Linux Audio

Check our new training course

Loading...
v6.13.7
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
 
 
 
 
   3 *
   4 *   Robert Olsson <robert.olsson@its.uu.se> Uppsala Universitet
   5 *     & Swedish University of Agricultural Sciences.
   6 *
   7 *   Jens Laas <jens.laas@data.slu.se> Swedish University of
   8 *     Agricultural Sciences.
   9 *
  10 *   Hans Liss <hans.liss@its.uu.se>  Uppsala Universitet
  11 *
  12 * This work is based on the LPC-trie which is originally described in:
  13 *
  14 * An experimental study of compression methods for dynamic tries
  15 * Stefan Nilsson and Matti Tikkanen. Algorithmica, 33(1):19-33, 2002.
  16 * https://www.csc.kth.se/~snilsson/software/dyntrie2/
 
  17 *
  18 * IP-address lookup using LC-tries. Stefan Nilsson and Gunnar Karlsson
  19 * IEEE Journal on Selected Areas in Communications, 17(6):1083-1092, June 1999
  20 *
 
  21 * Code from fib_hash has been reused which includes the following header:
  22 *
 
  23 * INET		An implementation of the TCP/IP protocol suite for the LINUX
  24 *		operating system.  INET is implemented using the  BSD Socket
  25 *		interface as the means of communication with the user level.
  26 *
  27 *		IPv4 FIB: lookup engine and maintenance routines.
  28 *
 
  29 * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  30 *
 
 
 
 
 
  31 * Substantial contributions to this work comes from:
  32 *
  33 *		David S. Miller, <davem@davemloft.net>
  34 *		Stephen Hemminger <shemminger@osdl.org>
  35 *		Paul E. McKenney <paulmck@us.ibm.com>
  36 *		Patrick McHardy <kaber@trash.net>
  37 */
  38#include <linux/cache.h>
 
 
  39#include <linux/uaccess.h>
  40#include <linux/bitops.h>
  41#include <linux/types.h>
  42#include <linux/kernel.h>
  43#include <linux/mm.h>
  44#include <linux/string.h>
  45#include <linux/socket.h>
  46#include <linux/sockios.h>
  47#include <linux/errno.h>
  48#include <linux/in.h>
  49#include <linux/inet.h>
  50#include <linux/inetdevice.h>
  51#include <linux/netdevice.h>
  52#include <linux/if_arp.h>
  53#include <linux/proc_fs.h>
  54#include <linux/rcupdate.h>
  55#include <linux/rcupdate_wait.h>
  56#include <linux/skbuff.h>
  57#include <linux/netlink.h>
  58#include <linux/init.h>
  59#include <linux/list.h>
  60#include <linux/slab.h>
  61#include <linux/export.h>
  62#include <linux/vmalloc.h>
  63#include <linux/notifier.h>
  64#include <net/net_namespace.h>
  65#include <net/inet_dscp.h>
  66#include <net/ip.h>
  67#include <net/protocol.h>
  68#include <net/route.h>
  69#include <net/tcp.h>
  70#include <net/sock.h>
  71#include <net/ip_fib.h>
  72#include <net/fib_notifier.h>
  73#include <trace/events/fib.h>
  74#include "fib_lookup.h"
  75
  76static int call_fib_entry_notifier(struct notifier_block *nb,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  77				   enum fib_event_type event_type, u32 dst,
  78				   int dst_len, struct fib_alias *fa,
  79				   struct netlink_ext_ack *extack)
  80{
  81	struct fib_entry_notifier_info info = {
  82		.info.extack = extack,
  83		.dst = dst,
  84		.dst_len = dst_len,
  85		.fi = fa->fa_info,
  86		.dscp = fa->fa_dscp,
  87		.type = fa->fa_type,
  88		.tb_id = fa->tb_id,
 
  89	};
  90	return call_fib4_notifier(nb, event_type, &info.info);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  91}
  92
  93static int call_fib_entry_notifiers(struct net *net,
  94				    enum fib_event_type event_type, u32 dst,
  95				    int dst_len, struct fib_alias *fa,
  96				    struct netlink_ext_ack *extack)
  97{
  98	struct fib_entry_notifier_info info = {
  99		.info.extack = extack,
 100		.dst = dst,
 101		.dst_len = dst_len,
 102		.fi = fa->fa_info,
 103		.dscp = fa->fa_dscp,
 104		.type = fa->fa_type,
 105		.tb_id = fa->tb_id,
 
 106	};
 107	return call_fib4_notifiers(net, event_type, &info.info);
 108}
 109
 110#define MAX_STAT_DEPTH 32
 111
 112#define KEYLENGTH	(8*sizeof(t_key))
 113#define KEY_MAX		((t_key)~0)
 114
 115typedef unsigned int t_key;
 116
 117#define IS_TRIE(n)	((n)->pos >= KEYLENGTH)
 118#define IS_TNODE(n)	((n)->bits)
 119#define IS_LEAF(n)	(!(n)->bits)
 120
 121struct key_vector {
 122	t_key key;
 123	unsigned char pos;		/* 2log(KEYLENGTH) bits needed */
 124	unsigned char bits;		/* 2log(KEYLENGTH) bits needed */
 125	unsigned char slen;
 126	union {
 127		/* This list pointer if valid if (pos | bits) == 0 (LEAF) */
 128		struct hlist_head leaf;
 129		/* This array is valid if (pos | bits) > 0 (TNODE) */
 130		DECLARE_FLEX_ARRAY(struct key_vector __rcu *, tnode);
 131	};
 132};
 133
 134struct tnode {
 135	struct rcu_head rcu;
 136	t_key empty_children;		/* KEYLENGTH bits needed */
 137	t_key full_children;		/* KEYLENGTH bits needed */
 138	struct key_vector __rcu *parent;
 139	struct key_vector kv[1];
 140#define tn_bits kv[0].bits
 141};
 142
 143#define TNODE_SIZE(n)	offsetof(struct tnode, kv[0].tnode[n])
 144#define LEAF_SIZE	TNODE_SIZE(1)
 145
 146#ifdef CONFIG_IP_FIB_TRIE_STATS
 147struct trie_use_stats {
 148	unsigned int gets;
 149	unsigned int backtrack;
 150	unsigned int semantic_match_passed;
 151	unsigned int semantic_match_miss;
 152	unsigned int null_node_hit;
 153	unsigned int resize_node_skipped;
 154};
 155#endif
 156
 157struct trie_stat {
 158	unsigned int totdepth;
 159	unsigned int maxdepth;
 160	unsigned int tnodes;
 161	unsigned int leaves;
 162	unsigned int nullpointers;
 163	unsigned int prefixes;
 164	unsigned int nodesizes[MAX_STAT_DEPTH];
 165};
 166
 167struct trie {
 168	struct key_vector kv[1];
 169#ifdef CONFIG_IP_FIB_TRIE_STATS
 170	struct trie_use_stats __percpu *stats;
 171#endif
 172};
 173
 174static struct key_vector *resize(struct trie *t, struct key_vector *tn);
 175static unsigned int tnode_free_size;
 176
 177/*
 178 * synchronize_rcu after call_rcu for outstanding dirty memory; it should be
 179 * especially useful before resizing the root node with PREEMPT_NONE configs;
 180 * the value was obtained experimentally, aiming to avoid visible slowdown.
 181 */
 182unsigned int sysctl_fib_sync_mem = 512 * 1024;
 183unsigned int sysctl_fib_sync_mem_min = 64 * 1024;
 184unsigned int sysctl_fib_sync_mem_max = 64 * 1024 * 1024;
 185
 186static struct kmem_cache *fn_alias_kmem __ro_after_init;
 187static struct kmem_cache *trie_leaf_kmem __ro_after_init;
 188
 189static inline struct tnode *tn_info(struct key_vector *kv)
 190{
 191	return container_of(kv, struct tnode, kv[0]);
 192}
 193
 194/* caller must hold RTNL */
 195#define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
 196#define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
 197
 198/* caller must hold RCU read lock or RTNL */
 199#define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
 200#define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
 201
 202/* wrapper for rcu_assign_pointer */
 203static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
 204{
 205	if (n)
 206		rcu_assign_pointer(tn_info(n)->parent, tp);
 207}
 208
 209#define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
 210
 211/* This provides us with the number of children in this node, in the case of a
 212 * leaf this will return 0 meaning none of the children are accessible.
 213 */
 214static inline unsigned long child_length(const struct key_vector *tn)
 215{
 216	return (1ul << tn->bits) & ~(1ul);
 217}
 218
 219#define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
 220
 221static inline unsigned long get_index(t_key key, struct key_vector *kv)
 222{
 223	unsigned long index = key ^ kv->key;
 224
 225	if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
 226		return 0;
 227
 228	return index >> kv->pos;
 229}
 230
 231/* To understand this stuff, an understanding of keys and all their bits is
 232 * necessary. Every node in the trie has a key associated with it, but not
 233 * all of the bits in that key are significant.
 234 *
 235 * Consider a node 'n' and its parent 'tp'.
 236 *
 237 * If n is a leaf, every bit in its key is significant. Its presence is
 238 * necessitated by path compression, since during a tree traversal (when
 239 * searching for a leaf - unless we are doing an insertion) we will completely
 240 * ignore all skipped bits we encounter. Thus we need to verify, at the end of
 241 * a potentially successful search, that we have indeed been walking the
 242 * correct key path.
 243 *
 244 * Note that we can never "miss" the correct key in the tree if present by
 245 * following the wrong path. Path compression ensures that segments of the key
 246 * that are the same for all keys with a given prefix are skipped, but the
 247 * skipped part *is* identical for each node in the subtrie below the skipped
 248 * bit! trie_insert() in this implementation takes care of that.
 249 *
 250 * if n is an internal node - a 'tnode' here, the various parts of its key
 251 * have many different meanings.
 252 *
 253 * Example:
 254 * _________________________________________________________________
 255 * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
 256 * -----------------------------------------------------------------
 257 *  31  30  29  28  27  26  25  24  23  22  21  20  19  18  17  16
 258 *
 259 * _________________________________________________________________
 260 * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
 261 * -----------------------------------------------------------------
 262 *  15  14  13  12  11  10   9   8   7   6   5   4   3   2   1   0
 263 *
 264 * tp->pos = 22
 265 * tp->bits = 3
 266 * n->pos = 13
 267 * n->bits = 4
 268 *
 269 * First, let's just ignore the bits that come before the parent tp, that is
 270 * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
 271 * point we do not use them for anything.
 272 *
 273 * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
 274 * index into the parent's child array. That is, they will be used to find
 275 * 'n' among tp's children.
 276 *
 277 * The bits from (n->pos + n->bits) to (tp->pos - 1) - "S" - are skipped bits
 278 * for the node n.
 279 *
 280 * All the bits we have seen so far are significant to the node n. The rest
 281 * of the bits are really not needed or indeed known in n->key.
 282 *
 283 * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
 284 * n's child array, and will of course be different for each child.
 285 *
 286 * The rest of the bits, from 0 to (n->pos -1) - "u" - are completely unknown
 287 * at this point.
 288 */
 289
 290static const int halve_threshold = 25;
 291static const int inflate_threshold = 50;
 292static const int halve_threshold_root = 15;
 293static const int inflate_threshold_root = 30;
 294
 
 
 
 
 
 
 295static inline void alias_free_mem_rcu(struct fib_alias *fa)
 296{
 297	kfree_rcu(fa, rcu);
 298}
 299
 
 
 300#define TNODE_VMALLOC_MAX \
 301	ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
 302
 303static void __node_free_rcu(struct rcu_head *head)
 304{
 305	struct tnode *n = container_of(head, struct tnode, rcu);
 306
 307	if (!n->tn_bits)
 308		kmem_cache_free(trie_leaf_kmem, n);
 309	else
 310		kvfree(n);
 311}
 312
 313#define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
 314
 315static struct tnode *tnode_alloc(int bits)
 316{
 317	size_t size;
 318
 319	/* verify bits is within bounds */
 320	if (bits > TNODE_VMALLOC_MAX)
 321		return NULL;
 322
 323	/* determine size and verify it is non-zero and didn't overflow */
 324	size = TNODE_SIZE(1ul << bits);
 325
 326	if (size <= PAGE_SIZE)
 327		return kzalloc(size, GFP_KERNEL);
 328	else
 329		return vzalloc(size);
 330}
 331
 332static inline void empty_child_inc(struct key_vector *n)
 333{
 334	tn_info(n)->empty_children++;
 335
 336	if (!tn_info(n)->empty_children)
 337		tn_info(n)->full_children++;
 338}
 339
 340static inline void empty_child_dec(struct key_vector *n)
 341{
 342	if (!tn_info(n)->empty_children)
 343		tn_info(n)->full_children--;
 344
 345	tn_info(n)->empty_children--;
 346}
 347
 348static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
 349{
 350	struct key_vector *l;
 351	struct tnode *kv;
 352
 353	kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
 354	if (!kv)
 355		return NULL;
 356
 357	/* initialize key vector */
 358	l = kv->kv;
 359	l->key = key;
 360	l->pos = 0;
 361	l->bits = 0;
 362	l->slen = fa->fa_slen;
 363
 364	/* link leaf to fib alias */
 365	INIT_HLIST_HEAD(&l->leaf);
 366	hlist_add_head(&fa->fa_list, &l->leaf);
 367
 368	return l;
 369}
 370
 371static struct key_vector *tnode_new(t_key key, int pos, int bits)
 372{
 373	unsigned int shift = pos + bits;
 374	struct key_vector *tn;
 375	struct tnode *tnode;
 376
 377	/* verify bits and pos their msb bits clear and values are valid */
 378	BUG_ON(!bits || (shift > KEYLENGTH));
 379
 380	tnode = tnode_alloc(bits);
 381	if (!tnode)
 382		return NULL;
 383
 384	pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
 385		 sizeof(struct key_vector *) << bits);
 386
 387	if (bits == KEYLENGTH)
 388		tnode->full_children = 1;
 389	else
 390		tnode->empty_children = 1ul << bits;
 391
 392	tn = tnode->kv;
 393	tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
 394	tn->pos = pos;
 395	tn->bits = bits;
 396	tn->slen = pos;
 397
 398	return tn;
 399}
 400
 401/* Check whether a tnode 'n' is "full", i.e. it is an internal node
 402 * and no bits are skipped. See discussion in dyntree paper p. 6
 403 */
 404static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
 405{
 406	return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
 407}
 408
 409/* Add a child at position i overwriting the old value.
 410 * Update the value of full_children and empty_children.
 411 */
 412static void put_child(struct key_vector *tn, unsigned long i,
 413		      struct key_vector *n)
 414{
 415	struct key_vector *chi = get_child(tn, i);
 416	int isfull, wasfull;
 417
 418	BUG_ON(i >= child_length(tn));
 419
 420	/* update emptyChildren, overflow into fullChildren */
 421	if (!n && chi)
 422		empty_child_inc(tn);
 423	if (n && !chi)
 424		empty_child_dec(tn);
 425
 426	/* update fullChildren */
 427	wasfull = tnode_full(tn, chi);
 428	isfull = tnode_full(tn, n);
 429
 430	if (wasfull && !isfull)
 431		tn_info(tn)->full_children--;
 432	else if (!wasfull && isfull)
 433		tn_info(tn)->full_children++;
 434
 435	if (n && (tn->slen < n->slen))
 436		tn->slen = n->slen;
 437
 438	rcu_assign_pointer(tn->tnode[i], n);
 439}
 440
 441static void update_children(struct key_vector *tn)
 442{
 443	unsigned long i;
 444
 445	/* update all of the child parent pointers */
 446	for (i = child_length(tn); i;) {
 447		struct key_vector *inode = get_child(tn, --i);
 448
 449		if (!inode)
 450			continue;
 451
 452		/* Either update the children of a tnode that
 453		 * already belongs to us or update the child
 454		 * to point to ourselves.
 455		 */
 456		if (node_parent(inode) == tn)
 457			update_children(inode);
 458		else
 459			node_set_parent(inode, tn);
 460	}
 461}
 462
 463static inline void put_child_root(struct key_vector *tp, t_key key,
 464				  struct key_vector *n)
 465{
 466	if (IS_TRIE(tp))
 467		rcu_assign_pointer(tp->tnode[0], n);
 468	else
 469		put_child(tp, get_index(key, tp), n);
 470}
 471
 472static inline void tnode_free_init(struct key_vector *tn)
 473{
 474	tn_info(tn)->rcu.next = NULL;
 475}
 476
 477static inline void tnode_free_append(struct key_vector *tn,
 478				     struct key_vector *n)
 479{
 480	tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
 481	tn_info(tn)->rcu.next = &tn_info(n)->rcu;
 482}
 483
 484static void tnode_free(struct key_vector *tn)
 485{
 486	struct callback_head *head = &tn_info(tn)->rcu;
 487
 488	while (head) {
 489		head = head->next;
 490		tnode_free_size += TNODE_SIZE(1ul << tn->bits);
 491		node_free(tn);
 492
 493		tn = container_of(head, struct tnode, rcu)->kv;
 494	}
 495
 496	if (tnode_free_size >= READ_ONCE(sysctl_fib_sync_mem)) {
 497		tnode_free_size = 0;
 498		synchronize_net();
 499	}
 500}
 501
 502static struct key_vector *replace(struct trie *t,
 503				  struct key_vector *oldtnode,
 504				  struct key_vector *tn)
 505{
 506	struct key_vector *tp = node_parent(oldtnode);
 507	unsigned long i;
 508
 509	/* setup the parent pointer out of and back into this node */
 510	NODE_INIT_PARENT(tn, tp);
 511	put_child_root(tp, tn->key, tn);
 512
 513	/* update all of the child parent pointers */
 514	update_children(tn);
 515
 516	/* all pointers should be clean so we are done */
 517	tnode_free(oldtnode);
 518
 519	/* resize children now that oldtnode is freed */
 520	for (i = child_length(tn); i;) {
 521		struct key_vector *inode = get_child(tn, --i);
 522
 523		/* resize child node */
 524		if (tnode_full(tn, inode))
 525			tn = resize(t, inode);
 526	}
 527
 528	return tp;
 529}
 530
 531static struct key_vector *inflate(struct trie *t,
 532				  struct key_vector *oldtnode)
 533{
 534	struct key_vector *tn;
 535	unsigned long i;
 536	t_key m;
 537
 538	pr_debug("In inflate\n");
 539
 540	tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
 541	if (!tn)
 542		goto notnode;
 543
 544	/* prepare oldtnode to be freed */
 545	tnode_free_init(oldtnode);
 546
 547	/* Assemble all of the pointers in our cluster, in this case that
 548	 * represents all of the pointers out of our allocated nodes that
 549	 * point to existing tnodes and the links between our allocated
 550	 * nodes.
 551	 */
 552	for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
 553		struct key_vector *inode = get_child(oldtnode, --i);
 554		struct key_vector *node0, *node1;
 555		unsigned long j, k;
 556
 557		/* An empty child */
 558		if (!inode)
 559			continue;
 560
 561		/* A leaf or an internal node with skipped bits */
 562		if (!tnode_full(oldtnode, inode)) {
 563			put_child(tn, get_index(inode->key, tn), inode);
 564			continue;
 565		}
 566
 567		/* drop the node in the old tnode free list */
 568		tnode_free_append(oldtnode, inode);
 569
 570		/* An internal node with two children */
 571		if (inode->bits == 1) {
 572			put_child(tn, 2 * i + 1, get_child(inode, 1));
 573			put_child(tn, 2 * i, get_child(inode, 0));
 574			continue;
 575		}
 576
 577		/* We will replace this node 'inode' with two new
 578		 * ones, 'node0' and 'node1', each with half of the
 579		 * original children. The two new nodes will have
 580		 * a position one bit further down the key and this
 581		 * means that the "significant" part of their keys
 582		 * (see the discussion near the top of this file)
 583		 * will differ by one bit, which will be "0" in
 584		 * node0's key and "1" in node1's key. Since we are
 585		 * moving the key position by one step, the bit that
 586		 * we are moving away from - the bit at position
 587		 * (tn->pos) - is the one that will differ between
 588		 * node0 and node1. So... we synthesize that bit in the
 589		 * two new keys.
 590		 */
 591		node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
 592		if (!node1)
 593			goto nomem;
 594		node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
 595
 596		tnode_free_append(tn, node1);
 597		if (!node0)
 598			goto nomem;
 599		tnode_free_append(tn, node0);
 600
 601		/* populate child pointers in new nodes */
 602		for (k = child_length(inode), j = k / 2; j;) {
 603			put_child(node1, --j, get_child(inode, --k));
 604			put_child(node0, j, get_child(inode, j));
 605			put_child(node1, --j, get_child(inode, --k));
 606			put_child(node0, j, get_child(inode, j));
 607		}
 608
 609		/* link new nodes to parent */
 610		NODE_INIT_PARENT(node1, tn);
 611		NODE_INIT_PARENT(node0, tn);
 612
 613		/* link parent to nodes */
 614		put_child(tn, 2 * i + 1, node1);
 615		put_child(tn, 2 * i, node0);
 616	}
 617
 618	/* setup the parent pointers into and out of this node */
 619	return replace(t, oldtnode, tn);
 620nomem:
 621	/* all pointers should be clean so we are done */
 622	tnode_free(tn);
 623notnode:
 624	return NULL;
 625}
 626
 627static struct key_vector *halve(struct trie *t,
 628				struct key_vector *oldtnode)
 629{
 630	struct key_vector *tn;
 631	unsigned long i;
 632
 633	pr_debug("In halve\n");
 634
 635	tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
 636	if (!tn)
 637		goto notnode;
 638
 639	/* prepare oldtnode to be freed */
 640	tnode_free_init(oldtnode);
 641
 642	/* Assemble all of the pointers in our cluster, in this case that
 643	 * represents all of the pointers out of our allocated nodes that
 644	 * point to existing tnodes and the links between our allocated
 645	 * nodes.
 646	 */
 647	for (i = child_length(oldtnode); i;) {
 648		struct key_vector *node1 = get_child(oldtnode, --i);
 649		struct key_vector *node0 = get_child(oldtnode, --i);
 650		struct key_vector *inode;
 651
 652		/* At least one of the children is empty */
 653		if (!node1 || !node0) {
 654			put_child(tn, i / 2, node1 ? : node0);
 655			continue;
 656		}
 657
 658		/* Two nonempty children */
 659		inode = tnode_new(node0->key, oldtnode->pos, 1);
 660		if (!inode)
 661			goto nomem;
 662		tnode_free_append(tn, inode);
 663
 664		/* initialize pointers out of node */
 665		put_child(inode, 1, node1);
 666		put_child(inode, 0, node0);
 667		NODE_INIT_PARENT(inode, tn);
 668
 669		/* link parent to node */
 670		put_child(tn, i / 2, inode);
 671	}
 672
 673	/* setup the parent pointers into and out of this node */
 674	return replace(t, oldtnode, tn);
 675nomem:
 676	/* all pointers should be clean so we are done */
 677	tnode_free(tn);
 678notnode:
 679	return NULL;
 680}
 681
 682static struct key_vector *collapse(struct trie *t,
 683				   struct key_vector *oldtnode)
 684{
 685	struct key_vector *n, *tp;
 686	unsigned long i;
 687
 688	/* scan the tnode looking for that one child that might still exist */
 689	for (n = NULL, i = child_length(oldtnode); !n && i;)
 690		n = get_child(oldtnode, --i);
 691
 692	/* compress one level */
 693	tp = node_parent(oldtnode);
 694	put_child_root(tp, oldtnode->key, n);
 695	node_set_parent(n, tp);
 696
 697	/* drop dead node */
 698	node_free(oldtnode);
 699
 700	return tp;
 701}
 702
 703static unsigned char update_suffix(struct key_vector *tn)
 704{
 705	unsigned char slen = tn->pos;
 706	unsigned long stride, i;
 707	unsigned char slen_max;
 708
 709	/* only vector 0 can have a suffix length greater than or equal to
 710	 * tn->pos + tn->bits, the second highest node will have a suffix
 711	 * length at most of tn->pos + tn->bits - 1
 712	 */
 713	slen_max = min_t(unsigned char, tn->pos + tn->bits - 1, tn->slen);
 714
 715	/* search though the list of children looking for nodes that might
 716	 * have a suffix greater than the one we currently have.  This is
 717	 * why we start with a stride of 2 since a stride of 1 would
 718	 * represent the nodes with suffix length equal to tn->pos
 719	 */
 720	for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
 721		struct key_vector *n = get_child(tn, i);
 722
 723		if (!n || (n->slen <= slen))
 724			continue;
 725
 726		/* update stride and slen based on new value */
 727		stride <<= (n->slen - slen);
 728		slen = n->slen;
 729		i &= ~(stride - 1);
 730
 731		/* stop searching if we have hit the maximum possible value */
 732		if (slen >= slen_max)
 733			break;
 734	}
 735
 736	tn->slen = slen;
 737
 738	return slen;
 739}
 740
 741/* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
 742 * the Helsinki University of Technology and Matti Tikkanen of Nokia
 743 * Telecommunications, page 6:
 744 * "A node is doubled if the ratio of non-empty children to all
 745 * children in the *doubled* node is at least 'high'."
 746 *
 747 * 'high' in this instance is the variable 'inflate_threshold'. It
 748 * is expressed as a percentage, so we multiply it with
 749 * child_length() and instead of multiplying by 2 (since the
 750 * child array will be doubled by inflate()) and multiplying
 751 * the left-hand side by 100 (to handle the percentage thing) we
 752 * multiply the left-hand side by 50.
 753 *
 754 * The left-hand side may look a bit weird: child_length(tn)
 755 * - tn->empty_children is of course the number of non-null children
 756 * in the current node. tn->full_children is the number of "full"
 757 * children, that is non-null tnodes with a skip value of 0.
 758 * All of those will be doubled in the resulting inflated tnode, so
 759 * we just count them one extra time here.
 760 *
 761 * A clearer way to write this would be:
 762 *
 763 * to_be_doubled = tn->full_children;
 764 * not_to_be_doubled = child_length(tn) - tn->empty_children -
 765 *     tn->full_children;
 766 *
 767 * new_child_length = child_length(tn) * 2;
 768 *
 769 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
 770 *      new_child_length;
 771 * if (new_fill_factor >= inflate_threshold)
 772 *
 773 * ...and so on, tho it would mess up the while () loop.
 774 *
 775 * anyway,
 776 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
 777 *      inflate_threshold
 778 *
 779 * avoid a division:
 780 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
 781 *      inflate_threshold * new_child_length
 782 *
 783 * expand not_to_be_doubled and to_be_doubled, and shorten:
 784 * 100 * (child_length(tn) - tn->empty_children +
 785 *    tn->full_children) >= inflate_threshold * new_child_length
 786 *
 787 * expand new_child_length:
 788 * 100 * (child_length(tn) - tn->empty_children +
 789 *    tn->full_children) >=
 790 *      inflate_threshold * child_length(tn) * 2
 791 *
 792 * shorten again:
 793 * 50 * (tn->full_children + child_length(tn) -
 794 *    tn->empty_children) >= inflate_threshold *
 795 *    child_length(tn)
 796 *
 797 */
 798static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
 799{
 800	unsigned long used = child_length(tn);
 801	unsigned long threshold = used;
 802
 803	/* Keep root node larger */
 804	threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
 805	used -= tn_info(tn)->empty_children;
 806	used += tn_info(tn)->full_children;
 807
 808	/* if bits == KEYLENGTH then pos = 0, and will fail below */
 809
 810	return (used > 1) && tn->pos && ((50 * used) >= threshold);
 811}
 812
 813static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
 814{
 815	unsigned long used = child_length(tn);
 816	unsigned long threshold = used;
 817
 818	/* Keep root node larger */
 819	threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
 820	used -= tn_info(tn)->empty_children;
 821
 822	/* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
 823
 824	return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
 825}
 826
 827static inline bool should_collapse(struct key_vector *tn)
 828{
 829	unsigned long used = child_length(tn);
 830
 831	used -= tn_info(tn)->empty_children;
 832
 833	/* account for bits == KEYLENGTH case */
 834	if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
 835		used -= KEY_MAX;
 836
 837	/* One child or none, time to drop us from the trie */
 838	return used < 2;
 839}
 840
 841#define MAX_WORK 10
 842static struct key_vector *resize(struct trie *t, struct key_vector *tn)
 843{
 844#ifdef CONFIG_IP_FIB_TRIE_STATS
 845	struct trie_use_stats __percpu *stats = t->stats;
 846#endif
 847	struct key_vector *tp = node_parent(tn);
 848	unsigned long cindex = get_index(tn->key, tp);
 849	int max_work = MAX_WORK;
 850
 851	pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
 852		 tn, inflate_threshold, halve_threshold);
 853
 854	/* track the tnode via the pointer from the parent instead of
 855	 * doing it ourselves.  This way we can let RCU fully do its
 856	 * thing without us interfering
 857	 */
 858	BUG_ON(tn != get_child(tp, cindex));
 859
 860	/* Double as long as the resulting node has a number of
 861	 * nonempty nodes that are above the threshold.
 862	 */
 863	while (should_inflate(tp, tn) && max_work) {
 864		tp = inflate(t, tn);
 865		if (!tp) {
 866#ifdef CONFIG_IP_FIB_TRIE_STATS
 867			this_cpu_inc(stats->resize_node_skipped);
 868#endif
 869			break;
 870		}
 871
 872		max_work--;
 873		tn = get_child(tp, cindex);
 874	}
 875
 876	/* update parent in case inflate failed */
 877	tp = node_parent(tn);
 878
 879	/* Return if at least one inflate is run */
 880	if (max_work != MAX_WORK)
 881		return tp;
 882
 883	/* Halve as long as the number of empty children in this
 884	 * node is above threshold.
 885	 */
 886	while (should_halve(tp, tn) && max_work) {
 887		tp = halve(t, tn);
 888		if (!tp) {
 889#ifdef CONFIG_IP_FIB_TRIE_STATS
 890			this_cpu_inc(stats->resize_node_skipped);
 891#endif
 892			break;
 893		}
 894
 895		max_work--;
 896		tn = get_child(tp, cindex);
 897	}
 898
 899	/* Only one child remains */
 900	if (should_collapse(tn))
 901		return collapse(t, tn);
 902
 903	/* update parent in case halve failed */
 904	return node_parent(tn);
 905}
 906
 907static void node_pull_suffix(struct key_vector *tn, unsigned char slen)
 908{
 909	unsigned char node_slen = tn->slen;
 910
 911	while ((node_slen > tn->pos) && (node_slen > slen)) {
 912		slen = update_suffix(tn);
 913		if (node_slen == slen)
 914			break;
 915
 916		tn = node_parent(tn);
 917		node_slen = tn->slen;
 918	}
 919}
 920
 921static void node_push_suffix(struct key_vector *tn, unsigned char slen)
 922{
 923	while (tn->slen < slen) {
 924		tn->slen = slen;
 925		tn = node_parent(tn);
 926	}
 927}
 928
 929/* rcu_read_lock needs to be hold by caller from readside */
 930static struct key_vector *fib_find_node(struct trie *t,
 931					struct key_vector **tp, u32 key)
 932{
 933	struct key_vector *pn, *n = t->kv;
 934	unsigned long index = 0;
 935
 936	do {
 937		pn = n;
 938		n = get_child_rcu(n, index);
 939
 940		if (!n)
 941			break;
 942
 943		index = get_cindex(key, n);
 944
 945		/* This bit of code is a bit tricky but it combines multiple
 946		 * checks into a single check.  The prefix consists of the
 947		 * prefix plus zeros for the bits in the cindex. The index
 948		 * is the difference between the key and this value.  From
 949		 * this we can actually derive several pieces of data.
 950		 *   if (index >= (1ul << bits))
 951		 *     we have a mismatch in skip bits and failed
 952		 *   else
 953		 *     we know the value is cindex
 954		 *
 955		 * This check is safe even if bits == KEYLENGTH due to the
 956		 * fact that we can only allocate a node with 32 bits if a
 957		 * long is greater than 32 bits.
 958		 */
 959		if (index >= (1ul << n->bits)) {
 960			n = NULL;
 961			break;
 962		}
 963
 964		/* keep searching until we find a perfect match leaf or NULL */
 965	} while (IS_TNODE(n));
 966
 967	*tp = pn;
 968
 969	return n;
 970}
 971
 972/* Return the first fib alias matching DSCP with
 973 * priority less than or equal to PRIO.
 974 * If 'find_first' is set, return the first matching
 975 * fib alias, regardless of DSCP and priority.
 976 */
 977static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
 978					dscp_t dscp, u32 prio, u32 tb_id,
 979					bool find_first)
 980{
 981	struct fib_alias *fa;
 982
 983	if (!fah)
 984		return NULL;
 985
 986	hlist_for_each_entry(fa, fah, fa_list) {
 987		/* Avoid Sparse warning when using dscp_t in inequalities */
 988		u8 __fa_dscp = inet_dscp_to_dsfield(fa->fa_dscp);
 989		u8 __dscp = inet_dscp_to_dsfield(dscp);
 990
 991		if (fa->fa_slen < slen)
 992			continue;
 993		if (fa->fa_slen != slen)
 994			break;
 995		if (fa->tb_id > tb_id)
 996			continue;
 997		if (fa->tb_id != tb_id)
 998			break;
 999		if (find_first)
1000			return fa;
1001		if (__fa_dscp > __dscp)
1002			continue;
1003		if (fa->fa_info->fib_priority >= prio || __fa_dscp < __dscp)
1004			return fa;
1005	}
1006
1007	return NULL;
1008}
1009
1010static struct fib_alias *
1011fib_find_matching_alias(struct net *net, const struct fib_rt_info *fri)
1012{
1013	u8 slen = KEYLENGTH - fri->dst_len;
1014	struct key_vector *l, *tp;
1015	struct fib_table *tb;
1016	struct fib_alias *fa;
1017	struct trie *t;
1018
1019	tb = fib_get_table(net, fri->tb_id);
1020	if (!tb)
1021		return NULL;
1022
1023	t = (struct trie *)tb->tb_data;
1024	l = fib_find_node(t, &tp, be32_to_cpu(fri->dst));
1025	if (!l)
1026		return NULL;
1027
1028	hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1029		if (fa->fa_slen == slen && fa->tb_id == fri->tb_id &&
1030		    fa->fa_dscp == fri->dscp && fa->fa_info == fri->fi &&
1031		    fa->fa_type == fri->type)
1032			return fa;
1033	}
1034
1035	return NULL;
1036}
1037
1038void fib_alias_hw_flags_set(struct net *net, const struct fib_rt_info *fri)
1039{
1040	u8 fib_notify_on_flag_change;
1041	struct fib_alias *fa_match;
1042	struct sk_buff *skb;
1043	int err;
1044
1045	rcu_read_lock();
1046
1047	fa_match = fib_find_matching_alias(net, fri);
1048	if (!fa_match)
1049		goto out;
1050
1051	/* These are paired with the WRITE_ONCE() happening in this function.
1052	 * The reason is that we are only protected by RCU at this point.
1053	 */
1054	if (READ_ONCE(fa_match->offload) == fri->offload &&
1055	    READ_ONCE(fa_match->trap) == fri->trap &&
1056	    READ_ONCE(fa_match->offload_failed) == fri->offload_failed)
1057		goto out;
1058
1059	WRITE_ONCE(fa_match->offload, fri->offload);
1060	WRITE_ONCE(fa_match->trap, fri->trap);
1061
1062	fib_notify_on_flag_change = READ_ONCE(net->ipv4.sysctl_fib_notify_on_flag_change);
1063
1064	/* 2 means send notifications only if offload_failed was changed. */
1065	if (fib_notify_on_flag_change == 2 &&
1066	    READ_ONCE(fa_match->offload_failed) == fri->offload_failed)
1067		goto out;
1068
1069	WRITE_ONCE(fa_match->offload_failed, fri->offload_failed);
1070
1071	if (!fib_notify_on_flag_change)
1072		goto out;
1073
1074	skb = nlmsg_new(fib_nlmsg_size(fa_match->fa_info), GFP_ATOMIC);
1075	if (!skb) {
1076		err = -ENOBUFS;
1077		goto errout;
1078	}
1079
1080	err = fib_dump_info(skb, 0, 0, RTM_NEWROUTE, fri, 0);
1081	if (err < 0) {
1082		/* -EMSGSIZE implies BUG in fib_nlmsg_size() */
1083		WARN_ON(err == -EMSGSIZE);
1084		kfree_skb(skb);
1085		goto errout;
1086	}
1087
1088	rtnl_notify(skb, net, 0, RTNLGRP_IPV4_ROUTE, NULL, GFP_ATOMIC);
1089	goto out;
1090
1091errout:
1092	rtnl_set_sk_err(net, RTNLGRP_IPV4_ROUTE, err);
1093out:
1094	rcu_read_unlock();
1095}
1096EXPORT_SYMBOL_GPL(fib_alias_hw_flags_set);
1097
1098static void trie_rebalance(struct trie *t, struct key_vector *tn)
1099{
1100	while (!IS_TRIE(tn))
1101		tn = resize(t, tn);
1102}
1103
1104static int fib_insert_node(struct trie *t, struct key_vector *tp,
1105			   struct fib_alias *new, t_key key)
1106{
1107	struct key_vector *n, *l;
1108
1109	l = leaf_new(key, new);
1110	if (!l)
1111		goto noleaf;
1112
1113	/* retrieve child from parent node */
1114	n = get_child(tp, get_index(key, tp));
1115
1116	/* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1117	 *
1118	 *  Add a new tnode here
1119	 *  first tnode need some special handling
1120	 *  leaves us in position for handling as case 3
1121	 */
1122	if (n) {
1123		struct key_vector *tn;
1124
1125		tn = tnode_new(key, __fls(key ^ n->key), 1);
1126		if (!tn)
1127			goto notnode;
1128
1129		/* initialize routes out of node */
1130		NODE_INIT_PARENT(tn, tp);
1131		put_child(tn, get_index(key, tn) ^ 1, n);
1132
1133		/* start adding routes into the node */
1134		put_child_root(tp, key, tn);
1135		node_set_parent(n, tn);
1136
1137		/* parent now has a NULL spot where the leaf can go */
1138		tp = tn;
1139	}
1140
1141	/* Case 3: n is NULL, and will just insert a new leaf */
1142	node_push_suffix(tp, new->fa_slen);
1143	NODE_INIT_PARENT(l, tp);
1144	put_child_root(tp, key, l);
1145	trie_rebalance(t, tp);
1146
1147	return 0;
1148notnode:
1149	node_free(l);
1150noleaf:
1151	return -ENOMEM;
1152}
1153
1154static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1155			    struct key_vector *l, struct fib_alias *new,
1156			    struct fib_alias *fa, t_key key)
1157{
1158	if (!l)
1159		return fib_insert_node(t, tp, new, key);
1160
1161	if (fa) {
1162		hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1163	} else {
1164		struct fib_alias *last;
1165
1166		hlist_for_each_entry(last, &l->leaf, fa_list) {
1167			if (new->fa_slen < last->fa_slen)
1168				break;
1169			if ((new->fa_slen == last->fa_slen) &&
1170			    (new->tb_id > last->tb_id))
1171				break;
1172			fa = last;
1173		}
1174
1175		if (fa)
1176			hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1177		else
1178			hlist_add_head_rcu(&new->fa_list, &l->leaf);
1179	}
1180
1181	/* if we added to the tail node then we need to update slen */
1182	if (l->slen < new->fa_slen) {
1183		l->slen = new->fa_slen;
1184		node_push_suffix(tp, new->fa_slen);
1185	}
1186
1187	return 0;
1188}
1189
1190static bool fib_valid_key_len(u32 key, u8 plen, struct netlink_ext_ack *extack)
1191{
1192	if (plen > KEYLENGTH) {
1193		NL_SET_ERR_MSG(extack, "Invalid prefix length");
1194		return false;
1195	}
1196
1197	if ((plen < KEYLENGTH) && (key << plen)) {
1198		NL_SET_ERR_MSG(extack,
1199			       "Invalid prefix for given prefix length");
1200		return false;
1201	}
1202
1203	return true;
1204}
1205
1206static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1207			     struct key_vector *l, struct fib_alias *old);
1208
1209/* Caller must hold RTNL. */
1210int fib_table_insert(struct net *net, struct fib_table *tb,
1211		     struct fib_config *cfg, struct netlink_ext_ack *extack)
1212{
1213	struct trie *t = (struct trie *)tb->tb_data;
1214	struct fib_alias *fa, *new_fa;
1215	struct key_vector *l, *tp;
1216	u16 nlflags = NLM_F_EXCL;
1217	struct fib_info *fi;
1218	u8 plen = cfg->fc_dst_len;
1219	u8 slen = KEYLENGTH - plen;
1220	dscp_t dscp;
1221	u32 key;
1222	int err;
1223
1224	key = ntohl(cfg->fc_dst);
1225
1226	if (!fib_valid_key_len(key, plen, extack))
1227		return -EINVAL;
1228
 
 
1229	pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1230
1231	fi = fib_create_info(cfg, extack);
 
 
 
1232	if (IS_ERR(fi)) {
1233		err = PTR_ERR(fi);
1234		goto err;
1235	}
1236
1237	dscp = cfg->fc_dscp;
1238	l = fib_find_node(t, &tp, key);
1239	fa = l ? fib_find_alias(&l->leaf, slen, dscp, fi->fib_priority,
1240				tb->tb_id, false) : NULL;
1241
1242	/* Now fa, if non-NULL, points to the first fib alias
1243	 * with the same keys [prefix,dscp,priority], if such key already
1244	 * exists or to the node before which we will insert new one.
1245	 *
1246	 * If fa is NULL, we will need to allocate a new one and
1247	 * insert to the tail of the section matching the suffix length
1248	 * of the new alias.
1249	 */
1250
1251	if (fa && fa->fa_dscp == dscp &&
1252	    fa->fa_info->fib_priority == fi->fib_priority) {
1253		struct fib_alias *fa_first, *fa_match;
1254
1255		err = -EEXIST;
1256		if (cfg->fc_nlflags & NLM_F_EXCL)
1257			goto out;
1258
1259		nlflags &= ~NLM_F_EXCL;
1260
1261		/* We have 2 goals:
1262		 * 1. Find exact match for type, scope, fib_info to avoid
1263		 * duplicate routes
1264		 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1265		 */
1266		fa_match = NULL;
1267		fa_first = fa;
1268		hlist_for_each_entry_from(fa, fa_list) {
1269			if ((fa->fa_slen != slen) ||
1270			    (fa->tb_id != tb->tb_id) ||
1271			    (fa->fa_dscp != dscp))
1272				break;
1273			if (fa->fa_info->fib_priority != fi->fib_priority)
1274				break;
1275			if (fa->fa_type == cfg->fc_type &&
1276			    fa->fa_info == fi) {
1277				fa_match = fa;
1278				break;
1279			}
1280		}
1281
1282		if (cfg->fc_nlflags & NLM_F_REPLACE) {
1283			struct fib_info *fi_drop;
1284			u8 state;
1285
1286			nlflags |= NLM_F_REPLACE;
1287			fa = fa_first;
1288			if (fa_match) {
1289				if (fa == fa_match)
1290					err = 0;
1291				goto out;
1292			}
1293			err = -ENOBUFS;
1294			new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1295			if (!new_fa)
1296				goto out;
1297
1298			fi_drop = fa->fa_info;
1299			new_fa->fa_dscp = fa->fa_dscp;
1300			new_fa->fa_info = fi;
1301			new_fa->fa_type = cfg->fc_type;
1302			state = fa->fa_state;
1303			new_fa->fa_state = state & ~FA_S_ACCESSED;
1304			new_fa->fa_slen = fa->fa_slen;
1305			new_fa->tb_id = tb->tb_id;
1306			new_fa->fa_default = -1;
1307			new_fa->offload = 0;
1308			new_fa->trap = 0;
1309			new_fa->offload_failed = 0;
1310
1311			hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1312
1313			if (fib_find_alias(&l->leaf, fa->fa_slen, 0, 0,
1314					   tb->tb_id, true) == new_fa) {
1315				enum fib_event_type fib_event;
1316
1317				fib_event = FIB_EVENT_ENTRY_REPLACE;
1318				err = call_fib_entry_notifiers(net, fib_event,
1319							       key, plen,
1320							       new_fa, extack);
1321				if (err) {
1322					hlist_replace_rcu(&new_fa->fa_list,
1323							  &fa->fa_list);
1324					goto out_free_new_fa;
1325				}
1326			}
1327
1328			rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1329				  tb->tb_id, &cfg->fc_nlinfo, nlflags);
1330
1331			alias_free_mem_rcu(fa);
1332
1333			fib_release_info(fi_drop);
1334			if (state & FA_S_ACCESSED)
1335				rt_cache_flush(cfg->fc_nlinfo.nl_net);
1336
 
 
 
 
 
 
 
1337			goto succeeded;
1338		}
1339		/* Error if we find a perfect match which
1340		 * uses the same scope, type, and nexthop
1341		 * information.
1342		 */
1343		if (fa_match)
1344			goto out;
1345
1346		if (cfg->fc_nlflags & NLM_F_APPEND)
1347			nlflags |= NLM_F_APPEND;
1348		else
1349			fa = fa_first;
1350	}
1351	err = -ENOENT;
1352	if (!(cfg->fc_nlflags & NLM_F_CREATE))
1353		goto out;
1354
1355	nlflags |= NLM_F_CREATE;
1356	err = -ENOBUFS;
1357	new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1358	if (!new_fa)
1359		goto out;
1360
1361	new_fa->fa_info = fi;
1362	new_fa->fa_dscp = dscp;
1363	new_fa->fa_type = cfg->fc_type;
1364	new_fa->fa_state = 0;
1365	new_fa->fa_slen = slen;
1366	new_fa->tb_id = tb->tb_id;
1367	new_fa->fa_default = -1;
1368	new_fa->offload = 0;
1369	new_fa->trap = 0;
1370	new_fa->offload_failed = 0;
1371
1372	/* Insert new entry to the list. */
1373	err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1374	if (err)
1375		goto out_free_new_fa;
1376
1377	/* The alias was already inserted, so the node must exist. */
1378	l = l ? l : fib_find_node(t, &tp, key);
1379	if (WARN_ON_ONCE(!l)) {
1380		err = -ENOENT;
1381		goto out_free_new_fa;
1382	}
1383
1384	if (fib_find_alias(&l->leaf, new_fa->fa_slen, 0, 0, tb->tb_id, true) ==
1385	    new_fa) {
1386		enum fib_event_type fib_event;
1387
1388		fib_event = FIB_EVENT_ENTRY_REPLACE;
1389		err = call_fib_entry_notifiers(net, fib_event, key, plen,
1390					       new_fa, extack);
1391		if (err)
1392			goto out_remove_new_fa;
1393	}
1394
1395	if (!plen)
1396		tb->tb_num_default++;
1397
1398	rt_cache_flush(cfg->fc_nlinfo.nl_net);
 
 
1399	rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1400		  &cfg->fc_nlinfo, nlflags);
1401succeeded:
1402	return 0;
1403
1404out_remove_new_fa:
1405	fib_remove_alias(t, tp, l, new_fa);
1406out_free_new_fa:
1407	kmem_cache_free(fn_alias_kmem, new_fa);
1408out:
1409	fib_release_info(fi);
1410err:
1411	return err;
1412}
1413
1414static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1415{
1416	t_key prefix = n->key;
1417
1418	return (key ^ prefix) & (prefix | -prefix);
1419}
1420
1421bool fib_lookup_good_nhc(const struct fib_nh_common *nhc, int fib_flags,
1422			 const struct flowi4 *flp)
1423{
1424	if (nhc->nhc_flags & RTNH_F_DEAD)
1425		return false;
1426
1427	if (ip_ignore_linkdown(nhc->nhc_dev) &&
1428	    nhc->nhc_flags & RTNH_F_LINKDOWN &&
1429	    !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1430		return false;
1431
1432	if (flp->flowi4_oif && flp->flowi4_oif != nhc->nhc_oif)
1433		return false;
1434
1435	return true;
1436}
1437
1438/* should be called with rcu_read_lock */
1439int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1440		     struct fib_result *res, int fib_flags)
1441{
1442	struct trie *t = (struct trie *) tb->tb_data;
1443#ifdef CONFIG_IP_FIB_TRIE_STATS
1444	struct trie_use_stats __percpu *stats = t->stats;
1445#endif
1446	const t_key key = ntohl(flp->daddr);
1447	struct key_vector *n, *pn;
1448	struct fib_alias *fa;
1449	unsigned long index;
1450	t_key cindex;
1451
 
 
1452	pn = t->kv;
1453	cindex = 0;
1454
1455	n = get_child_rcu(pn, cindex);
1456	if (!n) {
1457		trace_fib_table_lookup(tb->tb_id, flp, NULL, -EAGAIN);
1458		return -EAGAIN;
1459	}
1460
1461#ifdef CONFIG_IP_FIB_TRIE_STATS
1462	this_cpu_inc(stats->gets);
1463#endif
1464
1465	/* Step 1: Travel to the longest prefix match in the trie */
1466	for (;;) {
1467		index = get_cindex(key, n);
1468
1469		/* This bit of code is a bit tricky but it combines multiple
1470		 * checks into a single check.  The prefix consists of the
1471		 * prefix plus zeros for the "bits" in the prefix. The index
1472		 * is the difference between the key and this value.  From
1473		 * this we can actually derive several pieces of data.
1474		 *   if (index >= (1ul << bits))
1475		 *     we have a mismatch in skip bits and failed
1476		 *   else
1477		 *     we know the value is cindex
1478		 *
1479		 * This check is safe even if bits == KEYLENGTH due to the
1480		 * fact that we can only allocate a node with 32 bits if a
1481		 * long is greater than 32 bits.
1482		 */
1483		if (index >= (1ul << n->bits))
1484			break;
1485
1486		/* we have found a leaf. Prefixes have already been compared */
1487		if (IS_LEAF(n))
1488			goto found;
1489
1490		/* only record pn and cindex if we are going to be chopping
1491		 * bits later.  Otherwise we are just wasting cycles.
1492		 */
1493		if (n->slen > n->pos) {
1494			pn = n;
1495			cindex = index;
1496		}
1497
1498		n = get_child_rcu(n, index);
1499		if (unlikely(!n))
1500			goto backtrace;
1501	}
1502
1503	/* Step 2: Sort out leaves and begin backtracing for longest prefix */
1504	for (;;) {
1505		/* record the pointer where our next node pointer is stored */
1506		struct key_vector __rcu **cptr = n->tnode;
1507
1508		/* This test verifies that none of the bits that differ
1509		 * between the key and the prefix exist in the region of
1510		 * the lsb and higher in the prefix.
1511		 */
1512		if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1513			goto backtrace;
1514
1515		/* exit out and process leaf */
1516		if (unlikely(IS_LEAF(n)))
1517			break;
1518
1519		/* Don't bother recording parent info.  Since we are in
1520		 * prefix match mode we will have to come back to wherever
1521		 * we started this traversal anyway
1522		 */
1523
1524		while ((n = rcu_dereference(*cptr)) == NULL) {
1525backtrace:
1526#ifdef CONFIG_IP_FIB_TRIE_STATS
1527			if (!n)
1528				this_cpu_inc(stats->null_node_hit);
1529#endif
1530			/* If we are at cindex 0 there are no more bits for
1531			 * us to strip at this level so we must ascend back
1532			 * up one level to see if there are any more bits to
1533			 * be stripped there.
1534			 */
1535			while (!cindex) {
1536				t_key pkey = pn->key;
1537
1538				/* If we don't have a parent then there is
1539				 * nothing for us to do as we do not have any
1540				 * further nodes to parse.
1541				 */
1542				if (IS_TRIE(pn)) {
1543					trace_fib_table_lookup(tb->tb_id, flp,
1544							       NULL, -EAGAIN);
1545					return -EAGAIN;
1546				}
1547#ifdef CONFIG_IP_FIB_TRIE_STATS
1548				this_cpu_inc(stats->backtrack);
1549#endif
1550				/* Get Child's index */
1551				pn = node_parent_rcu(pn);
1552				cindex = get_index(pkey, pn);
1553			}
1554
1555			/* strip the least significant bit from the cindex */
1556			cindex &= cindex - 1;
1557
1558			/* grab pointer for next child node */
1559			cptr = &pn->tnode[cindex];
1560		}
1561	}
1562
1563found:
1564	/* this line carries forward the xor from earlier in the function */
1565	index = key ^ n->key;
1566
1567	/* Step 3: Process the leaf, if that fails fall back to backtracing */
1568	hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1569		struct fib_info *fi = fa->fa_info;
1570		struct fib_nh_common *nhc;
1571		int nhsel, err;
1572
1573		if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1574			if (index >= (1ul << fa->fa_slen))
1575				continue;
1576		}
1577		if (fa->fa_dscp && !fib_dscp_masked_match(fa->fa_dscp, flp))
1578			continue;
1579		/* Paired with WRITE_ONCE() in fib_release_info() */
1580		if (READ_ONCE(fi->fib_dead))
1581			continue;
1582		if (fa->fa_info->fib_scope < flp->flowi4_scope)
1583			continue;
1584		fib_alias_accessed(fa);
1585		err = fib_props[fa->fa_type].error;
1586		if (unlikely(err < 0)) {
1587out_reject:
1588#ifdef CONFIG_IP_FIB_TRIE_STATS
1589			this_cpu_inc(stats->semantic_match_passed);
1590#endif
1591			trace_fib_table_lookup(tb->tb_id, flp, NULL, err);
1592			return err;
1593		}
1594		if (fi->fib_flags & RTNH_F_DEAD)
1595			continue;
 
 
 
1596
1597		if (unlikely(fi->nh)) {
1598			if (nexthop_is_blackhole(fi->nh)) {
1599				err = fib_props[RTN_BLACKHOLE].error;
1600				goto out_reject;
 
 
 
 
 
 
 
1601			}
1602
1603			nhc = nexthop_get_nhc_lookup(fi->nh, fib_flags, flp,
1604						     &nhsel);
1605			if (nhc)
1606				goto set_result;
1607			goto miss;
1608		}
1609
1610		for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
1611			nhc = fib_info_nhc(fi, nhsel);
1612
1613			if (!fib_lookup_good_nhc(nhc, fib_flags, flp))
1614				continue;
1615set_result:
1616			if (!(fib_flags & FIB_LOOKUP_NOREF))
1617				refcount_inc(&fi->fib_clntref);
1618
1619			res->prefix = htonl(n->key);
1620			res->prefixlen = KEYLENGTH - fa->fa_slen;
1621			res->nh_sel = nhsel;
1622			res->nhc = nhc;
1623			res->type = fa->fa_type;
1624			res->scope = fi->fib_scope;
1625			res->dscp = fa->fa_dscp;
1626			res->fi = fi;
1627			res->table = tb;
1628			res->fa_head = &n->leaf;
1629#ifdef CONFIG_IP_FIB_TRIE_STATS
1630			this_cpu_inc(stats->semantic_match_passed);
1631#endif
1632			trace_fib_table_lookup(tb->tb_id, flp, nhc, err);
1633
1634			return err;
1635		}
1636	}
1637miss:
1638#ifdef CONFIG_IP_FIB_TRIE_STATS
1639	this_cpu_inc(stats->semantic_match_miss);
1640#endif
1641	goto backtrace;
1642}
1643EXPORT_SYMBOL_GPL(fib_table_lookup);
1644
1645static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1646			     struct key_vector *l, struct fib_alias *old)
1647{
1648	/* record the location of the previous list_info entry */
1649	struct hlist_node **pprev = old->fa_list.pprev;
1650	struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1651
1652	/* remove the fib_alias from the list */
1653	hlist_del_rcu(&old->fa_list);
1654
1655	/* if we emptied the list this leaf will be freed and we can sort
1656	 * out parent suffix lengths as a part of trie_rebalance
1657	 */
1658	if (hlist_empty(&l->leaf)) {
1659		if (tp->slen == l->slen)
1660			node_pull_suffix(tp, tp->pos);
1661		put_child_root(tp, l->key, NULL);
1662		node_free(l);
1663		trie_rebalance(t, tp);
1664		return;
1665	}
1666
1667	/* only access fa if it is pointing at the last valid hlist_node */
1668	if (*pprev)
1669		return;
1670
1671	/* update the trie with the latest suffix length */
1672	l->slen = fa->fa_slen;
1673	node_pull_suffix(tp, fa->fa_slen);
1674}
1675
1676static void fib_notify_alias_delete(struct net *net, u32 key,
1677				    struct hlist_head *fah,
1678				    struct fib_alias *fa_to_delete,
1679				    struct netlink_ext_ack *extack)
1680{
1681	struct fib_alias *fa_next, *fa_to_notify;
1682	u32 tb_id = fa_to_delete->tb_id;
1683	u8 slen = fa_to_delete->fa_slen;
1684	enum fib_event_type fib_event;
1685
1686	/* Do not notify if we do not care about the route. */
1687	if (fib_find_alias(fah, slen, 0, 0, tb_id, true) != fa_to_delete)
1688		return;
1689
1690	/* Determine if the route should be replaced by the next route in the
1691	 * list.
1692	 */
1693	fa_next = hlist_entry_safe(fa_to_delete->fa_list.next,
1694				   struct fib_alias, fa_list);
1695	if (fa_next && fa_next->fa_slen == slen && fa_next->tb_id == tb_id) {
1696		fib_event = FIB_EVENT_ENTRY_REPLACE;
1697		fa_to_notify = fa_next;
1698	} else {
1699		fib_event = FIB_EVENT_ENTRY_DEL;
1700		fa_to_notify = fa_to_delete;
1701	}
1702	call_fib_entry_notifiers(net, fib_event, key, KEYLENGTH - slen,
1703				 fa_to_notify, extack);
1704}
1705
1706/* Caller must hold RTNL. */
1707int fib_table_delete(struct net *net, struct fib_table *tb,
1708		     struct fib_config *cfg, struct netlink_ext_ack *extack)
1709{
1710	struct trie *t = (struct trie *) tb->tb_data;
1711	struct fib_alias *fa, *fa_to_delete;
1712	struct key_vector *l, *tp;
1713	u8 plen = cfg->fc_dst_len;
1714	u8 slen = KEYLENGTH - plen;
1715	dscp_t dscp;
1716	u32 key;
1717
 
 
 
1718	key = ntohl(cfg->fc_dst);
1719
1720	if (!fib_valid_key_len(key, plen, extack))
1721		return -EINVAL;
1722
1723	l = fib_find_node(t, &tp, key);
1724	if (!l)
1725		return -ESRCH;
1726
1727	dscp = cfg->fc_dscp;
1728	fa = fib_find_alias(&l->leaf, slen, dscp, 0, tb->tb_id, false);
1729	if (!fa)
1730		return -ESRCH;
1731
1732	pr_debug("Deleting %08x/%d dsfield=0x%02x t=%p\n", key, plen,
1733		 inet_dscp_to_dsfield(dscp), t);
1734
1735	fa_to_delete = NULL;
1736	hlist_for_each_entry_from(fa, fa_list) {
1737		struct fib_info *fi = fa->fa_info;
1738
1739		if ((fa->fa_slen != slen) ||
1740		    (fa->tb_id != tb->tb_id) ||
1741		    (fa->fa_dscp != dscp))
1742			break;
1743
1744		if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1745		    (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1746		     fa->fa_info->fib_scope == cfg->fc_scope) &&
1747		    (!cfg->fc_prefsrc ||
1748		     fi->fib_prefsrc == cfg->fc_prefsrc) &&
1749		    (!cfg->fc_protocol ||
1750		     fi->fib_protocol == cfg->fc_protocol) &&
1751		    fib_nh_match(net, cfg, fi, extack) == 0 &&
1752		    fib_metrics_match(cfg, fi)) {
1753			fa_to_delete = fa;
1754			break;
1755		}
1756	}
1757
1758	if (!fa_to_delete)
1759		return -ESRCH;
1760
1761	fib_notify_alias_delete(net, key, &l->leaf, fa_to_delete, extack);
 
 
1762	rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1763		  &cfg->fc_nlinfo, 0);
1764
1765	if (!plen)
1766		tb->tb_num_default--;
1767
1768	fib_remove_alias(t, tp, l, fa_to_delete);
1769
1770	if (fa_to_delete->fa_state & FA_S_ACCESSED)
1771		rt_cache_flush(cfg->fc_nlinfo.nl_net);
1772
1773	fib_release_info(fa_to_delete->fa_info);
1774	alias_free_mem_rcu(fa_to_delete);
1775	return 0;
1776}
1777
1778/* Scan for the next leaf starting at the provided key value */
1779static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1780{
1781	struct key_vector *pn, *n = *tn;
1782	unsigned long cindex;
1783
1784	/* this loop is meant to try and find the key in the trie */
1785	do {
1786		/* record parent and next child index */
1787		pn = n;
1788		cindex = (key > pn->key) ? get_index(key, pn) : 0;
1789
1790		if (cindex >> pn->bits)
1791			break;
1792
1793		/* descend into the next child */
1794		n = get_child_rcu(pn, cindex++);
1795		if (!n)
1796			break;
1797
1798		/* guarantee forward progress on the keys */
1799		if (IS_LEAF(n) && (n->key >= key))
1800			goto found;
1801	} while (IS_TNODE(n));
1802
1803	/* this loop will search for the next leaf with a greater key */
1804	while (!IS_TRIE(pn)) {
1805		/* if we exhausted the parent node we will need to climb */
1806		if (cindex >= (1ul << pn->bits)) {
1807			t_key pkey = pn->key;
1808
1809			pn = node_parent_rcu(pn);
1810			cindex = get_index(pkey, pn) + 1;
1811			continue;
1812		}
1813
1814		/* grab the next available node */
1815		n = get_child_rcu(pn, cindex++);
1816		if (!n)
1817			continue;
1818
1819		/* no need to compare keys since we bumped the index */
1820		if (IS_LEAF(n))
1821			goto found;
1822
1823		/* Rescan start scanning in new node */
1824		pn = n;
1825		cindex = 0;
1826	}
1827
1828	*tn = pn;
1829	return NULL; /* Root of trie */
1830found:
1831	/* if we are at the limit for keys just return NULL for the tnode */
1832	*tn = pn;
1833	return n;
1834}
1835
1836static void fib_trie_free(struct fib_table *tb)
1837{
1838	struct trie *t = (struct trie *)tb->tb_data;
1839	struct key_vector *pn = t->kv;
1840	unsigned long cindex = 1;
1841	struct hlist_node *tmp;
1842	struct fib_alias *fa;
1843
1844	/* walk trie in reverse order and free everything */
1845	for (;;) {
1846		struct key_vector *n;
1847
1848		if (!(cindex--)) {
1849			t_key pkey = pn->key;
1850
1851			if (IS_TRIE(pn))
1852				break;
1853
1854			n = pn;
1855			pn = node_parent(pn);
1856
1857			/* drop emptied tnode */
1858			put_child_root(pn, n->key, NULL);
1859			node_free(n);
1860
1861			cindex = get_index(pkey, pn);
1862
1863			continue;
1864		}
1865
1866		/* grab the next available node */
1867		n = get_child(pn, cindex);
1868		if (!n)
1869			continue;
1870
1871		if (IS_TNODE(n)) {
1872			/* record pn and cindex for leaf walking */
1873			pn = n;
1874			cindex = 1ul << n->bits;
1875
1876			continue;
1877		}
1878
1879		hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1880			hlist_del_rcu(&fa->fa_list);
1881			alias_free_mem_rcu(fa);
1882		}
1883
1884		put_child_root(pn, n->key, NULL);
1885		node_free(n);
1886	}
1887
1888#ifdef CONFIG_IP_FIB_TRIE_STATS
1889	free_percpu(t->stats);
1890#endif
1891	kfree(tb);
1892}
1893
1894struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1895{
1896	struct trie *ot = (struct trie *)oldtb->tb_data;
1897	struct key_vector *l, *tp = ot->kv;
1898	struct fib_table *local_tb;
1899	struct fib_alias *fa;
1900	struct trie *lt;
1901	t_key key = 0;
1902
1903	if (oldtb->tb_data == oldtb->__data)
1904		return oldtb;
1905
1906	local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1907	if (!local_tb)
1908		return NULL;
1909
1910	lt = (struct trie *)local_tb->tb_data;
1911
1912	while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1913		struct key_vector *local_l = NULL, *local_tp;
1914
1915		hlist_for_each_entry(fa, &l->leaf, fa_list) {
1916			struct fib_alias *new_fa;
1917
1918			if (local_tb->tb_id != fa->tb_id)
1919				continue;
1920
1921			/* clone fa for new local table */
1922			new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1923			if (!new_fa)
1924				goto out;
1925
1926			memcpy(new_fa, fa, sizeof(*fa));
1927
1928			/* insert clone into table */
1929			if (!local_l)
1930				local_l = fib_find_node(lt, &local_tp, l->key);
1931
1932			if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1933					     NULL, l->key)) {
1934				kmem_cache_free(fn_alias_kmem, new_fa);
1935				goto out;
1936			}
1937		}
1938
1939		/* stop loop if key wrapped back to 0 */
1940		key = l->key + 1;
1941		if (key < l->key)
1942			break;
1943	}
1944
1945	return local_tb;
1946out:
1947	fib_trie_free(local_tb);
1948
1949	return NULL;
1950}
1951
1952/* Caller must hold RTNL */
1953void fib_table_flush_external(struct fib_table *tb)
1954{
1955	struct trie *t = (struct trie *)tb->tb_data;
1956	struct key_vector *pn = t->kv;
1957	unsigned long cindex = 1;
1958	struct hlist_node *tmp;
1959	struct fib_alias *fa;
1960
1961	/* walk trie in reverse order */
1962	for (;;) {
1963		unsigned char slen = 0;
1964		struct key_vector *n;
1965
1966		if (!(cindex--)) {
1967			t_key pkey = pn->key;
1968
1969			/* cannot resize the trie vector */
1970			if (IS_TRIE(pn))
1971				break;
1972
1973			/* update the suffix to address pulled leaves */
1974			if (pn->slen > pn->pos)
1975				update_suffix(pn);
1976
1977			/* resize completed node */
1978			pn = resize(t, pn);
1979			cindex = get_index(pkey, pn);
1980
1981			continue;
1982		}
1983
1984		/* grab the next available node */
1985		n = get_child(pn, cindex);
1986		if (!n)
1987			continue;
1988
1989		if (IS_TNODE(n)) {
1990			/* record pn and cindex for leaf walking */
1991			pn = n;
1992			cindex = 1ul << n->bits;
1993
1994			continue;
1995		}
1996
1997		hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1998			/* if alias was cloned to local then we just
1999			 * need to remove the local copy from main
2000			 */
2001			if (tb->tb_id != fa->tb_id) {
2002				hlist_del_rcu(&fa->fa_list);
2003				alias_free_mem_rcu(fa);
2004				continue;
2005			}
2006
2007			/* record local slen */
2008			slen = fa->fa_slen;
2009		}
2010
2011		/* update leaf slen */
2012		n->slen = slen;
2013
2014		if (hlist_empty(&n->leaf)) {
2015			put_child_root(pn, n->key, NULL);
2016			node_free(n);
2017		}
2018	}
2019}
2020
2021/* Caller must hold RTNL. */
2022int fib_table_flush(struct net *net, struct fib_table *tb, bool flush_all)
2023{
2024	struct trie *t = (struct trie *)tb->tb_data;
2025	struct nl_info info = { .nl_net = net };
2026	struct key_vector *pn = t->kv;
2027	unsigned long cindex = 1;
2028	struct hlist_node *tmp;
2029	struct fib_alias *fa;
2030	int found = 0;
2031
2032	/* walk trie in reverse order */
2033	for (;;) {
2034		unsigned char slen = 0;
2035		struct key_vector *n;
2036
2037		if (!(cindex--)) {
2038			t_key pkey = pn->key;
2039
2040			/* cannot resize the trie vector */
2041			if (IS_TRIE(pn))
2042				break;
2043
2044			/* update the suffix to address pulled leaves */
2045			if (pn->slen > pn->pos)
2046				update_suffix(pn);
2047
2048			/* resize completed node */
2049			pn = resize(t, pn);
2050			cindex = get_index(pkey, pn);
2051
2052			continue;
2053		}
2054
2055		/* grab the next available node */
2056		n = get_child(pn, cindex);
2057		if (!n)
2058			continue;
2059
2060		if (IS_TNODE(n)) {
2061			/* record pn and cindex for leaf walking */
2062			pn = n;
2063			cindex = 1ul << n->bits;
2064
2065			continue;
2066		}
2067
2068		hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
2069			struct fib_info *fi = fa->fa_info;
2070
2071			if (!fi || tb->tb_id != fa->tb_id ||
2072			    (!(fi->fib_flags & RTNH_F_DEAD) &&
2073			     !fib_props[fa->fa_type].error)) {
2074				slen = fa->fa_slen;
2075				continue;
2076			}
2077
2078			/* Do not flush error routes if network namespace is
2079			 * not being dismantled
2080			 */
2081			if (!flush_all && fib_props[fa->fa_type].error) {
2082				slen = fa->fa_slen;
2083				continue;
2084			}
2085
2086			fib_notify_alias_delete(net, n->key, &n->leaf, fa,
2087						NULL);
2088			if (fi->pfsrc_removed)
2089				rtmsg_fib(RTM_DELROUTE, htonl(n->key), fa,
2090					  KEYLENGTH - fa->fa_slen, tb->tb_id, &info, 0);
2091			hlist_del_rcu(&fa->fa_list);
2092			fib_release_info(fa->fa_info);
2093			alias_free_mem_rcu(fa);
2094			found++;
2095		}
2096
2097		/* update leaf slen */
2098		n->slen = slen;
2099
2100		if (hlist_empty(&n->leaf)) {
2101			put_child_root(pn, n->key, NULL);
2102			node_free(n);
2103		}
2104	}
2105
2106	pr_debug("trie_flush found=%d\n", found);
2107	return found;
2108}
2109
2110/* derived from fib_trie_free */
2111static void __fib_info_notify_update(struct net *net, struct fib_table *tb,
2112				     struct nl_info *info)
2113{
2114	struct trie *t = (struct trie *)tb->tb_data;
2115	struct key_vector *pn = t->kv;
2116	unsigned long cindex = 1;
2117	struct fib_alias *fa;
2118
2119	for (;;) {
2120		struct key_vector *n;
2121
2122		if (!(cindex--)) {
2123			t_key pkey = pn->key;
2124
2125			if (IS_TRIE(pn))
2126				break;
2127
2128			pn = node_parent(pn);
2129			cindex = get_index(pkey, pn);
2130			continue;
2131		}
2132
2133		/* grab the next available node */
2134		n = get_child(pn, cindex);
2135		if (!n)
2136			continue;
2137
2138		if (IS_TNODE(n)) {
2139			/* record pn and cindex for leaf walking */
2140			pn = n;
2141			cindex = 1ul << n->bits;
2142
2143			continue;
2144		}
2145
2146		hlist_for_each_entry(fa, &n->leaf, fa_list) {
2147			struct fib_info *fi = fa->fa_info;
2148
2149			if (!fi || !fi->nh_updated || fa->tb_id != tb->tb_id)
2150				continue;
2151
2152			rtmsg_fib(RTM_NEWROUTE, htonl(n->key), fa,
2153				  KEYLENGTH - fa->fa_slen, tb->tb_id,
2154				  info, NLM_F_REPLACE);
2155		}
2156	}
2157}
2158
2159void fib_info_notify_update(struct net *net, struct nl_info *info)
2160{
2161	unsigned int h;
2162
2163	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2164		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2165		struct fib_table *tb;
2166
2167		hlist_for_each_entry_rcu(tb, head, tb_hlist,
2168					 lockdep_rtnl_is_held())
2169			__fib_info_notify_update(net, tb, info);
2170	}
2171}
2172
2173static int fib_leaf_notify(struct key_vector *l, struct fib_table *tb,
2174			   struct notifier_block *nb,
2175			   struct netlink_ext_ack *extack)
2176{
2177	struct fib_alias *fa;
2178	int last_slen = -1;
2179	int err;
2180
2181	hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2182		struct fib_info *fi = fa->fa_info;
2183
2184		if (!fi)
2185			continue;
2186
2187		/* local and main table can share the same trie,
2188		 * so don't notify twice for the same entry.
2189		 */
2190		if (tb->tb_id != fa->tb_id)
2191			continue;
2192
2193		if (fa->fa_slen == last_slen)
2194			continue;
2195
2196		last_slen = fa->fa_slen;
2197		err = call_fib_entry_notifier(nb, FIB_EVENT_ENTRY_REPLACE,
2198					      l->key, KEYLENGTH - fa->fa_slen,
2199					      fa, extack);
2200		if (err)
2201			return err;
2202	}
2203	return 0;
2204}
2205
2206static int fib_table_notify(struct fib_table *tb, struct notifier_block *nb,
2207			    struct netlink_ext_ack *extack)
 
2208{
2209	struct trie *t = (struct trie *)tb->tb_data;
2210	struct key_vector *l, *tp = t->kv;
2211	t_key key = 0;
2212	int err;
2213
2214	while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2215		err = fib_leaf_notify(l, tb, nb, extack);
2216		if (err)
2217			return err;
2218
2219		key = l->key + 1;
2220		/* stop in case of wrap around */
2221		if (key < l->key)
2222			break;
2223	}
2224	return 0;
2225}
2226
2227int fib_notify(struct net *net, struct notifier_block *nb,
2228	       struct netlink_ext_ack *extack)
2229{
2230	unsigned int h;
2231	int err;
2232
2233	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2234		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2235		struct fib_table *tb;
2236
2237		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2238			err = fib_table_notify(tb, nb, extack);
2239			if (err)
2240				return err;
2241		}
2242	}
2243	return 0;
2244}
2245
2246static void __trie_free_rcu(struct rcu_head *head)
2247{
2248	struct fib_table *tb = container_of(head, struct fib_table, rcu);
2249#ifdef CONFIG_IP_FIB_TRIE_STATS
2250	struct trie *t = (struct trie *)tb->tb_data;
2251
2252	if (tb->tb_data == tb->__data)
2253		free_percpu(t->stats);
2254#endif /* CONFIG_IP_FIB_TRIE_STATS */
2255	kfree(tb);
2256}
2257
2258void fib_free_table(struct fib_table *tb)
2259{
2260	call_rcu(&tb->rcu, __trie_free_rcu);
2261}
2262
2263static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
2264			     struct sk_buff *skb, struct netlink_callback *cb,
2265			     struct fib_dump_filter *filter)
2266{
2267	unsigned int flags = NLM_F_MULTI;
2268	__be32 xkey = htonl(l->key);
2269	int i, s_i, i_fa, s_fa, err;
2270	struct fib_alias *fa;
2271
2272	if (filter->filter_set ||
2273	    !filter->dump_exceptions || !filter->dump_routes)
2274		flags |= NLM_F_DUMP_FILTERED;
2275
2276	s_i = cb->args[4];
2277	s_fa = cb->args[5];
2278	i = 0;
2279
2280	/* rcu_read_lock is hold by caller */
2281	hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2282		struct fib_info *fi = fa->fa_info;
2283
2284		if (i < s_i)
2285			goto next;
2286
2287		i_fa = 0;
2288
2289		if (tb->tb_id != fa->tb_id)
2290			goto next;
2291
2292		if (filter->filter_set) {
2293			if (filter->rt_type && fa->fa_type != filter->rt_type)
2294				goto next;
2295
2296			if ((filter->protocol &&
2297			     fi->fib_protocol != filter->protocol))
2298				goto next;
2299
2300			if (filter->dev &&
2301			    !fib_info_nh_uses_dev(fi, filter->dev))
2302				goto next;
2303		}
2304
2305		if (filter->dump_routes) {
2306			if (!s_fa) {
2307				struct fib_rt_info fri;
2308
2309				fri.fi = fi;
2310				fri.tb_id = tb->tb_id;
2311				fri.dst = xkey;
2312				fri.dst_len = KEYLENGTH - fa->fa_slen;
2313				fri.dscp = fa->fa_dscp;
2314				fri.type = fa->fa_type;
2315				fri.offload = READ_ONCE(fa->offload);
2316				fri.trap = READ_ONCE(fa->trap);
2317				fri.offload_failed = READ_ONCE(fa->offload_failed);
2318				err = fib_dump_info(skb,
2319						    NETLINK_CB(cb->skb).portid,
2320						    cb->nlh->nlmsg_seq,
2321						    RTM_NEWROUTE, &fri, flags);
2322				if (err < 0)
2323					goto stop;
2324			}
2325
2326			i_fa++;
2327		}
2328
2329		if (filter->dump_exceptions) {
2330			err = fib_dump_info_fnhe(skb, cb, tb->tb_id, fi,
2331						 &i_fa, s_fa, flags);
2332			if (err < 0)
2333				goto stop;
2334		}
2335
2336next:
 
 
 
 
 
 
 
 
 
 
 
2337		i++;
2338	}
2339
2340	cb->args[4] = i;
2341	return skb->len;
2342
2343stop:
2344	cb->args[4] = i;
2345	cb->args[5] = i_fa;
2346	return err;
2347}
2348
2349/* rcu_read_lock needs to be hold by caller from readside */
2350int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
2351		   struct netlink_callback *cb, struct fib_dump_filter *filter)
2352{
2353	struct trie *t = (struct trie *)tb->tb_data;
2354	struct key_vector *l, *tp = t->kv;
2355	/* Dump starting at last key.
2356	 * Note: 0.0.0.0/0 (ie default) is first key.
2357	 */
2358	int count = cb->args[2];
2359	t_key key = cb->args[3];
2360
2361	/* First time here, count and key are both always 0. Count > 0
2362	 * and key == 0 means the dump has wrapped around and we are done.
2363	 */
2364	if (count && !key)
2365		return 0;
2366
2367	while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2368		int err;
2369
2370		err = fn_trie_dump_leaf(l, tb, skb, cb, filter);
2371		if (err < 0) {
2372			cb->args[3] = key;
2373			cb->args[2] = count;
2374			return err;
2375		}
2376
2377		++count;
2378		key = l->key + 1;
2379
2380		memset(&cb->args[4], 0,
2381		       sizeof(cb->args) - 4*sizeof(cb->args[0]));
2382
2383		/* stop loop if key wrapped back to 0 */
2384		if (key < l->key)
2385			break;
2386	}
2387
2388	cb->args[3] = key;
2389	cb->args[2] = count;
2390
2391	return 0;
2392}
2393
2394void __init fib_trie_init(void)
2395{
2396	fn_alias_kmem = kmem_cache_create("ip_fib_alias",
2397					  sizeof(struct fib_alias),
2398					  0, SLAB_PANIC | SLAB_ACCOUNT, NULL);
2399
2400	trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
2401					   LEAF_SIZE,
2402					   0, SLAB_PANIC | SLAB_ACCOUNT, NULL);
2403}
2404
2405struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
2406{
2407	struct fib_table *tb;
2408	struct trie *t;
2409	size_t sz = sizeof(*tb);
2410
2411	if (!alias)
2412		sz += sizeof(struct trie);
2413
2414	tb = kzalloc(sz, GFP_KERNEL);
2415	if (!tb)
2416		return NULL;
2417
2418	tb->tb_id = id;
2419	tb->tb_num_default = 0;
2420	tb->tb_data = (alias ? alias->__data : tb->__data);
2421
2422	if (alias)
2423		return tb;
2424
2425	t = (struct trie *) tb->tb_data;
2426	t->kv[0].pos = KEYLENGTH;
2427	t->kv[0].slen = KEYLENGTH;
2428#ifdef CONFIG_IP_FIB_TRIE_STATS
2429	t->stats = alloc_percpu(struct trie_use_stats);
2430	if (!t->stats) {
2431		kfree(tb);
2432		tb = NULL;
2433	}
2434#endif
2435
2436	return tb;
2437}
2438
2439#ifdef CONFIG_PROC_FS
2440/* Depth first Trie walk iterator */
2441struct fib_trie_iter {
2442	struct seq_net_private p;
2443	struct fib_table *tb;
2444	struct key_vector *tnode;
2445	unsigned int index;
2446	unsigned int depth;
2447};
2448
2449static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2450{
2451	unsigned long cindex = iter->index;
2452	struct key_vector *pn = iter->tnode;
2453	t_key pkey;
2454
2455	pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2456		 iter->tnode, iter->index, iter->depth);
2457
2458	while (!IS_TRIE(pn)) {
2459		while (cindex < child_length(pn)) {
2460			struct key_vector *n = get_child_rcu(pn, cindex++);
2461
2462			if (!n)
2463				continue;
2464
2465			if (IS_LEAF(n)) {
2466				iter->tnode = pn;
2467				iter->index = cindex;
2468			} else {
2469				/* push down one level */
2470				iter->tnode = n;
2471				iter->index = 0;
2472				++iter->depth;
2473			}
2474
2475			return n;
2476		}
2477
2478		/* Current node exhausted, pop back up */
2479		pkey = pn->key;
2480		pn = node_parent_rcu(pn);
2481		cindex = get_index(pkey, pn) + 1;
2482		--iter->depth;
2483	}
2484
2485	/* record root node so further searches know we are done */
2486	iter->tnode = pn;
2487	iter->index = 0;
2488
2489	return NULL;
2490}
2491
2492static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2493					     struct trie *t)
2494{
2495	struct key_vector *n, *pn;
2496
2497	if (!t)
2498		return NULL;
2499
2500	pn = t->kv;
2501	n = rcu_dereference(pn->tnode[0]);
2502	if (!n)
2503		return NULL;
2504
2505	if (IS_TNODE(n)) {
2506		iter->tnode = n;
2507		iter->index = 0;
2508		iter->depth = 1;
2509	} else {
2510		iter->tnode = pn;
2511		iter->index = 0;
2512		iter->depth = 0;
2513	}
2514
2515	return n;
2516}
2517
2518static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2519{
2520	struct key_vector *n;
2521	struct fib_trie_iter iter;
2522
2523	memset(s, 0, sizeof(*s));
2524
2525	rcu_read_lock();
2526	for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2527		if (IS_LEAF(n)) {
2528			struct fib_alias *fa;
2529
2530			s->leaves++;
2531			s->totdepth += iter.depth;
2532			if (iter.depth > s->maxdepth)
2533				s->maxdepth = iter.depth;
2534
2535			hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2536				++s->prefixes;
2537		} else {
2538			s->tnodes++;
2539			if (n->bits < MAX_STAT_DEPTH)
2540				s->nodesizes[n->bits]++;
2541			s->nullpointers += tn_info(n)->empty_children;
2542		}
2543	}
2544	rcu_read_unlock();
2545}
2546
2547/*
2548 *	This outputs /proc/net/fib_triestats
2549 */
2550static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2551{
2552	unsigned int i, max, pointers, bytes, avdepth;
2553
2554	if (stat->leaves)
2555		avdepth = stat->totdepth*100 / stat->leaves;
2556	else
2557		avdepth = 0;
2558
2559	seq_printf(seq, "\tAver depth:     %u.%02d\n",
2560		   avdepth / 100, avdepth % 100);
2561	seq_printf(seq, "\tMax depth:      %u\n", stat->maxdepth);
2562
2563	seq_printf(seq, "\tLeaves:         %u\n", stat->leaves);
2564	bytes = LEAF_SIZE * stat->leaves;
2565
2566	seq_printf(seq, "\tPrefixes:       %u\n", stat->prefixes);
2567	bytes += sizeof(struct fib_alias) * stat->prefixes;
2568
2569	seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2570	bytes += TNODE_SIZE(0) * stat->tnodes;
2571
2572	max = MAX_STAT_DEPTH;
2573	while (max > 0 && stat->nodesizes[max-1] == 0)
2574		max--;
2575
2576	pointers = 0;
2577	for (i = 1; i < max; i++)
2578		if (stat->nodesizes[i] != 0) {
2579			seq_printf(seq, "  %u: %u",  i, stat->nodesizes[i]);
2580			pointers += (1<<i) * stat->nodesizes[i];
2581		}
2582	seq_putc(seq, '\n');
2583	seq_printf(seq, "\tPointers: %u\n", pointers);
2584
2585	bytes += sizeof(struct key_vector *) * pointers;
2586	seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2587	seq_printf(seq, "Total size: %u  kB\n", (bytes + 1023) / 1024);
2588}
2589
2590#ifdef CONFIG_IP_FIB_TRIE_STATS
2591static void trie_show_usage(struct seq_file *seq,
2592			    const struct trie_use_stats __percpu *stats)
2593{
2594	struct trie_use_stats s = { 0 };
2595	int cpu;
2596
2597	/* loop through all of the CPUs and gather up the stats */
2598	for_each_possible_cpu(cpu) {
2599		const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2600
2601		s.gets += pcpu->gets;
2602		s.backtrack += pcpu->backtrack;
2603		s.semantic_match_passed += pcpu->semantic_match_passed;
2604		s.semantic_match_miss += pcpu->semantic_match_miss;
2605		s.null_node_hit += pcpu->null_node_hit;
2606		s.resize_node_skipped += pcpu->resize_node_skipped;
2607	}
2608
2609	seq_printf(seq, "\nCounters:\n---------\n");
2610	seq_printf(seq, "gets = %u\n", s.gets);
2611	seq_printf(seq, "backtracks = %u\n", s.backtrack);
2612	seq_printf(seq, "semantic match passed = %u\n",
2613		   s.semantic_match_passed);
2614	seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2615	seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2616	seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2617}
2618#endif /*  CONFIG_IP_FIB_TRIE_STATS */
2619
2620static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2621{
2622	if (tb->tb_id == RT_TABLE_LOCAL)
2623		seq_puts(seq, "Local:\n");
2624	else if (tb->tb_id == RT_TABLE_MAIN)
2625		seq_puts(seq, "Main:\n");
2626	else
2627		seq_printf(seq, "Id %d:\n", tb->tb_id);
2628}
2629
2630
2631static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2632{
2633	struct net *net = seq->private;
2634	unsigned int h;
2635
2636	seq_printf(seq,
2637		   "Basic info: size of leaf:"
2638		   " %zd bytes, size of tnode: %zd bytes.\n",
2639		   LEAF_SIZE, TNODE_SIZE(0));
2640
2641	rcu_read_lock();
2642	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2643		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2644		struct fib_table *tb;
2645
2646		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2647			struct trie *t = (struct trie *) tb->tb_data;
2648			struct trie_stat stat;
2649
2650			if (!t)
2651				continue;
2652
2653			fib_table_print(seq, tb);
2654
2655			trie_collect_stats(t, &stat);
2656			trie_show_stats(seq, &stat);
2657#ifdef CONFIG_IP_FIB_TRIE_STATS
2658			trie_show_usage(seq, t->stats);
2659#endif
2660		}
2661		cond_resched_rcu();
2662	}
2663	rcu_read_unlock();
2664
2665	return 0;
2666}
2667
 
 
 
 
 
 
 
 
 
 
 
 
 
2668static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2669{
2670	struct fib_trie_iter *iter = seq->private;
2671	struct net *net = seq_file_net(seq);
2672	loff_t idx = 0;
2673	unsigned int h;
2674
2675	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2676		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2677		struct fib_table *tb;
2678
2679		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2680			struct key_vector *n;
2681
2682			for (n = fib_trie_get_first(iter,
2683						    (struct trie *) tb->tb_data);
2684			     n; n = fib_trie_get_next(iter))
2685				if (pos == idx++) {
2686					iter->tb = tb;
2687					return n;
2688				}
2689		}
2690	}
2691
2692	return NULL;
2693}
2694
2695static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2696	__acquires(RCU)
2697{
2698	rcu_read_lock();
2699	return fib_trie_get_idx(seq, *pos);
2700}
2701
2702static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2703{
2704	struct fib_trie_iter *iter = seq->private;
2705	struct net *net = seq_file_net(seq);
2706	struct fib_table *tb = iter->tb;
2707	struct hlist_node *tb_node;
2708	unsigned int h;
2709	struct key_vector *n;
2710
2711	++*pos;
2712	/* next node in same table */
2713	n = fib_trie_get_next(iter);
2714	if (n)
2715		return n;
2716
2717	/* walk rest of this hash chain */
2718	h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2719	while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2720		tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2721		n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2722		if (n)
2723			goto found;
2724	}
2725
2726	/* new hash chain */
2727	while (++h < FIB_TABLE_HASHSZ) {
2728		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2729		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2730			n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2731			if (n)
2732				goto found;
2733		}
2734	}
2735	return NULL;
2736
2737found:
2738	iter->tb = tb;
2739	return n;
2740}
2741
2742static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2743	__releases(RCU)
2744{
2745	rcu_read_unlock();
2746}
2747
2748static void seq_indent(struct seq_file *seq, int n)
2749{
2750	while (n-- > 0)
2751		seq_puts(seq, "   ");
2752}
2753
2754static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2755{
2756	switch (s) {
2757	case RT_SCOPE_UNIVERSE: return "universe";
2758	case RT_SCOPE_SITE:	return "site";
2759	case RT_SCOPE_LINK:	return "link";
2760	case RT_SCOPE_HOST:	return "host";
2761	case RT_SCOPE_NOWHERE:	return "nowhere";
2762	default:
2763		snprintf(buf, len, "scope=%d", s);
2764		return buf;
2765	}
2766}
2767
2768static const char *const rtn_type_names[__RTN_MAX] = {
2769	[RTN_UNSPEC] = "UNSPEC",
2770	[RTN_UNICAST] = "UNICAST",
2771	[RTN_LOCAL] = "LOCAL",
2772	[RTN_BROADCAST] = "BROADCAST",
2773	[RTN_ANYCAST] = "ANYCAST",
2774	[RTN_MULTICAST] = "MULTICAST",
2775	[RTN_BLACKHOLE] = "BLACKHOLE",
2776	[RTN_UNREACHABLE] = "UNREACHABLE",
2777	[RTN_PROHIBIT] = "PROHIBIT",
2778	[RTN_THROW] = "THROW",
2779	[RTN_NAT] = "NAT",
2780	[RTN_XRESOLVE] = "XRESOLVE",
2781};
2782
2783static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2784{
2785	if (t < __RTN_MAX && rtn_type_names[t])
2786		return rtn_type_names[t];
2787	snprintf(buf, len, "type %u", t);
2788	return buf;
2789}
2790
2791/* Pretty print the trie */
2792static int fib_trie_seq_show(struct seq_file *seq, void *v)
2793{
2794	const struct fib_trie_iter *iter = seq->private;
2795	struct key_vector *n = v;
2796
2797	if (IS_TRIE(node_parent_rcu(n)))
2798		fib_table_print(seq, iter->tb);
2799
2800	if (IS_TNODE(n)) {
2801		__be32 prf = htonl(n->key);
2802
2803		seq_indent(seq, iter->depth-1);
2804		seq_printf(seq, "  +-- %pI4/%zu %u %u %u\n",
2805			   &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2806			   tn_info(n)->full_children,
2807			   tn_info(n)->empty_children);
2808	} else {
2809		__be32 val = htonl(n->key);
2810		struct fib_alias *fa;
2811
2812		seq_indent(seq, iter->depth);
2813		seq_printf(seq, "  |-- %pI4\n", &val);
2814
2815		hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2816			char buf1[32], buf2[32];
2817
2818			seq_indent(seq, iter->depth + 1);
2819			seq_printf(seq, "  /%zu %s %s",
2820				   KEYLENGTH - fa->fa_slen,
2821				   rtn_scope(buf1, sizeof(buf1),
2822					     fa->fa_info->fib_scope),
2823				   rtn_type(buf2, sizeof(buf2),
2824					    fa->fa_type));
2825			if (fa->fa_dscp)
2826				seq_printf(seq, " tos=%d",
2827					   inet_dscp_to_dsfield(fa->fa_dscp));
2828			seq_putc(seq, '\n');
2829		}
2830	}
2831
2832	return 0;
2833}
2834
2835static const struct seq_operations fib_trie_seq_ops = {
2836	.start  = fib_trie_seq_start,
2837	.next   = fib_trie_seq_next,
2838	.stop   = fib_trie_seq_stop,
2839	.show   = fib_trie_seq_show,
2840};
2841
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2842struct fib_route_iter {
2843	struct seq_net_private p;
2844	struct fib_table *main_tb;
2845	struct key_vector *tnode;
2846	loff_t	pos;
2847	t_key	key;
2848};
2849
2850static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2851					    loff_t pos)
2852{
2853	struct key_vector *l, **tp = &iter->tnode;
2854	t_key key;
2855
2856	/* use cached location of previously found key */
2857	if (iter->pos > 0 && pos >= iter->pos) {
2858		key = iter->key;
2859	} else {
2860		iter->pos = 1;
2861		key = 0;
2862	}
2863
2864	pos -= iter->pos;
2865
2866	while ((l = leaf_walk_rcu(tp, key)) && (pos-- > 0)) {
2867		key = l->key + 1;
2868		iter->pos++;
2869		l = NULL;
2870
2871		/* handle unlikely case of a key wrap */
2872		if (!key)
2873			break;
2874	}
2875
2876	if (l)
2877		iter->key = l->key;	/* remember it */
2878	else
2879		iter->pos = 0;		/* forget it */
2880
2881	return l;
2882}
2883
2884static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2885	__acquires(RCU)
2886{
2887	struct fib_route_iter *iter = seq->private;
2888	struct fib_table *tb;
2889	struct trie *t;
2890
2891	rcu_read_lock();
2892
2893	tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2894	if (!tb)
2895		return NULL;
2896
2897	iter->main_tb = tb;
2898	t = (struct trie *)tb->tb_data;
2899	iter->tnode = t->kv;
2900
2901	if (*pos != 0)
2902		return fib_route_get_idx(iter, *pos);
2903
2904	iter->pos = 0;
2905	iter->key = KEY_MAX;
2906
2907	return SEQ_START_TOKEN;
2908}
2909
2910static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2911{
2912	struct fib_route_iter *iter = seq->private;
2913	struct key_vector *l = NULL;
2914	t_key key = iter->key + 1;
2915
2916	++*pos;
2917
2918	/* only allow key of 0 for start of sequence */
2919	if ((v == SEQ_START_TOKEN) || key)
2920		l = leaf_walk_rcu(&iter->tnode, key);
2921
2922	if (l) {
2923		iter->key = l->key;
2924		iter->pos++;
2925	} else {
2926		iter->pos = 0;
2927	}
2928
2929	return l;
2930}
2931
2932static void fib_route_seq_stop(struct seq_file *seq, void *v)
2933	__releases(RCU)
2934{
2935	rcu_read_unlock();
2936}
2937
2938static unsigned int fib_flag_trans(int type, __be32 mask, struct fib_info *fi)
2939{
2940	unsigned int flags = 0;
2941
2942	if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2943		flags = RTF_REJECT;
2944	if (fi) {
2945		const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2946
2947		if (nhc->nhc_gw.ipv4)
2948			flags |= RTF_GATEWAY;
2949	}
2950	if (mask == htonl(0xFFFFFFFF))
2951		flags |= RTF_HOST;
2952	flags |= RTF_UP;
2953	return flags;
2954}
2955
2956/*
2957 *	This outputs /proc/net/route.
2958 *	The format of the file is not supposed to be changed
2959 *	and needs to be same as fib_hash output to avoid breaking
2960 *	legacy utilities
2961 */
2962static int fib_route_seq_show(struct seq_file *seq, void *v)
2963{
2964	struct fib_route_iter *iter = seq->private;
2965	struct fib_table *tb = iter->main_tb;
2966	struct fib_alias *fa;
2967	struct key_vector *l = v;
2968	__be32 prefix;
2969
2970	if (v == SEQ_START_TOKEN) {
2971		seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2972			   "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2973			   "\tWindow\tIRTT");
2974		return 0;
2975	}
2976
2977	prefix = htonl(l->key);
2978
2979	hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2980		struct fib_info *fi = fa->fa_info;
2981		__be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2982		unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2983
2984		if ((fa->fa_type == RTN_BROADCAST) ||
2985		    (fa->fa_type == RTN_MULTICAST))
2986			continue;
2987
2988		if (fa->tb_id != tb->tb_id)
2989			continue;
2990
2991		seq_setwidth(seq, 127);
2992
2993		if (fi) {
2994			struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2995			__be32 gw = 0;
2996
2997			if (nhc->nhc_gw_family == AF_INET)
2998				gw = nhc->nhc_gw.ipv4;
2999
3000			seq_printf(seq,
3001				   "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
3002				   "%d\t%08X\t%d\t%u\t%u",
3003				   nhc->nhc_dev ? nhc->nhc_dev->name : "*",
3004				   prefix, gw, flags, 0, 0,
 
3005				   fi->fib_priority,
3006				   mask,
3007				   (fi->fib_advmss ?
3008				    fi->fib_advmss + 40 : 0),
3009				   fi->fib_window,
3010				   fi->fib_rtt >> 3);
3011		} else {
3012			seq_printf(seq,
3013				   "*\t%08X\t%08X\t%04X\t%d\t%u\t"
3014				   "%d\t%08X\t%d\t%u\t%u",
3015				   prefix, 0, flags, 0, 0, 0,
3016				   mask, 0, 0, 0);
3017		}
3018		seq_pad(seq, '\n');
3019	}
3020
3021	return 0;
3022}
3023
3024static const struct seq_operations fib_route_seq_ops = {
3025	.start  = fib_route_seq_start,
3026	.next   = fib_route_seq_next,
3027	.stop   = fib_route_seq_stop,
3028	.show   = fib_route_seq_show,
3029};
3030
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3031int __net_init fib_proc_init(struct net *net)
3032{
3033	if (!proc_create_net("fib_trie", 0444, net->proc_net, &fib_trie_seq_ops,
3034			sizeof(struct fib_trie_iter)))
3035		goto out1;
3036
3037	if (!proc_create_net_single("fib_triestat", 0444, net->proc_net,
3038			fib_triestat_seq_show, NULL))
3039		goto out2;
3040
3041	if (!proc_create_net("route", 0444, net->proc_net, &fib_route_seq_ops,
3042			sizeof(struct fib_route_iter)))
3043		goto out3;
3044
3045	return 0;
3046
3047out3:
3048	remove_proc_entry("fib_triestat", net->proc_net);
3049out2:
3050	remove_proc_entry("fib_trie", net->proc_net);
3051out1:
3052	return -ENOMEM;
3053}
3054
3055void __net_exit fib_proc_exit(struct net *net)
3056{
3057	remove_proc_entry("fib_trie", net->proc_net);
3058	remove_proc_entry("fib_triestat", net->proc_net);
3059	remove_proc_entry("route", net->proc_net);
3060}
3061
3062#endif /* CONFIG_PROC_FS */
v4.10.11
 
   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/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/export.h>
  75#include <linux/vmalloc.h>
  76#include <linux/notifier.h>
  77#include <net/net_namespace.h>
 
  78#include <net/ip.h>
  79#include <net/protocol.h>
  80#include <net/route.h>
  81#include <net/tcp.h>
  82#include <net/sock.h>
  83#include <net/ip_fib.h>
 
  84#include <trace/events/fib.h>
  85#include "fib_lookup.h"
  86
  87static unsigned int fib_seq_sum(void)
  88{
  89	unsigned int fib_seq = 0;
  90	struct net *net;
  91
  92	rtnl_lock();
  93	for_each_net(net)
  94		fib_seq += net->ipv4.fib_seq;
  95	rtnl_unlock();
  96
  97	return fib_seq;
  98}
  99
 100static ATOMIC_NOTIFIER_HEAD(fib_chain);
 101
 102static int call_fib_notifier(struct notifier_block *nb, struct net *net,
 103			     enum fib_event_type event_type,
 104			     struct fib_notifier_info *info)
 105{
 106	info->net = net;
 107	return nb->notifier_call(nb, event_type, info);
 108}
 109
 110static void fib_rules_notify(struct net *net, struct notifier_block *nb,
 111			     enum fib_event_type event_type)
 112{
 113#ifdef CONFIG_IP_MULTIPLE_TABLES
 114	struct fib_notifier_info info;
 115
 116	if (net->ipv4.fib_has_custom_rules)
 117		call_fib_notifier(nb, net, event_type, &info);
 118#endif
 119}
 120
 121static void fib_notify(struct net *net, struct notifier_block *nb,
 122		       enum fib_event_type event_type);
 123
 124static int call_fib_entry_notifier(struct notifier_block *nb, struct net *net,
 125				   enum fib_event_type event_type, u32 dst,
 126				   int dst_len, struct fib_info *fi,
 127				   u8 tos, u8 type, u32 tb_id, u32 nlflags)
 128{
 129	struct fib_entry_notifier_info info = {
 
 130		.dst = dst,
 131		.dst_len = dst_len,
 132		.fi = fi,
 133		.tos = tos,
 134		.type = type,
 135		.tb_id = tb_id,
 136		.nlflags = nlflags,
 137	};
 138	return call_fib_notifier(nb, net, event_type, &info.info);
 139}
 140
 141static bool fib_dump_is_consistent(struct notifier_block *nb,
 142				   void (*cb)(struct notifier_block *nb),
 143				   unsigned int fib_seq)
 144{
 145	atomic_notifier_chain_register(&fib_chain, nb);
 146	if (fib_seq == fib_seq_sum())
 147		return true;
 148	atomic_notifier_chain_unregister(&fib_chain, nb);
 149	if (cb)
 150		cb(nb);
 151	return false;
 152}
 153
 154#define FIB_DUMP_MAX_RETRIES 5
 155int register_fib_notifier(struct notifier_block *nb,
 156			  void (*cb)(struct notifier_block *nb))
 157{
 158	int retries = 0;
 159
 160	do {
 161		unsigned int fib_seq = fib_seq_sum();
 162		struct net *net;
 163
 164		/* Mutex semantics guarantee that every change done to
 165		 * FIB tries before we read the change sequence counter
 166		 * is now visible to us.
 167		 */
 168		rcu_read_lock();
 169		for_each_net_rcu(net) {
 170			fib_rules_notify(net, nb, FIB_EVENT_RULE_ADD);
 171			fib_notify(net, nb, FIB_EVENT_ENTRY_ADD);
 172		}
 173		rcu_read_unlock();
 174
 175		if (fib_dump_is_consistent(nb, cb, fib_seq))
 176			return 0;
 177	} while (++retries < FIB_DUMP_MAX_RETRIES);
 178
 179	return -EBUSY;
 180}
 181EXPORT_SYMBOL(register_fib_notifier);
 182
 183int unregister_fib_notifier(struct notifier_block *nb)
 184{
 185	return atomic_notifier_chain_unregister(&fib_chain, nb);
 186}
 187EXPORT_SYMBOL(unregister_fib_notifier);
 188
 189int call_fib_notifiers(struct net *net, enum fib_event_type event_type,
 190		       struct fib_notifier_info *info)
 191{
 192	net->ipv4.fib_seq++;
 193	info->net = net;
 194	return atomic_notifier_call_chain(&fib_chain, event_type, info);
 195}
 196
 197static int call_fib_entry_notifiers(struct net *net,
 198				    enum fib_event_type event_type, u32 dst,
 199				    int dst_len, struct fib_info *fi,
 200				    u8 tos, u8 type, u32 tb_id, u32 nlflags)
 201{
 202	struct fib_entry_notifier_info info = {
 
 203		.dst = dst,
 204		.dst_len = dst_len,
 205		.fi = fi,
 206		.tos = tos,
 207		.type = type,
 208		.tb_id = tb_id,
 209		.nlflags = nlflags,
 210	};
 211	return call_fib_notifiers(net, event_type, &info.info);
 212}
 213
 214#define MAX_STAT_DEPTH 32
 215
 216#define KEYLENGTH	(8*sizeof(t_key))
 217#define KEY_MAX		((t_key)~0)
 218
 219typedef unsigned int t_key;
 220
 221#define IS_TRIE(n)	((n)->pos >= KEYLENGTH)
 222#define IS_TNODE(n)	((n)->bits)
 223#define IS_LEAF(n)	(!(n)->bits)
 224
 225struct key_vector {
 226	t_key key;
 227	unsigned char pos;		/* 2log(KEYLENGTH) bits needed */
 228	unsigned char bits;		/* 2log(KEYLENGTH) bits needed */
 229	unsigned char slen;
 230	union {
 231		/* This list pointer if valid if (pos | bits) == 0 (LEAF) */
 232		struct hlist_head leaf;
 233		/* This array is valid if (pos | bits) > 0 (TNODE) */
 234		struct key_vector __rcu *tnode[0];
 235	};
 236};
 237
 238struct tnode {
 239	struct rcu_head rcu;
 240	t_key empty_children;		/* KEYLENGTH bits needed */
 241	t_key full_children;		/* KEYLENGTH bits needed */
 242	struct key_vector __rcu *parent;
 243	struct key_vector kv[1];
 244#define tn_bits kv[0].bits
 245};
 246
 247#define TNODE_SIZE(n)	offsetof(struct tnode, kv[0].tnode[n])
 248#define LEAF_SIZE	TNODE_SIZE(1)
 249
 250#ifdef CONFIG_IP_FIB_TRIE_STATS
 251struct trie_use_stats {
 252	unsigned int gets;
 253	unsigned int backtrack;
 254	unsigned int semantic_match_passed;
 255	unsigned int semantic_match_miss;
 256	unsigned int null_node_hit;
 257	unsigned int resize_node_skipped;
 258};
 259#endif
 260
 261struct trie_stat {
 262	unsigned int totdepth;
 263	unsigned int maxdepth;
 264	unsigned int tnodes;
 265	unsigned int leaves;
 266	unsigned int nullpointers;
 267	unsigned int prefixes;
 268	unsigned int nodesizes[MAX_STAT_DEPTH];
 269};
 270
 271struct trie {
 272	struct key_vector kv[1];
 273#ifdef CONFIG_IP_FIB_TRIE_STATS
 274	struct trie_use_stats __percpu *stats;
 275#endif
 276};
 277
 278static struct key_vector *resize(struct trie *t, struct key_vector *tn);
 279static size_t tnode_free_size;
 280
 281/*
 282 * synchronize_rcu after call_rcu for that many pages; it should be especially
 283 * useful before resizing the root node with PREEMPT_NONE configs; the value was
 284 * obtained experimentally, aiming to avoid visible slowdown.
 285 */
 286static const int sync_pages = 128;
 
 
 287
 288static struct kmem_cache *fn_alias_kmem __read_mostly;
 289static struct kmem_cache *trie_leaf_kmem __read_mostly;
 290
 291static inline struct tnode *tn_info(struct key_vector *kv)
 292{
 293	return container_of(kv, struct tnode, kv[0]);
 294}
 295
 296/* caller must hold RTNL */
 297#define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
 298#define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
 299
 300/* caller must hold RCU read lock or RTNL */
 301#define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
 302#define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
 303
 304/* wrapper for rcu_assign_pointer */
 305static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
 306{
 307	if (n)
 308		rcu_assign_pointer(tn_info(n)->parent, tp);
 309}
 310
 311#define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
 312
 313/* This provides us with the number of children in this node, in the case of a
 314 * leaf this will return 0 meaning none of the children are accessible.
 315 */
 316static inline unsigned long child_length(const struct key_vector *tn)
 317{
 318	return (1ul << tn->bits) & ~(1ul);
 319}
 320
 321#define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
 322
 323static inline unsigned long get_index(t_key key, struct key_vector *kv)
 324{
 325	unsigned long index = key ^ kv->key;
 326
 327	if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
 328		return 0;
 329
 330	return index >> kv->pos;
 331}
 332
 333/* To understand this stuff, an understanding of keys and all their bits is
 334 * necessary. Every node in the trie has a key associated with it, but not
 335 * all of the bits in that key are significant.
 336 *
 337 * Consider a node 'n' and its parent 'tp'.
 338 *
 339 * If n is a leaf, every bit in its key is significant. Its presence is
 340 * necessitated by path compression, since during a tree traversal (when
 341 * searching for a leaf - unless we are doing an insertion) we will completely
 342 * ignore all skipped bits we encounter. Thus we need to verify, at the end of
 343 * a potentially successful search, that we have indeed been walking the
 344 * correct key path.
 345 *
 346 * Note that we can never "miss" the correct key in the tree if present by
 347 * following the wrong path. Path compression ensures that segments of the key
 348 * that are the same for all keys with a given prefix are skipped, but the
 349 * skipped part *is* identical for each node in the subtrie below the skipped
 350 * bit! trie_insert() in this implementation takes care of that.
 351 *
 352 * if n is an internal node - a 'tnode' here, the various parts of its key
 353 * have many different meanings.
 354 *
 355 * Example:
 356 * _________________________________________________________________
 357 * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
 358 * -----------------------------------------------------------------
 359 *  31  30  29  28  27  26  25  24  23  22  21  20  19  18  17  16
 360 *
 361 * _________________________________________________________________
 362 * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
 363 * -----------------------------------------------------------------
 364 *  15  14  13  12  11  10   9   8   7   6   5   4   3   2   1   0
 365 *
 366 * tp->pos = 22
 367 * tp->bits = 3
 368 * n->pos = 13
 369 * n->bits = 4
 370 *
 371 * First, let's just ignore the bits that come before the parent tp, that is
 372 * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
 373 * point we do not use them for anything.
 374 *
 375 * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
 376 * index into the parent's child array. That is, they will be used to find
 377 * 'n' among tp's children.
 378 *
 379 * The bits from (n->pos + n->bits) to (tp->pos - 1) - "S" - are skipped bits
 380 * for the node n.
 381 *
 382 * All the bits we have seen so far are significant to the node n. The rest
 383 * of the bits are really not needed or indeed known in n->key.
 384 *
 385 * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
 386 * n's child array, and will of course be different for each child.
 387 *
 388 * The rest of the bits, from 0 to (n->pos -1) - "u" - are completely unknown
 389 * at this point.
 390 */
 391
 392static const int halve_threshold = 25;
 393static const int inflate_threshold = 50;
 394static const int halve_threshold_root = 15;
 395static const int inflate_threshold_root = 30;
 396
 397static void __alias_free_mem(struct rcu_head *head)
 398{
 399	struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
 400	kmem_cache_free(fn_alias_kmem, fa);
 401}
 402
 403static inline void alias_free_mem_rcu(struct fib_alias *fa)
 404{
 405	call_rcu(&fa->rcu, __alias_free_mem);
 406}
 407
 408#define TNODE_KMALLOC_MAX \
 409	ilog2((PAGE_SIZE - TNODE_SIZE(0)) / sizeof(struct key_vector *))
 410#define TNODE_VMALLOC_MAX \
 411	ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
 412
 413static void __node_free_rcu(struct rcu_head *head)
 414{
 415	struct tnode *n = container_of(head, struct tnode, rcu);
 416
 417	if (!n->tn_bits)
 418		kmem_cache_free(trie_leaf_kmem, n);
 419	else
 420		kvfree(n);
 421}
 422
 423#define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
 424
 425static struct tnode *tnode_alloc(int bits)
 426{
 427	size_t size;
 428
 429	/* verify bits is within bounds */
 430	if (bits > TNODE_VMALLOC_MAX)
 431		return NULL;
 432
 433	/* determine size and verify it is non-zero and didn't overflow */
 434	size = TNODE_SIZE(1ul << bits);
 435
 436	if (size <= PAGE_SIZE)
 437		return kzalloc(size, GFP_KERNEL);
 438	else
 439		return vzalloc(size);
 440}
 441
 442static inline void empty_child_inc(struct key_vector *n)
 443{
 444	++tn_info(n)->empty_children ? : ++tn_info(n)->full_children;
 
 
 
 445}
 446
 447static inline void empty_child_dec(struct key_vector *n)
 448{
 449	tn_info(n)->empty_children-- ? : tn_info(n)->full_children--;
 
 
 
 450}
 451
 452static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
 453{
 454	struct key_vector *l;
 455	struct tnode *kv;
 456
 457	kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
 458	if (!kv)
 459		return NULL;
 460
 461	/* initialize key vector */
 462	l = kv->kv;
 463	l->key = key;
 464	l->pos = 0;
 465	l->bits = 0;
 466	l->slen = fa->fa_slen;
 467
 468	/* link leaf to fib alias */
 469	INIT_HLIST_HEAD(&l->leaf);
 470	hlist_add_head(&fa->fa_list, &l->leaf);
 471
 472	return l;
 473}
 474
 475static struct key_vector *tnode_new(t_key key, int pos, int bits)
 476{
 477	unsigned int shift = pos + bits;
 478	struct key_vector *tn;
 479	struct tnode *tnode;
 480
 481	/* verify bits and pos their msb bits clear and values are valid */
 482	BUG_ON(!bits || (shift > KEYLENGTH));
 483
 484	tnode = tnode_alloc(bits);
 485	if (!tnode)
 486		return NULL;
 487
 488	pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
 489		 sizeof(struct key_vector *) << bits);
 490
 491	if (bits == KEYLENGTH)
 492		tnode->full_children = 1;
 493	else
 494		tnode->empty_children = 1ul << bits;
 495
 496	tn = tnode->kv;
 497	tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
 498	tn->pos = pos;
 499	tn->bits = bits;
 500	tn->slen = pos;
 501
 502	return tn;
 503}
 504
 505/* Check whether a tnode 'n' is "full", i.e. it is an internal node
 506 * and no bits are skipped. See discussion in dyntree paper p. 6
 507 */
 508static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
 509{
 510	return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
 511}
 512
 513/* Add a child at position i overwriting the old value.
 514 * Update the value of full_children and empty_children.
 515 */
 516static void put_child(struct key_vector *tn, unsigned long i,
 517		      struct key_vector *n)
 518{
 519	struct key_vector *chi = get_child(tn, i);
 520	int isfull, wasfull;
 521
 522	BUG_ON(i >= child_length(tn));
 523
 524	/* update emptyChildren, overflow into fullChildren */
 525	if (!n && chi)
 526		empty_child_inc(tn);
 527	if (n && !chi)
 528		empty_child_dec(tn);
 529
 530	/* update fullChildren */
 531	wasfull = tnode_full(tn, chi);
 532	isfull = tnode_full(tn, n);
 533
 534	if (wasfull && !isfull)
 535		tn_info(tn)->full_children--;
 536	else if (!wasfull && isfull)
 537		tn_info(tn)->full_children++;
 538
 539	if (n && (tn->slen < n->slen))
 540		tn->slen = n->slen;
 541
 542	rcu_assign_pointer(tn->tnode[i], n);
 543}
 544
 545static void update_children(struct key_vector *tn)
 546{
 547	unsigned long i;
 548
 549	/* update all of the child parent pointers */
 550	for (i = child_length(tn); i;) {
 551		struct key_vector *inode = get_child(tn, --i);
 552
 553		if (!inode)
 554			continue;
 555
 556		/* Either update the children of a tnode that
 557		 * already belongs to us or update the child
 558		 * to point to ourselves.
 559		 */
 560		if (node_parent(inode) == tn)
 561			update_children(inode);
 562		else
 563			node_set_parent(inode, tn);
 564	}
 565}
 566
 567static inline void put_child_root(struct key_vector *tp, t_key key,
 568				  struct key_vector *n)
 569{
 570	if (IS_TRIE(tp))
 571		rcu_assign_pointer(tp->tnode[0], n);
 572	else
 573		put_child(tp, get_index(key, tp), n);
 574}
 575
 576static inline void tnode_free_init(struct key_vector *tn)
 577{
 578	tn_info(tn)->rcu.next = NULL;
 579}
 580
 581static inline void tnode_free_append(struct key_vector *tn,
 582				     struct key_vector *n)
 583{
 584	tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
 585	tn_info(tn)->rcu.next = &tn_info(n)->rcu;
 586}
 587
 588static void tnode_free(struct key_vector *tn)
 589{
 590	struct callback_head *head = &tn_info(tn)->rcu;
 591
 592	while (head) {
 593		head = head->next;
 594		tnode_free_size += TNODE_SIZE(1ul << tn->bits);
 595		node_free(tn);
 596
 597		tn = container_of(head, struct tnode, rcu)->kv;
 598	}
 599
 600	if (tnode_free_size >= PAGE_SIZE * sync_pages) {
 601		tnode_free_size = 0;
 602		synchronize_rcu();
 603	}
 604}
 605
 606static struct key_vector *replace(struct trie *t,
 607				  struct key_vector *oldtnode,
 608				  struct key_vector *tn)
 609{
 610	struct key_vector *tp = node_parent(oldtnode);
 611	unsigned long i;
 612
 613	/* setup the parent pointer out of and back into this node */
 614	NODE_INIT_PARENT(tn, tp);
 615	put_child_root(tp, tn->key, tn);
 616
 617	/* update all of the child parent pointers */
 618	update_children(tn);
 619
 620	/* all pointers should be clean so we are done */
 621	tnode_free(oldtnode);
 622
 623	/* resize children now that oldtnode is freed */
 624	for (i = child_length(tn); i;) {
 625		struct key_vector *inode = get_child(tn, --i);
 626
 627		/* resize child node */
 628		if (tnode_full(tn, inode))
 629			tn = resize(t, inode);
 630	}
 631
 632	return tp;
 633}
 634
 635static struct key_vector *inflate(struct trie *t,
 636				  struct key_vector *oldtnode)
 637{
 638	struct key_vector *tn;
 639	unsigned long i;
 640	t_key m;
 641
 642	pr_debug("In inflate\n");
 643
 644	tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
 645	if (!tn)
 646		goto notnode;
 647
 648	/* prepare oldtnode to be freed */
 649	tnode_free_init(oldtnode);
 650
 651	/* Assemble all of the pointers in our cluster, in this case that
 652	 * represents all of the pointers out of our allocated nodes that
 653	 * point to existing tnodes and the links between our allocated
 654	 * nodes.
 655	 */
 656	for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
 657		struct key_vector *inode = get_child(oldtnode, --i);
 658		struct key_vector *node0, *node1;
 659		unsigned long j, k;
 660
 661		/* An empty child */
 662		if (!inode)
 663			continue;
 664
 665		/* A leaf or an internal node with skipped bits */
 666		if (!tnode_full(oldtnode, inode)) {
 667			put_child(tn, get_index(inode->key, tn), inode);
 668			continue;
 669		}
 670
 671		/* drop the node in the old tnode free list */
 672		tnode_free_append(oldtnode, inode);
 673
 674		/* An internal node with two children */
 675		if (inode->bits == 1) {
 676			put_child(tn, 2 * i + 1, get_child(inode, 1));
 677			put_child(tn, 2 * i, get_child(inode, 0));
 678			continue;
 679		}
 680
 681		/* We will replace this node 'inode' with two new
 682		 * ones, 'node0' and 'node1', each with half of the
 683		 * original children. The two new nodes will have
 684		 * a position one bit further down the key and this
 685		 * means that the "significant" part of their keys
 686		 * (see the discussion near the top of this file)
 687		 * will differ by one bit, which will be "0" in
 688		 * node0's key and "1" in node1's key. Since we are
 689		 * moving the key position by one step, the bit that
 690		 * we are moving away from - the bit at position
 691		 * (tn->pos) - is the one that will differ between
 692		 * node0 and node1. So... we synthesize that bit in the
 693		 * two new keys.
 694		 */
 695		node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
 696		if (!node1)
 697			goto nomem;
 698		node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
 699
 700		tnode_free_append(tn, node1);
 701		if (!node0)
 702			goto nomem;
 703		tnode_free_append(tn, node0);
 704
 705		/* populate child pointers in new nodes */
 706		for (k = child_length(inode), j = k / 2; j;) {
 707			put_child(node1, --j, get_child(inode, --k));
 708			put_child(node0, j, get_child(inode, j));
 709			put_child(node1, --j, get_child(inode, --k));
 710			put_child(node0, j, get_child(inode, j));
 711		}
 712
 713		/* link new nodes to parent */
 714		NODE_INIT_PARENT(node1, tn);
 715		NODE_INIT_PARENT(node0, tn);
 716
 717		/* link parent to nodes */
 718		put_child(tn, 2 * i + 1, node1);
 719		put_child(tn, 2 * i, node0);
 720	}
 721
 722	/* setup the parent pointers into and out of this node */
 723	return replace(t, oldtnode, tn);
 724nomem:
 725	/* all pointers should be clean so we are done */
 726	tnode_free(tn);
 727notnode:
 728	return NULL;
 729}
 730
 731static struct key_vector *halve(struct trie *t,
 732				struct key_vector *oldtnode)
 733{
 734	struct key_vector *tn;
 735	unsigned long i;
 736
 737	pr_debug("In halve\n");
 738
 739	tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
 740	if (!tn)
 741		goto notnode;
 742
 743	/* prepare oldtnode to be freed */
 744	tnode_free_init(oldtnode);
 745
 746	/* Assemble all of the pointers in our cluster, in this case that
 747	 * represents all of the pointers out of our allocated nodes that
 748	 * point to existing tnodes and the links between our allocated
 749	 * nodes.
 750	 */
 751	for (i = child_length(oldtnode); i;) {
 752		struct key_vector *node1 = get_child(oldtnode, --i);
 753		struct key_vector *node0 = get_child(oldtnode, --i);
 754		struct key_vector *inode;
 755
 756		/* At least one of the children is empty */
 757		if (!node1 || !node0) {
 758			put_child(tn, i / 2, node1 ? : node0);
 759			continue;
 760		}
 761
 762		/* Two nonempty children */
 763		inode = tnode_new(node0->key, oldtnode->pos, 1);
 764		if (!inode)
 765			goto nomem;
 766		tnode_free_append(tn, inode);
 767
 768		/* initialize pointers out of node */
 769		put_child(inode, 1, node1);
 770		put_child(inode, 0, node0);
 771		NODE_INIT_PARENT(inode, tn);
 772
 773		/* link parent to node */
 774		put_child(tn, i / 2, inode);
 775	}
 776
 777	/* setup the parent pointers into and out of this node */
 778	return replace(t, oldtnode, tn);
 779nomem:
 780	/* all pointers should be clean so we are done */
 781	tnode_free(tn);
 782notnode:
 783	return NULL;
 784}
 785
 786static struct key_vector *collapse(struct trie *t,
 787				   struct key_vector *oldtnode)
 788{
 789	struct key_vector *n, *tp;
 790	unsigned long i;
 791
 792	/* scan the tnode looking for that one child that might still exist */
 793	for (n = NULL, i = child_length(oldtnode); !n && i;)
 794		n = get_child(oldtnode, --i);
 795
 796	/* compress one level */
 797	tp = node_parent(oldtnode);
 798	put_child_root(tp, oldtnode->key, n);
 799	node_set_parent(n, tp);
 800
 801	/* drop dead node */
 802	node_free(oldtnode);
 803
 804	return tp;
 805}
 806
 807static unsigned char update_suffix(struct key_vector *tn)
 808{
 809	unsigned char slen = tn->pos;
 810	unsigned long stride, i;
 811	unsigned char slen_max;
 812
 813	/* only vector 0 can have a suffix length greater than or equal to
 814	 * tn->pos + tn->bits, the second highest node will have a suffix
 815	 * length at most of tn->pos + tn->bits - 1
 816	 */
 817	slen_max = min_t(unsigned char, tn->pos + tn->bits - 1, tn->slen);
 818
 819	/* search though the list of children looking for nodes that might
 820	 * have a suffix greater than the one we currently have.  This is
 821	 * why we start with a stride of 2 since a stride of 1 would
 822	 * represent the nodes with suffix length equal to tn->pos
 823	 */
 824	for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
 825		struct key_vector *n = get_child(tn, i);
 826
 827		if (!n || (n->slen <= slen))
 828			continue;
 829
 830		/* update stride and slen based on new value */
 831		stride <<= (n->slen - slen);
 832		slen = n->slen;
 833		i &= ~(stride - 1);
 834
 835		/* stop searching if we have hit the maximum possible value */
 836		if (slen >= slen_max)
 837			break;
 838	}
 839
 840	tn->slen = slen;
 841
 842	return slen;
 843}
 844
 845/* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
 846 * the Helsinki University of Technology and Matti Tikkanen of Nokia
 847 * Telecommunications, page 6:
 848 * "A node is doubled if the ratio of non-empty children to all
 849 * children in the *doubled* node is at least 'high'."
 850 *
 851 * 'high' in this instance is the variable 'inflate_threshold'. It
 852 * is expressed as a percentage, so we multiply it with
 853 * child_length() and instead of multiplying by 2 (since the
 854 * child array will be doubled by inflate()) and multiplying
 855 * the left-hand side by 100 (to handle the percentage thing) we
 856 * multiply the left-hand side by 50.
 857 *
 858 * The left-hand side may look a bit weird: child_length(tn)
 859 * - tn->empty_children is of course the number of non-null children
 860 * in the current node. tn->full_children is the number of "full"
 861 * children, that is non-null tnodes with a skip value of 0.
 862 * All of those will be doubled in the resulting inflated tnode, so
 863 * we just count them one extra time here.
 864 *
 865 * A clearer way to write this would be:
 866 *
 867 * to_be_doubled = tn->full_children;
 868 * not_to_be_doubled = child_length(tn) - tn->empty_children -
 869 *     tn->full_children;
 870 *
 871 * new_child_length = child_length(tn) * 2;
 872 *
 873 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
 874 *      new_child_length;
 875 * if (new_fill_factor >= inflate_threshold)
 876 *
 877 * ...and so on, tho it would mess up the while () loop.
 878 *
 879 * anyway,
 880 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
 881 *      inflate_threshold
 882 *
 883 * avoid a division:
 884 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
 885 *      inflate_threshold * new_child_length
 886 *
 887 * expand not_to_be_doubled and to_be_doubled, and shorten:
 888 * 100 * (child_length(tn) - tn->empty_children +
 889 *    tn->full_children) >= inflate_threshold * new_child_length
 890 *
 891 * expand new_child_length:
 892 * 100 * (child_length(tn) - tn->empty_children +
 893 *    tn->full_children) >=
 894 *      inflate_threshold * child_length(tn) * 2
 895 *
 896 * shorten again:
 897 * 50 * (tn->full_children + child_length(tn) -
 898 *    tn->empty_children) >= inflate_threshold *
 899 *    child_length(tn)
 900 *
 901 */
 902static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
 903{
 904	unsigned long used = child_length(tn);
 905	unsigned long threshold = used;
 906
 907	/* Keep root node larger */
 908	threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
 909	used -= tn_info(tn)->empty_children;
 910	used += tn_info(tn)->full_children;
 911
 912	/* if bits == KEYLENGTH then pos = 0, and will fail below */
 913
 914	return (used > 1) && tn->pos && ((50 * used) >= threshold);
 915}
 916
 917static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
 918{
 919	unsigned long used = child_length(tn);
 920	unsigned long threshold = used;
 921
 922	/* Keep root node larger */
 923	threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
 924	used -= tn_info(tn)->empty_children;
 925
 926	/* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
 927
 928	return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
 929}
 930
 931static inline bool should_collapse(struct key_vector *tn)
 932{
 933	unsigned long used = child_length(tn);
 934
 935	used -= tn_info(tn)->empty_children;
 936
 937	/* account for bits == KEYLENGTH case */
 938	if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
 939		used -= KEY_MAX;
 940
 941	/* One child or none, time to drop us from the trie */
 942	return used < 2;
 943}
 944
 945#define MAX_WORK 10
 946static struct key_vector *resize(struct trie *t, struct key_vector *tn)
 947{
 948#ifdef CONFIG_IP_FIB_TRIE_STATS
 949	struct trie_use_stats __percpu *stats = t->stats;
 950#endif
 951	struct key_vector *tp = node_parent(tn);
 952	unsigned long cindex = get_index(tn->key, tp);
 953	int max_work = MAX_WORK;
 954
 955	pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
 956		 tn, inflate_threshold, halve_threshold);
 957
 958	/* track the tnode via the pointer from the parent instead of
 959	 * doing it ourselves.  This way we can let RCU fully do its
 960	 * thing without us interfering
 961	 */
 962	BUG_ON(tn != get_child(tp, cindex));
 963
 964	/* Double as long as the resulting node has a number of
 965	 * nonempty nodes that are above the threshold.
 966	 */
 967	while (should_inflate(tp, tn) && max_work) {
 968		tp = inflate(t, tn);
 969		if (!tp) {
 970#ifdef CONFIG_IP_FIB_TRIE_STATS
 971			this_cpu_inc(stats->resize_node_skipped);
 972#endif
 973			break;
 974		}
 975
 976		max_work--;
 977		tn = get_child(tp, cindex);
 978	}
 979
 980	/* update parent in case inflate failed */
 981	tp = node_parent(tn);
 982
 983	/* Return if at least one inflate is run */
 984	if (max_work != MAX_WORK)
 985		return tp;
 986
 987	/* Halve as long as the number of empty children in this
 988	 * node is above threshold.
 989	 */
 990	while (should_halve(tp, tn) && max_work) {
 991		tp = halve(t, tn);
 992		if (!tp) {
 993#ifdef CONFIG_IP_FIB_TRIE_STATS
 994			this_cpu_inc(stats->resize_node_skipped);
 995#endif
 996			break;
 997		}
 998
 999		max_work--;
1000		tn = get_child(tp, cindex);
1001	}
1002
1003	/* Only one child remains */
1004	if (should_collapse(tn))
1005		return collapse(t, tn);
1006
1007	/* update parent in case halve failed */
1008	return node_parent(tn);
1009}
1010
1011static void node_pull_suffix(struct key_vector *tn, unsigned char slen)
1012{
1013	unsigned char node_slen = tn->slen;
1014
1015	while ((node_slen > tn->pos) && (node_slen > slen)) {
1016		slen = update_suffix(tn);
1017		if (node_slen == slen)
1018			break;
1019
1020		tn = node_parent(tn);
1021		node_slen = tn->slen;
1022	}
1023}
1024
1025static void node_push_suffix(struct key_vector *tn, unsigned char slen)
1026{
1027	while (tn->slen < slen) {
1028		tn->slen = slen;
1029		tn = node_parent(tn);
1030	}
1031}
1032
1033/* rcu_read_lock needs to be hold by caller from readside */
1034static struct key_vector *fib_find_node(struct trie *t,
1035					struct key_vector **tp, u32 key)
1036{
1037	struct key_vector *pn, *n = t->kv;
1038	unsigned long index = 0;
1039
1040	do {
1041		pn = n;
1042		n = get_child_rcu(n, index);
1043
1044		if (!n)
1045			break;
1046
1047		index = get_cindex(key, n);
1048
1049		/* This bit of code is a bit tricky but it combines multiple
1050		 * checks into a single check.  The prefix consists of the
1051		 * prefix plus zeros for the bits in the cindex. The index
1052		 * is the difference between the key and this value.  From
1053		 * this we can actually derive several pieces of data.
1054		 *   if (index >= (1ul << bits))
1055		 *     we have a mismatch in skip bits and failed
1056		 *   else
1057		 *     we know the value is cindex
1058		 *
1059		 * This check is safe even if bits == KEYLENGTH due to the
1060		 * fact that we can only allocate a node with 32 bits if a
1061		 * long is greater than 32 bits.
1062		 */
1063		if (index >= (1ul << n->bits)) {
1064			n = NULL;
1065			break;
1066		}
1067
1068		/* keep searching until we find a perfect match leaf or NULL */
1069	} while (IS_TNODE(n));
1070
1071	*tp = pn;
1072
1073	return n;
1074}
1075
1076/* Return the first fib alias matching TOS with
1077 * priority less than or equal to PRIO.
 
 
1078 */
1079static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
1080					u8 tos, u32 prio, u32 tb_id)
 
1081{
1082	struct fib_alias *fa;
1083
1084	if (!fah)
1085		return NULL;
1086
1087	hlist_for_each_entry(fa, fah, fa_list) {
 
 
 
 
1088		if (fa->fa_slen < slen)
1089			continue;
1090		if (fa->fa_slen != slen)
1091			break;
1092		if (fa->tb_id > tb_id)
1093			continue;
1094		if (fa->tb_id != tb_id)
1095			break;
1096		if (fa->fa_tos > tos)
 
 
1097			continue;
1098		if (fa->fa_info->fib_priority >= prio || fa->fa_tos < tos)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1099			return fa;
1100	}
1101
1102	return NULL;
1103}
1104
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1105static void trie_rebalance(struct trie *t, struct key_vector *tn)
1106{
1107	while (!IS_TRIE(tn))
1108		tn = resize(t, tn);
1109}
1110
1111static int fib_insert_node(struct trie *t, struct key_vector *tp,
1112			   struct fib_alias *new, t_key key)
1113{
1114	struct key_vector *n, *l;
1115
1116	l = leaf_new(key, new);
1117	if (!l)
1118		goto noleaf;
1119
1120	/* retrieve child from parent node */
1121	n = get_child(tp, get_index(key, tp));
1122
1123	/* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1124	 *
1125	 *  Add a new tnode here
1126	 *  first tnode need some special handling
1127	 *  leaves us in position for handling as case 3
1128	 */
1129	if (n) {
1130		struct key_vector *tn;
1131
1132		tn = tnode_new(key, __fls(key ^ n->key), 1);
1133		if (!tn)
1134			goto notnode;
1135
1136		/* initialize routes out of node */
1137		NODE_INIT_PARENT(tn, tp);
1138		put_child(tn, get_index(key, tn) ^ 1, n);
1139
1140		/* start adding routes into the node */
1141		put_child_root(tp, key, tn);
1142		node_set_parent(n, tn);
1143
1144		/* parent now has a NULL spot where the leaf can go */
1145		tp = tn;
1146	}
1147
1148	/* Case 3: n is NULL, and will just insert a new leaf */
1149	node_push_suffix(tp, new->fa_slen);
1150	NODE_INIT_PARENT(l, tp);
1151	put_child_root(tp, key, l);
1152	trie_rebalance(t, tp);
1153
1154	return 0;
1155notnode:
1156	node_free(l);
1157noleaf:
1158	return -ENOMEM;
1159}
1160
1161static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1162			    struct key_vector *l, struct fib_alias *new,
1163			    struct fib_alias *fa, t_key key)
1164{
1165	if (!l)
1166		return fib_insert_node(t, tp, new, key);
1167
1168	if (fa) {
1169		hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1170	} else {
1171		struct fib_alias *last;
1172
1173		hlist_for_each_entry(last, &l->leaf, fa_list) {
1174			if (new->fa_slen < last->fa_slen)
1175				break;
1176			if ((new->fa_slen == last->fa_slen) &&
1177			    (new->tb_id > last->tb_id))
1178				break;
1179			fa = last;
1180		}
1181
1182		if (fa)
1183			hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1184		else
1185			hlist_add_head_rcu(&new->fa_list, &l->leaf);
1186	}
1187
1188	/* if we added to the tail node then we need to update slen */
1189	if (l->slen < new->fa_slen) {
1190		l->slen = new->fa_slen;
1191		node_push_suffix(tp, new->fa_slen);
1192	}
1193
1194	return 0;
1195}
1196
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1197/* Caller must hold RTNL. */
1198int fib_table_insert(struct net *net, struct fib_table *tb,
1199		     struct fib_config *cfg)
1200{
1201	struct trie *t = (struct trie *)tb->tb_data;
1202	struct fib_alias *fa, *new_fa;
1203	struct key_vector *l, *tp;
1204	u16 nlflags = NLM_F_EXCL;
1205	struct fib_info *fi;
1206	u8 plen = cfg->fc_dst_len;
1207	u8 slen = KEYLENGTH - plen;
1208	u8 tos = cfg->fc_tos;
1209	u32 key;
1210	int err;
1211
1212	if (plen > KEYLENGTH)
 
 
1213		return -EINVAL;
1214
1215	key = ntohl(cfg->fc_dst);
1216
1217	pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1218
1219	if ((plen < KEYLENGTH) && (key << plen))
1220		return -EINVAL;
1221
1222	fi = fib_create_info(cfg);
1223	if (IS_ERR(fi)) {
1224		err = PTR_ERR(fi);
1225		goto err;
1226	}
1227
 
1228	l = fib_find_node(t, &tp, key);
1229	fa = l ? fib_find_alias(&l->leaf, slen, tos, fi->fib_priority,
1230				tb->tb_id) : NULL;
1231
1232	/* Now fa, if non-NULL, points to the first fib alias
1233	 * with the same keys [prefix,tos,priority], if such key already
1234	 * exists or to the node before which we will insert new one.
1235	 *
1236	 * If fa is NULL, we will need to allocate a new one and
1237	 * insert to the tail of the section matching the suffix length
1238	 * of the new alias.
1239	 */
1240
1241	if (fa && fa->fa_tos == tos &&
1242	    fa->fa_info->fib_priority == fi->fib_priority) {
1243		struct fib_alias *fa_first, *fa_match;
1244
1245		err = -EEXIST;
1246		if (cfg->fc_nlflags & NLM_F_EXCL)
1247			goto out;
1248
1249		nlflags &= ~NLM_F_EXCL;
1250
1251		/* We have 2 goals:
1252		 * 1. Find exact match for type, scope, fib_info to avoid
1253		 * duplicate routes
1254		 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1255		 */
1256		fa_match = NULL;
1257		fa_first = fa;
1258		hlist_for_each_entry_from(fa, fa_list) {
1259			if ((fa->fa_slen != slen) ||
1260			    (fa->tb_id != tb->tb_id) ||
1261			    (fa->fa_tos != tos))
1262				break;
1263			if (fa->fa_info->fib_priority != fi->fib_priority)
1264				break;
1265			if (fa->fa_type == cfg->fc_type &&
1266			    fa->fa_info == fi) {
1267				fa_match = fa;
1268				break;
1269			}
1270		}
1271
1272		if (cfg->fc_nlflags & NLM_F_REPLACE) {
1273			struct fib_info *fi_drop;
1274			u8 state;
1275
1276			nlflags |= NLM_F_REPLACE;
1277			fa = fa_first;
1278			if (fa_match) {
1279				if (fa == fa_match)
1280					err = 0;
1281				goto out;
1282			}
1283			err = -ENOBUFS;
1284			new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1285			if (!new_fa)
1286				goto out;
1287
1288			fi_drop = fa->fa_info;
1289			new_fa->fa_tos = fa->fa_tos;
1290			new_fa->fa_info = fi;
1291			new_fa->fa_type = cfg->fc_type;
1292			state = fa->fa_state;
1293			new_fa->fa_state = state & ~FA_S_ACCESSED;
1294			new_fa->fa_slen = fa->fa_slen;
1295			new_fa->tb_id = tb->tb_id;
1296			new_fa->fa_default = -1;
 
 
 
1297
1298			hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1299
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1300			alias_free_mem_rcu(fa);
1301
1302			fib_release_info(fi_drop);
1303			if (state & FA_S_ACCESSED)
1304				rt_cache_flush(cfg->fc_nlinfo.nl_net);
1305
1306			call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_ADD,
1307						 key, plen, fi,
1308						 new_fa->fa_tos, cfg->fc_type,
1309						 tb->tb_id, cfg->fc_nlflags);
1310			rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1311				tb->tb_id, &cfg->fc_nlinfo, nlflags);
1312
1313			goto succeeded;
1314		}
1315		/* Error if we find a perfect match which
1316		 * uses the same scope, type, and nexthop
1317		 * information.
1318		 */
1319		if (fa_match)
1320			goto out;
1321
1322		if (cfg->fc_nlflags & NLM_F_APPEND)
1323			nlflags |= NLM_F_APPEND;
1324		else
1325			fa = fa_first;
1326	}
1327	err = -ENOENT;
1328	if (!(cfg->fc_nlflags & NLM_F_CREATE))
1329		goto out;
1330
1331	nlflags |= NLM_F_CREATE;
1332	err = -ENOBUFS;
1333	new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1334	if (!new_fa)
1335		goto out;
1336
1337	new_fa->fa_info = fi;
1338	new_fa->fa_tos = tos;
1339	new_fa->fa_type = cfg->fc_type;
1340	new_fa->fa_state = 0;
1341	new_fa->fa_slen = slen;
1342	new_fa->tb_id = tb->tb_id;
1343	new_fa->fa_default = -1;
 
 
 
1344
1345	/* Insert new entry to the list. */
1346	err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1347	if (err)
1348		goto out_free_new_fa;
1349
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1350	if (!plen)
1351		tb->tb_num_default++;
1352
1353	rt_cache_flush(cfg->fc_nlinfo.nl_net);
1354	call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_ADD, key, plen, fi, tos,
1355				 cfg->fc_type, tb->tb_id, cfg->fc_nlflags);
1356	rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1357		  &cfg->fc_nlinfo, nlflags);
1358succeeded:
1359	return 0;
1360
 
 
1361out_free_new_fa:
1362	kmem_cache_free(fn_alias_kmem, new_fa);
1363out:
1364	fib_release_info(fi);
1365err:
1366	return err;
1367}
1368
1369static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1370{
1371	t_key prefix = n->key;
1372
1373	return (key ^ prefix) & (prefix | -prefix);
1374}
1375
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1376/* should be called with rcu_read_lock */
1377int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1378		     struct fib_result *res, int fib_flags)
1379{
1380	struct trie *t = (struct trie *) tb->tb_data;
1381#ifdef CONFIG_IP_FIB_TRIE_STATS
1382	struct trie_use_stats __percpu *stats = t->stats;
1383#endif
1384	const t_key key = ntohl(flp->daddr);
1385	struct key_vector *n, *pn;
1386	struct fib_alias *fa;
1387	unsigned long index;
1388	t_key cindex;
1389
1390	trace_fib_table_lookup(tb->tb_id, flp);
1391
1392	pn = t->kv;
1393	cindex = 0;
1394
1395	n = get_child_rcu(pn, cindex);
1396	if (!n)
 
1397		return -EAGAIN;
 
1398
1399#ifdef CONFIG_IP_FIB_TRIE_STATS
1400	this_cpu_inc(stats->gets);
1401#endif
1402
1403	/* Step 1: Travel to the longest prefix match in the trie */
1404	for (;;) {
1405		index = get_cindex(key, n);
1406
1407		/* This bit of code is a bit tricky but it combines multiple
1408		 * checks into a single check.  The prefix consists of the
1409		 * prefix plus zeros for the "bits" in the prefix. The index
1410		 * is the difference between the key and this value.  From
1411		 * this we can actually derive several pieces of data.
1412		 *   if (index >= (1ul << bits))
1413		 *     we have a mismatch in skip bits and failed
1414		 *   else
1415		 *     we know the value is cindex
1416		 *
1417		 * This check is safe even if bits == KEYLENGTH due to the
1418		 * fact that we can only allocate a node with 32 bits if a
1419		 * long is greater than 32 bits.
1420		 */
1421		if (index >= (1ul << n->bits))
1422			break;
1423
1424		/* we have found a leaf. Prefixes have already been compared */
1425		if (IS_LEAF(n))
1426			goto found;
1427
1428		/* only record pn and cindex if we are going to be chopping
1429		 * bits later.  Otherwise we are just wasting cycles.
1430		 */
1431		if (n->slen > n->pos) {
1432			pn = n;
1433			cindex = index;
1434		}
1435
1436		n = get_child_rcu(n, index);
1437		if (unlikely(!n))
1438			goto backtrace;
1439	}
1440
1441	/* Step 2: Sort out leaves and begin backtracing for longest prefix */
1442	for (;;) {
1443		/* record the pointer where our next node pointer is stored */
1444		struct key_vector __rcu **cptr = n->tnode;
1445
1446		/* This test verifies that none of the bits that differ
1447		 * between the key and the prefix exist in the region of
1448		 * the lsb and higher in the prefix.
1449		 */
1450		if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1451			goto backtrace;
1452
1453		/* exit out and process leaf */
1454		if (unlikely(IS_LEAF(n)))
1455			break;
1456
1457		/* Don't bother recording parent info.  Since we are in
1458		 * prefix match mode we will have to come back to wherever
1459		 * we started this traversal anyway
1460		 */
1461
1462		while ((n = rcu_dereference(*cptr)) == NULL) {
1463backtrace:
1464#ifdef CONFIG_IP_FIB_TRIE_STATS
1465			if (!n)
1466				this_cpu_inc(stats->null_node_hit);
1467#endif
1468			/* If we are at cindex 0 there are no more bits for
1469			 * us to strip at this level so we must ascend back
1470			 * up one level to see if there are any more bits to
1471			 * be stripped there.
1472			 */
1473			while (!cindex) {
1474				t_key pkey = pn->key;
1475
1476				/* If we don't have a parent then there is
1477				 * nothing for us to do as we do not have any
1478				 * further nodes to parse.
1479				 */
1480				if (IS_TRIE(pn))
 
 
1481					return -EAGAIN;
 
1482#ifdef CONFIG_IP_FIB_TRIE_STATS
1483				this_cpu_inc(stats->backtrack);
1484#endif
1485				/* Get Child's index */
1486				pn = node_parent_rcu(pn);
1487				cindex = get_index(pkey, pn);
1488			}
1489
1490			/* strip the least significant bit from the cindex */
1491			cindex &= cindex - 1;
1492
1493			/* grab pointer for next child node */
1494			cptr = &pn->tnode[cindex];
1495		}
1496	}
1497
1498found:
1499	/* this line carries forward the xor from earlier in the function */
1500	index = key ^ n->key;
1501
1502	/* Step 3: Process the leaf, if that fails fall back to backtracing */
1503	hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1504		struct fib_info *fi = fa->fa_info;
 
1505		int nhsel, err;
1506
1507		if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1508			if (index >= (1ul << fa->fa_slen))
1509				continue;
1510		}
1511		if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1512			continue;
1513		if (fi->fib_dead)
 
1514			continue;
1515		if (fa->fa_info->fib_scope < flp->flowi4_scope)
1516			continue;
1517		fib_alias_accessed(fa);
1518		err = fib_props[fa->fa_type].error;
1519		if (unlikely(err < 0)) {
 
1520#ifdef CONFIG_IP_FIB_TRIE_STATS
1521			this_cpu_inc(stats->semantic_match_passed);
1522#endif
 
1523			return err;
1524		}
1525		if (fi->fib_flags & RTNH_F_DEAD)
1526			continue;
1527		for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
1528			const struct fib_nh *nh = &fi->fib_nh[nhsel];
1529			struct in_device *in_dev = __in_dev_get_rcu(nh->nh_dev);
1530
1531			if (nh->nh_flags & RTNH_F_DEAD)
1532				continue;
1533			if (in_dev &&
1534			    IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1535			    nh->nh_flags & RTNH_F_LINKDOWN &&
1536			    !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1537				continue;
1538			if (!(flp->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF)) {
1539				if (flp->flowi4_oif &&
1540				    flp->flowi4_oif != nh->nh_oif)
1541					continue;
1542			}
1543
 
 
 
 
 
 
 
 
 
 
 
 
 
1544			if (!(fib_flags & FIB_LOOKUP_NOREF))
1545				atomic_inc(&fi->fib_clntref);
1546
 
1547			res->prefixlen = KEYLENGTH - fa->fa_slen;
1548			res->nh_sel = nhsel;
 
1549			res->type = fa->fa_type;
1550			res->scope = fi->fib_scope;
 
1551			res->fi = fi;
1552			res->table = tb;
1553			res->fa_head = &n->leaf;
1554#ifdef CONFIG_IP_FIB_TRIE_STATS
1555			this_cpu_inc(stats->semantic_match_passed);
1556#endif
1557			trace_fib_table_lookup_nh(nh);
1558
1559			return err;
1560		}
1561	}
 
1562#ifdef CONFIG_IP_FIB_TRIE_STATS
1563	this_cpu_inc(stats->semantic_match_miss);
1564#endif
1565	goto backtrace;
1566}
1567EXPORT_SYMBOL_GPL(fib_table_lookup);
1568
1569static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1570			     struct key_vector *l, struct fib_alias *old)
1571{
1572	/* record the location of the previous list_info entry */
1573	struct hlist_node **pprev = old->fa_list.pprev;
1574	struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1575
1576	/* remove the fib_alias from the list */
1577	hlist_del_rcu(&old->fa_list);
1578
1579	/* if we emptied the list this leaf will be freed and we can sort
1580	 * out parent suffix lengths as a part of trie_rebalance
1581	 */
1582	if (hlist_empty(&l->leaf)) {
1583		if (tp->slen == l->slen)
1584			node_pull_suffix(tp, tp->pos);
1585		put_child_root(tp, l->key, NULL);
1586		node_free(l);
1587		trie_rebalance(t, tp);
1588		return;
1589	}
1590
1591	/* only access fa if it is pointing at the last valid hlist_node */
1592	if (*pprev)
1593		return;
1594
1595	/* update the trie with the latest suffix length */
1596	l->slen = fa->fa_slen;
1597	node_pull_suffix(tp, fa->fa_slen);
1598}
1599
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1600/* Caller must hold RTNL. */
1601int fib_table_delete(struct net *net, struct fib_table *tb,
1602		     struct fib_config *cfg)
1603{
1604	struct trie *t = (struct trie *) tb->tb_data;
1605	struct fib_alias *fa, *fa_to_delete;
1606	struct key_vector *l, *tp;
1607	u8 plen = cfg->fc_dst_len;
1608	u8 slen = KEYLENGTH - plen;
1609	u8 tos = cfg->fc_tos;
1610	u32 key;
1611
1612	if (plen > KEYLENGTH)
1613		return -EINVAL;
1614
1615	key = ntohl(cfg->fc_dst);
1616
1617	if ((plen < KEYLENGTH) && (key << plen))
1618		return -EINVAL;
1619
1620	l = fib_find_node(t, &tp, key);
1621	if (!l)
1622		return -ESRCH;
1623
1624	fa = fib_find_alias(&l->leaf, slen, tos, 0, tb->tb_id);
 
1625	if (!fa)
1626		return -ESRCH;
1627
1628	pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
 
1629
1630	fa_to_delete = NULL;
1631	hlist_for_each_entry_from(fa, fa_list) {
1632		struct fib_info *fi = fa->fa_info;
1633
1634		if ((fa->fa_slen != slen) ||
1635		    (fa->tb_id != tb->tb_id) ||
1636		    (fa->fa_tos != tos))
1637			break;
1638
1639		if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1640		    (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1641		     fa->fa_info->fib_scope == cfg->fc_scope) &&
1642		    (!cfg->fc_prefsrc ||
1643		     fi->fib_prefsrc == cfg->fc_prefsrc) &&
1644		    (!cfg->fc_protocol ||
1645		     fi->fib_protocol == cfg->fc_protocol) &&
1646		    fib_nh_match(cfg, fi) == 0) {
 
1647			fa_to_delete = fa;
1648			break;
1649		}
1650	}
1651
1652	if (!fa_to_delete)
1653		return -ESRCH;
1654
1655	call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, key, plen,
1656				 fa_to_delete->fa_info, tos, cfg->fc_type,
1657				 tb->tb_id, 0);
1658	rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1659		  &cfg->fc_nlinfo, 0);
1660
1661	if (!plen)
1662		tb->tb_num_default--;
1663
1664	fib_remove_alias(t, tp, l, fa_to_delete);
1665
1666	if (fa_to_delete->fa_state & FA_S_ACCESSED)
1667		rt_cache_flush(cfg->fc_nlinfo.nl_net);
1668
1669	fib_release_info(fa_to_delete->fa_info);
1670	alias_free_mem_rcu(fa_to_delete);
1671	return 0;
1672}
1673
1674/* Scan for the next leaf starting at the provided key value */
1675static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1676{
1677	struct key_vector *pn, *n = *tn;
1678	unsigned long cindex;
1679
1680	/* this loop is meant to try and find the key in the trie */
1681	do {
1682		/* record parent and next child index */
1683		pn = n;
1684		cindex = (key > pn->key) ? get_index(key, pn) : 0;
1685
1686		if (cindex >> pn->bits)
1687			break;
1688
1689		/* descend into the next child */
1690		n = get_child_rcu(pn, cindex++);
1691		if (!n)
1692			break;
1693
1694		/* guarantee forward progress on the keys */
1695		if (IS_LEAF(n) && (n->key >= key))
1696			goto found;
1697	} while (IS_TNODE(n));
1698
1699	/* this loop will search for the next leaf with a greater key */
1700	while (!IS_TRIE(pn)) {
1701		/* if we exhausted the parent node we will need to climb */
1702		if (cindex >= (1ul << pn->bits)) {
1703			t_key pkey = pn->key;
1704
1705			pn = node_parent_rcu(pn);
1706			cindex = get_index(pkey, pn) + 1;
1707			continue;
1708		}
1709
1710		/* grab the next available node */
1711		n = get_child_rcu(pn, cindex++);
1712		if (!n)
1713			continue;
1714
1715		/* no need to compare keys since we bumped the index */
1716		if (IS_LEAF(n))
1717			goto found;
1718
1719		/* Rescan start scanning in new node */
1720		pn = n;
1721		cindex = 0;
1722	}
1723
1724	*tn = pn;
1725	return NULL; /* Root of trie */
1726found:
1727	/* if we are at the limit for keys just return NULL for the tnode */
1728	*tn = pn;
1729	return n;
1730}
1731
1732static void fib_trie_free(struct fib_table *tb)
1733{
1734	struct trie *t = (struct trie *)tb->tb_data;
1735	struct key_vector *pn = t->kv;
1736	unsigned long cindex = 1;
1737	struct hlist_node *tmp;
1738	struct fib_alias *fa;
1739
1740	/* walk trie in reverse order and free everything */
1741	for (;;) {
1742		struct key_vector *n;
1743
1744		if (!(cindex--)) {
1745			t_key pkey = pn->key;
1746
1747			if (IS_TRIE(pn))
1748				break;
1749
1750			n = pn;
1751			pn = node_parent(pn);
1752
1753			/* drop emptied tnode */
1754			put_child_root(pn, n->key, NULL);
1755			node_free(n);
1756
1757			cindex = get_index(pkey, pn);
1758
1759			continue;
1760		}
1761
1762		/* grab the next available node */
1763		n = get_child(pn, cindex);
1764		if (!n)
1765			continue;
1766
1767		if (IS_TNODE(n)) {
1768			/* record pn and cindex for leaf walking */
1769			pn = n;
1770			cindex = 1ul << n->bits;
1771
1772			continue;
1773		}
1774
1775		hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1776			hlist_del_rcu(&fa->fa_list);
1777			alias_free_mem_rcu(fa);
1778		}
1779
1780		put_child_root(pn, n->key, NULL);
1781		node_free(n);
1782	}
1783
1784#ifdef CONFIG_IP_FIB_TRIE_STATS
1785	free_percpu(t->stats);
1786#endif
1787	kfree(tb);
1788}
1789
1790struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1791{
1792	struct trie *ot = (struct trie *)oldtb->tb_data;
1793	struct key_vector *l, *tp = ot->kv;
1794	struct fib_table *local_tb;
1795	struct fib_alias *fa;
1796	struct trie *lt;
1797	t_key key = 0;
1798
1799	if (oldtb->tb_data == oldtb->__data)
1800		return oldtb;
1801
1802	local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1803	if (!local_tb)
1804		return NULL;
1805
1806	lt = (struct trie *)local_tb->tb_data;
1807
1808	while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1809		struct key_vector *local_l = NULL, *local_tp;
1810
1811		hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1812			struct fib_alias *new_fa;
1813
1814			if (local_tb->tb_id != fa->tb_id)
1815				continue;
1816
1817			/* clone fa for new local table */
1818			new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1819			if (!new_fa)
1820				goto out;
1821
1822			memcpy(new_fa, fa, sizeof(*fa));
1823
1824			/* insert clone into table */
1825			if (!local_l)
1826				local_l = fib_find_node(lt, &local_tp, l->key);
1827
1828			if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1829					     NULL, l->key)) {
1830				kmem_cache_free(fn_alias_kmem, new_fa);
1831				goto out;
1832			}
1833		}
1834
1835		/* stop loop if key wrapped back to 0 */
1836		key = l->key + 1;
1837		if (key < l->key)
1838			break;
1839	}
1840
1841	return local_tb;
1842out:
1843	fib_trie_free(local_tb);
1844
1845	return NULL;
1846}
1847
1848/* Caller must hold RTNL */
1849void fib_table_flush_external(struct fib_table *tb)
1850{
1851	struct trie *t = (struct trie *)tb->tb_data;
1852	struct key_vector *pn = t->kv;
1853	unsigned long cindex = 1;
1854	struct hlist_node *tmp;
1855	struct fib_alias *fa;
1856
1857	/* walk trie in reverse order */
1858	for (;;) {
1859		unsigned char slen = 0;
1860		struct key_vector *n;
1861
1862		if (!(cindex--)) {
1863			t_key pkey = pn->key;
1864
1865			/* cannot resize the trie vector */
1866			if (IS_TRIE(pn))
1867				break;
1868
1869			/* update the suffix to address pulled leaves */
1870			if (pn->slen > pn->pos)
1871				update_suffix(pn);
1872
1873			/* resize completed node */
1874			pn = resize(t, pn);
1875			cindex = get_index(pkey, pn);
1876
1877			continue;
1878		}
1879
1880		/* grab the next available node */
1881		n = get_child(pn, cindex);
1882		if (!n)
1883			continue;
1884
1885		if (IS_TNODE(n)) {
1886			/* record pn and cindex for leaf walking */
1887			pn = n;
1888			cindex = 1ul << n->bits;
1889
1890			continue;
1891		}
1892
1893		hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1894			/* if alias was cloned to local then we just
1895			 * need to remove the local copy from main
1896			 */
1897			if (tb->tb_id != fa->tb_id) {
1898				hlist_del_rcu(&fa->fa_list);
1899				alias_free_mem_rcu(fa);
1900				continue;
1901			}
1902
1903			/* record local slen */
1904			slen = fa->fa_slen;
1905		}
1906
1907		/* update leaf slen */
1908		n->slen = slen;
1909
1910		if (hlist_empty(&n->leaf)) {
1911			put_child_root(pn, n->key, NULL);
1912			node_free(n);
1913		}
1914	}
1915}
1916
1917/* Caller must hold RTNL. */
1918int fib_table_flush(struct net *net, struct fib_table *tb)
1919{
1920	struct trie *t = (struct trie *)tb->tb_data;
 
1921	struct key_vector *pn = t->kv;
1922	unsigned long cindex = 1;
1923	struct hlist_node *tmp;
1924	struct fib_alias *fa;
1925	int found = 0;
1926
1927	/* walk trie in reverse order */
1928	for (;;) {
1929		unsigned char slen = 0;
1930		struct key_vector *n;
1931
1932		if (!(cindex--)) {
1933			t_key pkey = pn->key;
1934
1935			/* cannot resize the trie vector */
1936			if (IS_TRIE(pn))
1937				break;
1938
1939			/* update the suffix to address pulled leaves */
1940			if (pn->slen > pn->pos)
1941				update_suffix(pn);
1942
1943			/* resize completed node */
1944			pn = resize(t, pn);
1945			cindex = get_index(pkey, pn);
1946
1947			continue;
1948		}
1949
1950		/* grab the next available node */
1951		n = get_child(pn, cindex);
1952		if (!n)
1953			continue;
1954
1955		if (IS_TNODE(n)) {
1956			/* record pn and cindex for leaf walking */
1957			pn = n;
1958			cindex = 1ul << n->bits;
1959
1960			continue;
1961		}
1962
1963		hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1964			struct fib_info *fi = fa->fa_info;
1965
1966			if (!fi || !(fi->fib_flags & RTNH_F_DEAD)) {
 
 
1967				slen = fa->fa_slen;
1968				continue;
1969			}
1970
1971			call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL,
1972						 n->key,
1973						 KEYLENGTH - fa->fa_slen,
1974						 fi, fa->fa_tos, fa->fa_type,
1975						 tb->tb_id, 0);
 
 
 
 
 
 
 
 
1976			hlist_del_rcu(&fa->fa_list);
1977			fib_release_info(fa->fa_info);
1978			alias_free_mem_rcu(fa);
1979			found++;
1980		}
1981
1982		/* update leaf slen */
1983		n->slen = slen;
1984
1985		if (hlist_empty(&n->leaf)) {
1986			put_child_root(pn, n->key, NULL);
1987			node_free(n);
1988		}
1989	}
1990
1991	pr_debug("trie_flush found=%d\n", found);
1992	return found;
1993}
1994
1995static void fib_leaf_notify(struct net *net, struct key_vector *l,
1996			    struct fib_table *tb, struct notifier_block *nb,
1997			    enum fib_event_type event_type)
1998{
 
 
 
1999	struct fib_alias *fa;
2000
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2001	hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2002		struct fib_info *fi = fa->fa_info;
2003
2004		if (!fi)
2005			continue;
2006
2007		/* local and main table can share the same trie,
2008		 * so don't notify twice for the same entry.
2009		 */
2010		if (tb->tb_id != fa->tb_id)
2011			continue;
2012
2013		call_fib_entry_notifier(nb, net, event_type, l->key,
2014					KEYLENGTH - fa->fa_slen, fi, fa->fa_tos,
2015					fa->fa_type, fa->tb_id, 0);
 
 
 
 
 
 
2016	}
 
2017}
2018
2019static void fib_table_notify(struct net *net, struct fib_table *tb,
2020			     struct notifier_block *nb,
2021			     enum fib_event_type event_type)
2022{
2023	struct trie *t = (struct trie *)tb->tb_data;
2024	struct key_vector *l, *tp = t->kv;
2025	t_key key = 0;
 
2026
2027	while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2028		fib_leaf_notify(net, l, tb, nb, event_type);
 
 
2029
2030		key = l->key + 1;
2031		/* stop in case of wrap around */
2032		if (key < l->key)
2033			break;
2034	}
 
2035}
2036
2037static void fib_notify(struct net *net, struct notifier_block *nb,
2038		       enum fib_event_type event_type)
2039{
2040	unsigned int h;
 
2041
2042	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2043		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2044		struct fib_table *tb;
2045
2046		hlist_for_each_entry_rcu(tb, head, tb_hlist)
2047			fib_table_notify(net, tb, nb, event_type);
 
 
 
2048	}
 
2049}
2050
2051static void __trie_free_rcu(struct rcu_head *head)
2052{
2053	struct fib_table *tb = container_of(head, struct fib_table, rcu);
2054#ifdef CONFIG_IP_FIB_TRIE_STATS
2055	struct trie *t = (struct trie *)tb->tb_data;
2056
2057	if (tb->tb_data == tb->__data)
2058		free_percpu(t->stats);
2059#endif /* CONFIG_IP_FIB_TRIE_STATS */
2060	kfree(tb);
2061}
2062
2063void fib_free_table(struct fib_table *tb)
2064{
2065	call_rcu(&tb->rcu, __trie_free_rcu);
2066}
2067
2068static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
2069			     struct sk_buff *skb, struct netlink_callback *cb)
 
2070{
 
2071	__be32 xkey = htonl(l->key);
 
2072	struct fib_alias *fa;
2073	int i, s_i;
 
 
 
2074
2075	s_i = cb->args[4];
 
2076	i = 0;
2077
2078	/* rcu_read_lock is hold by caller */
2079	hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2080		if (i < s_i) {
2081			i++;
2082			continue;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2083		}
2084
2085		if (tb->tb_id != fa->tb_id) {
2086			i++;
2087			continue;
 
 
2088		}
2089
2090		if (fib_dump_info(skb, NETLINK_CB(cb->skb).portid,
2091				  cb->nlh->nlmsg_seq,
2092				  RTM_NEWROUTE,
2093				  tb->tb_id,
2094				  fa->fa_type,
2095				  xkey,
2096				  KEYLENGTH - fa->fa_slen,
2097				  fa->fa_tos,
2098				  fa->fa_info, NLM_F_MULTI) < 0) {
2099			cb->args[4] = i;
2100			return -1;
2101		}
2102		i++;
2103	}
2104
2105	cb->args[4] = i;
2106	return skb->len;
 
 
 
 
 
2107}
2108
2109/* rcu_read_lock needs to be hold by caller from readside */
2110int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
2111		   struct netlink_callback *cb)
2112{
2113	struct trie *t = (struct trie *)tb->tb_data;
2114	struct key_vector *l, *tp = t->kv;
2115	/* Dump starting at last key.
2116	 * Note: 0.0.0.0/0 (ie default) is first key.
2117	 */
2118	int count = cb->args[2];
2119	t_key key = cb->args[3];
2120
 
 
 
 
 
 
2121	while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2122		if (fn_trie_dump_leaf(l, tb, skb, cb) < 0) {
 
 
 
2123			cb->args[3] = key;
2124			cb->args[2] = count;
2125			return -1;
2126		}
2127
2128		++count;
2129		key = l->key + 1;
2130
2131		memset(&cb->args[4], 0,
2132		       sizeof(cb->args) - 4*sizeof(cb->args[0]));
2133
2134		/* stop loop if key wrapped back to 0 */
2135		if (key < l->key)
2136			break;
2137	}
2138
2139	cb->args[3] = key;
2140	cb->args[2] = count;
2141
2142	return skb->len;
2143}
2144
2145void __init fib_trie_init(void)
2146{
2147	fn_alias_kmem = kmem_cache_create("ip_fib_alias",
2148					  sizeof(struct fib_alias),
2149					  0, SLAB_PANIC, NULL);
2150
2151	trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
2152					   LEAF_SIZE,
2153					   0, SLAB_PANIC, NULL);
2154}
2155
2156struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
2157{
2158	struct fib_table *tb;
2159	struct trie *t;
2160	size_t sz = sizeof(*tb);
2161
2162	if (!alias)
2163		sz += sizeof(struct trie);
2164
2165	tb = kzalloc(sz, GFP_KERNEL);
2166	if (!tb)
2167		return NULL;
2168
2169	tb->tb_id = id;
2170	tb->tb_num_default = 0;
2171	tb->tb_data = (alias ? alias->__data : tb->__data);
2172
2173	if (alias)
2174		return tb;
2175
2176	t = (struct trie *) tb->tb_data;
2177	t->kv[0].pos = KEYLENGTH;
2178	t->kv[0].slen = KEYLENGTH;
2179#ifdef CONFIG_IP_FIB_TRIE_STATS
2180	t->stats = alloc_percpu(struct trie_use_stats);
2181	if (!t->stats) {
2182		kfree(tb);
2183		tb = NULL;
2184	}
2185#endif
2186
2187	return tb;
2188}
2189
2190#ifdef CONFIG_PROC_FS
2191/* Depth first Trie walk iterator */
2192struct fib_trie_iter {
2193	struct seq_net_private p;
2194	struct fib_table *tb;
2195	struct key_vector *tnode;
2196	unsigned int index;
2197	unsigned int depth;
2198};
2199
2200static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2201{
2202	unsigned long cindex = iter->index;
2203	struct key_vector *pn = iter->tnode;
2204	t_key pkey;
2205
2206	pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2207		 iter->tnode, iter->index, iter->depth);
2208
2209	while (!IS_TRIE(pn)) {
2210		while (cindex < child_length(pn)) {
2211			struct key_vector *n = get_child_rcu(pn, cindex++);
2212
2213			if (!n)
2214				continue;
2215
2216			if (IS_LEAF(n)) {
2217				iter->tnode = pn;
2218				iter->index = cindex;
2219			} else {
2220				/* push down one level */
2221				iter->tnode = n;
2222				iter->index = 0;
2223				++iter->depth;
2224			}
2225
2226			return n;
2227		}
2228
2229		/* Current node exhausted, pop back up */
2230		pkey = pn->key;
2231		pn = node_parent_rcu(pn);
2232		cindex = get_index(pkey, pn) + 1;
2233		--iter->depth;
2234	}
2235
2236	/* record root node so further searches know we are done */
2237	iter->tnode = pn;
2238	iter->index = 0;
2239
2240	return NULL;
2241}
2242
2243static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2244					     struct trie *t)
2245{
2246	struct key_vector *n, *pn;
2247
2248	if (!t)
2249		return NULL;
2250
2251	pn = t->kv;
2252	n = rcu_dereference(pn->tnode[0]);
2253	if (!n)
2254		return NULL;
2255
2256	if (IS_TNODE(n)) {
2257		iter->tnode = n;
2258		iter->index = 0;
2259		iter->depth = 1;
2260	} else {
2261		iter->tnode = pn;
2262		iter->index = 0;
2263		iter->depth = 0;
2264	}
2265
2266	return n;
2267}
2268
2269static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2270{
2271	struct key_vector *n;
2272	struct fib_trie_iter iter;
2273
2274	memset(s, 0, sizeof(*s));
2275
2276	rcu_read_lock();
2277	for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2278		if (IS_LEAF(n)) {
2279			struct fib_alias *fa;
2280
2281			s->leaves++;
2282			s->totdepth += iter.depth;
2283			if (iter.depth > s->maxdepth)
2284				s->maxdepth = iter.depth;
2285
2286			hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2287				++s->prefixes;
2288		} else {
2289			s->tnodes++;
2290			if (n->bits < MAX_STAT_DEPTH)
2291				s->nodesizes[n->bits]++;
2292			s->nullpointers += tn_info(n)->empty_children;
2293		}
2294	}
2295	rcu_read_unlock();
2296}
2297
2298/*
2299 *	This outputs /proc/net/fib_triestats
2300 */
2301static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2302{
2303	unsigned int i, max, pointers, bytes, avdepth;
2304
2305	if (stat->leaves)
2306		avdepth = stat->totdepth*100 / stat->leaves;
2307	else
2308		avdepth = 0;
2309
2310	seq_printf(seq, "\tAver depth:     %u.%02d\n",
2311		   avdepth / 100, avdepth % 100);
2312	seq_printf(seq, "\tMax depth:      %u\n", stat->maxdepth);
2313
2314	seq_printf(seq, "\tLeaves:         %u\n", stat->leaves);
2315	bytes = LEAF_SIZE * stat->leaves;
2316
2317	seq_printf(seq, "\tPrefixes:       %u\n", stat->prefixes);
2318	bytes += sizeof(struct fib_alias) * stat->prefixes;
2319
2320	seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2321	bytes += TNODE_SIZE(0) * stat->tnodes;
2322
2323	max = MAX_STAT_DEPTH;
2324	while (max > 0 && stat->nodesizes[max-1] == 0)
2325		max--;
2326
2327	pointers = 0;
2328	for (i = 1; i < max; i++)
2329		if (stat->nodesizes[i] != 0) {
2330			seq_printf(seq, "  %u: %u",  i, stat->nodesizes[i]);
2331			pointers += (1<<i) * stat->nodesizes[i];
2332		}
2333	seq_putc(seq, '\n');
2334	seq_printf(seq, "\tPointers: %u\n", pointers);
2335
2336	bytes += sizeof(struct key_vector *) * pointers;
2337	seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2338	seq_printf(seq, "Total size: %u  kB\n", (bytes + 1023) / 1024);
2339}
2340
2341#ifdef CONFIG_IP_FIB_TRIE_STATS
2342static void trie_show_usage(struct seq_file *seq,
2343			    const struct trie_use_stats __percpu *stats)
2344{
2345	struct trie_use_stats s = { 0 };
2346	int cpu;
2347
2348	/* loop through all of the CPUs and gather up the stats */
2349	for_each_possible_cpu(cpu) {
2350		const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2351
2352		s.gets += pcpu->gets;
2353		s.backtrack += pcpu->backtrack;
2354		s.semantic_match_passed += pcpu->semantic_match_passed;
2355		s.semantic_match_miss += pcpu->semantic_match_miss;
2356		s.null_node_hit += pcpu->null_node_hit;
2357		s.resize_node_skipped += pcpu->resize_node_skipped;
2358	}
2359
2360	seq_printf(seq, "\nCounters:\n---------\n");
2361	seq_printf(seq, "gets = %u\n", s.gets);
2362	seq_printf(seq, "backtracks = %u\n", s.backtrack);
2363	seq_printf(seq, "semantic match passed = %u\n",
2364		   s.semantic_match_passed);
2365	seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2366	seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2367	seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2368}
2369#endif /*  CONFIG_IP_FIB_TRIE_STATS */
2370
2371static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2372{
2373	if (tb->tb_id == RT_TABLE_LOCAL)
2374		seq_puts(seq, "Local:\n");
2375	else if (tb->tb_id == RT_TABLE_MAIN)
2376		seq_puts(seq, "Main:\n");
2377	else
2378		seq_printf(seq, "Id %d:\n", tb->tb_id);
2379}
2380
2381
2382static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2383{
2384	struct net *net = (struct net *)seq->private;
2385	unsigned int h;
2386
2387	seq_printf(seq,
2388		   "Basic info: size of leaf:"
2389		   " %Zd bytes, size of tnode: %Zd bytes.\n",
2390		   LEAF_SIZE, TNODE_SIZE(0));
2391
 
2392	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2393		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2394		struct fib_table *tb;
2395
2396		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2397			struct trie *t = (struct trie *) tb->tb_data;
2398			struct trie_stat stat;
2399
2400			if (!t)
2401				continue;
2402
2403			fib_table_print(seq, tb);
2404
2405			trie_collect_stats(t, &stat);
2406			trie_show_stats(seq, &stat);
2407#ifdef CONFIG_IP_FIB_TRIE_STATS
2408			trie_show_usage(seq, t->stats);
2409#endif
2410		}
 
2411	}
 
2412
2413	return 0;
2414}
2415
2416static int fib_triestat_seq_open(struct inode *inode, struct file *file)
2417{
2418	return single_open_net(inode, file, fib_triestat_seq_show);
2419}
2420
2421static const struct file_operations fib_triestat_fops = {
2422	.owner	= THIS_MODULE,
2423	.open	= fib_triestat_seq_open,
2424	.read	= seq_read,
2425	.llseek	= seq_lseek,
2426	.release = single_release_net,
2427};
2428
2429static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2430{
2431	struct fib_trie_iter *iter = seq->private;
2432	struct net *net = seq_file_net(seq);
2433	loff_t idx = 0;
2434	unsigned int h;
2435
2436	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2437		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2438		struct fib_table *tb;
2439
2440		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2441			struct key_vector *n;
2442
2443			for (n = fib_trie_get_first(iter,
2444						    (struct trie *) tb->tb_data);
2445			     n; n = fib_trie_get_next(iter))
2446				if (pos == idx++) {
2447					iter->tb = tb;
2448					return n;
2449				}
2450		}
2451	}
2452
2453	return NULL;
2454}
2455
2456static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2457	__acquires(RCU)
2458{
2459	rcu_read_lock();
2460	return fib_trie_get_idx(seq, *pos);
2461}
2462
2463static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2464{
2465	struct fib_trie_iter *iter = seq->private;
2466	struct net *net = seq_file_net(seq);
2467	struct fib_table *tb = iter->tb;
2468	struct hlist_node *tb_node;
2469	unsigned int h;
2470	struct key_vector *n;
2471
2472	++*pos;
2473	/* next node in same table */
2474	n = fib_trie_get_next(iter);
2475	if (n)
2476		return n;
2477
2478	/* walk rest of this hash chain */
2479	h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2480	while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2481		tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2482		n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2483		if (n)
2484			goto found;
2485	}
2486
2487	/* new hash chain */
2488	while (++h < FIB_TABLE_HASHSZ) {
2489		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2490		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2491			n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2492			if (n)
2493				goto found;
2494		}
2495	}
2496	return NULL;
2497
2498found:
2499	iter->tb = tb;
2500	return n;
2501}
2502
2503static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2504	__releases(RCU)
2505{
2506	rcu_read_unlock();
2507}
2508
2509static void seq_indent(struct seq_file *seq, int n)
2510{
2511	while (n-- > 0)
2512		seq_puts(seq, "   ");
2513}
2514
2515static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2516{
2517	switch (s) {
2518	case RT_SCOPE_UNIVERSE: return "universe";
2519	case RT_SCOPE_SITE:	return "site";
2520	case RT_SCOPE_LINK:	return "link";
2521	case RT_SCOPE_HOST:	return "host";
2522	case RT_SCOPE_NOWHERE:	return "nowhere";
2523	default:
2524		snprintf(buf, len, "scope=%d", s);
2525		return buf;
2526	}
2527}
2528
2529static const char *const rtn_type_names[__RTN_MAX] = {
2530	[RTN_UNSPEC] = "UNSPEC",
2531	[RTN_UNICAST] = "UNICAST",
2532	[RTN_LOCAL] = "LOCAL",
2533	[RTN_BROADCAST] = "BROADCAST",
2534	[RTN_ANYCAST] = "ANYCAST",
2535	[RTN_MULTICAST] = "MULTICAST",
2536	[RTN_BLACKHOLE] = "BLACKHOLE",
2537	[RTN_UNREACHABLE] = "UNREACHABLE",
2538	[RTN_PROHIBIT] = "PROHIBIT",
2539	[RTN_THROW] = "THROW",
2540	[RTN_NAT] = "NAT",
2541	[RTN_XRESOLVE] = "XRESOLVE",
2542};
2543
2544static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2545{
2546	if (t < __RTN_MAX && rtn_type_names[t])
2547		return rtn_type_names[t];
2548	snprintf(buf, len, "type %u", t);
2549	return buf;
2550}
2551
2552/* Pretty print the trie */
2553static int fib_trie_seq_show(struct seq_file *seq, void *v)
2554{
2555	const struct fib_trie_iter *iter = seq->private;
2556	struct key_vector *n = v;
2557
2558	if (IS_TRIE(node_parent_rcu(n)))
2559		fib_table_print(seq, iter->tb);
2560
2561	if (IS_TNODE(n)) {
2562		__be32 prf = htonl(n->key);
2563
2564		seq_indent(seq, iter->depth-1);
2565		seq_printf(seq, "  +-- %pI4/%zu %u %u %u\n",
2566			   &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2567			   tn_info(n)->full_children,
2568			   tn_info(n)->empty_children);
2569	} else {
2570		__be32 val = htonl(n->key);
2571		struct fib_alias *fa;
2572
2573		seq_indent(seq, iter->depth);
2574		seq_printf(seq, "  |-- %pI4\n", &val);
2575
2576		hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2577			char buf1[32], buf2[32];
2578
2579			seq_indent(seq, iter->depth + 1);
2580			seq_printf(seq, "  /%zu %s %s",
2581				   KEYLENGTH - fa->fa_slen,
2582				   rtn_scope(buf1, sizeof(buf1),
2583					     fa->fa_info->fib_scope),
2584				   rtn_type(buf2, sizeof(buf2),
2585					    fa->fa_type));
2586			if (fa->fa_tos)
2587				seq_printf(seq, " tos=%d", fa->fa_tos);
 
2588			seq_putc(seq, '\n');
2589		}
2590	}
2591
2592	return 0;
2593}
2594
2595static const struct seq_operations fib_trie_seq_ops = {
2596	.start  = fib_trie_seq_start,
2597	.next   = fib_trie_seq_next,
2598	.stop   = fib_trie_seq_stop,
2599	.show   = fib_trie_seq_show,
2600};
2601
2602static int fib_trie_seq_open(struct inode *inode, struct file *file)
2603{
2604	return seq_open_net(inode, file, &fib_trie_seq_ops,
2605			    sizeof(struct fib_trie_iter));
2606}
2607
2608static const struct file_operations fib_trie_fops = {
2609	.owner  = THIS_MODULE,
2610	.open   = fib_trie_seq_open,
2611	.read   = seq_read,
2612	.llseek = seq_lseek,
2613	.release = seq_release_net,
2614};
2615
2616struct fib_route_iter {
2617	struct seq_net_private p;
2618	struct fib_table *main_tb;
2619	struct key_vector *tnode;
2620	loff_t	pos;
2621	t_key	key;
2622};
2623
2624static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2625					    loff_t pos)
2626{
2627	struct key_vector *l, **tp = &iter->tnode;
2628	t_key key;
2629
2630	/* use cached location of previously found key */
2631	if (iter->pos > 0 && pos >= iter->pos) {
2632		key = iter->key;
2633	} else {
2634		iter->pos = 1;
2635		key = 0;
2636	}
2637
2638	pos -= iter->pos;
2639
2640	while ((l = leaf_walk_rcu(tp, key)) && (pos-- > 0)) {
2641		key = l->key + 1;
2642		iter->pos++;
2643		l = NULL;
2644
2645		/* handle unlikely case of a key wrap */
2646		if (!key)
2647			break;
2648	}
2649
2650	if (l)
2651		iter->key = l->key;	/* remember it */
2652	else
2653		iter->pos = 0;		/* forget it */
2654
2655	return l;
2656}
2657
2658static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2659	__acquires(RCU)
2660{
2661	struct fib_route_iter *iter = seq->private;
2662	struct fib_table *tb;
2663	struct trie *t;
2664
2665	rcu_read_lock();
2666
2667	tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2668	if (!tb)
2669		return NULL;
2670
2671	iter->main_tb = tb;
2672	t = (struct trie *)tb->tb_data;
2673	iter->tnode = t->kv;
2674
2675	if (*pos != 0)
2676		return fib_route_get_idx(iter, *pos);
2677
2678	iter->pos = 0;
2679	iter->key = KEY_MAX;
2680
2681	return SEQ_START_TOKEN;
2682}
2683
2684static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2685{
2686	struct fib_route_iter *iter = seq->private;
2687	struct key_vector *l = NULL;
2688	t_key key = iter->key + 1;
2689
2690	++*pos;
2691
2692	/* only allow key of 0 for start of sequence */
2693	if ((v == SEQ_START_TOKEN) || key)
2694		l = leaf_walk_rcu(&iter->tnode, key);
2695
2696	if (l) {
2697		iter->key = l->key;
2698		iter->pos++;
2699	} else {
2700		iter->pos = 0;
2701	}
2702
2703	return l;
2704}
2705
2706static void fib_route_seq_stop(struct seq_file *seq, void *v)
2707	__releases(RCU)
2708{
2709	rcu_read_unlock();
2710}
2711
2712static unsigned int fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2713{
2714	unsigned int flags = 0;
2715
2716	if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2717		flags = RTF_REJECT;
2718	if (fi && fi->fib_nh->nh_gw)
2719		flags |= RTF_GATEWAY;
 
 
 
 
2720	if (mask == htonl(0xFFFFFFFF))
2721		flags |= RTF_HOST;
2722	flags |= RTF_UP;
2723	return flags;
2724}
2725
2726/*
2727 *	This outputs /proc/net/route.
2728 *	The format of the file is not supposed to be changed
2729 *	and needs to be same as fib_hash output to avoid breaking
2730 *	legacy utilities
2731 */
2732static int fib_route_seq_show(struct seq_file *seq, void *v)
2733{
2734	struct fib_route_iter *iter = seq->private;
2735	struct fib_table *tb = iter->main_tb;
2736	struct fib_alias *fa;
2737	struct key_vector *l = v;
2738	__be32 prefix;
2739
2740	if (v == SEQ_START_TOKEN) {
2741		seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2742			   "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2743			   "\tWindow\tIRTT");
2744		return 0;
2745	}
2746
2747	prefix = htonl(l->key);
2748
2749	hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2750		const struct fib_info *fi = fa->fa_info;
2751		__be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2752		unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2753
2754		if ((fa->fa_type == RTN_BROADCAST) ||
2755		    (fa->fa_type == RTN_MULTICAST))
2756			continue;
2757
2758		if (fa->tb_id != tb->tb_id)
2759			continue;
2760
2761		seq_setwidth(seq, 127);
2762
2763		if (fi)
 
 
 
 
 
 
2764			seq_printf(seq,
2765				   "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2766				   "%d\t%08X\t%d\t%u\t%u",
2767				   fi->fib_dev ? fi->fib_dev->name : "*",
2768				   prefix,
2769				   fi->fib_nh->nh_gw, flags, 0, 0,
2770				   fi->fib_priority,
2771				   mask,
2772				   (fi->fib_advmss ?
2773				    fi->fib_advmss + 40 : 0),
2774				   fi->fib_window,
2775				   fi->fib_rtt >> 3);
2776		else
2777			seq_printf(seq,
2778				   "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2779				   "%d\t%08X\t%d\t%u\t%u",
2780				   prefix, 0, flags, 0, 0, 0,
2781				   mask, 0, 0, 0);
2782
2783		seq_pad(seq, '\n');
2784	}
2785
2786	return 0;
2787}
2788
2789static const struct seq_operations fib_route_seq_ops = {
2790	.start  = fib_route_seq_start,
2791	.next   = fib_route_seq_next,
2792	.stop   = fib_route_seq_stop,
2793	.show   = fib_route_seq_show,
2794};
2795
2796static int fib_route_seq_open(struct inode *inode, struct file *file)
2797{
2798	return seq_open_net(inode, file, &fib_route_seq_ops,
2799			    sizeof(struct fib_route_iter));
2800}
2801
2802static const struct file_operations fib_route_fops = {
2803	.owner  = THIS_MODULE,
2804	.open   = fib_route_seq_open,
2805	.read   = seq_read,
2806	.llseek = seq_lseek,
2807	.release = seq_release_net,
2808};
2809
2810int __net_init fib_proc_init(struct net *net)
2811{
2812	if (!proc_create("fib_trie", S_IRUGO, net->proc_net, &fib_trie_fops))
 
2813		goto out1;
2814
2815	if (!proc_create("fib_triestat", S_IRUGO, net->proc_net,
2816			 &fib_triestat_fops))
2817		goto out2;
2818
2819	if (!proc_create("route", S_IRUGO, net->proc_net, &fib_route_fops))
 
2820		goto out3;
2821
2822	return 0;
2823
2824out3:
2825	remove_proc_entry("fib_triestat", net->proc_net);
2826out2:
2827	remove_proc_entry("fib_trie", net->proc_net);
2828out1:
2829	return -ENOMEM;
2830}
2831
2832void __net_exit fib_proc_exit(struct net *net)
2833{
2834	remove_proc_entry("fib_trie", net->proc_net);
2835	remove_proc_entry("fib_triestat", net->proc_net);
2836	remove_proc_entry("route", net->proc_net);
2837}
2838
2839#endif /* CONFIG_PROC_FS */