xref: /openbmc/linux/net/ipv4/fib_trie.c (revision 2d972b6a)
1 /*
2  *   This program is free software; you can redistribute it and/or
3  *   modify it under the terms of the GNU General Public License
4  *   as published by the Free Software Foundation; either version
5  *   2 of the License, or (at your option) any later version.
6  *
7  *   Robert Olsson <robert.olsson@its.uu.se> Uppsala Universitet
8  *     & Swedish University of Agricultural Sciences.
9  *
10  *   Jens Laas <jens.laas@data.slu.se> Swedish University of
11  *     Agricultural Sciences.
12  *
13  *   Hans Liss <hans.liss@its.uu.se>  Uppsala Universitet
14  *
15  * This work is based on the LPC-trie which is originally described in:
16  *
17  * An experimental study of compression methods for dynamic tries
18  * Stefan Nilsson and Matti Tikkanen. Algorithmica, 33(1):19-33, 2002.
19  * http://www.csc.kth.se/~snilsson/software/dyntrie2/
20  *
21  *
22  * IP-address lookup using LC-tries. Stefan Nilsson and Gunnar Karlsson
23  * IEEE Journal on Selected Areas in Communications, 17(6):1083-1092, June 1999
24  *
25  *
26  * Code from fib_hash has been reused which includes the following header:
27  *
28  *
29  * INET		An implementation of the TCP/IP protocol suite for the LINUX
30  *		operating system.  INET is implemented using the  BSD Socket
31  *		interface as the means of communication with the user level.
32  *
33  *		IPv4 FIB: lookup engine and maintenance routines.
34  *
35  *
36  * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
37  *
38  *		This program is free software; you can redistribute it and/or
39  *		modify it under the terms of the GNU General Public License
40  *		as published by the Free Software Foundation; either version
41  *		2 of the License, or (at your option) any later version.
42  *
43  * Substantial contributions to this work comes from:
44  *
45  *		David S. Miller, <davem@davemloft.net>
46  *		Stephen Hemminger <shemminger@osdl.org>
47  *		Paul E. McKenney <paulmck@us.ibm.com>
48  *		Patrick McHardy <kaber@trash.net>
49  */
50 
51 #define VERSION "0.409"
52 
53 #include <linux/cache.h>
54 #include <linux/uaccess.h>
55 #include <linux/bitops.h>
56 #include <linux/types.h>
57 #include <linux/kernel.h>
58 #include <linux/mm.h>
59 #include <linux/string.h>
60 #include <linux/socket.h>
61 #include <linux/sockios.h>
62 #include <linux/errno.h>
63 #include <linux/in.h>
64 #include <linux/inet.h>
65 #include <linux/inetdevice.h>
66 #include <linux/netdevice.h>
67 #include <linux/if_arp.h>
68 #include <linux/proc_fs.h>
69 #include <linux/rcupdate.h>
70 #include <linux/skbuff.h>
71 #include <linux/netlink.h>
72 #include <linux/init.h>
73 #include <linux/list.h>
74 #include <linux/slab.h>
75 #include <linux/export.h>
76 #include <linux/vmalloc.h>
77 #include <linux/notifier.h>
78 #include <net/net_namespace.h>
79 #include <net/ip.h>
80 #include <net/protocol.h>
81 #include <net/route.h>
82 #include <net/tcp.h>
83 #include <net/sock.h>
84 #include <net/ip_fib.h>
85 #include <net/fib_notifier.h>
86 #include <trace/events/fib.h>
87 #include "fib_lookup.h"
88 
89 static int call_fib_entry_notifier(struct notifier_block *nb, struct net *net,
90 				   enum fib_event_type event_type, u32 dst,
91 				   int dst_len, struct fib_alias *fa)
92 {
93 	struct fib_entry_notifier_info info = {
94 		.dst = dst,
95 		.dst_len = dst_len,
96 		.fi = fa->fa_info,
97 		.tos = fa->fa_tos,
98 		.type = fa->fa_type,
99 		.tb_id = fa->tb_id,
100 	};
101 	return call_fib4_notifier(nb, net, event_type, &info.info);
102 }
103 
104 static int call_fib_entry_notifiers(struct net *net,
105 				    enum fib_event_type event_type, u32 dst,
106 				    int dst_len, struct fib_alias *fa,
107 				    struct netlink_ext_ack *extack)
108 {
109 	struct fib_entry_notifier_info info = {
110 		.info.extack = extack,
111 		.dst = dst,
112 		.dst_len = dst_len,
113 		.fi = fa->fa_info,
114 		.tos = fa->fa_tos,
115 		.type = fa->fa_type,
116 		.tb_id = fa->tb_id,
117 	};
118 	return call_fib4_notifiers(net, event_type, &info.info);
119 }
120 
121 #define MAX_STAT_DEPTH 32
122 
123 #define KEYLENGTH	(8*sizeof(t_key))
124 #define KEY_MAX		((t_key)~0)
125 
126 typedef unsigned int t_key;
127 
128 #define IS_TRIE(n)	((n)->pos >= KEYLENGTH)
129 #define IS_TNODE(n)	((n)->bits)
130 #define IS_LEAF(n)	(!(n)->bits)
131 
132 struct key_vector {
133 	t_key key;
134 	unsigned char pos;		/* 2log(KEYLENGTH) bits needed */
135 	unsigned char bits;		/* 2log(KEYLENGTH) bits needed */
136 	unsigned char slen;
137 	union {
138 		/* This list pointer if valid if (pos | bits) == 0 (LEAF) */
139 		struct hlist_head leaf;
140 		/* This array is valid if (pos | bits) > 0 (TNODE) */
141 		struct key_vector __rcu *tnode[0];
142 	};
143 };
144 
145 struct tnode {
146 	struct rcu_head rcu;
147 	t_key empty_children;		/* KEYLENGTH bits needed */
148 	t_key full_children;		/* KEYLENGTH bits needed */
149 	struct key_vector __rcu *parent;
150 	struct key_vector kv[1];
151 #define tn_bits kv[0].bits
152 };
153 
154 #define TNODE_SIZE(n)	offsetof(struct tnode, kv[0].tnode[n])
155 #define LEAF_SIZE	TNODE_SIZE(1)
156 
157 #ifdef CONFIG_IP_FIB_TRIE_STATS
158 struct trie_use_stats {
159 	unsigned int gets;
160 	unsigned int backtrack;
161 	unsigned int semantic_match_passed;
162 	unsigned int semantic_match_miss;
163 	unsigned int null_node_hit;
164 	unsigned int resize_node_skipped;
165 };
166 #endif
167 
168 struct trie_stat {
169 	unsigned int totdepth;
170 	unsigned int maxdepth;
171 	unsigned int tnodes;
172 	unsigned int leaves;
173 	unsigned int nullpointers;
174 	unsigned int prefixes;
175 	unsigned int nodesizes[MAX_STAT_DEPTH];
176 };
177 
178 struct trie {
179 	struct key_vector kv[1];
180 #ifdef CONFIG_IP_FIB_TRIE_STATS
181 	struct trie_use_stats __percpu *stats;
182 #endif
183 };
184 
185 static struct key_vector *resize(struct trie *t, struct key_vector *tn);
186 static size_t tnode_free_size;
187 
188 /*
189  * synchronize_rcu after call_rcu for that many pages; it should be especially
190  * useful before resizing the root node with PREEMPT_NONE configs; the value was
191  * obtained experimentally, aiming to avoid visible slowdown.
192  */
193 static const int sync_pages = 128;
194 
195 static struct kmem_cache *fn_alias_kmem __ro_after_init;
196 static struct kmem_cache *trie_leaf_kmem __ro_after_init;
197 
198 static inline struct tnode *tn_info(struct key_vector *kv)
199 {
200 	return container_of(kv, struct tnode, kv[0]);
201 }
202 
203 /* caller must hold RTNL */
204 #define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
205 #define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
206 
207 /* caller must hold RCU read lock or RTNL */
208 #define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
209 #define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
210 
211 /* wrapper for rcu_assign_pointer */
212 static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
213 {
214 	if (n)
215 		rcu_assign_pointer(tn_info(n)->parent, tp);
216 }
217 
218 #define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
219 
220 /* This provides us with the number of children in this node, in the case of a
221  * leaf this will return 0 meaning none of the children are accessible.
222  */
223 static inline unsigned long child_length(const struct key_vector *tn)
224 {
225 	return (1ul << tn->bits) & ~(1ul);
226 }
227 
228 #define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
229 
230 static inline unsigned long get_index(t_key key, struct key_vector *kv)
231 {
232 	unsigned long index = key ^ kv->key;
233 
234 	if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
235 		return 0;
236 
237 	return index >> kv->pos;
238 }
239 
240 /* To understand this stuff, an understanding of keys and all their bits is
241  * necessary. Every node in the trie has a key associated with it, but not
242  * all of the bits in that key are significant.
243  *
244  * Consider a node 'n' and its parent 'tp'.
245  *
246  * If n is a leaf, every bit in its key is significant. Its presence is
247  * necessitated by path compression, since during a tree traversal (when
248  * searching for a leaf - unless we are doing an insertion) we will completely
249  * ignore all skipped bits we encounter. Thus we need to verify, at the end of
250  * a potentially successful search, that we have indeed been walking the
251  * correct key path.
252  *
253  * Note that we can never "miss" the correct key in the tree if present by
254  * following the wrong path. Path compression ensures that segments of the key
255  * that are the same for all keys with a given prefix are skipped, but the
256  * skipped part *is* identical for each node in the subtrie below the skipped
257  * bit! trie_insert() in this implementation takes care of that.
258  *
259  * if n is an internal node - a 'tnode' here, the various parts of its key
260  * have many different meanings.
261  *
262  * Example:
263  * _________________________________________________________________
264  * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
265  * -----------------------------------------------------------------
266  *  31  30  29  28  27  26  25  24  23  22  21  20  19  18  17  16
267  *
268  * _________________________________________________________________
269  * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
270  * -----------------------------------------------------------------
271  *  15  14  13  12  11  10   9   8   7   6   5   4   3   2   1   0
272  *
273  * tp->pos = 22
274  * tp->bits = 3
275  * n->pos = 13
276  * n->bits = 4
277  *
278  * First, let's just ignore the bits that come before the parent tp, that is
279  * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
280  * point we do not use them for anything.
281  *
282  * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
283  * index into the parent's child array. That is, they will be used to find
284  * 'n' among tp's children.
285  *
286  * The bits from (n->pos + n->bits) to (tp->pos - 1) - "S" - are skipped bits
287  * for the node n.
288  *
289  * All the bits we have seen so far are significant to the node n. The rest
290  * of the bits are really not needed or indeed known in n->key.
291  *
292  * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
293  * n's child array, and will of course be different for each child.
294  *
295  * The rest of the bits, from 0 to (n->pos -1) - "u" - are completely unknown
296  * at this point.
297  */
298 
299 static const int halve_threshold = 25;
300 static const int inflate_threshold = 50;
301 static const int halve_threshold_root = 15;
302 static const int inflate_threshold_root = 30;
303 
304 static void __alias_free_mem(struct rcu_head *head)
305 {
306 	struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
307 	kmem_cache_free(fn_alias_kmem, fa);
308 }
309 
310 static inline void alias_free_mem_rcu(struct fib_alias *fa)
311 {
312 	call_rcu(&fa->rcu, __alias_free_mem);
313 }
314 
315 #define TNODE_KMALLOC_MAX \
316 	ilog2((PAGE_SIZE - TNODE_SIZE(0)) / sizeof(struct key_vector *))
317 #define TNODE_VMALLOC_MAX \
318 	ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
319 
320 static void __node_free_rcu(struct rcu_head *head)
321 {
322 	struct tnode *n = container_of(head, struct tnode, rcu);
323 
324 	if (!n->tn_bits)
325 		kmem_cache_free(trie_leaf_kmem, n);
326 	else
327 		kvfree(n);
328 }
329 
330 #define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
331 
332 static struct tnode *tnode_alloc(int bits)
333 {
334 	size_t size;
335 
336 	/* verify bits is within bounds */
337 	if (bits > TNODE_VMALLOC_MAX)
338 		return NULL;
339 
340 	/* determine size and verify it is non-zero and didn't overflow */
341 	size = TNODE_SIZE(1ul << bits);
342 
343 	if (size <= PAGE_SIZE)
344 		return kzalloc(size, GFP_KERNEL);
345 	else
346 		return vzalloc(size);
347 }
348 
349 static inline void empty_child_inc(struct key_vector *n)
350 {
351 	++tn_info(n)->empty_children ? : ++tn_info(n)->full_children;
352 }
353 
354 static inline void empty_child_dec(struct key_vector *n)
355 {
356 	tn_info(n)->empty_children-- ? : tn_info(n)->full_children--;
357 }
358 
359 static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
360 {
361 	struct key_vector *l;
362 	struct tnode *kv;
363 
364 	kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
365 	if (!kv)
366 		return NULL;
367 
368 	/* initialize key vector */
369 	l = kv->kv;
370 	l->key = key;
371 	l->pos = 0;
372 	l->bits = 0;
373 	l->slen = fa->fa_slen;
374 
375 	/* link leaf to fib alias */
376 	INIT_HLIST_HEAD(&l->leaf);
377 	hlist_add_head(&fa->fa_list, &l->leaf);
378 
379 	return l;
380 }
381 
382 static struct key_vector *tnode_new(t_key key, int pos, int bits)
383 {
384 	unsigned int shift = pos + bits;
385 	struct key_vector *tn;
386 	struct tnode *tnode;
387 
388 	/* verify bits and pos their msb bits clear and values are valid */
389 	BUG_ON(!bits || (shift > KEYLENGTH));
390 
391 	tnode = tnode_alloc(bits);
392 	if (!tnode)
393 		return NULL;
394 
395 	pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
396 		 sizeof(struct key_vector *) << bits);
397 
398 	if (bits == KEYLENGTH)
399 		tnode->full_children = 1;
400 	else
401 		tnode->empty_children = 1ul << bits;
402 
403 	tn = tnode->kv;
404 	tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
405 	tn->pos = pos;
406 	tn->bits = bits;
407 	tn->slen = pos;
408 
409 	return tn;
410 }
411 
412 /* Check whether a tnode 'n' is "full", i.e. it is an internal node
413  * and no bits are skipped. See discussion in dyntree paper p. 6
414  */
415 static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
416 {
417 	return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
418 }
419 
420 /* Add a child at position i overwriting the old value.
421  * Update the value of full_children and empty_children.
422  */
423 static void put_child(struct key_vector *tn, unsigned long i,
424 		      struct key_vector *n)
425 {
426 	struct key_vector *chi = get_child(tn, i);
427 	int isfull, wasfull;
428 
429 	BUG_ON(i >= child_length(tn));
430 
431 	/* update emptyChildren, overflow into fullChildren */
432 	if (!n && chi)
433 		empty_child_inc(tn);
434 	if (n && !chi)
435 		empty_child_dec(tn);
436 
437 	/* update fullChildren */
438 	wasfull = tnode_full(tn, chi);
439 	isfull = tnode_full(tn, n);
440 
441 	if (wasfull && !isfull)
442 		tn_info(tn)->full_children--;
443 	else if (!wasfull && isfull)
444 		tn_info(tn)->full_children++;
445 
446 	if (n && (tn->slen < n->slen))
447 		tn->slen = n->slen;
448 
449 	rcu_assign_pointer(tn->tnode[i], n);
450 }
451 
452 static void update_children(struct key_vector *tn)
453 {
454 	unsigned long i;
455 
456 	/* update all of the child parent pointers */
457 	for (i = child_length(tn); i;) {
458 		struct key_vector *inode = get_child(tn, --i);
459 
460 		if (!inode)
461 			continue;
462 
463 		/* Either update the children of a tnode that
464 		 * already belongs to us or update the child
465 		 * to point to ourselves.
466 		 */
467 		if (node_parent(inode) == tn)
468 			update_children(inode);
469 		else
470 			node_set_parent(inode, tn);
471 	}
472 }
473 
474 static inline void put_child_root(struct key_vector *tp, t_key key,
475 				  struct key_vector *n)
476 {
477 	if (IS_TRIE(tp))
478 		rcu_assign_pointer(tp->tnode[0], n);
479 	else
480 		put_child(tp, get_index(key, tp), n);
481 }
482 
483 static inline void tnode_free_init(struct key_vector *tn)
484 {
485 	tn_info(tn)->rcu.next = NULL;
486 }
487 
488 static inline void tnode_free_append(struct key_vector *tn,
489 				     struct key_vector *n)
490 {
491 	tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
492 	tn_info(tn)->rcu.next = &tn_info(n)->rcu;
493 }
494 
495 static void tnode_free(struct key_vector *tn)
496 {
497 	struct callback_head *head = &tn_info(tn)->rcu;
498 
499 	while (head) {
500 		head = head->next;
501 		tnode_free_size += TNODE_SIZE(1ul << tn->bits);
502 		node_free(tn);
503 
504 		tn = container_of(head, struct tnode, rcu)->kv;
505 	}
506 
507 	if (tnode_free_size >= PAGE_SIZE * sync_pages) {
508 		tnode_free_size = 0;
509 		synchronize_rcu();
510 	}
511 }
512 
513 static struct key_vector *replace(struct trie *t,
514 				  struct key_vector *oldtnode,
515 				  struct key_vector *tn)
516 {
517 	struct key_vector *tp = node_parent(oldtnode);
518 	unsigned long i;
519 
520 	/* setup the parent pointer out of and back into this node */
521 	NODE_INIT_PARENT(tn, tp);
522 	put_child_root(tp, tn->key, tn);
523 
524 	/* update all of the child parent pointers */
525 	update_children(tn);
526 
527 	/* all pointers should be clean so we are done */
528 	tnode_free(oldtnode);
529 
530 	/* resize children now that oldtnode is freed */
531 	for (i = child_length(tn); i;) {
532 		struct key_vector *inode = get_child(tn, --i);
533 
534 		/* resize child node */
535 		if (tnode_full(tn, inode))
536 			tn = resize(t, inode);
537 	}
538 
539 	return tp;
540 }
541 
542 static struct key_vector *inflate(struct trie *t,
543 				  struct key_vector *oldtnode)
544 {
545 	struct key_vector *tn;
546 	unsigned long i;
547 	t_key m;
548 
549 	pr_debug("In inflate\n");
550 
551 	tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
552 	if (!tn)
553 		goto notnode;
554 
555 	/* prepare oldtnode to be freed */
556 	tnode_free_init(oldtnode);
557 
558 	/* Assemble all of the pointers in our cluster, in this case that
559 	 * represents all of the pointers out of our allocated nodes that
560 	 * point to existing tnodes and the links between our allocated
561 	 * nodes.
562 	 */
563 	for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
564 		struct key_vector *inode = get_child(oldtnode, --i);
565 		struct key_vector *node0, *node1;
566 		unsigned long j, k;
567 
568 		/* An empty child */
569 		if (!inode)
570 			continue;
571 
572 		/* A leaf or an internal node with skipped bits */
573 		if (!tnode_full(oldtnode, inode)) {
574 			put_child(tn, get_index(inode->key, tn), inode);
575 			continue;
576 		}
577 
578 		/* drop the node in the old tnode free list */
579 		tnode_free_append(oldtnode, inode);
580 
581 		/* An internal node with two children */
582 		if (inode->bits == 1) {
583 			put_child(tn, 2 * i + 1, get_child(inode, 1));
584 			put_child(tn, 2 * i, get_child(inode, 0));
585 			continue;
586 		}
587 
588 		/* We will replace this node 'inode' with two new
589 		 * ones, 'node0' and 'node1', each with half of the
590 		 * original children. The two new nodes will have
591 		 * a position one bit further down the key and this
592 		 * means that the "significant" part of their keys
593 		 * (see the discussion near the top of this file)
594 		 * will differ by one bit, which will be "0" in
595 		 * node0's key and "1" in node1's key. Since we are
596 		 * moving the key position by one step, the bit that
597 		 * we are moving away from - the bit at position
598 		 * (tn->pos) - is the one that will differ between
599 		 * node0 and node1. So... we synthesize that bit in the
600 		 * two new keys.
601 		 */
602 		node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
603 		if (!node1)
604 			goto nomem;
605 		node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
606 
607 		tnode_free_append(tn, node1);
608 		if (!node0)
609 			goto nomem;
610 		tnode_free_append(tn, node0);
611 
612 		/* populate child pointers in new nodes */
613 		for (k = child_length(inode), j = k / 2; j;) {
614 			put_child(node1, --j, get_child(inode, --k));
615 			put_child(node0, j, get_child(inode, j));
616 			put_child(node1, --j, get_child(inode, --k));
617 			put_child(node0, j, get_child(inode, j));
618 		}
619 
620 		/* link new nodes to parent */
621 		NODE_INIT_PARENT(node1, tn);
622 		NODE_INIT_PARENT(node0, tn);
623 
624 		/* link parent to nodes */
625 		put_child(tn, 2 * i + 1, node1);
626 		put_child(tn, 2 * i, node0);
627 	}
628 
629 	/* setup the parent pointers into and out of this node */
630 	return replace(t, oldtnode, tn);
631 nomem:
632 	/* all pointers should be clean so we are done */
633 	tnode_free(tn);
634 notnode:
635 	return NULL;
636 }
637 
638 static struct key_vector *halve(struct trie *t,
639 				struct key_vector *oldtnode)
640 {
641 	struct key_vector *tn;
642 	unsigned long i;
643 
644 	pr_debug("In halve\n");
645 
646 	tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
647 	if (!tn)
648 		goto notnode;
649 
650 	/* prepare oldtnode to be freed */
651 	tnode_free_init(oldtnode);
652 
653 	/* Assemble all of the pointers in our cluster, in this case that
654 	 * represents all of the pointers out of our allocated nodes that
655 	 * point to existing tnodes and the links between our allocated
656 	 * nodes.
657 	 */
658 	for (i = child_length(oldtnode); i;) {
659 		struct key_vector *node1 = get_child(oldtnode, --i);
660 		struct key_vector *node0 = get_child(oldtnode, --i);
661 		struct key_vector *inode;
662 
663 		/* At least one of the children is empty */
664 		if (!node1 || !node0) {
665 			put_child(tn, i / 2, node1 ? : node0);
666 			continue;
667 		}
668 
669 		/* Two nonempty children */
670 		inode = tnode_new(node0->key, oldtnode->pos, 1);
671 		if (!inode)
672 			goto nomem;
673 		tnode_free_append(tn, inode);
674 
675 		/* initialize pointers out of node */
676 		put_child(inode, 1, node1);
677 		put_child(inode, 0, node0);
678 		NODE_INIT_PARENT(inode, tn);
679 
680 		/* link parent to node */
681 		put_child(tn, i / 2, inode);
682 	}
683 
684 	/* setup the parent pointers into and out of this node */
685 	return replace(t, oldtnode, tn);
686 nomem:
687 	/* all pointers should be clean so we are done */
688 	tnode_free(tn);
689 notnode:
690 	return NULL;
691 }
692 
693 static struct key_vector *collapse(struct trie *t,
694 				   struct key_vector *oldtnode)
695 {
696 	struct key_vector *n, *tp;
697 	unsigned long i;
698 
699 	/* scan the tnode looking for that one child that might still exist */
700 	for (n = NULL, i = child_length(oldtnode); !n && i;)
701 		n = get_child(oldtnode, --i);
702 
703 	/* compress one level */
704 	tp = node_parent(oldtnode);
705 	put_child_root(tp, oldtnode->key, n);
706 	node_set_parent(n, tp);
707 
708 	/* drop dead node */
709 	node_free(oldtnode);
710 
711 	return tp;
712 }
713 
714 static unsigned char update_suffix(struct key_vector *tn)
715 {
716 	unsigned char slen = tn->pos;
717 	unsigned long stride, i;
718 	unsigned char slen_max;
719 
720 	/* only vector 0 can have a suffix length greater than or equal to
721 	 * tn->pos + tn->bits, the second highest node will have a suffix
722 	 * length at most of tn->pos + tn->bits - 1
723 	 */
724 	slen_max = min_t(unsigned char, tn->pos + tn->bits - 1, tn->slen);
725 
726 	/* search though the list of children looking for nodes that might
727 	 * have a suffix greater than the one we currently have.  This is
728 	 * why we start with a stride of 2 since a stride of 1 would
729 	 * represent the nodes with suffix length equal to tn->pos
730 	 */
731 	for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
732 		struct key_vector *n = get_child(tn, i);
733 
734 		if (!n || (n->slen <= slen))
735 			continue;
736 
737 		/* update stride and slen based on new value */
738 		stride <<= (n->slen - slen);
739 		slen = n->slen;
740 		i &= ~(stride - 1);
741 
742 		/* stop searching if we have hit the maximum possible value */
743 		if (slen >= slen_max)
744 			break;
745 	}
746 
747 	tn->slen = slen;
748 
749 	return slen;
750 }
751 
752 /* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
753  * the Helsinki University of Technology and Matti Tikkanen of Nokia
754  * Telecommunications, page 6:
755  * "A node is doubled if the ratio of non-empty children to all
756  * children in the *doubled* node is at least 'high'."
757  *
758  * 'high' in this instance is the variable 'inflate_threshold'. It
759  * is expressed as a percentage, so we multiply it with
760  * child_length() and instead of multiplying by 2 (since the
761  * child array will be doubled by inflate()) and multiplying
762  * the left-hand side by 100 (to handle the percentage thing) we
763  * multiply the left-hand side by 50.
764  *
765  * The left-hand side may look a bit weird: child_length(tn)
766  * - tn->empty_children is of course the number of non-null children
767  * in the current node. tn->full_children is the number of "full"
768  * children, that is non-null tnodes with a skip value of 0.
769  * All of those will be doubled in the resulting inflated tnode, so
770  * we just count them one extra time here.
771  *
772  * A clearer way to write this would be:
773  *
774  * to_be_doubled = tn->full_children;
775  * not_to_be_doubled = child_length(tn) - tn->empty_children -
776  *     tn->full_children;
777  *
778  * new_child_length = child_length(tn) * 2;
779  *
780  * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
781  *      new_child_length;
782  * if (new_fill_factor >= inflate_threshold)
783  *
784  * ...and so on, tho it would mess up the while () loop.
785  *
786  * anyway,
787  * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
788  *      inflate_threshold
789  *
790  * avoid a division:
791  * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
792  *      inflate_threshold * new_child_length
793  *
794  * expand not_to_be_doubled and to_be_doubled, and shorten:
795  * 100 * (child_length(tn) - tn->empty_children +
796  *    tn->full_children) >= inflate_threshold * new_child_length
797  *
798  * expand new_child_length:
799  * 100 * (child_length(tn) - tn->empty_children +
800  *    tn->full_children) >=
801  *      inflate_threshold * child_length(tn) * 2
802  *
803  * shorten again:
804  * 50 * (tn->full_children + child_length(tn) -
805  *    tn->empty_children) >= inflate_threshold *
806  *    child_length(tn)
807  *
808  */
809 static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
810 {
811 	unsigned long used = child_length(tn);
812 	unsigned long threshold = used;
813 
814 	/* Keep root node larger */
815 	threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
816 	used -= tn_info(tn)->empty_children;
817 	used += tn_info(tn)->full_children;
818 
819 	/* if bits == KEYLENGTH then pos = 0, and will fail below */
820 
821 	return (used > 1) && tn->pos && ((50 * used) >= threshold);
822 }
823 
824 static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
825 {
826 	unsigned long used = child_length(tn);
827 	unsigned long threshold = used;
828 
829 	/* Keep root node larger */
830 	threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
831 	used -= tn_info(tn)->empty_children;
832 
833 	/* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
834 
835 	return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
836 }
837 
838 static inline bool should_collapse(struct key_vector *tn)
839 {
840 	unsigned long used = child_length(tn);
841 
842 	used -= tn_info(tn)->empty_children;
843 
844 	/* account for bits == KEYLENGTH case */
845 	if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
846 		used -= KEY_MAX;
847 
848 	/* One child or none, time to drop us from the trie */
849 	return used < 2;
850 }
851 
852 #define MAX_WORK 10
853 static struct key_vector *resize(struct trie *t, struct key_vector *tn)
854 {
855 #ifdef CONFIG_IP_FIB_TRIE_STATS
856 	struct trie_use_stats __percpu *stats = t->stats;
857 #endif
858 	struct key_vector *tp = node_parent(tn);
859 	unsigned long cindex = get_index(tn->key, tp);
860 	int max_work = MAX_WORK;
861 
862 	pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
863 		 tn, inflate_threshold, halve_threshold);
864 
865 	/* track the tnode via the pointer from the parent instead of
866 	 * doing it ourselves.  This way we can let RCU fully do its
867 	 * thing without us interfering
868 	 */
869 	BUG_ON(tn != get_child(tp, cindex));
870 
871 	/* Double as long as the resulting node has a number of
872 	 * nonempty nodes that are above the threshold.
873 	 */
874 	while (should_inflate(tp, tn) && max_work) {
875 		tp = inflate(t, tn);
876 		if (!tp) {
877 #ifdef CONFIG_IP_FIB_TRIE_STATS
878 			this_cpu_inc(stats->resize_node_skipped);
879 #endif
880 			break;
881 		}
882 
883 		max_work--;
884 		tn = get_child(tp, cindex);
885 	}
886 
887 	/* update parent in case inflate failed */
888 	tp = node_parent(tn);
889 
890 	/* Return if at least one inflate is run */
891 	if (max_work != MAX_WORK)
892 		return tp;
893 
894 	/* Halve as long as the number of empty children in this
895 	 * node is above threshold.
896 	 */
897 	while (should_halve(tp, tn) && max_work) {
898 		tp = halve(t, tn);
899 		if (!tp) {
900 #ifdef CONFIG_IP_FIB_TRIE_STATS
901 			this_cpu_inc(stats->resize_node_skipped);
902 #endif
903 			break;
904 		}
905 
906 		max_work--;
907 		tn = get_child(tp, cindex);
908 	}
909 
910 	/* Only one child remains */
911 	if (should_collapse(tn))
912 		return collapse(t, tn);
913 
914 	/* update parent in case halve failed */
915 	return node_parent(tn);
916 }
917 
918 static void node_pull_suffix(struct key_vector *tn, unsigned char slen)
919 {
920 	unsigned char node_slen = tn->slen;
921 
922 	while ((node_slen > tn->pos) && (node_slen > slen)) {
923 		slen = update_suffix(tn);
924 		if (node_slen == slen)
925 			break;
926 
927 		tn = node_parent(tn);
928 		node_slen = tn->slen;
929 	}
930 }
931 
932 static void node_push_suffix(struct key_vector *tn, unsigned char slen)
933 {
934 	while (tn->slen < slen) {
935 		tn->slen = slen;
936 		tn = node_parent(tn);
937 	}
938 }
939 
940 /* rcu_read_lock needs to be hold by caller from readside */
941 static struct key_vector *fib_find_node(struct trie *t,
942 					struct key_vector **tp, u32 key)
943 {
944 	struct key_vector *pn, *n = t->kv;
945 	unsigned long index = 0;
946 
947 	do {
948 		pn = n;
949 		n = get_child_rcu(n, index);
950 
951 		if (!n)
952 			break;
953 
954 		index = get_cindex(key, n);
955 
956 		/* This bit of code is a bit tricky but it combines multiple
957 		 * checks into a single check.  The prefix consists of the
958 		 * prefix plus zeros for the bits in the cindex. The index
959 		 * is the difference between the key and this value.  From
960 		 * this we can actually derive several pieces of data.
961 		 *   if (index >= (1ul << bits))
962 		 *     we have a mismatch in skip bits and failed
963 		 *   else
964 		 *     we know the value is cindex
965 		 *
966 		 * This check is safe even if bits == KEYLENGTH due to the
967 		 * fact that we can only allocate a node with 32 bits if a
968 		 * long is greater than 32 bits.
969 		 */
970 		if (index >= (1ul << n->bits)) {
971 			n = NULL;
972 			break;
973 		}
974 
975 		/* keep searching until we find a perfect match leaf or NULL */
976 	} while (IS_TNODE(n));
977 
978 	*tp = pn;
979 
980 	return n;
981 }
982 
983 /* Return the first fib alias matching TOS with
984  * priority less than or equal to PRIO.
985  */
986 static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
987 					u8 tos, u32 prio, u32 tb_id)
988 {
989 	struct fib_alias *fa;
990 
991 	if (!fah)
992 		return NULL;
993 
994 	hlist_for_each_entry(fa, fah, fa_list) {
995 		if (fa->fa_slen < slen)
996 			continue;
997 		if (fa->fa_slen != slen)
998 			break;
999 		if (fa->tb_id > tb_id)
1000 			continue;
1001 		if (fa->tb_id != tb_id)
1002 			break;
1003 		if (fa->fa_tos > tos)
1004 			continue;
1005 		if (fa->fa_info->fib_priority >= prio || fa->fa_tos < tos)
1006 			return fa;
1007 	}
1008 
1009 	return NULL;
1010 }
1011 
1012 static void trie_rebalance(struct trie *t, struct key_vector *tn)
1013 {
1014 	while (!IS_TRIE(tn))
1015 		tn = resize(t, tn);
1016 }
1017 
1018 static int fib_insert_node(struct trie *t, struct key_vector *tp,
1019 			   struct fib_alias *new, t_key key)
1020 {
1021 	struct key_vector *n, *l;
1022 
1023 	l = leaf_new(key, new);
1024 	if (!l)
1025 		goto noleaf;
1026 
1027 	/* retrieve child from parent node */
1028 	n = get_child(tp, get_index(key, tp));
1029 
1030 	/* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1031 	 *
1032 	 *  Add a new tnode here
1033 	 *  first tnode need some special handling
1034 	 *  leaves us in position for handling as case 3
1035 	 */
1036 	if (n) {
1037 		struct key_vector *tn;
1038 
1039 		tn = tnode_new(key, __fls(key ^ n->key), 1);
1040 		if (!tn)
1041 			goto notnode;
1042 
1043 		/* initialize routes out of node */
1044 		NODE_INIT_PARENT(tn, tp);
1045 		put_child(tn, get_index(key, tn) ^ 1, n);
1046 
1047 		/* start adding routes into the node */
1048 		put_child_root(tp, key, tn);
1049 		node_set_parent(n, tn);
1050 
1051 		/* parent now has a NULL spot where the leaf can go */
1052 		tp = tn;
1053 	}
1054 
1055 	/* Case 3: n is NULL, and will just insert a new leaf */
1056 	node_push_suffix(tp, new->fa_slen);
1057 	NODE_INIT_PARENT(l, tp);
1058 	put_child_root(tp, key, l);
1059 	trie_rebalance(t, tp);
1060 
1061 	return 0;
1062 notnode:
1063 	node_free(l);
1064 noleaf:
1065 	return -ENOMEM;
1066 }
1067 
1068 /* fib notifier for ADD is sent before calling fib_insert_alias with
1069  * the expectation that the only possible failure ENOMEM
1070  */
1071 static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1072 			    struct key_vector *l, struct fib_alias *new,
1073 			    struct fib_alias *fa, t_key key)
1074 {
1075 	if (!l)
1076 		return fib_insert_node(t, tp, new, key);
1077 
1078 	if (fa) {
1079 		hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1080 	} else {
1081 		struct fib_alias *last;
1082 
1083 		hlist_for_each_entry(last, &l->leaf, fa_list) {
1084 			if (new->fa_slen < last->fa_slen)
1085 				break;
1086 			if ((new->fa_slen == last->fa_slen) &&
1087 			    (new->tb_id > last->tb_id))
1088 				break;
1089 			fa = last;
1090 		}
1091 
1092 		if (fa)
1093 			hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1094 		else
1095 			hlist_add_head_rcu(&new->fa_list, &l->leaf);
1096 	}
1097 
1098 	/* if we added to the tail node then we need to update slen */
1099 	if (l->slen < new->fa_slen) {
1100 		l->slen = new->fa_slen;
1101 		node_push_suffix(tp, new->fa_slen);
1102 	}
1103 
1104 	return 0;
1105 }
1106 
1107 static bool fib_valid_key_len(u32 key, u8 plen, struct netlink_ext_ack *extack)
1108 {
1109 	if (plen > KEYLENGTH) {
1110 		NL_SET_ERR_MSG(extack, "Invalid prefix length");
1111 		return false;
1112 	}
1113 
1114 	if ((plen < KEYLENGTH) && (key << plen)) {
1115 		NL_SET_ERR_MSG(extack,
1116 			       "Invalid prefix for given prefix length");
1117 		return false;
1118 	}
1119 
1120 	return true;
1121 }
1122 
1123 /* Caller must hold RTNL. */
1124 int fib_table_insert(struct net *net, struct fib_table *tb,
1125 		     struct fib_config *cfg, struct netlink_ext_ack *extack)
1126 {
1127 	enum fib_event_type event = FIB_EVENT_ENTRY_ADD;
1128 	struct trie *t = (struct trie *)tb->tb_data;
1129 	struct fib_alias *fa, *new_fa;
1130 	struct key_vector *l, *tp;
1131 	u16 nlflags = NLM_F_EXCL;
1132 	struct fib_info *fi;
1133 	u8 plen = cfg->fc_dst_len;
1134 	u8 slen = KEYLENGTH - plen;
1135 	u8 tos = cfg->fc_tos;
1136 	u32 key;
1137 	int err;
1138 
1139 	key = ntohl(cfg->fc_dst);
1140 
1141 	if (!fib_valid_key_len(key, plen, extack))
1142 		return -EINVAL;
1143 
1144 	pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1145 
1146 	fi = fib_create_info(cfg, extack);
1147 	if (IS_ERR(fi)) {
1148 		err = PTR_ERR(fi);
1149 		goto err;
1150 	}
1151 
1152 	l = fib_find_node(t, &tp, key);
1153 	fa = l ? fib_find_alias(&l->leaf, slen, tos, fi->fib_priority,
1154 				tb->tb_id) : NULL;
1155 
1156 	/* Now fa, if non-NULL, points to the first fib alias
1157 	 * with the same keys [prefix,tos,priority], if such key already
1158 	 * exists or to the node before which we will insert new one.
1159 	 *
1160 	 * If fa is NULL, we will need to allocate a new one and
1161 	 * insert to the tail of the section matching the suffix length
1162 	 * of the new alias.
1163 	 */
1164 
1165 	if (fa && fa->fa_tos == tos &&
1166 	    fa->fa_info->fib_priority == fi->fib_priority) {
1167 		struct fib_alias *fa_first, *fa_match;
1168 
1169 		err = -EEXIST;
1170 		if (cfg->fc_nlflags & NLM_F_EXCL)
1171 			goto out;
1172 
1173 		nlflags &= ~NLM_F_EXCL;
1174 
1175 		/* We have 2 goals:
1176 		 * 1. Find exact match for type, scope, fib_info to avoid
1177 		 * duplicate routes
1178 		 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1179 		 */
1180 		fa_match = NULL;
1181 		fa_first = fa;
1182 		hlist_for_each_entry_from(fa, fa_list) {
1183 			if ((fa->fa_slen != slen) ||
1184 			    (fa->tb_id != tb->tb_id) ||
1185 			    (fa->fa_tos != tos))
1186 				break;
1187 			if (fa->fa_info->fib_priority != fi->fib_priority)
1188 				break;
1189 			if (fa->fa_type == cfg->fc_type &&
1190 			    fa->fa_info == fi) {
1191 				fa_match = fa;
1192 				break;
1193 			}
1194 		}
1195 
1196 		if (cfg->fc_nlflags & NLM_F_REPLACE) {
1197 			struct fib_info *fi_drop;
1198 			u8 state;
1199 
1200 			nlflags |= NLM_F_REPLACE;
1201 			fa = fa_first;
1202 			if (fa_match) {
1203 				if (fa == fa_match)
1204 					err = 0;
1205 				goto out;
1206 			}
1207 			err = -ENOBUFS;
1208 			new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1209 			if (!new_fa)
1210 				goto out;
1211 
1212 			fi_drop = fa->fa_info;
1213 			new_fa->fa_tos = fa->fa_tos;
1214 			new_fa->fa_info = fi;
1215 			new_fa->fa_type = cfg->fc_type;
1216 			state = fa->fa_state;
1217 			new_fa->fa_state = state & ~FA_S_ACCESSED;
1218 			new_fa->fa_slen = fa->fa_slen;
1219 			new_fa->tb_id = tb->tb_id;
1220 			new_fa->fa_default = -1;
1221 
1222 			err = call_fib_entry_notifiers(net,
1223 						       FIB_EVENT_ENTRY_REPLACE,
1224 						       key, plen, new_fa,
1225 						       extack);
1226 			if (err)
1227 				goto out_free_new_fa;
1228 
1229 			rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1230 				  tb->tb_id, &cfg->fc_nlinfo, nlflags);
1231 
1232 			hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1233 
1234 			alias_free_mem_rcu(fa);
1235 
1236 			fib_release_info(fi_drop);
1237 			if (state & FA_S_ACCESSED)
1238 				rt_cache_flush(cfg->fc_nlinfo.nl_net);
1239 
1240 			goto succeeded;
1241 		}
1242 		/* Error if we find a perfect match which
1243 		 * uses the same scope, type, and nexthop
1244 		 * information.
1245 		 */
1246 		if (fa_match)
1247 			goto out;
1248 
1249 		if (cfg->fc_nlflags & NLM_F_APPEND) {
1250 			event = FIB_EVENT_ENTRY_APPEND;
1251 			nlflags |= NLM_F_APPEND;
1252 		} else {
1253 			fa = fa_first;
1254 		}
1255 	}
1256 	err = -ENOENT;
1257 	if (!(cfg->fc_nlflags & NLM_F_CREATE))
1258 		goto out;
1259 
1260 	nlflags |= NLM_F_CREATE;
1261 	err = -ENOBUFS;
1262 	new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1263 	if (!new_fa)
1264 		goto out;
1265 
1266 	new_fa->fa_info = fi;
1267 	new_fa->fa_tos = tos;
1268 	new_fa->fa_type = cfg->fc_type;
1269 	new_fa->fa_state = 0;
1270 	new_fa->fa_slen = slen;
1271 	new_fa->tb_id = tb->tb_id;
1272 	new_fa->fa_default = -1;
1273 
1274 	err = call_fib_entry_notifiers(net, event, key, plen, new_fa, extack);
1275 	if (err)
1276 		goto out_free_new_fa;
1277 
1278 	/* Insert new entry to the list. */
1279 	err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1280 	if (err)
1281 		goto out_fib_notif;
1282 
1283 	if (!plen)
1284 		tb->tb_num_default++;
1285 
1286 	rt_cache_flush(cfg->fc_nlinfo.nl_net);
1287 	rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1288 		  &cfg->fc_nlinfo, nlflags);
1289 succeeded:
1290 	return 0;
1291 
1292 out_fib_notif:
1293 	/* notifier was sent that entry would be added to trie, but
1294 	 * the add failed and need to recover. Only failure for
1295 	 * fib_insert_alias is ENOMEM.
1296 	 */
1297 	NL_SET_ERR_MSG(extack, "Failed to insert route into trie");
1298 	call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, key,
1299 				 plen, new_fa, NULL);
1300 out_free_new_fa:
1301 	kmem_cache_free(fn_alias_kmem, new_fa);
1302 out:
1303 	fib_release_info(fi);
1304 err:
1305 	return err;
1306 }
1307 
1308 static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1309 {
1310 	t_key prefix = n->key;
1311 
1312 	return (key ^ prefix) & (prefix | -prefix);
1313 }
1314 
1315 /* should be called with rcu_read_lock */
1316 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1317 		     struct fib_result *res, int fib_flags)
1318 {
1319 	struct trie *t = (struct trie *) tb->tb_data;
1320 #ifdef CONFIG_IP_FIB_TRIE_STATS
1321 	struct trie_use_stats __percpu *stats = t->stats;
1322 #endif
1323 	const t_key key = ntohl(flp->daddr);
1324 	struct key_vector *n, *pn;
1325 	struct fib_alias *fa;
1326 	unsigned long index;
1327 	t_key cindex;
1328 
1329 	trace_fib_table_lookup(tb->tb_id, flp);
1330 
1331 	pn = t->kv;
1332 	cindex = 0;
1333 
1334 	n = get_child_rcu(pn, cindex);
1335 	if (!n)
1336 		return -EAGAIN;
1337 
1338 #ifdef CONFIG_IP_FIB_TRIE_STATS
1339 	this_cpu_inc(stats->gets);
1340 #endif
1341 
1342 	/* Step 1: Travel to the longest prefix match in the trie */
1343 	for (;;) {
1344 		index = get_cindex(key, n);
1345 
1346 		/* This bit of code is a bit tricky but it combines multiple
1347 		 * checks into a single check.  The prefix consists of the
1348 		 * prefix plus zeros for the "bits" in the prefix. The index
1349 		 * is the difference between the key and this value.  From
1350 		 * this we can actually derive several pieces of data.
1351 		 *   if (index >= (1ul << bits))
1352 		 *     we have a mismatch in skip bits and failed
1353 		 *   else
1354 		 *     we know the value is cindex
1355 		 *
1356 		 * This check is safe even if bits == KEYLENGTH due to the
1357 		 * fact that we can only allocate a node with 32 bits if a
1358 		 * long is greater than 32 bits.
1359 		 */
1360 		if (index >= (1ul << n->bits))
1361 			break;
1362 
1363 		/* we have found a leaf. Prefixes have already been compared */
1364 		if (IS_LEAF(n))
1365 			goto found;
1366 
1367 		/* only record pn and cindex if we are going to be chopping
1368 		 * bits later.  Otherwise we are just wasting cycles.
1369 		 */
1370 		if (n->slen > n->pos) {
1371 			pn = n;
1372 			cindex = index;
1373 		}
1374 
1375 		n = get_child_rcu(n, index);
1376 		if (unlikely(!n))
1377 			goto backtrace;
1378 	}
1379 
1380 	/* Step 2: Sort out leaves and begin backtracing for longest prefix */
1381 	for (;;) {
1382 		/* record the pointer where our next node pointer is stored */
1383 		struct key_vector __rcu **cptr = n->tnode;
1384 
1385 		/* This test verifies that none of the bits that differ
1386 		 * between the key and the prefix exist in the region of
1387 		 * the lsb and higher in the prefix.
1388 		 */
1389 		if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1390 			goto backtrace;
1391 
1392 		/* exit out and process leaf */
1393 		if (unlikely(IS_LEAF(n)))
1394 			break;
1395 
1396 		/* Don't bother recording parent info.  Since we are in
1397 		 * prefix match mode we will have to come back to wherever
1398 		 * we started this traversal anyway
1399 		 */
1400 
1401 		while ((n = rcu_dereference(*cptr)) == NULL) {
1402 backtrace:
1403 #ifdef CONFIG_IP_FIB_TRIE_STATS
1404 			if (!n)
1405 				this_cpu_inc(stats->null_node_hit);
1406 #endif
1407 			/* If we are at cindex 0 there are no more bits for
1408 			 * us to strip at this level so we must ascend back
1409 			 * up one level to see if there are any more bits to
1410 			 * be stripped there.
1411 			 */
1412 			while (!cindex) {
1413 				t_key pkey = pn->key;
1414 
1415 				/* If we don't have a parent then there is
1416 				 * nothing for us to do as we do not have any
1417 				 * further nodes to parse.
1418 				 */
1419 				if (IS_TRIE(pn))
1420 					return -EAGAIN;
1421 #ifdef CONFIG_IP_FIB_TRIE_STATS
1422 				this_cpu_inc(stats->backtrack);
1423 #endif
1424 				/* Get Child's index */
1425 				pn = node_parent_rcu(pn);
1426 				cindex = get_index(pkey, pn);
1427 			}
1428 
1429 			/* strip the least significant bit from the cindex */
1430 			cindex &= cindex - 1;
1431 
1432 			/* grab pointer for next child node */
1433 			cptr = &pn->tnode[cindex];
1434 		}
1435 	}
1436 
1437 found:
1438 	/* this line carries forward the xor from earlier in the function */
1439 	index = key ^ n->key;
1440 
1441 	/* Step 3: Process the leaf, if that fails fall back to backtracing */
1442 	hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1443 		struct fib_info *fi = fa->fa_info;
1444 		int nhsel, err;
1445 
1446 		if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1447 			if (index >= (1ul << fa->fa_slen))
1448 				continue;
1449 		}
1450 		if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1451 			continue;
1452 		if (fi->fib_dead)
1453 			continue;
1454 		if (fa->fa_info->fib_scope < flp->flowi4_scope)
1455 			continue;
1456 		fib_alias_accessed(fa);
1457 		err = fib_props[fa->fa_type].error;
1458 		if (unlikely(err < 0)) {
1459 #ifdef CONFIG_IP_FIB_TRIE_STATS
1460 			this_cpu_inc(stats->semantic_match_passed);
1461 #endif
1462 			return err;
1463 		}
1464 		if (fi->fib_flags & RTNH_F_DEAD)
1465 			continue;
1466 		for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
1467 			const struct fib_nh *nh = &fi->fib_nh[nhsel];
1468 			struct in_device *in_dev = __in_dev_get_rcu(nh->nh_dev);
1469 
1470 			if (nh->nh_flags & RTNH_F_DEAD)
1471 				continue;
1472 			if (in_dev &&
1473 			    IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1474 			    nh->nh_flags & RTNH_F_LINKDOWN &&
1475 			    !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1476 				continue;
1477 			if (!(flp->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF)) {
1478 				if (flp->flowi4_oif &&
1479 				    flp->flowi4_oif != nh->nh_oif)
1480 					continue;
1481 			}
1482 
1483 			if (!(fib_flags & FIB_LOOKUP_NOREF))
1484 				refcount_inc(&fi->fib_clntref);
1485 
1486 			res->prefix = htonl(n->key);
1487 			res->prefixlen = KEYLENGTH - fa->fa_slen;
1488 			res->nh_sel = nhsel;
1489 			res->type = fa->fa_type;
1490 			res->scope = fi->fib_scope;
1491 			res->fi = fi;
1492 			res->table = tb;
1493 			res->fa_head = &n->leaf;
1494 #ifdef CONFIG_IP_FIB_TRIE_STATS
1495 			this_cpu_inc(stats->semantic_match_passed);
1496 #endif
1497 			trace_fib_table_lookup_nh(nh);
1498 
1499 			return err;
1500 		}
1501 	}
1502 #ifdef CONFIG_IP_FIB_TRIE_STATS
1503 	this_cpu_inc(stats->semantic_match_miss);
1504 #endif
1505 	goto backtrace;
1506 }
1507 EXPORT_SYMBOL_GPL(fib_table_lookup);
1508 
1509 static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1510 			     struct key_vector *l, struct fib_alias *old)
1511 {
1512 	/* record the location of the previous list_info entry */
1513 	struct hlist_node **pprev = old->fa_list.pprev;
1514 	struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1515 
1516 	/* remove the fib_alias from the list */
1517 	hlist_del_rcu(&old->fa_list);
1518 
1519 	/* if we emptied the list this leaf will be freed and we can sort
1520 	 * out parent suffix lengths as a part of trie_rebalance
1521 	 */
1522 	if (hlist_empty(&l->leaf)) {
1523 		if (tp->slen == l->slen)
1524 			node_pull_suffix(tp, tp->pos);
1525 		put_child_root(tp, l->key, NULL);
1526 		node_free(l);
1527 		trie_rebalance(t, tp);
1528 		return;
1529 	}
1530 
1531 	/* only access fa if it is pointing at the last valid hlist_node */
1532 	if (*pprev)
1533 		return;
1534 
1535 	/* update the trie with the latest suffix length */
1536 	l->slen = fa->fa_slen;
1537 	node_pull_suffix(tp, fa->fa_slen);
1538 }
1539 
1540 /* Caller must hold RTNL. */
1541 int fib_table_delete(struct net *net, struct fib_table *tb,
1542 		     struct fib_config *cfg, struct netlink_ext_ack *extack)
1543 {
1544 	struct trie *t = (struct trie *) tb->tb_data;
1545 	struct fib_alias *fa, *fa_to_delete;
1546 	struct key_vector *l, *tp;
1547 	u8 plen = cfg->fc_dst_len;
1548 	u8 slen = KEYLENGTH - plen;
1549 	u8 tos = cfg->fc_tos;
1550 	u32 key;
1551 
1552 	key = ntohl(cfg->fc_dst);
1553 
1554 	if (!fib_valid_key_len(key, plen, extack))
1555 		return -EINVAL;
1556 
1557 	l = fib_find_node(t, &tp, key);
1558 	if (!l)
1559 		return -ESRCH;
1560 
1561 	fa = fib_find_alias(&l->leaf, slen, tos, 0, tb->tb_id);
1562 	if (!fa)
1563 		return -ESRCH;
1564 
1565 	pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1566 
1567 	fa_to_delete = NULL;
1568 	hlist_for_each_entry_from(fa, fa_list) {
1569 		struct fib_info *fi = fa->fa_info;
1570 
1571 		if ((fa->fa_slen != slen) ||
1572 		    (fa->tb_id != tb->tb_id) ||
1573 		    (fa->fa_tos != tos))
1574 			break;
1575 
1576 		if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1577 		    (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1578 		     fa->fa_info->fib_scope == cfg->fc_scope) &&
1579 		    (!cfg->fc_prefsrc ||
1580 		     fi->fib_prefsrc == cfg->fc_prefsrc) &&
1581 		    (!cfg->fc_protocol ||
1582 		     fi->fib_protocol == cfg->fc_protocol) &&
1583 		    fib_nh_match(cfg, fi, extack) == 0 &&
1584 		    fib_metrics_match(cfg, fi)) {
1585 			fa_to_delete = fa;
1586 			break;
1587 		}
1588 	}
1589 
1590 	if (!fa_to_delete)
1591 		return -ESRCH;
1592 
1593 	call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, key, plen,
1594 				 fa_to_delete, extack);
1595 	rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1596 		  &cfg->fc_nlinfo, 0);
1597 
1598 	if (!plen)
1599 		tb->tb_num_default--;
1600 
1601 	fib_remove_alias(t, tp, l, fa_to_delete);
1602 
1603 	if (fa_to_delete->fa_state & FA_S_ACCESSED)
1604 		rt_cache_flush(cfg->fc_nlinfo.nl_net);
1605 
1606 	fib_release_info(fa_to_delete->fa_info);
1607 	alias_free_mem_rcu(fa_to_delete);
1608 	return 0;
1609 }
1610 
1611 /* Scan for the next leaf starting at the provided key value */
1612 static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1613 {
1614 	struct key_vector *pn, *n = *tn;
1615 	unsigned long cindex;
1616 
1617 	/* this loop is meant to try and find the key in the trie */
1618 	do {
1619 		/* record parent and next child index */
1620 		pn = n;
1621 		cindex = (key > pn->key) ? get_index(key, pn) : 0;
1622 
1623 		if (cindex >> pn->bits)
1624 			break;
1625 
1626 		/* descend into the next child */
1627 		n = get_child_rcu(pn, cindex++);
1628 		if (!n)
1629 			break;
1630 
1631 		/* guarantee forward progress on the keys */
1632 		if (IS_LEAF(n) && (n->key >= key))
1633 			goto found;
1634 	} while (IS_TNODE(n));
1635 
1636 	/* this loop will search for the next leaf with a greater key */
1637 	while (!IS_TRIE(pn)) {
1638 		/* if we exhausted the parent node we will need to climb */
1639 		if (cindex >= (1ul << pn->bits)) {
1640 			t_key pkey = pn->key;
1641 
1642 			pn = node_parent_rcu(pn);
1643 			cindex = get_index(pkey, pn) + 1;
1644 			continue;
1645 		}
1646 
1647 		/* grab the next available node */
1648 		n = get_child_rcu(pn, cindex++);
1649 		if (!n)
1650 			continue;
1651 
1652 		/* no need to compare keys since we bumped the index */
1653 		if (IS_LEAF(n))
1654 			goto found;
1655 
1656 		/* Rescan start scanning in new node */
1657 		pn = n;
1658 		cindex = 0;
1659 	}
1660 
1661 	*tn = pn;
1662 	return NULL; /* Root of trie */
1663 found:
1664 	/* if we are at the limit for keys just return NULL for the tnode */
1665 	*tn = pn;
1666 	return n;
1667 }
1668 
1669 static void fib_trie_free(struct fib_table *tb)
1670 {
1671 	struct trie *t = (struct trie *)tb->tb_data;
1672 	struct key_vector *pn = t->kv;
1673 	unsigned long cindex = 1;
1674 	struct hlist_node *tmp;
1675 	struct fib_alias *fa;
1676 
1677 	/* walk trie in reverse order and free everything */
1678 	for (;;) {
1679 		struct key_vector *n;
1680 
1681 		if (!(cindex--)) {
1682 			t_key pkey = pn->key;
1683 
1684 			if (IS_TRIE(pn))
1685 				break;
1686 
1687 			n = pn;
1688 			pn = node_parent(pn);
1689 
1690 			/* drop emptied tnode */
1691 			put_child_root(pn, n->key, NULL);
1692 			node_free(n);
1693 
1694 			cindex = get_index(pkey, pn);
1695 
1696 			continue;
1697 		}
1698 
1699 		/* grab the next available node */
1700 		n = get_child(pn, cindex);
1701 		if (!n)
1702 			continue;
1703 
1704 		if (IS_TNODE(n)) {
1705 			/* record pn and cindex for leaf walking */
1706 			pn = n;
1707 			cindex = 1ul << n->bits;
1708 
1709 			continue;
1710 		}
1711 
1712 		hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1713 			hlist_del_rcu(&fa->fa_list);
1714 			alias_free_mem_rcu(fa);
1715 		}
1716 
1717 		put_child_root(pn, n->key, NULL);
1718 		node_free(n);
1719 	}
1720 
1721 #ifdef CONFIG_IP_FIB_TRIE_STATS
1722 	free_percpu(t->stats);
1723 #endif
1724 	kfree(tb);
1725 }
1726 
1727 struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1728 {
1729 	struct trie *ot = (struct trie *)oldtb->tb_data;
1730 	struct key_vector *l, *tp = ot->kv;
1731 	struct fib_table *local_tb;
1732 	struct fib_alias *fa;
1733 	struct trie *lt;
1734 	t_key key = 0;
1735 
1736 	if (oldtb->tb_data == oldtb->__data)
1737 		return oldtb;
1738 
1739 	local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1740 	if (!local_tb)
1741 		return NULL;
1742 
1743 	lt = (struct trie *)local_tb->tb_data;
1744 
1745 	while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1746 		struct key_vector *local_l = NULL, *local_tp;
1747 
1748 		hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1749 			struct fib_alias *new_fa;
1750 
1751 			if (local_tb->tb_id != fa->tb_id)
1752 				continue;
1753 
1754 			/* clone fa for new local table */
1755 			new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1756 			if (!new_fa)
1757 				goto out;
1758 
1759 			memcpy(new_fa, fa, sizeof(*fa));
1760 
1761 			/* insert clone into table */
1762 			if (!local_l)
1763 				local_l = fib_find_node(lt, &local_tp, l->key);
1764 
1765 			if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1766 					     NULL, l->key)) {
1767 				kmem_cache_free(fn_alias_kmem, new_fa);
1768 				goto out;
1769 			}
1770 		}
1771 
1772 		/* stop loop if key wrapped back to 0 */
1773 		key = l->key + 1;
1774 		if (key < l->key)
1775 			break;
1776 	}
1777 
1778 	return local_tb;
1779 out:
1780 	fib_trie_free(local_tb);
1781 
1782 	return NULL;
1783 }
1784 
1785 /* Caller must hold RTNL */
1786 void fib_table_flush_external(struct fib_table *tb)
1787 {
1788 	struct trie *t = (struct trie *)tb->tb_data;
1789 	struct key_vector *pn = t->kv;
1790 	unsigned long cindex = 1;
1791 	struct hlist_node *tmp;
1792 	struct fib_alias *fa;
1793 
1794 	/* walk trie in reverse order */
1795 	for (;;) {
1796 		unsigned char slen = 0;
1797 		struct key_vector *n;
1798 
1799 		if (!(cindex--)) {
1800 			t_key pkey = pn->key;
1801 
1802 			/* cannot resize the trie vector */
1803 			if (IS_TRIE(pn))
1804 				break;
1805 
1806 			/* update the suffix to address pulled leaves */
1807 			if (pn->slen > pn->pos)
1808 				update_suffix(pn);
1809 
1810 			/* resize completed node */
1811 			pn = resize(t, pn);
1812 			cindex = get_index(pkey, pn);
1813 
1814 			continue;
1815 		}
1816 
1817 		/* grab the next available node */
1818 		n = get_child(pn, cindex);
1819 		if (!n)
1820 			continue;
1821 
1822 		if (IS_TNODE(n)) {
1823 			/* record pn and cindex for leaf walking */
1824 			pn = n;
1825 			cindex = 1ul << n->bits;
1826 
1827 			continue;
1828 		}
1829 
1830 		hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1831 			/* if alias was cloned to local then we just
1832 			 * need to remove the local copy from main
1833 			 */
1834 			if (tb->tb_id != fa->tb_id) {
1835 				hlist_del_rcu(&fa->fa_list);
1836 				alias_free_mem_rcu(fa);
1837 				continue;
1838 			}
1839 
1840 			/* record local slen */
1841 			slen = fa->fa_slen;
1842 		}
1843 
1844 		/* update leaf slen */
1845 		n->slen = slen;
1846 
1847 		if (hlist_empty(&n->leaf)) {
1848 			put_child_root(pn, n->key, NULL);
1849 			node_free(n);
1850 		}
1851 	}
1852 }
1853 
1854 /* Caller must hold RTNL. */
1855 int fib_table_flush(struct net *net, struct fib_table *tb)
1856 {
1857 	struct trie *t = (struct trie *)tb->tb_data;
1858 	struct key_vector *pn = t->kv;
1859 	unsigned long cindex = 1;
1860 	struct hlist_node *tmp;
1861 	struct fib_alias *fa;
1862 	int found = 0;
1863 
1864 	/* walk trie in reverse order */
1865 	for (;;) {
1866 		unsigned char slen = 0;
1867 		struct key_vector *n;
1868 
1869 		if (!(cindex--)) {
1870 			t_key pkey = pn->key;
1871 
1872 			/* cannot resize the trie vector */
1873 			if (IS_TRIE(pn))
1874 				break;
1875 
1876 			/* update the suffix to address pulled leaves */
1877 			if (pn->slen > pn->pos)
1878 				update_suffix(pn);
1879 
1880 			/* resize completed node */
1881 			pn = resize(t, pn);
1882 			cindex = get_index(pkey, pn);
1883 
1884 			continue;
1885 		}
1886 
1887 		/* grab the next available node */
1888 		n = get_child(pn, cindex);
1889 		if (!n)
1890 			continue;
1891 
1892 		if (IS_TNODE(n)) {
1893 			/* record pn and cindex for leaf walking */
1894 			pn = n;
1895 			cindex = 1ul << n->bits;
1896 
1897 			continue;
1898 		}
1899 
1900 		hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1901 			struct fib_info *fi = fa->fa_info;
1902 
1903 			if (!fi || !(fi->fib_flags & RTNH_F_DEAD) ||
1904 			    tb->tb_id != fa->tb_id) {
1905 				slen = fa->fa_slen;
1906 				continue;
1907 			}
1908 
1909 			call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL,
1910 						 n->key,
1911 						 KEYLENGTH - fa->fa_slen, fa,
1912 						 NULL);
1913 			hlist_del_rcu(&fa->fa_list);
1914 			fib_release_info(fa->fa_info);
1915 			alias_free_mem_rcu(fa);
1916 			found++;
1917 		}
1918 
1919 		/* update leaf slen */
1920 		n->slen = slen;
1921 
1922 		if (hlist_empty(&n->leaf)) {
1923 			put_child_root(pn, n->key, NULL);
1924 			node_free(n);
1925 		}
1926 	}
1927 
1928 	pr_debug("trie_flush found=%d\n", found);
1929 	return found;
1930 }
1931 
1932 static void fib_leaf_notify(struct net *net, struct key_vector *l,
1933 			    struct fib_table *tb, struct notifier_block *nb)
1934 {
1935 	struct fib_alias *fa;
1936 
1937 	hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1938 		struct fib_info *fi = fa->fa_info;
1939 
1940 		if (!fi)
1941 			continue;
1942 
1943 		/* local and main table can share the same trie,
1944 		 * so don't notify twice for the same entry.
1945 		 */
1946 		if (tb->tb_id != fa->tb_id)
1947 			continue;
1948 
1949 		call_fib_entry_notifier(nb, net, FIB_EVENT_ENTRY_ADD, l->key,
1950 					KEYLENGTH - fa->fa_slen, fa);
1951 	}
1952 }
1953 
1954 static void fib_table_notify(struct net *net, struct fib_table *tb,
1955 			     struct notifier_block *nb)
1956 {
1957 	struct trie *t = (struct trie *)tb->tb_data;
1958 	struct key_vector *l, *tp = t->kv;
1959 	t_key key = 0;
1960 
1961 	while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1962 		fib_leaf_notify(net, l, tb, nb);
1963 
1964 		key = l->key + 1;
1965 		/* stop in case of wrap around */
1966 		if (key < l->key)
1967 			break;
1968 	}
1969 }
1970 
1971 void fib_notify(struct net *net, struct notifier_block *nb)
1972 {
1973 	unsigned int h;
1974 
1975 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
1976 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
1977 		struct fib_table *tb;
1978 
1979 		hlist_for_each_entry_rcu(tb, head, tb_hlist)
1980 			fib_table_notify(net, tb, nb);
1981 	}
1982 }
1983 
1984 static void __trie_free_rcu(struct rcu_head *head)
1985 {
1986 	struct fib_table *tb = container_of(head, struct fib_table, rcu);
1987 #ifdef CONFIG_IP_FIB_TRIE_STATS
1988 	struct trie *t = (struct trie *)tb->tb_data;
1989 
1990 	if (tb->tb_data == tb->__data)
1991 		free_percpu(t->stats);
1992 #endif /* CONFIG_IP_FIB_TRIE_STATS */
1993 	kfree(tb);
1994 }
1995 
1996 void fib_free_table(struct fib_table *tb)
1997 {
1998 	call_rcu(&tb->rcu, __trie_free_rcu);
1999 }
2000 
2001 static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
2002 			     struct sk_buff *skb, struct netlink_callback *cb)
2003 {
2004 	__be32 xkey = htonl(l->key);
2005 	struct fib_alias *fa;
2006 	int i, s_i;
2007 
2008 	s_i = cb->args[4];
2009 	i = 0;
2010 
2011 	/* rcu_read_lock is hold by caller */
2012 	hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2013 		int err;
2014 
2015 		if (i < s_i) {
2016 			i++;
2017 			continue;
2018 		}
2019 
2020 		if (tb->tb_id != fa->tb_id) {
2021 			i++;
2022 			continue;
2023 		}
2024 
2025 		err = fib_dump_info(skb, NETLINK_CB(cb->skb).portid,
2026 				    cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2027 				    tb->tb_id, fa->fa_type,
2028 				    xkey, KEYLENGTH - fa->fa_slen,
2029 				    fa->fa_tos, fa->fa_info, NLM_F_MULTI);
2030 		if (err < 0) {
2031 			cb->args[4] = i;
2032 			return err;
2033 		}
2034 		i++;
2035 	}
2036 
2037 	cb->args[4] = i;
2038 	return skb->len;
2039 }
2040 
2041 /* rcu_read_lock needs to be hold by caller from readside */
2042 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
2043 		   struct netlink_callback *cb)
2044 {
2045 	struct trie *t = (struct trie *)tb->tb_data;
2046 	struct key_vector *l, *tp = t->kv;
2047 	/* Dump starting at last key.
2048 	 * Note: 0.0.0.0/0 (ie default) is first key.
2049 	 */
2050 	int count = cb->args[2];
2051 	t_key key = cb->args[3];
2052 
2053 	while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2054 		int err;
2055 
2056 		err = fn_trie_dump_leaf(l, tb, skb, cb);
2057 		if (err < 0) {
2058 			cb->args[3] = key;
2059 			cb->args[2] = count;
2060 			return err;
2061 		}
2062 
2063 		++count;
2064 		key = l->key + 1;
2065 
2066 		memset(&cb->args[4], 0,
2067 		       sizeof(cb->args) - 4*sizeof(cb->args[0]));
2068 
2069 		/* stop loop if key wrapped back to 0 */
2070 		if (key < l->key)
2071 			break;
2072 	}
2073 
2074 	cb->args[3] = key;
2075 	cb->args[2] = count;
2076 
2077 	return skb->len;
2078 }
2079 
2080 void __init fib_trie_init(void)
2081 {
2082 	fn_alias_kmem = kmem_cache_create("ip_fib_alias",
2083 					  sizeof(struct fib_alias),
2084 					  0, SLAB_PANIC, NULL);
2085 
2086 	trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
2087 					   LEAF_SIZE,
2088 					   0, SLAB_PANIC, NULL);
2089 }
2090 
2091 struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
2092 {
2093 	struct fib_table *tb;
2094 	struct trie *t;
2095 	size_t sz = sizeof(*tb);
2096 
2097 	if (!alias)
2098 		sz += sizeof(struct trie);
2099 
2100 	tb = kzalloc(sz, GFP_KERNEL);
2101 	if (!tb)
2102 		return NULL;
2103 
2104 	tb->tb_id = id;
2105 	tb->tb_num_default = 0;
2106 	tb->tb_data = (alias ? alias->__data : tb->__data);
2107 
2108 	if (alias)
2109 		return tb;
2110 
2111 	t = (struct trie *) tb->tb_data;
2112 	t->kv[0].pos = KEYLENGTH;
2113 	t->kv[0].slen = KEYLENGTH;
2114 #ifdef CONFIG_IP_FIB_TRIE_STATS
2115 	t->stats = alloc_percpu(struct trie_use_stats);
2116 	if (!t->stats) {
2117 		kfree(tb);
2118 		tb = NULL;
2119 	}
2120 #endif
2121 
2122 	return tb;
2123 }
2124 
2125 #ifdef CONFIG_PROC_FS
2126 /* Depth first Trie walk iterator */
2127 struct fib_trie_iter {
2128 	struct seq_net_private p;
2129 	struct fib_table *tb;
2130 	struct key_vector *tnode;
2131 	unsigned int index;
2132 	unsigned int depth;
2133 };
2134 
2135 static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2136 {
2137 	unsigned long cindex = iter->index;
2138 	struct key_vector *pn = iter->tnode;
2139 	t_key pkey;
2140 
2141 	pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2142 		 iter->tnode, iter->index, iter->depth);
2143 
2144 	while (!IS_TRIE(pn)) {
2145 		while (cindex < child_length(pn)) {
2146 			struct key_vector *n = get_child_rcu(pn, cindex++);
2147 
2148 			if (!n)
2149 				continue;
2150 
2151 			if (IS_LEAF(n)) {
2152 				iter->tnode = pn;
2153 				iter->index = cindex;
2154 			} else {
2155 				/* push down one level */
2156 				iter->tnode = n;
2157 				iter->index = 0;
2158 				++iter->depth;
2159 			}
2160 
2161 			return n;
2162 		}
2163 
2164 		/* Current node exhausted, pop back up */
2165 		pkey = pn->key;
2166 		pn = node_parent_rcu(pn);
2167 		cindex = get_index(pkey, pn) + 1;
2168 		--iter->depth;
2169 	}
2170 
2171 	/* record root node so further searches know we are done */
2172 	iter->tnode = pn;
2173 	iter->index = 0;
2174 
2175 	return NULL;
2176 }
2177 
2178 static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2179 					     struct trie *t)
2180 {
2181 	struct key_vector *n, *pn;
2182 
2183 	if (!t)
2184 		return NULL;
2185 
2186 	pn = t->kv;
2187 	n = rcu_dereference(pn->tnode[0]);
2188 	if (!n)
2189 		return NULL;
2190 
2191 	if (IS_TNODE(n)) {
2192 		iter->tnode = n;
2193 		iter->index = 0;
2194 		iter->depth = 1;
2195 	} else {
2196 		iter->tnode = pn;
2197 		iter->index = 0;
2198 		iter->depth = 0;
2199 	}
2200 
2201 	return n;
2202 }
2203 
2204 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2205 {
2206 	struct key_vector *n;
2207 	struct fib_trie_iter iter;
2208 
2209 	memset(s, 0, sizeof(*s));
2210 
2211 	rcu_read_lock();
2212 	for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2213 		if (IS_LEAF(n)) {
2214 			struct fib_alias *fa;
2215 
2216 			s->leaves++;
2217 			s->totdepth += iter.depth;
2218 			if (iter.depth > s->maxdepth)
2219 				s->maxdepth = iter.depth;
2220 
2221 			hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2222 				++s->prefixes;
2223 		} else {
2224 			s->tnodes++;
2225 			if (n->bits < MAX_STAT_DEPTH)
2226 				s->nodesizes[n->bits]++;
2227 			s->nullpointers += tn_info(n)->empty_children;
2228 		}
2229 	}
2230 	rcu_read_unlock();
2231 }
2232 
2233 /*
2234  *	This outputs /proc/net/fib_triestats
2235  */
2236 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2237 {
2238 	unsigned int i, max, pointers, bytes, avdepth;
2239 
2240 	if (stat->leaves)
2241 		avdepth = stat->totdepth*100 / stat->leaves;
2242 	else
2243 		avdepth = 0;
2244 
2245 	seq_printf(seq, "\tAver depth:     %u.%02d\n",
2246 		   avdepth / 100, avdepth % 100);
2247 	seq_printf(seq, "\tMax depth:      %u\n", stat->maxdepth);
2248 
2249 	seq_printf(seq, "\tLeaves:         %u\n", stat->leaves);
2250 	bytes = LEAF_SIZE * stat->leaves;
2251 
2252 	seq_printf(seq, "\tPrefixes:       %u\n", stat->prefixes);
2253 	bytes += sizeof(struct fib_alias) * stat->prefixes;
2254 
2255 	seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2256 	bytes += TNODE_SIZE(0) * stat->tnodes;
2257 
2258 	max = MAX_STAT_DEPTH;
2259 	while (max > 0 && stat->nodesizes[max-1] == 0)
2260 		max--;
2261 
2262 	pointers = 0;
2263 	for (i = 1; i < max; i++)
2264 		if (stat->nodesizes[i] != 0) {
2265 			seq_printf(seq, "  %u: %u",  i, stat->nodesizes[i]);
2266 			pointers += (1<<i) * stat->nodesizes[i];
2267 		}
2268 	seq_putc(seq, '\n');
2269 	seq_printf(seq, "\tPointers: %u\n", pointers);
2270 
2271 	bytes += sizeof(struct key_vector *) * pointers;
2272 	seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2273 	seq_printf(seq, "Total size: %u  kB\n", (bytes + 1023) / 1024);
2274 }
2275 
2276 #ifdef CONFIG_IP_FIB_TRIE_STATS
2277 static void trie_show_usage(struct seq_file *seq,
2278 			    const struct trie_use_stats __percpu *stats)
2279 {
2280 	struct trie_use_stats s = { 0 };
2281 	int cpu;
2282 
2283 	/* loop through all of the CPUs and gather up the stats */
2284 	for_each_possible_cpu(cpu) {
2285 		const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2286 
2287 		s.gets += pcpu->gets;
2288 		s.backtrack += pcpu->backtrack;
2289 		s.semantic_match_passed += pcpu->semantic_match_passed;
2290 		s.semantic_match_miss += pcpu->semantic_match_miss;
2291 		s.null_node_hit += pcpu->null_node_hit;
2292 		s.resize_node_skipped += pcpu->resize_node_skipped;
2293 	}
2294 
2295 	seq_printf(seq, "\nCounters:\n---------\n");
2296 	seq_printf(seq, "gets = %u\n", s.gets);
2297 	seq_printf(seq, "backtracks = %u\n", s.backtrack);
2298 	seq_printf(seq, "semantic match passed = %u\n",
2299 		   s.semantic_match_passed);
2300 	seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2301 	seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2302 	seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2303 }
2304 #endif /*  CONFIG_IP_FIB_TRIE_STATS */
2305 
2306 static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2307 {
2308 	if (tb->tb_id == RT_TABLE_LOCAL)
2309 		seq_puts(seq, "Local:\n");
2310 	else if (tb->tb_id == RT_TABLE_MAIN)
2311 		seq_puts(seq, "Main:\n");
2312 	else
2313 		seq_printf(seq, "Id %d:\n", tb->tb_id);
2314 }
2315 
2316 
2317 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2318 {
2319 	struct net *net = (struct net *)seq->private;
2320 	unsigned int h;
2321 
2322 	seq_printf(seq,
2323 		   "Basic info: size of leaf:"
2324 		   " %zd bytes, size of tnode: %zd bytes.\n",
2325 		   LEAF_SIZE, TNODE_SIZE(0));
2326 
2327 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2328 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2329 		struct fib_table *tb;
2330 
2331 		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2332 			struct trie *t = (struct trie *) tb->tb_data;
2333 			struct trie_stat stat;
2334 
2335 			if (!t)
2336 				continue;
2337 
2338 			fib_table_print(seq, tb);
2339 
2340 			trie_collect_stats(t, &stat);
2341 			trie_show_stats(seq, &stat);
2342 #ifdef CONFIG_IP_FIB_TRIE_STATS
2343 			trie_show_usage(seq, t->stats);
2344 #endif
2345 		}
2346 	}
2347 
2348 	return 0;
2349 }
2350 
2351 static int fib_triestat_seq_open(struct inode *inode, struct file *file)
2352 {
2353 	return single_open_net(inode, file, fib_triestat_seq_show);
2354 }
2355 
2356 static const struct file_operations fib_triestat_fops = {
2357 	.open	= fib_triestat_seq_open,
2358 	.read	= seq_read,
2359 	.llseek	= seq_lseek,
2360 	.release = single_release_net,
2361 };
2362 
2363 static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2364 {
2365 	struct fib_trie_iter *iter = seq->private;
2366 	struct net *net = seq_file_net(seq);
2367 	loff_t idx = 0;
2368 	unsigned int h;
2369 
2370 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2371 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2372 		struct fib_table *tb;
2373 
2374 		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2375 			struct key_vector *n;
2376 
2377 			for (n = fib_trie_get_first(iter,
2378 						    (struct trie *) tb->tb_data);
2379 			     n; n = fib_trie_get_next(iter))
2380 				if (pos == idx++) {
2381 					iter->tb = tb;
2382 					return n;
2383 				}
2384 		}
2385 	}
2386 
2387 	return NULL;
2388 }
2389 
2390 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2391 	__acquires(RCU)
2392 {
2393 	rcu_read_lock();
2394 	return fib_trie_get_idx(seq, *pos);
2395 }
2396 
2397 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2398 {
2399 	struct fib_trie_iter *iter = seq->private;
2400 	struct net *net = seq_file_net(seq);
2401 	struct fib_table *tb = iter->tb;
2402 	struct hlist_node *tb_node;
2403 	unsigned int h;
2404 	struct key_vector *n;
2405 
2406 	++*pos;
2407 	/* next node in same table */
2408 	n = fib_trie_get_next(iter);
2409 	if (n)
2410 		return n;
2411 
2412 	/* walk rest of this hash chain */
2413 	h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2414 	while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2415 		tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2416 		n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2417 		if (n)
2418 			goto found;
2419 	}
2420 
2421 	/* new hash chain */
2422 	while (++h < FIB_TABLE_HASHSZ) {
2423 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2424 		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2425 			n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2426 			if (n)
2427 				goto found;
2428 		}
2429 	}
2430 	return NULL;
2431 
2432 found:
2433 	iter->tb = tb;
2434 	return n;
2435 }
2436 
2437 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2438 	__releases(RCU)
2439 {
2440 	rcu_read_unlock();
2441 }
2442 
2443 static void seq_indent(struct seq_file *seq, int n)
2444 {
2445 	while (n-- > 0)
2446 		seq_puts(seq, "   ");
2447 }
2448 
2449 static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2450 {
2451 	switch (s) {
2452 	case RT_SCOPE_UNIVERSE: return "universe";
2453 	case RT_SCOPE_SITE:	return "site";
2454 	case RT_SCOPE_LINK:	return "link";
2455 	case RT_SCOPE_HOST:	return "host";
2456 	case RT_SCOPE_NOWHERE:	return "nowhere";
2457 	default:
2458 		snprintf(buf, len, "scope=%d", s);
2459 		return buf;
2460 	}
2461 }
2462 
2463 static const char *const rtn_type_names[__RTN_MAX] = {
2464 	[RTN_UNSPEC] = "UNSPEC",
2465 	[RTN_UNICAST] = "UNICAST",
2466 	[RTN_LOCAL] = "LOCAL",
2467 	[RTN_BROADCAST] = "BROADCAST",
2468 	[RTN_ANYCAST] = "ANYCAST",
2469 	[RTN_MULTICAST] = "MULTICAST",
2470 	[RTN_BLACKHOLE] = "BLACKHOLE",
2471 	[RTN_UNREACHABLE] = "UNREACHABLE",
2472 	[RTN_PROHIBIT] = "PROHIBIT",
2473 	[RTN_THROW] = "THROW",
2474 	[RTN_NAT] = "NAT",
2475 	[RTN_XRESOLVE] = "XRESOLVE",
2476 };
2477 
2478 static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2479 {
2480 	if (t < __RTN_MAX && rtn_type_names[t])
2481 		return rtn_type_names[t];
2482 	snprintf(buf, len, "type %u", t);
2483 	return buf;
2484 }
2485 
2486 /* Pretty print the trie */
2487 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2488 {
2489 	const struct fib_trie_iter *iter = seq->private;
2490 	struct key_vector *n = v;
2491 
2492 	if (IS_TRIE(node_parent_rcu(n)))
2493 		fib_table_print(seq, iter->tb);
2494 
2495 	if (IS_TNODE(n)) {
2496 		__be32 prf = htonl(n->key);
2497 
2498 		seq_indent(seq, iter->depth-1);
2499 		seq_printf(seq, "  +-- %pI4/%zu %u %u %u\n",
2500 			   &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2501 			   tn_info(n)->full_children,
2502 			   tn_info(n)->empty_children);
2503 	} else {
2504 		__be32 val = htonl(n->key);
2505 		struct fib_alias *fa;
2506 
2507 		seq_indent(seq, iter->depth);
2508 		seq_printf(seq, "  |-- %pI4\n", &val);
2509 
2510 		hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2511 			char buf1[32], buf2[32];
2512 
2513 			seq_indent(seq, iter->depth + 1);
2514 			seq_printf(seq, "  /%zu %s %s",
2515 				   KEYLENGTH - fa->fa_slen,
2516 				   rtn_scope(buf1, sizeof(buf1),
2517 					     fa->fa_info->fib_scope),
2518 				   rtn_type(buf2, sizeof(buf2),
2519 					    fa->fa_type));
2520 			if (fa->fa_tos)
2521 				seq_printf(seq, " tos=%d", fa->fa_tos);
2522 			seq_putc(seq, '\n');
2523 		}
2524 	}
2525 
2526 	return 0;
2527 }
2528 
2529 static const struct seq_operations fib_trie_seq_ops = {
2530 	.start  = fib_trie_seq_start,
2531 	.next   = fib_trie_seq_next,
2532 	.stop   = fib_trie_seq_stop,
2533 	.show   = fib_trie_seq_show,
2534 };
2535 
2536 static int fib_trie_seq_open(struct inode *inode, struct file *file)
2537 {
2538 	return seq_open_net(inode, file, &fib_trie_seq_ops,
2539 			    sizeof(struct fib_trie_iter));
2540 }
2541 
2542 static const struct file_operations fib_trie_fops = {
2543 	.open   = fib_trie_seq_open,
2544 	.read   = seq_read,
2545 	.llseek = seq_lseek,
2546 	.release = seq_release_net,
2547 };
2548 
2549 struct fib_route_iter {
2550 	struct seq_net_private p;
2551 	struct fib_table *main_tb;
2552 	struct key_vector *tnode;
2553 	loff_t	pos;
2554 	t_key	key;
2555 };
2556 
2557 static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2558 					    loff_t pos)
2559 {
2560 	struct key_vector *l, **tp = &iter->tnode;
2561 	t_key key;
2562 
2563 	/* use cached location of previously found key */
2564 	if (iter->pos > 0 && pos >= iter->pos) {
2565 		key = iter->key;
2566 	} else {
2567 		iter->pos = 1;
2568 		key = 0;
2569 	}
2570 
2571 	pos -= iter->pos;
2572 
2573 	while ((l = leaf_walk_rcu(tp, key)) && (pos-- > 0)) {
2574 		key = l->key + 1;
2575 		iter->pos++;
2576 		l = NULL;
2577 
2578 		/* handle unlikely case of a key wrap */
2579 		if (!key)
2580 			break;
2581 	}
2582 
2583 	if (l)
2584 		iter->key = l->key;	/* remember it */
2585 	else
2586 		iter->pos = 0;		/* forget it */
2587 
2588 	return l;
2589 }
2590 
2591 static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2592 	__acquires(RCU)
2593 {
2594 	struct fib_route_iter *iter = seq->private;
2595 	struct fib_table *tb;
2596 	struct trie *t;
2597 
2598 	rcu_read_lock();
2599 
2600 	tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2601 	if (!tb)
2602 		return NULL;
2603 
2604 	iter->main_tb = tb;
2605 	t = (struct trie *)tb->tb_data;
2606 	iter->tnode = t->kv;
2607 
2608 	if (*pos != 0)
2609 		return fib_route_get_idx(iter, *pos);
2610 
2611 	iter->pos = 0;
2612 	iter->key = KEY_MAX;
2613 
2614 	return SEQ_START_TOKEN;
2615 }
2616 
2617 static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2618 {
2619 	struct fib_route_iter *iter = seq->private;
2620 	struct key_vector *l = NULL;
2621 	t_key key = iter->key + 1;
2622 
2623 	++*pos;
2624 
2625 	/* only allow key of 0 for start of sequence */
2626 	if ((v == SEQ_START_TOKEN) || key)
2627 		l = leaf_walk_rcu(&iter->tnode, key);
2628 
2629 	if (l) {
2630 		iter->key = l->key;
2631 		iter->pos++;
2632 	} else {
2633 		iter->pos = 0;
2634 	}
2635 
2636 	return l;
2637 }
2638 
2639 static void fib_route_seq_stop(struct seq_file *seq, void *v)
2640 	__releases(RCU)
2641 {
2642 	rcu_read_unlock();
2643 }
2644 
2645 static unsigned int fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2646 {
2647 	unsigned int flags = 0;
2648 
2649 	if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2650 		flags = RTF_REJECT;
2651 	if (fi && fi->fib_nh->nh_gw)
2652 		flags |= RTF_GATEWAY;
2653 	if (mask == htonl(0xFFFFFFFF))
2654 		flags |= RTF_HOST;
2655 	flags |= RTF_UP;
2656 	return flags;
2657 }
2658 
2659 /*
2660  *	This outputs /proc/net/route.
2661  *	The format of the file is not supposed to be changed
2662  *	and needs to be same as fib_hash output to avoid breaking
2663  *	legacy utilities
2664  */
2665 static int fib_route_seq_show(struct seq_file *seq, void *v)
2666 {
2667 	struct fib_route_iter *iter = seq->private;
2668 	struct fib_table *tb = iter->main_tb;
2669 	struct fib_alias *fa;
2670 	struct key_vector *l = v;
2671 	__be32 prefix;
2672 
2673 	if (v == SEQ_START_TOKEN) {
2674 		seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2675 			   "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2676 			   "\tWindow\tIRTT");
2677 		return 0;
2678 	}
2679 
2680 	prefix = htonl(l->key);
2681 
2682 	hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2683 		const struct fib_info *fi = fa->fa_info;
2684 		__be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2685 		unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2686 
2687 		if ((fa->fa_type == RTN_BROADCAST) ||
2688 		    (fa->fa_type == RTN_MULTICAST))
2689 			continue;
2690 
2691 		if (fa->tb_id != tb->tb_id)
2692 			continue;
2693 
2694 		seq_setwidth(seq, 127);
2695 
2696 		if (fi)
2697 			seq_printf(seq,
2698 				   "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2699 				   "%d\t%08X\t%d\t%u\t%u",
2700 				   fi->fib_dev ? fi->fib_dev->name : "*",
2701 				   prefix,
2702 				   fi->fib_nh->nh_gw, flags, 0, 0,
2703 				   fi->fib_priority,
2704 				   mask,
2705 				   (fi->fib_advmss ?
2706 				    fi->fib_advmss + 40 : 0),
2707 				   fi->fib_window,
2708 				   fi->fib_rtt >> 3);
2709 		else
2710 			seq_printf(seq,
2711 				   "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2712 				   "%d\t%08X\t%d\t%u\t%u",
2713 				   prefix, 0, flags, 0, 0, 0,
2714 				   mask, 0, 0, 0);
2715 
2716 		seq_pad(seq, '\n');
2717 	}
2718 
2719 	return 0;
2720 }
2721 
2722 static const struct seq_operations fib_route_seq_ops = {
2723 	.start  = fib_route_seq_start,
2724 	.next   = fib_route_seq_next,
2725 	.stop   = fib_route_seq_stop,
2726 	.show   = fib_route_seq_show,
2727 };
2728 
2729 static int fib_route_seq_open(struct inode *inode, struct file *file)
2730 {
2731 	return seq_open_net(inode, file, &fib_route_seq_ops,
2732 			    sizeof(struct fib_route_iter));
2733 }
2734 
2735 static const struct file_operations fib_route_fops = {
2736 	.open   = fib_route_seq_open,
2737 	.read   = seq_read,
2738 	.llseek = seq_lseek,
2739 	.release = seq_release_net,
2740 };
2741 
2742 int __net_init fib_proc_init(struct net *net)
2743 {
2744 	if (!proc_create("fib_trie", 0444, net->proc_net, &fib_trie_fops))
2745 		goto out1;
2746 
2747 	if (!proc_create("fib_triestat", 0444, net->proc_net,
2748 			 &fib_triestat_fops))
2749 		goto out2;
2750 
2751 	if (!proc_create("route", 0444, net->proc_net, &fib_route_fops))
2752 		goto out3;
2753 
2754 	return 0;
2755 
2756 out3:
2757 	remove_proc_entry("fib_triestat", net->proc_net);
2758 out2:
2759 	remove_proc_entry("fib_trie", net->proc_net);
2760 out1:
2761 	return -ENOMEM;
2762 }
2763 
2764 void __net_exit fib_proc_exit(struct net *net)
2765 {
2766 	remove_proc_entry("fib_trie", net->proc_net);
2767 	remove_proc_entry("fib_triestat", net->proc_net);
2768 	remove_proc_entry("route", net->proc_net);
2769 }
2770 
2771 #endif /* CONFIG_PROC_FS */
2772