xref: /openbmc/linux/net/ipv4/fib_trie.c (revision 11a163f2)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *
4  *   Robert Olsson <robert.olsson@its.uu.se> Uppsala Universitet
5  *     & Swedish University of Agricultural Sciences.
6  *
7  *   Jens Laas <jens.laas@data.slu.se> Swedish University of
8  *     Agricultural Sciences.
9  *
10  *   Hans Liss <hans.liss@its.uu.se>  Uppsala Universitet
11  *
12  * This work is based on the LPC-trie which is originally described in:
13  *
14  * An experimental study of compression methods for dynamic tries
15  * Stefan Nilsson and Matti Tikkanen. Algorithmica, 33(1):19-33, 2002.
16  * https://www.csc.kth.se/~snilsson/software/dyntrie2/
17  *
18  * IP-address lookup using LC-tries. Stefan Nilsson and Gunnar Karlsson
19  * IEEE Journal on Selected Areas in Communications, 17(6):1083-1092, June 1999
20  *
21  * Code from fib_hash has been reused which includes the following header:
22  *
23  * INET		An implementation of the TCP/IP protocol suite for the LINUX
24  *		operating system.  INET is implemented using the  BSD Socket
25  *		interface as the means of communication with the user level.
26  *
27  *		IPv4 FIB: lookup engine and maintenance routines.
28  *
29  * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
30  *
31  * Substantial contributions to this work comes from:
32  *
33  *		David S. Miller, <davem@davemloft.net>
34  *		Stephen Hemminger <shemminger@osdl.org>
35  *		Paul E. McKenney <paulmck@us.ibm.com>
36  *		Patrick McHardy <kaber@trash.net>
37  */
38 #include <linux/cache.h>
39 #include <linux/uaccess.h>
40 #include <linux/bitops.h>
41 #include <linux/types.h>
42 #include <linux/kernel.h>
43 #include <linux/mm.h>
44 #include <linux/string.h>
45 #include <linux/socket.h>
46 #include <linux/sockios.h>
47 #include <linux/errno.h>
48 #include <linux/in.h>
49 #include <linux/inet.h>
50 #include <linux/inetdevice.h>
51 #include <linux/netdevice.h>
52 #include <linux/if_arp.h>
53 #include <linux/proc_fs.h>
54 #include <linux/rcupdate.h>
55 #include <linux/skbuff.h>
56 #include <linux/netlink.h>
57 #include <linux/init.h>
58 #include <linux/list.h>
59 #include <linux/slab.h>
60 #include <linux/export.h>
61 #include <linux/vmalloc.h>
62 #include <linux/notifier.h>
63 #include <net/net_namespace.h>
64 #include <net/ip.h>
65 #include <net/protocol.h>
66 #include <net/route.h>
67 #include <net/tcp.h>
68 #include <net/sock.h>
69 #include <net/ip_fib.h>
70 #include <net/fib_notifier.h>
71 #include <trace/events/fib.h>
72 #include "fib_lookup.h"
73 
74 static int call_fib_entry_notifier(struct notifier_block *nb,
75 				   enum fib_event_type event_type, u32 dst,
76 				   int dst_len, struct fib_alias *fa,
77 				   struct netlink_ext_ack *extack)
78 {
79 	struct fib_entry_notifier_info info = {
80 		.info.extack = extack,
81 		.dst = dst,
82 		.dst_len = dst_len,
83 		.fi = fa->fa_info,
84 		.tos = fa->fa_tos,
85 		.type = fa->fa_type,
86 		.tb_id = fa->tb_id,
87 	};
88 	return call_fib4_notifier(nb, event_type, &info.info);
89 }
90 
91 static int call_fib_entry_notifiers(struct net *net,
92 				    enum fib_event_type event_type, u32 dst,
93 				    int dst_len, struct fib_alias *fa,
94 				    struct netlink_ext_ack *extack)
95 {
96 	struct fib_entry_notifier_info info = {
97 		.info.extack = extack,
98 		.dst = dst,
99 		.dst_len = dst_len,
100 		.fi = fa->fa_info,
101 		.tos = fa->fa_tos,
102 		.type = fa->fa_type,
103 		.tb_id = fa->tb_id,
104 	};
105 	return call_fib4_notifiers(net, event_type, &info.info);
106 }
107 
108 #define MAX_STAT_DEPTH 32
109 
110 #define KEYLENGTH	(8*sizeof(t_key))
111 #define KEY_MAX		((t_key)~0)
112 
113 typedef unsigned int t_key;
114 
115 #define IS_TRIE(n)	((n)->pos >= KEYLENGTH)
116 #define IS_TNODE(n)	((n)->bits)
117 #define IS_LEAF(n)	(!(n)->bits)
118 
119 struct key_vector {
120 	t_key key;
121 	unsigned char pos;		/* 2log(KEYLENGTH) bits needed */
122 	unsigned char bits;		/* 2log(KEYLENGTH) bits needed */
123 	unsigned char slen;
124 	union {
125 		/* This list pointer if valid if (pos | bits) == 0 (LEAF) */
126 		struct hlist_head leaf;
127 		/* This array is valid if (pos | bits) > 0 (TNODE) */
128 		struct key_vector __rcu *tnode[0];
129 	};
130 };
131 
132 struct tnode {
133 	struct rcu_head rcu;
134 	t_key empty_children;		/* KEYLENGTH bits needed */
135 	t_key full_children;		/* KEYLENGTH bits needed */
136 	struct key_vector __rcu *parent;
137 	struct key_vector kv[1];
138 #define tn_bits kv[0].bits
139 };
140 
141 #define TNODE_SIZE(n)	offsetof(struct tnode, kv[0].tnode[n])
142 #define LEAF_SIZE	TNODE_SIZE(1)
143 
144 #ifdef CONFIG_IP_FIB_TRIE_STATS
145 struct trie_use_stats {
146 	unsigned int gets;
147 	unsigned int backtrack;
148 	unsigned int semantic_match_passed;
149 	unsigned int semantic_match_miss;
150 	unsigned int null_node_hit;
151 	unsigned int resize_node_skipped;
152 };
153 #endif
154 
155 struct trie_stat {
156 	unsigned int totdepth;
157 	unsigned int maxdepth;
158 	unsigned int tnodes;
159 	unsigned int leaves;
160 	unsigned int nullpointers;
161 	unsigned int prefixes;
162 	unsigned int nodesizes[MAX_STAT_DEPTH];
163 };
164 
165 struct trie {
166 	struct key_vector kv[1];
167 #ifdef CONFIG_IP_FIB_TRIE_STATS
168 	struct trie_use_stats __percpu *stats;
169 #endif
170 };
171 
172 static struct key_vector *resize(struct trie *t, struct key_vector *tn);
173 static unsigned int tnode_free_size;
174 
175 /*
176  * synchronize_rcu after call_rcu for outstanding dirty memory; it should be
177  * especially useful before resizing the root node with PREEMPT_NONE configs;
178  * the value was obtained experimentally, aiming to avoid visible slowdown.
179  */
180 unsigned int sysctl_fib_sync_mem = 512 * 1024;
181 unsigned int sysctl_fib_sync_mem_min = 64 * 1024;
182 unsigned int sysctl_fib_sync_mem_max = 64 * 1024 * 1024;
183 
184 static struct kmem_cache *fn_alias_kmem __ro_after_init;
185 static struct kmem_cache *trie_leaf_kmem __ro_after_init;
186 
187 static inline struct tnode *tn_info(struct key_vector *kv)
188 {
189 	return container_of(kv, struct tnode, kv[0]);
190 }
191 
192 /* caller must hold RTNL */
193 #define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
194 #define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
195 
196 /* caller must hold RCU read lock or RTNL */
197 #define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
198 #define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
199 
200 /* wrapper for rcu_assign_pointer */
201 static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
202 {
203 	if (n)
204 		rcu_assign_pointer(tn_info(n)->parent, tp);
205 }
206 
207 #define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
208 
209 /* This provides us with the number of children in this node, in the case of a
210  * leaf this will return 0 meaning none of the children are accessible.
211  */
212 static inline unsigned long child_length(const struct key_vector *tn)
213 {
214 	return (1ul << tn->bits) & ~(1ul);
215 }
216 
217 #define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
218 
219 static inline unsigned long get_index(t_key key, struct key_vector *kv)
220 {
221 	unsigned long index = key ^ kv->key;
222 
223 	if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
224 		return 0;
225 
226 	return index >> kv->pos;
227 }
228 
229 /* To understand this stuff, an understanding of keys and all their bits is
230  * necessary. Every node in the trie has a key associated with it, but not
231  * all of the bits in that key are significant.
232  *
233  * Consider a node 'n' and its parent 'tp'.
234  *
235  * If n is a leaf, every bit in its key is significant. Its presence is
236  * necessitated by path compression, since during a tree traversal (when
237  * searching for a leaf - unless we are doing an insertion) we will completely
238  * ignore all skipped bits we encounter. Thus we need to verify, at the end of
239  * a potentially successful search, that we have indeed been walking the
240  * correct key path.
241  *
242  * Note that we can never "miss" the correct key in the tree if present by
243  * following the wrong path. Path compression ensures that segments of the key
244  * that are the same for all keys with a given prefix are skipped, but the
245  * skipped part *is* identical for each node in the subtrie below the skipped
246  * bit! trie_insert() in this implementation takes care of that.
247  *
248  * if n is an internal node - a 'tnode' here, the various parts of its key
249  * have many different meanings.
250  *
251  * Example:
252  * _________________________________________________________________
253  * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
254  * -----------------------------------------------------------------
255  *  31  30  29  28  27  26  25  24  23  22  21  20  19  18  17  16
256  *
257  * _________________________________________________________________
258  * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
259  * -----------------------------------------------------------------
260  *  15  14  13  12  11  10   9   8   7   6   5   4   3   2   1   0
261  *
262  * tp->pos = 22
263  * tp->bits = 3
264  * n->pos = 13
265  * n->bits = 4
266  *
267  * First, let's just ignore the bits that come before the parent tp, that is
268  * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
269  * point we do not use them for anything.
270  *
271  * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
272  * index into the parent's child array. That is, they will be used to find
273  * 'n' among tp's children.
274  *
275  * The bits from (n->pos + n->bits) to (tp->pos - 1) - "S" - are skipped bits
276  * for the node n.
277  *
278  * All the bits we have seen so far are significant to the node n. The rest
279  * of the bits are really not needed or indeed known in n->key.
280  *
281  * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
282  * n's child array, and will of course be different for each child.
283  *
284  * The rest of the bits, from 0 to (n->pos -1) - "u" - are completely unknown
285  * at this point.
286  */
287 
288 static const int halve_threshold = 25;
289 static const int inflate_threshold = 50;
290 static const int halve_threshold_root = 15;
291 static const int inflate_threshold_root = 30;
292 
293 static void __alias_free_mem(struct rcu_head *head)
294 {
295 	struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
296 	kmem_cache_free(fn_alias_kmem, fa);
297 }
298 
299 static inline void alias_free_mem_rcu(struct fib_alias *fa)
300 {
301 	call_rcu(&fa->rcu, __alias_free_mem);
302 }
303 
304 #define TNODE_VMALLOC_MAX \
305 	ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
306 
307 static void __node_free_rcu(struct rcu_head *head)
308 {
309 	struct tnode *n = container_of(head, struct tnode, rcu);
310 
311 	if (!n->tn_bits)
312 		kmem_cache_free(trie_leaf_kmem, n);
313 	else
314 		kvfree(n);
315 }
316 
317 #define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
318 
319 static struct tnode *tnode_alloc(int bits)
320 {
321 	size_t size;
322 
323 	/* verify bits is within bounds */
324 	if (bits > TNODE_VMALLOC_MAX)
325 		return NULL;
326 
327 	/* determine size and verify it is non-zero and didn't overflow */
328 	size = TNODE_SIZE(1ul << bits);
329 
330 	if (size <= PAGE_SIZE)
331 		return kzalloc(size, GFP_KERNEL);
332 	else
333 		return vzalloc(size);
334 }
335 
336 static inline void empty_child_inc(struct key_vector *n)
337 {
338 	tn_info(n)->empty_children++;
339 
340 	if (!tn_info(n)->empty_children)
341 		tn_info(n)->full_children++;
342 }
343 
344 static inline void empty_child_dec(struct key_vector *n)
345 {
346 	if (!tn_info(n)->empty_children)
347 		tn_info(n)->full_children--;
348 
349 	tn_info(n)->empty_children--;
350 }
351 
352 static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
353 {
354 	struct key_vector *l;
355 	struct tnode *kv;
356 
357 	kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
358 	if (!kv)
359 		return NULL;
360 
361 	/* initialize key vector */
362 	l = kv->kv;
363 	l->key = key;
364 	l->pos = 0;
365 	l->bits = 0;
366 	l->slen = fa->fa_slen;
367 
368 	/* link leaf to fib alias */
369 	INIT_HLIST_HEAD(&l->leaf);
370 	hlist_add_head(&fa->fa_list, &l->leaf);
371 
372 	return l;
373 }
374 
375 static struct key_vector *tnode_new(t_key key, int pos, int bits)
376 {
377 	unsigned int shift = pos + bits;
378 	struct key_vector *tn;
379 	struct tnode *tnode;
380 
381 	/* verify bits and pos their msb bits clear and values are valid */
382 	BUG_ON(!bits || (shift > KEYLENGTH));
383 
384 	tnode = tnode_alloc(bits);
385 	if (!tnode)
386 		return NULL;
387 
388 	pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
389 		 sizeof(struct key_vector *) << bits);
390 
391 	if (bits == KEYLENGTH)
392 		tnode->full_children = 1;
393 	else
394 		tnode->empty_children = 1ul << bits;
395 
396 	tn = tnode->kv;
397 	tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
398 	tn->pos = pos;
399 	tn->bits = bits;
400 	tn->slen = pos;
401 
402 	return tn;
403 }
404 
405 /* Check whether a tnode 'n' is "full", i.e. it is an internal node
406  * and no bits are skipped. See discussion in dyntree paper p. 6
407  */
408 static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
409 {
410 	return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
411 }
412 
413 /* Add a child at position i overwriting the old value.
414  * Update the value of full_children and empty_children.
415  */
416 static void put_child(struct key_vector *tn, unsigned long i,
417 		      struct key_vector *n)
418 {
419 	struct key_vector *chi = get_child(tn, i);
420 	int isfull, wasfull;
421 
422 	BUG_ON(i >= child_length(tn));
423 
424 	/* update emptyChildren, overflow into fullChildren */
425 	if (!n && chi)
426 		empty_child_inc(tn);
427 	if (n && !chi)
428 		empty_child_dec(tn);
429 
430 	/* update fullChildren */
431 	wasfull = tnode_full(tn, chi);
432 	isfull = tnode_full(tn, n);
433 
434 	if (wasfull && !isfull)
435 		tn_info(tn)->full_children--;
436 	else if (!wasfull && isfull)
437 		tn_info(tn)->full_children++;
438 
439 	if (n && (tn->slen < n->slen))
440 		tn->slen = n->slen;
441 
442 	rcu_assign_pointer(tn->tnode[i], n);
443 }
444 
445 static void update_children(struct key_vector *tn)
446 {
447 	unsigned long i;
448 
449 	/* update all of the child parent pointers */
450 	for (i = child_length(tn); i;) {
451 		struct key_vector *inode = get_child(tn, --i);
452 
453 		if (!inode)
454 			continue;
455 
456 		/* Either update the children of a tnode that
457 		 * already belongs to us or update the child
458 		 * to point to ourselves.
459 		 */
460 		if (node_parent(inode) == tn)
461 			update_children(inode);
462 		else
463 			node_set_parent(inode, tn);
464 	}
465 }
466 
467 static inline void put_child_root(struct key_vector *tp, t_key key,
468 				  struct key_vector *n)
469 {
470 	if (IS_TRIE(tp))
471 		rcu_assign_pointer(tp->tnode[0], n);
472 	else
473 		put_child(tp, get_index(key, tp), n);
474 }
475 
476 static inline void tnode_free_init(struct key_vector *tn)
477 {
478 	tn_info(tn)->rcu.next = NULL;
479 }
480 
481 static inline void tnode_free_append(struct key_vector *tn,
482 				     struct key_vector *n)
483 {
484 	tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
485 	tn_info(tn)->rcu.next = &tn_info(n)->rcu;
486 }
487 
488 static void tnode_free(struct key_vector *tn)
489 {
490 	struct callback_head *head = &tn_info(tn)->rcu;
491 
492 	while (head) {
493 		head = head->next;
494 		tnode_free_size += TNODE_SIZE(1ul << tn->bits);
495 		node_free(tn);
496 
497 		tn = container_of(head, struct tnode, rcu)->kv;
498 	}
499 
500 	if (tnode_free_size >= sysctl_fib_sync_mem) {
501 		tnode_free_size = 0;
502 		synchronize_rcu();
503 	}
504 }
505 
506 static struct key_vector *replace(struct trie *t,
507 				  struct key_vector *oldtnode,
508 				  struct key_vector *tn)
509 {
510 	struct key_vector *tp = node_parent(oldtnode);
511 	unsigned long i;
512 
513 	/* setup the parent pointer out of and back into this node */
514 	NODE_INIT_PARENT(tn, tp);
515 	put_child_root(tp, tn->key, tn);
516 
517 	/* update all of the child parent pointers */
518 	update_children(tn);
519 
520 	/* all pointers should be clean so we are done */
521 	tnode_free(oldtnode);
522 
523 	/* resize children now that oldtnode is freed */
524 	for (i = child_length(tn); i;) {
525 		struct key_vector *inode = get_child(tn, --i);
526 
527 		/* resize child node */
528 		if (tnode_full(tn, inode))
529 			tn = resize(t, inode);
530 	}
531 
532 	return tp;
533 }
534 
535 static struct key_vector *inflate(struct trie *t,
536 				  struct key_vector *oldtnode)
537 {
538 	struct key_vector *tn;
539 	unsigned long i;
540 	t_key m;
541 
542 	pr_debug("In inflate\n");
543 
544 	tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
545 	if (!tn)
546 		goto notnode;
547 
548 	/* prepare oldtnode to be freed */
549 	tnode_free_init(oldtnode);
550 
551 	/* Assemble all of the pointers in our cluster, in this case that
552 	 * represents all of the pointers out of our allocated nodes that
553 	 * point to existing tnodes and the links between our allocated
554 	 * nodes.
555 	 */
556 	for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
557 		struct key_vector *inode = get_child(oldtnode, --i);
558 		struct key_vector *node0, *node1;
559 		unsigned long j, k;
560 
561 		/* An empty child */
562 		if (!inode)
563 			continue;
564 
565 		/* A leaf or an internal node with skipped bits */
566 		if (!tnode_full(oldtnode, inode)) {
567 			put_child(tn, get_index(inode->key, tn), inode);
568 			continue;
569 		}
570 
571 		/* drop the node in the old tnode free list */
572 		tnode_free_append(oldtnode, inode);
573 
574 		/* An internal node with two children */
575 		if (inode->bits == 1) {
576 			put_child(tn, 2 * i + 1, get_child(inode, 1));
577 			put_child(tn, 2 * i, get_child(inode, 0));
578 			continue;
579 		}
580 
581 		/* We will replace this node 'inode' with two new
582 		 * ones, 'node0' and 'node1', each with half of the
583 		 * original children. The two new nodes will have
584 		 * a position one bit further down the key and this
585 		 * means that the "significant" part of their keys
586 		 * (see the discussion near the top of this file)
587 		 * will differ by one bit, which will be "0" in
588 		 * node0's key and "1" in node1's key. Since we are
589 		 * moving the key position by one step, the bit that
590 		 * we are moving away from - the bit at position
591 		 * (tn->pos) - is the one that will differ between
592 		 * node0 and node1. So... we synthesize that bit in the
593 		 * two new keys.
594 		 */
595 		node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
596 		if (!node1)
597 			goto nomem;
598 		node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
599 
600 		tnode_free_append(tn, node1);
601 		if (!node0)
602 			goto nomem;
603 		tnode_free_append(tn, node0);
604 
605 		/* populate child pointers in new nodes */
606 		for (k = child_length(inode), j = k / 2; j;) {
607 			put_child(node1, --j, get_child(inode, --k));
608 			put_child(node0, j, get_child(inode, j));
609 			put_child(node1, --j, get_child(inode, --k));
610 			put_child(node0, j, get_child(inode, j));
611 		}
612 
613 		/* link new nodes to parent */
614 		NODE_INIT_PARENT(node1, tn);
615 		NODE_INIT_PARENT(node0, tn);
616 
617 		/* link parent to nodes */
618 		put_child(tn, 2 * i + 1, node1);
619 		put_child(tn, 2 * i, node0);
620 	}
621 
622 	/* setup the parent pointers into and out of this node */
623 	return replace(t, oldtnode, tn);
624 nomem:
625 	/* all pointers should be clean so we are done */
626 	tnode_free(tn);
627 notnode:
628 	return NULL;
629 }
630 
631 static struct key_vector *halve(struct trie *t,
632 				struct key_vector *oldtnode)
633 {
634 	struct key_vector *tn;
635 	unsigned long i;
636 
637 	pr_debug("In halve\n");
638 
639 	tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
640 	if (!tn)
641 		goto notnode;
642 
643 	/* prepare oldtnode to be freed */
644 	tnode_free_init(oldtnode);
645 
646 	/* Assemble all of the pointers in our cluster, in this case that
647 	 * represents all of the pointers out of our allocated nodes that
648 	 * point to existing tnodes and the links between our allocated
649 	 * nodes.
650 	 */
651 	for (i = child_length(oldtnode); i;) {
652 		struct key_vector *node1 = get_child(oldtnode, --i);
653 		struct key_vector *node0 = get_child(oldtnode, --i);
654 		struct key_vector *inode;
655 
656 		/* At least one of the children is empty */
657 		if (!node1 || !node0) {
658 			put_child(tn, i / 2, node1 ? : node0);
659 			continue;
660 		}
661 
662 		/* Two nonempty children */
663 		inode = tnode_new(node0->key, oldtnode->pos, 1);
664 		if (!inode)
665 			goto nomem;
666 		tnode_free_append(tn, inode);
667 
668 		/* initialize pointers out of node */
669 		put_child(inode, 1, node1);
670 		put_child(inode, 0, node0);
671 		NODE_INIT_PARENT(inode, tn);
672 
673 		/* link parent to node */
674 		put_child(tn, i / 2, inode);
675 	}
676 
677 	/* setup the parent pointers into and out of this node */
678 	return replace(t, oldtnode, tn);
679 nomem:
680 	/* all pointers should be clean so we are done */
681 	tnode_free(tn);
682 notnode:
683 	return NULL;
684 }
685 
686 static struct key_vector *collapse(struct trie *t,
687 				   struct key_vector *oldtnode)
688 {
689 	struct key_vector *n, *tp;
690 	unsigned long i;
691 
692 	/* scan the tnode looking for that one child that might still exist */
693 	for (n = NULL, i = child_length(oldtnode); !n && i;)
694 		n = get_child(oldtnode, --i);
695 
696 	/* compress one level */
697 	tp = node_parent(oldtnode);
698 	put_child_root(tp, oldtnode->key, n);
699 	node_set_parent(n, tp);
700 
701 	/* drop dead node */
702 	node_free(oldtnode);
703 
704 	return tp;
705 }
706 
707 static unsigned char update_suffix(struct key_vector *tn)
708 {
709 	unsigned char slen = tn->pos;
710 	unsigned long stride, i;
711 	unsigned char slen_max;
712 
713 	/* only vector 0 can have a suffix length greater than or equal to
714 	 * tn->pos + tn->bits, the second highest node will have a suffix
715 	 * length at most of tn->pos + tn->bits - 1
716 	 */
717 	slen_max = min_t(unsigned char, tn->pos + tn->bits - 1, tn->slen);
718 
719 	/* search though the list of children looking for nodes that might
720 	 * have a suffix greater than the one we currently have.  This is
721 	 * why we start with a stride of 2 since a stride of 1 would
722 	 * represent the nodes with suffix length equal to tn->pos
723 	 */
724 	for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
725 		struct key_vector *n = get_child(tn, i);
726 
727 		if (!n || (n->slen <= slen))
728 			continue;
729 
730 		/* update stride and slen based on new value */
731 		stride <<= (n->slen - slen);
732 		slen = n->slen;
733 		i &= ~(stride - 1);
734 
735 		/* stop searching if we have hit the maximum possible value */
736 		if (slen >= slen_max)
737 			break;
738 	}
739 
740 	tn->slen = slen;
741 
742 	return slen;
743 }
744 
745 /* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
746  * the Helsinki University of Technology and Matti Tikkanen of Nokia
747  * Telecommunications, page 6:
748  * "A node is doubled if the ratio of non-empty children to all
749  * children in the *doubled* node is at least 'high'."
750  *
751  * 'high' in this instance is the variable 'inflate_threshold'. It
752  * is expressed as a percentage, so we multiply it with
753  * child_length() and instead of multiplying by 2 (since the
754  * child array will be doubled by inflate()) and multiplying
755  * the left-hand side by 100 (to handle the percentage thing) we
756  * multiply the left-hand side by 50.
757  *
758  * The left-hand side may look a bit weird: child_length(tn)
759  * - tn->empty_children is of course the number of non-null children
760  * in the current node. tn->full_children is the number of "full"
761  * children, that is non-null tnodes with a skip value of 0.
762  * All of those will be doubled in the resulting inflated tnode, so
763  * we just count them one extra time here.
764  *
765  * A clearer way to write this would be:
766  *
767  * to_be_doubled = tn->full_children;
768  * not_to_be_doubled = child_length(tn) - tn->empty_children -
769  *     tn->full_children;
770  *
771  * new_child_length = child_length(tn) * 2;
772  *
773  * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
774  *      new_child_length;
775  * if (new_fill_factor >= inflate_threshold)
776  *
777  * ...and so on, tho it would mess up the while () loop.
778  *
779  * anyway,
780  * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
781  *      inflate_threshold
782  *
783  * avoid a division:
784  * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
785  *      inflate_threshold * new_child_length
786  *
787  * expand not_to_be_doubled and to_be_doubled, and shorten:
788  * 100 * (child_length(tn) - tn->empty_children +
789  *    tn->full_children) >= inflate_threshold * new_child_length
790  *
791  * expand new_child_length:
792  * 100 * (child_length(tn) - tn->empty_children +
793  *    tn->full_children) >=
794  *      inflate_threshold * child_length(tn) * 2
795  *
796  * shorten again:
797  * 50 * (tn->full_children + child_length(tn) -
798  *    tn->empty_children) >= inflate_threshold *
799  *    child_length(tn)
800  *
801  */
802 static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
803 {
804 	unsigned long used = child_length(tn);
805 	unsigned long threshold = used;
806 
807 	/* Keep root node larger */
808 	threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
809 	used -= tn_info(tn)->empty_children;
810 	used += tn_info(tn)->full_children;
811 
812 	/* if bits == KEYLENGTH then pos = 0, and will fail below */
813 
814 	return (used > 1) && tn->pos && ((50 * used) >= threshold);
815 }
816 
817 static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
818 {
819 	unsigned long used = child_length(tn);
820 	unsigned long threshold = used;
821 
822 	/* Keep root node larger */
823 	threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
824 	used -= tn_info(tn)->empty_children;
825 
826 	/* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
827 
828 	return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
829 }
830 
831 static inline bool should_collapse(struct key_vector *tn)
832 {
833 	unsigned long used = child_length(tn);
834 
835 	used -= tn_info(tn)->empty_children;
836 
837 	/* account for bits == KEYLENGTH case */
838 	if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
839 		used -= KEY_MAX;
840 
841 	/* One child or none, time to drop us from the trie */
842 	return used < 2;
843 }
844 
845 #define MAX_WORK 10
846 static struct key_vector *resize(struct trie *t, struct key_vector *tn)
847 {
848 #ifdef CONFIG_IP_FIB_TRIE_STATS
849 	struct trie_use_stats __percpu *stats = t->stats;
850 #endif
851 	struct key_vector *tp = node_parent(tn);
852 	unsigned long cindex = get_index(tn->key, tp);
853 	int max_work = MAX_WORK;
854 
855 	pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
856 		 tn, inflate_threshold, halve_threshold);
857 
858 	/* track the tnode via the pointer from the parent instead of
859 	 * doing it ourselves.  This way we can let RCU fully do its
860 	 * thing without us interfering
861 	 */
862 	BUG_ON(tn != get_child(tp, cindex));
863 
864 	/* Double as long as the resulting node has a number of
865 	 * nonempty nodes that are above the threshold.
866 	 */
867 	while (should_inflate(tp, tn) && max_work) {
868 		tp = inflate(t, tn);
869 		if (!tp) {
870 #ifdef CONFIG_IP_FIB_TRIE_STATS
871 			this_cpu_inc(stats->resize_node_skipped);
872 #endif
873 			break;
874 		}
875 
876 		max_work--;
877 		tn = get_child(tp, cindex);
878 	}
879 
880 	/* update parent in case inflate failed */
881 	tp = node_parent(tn);
882 
883 	/* Return if at least one inflate is run */
884 	if (max_work != MAX_WORK)
885 		return tp;
886 
887 	/* Halve as long as the number of empty children in this
888 	 * node is above threshold.
889 	 */
890 	while (should_halve(tp, tn) && max_work) {
891 		tp = halve(t, tn);
892 		if (!tp) {
893 #ifdef CONFIG_IP_FIB_TRIE_STATS
894 			this_cpu_inc(stats->resize_node_skipped);
895 #endif
896 			break;
897 		}
898 
899 		max_work--;
900 		tn = get_child(tp, cindex);
901 	}
902 
903 	/* Only one child remains */
904 	if (should_collapse(tn))
905 		return collapse(t, tn);
906 
907 	/* update parent in case halve failed */
908 	return node_parent(tn);
909 }
910 
911 static void node_pull_suffix(struct key_vector *tn, unsigned char slen)
912 {
913 	unsigned char node_slen = tn->slen;
914 
915 	while ((node_slen > tn->pos) && (node_slen > slen)) {
916 		slen = update_suffix(tn);
917 		if (node_slen == slen)
918 			break;
919 
920 		tn = node_parent(tn);
921 		node_slen = tn->slen;
922 	}
923 }
924 
925 static void node_push_suffix(struct key_vector *tn, unsigned char slen)
926 {
927 	while (tn->slen < slen) {
928 		tn->slen = slen;
929 		tn = node_parent(tn);
930 	}
931 }
932 
933 /* rcu_read_lock needs to be hold by caller from readside */
934 static struct key_vector *fib_find_node(struct trie *t,
935 					struct key_vector **tp, u32 key)
936 {
937 	struct key_vector *pn, *n = t->kv;
938 	unsigned long index = 0;
939 
940 	do {
941 		pn = n;
942 		n = get_child_rcu(n, index);
943 
944 		if (!n)
945 			break;
946 
947 		index = get_cindex(key, n);
948 
949 		/* This bit of code is a bit tricky but it combines multiple
950 		 * checks into a single check.  The prefix consists of the
951 		 * prefix plus zeros for the bits in the cindex. The index
952 		 * is the difference between the key and this value.  From
953 		 * this we can actually derive several pieces of data.
954 		 *   if (index >= (1ul << bits))
955 		 *     we have a mismatch in skip bits and failed
956 		 *   else
957 		 *     we know the value is cindex
958 		 *
959 		 * This check is safe even if bits == KEYLENGTH due to the
960 		 * fact that we can only allocate a node with 32 bits if a
961 		 * long is greater than 32 bits.
962 		 */
963 		if (index >= (1ul << n->bits)) {
964 			n = NULL;
965 			break;
966 		}
967 
968 		/* keep searching until we find a perfect match leaf or NULL */
969 	} while (IS_TNODE(n));
970 
971 	*tp = pn;
972 
973 	return n;
974 }
975 
976 /* Return the first fib alias matching TOS with
977  * priority less than or equal to PRIO.
978  * If 'find_first' is set, return the first matching
979  * fib alias, regardless of TOS and priority.
980  */
981 static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
982 					u8 tos, u32 prio, u32 tb_id,
983 					bool find_first)
984 {
985 	struct fib_alias *fa;
986 
987 	if (!fah)
988 		return NULL;
989 
990 	hlist_for_each_entry(fa, fah, fa_list) {
991 		if (fa->fa_slen < slen)
992 			continue;
993 		if (fa->fa_slen != slen)
994 			break;
995 		if (fa->tb_id > tb_id)
996 			continue;
997 		if (fa->tb_id != tb_id)
998 			break;
999 		if (find_first)
1000 			return fa;
1001 		if (fa->fa_tos > tos)
1002 			continue;
1003 		if (fa->fa_info->fib_priority >= prio || fa->fa_tos < tos)
1004 			return fa;
1005 	}
1006 
1007 	return NULL;
1008 }
1009 
1010 static struct fib_alias *
1011 fib_find_matching_alias(struct net *net, const struct fib_rt_info *fri)
1012 {
1013 	u8 slen = KEYLENGTH - fri->dst_len;
1014 	struct key_vector *l, *tp;
1015 	struct fib_table *tb;
1016 	struct fib_alias *fa;
1017 	struct trie *t;
1018 
1019 	tb = fib_get_table(net, fri->tb_id);
1020 	if (!tb)
1021 		return NULL;
1022 
1023 	t = (struct trie *)tb->tb_data;
1024 	l = fib_find_node(t, &tp, be32_to_cpu(fri->dst));
1025 	if (!l)
1026 		return NULL;
1027 
1028 	hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1029 		if (fa->fa_slen == slen && fa->tb_id == fri->tb_id &&
1030 		    fa->fa_tos == fri->tos && fa->fa_info == fri->fi &&
1031 		    fa->fa_type == fri->type)
1032 			return fa;
1033 	}
1034 
1035 	return NULL;
1036 }
1037 
1038 void fib_alias_hw_flags_set(struct net *net, const struct fib_rt_info *fri)
1039 {
1040 	struct fib_alias *fa_match;
1041 
1042 	rcu_read_lock();
1043 
1044 	fa_match = fib_find_matching_alias(net, fri);
1045 	if (!fa_match)
1046 		goto out;
1047 
1048 	fa_match->offload = fri->offload;
1049 	fa_match->trap = fri->trap;
1050 
1051 out:
1052 	rcu_read_unlock();
1053 }
1054 EXPORT_SYMBOL_GPL(fib_alias_hw_flags_set);
1055 
1056 static void trie_rebalance(struct trie *t, struct key_vector *tn)
1057 {
1058 	while (!IS_TRIE(tn))
1059 		tn = resize(t, tn);
1060 }
1061 
1062 static int fib_insert_node(struct trie *t, struct key_vector *tp,
1063 			   struct fib_alias *new, t_key key)
1064 {
1065 	struct key_vector *n, *l;
1066 
1067 	l = leaf_new(key, new);
1068 	if (!l)
1069 		goto noleaf;
1070 
1071 	/* retrieve child from parent node */
1072 	n = get_child(tp, get_index(key, tp));
1073 
1074 	/* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1075 	 *
1076 	 *  Add a new tnode here
1077 	 *  first tnode need some special handling
1078 	 *  leaves us in position for handling as case 3
1079 	 */
1080 	if (n) {
1081 		struct key_vector *tn;
1082 
1083 		tn = tnode_new(key, __fls(key ^ n->key), 1);
1084 		if (!tn)
1085 			goto notnode;
1086 
1087 		/* initialize routes out of node */
1088 		NODE_INIT_PARENT(tn, tp);
1089 		put_child(tn, get_index(key, tn) ^ 1, n);
1090 
1091 		/* start adding routes into the node */
1092 		put_child_root(tp, key, tn);
1093 		node_set_parent(n, tn);
1094 
1095 		/* parent now has a NULL spot where the leaf can go */
1096 		tp = tn;
1097 	}
1098 
1099 	/* Case 3: n is NULL, and will just insert a new leaf */
1100 	node_push_suffix(tp, new->fa_slen);
1101 	NODE_INIT_PARENT(l, tp);
1102 	put_child_root(tp, key, l);
1103 	trie_rebalance(t, tp);
1104 
1105 	return 0;
1106 notnode:
1107 	node_free(l);
1108 noleaf:
1109 	return -ENOMEM;
1110 }
1111 
1112 static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1113 			    struct key_vector *l, struct fib_alias *new,
1114 			    struct fib_alias *fa, t_key key)
1115 {
1116 	if (!l)
1117 		return fib_insert_node(t, tp, new, key);
1118 
1119 	if (fa) {
1120 		hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1121 	} else {
1122 		struct fib_alias *last;
1123 
1124 		hlist_for_each_entry(last, &l->leaf, fa_list) {
1125 			if (new->fa_slen < last->fa_slen)
1126 				break;
1127 			if ((new->fa_slen == last->fa_slen) &&
1128 			    (new->tb_id > last->tb_id))
1129 				break;
1130 			fa = last;
1131 		}
1132 
1133 		if (fa)
1134 			hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1135 		else
1136 			hlist_add_head_rcu(&new->fa_list, &l->leaf);
1137 	}
1138 
1139 	/* if we added to the tail node then we need to update slen */
1140 	if (l->slen < new->fa_slen) {
1141 		l->slen = new->fa_slen;
1142 		node_push_suffix(tp, new->fa_slen);
1143 	}
1144 
1145 	return 0;
1146 }
1147 
1148 static bool fib_valid_key_len(u32 key, u8 plen, struct netlink_ext_ack *extack)
1149 {
1150 	if (plen > KEYLENGTH) {
1151 		NL_SET_ERR_MSG(extack, "Invalid prefix length");
1152 		return false;
1153 	}
1154 
1155 	if ((plen < KEYLENGTH) && (key << plen)) {
1156 		NL_SET_ERR_MSG(extack,
1157 			       "Invalid prefix for given prefix length");
1158 		return false;
1159 	}
1160 
1161 	return true;
1162 }
1163 
1164 static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1165 			     struct key_vector *l, struct fib_alias *old);
1166 
1167 /* Caller must hold RTNL. */
1168 int fib_table_insert(struct net *net, struct fib_table *tb,
1169 		     struct fib_config *cfg, struct netlink_ext_ack *extack)
1170 {
1171 	struct trie *t = (struct trie *)tb->tb_data;
1172 	struct fib_alias *fa, *new_fa;
1173 	struct key_vector *l, *tp;
1174 	u16 nlflags = NLM_F_EXCL;
1175 	struct fib_info *fi;
1176 	u8 plen = cfg->fc_dst_len;
1177 	u8 slen = KEYLENGTH - plen;
1178 	u8 tos = cfg->fc_tos;
1179 	u32 key;
1180 	int err;
1181 
1182 	key = ntohl(cfg->fc_dst);
1183 
1184 	if (!fib_valid_key_len(key, plen, extack))
1185 		return -EINVAL;
1186 
1187 	pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1188 
1189 	fi = fib_create_info(cfg, extack);
1190 	if (IS_ERR(fi)) {
1191 		err = PTR_ERR(fi);
1192 		goto err;
1193 	}
1194 
1195 	l = fib_find_node(t, &tp, key);
1196 	fa = l ? fib_find_alias(&l->leaf, slen, tos, fi->fib_priority,
1197 				tb->tb_id, false) : NULL;
1198 
1199 	/* Now fa, if non-NULL, points to the first fib alias
1200 	 * with the same keys [prefix,tos,priority], if such key already
1201 	 * exists or to the node before which we will insert new one.
1202 	 *
1203 	 * If fa is NULL, we will need to allocate a new one and
1204 	 * insert to the tail of the section matching the suffix length
1205 	 * of the new alias.
1206 	 */
1207 
1208 	if (fa && fa->fa_tos == tos &&
1209 	    fa->fa_info->fib_priority == fi->fib_priority) {
1210 		struct fib_alias *fa_first, *fa_match;
1211 
1212 		err = -EEXIST;
1213 		if (cfg->fc_nlflags & NLM_F_EXCL)
1214 			goto out;
1215 
1216 		nlflags &= ~NLM_F_EXCL;
1217 
1218 		/* We have 2 goals:
1219 		 * 1. Find exact match for type, scope, fib_info to avoid
1220 		 * duplicate routes
1221 		 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1222 		 */
1223 		fa_match = NULL;
1224 		fa_first = fa;
1225 		hlist_for_each_entry_from(fa, fa_list) {
1226 			if ((fa->fa_slen != slen) ||
1227 			    (fa->tb_id != tb->tb_id) ||
1228 			    (fa->fa_tos != tos))
1229 				break;
1230 			if (fa->fa_info->fib_priority != fi->fib_priority)
1231 				break;
1232 			if (fa->fa_type == cfg->fc_type &&
1233 			    fa->fa_info == fi) {
1234 				fa_match = fa;
1235 				break;
1236 			}
1237 		}
1238 
1239 		if (cfg->fc_nlflags & NLM_F_REPLACE) {
1240 			struct fib_info *fi_drop;
1241 			u8 state;
1242 
1243 			nlflags |= NLM_F_REPLACE;
1244 			fa = fa_first;
1245 			if (fa_match) {
1246 				if (fa == fa_match)
1247 					err = 0;
1248 				goto out;
1249 			}
1250 			err = -ENOBUFS;
1251 			new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1252 			if (!new_fa)
1253 				goto out;
1254 
1255 			fi_drop = fa->fa_info;
1256 			new_fa->fa_tos = fa->fa_tos;
1257 			new_fa->fa_info = fi;
1258 			new_fa->fa_type = cfg->fc_type;
1259 			state = fa->fa_state;
1260 			new_fa->fa_state = state & ~FA_S_ACCESSED;
1261 			new_fa->fa_slen = fa->fa_slen;
1262 			new_fa->tb_id = tb->tb_id;
1263 			new_fa->fa_default = -1;
1264 			new_fa->offload = 0;
1265 			new_fa->trap = 0;
1266 
1267 			hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1268 
1269 			if (fib_find_alias(&l->leaf, fa->fa_slen, 0, 0,
1270 					   tb->tb_id, true) == new_fa) {
1271 				enum fib_event_type fib_event;
1272 
1273 				fib_event = FIB_EVENT_ENTRY_REPLACE;
1274 				err = call_fib_entry_notifiers(net, fib_event,
1275 							       key, plen,
1276 							       new_fa, extack);
1277 				if (err) {
1278 					hlist_replace_rcu(&new_fa->fa_list,
1279 							  &fa->fa_list);
1280 					goto out_free_new_fa;
1281 				}
1282 			}
1283 
1284 			rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1285 				  tb->tb_id, &cfg->fc_nlinfo, nlflags);
1286 
1287 			alias_free_mem_rcu(fa);
1288 
1289 			fib_release_info(fi_drop);
1290 			if (state & FA_S_ACCESSED)
1291 				rt_cache_flush(cfg->fc_nlinfo.nl_net);
1292 
1293 			goto succeeded;
1294 		}
1295 		/* Error if we find a perfect match which
1296 		 * uses the same scope, type, and nexthop
1297 		 * information.
1298 		 */
1299 		if (fa_match)
1300 			goto out;
1301 
1302 		if (cfg->fc_nlflags & NLM_F_APPEND)
1303 			nlflags |= NLM_F_APPEND;
1304 		else
1305 			fa = fa_first;
1306 	}
1307 	err = -ENOENT;
1308 	if (!(cfg->fc_nlflags & NLM_F_CREATE))
1309 		goto out;
1310 
1311 	nlflags |= NLM_F_CREATE;
1312 	err = -ENOBUFS;
1313 	new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1314 	if (!new_fa)
1315 		goto out;
1316 
1317 	new_fa->fa_info = fi;
1318 	new_fa->fa_tos = tos;
1319 	new_fa->fa_type = cfg->fc_type;
1320 	new_fa->fa_state = 0;
1321 	new_fa->fa_slen = slen;
1322 	new_fa->tb_id = tb->tb_id;
1323 	new_fa->fa_default = -1;
1324 	new_fa->offload = 0;
1325 	new_fa->trap = 0;
1326 
1327 	/* Insert new entry to the list. */
1328 	err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1329 	if (err)
1330 		goto out_free_new_fa;
1331 
1332 	/* The alias was already inserted, so the node must exist. */
1333 	l = l ? l : fib_find_node(t, &tp, key);
1334 	if (WARN_ON_ONCE(!l))
1335 		goto out_free_new_fa;
1336 
1337 	if (fib_find_alias(&l->leaf, new_fa->fa_slen, 0, 0, tb->tb_id, true) ==
1338 	    new_fa) {
1339 		enum fib_event_type fib_event;
1340 
1341 		fib_event = FIB_EVENT_ENTRY_REPLACE;
1342 		err = call_fib_entry_notifiers(net, fib_event, key, plen,
1343 					       new_fa, extack);
1344 		if (err)
1345 			goto out_remove_new_fa;
1346 	}
1347 
1348 	if (!plen)
1349 		tb->tb_num_default++;
1350 
1351 	rt_cache_flush(cfg->fc_nlinfo.nl_net);
1352 	rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1353 		  &cfg->fc_nlinfo, nlflags);
1354 succeeded:
1355 	return 0;
1356 
1357 out_remove_new_fa:
1358 	fib_remove_alias(t, tp, l, new_fa);
1359 out_free_new_fa:
1360 	kmem_cache_free(fn_alias_kmem, new_fa);
1361 out:
1362 	fib_release_info(fi);
1363 err:
1364 	return err;
1365 }
1366 
1367 static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1368 {
1369 	t_key prefix = n->key;
1370 
1371 	return (key ^ prefix) & (prefix | -prefix);
1372 }
1373 
1374 bool fib_lookup_good_nhc(const struct fib_nh_common *nhc, int fib_flags,
1375 			 const struct flowi4 *flp)
1376 {
1377 	if (nhc->nhc_flags & RTNH_F_DEAD)
1378 		return false;
1379 
1380 	if (ip_ignore_linkdown(nhc->nhc_dev) &&
1381 	    nhc->nhc_flags & RTNH_F_LINKDOWN &&
1382 	    !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1383 		return false;
1384 
1385 	if (!(flp->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF)) {
1386 		if (flp->flowi4_oif &&
1387 		    flp->flowi4_oif != nhc->nhc_oif)
1388 			return false;
1389 	}
1390 
1391 	return true;
1392 }
1393 
1394 /* should be called with rcu_read_lock */
1395 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1396 		     struct fib_result *res, int fib_flags)
1397 {
1398 	struct trie *t = (struct trie *) tb->tb_data;
1399 #ifdef CONFIG_IP_FIB_TRIE_STATS
1400 	struct trie_use_stats __percpu *stats = t->stats;
1401 #endif
1402 	const t_key key = ntohl(flp->daddr);
1403 	struct key_vector *n, *pn;
1404 	struct fib_alias *fa;
1405 	unsigned long index;
1406 	t_key cindex;
1407 
1408 	pn = t->kv;
1409 	cindex = 0;
1410 
1411 	n = get_child_rcu(pn, cindex);
1412 	if (!n) {
1413 		trace_fib_table_lookup(tb->tb_id, flp, NULL, -EAGAIN);
1414 		return -EAGAIN;
1415 	}
1416 
1417 #ifdef CONFIG_IP_FIB_TRIE_STATS
1418 	this_cpu_inc(stats->gets);
1419 #endif
1420 
1421 	/* Step 1: Travel to the longest prefix match in the trie */
1422 	for (;;) {
1423 		index = get_cindex(key, n);
1424 
1425 		/* This bit of code is a bit tricky but it combines multiple
1426 		 * checks into a single check.  The prefix consists of the
1427 		 * prefix plus zeros for the "bits" in the prefix. The index
1428 		 * is the difference between the key and this value.  From
1429 		 * this we can actually derive several pieces of data.
1430 		 *   if (index >= (1ul << bits))
1431 		 *     we have a mismatch in skip bits and failed
1432 		 *   else
1433 		 *     we know the value is cindex
1434 		 *
1435 		 * This check is safe even if bits == KEYLENGTH due to the
1436 		 * fact that we can only allocate a node with 32 bits if a
1437 		 * long is greater than 32 bits.
1438 		 */
1439 		if (index >= (1ul << n->bits))
1440 			break;
1441 
1442 		/* we have found a leaf. Prefixes have already been compared */
1443 		if (IS_LEAF(n))
1444 			goto found;
1445 
1446 		/* only record pn and cindex if we are going to be chopping
1447 		 * bits later.  Otherwise we are just wasting cycles.
1448 		 */
1449 		if (n->slen > n->pos) {
1450 			pn = n;
1451 			cindex = index;
1452 		}
1453 
1454 		n = get_child_rcu(n, index);
1455 		if (unlikely(!n))
1456 			goto backtrace;
1457 	}
1458 
1459 	/* Step 2: Sort out leaves and begin backtracing for longest prefix */
1460 	for (;;) {
1461 		/* record the pointer where our next node pointer is stored */
1462 		struct key_vector __rcu **cptr = n->tnode;
1463 
1464 		/* This test verifies that none of the bits that differ
1465 		 * between the key and the prefix exist in the region of
1466 		 * the lsb and higher in the prefix.
1467 		 */
1468 		if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1469 			goto backtrace;
1470 
1471 		/* exit out and process leaf */
1472 		if (unlikely(IS_LEAF(n)))
1473 			break;
1474 
1475 		/* Don't bother recording parent info.  Since we are in
1476 		 * prefix match mode we will have to come back to wherever
1477 		 * we started this traversal anyway
1478 		 */
1479 
1480 		while ((n = rcu_dereference(*cptr)) == NULL) {
1481 backtrace:
1482 #ifdef CONFIG_IP_FIB_TRIE_STATS
1483 			if (!n)
1484 				this_cpu_inc(stats->null_node_hit);
1485 #endif
1486 			/* If we are at cindex 0 there are no more bits for
1487 			 * us to strip at this level so we must ascend back
1488 			 * up one level to see if there are any more bits to
1489 			 * be stripped there.
1490 			 */
1491 			while (!cindex) {
1492 				t_key pkey = pn->key;
1493 
1494 				/* If we don't have a parent then there is
1495 				 * nothing for us to do as we do not have any
1496 				 * further nodes to parse.
1497 				 */
1498 				if (IS_TRIE(pn)) {
1499 					trace_fib_table_lookup(tb->tb_id, flp,
1500 							       NULL, -EAGAIN);
1501 					return -EAGAIN;
1502 				}
1503 #ifdef CONFIG_IP_FIB_TRIE_STATS
1504 				this_cpu_inc(stats->backtrack);
1505 #endif
1506 				/* Get Child's index */
1507 				pn = node_parent_rcu(pn);
1508 				cindex = get_index(pkey, pn);
1509 			}
1510 
1511 			/* strip the least significant bit from the cindex */
1512 			cindex &= cindex - 1;
1513 
1514 			/* grab pointer for next child node */
1515 			cptr = &pn->tnode[cindex];
1516 		}
1517 	}
1518 
1519 found:
1520 	/* this line carries forward the xor from earlier in the function */
1521 	index = key ^ n->key;
1522 
1523 	/* Step 3: Process the leaf, if that fails fall back to backtracing */
1524 	hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1525 		struct fib_info *fi = fa->fa_info;
1526 		struct fib_nh_common *nhc;
1527 		int nhsel, err;
1528 
1529 		if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1530 			if (index >= (1ul << fa->fa_slen))
1531 				continue;
1532 		}
1533 		if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1534 			continue;
1535 		if (fi->fib_dead)
1536 			continue;
1537 		if (fa->fa_info->fib_scope < flp->flowi4_scope)
1538 			continue;
1539 		fib_alias_accessed(fa);
1540 		err = fib_props[fa->fa_type].error;
1541 		if (unlikely(err < 0)) {
1542 out_reject:
1543 #ifdef CONFIG_IP_FIB_TRIE_STATS
1544 			this_cpu_inc(stats->semantic_match_passed);
1545 #endif
1546 			trace_fib_table_lookup(tb->tb_id, flp, NULL, err);
1547 			return err;
1548 		}
1549 		if (fi->fib_flags & RTNH_F_DEAD)
1550 			continue;
1551 
1552 		if (unlikely(fi->nh)) {
1553 			if (nexthop_is_blackhole(fi->nh)) {
1554 				err = fib_props[RTN_BLACKHOLE].error;
1555 				goto out_reject;
1556 			}
1557 
1558 			nhc = nexthop_get_nhc_lookup(fi->nh, fib_flags, flp,
1559 						     &nhsel);
1560 			if (nhc)
1561 				goto set_result;
1562 			goto miss;
1563 		}
1564 
1565 		for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
1566 			nhc = fib_info_nhc(fi, nhsel);
1567 
1568 			if (!fib_lookup_good_nhc(nhc, fib_flags, flp))
1569 				continue;
1570 set_result:
1571 			if (!(fib_flags & FIB_LOOKUP_NOREF))
1572 				refcount_inc(&fi->fib_clntref);
1573 
1574 			res->prefix = htonl(n->key);
1575 			res->prefixlen = KEYLENGTH - fa->fa_slen;
1576 			res->nh_sel = nhsel;
1577 			res->nhc = nhc;
1578 			res->type = fa->fa_type;
1579 			res->scope = fi->fib_scope;
1580 			res->fi = fi;
1581 			res->table = tb;
1582 			res->fa_head = &n->leaf;
1583 #ifdef CONFIG_IP_FIB_TRIE_STATS
1584 			this_cpu_inc(stats->semantic_match_passed);
1585 #endif
1586 			trace_fib_table_lookup(tb->tb_id, flp, nhc, err);
1587 
1588 			return err;
1589 		}
1590 	}
1591 miss:
1592 #ifdef CONFIG_IP_FIB_TRIE_STATS
1593 	this_cpu_inc(stats->semantic_match_miss);
1594 #endif
1595 	goto backtrace;
1596 }
1597 EXPORT_SYMBOL_GPL(fib_table_lookup);
1598 
1599 static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1600 			     struct key_vector *l, struct fib_alias *old)
1601 {
1602 	/* record the location of the previous list_info entry */
1603 	struct hlist_node **pprev = old->fa_list.pprev;
1604 	struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1605 
1606 	/* remove the fib_alias from the list */
1607 	hlist_del_rcu(&old->fa_list);
1608 
1609 	/* if we emptied the list this leaf will be freed and we can sort
1610 	 * out parent suffix lengths as a part of trie_rebalance
1611 	 */
1612 	if (hlist_empty(&l->leaf)) {
1613 		if (tp->slen == l->slen)
1614 			node_pull_suffix(tp, tp->pos);
1615 		put_child_root(tp, l->key, NULL);
1616 		node_free(l);
1617 		trie_rebalance(t, tp);
1618 		return;
1619 	}
1620 
1621 	/* only access fa if it is pointing at the last valid hlist_node */
1622 	if (*pprev)
1623 		return;
1624 
1625 	/* update the trie with the latest suffix length */
1626 	l->slen = fa->fa_slen;
1627 	node_pull_suffix(tp, fa->fa_slen);
1628 }
1629 
1630 static void fib_notify_alias_delete(struct net *net, u32 key,
1631 				    struct hlist_head *fah,
1632 				    struct fib_alias *fa_to_delete,
1633 				    struct netlink_ext_ack *extack)
1634 {
1635 	struct fib_alias *fa_next, *fa_to_notify;
1636 	u32 tb_id = fa_to_delete->tb_id;
1637 	u8 slen = fa_to_delete->fa_slen;
1638 	enum fib_event_type fib_event;
1639 
1640 	/* Do not notify if we do not care about the route. */
1641 	if (fib_find_alias(fah, slen, 0, 0, tb_id, true) != fa_to_delete)
1642 		return;
1643 
1644 	/* Determine if the route should be replaced by the next route in the
1645 	 * list.
1646 	 */
1647 	fa_next = hlist_entry_safe(fa_to_delete->fa_list.next,
1648 				   struct fib_alias, fa_list);
1649 	if (fa_next && fa_next->fa_slen == slen && fa_next->tb_id == tb_id) {
1650 		fib_event = FIB_EVENT_ENTRY_REPLACE;
1651 		fa_to_notify = fa_next;
1652 	} else {
1653 		fib_event = FIB_EVENT_ENTRY_DEL;
1654 		fa_to_notify = fa_to_delete;
1655 	}
1656 	call_fib_entry_notifiers(net, fib_event, key, KEYLENGTH - slen,
1657 				 fa_to_notify, extack);
1658 }
1659 
1660 /* Caller must hold RTNL. */
1661 int fib_table_delete(struct net *net, struct fib_table *tb,
1662 		     struct fib_config *cfg, struct netlink_ext_ack *extack)
1663 {
1664 	struct trie *t = (struct trie *) tb->tb_data;
1665 	struct fib_alias *fa, *fa_to_delete;
1666 	struct key_vector *l, *tp;
1667 	u8 plen = cfg->fc_dst_len;
1668 	u8 slen = KEYLENGTH - plen;
1669 	u8 tos = cfg->fc_tos;
1670 	u32 key;
1671 
1672 	key = ntohl(cfg->fc_dst);
1673 
1674 	if (!fib_valid_key_len(key, plen, extack))
1675 		return -EINVAL;
1676 
1677 	l = fib_find_node(t, &tp, key);
1678 	if (!l)
1679 		return -ESRCH;
1680 
1681 	fa = fib_find_alias(&l->leaf, slen, tos, 0, tb->tb_id, false);
1682 	if (!fa)
1683 		return -ESRCH;
1684 
1685 	pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1686 
1687 	fa_to_delete = NULL;
1688 	hlist_for_each_entry_from(fa, fa_list) {
1689 		struct fib_info *fi = fa->fa_info;
1690 
1691 		if ((fa->fa_slen != slen) ||
1692 		    (fa->tb_id != tb->tb_id) ||
1693 		    (fa->fa_tos != tos))
1694 			break;
1695 
1696 		if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1697 		    (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1698 		     fa->fa_info->fib_scope == cfg->fc_scope) &&
1699 		    (!cfg->fc_prefsrc ||
1700 		     fi->fib_prefsrc == cfg->fc_prefsrc) &&
1701 		    (!cfg->fc_protocol ||
1702 		     fi->fib_protocol == cfg->fc_protocol) &&
1703 		    fib_nh_match(net, cfg, fi, extack) == 0 &&
1704 		    fib_metrics_match(cfg, fi)) {
1705 			fa_to_delete = fa;
1706 			break;
1707 		}
1708 	}
1709 
1710 	if (!fa_to_delete)
1711 		return -ESRCH;
1712 
1713 	fib_notify_alias_delete(net, key, &l->leaf, fa_to_delete, extack);
1714 	rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1715 		  &cfg->fc_nlinfo, 0);
1716 
1717 	if (!plen)
1718 		tb->tb_num_default--;
1719 
1720 	fib_remove_alias(t, tp, l, fa_to_delete);
1721 
1722 	if (fa_to_delete->fa_state & FA_S_ACCESSED)
1723 		rt_cache_flush(cfg->fc_nlinfo.nl_net);
1724 
1725 	fib_release_info(fa_to_delete->fa_info);
1726 	alias_free_mem_rcu(fa_to_delete);
1727 	return 0;
1728 }
1729 
1730 /* Scan for the next leaf starting at the provided key value */
1731 static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1732 {
1733 	struct key_vector *pn, *n = *tn;
1734 	unsigned long cindex;
1735 
1736 	/* this loop is meant to try and find the key in the trie */
1737 	do {
1738 		/* record parent and next child index */
1739 		pn = n;
1740 		cindex = (key > pn->key) ? get_index(key, pn) : 0;
1741 
1742 		if (cindex >> pn->bits)
1743 			break;
1744 
1745 		/* descend into the next child */
1746 		n = get_child_rcu(pn, cindex++);
1747 		if (!n)
1748 			break;
1749 
1750 		/* guarantee forward progress on the keys */
1751 		if (IS_LEAF(n) && (n->key >= key))
1752 			goto found;
1753 	} while (IS_TNODE(n));
1754 
1755 	/* this loop will search for the next leaf with a greater key */
1756 	while (!IS_TRIE(pn)) {
1757 		/* if we exhausted the parent node we will need to climb */
1758 		if (cindex >= (1ul << pn->bits)) {
1759 			t_key pkey = pn->key;
1760 
1761 			pn = node_parent_rcu(pn);
1762 			cindex = get_index(pkey, pn) + 1;
1763 			continue;
1764 		}
1765 
1766 		/* grab the next available node */
1767 		n = get_child_rcu(pn, cindex++);
1768 		if (!n)
1769 			continue;
1770 
1771 		/* no need to compare keys since we bumped the index */
1772 		if (IS_LEAF(n))
1773 			goto found;
1774 
1775 		/* Rescan start scanning in new node */
1776 		pn = n;
1777 		cindex = 0;
1778 	}
1779 
1780 	*tn = pn;
1781 	return NULL; /* Root of trie */
1782 found:
1783 	/* if we are at the limit for keys just return NULL for the tnode */
1784 	*tn = pn;
1785 	return n;
1786 }
1787 
1788 static void fib_trie_free(struct fib_table *tb)
1789 {
1790 	struct trie *t = (struct trie *)tb->tb_data;
1791 	struct key_vector *pn = t->kv;
1792 	unsigned long cindex = 1;
1793 	struct hlist_node *tmp;
1794 	struct fib_alias *fa;
1795 
1796 	/* walk trie in reverse order and free everything */
1797 	for (;;) {
1798 		struct key_vector *n;
1799 
1800 		if (!(cindex--)) {
1801 			t_key pkey = pn->key;
1802 
1803 			if (IS_TRIE(pn))
1804 				break;
1805 
1806 			n = pn;
1807 			pn = node_parent(pn);
1808 
1809 			/* drop emptied tnode */
1810 			put_child_root(pn, n->key, NULL);
1811 			node_free(n);
1812 
1813 			cindex = get_index(pkey, pn);
1814 
1815 			continue;
1816 		}
1817 
1818 		/* grab the next available node */
1819 		n = get_child(pn, cindex);
1820 		if (!n)
1821 			continue;
1822 
1823 		if (IS_TNODE(n)) {
1824 			/* record pn and cindex for leaf walking */
1825 			pn = n;
1826 			cindex = 1ul << n->bits;
1827 
1828 			continue;
1829 		}
1830 
1831 		hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1832 			hlist_del_rcu(&fa->fa_list);
1833 			alias_free_mem_rcu(fa);
1834 		}
1835 
1836 		put_child_root(pn, n->key, NULL);
1837 		node_free(n);
1838 	}
1839 
1840 #ifdef CONFIG_IP_FIB_TRIE_STATS
1841 	free_percpu(t->stats);
1842 #endif
1843 	kfree(tb);
1844 }
1845 
1846 struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1847 {
1848 	struct trie *ot = (struct trie *)oldtb->tb_data;
1849 	struct key_vector *l, *tp = ot->kv;
1850 	struct fib_table *local_tb;
1851 	struct fib_alias *fa;
1852 	struct trie *lt;
1853 	t_key key = 0;
1854 
1855 	if (oldtb->tb_data == oldtb->__data)
1856 		return oldtb;
1857 
1858 	local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1859 	if (!local_tb)
1860 		return NULL;
1861 
1862 	lt = (struct trie *)local_tb->tb_data;
1863 
1864 	while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1865 		struct key_vector *local_l = NULL, *local_tp;
1866 
1867 		hlist_for_each_entry(fa, &l->leaf, fa_list) {
1868 			struct fib_alias *new_fa;
1869 
1870 			if (local_tb->tb_id != fa->tb_id)
1871 				continue;
1872 
1873 			/* clone fa for new local table */
1874 			new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1875 			if (!new_fa)
1876 				goto out;
1877 
1878 			memcpy(new_fa, fa, sizeof(*fa));
1879 
1880 			/* insert clone into table */
1881 			if (!local_l)
1882 				local_l = fib_find_node(lt, &local_tp, l->key);
1883 
1884 			if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1885 					     NULL, l->key)) {
1886 				kmem_cache_free(fn_alias_kmem, new_fa);
1887 				goto out;
1888 			}
1889 		}
1890 
1891 		/* stop loop if key wrapped back to 0 */
1892 		key = l->key + 1;
1893 		if (key < l->key)
1894 			break;
1895 	}
1896 
1897 	return local_tb;
1898 out:
1899 	fib_trie_free(local_tb);
1900 
1901 	return NULL;
1902 }
1903 
1904 /* Caller must hold RTNL */
1905 void fib_table_flush_external(struct fib_table *tb)
1906 {
1907 	struct trie *t = (struct trie *)tb->tb_data;
1908 	struct key_vector *pn = t->kv;
1909 	unsigned long cindex = 1;
1910 	struct hlist_node *tmp;
1911 	struct fib_alias *fa;
1912 
1913 	/* walk trie in reverse order */
1914 	for (;;) {
1915 		unsigned char slen = 0;
1916 		struct key_vector *n;
1917 
1918 		if (!(cindex--)) {
1919 			t_key pkey = pn->key;
1920 
1921 			/* cannot resize the trie vector */
1922 			if (IS_TRIE(pn))
1923 				break;
1924 
1925 			/* update the suffix to address pulled leaves */
1926 			if (pn->slen > pn->pos)
1927 				update_suffix(pn);
1928 
1929 			/* resize completed node */
1930 			pn = resize(t, pn);
1931 			cindex = get_index(pkey, pn);
1932 
1933 			continue;
1934 		}
1935 
1936 		/* grab the next available node */
1937 		n = get_child(pn, cindex);
1938 		if (!n)
1939 			continue;
1940 
1941 		if (IS_TNODE(n)) {
1942 			/* record pn and cindex for leaf walking */
1943 			pn = n;
1944 			cindex = 1ul << n->bits;
1945 
1946 			continue;
1947 		}
1948 
1949 		hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1950 			/* if alias was cloned to local then we just
1951 			 * need to remove the local copy from main
1952 			 */
1953 			if (tb->tb_id != fa->tb_id) {
1954 				hlist_del_rcu(&fa->fa_list);
1955 				alias_free_mem_rcu(fa);
1956 				continue;
1957 			}
1958 
1959 			/* record local slen */
1960 			slen = fa->fa_slen;
1961 		}
1962 
1963 		/* update leaf slen */
1964 		n->slen = slen;
1965 
1966 		if (hlist_empty(&n->leaf)) {
1967 			put_child_root(pn, n->key, NULL);
1968 			node_free(n);
1969 		}
1970 	}
1971 }
1972 
1973 /* Caller must hold RTNL. */
1974 int fib_table_flush(struct net *net, struct fib_table *tb, bool flush_all)
1975 {
1976 	struct trie *t = (struct trie *)tb->tb_data;
1977 	struct key_vector *pn = t->kv;
1978 	unsigned long cindex = 1;
1979 	struct hlist_node *tmp;
1980 	struct fib_alias *fa;
1981 	int found = 0;
1982 
1983 	/* walk trie in reverse order */
1984 	for (;;) {
1985 		unsigned char slen = 0;
1986 		struct key_vector *n;
1987 
1988 		if (!(cindex--)) {
1989 			t_key pkey = pn->key;
1990 
1991 			/* cannot resize the trie vector */
1992 			if (IS_TRIE(pn))
1993 				break;
1994 
1995 			/* update the suffix to address pulled leaves */
1996 			if (pn->slen > pn->pos)
1997 				update_suffix(pn);
1998 
1999 			/* resize completed node */
2000 			pn = resize(t, pn);
2001 			cindex = get_index(pkey, pn);
2002 
2003 			continue;
2004 		}
2005 
2006 		/* grab the next available node */
2007 		n = get_child(pn, cindex);
2008 		if (!n)
2009 			continue;
2010 
2011 		if (IS_TNODE(n)) {
2012 			/* record pn and cindex for leaf walking */
2013 			pn = n;
2014 			cindex = 1ul << n->bits;
2015 
2016 			continue;
2017 		}
2018 
2019 		hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
2020 			struct fib_info *fi = fa->fa_info;
2021 
2022 			if (!fi || tb->tb_id != fa->tb_id ||
2023 			    (!(fi->fib_flags & RTNH_F_DEAD) &&
2024 			     !fib_props[fa->fa_type].error)) {
2025 				slen = fa->fa_slen;
2026 				continue;
2027 			}
2028 
2029 			/* Do not flush error routes if network namespace is
2030 			 * not being dismantled
2031 			 */
2032 			if (!flush_all && fib_props[fa->fa_type].error) {
2033 				slen = fa->fa_slen;
2034 				continue;
2035 			}
2036 
2037 			fib_notify_alias_delete(net, n->key, &n->leaf, fa,
2038 						NULL);
2039 			hlist_del_rcu(&fa->fa_list);
2040 			fib_release_info(fa->fa_info);
2041 			alias_free_mem_rcu(fa);
2042 			found++;
2043 		}
2044 
2045 		/* update leaf slen */
2046 		n->slen = slen;
2047 
2048 		if (hlist_empty(&n->leaf)) {
2049 			put_child_root(pn, n->key, NULL);
2050 			node_free(n);
2051 		}
2052 	}
2053 
2054 	pr_debug("trie_flush found=%d\n", found);
2055 	return found;
2056 }
2057 
2058 /* derived from fib_trie_free */
2059 static void __fib_info_notify_update(struct net *net, struct fib_table *tb,
2060 				     struct nl_info *info)
2061 {
2062 	struct trie *t = (struct trie *)tb->tb_data;
2063 	struct key_vector *pn = t->kv;
2064 	unsigned long cindex = 1;
2065 	struct fib_alias *fa;
2066 
2067 	for (;;) {
2068 		struct key_vector *n;
2069 
2070 		if (!(cindex--)) {
2071 			t_key pkey = pn->key;
2072 
2073 			if (IS_TRIE(pn))
2074 				break;
2075 
2076 			pn = node_parent(pn);
2077 			cindex = get_index(pkey, pn);
2078 			continue;
2079 		}
2080 
2081 		/* grab the next available node */
2082 		n = get_child(pn, cindex);
2083 		if (!n)
2084 			continue;
2085 
2086 		if (IS_TNODE(n)) {
2087 			/* record pn and cindex for leaf walking */
2088 			pn = n;
2089 			cindex = 1ul << n->bits;
2090 
2091 			continue;
2092 		}
2093 
2094 		hlist_for_each_entry(fa, &n->leaf, fa_list) {
2095 			struct fib_info *fi = fa->fa_info;
2096 
2097 			if (!fi || !fi->nh_updated || fa->tb_id != tb->tb_id)
2098 				continue;
2099 
2100 			rtmsg_fib(RTM_NEWROUTE, htonl(n->key), fa,
2101 				  KEYLENGTH - fa->fa_slen, tb->tb_id,
2102 				  info, NLM_F_REPLACE);
2103 
2104 			/* call_fib_entry_notifiers will be removed when
2105 			 * in-kernel notifier is implemented and supported
2106 			 * for nexthop objects
2107 			 */
2108 			call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_REPLACE,
2109 						 n->key,
2110 						 KEYLENGTH - fa->fa_slen, fa,
2111 						 NULL);
2112 		}
2113 	}
2114 }
2115 
2116 void fib_info_notify_update(struct net *net, struct nl_info *info)
2117 {
2118 	unsigned int h;
2119 
2120 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2121 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2122 		struct fib_table *tb;
2123 
2124 		hlist_for_each_entry_rcu(tb, head, tb_hlist,
2125 					 lockdep_rtnl_is_held())
2126 			__fib_info_notify_update(net, tb, info);
2127 	}
2128 }
2129 
2130 static int fib_leaf_notify(struct key_vector *l, struct fib_table *tb,
2131 			   struct notifier_block *nb,
2132 			   struct netlink_ext_ack *extack)
2133 {
2134 	struct fib_alias *fa;
2135 	int last_slen = -1;
2136 	int err;
2137 
2138 	hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2139 		struct fib_info *fi = fa->fa_info;
2140 
2141 		if (!fi)
2142 			continue;
2143 
2144 		/* local and main table can share the same trie,
2145 		 * so don't notify twice for the same entry.
2146 		 */
2147 		if (tb->tb_id != fa->tb_id)
2148 			continue;
2149 
2150 		if (fa->fa_slen == last_slen)
2151 			continue;
2152 
2153 		last_slen = fa->fa_slen;
2154 		err = call_fib_entry_notifier(nb, FIB_EVENT_ENTRY_REPLACE,
2155 					      l->key, KEYLENGTH - fa->fa_slen,
2156 					      fa, extack);
2157 		if (err)
2158 			return err;
2159 	}
2160 	return 0;
2161 }
2162 
2163 static int fib_table_notify(struct fib_table *tb, struct notifier_block *nb,
2164 			    struct netlink_ext_ack *extack)
2165 {
2166 	struct trie *t = (struct trie *)tb->tb_data;
2167 	struct key_vector *l, *tp = t->kv;
2168 	t_key key = 0;
2169 	int err;
2170 
2171 	while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2172 		err = fib_leaf_notify(l, tb, nb, extack);
2173 		if (err)
2174 			return err;
2175 
2176 		key = l->key + 1;
2177 		/* stop in case of wrap around */
2178 		if (key < l->key)
2179 			break;
2180 	}
2181 	return 0;
2182 }
2183 
2184 int fib_notify(struct net *net, struct notifier_block *nb,
2185 	       struct netlink_ext_ack *extack)
2186 {
2187 	unsigned int h;
2188 	int err;
2189 
2190 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2191 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2192 		struct fib_table *tb;
2193 
2194 		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2195 			err = fib_table_notify(tb, nb, extack);
2196 			if (err)
2197 				return err;
2198 		}
2199 	}
2200 	return 0;
2201 }
2202 
2203 static void __trie_free_rcu(struct rcu_head *head)
2204 {
2205 	struct fib_table *tb = container_of(head, struct fib_table, rcu);
2206 #ifdef CONFIG_IP_FIB_TRIE_STATS
2207 	struct trie *t = (struct trie *)tb->tb_data;
2208 
2209 	if (tb->tb_data == tb->__data)
2210 		free_percpu(t->stats);
2211 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2212 	kfree(tb);
2213 }
2214 
2215 void fib_free_table(struct fib_table *tb)
2216 {
2217 	call_rcu(&tb->rcu, __trie_free_rcu);
2218 }
2219 
2220 static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
2221 			     struct sk_buff *skb, struct netlink_callback *cb,
2222 			     struct fib_dump_filter *filter)
2223 {
2224 	unsigned int flags = NLM_F_MULTI;
2225 	__be32 xkey = htonl(l->key);
2226 	int i, s_i, i_fa, s_fa, err;
2227 	struct fib_alias *fa;
2228 
2229 	if (filter->filter_set ||
2230 	    !filter->dump_exceptions || !filter->dump_routes)
2231 		flags |= NLM_F_DUMP_FILTERED;
2232 
2233 	s_i = cb->args[4];
2234 	s_fa = cb->args[5];
2235 	i = 0;
2236 
2237 	/* rcu_read_lock is hold by caller */
2238 	hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2239 		struct fib_info *fi = fa->fa_info;
2240 
2241 		if (i < s_i)
2242 			goto next;
2243 
2244 		i_fa = 0;
2245 
2246 		if (tb->tb_id != fa->tb_id)
2247 			goto next;
2248 
2249 		if (filter->filter_set) {
2250 			if (filter->rt_type && fa->fa_type != filter->rt_type)
2251 				goto next;
2252 
2253 			if ((filter->protocol &&
2254 			     fi->fib_protocol != filter->protocol))
2255 				goto next;
2256 
2257 			if (filter->dev &&
2258 			    !fib_info_nh_uses_dev(fi, filter->dev))
2259 				goto next;
2260 		}
2261 
2262 		if (filter->dump_routes) {
2263 			if (!s_fa) {
2264 				struct fib_rt_info fri;
2265 
2266 				fri.fi = fi;
2267 				fri.tb_id = tb->tb_id;
2268 				fri.dst = xkey;
2269 				fri.dst_len = KEYLENGTH - fa->fa_slen;
2270 				fri.tos = fa->fa_tos;
2271 				fri.type = fa->fa_type;
2272 				fri.offload = fa->offload;
2273 				fri.trap = fa->trap;
2274 				err = fib_dump_info(skb,
2275 						    NETLINK_CB(cb->skb).portid,
2276 						    cb->nlh->nlmsg_seq,
2277 						    RTM_NEWROUTE, &fri, flags);
2278 				if (err < 0)
2279 					goto stop;
2280 			}
2281 
2282 			i_fa++;
2283 		}
2284 
2285 		if (filter->dump_exceptions) {
2286 			err = fib_dump_info_fnhe(skb, cb, tb->tb_id, fi,
2287 						 &i_fa, s_fa, flags);
2288 			if (err < 0)
2289 				goto stop;
2290 		}
2291 
2292 next:
2293 		i++;
2294 	}
2295 
2296 	cb->args[4] = i;
2297 	return skb->len;
2298 
2299 stop:
2300 	cb->args[4] = i;
2301 	cb->args[5] = i_fa;
2302 	return err;
2303 }
2304 
2305 /* rcu_read_lock needs to be hold by caller from readside */
2306 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
2307 		   struct netlink_callback *cb, struct fib_dump_filter *filter)
2308 {
2309 	struct trie *t = (struct trie *)tb->tb_data;
2310 	struct key_vector *l, *tp = t->kv;
2311 	/* Dump starting at last key.
2312 	 * Note: 0.0.0.0/0 (ie default) is first key.
2313 	 */
2314 	int count = cb->args[2];
2315 	t_key key = cb->args[3];
2316 
2317 	/* First time here, count and key are both always 0. Count > 0
2318 	 * and key == 0 means the dump has wrapped around and we are done.
2319 	 */
2320 	if (count && !key)
2321 		return skb->len;
2322 
2323 	while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2324 		int err;
2325 
2326 		err = fn_trie_dump_leaf(l, tb, skb, cb, filter);
2327 		if (err < 0) {
2328 			cb->args[3] = key;
2329 			cb->args[2] = count;
2330 			return err;
2331 		}
2332 
2333 		++count;
2334 		key = l->key + 1;
2335 
2336 		memset(&cb->args[4], 0,
2337 		       sizeof(cb->args) - 4*sizeof(cb->args[0]));
2338 
2339 		/* stop loop if key wrapped back to 0 */
2340 		if (key < l->key)
2341 			break;
2342 	}
2343 
2344 	cb->args[3] = key;
2345 	cb->args[2] = count;
2346 
2347 	return skb->len;
2348 }
2349 
2350 void __init fib_trie_init(void)
2351 {
2352 	fn_alias_kmem = kmem_cache_create("ip_fib_alias",
2353 					  sizeof(struct fib_alias),
2354 					  0, SLAB_PANIC, NULL);
2355 
2356 	trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
2357 					   LEAF_SIZE,
2358 					   0, SLAB_PANIC, NULL);
2359 }
2360 
2361 struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
2362 {
2363 	struct fib_table *tb;
2364 	struct trie *t;
2365 	size_t sz = sizeof(*tb);
2366 
2367 	if (!alias)
2368 		sz += sizeof(struct trie);
2369 
2370 	tb = kzalloc(sz, GFP_KERNEL);
2371 	if (!tb)
2372 		return NULL;
2373 
2374 	tb->tb_id = id;
2375 	tb->tb_num_default = 0;
2376 	tb->tb_data = (alias ? alias->__data : tb->__data);
2377 
2378 	if (alias)
2379 		return tb;
2380 
2381 	t = (struct trie *) tb->tb_data;
2382 	t->kv[0].pos = KEYLENGTH;
2383 	t->kv[0].slen = KEYLENGTH;
2384 #ifdef CONFIG_IP_FIB_TRIE_STATS
2385 	t->stats = alloc_percpu(struct trie_use_stats);
2386 	if (!t->stats) {
2387 		kfree(tb);
2388 		tb = NULL;
2389 	}
2390 #endif
2391 
2392 	return tb;
2393 }
2394 
2395 #ifdef CONFIG_PROC_FS
2396 /* Depth first Trie walk iterator */
2397 struct fib_trie_iter {
2398 	struct seq_net_private p;
2399 	struct fib_table *tb;
2400 	struct key_vector *tnode;
2401 	unsigned int index;
2402 	unsigned int depth;
2403 };
2404 
2405 static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2406 {
2407 	unsigned long cindex = iter->index;
2408 	struct key_vector *pn = iter->tnode;
2409 	t_key pkey;
2410 
2411 	pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2412 		 iter->tnode, iter->index, iter->depth);
2413 
2414 	while (!IS_TRIE(pn)) {
2415 		while (cindex < child_length(pn)) {
2416 			struct key_vector *n = get_child_rcu(pn, cindex++);
2417 
2418 			if (!n)
2419 				continue;
2420 
2421 			if (IS_LEAF(n)) {
2422 				iter->tnode = pn;
2423 				iter->index = cindex;
2424 			} else {
2425 				/* push down one level */
2426 				iter->tnode = n;
2427 				iter->index = 0;
2428 				++iter->depth;
2429 			}
2430 
2431 			return n;
2432 		}
2433 
2434 		/* Current node exhausted, pop back up */
2435 		pkey = pn->key;
2436 		pn = node_parent_rcu(pn);
2437 		cindex = get_index(pkey, pn) + 1;
2438 		--iter->depth;
2439 	}
2440 
2441 	/* record root node so further searches know we are done */
2442 	iter->tnode = pn;
2443 	iter->index = 0;
2444 
2445 	return NULL;
2446 }
2447 
2448 static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2449 					     struct trie *t)
2450 {
2451 	struct key_vector *n, *pn;
2452 
2453 	if (!t)
2454 		return NULL;
2455 
2456 	pn = t->kv;
2457 	n = rcu_dereference(pn->tnode[0]);
2458 	if (!n)
2459 		return NULL;
2460 
2461 	if (IS_TNODE(n)) {
2462 		iter->tnode = n;
2463 		iter->index = 0;
2464 		iter->depth = 1;
2465 	} else {
2466 		iter->tnode = pn;
2467 		iter->index = 0;
2468 		iter->depth = 0;
2469 	}
2470 
2471 	return n;
2472 }
2473 
2474 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2475 {
2476 	struct key_vector *n;
2477 	struct fib_trie_iter iter;
2478 
2479 	memset(s, 0, sizeof(*s));
2480 
2481 	rcu_read_lock();
2482 	for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2483 		if (IS_LEAF(n)) {
2484 			struct fib_alias *fa;
2485 
2486 			s->leaves++;
2487 			s->totdepth += iter.depth;
2488 			if (iter.depth > s->maxdepth)
2489 				s->maxdepth = iter.depth;
2490 
2491 			hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2492 				++s->prefixes;
2493 		} else {
2494 			s->tnodes++;
2495 			if (n->bits < MAX_STAT_DEPTH)
2496 				s->nodesizes[n->bits]++;
2497 			s->nullpointers += tn_info(n)->empty_children;
2498 		}
2499 	}
2500 	rcu_read_unlock();
2501 }
2502 
2503 /*
2504  *	This outputs /proc/net/fib_triestats
2505  */
2506 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2507 {
2508 	unsigned int i, max, pointers, bytes, avdepth;
2509 
2510 	if (stat->leaves)
2511 		avdepth = stat->totdepth*100 / stat->leaves;
2512 	else
2513 		avdepth = 0;
2514 
2515 	seq_printf(seq, "\tAver depth:     %u.%02d\n",
2516 		   avdepth / 100, avdepth % 100);
2517 	seq_printf(seq, "\tMax depth:      %u\n", stat->maxdepth);
2518 
2519 	seq_printf(seq, "\tLeaves:         %u\n", stat->leaves);
2520 	bytes = LEAF_SIZE * stat->leaves;
2521 
2522 	seq_printf(seq, "\tPrefixes:       %u\n", stat->prefixes);
2523 	bytes += sizeof(struct fib_alias) * stat->prefixes;
2524 
2525 	seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2526 	bytes += TNODE_SIZE(0) * stat->tnodes;
2527 
2528 	max = MAX_STAT_DEPTH;
2529 	while (max > 0 && stat->nodesizes[max-1] == 0)
2530 		max--;
2531 
2532 	pointers = 0;
2533 	for (i = 1; i < max; i++)
2534 		if (stat->nodesizes[i] != 0) {
2535 			seq_printf(seq, "  %u: %u",  i, stat->nodesizes[i]);
2536 			pointers += (1<<i) * stat->nodesizes[i];
2537 		}
2538 	seq_putc(seq, '\n');
2539 	seq_printf(seq, "\tPointers: %u\n", pointers);
2540 
2541 	bytes += sizeof(struct key_vector *) * pointers;
2542 	seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2543 	seq_printf(seq, "Total size: %u  kB\n", (bytes + 1023) / 1024);
2544 }
2545 
2546 #ifdef CONFIG_IP_FIB_TRIE_STATS
2547 static void trie_show_usage(struct seq_file *seq,
2548 			    const struct trie_use_stats __percpu *stats)
2549 {
2550 	struct trie_use_stats s = { 0 };
2551 	int cpu;
2552 
2553 	/* loop through all of the CPUs and gather up the stats */
2554 	for_each_possible_cpu(cpu) {
2555 		const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2556 
2557 		s.gets += pcpu->gets;
2558 		s.backtrack += pcpu->backtrack;
2559 		s.semantic_match_passed += pcpu->semantic_match_passed;
2560 		s.semantic_match_miss += pcpu->semantic_match_miss;
2561 		s.null_node_hit += pcpu->null_node_hit;
2562 		s.resize_node_skipped += pcpu->resize_node_skipped;
2563 	}
2564 
2565 	seq_printf(seq, "\nCounters:\n---------\n");
2566 	seq_printf(seq, "gets = %u\n", s.gets);
2567 	seq_printf(seq, "backtracks = %u\n", s.backtrack);
2568 	seq_printf(seq, "semantic match passed = %u\n",
2569 		   s.semantic_match_passed);
2570 	seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2571 	seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2572 	seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2573 }
2574 #endif /*  CONFIG_IP_FIB_TRIE_STATS */
2575 
2576 static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2577 {
2578 	if (tb->tb_id == RT_TABLE_LOCAL)
2579 		seq_puts(seq, "Local:\n");
2580 	else if (tb->tb_id == RT_TABLE_MAIN)
2581 		seq_puts(seq, "Main:\n");
2582 	else
2583 		seq_printf(seq, "Id %d:\n", tb->tb_id);
2584 }
2585 
2586 
2587 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2588 {
2589 	struct net *net = (struct net *)seq->private;
2590 	unsigned int h;
2591 
2592 	seq_printf(seq,
2593 		   "Basic info: size of leaf:"
2594 		   " %zd bytes, size of tnode: %zd bytes.\n",
2595 		   LEAF_SIZE, TNODE_SIZE(0));
2596 
2597 	rcu_read_lock();
2598 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2599 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2600 		struct fib_table *tb;
2601 
2602 		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2603 			struct trie *t = (struct trie *) tb->tb_data;
2604 			struct trie_stat stat;
2605 
2606 			if (!t)
2607 				continue;
2608 
2609 			fib_table_print(seq, tb);
2610 
2611 			trie_collect_stats(t, &stat);
2612 			trie_show_stats(seq, &stat);
2613 #ifdef CONFIG_IP_FIB_TRIE_STATS
2614 			trie_show_usage(seq, t->stats);
2615 #endif
2616 		}
2617 		cond_resched_rcu();
2618 	}
2619 	rcu_read_unlock();
2620 
2621 	return 0;
2622 }
2623 
2624 static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2625 {
2626 	struct fib_trie_iter *iter = seq->private;
2627 	struct net *net = seq_file_net(seq);
2628 	loff_t idx = 0;
2629 	unsigned int h;
2630 
2631 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2632 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2633 		struct fib_table *tb;
2634 
2635 		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2636 			struct key_vector *n;
2637 
2638 			for (n = fib_trie_get_first(iter,
2639 						    (struct trie *) tb->tb_data);
2640 			     n; n = fib_trie_get_next(iter))
2641 				if (pos == idx++) {
2642 					iter->tb = tb;
2643 					return n;
2644 				}
2645 		}
2646 	}
2647 
2648 	return NULL;
2649 }
2650 
2651 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2652 	__acquires(RCU)
2653 {
2654 	rcu_read_lock();
2655 	return fib_trie_get_idx(seq, *pos);
2656 }
2657 
2658 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2659 {
2660 	struct fib_trie_iter *iter = seq->private;
2661 	struct net *net = seq_file_net(seq);
2662 	struct fib_table *tb = iter->tb;
2663 	struct hlist_node *tb_node;
2664 	unsigned int h;
2665 	struct key_vector *n;
2666 
2667 	++*pos;
2668 	/* next node in same table */
2669 	n = fib_trie_get_next(iter);
2670 	if (n)
2671 		return n;
2672 
2673 	/* walk rest of this hash chain */
2674 	h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2675 	while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2676 		tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2677 		n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2678 		if (n)
2679 			goto found;
2680 	}
2681 
2682 	/* new hash chain */
2683 	while (++h < FIB_TABLE_HASHSZ) {
2684 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2685 		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2686 			n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2687 			if (n)
2688 				goto found;
2689 		}
2690 	}
2691 	return NULL;
2692 
2693 found:
2694 	iter->tb = tb;
2695 	return n;
2696 }
2697 
2698 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2699 	__releases(RCU)
2700 {
2701 	rcu_read_unlock();
2702 }
2703 
2704 static void seq_indent(struct seq_file *seq, int n)
2705 {
2706 	while (n-- > 0)
2707 		seq_puts(seq, "   ");
2708 }
2709 
2710 static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2711 {
2712 	switch (s) {
2713 	case RT_SCOPE_UNIVERSE: return "universe";
2714 	case RT_SCOPE_SITE:	return "site";
2715 	case RT_SCOPE_LINK:	return "link";
2716 	case RT_SCOPE_HOST:	return "host";
2717 	case RT_SCOPE_NOWHERE:	return "nowhere";
2718 	default:
2719 		snprintf(buf, len, "scope=%d", s);
2720 		return buf;
2721 	}
2722 }
2723 
2724 static const char *const rtn_type_names[__RTN_MAX] = {
2725 	[RTN_UNSPEC] = "UNSPEC",
2726 	[RTN_UNICAST] = "UNICAST",
2727 	[RTN_LOCAL] = "LOCAL",
2728 	[RTN_BROADCAST] = "BROADCAST",
2729 	[RTN_ANYCAST] = "ANYCAST",
2730 	[RTN_MULTICAST] = "MULTICAST",
2731 	[RTN_BLACKHOLE] = "BLACKHOLE",
2732 	[RTN_UNREACHABLE] = "UNREACHABLE",
2733 	[RTN_PROHIBIT] = "PROHIBIT",
2734 	[RTN_THROW] = "THROW",
2735 	[RTN_NAT] = "NAT",
2736 	[RTN_XRESOLVE] = "XRESOLVE",
2737 };
2738 
2739 static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2740 {
2741 	if (t < __RTN_MAX && rtn_type_names[t])
2742 		return rtn_type_names[t];
2743 	snprintf(buf, len, "type %u", t);
2744 	return buf;
2745 }
2746 
2747 /* Pretty print the trie */
2748 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2749 {
2750 	const struct fib_trie_iter *iter = seq->private;
2751 	struct key_vector *n = v;
2752 
2753 	if (IS_TRIE(node_parent_rcu(n)))
2754 		fib_table_print(seq, iter->tb);
2755 
2756 	if (IS_TNODE(n)) {
2757 		__be32 prf = htonl(n->key);
2758 
2759 		seq_indent(seq, iter->depth-1);
2760 		seq_printf(seq, "  +-- %pI4/%zu %u %u %u\n",
2761 			   &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2762 			   tn_info(n)->full_children,
2763 			   tn_info(n)->empty_children);
2764 	} else {
2765 		__be32 val = htonl(n->key);
2766 		struct fib_alias *fa;
2767 
2768 		seq_indent(seq, iter->depth);
2769 		seq_printf(seq, "  |-- %pI4\n", &val);
2770 
2771 		hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2772 			char buf1[32], buf2[32];
2773 
2774 			seq_indent(seq, iter->depth + 1);
2775 			seq_printf(seq, "  /%zu %s %s",
2776 				   KEYLENGTH - fa->fa_slen,
2777 				   rtn_scope(buf1, sizeof(buf1),
2778 					     fa->fa_info->fib_scope),
2779 				   rtn_type(buf2, sizeof(buf2),
2780 					    fa->fa_type));
2781 			if (fa->fa_tos)
2782 				seq_printf(seq, " tos=%d", fa->fa_tos);
2783 			seq_putc(seq, '\n');
2784 		}
2785 	}
2786 
2787 	return 0;
2788 }
2789 
2790 static const struct seq_operations fib_trie_seq_ops = {
2791 	.start  = fib_trie_seq_start,
2792 	.next   = fib_trie_seq_next,
2793 	.stop   = fib_trie_seq_stop,
2794 	.show   = fib_trie_seq_show,
2795 };
2796 
2797 struct fib_route_iter {
2798 	struct seq_net_private p;
2799 	struct fib_table *main_tb;
2800 	struct key_vector *tnode;
2801 	loff_t	pos;
2802 	t_key	key;
2803 };
2804 
2805 static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2806 					    loff_t pos)
2807 {
2808 	struct key_vector *l, **tp = &iter->tnode;
2809 	t_key key;
2810 
2811 	/* use cached location of previously found key */
2812 	if (iter->pos > 0 && pos >= iter->pos) {
2813 		key = iter->key;
2814 	} else {
2815 		iter->pos = 1;
2816 		key = 0;
2817 	}
2818 
2819 	pos -= iter->pos;
2820 
2821 	while ((l = leaf_walk_rcu(tp, key)) && (pos-- > 0)) {
2822 		key = l->key + 1;
2823 		iter->pos++;
2824 		l = NULL;
2825 
2826 		/* handle unlikely case of a key wrap */
2827 		if (!key)
2828 			break;
2829 	}
2830 
2831 	if (l)
2832 		iter->key = l->key;	/* remember it */
2833 	else
2834 		iter->pos = 0;		/* forget it */
2835 
2836 	return l;
2837 }
2838 
2839 static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2840 	__acquires(RCU)
2841 {
2842 	struct fib_route_iter *iter = seq->private;
2843 	struct fib_table *tb;
2844 	struct trie *t;
2845 
2846 	rcu_read_lock();
2847 
2848 	tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2849 	if (!tb)
2850 		return NULL;
2851 
2852 	iter->main_tb = tb;
2853 	t = (struct trie *)tb->tb_data;
2854 	iter->tnode = t->kv;
2855 
2856 	if (*pos != 0)
2857 		return fib_route_get_idx(iter, *pos);
2858 
2859 	iter->pos = 0;
2860 	iter->key = KEY_MAX;
2861 
2862 	return SEQ_START_TOKEN;
2863 }
2864 
2865 static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2866 {
2867 	struct fib_route_iter *iter = seq->private;
2868 	struct key_vector *l = NULL;
2869 	t_key key = iter->key + 1;
2870 
2871 	++*pos;
2872 
2873 	/* only allow key of 0 for start of sequence */
2874 	if ((v == SEQ_START_TOKEN) || key)
2875 		l = leaf_walk_rcu(&iter->tnode, key);
2876 
2877 	if (l) {
2878 		iter->key = l->key;
2879 		iter->pos++;
2880 	} else {
2881 		iter->pos = 0;
2882 	}
2883 
2884 	return l;
2885 }
2886 
2887 static void fib_route_seq_stop(struct seq_file *seq, void *v)
2888 	__releases(RCU)
2889 {
2890 	rcu_read_unlock();
2891 }
2892 
2893 static unsigned int fib_flag_trans(int type, __be32 mask, struct fib_info *fi)
2894 {
2895 	unsigned int flags = 0;
2896 
2897 	if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2898 		flags = RTF_REJECT;
2899 	if (fi) {
2900 		const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2901 
2902 		if (nhc->nhc_gw.ipv4)
2903 			flags |= RTF_GATEWAY;
2904 	}
2905 	if (mask == htonl(0xFFFFFFFF))
2906 		flags |= RTF_HOST;
2907 	flags |= RTF_UP;
2908 	return flags;
2909 }
2910 
2911 /*
2912  *	This outputs /proc/net/route.
2913  *	The format of the file is not supposed to be changed
2914  *	and needs to be same as fib_hash output to avoid breaking
2915  *	legacy utilities
2916  */
2917 static int fib_route_seq_show(struct seq_file *seq, void *v)
2918 {
2919 	struct fib_route_iter *iter = seq->private;
2920 	struct fib_table *tb = iter->main_tb;
2921 	struct fib_alias *fa;
2922 	struct key_vector *l = v;
2923 	__be32 prefix;
2924 
2925 	if (v == SEQ_START_TOKEN) {
2926 		seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2927 			   "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2928 			   "\tWindow\tIRTT");
2929 		return 0;
2930 	}
2931 
2932 	prefix = htonl(l->key);
2933 
2934 	hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2935 		struct fib_info *fi = fa->fa_info;
2936 		__be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2937 		unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2938 
2939 		if ((fa->fa_type == RTN_BROADCAST) ||
2940 		    (fa->fa_type == RTN_MULTICAST))
2941 			continue;
2942 
2943 		if (fa->tb_id != tb->tb_id)
2944 			continue;
2945 
2946 		seq_setwidth(seq, 127);
2947 
2948 		if (fi) {
2949 			struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2950 			__be32 gw = 0;
2951 
2952 			if (nhc->nhc_gw_family == AF_INET)
2953 				gw = nhc->nhc_gw.ipv4;
2954 
2955 			seq_printf(seq,
2956 				   "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2957 				   "%d\t%08X\t%d\t%u\t%u",
2958 				   nhc->nhc_dev ? nhc->nhc_dev->name : "*",
2959 				   prefix, gw, flags, 0, 0,
2960 				   fi->fib_priority,
2961 				   mask,
2962 				   (fi->fib_advmss ?
2963 				    fi->fib_advmss + 40 : 0),
2964 				   fi->fib_window,
2965 				   fi->fib_rtt >> 3);
2966 		} else {
2967 			seq_printf(seq,
2968 				   "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2969 				   "%d\t%08X\t%d\t%u\t%u",
2970 				   prefix, 0, flags, 0, 0, 0,
2971 				   mask, 0, 0, 0);
2972 		}
2973 		seq_pad(seq, '\n');
2974 	}
2975 
2976 	return 0;
2977 }
2978 
2979 static const struct seq_operations fib_route_seq_ops = {
2980 	.start  = fib_route_seq_start,
2981 	.next   = fib_route_seq_next,
2982 	.stop   = fib_route_seq_stop,
2983 	.show   = fib_route_seq_show,
2984 };
2985 
2986 int __net_init fib_proc_init(struct net *net)
2987 {
2988 	if (!proc_create_net("fib_trie", 0444, net->proc_net, &fib_trie_seq_ops,
2989 			sizeof(struct fib_trie_iter)))
2990 		goto out1;
2991 
2992 	if (!proc_create_net_single("fib_triestat", 0444, net->proc_net,
2993 			fib_triestat_seq_show, NULL))
2994 		goto out2;
2995 
2996 	if (!proc_create_net("route", 0444, net->proc_net, &fib_route_seq_ops,
2997 			sizeof(struct fib_route_iter)))
2998 		goto out3;
2999 
3000 	return 0;
3001 
3002 out3:
3003 	remove_proc_entry("fib_triestat", net->proc_net);
3004 out2:
3005 	remove_proc_entry("fib_trie", net->proc_net);
3006 out1:
3007 	return -ENOMEM;
3008 }
3009 
3010 void __net_exit fib_proc_exit(struct net *net)
3011 {
3012 	remove_proc_entry("fib_trie", net->proc_net);
3013 	remove_proc_entry("fib_triestat", net->proc_net);
3014 	remove_proc_entry("route", net->proc_net);
3015 }
3016 
3017 #endif /* CONFIG_PROC_FS */
3018