 218#define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
 
 
 
 
 
 
 
 
 219
 220/* This provides us with the number of children in this node, in the case of a
 221 * leaf this will return 0 meaning none of the children are accessible.
 222 */
 223static inline unsigned long child_length(const struct key_vector *tn)
 224{
 225	return (1ul << tn->bits) & ~(1ul);
 
 
 226}
 227
 228#define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
 
 
 
 229
 230static inline unsigned long get_index(t_key key, struct key_vector *kv)
 231{
 232	unsigned long index = key ^ kv->key;
 
 233
 234	if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
 
 
 
 
 235		return 0;
 
 236
 237	return index >> kv->pos;
 
 
 238}
 239
 240/* To understand this stuff, an understanding of keys and all their bits is
 241 * necessary. Every node in the trie has a key associated with it, but not
 242 * all of the bits in that key are significant.
 243 *
 244 * Consider a node 'n' and its parent 'tp'.
 245 *
 246 * If n is a leaf, every bit in its key is significant. Its presence is
 247 * necessitated by path compression, since during a tree traversal (when
 248 * searching for a leaf - unless we are doing an insertion) we will completely
 249 * ignore all skipped bits we encounter. Thus we need to verify, at the end of
 250 * a potentially successful search, that we have indeed been walking the
 251 * correct key path.
 252 *
 253 * Note that we can never "miss" the correct key in the tree if present by
 254 * following the wrong path. Path compression ensures that segments of the key
 255 * that are the same for all keys with a given prefix are skipped, but the
 256 * skipped part *is* identical for each node in the subtrie below the skipped
 257 * bit! trie_insert() in this implementation takes care of that.
 258 *
 259 * if n is an internal node - a 'tnode' here, the various parts of its key
 260 * have many different meanings.
 261 *
 262 * Example:
 263 * _________________________________________________________________
 264 * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
 265 * -----------------------------------------------------------------
 266 *  31  30  29  28  27  26  25  24  23  22  21  20  19  18  17  16
 267 *
 268 * _________________________________________________________________
 269 * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
 270 * -----------------------------------------------------------------
 271 *  15  14  13  12  11  10   9   8   7   6   5   4   3   2   1   0
 272 *
 273 * tp->pos = 22
 274 * tp->bits = 3
 275 * n->pos = 13
 276 * n->bits = 4
 277 *
 278 * First, let's just ignore the bits that come before the parent tp, that is
 279 * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
 280 * point we do not use them for anything.
 281 *
 282 * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
 283 * index into the parent's child array. That is, they will be used to find
 284 * 'n' among tp's children.
 285 *
 286 * The bits from (n->pos + n->bits) to (tp->pos - 1) - "S" - are skipped bits
 287 * for the node n.
 288 *
 289 * All the bits we have seen so far are significant to the node n. The rest
 290 * of the bits are really not needed or indeed known in n->key.
 291 *
 292 * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
 293 * n's child array, and will of course be different for each child.
 294 *
 295 * The rest of the bits, from 0 to (n->pos -1) - "u" - are completely unknown
 296 * at this point.
 297 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 298
 299static const int halve_threshold = 25;
 300static const int inflate_threshold = 50;
 301static const int halve_threshold_root = 15;
 302static const int inflate_threshold_root = 30;
 303
 304static void __alias_free_mem(struct rcu_head *head)
 305{
 306	struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
 307	kmem_cache_free(fn_alias_kmem, fa);
 308}
 309
 310static inline void alias_free_mem_rcu(struct fib_alias *fa)
 311{
 312	call_rcu(&fa->rcu, __alias_free_mem);
 313}
 314
 315#define TNODE_KMALLOC_MAX \
 316	ilog2((PAGE_SIZE - TNODE_SIZE(0)) / sizeof(struct key_vector *))
 317#define TNODE_VMALLOC_MAX \
 318	ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
 319
 320static void __node_free_rcu(struct rcu_head *head)
 321{
 322	struct tnode *n = container_of(head, struct tnode, rcu);
 323
 324	if (!n->tn_bits)
 325		kmem_cache_free(trie_leaf_kmem, n);
 326	else
 327		kvfree(n);
 328}
 329
 330#define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
 331
 332static struct tnode *tnode_alloc(int bits)
 333{
 334	size_t size;
 335
 336	/* verify bits is within bounds */
 337	if (bits > TNODE_VMALLOC_MAX)
 338		return NULL;
 339
 340	/* determine size and verify it is non-zero and didn't overflow */
 341	size = TNODE_SIZE(1ul << bits);
 
 
 342
 
 
 343	if (size <= PAGE_SIZE)
 344		return kzalloc(size, GFP_KERNEL);
 345	else
 346		return vzalloc(size);
 347}
 348
 349static inline void empty_child_inc(struct key_vector *n)
 350{
 351	++tn_info(n)->empty_children ? : ++tn_info(n)->full_children;
 
 352}
 353
 354static inline void empty_child_dec(struct key_vector *n)
 355{
 356	tn_info(n)->empty_children-- ? : tn_info(n)->full_children--;
 
 
 
 
 
 
 
 
 
 357}
 358
 359static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
 360{
 361	struct key_vector *l;
 362	struct tnode *kv;
 363
 364	kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
 365	if (!kv)
 366		return NULL;
 367
 368	/* initialize key vector */
 369	l = kv->kv;
 370	l->key = key;
 371	l->pos = 0;
 372	l->bits = 0;
 373	l->slen = fa->fa_slen;
 374
 375	/* link leaf to fib alias */
 376	INIT_HLIST_HEAD(&l->leaf);
 377	hlist_add_head(&fa->fa_list, &l->leaf);
 378
 379	return l;
 
 
 
 
 
 
 380}
 381
 382static struct key_vector *tnode_new(t_key key, int pos, int bits)
 383{
 384	unsigned int shift = pos + bits;
 385	struct key_vector *tn;
 386	struct tnode *tnode;
 387
 388	/* verify bits and pos their msb bits clear and values are valid */
 389	BUG_ON(!bits || (shift > KEYLENGTH));
 
 
 
 390
 391	tnode = tnode_alloc(bits);
 392	if (!tnode)
 393		return NULL;
 
 
 394
 395	pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
 396		 sizeof(struct key_vector *) << bits);
 
 
 
 
 
 
 
 397
 398	if (bits == KEYLENGTH)
 399		tnode->full_children = 1;
 400	else
 401		tnode->empty_children = 1ul << bits;
 
 
 
 
 
 
 402
 403	tn = tnode->kv;
 404	tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
 405	tn->pos = pos;
 406	tn->bits = bits;
 407	tn->slen = pos;
 
 
 
 
 
 
 
 
 408
 
 
 409	return tn;
 410}
 411
 412/* Check whether a tnode 'n' is "full", i.e. it is an internal node
 
 413 * and no bits are skipped. See discussion in dyntree paper p. 6
 414 */
 415static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
 
 416{
 417	return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
 
 
 
 418}
 419
 420/* Add a child at position i overwriting the old value.
 421 * Update the value of full_children and empty_children.
 422 */
 423static void put_child(struct key_vector *tn, unsigned long i,
 424		      struct key_vector *n)
 
 
 
 
 
 
 
 
 425{
 426	struct key_vector *chi = get_child(tn, i);
 427	int isfull, wasfull;
 428
 429	BUG_ON(i >= child_length(tn));
 430
 431	/* update emptyChildren, overflow into fullChildren */
 432	if (!n && chi)
 433		empty_child_inc(tn);
 434	if (n && !chi)
 435		empty_child_dec(tn);
 436
 437	/* update fullChildren */
 438	wasfull = tnode_full(tn, chi);
 439	isfull = tnode_full(tn, n);
 440
 
 441	if (wasfull && !isfull)
 442		tn_info(tn)->full_children--;
 443	else if (!wasfull && isfull)
 444		tn_info(tn)->full_children++;
 445
 446	if (n && (tn->slen < n->slen))
 447		tn->slen = n->slen;
 448
 449	rcu_assign_pointer(tn->tnode[i], n);
 450}
 451
 452static void update_children(struct key_vector *tn)
 
 453{
 454	unsigned long i;
 
 
 
 
 455
 456	/* update all of the child parent pointers */
 457	for (i = child_length(tn); i;) {
 458		struct key_vector *inode = get_child(tn, --i);
 459
 460		if (!inode)
 461			continue;
 462
 463		/* Either update the children of a tnode that
 464		 * already belongs to us or update the child
 465		 * to point to ourselves.
 466		 */
 467		if (node_parent(inode) == tn)
 468			update_children(inode);
 469		else
 470			node_set_parent(inode, tn);
 471	}
 472}
 
 
 
 
 
 
 473
 474static inline void put_child_root(struct key_vector *tp, t_key key,
 475				  struct key_vector *n)
 476{
 477	if (IS_TRIE(tp))
 478		rcu_assign_pointer(tp->tnode[0], n);
 479	else
 480		put_child(tp, get_index(key, tp), n);
 481}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 482
 483static inline void tnode_free_init(struct key_vector *tn)
 484{
 485	tn_info(tn)->rcu.next = NULL;
 486}
 487
 488static inline void tnode_free_append(struct key_vector *tn,
 489				     struct key_vector *n)
 490{
 491	tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
 492	tn_info(tn)->rcu.next = &tn_info(n)->rcu;
 493}
 494
 495static void tnode_free(struct key_vector *tn)
 496{
 497	struct callback_head *head = &tn_info(tn)->rcu;
 
 
 
 
 498
 499	while (head) {
 500		head = head->next;
 501		tnode_free_size += TNODE_SIZE(1ul << tn->bits);
 502		node_free(tn);
 
 503
 504		tn = container_of(head, struct tnode, rcu)->kv;
 505	}
 506
 507	if (tnode_free_size >= PAGE_SIZE * sync_pages) {
 508		tnode_free_size = 0;
 509		synchronize_rcu();
 
 
 
 
 510	}
 511}
 512
 513static struct key_vector *replace(struct trie *t,
 514				  struct key_vector *oldtnode,
 515				  struct key_vector *tn)
 516{
 517	struct key_vector *tp = node_parent(oldtnode);
 518	unsigned long i;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 519
 520	/* setup the parent pointer out of and back into this node */
 521	NODE_INIT_PARENT(tn, tp);
 522	put_child_root(tp, tn->key, tn);
 523
 524	/* update all of the child parent pointers */
 525	update_children(tn);
 
 
 
 526
 527	/* all pointers should be clean so we are done */
 528	tnode_free(oldtnode);
 
 529
 530	/* resize children now that oldtnode is freed */
 531	for (i = child_length(tn); i;) {
 532		struct key_vector *inode = get_child(tn, --i);
 533
 534		/* resize child node */
 535		if (tnode_full(tn, inode))
 536			tn = resize(t, inode);
 
 537	}
 
 
 
 
 
 
 
 
 538
 539	return tp;
 
 
 
 
 
 540}
 541
 542static struct key_vector *inflate(struct trie *t,
 543				  struct key_vector *oldtnode)
 544{
 545	struct key_vector *tn;
 546	unsigned long i;
 547	t_key m;
 548
 549	pr_debug("In inflate\n");
 550
 551	tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
 552	if (!tn)
 553		goto notnode;
 554
 555	/* prepare oldtnode to be freed */
 556	tnode_free_init(oldtnode);
 557
 558	/* Assemble all of the pointers in our cluster, in this case that
 559	 * represents all of the pointers out of our allocated nodes that
 560	 * point to existing tnodes and the links between our allocated
 561	 * nodes.
 
 562	 */
 563	for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
 564		struct key_vector *inode = get_child(oldtnode, --i);
 565		struct key_vector *node0, *node1;
 566		unsigned long j, k;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 567
 568		/* An empty child */
 569		if (!inode)
 570			continue;
 571
 572		/* A leaf or an internal node with skipped bits */
 573		if (!tnode_full(oldtnode, inode)) {
 574			put_child(tn, get_index(inode->key, tn), inode);
 
 
 
 
 
 
 
 575			continue;
 576		}
 577
 578		/* drop the node in the old tnode free list */
 579		tnode_free_append(oldtnode, inode);
 580
 581		/* An internal node with two children */
 
 
 582		if (inode->bits == 1) {
 583			put_child(tn, 2 * i + 1, get_child(inode, 1));
 584			put_child(tn, 2 * i, get_child(inode, 0));
 
 
 585			continue;
 586		}
 587
 
 
 588		/* We will replace this node 'inode' with two new
 589		 * ones, 'node0' and 'node1', each with half of the
 590		 * original children. The two new nodes will have
 591		 * a position one bit further down the key and this
 592		 * means that the "significant" part of their keys
 593		 * (see the discussion near the top of this file)
 594		 * will differ by one bit, which will be "0" in
 595		 * node0's key and "1" in node1's key. Since we are
 596		 * moving the key position by one step, the bit that
 597		 * we are moving away from - the bit at position
 598		 * (tn->pos) - is the one that will differ between
 599		 * node0 and node1. So... we synthesize that bit in the
 600		 * two new keys.
 
 
 601		 */
 602		node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
 603		if (!node1)
 604			goto nomem;
 605		node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
 606
 607		tnode_free_append(tn, node1);
 608		if (!node0)
 609			goto nomem;
 610		tnode_free_append(tn, node0);
 611
 612		/* populate child pointers in new nodes */
 613		for (k = child_length(inode), j = k / 2; j;) {
 614			put_child(node1, --j, get_child(inode, --k));
 615			put_child(node0, j, get_child(inode, j));
 616			put_child(node1, --j, get_child(inode, --k));
 617			put_child(node0, j, get_child(inode, j));
 618		}
 619
 620		/* link new nodes to parent */
 621		NODE_INIT_PARENT(node1, tn);
 622		NODE_INIT_PARENT(node0, tn);
 623
 624		/* link parent to nodes */
 625		put_child(tn, 2 * i + 1, node1);
 626		put_child(tn, 2 * i, node0);
 627	}
 628
 629	/* setup the parent pointers into and out of this node */
 630	return replace(t, oldtnode, tn);
 631nomem:
 632	/* all pointers should be clean so we are done */
 633	tnode_free(tn);
 634notnode:
 635	return NULL;
 636}
 637
 638static struct key_vector *halve(struct trie *t,
 639				struct key_vector *oldtnode)
 640{
 641	struct key_vector *tn;
 642	unsigned long i;
 643
 644	pr_debug("In halve\n");
 645
 646	tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
 647	if (!tn)
 648		goto notnode;
 649
 650	/* prepare oldtnode to be freed */
 651	tnode_free_init(oldtnode);
 652
 653	/* Assemble all of the pointers in our cluster, in this case that
 654	 * represents all of the pointers out of our allocated nodes that
 655	 * point to existing tnodes and the links between our allocated
 656	 * nodes.
 657	 */
 658	for (i = child_length(oldtnode); i;) {
 659		struct key_vector *node1 = get_child(oldtnode, --i);
 660		struct key_vector *node0 = get_child(oldtnode, --i);
 661		struct key_vector *inode;
 662
 663		/* At least one of the children is empty */
 664		if (!node1 || !node0) {
 665			put_child(tn, i / 2, node1 ? : node0);
 666			continue;
 667		}
 
 
 668
 669		/* Two nonempty children */
 670		inode = tnode_new(node0->key, oldtnode->pos, 1);
 671		if (!inode)
 672			goto nomem;
 673		tnode_free_append(tn, inode);
 674
 675		/* initialize pointers out of node */
 676		put_child(inode, 1, node1);
 677		put_child(inode, 0, node0);
 678		NODE_INIT_PARENT(inode, tn);
 679
 680		/* link parent to node */
 681		put_child(tn, i / 2, inode);
 682	}
 683
 684	/* setup the parent pointers into and out of this node */
 685	return replace(t, oldtnode, tn);
 686nomem:
 687	/* all pointers should be clean so we are done */
 688	tnode_free(tn);
 689notnode:
 690	return NULL;
 691}
 692
 693static struct key_vector *collapse(struct trie *t,
 694				   struct key_vector *oldtnode)
 695{
 696	struct key_vector *n, *tp;
 697	unsigned long i;
 
 
 698
 699	/* scan the tnode looking for that one child that might still exist */
 700	for (n = NULL, i = child_length(oldtnode); !n && i;)
 701		n = get_child(oldtnode, --i);
 702
 703	/* compress one level */
 704	tp = node_parent(oldtnode);
 705	put_child_root(tp, oldtnode->key, n);
 706	node_set_parent(n, tp);
 707
 708	/* drop dead node */
 709	node_free(oldtnode);
 710
 711	return tp;
 712}
 
 
 
 
 713
 714static unsigned char update_suffix(struct key_vector *tn)
 715{
 716	unsigned char slen = tn->pos;
 717	unsigned long stride, i;
 718	unsigned char slen_max;
 719
 720	/* only vector 0 can have a suffix length greater than or equal to
 721	 * tn->pos + tn->bits, the second highest node will have a suffix
 722	 * length at most of tn->pos + tn->bits - 1
 723	 */
 724	slen_max = min_t(unsigned char, tn->pos + tn->bits - 1, tn->slen);
 725
 726	/* search though the list of children looking for nodes that might
 727	 * have a suffix greater than the one we currently have.  This is
 728	 * why we start with a stride of 2 since a stride of 1 would
 729	 * represent the nodes with suffix length equal to tn->pos
 730	 */
 731	for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
 732		struct key_vector *n = get_child(tn, i);
 733
 734		if (!n || (n->slen <= slen))
 735			continue;
 736
 737		/* update stride and slen based on new value */
 738		stride <<= (n->slen - slen);
 739		slen = n->slen;
 740		i &= ~(stride - 1);
 741
 742		/* stop searching if we have hit the maximum possible value */
 743		if (slen >= slen_max)
 744			break;
 745	}
 746
 747	tn->slen = slen;
 
 748
 749	return slen;
 750}
 751
 752/* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
 753 * the Helsinki University of Technology and Matti Tikkanen of Nokia
 754 * Telecommunications, page 6:
 755 * "A node is doubled if the ratio of non-empty children to all
 756 * children in the *doubled* node is at least 'high'."
 757 *
 758 * 'high' in this instance is the variable 'inflate_threshold'. It
 759 * is expressed as a percentage, so we multiply it with
 760 * child_length() and instead of multiplying by 2 (since the
 761 * child array will be doubled by inflate()) and multiplying
 762 * the left-hand side by 100 (to handle the percentage thing) we
 763 * multiply the left-hand side by 50.
 764 *
 765 * The left-hand side may look a bit weird: child_length(tn)
 766 * - tn->empty_children is of course the number of non-null children
 767 * in the current node. tn->full_children is the number of "full"
 768 * children, that is non-null tnodes with a skip value of 0.
 769 * All of those will be doubled in the resulting inflated tnode, so
 770 * we just count them one extra time here.
 771 *
 772 * A clearer way to write this would be:
 773 *
 774 * to_be_doubled = tn->full_children;
 775 * not_to_be_doubled = child_length(tn) - tn->empty_children -
 776 *     tn->full_children;
 777 *
 778 * new_child_length = child_length(tn) * 2;
 779 *
 780 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
 781 *      new_child_length;
 782 * if (new_fill_factor >= inflate_threshold)
 783 *
 784 * ...and so on, tho it would mess up the while () loop.
 785 *
 786 * anyway,
 787 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
 788 *      inflate_threshold
 789 *
 790 * avoid a division:
 791 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
 792 *      inflate_threshold * new_child_length
 793 *
 794 * expand not_to_be_doubled and to_be_doubled, and shorten:
 795 * 100 * (child_length(tn) - tn->empty_children +
 796 *    tn->full_children) >= inflate_threshold * new_child_length
 797 *
 798 * expand new_child_length:
 799 * 100 * (child_length(tn) - tn->empty_children +
 800 *    tn->full_children) >=
 801 *      inflate_threshold * child_length(tn) * 2
 802 *
 803 * shorten again:
 804 * 50 * (tn->full_children + child_length(tn) -
 805 *    tn->empty_children) >= inflate_threshold *
 806 *    child_length(tn)
 807 *
 808 */
 809static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
 810{
 811	unsigned long used = child_length(tn);
 812	unsigned long threshold = used;
 813
 814	/* Keep root node larger */
 815	threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
 816	used -= tn_info(tn)->empty_children;
 817	used += tn_info(tn)->full_children;
 
 
 
 818
 819	/* if bits == KEYLENGTH then pos = 0, and will fail below */
 
 
 
 820
 821	return (used > 1) && tn->pos && ((50 * used) >= threshold);
 
 
 
 
 
 
 
 
 
 
 
 822}
 823
 824static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
 825{
 826	unsigned long used = child_length(tn);
 827	unsigned long threshold = used;
 828
 829	/* Keep root node larger */
 830	threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
 831	used -= tn_info(tn)->empty_children;
 
 
 832
 833	/* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
 
 
 834
 835	return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
 836}
 837
 838static inline bool should_collapse(struct key_vector *tn)
 839{
 840	unsigned long used = child_length(tn);
 841
 842	used -= tn_info(tn)->empty_children;
 843
 844	/* account for bits == KEYLENGTH case */
 845	if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
 846		used -= KEY_MAX;
 847
 848	/* One child or none, time to drop us from the trie */
 849	return used < 2;
 850}
 851
 852#define MAX_WORK 10
 853static struct key_vector *resize(struct trie *t, struct key_vector *tn)
 854{
 855#ifdef CONFIG_IP_FIB_TRIE_STATS
 856	struct trie_use_stats __percpu *stats = t->stats;
 857#endif
 858	struct key_vector *tp = node_parent(tn);
 859	unsigned long cindex = get_index(tn->key, tp);
 860	int max_work = MAX_WORK;
 861
 862	pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
 863		 tn, inflate_threshold, halve_threshold);
 864
 865	/* track the tnode via the pointer from the parent instead of
 866	 * doing it ourselves.  This way we can let RCU fully do its
 867	 * thing without us interfering
 868	 */
 869	BUG_ON(tn != get_child(tp, cindex));
 
 870
 871	/* Double as long as the resulting node has a number of
 872	 * nonempty nodes that are above the threshold.
 873	 */
 874	while (should_inflate(tp, tn) && max_work) {
 875		tp = inflate(t, tn);
 876		if (!tp) {
 877#ifdef CONFIG_IP_FIB_TRIE_STATS
 878			this_cpu_inc(stats->resize_node_skipped);
 879#endif
 880			break;
 881		}
 882
 883		max_work--;
 884		tn = get_child(tp, cindex);
 
 885	}
 
 886
 887	/* update parent in case inflate failed */
 888	tp = node_parent(tn);
 889
 890	/* Return if at least one inflate is run */
 891	if (max_work != MAX_WORK)
 892		return tp;
 893
 894	/* Halve as long as the number of empty children in this
 895	 * node is above threshold.
 896	 */
 897	while (should_halve(tp, tn) && max_work) {
 898		tp = halve(t, tn);
 899		if (!tp) {
 900#ifdef CONFIG_IP_FIB_TRIE_STATS
 901			this_cpu_inc(stats->resize_node_skipped);
 902#endif
 
 
 
 
 
 
 
 
 
 
 
 
 
 903			break;
 904		}
 905
 906		max_work--;
 907		tn = get_child(tp, cindex);
 908	}
 
 909
 910	/* Only one child remains */
 911	if (should_collapse(tn))
 912		return collapse(t, tn);
 913
 914	/* update parent in case halve failed */
 915	return node_parent(tn);
 916}
 917
 918static void node_pull_suffix(struct key_vector *tn, unsigned char slen)
 919{
 920	unsigned char node_slen = tn->slen;
 921
 922	while ((node_slen > tn->pos) && (node_slen > slen)) {
 923		slen = update_suffix(tn);
 924		if (node_slen == slen)
 925			break;
 926
 927		tn = node_parent(tn);
 928		node_slen = tn->slen;
 929	}
 930}
 931
 932static void node_push_suffix(struct key_vector *tn, unsigned char slen)
 933{
 934	while (tn->slen < slen) {
 935		tn->slen = slen;
 936		tn = node_parent(tn);
 937	}
 938}
 939
 940/* rcu_read_lock needs to be hold by caller from readside */
 941static struct key_vector *fib_find_node(struct trie *t,
 942					struct key_vector **tp, u32 key)
 943{
 944	struct key_vector *pn, *n = t->kv;
 945	unsigned long index = 0;
 946
 947	do {
 948		pn = n;
 949		n = get_child_rcu(n, index);
 950
 951		if (!n)
 
 952			break;
 
 
 953
 954		index = get_cindex(key, n);
 
 
 955
 956		/* This bit of code is a bit tricky but it combines multiple
 957		 * checks into a single check.  The prefix consists of the
 958		 * prefix plus zeros for the bits in the cindex. The index
 959		 * is the difference between the key and this value.  From
 960		 * this we can actually derive several pieces of data.
 961		 *   if (index >= (1ul << bits))
 962		 *     we have a mismatch in skip bits and failed
 963		 *   else
 964		 *     we know the value is cindex
 965		 *
 966		 * This check is safe even if bits == KEYLENGTH due to the
 967		 * fact that we can only allocate a node with 32 bits if a
 968		 * long is greater than 32 bits.
 969		 */
 970		if (index >= (1ul << n->bits)) {
 971			n = NULL;
 972			break;
 973		}
 974
 975		/* keep searching until we find a perfect match leaf or NULL */
 976	} while (IS_TNODE(n));
 977
 978	*tp = pn;
 
 
 
 
 
 
 
 
 
 979
 980	return n;
 981}
 982
 983/* Return the first fib alias matching TOS with
 984 * priority less than or equal to PRIO.
 985 */
 986static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
 987					u8 tos, u32 prio, u32 tb_id)
 988{
 989	struct fib_alias *fa;
 
 
 
 
 
 
 
 
 
 
 990
 991	if (!fah)
 992		return NULL;
 993
 994	hlist_for_each_entry(fa, fah, fa_list) {
 995		if (fa->fa_slen < slen)
 996			continue;
 997		if (fa->fa_slen != slen)
 998			break;
 999		if (fa->tb_id > tb_id)
1000			continue;
1001		if (fa->tb_id != tb_id)
 
 
 
 
1002			break;
1003		if (fa->fa_tos > tos)
1004			continue;
1005		if (fa->fa_info->fib_priority >= prio || fa->fa_tos < tos)
1006			return fa;
1007	}
1008
1009	return NULL;
1010}
 
 
 
1011
1012static void trie_rebalance(struct trie *t, struct key_vector *tn)
1013{
1014	while (!IS_TRIE(tn))
1015		tn = resize(t, tn);
1016}
1017
1018static int fib_insert_node(struct trie *t, struct key_vector *tp,
1019			   struct fib_alias *new, t_key key)
1020{
1021	struct key_vector *n, *l;
1022
1023	l = leaf_new(key, new);
1024	if (!l)
1025		goto noleaf;
1026
1027	/* retrieve child from parent node */
1028	n = get_child(tp, get_index(key, tp));
1029
1030	/* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1031	 *
1032	 *  Add a new tnode here
1033	 *  first tnode need some special handling
1034	 *  leaves us in position for handling as case 3
1035	 */
1036	if (n) {
1037		struct key_vector *tn;
1038
1039		tn = tnode_new(key, __fls(key ^ n->key), 1);
1040		if (!tn)
1041			goto notnode;
1042
1043		/* initialize routes out of node */
1044		NODE_INIT_PARENT(tn, tp);
1045		put_child(tn, get_index(key, tn) ^ 1, n);
1046
1047		/* start adding routes into the node */
1048		put_child_root(tp, key, tn);
1049		node_set_parent(n, tn);
1050
1051		/* parent now has a NULL spot where the leaf can go */
1052		tp = tn;
 
1053	}
1054
1055	/* Case 3: n is NULL, and will just insert a new leaf */
1056	node_push_suffix(tp, new->fa_slen);
1057	NODE_INIT_PARENT(l, tp);
1058	put_child_root(tp, key, l);
1059	trie_rebalance(t, tp);
1060
1061	return 0;
1062notnode:
1063	node_free(l);
1064noleaf:
1065	return -ENOMEM;
1066}
1067
1068/* fib notifier for ADD is sent before calling fib_insert_alias with
1069 * the expectation that the only possible failure ENOMEM
1070 */
1071static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1072			    struct key_vector *l, struct fib_alias *new,
1073			    struct fib_alias *fa, t_key key)
1074{
1075	if (!l)
1076		return fib_insert_node(t, tp, new, key);
1077
1078	if (fa) {
1079		hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1080	} else {
1081		struct fib_alias *last;
 
 
 
 
1082
1083		hlist_for_each_entry(last, &l->leaf, fa_list) {
1084			if (new->fa_slen < last->fa_slen)
1085				break;
1086			if ((new->fa_slen == last->fa_slen) &&
1087			    (new->tb_id > last->tb_id))
1088				break;
1089			fa = last;
 
 
 
 
1090		}
1091
1092		if (fa)
1093			hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1094		else
1095			hlist_add_head_rcu(&new->fa_list, &l->leaf);
1096	}
1097
1098	/* if we added to the tail node then we need to update slen */
1099	if (l->slen < new->fa_slen) {
1100		l->slen = new->fa_slen;
1101		node_push_suffix(tp, new->fa_slen);
1102	}
1103
1104	return 0;
1105}
 
1106
1107static bool fib_valid_key_len(u32 key, u8 plen, struct netlink_ext_ack *extack)
1108{
1109	if (plen > KEYLENGTH) {
1110		NL_SET_ERR_MSG(extack, "Invalid prefix length");
1111		return false;
 
 
 
1112	}
1113
1114	if ((plen < KEYLENGTH) && (key << plen)) {
1115		NL_SET_ERR_MSG(extack,
1116			       "Invalid prefix for given prefix length");
1117		return false;
1118	}
1119
1120	return true;
 
 
1121}
1122
1123/* Caller must hold RTNL. */
1124int fib_table_insert(struct net *net, struct fib_table *tb,
1125		     struct fib_config *cfg, struct netlink_ext_ack *extack)
 
1126{
1127	enum fib_event_type event = FIB_EVENT_ENTRY_ADD;
1128	struct trie *t = (struct trie *)tb->tb_data;
1129	struct fib_alias *fa, *new_fa;
1130	struct key_vector *l, *tp;
1131	u16 nlflags = NLM_F_EXCL;
1132	struct fib_info *fi;
1133	u8 plen = cfg->fc_dst_len;
1134	u8 slen = KEYLENGTH - plen;
1135	u8 tos = cfg->fc_tos;
1136	u32 key;
1137	int err;
 
 
 
 
1138
1139	key = ntohl(cfg->fc_dst);
1140
1141	if (!fib_valid_key_len(key, plen, extack))
 
 
 
 
1142		return -EINVAL;
1143
1144	pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1145
1146	fi = fib_create_info(cfg, extack);
1147	if (IS_ERR(fi)) {
1148		err = PTR_ERR(fi);
1149		goto err;
1150	}
1151
1152	l = fib_find_node(t, &tp, key);
1153	fa = l ? fib_find_alias(&l->leaf, slen, tos, fi->fib_priority,
1154				tb->tb_id) : NULL;
 
 
 
 
1155
1156	/* Now fa, if non-NULL, points to the first fib alias
1157	 * with the same keys [prefix,tos,priority], if such key already
1158	 * exists or to the node before which we will insert new one.
1159	 *
1160	 * If fa is NULL, we will need to allocate a new one and
1161	 * insert to the tail of the section matching the suffix length
1162	 * of the new alias.
 
 
1163	 */
1164
1165	if (fa && fa->fa_tos == tos &&
1166	    fa->fa_info->fib_priority == fi->fib_priority) {
1167		struct fib_alias *fa_first, *fa_match;
1168
1169		err = -EEXIST;
1170		if (cfg->fc_nlflags & NLM_F_EXCL)
1171			goto out;
1172
1173		nlflags &= ~NLM_F_EXCL;
1174
1175		/* We have 2 goals:
1176		 * 1. Find exact match for type, scope, fib_info to avoid
1177		 * duplicate routes
1178		 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1179		 */
1180		fa_match = NULL;
1181		fa_first = fa;
1182		hlist_for_each_entry_from(fa, fa_list) {
1183			if ((fa->fa_slen != slen) ||
1184			    (fa->tb_id != tb->tb_id) ||
1185			    (fa->fa_tos != tos))
1186				break;
1187			if (fa->fa_info->fib_priority != fi->fib_priority)
1188				break;
1189			if (fa->fa_type == cfg->fc_type &&
1190			    fa->fa_info == fi) {
1191				fa_match = fa;
1192				break;
1193			}
1194		}
1195
1196		if (cfg->fc_nlflags & NLM_F_REPLACE) {
1197			struct fib_info *fi_drop;
1198			u8 state;
1199
1200			nlflags |= NLM_F_REPLACE;
1201			fa = fa_first;
1202			if (fa_match) {
1203				if (fa == fa_match)
1204					err = 0;
1205				goto out;
1206			}
1207			err = -ENOBUFS;
1208			new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1209			if (!new_fa)
1210				goto out;
1211
1212			fi_drop = fa->fa_info;
1213			new_fa->fa_tos = fa->fa_tos;
1214			new_fa->fa_info = fi;
1215			new_fa->fa_type = cfg->fc_type;
1216			state = fa->fa_state;
1217			new_fa->fa_state = state & ~FA_S_ACCESSED;
1218			new_fa->fa_slen = fa->fa_slen;
1219			new_fa->tb_id = tb->tb_id;
1220			new_fa->fa_default = -1;
1221
1222			err = call_fib_entry_notifiers(net,
1223						       FIB_EVENT_ENTRY_REPLACE,
1224						       key, plen, new_fa,
1225						       extack);
1226			if (err)
1227				goto out_free_new_fa;
1228
1229			rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1230				  tb->tb_id, &cfg->fc_nlinfo, nlflags);
1231
1232			hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1233
 
1234			alias_free_mem_rcu(fa);
1235
1236			fib_release_info(fi_drop);
1237			if (state & FA_S_ACCESSED)
1238				rt_cache_flush(cfg->fc_nlinfo.nl_net);
 
 
1239
1240			goto succeeded;
1241		}
1242		/* Error if we find a perfect match which
1243		 * uses the same scope, type, and nexthop
1244		 * information.
1245		 */
1246		if (fa_match)
1247			goto out;
1248
1249		if (cfg->fc_nlflags & NLM_F_APPEND) {
1250			event = FIB_EVENT_ENTRY_APPEND;
1251			nlflags |= NLM_F_APPEND;
1252		} else {
1253			fa = fa_first;
1254		}
1255	}
1256	err = -ENOENT;
1257	if (!(cfg->fc_nlflags & NLM_F_CREATE))
1258		goto out;
1259
1260	nlflags |= NLM_F_CREATE;
1261	err = -ENOBUFS;
1262	new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1263	if (!new_fa)
1264		goto out;
1265
1266	new_fa->fa_info = fi;
1267	new_fa->fa_tos = tos;
1268	new_fa->fa_type = cfg->fc_type;
1269	new_fa->fa_state = 0;
1270	new_fa->fa_slen = slen;
1271	new_fa->tb_id = tb->tb_id;
1272	new_fa->fa_default = -1;
1273
1274	err = call_fib_entry_notifiers(net, event, key, plen, new_fa, extack);
1275	if (err)
1276		goto out_free_new_fa;
1277
1278	/* Insert new entry to the list. */
1279	err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1280	if (err)
1281		goto out_fib_notif;
1282
1283	if (!plen)
1284		tb->tb_num_default++;
1285
1286	rt_cache_flush(cfg->fc_nlinfo.nl_net);
1287	rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1288		  &cfg->fc_nlinfo, nlflags);
 
 
 
1289succeeded:
1290	return 0;
1291
1292out_fib_notif:
1293	/* notifier was sent that entry would be added to trie, but
1294	 * the add failed and need to recover. Only failure for
1295	 * fib_insert_alias is ENOMEM.
1296	 */
1297	NL_SET_ERR_MSG(extack, "Failed to insert route into trie");
1298	call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, key,
1299				 plen, new_fa, NULL);
1300out_free_new_fa:
1301	kmem_cache_free(fn_alias_kmem, new_fa);
1302out:
1303	fib_release_info(fi);
1304err:
1305	return err;
1306}
1307
1308static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1309{
1310	t_key prefix = n->key;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1311
1312	return (key ^ prefix) & (prefix | -prefix);
1313}
1314
1315/* should be called with rcu_read_lock */
1316int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1317		     struct fib_result *res, int fib_flags)
1318{
1319	struct trie *t = (struct trie *) tb->tb_data;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1320#ifdef CONFIG_IP_FIB_TRIE_STATS
1321	struct trie_use_stats __percpu *stats = t->stats;
1322#endif
1323	const t_key key = ntohl(flp->daddr);
1324	struct key_vector *n, *pn;
1325	struct fib_alias *fa;
1326	unsigned long index;
1327	t_key cindex;
1328
1329	trace_fib_table_lookup(tb->tb_id, flp);
 
 
 
 
 
 
 
1330
1331	pn = t->kv;
1332	cindex = 0;
 
1333
1334	n = get_child_rcu(pn, cindex);
1335	if (!n)
1336		return -EAGAIN;
 
 
1337
 
1338#ifdef CONFIG_IP_FIB_TRIE_STATS
1339	this_cpu_inc(stats->gets);
1340#endif
 
 
 
 
 
 
 
 
 
1341
1342	/* Step 1: Travel to the longest prefix match in the trie */
1343	for (;;) {
1344		index = get_cindex(key, n);
1345
1346		/* This bit of code is a bit tricky but it combines multiple
1347		 * checks into a single check.  The prefix consists of the
1348		 * prefix plus zeros for the "bits" in the prefix. The index
1349		 * is the difference between the key and this value.  From
1350		 * this we can actually derive several pieces of data.
1351		 *   if (index >= (1ul << bits))
1352		 *     we have a mismatch in skip bits and failed
1353		 *   else
1354		 *     we know the value is cindex
 
1355		 *
1356		 * This check is safe even if bits == KEYLENGTH due to the
1357		 * fact that we can only allocate a node with 32 bits if a
1358		 * long is greater than 32 bits.
1359		 */
1360		if (index >= (1ul << n->bits))
1361			break;
1362
1363		/* we have found a leaf. Prefixes have already been compared */
1364		if (IS_LEAF(n))
1365			goto found;
1366
1367		/* only record pn and cindex if we are going to be chopping
1368		 * bits later.  Otherwise we are just wasting cycles.
 
 
 
 
 
 
1369		 */
1370		if (n->slen > n->pos) {
1371			pn = n;
1372			cindex = index;
1373		}
1374
1375		n = get_child_rcu(n, index);
1376		if (unlikely(!n))
1377			goto backtrace;
1378	}
1379
1380	/* Step 2: Sort out leaves and begin backtracing for longest prefix */
1381	for (;;) {
1382		/* record the pointer where our next node pointer is stored */
1383		struct key_vector __rcu **cptr = n->tnode;
1384
1385		/* This test verifies that none of the bits that differ
1386		 * between the key and the prefix exist in the region of
1387		 * the lsb and higher in the prefix.
 
 
 
 
 
 
 
1388		 */
1389		if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1390			goto backtrace;
1391
1392		/* exit out and process leaf */
1393		if (unlikely(IS_LEAF(n)))
1394			break;
 
 
 
 
 
1395
1396		/* Don't bother recording parent info.  Since we are in
1397		 * prefix match mode we will have to come back to wherever
1398		 * we started this traversal anyway
 
 
 
 
 
 
 
 
1399		 */
1400
1401		while ((n = rcu_dereference(*cptr)) == NULL) {
1402backtrace:
1403#ifdef CONFIG_IP_FIB_TRIE_STATS
1404			if (!n)
1405				this_cpu_inc(stats->null_node_hit);
1406#endif
1407			/* If we are at cindex 0 there are no more bits for
1408			 * us to strip at this level so we must ascend back
1409			 * up one level to see if there are any more bits to
1410			 * be stripped there.
1411			 */
1412			while (!cindex) {
1413				t_key pkey = pn->key;
1414
1415				/* If we don't have a parent then there is
1416				 * nothing for us to do as we do not have any
1417				 * further nodes to parse.
1418				 */
1419				if (IS_TRIE(pn))
1420					return -EAGAIN;
1421#ifdef CONFIG_IP_FIB_TRIE_STATS
1422				this_cpu_inc(stats->backtrack);
1423#endif
1424				/* Get Child's index */
1425				pn = node_parent_rcu(pn);
1426				cindex = get_index(pkey, pn);
1427			}
1428
1429			/* strip the least significant bit from the cindex */
1430			cindex &= cindex - 1;
 
 
 
 
 
1431
1432			/* grab pointer for next child node */
1433			cptr = &pn->tnode[cindex];
1434		}
1435	}
1436
1437found:
1438	/* this line carries forward the xor from earlier in the function */
1439	index = key ^ n->key;
1440
1441	/* Step 3: Process the leaf, if that fails fall back to backtracing */
1442	hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1443		struct fib_info *fi = fa->fa_info;
1444		int nhsel, err;
1445
1446		if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1447			if (index >= (1ul << fa->fa_slen))
1448				continue;
1449		}
1450		if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1451			continue;
1452		if (fi->fib_dead)
1453			continue;
1454		if (fa->fa_info->fib_scope < flp->flowi4_scope)
1455			continue;
1456		fib_alias_accessed(fa);
1457		err = fib_props[fa->fa_type].error;
1458		if (unlikely(err < 0)) {
1459#ifdef CONFIG_IP_FIB_TRIE_STATS
1460			this_cpu_inc(stats->semantic_match_passed);
1461#endif
1462			return err;
1463		}
1464		if (fi->fib_flags & RTNH_F_DEAD)
1465			continue;
1466		for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
1467			const struct fib_nh *nh = &fi->fib_nh[nhsel];
1468			struct in_device *in_dev = __in_dev_get_rcu(nh->nh_dev);
1469
1470			if (nh->nh_flags & RTNH_F_DEAD)
1471				continue;
1472			if (in_dev &&
1473			    IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1474			    nh->nh_flags & RTNH_F_LINKDOWN &&
1475			    !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1476				continue;
1477			if (!(flp->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF)) {
1478				if (flp->flowi4_oif &&
1479				    flp->flowi4_oif != nh->nh_oif)
1480					continue;
1481			}
 
 
1482
1483			if (!(fib_flags & FIB_LOOKUP_NOREF))
1484				refcount_inc(&fi->fib_clntref);
 
 
 
 
 
 
 
 
 
1485
1486			res->prefix = htonl(n->key);
1487			res->prefixlen = KEYLENGTH - fa->fa_slen;
1488			res->nh_sel = nhsel;
1489			res->type = fa->fa_type;
1490			res->scope = fi->fib_scope;
1491			res->fi = fi;
1492			res->table = tb;
1493			res->fa_head = &n->leaf;
1494#ifdef CONFIG_IP_FIB_TRIE_STATS
1495			this_cpu_inc(stats->semantic_match_passed);
1496#endif
1497			trace_fib_table_lookup_nh(nh);
1498
1499			return err;
1500		}
1501	}
1502#ifdef CONFIG_IP_FIB_TRIE_STATS
1503	this_cpu_inc(stats->semantic_match_miss);
1504#endif
1505	goto backtrace;
 
1506}
1507EXPORT_SYMBOL_GPL(fib_table_lookup);
1508
1509static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1510			     struct key_vector *l, struct fib_alias *old)
 
 
1511{
1512	/* record the location of the previous list_info entry */
1513	struct hlist_node **pprev = old->fa_list.pprev;
1514	struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1515
1516	/* remove the fib_alias from the list */
1517	hlist_del_rcu(&old->fa_list);
1518
1519	/* if we emptied the list this leaf will be freed and we can sort
1520	 * out parent suffix lengths as a part of trie_rebalance
1521	 */
1522	if (hlist_empty(&l->leaf)) {
1523		if (tp->slen == l->slen)
1524			node_pull_suffix(tp, tp->pos);
1525		put_child_root(tp, l->key, NULL);
1526		node_free(l);
1527		trie_rebalance(t, tp);
1528		return;
1529	}
1530
1531	/* only access fa if it is pointing at the last valid hlist_node */
1532	if (*pprev)
1533		return;
1534
1535	/* update the trie with the latest suffix length */
1536	l->slen = fa->fa_slen;
1537	node_pull_suffix(tp, fa->fa_slen);
1538}
1539
1540/* Caller must hold RTNL. */
1541int fib_table_delete(struct net *net, struct fib_table *tb,
1542		     struct fib_config *cfg, struct netlink_ext_ack *extack)
 
1543{
1544	struct trie *t = (struct trie *) tb->tb_data;
1545	struct fib_alias *fa, *fa_to_delete;
1546	struct key_vector *l, *tp;
1547	u8 plen = cfg->fc_dst_len;
1548	u8 slen = KEYLENGTH - plen;
1549	u8 tos = cfg->fc_tos;
1550	u32 key;
 
 
 
 
 
 
1551
1552	key = ntohl(cfg->fc_dst);
 
1553
1554	if (!fib_valid_key_len(key, plen, extack))
1555		return -EINVAL;
1556
1557	l = fib_find_node(t, &tp, key);
 
 
1558	if (!l)
1559		return -ESRCH;
1560
1561	fa = fib_find_alias(&l->leaf, slen, tos, 0, tb->tb_id);
 
 
1562	if (!fa)
1563		return -ESRCH;
1564
1565	pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1566
1567	fa_to_delete = NULL;
1568	hlist_for_each_entry_from(fa, fa_list) {
 
1569		struct fib_info *fi = fa->fa_info;
1570
1571		if ((fa->fa_slen != slen) ||
1572		    (fa->tb_id != tb->tb_id) ||
1573		    (fa->fa_tos != tos))
1574			break;
1575
1576		if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1577		    (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1578		     fa->fa_info->fib_scope == cfg->fc_scope) &&
1579		    (!cfg->fc_prefsrc ||
1580		     fi->fib_prefsrc == cfg->fc_prefsrc) &&
1581		    (!cfg->fc_protocol ||
1582		     fi->fib_protocol == cfg->fc_protocol) &&
1583		    fib_nh_match(cfg, fi, extack) == 0 &&
1584		    fib_metrics_match(cfg, fi)) {
1585			fa_to_delete = fa;
1586			break;
1587		}
1588	}
1589
1590	if (!fa_to_delete)
1591		return -ESRCH;
1592
1593	call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, key, plen,
1594				 fa_to_delete, extack);
1595	rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1596		  &cfg->fc_nlinfo, 0);
1597
 
 
 
 
 
1598	if (!plen)
1599		tb->tb_num_default--;
1600
1601	fib_remove_alias(t, tp, l, fa_to_delete);
 
 
 
 
 
 
1602
1603	if (fa_to_delete->fa_state & FA_S_ACCESSED)
1604		rt_cache_flush(cfg->fc_nlinfo.nl_net);
1605
1606	fib_release_info(fa_to_delete->fa_info);
1607	alias_free_mem_rcu(fa_to_delete);
1608	return 0;
1609}
1610
1611/* Scan for the next leaf starting at the provided key value */
1612static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1613{
1614	struct key_vector *pn, *n = *tn;
1615	unsigned long cindex;
1616
1617	/* this loop is meant to try and find the key in the trie */
1618	do {
1619		/* record parent and next child index */
1620		pn = n;
1621		cindex = (key > pn->key) ? get_index(key, pn) : 0;
1622
1623		if (cindex >> pn->bits)
1624			break;
1625
1626		/* descend into the next child */
1627		n = get_child_rcu(pn, cindex++);
1628		if (!n)
1629			break;
1630
1631		/* guarantee forward progress on the keys */
1632		if (IS_LEAF(n) && (n->key >= key))
1633			goto found;
1634	} while (IS_TNODE(n));
1635
1636	/* this loop will search for the next leaf with a greater key */
1637	while (!IS_TRIE(pn)) {
1638		/* if we exhausted the parent node we will need to climb */
1639		if (cindex >= (1ul << pn->bits)) {
1640			t_key pkey = pn->key;
1641
1642			pn = node_parent_rcu(pn);
1643			cindex = get_index(pkey, pn) + 1;
1644			continue;
 
 
1645		}
1646
1647		/* grab the next available node */
1648		n = get_child_rcu(pn, cindex++);
1649		if (!n)
1650			continue;
1651
1652		/* no need to compare keys since we bumped the index */
1653		if (IS_LEAF(n))
1654			goto found;
1655
1656		/* Rescan start scanning in new node */
1657		pn = n;
1658		cindex = 0;
1659	}
1660
1661	*tn = pn;
1662	return NULL; /* Root of trie */
1663found:
1664	/* if we are at the limit for keys just return NULL for the tnode */
1665	*tn = pn;
1666	return n;
1667}
1668
1669static void fib_trie_free(struct fib_table *tb)
1670{
1671	struct trie *t = (struct trie *)tb->tb_data;
1672	struct key_vector *pn = t->kv;
1673	unsigned long cindex = 1;
1674	struct hlist_node *tmp;
1675	struct fib_alias *fa;
1676
1677	/* walk trie in reverse order and free everything */
1678	for (;;) {
1679		struct key_vector *n;
1680
1681		if (!(cindex--)) {
1682			t_key pkey = pn->key;
1683
1684			if (IS_TRIE(pn))
1685				break;
1686
1687			n = pn;
1688			pn = node_parent(pn);
1689
1690			/* drop emptied tnode */
1691			put_child_root(pn, n->key, NULL);
1692			node_free(n);
1693
1694			cindex = get_index(pkey, pn);
1695
1696			continue;
1697		}
1698
1699		/* grab the next available node */
1700		n = get_child(pn, cindex);
1701		if (!n)
1702			continue;
1703
1704		if (IS_TNODE(n)) {
1705			/* record pn and cindex for leaf walking */
1706			pn = n;
1707			cindex = 1ul << n->bits;
1708
1709			continue;
1710		}
1711
1712		hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1713			hlist_del_rcu(&fa->fa_list);
1714			alias_free_mem_rcu(fa);
1715		}
1716
1717		put_child_root(pn, n->key, NULL);
1718		node_free(n);
1719	}
1720
1721#ifdef CONFIG_IP_FIB_TRIE_STATS
1722	free_percpu(t->stats);
1723#endif
1724	kfree(tb);
1725}
1726
1727struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
 
 
 
 
1728{
1729	struct trie *ot = (struct trie *)oldtb->tb_data;
1730	struct key_vector *l, *tp = ot->kv;
1731	struct fib_table *local_tb;
1732	struct fib_alias *fa;
1733	struct trie *lt;
1734	t_key key = 0;
1735
1736	if (oldtb->tb_data == oldtb->__data)
1737		return oldtb;
1738
1739	local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1740	if (!local_tb)
1741		return NULL;
1742
1743	lt = (struct trie *)local_tb->tb_data;
1744
1745	while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1746		struct key_vector *local_l = NULL, *local_tp;
1747
1748		hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1749			struct fib_alias *new_fa;
 
 
1750
1751			if (local_tb->tb_id != fa->tb_id)
 
 
1752				continue;
1753
1754			/* clone fa for new local table */
1755			new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1756			if (!new_fa)
1757				goto out;
1758
1759			memcpy(new_fa, fa, sizeof(*fa));
1760
1761			/* insert clone into table */
1762			if (!local_l)
1763				local_l = fib_find_node(lt, &local_tp, l->key);
1764
1765			if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1766					     NULL, l->key)) {
1767				kmem_cache_free(fn_alias_kmem, new_fa);
1768				goto out;
1769			}
1770		}
1771
1772		/* stop loop if key wrapped back to 0 */
1773		key = l->key + 1;
1774		if (key < l->key)
1775			break;
1776	}
1777
1778	return local_tb;
1779out:
1780	fib_trie_free(local_tb);
1781
1782	return NULL;
1783}
1784
1785/* Caller must hold RTNL */
1786void fib_table_flush_external(struct fib_table *tb)
1787{
1788	struct trie *t = (struct trie *)tb->tb_data;
1789	struct key_vector *pn = t->kv;
1790	unsigned long cindex = 1;
1791	struct hlist_node *tmp;
1792	struct fib_alias *fa;
1793
1794	/* walk trie in reverse order */
1795	for (;;) {
1796		unsigned char slen = 0;
1797		struct key_vector *n;
1798
1799		if (!(cindex--)) {
1800			t_key pkey = pn->key;
1801
1802			/* cannot resize the trie vector */
1803			if (IS_TRIE(pn))
1804				break;
1805
1806			/* update the suffix to address pulled leaves */
1807			if (pn->slen > pn->pos)
1808				update_suffix(pn);
1809
1810			/* resize completed node */
1811			pn = resize(t, pn);
1812			cindex = get_index(pkey, pn);
1813
1814			continue;
1815		}
1816
1817		/* grab the next available node */
1818		n = get_child(pn, cindex);
1819		if (!n)
1820			continue;
1821
1822		if (IS_TNODE(n)) {
1823			/* record pn and cindex for leaf walking */
1824			pn = n;
1825			cindex = 1ul << n->bits;
1826
1827			continue;
1828		}
1829
1830		hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1831			/* if alias was cloned to local then we just
1832			 * need to remove the local copy from main
1833			 */
1834			if (tb->tb_id != fa->tb_id) {
1835				hlist_del_rcu(&fa->fa_list);
1836				alias_free_mem_rcu(fa);
1837				continue;
1838			}
1839
1840			/* record local slen */
1841			slen = fa->fa_slen;
1842		}
1843
1844		/* update leaf slen */
1845		n->slen = slen;
1846
1847		if (hlist_empty(&n->leaf)) {
1848			put_child_root(pn, n->key, NULL);
1849			node_free(n);
1850		}
1851	}
1852}
1853
1854/* Caller must hold RTNL. */
1855int fib_table_flush(struct net *net, struct fib_table *tb)
1856{
1857	struct trie *t = (struct trie *)tb->tb_data;
1858	struct key_vector *pn = t->kv;
1859	unsigned long cindex = 1;
1860	struct hlist_node *tmp;
1861	struct fib_alias *fa;
1862	int found = 0;
1863
1864	/* walk trie in reverse order */
1865	for (;;) {
1866		unsigned char slen = 0;
1867		struct key_vector *n;
1868
1869		if (!(cindex--)) {
1870			t_key pkey = pn->key;
1871
1872			/* cannot resize the trie vector */
1873			if (IS_TRIE(pn))
1874				break;
1875
1876			/* update the suffix to address pulled leaves */
1877			if (pn->slen > pn->pos)
1878				update_suffix(pn);
1879
1880			/* resize completed node */
1881			pn = resize(t, pn);
1882			cindex = get_index(pkey, pn);
1883
1884			continue;
1885		}
1886
1887		/* grab the next available node */
1888		n = get_child(pn, cindex);
1889		if (!n)
1890			continue;
1891
1892		if (IS_TNODE(n)) {
1893			/* record pn and cindex for leaf walking */
1894			pn = n;
1895			cindex = 1ul << n->bits;
1896
1897			continue;
1898		}
1899
1900		hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1901			struct fib_info *fi = fa->fa_info;
1902
1903			if (!fi || !(fi->fib_flags & RTNH_F_DEAD) ||
1904			    tb->tb_id != fa->tb_id) {
1905				slen = fa->fa_slen;
1906				continue;
1907			}
1908
1909			call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL,
1910						 n->key,
1911						 KEYLENGTH - fa->fa_slen, fa,
1912						 NULL);
1913			hlist_del_rcu(&fa->fa_list);
1914			fib_release_info(fa->fa_info);
1915			alias_free_mem_rcu(fa);
1916			found++;
1917		}
1918
1919		/* update leaf slen */
1920		n->slen = slen;
1921
1922		if (hlist_empty(&n->leaf)) {
1923			put_child_root(pn, n->key, NULL);
1924			node_free(n);
1925		}
1926	}
1927
 
 
 
1928	pr_debug("trie_flush found=%d\n", found);
1929	return found;
1930}
1931
1932static void fib_leaf_notify(struct net *net, struct key_vector *l,
1933			    struct fib_table *tb, struct notifier_block *nb)
1934{
1935	struct fib_alias *fa;
1936
1937	hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1938		struct fib_info *fi = fa->fa_info;
1939
1940		if (!fi)
1941			continue;
1942
1943		/* local and main table can share the same trie,
1944		 * so don't notify twice for the same entry.
1945		 */
1946		if (tb->tb_id != fa->tb_id)
1947			continue;
1948
1949		call_fib_entry_notifier(nb, net, FIB_EVENT_ENTRY_ADD, l->key,
1950					KEYLENGTH - fa->fa_slen, fa);
1951	}
1952}
1953
1954static void fib_table_notify(struct net *net, struct fib_table *tb,
1955			     struct notifier_block *nb)
 
1956{
1957	struct trie *t = (struct trie *)tb->tb_data;
1958	struct key_vector *l, *tp = t->kv;
1959	t_key key = 0;
1960
1961	while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1962		fib_leaf_notify(net, l, tb, nb);
1963
1964		key = l->key + 1;
1965		/* stop in case of wrap around */
1966		if (key < l->key)
1967			break;
1968	}
1969}
1970
1971void fib_notify(struct net *net, struct notifier_block *nb)
1972{
1973	unsigned int h;
1974
1975	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
1976		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
1977		struct fib_table *tb;
 
 
1978
1979		hlist_for_each_entry_rcu(tb, head, tb_hlist)
1980			fib_table_notify(net, tb, nb);
 
 
 
 
 
 
 
 
 
 
 
1981	}
 
 
1982}
1983
1984static void __trie_free_rcu(struct rcu_head *head)
1985{
1986	struct fib_table *tb = container_of(head, struct fib_table, rcu);
1987#ifdef CONFIG_IP_FIB_TRIE_STATS
1988	struct trie *t = (struct trie *)tb->tb_data;
1989
1990	if (tb->tb_data == tb->__data)
1991		free_percpu(t->stats);
1992#endif /* CONFIG_IP_FIB_TRIE_STATS */
1993	kfree(tb);
1994}
1995
1996void fib_free_table(struct fib_table *tb)
1997{
1998	call_rcu(&tb->rcu, __trie_free_rcu);
1999}
2000
2001static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
2002			     struct sk_buff *skb, struct netlink_callback *cb)
2003{
2004	__be32 xkey = htonl(l->key);
2005	struct fib_alias *fa;
2006	int i, s_i;
2007
2008	s_i = cb->args[4];
2009	i = 0;
2010
2011	/* rcu_read_lock is hold by caller */
2012	hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2013		int err;
2014
2015		if (i < s_i) {
2016			i++;
2017			continue;
2018		}
2019
2020		if (tb->tb_id != fa->tb_id) {
2021			i++;
 
 
2022			continue;
2023		}
2024
2025		err = fib_dump_info(skb, NETLINK_CB(cb->skb).portid,
2026				    cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2027				    tb->tb_id, fa->fa_type,
2028				    xkey, KEYLENGTH - fa->fa_slen,
2029				    fa->fa_tos, fa->fa_info, NLM_F_MULTI);
2030		if (err < 0) {
2031			cb->args[4] = i;
2032			return err;
2033		}
2034		i++;
2035	}
2036
2037	cb->args[4] = i;
2038	return skb->len;
2039}
2040
2041/* rcu_read_lock needs to be hold by caller from readside */
2042int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
2043		   struct netlink_callback *cb)
2044{
2045	struct trie *t = (struct trie *)tb->tb_data;
2046	struct key_vector *l, *tp = t->kv;
 
 
 
 
2047	/* Dump starting at last key.
2048	 * Note: 0.0.0.0/0 (ie default) is first key.
2049	 */
2050	int count = cb->args[2];
2051	t_key key = cb->args[3];
2052
2053	while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2054		int err;
 
 
 
 
 
2055
2056		err = fn_trie_dump_leaf(l, tb, skb, cb);
2057		if (err < 0) {
2058			cb->args[3] = key;
2059			cb->args[2] = count;
2060			return err;
 
2061		}
2062
2063		++count;
2064		key = l->key + 1;
2065
2066		memset(&cb->args[4], 0,
2067		       sizeof(cb->args) - 4*sizeof(cb->args[0]));
2068
2069		/* stop loop if key wrapped back to 0 */
2070		if (key < l->key)
2071			break;
2072	}
2073
2074	cb->args[3] = key;
2075	cb->args[2] = count;
2076
2077	return skb->len;
2078}
2079
2080void __init fib_trie_init(void)
2081{
2082	fn_alias_kmem = kmem_cache_create("ip_fib_alias",
2083					  sizeof(struct fib_alias),
2084					  0, SLAB_PANIC, NULL);
2085
2086	trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
2087					   LEAF_SIZE,
 
2088					   0, SLAB_PANIC, NULL);
2089}
2090
2091struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
 
2092{
2093	struct fib_table *tb;
2094	struct trie *t;
2095	size_t sz = sizeof(*tb);
2096
2097	if (!alias)
2098		sz += sizeof(struct trie);
2099
2100	tb = kzalloc(sz, GFP_KERNEL);
2101	if (!tb)
2102		return NULL;
2103
2104	tb->tb_id = id;
 
2105	tb->tb_num_default = 0;
2106	tb->tb_data = (alias ? alias->__data : tb->__data);
2107
2108	if (alias)
2109		return tb;
2110
2111	t = (struct trie *) tb->tb_data;
2112	t->kv[0].pos = KEYLENGTH;
2113	t->kv[0].slen = KEYLENGTH;
2114#ifdef CONFIG_IP_FIB_TRIE_STATS
2115	t->stats = alloc_percpu(struct trie_use_stats);
2116	if (!t->stats) {
2117		kfree(tb);
2118		tb = NULL;
2119	}
2120#endif
2121
2122	return tb;
2123}
2124
2125#ifdef CONFIG_PROC_FS
2126/* Depth first Trie walk iterator */
2127struct fib_trie_iter {
2128	struct seq_net_private p;
2129	struct fib_table *tb;
2130	struct key_vector *tnode;
2131	unsigned int index;
2132	unsigned int depth;
2133};
2134
2135static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2136{
2137	unsigned long cindex = iter->index;
2138	struct key_vector *pn = iter->tnode;
2139	t_key pkey;
 
 
 
 
2140
2141	pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2142		 iter->tnode, iter->index, iter->depth);
 
 
 
2143
2144	while (!IS_TRIE(pn)) {
2145		while (cindex < child_length(pn)) {
2146			struct key_vector *n = get_child_rcu(pn, cindex++);
2147
2148			if (!n)
2149				continue;
2150
2151			if (IS_LEAF(n)) {
2152				iter->tnode = pn;
2153				iter->index = cindex;
2154			} else {
2155				/* push down one level */
2156				iter->tnode = n;
2157				iter->index = 0;
2158				++iter->depth;
2159			}
2160
2161			return n;
2162		}
2163
2164		/* Current node exhausted, pop back up */
2165		pkey = pn->key;
2166		pn = node_parent_rcu(pn);
2167		cindex = get_index(pkey, pn) + 1;
2168		--iter->depth;
2169	}
2170
2171	/* record root node so further searches know we are done */
2172	iter->tnode = pn;
2173	iter->index = 0;
 
 
 
 
 
2174
 
2175	return NULL;
2176}
2177
2178static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2179					     struct trie *t)
2180{
2181	struct key_vector *n, *pn;
2182
2183	if (!t)
2184		return NULL;
2185
2186	pn = t->kv;
2187	n = rcu_dereference(pn->tnode[0]);
2188	if (!n)
2189		return NULL;
2190
2191	if (IS_TNODE(n)) {
2192		iter->tnode = n;
2193		iter->index = 0;
2194		iter->depth = 1;
2195	} else {
2196		iter->tnode = pn;
2197		iter->index = 0;
2198		iter->depth = 0;
2199	}
2200
2201	return n;
2202}
2203
2204static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2205{
2206	struct key_vector *n;
2207	struct fib_trie_iter iter;
2208
2209	memset(s, 0, sizeof(*s));
2210
2211	rcu_read_lock();
2212	for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2213		if (IS_LEAF(n)) {
2214			struct fib_alias *fa;
 
 
2215
2216			s->leaves++;
2217			s->totdepth += iter.depth;
2218			if (iter.depth > s->maxdepth)
2219				s->maxdepth = iter.depth;
2220
2221			hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2222				++s->prefixes;
2223		} else {
 
 
 
2224			s->tnodes++;
2225			if (n->bits < MAX_STAT_DEPTH)
2226				s->nodesizes[n->bits]++;
2227			s->nullpointers += tn_info(n)->empty_children;
 
 
 
2228		}
2229	}
2230	rcu_read_unlock();
2231}
2232
2233/*
2234 *	This outputs /proc/net/fib_triestats
2235 */
2236static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2237{
2238	unsigned int i, max, pointers, bytes, avdepth;
2239
2240	if (stat->leaves)
2241		avdepth = stat->totdepth*100 / stat->leaves;
2242	else
2243		avdepth = 0;
2244
2245	seq_printf(seq, "\tAver depth:     %u.%02d\n",
2246		   avdepth / 100, avdepth % 100);
2247	seq_printf(seq, "\tMax depth:      %u\n", stat->maxdepth);
2248
2249	seq_printf(seq, "\tLeaves:         %u\n", stat->leaves);
2250	bytes = LEAF_SIZE * stat->leaves;
2251
2252	seq_printf(seq, "\tPrefixes:       %u\n", stat->prefixes);
2253	bytes += sizeof(struct fib_alias) * stat->prefixes;
2254
2255	seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2256	bytes += TNODE_SIZE(0) * stat->tnodes;
2257
2258	max = MAX_STAT_DEPTH;
2259	while (max > 0 && stat->nodesizes[max-1] == 0)
2260		max--;
2261
2262	pointers = 0;
2263	for (i = 1; i < max; i++)
2264		if (stat->nodesizes[i] != 0) {
2265			seq_printf(seq, "  %u: %u",  i, stat->nodesizes[i]);
2266			pointers += (1<<i) * stat->nodesizes[i];
2267		}
2268	seq_putc(seq, '\n');
2269	seq_printf(seq, "\tPointers: %u\n", pointers);
2270
2271	bytes += sizeof(struct key_vector *) * pointers;
2272	seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2273	seq_printf(seq, "Total size: %u  kB\n", (bytes + 1023) / 1024);
2274}
2275
2276#ifdef CONFIG_IP_FIB_TRIE_STATS
2277static void trie_show_usage(struct seq_file *seq,
2278			    const struct trie_use_stats __percpu *stats)
2279{
2280	struct trie_use_stats s = { 0 };
2281	int cpu;
2282
2283	/* loop through all of the CPUs and gather up the stats */
2284	for_each_possible_cpu(cpu) {
2285		const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2286
2287		s.gets += pcpu->gets;
2288		s.backtrack += pcpu->backtrack;
2289		s.semantic_match_passed += pcpu->semantic_match_passed;
2290		s.semantic_match_miss += pcpu->semantic_match_miss;
2291		s.null_node_hit += pcpu->null_node_hit;
2292		s.resize_node_skipped += pcpu->resize_node_skipped;
2293	}
2294
2295	seq_printf(seq, "\nCounters:\n---------\n");
2296	seq_printf(seq, "gets = %u\n", s.gets);
2297	seq_printf(seq, "backtracks = %u\n", s.backtrack);
2298	seq_printf(seq, "semantic match passed = %u\n",
2299		   s.semantic_match_passed);
2300	seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2301	seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2302	seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
 
 
2303}
2304#endif /*  CONFIG_IP_FIB_TRIE_STATS */
2305
2306static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2307{
2308	if (tb->tb_id == RT_TABLE_LOCAL)
2309		seq_puts(seq, "Local:\n");
2310	else if (tb->tb_id == RT_TABLE_MAIN)
2311		seq_puts(seq, "Main:\n");
2312	else
2313		seq_printf(seq, "Id %d:\n", tb->tb_id);
2314}
2315
2316
2317static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2318{
2319	struct net *net = (struct net *)seq->private;
2320	unsigned int h;
2321
2322	seq_printf(seq,
2323		   "Basic info: size of leaf:"
2324		   " %zd bytes, size of tnode: %zd bytes.\n",
2325		   LEAF_SIZE, TNODE_SIZE(0));
2326
2327	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2328		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
 
2329		struct fib_table *tb;
2330
2331		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2332			struct trie *t = (struct trie *) tb->tb_data;
2333			struct trie_stat stat;
2334
2335			if (!t)
2336				continue;
2337
2338			fib_table_print(seq, tb);
2339
2340			trie_collect_stats(t, &stat);
2341			trie_show_stats(seq, &stat);
2342#ifdef CONFIG_IP_FIB_TRIE_STATS
2343			trie_show_usage(seq, t->stats);
2344#endif
2345		}
2346	}
2347
2348	return 0;
2349}
2350
2351static int fib_triestat_seq_open(struct inode *inode, struct file *file)
2352{
2353	return single_open_net(inode, file, fib_triestat_seq_show);
2354}
2355
2356static const struct file_operations fib_triestat_fops = {
 
2357	.open	= fib_triestat_seq_open,
2358	.read	= seq_read,
2359	.llseek	= seq_lseek,
2360	.release = single_release_net,
2361};
2362
2363static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2364{
2365	struct fib_trie_iter *iter = seq->private;
2366	struct net *net = seq_file_net(seq);
2367	loff_t idx = 0;
2368	unsigned int h;
2369
2370	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2371		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
 
2372		struct fib_table *tb;
2373
2374		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2375			struct key_vector *n;
2376
2377			for (n = fib_trie_get_first(iter,
2378						    (struct trie *) tb->tb_data);
2379			     n; n = fib_trie_get_next(iter))
2380				if (pos == idx++) {
2381					iter->tb = tb;
2382					return n;
2383				}
2384		}
2385	}
2386
2387	return NULL;
2388}
2389
2390static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2391	__acquires(RCU)
2392{
2393	rcu_read_lock();
2394	return fib_trie_get_idx(seq, *pos);
2395}
2396
2397static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2398{
2399	struct fib_trie_iter *iter = seq->private;
2400	struct net *net = seq_file_net(seq);
2401	struct fib_table *tb = iter->tb;
2402	struct hlist_node *tb_node;
2403	unsigned int h;
2404	struct key_vector *n;
2405
2406	++*pos;
2407	/* next node in same table */
2408	n = fib_trie_get_next(iter);
2409	if (n)
2410		return n;
2411
2412	/* walk rest of this hash chain */
2413	h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2414	while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2415		tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2416		n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2417		if (n)
2418			goto found;
2419	}
2420
2421	/* new hash chain */
2422	while (++h < FIB_TABLE_HASHSZ) {
2423		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2424		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2425			n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2426			if (n)
2427				goto found;
2428		}
2429	}
2430	return NULL;
2431
2432found:
2433	iter->tb = tb;
2434	return n;
2435}
2436
2437static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2438	__releases(RCU)
2439{
2440	rcu_read_unlock();
2441}
2442
2443static void seq_indent(struct seq_file *seq, int n)
2444{
2445	while (n-- > 0)
2446		seq_puts(seq, "   ");
2447}
2448
2449static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2450{
2451	switch (s) {
2452	case RT_SCOPE_UNIVERSE: return "universe";
2453	case RT_SCOPE_SITE:	return "site";
2454	case RT_SCOPE_LINK:	return "link";
2455	case RT_SCOPE_HOST:	return "host";
2456	case RT_SCOPE_NOWHERE:	return "nowhere";
2457	default:
2458		snprintf(buf, len, "scope=%d", s);
2459		return buf;
2460	}
2461}
2462
2463static const char *const rtn_type_names[__RTN_MAX] = {
2464	[RTN_UNSPEC] = "UNSPEC",
2465	[RTN_UNICAST] = "UNICAST",
2466	[RTN_LOCAL] = "LOCAL",
2467	[RTN_BROADCAST] = "BROADCAST",
2468	[RTN_ANYCAST] = "ANYCAST",
2469	[RTN_MULTICAST] = "MULTICAST",
2470	[RTN_BLACKHOLE] = "BLACKHOLE",
2471	[RTN_UNREACHABLE] = "UNREACHABLE",
2472	[RTN_PROHIBIT] = "PROHIBIT",
2473	[RTN_THROW] = "THROW",
2474	[RTN_NAT] = "NAT",
2475	[RTN_XRESOLVE] = "XRESOLVE",
2476};
2477
2478static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2479{
2480	if (t < __RTN_MAX && rtn_type_names[t])
2481		return rtn_type_names[t];
2482	snprintf(buf, len, "type %u", t);
2483	return buf;
2484}
2485
2486/* Pretty print the trie */
2487static int fib_trie_seq_show(struct seq_file *seq, void *v)
2488{
2489	const struct fib_trie_iter *iter = seq->private;
2490	struct key_vector *n = v;
2491
2492	if (IS_TRIE(node_parent_rcu(n)))
2493		fib_table_print(seq, iter->tb);
2494
2495	if (IS_TNODE(n)) {
2496		__be32 prf = htonl(n->key);
 
2497
2498		seq_indent(seq, iter->depth-1);
2499		seq_printf(seq, "  +-- %pI4/%zu %u %u %u\n",
2500			   &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2501			   tn_info(n)->full_children,
2502			   tn_info(n)->empty_children);
2503	} else {
2504		__be32 val = htonl(n->key);
2505		struct fib_alias *fa;
 
 
2506
2507		seq_indent(seq, iter->depth);
2508		seq_printf(seq, "  |-- %pI4\n", &val);
2509
2510		hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2511			char buf1[32], buf2[32];
 
 
 
2512
2513			seq_indent(seq, iter->depth + 1);
2514			seq_printf(seq, "  /%zu %s %s",
2515				   KEYLENGTH - fa->fa_slen,
2516				   rtn_scope(buf1, sizeof(buf1),
2517					     fa->fa_info->fib_scope),
2518				   rtn_type(buf2, sizeof(buf2),
2519					    fa->fa_type));
2520			if (fa->fa_tos)
2521				seq_printf(seq, " tos=%d", fa->fa_tos);
2522			seq_putc(seq, '\n');
2523		}
2524	}
2525
2526	return 0;
2527}
2528
2529static const struct seq_operations fib_trie_seq_ops = {
2530	.start  = fib_trie_seq_start,
2531	.next   = fib_trie_seq_next,
2532	.stop   = fib_trie_seq_stop,
2533	.show   = fib_trie_seq_show,
2534};
2535
2536static int fib_trie_seq_open(struct inode *inode, struct file *file)
2537{
2538	return seq_open_net(inode, file, &fib_trie_seq_ops,
2539			    sizeof(struct fib_trie_iter));
2540}
2541
2542static const struct file_operations fib_trie_fops = {
 
2543	.open   = fib_trie_seq_open,
2544	.read   = seq_read,
2545	.llseek = seq_lseek,
2546	.release = seq_release_net,
2547};
2548
2549struct fib_route_iter {
2550	struct seq_net_private p;
2551	struct fib_table *main_tb;
2552	struct key_vector *tnode;
2553	loff_t	pos;
2554	t_key	key;
2555};
2556
2557static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2558					    loff_t pos)
2559{
2560	struct key_vector *l, **tp = &iter->tnode;
2561	t_key key;
2562
2563	/* use cached location of previously found key */
2564	if (iter->pos > 0 && pos >= iter->pos) {
2565		key = iter->key;
2566	} else {
2567		iter->pos = 1;
2568		key = 0;
2569	}
2570
2571	pos -= iter->pos;
2572
2573	while ((l = leaf_walk_rcu(tp, key)) && (pos-- > 0)) {
2574		key = l->key + 1;
2575		iter->pos++;
2576		l = NULL;
2577
2578		/* handle unlikely case of a key wrap */
2579		if (!key)
2580			break;
2581	}
2582
2583	if (l)
2584		iter->key = l->key;	/* remember it */
2585	else
2586		iter->pos = 0;		/* forget it */
2587
2588	return l;
2589}
2590
2591static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2592	__acquires(RCU)
2593{
2594	struct fib_route_iter *iter = seq->private;
2595	struct fib_table *tb;
2596	struct trie *t;
2597
2598	rcu_read_lock();
2599
2600	tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2601	if (!tb)
2602		return NULL;
2603
2604	iter->main_tb = tb;
2605	t = (struct trie *)tb->tb_data;
2606	iter->tnode = t->kv;
2607
2608	if (*pos != 0)
2609		return fib_route_get_idx(iter, *pos);
2610
2611	iter->pos = 0;
2612	iter->key = KEY_MAX;
2613
2614	return SEQ_START_TOKEN;
2615}
2616
2617static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2618{
2619	struct fib_route_iter *iter = seq->private;
2620	struct key_vector *l = NULL;
2621	t_key key = iter->key + 1;
2622
2623	++*pos;
2624
2625	/* only allow key of 0 for start of sequence */
2626	if ((v == SEQ_START_TOKEN) || key)
2627		l = leaf_walk_rcu(&iter->tnode, key);
2628
2629	if (l) {
2630		iter->key = l->key;
2631		iter->pos++;
2632	} else {
2633		iter->pos = 0;
 
 
 
 
2634	}
2635
 
 
 
 
2636	return l;
2637}
2638
2639static void fib_route_seq_stop(struct seq_file *seq, void *v)
2640	__releases(RCU)
2641{
2642	rcu_read_unlock();
2643}
2644
2645static unsigned int fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2646{
2647	unsigned int flags = 0;
2648
2649	if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2650		flags = RTF_REJECT;
2651	if (fi && fi->fib_nh->nh_gw)
2652		flags |= RTF_GATEWAY;
2653	if (mask == htonl(0xFFFFFFFF))
2654		flags |= RTF_HOST;
2655	flags |= RTF_UP;
2656	return flags;
2657}
2658
2659/*
2660 *	This outputs /proc/net/route.
2661 *	The format of the file is not supposed to be changed
2662 *	and needs to be same as fib_hash output to avoid breaking
2663 *	legacy utilities
2664 */
2665static int fib_route_seq_show(struct seq_file *seq, void *v)
2666{
2667	struct fib_route_iter *iter = seq->private;
2668	struct fib_table *tb = iter->main_tb;
2669	struct fib_alias *fa;
2670	struct key_vector *l = v;
2671	__be32 prefix;
2672
2673	if (v == SEQ_START_TOKEN) {
2674		seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2675			   "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2676			   "\tWindow\tIRTT");
2677		return 0;
2678	}
2679
2680	prefix = htonl(l->key);
 
 
2681
2682	hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2683		const struct fib_info *fi = fa->fa_info;
2684		__be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2685		unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2686
2687		if ((fa->fa_type == RTN_BROADCAST) ||
2688		    (fa->fa_type == RTN_MULTICAST))
2689			continue;
 
2690
2691		if (fa->tb_id != tb->tb_id)
2692			continue;
 
2693
2694		seq_setwidth(seq, 127);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2695
2696		if (fi)
2697			seq_printf(seq,
2698				   "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2699				   "%d\t%08X\t%d\t%u\t%u",
2700				   fi->fib_dev ? fi->fib_dev->name : "*",
2701				   prefix,
2702				   fi->fib_nh->nh_gw, flags, 0, 0,
2703				   fi->fib_priority,
2704				   mask,
2705				   (fi->fib_advmss ?
2706				    fi->fib_advmss + 40 : 0),
2707				   fi->fib_window,
2708				   fi->fib_rtt >> 3);
2709		else
2710			seq_printf(seq,
2711				   "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2712				   "%d\t%08X\t%d\t%u\t%u",
2713				   prefix, 0, flags, 0, 0, 0,
2714				   mask, 0, 0, 0);
2715
2716		seq_pad(seq, '\n');
2717	}
2718
2719	return 0;
2720}
2721
2722static const struct seq_operations fib_route_seq_ops = {
2723	.start  = fib_route_seq_start,
2724	.next   = fib_route_seq_next,
2725	.stop   = fib_route_seq_stop,
2726	.show   = fib_route_seq_show,
2727};
2728
2729static int fib_route_seq_open(struct inode *inode, struct file *file)
2730{
2731	return seq_open_net(inode, file, &fib_route_seq_ops,
2732			    sizeof(struct fib_route_iter));
2733}
2734
2735static const struct file_operations fib_route_fops = {
 
2736	.open   = fib_route_seq_open,
2737	.read   = seq_read,
2738	.llseek = seq_lseek,
2739	.release = seq_release_net,
2740};
2741
2742int __net_init fib_proc_init(struct net *net)
2743{
2744	if (!proc_create("fib_trie", 0444, net->proc_net, &fib_trie_fops))
2745		goto out1;
2746
2747	if (!proc_create("fib_triestat", 0444, net->proc_net,
2748			 &fib_triestat_fops))
2749		goto out2;
2750
2751	if (!proc_create("route", 0444, net->proc_net, &fib_route_fops))
2752		goto out3;
2753
2754	return 0;
2755
2756out3:
2757	remove_proc_entry("fib_triestat", net->proc_net);
2758out2:
2759	remove_proc_entry("fib_trie", net->proc_net);
2760out1:
2761	return -ENOMEM;
2762}
2763
2764void __net_exit fib_proc_exit(struct net *net)
2765{
2766	remove_proc_entry("fib_trie", net->proc_net);
2767	remove_proc_entry("fib_triestat", net->proc_net);
2768	remove_proc_entry("route", net->proc_net);
2769}
2770
2771#endif /* CONFIG_PROC_FS */
v3.5.6
   1/*
   2 *   This program is free software; you can redistribute it and/or
   3 *   modify it under the terms of the GNU General Public License
   4 *   as published by the Free Software Foundation; either version
   5 *   2 of the License, or (at your option) any later version.
   6 *
   7 *   Robert Olsson <robert.olsson@its.uu.se> Uppsala Universitet
   8 *     & Swedish University of Agricultural Sciences.
   9 *
  10 *   Jens Laas <jens.laas@data.slu.se> Swedish University of
  11 *     Agricultural Sciences.
  12 *
  13 *   Hans Liss <hans.liss@its.uu.se>  Uppsala Universitet
  14 *
  15 * This work is based on the LPC-trie which is originally described in:
  16 *
  17 * An experimental study of compression methods for dynamic tries
  18 * Stefan Nilsson and Matti Tikkanen. Algorithmica, 33(1):19-33, 2002.
  19 * http://www.csc.kth.se/~snilsson/software/dyntrie2/
  20 *
  21 *
  22 * IP-address lookup using LC-tries. Stefan Nilsson and Gunnar Karlsson
  23 * IEEE Journal on Selected Areas in Communications, 17(6):1083-1092, June 1999
  24 *
  25 *
  26 * Code from fib_hash has been reused which includes the following header:
  27 *
  28 *
  29 * INET		An implementation of the TCP/IP protocol suite for the LINUX
  30 *		operating system.  INET is implemented using the  BSD Socket
  31 *		interface as the means of communication with the user level.
  32 *
  33 *		IPv4 FIB: lookup engine and maintenance routines.
  34 *
  35 *
  36 * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  37 *
  38 *		This program is free software; you can redistribute it and/or
  39 *		modify it under the terms of the GNU General Public License
  40 *		as published by the Free Software Foundation; either version
  41 *		2 of the License, or (at your option) any later version.
  42 *
  43 * Substantial contributions to this work comes from:
  44 *
  45 *		David S. Miller, <davem@davemloft.net>
  46 *		Stephen Hemminger <shemminger@osdl.org>
  47 *		Paul E. McKenney <paulmck@us.ibm.com>
  48 *		Patrick McHardy <kaber@trash.net>
  49 */
  50
  51#define VERSION "0.409"
  52
  53#include <asm/uaccess.h>
 
  54#include <linux/bitops.h>
  55#include <linux/types.h>
  56#include <linux/kernel.h>
  57#include <linux/mm.h>
  58#include <linux/string.h>
  59#include <linux/socket.h>
  60#include <linux/sockios.h>
  61#include <linux/errno.h>
  62#include <linux/in.h>
  63#include <linux/inet.h>
  64#include <linux/inetdevice.h>
  65#include <linux/netdevice.h>
  66#include <linux/if_arp.h>
  67#include <linux/proc_fs.h>
  68#include <linux/rcupdate.h>
  69#include <linux/skbuff.h>
  70#include <linux/netlink.h>
  71#include <linux/init.h>
  72#include <linux/list.h>
  73#include <linux/slab.h>
  74#include <linux/prefetch.h>
  75#include <linux/export.h>
 
 
  76#include <net/net_namespace.h>
  77#include <net/ip.h>
  78#include <net/protocol.h>
  79#include <net/route.h>
  80#include <net/tcp.h>
  81#include <net/sock.h>
  82#include <net/ip_fib.h>
 
 
  83#include "fib_lookup.h"
  84
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  85#define MAX_STAT_DEPTH 32
  86
  87#define KEYLENGTH (8*sizeof(t_key))
 
  88
  89typedef unsigned int t_key;
  90
  91#define T_TNODE 0
  92#define T_LEAF  1
  93#define NODE_TYPE_MASK	0x1UL
  94#define NODE_TYPE(node) ((node)->parent & NODE_TYPE_MASK)
  95
  96#define IS_TNODE(n) (!(n->parent & T_LEAF))
  97#define IS_LEAF(n) (n->parent & T_LEAF)
  98
  99struct rt_trie_node {
 100	unsigned long parent;
 101	t_key key;
 
 
 
 
 
 
 
 
 
 102};
 103
 104struct leaf {
 105	unsigned long parent;
 106	t_key key;
 107	struct hlist_head list;
 108	struct rcu_head rcu;
 109};
 110
 111struct leaf_info {
 112	struct hlist_node hlist;
 113	int plen;
 114	u32 mask_plen; /* ntohl(inet_make_mask(plen)) */
 115	struct list_head falh;
 116	struct rcu_head rcu;
 
 
 
 
 
 117};
 118
 119struct tnode {
 120	unsigned long parent;
 121	t_key key;
 122	unsigned char pos;		/* 2log(KEYLENGTH) bits needed */
 123	unsigned char bits;		/* 2log(KEYLENGTH) bits needed */
 124	unsigned int full_children;	/* KEYLENGTH bits needed */
 125	unsigned int empty_children;	/* KEYLENGTH bits needed */
 126	union {
 127		struct rcu_head rcu;
 128		struct work_struct work;
 129		struct tnode *tnode_free;
 130	};
 131	struct rt_trie_node __rcu *child[0];
 132};
 133
 134#ifdef CONFIG_IP_FIB_TRIE_STATS
 135struct trie_use_stats {
 136	unsigned int gets;
 137	unsigned int backtrack;
 138	unsigned int semantic_match_passed;
 139	unsigned int semantic_match_miss;
 140	unsigned int null_node_hit;
 141	unsigned int resize_node_skipped;
 142};
 143#endif
 144
 145struct trie_stat {
 146	unsigned int totdepth;
 147	unsigned int maxdepth;
 148	unsigned int tnodes;
 149	unsigned int leaves;
 150	unsigned int nullpointers;
 151	unsigned int prefixes;
 152	unsigned int nodesizes[MAX_STAT_DEPTH];
 153};
 154
 155struct trie {
 156	struct rt_trie_node __rcu *trie;
 157#ifdef CONFIG_IP_FIB_TRIE_STATS
 158	struct trie_use_stats stats;
 159#endif
 160};
 161
 162static void put_child(struct trie *t, struct tnode *tn, int i, struct rt_trie_node *n);
 163static void tnode_put_child_reorg(struct tnode *tn, int i, struct rt_trie_node *n,
 164				  int wasfull);
 165static struct rt_trie_node *resize(struct trie *t, struct tnode *tn);
 166static struct tnode *inflate(struct trie *t, struct tnode *tn);
 167static struct tnode *halve(struct trie *t, struct tnode *tn);
 168/* tnodes to free after resize(); protected by RTNL */
 169static struct tnode *tnode_free_head;
 170static size_t tnode_free_size;
 171
 172/*
 173 * synchronize_rcu after call_rcu for that many pages; it should be especially
 174 * useful before resizing the root node with PREEMPT_NONE configs; the value was
 175 * obtained experimentally, aiming to avoid visible slowdown.
 176 */
 177static const int sync_pages = 128;
 178
 179static struct kmem_cache *fn_alias_kmem __read_mostly;
 180static struct kmem_cache *trie_leaf_kmem __read_mostly;
 181
 182/*
 183 * caller must hold RTNL
 184 */
 185static inline struct tnode *node_parent(const struct rt_trie_node *node)
 186{
 187	unsigned long parent;
 188
 189	parent = rcu_dereference_index_check(node->parent, lockdep_rtnl_is_held());
 190
 191	return (struct tnode *)(parent & ~NODE_TYPE_MASK);
 192}
 193
 194/*
 195 * caller must hold RCU read lock or RTNL
 196 */
 197static inline struct tnode *node_parent_rcu(const struct rt_trie_node *node)
 198{
 199	unsigned long parent;
 200
 201	parent = rcu_dereference_index_check(node->parent, rcu_read_lock_held() ||
 202							   lockdep_rtnl_is_held());
 203
 204	return (struct tnode *)(parent & ~NODE_TYPE_MASK);
 205}
 
 206
 207/* Same as rcu_assign_pointer
 208 * but that macro() assumes that value is a pointer.
 209 */
 210static inline void node_set_parent(struct rt_trie_node *node, struct tnode *ptr)
 211{
 212	smp_wmb();
 213	node->parent = (unsigned long)ptr | NODE_TYPE(node);
 214}
 215
 216/*
 217 * caller must hold RTNL
 218 */
 219static inline struct rt_trie_node *tnode_get_child(const struct tnode *tn, unsigned int i)
 220{
 221	BUG_ON(i >= 1U << tn->bits);
 222
 223	return rtnl_dereference(tn->child[i]);
 224}
 225
 226/*
 227 * caller must hold RCU read lock or RTNL
 228 */
 229static inline struct rt_trie_node *tnode_get_child_rcu(const struct tnode *tn, unsigned int i)
 230{
 231	BUG_ON(i >= 1U << tn->bits);
 232
 233	return rcu_dereference_rtnl(tn->child[i]);
 234}
 235
 236static inline int tnode_child_length(const struct tnode *tn)
 237{
 238	return 1 << tn->bits;
 239}
 240
 241static inline t_key mask_pfx(t_key k, unsigned int l)
 242{
 243	return (l == 0) ? 0 : k >> (KEYLENGTH-l) << (KEYLENGTH-l);
 244}
 245
 246static inline t_key tkey_extract_bits(t_key a, unsigned int offset, unsigned int bits)
 247{
 248	if (offset < KEYLENGTH)
 249		return ((t_key)(a << offset)) >> (KEYLENGTH - bits);
 250	else
 251		return 0;
 252}
 253
 254static inline int tkey_equals(t_key a, t_key b)
 255{
 256	return a == b;
 257}
 258
 259static inline int tkey_sub_equals(t_key a, int offset, int bits, t_key b)
 260{
 261	if (bits == 0 || offset >= KEYLENGTH)
 262		return 1;
 263	bits = bits > KEYLENGTH ? KEYLENGTH : bits;
 264	return ((a ^ b) << offset) >> (KEYLENGTH - bits) == 0;
 265}
 266
 267static inline int tkey_mismatch(t_key a, int offset, t_key b)
 268{
 269	t_key diff = a ^ b;
 270	int i = offset;
 271
 272	if (!diff)
 273		return 0;
 274	while ((diff << i) >> (KEYLENGTH-1) == 0)
 275		i++;
 276	return i;
 277}
 278
 279/*
 280  To understand this stuff, an understanding of keys and all their bits is
 281  necessary. Every node in the trie has a key associated with it, but not
 282  all of the bits in that key are significant.
 283
 284  Consider a node 'n' and its parent 'tp'.
 285
 286  If n is a leaf, every bit in its key is significant. Its presence is
 287  necessitated by path compression, since during a tree traversal (when
 288  searching for a leaf - unless we are doing an insertion) we will completely
 289  ignore all skipped bits we encounter. Thus we need to verify, at the end of
 290  a potentially successful search, that we have indeed been walking the
 291  correct key path.
 292
 293  Note that we can never "miss" the correct key in the tree if present by
 294  following the wrong path. Path compression ensures that segments of the key
 295  that are the same for all keys with a given prefix are skipped, but the
 296  skipped part *is* identical for each node in the subtrie below the skipped
 297  bit! trie_insert() in this implementation takes care of that - note the
 298  call to tkey_sub_equals() in trie_insert().
 299
 300  if n is an internal node - a 'tnode' here, the various parts of its key
 301  have many different meanings.
 302
 303  Example:
 304  _________________________________________________________________
 305  | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
 306  -----------------------------------------------------------------
 307    0   1   2   3   4   5   6   7   8   9  10  11  12  13  14  15
 308
 309  _________________________________________________________________
 310  | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
 311  -----------------------------------------------------------------
 312   16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31
 313
 314  tp->pos = 7
 315  tp->bits = 3
 316  n->pos = 15
 317  n->bits = 4
 318
 319  First, let's just ignore the bits that come before the parent tp, that is
 320  the bits from 0 to (tp->pos-1). They are *known* but at this point we do
 321  not use them for anything.
 322
 323  The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
 324  index into the parent's child array. That is, they will be used to find
 325  'n' among tp's children.
 326
 327  The bits from (tp->pos + tp->bits) to (n->pos - 1) - "S" - are skipped bits
 328  for the node n.
 329
 330  All the bits we have seen so far are significant to the node n. The rest
 331  of the bits are really not needed or indeed known in n->key.
 332
 333  The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
 334  n's child array, and will of course be different for each child.
 335
 336
 337  The rest of the bits, from (n->pos + n->bits) onward, are completely unknown
 338  at this point.
 339
 340*/
 341
 342static inline void check_tnode(const struct tnode *tn)
 343{
 344	WARN_ON(tn && tn->pos+tn->bits > 32);
 345}
 346
 347static const int halve_threshold = 25;
 348static const int inflate_threshold = 50;
 349static const int halve_threshold_root = 15;
 350static const int inflate_threshold_root = 30;
 351
 352static void __alias_free_mem(struct rcu_head *head)
 353{
 354	struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
 355	kmem_cache_free(fn_alias_kmem, fa);
 356}
 357
 358static inline void alias_free_mem_rcu(struct fib_alias *fa)
 359{
 360	call_rcu(&fa->rcu, __alias_free_mem);
 361}
 362
 363static void __leaf_free_rcu(struct rcu_head *head)
 
 
 
 
 
 364{
 365	struct leaf *l = container_of(head, struct leaf, rcu);
 366	kmem_cache_free(trie_leaf_kmem, l);
 
 
 
 
 367}
 368
 369static inline void free_leaf(struct leaf *l)
 
 
 370{
 371	call_rcu_bh(&l->rcu, __leaf_free_rcu);
 372}
 
 
 
 373
 374static inline void free_leaf_info(struct leaf_info *leaf)
 375{
 376	kfree_rcu(leaf, rcu);
 377}
 378
 379static struct tnode *tnode_alloc(size_t size)
 380{
 381	if (size <= PAGE_SIZE)
 382		return kzalloc(size, GFP_KERNEL);
 383	else
 384		return vzalloc(size);
 385}
 386
 387static void __tnode_vfree(struct work_struct *arg)
 388{
 389	struct tnode *tn = container_of(arg, struct tnode, work);
 390	vfree(tn);
 391}
 392
 393static void __tnode_free_rcu(struct rcu_head *head)
 394{
 395	struct tnode *tn = container_of(head, struct tnode, rcu);
 396	size_t size = sizeof(struct tnode) +
 397		      (sizeof(struct rt_trie_node *) << tn->bits);
 398
 399	if (size <= PAGE_SIZE)
 400		kfree(tn);
 401	else {
 402		INIT_WORK(&tn->work, __tnode_vfree);
 403		schedule_work(&tn->work);
 404	}
 405}
 406
 407static inline void tnode_free(struct tnode *tn)
 408{
 409	if (IS_LEAF(tn))
 410		free_leaf((struct leaf *) tn);
 411	else
 412		call_rcu(&tn->rcu, __tnode_free_rcu);
 413}
 
 
 
 
 
 
 
 
 
 
 
 
 414
 415static void tnode_free_safe(struct tnode *tn)
 416{
 417	BUG_ON(IS_LEAF(tn));
 418	tn->tnode_free = tnode_free_head;
 419	tnode_free_head = tn;
 420	tnode_free_size += sizeof(struct tnode) +
 421			   (sizeof(struct rt_trie_node *) << tn->bits);
 422}
 423
 424static void tnode_free_flush(void)
 425{
 426	struct tnode *tn;
 
 
 427
 428	while ((tn = tnode_free_head)) {
 429		tnode_free_head = tn->tnode_free;
 430		tn->tnode_free = NULL;
 431		tnode_free(tn);
 432	}
 433
 434	if (tnode_free_size >= PAGE_SIZE * sync_pages) {
 435		tnode_free_size = 0;
 436		synchronize_rcu();
 437	}
 438}
 439
 440static struct leaf *leaf_new(void)
 441{
 442	struct leaf *l = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
 443	if (l) {
 444		l->parent = T_LEAF;
 445		INIT_HLIST_HEAD(&l->list);
 446	}
 447	return l;
 448}
 449
 450static struct leaf_info *leaf_info_new(int plen)
 451{
 452	struct leaf_info *li = kmalloc(sizeof(struct leaf_info),  GFP_KERNEL);
 453	if (li) {
 454		li->plen = plen;
 455		li->mask_plen = ntohl(inet_make_mask(plen));
 456		INIT_LIST_HEAD(&li->falh);
 457	}
 458	return li;
 459}
 460
 461static struct tnode *tnode_new(t_key key, int pos, int bits)
 462{
 463	size_t sz = sizeof(struct tnode) + (sizeof(struct rt_trie_node *) << bits);
 464	struct tnode *tn = tnode_alloc(sz);
 465
 466	if (tn) {
 467		tn->parent = T_TNODE;
 468		tn->pos = pos;
 469		tn->bits = bits;
 470		tn->key = key;
 471		tn->full_children = 0;
 472		tn->empty_children = 1<<bits;
 473	}
 474
 475	pr_debug("AT %p s=%zu %zu\n", tn, sizeof(struct tnode),
 476		 sizeof(struct rt_trie_node) << bits);
 477	return tn;
 478}
 479
 480/*
 481 * Check whether a tnode 'n' is "full", i.e. it is an internal node
 482 * and no bits are skipped. See discussion in dyntree paper p. 6
 483 */
 484
 485static inline int tnode_full(const struct tnode *tn, const struct rt_trie_node *n)
 486{
 487	if (n == NULL || IS_LEAF(n))
 488		return 0;
 489
 490	return ((struct tnode *) n)->pos == tn->pos + tn->bits;
 491}
 492
 493static inline void put_child(struct trie *t, struct tnode *tn, int i,
 494			     struct rt_trie_node *n)
 495{
 496	tnode_put_child_reorg(tn, i, n, -1);
 497}
 498
 499 /*
 500  * Add a child at position i overwriting the old value.
 501  * Update the value of full_children and empty_children.
 502  */
 503
 504static void tnode_put_child_reorg(struct tnode *tn, int i, struct rt_trie_node *n,
 505				  int wasfull)
 506{
 507	struct rt_trie_node *chi = rtnl_dereference(tn->child[i]);
 508	int isfull;
 509
 510	BUG_ON(i >= 1<<tn->bits);
 511
 512	/* update emptyChildren */
 513	if (n == NULL && chi != NULL)
 514		tn->empty_children++;
 515	else if (n != NULL && chi == NULL)
 516		tn->empty_children--;
 517
 518	/* update fullChildren */
 519	if (wasfull == -1)
 520		wasfull = tnode_full(tn, chi);
 521
 522	isfull = tnode_full(tn, n);
 523	if (wasfull && !isfull)
 524		tn->full_children--;
 525	else if (!wasfull && isfull)
 526		tn->full_children++;
 527
 528	if (n)
 529		node_set_parent(n, tn);
 530
 531	rcu_assign_pointer(tn->child[i], n);
 532}
 533
 534#define MAX_WORK 10
 535static struct rt_trie_node *resize(struct trie *t, struct tnode *tn)
 536{
 537	int i;
 538	struct tnode *old_tn;
 539	int inflate_threshold_use;
 540	int halve_threshold_use;
 541	int max_work;
 542
 543	if (!tn)
 544		return NULL;
 
 545
 546	pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
 547		 tn, inflate_threshold, halve_threshold);
 548
 549	/* No children */
 550	if (tn->empty_children == tnode_child_length(tn)) {
 551		tnode_free_safe(tn);
 552		return NULL;
 
 
 
 
 553	}
 554	/* One child */
 555	if (tn->empty_children == tnode_child_length(tn) - 1)
 556		goto one_child;
 557	/*
 558	 * Double as long as the resulting node has a number of
 559	 * nonempty nodes that are above the threshold.
 560	 */
 561
 562	/*
 563	 * From "Implementing a dynamic compressed trie" by Stefan Nilsson of
 564	 * the Helsinki University of Technology and Matti Tikkanen of Nokia
 565	 * Telecommunications, page 6:
 566	 * "A node is doubled if the ratio of non-empty children to all
 567	 * children in the *doubled* node is at least 'high'."
 568	 *
 569	 * 'high' in this instance is the variable 'inflate_threshold'. It
 570	 * is expressed as a percentage, so we multiply it with
 571	 * tnode_child_length() and instead of multiplying by 2 (since the
 572	 * child array will be doubled by inflate()) and multiplying
 573	 * the left-hand side by 100 (to handle the percentage thing) we
 574	 * multiply the left-hand side by 50.
 575	 *
 576	 * The left-hand side may look a bit weird: tnode_child_length(tn)
 577	 * - tn->empty_children is of course the number of non-null children
 578	 * in the current node. tn->full_children is the number of "full"
 579	 * children, that is non-null tnodes with a skip value of 0.
 580	 * All of those will be doubled in the resulting inflated tnode, so
 581	 * we just count them one extra time here.
 582	 *
 583	 * A clearer way to write this would be:
 584	 *
 585	 * to_be_doubled = tn->full_children;
 586	 * not_to_be_doubled = tnode_child_length(tn) - tn->empty_children -
 587	 *     tn->full_children;
 588	 *
 589	 * new_child_length = tnode_child_length(tn) * 2;
 590	 *
 591	 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
 592	 *      new_child_length;
 593	 * if (new_fill_factor >= inflate_threshold)
 594	 *
 595	 * ...and so on, tho it would mess up the while () loop.
 596	 *
 597	 * anyway,
 598	 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
 599	 *      inflate_threshold
 600	 *
 601	 * avoid a division:
 602	 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
 603	 *      inflate_threshold * new_child_length
 604	 *
 605	 * expand not_to_be_doubled and to_be_doubled, and shorten:
 606	 * 100 * (tnode_child_length(tn) - tn->empty_children +
 607	 *    tn->full_children) >= inflate_threshold * new_child_length
 608	 *
 609	 * expand new_child_length:
 610	 * 100 * (tnode_child_length(tn) - tn->empty_children +
 611	 *    tn->full_children) >=
 612	 *      inflate_threshold * tnode_child_length(tn) * 2
 613	 *
 614	 * shorten again:
 615	 * 50 * (tn->full_children + tnode_child_length(tn) -
 616	 *    tn->empty_children) >= inflate_threshold *
 617	 *    tnode_child_length(tn)
 618	 *
 619	 */
 620
 621	check_tnode(tn);
 
 
 
 622
 623	/* Keep root node larger  */
 
 
 
 
 
 624
 625	if (!node_parent((struct rt_trie_node *)tn)) {
 626		inflate_threshold_use = inflate_threshold_root;
 627		halve_threshold_use = halve_threshold_root;
 628	} else {
 629		inflate_threshold_use = inflate_threshold;
 630		halve_threshold_use = halve_threshold;
 631	}
 632
 633	max_work = MAX_WORK;
 634	while ((tn->full_children > 0 &&  max_work-- &&
 635		50 * (tn->full_children + tnode_child_length(tn)
 636		      - tn->empty_children)
 637		>= inflate_threshold_use * tnode_child_length(tn))) {
 638
 639		old_tn = tn;
 640		tn = inflate(t, tn);
 641
 642		if (IS_ERR(tn)) {
 643			tn = old_tn;
 644#ifdef CONFIG_IP_FIB_TRIE_STATS
 645			t->stats.resize_node_skipped++;
 646#endif
 647			break;
 648		}
 649	}
 
 650
 651	check_tnode(tn);
 652
 653	/* Return if at least one inflate is run */
 654	if (max_work != MAX_WORK)
 655		return (struct rt_trie_node *) tn;
 656
 657	/*
 658	 * Halve as long as the number of empty children in this
 659	 * node is above threshold.
 660	 */
 661
 662	max_work = MAX_WORK;
 663	while (tn->bits > 1 &&  max_work-- &&
 664	       100 * (tnode_child_length(tn) - tn->empty_children) <
 665	       halve_threshold_use * tnode_child_length(tn)) {
 666
 667		old_tn = tn;
 668		tn = halve(t, tn);
 669		if (IS_ERR(tn)) {
 670			tn = old_tn;
 671#ifdef CONFIG_IP_FIB_TRIE_STATS
 672			t->stats.resize_node_skipped++;
 673#endif
 674			break;
 675		}
 676	}
 677
 
 
 
 678
 679	/* Only one child remains */
 680	if (tn->empty_children == tnode_child_length(tn) - 1) {
 681one_child:
 682		for (i = 0; i < tnode_child_length(tn); i++) {
 683			struct rt_trie_node *n;
 684
 685			n = rtnl_dereference(tn->child[i]);
 686			if (!n)
 687				continue;
 688
 689			/* compress one level */
 
 
 690
 691			node_set_parent(n, NULL);
 692			tnode_free_safe(tn);
 693			return n;
 694		}
 695	}
 696	return (struct rt_trie_node *) tn;
 697}
 698
 699
 700static void tnode_clean_free(struct tnode *tn)
 701{
 702	int i;
 703	struct tnode *tofree;
 704
 705	for (i = 0; i < tnode_child_length(tn); i++) {
 706		tofree = (struct tnode *)rtnl_dereference(tn->child[i]);
 707		if (tofree)
 708			tnode_free(tofree);
 709	}
 710	tnode_free(tn);
 711}
 712
 713static struct tnode *inflate(struct trie *t, struct tnode *tn)
 
 714{
 715	struct tnode *oldtnode = tn;
 716	int olen = tnode_child_length(tn);
 717	int i;
 718
 719	pr_debug("In inflate\n");
 720
 721	tn = tnode_new(oldtnode->key, oldtnode->pos, oldtnode->bits + 1);
 
 
 722
 723	if (!tn)
 724		return ERR_PTR(-ENOMEM);
 725
 726	/*
 727	 * Preallocate and store tnodes before the actual work so we
 728	 * don't get into an inconsistent state if memory allocation
 729	 * fails. In case of failure we return the oldnode and  inflate
 730	 * of tnode is ignored.
 731	 */
 732
 733	for (i = 0; i < olen; i++) {
 734		struct tnode *inode;
 735
 736		inode = (struct tnode *) tnode_get_child(oldtnode, i);
 737		if (inode &&
 738		    IS_TNODE(inode) &&
 739		    inode->pos == oldtnode->pos + oldtnode->bits &&
 740		    inode->bits > 1) {
 741			struct tnode *left, *right;
 742			t_key m = ~0U << (KEYLENGTH - 1) >> inode->pos;
 743
 744			left = tnode_new(inode->key&(~m), inode->pos + 1,
 745					 inode->bits - 1);
 746			if (!left)
 747				goto nomem;
 748
 749			right = tnode_new(inode->key|m, inode->pos + 1,
 750					  inode->bits - 1);
 751
 752			if (!right) {
 753				tnode_free(left);
 754				goto nomem;
 755			}
 756
 757			put_child(t, tn, 2*i, (struct rt_trie_node *) left);
 758			put_child(t, tn, 2*i+1, (struct rt_trie_node *) right);
 759		}
 760	}
 761
 762	for (i = 0; i < olen; i++) {
 763		struct tnode *inode;
 764		struct rt_trie_node *node = tnode_get_child(oldtnode, i);
 765		struct tnode *left, *right;
 766		int size, j;
 767
 768		/* An empty child */
 769		if (node == NULL)
 770			continue;
 771
 772		/* A leaf or an internal node with skipped bits */
 773
 774		if (IS_LEAF(node) || ((struct tnode *) node)->pos >
 775		   tn->pos + tn->bits - 1) {
 776			if (tkey_extract_bits(node->key,
 777					      oldtnode->pos + oldtnode->bits,
 778					      1) == 0)
 779				put_child(t, tn, 2*i, node);
 780			else
 781				put_child(t, tn, 2*i+1, node);
 782			continue;
 783		}
 784
 
 
 
 785		/* An internal node with two children */
 786		inode = (struct tnode *) node;
 787
 788		if (inode->bits == 1) {
 789			put_child(t, tn, 2*i, rtnl_dereference(inode->child[0]));
 790			put_child(t, tn, 2*i+1, rtnl_dereference(inode->child[1]));
 791
 792			tnode_free_safe(inode);
 793			continue;
 794		}
 795
 796		/* An internal node with more than two children */
 797
 798		/* We will replace this node 'inode' with two new
 799		 * ones, 'left' and 'right', each with half of the
 800		 * original children. The two new nodes will have
 801		 * a position one bit further down the key and this
 802		 * means that the "significant" part of their keys
 803		 * (see the discussion near the top of this file)
 804		 * will differ by one bit, which will be "0" in
 805		 * left's key and "1" in right's key. Since we are
 806		 * moving the key position by one step, the bit that
 807		 * we are moving away from - the bit at position
 808		 * (inode->pos) - is the one that will differ between
 809		 * left and right. So... we synthesize that bit in the
 810		 * two  new keys.
 811		 * The mask 'm' below will be a single "one" bit at
 812		 * the position (inode->pos)
 813		 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 814
 815		/* Use the old key, but set the new significant
 816		 *   bit to zero.
 817		 */
 
 
 
 
 
 818
 819		left = (struct tnode *) tnode_get_child(tn, 2*i);
 820		put_child(t, tn, 2*i, NULL);
 
 
 
 821
 822		BUG_ON(!left);
 823
 824		right = (struct tnode *) tnode_get_child(tn, 2*i+1);
 825		put_child(t, tn, 2*i+1, NULL);
 
 826
 827		BUG_ON(!right);
 
 828
 829		size = tnode_child_length(left);
 830		for (j = 0; j < size; j++) {
 831			put_child(t, left, j, rtnl_dereference(inode->child[j]));
 832			put_child(t, right, j, rtnl_dereference(inode->child[j + size]));
 
 
 
 
 
 
 
 
 
 
 833		}
 834		put_child(t, tn, 2*i, resize(t, left));
 835		put_child(t, tn, 2*i+1, resize(t, right));
 836
 837		tnode_free_safe(inode);
 
 
 
 
 
 
 
 
 
 
 
 
 838	}
 839	tnode_free_safe(oldtnode);
 840	return tn;
 
 841nomem:
 842	tnode_clean_free(tn);
 843	return ERR_PTR(-ENOMEM);
 
 
 844}
 845
 846static struct tnode *halve(struct trie *t, struct tnode *tn)
 
 847{
 848	struct tnode *oldtnode = tn;
 849	struct rt_trie_node *left, *right;
 850	int i;
 851	int olen = tnode_child_length(tn);
 852
 853	pr_debug("In halve\n");
 
 
 854
 855	tn = tnode_new(oldtnode->key, oldtnode->pos, oldtnode->bits - 1);
 
 
 
 856
 857	if (!tn)
 858		return ERR_PTR(-ENOMEM);
 859
 860	/*
 861	 * Preallocate and store tnodes before the actual work so we
 862	 * don't get into an inconsistent state if memory allocation
 863	 * fails. In case of failure we return the oldnode and halve
 864	 * of tnode is ignored.
 865	 */
 866
 867	for (i = 0; i < olen; i += 2) {
 868		left = tnode_get_child(oldtnode, i);
 869		right = tnode_get_child(oldtnode, i+1);
 
 
 870
 871		/* Two nonempty children */
 872		if (left && right) {
 873			struct tnode *newn;
 
 
 874
 875			newn = tnode_new(left->key, tn->pos + tn->bits, 1);
 
 
 
 
 
 
 876
 877			if (!newn)
 878				goto nomem;
 879
 880			put_child(t, tn, i/2, (struct rt_trie_node *)newn);
 881		}
 
 
 882
 
 
 
 883	}
 884
 885	for (i = 0; i < olen; i += 2) {
 886		struct tnode *newBinNode;
 887
 888		left = tnode_get_child(oldtnode, i);
 889		right = tnode_get_child(oldtnode, i+1);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 890
 891		/* At least one of the children is empty */
 892		if (left == NULL) {
 893			if (right == NULL)    /* Both are empty */
 894				continue;
 895			put_child(t, tn, i/2, right);
 896			continue;
 897		}
 898
 899		if (right == NULL) {
 900			put_child(t, tn, i/2, left);
 901			continue;
 902		}
 903
 904		/* Two nonempty children */
 905		newBinNode = (struct tnode *) tnode_get_child(tn, i/2);
 906		put_child(t, tn, i/2, NULL);
 907		put_child(t, newBinNode, 0, left);
 908		put_child(t, newBinNode, 1, right);
 909		put_child(t, tn, i/2, resize(t, newBinNode));
 910	}
 911	tnode_free_safe(oldtnode);
 912	return tn;
 913nomem:
 914	tnode_clean_free(tn);
 915	return ERR_PTR(-ENOMEM);
 916}
 917
 918/* readside must use rcu_read_lock currently dump routines
 919 via get_fa_head and dump */
 
 
 920
 921static struct leaf_info *find_leaf_info(struct leaf *l, int plen)
 922{
 923	struct hlist_head *head = &l->list;
 924	struct hlist_node *node;
 925	struct leaf_info *li;
 926
 927	hlist_for_each_entry_rcu(li, node, head, hlist)
 928		if (li->plen == plen)
 929			return li;
 930
 931	return NULL;
 932}
 933
 934static inline struct list_head *get_fa_head(struct leaf *l, int plen)
 935{
 936	struct leaf_info *li = find_leaf_info(l, plen);
 
 
 937
 938	if (!li)
 939		return NULL;
 
 940
 941	return &li->falh;
 
 942}
 943
 944static void insert_leaf_info(struct hlist_head *head, struct leaf_info *new)
 
 945{
 946	struct leaf_info *li = NULL, *last = NULL;
 947	struct hlist_node *node;
 
 
 
 
 
 
 
 948
 949	if (hlist_empty(head)) {
 950		hlist_add_head_rcu(&new->hlist, head);
 951	} else {
 952		hlist_for_each_entry(li, node, head, hlist) {
 953			if (new->plen > li->plen)
 954				break;
 955
 956			last = li;
 
 
 
 
 
 
 
 
 
 957		}
 958		if (last)
 959			hlist_add_after_rcu(&last->hlist, &new->hlist);
 960		else
 961			hlist_add_before_rcu(&new->hlist, &li->hlist);
 962	}
 963}
 964
 965/* rcu_read_lock needs to be hold by caller from readside */
 
 
 
 
 
 966
 967static struct leaf *
 968fib_find_node(struct trie *t, u32 key)
 969{
 970	int pos;
 971	struct tnode *tn;
 972	struct rt_trie_node *n;
 973
 974	pos = 0;
 975	n = rcu_dereference_rtnl(t->trie);
 976
 977	while (n != NULL &&  NODE_TYPE(n) == T_TNODE) {
 978		tn = (struct tnode *) n;
 979
 980		check_tnode(tn);
 981
 982		if (tkey_sub_equals(tn->key, pos, tn->pos-pos, key)) {
 983			pos = tn->pos + tn->bits;
 984			n = tnode_get_child_rcu(tn,
 985						tkey_extract_bits(key,
 986								  tn->pos,
 987								  tn->bits));
 988		} else
 989			break;
 
 
 
 
 990	}
 991	/* Case we have found a leaf. Compare prefixes */
 992
 993	if (n != NULL && IS_LEAF(n) && tkey_equals(key, n->key))
 994		return (struct leaf *)n;
 
 995
 996	return NULL;
 
 997}
 998
 999static void trie_rebalance(struct trie *t, struct tnode *tn)
1000{
1001	int wasfull;
1002	t_key cindex, key;
1003	struct tnode *tp;
 
 
 
1004
1005	key = tn->key;
 
 
 
1006
1007	while (tn != NULL && (tp = node_parent((struct rt_trie_node *)tn)) != NULL) {
1008		cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1009		wasfull = tnode_full(tp, tnode_get_child(tp, cindex));
1010		tn = (struct tnode *) resize(t, (struct tnode *)tn);
 
 
 
1011
1012		tnode_put_child_reorg((struct tnode *)tp, cindex,
1013				      (struct rt_trie_node *)tn, wasfull);
 
 
 
 
1014
1015		tp = node_parent((struct rt_trie_node *) tn);
1016		if (!tp)
1017			rcu_assign_pointer(t->trie, (struct rt_trie_node *)tn);
1018
1019		tnode_free_flush();
1020		if (!tp)
1021			break;
1022		tn = tp;
1023	}
1024
1025	/* Handle last (top) tnode */
1026	if (IS_TNODE(tn))
1027		tn = (struct tnode *)resize(t, (struct tnode *)tn);
1028
1029	rcu_assign_pointer(t->trie, (struct rt_trie_node *)tn);
1030	tnode_free_flush();
1031}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1032
1033/* only used from updater-side */
 
1034
1035static struct list_head *fib_insert_node(struct trie *t, u32 key, int plen)
1036{
1037	int pos, newpos;
1038	struct tnode *tp = NULL, *tn = NULL;
1039	struct rt_trie_node *n;
1040	struct leaf *l;
1041	int missbit;
1042	struct list_head *fa_head = NULL;
1043	struct leaf_info *li;
1044	t_key cindex;
1045
1046	pos = 0;
1047	n = rtnl_dereference(t->trie);
1048
1049	/* If we point to NULL, stop. Either the tree is empty and we should
1050	 * just put a new leaf in if, or we have reached an empty child slot,
1051	 * and we should just put our new leaf in that.
1052	 * If we point to a T_TNODE, check if it matches our key. Note that
1053	 * a T_TNODE might be skipping any number of bits - its 'pos' need
1054	 * not be the parent's 'pos'+'bits'!
1055	 *
1056	 * If it does match the current key, get pos/bits from it, extract
1057	 * the index from our key, push the T_TNODE and walk the tree.
1058	 *
1059	 * If it doesn't, we have to replace it with a new T_TNODE.
1060	 *
1061	 * If we point to a T_LEAF, it might or might not have the same key
1062	 * as we do. If it does, just change the value, update the T_LEAF's
1063	 * value, and return it.
1064	 * If it doesn't, we need to replace it with a T_TNODE.
1065	 */
1066
1067	while (n != NULL &&  NODE_TYPE(n) == T_TNODE) {
1068		tn = (struct tnode *) n;
1069
1070		check_tnode(tn);
1071
1072		if (tkey_sub_equals(tn->key, pos, tn->pos-pos, key)) {
1073			tp = tn;
1074			pos = tn->pos + tn->bits;
1075			n = tnode_get_child(tn,
1076					    tkey_extract_bits(key,
1077							      tn->pos,
1078							      tn->bits));
1079
1080			BUG_ON(n && node_parent(n) != tn);
1081		} else
1082			break;
 
 
 
 
1083	}
1084
1085	/*
1086	 * n  ----> NULL, LEAF or TNODE
1087	 *
1088	 * tp is n's (parent) ----> NULL or TNODE
1089	 */
1090
1091	BUG_ON(tp && IS_LEAF(tp));
 
 
 
 
1092
1093	/* Case 1: n is a leaf. Compare prefixes */
 
 
 
1094
1095	if (n != NULL && IS_LEAF(n) && tkey_equals(key, n->key)) {
1096		l = (struct leaf *) n;
1097		li = leaf_info_new(plen);
1098
1099		if (!li)
1100			return NULL;
1101
1102		fa_head = &li->falh;
1103		insert_leaf_info(&l->list, li);
1104		goto done;
1105	}
1106	l = leaf_new();
 
 
 
1107
1108	if (!l)
1109		return NULL;
 
 
 
 
 
1110
1111	l->key = key;
1112	li = leaf_info_new(plen);
 
1113
1114	if (!li) {
1115		free_leaf(l);
1116		return NULL;
1117	}
1118
1119	fa_head = &li->falh;
1120	insert_leaf_info(&l->list, li);
 
 
 
1121
1122	if (t->trie && n == NULL) {
1123		/* Case 2: n is NULL, and will just insert a new leaf */
 
 
 
 
1124
1125		node_set_parent((struct rt_trie_node *)l, tp);
 
 
 
 
 
 
 
 
1126
1127		cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1128		put_child(t, (struct tnode *)tp, cindex, (struct rt_trie_node *)l);
1129	} else {
1130		/* Case 3: n is a LEAF or a TNODE and the key doesn't match. */
1131		/*
1132		 *  Add a new tnode here
1133		 *  first tnode need some special handling
1134		 */
1135
1136		if (tp)
1137			pos = tp->pos+tp->bits;
1138		else
1139			pos = 0;
1140
1141		if (n) {
1142			newpos = tkey_mismatch(key, pos, n->key);
1143			tn = tnode_new(n->key, newpos, 1);
1144		} else {
1145			newpos = 0;
1146			tn = tnode_new(key, newpos, 1); /* First tnode */
1147		}
1148
1149		if (!tn) {
1150			free_leaf_info(li);
1151			free_leaf(l);
1152			return NULL;
1153		}
1154
1155		node_set_parent((struct rt_trie_node *)tn, tp);
 
 
 
 
1156
1157		missbit = tkey_extract_bits(key, newpos, 1);
1158		put_child(t, tn, missbit, (struct rt_trie_node *)l);
1159		put_child(t, tn, 1-missbit, n);
1160
1161		if (tp) {
1162			cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1163			put_child(t, (struct tnode *)tp, cindex,
1164				  (struct rt_trie_node *)tn);
1165		} else {
1166			rcu_assign_pointer(t->trie, (struct rt_trie_node *)tn);
1167			tp = tn;
1168		}
1169	}
1170
1171	if (tp && tp->pos + tp->bits > 32)
1172		pr_warn("fib_trie tp=%p pos=%d, bits=%d, key=%0x plen=%d\n",
1173			tp, tp->pos, tp->bits, key, plen);
1174
1175	/* Rebalance the trie */
1176
1177	trie_rebalance(t, tp);
1178done:
1179	return fa_head;
1180}
1181
1182/*
1183 * Caller must hold RTNL.
1184 */
1185int fib_table_insert(struct fib_table *tb, struct fib_config *cfg)
1186{
1187	struct trie *t = (struct trie *) tb->tb_data;
 
1188	struct fib_alias *fa, *new_fa;
1189	struct list_head *fa_head = NULL;
 
1190	struct fib_info *fi;
1191	int plen = cfg->fc_dst_len;
 
1192	u8 tos = cfg->fc_tos;
1193	u32 key, mask;
1194	int err;
1195	struct leaf *l;
1196
1197	if (plen > 32)
1198		return -EINVAL;
1199
1200	key = ntohl(cfg->fc_dst);
1201
1202	pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1203
1204	mask = ntohl(inet_make_mask(plen));
1205
1206	if (key & ~mask)
1207		return -EINVAL;
1208
1209	key = key & mask;
1210
1211	fi = fib_create_info(cfg);
1212	if (IS_ERR(fi)) {
1213		err = PTR_ERR(fi);
1214		goto err;
1215	}
1216
1217	l = fib_find_node(t, key);
1218	fa = NULL;
1219
1220	if (l) {
1221		fa_head = get_fa_head(l, plen);
1222		fa = fib_find_alias(fa_head, tos, fi->fib_priority);
1223	}
1224
1225	/* Now fa, if non-NULL, points to the first fib alias
1226	 * with the same keys [prefix,tos,priority], if such key already
1227	 * exists or to the node before which we will insert new one.
1228	 *
1229	 * If fa is NULL, we will need to allocate a new one and
1230	 * insert to the head of f.
1231	 *
1232	 * If f is NULL, no fib node matched the destination key
1233	 * and we need to allocate a new one of those as well.
1234	 */
1235
1236	if (fa && fa->fa_tos == tos &&
1237	    fa->fa_info->fib_priority == fi->fib_priority) {
1238		struct fib_alias *fa_first, *fa_match;
1239
1240		err = -EEXIST;
1241		if (cfg->fc_nlflags & NLM_F_EXCL)
1242			goto out;
1243
 
 
1244		/* We have 2 goals:
1245		 * 1. Find exact match for type, scope, fib_info to avoid
1246		 * duplicate routes
1247		 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1248		 */
1249		fa_match = NULL;
1250		fa_first = fa;
1251		fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
1252		list_for_each_entry_continue(fa, fa_head, fa_list) {
1253			if (fa->fa_tos != tos)
 
1254				break;
1255			if (fa->fa_info->fib_priority != fi->fib_priority)
1256				break;
1257			if (fa->fa_type == cfg->fc_type &&
1258			    fa->fa_info == fi) {
1259				fa_match = fa;
1260				break;
1261			}
1262		}
1263
1264		if (cfg->fc_nlflags & NLM_F_REPLACE) {
1265			struct fib_info *fi_drop;
1266			u8 state;
1267
 
1268			fa = fa_first;
1269			if (fa_match) {
1270				if (fa == fa_match)
1271					err = 0;
1272				goto out;
1273			}
1274			err = -ENOBUFS;
1275			new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1276			if (new_fa == NULL)
1277				goto out;
1278
1279			fi_drop = fa->fa_info;
1280			new_fa->fa_tos = fa->fa_tos;
1281			new_fa->fa_info = fi;
1282			new_fa->fa_type = cfg->fc_type;
1283			state = fa->fa_state;
1284			new_fa->fa_state = state & ~FA_S_ACCESSED;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1285
1286			list_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1287			alias_free_mem_rcu(fa);
1288
1289			fib_release_info(fi_drop);
1290			if (state & FA_S_ACCESSED)
1291				rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1292			rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1293				tb->tb_id, &cfg->fc_nlinfo, NLM_F_REPLACE);
1294
1295			goto succeeded;
1296		}
1297		/* Error if we find a perfect match which
1298		 * uses the same scope, type, and nexthop
1299		 * information.
1300		 */
1301		if (fa_match)
1302			goto out;
1303
1304		if (!(cfg->fc_nlflags & NLM_F_APPEND))
 
 
 
1305			fa = fa_first;
 
1306	}
1307	err = -ENOENT;
1308	if (!(cfg->fc_nlflags & NLM_F_CREATE))
1309		goto out;
1310
 
1311	err = -ENOBUFS;
1312	new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1313	if (new_fa == NULL)
1314		goto out;
1315
1316	new_fa->fa_info = fi;
1317	new_fa->fa_tos = tos;
1318	new_fa->fa_type = cfg->fc_type;
1319	new_fa->fa_state = 0;
1320	/*
1321	 * Insert new entry to the list.
1322	 */
1323
1324	if (!fa_head) {
1325		fa_head = fib_insert_node(t, key, plen);
1326		if (unlikely(!fa_head)) {
1327			err = -ENOMEM;
1328			goto out_free_new_fa;
1329		}
1330	}
 
1331
1332	if (!plen)
1333		tb->tb_num_default++;
1334
1335	list_add_tail_rcu(&new_fa->fa_list,
1336			  (fa ? &fa->fa_list : fa_head));
1337
1338	rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1339	rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, tb->tb_id,
1340		  &cfg->fc_nlinfo, 0);
1341succeeded:
1342	return 0;
1343
 
 
 
 
 
 
 
 
1344out_free_new_fa:
1345	kmem_cache_free(fn_alias_kmem, new_fa);
1346out:
1347	fib_release_info(fi);
1348err:
1349	return err;
1350}
1351
1352/* should be called with rcu_read_lock */
1353static int check_leaf(struct fib_table *tb, struct trie *t, struct leaf *l,
1354		      t_key key,  const struct flowi4 *flp,
1355		      struct fib_result *res, int fib_flags)
1356{
1357	struct leaf_info *li;
1358	struct hlist_head *hhead = &l->list;
1359	struct hlist_node *node;
1360
1361	hlist_for_each_entry_rcu(li, node, hhead, hlist) {
1362		struct fib_alias *fa;
1363
1364		if (l->key != (key & li->mask_plen))
1365			continue;
1366
1367		list_for_each_entry_rcu(fa, &li->falh, fa_list) {
1368			struct fib_info *fi = fa->fa_info;
1369			int nhsel, err;
1370
1371			if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1372				continue;
1373			if (fi->fib_dead)
1374				continue;
1375			if (fa->fa_info->fib_scope < flp->flowi4_scope)
1376				continue;
1377			fib_alias_accessed(fa);
1378			err = fib_props[fa->fa_type].error;
1379			if (err) {
1380#ifdef CONFIG_IP_FIB_TRIE_STATS
1381				t->stats.semantic_match_passed++;
1382#endif
1383				return err;
1384			}
1385			if (fi->fib_flags & RTNH_F_DEAD)
1386				continue;
1387			for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
1388				const struct fib_nh *nh = &fi->fib_nh[nhsel];
1389
1390				if (nh->nh_flags & RTNH_F_DEAD)
1391					continue;
1392				if (flp->flowi4_oif && flp->flowi4_oif != nh->nh_oif)
1393					continue;
1394
1395#ifdef CONFIG_IP_FIB_TRIE_STATS
1396				t->stats.semantic_match_passed++;
1397#endif
1398				res->prefixlen = li->plen;
1399				res->nh_sel = nhsel;
1400				res->type = fa->fa_type;
1401				res->scope = fa->fa_info->fib_scope;
1402				res->fi = fi;
1403				res->table = tb;
1404				res->fa_head = &li->falh;
1405				if (!(fib_flags & FIB_LOOKUP_NOREF))
1406					atomic_inc(&fi->fib_clntref);
1407				return 0;
1408			}
1409		}
1410
1411#ifdef CONFIG_IP_FIB_TRIE_STATS
1412		t->stats.semantic_match_miss++;
1413#endif
1414	}
1415
1416	return 1;
1417}
1418
 
1419int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1420		     struct fib_result *res, int fib_flags)
1421{
1422	struct trie *t = (struct trie *) tb->tb_data;
1423	int ret;
1424	struct rt_trie_node *n;
1425	struct tnode *pn;
1426	unsigned int pos, bits;
1427	t_key key = ntohl(flp->daddr);
1428	unsigned int chopped_off;
1429	t_key cindex = 0;
1430	unsigned int current_prefix_length = KEYLENGTH;
1431	struct tnode *cn;
1432	t_key pref_mismatch;
1433
1434	rcu_read_lock();
1435
1436	n = rcu_dereference(t->trie);
1437	if (!n)
1438		goto failed;
1439
1440#ifdef CONFIG_IP_FIB_TRIE_STATS
1441	t->stats.gets++;
1442#endif
 
 
 
 
 
1443
1444	/* Just a leaf? */
1445	if (IS_LEAF(n)) {
1446		ret = check_leaf(tb, t, (struct leaf *)n, key, flp, res, fib_flags);
1447		goto found;
1448	}
1449
1450	pn = (struct tnode *) n;
1451	chopped_off = 0;
1452
1453	while (pn) {
1454		pos = pn->pos;
1455		bits = pn->bits;
1456
1457		if (!chopped_off)
1458			cindex = tkey_extract_bits(mask_pfx(key, current_prefix_length),
1459						   pos, bits);
1460
1461		n = tnode_get_child_rcu(pn, cindex);
1462
1463		if (n == NULL) {
1464#ifdef CONFIG_IP_FIB_TRIE_STATS
1465			t->stats.null_node_hit++;
1466#endif
1467			goto backtrace;
1468		}
1469
1470		if (IS_LEAF(n)) {
1471			ret = check_leaf(tb, t, (struct leaf *)n, key, flp, res, fib_flags);
1472			if (ret > 0)
1473				goto backtrace;
1474			goto found;
1475		}
1476
1477		cn = (struct tnode *)n;
1478
1479		/*
1480		 * It's a tnode, and we can do some extra checks here if we
1481		 * like, to avoid descending into a dead-end branch.
1482		 * This tnode is in the parent's child array at index
1483		 * key[p_pos..p_pos+p_bits] but potentially with some bits
1484		 * chopped off, so in reality the index may be just a
1485		 * subprefix, padded with zero at the end.
1486		 * We can also take a look at any skipped bits in this
1487		 * tnode - everything up to p_pos is supposed to be ok,
1488		 * and the non-chopped bits of the index (se previous
1489		 * paragraph) are also guaranteed ok, but the rest is
1490		 * considered unknown.
1491		 *
1492		 * The skipped bits are key[pos+bits..cn->pos].
 
 
1493		 */
 
 
 
 
 
 
1494
1495		/* If current_prefix_length < pos+bits, we are already doing
1496		 * actual prefix  matching, which means everything from
1497		 * pos+(bits-chopped_off) onward must be zero along some
1498		 * branch of this subtree - otherwise there is *no* valid
1499		 * prefix present. Here we can only check the skipped
1500		 * bits. Remember, since we have already indexed into the
1501		 * parent's child array, we know that the bits we chopped of
1502		 * *are* zero.
1503		 */
 
 
 
 
1504
1505		/* NOTA BENE: Checking only skipped bits
1506		   for the new node here */
 
 
1507
1508		if (current_prefix_length < pos+bits) {
1509			if (tkey_extract_bits(cn->key, current_prefix_length,
1510						cn->pos - current_prefix_length)
1511			    || !(cn->child[0]))
1512				goto backtrace;
1513		}
1514
1515		/*
1516		 * If chopped_off=0, the index is fully validated and we
1517		 * only need to look at the skipped bits for this, the new,
1518		 * tnode. What we actually want to do is to find out if
1519		 * these skipped bits match our key perfectly, or if we will
1520		 * have to count on finding a matching prefix further down,
1521		 * because if we do, we would like to have some way of
1522		 * verifying the existence of such a prefix at this point.
1523		 */
 
 
1524
1525		/* The only thing we can do at this point is to verify that
1526		 * any such matching prefix can indeed be a prefix to our
1527		 * key, and if the bits in the node we are inspecting that
1528		 * do not match our key are not ZERO, this cannot be true.
1529		 * Thus, find out where there is a mismatch (before cn->pos)
1530		 * and verify that all the mismatching bits are zero in the
1531		 * new tnode's key.
1532		 */
1533
1534		/*
1535		 * Note: We aren't very concerned about the piece of
1536		 * the key that precede pn->pos+pn->bits, since these
1537		 * have already been checked. The bits after cn->pos
1538		 * aren't checked since these are by definition
1539		 * "unknown" at this point. Thus, what we want to see
1540		 * is if we are about to enter the "prefix matching"
1541		 * state, and in that case verify that the skipped
1542		 * bits that will prevail throughout this subtree are
1543		 * zero, as they have to be if we are to find a
1544		 * matching prefix.
1545		 */
1546
1547		pref_mismatch = mask_pfx(cn->key ^ key, cn->pos);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1548
1549		/*
1550		 * In short: If skipped bits in this node do not match
1551		 * the search key, enter the "prefix matching"
1552		 * state.directly.
1553		 */
1554		if (pref_mismatch) {
1555			int mp = KEYLENGTH - fls(pref_mismatch);
1556
1557			if (tkey_extract_bits(cn->key, mp, cn->pos - mp) != 0)
1558				goto backtrace;
 
 
1559
1560			if (current_prefix_length >= cn->pos)
1561				current_prefix_length = mp;
1562		}
1563
1564		pn = (struct tnode *)n; /* Descend */
1565		chopped_off = 0;
1566		continue;
 
1567
1568backtrace:
1569		chopped_off++;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1570
1571		/* As zero don't change the child key (cindex) */
1572		while ((chopped_off <= pn->bits)
1573		       && !(cindex & (1<<(chopped_off-1))))
1574			chopped_off++;
1575
1576		/* Decrease current_... with bits chopped off */
1577		if (current_prefix_length > pn->pos + pn->bits - chopped_off)
1578			current_prefix_length = pn->pos + pn->bits
1579				- chopped_off;
1580
1581		/*
1582		 * Either we do the actual chop off according or if we have
1583		 * chopped off all bits in this tnode walk up to our parent.
1584		 */
1585
1586		if (chopped_off <= pn->bits) {
1587			cindex &= ~(1 << (chopped_off-1));
1588		} else {
1589			struct tnode *parent = node_parent_rcu((struct rt_trie_node *) pn);
1590			if (!parent)
1591				goto failed;
1592
1593			/* Get Child's index */
1594			cindex = tkey_extract_bits(pn->key, parent->pos, parent->bits);
1595			pn = parent;
1596			chopped_off = 0;
1597
 
 
 
 
 
 
 
 
1598#ifdef CONFIG_IP_FIB_TRIE_STATS
1599			t->stats.backtrack++;
1600#endif
1601			goto backtrace;
 
 
1602		}
1603	}
1604failed:
1605	ret = 1;
1606found:
1607	rcu_read_unlock();
1608	return ret;
1609}
1610EXPORT_SYMBOL_GPL(fib_table_lookup);
1611
1612/*
1613 * Remove the leaf and return parent.
1614 */
1615static void trie_leaf_remove(struct trie *t, struct leaf *l)
1616{
1617	struct tnode *tp = node_parent((struct rt_trie_node *) l);
 
 
1618
1619	pr_debug("entering trie_leaf_remove(%p)\n", l);
 
1620
1621	if (tp) {
1622		t_key cindex = tkey_extract_bits(l->key, tp->pos, tp->bits);
1623		put_child(t, (struct tnode *)tp, cindex, NULL);
 
 
 
 
 
1624		trie_rebalance(t, tp);
1625	} else
1626		RCU_INIT_POINTER(t->trie, NULL);
 
 
 
 
1627
1628	free_leaf(l);
 
 
1629}
1630
1631/*
1632 * Caller must hold RTNL.
1633 */
1634int fib_table_delete(struct fib_table *tb, struct fib_config *cfg)
1635{
1636	struct trie *t = (struct trie *) tb->tb_data;
1637	u32 key, mask;
1638	int plen = cfg->fc_dst_len;
 
 
1639	u8 tos = cfg->fc_tos;
1640	struct fib_alias *fa, *fa_to_delete;
1641	struct list_head *fa_head;
1642	struct leaf *l;
1643	struct leaf_info *li;
1644
1645	if (plen > 32)
1646		return -EINVAL;
1647
1648	key = ntohl(cfg->fc_dst);
1649	mask = ntohl(inet_make_mask(plen));
1650
1651	if (key & ~mask)
1652		return -EINVAL;
1653
1654	key = key & mask;
1655	l = fib_find_node(t, key);
1656
1657	if (!l)
1658		return -ESRCH;
1659
1660	fa_head = get_fa_head(l, plen);
1661	fa = fib_find_alias(fa_head, tos, 0);
1662
1663	if (!fa)
1664		return -ESRCH;
1665
1666	pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1667
1668	fa_to_delete = NULL;
1669	fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
1670	list_for_each_entry_continue(fa, fa_head, fa_list) {
1671		struct fib_info *fi = fa->fa_info;
1672
1673		if (fa->fa_tos != tos)
 
 
1674			break;
1675
1676		if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1677		    (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1678		     fa->fa_info->fib_scope == cfg->fc_scope) &&
1679		    (!cfg->fc_prefsrc ||
1680		     fi->fib_prefsrc == cfg->fc_prefsrc) &&
1681		    (!cfg->fc_protocol ||
1682		     fi->fib_protocol == cfg->fc_protocol) &&
1683		    fib_nh_match(cfg, fi) == 0) {
 
1684			fa_to_delete = fa;
1685			break;
1686		}
1687	}
1688
1689	if (!fa_to_delete)
1690		return -ESRCH;
1691
1692	fa = fa_to_delete;
1693	rtmsg_fib(RTM_DELROUTE, htonl(key), fa, plen, tb->tb_id,
 
1694		  &cfg->fc_nlinfo, 0);
1695
1696	l = fib_find_node(t, key);
1697	li = find_leaf_info(l, plen);
1698
1699	list_del_rcu(&fa->fa_list);
1700
1701	if (!plen)
1702		tb->tb_num_default--;
1703
1704	if (list_empty(fa_head)) {
1705		hlist_del_rcu(&li->hlist);
1706		free_leaf_info(li);
1707	}
1708
1709	if (hlist_empty(&l->list))
1710		trie_leaf_remove(t, l);
1711
1712	if (fa->fa_state & FA_S_ACCESSED)
1713		rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1714
1715	fib_release_info(fa->fa_info);
1716	alias_free_mem_rcu(fa);
1717	return 0;
1718}
1719
1720static int trie_flush_list(struct list_head *head)
 
1721{
1722	struct fib_alias *fa, *fa_node;
1723	int found = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1724
1725	list_for_each_entry_safe(fa, fa_node, head, fa_list) {
1726		struct fib_info *fi = fa->fa_info;
 
 
 
1727
1728		if (fi && (fi->fib_flags & RTNH_F_DEAD)) {
1729			list_del_rcu(&fa->fa_list);
1730			fib_release_info(fa->fa_info);
1731			alias_free_mem_rcu(fa);
1732			found++;
1733		}
 
 
 
 
 
 
 
 
 
 
 
 
 
1734	}
1735	return found;
 
 
 
 
 
 
1736}
1737
1738static int trie_flush_leaf(struct leaf *l)
1739{
1740	int found = 0;
1741	struct hlist_head *lih = &l->list;
1742	struct hlist_node *node, *tmp;
1743	struct leaf_info *li = NULL;
1744
1745	hlist_for_each_entry_safe(li, node, tmp, lih, hlist) {
1746		found += trie_flush_list(&li->falh);
1747
1748		if (list_empty(&li->falh)) {
1749			hlist_del_rcu(&li->hlist);
1750			free_leaf_info(li);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1751		}
 
 
 
1752	}
1753	return found;
 
 
 
 
1754}
1755
1756/*
1757 * Scan for the next right leaf starting at node p->child[idx]
1758 * Since we have back pointer, no recursion necessary.
1759 */
1760static struct leaf *leaf_walk_rcu(struct tnode *p, struct rt_trie_node *c)
1761{
1762	do {
1763		t_key idx;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1764
1765		if (c)
1766			idx = tkey_extract_bits(c->key, p->pos, p->bits) + 1;
1767		else
1768			idx = 0;
1769
1770		while (idx < 1u << p->bits) {
1771			c = tnode_get_child_rcu(p, idx++);
1772			if (!c)
1773				continue;
1774
1775			if (IS_LEAF(c)) {
1776				prefetch(rcu_dereference_rtnl(p->child[idx]));
1777				return (struct leaf *) c;
 
 
 
 
 
 
 
 
 
 
 
 
1778			}
 
1779
1780			/* Rescan start scanning in new node */
1781			p = (struct tnode *) c;
1782			idx = 0;
1783		}
 
1784
1785		/* Node empty, walk back up to parent */
1786		c = (struct rt_trie_node *) p;
1787	} while ((p = node_parent_rcu(c)) != NULL);
1788
1789	return NULL; /* Root of trie */
1790}
1791
1792static struct leaf *trie_firstleaf(struct trie *t)
 
1793{
1794	struct tnode *n = (struct tnode *)rcu_dereference_rtnl(t->trie);
 
 
 
 
1795
1796	if (!n)
1797		return NULL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1798
1799	if (IS_LEAF(n))          /* trie is just a leaf */
1800		return (struct leaf *) n;
1801
1802	return leaf_walk_rcu(n, NULL);
 
 
 
 
1803}
1804
1805static struct leaf *trie_nextleaf(struct leaf *l)
 
1806{
1807	struct rt_trie_node *c = (struct rt_trie_node *) l;
1808	struct tnode *p = node_parent_rcu(c);
 
 
 
 
1809
1810	if (!p)
1811		return NULL;	/* trie with just one leaf */
 
 
1812
1813	return leaf_walk_rcu(p, c);
1814}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1815
1816static struct leaf *trie_leafindex(struct trie *t, int index)
1817{
1818	struct leaf *l = trie_firstleaf(t);
 
1819
1820	while (l && index-- > 0)
1821		l = trie_nextleaf(l);
1822
1823	return l;
1824}
1825
 
 
 
 
 
1826
1827/*
1828 * Caller must hold RTNL.
1829 */
1830int fib_table_flush(struct fib_table *tb)
1831{
1832	struct trie *t = (struct trie *) tb->tb_data;
1833	struct leaf *l, *ll = NULL;
1834	int found = 0;
 
1835
1836	for (l = trie_firstleaf(t); l; l = trie_nextleaf(l)) {
1837		found += trie_flush_leaf(l);
1838
1839		if (ll && hlist_empty(&ll->list))
1840			trie_leaf_remove(t, ll);
1841		ll = l;
 
1842	}
1843
1844	if (ll && hlist_empty(&ll->list))
1845		trie_leaf_remove(t, ll);
1846
1847	pr_debug("trie_flush found=%d\n", found);
1848	return found;
1849}
1850
1851void fib_free_table(struct fib_table *tb)
 
1852{
1853	kfree(tb);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1854}
1855
1856static int fn_trie_dump_fa(t_key key, int plen, struct list_head *fah,
1857			   struct fib_table *tb,
1858			   struct sk_buff *skb, struct netlink_callback *cb)
1859{
1860	int i, s_i;
1861	struct fib_alias *fa;
1862	__be32 xkey = htonl(key);
 
 
 
1863
1864	s_i = cb->args[5];
1865	i = 0;
 
 
 
 
1866
1867	/* rcu_read_lock is hold by caller */
 
 
1868
1869	list_for_each_entry_rcu(fa, fah, fa_list) {
1870		if (i < s_i) {
1871			i++;
1872			continue;
1873		}
1874
1875		if (fib_dump_info(skb, NETLINK_CB(cb->skb).pid,
1876				  cb->nlh->nlmsg_seq,
1877				  RTM_NEWROUTE,
1878				  tb->tb_id,
1879				  fa->fa_type,
1880				  xkey,
1881				  plen,
1882				  fa->fa_tos,
1883				  fa->fa_info, NLM_F_MULTI) < 0) {
1884			cb->args[5] = i;
1885			return -1;
1886		}
1887		i++;
1888	}
1889	cb->args[5] = i;
1890	return skb->len;
1891}
1892
1893static int fn_trie_dump_leaf(struct leaf *l, struct fib_table *tb,
1894			struct sk_buff *skb, struct netlink_callback *cb)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1895{
1896	struct leaf_info *li;
1897	struct hlist_node *node;
1898	int i, s_i;
1899
1900	s_i = cb->args[4];
1901	i = 0;
1902
1903	/* rcu_read_lock is hold by caller */
1904	hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
 
 
1905		if (i < s_i) {
1906			i++;
1907			continue;
1908		}
1909
1910		if (i > s_i)
1911			cb->args[5] = 0;
1912
1913		if (list_empty(&li->falh))
1914			continue;
 
1915
1916		if (fn_trie_dump_fa(l->key, li->plen, &li->falh, tb, skb, cb) < 0) {
 
 
 
 
 
1917			cb->args[4] = i;
1918			return -1;
1919		}
1920		i++;
1921	}
1922
1923	cb->args[4] = i;
1924	return skb->len;
1925}
1926
 
1927int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
1928		   struct netlink_callback *cb)
1929{
1930	struct leaf *l;
1931	struct trie *t = (struct trie *) tb->tb_data;
1932	t_key key = cb->args[2];
1933	int count = cb->args[3];
1934
1935	rcu_read_lock();
1936	/* Dump starting at last key.
1937	 * Note: 0.0.0.0/0 (ie default) is first key.
1938	 */
1939	if (count == 0)
1940		l = trie_firstleaf(t);
1941	else {
1942		/* Normally, continue from last key, but if that is missing
1943		 * fallback to using slow rescan
1944		 */
1945		l = fib_find_node(t, key);
1946		if (!l)
1947			l = trie_leafindex(t, count);
1948	}
1949
1950	while (l) {
1951		cb->args[2] = l->key;
1952		if (fn_trie_dump_leaf(l, tb, skb, cb) < 0) {
1953			cb->args[3] = count;
1954			rcu_read_unlock();
1955			return -1;
1956		}
1957
1958		++count;
1959		l = trie_nextleaf(l);
 
1960		memset(&cb->args[4], 0,
1961		       sizeof(cb->args) - 4*sizeof(cb->args[0]));
 
 
 
 
1962	}
1963	cb->args[3] = count;
1964	rcu_read_unlock();
 
1965
1966	return skb->len;
1967}
1968
1969void __init fib_trie_init(void)
1970{
1971	fn_alias_kmem = kmem_cache_create("ip_fib_alias",
1972					  sizeof(struct fib_alias),
1973					  0, SLAB_PANIC, NULL);
1974
1975	trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
1976					   max(sizeof(struct leaf),
1977					       sizeof(struct leaf_info)),
1978					   0, SLAB_PANIC, NULL);
1979}
1980
1981
1982struct fib_table *fib_trie_table(u32 id)
1983{
1984	struct fib_table *tb;
1985	struct trie *t;
 
1986
1987	tb = kmalloc(sizeof(struct fib_table) + sizeof(struct trie),
1988		     GFP_KERNEL);
1989	if (tb == NULL)
 
 
1990		return NULL;
1991
1992	tb->tb_id = id;
1993	tb->tb_default = -1;
1994	tb->tb_num_default = 0;
 
 
 
 
1995
1996	t = (struct trie *) tb->tb_data;
1997	memset(t, 0, sizeof(*t));
 
 
 
 
 
 
 
 
1998
1999	return tb;
2000}
2001
2002#ifdef CONFIG_PROC_FS
2003/* Depth first Trie walk iterator */
2004struct fib_trie_iter {
2005	struct seq_net_private p;
2006	struct fib_table *tb;
2007	struct tnode *tnode;
2008	unsigned int index;
2009	unsigned int depth;
2010};
2011
2012static struct rt_trie_node *fib_trie_get_next(struct fib_trie_iter *iter)
2013{
2014	struct tnode *tn = iter->tnode;
2015	unsigned int cindex = iter->index;
2016	struct tnode *p;
2017
2018	/* A single entry routing table */
2019	if (!tn)
2020		return NULL;
2021
2022	pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2023		 iter->tnode, iter->index, iter->depth);
2024rescan:
2025	while (cindex < (1<<tn->bits)) {
2026		struct rt_trie_node *n = tnode_get_child_rcu(tn, cindex);
2027
2028		if (n) {
 
 
 
 
 
 
2029			if (IS_LEAF(n)) {
2030				iter->tnode = tn;
2031				iter->index = cindex + 1;
2032			} else {
2033				/* push down one level */
2034				iter->tnode = (struct tnode *) n;
2035				iter->index = 0;
2036				++iter->depth;
2037			}
 
2038			return n;
2039		}
2040
2041		++cindex;
 
 
 
 
2042	}
2043
2044	/* Current node exhausted, pop back up */
2045	p = node_parent_rcu((struct rt_trie_node *)tn);
2046	if (p) {
2047		cindex = tkey_extract_bits(tn->key, p->pos, p->bits)+1;
2048		tn = p;
2049		--iter->depth;
2050		goto rescan;
2051	}
2052
2053	/* got root? */
2054	return NULL;
2055}
2056
2057static struct rt_trie_node *fib_trie_get_first(struct fib_trie_iter *iter,
2058				       struct trie *t)
2059{
2060	struct rt_trie_node *n;
2061
2062	if (!t)
2063		return NULL;
2064
2065	n = rcu_dereference(t->trie);
 
2066	if (!n)
2067		return NULL;
2068
2069	if (IS_TNODE(n)) {
2070		iter->tnode = (struct tnode *) n;
2071		iter->index = 0;
2072		iter->depth = 1;
2073	} else {
2074		iter->tnode = NULL;
2075		iter->index = 0;
2076		iter->depth = 0;
2077	}
2078
2079	return n;
2080}
2081
2082static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2083{
2084	struct rt_trie_node *n;
2085	struct fib_trie_iter iter;
2086
2087	memset(s, 0, sizeof(*s));
2088
2089	rcu_read_lock();
2090	for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2091		if (IS_LEAF(n)) {
2092			struct leaf *l = (struct leaf *)n;
2093			struct leaf_info *li;
2094			struct hlist_node *tmp;
2095
2096			s->leaves++;
2097			s->totdepth += iter.depth;
2098			if (iter.depth > s->maxdepth)
2099				s->maxdepth = iter.depth;
2100
2101			hlist_for_each_entry_rcu(li, tmp, &l->list, hlist)
2102				++s->prefixes;
2103		} else {
2104			const struct tnode *tn = (const struct tnode *) n;
2105			int i;
2106
2107			s->tnodes++;
2108			if (tn->bits < MAX_STAT_DEPTH)
2109				s->nodesizes[tn->bits]++;
2110
2111			for (i = 0; i < (1<<tn->bits); i++)
2112				if (!tn->child[i])
2113					s->nullpointers++;
2114		}
2115	}
2116	rcu_read_unlock();
2117}
2118
2119/*
2120 *	This outputs /proc/net/fib_triestats
2121 */
2122static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2123{
2124	unsigned int i, max, pointers, bytes, avdepth;
2125
2126	if (stat->leaves)
2127		avdepth = stat->totdepth*100 / stat->leaves;
2128	else
2129		avdepth = 0;
2130
2131	seq_printf(seq, "\tAver depth:     %u.%02d\n",
2132		   avdepth / 100, avdepth % 100);
2133	seq_printf(seq, "\tMax depth:      %u\n", stat->maxdepth);
2134
2135	seq_printf(seq, "\tLeaves:         %u\n", stat->leaves);
2136	bytes = sizeof(struct leaf) * stat->leaves;
2137
2138	seq_printf(seq, "\tPrefixes:       %u\n", stat->prefixes);
2139	bytes += sizeof(struct leaf_info) * stat->prefixes;
2140
2141	seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2142	bytes += sizeof(struct tnode) * stat->tnodes;
2143
2144	max = MAX_STAT_DEPTH;
2145	while (max > 0 && stat->nodesizes[max-1] == 0)
2146		max--;
2147
2148	pointers = 0;
2149	for (i = 1; i <= max; i++)
2150		if (stat->nodesizes[i] != 0) {
2151			seq_printf(seq, "  %u: %u",  i, stat->nodesizes[i]);
2152			pointers += (1<<i) * stat->nodesizes[i];
2153		}
2154	seq_putc(seq, '\n');
2155	seq_printf(seq, "\tPointers: %u\n", pointers);
2156
2157	bytes += sizeof(struct rt_trie_node *) * pointers;
2158	seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2159	seq_printf(seq, "Total size: %u  kB\n", (bytes + 1023) / 1024);
2160}
2161
2162#ifdef CONFIG_IP_FIB_TRIE_STATS
2163static void trie_show_usage(struct seq_file *seq,
2164			    const struct trie_use_stats *stats)
2165{
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2166	seq_printf(seq, "\nCounters:\n---------\n");
2167	seq_printf(seq, "gets = %u\n", stats->gets);
2168	seq_printf(seq, "backtracks = %u\n", stats->backtrack);
2169	seq_printf(seq, "semantic match passed = %u\n",
2170		   stats->semantic_match_passed);
2171	seq_printf(seq, "semantic match miss = %u\n",
2172		   stats->semantic_match_miss);
2173	seq_printf(seq, "null node hit= %u\n", stats->null_node_hit);
2174	seq_printf(seq, "skipped node resize = %u\n\n",
2175		   stats->resize_node_skipped);
2176}
2177#endif /*  CONFIG_IP_FIB_TRIE_STATS */
2178
2179static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2180{
2181	if (tb->tb_id == RT_TABLE_LOCAL)
2182		seq_puts(seq, "Local:\n");
2183	else if (tb->tb_id == RT_TABLE_MAIN)
2184		seq_puts(seq, "Main:\n");
2185	else
2186		seq_printf(seq, "Id %d:\n", tb->tb_id);
2187}
2188
2189
2190static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2191{
2192	struct net *net = (struct net *)seq->private;
2193	unsigned int h;
2194
2195	seq_printf(seq,
2196		   "Basic info: size of leaf:"
2197		   " %Zd bytes, size of tnode: %Zd bytes.\n",
2198		   sizeof(struct leaf), sizeof(struct tnode));
2199
2200	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2201		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2202		struct hlist_node *node;
2203		struct fib_table *tb;
2204
2205		hlist_for_each_entry_rcu(tb, node, head, tb_hlist) {
2206			struct trie *t = (struct trie *) tb->tb_data;
2207			struct trie_stat stat;
2208
2209			if (!t)
2210				continue;
2211
2212			fib_table_print(seq, tb);
2213
2214			trie_collect_stats(t, &stat);
2215			trie_show_stats(seq, &stat);
2216#ifdef CONFIG_IP_FIB_TRIE_STATS
2217			trie_show_usage(seq, &t->stats);
2218#endif
2219		}
2220	}
2221
2222	return 0;
2223}
2224
2225static int fib_triestat_seq_open(struct inode *inode, struct file *file)
2226{
2227	return single_open_net(inode, file, fib_triestat_seq_show);
2228}
2229
2230static const struct file_operations fib_triestat_fops = {
2231	.owner	= THIS_MODULE,
2232	.open	= fib_triestat_seq_open,
2233	.read	= seq_read,
2234	.llseek	= seq_lseek,
2235	.release = single_release_net,
2236};
2237
2238static struct rt_trie_node *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2239{
2240	struct fib_trie_iter *iter = seq->private;
2241	struct net *net = seq_file_net(seq);
2242	loff_t idx = 0;
2243	unsigned int h;
2244
2245	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2246		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2247		struct hlist_node *node;
2248		struct fib_table *tb;
2249
2250		hlist_for_each_entry_rcu(tb, node, head, tb_hlist) {
2251			struct rt_trie_node *n;
2252
2253			for (n = fib_trie_get_first(iter,
2254						    (struct trie *) tb->tb_data);
2255			     n; n = fib_trie_get_next(iter))
2256				if (pos == idx++) {
2257					iter->tb = tb;
2258					return n;
2259				}
2260		}
2261	}
2262
2263	return NULL;
2264}
2265
2266static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2267	__acquires(RCU)
2268{
2269	rcu_read_lock();
2270	return fib_trie_get_idx(seq, *pos);
2271}
2272
2273static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2274{
2275	struct fib_trie_iter *iter = seq->private;
2276	struct net *net = seq_file_net(seq);
2277	struct fib_table *tb = iter->tb;
2278	struct hlist_node *tb_node;
2279	unsigned int h;
2280	struct rt_trie_node *n;
2281
2282	++*pos;
2283	/* next node in same table */
2284	n = fib_trie_get_next(iter);
2285	if (n)
2286		return n;
2287
2288	/* walk rest of this hash chain */
2289	h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2290	while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2291		tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2292		n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2293		if (n)
2294			goto found;
2295	}
2296
2297	/* new hash chain */
2298	while (++h < FIB_TABLE_HASHSZ) {
2299		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2300		hlist_for_each_entry_rcu(tb, tb_node, head, tb_hlist) {
2301			n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2302			if (n)
2303				goto found;
2304		}
2305	}
2306	return NULL;
2307
2308found:
2309	iter->tb = tb;
2310	return n;
2311}
2312
2313static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2314	__releases(RCU)
2315{
2316	rcu_read_unlock();
2317}
2318
2319static void seq_indent(struct seq_file *seq, int n)
2320{
2321	while (n-- > 0)
2322		seq_puts(seq, "   ");
2323}
2324
2325static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2326{
2327	switch (s) {
2328	case RT_SCOPE_UNIVERSE: return "universe";
2329	case RT_SCOPE_SITE:	return "site";
2330	case RT_SCOPE_LINK:	return "link";
2331	case RT_SCOPE_HOST:	return "host";
2332	case RT_SCOPE_NOWHERE:	return "nowhere";
2333	default:
2334		snprintf(buf, len, "scope=%d", s);
2335		return buf;
2336	}
2337}
2338
2339static const char *const rtn_type_names[__RTN_MAX] = {
2340	[RTN_UNSPEC] = "UNSPEC",
2341	[RTN_UNICAST] = "UNICAST",
2342	[RTN_LOCAL] = "LOCAL",
2343	[RTN_BROADCAST] = "BROADCAST",
2344	[RTN_ANYCAST] = "ANYCAST",
2345	[RTN_MULTICAST] = "MULTICAST",
2346	[RTN_BLACKHOLE] = "BLACKHOLE",
2347	[RTN_UNREACHABLE] = "UNREACHABLE",
2348	[RTN_PROHIBIT] = "PROHIBIT",
2349	[RTN_THROW] = "THROW",
2350	[RTN_NAT] = "NAT",
2351	[RTN_XRESOLVE] = "XRESOLVE",
2352};
2353
2354static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2355{
2356	if (t < __RTN_MAX && rtn_type_names[t])
2357		return rtn_type_names[t];
2358	snprintf(buf, len, "type %u", t);
2359	return buf;
2360}
2361
2362/* Pretty print the trie */
2363static int fib_trie_seq_show(struct seq_file *seq, void *v)
2364{
2365	const struct fib_trie_iter *iter = seq->private;
2366	struct rt_trie_node *n = v;
2367
2368	if (!node_parent_rcu(n))
2369		fib_table_print(seq, iter->tb);
2370
2371	if (IS_TNODE(n)) {
2372		struct tnode *tn = (struct tnode *) n;
2373		__be32 prf = htonl(mask_pfx(tn->key, tn->pos));
2374
2375		seq_indent(seq, iter->depth-1);
2376		seq_printf(seq, "  +-- %pI4/%d %d %d %d\n",
2377			   &prf, tn->pos, tn->bits, tn->full_children,
2378			   tn->empty_children);
2379
2380	} else {
2381		struct leaf *l = (struct leaf *) n;
2382		struct leaf_info *li;
2383		struct hlist_node *node;
2384		__be32 val = htonl(l->key);
2385
2386		seq_indent(seq, iter->depth);
2387		seq_printf(seq, "  |-- %pI4\n", &val);
2388
2389		hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
2390			struct fib_alias *fa;
2391
2392			list_for_each_entry_rcu(fa, &li->falh, fa_list) {
2393				char buf1[32], buf2[32];
2394
2395				seq_indent(seq, iter->depth+1);
2396				seq_printf(seq, "  /%d %s %s", li->plen,
2397					   rtn_scope(buf1, sizeof(buf1),
2398						     fa->fa_info->fib_scope),
2399					   rtn_type(buf2, sizeof(buf2),
2400						    fa->fa_type));
2401				if (fa->fa_tos)
2402					seq_printf(seq, " tos=%d", fa->fa_tos);
2403				seq_putc(seq, '\n');
2404			}
2405		}
2406	}
2407
2408	return 0;
2409}
2410
2411static const struct seq_operations fib_trie_seq_ops = {
2412	.start  = fib_trie_seq_start,
2413	.next   = fib_trie_seq_next,
2414	.stop   = fib_trie_seq_stop,
2415	.show   = fib_trie_seq_show,
2416};
2417
2418static int fib_trie_seq_open(struct inode *inode, struct file *file)
2419{
2420	return seq_open_net(inode, file, &fib_trie_seq_ops,
2421			    sizeof(struct fib_trie_iter));
2422}
2423
2424static const struct file_operations fib_trie_fops = {
2425	.owner  = THIS_MODULE,
2426	.open   = fib_trie_seq_open,
2427	.read   = seq_read,
2428	.llseek = seq_lseek,
2429	.release = seq_release_net,
2430};
2431
2432struct fib_route_iter {
2433	struct seq_net_private p;
2434	struct trie *main_trie;
 
2435	loff_t	pos;
2436	t_key	key;
2437};
2438
2439static struct leaf *fib_route_get_idx(struct fib_route_iter *iter, loff_t pos)
 
2440{
2441	struct leaf *l = NULL;
2442	struct trie *t = iter->main_trie;
2443
2444	/* use cache location of last found key */
2445	if (iter->pos > 0 && pos >= iter->pos && (l = fib_find_node(t, iter->key)))
2446		pos -= iter->pos;
2447	else {
2448		iter->pos = 0;
2449		l = trie_firstleaf(t);
2450	}
2451
2452	while (l && pos-- > 0) {
 
 
 
2453		iter->pos++;
2454		l = trie_nextleaf(l);
 
 
 
 
2455	}
2456
2457	if (l)
2458		iter->key = pos;	/* remember it */
2459	else
2460		iter->pos = 0;		/* forget it */
2461
2462	return l;
2463}
2464
2465static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2466	__acquires(RCU)
2467{
2468	struct fib_route_iter *iter = seq->private;
2469	struct fib_table *tb;
 
2470
2471	rcu_read_lock();
 
2472	tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2473	if (!tb)
2474		return NULL;
2475
2476	iter->main_trie = (struct trie *) tb->tb_data;
2477	if (*pos == 0)
2478		return SEQ_START_TOKEN;
2479	else
2480		return fib_route_get_idx(iter, *pos - 1);
 
 
 
 
 
 
2481}
2482
2483static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2484{
2485	struct fib_route_iter *iter = seq->private;
2486	struct leaf *l = v;
 
2487
2488	++*pos;
2489	if (v == SEQ_START_TOKEN) {
 
 
 
 
 
 
 
 
2490		iter->pos = 0;
2491		l = trie_firstleaf(iter->main_trie);
2492	} else {
2493		iter->pos++;
2494		l = trie_nextleaf(l);
2495	}
2496
2497	if (l)
2498		iter->key = l->key;
2499	else
2500		iter->pos = 0;
2501	return l;
2502}
2503
2504static void fib_route_seq_stop(struct seq_file *seq, void *v)
2505	__releases(RCU)
2506{
2507	rcu_read_unlock();
2508}
2509
2510static unsigned int fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2511{
2512	unsigned int flags = 0;
2513
2514	if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2515		flags = RTF_REJECT;
2516	if (fi && fi->fib_nh->nh_gw)
2517		flags |= RTF_GATEWAY;
2518	if (mask == htonl(0xFFFFFFFF))
2519		flags |= RTF_HOST;
2520	flags |= RTF_UP;
2521	return flags;
2522}
2523
2524/*
2525 *	This outputs /proc/net/route.
2526 *	The format of the file is not supposed to be changed
2527 *	and needs to be same as fib_hash output to avoid breaking
2528 *	legacy utilities
2529 */
2530static int fib_route_seq_show(struct seq_file *seq, void *v)
2531{
2532	struct leaf *l = v;
2533	struct leaf_info *li;
2534	struct hlist_node *node;
 
 
2535
2536	if (v == SEQ_START_TOKEN) {
2537		seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2538			   "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2539			   "\tWindow\tIRTT");
2540		return 0;
2541	}
2542
2543	hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
2544		struct fib_alias *fa;
2545		__be32 mask, prefix;
2546
2547		mask = inet_make_mask(li->plen);
2548		prefix = htonl(l->key);
 
 
2549
2550		list_for_each_entry_rcu(fa, &li->falh, fa_list) {
2551			const struct fib_info *fi = fa->fa_info;
2552			unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2553			int len;
2554
2555			if (fa->fa_type == RTN_BROADCAST
2556			    || fa->fa_type == RTN_MULTICAST)
2557				continue;
2558
2559			if (fi)
2560				seq_printf(seq,
2561					 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2562					 "%d\t%08X\t%d\t%u\t%u%n",
2563					 fi->fib_dev ? fi->fib_dev->name : "*",
2564					 prefix,
2565					 fi->fib_nh->nh_gw, flags, 0, 0,
2566					 fi->fib_priority,
2567					 mask,
2568					 (fi->fib_advmss ?
2569					  fi->fib_advmss + 40 : 0),
2570					 fi->fib_window,
2571					 fi->fib_rtt >> 3, &len);
2572			else
2573				seq_printf(seq,
2574					 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2575					 "%d\t%08X\t%d\t%u\t%u%n",
2576					 prefix, 0, flags, 0, 0, 0,
2577					 mask, 0, 0, 0, &len);
2578
2579			seq_printf(seq, "%*s\n", 127 - len, "");
2580		}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2581	}
2582
2583	return 0;
2584}
2585
2586static const struct seq_operations fib_route_seq_ops = {
2587	.start  = fib_route_seq_start,
2588	.next   = fib_route_seq_next,
2589	.stop   = fib_route_seq_stop,
2590	.show   = fib_route_seq_show,
2591};
2592
2593static int fib_route_seq_open(struct inode *inode, struct file *file)
2594{
2595	return seq_open_net(inode, file, &fib_route_seq_ops,
2596			    sizeof(struct fib_route_iter));
2597}
2598
2599static const struct file_operations fib_route_fops = {
2600	.owner  = THIS_MODULE,
2601	.open   = fib_route_seq_open,
2602	.read   = seq_read,
2603	.llseek = seq_lseek,
2604	.release = seq_release_net,
2605};
2606
2607int __net_init fib_proc_init(struct net *net)
2608{
2609	if (!proc_net_fops_create(net, "fib_trie", S_IRUGO, &fib_trie_fops))
2610		goto out1;
2611
2612	if (!proc_net_fops_create(net, "fib_triestat", S_IRUGO,
2613				  &fib_triestat_fops))
2614		goto out2;
2615
2616	if (!proc_net_fops_create(net, "route", S_IRUGO, &fib_route_fops))
2617		goto out3;
2618
2619	return 0;
2620
2621out3:
2622	proc_net_remove(net, "fib_triestat");
2623out2:
2624	proc_net_remove(net, "fib_trie");
2625out1:
2626	return -ENOMEM;
2627}
2628
2629void __net_exit fib_proc_exit(struct net *net)
2630{
2631	proc_net_remove(net, "fib_trie");
2632	proc_net_remove(net, "fib_triestat");
2633	proc_net_remove(net, "route");
2634}
2635
2636#endif /* CONFIG_PROC_FS */