xref: /kernel/linux/linux-5.10/net/ipv4/fib_trie.c (revision 8c2ecf20)
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/skbuff.h>
56#include <linux/netlink.h>
57#include <linux/init.h>
58#include <linux/list.h>
59#include <linux/slab.h>
60#include <linux/export.h>
61#include <linux/vmalloc.h>
62#include <linux/notifier.h>
63#include <net/net_namespace.h>
64#include <net/ip.h>
65#include <net/protocol.h>
66#include <net/route.h>
67#include <net/tcp.h>
68#include <net/sock.h>
69#include <net/ip_fib.h>
70#include <net/fib_notifier.h>
71#include <trace/events/fib.h>
72#include "fib_lookup.h"
73
74static int call_fib_entry_notifier(struct notifier_block *nb,
75				   enum fib_event_type event_type, u32 dst,
76				   int dst_len, struct fib_alias *fa,
77				   struct netlink_ext_ack *extack)
78{
79	struct fib_entry_notifier_info info = {
80		.info.extack = extack,
81		.dst = dst,
82		.dst_len = dst_len,
83		.fi = fa->fa_info,
84		.tos = fa->fa_tos,
85		.type = fa->fa_type,
86		.tb_id = fa->tb_id,
87	};
88	return call_fib4_notifier(nb, event_type, &info.info);
89}
90
91static int call_fib_entry_notifiers(struct net *net,
92				    enum fib_event_type event_type, u32 dst,
93				    int dst_len, struct fib_alias *fa,
94				    struct netlink_ext_ack *extack)
95{
96	struct fib_entry_notifier_info info = {
97		.info.extack = extack,
98		.dst = dst,
99		.dst_len = dst_len,
100		.fi = fa->fa_info,
101		.tos = fa->fa_tos,
102		.type = fa->fa_type,
103		.tb_id = fa->tb_id,
104	};
105	return call_fib4_notifiers(net, event_type, &info.info);
106}
107
108#define MAX_STAT_DEPTH 32
109
110#define KEYLENGTH	(8*sizeof(t_key))
111#define KEY_MAX		((t_key)~0)
112
113typedef unsigned int t_key;
114
115#define IS_TRIE(n)	((n)->pos >= KEYLENGTH)
116#define IS_TNODE(n)	((n)->bits)
117#define IS_LEAF(n)	(!(n)->bits)
118
119struct key_vector {
120	t_key key;
121	unsigned char pos;		/* 2log(KEYLENGTH) bits needed */
122	unsigned char bits;		/* 2log(KEYLENGTH) bits needed */
123	unsigned char slen;
124	union {
125		/* This list pointer if valid if (pos | bits) == 0 (LEAF) */
126		struct hlist_head leaf;
127		/* This array is valid if (pos | bits) > 0 (TNODE) */
128		struct key_vector __rcu *tnode[0];
129	};
130};
131
132struct tnode {
133	struct rcu_head rcu;
134	t_key empty_children;		/* KEYLENGTH bits needed */
135	t_key full_children;		/* KEYLENGTH bits needed */
136	struct key_vector __rcu *parent;
137	struct key_vector kv[1];
138#define tn_bits kv[0].bits
139};
140
141#define TNODE_SIZE(n)	offsetof(struct tnode, kv[0].tnode[n])
142#define LEAF_SIZE	TNODE_SIZE(1)
143
144#ifdef CONFIG_IP_FIB_TRIE_STATS
145struct trie_use_stats {
146	unsigned int gets;
147	unsigned int backtrack;
148	unsigned int semantic_match_passed;
149	unsigned int semantic_match_miss;
150	unsigned int null_node_hit;
151	unsigned int resize_node_skipped;
152};
153#endif
154
155struct trie_stat {
156	unsigned int totdepth;
157	unsigned int maxdepth;
158	unsigned int tnodes;
159	unsigned int leaves;
160	unsigned int nullpointers;
161	unsigned int prefixes;
162	unsigned int nodesizes[MAX_STAT_DEPTH];
163};
164
165struct trie {
166	struct key_vector kv[1];
167#ifdef CONFIG_IP_FIB_TRIE_STATS
168	struct trie_use_stats __percpu *stats;
169#endif
170};
171
172static struct key_vector *resize(struct trie *t, struct key_vector *tn);
173static unsigned int tnode_free_size;
174
175/*
176 * synchronize_rcu after call_rcu for outstanding dirty memory; it should be
177 * especially useful before resizing the root node with PREEMPT_NONE configs;
178 * the value was obtained experimentally, aiming to avoid visible slowdown.
179 */
180unsigned int sysctl_fib_sync_mem = 512 * 1024;
181unsigned int sysctl_fib_sync_mem_min = 64 * 1024;
182unsigned int sysctl_fib_sync_mem_max = 64 * 1024 * 1024;
183
184static struct kmem_cache *fn_alias_kmem __ro_after_init;
185static struct kmem_cache *trie_leaf_kmem __ro_after_init;
186
187static inline struct tnode *tn_info(struct key_vector *kv)
188{
189	return container_of(kv, struct tnode, kv[0]);
190}
191
192/* caller must hold RTNL */
193#define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
194#define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
195
196/* caller must hold RCU read lock or RTNL */
197#define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
198#define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
199
200/* wrapper for rcu_assign_pointer */
201static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
202{
203	if (n)
204		rcu_assign_pointer(tn_info(n)->parent, tp);
205}
206
207#define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
208
209/* This provides us with the number of children in this node, in the case of a
210 * leaf this will return 0 meaning none of the children are accessible.
211 */
212static inline unsigned long child_length(const struct key_vector *tn)
213{
214	return (1ul << tn->bits) & ~(1ul);
215}
216
217#define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
218
219static inline unsigned long get_index(t_key key, struct key_vector *kv)
220{
221	unsigned long index = key ^ kv->key;
222
223	if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
224		return 0;
225
226	return index >> kv->pos;
227}
228
229/* To understand this stuff, an understanding of keys and all their bits is
230 * necessary. Every node in the trie has a key associated with it, but not
231 * all of the bits in that key are significant.
232 *
233 * Consider a node 'n' and its parent 'tp'.
234 *
235 * If n is a leaf, every bit in its key is significant. Its presence is
236 * necessitated by path compression, since during a tree traversal (when
237 * searching for a leaf - unless we are doing an insertion) we will completely
238 * ignore all skipped bits we encounter. Thus we need to verify, at the end of
239 * a potentially successful search, that we have indeed been walking the
240 * correct key path.
241 *
242 * Note that we can never "miss" the correct key in the tree if present by
243 * following the wrong path. Path compression ensures that segments of the key
244 * that are the same for all keys with a given prefix are skipped, but the
245 * skipped part *is* identical for each node in the subtrie below the skipped
246 * bit! trie_insert() in this implementation takes care of that.
247 *
248 * if n is an internal node - a 'tnode' here, the various parts of its key
249 * have many different meanings.
250 *
251 * Example:
252 * _________________________________________________________________
253 * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
254 * -----------------------------------------------------------------
255 *  31  30  29  28  27  26  25  24  23  22  21  20  19  18  17  16
256 *
257 * _________________________________________________________________
258 * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
259 * -----------------------------------------------------------------
260 *  15  14  13  12  11  10   9   8   7   6   5   4   3   2   1   0
261 *
262 * tp->pos = 22
263 * tp->bits = 3
264 * n->pos = 13
265 * n->bits = 4
266 *
267 * First, let's just ignore the bits that come before the parent tp, that is
268 * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
269 * point we do not use them for anything.
270 *
271 * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
272 * index into the parent's child array. That is, they will be used to find
273 * 'n' among tp's children.
274 *
275 * The bits from (n->pos + n->bits) to (tp->pos - 1) - "S" - are skipped bits
276 * for the node n.
277 *
278 * All the bits we have seen so far are significant to the node n. The rest
279 * of the bits are really not needed or indeed known in n->key.
280 *
281 * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
282 * n's child array, and will of course be different for each child.
283 *
284 * The rest of the bits, from 0 to (n->pos -1) - "u" - are completely unknown
285 * at this point.
286 */
287
288static const int halve_threshold = 25;
289static const int inflate_threshold = 50;
290static const int halve_threshold_root = 15;
291static const int inflate_threshold_root = 30;
292
293static void __alias_free_mem(struct rcu_head *head)
294{
295	struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
296	kmem_cache_free(fn_alias_kmem, fa);
297}
298
299static inline void alias_free_mem_rcu(struct fib_alias *fa)
300{
301	call_rcu(&fa->rcu, __alias_free_mem);
302}
303
304#define TNODE_VMALLOC_MAX \
305	ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
306
307static void __node_free_rcu(struct rcu_head *head)
308{
309	struct tnode *n = container_of(head, struct tnode, rcu);
310
311	if (!n->tn_bits)
312		kmem_cache_free(trie_leaf_kmem, n);
313	else
314		kvfree(n);
315}
316
317#define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
318
319static struct tnode *tnode_alloc(int bits)
320{
321	size_t size;
322
323	/* verify bits is within bounds */
324	if (bits > TNODE_VMALLOC_MAX)
325		return NULL;
326
327	/* determine size and verify it is non-zero and didn't overflow */
328	size = TNODE_SIZE(1ul << bits);
329
330	if (size <= PAGE_SIZE)
331		return kzalloc(size, GFP_KERNEL);
332	else
333		return vzalloc(size);
334}
335
336static inline void empty_child_inc(struct key_vector *n)
337{
338	tn_info(n)->empty_children++;
339
340	if (!tn_info(n)->empty_children)
341		tn_info(n)->full_children++;
342}
343
344static inline void empty_child_dec(struct key_vector *n)
345{
346	if (!tn_info(n)->empty_children)
347		tn_info(n)->full_children--;
348
349	tn_info(n)->empty_children--;
350}
351
352static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
353{
354	struct key_vector *l;
355	struct tnode *kv;
356
357	kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
358	if (!kv)
359		return NULL;
360
361	/* initialize key vector */
362	l = kv->kv;
363	l->key = key;
364	l->pos = 0;
365	l->bits = 0;
366	l->slen = fa->fa_slen;
367
368	/* link leaf to fib alias */
369	INIT_HLIST_HEAD(&l->leaf);
370	hlist_add_head(&fa->fa_list, &l->leaf);
371
372	return l;
373}
374
375static struct key_vector *tnode_new(t_key key, int pos, int bits)
376{
377	unsigned int shift = pos + bits;
378	struct key_vector *tn;
379	struct tnode *tnode;
380
381	/* verify bits and pos their msb bits clear and values are valid */
382	BUG_ON(!bits || (shift > KEYLENGTH));
383
384	tnode = tnode_alloc(bits);
385	if (!tnode)
386		return NULL;
387
388	pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
389		 sizeof(struct key_vector *) << bits);
390
391	if (bits == KEYLENGTH)
392		tnode->full_children = 1;
393	else
394		tnode->empty_children = 1ul << bits;
395
396	tn = tnode->kv;
397	tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
398	tn->pos = pos;
399	tn->bits = bits;
400	tn->slen = pos;
401
402	return tn;
403}
404
405/* Check whether a tnode 'n' is "full", i.e. it is an internal node
406 * and no bits are skipped. See discussion in dyntree paper p. 6
407 */
408static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
409{
410	return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
411}
412
413/* Add a child at position i overwriting the old value.
414 * Update the value of full_children and empty_children.
415 */
416static void put_child(struct key_vector *tn, unsigned long i,
417		      struct key_vector *n)
418{
419	struct key_vector *chi = get_child(tn, i);
420	int isfull, wasfull;
421
422	BUG_ON(i >= child_length(tn));
423
424	/* update emptyChildren, overflow into fullChildren */
425	if (!n && chi)
426		empty_child_inc(tn);
427	if (n && !chi)
428		empty_child_dec(tn);
429
430	/* update fullChildren */
431	wasfull = tnode_full(tn, chi);
432	isfull = tnode_full(tn, n);
433
434	if (wasfull && !isfull)
435		tn_info(tn)->full_children--;
436	else if (!wasfull && isfull)
437		tn_info(tn)->full_children++;
438
439	if (n && (tn->slen < n->slen))
440		tn->slen = n->slen;
441
442	rcu_assign_pointer(tn->tnode[i], n);
443}
444
445static void update_children(struct key_vector *tn)
446{
447	unsigned long i;
448
449	/* update all of the child parent pointers */
450	for (i = child_length(tn); i;) {
451		struct key_vector *inode = get_child(tn, --i);
452
453		if (!inode)
454			continue;
455
456		/* Either update the children of a tnode that
457		 * already belongs to us or update the child
458		 * to point to ourselves.
459		 */
460		if (node_parent(inode) == tn)
461			update_children(inode);
462		else
463			node_set_parent(inode, tn);
464	}
465}
466
467static inline void put_child_root(struct key_vector *tp, t_key key,
468				  struct key_vector *n)
469{
470	if (IS_TRIE(tp))
471		rcu_assign_pointer(tp->tnode[0], n);
472	else
473		put_child(tp, get_index(key, tp), n);
474}
475
476static inline void tnode_free_init(struct key_vector *tn)
477{
478	tn_info(tn)->rcu.next = NULL;
479}
480
481static inline void tnode_free_append(struct key_vector *tn,
482				     struct key_vector *n)
483{
484	tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
485	tn_info(tn)->rcu.next = &tn_info(n)->rcu;
486}
487
488static void tnode_free(struct key_vector *tn)
489{
490	struct callback_head *head = &tn_info(tn)->rcu;
491
492	while (head) {
493		head = head->next;
494		tnode_free_size += TNODE_SIZE(1ul << tn->bits);
495		node_free(tn);
496
497		tn = container_of(head, struct tnode, rcu)->kv;
498	}
499
500	if (tnode_free_size >= READ_ONCE(sysctl_fib_sync_mem)) {
501		tnode_free_size = 0;
502		synchronize_rcu();
503	}
504}
505
506static struct key_vector *replace(struct trie *t,
507				  struct key_vector *oldtnode,
508				  struct key_vector *tn)
509{
510	struct key_vector *tp = node_parent(oldtnode);
511	unsigned long i;
512
513	/* setup the parent pointer out of and back into this node */
514	NODE_INIT_PARENT(tn, tp);
515	put_child_root(tp, tn->key, tn);
516
517	/* update all of the child parent pointers */
518	update_children(tn);
519
520	/* all pointers should be clean so we are done */
521	tnode_free(oldtnode);
522
523	/* resize children now that oldtnode is freed */
524	for (i = child_length(tn); i;) {
525		struct key_vector *inode = get_child(tn, --i);
526
527		/* resize child node */
528		if (tnode_full(tn, inode))
529			tn = resize(t, inode);
530	}
531
532	return tp;
533}
534
535static struct key_vector *inflate(struct trie *t,
536				  struct key_vector *oldtnode)
537{
538	struct key_vector *tn;
539	unsigned long i;
540	t_key m;
541
542	pr_debug("In inflate\n");
543
544	tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
545	if (!tn)
546		goto notnode;
547
548	/* prepare oldtnode to be freed */
549	tnode_free_init(oldtnode);
550
551	/* Assemble all of the pointers in our cluster, in this case that
552	 * represents all of the pointers out of our allocated nodes that
553	 * point to existing tnodes and the links between our allocated
554	 * nodes.
555	 */
556	for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
557		struct key_vector *inode = get_child(oldtnode, --i);
558		struct key_vector *node0, *node1;
559		unsigned long j, k;
560
561		/* An empty child */
562		if (!inode)
563			continue;
564
565		/* A leaf or an internal node with skipped bits */
566		if (!tnode_full(oldtnode, inode)) {
567			put_child(tn, get_index(inode->key, tn), inode);
568			continue;
569		}
570
571		/* drop the node in the old tnode free list */
572		tnode_free_append(oldtnode, inode);
573
574		/* An internal node with two children */
575		if (inode->bits == 1) {
576			put_child(tn, 2 * i + 1, get_child(inode, 1));
577			put_child(tn, 2 * i, get_child(inode, 0));
578			continue;
579		}
580
581		/* We will replace this node 'inode' with two new
582		 * ones, 'node0' and 'node1', each with half of the
583		 * original children. The two new nodes will have
584		 * a position one bit further down the key and this
585		 * means that the "significant" part of their keys
586		 * (see the discussion near the top of this file)
587		 * will differ by one bit, which will be "0" in
588		 * node0's key and "1" in node1's key. Since we are
589		 * moving the key position by one step, the bit that
590		 * we are moving away from - the bit at position
591		 * (tn->pos) - is the one that will differ between
592		 * node0 and node1. So... we synthesize that bit in the
593		 * two new keys.
594		 */
595		node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
596		if (!node1)
597			goto nomem;
598		node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
599
600		tnode_free_append(tn, node1);
601		if (!node0)
602			goto nomem;
603		tnode_free_append(tn, node0);
604
605		/* populate child pointers in new nodes */
606		for (k = child_length(inode), j = k / 2; j;) {
607			put_child(node1, --j, get_child(inode, --k));
608			put_child(node0, j, get_child(inode, j));
609			put_child(node1, --j, get_child(inode, --k));
610			put_child(node0, j, get_child(inode, j));
611		}
612
613		/* link new nodes to parent */
614		NODE_INIT_PARENT(node1, tn);
615		NODE_INIT_PARENT(node0, tn);
616
617		/* link parent to nodes */
618		put_child(tn, 2 * i + 1, node1);
619		put_child(tn, 2 * i, node0);
620	}
621
622	/* setup the parent pointers into and out of this node */
623	return replace(t, oldtnode, tn);
624nomem:
625	/* all pointers should be clean so we are done */
626	tnode_free(tn);
627notnode:
628	return NULL;
629}
630
631static struct key_vector *halve(struct trie *t,
632				struct key_vector *oldtnode)
633{
634	struct key_vector *tn;
635	unsigned long i;
636
637	pr_debug("In halve\n");
638
639	tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
640	if (!tn)
641		goto notnode;
642
643	/* prepare oldtnode to be freed */
644	tnode_free_init(oldtnode);
645
646	/* Assemble all of the pointers in our cluster, in this case that
647	 * represents all of the pointers out of our allocated nodes that
648	 * point to existing tnodes and the links between our allocated
649	 * nodes.
650	 */
651	for (i = child_length(oldtnode); i;) {
652		struct key_vector *node1 = get_child(oldtnode, --i);
653		struct key_vector *node0 = get_child(oldtnode, --i);
654		struct key_vector *inode;
655
656		/* At least one of the children is empty */
657		if (!node1 || !node0) {
658			put_child(tn, i / 2, node1 ? : node0);
659			continue;
660		}
661
662		/* Two nonempty children */
663		inode = tnode_new(node0->key, oldtnode->pos, 1);
664		if (!inode)
665			goto nomem;
666		tnode_free_append(tn, inode);
667
668		/* initialize pointers out of node */
669		put_child(inode, 1, node1);
670		put_child(inode, 0, node0);
671		NODE_INIT_PARENT(inode, tn);
672
673		/* link parent to node */
674		put_child(tn, i / 2, inode);
675	}
676
677	/* setup the parent pointers into and out of this node */
678	return replace(t, oldtnode, tn);
679nomem:
680	/* all pointers should be clean so we are done */
681	tnode_free(tn);
682notnode:
683	return NULL;
684}
685
686static struct key_vector *collapse(struct trie *t,
687				   struct key_vector *oldtnode)
688{
689	struct key_vector *n, *tp;
690	unsigned long i;
691
692	/* scan the tnode looking for that one child that might still exist */
693	for (n = NULL, i = child_length(oldtnode); !n && i;)
694		n = get_child(oldtnode, --i);
695
696	/* compress one level */
697	tp = node_parent(oldtnode);
698	put_child_root(tp, oldtnode->key, n);
699	node_set_parent(n, tp);
700
701	/* drop dead node */
702	node_free(oldtnode);
703
704	return tp;
705}
706
707static unsigned char update_suffix(struct key_vector *tn)
708{
709	unsigned char slen = tn->pos;
710	unsigned long stride, i;
711	unsigned char slen_max;
712
713	/* only vector 0 can have a suffix length greater than or equal to
714	 * tn->pos + tn->bits, the second highest node will have a suffix
715	 * length at most of tn->pos + tn->bits - 1
716	 */
717	slen_max = min_t(unsigned char, tn->pos + tn->bits - 1, tn->slen);
718
719	/* search though the list of children looking for nodes that might
720	 * have a suffix greater than the one we currently have.  This is
721	 * why we start with a stride of 2 since a stride of 1 would
722	 * represent the nodes with suffix length equal to tn->pos
723	 */
724	for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
725		struct key_vector *n = get_child(tn, i);
726
727		if (!n || (n->slen <= slen))
728			continue;
729
730		/* update stride and slen based on new value */
731		stride <<= (n->slen - slen);
732		slen = n->slen;
733		i &= ~(stride - 1);
734
735		/* stop searching if we have hit the maximum possible value */
736		if (slen >= slen_max)
737			break;
738	}
739
740	tn->slen = slen;
741
742	return slen;
743}
744
745/* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
746 * the Helsinki University of Technology and Matti Tikkanen of Nokia
747 * Telecommunications, page 6:
748 * "A node is doubled if the ratio of non-empty children to all
749 * children in the *doubled* node is at least 'high'."
750 *
751 * 'high' in this instance is the variable 'inflate_threshold'. It
752 * is expressed as a percentage, so we multiply it with
753 * child_length() and instead of multiplying by 2 (since the
754 * child array will be doubled by inflate()) and multiplying
755 * the left-hand side by 100 (to handle the percentage thing) we
756 * multiply the left-hand side by 50.
757 *
758 * The left-hand side may look a bit weird: child_length(tn)
759 * - tn->empty_children is of course the number of non-null children
760 * in the current node. tn->full_children is the number of "full"
761 * children, that is non-null tnodes with a skip value of 0.
762 * All of those will be doubled in the resulting inflated tnode, so
763 * we just count them one extra time here.
764 *
765 * A clearer way to write this would be:
766 *
767 * to_be_doubled = tn->full_children;
768 * not_to_be_doubled = child_length(tn) - tn->empty_children -
769 *     tn->full_children;
770 *
771 * new_child_length = child_length(tn) * 2;
772 *
773 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
774 *      new_child_length;
775 * if (new_fill_factor >= inflate_threshold)
776 *
777 * ...and so on, tho it would mess up the while () loop.
778 *
779 * anyway,
780 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
781 *      inflate_threshold
782 *
783 * avoid a division:
784 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
785 *      inflate_threshold * new_child_length
786 *
787 * expand not_to_be_doubled and to_be_doubled, and shorten:
788 * 100 * (child_length(tn) - tn->empty_children +
789 *    tn->full_children) >= inflate_threshold * new_child_length
790 *
791 * expand new_child_length:
792 * 100 * (child_length(tn) - tn->empty_children +
793 *    tn->full_children) >=
794 *      inflate_threshold * child_length(tn) * 2
795 *
796 * shorten again:
797 * 50 * (tn->full_children + child_length(tn) -
798 *    tn->empty_children) >= inflate_threshold *
799 *    child_length(tn)
800 *
801 */
802static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
803{
804	unsigned long used = child_length(tn);
805	unsigned long threshold = used;
806
807	/* Keep root node larger */
808	threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
809	used -= tn_info(tn)->empty_children;
810	used += tn_info(tn)->full_children;
811
812	/* if bits == KEYLENGTH then pos = 0, and will fail below */
813
814	return (used > 1) && tn->pos && ((50 * used) >= threshold);
815}
816
817static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
818{
819	unsigned long used = child_length(tn);
820	unsigned long threshold = used;
821
822	/* Keep root node larger */
823	threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
824	used -= tn_info(tn)->empty_children;
825
826	/* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
827
828	return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
829}
830
831static inline bool should_collapse(struct key_vector *tn)
832{
833	unsigned long used = child_length(tn);
834
835	used -= tn_info(tn)->empty_children;
836
837	/* account for bits == KEYLENGTH case */
838	if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
839		used -= KEY_MAX;
840
841	/* One child or none, time to drop us from the trie */
842	return used < 2;
843}
844
845#define MAX_WORK 10
846static struct key_vector *resize(struct trie *t, struct key_vector *tn)
847{
848#ifdef CONFIG_IP_FIB_TRIE_STATS
849	struct trie_use_stats __percpu *stats = t->stats;
850#endif
851	struct key_vector *tp = node_parent(tn);
852	unsigned long cindex = get_index(tn->key, tp);
853	int max_work = MAX_WORK;
854
855	pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
856		 tn, inflate_threshold, halve_threshold);
857
858	/* track the tnode via the pointer from the parent instead of
859	 * doing it ourselves.  This way we can let RCU fully do its
860	 * thing without us interfering
861	 */
862	BUG_ON(tn != get_child(tp, cindex));
863
864	/* Double as long as the resulting node has a number of
865	 * nonempty nodes that are above the threshold.
866	 */
867	while (should_inflate(tp, tn) && max_work) {
868		tp = inflate(t, tn);
869		if (!tp) {
870#ifdef CONFIG_IP_FIB_TRIE_STATS
871			this_cpu_inc(stats->resize_node_skipped);
872#endif
873			break;
874		}
875
876		max_work--;
877		tn = get_child(tp, cindex);
878	}
879
880	/* update parent in case inflate failed */
881	tp = node_parent(tn);
882
883	/* Return if at least one inflate is run */
884	if (max_work != MAX_WORK)
885		return tp;
886
887	/* Halve as long as the number of empty children in this
888	 * node is above threshold.
889	 */
890	while (should_halve(tp, tn) && max_work) {
891		tp = halve(t, tn);
892		if (!tp) {
893#ifdef CONFIG_IP_FIB_TRIE_STATS
894			this_cpu_inc(stats->resize_node_skipped);
895#endif
896			break;
897		}
898
899		max_work--;
900		tn = get_child(tp, cindex);
901	}
902
903	/* Only one child remains */
904	if (should_collapse(tn))
905		return collapse(t, tn);
906
907	/* update parent in case halve failed */
908	return node_parent(tn);
909}
910
911static void node_pull_suffix(struct key_vector *tn, unsigned char slen)
912{
913	unsigned char node_slen = tn->slen;
914
915	while ((node_slen > tn->pos) && (node_slen > slen)) {
916		slen = update_suffix(tn);
917		if (node_slen == slen)
918			break;
919
920		tn = node_parent(tn);
921		node_slen = tn->slen;
922	}
923}
924
925static void node_push_suffix(struct key_vector *tn, unsigned char slen)
926{
927	while (tn->slen < slen) {
928		tn->slen = slen;
929		tn = node_parent(tn);
930	}
931}
932
933/* rcu_read_lock needs to be hold by caller from readside */
934static struct key_vector *fib_find_node(struct trie *t,
935					struct key_vector **tp, u32 key)
936{
937	struct key_vector *pn, *n = t->kv;
938	unsigned long index = 0;
939
940	do {
941		pn = n;
942		n = get_child_rcu(n, index);
943
944		if (!n)
945			break;
946
947		index = get_cindex(key, n);
948
949		/* This bit of code is a bit tricky but it combines multiple
950		 * checks into a single check.  The prefix consists of the
951		 * prefix plus zeros for the bits in the cindex. The index
952		 * is the difference between the key and this value.  From
953		 * this we can actually derive several pieces of data.
954		 *   if (index >= (1ul << bits))
955		 *     we have a mismatch in skip bits and failed
956		 *   else
957		 *     we know the value is cindex
958		 *
959		 * This check is safe even if bits == KEYLENGTH due to the
960		 * fact that we can only allocate a node with 32 bits if a
961		 * long is greater than 32 bits.
962		 */
963		if (index >= (1ul << n->bits)) {
964			n = NULL;
965			break;
966		}
967
968		/* keep searching until we find a perfect match leaf or NULL */
969	} while (IS_TNODE(n));
970
971	*tp = pn;
972
973	return n;
974}
975
976/* Return the first fib alias matching TOS with
977 * priority less than or equal to PRIO.
978 * If 'find_first' is set, return the first matching
979 * fib alias, regardless of TOS and priority.
980 */
981static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
982					u8 tos, u32 prio, u32 tb_id,
983					bool find_first)
984{
985	struct fib_alias *fa;
986
987	if (!fah)
988		return NULL;
989
990	hlist_for_each_entry(fa, fah, fa_list) {
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->fa_tos > tos)
1002			continue;
1003		if (fa->fa_info->fib_priority >= prio || fa->fa_tos < tos)
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_tos == fri->tos && 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	struct fib_alias *fa_match;
1041
1042	rcu_read_lock();
1043
1044	fa_match = fib_find_matching_alias(net, fri);
1045	if (!fa_match)
1046		goto out;
1047
1048	fa_match->offload = fri->offload;
1049	fa_match->trap = fri->trap;
1050
1051out:
1052	rcu_read_unlock();
1053}
1054EXPORT_SYMBOL_GPL(fib_alias_hw_flags_set);
1055
1056static void trie_rebalance(struct trie *t, struct key_vector *tn)
1057{
1058	while (!IS_TRIE(tn))
1059		tn = resize(t, tn);
1060}
1061
1062static int fib_insert_node(struct trie *t, struct key_vector *tp,
1063			   struct fib_alias *new, t_key key)
1064{
1065	struct key_vector *n, *l;
1066
1067	l = leaf_new(key, new);
1068	if (!l)
1069		goto noleaf;
1070
1071	/* retrieve child from parent node */
1072	n = get_child(tp, get_index(key, tp));
1073
1074	/* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1075	 *
1076	 *  Add a new tnode here
1077	 *  first tnode need some special handling
1078	 *  leaves us in position for handling as case 3
1079	 */
1080	if (n) {
1081		struct key_vector *tn;
1082
1083		tn = tnode_new(key, __fls(key ^ n->key), 1);
1084		if (!tn)
1085			goto notnode;
1086
1087		/* initialize routes out of node */
1088		NODE_INIT_PARENT(tn, tp);
1089		put_child(tn, get_index(key, tn) ^ 1, n);
1090
1091		/* start adding routes into the node */
1092		put_child_root(tp, key, tn);
1093		node_set_parent(n, tn);
1094
1095		/* parent now has a NULL spot where the leaf can go */
1096		tp = tn;
1097	}
1098
1099	/* Case 3: n is NULL, and will just insert a new leaf */
1100	node_push_suffix(tp, new->fa_slen);
1101	NODE_INIT_PARENT(l, tp);
1102	put_child_root(tp, key, l);
1103	trie_rebalance(t, tp);
1104
1105	return 0;
1106notnode:
1107	node_free(l);
1108noleaf:
1109	return -ENOMEM;
1110}
1111
1112static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1113			    struct key_vector *l, struct fib_alias *new,
1114			    struct fib_alias *fa, t_key key)
1115{
1116	if (!l)
1117		return fib_insert_node(t, tp, new, key);
1118
1119	if (fa) {
1120		hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1121	} else {
1122		struct fib_alias *last;
1123
1124		hlist_for_each_entry(last, &l->leaf, fa_list) {
1125			if (new->fa_slen < last->fa_slen)
1126				break;
1127			if ((new->fa_slen == last->fa_slen) &&
1128			    (new->tb_id > last->tb_id))
1129				break;
1130			fa = last;
1131		}
1132
1133		if (fa)
1134			hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1135		else
1136			hlist_add_head_rcu(&new->fa_list, &l->leaf);
1137	}
1138
1139	/* if we added to the tail node then we need to update slen */
1140	if (l->slen < new->fa_slen) {
1141		l->slen = new->fa_slen;
1142		node_push_suffix(tp, new->fa_slen);
1143	}
1144
1145	return 0;
1146}
1147
1148static bool fib_valid_key_len(u32 key, u8 plen, struct netlink_ext_ack *extack)
1149{
1150	if (plen > KEYLENGTH) {
1151		NL_SET_ERR_MSG(extack, "Invalid prefix length");
1152		return false;
1153	}
1154
1155	if ((plen < KEYLENGTH) && (key << plen)) {
1156		NL_SET_ERR_MSG(extack,
1157			       "Invalid prefix for given prefix length");
1158		return false;
1159	}
1160
1161	return true;
1162}
1163
1164static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1165			     struct key_vector *l, struct fib_alias *old);
1166
1167/* Caller must hold RTNL. */
1168int fib_table_insert(struct net *net, struct fib_table *tb,
1169		     struct fib_config *cfg, struct netlink_ext_ack *extack)
1170{
1171	struct trie *t = (struct trie *)tb->tb_data;
1172	struct fib_alias *fa, *new_fa;
1173	struct key_vector *l, *tp;
1174	u16 nlflags = NLM_F_EXCL;
1175	struct fib_info *fi;
1176	u8 plen = cfg->fc_dst_len;
1177	u8 slen = KEYLENGTH - plen;
1178	u8 tos = cfg->fc_tos;
1179	u32 key;
1180	int err;
1181
1182	key = ntohl(cfg->fc_dst);
1183
1184	if (!fib_valid_key_len(key, plen, extack))
1185		return -EINVAL;
1186
1187	pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1188
1189	fi = fib_create_info(cfg, extack);
1190	if (IS_ERR(fi)) {
1191		err = PTR_ERR(fi);
1192		goto err;
1193	}
1194
1195	l = fib_find_node(t, &tp, key);
1196	fa = l ? fib_find_alias(&l->leaf, slen, tos, fi->fib_priority,
1197				tb->tb_id, false) : NULL;
1198
1199	/* Now fa, if non-NULL, points to the first fib alias
1200	 * with the same keys [prefix,tos,priority], if such key already
1201	 * exists or to the node before which we will insert new one.
1202	 *
1203	 * If fa is NULL, we will need to allocate a new one and
1204	 * insert to the tail of the section matching the suffix length
1205	 * of the new alias.
1206	 */
1207
1208	if (fa && fa->fa_tos == tos &&
1209	    fa->fa_info->fib_priority == fi->fib_priority) {
1210		struct fib_alias *fa_first, *fa_match;
1211
1212		err = -EEXIST;
1213		if (cfg->fc_nlflags & NLM_F_EXCL)
1214			goto out;
1215
1216		nlflags &= ~NLM_F_EXCL;
1217
1218		/* We have 2 goals:
1219		 * 1. Find exact match for type, scope, fib_info to avoid
1220		 * duplicate routes
1221		 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1222		 */
1223		fa_match = NULL;
1224		fa_first = fa;
1225		hlist_for_each_entry_from(fa, fa_list) {
1226			if ((fa->fa_slen != slen) ||
1227			    (fa->tb_id != tb->tb_id) ||
1228			    (fa->fa_tos != tos))
1229				break;
1230			if (fa->fa_info->fib_priority != fi->fib_priority)
1231				break;
1232			if (fa->fa_type == cfg->fc_type &&
1233			    fa->fa_info == fi) {
1234				fa_match = fa;
1235				break;
1236			}
1237		}
1238
1239		if (cfg->fc_nlflags & NLM_F_REPLACE) {
1240			struct fib_info *fi_drop;
1241			u8 state;
1242
1243			nlflags |= NLM_F_REPLACE;
1244			fa = fa_first;
1245			if (fa_match) {
1246				if (fa == fa_match)
1247					err = 0;
1248				goto out;
1249			}
1250			err = -ENOBUFS;
1251			new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1252			if (!new_fa)
1253				goto out;
1254
1255			fi_drop = fa->fa_info;
1256			new_fa->fa_tos = fa->fa_tos;
1257			new_fa->fa_info = fi;
1258			new_fa->fa_type = cfg->fc_type;
1259			state = fa->fa_state;
1260			new_fa->fa_state = state & ~FA_S_ACCESSED;
1261			new_fa->fa_slen = fa->fa_slen;
1262			new_fa->tb_id = tb->tb_id;
1263			new_fa->fa_default = -1;
1264			new_fa->offload = 0;
1265			new_fa->trap = 0;
1266
1267			hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1268
1269			if (fib_find_alias(&l->leaf, fa->fa_slen, 0, 0,
1270					   tb->tb_id, true) == new_fa) {
1271				enum fib_event_type fib_event;
1272
1273				fib_event = FIB_EVENT_ENTRY_REPLACE;
1274				err = call_fib_entry_notifiers(net, fib_event,
1275							       key, plen,
1276							       new_fa, extack);
1277				if (err) {
1278					hlist_replace_rcu(&new_fa->fa_list,
1279							  &fa->fa_list);
1280					goto out_free_new_fa;
1281				}
1282			}
1283
1284			rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1285				  tb->tb_id, &cfg->fc_nlinfo, nlflags);
1286
1287			alias_free_mem_rcu(fa);
1288
1289			fib_release_info(fi_drop);
1290			if (state & FA_S_ACCESSED)
1291				rt_cache_flush(cfg->fc_nlinfo.nl_net);
1292
1293			goto succeeded;
1294		}
1295		/* Error if we find a perfect match which
1296		 * uses the same scope, type, and nexthop
1297		 * information.
1298		 */
1299		if (fa_match)
1300			goto out;
1301
1302		if (cfg->fc_nlflags & NLM_F_APPEND)
1303			nlflags |= NLM_F_APPEND;
1304		else
1305			fa = fa_first;
1306	}
1307	err = -ENOENT;
1308	if (!(cfg->fc_nlflags & NLM_F_CREATE))
1309		goto out;
1310
1311	nlflags |= NLM_F_CREATE;
1312	err = -ENOBUFS;
1313	new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1314	if (!new_fa)
1315		goto out;
1316
1317	new_fa->fa_info = fi;
1318	new_fa->fa_tos = tos;
1319	new_fa->fa_type = cfg->fc_type;
1320	new_fa->fa_state = 0;
1321	new_fa->fa_slen = slen;
1322	new_fa->tb_id = tb->tb_id;
1323	new_fa->fa_default = -1;
1324	new_fa->offload = 0;
1325	new_fa->trap = 0;
1326
1327	/* Insert new entry to the list. */
1328	err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1329	if (err)
1330		goto out_free_new_fa;
1331
1332	/* The alias was already inserted, so the node must exist. */
1333	l = l ? l : fib_find_node(t, &tp, key);
1334	if (WARN_ON_ONCE(!l)) {
1335		err = -ENOENT;
1336		goto out_free_new_fa;
1337	}
1338
1339	if (fib_find_alias(&l->leaf, new_fa->fa_slen, 0, 0, tb->tb_id, true) ==
1340	    new_fa) {
1341		enum fib_event_type fib_event;
1342
1343		fib_event = FIB_EVENT_ENTRY_REPLACE;
1344		err = call_fib_entry_notifiers(net, fib_event, key, plen,
1345					       new_fa, extack);
1346		if (err)
1347			goto out_remove_new_fa;
1348	}
1349
1350	if (!plen)
1351		tb->tb_num_default++;
1352
1353	rt_cache_flush(cfg->fc_nlinfo.nl_net);
1354	rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1355		  &cfg->fc_nlinfo, nlflags);
1356succeeded:
1357	return 0;
1358
1359out_remove_new_fa:
1360	fib_remove_alias(t, tp, l, new_fa);
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
1376bool fib_lookup_good_nhc(const struct fib_nh_common *nhc, int fib_flags,
1377			 const struct flowi4 *flp)
1378{
1379	if (nhc->nhc_flags & RTNH_F_DEAD)
1380		return false;
1381
1382	if (ip_ignore_linkdown(nhc->nhc_dev) &&
1383	    nhc->nhc_flags & RTNH_F_LINKDOWN &&
1384	    !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1385		return false;
1386
1387	if (!(flp->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF)) {
1388		if (flp->flowi4_oif &&
1389		    flp->flowi4_oif != nhc->nhc_oif)
1390			return false;
1391	}
1392
1393	return true;
1394}
1395
1396/* should be called with rcu_read_lock */
1397int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1398		     struct fib_result *res, int fib_flags)
1399{
1400	struct trie *t = (struct trie *) tb->tb_data;
1401#ifdef CONFIG_IP_FIB_TRIE_STATS
1402	struct trie_use_stats __percpu *stats = t->stats;
1403#endif
1404	const t_key key = ntohl(flp->daddr);
1405	struct key_vector *n, *pn;
1406	struct fib_alias *fa;
1407	unsigned long index;
1408	t_key cindex;
1409
1410	pn = t->kv;
1411	cindex = 0;
1412
1413	n = get_child_rcu(pn, cindex);
1414	if (!n) {
1415		trace_fib_table_lookup(tb->tb_id, flp, NULL, -EAGAIN);
1416		return -EAGAIN;
1417	}
1418
1419#ifdef CONFIG_IP_FIB_TRIE_STATS
1420	this_cpu_inc(stats->gets);
1421#endif
1422
1423	/* Step 1: Travel to the longest prefix match in the trie */
1424	for (;;) {
1425		index = get_cindex(key, n);
1426
1427		/* This bit of code is a bit tricky but it combines multiple
1428		 * checks into a single check.  The prefix consists of the
1429		 * prefix plus zeros for the "bits" in the prefix. The index
1430		 * is the difference between the key and this value.  From
1431		 * this we can actually derive several pieces of data.
1432		 *   if (index >= (1ul << bits))
1433		 *     we have a mismatch in skip bits and failed
1434		 *   else
1435		 *     we know the value is cindex
1436		 *
1437		 * This check is safe even if bits == KEYLENGTH due to the
1438		 * fact that we can only allocate a node with 32 bits if a
1439		 * long is greater than 32 bits.
1440		 */
1441		if (index >= (1ul << n->bits))
1442			break;
1443
1444		/* we have found a leaf. Prefixes have already been compared */
1445		if (IS_LEAF(n))
1446			goto found;
1447
1448		/* only record pn and cindex if we are going to be chopping
1449		 * bits later.  Otherwise we are just wasting cycles.
1450		 */
1451		if (n->slen > n->pos) {
1452			pn = n;
1453			cindex = index;
1454		}
1455
1456		n = get_child_rcu(n, index);
1457		if (unlikely(!n))
1458			goto backtrace;
1459	}
1460
1461	/* Step 2: Sort out leaves and begin backtracing for longest prefix */
1462	for (;;) {
1463		/* record the pointer where our next node pointer is stored */
1464		struct key_vector __rcu **cptr = n->tnode;
1465
1466		/* This test verifies that none of the bits that differ
1467		 * between the key and the prefix exist in the region of
1468		 * the lsb and higher in the prefix.
1469		 */
1470		if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1471			goto backtrace;
1472
1473		/* exit out and process leaf */
1474		if (unlikely(IS_LEAF(n)))
1475			break;
1476
1477		/* Don't bother recording parent info.  Since we are in
1478		 * prefix match mode we will have to come back to wherever
1479		 * we started this traversal anyway
1480		 */
1481
1482		while ((n = rcu_dereference(*cptr)) == NULL) {
1483backtrace:
1484#ifdef CONFIG_IP_FIB_TRIE_STATS
1485			if (!n)
1486				this_cpu_inc(stats->null_node_hit);
1487#endif
1488			/* If we are at cindex 0 there are no more bits for
1489			 * us to strip at this level so we must ascend back
1490			 * up one level to see if there are any more bits to
1491			 * be stripped there.
1492			 */
1493			while (!cindex) {
1494				t_key pkey = pn->key;
1495
1496				/* If we don't have a parent then there is
1497				 * nothing for us to do as we do not have any
1498				 * further nodes to parse.
1499				 */
1500				if (IS_TRIE(pn)) {
1501					trace_fib_table_lookup(tb->tb_id, flp,
1502							       NULL, -EAGAIN);
1503					return -EAGAIN;
1504				}
1505#ifdef CONFIG_IP_FIB_TRIE_STATS
1506				this_cpu_inc(stats->backtrack);
1507#endif
1508				/* Get Child's index */
1509				pn = node_parent_rcu(pn);
1510				cindex = get_index(pkey, pn);
1511			}
1512
1513			/* strip the least significant bit from the cindex */
1514			cindex &= cindex - 1;
1515
1516			/* grab pointer for next child node */
1517			cptr = &pn->tnode[cindex];
1518		}
1519	}
1520
1521found:
1522	/* this line carries forward the xor from earlier in the function */
1523	index = key ^ n->key;
1524
1525	/* Step 3: Process the leaf, if that fails fall back to backtracing */
1526	hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1527		struct fib_info *fi = fa->fa_info;
1528		struct fib_nh_common *nhc;
1529		int nhsel, err;
1530
1531		if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1532			if (index >= (1ul << fa->fa_slen))
1533				continue;
1534		}
1535		if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1536			continue;
1537		/* Paired with WRITE_ONCE() in fib_release_info() */
1538		if (READ_ONCE(fi->fib_dead))
1539			continue;
1540		if (fa->fa_info->fib_scope < flp->flowi4_scope)
1541			continue;
1542		fib_alias_accessed(fa);
1543		err = fib_props[fa->fa_type].error;
1544		if (unlikely(err < 0)) {
1545out_reject:
1546#ifdef CONFIG_IP_FIB_TRIE_STATS
1547			this_cpu_inc(stats->semantic_match_passed);
1548#endif
1549			trace_fib_table_lookup(tb->tb_id, flp, NULL, err);
1550			return err;
1551		}
1552		if (fi->fib_flags & RTNH_F_DEAD)
1553			continue;
1554
1555		if (unlikely(fi->nh)) {
1556			if (nexthop_is_blackhole(fi->nh)) {
1557				err = fib_props[RTN_BLACKHOLE].error;
1558				goto out_reject;
1559			}
1560
1561			nhc = nexthop_get_nhc_lookup(fi->nh, fib_flags, flp,
1562						     &nhsel);
1563			if (nhc)
1564				goto set_result;
1565			goto miss;
1566		}
1567
1568		for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
1569			nhc = fib_info_nhc(fi, nhsel);
1570
1571			if (!fib_lookup_good_nhc(nhc, fib_flags, flp))
1572				continue;
1573set_result:
1574			if (!(fib_flags & FIB_LOOKUP_NOREF))
1575				refcount_inc(&fi->fib_clntref);
1576
1577			res->prefix = htonl(n->key);
1578			res->prefixlen = KEYLENGTH - fa->fa_slen;
1579			res->nh_sel = nhsel;
1580			res->nhc = nhc;
1581			res->type = fa->fa_type;
1582			res->scope = fi->fib_scope;
1583			res->fi = fi;
1584			res->table = tb;
1585			res->fa_head = &n->leaf;
1586#ifdef CONFIG_IP_FIB_TRIE_STATS
1587			this_cpu_inc(stats->semantic_match_passed);
1588#endif
1589			trace_fib_table_lookup(tb->tb_id, flp, nhc, err);
1590
1591			return err;
1592		}
1593	}
1594miss:
1595#ifdef CONFIG_IP_FIB_TRIE_STATS
1596	this_cpu_inc(stats->semantic_match_miss);
1597#endif
1598	goto backtrace;
1599}
1600EXPORT_SYMBOL_GPL(fib_table_lookup);
1601
1602static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1603			     struct key_vector *l, struct fib_alias *old)
1604{
1605	/* record the location of the previous list_info entry */
1606	struct hlist_node **pprev = old->fa_list.pprev;
1607	struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1608
1609	/* remove the fib_alias from the list */
1610	hlist_del_rcu(&old->fa_list);
1611
1612	/* if we emptied the list this leaf will be freed and we can sort
1613	 * out parent suffix lengths as a part of trie_rebalance
1614	 */
1615	if (hlist_empty(&l->leaf)) {
1616		if (tp->slen == l->slen)
1617			node_pull_suffix(tp, tp->pos);
1618		put_child_root(tp, l->key, NULL);
1619		node_free(l);
1620		trie_rebalance(t, tp);
1621		return;
1622	}
1623
1624	/* only access fa if it is pointing at the last valid hlist_node */
1625	if (*pprev)
1626		return;
1627
1628	/* update the trie with the latest suffix length */
1629	l->slen = fa->fa_slen;
1630	node_pull_suffix(tp, fa->fa_slen);
1631}
1632
1633static void fib_notify_alias_delete(struct net *net, u32 key,
1634				    struct hlist_head *fah,
1635				    struct fib_alias *fa_to_delete,
1636				    struct netlink_ext_ack *extack)
1637{
1638	struct fib_alias *fa_next, *fa_to_notify;
1639	u32 tb_id = fa_to_delete->tb_id;
1640	u8 slen = fa_to_delete->fa_slen;
1641	enum fib_event_type fib_event;
1642
1643	/* Do not notify if we do not care about the route. */
1644	if (fib_find_alias(fah, slen, 0, 0, tb_id, true) != fa_to_delete)
1645		return;
1646
1647	/* Determine if the route should be replaced by the next route in the
1648	 * list.
1649	 */
1650	fa_next = hlist_entry_safe(fa_to_delete->fa_list.next,
1651				   struct fib_alias, fa_list);
1652	if (fa_next && fa_next->fa_slen == slen && fa_next->tb_id == tb_id) {
1653		fib_event = FIB_EVENT_ENTRY_REPLACE;
1654		fa_to_notify = fa_next;
1655	} else {
1656		fib_event = FIB_EVENT_ENTRY_DEL;
1657		fa_to_notify = fa_to_delete;
1658	}
1659	call_fib_entry_notifiers(net, fib_event, key, KEYLENGTH - slen,
1660				 fa_to_notify, extack);
1661}
1662
1663/* Caller must hold RTNL. */
1664int fib_table_delete(struct net *net, struct fib_table *tb,
1665		     struct fib_config *cfg, struct netlink_ext_ack *extack)
1666{
1667	struct trie *t = (struct trie *) tb->tb_data;
1668	struct fib_alias *fa, *fa_to_delete;
1669	struct key_vector *l, *tp;
1670	u8 plen = cfg->fc_dst_len;
1671	u8 slen = KEYLENGTH - plen;
1672	u8 tos = cfg->fc_tos;
1673	u32 key;
1674
1675	key = ntohl(cfg->fc_dst);
1676
1677	if (!fib_valid_key_len(key, plen, extack))
1678		return -EINVAL;
1679
1680	l = fib_find_node(t, &tp, key);
1681	if (!l)
1682		return -ESRCH;
1683
1684	fa = fib_find_alias(&l->leaf, slen, tos, 0, tb->tb_id, false);
1685	if (!fa)
1686		return -ESRCH;
1687
1688	pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1689
1690	fa_to_delete = NULL;
1691	hlist_for_each_entry_from(fa, fa_list) {
1692		struct fib_info *fi = fa->fa_info;
1693
1694		if ((fa->fa_slen != slen) ||
1695		    (fa->tb_id != tb->tb_id) ||
1696		    (fa->fa_tos != tos))
1697			break;
1698
1699		if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1700		    (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1701		     fa->fa_info->fib_scope == cfg->fc_scope) &&
1702		    (!cfg->fc_prefsrc ||
1703		     fi->fib_prefsrc == cfg->fc_prefsrc) &&
1704		    (!cfg->fc_protocol ||
1705		     fi->fib_protocol == cfg->fc_protocol) &&
1706		    fib_nh_match(net, cfg, fi, extack) == 0 &&
1707		    fib_metrics_match(cfg, fi)) {
1708			fa_to_delete = fa;
1709			break;
1710		}
1711	}
1712
1713	if (!fa_to_delete)
1714		return -ESRCH;
1715
1716	fib_notify_alias_delete(net, key, &l->leaf, fa_to_delete, extack);
1717	rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1718		  &cfg->fc_nlinfo, 0);
1719
1720	if (!plen)
1721		tb->tb_num_default--;
1722
1723	fib_remove_alias(t, tp, l, fa_to_delete);
1724
1725	if (fa_to_delete->fa_state & FA_S_ACCESSED)
1726		rt_cache_flush(cfg->fc_nlinfo.nl_net);
1727
1728	fib_release_info(fa_to_delete->fa_info);
1729	alias_free_mem_rcu(fa_to_delete);
1730	return 0;
1731}
1732
1733/* Scan for the next leaf starting at the provided key value */
1734static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1735{
1736	struct key_vector *pn, *n = *tn;
1737	unsigned long cindex;
1738
1739	/* this loop is meant to try and find the key in the trie */
1740	do {
1741		/* record parent and next child index */
1742		pn = n;
1743		cindex = (key > pn->key) ? get_index(key, pn) : 0;
1744
1745		if (cindex >> pn->bits)
1746			break;
1747
1748		/* descend into the next child */
1749		n = get_child_rcu(pn, cindex++);
1750		if (!n)
1751			break;
1752
1753		/* guarantee forward progress on the keys */
1754		if (IS_LEAF(n) && (n->key >= key))
1755			goto found;
1756	} while (IS_TNODE(n));
1757
1758	/* this loop will search for the next leaf with a greater key */
1759	while (!IS_TRIE(pn)) {
1760		/* if we exhausted the parent node we will need to climb */
1761		if (cindex >= (1ul << pn->bits)) {
1762			t_key pkey = pn->key;
1763
1764			pn = node_parent_rcu(pn);
1765			cindex = get_index(pkey, pn) + 1;
1766			continue;
1767		}
1768
1769		/* grab the next available node */
1770		n = get_child_rcu(pn, cindex++);
1771		if (!n)
1772			continue;
1773
1774		/* no need to compare keys since we bumped the index */
1775		if (IS_LEAF(n))
1776			goto found;
1777
1778		/* Rescan start scanning in new node */
1779		pn = n;
1780		cindex = 0;
1781	}
1782
1783	*tn = pn;
1784	return NULL; /* Root of trie */
1785found:
1786	/* if we are at the limit for keys just return NULL for the tnode */
1787	*tn = pn;
1788	return n;
1789}
1790
1791static void fib_trie_free(struct fib_table *tb)
1792{
1793	struct trie *t = (struct trie *)tb->tb_data;
1794	struct key_vector *pn = t->kv;
1795	unsigned long cindex = 1;
1796	struct hlist_node *tmp;
1797	struct fib_alias *fa;
1798
1799	/* walk trie in reverse order and free everything */
1800	for (;;) {
1801		struct key_vector *n;
1802
1803		if (!(cindex--)) {
1804			t_key pkey = pn->key;
1805
1806			if (IS_TRIE(pn))
1807				break;
1808
1809			n = pn;
1810			pn = node_parent(pn);
1811
1812			/* drop emptied tnode */
1813			put_child_root(pn, n->key, NULL);
1814			node_free(n);
1815
1816			cindex = get_index(pkey, pn);
1817
1818			continue;
1819		}
1820
1821		/* grab the next available node */
1822		n = get_child(pn, cindex);
1823		if (!n)
1824			continue;
1825
1826		if (IS_TNODE(n)) {
1827			/* record pn and cindex for leaf walking */
1828			pn = n;
1829			cindex = 1ul << n->bits;
1830
1831			continue;
1832		}
1833
1834		hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1835			hlist_del_rcu(&fa->fa_list);
1836			alias_free_mem_rcu(fa);
1837		}
1838
1839		put_child_root(pn, n->key, NULL);
1840		node_free(n);
1841	}
1842
1843#ifdef CONFIG_IP_FIB_TRIE_STATS
1844	free_percpu(t->stats);
1845#endif
1846	kfree(tb);
1847}
1848
1849struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1850{
1851	struct trie *ot = (struct trie *)oldtb->tb_data;
1852	struct key_vector *l, *tp = ot->kv;
1853	struct fib_table *local_tb;
1854	struct fib_alias *fa;
1855	struct trie *lt;
1856	t_key key = 0;
1857
1858	if (oldtb->tb_data == oldtb->__data)
1859		return oldtb;
1860
1861	local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1862	if (!local_tb)
1863		return NULL;
1864
1865	lt = (struct trie *)local_tb->tb_data;
1866
1867	while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1868		struct key_vector *local_l = NULL, *local_tp;
1869
1870		hlist_for_each_entry(fa, &l->leaf, fa_list) {
1871			struct fib_alias *new_fa;
1872
1873			if (local_tb->tb_id != fa->tb_id)
1874				continue;
1875
1876			/* clone fa for new local table */
1877			new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1878			if (!new_fa)
1879				goto out;
1880
1881			memcpy(new_fa, fa, sizeof(*fa));
1882
1883			/* insert clone into table */
1884			if (!local_l)
1885				local_l = fib_find_node(lt, &local_tp, l->key);
1886
1887			if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1888					     NULL, l->key)) {
1889				kmem_cache_free(fn_alias_kmem, new_fa);
1890				goto out;
1891			}
1892		}
1893
1894		/* stop loop if key wrapped back to 0 */
1895		key = l->key + 1;
1896		if (key < l->key)
1897			break;
1898	}
1899
1900	return local_tb;
1901out:
1902	fib_trie_free(local_tb);
1903
1904	return NULL;
1905}
1906
1907/* Caller must hold RTNL */
1908void fib_table_flush_external(struct fib_table *tb)
1909{
1910	struct trie *t = (struct trie *)tb->tb_data;
1911	struct key_vector *pn = t->kv;
1912	unsigned long cindex = 1;
1913	struct hlist_node *tmp;
1914	struct fib_alias *fa;
1915
1916	/* walk trie in reverse order */
1917	for (;;) {
1918		unsigned char slen = 0;
1919		struct key_vector *n;
1920
1921		if (!(cindex--)) {
1922			t_key pkey = pn->key;
1923
1924			/* cannot resize the trie vector */
1925			if (IS_TRIE(pn))
1926				break;
1927
1928			/* update the suffix to address pulled leaves */
1929			if (pn->slen > pn->pos)
1930				update_suffix(pn);
1931
1932			/* resize completed node */
1933			pn = resize(t, pn);
1934			cindex = get_index(pkey, pn);
1935
1936			continue;
1937		}
1938
1939		/* grab the next available node */
1940		n = get_child(pn, cindex);
1941		if (!n)
1942			continue;
1943
1944		if (IS_TNODE(n)) {
1945			/* record pn and cindex for leaf walking */
1946			pn = n;
1947			cindex = 1ul << n->bits;
1948
1949			continue;
1950		}
1951
1952		hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1953			/* if alias was cloned to local then we just
1954			 * need to remove the local copy from main
1955			 */
1956			if (tb->tb_id != fa->tb_id) {
1957				hlist_del_rcu(&fa->fa_list);
1958				alias_free_mem_rcu(fa);
1959				continue;
1960			}
1961
1962			/* record local slen */
1963			slen = fa->fa_slen;
1964		}
1965
1966		/* update leaf slen */
1967		n->slen = slen;
1968
1969		if (hlist_empty(&n->leaf)) {
1970			put_child_root(pn, n->key, NULL);
1971			node_free(n);
1972		}
1973	}
1974}
1975
1976/* Caller must hold RTNL. */
1977int fib_table_flush(struct net *net, struct fib_table *tb, bool flush_all)
1978{
1979	struct trie *t = (struct trie *)tb->tb_data;
1980	struct nl_info info = { .nl_net = net };
1981	struct key_vector *pn = t->kv;
1982	unsigned long cindex = 1;
1983	struct hlist_node *tmp;
1984	struct fib_alias *fa;
1985	int found = 0;
1986
1987	/* walk trie in reverse order */
1988	for (;;) {
1989		unsigned char slen = 0;
1990		struct key_vector *n;
1991
1992		if (!(cindex--)) {
1993			t_key pkey = pn->key;
1994
1995			/* cannot resize the trie vector */
1996			if (IS_TRIE(pn))
1997				break;
1998
1999			/* update the suffix to address pulled leaves */
2000			if (pn->slen > pn->pos)
2001				update_suffix(pn);
2002
2003			/* resize completed node */
2004			pn = resize(t, pn);
2005			cindex = get_index(pkey, pn);
2006
2007			continue;
2008		}
2009
2010		/* grab the next available node */
2011		n = get_child(pn, cindex);
2012		if (!n)
2013			continue;
2014
2015		if (IS_TNODE(n)) {
2016			/* record pn and cindex for leaf walking */
2017			pn = n;
2018			cindex = 1ul << n->bits;
2019
2020			continue;
2021		}
2022
2023		hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
2024			struct fib_info *fi = fa->fa_info;
2025
2026			if (!fi || tb->tb_id != fa->tb_id ||
2027			    (!(fi->fib_flags & RTNH_F_DEAD) &&
2028			     !fib_props[fa->fa_type].error)) {
2029				slen = fa->fa_slen;
2030				continue;
2031			}
2032
2033			/* Do not flush error routes if network namespace is
2034			 * not being dismantled
2035			 */
2036			if (!flush_all && fib_props[fa->fa_type].error) {
2037				slen = fa->fa_slen;
2038				continue;
2039			}
2040
2041			fib_notify_alias_delete(net, n->key, &n->leaf, fa,
2042						NULL);
2043			if (fi->pfsrc_removed)
2044				rtmsg_fib(RTM_DELROUTE, htonl(n->key), fa,
2045					  KEYLENGTH - fa->fa_slen, tb->tb_id, &info, 0);
2046			hlist_del_rcu(&fa->fa_list);
2047			fib_release_info(fa->fa_info);
2048			alias_free_mem_rcu(fa);
2049			found++;
2050		}
2051
2052		/* update leaf slen */
2053		n->slen = slen;
2054
2055		if (hlist_empty(&n->leaf)) {
2056			put_child_root(pn, n->key, NULL);
2057			node_free(n);
2058		}
2059	}
2060
2061	pr_debug("trie_flush found=%d\n", found);
2062	return found;
2063}
2064
2065/* derived from fib_trie_free */
2066static void __fib_info_notify_update(struct net *net, struct fib_table *tb,
2067				     struct nl_info *info)
2068{
2069	struct trie *t = (struct trie *)tb->tb_data;
2070	struct key_vector *pn = t->kv;
2071	unsigned long cindex = 1;
2072	struct fib_alias *fa;
2073
2074	for (;;) {
2075		struct key_vector *n;
2076
2077		if (!(cindex--)) {
2078			t_key pkey = pn->key;
2079
2080			if (IS_TRIE(pn))
2081				break;
2082
2083			pn = node_parent(pn);
2084			cindex = get_index(pkey, pn);
2085			continue;
2086		}
2087
2088		/* grab the next available node */
2089		n = get_child(pn, cindex);
2090		if (!n)
2091			continue;
2092
2093		if (IS_TNODE(n)) {
2094			/* record pn and cindex for leaf walking */
2095			pn = n;
2096			cindex = 1ul << n->bits;
2097
2098			continue;
2099		}
2100
2101		hlist_for_each_entry(fa, &n->leaf, fa_list) {
2102			struct fib_info *fi = fa->fa_info;
2103
2104			if (!fi || !fi->nh_updated || fa->tb_id != tb->tb_id)
2105				continue;
2106
2107			rtmsg_fib(RTM_NEWROUTE, htonl(n->key), fa,
2108				  KEYLENGTH - fa->fa_slen, tb->tb_id,
2109				  info, NLM_F_REPLACE);
2110
2111			/* call_fib_entry_notifiers will be removed when
2112			 * in-kernel notifier is implemented and supported
2113			 * for nexthop objects
2114			 */
2115			call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_REPLACE,
2116						 n->key,
2117						 KEYLENGTH - fa->fa_slen, fa,
2118						 NULL);
2119		}
2120	}
2121}
2122
2123void fib_info_notify_update(struct net *net, struct nl_info *info)
2124{
2125	unsigned int h;
2126
2127	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2128		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2129		struct fib_table *tb;
2130
2131		hlist_for_each_entry_rcu(tb, head, tb_hlist,
2132					 lockdep_rtnl_is_held())
2133			__fib_info_notify_update(net, tb, info);
2134	}
2135}
2136
2137static int fib_leaf_notify(struct key_vector *l, struct fib_table *tb,
2138			   struct notifier_block *nb,
2139			   struct netlink_ext_ack *extack)
2140{
2141	struct fib_alias *fa;
2142	int last_slen = -1;
2143	int err;
2144
2145	hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2146		struct fib_info *fi = fa->fa_info;
2147
2148		if (!fi)
2149			continue;
2150
2151		/* local and main table can share the same trie,
2152		 * so don't notify twice for the same entry.
2153		 */
2154		if (tb->tb_id != fa->tb_id)
2155			continue;
2156
2157		if (fa->fa_slen == last_slen)
2158			continue;
2159
2160		last_slen = fa->fa_slen;
2161		err = call_fib_entry_notifier(nb, FIB_EVENT_ENTRY_REPLACE,
2162					      l->key, KEYLENGTH - fa->fa_slen,
2163					      fa, extack);
2164		if (err)
2165			return err;
2166	}
2167	return 0;
2168}
2169
2170static int fib_table_notify(struct fib_table *tb, struct notifier_block *nb,
2171			    struct netlink_ext_ack *extack)
2172{
2173	struct trie *t = (struct trie *)tb->tb_data;
2174	struct key_vector *l, *tp = t->kv;
2175	t_key key = 0;
2176	int err;
2177
2178	while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2179		err = fib_leaf_notify(l, tb, nb, extack);
2180		if (err)
2181			return err;
2182
2183		key = l->key + 1;
2184		/* stop in case of wrap around */
2185		if (key < l->key)
2186			break;
2187	}
2188	return 0;
2189}
2190
2191int fib_notify(struct net *net, struct notifier_block *nb,
2192	       struct netlink_ext_ack *extack)
2193{
2194	unsigned int h;
2195	int err;
2196
2197	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2198		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2199		struct fib_table *tb;
2200
2201		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2202			err = fib_table_notify(tb, nb, extack);
2203			if (err)
2204				return err;
2205		}
2206	}
2207	return 0;
2208}
2209
2210static void __trie_free_rcu(struct rcu_head *head)
2211{
2212	struct fib_table *tb = container_of(head, struct fib_table, rcu);
2213#ifdef CONFIG_IP_FIB_TRIE_STATS
2214	struct trie *t = (struct trie *)tb->tb_data;
2215
2216	if (tb->tb_data == tb->__data)
2217		free_percpu(t->stats);
2218#endif /* CONFIG_IP_FIB_TRIE_STATS */
2219	kfree(tb);
2220}
2221
2222void fib_free_table(struct fib_table *tb)
2223{
2224	call_rcu(&tb->rcu, __trie_free_rcu);
2225}
2226
2227static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
2228			     struct sk_buff *skb, struct netlink_callback *cb,
2229			     struct fib_dump_filter *filter)
2230{
2231	unsigned int flags = NLM_F_MULTI;
2232	__be32 xkey = htonl(l->key);
2233	int i, s_i, i_fa, s_fa, err;
2234	struct fib_alias *fa;
2235
2236	if (filter->filter_set ||
2237	    !filter->dump_exceptions || !filter->dump_routes)
2238		flags |= NLM_F_DUMP_FILTERED;
2239
2240	s_i = cb->args[4];
2241	s_fa = cb->args[5];
2242	i = 0;
2243
2244	/* rcu_read_lock is hold by caller */
2245	hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2246		struct fib_info *fi = fa->fa_info;
2247
2248		if (i < s_i)
2249			goto next;
2250
2251		i_fa = 0;
2252
2253		if (tb->tb_id != fa->tb_id)
2254			goto next;
2255
2256		if (filter->filter_set) {
2257			if (filter->rt_type && fa->fa_type != filter->rt_type)
2258				goto next;
2259
2260			if ((filter->protocol &&
2261			     fi->fib_protocol != filter->protocol))
2262				goto next;
2263
2264			if (filter->dev &&
2265			    !fib_info_nh_uses_dev(fi, filter->dev))
2266				goto next;
2267		}
2268
2269		if (filter->dump_routes) {
2270			if (!s_fa) {
2271				struct fib_rt_info fri;
2272
2273				fri.fi = fi;
2274				fri.tb_id = tb->tb_id;
2275				fri.dst = xkey;
2276				fri.dst_len = KEYLENGTH - fa->fa_slen;
2277				fri.tos = fa->fa_tos;
2278				fri.type = fa->fa_type;
2279				fri.offload = fa->offload;
2280				fri.trap = fa->trap;
2281				err = fib_dump_info(skb,
2282						    NETLINK_CB(cb->skb).portid,
2283						    cb->nlh->nlmsg_seq,
2284						    RTM_NEWROUTE, &fri, flags);
2285				if (err < 0)
2286					goto stop;
2287			}
2288
2289			i_fa++;
2290		}
2291
2292		if (filter->dump_exceptions) {
2293			err = fib_dump_info_fnhe(skb, cb, tb->tb_id, fi,
2294						 &i_fa, s_fa, flags);
2295			if (err < 0)
2296				goto stop;
2297		}
2298
2299next:
2300		i++;
2301	}
2302
2303	cb->args[4] = i;
2304	return skb->len;
2305
2306stop:
2307	cb->args[4] = i;
2308	cb->args[5] = i_fa;
2309	return err;
2310}
2311
2312/* rcu_read_lock needs to be hold by caller from readside */
2313int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
2314		   struct netlink_callback *cb, struct fib_dump_filter *filter)
2315{
2316	struct trie *t = (struct trie *)tb->tb_data;
2317	struct key_vector *l, *tp = t->kv;
2318	/* Dump starting at last key.
2319	 * Note: 0.0.0.0/0 (ie default) is first key.
2320	 */
2321	int count = cb->args[2];
2322	t_key key = cb->args[3];
2323
2324	/* First time here, count and key are both always 0. Count > 0
2325	 * and key == 0 means the dump has wrapped around and we are done.
2326	 */
2327	if (count && !key)
2328		return skb->len;
2329
2330	while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2331		int err;
2332
2333		err = fn_trie_dump_leaf(l, tb, skb, cb, filter);
2334		if (err < 0) {
2335			cb->args[3] = key;
2336			cb->args[2] = count;
2337			return err;
2338		}
2339
2340		++count;
2341		key = l->key + 1;
2342
2343		memset(&cb->args[4], 0,
2344		       sizeof(cb->args) - 4*sizeof(cb->args[0]));
2345
2346		/* stop loop if key wrapped back to 0 */
2347		if (key < l->key)
2348			break;
2349	}
2350
2351	cb->args[3] = key;
2352	cb->args[2] = count;
2353
2354	return skb->len;
2355}
2356
2357void __init fib_trie_init(void)
2358{
2359	fn_alias_kmem = kmem_cache_create("ip_fib_alias",
2360					  sizeof(struct fib_alias),
2361					  0, SLAB_PANIC, NULL);
2362
2363	trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
2364					   LEAF_SIZE,
2365					   0, SLAB_PANIC, NULL);
2366}
2367
2368struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
2369{
2370	struct fib_table *tb;
2371	struct trie *t;
2372	size_t sz = sizeof(*tb);
2373
2374	if (!alias)
2375		sz += sizeof(struct trie);
2376
2377	tb = kzalloc(sz, GFP_KERNEL);
2378	if (!tb)
2379		return NULL;
2380
2381	tb->tb_id = id;
2382	tb->tb_num_default = 0;
2383	tb->tb_data = (alias ? alias->__data : tb->__data);
2384
2385	if (alias)
2386		return tb;
2387
2388	t = (struct trie *) tb->tb_data;
2389	t->kv[0].pos = KEYLENGTH;
2390	t->kv[0].slen = KEYLENGTH;
2391#ifdef CONFIG_IP_FIB_TRIE_STATS
2392	t->stats = alloc_percpu(struct trie_use_stats);
2393	if (!t->stats) {
2394		kfree(tb);
2395		tb = NULL;
2396	}
2397#endif
2398
2399	return tb;
2400}
2401
2402#ifdef CONFIG_PROC_FS
2403/* Depth first Trie walk iterator */
2404struct fib_trie_iter {
2405	struct seq_net_private p;
2406	struct fib_table *tb;
2407	struct key_vector *tnode;
2408	unsigned int index;
2409	unsigned int depth;
2410};
2411
2412static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2413{
2414	unsigned long cindex = iter->index;
2415	struct key_vector *pn = iter->tnode;
2416	t_key pkey;
2417
2418	pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2419		 iter->tnode, iter->index, iter->depth);
2420
2421	while (!IS_TRIE(pn)) {
2422		while (cindex < child_length(pn)) {
2423			struct key_vector *n = get_child_rcu(pn, cindex++);
2424
2425			if (!n)
2426				continue;
2427
2428			if (IS_LEAF(n)) {
2429				iter->tnode = pn;
2430				iter->index = cindex;
2431			} else {
2432				/* push down one level */
2433				iter->tnode = n;
2434				iter->index = 0;
2435				++iter->depth;
2436			}
2437
2438			return n;
2439		}
2440
2441		/* Current node exhausted, pop back up */
2442		pkey = pn->key;
2443		pn = node_parent_rcu(pn);
2444		cindex = get_index(pkey, pn) + 1;
2445		--iter->depth;
2446	}
2447
2448	/* record root node so further searches know we are done */
2449	iter->tnode = pn;
2450	iter->index = 0;
2451
2452	return NULL;
2453}
2454
2455static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2456					     struct trie *t)
2457{
2458	struct key_vector *n, *pn;
2459
2460	if (!t)
2461		return NULL;
2462
2463	pn = t->kv;
2464	n = rcu_dereference(pn->tnode[0]);
2465	if (!n)
2466		return NULL;
2467
2468	if (IS_TNODE(n)) {
2469		iter->tnode = n;
2470		iter->index = 0;
2471		iter->depth = 1;
2472	} else {
2473		iter->tnode = pn;
2474		iter->index = 0;
2475		iter->depth = 0;
2476	}
2477
2478	return n;
2479}
2480
2481static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2482{
2483	struct key_vector *n;
2484	struct fib_trie_iter iter;
2485
2486	memset(s, 0, sizeof(*s));
2487
2488	rcu_read_lock();
2489	for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2490		if (IS_LEAF(n)) {
2491			struct fib_alias *fa;
2492
2493			s->leaves++;
2494			s->totdepth += iter.depth;
2495			if (iter.depth > s->maxdepth)
2496				s->maxdepth = iter.depth;
2497
2498			hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2499				++s->prefixes;
2500		} else {
2501			s->tnodes++;
2502			if (n->bits < MAX_STAT_DEPTH)
2503				s->nodesizes[n->bits]++;
2504			s->nullpointers += tn_info(n)->empty_children;
2505		}
2506	}
2507	rcu_read_unlock();
2508}
2509
2510/*
2511 *	This outputs /proc/net/fib_triestats
2512 */
2513static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2514{
2515	unsigned int i, max, pointers, bytes, avdepth;
2516
2517	if (stat->leaves)
2518		avdepth = stat->totdepth*100 / stat->leaves;
2519	else
2520		avdepth = 0;
2521
2522	seq_printf(seq, "\tAver depth:     %u.%02d\n",
2523		   avdepth / 100, avdepth % 100);
2524	seq_printf(seq, "\tMax depth:      %u\n", stat->maxdepth);
2525
2526	seq_printf(seq, "\tLeaves:         %u\n", stat->leaves);
2527	bytes = LEAF_SIZE * stat->leaves;
2528
2529	seq_printf(seq, "\tPrefixes:       %u\n", stat->prefixes);
2530	bytes += sizeof(struct fib_alias) * stat->prefixes;
2531
2532	seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2533	bytes += TNODE_SIZE(0) * stat->tnodes;
2534
2535	max = MAX_STAT_DEPTH;
2536	while (max > 0 && stat->nodesizes[max-1] == 0)
2537		max--;
2538
2539	pointers = 0;
2540	for (i = 1; i < max; i++)
2541		if (stat->nodesizes[i] != 0) {
2542			seq_printf(seq, "  %u: %u",  i, stat->nodesizes[i]);
2543			pointers += (1<<i) * stat->nodesizes[i];
2544		}
2545	seq_putc(seq, '\n');
2546	seq_printf(seq, "\tPointers: %u\n", pointers);
2547
2548	bytes += sizeof(struct key_vector *) * pointers;
2549	seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2550	seq_printf(seq, "Total size: %u  kB\n", (bytes + 1023) / 1024);
2551}
2552
2553#ifdef CONFIG_IP_FIB_TRIE_STATS
2554static void trie_show_usage(struct seq_file *seq,
2555			    const struct trie_use_stats __percpu *stats)
2556{
2557	struct trie_use_stats s = { 0 };
2558	int cpu;
2559
2560	/* loop through all of the CPUs and gather up the stats */
2561	for_each_possible_cpu(cpu) {
2562		const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2563
2564		s.gets += pcpu->gets;
2565		s.backtrack += pcpu->backtrack;
2566		s.semantic_match_passed += pcpu->semantic_match_passed;
2567		s.semantic_match_miss += pcpu->semantic_match_miss;
2568		s.null_node_hit += pcpu->null_node_hit;
2569		s.resize_node_skipped += pcpu->resize_node_skipped;
2570	}
2571
2572	seq_printf(seq, "\nCounters:\n---------\n");
2573	seq_printf(seq, "gets = %u\n", s.gets);
2574	seq_printf(seq, "backtracks = %u\n", s.backtrack);
2575	seq_printf(seq, "semantic match passed = %u\n",
2576		   s.semantic_match_passed);
2577	seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2578	seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2579	seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2580}
2581#endif /*  CONFIG_IP_FIB_TRIE_STATS */
2582
2583static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2584{
2585	if (tb->tb_id == RT_TABLE_LOCAL)
2586		seq_puts(seq, "Local:\n");
2587	else if (tb->tb_id == RT_TABLE_MAIN)
2588		seq_puts(seq, "Main:\n");
2589	else
2590		seq_printf(seq, "Id %d:\n", tb->tb_id);
2591}
2592
2593
2594static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2595{
2596	struct net *net = (struct net *)seq->private;
2597	unsigned int h;
2598
2599	seq_printf(seq,
2600		   "Basic info: size of leaf:"
2601		   " %zd bytes, size of tnode: %zd bytes.\n",
2602		   LEAF_SIZE, TNODE_SIZE(0));
2603
2604	rcu_read_lock();
2605	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2606		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2607		struct fib_table *tb;
2608
2609		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2610			struct trie *t = (struct trie *) tb->tb_data;
2611			struct trie_stat stat;
2612
2613			if (!t)
2614				continue;
2615
2616			fib_table_print(seq, tb);
2617
2618			trie_collect_stats(t, &stat);
2619			trie_show_stats(seq, &stat);
2620#ifdef CONFIG_IP_FIB_TRIE_STATS
2621			trie_show_usage(seq, t->stats);
2622#endif
2623		}
2624		cond_resched_rcu();
2625	}
2626	rcu_read_unlock();
2627
2628	return 0;
2629}
2630
2631static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2632{
2633	struct fib_trie_iter *iter = seq->private;
2634	struct net *net = seq_file_net(seq);
2635	loff_t idx = 0;
2636	unsigned int h;
2637
2638	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2639		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2640		struct fib_table *tb;
2641
2642		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2643			struct key_vector *n;
2644
2645			for (n = fib_trie_get_first(iter,
2646						    (struct trie *) tb->tb_data);
2647			     n; n = fib_trie_get_next(iter))
2648				if (pos == idx++) {
2649					iter->tb = tb;
2650					return n;
2651				}
2652		}
2653	}
2654
2655	return NULL;
2656}
2657
2658static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2659	__acquires(RCU)
2660{
2661	rcu_read_lock();
2662	return fib_trie_get_idx(seq, *pos);
2663}
2664
2665static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2666{
2667	struct fib_trie_iter *iter = seq->private;
2668	struct net *net = seq_file_net(seq);
2669	struct fib_table *tb = iter->tb;
2670	struct hlist_node *tb_node;
2671	unsigned int h;
2672	struct key_vector *n;
2673
2674	++*pos;
2675	/* next node in same table */
2676	n = fib_trie_get_next(iter);
2677	if (n)
2678		return n;
2679
2680	/* walk rest of this hash chain */
2681	h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2682	while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2683		tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2684		n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2685		if (n)
2686			goto found;
2687	}
2688
2689	/* new hash chain */
2690	while (++h < FIB_TABLE_HASHSZ) {
2691		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2692		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2693			n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2694			if (n)
2695				goto found;
2696		}
2697	}
2698	return NULL;
2699
2700found:
2701	iter->tb = tb;
2702	return n;
2703}
2704
2705static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2706	__releases(RCU)
2707{
2708	rcu_read_unlock();
2709}
2710
2711static void seq_indent(struct seq_file *seq, int n)
2712{
2713	while (n-- > 0)
2714		seq_puts(seq, "   ");
2715}
2716
2717static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2718{
2719	switch (s) {
2720	case RT_SCOPE_UNIVERSE: return "universe";
2721	case RT_SCOPE_SITE:	return "site";
2722	case RT_SCOPE_LINK:	return "link";
2723	case RT_SCOPE_HOST:	return "host";
2724	case RT_SCOPE_NOWHERE:	return "nowhere";
2725	default:
2726		snprintf(buf, len, "scope=%d", s);
2727		return buf;
2728	}
2729}
2730
2731static const char *const rtn_type_names[__RTN_MAX] = {
2732	[RTN_UNSPEC] = "UNSPEC",
2733	[RTN_UNICAST] = "UNICAST",
2734	[RTN_LOCAL] = "LOCAL",
2735	[RTN_BROADCAST] = "BROADCAST",
2736	[RTN_ANYCAST] = "ANYCAST",
2737	[RTN_MULTICAST] = "MULTICAST",
2738	[RTN_BLACKHOLE] = "BLACKHOLE",
2739	[RTN_UNREACHABLE] = "UNREACHABLE",
2740	[RTN_PROHIBIT] = "PROHIBIT",
2741	[RTN_THROW] = "THROW",
2742	[RTN_NAT] = "NAT",
2743	[RTN_XRESOLVE] = "XRESOLVE",
2744};
2745
2746static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2747{
2748	if (t < __RTN_MAX && rtn_type_names[t])
2749		return rtn_type_names[t];
2750	snprintf(buf, len, "type %u", t);
2751	return buf;
2752}
2753
2754/* Pretty print the trie */
2755static int fib_trie_seq_show(struct seq_file *seq, void *v)
2756{
2757	const struct fib_trie_iter *iter = seq->private;
2758	struct key_vector *n = v;
2759
2760	if (IS_TRIE(node_parent_rcu(n)))
2761		fib_table_print(seq, iter->tb);
2762
2763	if (IS_TNODE(n)) {
2764		__be32 prf = htonl(n->key);
2765
2766		seq_indent(seq, iter->depth-1);
2767		seq_printf(seq, "  +-- %pI4/%zu %u %u %u\n",
2768			   &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2769			   tn_info(n)->full_children,
2770			   tn_info(n)->empty_children);
2771	} else {
2772		__be32 val = htonl(n->key);
2773		struct fib_alias *fa;
2774
2775		seq_indent(seq, iter->depth);
2776		seq_printf(seq, "  |-- %pI4\n", &val);
2777
2778		hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2779			char buf1[32], buf2[32];
2780
2781			seq_indent(seq, iter->depth + 1);
2782			seq_printf(seq, "  /%zu %s %s",
2783				   KEYLENGTH - fa->fa_slen,
2784				   rtn_scope(buf1, sizeof(buf1),
2785					     fa->fa_info->fib_scope),
2786				   rtn_type(buf2, sizeof(buf2),
2787					    fa->fa_type));
2788			if (fa->fa_tos)
2789				seq_printf(seq, " tos=%d", fa->fa_tos);
2790			seq_putc(seq, '\n');
2791		}
2792	}
2793
2794	return 0;
2795}
2796
2797static const struct seq_operations fib_trie_seq_ops = {
2798	.start  = fib_trie_seq_start,
2799	.next   = fib_trie_seq_next,
2800	.stop   = fib_trie_seq_stop,
2801	.show   = fib_trie_seq_show,
2802};
2803
2804struct fib_route_iter {
2805	struct seq_net_private p;
2806	struct fib_table *main_tb;
2807	struct key_vector *tnode;
2808	loff_t	pos;
2809	t_key	key;
2810};
2811
2812static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2813					    loff_t pos)
2814{
2815	struct key_vector *l, **tp = &iter->tnode;
2816	t_key key;
2817
2818	/* use cached location of previously found key */
2819	if (iter->pos > 0 && pos >= iter->pos) {
2820		key = iter->key;
2821	} else {
2822		iter->pos = 1;
2823		key = 0;
2824	}
2825
2826	pos -= iter->pos;
2827
2828	while ((l = leaf_walk_rcu(tp, key)) && (pos-- > 0)) {
2829		key = l->key + 1;
2830		iter->pos++;
2831		l = NULL;
2832
2833		/* handle unlikely case of a key wrap */
2834		if (!key)
2835			break;
2836	}
2837
2838	if (l)
2839		iter->key = l->key;	/* remember it */
2840	else
2841		iter->pos = 0;		/* forget it */
2842
2843	return l;
2844}
2845
2846static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2847	__acquires(RCU)
2848{
2849	struct fib_route_iter *iter = seq->private;
2850	struct fib_table *tb;
2851	struct trie *t;
2852
2853	rcu_read_lock();
2854
2855	tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2856	if (!tb)
2857		return NULL;
2858
2859	iter->main_tb = tb;
2860	t = (struct trie *)tb->tb_data;
2861	iter->tnode = t->kv;
2862
2863	if (*pos != 0)
2864		return fib_route_get_idx(iter, *pos);
2865
2866	iter->pos = 0;
2867	iter->key = KEY_MAX;
2868
2869	return SEQ_START_TOKEN;
2870}
2871
2872static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2873{
2874	struct fib_route_iter *iter = seq->private;
2875	struct key_vector *l = NULL;
2876	t_key key = iter->key + 1;
2877
2878	++*pos;
2879
2880	/* only allow key of 0 for start of sequence */
2881	if ((v == SEQ_START_TOKEN) || key)
2882		l = leaf_walk_rcu(&iter->tnode, key);
2883
2884	if (l) {
2885		iter->key = l->key;
2886		iter->pos++;
2887	} else {
2888		iter->pos = 0;
2889	}
2890
2891	return l;
2892}
2893
2894static void fib_route_seq_stop(struct seq_file *seq, void *v)
2895	__releases(RCU)
2896{
2897	rcu_read_unlock();
2898}
2899
2900static unsigned int fib_flag_trans(int type, __be32 mask, struct fib_info *fi)
2901{
2902	unsigned int flags = 0;
2903
2904	if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2905		flags = RTF_REJECT;
2906	if (fi) {
2907		const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2908
2909		if (nhc->nhc_gw.ipv4)
2910			flags |= RTF_GATEWAY;
2911	}
2912	if (mask == htonl(0xFFFFFFFF))
2913		flags |= RTF_HOST;
2914	flags |= RTF_UP;
2915	return flags;
2916}
2917
2918/*
2919 *	This outputs /proc/net/route.
2920 *	The format of the file is not supposed to be changed
2921 *	and needs to be same as fib_hash output to avoid breaking
2922 *	legacy utilities
2923 */
2924static int fib_route_seq_show(struct seq_file *seq, void *v)
2925{
2926	struct fib_route_iter *iter = seq->private;
2927	struct fib_table *tb = iter->main_tb;
2928	struct fib_alias *fa;
2929	struct key_vector *l = v;
2930	__be32 prefix;
2931
2932	if (v == SEQ_START_TOKEN) {
2933		seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2934			   "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2935			   "\tWindow\tIRTT");
2936		return 0;
2937	}
2938
2939	prefix = htonl(l->key);
2940
2941	hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2942		struct fib_info *fi = fa->fa_info;
2943		__be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2944		unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2945
2946		if ((fa->fa_type == RTN_BROADCAST) ||
2947		    (fa->fa_type == RTN_MULTICAST))
2948			continue;
2949
2950		if (fa->tb_id != tb->tb_id)
2951			continue;
2952
2953		seq_setwidth(seq, 127);
2954
2955		if (fi) {
2956			struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2957			__be32 gw = 0;
2958
2959			if (nhc->nhc_gw_family == AF_INET)
2960				gw = nhc->nhc_gw.ipv4;
2961
2962			seq_printf(seq,
2963				   "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2964				   "%d\t%08X\t%d\t%u\t%u",
2965				   nhc->nhc_dev ? nhc->nhc_dev->name : "*",
2966				   prefix, gw, flags, 0, 0,
2967				   fi->fib_priority,
2968				   mask,
2969				   (fi->fib_advmss ?
2970				    fi->fib_advmss + 40 : 0),
2971				   fi->fib_window,
2972				   fi->fib_rtt >> 3);
2973		} else {
2974			seq_printf(seq,
2975				   "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2976				   "%d\t%08X\t%d\t%u\t%u",
2977				   prefix, 0, flags, 0, 0, 0,
2978				   mask, 0, 0, 0);
2979		}
2980		seq_pad(seq, '\n');
2981	}
2982
2983	return 0;
2984}
2985
2986static const struct seq_operations fib_route_seq_ops = {
2987	.start  = fib_route_seq_start,
2988	.next   = fib_route_seq_next,
2989	.stop   = fib_route_seq_stop,
2990	.show   = fib_route_seq_show,
2991};
2992
2993int __net_init fib_proc_init(struct net *net)
2994{
2995	if (!proc_create_net("fib_trie", 0444, net->proc_net, &fib_trie_seq_ops,
2996			sizeof(struct fib_trie_iter)))
2997		goto out1;
2998
2999	if (!proc_create_net_single("fib_triestat", 0444, net->proc_net,
3000			fib_triestat_seq_show, NULL))
3001		goto out2;
3002
3003	if (!proc_create_net("route", 0444, net->proc_net, &fib_route_seq_ops,
3004			sizeof(struct fib_route_iter)))
3005		goto out3;
3006
3007	return 0;
3008
3009out3:
3010	remove_proc_entry("fib_triestat", net->proc_net);
3011out2:
3012	remove_proc_entry("fib_trie", net->proc_net);
3013out1:
3014	return -ENOMEM;
3015}
3016
3017void __net_exit fib_proc_exit(struct net *net)
3018{
3019	remove_proc_entry("fib_trie", net->proc_net);
3020	remove_proc_entry("fib_triestat", net->proc_net);
3021	remove_proc_entry("route", net->proc_net);
3022}
3023
3024#endif /* CONFIG_PROC_FS */
3025