xref: /openbmc/linux/net/ipv4/fib_trie.c (revision 2127c01b)
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  * http://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 
39 #define VERSION "0.409"
40 
41 #include <linux/cache.h>
42 #include <linux/uaccess.h>
43 #include <linux/bitops.h>
44 #include <linux/types.h>
45 #include <linux/kernel.h>
46 #include <linux/mm.h>
47 #include <linux/string.h>
48 #include <linux/socket.h>
49 #include <linux/sockios.h>
50 #include <linux/errno.h>
51 #include <linux/in.h>
52 #include <linux/inet.h>
53 #include <linux/inetdevice.h>
54 #include <linux/netdevice.h>
55 #include <linux/if_arp.h>
56 #include <linux/proc_fs.h>
57 #include <linux/rcupdate.h>
58 #include <linux/skbuff.h>
59 #include <linux/netlink.h>
60 #include <linux/init.h>
61 #include <linux/list.h>
62 #include <linux/slab.h>
63 #include <linux/export.h>
64 #include <linux/vmalloc.h>
65 #include <linux/notifier.h>
66 #include <net/net_namespace.h>
67 #include <net/ip.h>
68 #include <net/protocol.h>
69 #include <net/route.h>
70 #include <net/tcp.h>
71 #include <net/sock.h>
72 #include <net/ip_fib.h>
73 #include <net/fib_notifier.h>
74 #include <trace/events/fib.h>
75 #include "fib_lookup.h"
76 
77 static int call_fib_entry_notifier(struct notifier_block *nb, struct net *net,
78 				   enum fib_event_type event_type, u32 dst,
79 				   int dst_len, struct fib_alias *fa)
80 {
81 	struct fib_entry_notifier_info info = {
82 		.dst = dst,
83 		.dst_len = dst_len,
84 		.fi = fa->fa_info,
85 		.tos = fa->fa_tos,
86 		.type = fa->fa_type,
87 		.tb_id = fa->tb_id,
88 	};
89 	return call_fib4_notifier(nb, net, event_type, &info.info);
90 }
91 
92 static int call_fib_entry_notifiers(struct net *net,
93 				    enum fib_event_type event_type, u32 dst,
94 				    int dst_len, struct fib_alias *fa,
95 				    struct netlink_ext_ack *extack)
96 {
97 	struct fib_entry_notifier_info info = {
98 		.info.extack = extack,
99 		.dst = dst,
100 		.dst_len = dst_len,
101 		.fi = fa->fa_info,
102 		.tos = fa->fa_tos,
103 		.type = fa->fa_type,
104 		.tb_id = fa->tb_id,
105 	};
106 	return call_fib4_notifiers(net, event_type, &info.info);
107 }
108 
109 #define MAX_STAT_DEPTH 32
110 
111 #define KEYLENGTH	(8*sizeof(t_key))
112 #define KEY_MAX		((t_key)~0)
113 
114 typedef unsigned int t_key;
115 
116 #define IS_TRIE(n)	((n)->pos >= KEYLENGTH)
117 #define IS_TNODE(n)	((n)->bits)
118 #define IS_LEAF(n)	(!(n)->bits)
119 
120 struct key_vector {
121 	t_key key;
122 	unsigned char pos;		/* 2log(KEYLENGTH) bits needed */
123 	unsigned char bits;		/* 2log(KEYLENGTH) bits needed */
124 	unsigned char slen;
125 	union {
126 		/* This list pointer if valid if (pos | bits) == 0 (LEAF) */
127 		struct hlist_head leaf;
128 		/* This array is valid if (pos | bits) > 0 (TNODE) */
129 		struct key_vector __rcu *tnode[0];
130 	};
131 };
132 
133 struct tnode {
134 	struct rcu_head rcu;
135 	t_key empty_children;		/* KEYLENGTH bits needed */
136 	t_key full_children;		/* KEYLENGTH bits needed */
137 	struct key_vector __rcu *parent;
138 	struct key_vector kv[1];
139 #define tn_bits kv[0].bits
140 };
141 
142 #define TNODE_SIZE(n)	offsetof(struct tnode, kv[0].tnode[n])
143 #define LEAF_SIZE	TNODE_SIZE(1)
144 
145 #ifdef CONFIG_IP_FIB_TRIE_STATS
146 struct trie_use_stats {
147 	unsigned int gets;
148 	unsigned int backtrack;
149 	unsigned int semantic_match_passed;
150 	unsigned int semantic_match_miss;
151 	unsigned int null_node_hit;
152 	unsigned int resize_node_skipped;
153 };
154 #endif
155 
156 struct trie_stat {
157 	unsigned int totdepth;
158 	unsigned int maxdepth;
159 	unsigned int tnodes;
160 	unsigned int leaves;
161 	unsigned int nullpointers;
162 	unsigned int prefixes;
163 	unsigned int nodesizes[MAX_STAT_DEPTH];
164 };
165 
166 struct trie {
167 	struct key_vector kv[1];
168 #ifdef CONFIG_IP_FIB_TRIE_STATS
169 	struct trie_use_stats __percpu *stats;
170 #endif
171 };
172 
173 static struct key_vector *resize(struct trie *t, struct key_vector *tn);
174 static unsigned int tnode_free_size;
175 
176 /*
177  * synchronize_rcu after call_rcu for outstanding dirty memory; it should be
178  * especially useful before resizing the root node with PREEMPT_NONE configs;
179  * the value was obtained experimentally, aiming to avoid visible slowdown.
180  */
181 unsigned int sysctl_fib_sync_mem = 512 * 1024;
182 unsigned int sysctl_fib_sync_mem_min = 64 * 1024;
183 unsigned int sysctl_fib_sync_mem_max = 64 * 1024 * 1024;
184 
185 static struct kmem_cache *fn_alias_kmem __ro_after_init;
186 static struct kmem_cache *trie_leaf_kmem __ro_after_init;
187 
188 static inline struct tnode *tn_info(struct key_vector *kv)
189 {
190 	return container_of(kv, struct tnode, kv[0]);
191 }
192 
193 /* caller must hold RTNL */
194 #define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
195 #define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
196 
197 /* caller must hold RCU read lock or RTNL */
198 #define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
199 #define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
200 
201 /* wrapper for rcu_assign_pointer */
202 static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
203 {
204 	if (n)
205 		rcu_assign_pointer(tn_info(n)->parent, tp);
206 }
207 
208 #define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
209 
210 /* This provides us with the number of children in this node, in the case of a
211  * leaf this will return 0 meaning none of the children are accessible.
212  */
213 static inline unsigned long child_length(const struct key_vector *tn)
214 {
215 	return (1ul << tn->bits) & ~(1ul);
216 }
217 
218 #define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
219 
220 static inline unsigned long get_index(t_key key, struct key_vector *kv)
221 {
222 	unsigned long index = key ^ kv->key;
223 
224 	if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
225 		return 0;
226 
227 	return index >> kv->pos;
228 }
229 
230 /* To understand this stuff, an understanding of keys and all their bits is
231  * necessary. Every node in the trie has a key associated with it, but not
232  * all of the bits in that key are significant.
233  *
234  * Consider a node 'n' and its parent 'tp'.
235  *
236  * If n is a leaf, every bit in its key is significant. Its presence is
237  * necessitated by path compression, since during a tree traversal (when
238  * searching for a leaf - unless we are doing an insertion) we will completely
239  * ignore all skipped bits we encounter. Thus we need to verify, at the end of
240  * a potentially successful search, that we have indeed been walking the
241  * correct key path.
242  *
243  * Note that we can never "miss" the correct key in the tree if present by
244  * following the wrong path. Path compression ensures that segments of the key
245  * that are the same for all keys with a given prefix are skipped, but the
246  * skipped part *is* identical for each node in the subtrie below the skipped
247  * bit! trie_insert() in this implementation takes care of that.
248  *
249  * if n is an internal node - a 'tnode' here, the various parts of its key
250  * have many different meanings.
251  *
252  * Example:
253  * _________________________________________________________________
254  * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
255  * -----------------------------------------------------------------
256  *  31  30  29  28  27  26  25  24  23  22  21  20  19  18  17  16
257  *
258  * _________________________________________________________________
259  * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
260  * -----------------------------------------------------------------
261  *  15  14  13  12  11  10   9   8   7   6   5   4   3   2   1   0
262  *
263  * tp->pos = 22
264  * tp->bits = 3
265  * n->pos = 13
266  * n->bits = 4
267  *
268  * First, let's just ignore the bits that come before the parent tp, that is
269  * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
270  * point we do not use them for anything.
271  *
272  * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
273  * index into the parent's child array. That is, they will be used to find
274  * 'n' among tp's children.
275  *
276  * The bits from (n->pos + n->bits) to (tp->pos - 1) - "S" - are skipped bits
277  * for the node n.
278  *
279  * All the bits we have seen so far are significant to the node n. The rest
280  * of the bits are really not needed or indeed known in n->key.
281  *
282  * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
283  * n's child array, and will of course be different for each child.
284  *
285  * The rest of the bits, from 0 to (n->pos -1) - "u" - are completely unknown
286  * at this point.
287  */
288 
289 static const int halve_threshold = 25;
290 static const int inflate_threshold = 50;
291 static const int halve_threshold_root = 15;
292 static const int inflate_threshold_root = 30;
293 
294 static void __alias_free_mem(struct rcu_head *head)
295 {
296 	struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
297 	kmem_cache_free(fn_alias_kmem, fa);
298 }
299 
300 static inline void alias_free_mem_rcu(struct fib_alias *fa)
301 {
302 	call_rcu(&fa->rcu, __alias_free_mem);
303 }
304 
305 #define TNODE_KMALLOC_MAX \
306 	ilog2((PAGE_SIZE - TNODE_SIZE(0)) / sizeof(struct key_vector *))
307 #define TNODE_VMALLOC_MAX \
308 	ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
309 
310 static void __node_free_rcu(struct rcu_head *head)
311 {
312 	struct tnode *n = container_of(head, struct tnode, rcu);
313 
314 	if (!n->tn_bits)
315 		kmem_cache_free(trie_leaf_kmem, n);
316 	else
317 		kvfree(n);
318 }
319 
320 #define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
321 
322 static struct tnode *tnode_alloc(int bits)
323 {
324 	size_t size;
325 
326 	/* verify bits is within bounds */
327 	if (bits > TNODE_VMALLOC_MAX)
328 		return NULL;
329 
330 	/* determine size and verify it is non-zero and didn't overflow */
331 	size = TNODE_SIZE(1ul << bits);
332 
333 	if (size <= PAGE_SIZE)
334 		return kzalloc(size, GFP_KERNEL);
335 	else
336 		return vzalloc(size);
337 }
338 
339 static inline void empty_child_inc(struct key_vector *n)
340 {
341 	tn_info(n)->empty_children++;
342 
343 	if (!tn_info(n)->empty_children)
344 		tn_info(n)->full_children++;
345 }
346 
347 static inline void empty_child_dec(struct key_vector *n)
348 {
349 	if (!tn_info(n)->empty_children)
350 		tn_info(n)->full_children--;
351 
352 	tn_info(n)->empty_children--;
353 }
354 
355 static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
356 {
357 	struct key_vector *l;
358 	struct tnode *kv;
359 
360 	kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
361 	if (!kv)
362 		return NULL;
363 
364 	/* initialize key vector */
365 	l = kv->kv;
366 	l->key = key;
367 	l->pos = 0;
368 	l->bits = 0;
369 	l->slen = fa->fa_slen;
370 
371 	/* link leaf to fib alias */
372 	INIT_HLIST_HEAD(&l->leaf);
373 	hlist_add_head(&fa->fa_list, &l->leaf);
374 
375 	return l;
376 }
377 
378 static struct key_vector *tnode_new(t_key key, int pos, int bits)
379 {
380 	unsigned int shift = pos + bits;
381 	struct key_vector *tn;
382 	struct tnode *tnode;
383 
384 	/* verify bits and pos their msb bits clear and values are valid */
385 	BUG_ON(!bits || (shift > KEYLENGTH));
386 
387 	tnode = tnode_alloc(bits);
388 	if (!tnode)
389 		return NULL;
390 
391 	pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
392 		 sizeof(struct key_vector *) << bits);
393 
394 	if (bits == KEYLENGTH)
395 		tnode->full_children = 1;
396 	else
397 		tnode->empty_children = 1ul << bits;
398 
399 	tn = tnode->kv;
400 	tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
401 	tn->pos = pos;
402 	tn->bits = bits;
403 	tn->slen = pos;
404 
405 	return tn;
406 }
407 
408 /* Check whether a tnode 'n' is "full", i.e. it is an internal node
409  * and no bits are skipped. See discussion in dyntree paper p. 6
410  */
411 static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
412 {
413 	return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
414 }
415 
416 /* Add a child at position i overwriting the old value.
417  * Update the value of full_children and empty_children.
418  */
419 static void put_child(struct key_vector *tn, unsigned long i,
420 		      struct key_vector *n)
421 {
422 	struct key_vector *chi = get_child(tn, i);
423 	int isfull, wasfull;
424 
425 	BUG_ON(i >= child_length(tn));
426 
427 	/* update emptyChildren, overflow into fullChildren */
428 	if (!n && chi)
429 		empty_child_inc(tn);
430 	if (n && !chi)
431 		empty_child_dec(tn);
432 
433 	/* update fullChildren */
434 	wasfull = tnode_full(tn, chi);
435 	isfull = tnode_full(tn, n);
436 
437 	if (wasfull && !isfull)
438 		tn_info(tn)->full_children--;
439 	else if (!wasfull && isfull)
440 		tn_info(tn)->full_children++;
441 
442 	if (n && (tn->slen < n->slen))
443 		tn->slen = n->slen;
444 
445 	rcu_assign_pointer(tn->tnode[i], n);
446 }
447 
448 static void update_children(struct key_vector *tn)
449 {
450 	unsigned long i;
451 
452 	/* update all of the child parent pointers */
453 	for (i = child_length(tn); i;) {
454 		struct key_vector *inode = get_child(tn, --i);
455 
456 		if (!inode)
457 			continue;
458 
459 		/* Either update the children of a tnode that
460 		 * already belongs to us or update the child
461 		 * to point to ourselves.
462 		 */
463 		if (node_parent(inode) == tn)
464 			update_children(inode);
465 		else
466 			node_set_parent(inode, tn);
467 	}
468 }
469 
470 static inline void put_child_root(struct key_vector *tp, t_key key,
471 				  struct key_vector *n)
472 {
473 	if (IS_TRIE(tp))
474 		rcu_assign_pointer(tp->tnode[0], n);
475 	else
476 		put_child(tp, get_index(key, tp), n);
477 }
478 
479 static inline void tnode_free_init(struct key_vector *tn)
480 {
481 	tn_info(tn)->rcu.next = NULL;
482 }
483 
484 static inline void tnode_free_append(struct key_vector *tn,
485 				     struct key_vector *n)
486 {
487 	tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
488 	tn_info(tn)->rcu.next = &tn_info(n)->rcu;
489 }
490 
491 static void tnode_free(struct key_vector *tn)
492 {
493 	struct callback_head *head = &tn_info(tn)->rcu;
494 
495 	while (head) {
496 		head = head->next;
497 		tnode_free_size += TNODE_SIZE(1ul << tn->bits);
498 		node_free(tn);
499 
500 		tn = container_of(head, struct tnode, rcu)->kv;
501 	}
502 
503 	if (tnode_free_size >= sysctl_fib_sync_mem) {
504 		tnode_free_size = 0;
505 		synchronize_rcu();
506 	}
507 }
508 
509 static struct key_vector *replace(struct trie *t,
510 				  struct key_vector *oldtnode,
511 				  struct key_vector *tn)
512 {
513 	struct key_vector *tp = node_parent(oldtnode);
514 	unsigned long i;
515 
516 	/* setup the parent pointer out of and back into this node */
517 	NODE_INIT_PARENT(tn, tp);
518 	put_child_root(tp, tn->key, tn);
519 
520 	/* update all of the child parent pointers */
521 	update_children(tn);
522 
523 	/* all pointers should be clean so we are done */
524 	tnode_free(oldtnode);
525 
526 	/* resize children now that oldtnode is freed */
527 	for (i = child_length(tn); i;) {
528 		struct key_vector *inode = get_child(tn, --i);
529 
530 		/* resize child node */
531 		if (tnode_full(tn, inode))
532 			tn = resize(t, inode);
533 	}
534 
535 	return tp;
536 }
537 
538 static struct key_vector *inflate(struct trie *t,
539 				  struct key_vector *oldtnode)
540 {
541 	struct key_vector *tn;
542 	unsigned long i;
543 	t_key m;
544 
545 	pr_debug("In inflate\n");
546 
547 	tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
548 	if (!tn)
549 		goto notnode;
550 
551 	/* prepare oldtnode to be freed */
552 	tnode_free_init(oldtnode);
553 
554 	/* Assemble all of the pointers in our cluster, in this case that
555 	 * represents all of the pointers out of our allocated nodes that
556 	 * point to existing tnodes and the links between our allocated
557 	 * nodes.
558 	 */
559 	for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
560 		struct key_vector *inode = get_child(oldtnode, --i);
561 		struct key_vector *node0, *node1;
562 		unsigned long j, k;
563 
564 		/* An empty child */
565 		if (!inode)
566 			continue;
567 
568 		/* A leaf or an internal node with skipped bits */
569 		if (!tnode_full(oldtnode, inode)) {
570 			put_child(tn, get_index(inode->key, tn), inode);
571 			continue;
572 		}
573 
574 		/* drop the node in the old tnode free list */
575 		tnode_free_append(oldtnode, inode);
576 
577 		/* An internal node with two children */
578 		if (inode->bits == 1) {
579 			put_child(tn, 2 * i + 1, get_child(inode, 1));
580 			put_child(tn, 2 * i, get_child(inode, 0));
581 			continue;
582 		}
583 
584 		/* We will replace this node 'inode' with two new
585 		 * ones, 'node0' and 'node1', each with half of the
586 		 * original children. The two new nodes will have
587 		 * a position one bit further down the key and this
588 		 * means that the "significant" part of their keys
589 		 * (see the discussion near the top of this file)
590 		 * will differ by one bit, which will be "0" in
591 		 * node0's key and "1" in node1's key. Since we are
592 		 * moving the key position by one step, the bit that
593 		 * we are moving away from - the bit at position
594 		 * (tn->pos) - is the one that will differ between
595 		 * node0 and node1. So... we synthesize that bit in the
596 		 * two new keys.
597 		 */
598 		node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
599 		if (!node1)
600 			goto nomem;
601 		node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
602 
603 		tnode_free_append(tn, node1);
604 		if (!node0)
605 			goto nomem;
606 		tnode_free_append(tn, node0);
607 
608 		/* populate child pointers in new nodes */
609 		for (k = child_length(inode), j = k / 2; j;) {
610 			put_child(node1, --j, get_child(inode, --k));
611 			put_child(node0, j, get_child(inode, j));
612 			put_child(node1, --j, get_child(inode, --k));
613 			put_child(node0, j, get_child(inode, j));
614 		}
615 
616 		/* link new nodes to parent */
617 		NODE_INIT_PARENT(node1, tn);
618 		NODE_INIT_PARENT(node0, tn);
619 
620 		/* link parent to nodes */
621 		put_child(tn, 2 * i + 1, node1);
622 		put_child(tn, 2 * i, node0);
623 	}
624 
625 	/* setup the parent pointers into and out of this node */
626 	return replace(t, oldtnode, tn);
627 nomem:
628 	/* all pointers should be clean so we are done */
629 	tnode_free(tn);
630 notnode:
631 	return NULL;
632 }
633 
634 static struct key_vector *halve(struct trie *t,
635 				struct key_vector *oldtnode)
636 {
637 	struct key_vector *tn;
638 	unsigned long i;
639 
640 	pr_debug("In halve\n");
641 
642 	tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
643 	if (!tn)
644 		goto notnode;
645 
646 	/* prepare oldtnode to be freed */
647 	tnode_free_init(oldtnode);
648 
649 	/* Assemble all of the pointers in our cluster, in this case that
650 	 * represents all of the pointers out of our allocated nodes that
651 	 * point to existing tnodes and the links between our allocated
652 	 * nodes.
653 	 */
654 	for (i = child_length(oldtnode); i;) {
655 		struct key_vector *node1 = get_child(oldtnode, --i);
656 		struct key_vector *node0 = get_child(oldtnode, --i);
657 		struct key_vector *inode;
658 
659 		/* At least one of the children is empty */
660 		if (!node1 || !node0) {
661 			put_child(tn, i / 2, node1 ? : node0);
662 			continue;
663 		}
664 
665 		/* Two nonempty children */
666 		inode = tnode_new(node0->key, oldtnode->pos, 1);
667 		if (!inode)
668 			goto nomem;
669 		tnode_free_append(tn, inode);
670 
671 		/* initialize pointers out of node */
672 		put_child(inode, 1, node1);
673 		put_child(inode, 0, node0);
674 		NODE_INIT_PARENT(inode, tn);
675 
676 		/* link parent to node */
677 		put_child(tn, i / 2, inode);
678 	}
679 
680 	/* setup the parent pointers into and out of this node */
681 	return replace(t, oldtnode, tn);
682 nomem:
683 	/* all pointers should be clean so we are done */
684 	tnode_free(tn);
685 notnode:
686 	return NULL;
687 }
688 
689 static struct key_vector *collapse(struct trie *t,
690 				   struct key_vector *oldtnode)
691 {
692 	struct key_vector *n, *tp;
693 	unsigned long i;
694 
695 	/* scan the tnode looking for that one child that might still exist */
696 	for (n = NULL, i = child_length(oldtnode); !n && i;)
697 		n = get_child(oldtnode, --i);
698 
699 	/* compress one level */
700 	tp = node_parent(oldtnode);
701 	put_child_root(tp, oldtnode->key, n);
702 	node_set_parent(n, tp);
703 
704 	/* drop dead node */
705 	node_free(oldtnode);
706 
707 	return tp;
708 }
709 
710 static unsigned char update_suffix(struct key_vector *tn)
711 {
712 	unsigned char slen = tn->pos;
713 	unsigned long stride, i;
714 	unsigned char slen_max;
715 
716 	/* only vector 0 can have a suffix length greater than or equal to
717 	 * tn->pos + tn->bits, the second highest node will have a suffix
718 	 * length at most of tn->pos + tn->bits - 1
719 	 */
720 	slen_max = min_t(unsigned char, tn->pos + tn->bits - 1, tn->slen);
721 
722 	/* search though the list of children looking for nodes that might
723 	 * have a suffix greater than the one we currently have.  This is
724 	 * why we start with a stride of 2 since a stride of 1 would
725 	 * represent the nodes with suffix length equal to tn->pos
726 	 */
727 	for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
728 		struct key_vector *n = get_child(tn, i);
729 
730 		if (!n || (n->slen <= slen))
731 			continue;
732 
733 		/* update stride and slen based on new value */
734 		stride <<= (n->slen - slen);
735 		slen = n->slen;
736 		i &= ~(stride - 1);
737 
738 		/* stop searching if we have hit the maximum possible value */
739 		if (slen >= slen_max)
740 			break;
741 	}
742 
743 	tn->slen = slen;
744 
745 	return slen;
746 }
747 
748 /* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
749  * the Helsinki University of Technology and Matti Tikkanen of Nokia
750  * Telecommunications, page 6:
751  * "A node is doubled if the ratio of non-empty children to all
752  * children in the *doubled* node is at least 'high'."
753  *
754  * 'high' in this instance is the variable 'inflate_threshold'. It
755  * is expressed as a percentage, so we multiply it with
756  * child_length() and instead of multiplying by 2 (since the
757  * child array will be doubled by inflate()) and multiplying
758  * the left-hand side by 100 (to handle the percentage thing) we
759  * multiply the left-hand side by 50.
760  *
761  * The left-hand side may look a bit weird: child_length(tn)
762  * - tn->empty_children is of course the number of non-null children
763  * in the current node. tn->full_children is the number of "full"
764  * children, that is non-null tnodes with a skip value of 0.
765  * All of those will be doubled in the resulting inflated tnode, so
766  * we just count them one extra time here.
767  *
768  * A clearer way to write this would be:
769  *
770  * to_be_doubled = tn->full_children;
771  * not_to_be_doubled = child_length(tn) - tn->empty_children -
772  *     tn->full_children;
773  *
774  * new_child_length = child_length(tn) * 2;
775  *
776  * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
777  *      new_child_length;
778  * if (new_fill_factor >= inflate_threshold)
779  *
780  * ...and so on, tho it would mess up the while () loop.
781  *
782  * anyway,
783  * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
784  *      inflate_threshold
785  *
786  * avoid a division:
787  * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
788  *      inflate_threshold * new_child_length
789  *
790  * expand not_to_be_doubled and to_be_doubled, and shorten:
791  * 100 * (child_length(tn) - tn->empty_children +
792  *    tn->full_children) >= inflate_threshold * new_child_length
793  *
794  * expand new_child_length:
795  * 100 * (child_length(tn) - tn->empty_children +
796  *    tn->full_children) >=
797  *      inflate_threshold * child_length(tn) * 2
798  *
799  * shorten again:
800  * 50 * (tn->full_children + child_length(tn) -
801  *    tn->empty_children) >= inflate_threshold *
802  *    child_length(tn)
803  *
804  */
805 static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
806 {
807 	unsigned long used = child_length(tn);
808 	unsigned long threshold = used;
809 
810 	/* Keep root node larger */
811 	threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
812 	used -= tn_info(tn)->empty_children;
813 	used += tn_info(tn)->full_children;
814 
815 	/* if bits == KEYLENGTH then pos = 0, and will fail below */
816 
817 	return (used > 1) && tn->pos && ((50 * used) >= threshold);
818 }
819 
820 static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
821 {
822 	unsigned long used = child_length(tn);
823 	unsigned long threshold = used;
824 
825 	/* Keep root node larger */
826 	threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
827 	used -= tn_info(tn)->empty_children;
828 
829 	/* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
830 
831 	return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
832 }
833 
834 static inline bool should_collapse(struct key_vector *tn)
835 {
836 	unsigned long used = child_length(tn);
837 
838 	used -= tn_info(tn)->empty_children;
839 
840 	/* account for bits == KEYLENGTH case */
841 	if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
842 		used -= KEY_MAX;
843 
844 	/* One child or none, time to drop us from the trie */
845 	return used < 2;
846 }
847 
848 #define MAX_WORK 10
849 static struct key_vector *resize(struct trie *t, struct key_vector *tn)
850 {
851 #ifdef CONFIG_IP_FIB_TRIE_STATS
852 	struct trie_use_stats __percpu *stats = t->stats;
853 #endif
854 	struct key_vector *tp = node_parent(tn);
855 	unsigned long cindex = get_index(tn->key, tp);
856 	int max_work = MAX_WORK;
857 
858 	pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
859 		 tn, inflate_threshold, halve_threshold);
860 
861 	/* track the tnode via the pointer from the parent instead of
862 	 * doing it ourselves.  This way we can let RCU fully do its
863 	 * thing without us interfering
864 	 */
865 	BUG_ON(tn != get_child(tp, cindex));
866 
867 	/* Double as long as the resulting node has a number of
868 	 * nonempty nodes that are above the threshold.
869 	 */
870 	while (should_inflate(tp, tn) && max_work) {
871 		tp = inflate(t, tn);
872 		if (!tp) {
873 #ifdef CONFIG_IP_FIB_TRIE_STATS
874 			this_cpu_inc(stats->resize_node_skipped);
875 #endif
876 			break;
877 		}
878 
879 		max_work--;
880 		tn = get_child(tp, cindex);
881 	}
882 
883 	/* update parent in case inflate failed */
884 	tp = node_parent(tn);
885 
886 	/* Return if at least one inflate is run */
887 	if (max_work != MAX_WORK)
888 		return tp;
889 
890 	/* Halve as long as the number of empty children in this
891 	 * node is above threshold.
892 	 */
893 	while (should_halve(tp, tn) && max_work) {
894 		tp = halve(t, tn);
895 		if (!tp) {
896 #ifdef CONFIG_IP_FIB_TRIE_STATS
897 			this_cpu_inc(stats->resize_node_skipped);
898 #endif
899 			break;
900 		}
901 
902 		max_work--;
903 		tn = get_child(tp, cindex);
904 	}
905 
906 	/* Only one child remains */
907 	if (should_collapse(tn))
908 		return collapse(t, tn);
909 
910 	/* update parent in case halve failed */
911 	return node_parent(tn);
912 }
913 
914 static void node_pull_suffix(struct key_vector *tn, unsigned char slen)
915 {
916 	unsigned char node_slen = tn->slen;
917 
918 	while ((node_slen > tn->pos) && (node_slen > slen)) {
919 		slen = update_suffix(tn);
920 		if (node_slen == slen)
921 			break;
922 
923 		tn = node_parent(tn);
924 		node_slen = tn->slen;
925 	}
926 }
927 
928 static void node_push_suffix(struct key_vector *tn, unsigned char slen)
929 {
930 	while (tn->slen < slen) {
931 		tn->slen = slen;
932 		tn = node_parent(tn);
933 	}
934 }
935 
936 /* rcu_read_lock needs to be hold by caller from readside */
937 static struct key_vector *fib_find_node(struct trie *t,
938 					struct key_vector **tp, u32 key)
939 {
940 	struct key_vector *pn, *n = t->kv;
941 	unsigned long index = 0;
942 
943 	do {
944 		pn = n;
945 		n = get_child_rcu(n, index);
946 
947 		if (!n)
948 			break;
949 
950 		index = get_cindex(key, n);
951 
952 		/* This bit of code is a bit tricky but it combines multiple
953 		 * checks into a single check.  The prefix consists of the
954 		 * prefix plus zeros for the bits in the cindex. The index
955 		 * is the difference between the key and this value.  From
956 		 * this we can actually derive several pieces of data.
957 		 *   if (index >= (1ul << bits))
958 		 *     we have a mismatch in skip bits and failed
959 		 *   else
960 		 *     we know the value is cindex
961 		 *
962 		 * This check is safe even if bits == KEYLENGTH due to the
963 		 * fact that we can only allocate a node with 32 bits if a
964 		 * long is greater than 32 bits.
965 		 */
966 		if (index >= (1ul << n->bits)) {
967 			n = NULL;
968 			break;
969 		}
970 
971 		/* keep searching until we find a perfect match leaf or NULL */
972 	} while (IS_TNODE(n));
973 
974 	*tp = pn;
975 
976 	return n;
977 }
978 
979 /* Return the first fib alias matching TOS with
980  * priority less than or equal to PRIO.
981  */
982 static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
983 					u8 tos, u32 prio, u32 tb_id)
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 (fa->fa_tos > tos)
1000 			continue;
1001 		if (fa->fa_info->fib_priority >= prio || fa->fa_tos < tos)
1002 			return fa;
1003 	}
1004 
1005 	return NULL;
1006 }
1007 
1008 static void trie_rebalance(struct trie *t, struct key_vector *tn)
1009 {
1010 	while (!IS_TRIE(tn))
1011 		tn = resize(t, tn);
1012 }
1013 
1014 static int fib_insert_node(struct trie *t, struct key_vector *tp,
1015 			   struct fib_alias *new, t_key key)
1016 {
1017 	struct key_vector *n, *l;
1018 
1019 	l = leaf_new(key, new);
1020 	if (!l)
1021 		goto noleaf;
1022 
1023 	/* retrieve child from parent node */
1024 	n = get_child(tp, get_index(key, tp));
1025 
1026 	/* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1027 	 *
1028 	 *  Add a new tnode here
1029 	 *  first tnode need some special handling
1030 	 *  leaves us in position for handling as case 3
1031 	 */
1032 	if (n) {
1033 		struct key_vector *tn;
1034 
1035 		tn = tnode_new(key, __fls(key ^ n->key), 1);
1036 		if (!tn)
1037 			goto notnode;
1038 
1039 		/* initialize routes out of node */
1040 		NODE_INIT_PARENT(tn, tp);
1041 		put_child(tn, get_index(key, tn) ^ 1, n);
1042 
1043 		/* start adding routes into the node */
1044 		put_child_root(tp, key, tn);
1045 		node_set_parent(n, tn);
1046 
1047 		/* parent now has a NULL spot where the leaf can go */
1048 		tp = tn;
1049 	}
1050 
1051 	/* Case 3: n is NULL, and will just insert a new leaf */
1052 	node_push_suffix(tp, new->fa_slen);
1053 	NODE_INIT_PARENT(l, tp);
1054 	put_child_root(tp, key, l);
1055 	trie_rebalance(t, tp);
1056 
1057 	return 0;
1058 notnode:
1059 	node_free(l);
1060 noleaf:
1061 	return -ENOMEM;
1062 }
1063 
1064 /* fib notifier for ADD is sent before calling fib_insert_alias with
1065  * the expectation that the only possible failure ENOMEM
1066  */
1067 static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1068 			    struct key_vector *l, struct fib_alias *new,
1069 			    struct fib_alias *fa, t_key key)
1070 {
1071 	if (!l)
1072 		return fib_insert_node(t, tp, new, key);
1073 
1074 	if (fa) {
1075 		hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1076 	} else {
1077 		struct fib_alias *last;
1078 
1079 		hlist_for_each_entry(last, &l->leaf, fa_list) {
1080 			if (new->fa_slen < last->fa_slen)
1081 				break;
1082 			if ((new->fa_slen == last->fa_slen) &&
1083 			    (new->tb_id > last->tb_id))
1084 				break;
1085 			fa = last;
1086 		}
1087 
1088 		if (fa)
1089 			hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1090 		else
1091 			hlist_add_head_rcu(&new->fa_list, &l->leaf);
1092 	}
1093 
1094 	/* if we added to the tail node then we need to update slen */
1095 	if (l->slen < new->fa_slen) {
1096 		l->slen = new->fa_slen;
1097 		node_push_suffix(tp, new->fa_slen);
1098 	}
1099 
1100 	return 0;
1101 }
1102 
1103 static bool fib_valid_key_len(u32 key, u8 plen, struct netlink_ext_ack *extack)
1104 {
1105 	if (plen > KEYLENGTH) {
1106 		NL_SET_ERR_MSG(extack, "Invalid prefix length");
1107 		return false;
1108 	}
1109 
1110 	if ((plen < KEYLENGTH) && (key << plen)) {
1111 		NL_SET_ERR_MSG(extack,
1112 			       "Invalid prefix for given prefix length");
1113 		return false;
1114 	}
1115 
1116 	return true;
1117 }
1118 
1119 /* Caller must hold RTNL. */
1120 int fib_table_insert(struct net *net, struct fib_table *tb,
1121 		     struct fib_config *cfg, struct netlink_ext_ack *extack)
1122 {
1123 	enum fib_event_type event = FIB_EVENT_ENTRY_ADD;
1124 	struct trie *t = (struct trie *)tb->tb_data;
1125 	struct fib_alias *fa, *new_fa;
1126 	struct key_vector *l, *tp;
1127 	u16 nlflags = NLM_F_EXCL;
1128 	struct fib_info *fi;
1129 	u8 plen = cfg->fc_dst_len;
1130 	u8 slen = KEYLENGTH - plen;
1131 	u8 tos = cfg->fc_tos;
1132 	u32 key;
1133 	int err;
1134 
1135 	key = ntohl(cfg->fc_dst);
1136 
1137 	if (!fib_valid_key_len(key, plen, extack))
1138 		return -EINVAL;
1139 
1140 	pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1141 
1142 	fi = fib_create_info(cfg, extack);
1143 	if (IS_ERR(fi)) {
1144 		err = PTR_ERR(fi);
1145 		goto err;
1146 	}
1147 
1148 	l = fib_find_node(t, &tp, key);
1149 	fa = l ? fib_find_alias(&l->leaf, slen, tos, fi->fib_priority,
1150 				tb->tb_id) : NULL;
1151 
1152 	/* Now fa, if non-NULL, points to the first fib alias
1153 	 * with the same keys [prefix,tos,priority], if such key already
1154 	 * exists or to the node before which we will insert new one.
1155 	 *
1156 	 * If fa is NULL, we will need to allocate a new one and
1157 	 * insert to the tail of the section matching the suffix length
1158 	 * of the new alias.
1159 	 */
1160 
1161 	if (fa && fa->fa_tos == tos &&
1162 	    fa->fa_info->fib_priority == fi->fib_priority) {
1163 		struct fib_alias *fa_first, *fa_match;
1164 
1165 		err = -EEXIST;
1166 		if (cfg->fc_nlflags & NLM_F_EXCL)
1167 			goto out;
1168 
1169 		nlflags &= ~NLM_F_EXCL;
1170 
1171 		/* We have 2 goals:
1172 		 * 1. Find exact match for type, scope, fib_info to avoid
1173 		 * duplicate routes
1174 		 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1175 		 */
1176 		fa_match = NULL;
1177 		fa_first = fa;
1178 		hlist_for_each_entry_from(fa, fa_list) {
1179 			if ((fa->fa_slen != slen) ||
1180 			    (fa->tb_id != tb->tb_id) ||
1181 			    (fa->fa_tos != tos))
1182 				break;
1183 			if (fa->fa_info->fib_priority != fi->fib_priority)
1184 				break;
1185 			if (fa->fa_type == cfg->fc_type &&
1186 			    fa->fa_info == fi) {
1187 				fa_match = fa;
1188 				break;
1189 			}
1190 		}
1191 
1192 		if (cfg->fc_nlflags & NLM_F_REPLACE) {
1193 			struct fib_info *fi_drop;
1194 			u8 state;
1195 
1196 			nlflags |= NLM_F_REPLACE;
1197 			fa = fa_first;
1198 			if (fa_match) {
1199 				if (fa == fa_match)
1200 					err = 0;
1201 				goto out;
1202 			}
1203 			err = -ENOBUFS;
1204 			new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1205 			if (!new_fa)
1206 				goto out;
1207 
1208 			fi_drop = fa->fa_info;
1209 			new_fa->fa_tos = fa->fa_tos;
1210 			new_fa->fa_info = fi;
1211 			new_fa->fa_type = cfg->fc_type;
1212 			state = fa->fa_state;
1213 			new_fa->fa_state = state & ~FA_S_ACCESSED;
1214 			new_fa->fa_slen = fa->fa_slen;
1215 			new_fa->tb_id = tb->tb_id;
1216 			new_fa->fa_default = -1;
1217 
1218 			err = call_fib_entry_notifiers(net,
1219 						       FIB_EVENT_ENTRY_REPLACE,
1220 						       key, plen, new_fa,
1221 						       extack);
1222 			if (err)
1223 				goto out_free_new_fa;
1224 
1225 			rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1226 				  tb->tb_id, &cfg->fc_nlinfo, nlflags);
1227 
1228 			hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1229 
1230 			alias_free_mem_rcu(fa);
1231 
1232 			fib_release_info(fi_drop);
1233 			if (state & FA_S_ACCESSED)
1234 				rt_cache_flush(cfg->fc_nlinfo.nl_net);
1235 
1236 			goto succeeded;
1237 		}
1238 		/* Error if we find a perfect match which
1239 		 * uses the same scope, type, and nexthop
1240 		 * information.
1241 		 */
1242 		if (fa_match)
1243 			goto out;
1244 
1245 		if (cfg->fc_nlflags & NLM_F_APPEND) {
1246 			event = FIB_EVENT_ENTRY_APPEND;
1247 			nlflags |= NLM_F_APPEND;
1248 		} else {
1249 			fa = fa_first;
1250 		}
1251 	}
1252 	err = -ENOENT;
1253 	if (!(cfg->fc_nlflags & NLM_F_CREATE))
1254 		goto out;
1255 
1256 	nlflags |= NLM_F_CREATE;
1257 	err = -ENOBUFS;
1258 	new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1259 	if (!new_fa)
1260 		goto out;
1261 
1262 	new_fa->fa_info = fi;
1263 	new_fa->fa_tos = tos;
1264 	new_fa->fa_type = cfg->fc_type;
1265 	new_fa->fa_state = 0;
1266 	new_fa->fa_slen = slen;
1267 	new_fa->tb_id = tb->tb_id;
1268 	new_fa->fa_default = -1;
1269 
1270 	err = call_fib_entry_notifiers(net, event, key, plen, new_fa, extack);
1271 	if (err)
1272 		goto out_free_new_fa;
1273 
1274 	/* Insert new entry to the list. */
1275 	err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1276 	if (err)
1277 		goto out_fib_notif;
1278 
1279 	if (!plen)
1280 		tb->tb_num_default++;
1281 
1282 	rt_cache_flush(cfg->fc_nlinfo.nl_net);
1283 	rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1284 		  &cfg->fc_nlinfo, nlflags);
1285 succeeded:
1286 	return 0;
1287 
1288 out_fib_notif:
1289 	/* notifier was sent that entry would be added to trie, but
1290 	 * the add failed and need to recover. Only failure for
1291 	 * fib_insert_alias is ENOMEM.
1292 	 */
1293 	NL_SET_ERR_MSG(extack, "Failed to insert route into trie");
1294 	call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, key,
1295 				 plen, new_fa, NULL);
1296 out_free_new_fa:
1297 	kmem_cache_free(fn_alias_kmem, new_fa);
1298 out:
1299 	fib_release_info(fi);
1300 err:
1301 	return err;
1302 }
1303 
1304 static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1305 {
1306 	t_key prefix = n->key;
1307 
1308 	return (key ^ prefix) & (prefix | -prefix);
1309 }
1310 
1311 /* should be called with rcu_read_lock */
1312 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1313 		     struct fib_result *res, int fib_flags)
1314 {
1315 	struct trie *t = (struct trie *) tb->tb_data;
1316 #ifdef CONFIG_IP_FIB_TRIE_STATS
1317 	struct trie_use_stats __percpu *stats = t->stats;
1318 #endif
1319 	const t_key key = ntohl(flp->daddr);
1320 	struct key_vector *n, *pn;
1321 	struct fib_alias *fa;
1322 	unsigned long index;
1323 	t_key cindex;
1324 
1325 	pn = t->kv;
1326 	cindex = 0;
1327 
1328 	n = get_child_rcu(pn, cindex);
1329 	if (!n) {
1330 		trace_fib_table_lookup(tb->tb_id, flp, NULL, -EAGAIN);
1331 		return -EAGAIN;
1332 	}
1333 
1334 #ifdef CONFIG_IP_FIB_TRIE_STATS
1335 	this_cpu_inc(stats->gets);
1336 #endif
1337 
1338 	/* Step 1: Travel to the longest prefix match in the trie */
1339 	for (;;) {
1340 		index = get_cindex(key, n);
1341 
1342 		/* This bit of code is a bit tricky but it combines multiple
1343 		 * checks into a single check.  The prefix consists of the
1344 		 * prefix plus zeros for the "bits" in the prefix. The index
1345 		 * is the difference between the key and this value.  From
1346 		 * this we can actually derive several pieces of data.
1347 		 *   if (index >= (1ul << bits))
1348 		 *     we have a mismatch in skip bits and failed
1349 		 *   else
1350 		 *     we know the value is cindex
1351 		 *
1352 		 * This check is safe even if bits == KEYLENGTH due to the
1353 		 * fact that we can only allocate a node with 32 bits if a
1354 		 * long is greater than 32 bits.
1355 		 */
1356 		if (index >= (1ul << n->bits))
1357 			break;
1358 
1359 		/* we have found a leaf. Prefixes have already been compared */
1360 		if (IS_LEAF(n))
1361 			goto found;
1362 
1363 		/* only record pn and cindex if we are going to be chopping
1364 		 * bits later.  Otherwise we are just wasting cycles.
1365 		 */
1366 		if (n->slen > n->pos) {
1367 			pn = n;
1368 			cindex = index;
1369 		}
1370 
1371 		n = get_child_rcu(n, index);
1372 		if (unlikely(!n))
1373 			goto backtrace;
1374 	}
1375 
1376 	/* Step 2: Sort out leaves and begin backtracing for longest prefix */
1377 	for (;;) {
1378 		/* record the pointer where our next node pointer is stored */
1379 		struct key_vector __rcu **cptr = n->tnode;
1380 
1381 		/* This test verifies that none of the bits that differ
1382 		 * between the key and the prefix exist in the region of
1383 		 * the lsb and higher in the prefix.
1384 		 */
1385 		if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1386 			goto backtrace;
1387 
1388 		/* exit out and process leaf */
1389 		if (unlikely(IS_LEAF(n)))
1390 			break;
1391 
1392 		/* Don't bother recording parent info.  Since we are in
1393 		 * prefix match mode we will have to come back to wherever
1394 		 * we started this traversal anyway
1395 		 */
1396 
1397 		while ((n = rcu_dereference(*cptr)) == NULL) {
1398 backtrace:
1399 #ifdef CONFIG_IP_FIB_TRIE_STATS
1400 			if (!n)
1401 				this_cpu_inc(stats->null_node_hit);
1402 #endif
1403 			/* If we are at cindex 0 there are no more bits for
1404 			 * us to strip at this level so we must ascend back
1405 			 * up one level to see if there are any more bits to
1406 			 * be stripped there.
1407 			 */
1408 			while (!cindex) {
1409 				t_key pkey = pn->key;
1410 
1411 				/* If we don't have a parent then there is
1412 				 * nothing for us to do as we do not have any
1413 				 * further nodes to parse.
1414 				 */
1415 				if (IS_TRIE(pn)) {
1416 					trace_fib_table_lookup(tb->tb_id, flp,
1417 							       NULL, -EAGAIN);
1418 					return -EAGAIN;
1419 				}
1420 #ifdef CONFIG_IP_FIB_TRIE_STATS
1421 				this_cpu_inc(stats->backtrack);
1422 #endif
1423 				/* Get Child's index */
1424 				pn = node_parent_rcu(pn);
1425 				cindex = get_index(pkey, pn);
1426 			}
1427 
1428 			/* strip the least significant bit from the cindex */
1429 			cindex &= cindex - 1;
1430 
1431 			/* grab pointer for next child node */
1432 			cptr = &pn->tnode[cindex];
1433 		}
1434 	}
1435 
1436 found:
1437 	/* this line carries forward the xor from earlier in the function */
1438 	index = key ^ n->key;
1439 
1440 	/* Step 3: Process the leaf, if that fails fall back to backtracing */
1441 	hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1442 		struct fib_info *fi = fa->fa_info;
1443 		int nhsel, err;
1444 
1445 		if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1446 			if (index >= (1ul << fa->fa_slen))
1447 				continue;
1448 		}
1449 		if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1450 			continue;
1451 		if (fi->fib_dead)
1452 			continue;
1453 		if (fa->fa_info->fib_scope < flp->flowi4_scope)
1454 			continue;
1455 		fib_alias_accessed(fa);
1456 		err = fib_props[fa->fa_type].error;
1457 		if (unlikely(err < 0)) {
1458 out_reject:
1459 #ifdef CONFIG_IP_FIB_TRIE_STATS
1460 			this_cpu_inc(stats->semantic_match_passed);
1461 #endif
1462 			trace_fib_table_lookup(tb->tb_id, flp, NULL, err);
1463 			return err;
1464 		}
1465 		if (fi->fib_flags & RTNH_F_DEAD)
1466 			continue;
1467 
1468 		if (unlikely(fi->nh && nexthop_is_blackhole(fi->nh))) {
1469 			err = fib_props[RTN_BLACKHOLE].error;
1470 			goto out_reject;
1471 		}
1472 
1473 		for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
1474 			struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
1475 
1476 			if (nhc->nhc_flags & RTNH_F_DEAD)
1477 				continue;
1478 			if (ip_ignore_linkdown(nhc->nhc_dev) &&
1479 			    nhc->nhc_flags & RTNH_F_LINKDOWN &&
1480 			    !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1481 				continue;
1482 			if (!(flp->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF)) {
1483 				if (flp->flowi4_oif &&
1484 				    flp->flowi4_oif != nhc->nhc_oif)
1485 					continue;
1486 			}
1487 
1488 			if (!(fib_flags & FIB_LOOKUP_NOREF))
1489 				refcount_inc(&fi->fib_clntref);
1490 
1491 			res->prefix = htonl(n->key);
1492 			res->prefixlen = KEYLENGTH - fa->fa_slen;
1493 			res->nh_sel = nhsel;
1494 			res->nhc = nhc;
1495 			res->type = fa->fa_type;
1496 			res->scope = fi->fib_scope;
1497 			res->fi = fi;
1498 			res->table = tb;
1499 			res->fa_head = &n->leaf;
1500 #ifdef CONFIG_IP_FIB_TRIE_STATS
1501 			this_cpu_inc(stats->semantic_match_passed);
1502 #endif
1503 			trace_fib_table_lookup(tb->tb_id, flp, nhc, err);
1504 
1505 			return err;
1506 		}
1507 	}
1508 #ifdef CONFIG_IP_FIB_TRIE_STATS
1509 	this_cpu_inc(stats->semantic_match_miss);
1510 #endif
1511 	goto backtrace;
1512 }
1513 EXPORT_SYMBOL_GPL(fib_table_lookup);
1514 
1515 static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1516 			     struct key_vector *l, struct fib_alias *old)
1517 {
1518 	/* record the location of the previous list_info entry */
1519 	struct hlist_node **pprev = old->fa_list.pprev;
1520 	struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1521 
1522 	/* remove the fib_alias from the list */
1523 	hlist_del_rcu(&old->fa_list);
1524 
1525 	/* if we emptied the list this leaf will be freed and we can sort
1526 	 * out parent suffix lengths as a part of trie_rebalance
1527 	 */
1528 	if (hlist_empty(&l->leaf)) {
1529 		if (tp->slen == l->slen)
1530 			node_pull_suffix(tp, tp->pos);
1531 		put_child_root(tp, l->key, NULL);
1532 		node_free(l);
1533 		trie_rebalance(t, tp);
1534 		return;
1535 	}
1536 
1537 	/* only access fa if it is pointing at the last valid hlist_node */
1538 	if (*pprev)
1539 		return;
1540 
1541 	/* update the trie with the latest suffix length */
1542 	l->slen = fa->fa_slen;
1543 	node_pull_suffix(tp, fa->fa_slen);
1544 }
1545 
1546 /* Caller must hold RTNL. */
1547 int fib_table_delete(struct net *net, struct fib_table *tb,
1548 		     struct fib_config *cfg, struct netlink_ext_ack *extack)
1549 {
1550 	struct trie *t = (struct trie *) tb->tb_data;
1551 	struct fib_alias *fa, *fa_to_delete;
1552 	struct key_vector *l, *tp;
1553 	u8 plen = cfg->fc_dst_len;
1554 	u8 slen = KEYLENGTH - plen;
1555 	u8 tos = cfg->fc_tos;
1556 	u32 key;
1557 
1558 	key = ntohl(cfg->fc_dst);
1559 
1560 	if (!fib_valid_key_len(key, plen, extack))
1561 		return -EINVAL;
1562 
1563 	l = fib_find_node(t, &tp, key);
1564 	if (!l)
1565 		return -ESRCH;
1566 
1567 	fa = fib_find_alias(&l->leaf, slen, tos, 0, tb->tb_id);
1568 	if (!fa)
1569 		return -ESRCH;
1570 
1571 	pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1572 
1573 	fa_to_delete = NULL;
1574 	hlist_for_each_entry_from(fa, fa_list) {
1575 		struct fib_info *fi = fa->fa_info;
1576 
1577 		if ((fa->fa_slen != slen) ||
1578 		    (fa->tb_id != tb->tb_id) ||
1579 		    (fa->fa_tos != tos))
1580 			break;
1581 
1582 		if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1583 		    (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1584 		     fa->fa_info->fib_scope == cfg->fc_scope) &&
1585 		    (!cfg->fc_prefsrc ||
1586 		     fi->fib_prefsrc == cfg->fc_prefsrc) &&
1587 		    (!cfg->fc_protocol ||
1588 		     fi->fib_protocol == cfg->fc_protocol) &&
1589 		    fib_nh_match(cfg, fi, extack) == 0 &&
1590 		    fib_metrics_match(cfg, fi)) {
1591 			fa_to_delete = fa;
1592 			break;
1593 		}
1594 	}
1595 
1596 	if (!fa_to_delete)
1597 		return -ESRCH;
1598 
1599 	call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, key, plen,
1600 				 fa_to_delete, extack);
1601 	rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1602 		  &cfg->fc_nlinfo, 0);
1603 
1604 	if (!plen)
1605 		tb->tb_num_default--;
1606 
1607 	fib_remove_alias(t, tp, l, fa_to_delete);
1608 
1609 	if (fa_to_delete->fa_state & FA_S_ACCESSED)
1610 		rt_cache_flush(cfg->fc_nlinfo.nl_net);
1611 
1612 	fib_release_info(fa_to_delete->fa_info);
1613 	alias_free_mem_rcu(fa_to_delete);
1614 	return 0;
1615 }
1616 
1617 /* Scan for the next leaf starting at the provided key value */
1618 static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1619 {
1620 	struct key_vector *pn, *n = *tn;
1621 	unsigned long cindex;
1622 
1623 	/* this loop is meant to try and find the key in the trie */
1624 	do {
1625 		/* record parent and next child index */
1626 		pn = n;
1627 		cindex = (key > pn->key) ? get_index(key, pn) : 0;
1628 
1629 		if (cindex >> pn->bits)
1630 			break;
1631 
1632 		/* descend into the next child */
1633 		n = get_child_rcu(pn, cindex++);
1634 		if (!n)
1635 			break;
1636 
1637 		/* guarantee forward progress on the keys */
1638 		if (IS_LEAF(n) && (n->key >= key))
1639 			goto found;
1640 	} while (IS_TNODE(n));
1641 
1642 	/* this loop will search for the next leaf with a greater key */
1643 	while (!IS_TRIE(pn)) {
1644 		/* if we exhausted the parent node we will need to climb */
1645 		if (cindex >= (1ul << pn->bits)) {
1646 			t_key pkey = pn->key;
1647 
1648 			pn = node_parent_rcu(pn);
1649 			cindex = get_index(pkey, pn) + 1;
1650 			continue;
1651 		}
1652 
1653 		/* grab the next available node */
1654 		n = get_child_rcu(pn, cindex++);
1655 		if (!n)
1656 			continue;
1657 
1658 		/* no need to compare keys since we bumped the index */
1659 		if (IS_LEAF(n))
1660 			goto found;
1661 
1662 		/* Rescan start scanning in new node */
1663 		pn = n;
1664 		cindex = 0;
1665 	}
1666 
1667 	*tn = pn;
1668 	return NULL; /* Root of trie */
1669 found:
1670 	/* if we are at the limit for keys just return NULL for the tnode */
1671 	*tn = pn;
1672 	return n;
1673 }
1674 
1675 static void fib_trie_free(struct fib_table *tb)
1676 {
1677 	struct trie *t = (struct trie *)tb->tb_data;
1678 	struct key_vector *pn = t->kv;
1679 	unsigned long cindex = 1;
1680 	struct hlist_node *tmp;
1681 	struct fib_alias *fa;
1682 
1683 	/* walk trie in reverse order and free everything */
1684 	for (;;) {
1685 		struct key_vector *n;
1686 
1687 		if (!(cindex--)) {
1688 			t_key pkey = pn->key;
1689 
1690 			if (IS_TRIE(pn))
1691 				break;
1692 
1693 			n = pn;
1694 			pn = node_parent(pn);
1695 
1696 			/* drop emptied tnode */
1697 			put_child_root(pn, n->key, NULL);
1698 			node_free(n);
1699 
1700 			cindex = get_index(pkey, pn);
1701 
1702 			continue;
1703 		}
1704 
1705 		/* grab the next available node */
1706 		n = get_child(pn, cindex);
1707 		if (!n)
1708 			continue;
1709 
1710 		if (IS_TNODE(n)) {
1711 			/* record pn and cindex for leaf walking */
1712 			pn = n;
1713 			cindex = 1ul << n->bits;
1714 
1715 			continue;
1716 		}
1717 
1718 		hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1719 			hlist_del_rcu(&fa->fa_list);
1720 			alias_free_mem_rcu(fa);
1721 		}
1722 
1723 		put_child_root(pn, n->key, NULL);
1724 		node_free(n);
1725 	}
1726 
1727 #ifdef CONFIG_IP_FIB_TRIE_STATS
1728 	free_percpu(t->stats);
1729 #endif
1730 	kfree(tb);
1731 }
1732 
1733 struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1734 {
1735 	struct trie *ot = (struct trie *)oldtb->tb_data;
1736 	struct key_vector *l, *tp = ot->kv;
1737 	struct fib_table *local_tb;
1738 	struct fib_alias *fa;
1739 	struct trie *lt;
1740 	t_key key = 0;
1741 
1742 	if (oldtb->tb_data == oldtb->__data)
1743 		return oldtb;
1744 
1745 	local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1746 	if (!local_tb)
1747 		return NULL;
1748 
1749 	lt = (struct trie *)local_tb->tb_data;
1750 
1751 	while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1752 		struct key_vector *local_l = NULL, *local_tp;
1753 
1754 		hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1755 			struct fib_alias *new_fa;
1756 
1757 			if (local_tb->tb_id != fa->tb_id)
1758 				continue;
1759 
1760 			/* clone fa for new local table */
1761 			new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1762 			if (!new_fa)
1763 				goto out;
1764 
1765 			memcpy(new_fa, fa, sizeof(*fa));
1766 
1767 			/* insert clone into table */
1768 			if (!local_l)
1769 				local_l = fib_find_node(lt, &local_tp, l->key);
1770 
1771 			if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1772 					     NULL, l->key)) {
1773 				kmem_cache_free(fn_alias_kmem, new_fa);
1774 				goto out;
1775 			}
1776 		}
1777 
1778 		/* stop loop if key wrapped back to 0 */
1779 		key = l->key + 1;
1780 		if (key < l->key)
1781 			break;
1782 	}
1783 
1784 	return local_tb;
1785 out:
1786 	fib_trie_free(local_tb);
1787 
1788 	return NULL;
1789 }
1790 
1791 /* Caller must hold RTNL */
1792 void fib_table_flush_external(struct fib_table *tb)
1793 {
1794 	struct trie *t = (struct trie *)tb->tb_data;
1795 	struct key_vector *pn = t->kv;
1796 	unsigned long cindex = 1;
1797 	struct hlist_node *tmp;
1798 	struct fib_alias *fa;
1799 
1800 	/* walk trie in reverse order */
1801 	for (;;) {
1802 		unsigned char slen = 0;
1803 		struct key_vector *n;
1804 
1805 		if (!(cindex--)) {
1806 			t_key pkey = pn->key;
1807 
1808 			/* cannot resize the trie vector */
1809 			if (IS_TRIE(pn))
1810 				break;
1811 
1812 			/* update the suffix to address pulled leaves */
1813 			if (pn->slen > pn->pos)
1814 				update_suffix(pn);
1815 
1816 			/* resize completed node */
1817 			pn = resize(t, pn);
1818 			cindex = get_index(pkey, pn);
1819 
1820 			continue;
1821 		}
1822 
1823 		/* grab the next available node */
1824 		n = get_child(pn, cindex);
1825 		if (!n)
1826 			continue;
1827 
1828 		if (IS_TNODE(n)) {
1829 			/* record pn and cindex for leaf walking */
1830 			pn = n;
1831 			cindex = 1ul << n->bits;
1832 
1833 			continue;
1834 		}
1835 
1836 		hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1837 			/* if alias was cloned to local then we just
1838 			 * need to remove the local copy from main
1839 			 */
1840 			if (tb->tb_id != fa->tb_id) {
1841 				hlist_del_rcu(&fa->fa_list);
1842 				alias_free_mem_rcu(fa);
1843 				continue;
1844 			}
1845 
1846 			/* record local slen */
1847 			slen = fa->fa_slen;
1848 		}
1849 
1850 		/* update leaf slen */
1851 		n->slen = slen;
1852 
1853 		if (hlist_empty(&n->leaf)) {
1854 			put_child_root(pn, n->key, NULL);
1855 			node_free(n);
1856 		}
1857 	}
1858 }
1859 
1860 /* Caller must hold RTNL. */
1861 int fib_table_flush(struct net *net, struct fib_table *tb, bool flush_all)
1862 {
1863 	struct trie *t = (struct trie *)tb->tb_data;
1864 	struct key_vector *pn = t->kv;
1865 	unsigned long cindex = 1;
1866 	struct hlist_node *tmp;
1867 	struct fib_alias *fa;
1868 	int found = 0;
1869 
1870 	/* walk trie in reverse order */
1871 	for (;;) {
1872 		unsigned char slen = 0;
1873 		struct key_vector *n;
1874 
1875 		if (!(cindex--)) {
1876 			t_key pkey = pn->key;
1877 
1878 			/* cannot resize the trie vector */
1879 			if (IS_TRIE(pn))
1880 				break;
1881 
1882 			/* update the suffix to address pulled leaves */
1883 			if (pn->slen > pn->pos)
1884 				update_suffix(pn);
1885 
1886 			/* resize completed node */
1887 			pn = resize(t, pn);
1888 			cindex = get_index(pkey, pn);
1889 
1890 			continue;
1891 		}
1892 
1893 		/* grab the next available node */
1894 		n = get_child(pn, cindex);
1895 		if (!n)
1896 			continue;
1897 
1898 		if (IS_TNODE(n)) {
1899 			/* record pn and cindex for leaf walking */
1900 			pn = n;
1901 			cindex = 1ul << n->bits;
1902 
1903 			continue;
1904 		}
1905 
1906 		hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1907 			struct fib_info *fi = fa->fa_info;
1908 
1909 			if (!fi || tb->tb_id != fa->tb_id ||
1910 			    (!(fi->fib_flags & RTNH_F_DEAD) &&
1911 			     !fib_props[fa->fa_type].error)) {
1912 				slen = fa->fa_slen;
1913 				continue;
1914 			}
1915 
1916 			/* Do not flush error routes if network namespace is
1917 			 * not being dismantled
1918 			 */
1919 			if (!flush_all && fib_props[fa->fa_type].error) {
1920 				slen = fa->fa_slen;
1921 				continue;
1922 			}
1923 
1924 			call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL,
1925 						 n->key,
1926 						 KEYLENGTH - fa->fa_slen, fa,
1927 						 NULL);
1928 			hlist_del_rcu(&fa->fa_list);
1929 			fib_release_info(fa->fa_info);
1930 			alias_free_mem_rcu(fa);
1931 			found++;
1932 		}
1933 
1934 		/* update leaf slen */
1935 		n->slen = slen;
1936 
1937 		if (hlist_empty(&n->leaf)) {
1938 			put_child_root(pn, n->key, NULL);
1939 			node_free(n);
1940 		}
1941 	}
1942 
1943 	pr_debug("trie_flush found=%d\n", found);
1944 	return found;
1945 }
1946 
1947 /* derived from fib_trie_free */
1948 static void __fib_info_notify_update(struct net *net, struct fib_table *tb,
1949 				     struct nl_info *info)
1950 {
1951 	struct trie *t = (struct trie *)tb->tb_data;
1952 	struct key_vector *pn = t->kv;
1953 	unsigned long cindex = 1;
1954 	struct fib_alias *fa;
1955 
1956 	for (;;) {
1957 		struct key_vector *n;
1958 
1959 		if (!(cindex--)) {
1960 			t_key pkey = pn->key;
1961 
1962 			if (IS_TRIE(pn))
1963 				break;
1964 
1965 			pn = node_parent(pn);
1966 			cindex = get_index(pkey, pn);
1967 			continue;
1968 		}
1969 
1970 		/* grab the next available node */
1971 		n = get_child(pn, cindex);
1972 		if (!n)
1973 			continue;
1974 
1975 		if (IS_TNODE(n)) {
1976 			/* record pn and cindex for leaf walking */
1977 			pn = n;
1978 			cindex = 1ul << n->bits;
1979 
1980 			continue;
1981 		}
1982 
1983 		hlist_for_each_entry(fa, &n->leaf, fa_list) {
1984 			struct fib_info *fi = fa->fa_info;
1985 
1986 			if (!fi || !fi->nh_updated || fa->tb_id != tb->tb_id)
1987 				continue;
1988 
1989 			rtmsg_fib(RTM_NEWROUTE, htonl(n->key), fa,
1990 				  KEYLENGTH - fa->fa_slen, tb->tb_id,
1991 				  info, NLM_F_REPLACE);
1992 
1993 			/* call_fib_entry_notifiers will be removed when
1994 			 * in-kernel notifier is implemented and supported
1995 			 * for nexthop objects
1996 			 */
1997 			call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_REPLACE,
1998 						 n->key,
1999 						 KEYLENGTH - fa->fa_slen, fa,
2000 						 NULL);
2001 		}
2002 	}
2003 }
2004 
2005 void fib_info_notify_update(struct net *net, struct nl_info *info)
2006 {
2007 	unsigned int h;
2008 
2009 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2010 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2011 		struct fib_table *tb;
2012 
2013 		hlist_for_each_entry_rcu(tb, head, tb_hlist)
2014 			__fib_info_notify_update(net, tb, info);
2015 	}
2016 }
2017 
2018 static void fib_leaf_notify(struct net *net, struct key_vector *l,
2019 			    struct fib_table *tb, struct notifier_block *nb)
2020 {
2021 	struct fib_alias *fa;
2022 
2023 	hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2024 		struct fib_info *fi = fa->fa_info;
2025 
2026 		if (!fi)
2027 			continue;
2028 
2029 		/* local and main table can share the same trie,
2030 		 * so don't notify twice for the same entry.
2031 		 */
2032 		if (tb->tb_id != fa->tb_id)
2033 			continue;
2034 
2035 		call_fib_entry_notifier(nb, net, FIB_EVENT_ENTRY_ADD, l->key,
2036 					KEYLENGTH - fa->fa_slen, fa);
2037 	}
2038 }
2039 
2040 static void fib_table_notify(struct net *net, struct fib_table *tb,
2041 			     struct notifier_block *nb)
2042 {
2043 	struct trie *t = (struct trie *)tb->tb_data;
2044 	struct key_vector *l, *tp = t->kv;
2045 	t_key key = 0;
2046 
2047 	while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2048 		fib_leaf_notify(net, l, tb, nb);
2049 
2050 		key = l->key + 1;
2051 		/* stop in case of wrap around */
2052 		if (key < l->key)
2053 			break;
2054 	}
2055 }
2056 
2057 void fib_notify(struct net *net, struct notifier_block *nb)
2058 {
2059 	unsigned int h;
2060 
2061 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2062 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2063 		struct fib_table *tb;
2064 
2065 		hlist_for_each_entry_rcu(tb, head, tb_hlist)
2066 			fib_table_notify(net, tb, nb);
2067 	}
2068 }
2069 
2070 static void __trie_free_rcu(struct rcu_head *head)
2071 {
2072 	struct fib_table *tb = container_of(head, struct fib_table, rcu);
2073 #ifdef CONFIG_IP_FIB_TRIE_STATS
2074 	struct trie *t = (struct trie *)tb->tb_data;
2075 
2076 	if (tb->tb_data == tb->__data)
2077 		free_percpu(t->stats);
2078 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2079 	kfree(tb);
2080 }
2081 
2082 void fib_free_table(struct fib_table *tb)
2083 {
2084 	call_rcu(&tb->rcu, __trie_free_rcu);
2085 }
2086 
2087 static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
2088 			     struct sk_buff *skb, struct netlink_callback *cb,
2089 			     struct fib_dump_filter *filter)
2090 {
2091 	unsigned int flags = NLM_F_MULTI;
2092 	__be32 xkey = htonl(l->key);
2093 	int i, s_i, i_fa, s_fa, err;
2094 	struct fib_alias *fa;
2095 
2096 	if (filter->filter_set ||
2097 	    !filter->dump_exceptions || !filter->dump_routes)
2098 		flags |= NLM_F_DUMP_FILTERED;
2099 
2100 	s_i = cb->args[4];
2101 	s_fa = cb->args[5];
2102 	i = 0;
2103 
2104 	/* rcu_read_lock is hold by caller */
2105 	hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2106 		struct fib_info *fi = fa->fa_info;
2107 
2108 		if (i < s_i)
2109 			goto next;
2110 
2111 		i_fa = 0;
2112 
2113 		if (tb->tb_id != fa->tb_id)
2114 			goto next;
2115 
2116 		if (filter->filter_set) {
2117 			if (filter->rt_type && fa->fa_type != filter->rt_type)
2118 				goto next;
2119 
2120 			if ((filter->protocol &&
2121 			     fi->fib_protocol != filter->protocol))
2122 				goto next;
2123 
2124 			if (filter->dev &&
2125 			    !fib_info_nh_uses_dev(fi, filter->dev))
2126 				goto next;
2127 		}
2128 
2129 		if (filter->dump_routes) {
2130 			if (!s_fa) {
2131 				err = fib_dump_info(skb,
2132 						    NETLINK_CB(cb->skb).portid,
2133 						    cb->nlh->nlmsg_seq,
2134 						    RTM_NEWROUTE,
2135 						    tb->tb_id, fa->fa_type,
2136 						    xkey,
2137 						    KEYLENGTH - fa->fa_slen,
2138 						    fa->fa_tos, fi, flags);
2139 				if (err < 0)
2140 					goto stop;
2141 			}
2142 
2143 			i_fa++;
2144 		}
2145 
2146 		if (filter->dump_exceptions) {
2147 			err = fib_dump_info_fnhe(skb, cb, tb->tb_id, fi,
2148 						 &i_fa, s_fa);
2149 			if (err < 0)
2150 				goto stop;
2151 		}
2152 
2153 next:
2154 		i++;
2155 	}
2156 
2157 	cb->args[4] = i;
2158 	return skb->len;
2159 
2160 stop:
2161 	cb->args[4] = i;
2162 	cb->args[5] = i_fa;
2163 	return err;
2164 }
2165 
2166 /* rcu_read_lock needs to be hold by caller from readside */
2167 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
2168 		   struct netlink_callback *cb, struct fib_dump_filter *filter)
2169 {
2170 	struct trie *t = (struct trie *)tb->tb_data;
2171 	struct key_vector *l, *tp = t->kv;
2172 	/* Dump starting at last key.
2173 	 * Note: 0.0.0.0/0 (ie default) is first key.
2174 	 */
2175 	int count = cb->args[2];
2176 	t_key key = cb->args[3];
2177 
2178 	while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2179 		int err;
2180 
2181 		err = fn_trie_dump_leaf(l, tb, skb, cb, filter);
2182 		if (err < 0) {
2183 			cb->args[3] = key;
2184 			cb->args[2] = count;
2185 			return err;
2186 		}
2187 
2188 		++count;
2189 		key = l->key + 1;
2190 
2191 		memset(&cb->args[4], 0,
2192 		       sizeof(cb->args) - 4*sizeof(cb->args[0]));
2193 
2194 		/* stop loop if key wrapped back to 0 */
2195 		if (key < l->key)
2196 			break;
2197 	}
2198 
2199 	cb->args[3] = key;
2200 	cb->args[2] = count;
2201 
2202 	return skb->len;
2203 }
2204 
2205 void __init fib_trie_init(void)
2206 {
2207 	fn_alias_kmem = kmem_cache_create("ip_fib_alias",
2208 					  sizeof(struct fib_alias),
2209 					  0, SLAB_PANIC, NULL);
2210 
2211 	trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
2212 					   LEAF_SIZE,
2213 					   0, SLAB_PANIC, NULL);
2214 }
2215 
2216 struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
2217 {
2218 	struct fib_table *tb;
2219 	struct trie *t;
2220 	size_t sz = sizeof(*tb);
2221 
2222 	if (!alias)
2223 		sz += sizeof(struct trie);
2224 
2225 	tb = kzalloc(sz, GFP_KERNEL);
2226 	if (!tb)
2227 		return NULL;
2228 
2229 	tb->tb_id = id;
2230 	tb->tb_num_default = 0;
2231 	tb->tb_data = (alias ? alias->__data : tb->__data);
2232 
2233 	if (alias)
2234 		return tb;
2235 
2236 	t = (struct trie *) tb->tb_data;
2237 	t->kv[0].pos = KEYLENGTH;
2238 	t->kv[0].slen = KEYLENGTH;
2239 #ifdef CONFIG_IP_FIB_TRIE_STATS
2240 	t->stats = alloc_percpu(struct trie_use_stats);
2241 	if (!t->stats) {
2242 		kfree(tb);
2243 		tb = NULL;
2244 	}
2245 #endif
2246 
2247 	return tb;
2248 }
2249 
2250 #ifdef CONFIG_PROC_FS
2251 /* Depth first Trie walk iterator */
2252 struct fib_trie_iter {
2253 	struct seq_net_private p;
2254 	struct fib_table *tb;
2255 	struct key_vector *tnode;
2256 	unsigned int index;
2257 	unsigned int depth;
2258 };
2259 
2260 static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2261 {
2262 	unsigned long cindex = iter->index;
2263 	struct key_vector *pn = iter->tnode;
2264 	t_key pkey;
2265 
2266 	pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2267 		 iter->tnode, iter->index, iter->depth);
2268 
2269 	while (!IS_TRIE(pn)) {
2270 		while (cindex < child_length(pn)) {
2271 			struct key_vector *n = get_child_rcu(pn, cindex++);
2272 
2273 			if (!n)
2274 				continue;
2275 
2276 			if (IS_LEAF(n)) {
2277 				iter->tnode = pn;
2278 				iter->index = cindex;
2279 			} else {
2280 				/* push down one level */
2281 				iter->tnode = n;
2282 				iter->index = 0;
2283 				++iter->depth;
2284 			}
2285 
2286 			return n;
2287 		}
2288 
2289 		/* Current node exhausted, pop back up */
2290 		pkey = pn->key;
2291 		pn = node_parent_rcu(pn);
2292 		cindex = get_index(pkey, pn) + 1;
2293 		--iter->depth;
2294 	}
2295 
2296 	/* record root node so further searches know we are done */
2297 	iter->tnode = pn;
2298 	iter->index = 0;
2299 
2300 	return NULL;
2301 }
2302 
2303 static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2304 					     struct trie *t)
2305 {
2306 	struct key_vector *n, *pn;
2307 
2308 	if (!t)
2309 		return NULL;
2310 
2311 	pn = t->kv;
2312 	n = rcu_dereference(pn->tnode[0]);
2313 	if (!n)
2314 		return NULL;
2315 
2316 	if (IS_TNODE(n)) {
2317 		iter->tnode = n;
2318 		iter->index = 0;
2319 		iter->depth = 1;
2320 	} else {
2321 		iter->tnode = pn;
2322 		iter->index = 0;
2323 		iter->depth = 0;
2324 	}
2325 
2326 	return n;
2327 }
2328 
2329 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2330 {
2331 	struct key_vector *n;
2332 	struct fib_trie_iter iter;
2333 
2334 	memset(s, 0, sizeof(*s));
2335 
2336 	rcu_read_lock();
2337 	for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2338 		if (IS_LEAF(n)) {
2339 			struct fib_alias *fa;
2340 
2341 			s->leaves++;
2342 			s->totdepth += iter.depth;
2343 			if (iter.depth > s->maxdepth)
2344 				s->maxdepth = iter.depth;
2345 
2346 			hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2347 				++s->prefixes;
2348 		} else {
2349 			s->tnodes++;
2350 			if (n->bits < MAX_STAT_DEPTH)
2351 				s->nodesizes[n->bits]++;
2352 			s->nullpointers += tn_info(n)->empty_children;
2353 		}
2354 	}
2355 	rcu_read_unlock();
2356 }
2357 
2358 /*
2359  *	This outputs /proc/net/fib_triestats
2360  */
2361 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2362 {
2363 	unsigned int i, max, pointers, bytes, avdepth;
2364 
2365 	if (stat->leaves)
2366 		avdepth = stat->totdepth*100 / stat->leaves;
2367 	else
2368 		avdepth = 0;
2369 
2370 	seq_printf(seq, "\tAver depth:     %u.%02d\n",
2371 		   avdepth / 100, avdepth % 100);
2372 	seq_printf(seq, "\tMax depth:      %u\n", stat->maxdepth);
2373 
2374 	seq_printf(seq, "\tLeaves:         %u\n", stat->leaves);
2375 	bytes = LEAF_SIZE * stat->leaves;
2376 
2377 	seq_printf(seq, "\tPrefixes:       %u\n", stat->prefixes);
2378 	bytes += sizeof(struct fib_alias) * stat->prefixes;
2379 
2380 	seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2381 	bytes += TNODE_SIZE(0) * stat->tnodes;
2382 
2383 	max = MAX_STAT_DEPTH;
2384 	while (max > 0 && stat->nodesizes[max-1] == 0)
2385 		max--;
2386 
2387 	pointers = 0;
2388 	for (i = 1; i < max; i++)
2389 		if (stat->nodesizes[i] != 0) {
2390 			seq_printf(seq, "  %u: %u",  i, stat->nodesizes[i]);
2391 			pointers += (1<<i) * stat->nodesizes[i];
2392 		}
2393 	seq_putc(seq, '\n');
2394 	seq_printf(seq, "\tPointers: %u\n", pointers);
2395 
2396 	bytes += sizeof(struct key_vector *) * pointers;
2397 	seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2398 	seq_printf(seq, "Total size: %u  kB\n", (bytes + 1023) / 1024);
2399 }
2400 
2401 #ifdef CONFIG_IP_FIB_TRIE_STATS
2402 static void trie_show_usage(struct seq_file *seq,
2403 			    const struct trie_use_stats __percpu *stats)
2404 {
2405 	struct trie_use_stats s = { 0 };
2406 	int cpu;
2407 
2408 	/* loop through all of the CPUs and gather up the stats */
2409 	for_each_possible_cpu(cpu) {
2410 		const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2411 
2412 		s.gets += pcpu->gets;
2413 		s.backtrack += pcpu->backtrack;
2414 		s.semantic_match_passed += pcpu->semantic_match_passed;
2415 		s.semantic_match_miss += pcpu->semantic_match_miss;
2416 		s.null_node_hit += pcpu->null_node_hit;
2417 		s.resize_node_skipped += pcpu->resize_node_skipped;
2418 	}
2419 
2420 	seq_printf(seq, "\nCounters:\n---------\n");
2421 	seq_printf(seq, "gets = %u\n", s.gets);
2422 	seq_printf(seq, "backtracks = %u\n", s.backtrack);
2423 	seq_printf(seq, "semantic match passed = %u\n",
2424 		   s.semantic_match_passed);
2425 	seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2426 	seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2427 	seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2428 }
2429 #endif /*  CONFIG_IP_FIB_TRIE_STATS */
2430 
2431 static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2432 {
2433 	if (tb->tb_id == RT_TABLE_LOCAL)
2434 		seq_puts(seq, "Local:\n");
2435 	else if (tb->tb_id == RT_TABLE_MAIN)
2436 		seq_puts(seq, "Main:\n");
2437 	else
2438 		seq_printf(seq, "Id %d:\n", tb->tb_id);
2439 }
2440 
2441 
2442 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2443 {
2444 	struct net *net = (struct net *)seq->private;
2445 	unsigned int h;
2446 
2447 	seq_printf(seq,
2448 		   "Basic info: size of leaf:"
2449 		   " %zd bytes, size of tnode: %zd bytes.\n",
2450 		   LEAF_SIZE, TNODE_SIZE(0));
2451 
2452 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2453 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2454 		struct fib_table *tb;
2455 
2456 		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2457 			struct trie *t = (struct trie *) tb->tb_data;
2458 			struct trie_stat stat;
2459 
2460 			if (!t)
2461 				continue;
2462 
2463 			fib_table_print(seq, tb);
2464 
2465 			trie_collect_stats(t, &stat);
2466 			trie_show_stats(seq, &stat);
2467 #ifdef CONFIG_IP_FIB_TRIE_STATS
2468 			trie_show_usage(seq, t->stats);
2469 #endif
2470 		}
2471 	}
2472 
2473 	return 0;
2474 }
2475 
2476 static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2477 {
2478 	struct fib_trie_iter *iter = seq->private;
2479 	struct net *net = seq_file_net(seq);
2480 	loff_t idx = 0;
2481 	unsigned int h;
2482 
2483 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2484 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2485 		struct fib_table *tb;
2486 
2487 		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2488 			struct key_vector *n;
2489 
2490 			for (n = fib_trie_get_first(iter,
2491 						    (struct trie *) tb->tb_data);
2492 			     n; n = fib_trie_get_next(iter))
2493 				if (pos == idx++) {
2494 					iter->tb = tb;
2495 					return n;
2496 				}
2497 		}
2498 	}
2499 
2500 	return NULL;
2501 }
2502 
2503 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2504 	__acquires(RCU)
2505 {
2506 	rcu_read_lock();
2507 	return fib_trie_get_idx(seq, *pos);
2508 }
2509 
2510 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2511 {
2512 	struct fib_trie_iter *iter = seq->private;
2513 	struct net *net = seq_file_net(seq);
2514 	struct fib_table *tb = iter->tb;
2515 	struct hlist_node *tb_node;
2516 	unsigned int h;
2517 	struct key_vector *n;
2518 
2519 	++*pos;
2520 	/* next node in same table */
2521 	n = fib_trie_get_next(iter);
2522 	if (n)
2523 		return n;
2524 
2525 	/* walk rest of this hash chain */
2526 	h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2527 	while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2528 		tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2529 		n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2530 		if (n)
2531 			goto found;
2532 	}
2533 
2534 	/* new hash chain */
2535 	while (++h < FIB_TABLE_HASHSZ) {
2536 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2537 		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2538 			n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2539 			if (n)
2540 				goto found;
2541 		}
2542 	}
2543 	return NULL;
2544 
2545 found:
2546 	iter->tb = tb;
2547 	return n;
2548 }
2549 
2550 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2551 	__releases(RCU)
2552 {
2553 	rcu_read_unlock();
2554 }
2555 
2556 static void seq_indent(struct seq_file *seq, int n)
2557 {
2558 	while (n-- > 0)
2559 		seq_puts(seq, "   ");
2560 }
2561 
2562 static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2563 {
2564 	switch (s) {
2565 	case RT_SCOPE_UNIVERSE: return "universe";
2566 	case RT_SCOPE_SITE:	return "site";
2567 	case RT_SCOPE_LINK:	return "link";
2568 	case RT_SCOPE_HOST:	return "host";
2569 	case RT_SCOPE_NOWHERE:	return "nowhere";
2570 	default:
2571 		snprintf(buf, len, "scope=%d", s);
2572 		return buf;
2573 	}
2574 }
2575 
2576 static const char *const rtn_type_names[__RTN_MAX] = {
2577 	[RTN_UNSPEC] = "UNSPEC",
2578 	[RTN_UNICAST] = "UNICAST",
2579 	[RTN_LOCAL] = "LOCAL",
2580 	[RTN_BROADCAST] = "BROADCAST",
2581 	[RTN_ANYCAST] = "ANYCAST",
2582 	[RTN_MULTICAST] = "MULTICAST",
2583 	[RTN_BLACKHOLE] = "BLACKHOLE",
2584 	[RTN_UNREACHABLE] = "UNREACHABLE",
2585 	[RTN_PROHIBIT] = "PROHIBIT",
2586 	[RTN_THROW] = "THROW",
2587 	[RTN_NAT] = "NAT",
2588 	[RTN_XRESOLVE] = "XRESOLVE",
2589 };
2590 
2591 static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2592 {
2593 	if (t < __RTN_MAX && rtn_type_names[t])
2594 		return rtn_type_names[t];
2595 	snprintf(buf, len, "type %u", t);
2596 	return buf;
2597 }
2598 
2599 /* Pretty print the trie */
2600 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2601 {
2602 	const struct fib_trie_iter *iter = seq->private;
2603 	struct key_vector *n = v;
2604 
2605 	if (IS_TRIE(node_parent_rcu(n)))
2606 		fib_table_print(seq, iter->tb);
2607 
2608 	if (IS_TNODE(n)) {
2609 		__be32 prf = htonl(n->key);
2610 
2611 		seq_indent(seq, iter->depth-1);
2612 		seq_printf(seq, "  +-- %pI4/%zu %u %u %u\n",
2613 			   &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2614 			   tn_info(n)->full_children,
2615 			   tn_info(n)->empty_children);
2616 	} else {
2617 		__be32 val = htonl(n->key);
2618 		struct fib_alias *fa;
2619 
2620 		seq_indent(seq, iter->depth);
2621 		seq_printf(seq, "  |-- %pI4\n", &val);
2622 
2623 		hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2624 			char buf1[32], buf2[32];
2625 
2626 			seq_indent(seq, iter->depth + 1);
2627 			seq_printf(seq, "  /%zu %s %s",
2628 				   KEYLENGTH - fa->fa_slen,
2629 				   rtn_scope(buf1, sizeof(buf1),
2630 					     fa->fa_info->fib_scope),
2631 				   rtn_type(buf2, sizeof(buf2),
2632 					    fa->fa_type));
2633 			if (fa->fa_tos)
2634 				seq_printf(seq, " tos=%d", fa->fa_tos);
2635 			seq_putc(seq, '\n');
2636 		}
2637 	}
2638 
2639 	return 0;
2640 }
2641 
2642 static const struct seq_operations fib_trie_seq_ops = {
2643 	.start  = fib_trie_seq_start,
2644 	.next   = fib_trie_seq_next,
2645 	.stop   = fib_trie_seq_stop,
2646 	.show   = fib_trie_seq_show,
2647 };
2648 
2649 struct fib_route_iter {
2650 	struct seq_net_private p;
2651 	struct fib_table *main_tb;
2652 	struct key_vector *tnode;
2653 	loff_t	pos;
2654 	t_key	key;
2655 };
2656 
2657 static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2658 					    loff_t pos)
2659 {
2660 	struct key_vector *l, **tp = &iter->tnode;
2661 	t_key key;
2662 
2663 	/* use cached location of previously found key */
2664 	if (iter->pos > 0 && pos >= iter->pos) {
2665 		key = iter->key;
2666 	} else {
2667 		iter->pos = 1;
2668 		key = 0;
2669 	}
2670 
2671 	pos -= iter->pos;
2672 
2673 	while ((l = leaf_walk_rcu(tp, key)) && (pos-- > 0)) {
2674 		key = l->key + 1;
2675 		iter->pos++;
2676 		l = NULL;
2677 
2678 		/* handle unlikely case of a key wrap */
2679 		if (!key)
2680 			break;
2681 	}
2682 
2683 	if (l)
2684 		iter->key = l->key;	/* remember it */
2685 	else
2686 		iter->pos = 0;		/* forget it */
2687 
2688 	return l;
2689 }
2690 
2691 static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2692 	__acquires(RCU)
2693 {
2694 	struct fib_route_iter *iter = seq->private;
2695 	struct fib_table *tb;
2696 	struct trie *t;
2697 
2698 	rcu_read_lock();
2699 
2700 	tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2701 	if (!tb)
2702 		return NULL;
2703 
2704 	iter->main_tb = tb;
2705 	t = (struct trie *)tb->tb_data;
2706 	iter->tnode = t->kv;
2707 
2708 	if (*pos != 0)
2709 		return fib_route_get_idx(iter, *pos);
2710 
2711 	iter->pos = 0;
2712 	iter->key = KEY_MAX;
2713 
2714 	return SEQ_START_TOKEN;
2715 }
2716 
2717 static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2718 {
2719 	struct fib_route_iter *iter = seq->private;
2720 	struct key_vector *l = NULL;
2721 	t_key key = iter->key + 1;
2722 
2723 	++*pos;
2724 
2725 	/* only allow key of 0 for start of sequence */
2726 	if ((v == SEQ_START_TOKEN) || key)
2727 		l = leaf_walk_rcu(&iter->tnode, key);
2728 
2729 	if (l) {
2730 		iter->key = l->key;
2731 		iter->pos++;
2732 	} else {
2733 		iter->pos = 0;
2734 	}
2735 
2736 	return l;
2737 }
2738 
2739 static void fib_route_seq_stop(struct seq_file *seq, void *v)
2740 	__releases(RCU)
2741 {
2742 	rcu_read_unlock();
2743 }
2744 
2745 static unsigned int fib_flag_trans(int type, __be32 mask, struct fib_info *fi)
2746 {
2747 	unsigned int flags = 0;
2748 
2749 	if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2750 		flags = RTF_REJECT;
2751 	if (fi) {
2752 		const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2753 
2754 		if (nhc->nhc_gw.ipv4)
2755 			flags |= RTF_GATEWAY;
2756 	}
2757 	if (mask == htonl(0xFFFFFFFF))
2758 		flags |= RTF_HOST;
2759 	flags |= RTF_UP;
2760 	return flags;
2761 }
2762 
2763 /*
2764  *	This outputs /proc/net/route.
2765  *	The format of the file is not supposed to be changed
2766  *	and needs to be same as fib_hash output to avoid breaking
2767  *	legacy utilities
2768  */
2769 static int fib_route_seq_show(struct seq_file *seq, void *v)
2770 {
2771 	struct fib_route_iter *iter = seq->private;
2772 	struct fib_table *tb = iter->main_tb;
2773 	struct fib_alias *fa;
2774 	struct key_vector *l = v;
2775 	__be32 prefix;
2776 
2777 	if (v == SEQ_START_TOKEN) {
2778 		seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2779 			   "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2780 			   "\tWindow\tIRTT");
2781 		return 0;
2782 	}
2783 
2784 	prefix = htonl(l->key);
2785 
2786 	hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2787 		struct fib_info *fi = fa->fa_info;
2788 		__be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2789 		unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2790 
2791 		if ((fa->fa_type == RTN_BROADCAST) ||
2792 		    (fa->fa_type == RTN_MULTICAST))
2793 			continue;
2794 
2795 		if (fa->tb_id != tb->tb_id)
2796 			continue;
2797 
2798 		seq_setwidth(seq, 127);
2799 
2800 		if (fi) {
2801 			struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2802 			__be32 gw = 0;
2803 
2804 			if (nhc->nhc_gw_family == AF_INET)
2805 				gw = nhc->nhc_gw.ipv4;
2806 
2807 			seq_printf(seq,
2808 				   "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2809 				   "%d\t%08X\t%d\t%u\t%u",
2810 				   nhc->nhc_dev ? nhc->nhc_dev->name : "*",
2811 				   prefix, gw, flags, 0, 0,
2812 				   fi->fib_priority,
2813 				   mask,
2814 				   (fi->fib_advmss ?
2815 				    fi->fib_advmss + 40 : 0),
2816 				   fi->fib_window,
2817 				   fi->fib_rtt >> 3);
2818 		} else {
2819 			seq_printf(seq,
2820 				   "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2821 				   "%d\t%08X\t%d\t%u\t%u",
2822 				   prefix, 0, flags, 0, 0, 0,
2823 				   mask, 0, 0, 0);
2824 		}
2825 		seq_pad(seq, '\n');
2826 	}
2827 
2828 	return 0;
2829 }
2830 
2831 static const struct seq_operations fib_route_seq_ops = {
2832 	.start  = fib_route_seq_start,
2833 	.next   = fib_route_seq_next,
2834 	.stop   = fib_route_seq_stop,
2835 	.show   = fib_route_seq_show,
2836 };
2837 
2838 int __net_init fib_proc_init(struct net *net)
2839 {
2840 	if (!proc_create_net("fib_trie", 0444, net->proc_net, &fib_trie_seq_ops,
2841 			sizeof(struct fib_trie_iter)))
2842 		goto out1;
2843 
2844 	if (!proc_create_net_single("fib_triestat", 0444, net->proc_net,
2845 			fib_triestat_seq_show, NULL))
2846 		goto out2;
2847 
2848 	if (!proc_create_net("route", 0444, net->proc_net, &fib_route_seq_ops,
2849 			sizeof(struct fib_route_iter)))
2850 		goto out3;
2851 
2852 	return 0;
2853 
2854 out3:
2855 	remove_proc_entry("fib_triestat", net->proc_net);
2856 out2:
2857 	remove_proc_entry("fib_trie", net->proc_net);
2858 out1:
2859 	return -ENOMEM;
2860 }
2861 
2862 void __net_exit fib_proc_exit(struct net *net)
2863 {
2864 	remove_proc_entry("fib_trie", net->proc_net);
2865 	remove_proc_entry("fib_triestat", net->proc_net);
2866 	remove_proc_entry("route", net->proc_net);
2867 }
2868 
2869 #endif /* CONFIG_PROC_FS */
2870