xref: /linux/net/ipv4/fib_trie.c (revision fd3d73286662950656f54616d0b3fea29e2dc99a)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *
4  *   Robert Olsson <robert.olsson@its.uu.se> Uppsala Universitet
5  *     & Swedish University of Agricultural Sciences.
6  *
7  *   Jens Laas <jens.laas@data.slu.se> Swedish University of
8  *     Agricultural Sciences.
9  *
10  *   Hans Liss <hans.liss@its.uu.se>  Uppsala Universitet
11  *
12  * This work is based on the LPC-trie which is originally described in:
13  *
14  * An experimental study of compression methods for dynamic tries
15  * Stefan Nilsson and Matti Tikkanen. Algorithmica, 33(1):19-33, 2002.
16  * https://www.csc.kth.se/~snilsson/software/dyntrie2/
17  *
18  * IP-address lookup using LC-tries. Stefan Nilsson and Gunnar Karlsson
19  * IEEE Journal on Selected Areas in Communications, 17(6):1083-1092, June 1999
20  *
21  * Code from fib_hash has been reused which includes the following header:
22  *
23  * INET		An implementation of the TCP/IP protocol suite for the LINUX
24  *		operating system.  INET is implemented using the  BSD Socket
25  *		interface as the means of communication with the user level.
26  *
27  *		IPv4 FIB: lookup engine and maintenance routines.
28  *
29  * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
30  *
31  * Substantial contributions to this work comes from:
32  *
33  *		David S. Miller, <davem@davemloft.net>
34  *		Stephen Hemminger <shemminger@osdl.org>
35  *		Paul E. McKenney <paulmck@us.ibm.com>
36  *		Patrick McHardy <kaber@trash.net>
37  */
38 #include <linux/cache.h>
39 #include <linux/uaccess.h>
40 #include <linux/bitops.h>
41 #include <linux/types.h>
42 #include <linux/kernel.h>
43 #include <linux/mm.h>
44 #include <linux/string.h>
45 #include <linux/socket.h>
46 #include <linux/sockios.h>
47 #include <linux/errno.h>
48 #include <linux/in.h>
49 #include <linux/inet.h>
50 #include <linux/inetdevice.h>
51 #include <linux/netdevice.h>
52 #include <linux/if_arp.h>
53 #include <linux/proc_fs.h>
54 #include <linux/rcupdate.h>
55 #include <linux/rcupdate_wait.h>
56 #include <linux/skbuff.h>
57 #include <linux/netlink.h>
58 #include <linux/init.h>
59 #include <linux/list.h>
60 #include <linux/slab.h>
61 #include <linux/export.h>
62 #include <linux/vmalloc.h>
63 #include <linux/notifier.h>
64 #include <linux/nospec.h>
65 #include <net/net_namespace.h>
66 #include <net/inet_dscp.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,
78 				   enum fib_event_type event_type, u32 dst,
79 				   int dst_len, struct fib_alias *fa,
80 				   struct netlink_ext_ack *extack)
81 {
82 	struct fib_entry_notifier_info info = {
83 		.info.extack = extack,
84 		.dst = dst,
85 		.dst_len = dst_len,
86 		.fi = fa->fa_info,
87 		.dscp = fa->fa_dscp,
88 		.type = fa->fa_type,
89 		.tb_id = fa->tb_id,
90 	};
91 	return call_fib4_notifier(nb, event_type, &info.info);
92 }
93 
94 static int call_fib_entry_notifiers(struct net *net,
95 				    enum fib_event_type event_type, u32 dst,
96 				    int dst_len, struct fib_alias *fa,
97 				    struct netlink_ext_ack *extack)
98 {
99 	struct fib_entry_notifier_info info = {
100 		.info.extack = extack,
101 		.dst = dst,
102 		.dst_len = dst_len,
103 		.fi = fa->fa_info,
104 		.dscp = fa->fa_dscp,
105 		.type = fa->fa_type,
106 		.tb_id = fa->tb_id,
107 	};
108 	return call_fib4_notifiers(net, event_type, &info.info);
109 }
110 
111 #define MAX_STAT_DEPTH 32
112 
113 #define KEYLENGTH	(8*sizeof(t_key))
114 #define KEY_MAX		((t_key)~0)
115 
116 typedef unsigned int t_key;
117 
118 #define IS_TRIE(n)	((n)->pos >= KEYLENGTH)
119 #define IS_TNODE(n)	((n)->bits)
120 #define IS_LEAF(n)	(!(n)->bits)
121 
122 struct key_vector {
123 	t_key key;
124 	unsigned char pos;		/* 2log(KEYLENGTH) bits needed */
125 	unsigned char bits;		/* 2log(KEYLENGTH) bits needed */
126 	unsigned char slen;
127 	union {
128 		/* This list pointer if valid if (pos | bits) == 0 (LEAF) */
129 		struct hlist_head leaf;
130 		/* This array is valid if (pos | bits) > 0 (TNODE) */
131 		DECLARE_FLEX_ARRAY(struct key_vector __rcu *, tnode);
132 	};
133 };
134 
135 struct tnode {
136 	struct rcu_head rcu;
137 	t_key empty_children;		/* KEYLENGTH bits needed */
138 	t_key full_children;		/* KEYLENGTH bits needed */
139 	struct key_vector __rcu *parent;
140 	struct key_vector kv[1];
141 #define tn_bits kv[0].bits
142 };
143 
144 #define TNODE_SIZE(n)	offsetof(struct tnode, kv[0].tnode[n])
145 #define LEAF_SIZE	TNODE_SIZE(1)
146 
147 #ifdef CONFIG_IP_FIB_TRIE_STATS
148 struct trie_use_stats {
149 	unsigned int gets;
150 	unsigned int backtrack;
151 	unsigned int semantic_match_passed;
152 	unsigned int semantic_match_miss;
153 	unsigned int null_node_hit;
154 	unsigned int resize_node_skipped;
155 };
156 #endif
157 
158 struct trie_stat {
159 	unsigned int totdepth;
160 	unsigned int maxdepth;
161 	unsigned int tnodes;
162 	unsigned int leaves;
163 	unsigned int nullpointers;
164 	unsigned int prefixes;
165 	unsigned int nodesizes[MAX_STAT_DEPTH];
166 };
167 
168 struct trie {
169 	struct key_vector kv[1];
170 #ifdef CONFIG_IP_FIB_TRIE_STATS
171 	struct trie_use_stats __percpu *stats;
172 #endif
173 };
174 
175 static struct key_vector *resize(struct trie *t, struct key_vector *tn);
176 static unsigned int tnode_free_size;
177 
178 /*
179  * synchronize_rcu after call_rcu for outstanding dirty memory; it should be
180  * especially useful before resizing the root node with PREEMPT_NONE configs;
181  * the value was obtained experimentally, aiming to avoid visible slowdown.
182  */
183 unsigned int sysctl_fib_sync_mem = 512 * 1024;
184 unsigned int sysctl_fib_sync_mem_min = 64 * 1024;
185 unsigned int sysctl_fib_sync_mem_max = 64 * 1024 * 1024;
186 
187 static struct kmem_cache *fn_alias_kmem __ro_after_init;
188 static struct kmem_cache *trie_leaf_kmem __ro_after_init;
189 
190 static inline struct tnode *tn_info(struct key_vector *kv)
191 {
192 	return container_of(kv, struct tnode, kv[0]);
193 }
194 
195 /* caller must hold RTNL */
196 #define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
197 #define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
198 
199 /* caller must hold RCU read lock or RTNL */
200 #define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
201 #define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
202 
203 /* wrapper for rcu_assign_pointer */
204 static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
205 {
206 	if (n)
207 		rcu_assign_pointer(tn_info(n)->parent, tp);
208 }
209 
210 #define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
211 
212 /* This provides us with the number of children in this node, in the case of a
213  * leaf this will return 0 meaning none of the children are accessible.
214  */
215 static inline unsigned long child_length(const struct key_vector *tn)
216 {
217 	return (1ul << tn->bits) & ~(1ul);
218 }
219 
220 #define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
221 
222 static inline unsigned long get_index(t_key key, struct key_vector *kv)
223 {
224 	unsigned long index = key ^ kv->key;
225 
226 	if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
227 		return 0;
228 
229 	return index >> kv->pos;
230 }
231 
232 /* To understand this stuff, an understanding of keys and all their bits is
233  * necessary. Every node in the trie has a key associated with it, but not
234  * all of the bits in that key are significant.
235  *
236  * Consider a node 'n' and its parent 'tp'.
237  *
238  * If n is a leaf, every bit in its key is significant. Its presence is
239  * necessitated by path compression, since during a tree traversal (when
240  * searching for a leaf - unless we are doing an insertion) we will completely
241  * ignore all skipped bits we encounter. Thus we need to verify, at the end of
242  * a potentially successful search, that we have indeed been walking the
243  * correct key path.
244  *
245  * Note that we can never "miss" the correct key in the tree if present by
246  * following the wrong path. Path compression ensures that segments of the key
247  * that are the same for all keys with a given prefix are skipped, but the
248  * skipped part *is* identical for each node in the subtrie below the skipped
249  * bit! trie_insert() in this implementation takes care of that.
250  *
251  * if n is an internal node - a 'tnode' here, the various parts of its key
252  * have many different meanings.
253  *
254  * Example:
255  * _________________________________________________________________
256  * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
257  * -----------------------------------------------------------------
258  *  31  30  29  28  27  26  25  24  23  22  21  20  19  18  17  16
259  *
260  * _________________________________________________________________
261  * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
262  * -----------------------------------------------------------------
263  *  15  14  13  12  11  10   9   8   7   6   5   4   3   2   1   0
264  *
265  * tp->pos = 22
266  * tp->bits = 3
267  * n->pos = 13
268  * n->bits = 4
269  *
270  * First, let's just ignore the bits that come before the parent tp, that is
271  * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
272  * point we do not use them for anything.
273  *
274  * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
275  * index into the parent's child array. That is, they will be used to find
276  * 'n' among tp's children.
277  *
278  * The bits from (n->pos + n->bits) to (tp->pos - 1) - "S" - are skipped bits
279  * for the node n.
280  *
281  * All the bits we have seen so far are significant to the node n. The rest
282  * of the bits are really not needed or indeed known in n->key.
283  *
284  * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
285  * n's child array, and will of course be different for each child.
286  *
287  * The rest of the bits, from 0 to (n->pos -1) - "u" - are completely unknown
288  * at this point.
289  */
290 
291 static const int halve_threshold = 25;
292 static const int inflate_threshold = 50;
293 static const int halve_threshold_root = 15;
294 static const int inflate_threshold_root = 30;
295 
296 static inline void alias_free_mem_rcu(struct fib_alias *fa)
297 {
298 	kfree_rcu(fa, rcu);
299 }
300 
301 #define TNODE_VMALLOC_MAX \
302 	ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
303 
304 static void __node_free_rcu(struct rcu_head *head)
305 {
306 	struct tnode *n = container_of(head, struct tnode, rcu);
307 
308 	if (!n->tn_bits)
309 		kmem_cache_free(trie_leaf_kmem, n);
310 	else
311 		kvfree(n);
312 }
313 
314 #define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
315 
316 static struct tnode *tnode_alloc(int bits)
317 {
318 	size_t size;
319 
320 	/* verify bits is within bounds */
321 	if (bits > TNODE_VMALLOC_MAX)
322 		return NULL;
323 
324 	/* determine size and verify it is non-zero and didn't overflow */
325 	size = TNODE_SIZE(1ul << bits);
326 
327 	if (size <= PAGE_SIZE)
328 		return kzalloc(size, GFP_KERNEL);
329 	else
330 		return vzalloc(size);
331 }
332 
333 static inline void empty_child_inc(struct key_vector *n)
334 {
335 	tn_info(n)->empty_children++;
336 
337 	if (!tn_info(n)->empty_children)
338 		tn_info(n)->full_children++;
339 }
340 
341 static inline void empty_child_dec(struct key_vector *n)
342 {
343 	if (!tn_info(n)->empty_children)
344 		tn_info(n)->full_children--;
345 
346 	tn_info(n)->empty_children--;
347 }
348 
349 static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
350 {
351 	struct key_vector *l;
352 	struct tnode *kv;
353 
354 	kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
355 	if (!kv)
356 		return NULL;
357 
358 	/* initialize key vector */
359 	l = kv->kv;
360 	l->key = key;
361 	l->pos = 0;
362 	l->bits = 0;
363 	l->slen = fa->fa_slen;
364 
365 	/* link leaf to fib alias */
366 	INIT_HLIST_HEAD(&l->leaf);
367 	hlist_add_head(&fa->fa_list, &l->leaf);
368 
369 	return l;
370 }
371 
372 static struct key_vector *tnode_new(t_key key, int pos, int bits)
373 {
374 	unsigned int shift = pos + bits;
375 	struct key_vector *tn;
376 	struct tnode *tnode;
377 
378 	/* verify bits and pos their msb bits clear and values are valid */
379 	BUG_ON(!bits || (shift > KEYLENGTH));
380 
381 	tnode = tnode_alloc(bits);
382 	if (!tnode)
383 		return NULL;
384 
385 	pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
386 		 sizeof(struct key_vector *) << bits);
387 
388 	if (bits == KEYLENGTH)
389 		tnode->full_children = 1;
390 	else
391 		tnode->empty_children = 1ul << bits;
392 
393 	tn = tnode->kv;
394 	tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
395 	tn->pos = pos;
396 	tn->bits = bits;
397 	tn->slen = pos;
398 
399 	return tn;
400 }
401 
402 /* Check whether a tnode 'n' is "full", i.e. it is an internal node
403  * and no bits are skipped. See discussion in dyntree paper p. 6
404  */
405 static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
406 {
407 	return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
408 }
409 
410 /* Add a child at position i overwriting the old value.
411  * Update the value of full_children and empty_children.
412  */
413 static void put_child(struct key_vector *tn, unsigned long i,
414 		      struct key_vector *n)
415 {
416 	struct key_vector *chi = get_child(tn, i);
417 	int isfull, wasfull;
418 
419 	BUG_ON(i >= child_length(tn));
420 
421 	/* update emptyChildren, overflow into fullChildren */
422 	if (!n && chi)
423 		empty_child_inc(tn);
424 	if (n && !chi)
425 		empty_child_dec(tn);
426 
427 	/* update fullChildren */
428 	wasfull = tnode_full(tn, chi);
429 	isfull = tnode_full(tn, n);
430 
431 	if (wasfull && !isfull)
432 		tn_info(tn)->full_children--;
433 	else if (!wasfull && isfull)
434 		tn_info(tn)->full_children++;
435 
436 	if (n && (tn->slen < n->slen))
437 		tn->slen = n->slen;
438 
439 	rcu_assign_pointer(tn->tnode[i], n);
440 }
441 
442 static void update_children(struct key_vector *tn)
443 {
444 	unsigned long i;
445 
446 	/* update all of the child parent pointers */
447 	for (i = child_length(tn); i;) {
448 		struct key_vector *inode = get_child(tn, --i);
449 
450 		if (!inode)
451 			continue;
452 
453 		/* Either update the children of a tnode that
454 		 * already belongs to us or update the child
455 		 * to point to ourselves.
456 		 */
457 		if (node_parent(inode) == tn)
458 			update_children(inode);
459 		else
460 			node_set_parent(inode, tn);
461 	}
462 }
463 
464 static inline void put_child_root(struct key_vector *tp, t_key key,
465 				  struct key_vector *n)
466 {
467 	if (IS_TRIE(tp))
468 		rcu_assign_pointer(tp->tnode[0], n);
469 	else
470 		put_child(tp, get_index(key, tp), n);
471 }
472 
473 static inline void tnode_free_init(struct key_vector *tn)
474 {
475 	tn_info(tn)->rcu.next = NULL;
476 }
477 
478 static inline void tnode_free_append(struct key_vector *tn,
479 				     struct key_vector *n)
480 {
481 	tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
482 	tn_info(tn)->rcu.next = &tn_info(n)->rcu;
483 }
484 
485 static void tnode_free(struct key_vector *tn)
486 {
487 	struct callback_head *head = &tn_info(tn)->rcu;
488 
489 	while (head) {
490 		head = head->next;
491 		tnode_free_size += TNODE_SIZE(1ul << tn->bits);
492 		node_free(tn);
493 
494 		tn = container_of(head, struct tnode, rcu)->kv;
495 	}
496 
497 	if (tnode_free_size >= READ_ONCE(sysctl_fib_sync_mem)) {
498 		tnode_free_size = 0;
499 		synchronize_net();
500 	}
501 }
502 
503 static struct key_vector *replace(struct trie *t,
504 				  struct key_vector *oldtnode,
505 				  struct key_vector *tn)
506 {
507 	struct key_vector *tp = node_parent(oldtnode);
508 	unsigned long i;
509 
510 	/* setup the parent pointer out of and back into this node */
511 	NODE_INIT_PARENT(tn, tp);
512 	put_child_root(tp, tn->key, tn);
513 
514 	/* update all of the child parent pointers */
515 	update_children(tn);
516 
517 	/* all pointers should be clean so we are done */
518 	tnode_free(oldtnode);
519 
520 	/* resize children now that oldtnode is freed */
521 	for (i = child_length(tn); i;) {
522 		struct key_vector *inode = get_child(tn, --i);
523 
524 		/* resize child node */
525 		if (tnode_full(tn, inode))
526 			tn = resize(t, inode);
527 	}
528 
529 	return tp;
530 }
531 
532 static struct key_vector *inflate(struct trie *t,
533 				  struct key_vector *oldtnode)
534 {
535 	struct key_vector *tn;
536 	unsigned long i;
537 	t_key m;
538 
539 	pr_debug("In inflate\n");
540 
541 	tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
542 	if (!tn)
543 		goto notnode;
544 
545 	/* prepare oldtnode to be freed */
546 	tnode_free_init(oldtnode);
547 
548 	/* Assemble all of the pointers in our cluster, in this case that
549 	 * represents all of the pointers out of our allocated nodes that
550 	 * point to existing tnodes and the links between our allocated
551 	 * nodes.
552 	 */
553 	for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
554 		struct key_vector *inode = get_child(oldtnode, --i);
555 		struct key_vector *node0, *node1;
556 		unsigned long j, k;
557 
558 		/* An empty child */
559 		if (!inode)
560 			continue;
561 
562 		/* A leaf or an internal node with skipped bits */
563 		if (!tnode_full(oldtnode, inode)) {
564 			put_child(tn, get_index(inode->key, tn), inode);
565 			continue;
566 		}
567 
568 		/* drop the node in the old tnode free list */
569 		tnode_free_append(oldtnode, inode);
570 
571 		/* An internal node with two children */
572 		if (inode->bits == 1) {
573 			put_child(tn, 2 * i + 1, get_child(inode, 1));
574 			put_child(tn, 2 * i, get_child(inode, 0));
575 			continue;
576 		}
577 
578 		/* We will replace this node 'inode' with two new
579 		 * ones, 'node0' and 'node1', each with half of the
580 		 * original children. The two new nodes will have
581 		 * a position one bit further down the key and this
582 		 * means that the "significant" part of their keys
583 		 * (see the discussion near the top of this file)
584 		 * will differ by one bit, which will be "0" in
585 		 * node0's key and "1" in node1's key. Since we are
586 		 * moving the key position by one step, the bit that
587 		 * we are moving away from - the bit at position
588 		 * (tn->pos) - is the one that will differ between
589 		 * node0 and node1. So... we synthesize that bit in the
590 		 * two new keys.
591 		 */
592 		node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
593 		if (!node1)
594 			goto nomem;
595 		node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
596 
597 		tnode_free_append(tn, node1);
598 		if (!node0)
599 			goto nomem;
600 		tnode_free_append(tn, node0);
601 
602 		/* populate child pointers in new nodes */
603 		for (k = child_length(inode), j = k / 2; j;) {
604 			put_child(node1, --j, get_child(inode, --k));
605 			put_child(node0, j, get_child(inode, j));
606 			put_child(node1, --j, get_child(inode, --k));
607 			put_child(node0, j, get_child(inode, j));
608 		}
609 
610 		/* link new nodes to parent */
611 		NODE_INIT_PARENT(node1, tn);
612 		NODE_INIT_PARENT(node0, tn);
613 
614 		/* link parent to nodes */
615 		put_child(tn, 2 * i + 1, node1);
616 		put_child(tn, 2 * i, node0);
617 	}
618 
619 	/* setup the parent pointers into and out of this node */
620 	return replace(t, oldtnode, tn);
621 nomem:
622 	/* all pointers should be clean so we are done */
623 	tnode_free(tn);
624 notnode:
625 	return NULL;
626 }
627 
628 static struct key_vector *halve(struct trie *t,
629 				struct key_vector *oldtnode)
630 {
631 	struct key_vector *tn;
632 	unsigned long i;
633 
634 	pr_debug("In halve\n");
635 
636 	tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
637 	if (!tn)
638 		goto notnode;
639 
640 	/* prepare oldtnode to be freed */
641 	tnode_free_init(oldtnode);
642 
643 	/* Assemble all of the pointers in our cluster, in this case that
644 	 * represents all of the pointers out of our allocated nodes that
645 	 * point to existing tnodes and the links between our allocated
646 	 * nodes.
647 	 */
648 	for (i = child_length(oldtnode); i;) {
649 		struct key_vector *node1 = get_child(oldtnode, --i);
650 		struct key_vector *node0 = get_child(oldtnode, --i);
651 		struct key_vector *inode;
652 
653 		/* At least one of the children is empty */
654 		if (!node1 || !node0) {
655 			put_child(tn, i / 2, node1 ? : node0);
656 			continue;
657 		}
658 
659 		/* Two nonempty children */
660 		inode = tnode_new(node0->key, oldtnode->pos, 1);
661 		if (!inode)
662 			goto nomem;
663 		tnode_free_append(tn, inode);
664 
665 		/* initialize pointers out of node */
666 		put_child(inode, 1, node1);
667 		put_child(inode, 0, node0);
668 		NODE_INIT_PARENT(inode, tn);
669 
670 		/* link parent to node */
671 		put_child(tn, i / 2, inode);
672 	}
673 
674 	/* setup the parent pointers into and out of this node */
675 	return replace(t, oldtnode, tn);
676 nomem:
677 	/* all pointers should be clean so we are done */
678 	tnode_free(tn);
679 notnode:
680 	return NULL;
681 }
682 
683 static struct key_vector *collapse(struct trie *t,
684 				   struct key_vector *oldtnode)
685 {
686 	struct key_vector *n, *tp;
687 	unsigned long i;
688 
689 	/* scan the tnode looking for that one child that might still exist */
690 	for (n = NULL, i = child_length(oldtnode); !n && i;)
691 		n = get_child(oldtnode, --i);
692 
693 	/* compress one level */
694 	tp = node_parent(oldtnode);
695 	put_child_root(tp, oldtnode->key, n);
696 	node_set_parent(n, tp);
697 
698 	/* drop dead node */
699 	node_free(oldtnode);
700 
701 	return tp;
702 }
703 
704 static unsigned char update_suffix(struct key_vector *tn)
705 {
706 	unsigned char slen = tn->pos;
707 	unsigned long stride, i;
708 	unsigned char slen_max;
709 
710 	/* only vector 0 can have a suffix length greater than or equal to
711 	 * tn->pos + tn->bits, the second highest node will have a suffix
712 	 * length at most of tn->pos + tn->bits - 1
713 	 */
714 	slen_max = min_t(unsigned char, tn->pos + tn->bits - 1, tn->slen);
715 
716 	/* search though the list of children looking for nodes that might
717 	 * have a suffix greater than the one we currently have.  This is
718 	 * why we start with a stride of 2 since a stride of 1 would
719 	 * represent the nodes with suffix length equal to tn->pos
720 	 */
721 	for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
722 		struct key_vector *n = get_child(tn, i);
723 
724 		if (!n || (n->slen <= slen))
725 			continue;
726 
727 		/* update stride and slen based on new value */
728 		stride <<= (n->slen - slen);
729 		slen = n->slen;
730 		i &= ~(stride - 1);
731 
732 		/* stop searching if we have hit the maximum possible value */
733 		if (slen >= slen_max)
734 			break;
735 	}
736 
737 	tn->slen = slen;
738 
739 	return slen;
740 }
741 
742 /* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
743  * the Helsinki University of Technology and Matti Tikkanen of Nokia
744  * Telecommunications, page 6:
745  * "A node is doubled if the ratio of non-empty children to all
746  * children in the *doubled* node is at least 'high'."
747  *
748  * 'high' in this instance is the variable 'inflate_threshold'. It
749  * is expressed as a percentage, so we multiply it with
750  * child_length() and instead of multiplying by 2 (since the
751  * child array will be doubled by inflate()) and multiplying
752  * the left-hand side by 100 (to handle the percentage thing) we
753  * multiply the left-hand side by 50.
754  *
755  * The left-hand side may look a bit weird: child_length(tn)
756  * - tn->empty_children is of course the number of non-null children
757  * in the current node. tn->full_children is the number of "full"
758  * children, that is non-null tnodes with a skip value of 0.
759  * All of those will be doubled in the resulting inflated tnode, so
760  * we just count them one extra time here.
761  *
762  * A clearer way to write this would be:
763  *
764  * to_be_doubled = tn->full_children;
765  * not_to_be_doubled = child_length(tn) - tn->empty_children -
766  *     tn->full_children;
767  *
768  * new_child_length = child_length(tn) * 2;
769  *
770  * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
771  *      new_child_length;
772  * if (new_fill_factor >= inflate_threshold)
773  *
774  * ...and so on, tho it would mess up the while () loop.
775  *
776  * anyway,
777  * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
778  *      inflate_threshold
779  *
780  * avoid a division:
781  * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
782  *      inflate_threshold * new_child_length
783  *
784  * expand not_to_be_doubled and to_be_doubled, and shorten:
785  * 100 * (child_length(tn) - tn->empty_children +
786  *    tn->full_children) >= inflate_threshold * new_child_length
787  *
788  * expand new_child_length:
789  * 100 * (child_length(tn) - tn->empty_children +
790  *    tn->full_children) >=
791  *      inflate_threshold * child_length(tn) * 2
792  *
793  * shorten again:
794  * 50 * (tn->full_children + child_length(tn) -
795  *    tn->empty_children) >= inflate_threshold *
796  *    child_length(tn)
797  *
798  */
799 static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
800 {
801 	unsigned long used = child_length(tn);
802 	unsigned long threshold = used;
803 
804 	/* Keep root node larger */
805 	threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
806 	used -= tn_info(tn)->empty_children;
807 	used += tn_info(tn)->full_children;
808 
809 	/* if bits == KEYLENGTH then pos = 0, and will fail below */
810 
811 	return (used > 1) && tn->pos && ((50 * used) >= threshold);
812 }
813 
814 static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
815 {
816 	unsigned long used = child_length(tn);
817 	unsigned long threshold = used;
818 
819 	/* Keep root node larger */
820 	threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
821 	used -= tn_info(tn)->empty_children;
822 
823 	/* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
824 
825 	return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
826 }
827 
828 static inline bool should_collapse(struct key_vector *tn)
829 {
830 	unsigned long used = child_length(tn);
831 
832 	used -= tn_info(tn)->empty_children;
833 
834 	/* account for bits == KEYLENGTH case */
835 	if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
836 		used -= KEY_MAX;
837 
838 	/* One child or none, time to drop us from the trie */
839 	return used < 2;
840 }
841 
842 #define MAX_WORK 10
843 static struct key_vector *resize(struct trie *t, struct key_vector *tn)
844 {
845 #ifdef CONFIG_IP_FIB_TRIE_STATS
846 	struct trie_use_stats __percpu *stats = t->stats;
847 #endif
848 	struct key_vector *tp = node_parent(tn);
849 	unsigned long cindex = get_index(tn->key, tp);
850 	int max_work = MAX_WORK;
851 
852 	pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
853 		 tn, inflate_threshold, halve_threshold);
854 
855 	/* track the tnode via the pointer from the parent instead of
856 	 * doing it ourselves.  This way we can let RCU fully do its
857 	 * thing without us interfering
858 	 */
859 	BUG_ON(tn != get_child(tp, cindex));
860 
861 	/* Double as long as the resulting node has a number of
862 	 * nonempty nodes that are above the threshold.
863 	 */
864 	while (should_inflate(tp, tn) && max_work) {
865 		tp = inflate(t, tn);
866 		if (!tp) {
867 #ifdef CONFIG_IP_FIB_TRIE_STATS
868 			this_cpu_inc(stats->resize_node_skipped);
869 #endif
870 			break;
871 		}
872 
873 		max_work--;
874 		tn = get_child(tp, cindex);
875 	}
876 
877 	/* update parent in case inflate failed */
878 	tp = node_parent(tn);
879 
880 	/* Return if at least one inflate is run */
881 	if (max_work != MAX_WORK)
882 		return tp;
883 
884 	/* Halve as long as the number of empty children in this
885 	 * node is above threshold.
886 	 */
887 	while (should_halve(tp, tn) && max_work) {
888 		tp = halve(t, tn);
889 		if (!tp) {
890 #ifdef CONFIG_IP_FIB_TRIE_STATS
891 			this_cpu_inc(stats->resize_node_skipped);
892 #endif
893 			break;
894 		}
895 
896 		max_work--;
897 		tn = get_child(tp, cindex);
898 	}
899 
900 	/* Only one child remains */
901 	if (should_collapse(tn))
902 		return collapse(t, tn);
903 
904 	/* update parent in case halve failed */
905 	return node_parent(tn);
906 }
907 
908 static void node_pull_suffix(struct key_vector *tn, unsigned char slen)
909 {
910 	unsigned char node_slen = tn->slen;
911 
912 	while ((node_slen > tn->pos) && (node_slen > slen)) {
913 		slen = update_suffix(tn);
914 		if (node_slen == slen)
915 			break;
916 
917 		tn = node_parent(tn);
918 		node_slen = tn->slen;
919 	}
920 }
921 
922 static void node_push_suffix(struct key_vector *tn, unsigned char slen)
923 {
924 	while (tn->slen < slen) {
925 		tn->slen = slen;
926 		tn = node_parent(tn);
927 	}
928 }
929 
930 /* rcu_read_lock needs to be hold by caller from readside */
931 static struct key_vector *fib_find_node(struct trie *t,
932 					struct key_vector **tp, u32 key)
933 {
934 	struct key_vector *pn, *n = t->kv;
935 	unsigned long index = 0;
936 
937 	do {
938 		pn = n;
939 		n = get_child_rcu(n, index);
940 
941 		if (!n)
942 			break;
943 
944 		index = get_cindex(key, n);
945 
946 		/* This bit of code is a bit tricky but it combines multiple
947 		 * checks into a single check.  The prefix consists of the
948 		 * prefix plus zeros for the bits in the cindex. The index
949 		 * is the difference between the key and this value.  From
950 		 * this we can actually derive several pieces of data.
951 		 *   if (index >= (1ul << bits))
952 		 *     we have a mismatch in skip bits and failed
953 		 *   else
954 		 *     we know the value is cindex
955 		 *
956 		 * This check is safe even if bits == KEYLENGTH due to the
957 		 * fact that we can only allocate a node with 32 bits if a
958 		 * long is greater than 32 bits.
959 		 */
960 		if (index >= (1ul << n->bits)) {
961 			n = NULL;
962 			break;
963 		}
964 
965 		index = array_index_nospec(index, 1ul << n->bits);
966 
967 		/* keep searching until we find a perfect match leaf or NULL */
968 	} while (IS_TNODE(n));
969 
970 	*tp = pn;
971 
972 	return n;
973 }
974 
975 /* Return the first fib alias matching DSCP with
976  * priority less than or equal to PRIO.
977  * If 'find_first' is set, return the first matching
978  * fib alias, regardless of DSCP and priority.
979  */
980 static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
981 					dscp_t dscp, u32 prio, u32 tb_id,
982 					bool find_first)
983 {
984 	struct fib_alias *fa;
985 
986 	if (!fah)
987 		return NULL;
988 
989 	hlist_for_each_entry(fa, fah, fa_list) {
990 		/* Avoid Sparse warning when using dscp_t in inequalities */
991 		u8 __fa_dscp = inet_dscp_to_dsfield(fa->fa_dscp);
992 		u8 __dscp = inet_dscp_to_dsfield(dscp);
993 
994 		if (fa->fa_slen < slen)
995 			continue;
996 		if (fa->fa_slen != slen)
997 			break;
998 		if (fa->tb_id > tb_id)
999 			continue;
1000 		if (fa->tb_id != tb_id)
1001 			break;
1002 		if (find_first)
1003 			return fa;
1004 		if (__fa_dscp > __dscp)
1005 			continue;
1006 		if (fa->fa_info->fib_priority >= prio || __fa_dscp < __dscp)
1007 			return fa;
1008 	}
1009 
1010 	return NULL;
1011 }
1012 
1013 static struct fib_alias *
1014 fib_find_matching_alias(struct net *net, const struct fib_rt_info *fri)
1015 {
1016 	u8 slen = KEYLENGTH - fri->dst_len;
1017 	struct key_vector *l, *tp;
1018 	struct fib_table *tb;
1019 	struct fib_alias *fa;
1020 	struct trie *t;
1021 
1022 	tb = fib_get_table(net, fri->tb_id);
1023 	if (!tb)
1024 		return NULL;
1025 
1026 	t = (struct trie *)tb->tb_data;
1027 	l = fib_find_node(t, &tp, be32_to_cpu(fri->dst));
1028 	if (!l)
1029 		return NULL;
1030 
1031 	hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1032 		if (fa->fa_slen == slen && fa->tb_id == fri->tb_id &&
1033 		    fa->fa_dscp == fri->dscp && fa->fa_info == fri->fi &&
1034 		    fa->fa_type == fri->type)
1035 			return fa;
1036 	}
1037 
1038 	return NULL;
1039 }
1040 
1041 void fib_alias_hw_flags_set(struct net *net, const struct fib_rt_info *fri)
1042 {
1043 	u8 fib_notify_on_flag_change;
1044 	struct fib_alias *fa_match;
1045 	struct sk_buff *skb;
1046 	int err;
1047 
1048 	rcu_read_lock();
1049 
1050 	fa_match = fib_find_matching_alias(net, fri);
1051 	if (!fa_match)
1052 		goto out;
1053 
1054 	/* These are paired with the WRITE_ONCE() happening in this function.
1055 	 * The reason is that we are only protected by RCU at this point.
1056 	 */
1057 	if (READ_ONCE(fa_match->offload) == fri->offload &&
1058 	    READ_ONCE(fa_match->trap) == fri->trap &&
1059 	    READ_ONCE(fa_match->offload_failed) == fri->offload_failed)
1060 		goto out;
1061 
1062 	WRITE_ONCE(fa_match->offload, fri->offload);
1063 	WRITE_ONCE(fa_match->trap, fri->trap);
1064 
1065 	fib_notify_on_flag_change = READ_ONCE(net->ipv4.sysctl_fib_notify_on_flag_change);
1066 
1067 	/* 2 means send notifications only if offload_failed was changed. */
1068 	if (fib_notify_on_flag_change == 2 &&
1069 	    READ_ONCE(fa_match->offload_failed) == fri->offload_failed)
1070 		goto out;
1071 
1072 	WRITE_ONCE(fa_match->offload_failed, fri->offload_failed);
1073 
1074 	if (!fib_notify_on_flag_change)
1075 		goto out;
1076 
1077 	skb = nlmsg_new(fib_nlmsg_size(fa_match->fa_info), GFP_ATOMIC);
1078 	if (!skb) {
1079 		err = -ENOBUFS;
1080 		goto errout;
1081 	}
1082 
1083 	err = fib_dump_info(skb, 0, 0, RTM_NEWROUTE, fri, 0);
1084 	if (err < 0) {
1085 		kfree_skb(skb);
1086 		goto errout;
1087 	}
1088 
1089 	rtnl_notify(skb, net, 0, RTNLGRP_IPV4_ROUTE, NULL, GFP_ATOMIC);
1090 	goto out;
1091 
1092 errout:
1093 	rtnl_set_sk_err(net, RTNLGRP_IPV4_ROUTE, err);
1094 out:
1095 	rcu_read_unlock();
1096 }
1097 EXPORT_SYMBOL_GPL(fib_alias_hw_flags_set);
1098 
1099 static void trie_rebalance(struct trie *t, struct key_vector *tn)
1100 {
1101 	while (!IS_TRIE(tn))
1102 		tn = resize(t, tn);
1103 }
1104 
1105 static int fib_insert_node(struct trie *t, struct key_vector *tp,
1106 			   struct fib_alias *new, t_key key)
1107 {
1108 	struct key_vector *n, *l;
1109 
1110 	l = leaf_new(key, new);
1111 	if (!l)
1112 		goto noleaf;
1113 
1114 	/* retrieve child from parent node */
1115 	n = get_child(tp, get_index(key, tp));
1116 
1117 	/* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1118 	 *
1119 	 *  Add a new tnode here
1120 	 *  first tnode need some special handling
1121 	 *  leaves us in position for handling as case 3
1122 	 */
1123 	if (n) {
1124 		struct key_vector *tn;
1125 
1126 		tn = tnode_new(key, __fls(key ^ n->key), 1);
1127 		if (!tn)
1128 			goto notnode;
1129 
1130 		/* initialize routes out of node */
1131 		NODE_INIT_PARENT(tn, tp);
1132 		put_child(tn, get_index(key, tn) ^ 1, n);
1133 
1134 		/* start adding routes into the node */
1135 		put_child_root(tp, key, tn);
1136 		node_set_parent(n, tn);
1137 
1138 		/* parent now has a NULL spot where the leaf can go */
1139 		tp = tn;
1140 	}
1141 
1142 	/* Case 3: n is NULL, and will just insert a new leaf */
1143 	node_push_suffix(tp, new->fa_slen);
1144 	NODE_INIT_PARENT(l, tp);
1145 	put_child_root(tp, key, l);
1146 	trie_rebalance(t, tp);
1147 
1148 	return 0;
1149 notnode:
1150 	node_free(l);
1151 noleaf:
1152 	return -ENOMEM;
1153 }
1154 
1155 static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1156 			    struct key_vector *l, struct fib_alias *new,
1157 			    struct fib_alias *fa, t_key key)
1158 {
1159 	if (!l)
1160 		return fib_insert_node(t, tp, new, key);
1161 
1162 	if (fa) {
1163 		hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1164 	} else {
1165 		struct fib_alias *last;
1166 
1167 		hlist_for_each_entry(last, &l->leaf, fa_list) {
1168 			if (new->fa_slen < last->fa_slen)
1169 				break;
1170 			if ((new->fa_slen == last->fa_slen) &&
1171 			    (new->tb_id > last->tb_id))
1172 				break;
1173 			fa = last;
1174 		}
1175 
1176 		if (fa)
1177 			hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1178 		else
1179 			hlist_add_head_rcu(&new->fa_list, &l->leaf);
1180 	}
1181 
1182 	/* if we added to the tail node then we need to update slen */
1183 	if (l->slen < new->fa_slen) {
1184 		l->slen = new->fa_slen;
1185 		node_push_suffix(tp, new->fa_slen);
1186 	}
1187 
1188 	return 0;
1189 }
1190 
1191 static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1192 			     struct key_vector *l, struct fib_alias *old);
1193 
1194 /* Caller must hold RTNL. */
1195 int fib_table_insert(struct net *net, struct fib_table *tb,
1196 		     struct fib_config *cfg, struct netlink_ext_ack *extack)
1197 {
1198 	struct trie *t = (struct trie *)tb->tb_data;
1199 	struct fib_alias *fa, *new_fa;
1200 	struct key_vector *l, *tp;
1201 	u16 nlflags = NLM_F_EXCL;
1202 	struct fib_info *fi;
1203 	u8 plen = cfg->fc_dst_len;
1204 	u8 slen = KEYLENGTH - plen;
1205 	dscp_t dscp;
1206 	u32 key;
1207 	int err;
1208 
1209 	key = ntohl(cfg->fc_dst);
1210 
1211 	pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1212 
1213 	fi = fib_create_info(cfg, extack);
1214 	if (IS_ERR(fi)) {
1215 		err = PTR_ERR(fi);
1216 		goto err;
1217 	}
1218 
1219 	dscp = cfg->fc_dscp;
1220 	l = fib_find_node(t, &tp, key);
1221 	fa = l ? fib_find_alias(&l->leaf, slen, dscp, fi->fib_priority,
1222 				tb->tb_id, false) : NULL;
1223 
1224 	/* Now fa, if non-NULL, points to the first fib alias
1225 	 * with the same keys [prefix,dscp,priority], if such key already
1226 	 * exists or to the node before which we will insert new one.
1227 	 *
1228 	 * If fa is NULL, we will need to allocate a new one and
1229 	 * insert to the tail of the section matching the suffix length
1230 	 * of the new alias.
1231 	 */
1232 
1233 	if (fa && fa->fa_dscp == dscp &&
1234 	    fa->fa_info->fib_priority == fi->fib_priority) {
1235 		struct fib_alias *fa_first, *fa_match;
1236 
1237 		err = -EEXIST;
1238 		if (cfg->fc_nlflags & NLM_F_EXCL)
1239 			goto out;
1240 
1241 		nlflags &= ~NLM_F_EXCL;
1242 
1243 		/* We have 2 goals:
1244 		 * 1. Find exact match for type, scope, fib_info to avoid
1245 		 * duplicate routes
1246 		 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1247 		 */
1248 		fa_match = NULL;
1249 		fa_first = fa;
1250 		hlist_for_each_entry_from(fa, fa_list) {
1251 			if ((fa->fa_slen != slen) ||
1252 			    (fa->tb_id != tb->tb_id) ||
1253 			    (fa->fa_dscp != dscp))
1254 				break;
1255 			if (fa->fa_info->fib_priority != fi->fib_priority)
1256 				break;
1257 			if (fa->fa_type == cfg->fc_type &&
1258 			    fa->fa_info == fi) {
1259 				fa_match = fa;
1260 				break;
1261 			}
1262 		}
1263 
1264 		if (cfg->fc_nlflags & NLM_F_REPLACE) {
1265 			struct fib_info *fi_drop;
1266 			u8 state;
1267 
1268 			nlflags |= NLM_F_REPLACE;
1269 			fa = fa_first;
1270 			if (fa_match) {
1271 				if (fa == fa_match)
1272 					err = 0;
1273 				goto out;
1274 			}
1275 			err = -ENOBUFS;
1276 			new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1277 			if (!new_fa)
1278 				goto out;
1279 
1280 			fi_drop = fa->fa_info;
1281 			new_fa->fa_dscp = fa->fa_dscp;
1282 			new_fa->fa_info = fi;
1283 			new_fa->fa_type = cfg->fc_type;
1284 			state = READ_ONCE(fa->fa_state);
1285 			new_fa->fa_state = state & ~FA_S_ACCESSED;
1286 			new_fa->fa_slen = fa->fa_slen;
1287 			new_fa->tb_id = tb->tb_id;
1288 			new_fa->fa_default = -1;
1289 			new_fa->offload = 0;
1290 			new_fa->trap = 0;
1291 			new_fa->offload_failed = 0;
1292 
1293 			hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1294 
1295 			if (fib_find_alias(&l->leaf, fa->fa_slen, 0, 0,
1296 					   tb->tb_id, true) == new_fa) {
1297 				enum fib_event_type fib_event;
1298 
1299 				fib_event = FIB_EVENT_ENTRY_REPLACE;
1300 				err = call_fib_entry_notifiers(net, fib_event,
1301 							       key, plen,
1302 							       new_fa, extack);
1303 				if (err) {
1304 					hlist_replace_rcu(&new_fa->fa_list,
1305 							  &fa->fa_list);
1306 					goto out_free_new_fa;
1307 				}
1308 			}
1309 
1310 			rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1311 				  tb->tb_id, &cfg->fc_nlinfo, nlflags);
1312 
1313 			alias_free_mem_rcu(fa);
1314 
1315 			fib_release_info(fi_drop);
1316 			if (state & FA_S_ACCESSED)
1317 				rt_cache_flush(cfg->fc_nlinfo.nl_net);
1318 
1319 			goto succeeded;
1320 		}
1321 		/* Error if we find a perfect match which
1322 		 * uses the same scope, type, and nexthop
1323 		 * information.
1324 		 */
1325 		if (fa_match)
1326 			goto out;
1327 
1328 		if (cfg->fc_nlflags & NLM_F_APPEND)
1329 			nlflags |= NLM_F_APPEND;
1330 		else
1331 			fa = fa_first;
1332 	}
1333 	err = -ENOENT;
1334 	if (!(cfg->fc_nlflags & NLM_F_CREATE))
1335 		goto out;
1336 
1337 	nlflags |= NLM_F_CREATE;
1338 	err = -ENOBUFS;
1339 	new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1340 	if (!new_fa)
1341 		goto out;
1342 
1343 	new_fa->fa_info = fi;
1344 	new_fa->fa_dscp = dscp;
1345 	new_fa->fa_type = cfg->fc_type;
1346 	new_fa->fa_state = 0;
1347 	new_fa->fa_slen = slen;
1348 	new_fa->tb_id = tb->tb_id;
1349 	new_fa->fa_default = -1;
1350 	new_fa->offload = 0;
1351 	new_fa->trap = 0;
1352 	new_fa->offload_failed = 0;
1353 
1354 	/* Insert new entry to the list. */
1355 	err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1356 	if (err)
1357 		goto out_free_new_fa;
1358 
1359 	/* The alias was already inserted, so the node must exist. */
1360 	l = l ? l : fib_find_node(t, &tp, key);
1361 	if (WARN_ON_ONCE(!l)) {
1362 		err = -ENOENT;
1363 		goto out_free_new_fa;
1364 	}
1365 
1366 	if (fib_find_alias(&l->leaf, new_fa->fa_slen, 0, 0, tb->tb_id, true) ==
1367 	    new_fa) {
1368 		enum fib_event_type fib_event;
1369 
1370 		fib_event = FIB_EVENT_ENTRY_REPLACE;
1371 		err = call_fib_entry_notifiers(net, fib_event, key, plen,
1372 					       new_fa, extack);
1373 		if (err)
1374 			goto out_remove_new_fa;
1375 	}
1376 
1377 	if (!plen)
1378 		tb->tb_num_default++;
1379 
1380 	rt_cache_flush(cfg->fc_nlinfo.nl_net);
1381 	rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1382 		  &cfg->fc_nlinfo, nlflags);
1383 succeeded:
1384 	return 0;
1385 
1386 out_remove_new_fa:
1387 	fib_remove_alias(t, tp, l, new_fa);
1388 out_free_new_fa:
1389 	alias_free_mem_rcu(new_fa);
1390 out:
1391 	fib_release_info(fi);
1392 err:
1393 	return err;
1394 }
1395 
1396 static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1397 {
1398 	t_key prefix = n->key;
1399 
1400 	return (key ^ prefix) & (prefix | -prefix);
1401 }
1402 
1403 bool fib_lookup_good_nhc(const struct fib_nh_common *nhc, int fib_flags,
1404 			 const struct flowi4 *flp)
1405 {
1406 	if (nhc->nhc_flags & RTNH_F_DEAD)
1407 		return false;
1408 
1409 	if (ip_ignore_linkdown(nhc->nhc_dev) &&
1410 	    nhc->nhc_flags & RTNH_F_LINKDOWN &&
1411 	    !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1412 		return false;
1413 
1414 	if (flp->flowi4_oif && flp->flowi4_oif != nhc->nhc_oif)
1415 		return false;
1416 
1417 	return true;
1418 }
1419 
1420 /* should be called with rcu_read_lock */
1421 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1422 		     struct fib_result *res, int fib_flags)
1423 {
1424 	struct trie *t = (struct trie *) tb->tb_data;
1425 #ifdef CONFIG_IP_FIB_TRIE_STATS
1426 	struct trie_use_stats __percpu *stats = t->stats;
1427 #endif
1428 	const t_key key = ntohl(flp->daddr);
1429 	struct key_vector *n, *pn;
1430 	struct fib_alias *fa;
1431 	unsigned long index;
1432 	t_key cindex;
1433 
1434 	pn = t->kv;
1435 	cindex = 0;
1436 
1437 	n = get_child_rcu(pn, cindex);
1438 	if (!n) {
1439 		trace_fib_table_lookup(tb->tb_id, flp, NULL, -EAGAIN,
1440 				       tb->tb_net);
1441 		return -EAGAIN;
1442 	}
1443 
1444 #ifdef CONFIG_IP_FIB_TRIE_STATS
1445 	this_cpu_inc(stats->gets);
1446 #endif
1447 
1448 	/* Step 1: Travel to the longest prefix match in the trie */
1449 	for (;;) {
1450 		index = get_cindex(key, n);
1451 
1452 		/* This bit of code is a bit tricky but it combines multiple
1453 		 * checks into a single check.  The prefix consists of the
1454 		 * prefix plus zeros for the "bits" in the prefix. The index
1455 		 * is the difference between the key and this value.  From
1456 		 * this we can actually derive several pieces of data.
1457 		 *   if (index >= (1ul << bits))
1458 		 *     we have a mismatch in skip bits and failed
1459 		 *   else
1460 		 *     we know the value is cindex
1461 		 *
1462 		 * This check is safe even if bits == KEYLENGTH due to the
1463 		 * fact that we can only allocate a node with 32 bits if a
1464 		 * long is greater than 32 bits.
1465 		 */
1466 		if (index >= (1ul << n->bits))
1467 			break;
1468 
1469 		/* we have found a leaf. Prefixes have already been compared */
1470 		if (IS_LEAF(n))
1471 			goto found;
1472 
1473 		index = array_index_nospec(index, 1ul << n->bits);
1474 
1475 		/* only record pn and cindex if we are going to be chopping
1476 		 * bits later.  Otherwise we are just wasting cycles.
1477 		 */
1478 		if (n->slen > n->pos) {
1479 			pn = n;
1480 			cindex = index;
1481 		}
1482 
1483 		n = get_child_rcu(n, index);
1484 		if (unlikely(!n))
1485 			goto backtrace;
1486 	}
1487 
1488 	/* Step 2: Sort out leaves and begin backtracing for longest prefix */
1489 	for (;;) {
1490 		/* record the pointer where our next node pointer is stored */
1491 		struct key_vector __rcu **cptr = n->tnode;
1492 
1493 		/* This test verifies that none of the bits that differ
1494 		 * between the key and the prefix exist in the region of
1495 		 * the lsb and higher in the prefix.
1496 		 */
1497 		if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1498 			goto backtrace;
1499 
1500 		/* exit out and process leaf */
1501 		if (unlikely(IS_LEAF(n)))
1502 			break;
1503 
1504 		/* Don't bother recording parent info.  Since we are in
1505 		 * prefix match mode we will have to come back to wherever
1506 		 * we started this traversal anyway
1507 		 */
1508 
1509 		while ((n = rcu_dereference(*cptr)) == NULL) {
1510 backtrace:
1511 #ifdef CONFIG_IP_FIB_TRIE_STATS
1512 			if (!n)
1513 				this_cpu_inc(stats->null_node_hit);
1514 #endif
1515 			/* If we are at cindex 0 there are no more bits for
1516 			 * us to strip at this level so we must ascend back
1517 			 * up one level to see if there are any more bits to
1518 			 * be stripped there.
1519 			 */
1520 			while (!cindex) {
1521 				t_key pkey = pn->key;
1522 
1523 				/* If we don't have a parent then there is
1524 				 * nothing for us to do as we do not have any
1525 				 * further nodes to parse.
1526 				 */
1527 				if (IS_TRIE(pn)) {
1528 					trace_fib_table_lookup(tb->tb_id, flp,
1529 							       NULL, -EAGAIN,
1530 							       tb->tb_net);
1531 					return -EAGAIN;
1532 				}
1533 #ifdef CONFIG_IP_FIB_TRIE_STATS
1534 				this_cpu_inc(stats->backtrack);
1535 #endif
1536 				/* Get Child's index */
1537 				pn = node_parent_rcu(pn);
1538 				cindex = get_index(pkey, pn);
1539 			}
1540 
1541 			/* strip the least significant bit from the cindex */
1542 			cindex &= cindex - 1;
1543 
1544 			/* grab pointer for next child node */
1545 			cptr = &pn->tnode[cindex];
1546 		}
1547 	}
1548 
1549 found:
1550 	/* this line carries forward the xor from earlier in the function */
1551 	index = key ^ n->key;
1552 
1553 	/* Step 3: Process the leaf, if that fails fall back to backtracing */
1554 	hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1555 		struct fib_info *fi = fa->fa_info;
1556 		struct fib_nh_common *nhc;
1557 		int nhsel, err;
1558 
1559 		if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1560 			if (index >= (1ul << fa->fa_slen))
1561 				continue;
1562 		}
1563 		if (fa->fa_dscp && !fib_dscp_masked_match(fa->fa_dscp, flp))
1564 			continue;
1565 		/* Paired with WRITE_ONCE() in fib_release_info() */
1566 		if (READ_ONCE(fi->fib_dead))
1567 			continue;
1568 		if (fa->fa_info->fib_scope < flp->flowi4_scope)
1569 			continue;
1570 		fib_alias_accessed(fa);
1571 		err = fib_props[fa->fa_type].error;
1572 		if (unlikely(err < 0)) {
1573 out_reject:
1574 #ifdef CONFIG_IP_FIB_TRIE_STATS
1575 			this_cpu_inc(stats->semantic_match_passed);
1576 #endif
1577 			trace_fib_table_lookup(tb->tb_id, flp, NULL, err,
1578 					       tb->tb_net);
1579 			return err;
1580 		}
1581 		if (fi->fib_flags & RTNH_F_DEAD)
1582 			continue;
1583 
1584 		if (unlikely(fi->nh)) {
1585 			if (nexthop_is_blackhole(fi->nh)) {
1586 				err = fib_props[RTN_BLACKHOLE].error;
1587 				goto out_reject;
1588 			}
1589 
1590 			nhc = nexthop_get_nhc_lookup(fi->nh, fib_flags, flp,
1591 						     &nhsel);
1592 			if (nhc)
1593 				goto set_result;
1594 			goto miss;
1595 		}
1596 
1597 		for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
1598 			nhc = fib_info_nhc(fi, nhsel);
1599 
1600 			if (!fib_lookup_good_nhc(nhc, fib_flags, flp))
1601 				continue;
1602 set_result:
1603 			if (!(fib_flags & FIB_LOOKUP_NOREF))
1604 				refcount_inc(&fi->fib_clntref);
1605 
1606 			res->prefix = htonl(n->key);
1607 			res->prefixlen = KEYLENGTH - fa->fa_slen;
1608 			res->nh_sel = nhsel;
1609 			res->nhc = nhc;
1610 			res->type = fa->fa_type;
1611 			res->scope = fi->fib_scope;
1612 			res->dscp = fa->fa_dscp;
1613 			res->fi = fi;
1614 			res->table = tb;
1615 			res->fa_head = &n->leaf;
1616 #ifdef CONFIG_IP_FIB_TRIE_STATS
1617 			this_cpu_inc(stats->semantic_match_passed);
1618 #endif
1619 			trace_fib_table_lookup(tb->tb_id, flp, nhc, err,
1620 					       tb->tb_net);
1621 
1622 			return err;
1623 		}
1624 	}
1625 miss:
1626 #ifdef CONFIG_IP_FIB_TRIE_STATS
1627 	this_cpu_inc(stats->semantic_match_miss);
1628 #endif
1629 	goto backtrace;
1630 }
1631 EXPORT_SYMBOL_GPL(fib_table_lookup);
1632 
1633 static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1634 			     struct key_vector *l, struct fib_alias *old)
1635 {
1636 	/* record the location of the previous list_info entry */
1637 	struct hlist_node **pprev = old->fa_list.pprev;
1638 	struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1639 
1640 	/* remove the fib_alias from the list */
1641 	hlist_del_rcu(&old->fa_list);
1642 
1643 	/* if we emptied the list this leaf will be freed and we can sort
1644 	 * out parent suffix lengths as a part of trie_rebalance
1645 	 */
1646 	if (hlist_empty(&l->leaf)) {
1647 		if (tp->slen == l->slen)
1648 			node_pull_suffix(tp, tp->pos);
1649 		put_child_root(tp, l->key, NULL);
1650 		node_free(l);
1651 		trie_rebalance(t, tp);
1652 		return;
1653 	}
1654 
1655 	/* only access fa if it is pointing at the last valid hlist_node */
1656 	if (*pprev)
1657 		return;
1658 
1659 	/* update the trie with the latest suffix length */
1660 	l->slen = fa->fa_slen;
1661 	node_pull_suffix(tp, fa->fa_slen);
1662 }
1663 
1664 static void fib_notify_alias_delete(struct net *net, u32 key,
1665 				    struct hlist_head *fah,
1666 				    struct fib_alias *fa_to_delete,
1667 				    struct netlink_ext_ack *extack)
1668 {
1669 	struct fib_alias *fa_next, *fa_to_notify;
1670 	u32 tb_id = fa_to_delete->tb_id;
1671 	u8 slen = fa_to_delete->fa_slen;
1672 	enum fib_event_type fib_event;
1673 
1674 	/* Do not notify if we do not care about the route. */
1675 	if (fib_find_alias(fah, slen, 0, 0, tb_id, true) != fa_to_delete)
1676 		return;
1677 
1678 	/* Determine if the route should be replaced by the next route in the
1679 	 * list.
1680 	 */
1681 	fa_next = hlist_entry_safe(fa_to_delete->fa_list.next,
1682 				   struct fib_alias, fa_list);
1683 	if (fa_next && fa_next->fa_slen == slen && fa_next->tb_id == tb_id) {
1684 		fib_event = FIB_EVENT_ENTRY_REPLACE;
1685 		fa_to_notify = fa_next;
1686 	} else {
1687 		fib_event = FIB_EVENT_ENTRY_DEL;
1688 		fa_to_notify = fa_to_delete;
1689 	}
1690 	call_fib_entry_notifiers(net, fib_event, key, KEYLENGTH - slen,
1691 				 fa_to_notify, extack);
1692 }
1693 
1694 /* Caller must hold RTNL. */
1695 int fib_table_delete(struct net *net, struct fib_table *tb,
1696 		     struct fib_config *cfg, struct netlink_ext_ack *extack)
1697 {
1698 	struct trie *t = (struct trie *) tb->tb_data;
1699 	struct fib_alias *fa, *fa_to_delete;
1700 	struct key_vector *l, *tp;
1701 	u8 plen = cfg->fc_dst_len;
1702 	u8 slen = KEYLENGTH - plen;
1703 	dscp_t dscp;
1704 	u32 key;
1705 
1706 	key = ntohl(cfg->fc_dst);
1707 
1708 	l = fib_find_node(t, &tp, key);
1709 	if (!l)
1710 		return -ESRCH;
1711 
1712 	dscp = cfg->fc_dscp;
1713 	fa = fib_find_alias(&l->leaf, slen, dscp, 0, tb->tb_id, false);
1714 	if (!fa)
1715 		return -ESRCH;
1716 
1717 	pr_debug("Deleting %08x/%d dsfield=0x%02x t=%p\n", key, plen,
1718 		 inet_dscp_to_dsfield(dscp), t);
1719 
1720 	fa_to_delete = NULL;
1721 	hlist_for_each_entry_from(fa, fa_list) {
1722 		struct fib_info *fi = fa->fa_info;
1723 
1724 		if ((fa->fa_slen != slen) ||
1725 		    (fa->tb_id != tb->tb_id) ||
1726 		    (fa->fa_dscp != dscp))
1727 			break;
1728 
1729 		if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1730 		    (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1731 		     fa->fa_info->fib_scope == cfg->fc_scope) &&
1732 		    (!cfg->fc_prefsrc ||
1733 		     fi->fib_prefsrc == cfg->fc_prefsrc) &&
1734 		    (!cfg->fc_protocol ||
1735 		     fi->fib_protocol == cfg->fc_protocol) &&
1736 		    fib_nh_match(net, cfg, fi, extack) == 0 &&
1737 		    fib_metrics_match(cfg, fi)) {
1738 			fa_to_delete = fa;
1739 			break;
1740 		}
1741 	}
1742 
1743 	if (!fa_to_delete)
1744 		return -ESRCH;
1745 
1746 	fib_notify_alias_delete(net, key, &l->leaf, fa_to_delete, extack);
1747 	rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1748 		  &cfg->fc_nlinfo, 0);
1749 
1750 	if (!plen)
1751 		tb->tb_num_default--;
1752 
1753 	fib_remove_alias(t, tp, l, fa_to_delete);
1754 
1755 	if (READ_ONCE(fa_to_delete->fa_state) & FA_S_ACCESSED)
1756 		rt_cache_flush(cfg->fc_nlinfo.nl_net);
1757 
1758 	fib_release_info(fa_to_delete->fa_info);
1759 	alias_free_mem_rcu(fa_to_delete);
1760 	return 0;
1761 }
1762 
1763 /* Scan for the next leaf starting at the provided key value */
1764 static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1765 {
1766 	struct key_vector *pn, *n = *tn;
1767 	unsigned long cindex;
1768 
1769 	/* this loop is meant to try and find the key in the trie */
1770 	do {
1771 		/* record parent and next child index */
1772 		pn = n;
1773 		cindex = (key > pn->key) ? get_index(key, pn) : 0;
1774 
1775 		if (cindex >> pn->bits)
1776 			break;
1777 
1778 		cindex = array_index_nospec(cindex, 1ul << pn->bits);
1779 
1780 		/* descend into the next child */
1781 		n = get_child_rcu(pn, cindex++);
1782 		if (!n)
1783 			break;
1784 
1785 		/* guarantee forward progress on the keys */
1786 		if (IS_LEAF(n) && (n->key >= key))
1787 			goto found;
1788 	} while (IS_TNODE(n));
1789 
1790 	/* this loop will search for the next leaf with a greater key */
1791 	while (!IS_TRIE(pn)) {
1792 		/* if we exhausted the parent node we will need to climb */
1793 		if (cindex >= (1ul << pn->bits)) {
1794 			t_key pkey = pn->key;
1795 
1796 			pn = node_parent_rcu(pn);
1797 			cindex = get_index(pkey, pn) + 1;
1798 			continue;
1799 		}
1800 
1801 		cindex = array_index_nospec(cindex, 1ul << pn->bits);
1802 
1803 		/* grab the next available node */
1804 		n = get_child_rcu(pn, cindex++);
1805 		if (!n)
1806 			continue;
1807 
1808 		/* no need to compare keys since we bumped the index */
1809 		if (IS_LEAF(n))
1810 			goto found;
1811 
1812 		/* Rescan start scanning in new node */
1813 		pn = n;
1814 		cindex = 0;
1815 	}
1816 
1817 	*tn = pn;
1818 	return NULL; /* Root of trie */
1819 found:
1820 	/* if we are at the limit for keys just return NULL for the tnode */
1821 	*tn = pn;
1822 	return n;
1823 }
1824 
1825 static void fib_trie_free(struct fib_table *tb)
1826 {
1827 	struct trie *t = (struct trie *)tb->tb_data;
1828 	struct key_vector *pn = t->kv;
1829 	unsigned long cindex = 1;
1830 	struct hlist_node *tmp;
1831 	struct fib_alias *fa;
1832 
1833 	/* walk trie in reverse order and free everything */
1834 	for (;;) {
1835 		struct key_vector *n;
1836 
1837 		if (!(cindex--)) {
1838 			t_key pkey = pn->key;
1839 
1840 			if (IS_TRIE(pn))
1841 				break;
1842 
1843 			n = pn;
1844 			pn = node_parent(pn);
1845 
1846 			/* drop emptied tnode */
1847 			put_child_root(pn, n->key, NULL);
1848 			node_free(n);
1849 
1850 			cindex = get_index(pkey, pn);
1851 
1852 			continue;
1853 		}
1854 
1855 		/* grab the next available node */
1856 		n = get_child(pn, cindex);
1857 		if (!n)
1858 			continue;
1859 
1860 		if (IS_TNODE(n)) {
1861 			/* record pn and cindex for leaf walking */
1862 			pn = n;
1863 			cindex = 1ul << n->bits;
1864 
1865 			continue;
1866 		}
1867 
1868 		hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1869 			hlist_del_rcu(&fa->fa_list);
1870 			alias_free_mem_rcu(fa);
1871 		}
1872 
1873 		put_child_root(pn, n->key, NULL);
1874 		node_free(n);
1875 	}
1876 
1877 #ifdef CONFIG_IP_FIB_TRIE_STATS
1878 	free_percpu(t->stats);
1879 #endif
1880 	kfree(tb);
1881 }
1882 
1883 struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1884 {
1885 	struct trie *ot = (struct trie *)oldtb->tb_data;
1886 	struct key_vector *l, *tp = ot->kv;
1887 	struct fib_table *local_tb;
1888 	struct fib_alias *fa;
1889 	struct trie *lt;
1890 	t_key key = 0;
1891 
1892 	if (oldtb->tb_data == oldtb->__data)
1893 		return oldtb;
1894 
1895 	local_tb = fib_trie_table(oldtb->tb_net, RT_TABLE_LOCAL, NULL);
1896 	if (!local_tb)
1897 		return NULL;
1898 
1899 	lt = (struct trie *)local_tb->tb_data;
1900 
1901 	while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1902 		struct key_vector *local_l = NULL, *local_tp;
1903 
1904 		hlist_for_each_entry(fa, &l->leaf, fa_list) {
1905 			struct fib_alias *new_fa;
1906 
1907 			if (local_tb->tb_id != fa->tb_id)
1908 				continue;
1909 
1910 			/* clone fa for new local table */
1911 			new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1912 			if (!new_fa)
1913 				goto out;
1914 
1915 			memcpy(new_fa, fa, sizeof(*fa));
1916 
1917 			/* insert clone into table */
1918 			if (!local_l)
1919 				local_l = fib_find_node(lt, &local_tp, l->key);
1920 
1921 			if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1922 					     NULL, l->key)) {
1923 				kmem_cache_free(fn_alias_kmem, new_fa);
1924 				goto out;
1925 			}
1926 		}
1927 
1928 		/* stop loop if key wrapped back to 0 */
1929 		key = l->key + 1;
1930 		if (key < l->key)
1931 			break;
1932 	}
1933 
1934 	return local_tb;
1935 out:
1936 	fib_trie_free(local_tb);
1937 
1938 	return NULL;
1939 }
1940 
1941 /* Caller must hold RTNL */
1942 void fib_table_flush_external(struct fib_table *tb)
1943 {
1944 	struct trie *t = (struct trie *)tb->tb_data;
1945 	struct key_vector *pn = t->kv;
1946 	unsigned long cindex = 1;
1947 	struct hlist_node *tmp;
1948 	struct fib_alias *fa;
1949 
1950 	/* walk trie in reverse order */
1951 	for (;;) {
1952 		unsigned char slen = 0;
1953 		struct key_vector *n;
1954 
1955 		if (!(cindex--)) {
1956 			t_key pkey = pn->key;
1957 
1958 			/* cannot resize the trie vector */
1959 			if (IS_TRIE(pn))
1960 				break;
1961 
1962 			/* update the suffix to address pulled leaves */
1963 			if (pn->slen > pn->pos)
1964 				update_suffix(pn);
1965 
1966 			/* resize completed node */
1967 			pn = resize(t, pn);
1968 			cindex = get_index(pkey, pn);
1969 
1970 			continue;
1971 		}
1972 
1973 		/* grab the next available node */
1974 		n = get_child(pn, cindex);
1975 		if (!n)
1976 			continue;
1977 
1978 		if (IS_TNODE(n)) {
1979 			/* record pn and cindex for leaf walking */
1980 			pn = n;
1981 			cindex = 1ul << n->bits;
1982 
1983 			continue;
1984 		}
1985 
1986 		hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1987 			/* if alias was cloned to local then we just
1988 			 * need to remove the local copy from main
1989 			 */
1990 			if (tb->tb_id != fa->tb_id) {
1991 				hlist_del_rcu(&fa->fa_list);
1992 				alias_free_mem_rcu(fa);
1993 				continue;
1994 			}
1995 
1996 			/* record local slen */
1997 			slen = fa->fa_slen;
1998 		}
1999 
2000 		/* update leaf slen */
2001 		n->slen = slen;
2002 
2003 		if (hlist_empty(&n->leaf)) {
2004 			put_child_root(pn, n->key, NULL);
2005 			node_free(n);
2006 		}
2007 	}
2008 }
2009 
2010 /* Caller must hold RTNL. */
2011 int fib_table_flush(struct net *net, struct fib_table *tb, bool flush_all)
2012 {
2013 	struct trie *t = (struct trie *)tb->tb_data;
2014 	struct nl_info info = { .nl_net = net };
2015 	struct key_vector *pn = t->kv;
2016 	unsigned long cindex = 1;
2017 	struct hlist_node *tmp;
2018 	struct fib_alias *fa;
2019 	int found = 0;
2020 
2021 	/* walk trie in reverse order */
2022 	for (;;) {
2023 		unsigned char slen = 0;
2024 		struct key_vector *n;
2025 
2026 		if (!(cindex--)) {
2027 			t_key pkey = pn->key;
2028 
2029 			/* cannot resize the trie vector */
2030 			if (IS_TRIE(pn))
2031 				break;
2032 
2033 			/* update the suffix to address pulled leaves */
2034 			if (pn->slen > pn->pos)
2035 				update_suffix(pn);
2036 
2037 			/* resize completed node */
2038 			pn = resize(t, pn);
2039 			cindex = get_index(pkey, pn);
2040 
2041 			continue;
2042 		}
2043 
2044 		/* grab the next available node */
2045 		n = get_child(pn, cindex);
2046 		if (!n)
2047 			continue;
2048 
2049 		if (IS_TNODE(n)) {
2050 			/* record pn and cindex for leaf walking */
2051 			pn = n;
2052 			cindex = 1ul << n->bits;
2053 
2054 			continue;
2055 		}
2056 
2057 		hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
2058 			struct fib_info *fi = fa->fa_info;
2059 
2060 			if (!fi || tb->tb_id != fa->tb_id) {
2061 				slen = fa->fa_slen;
2062 				continue;
2063 			}
2064 
2065 			if (!flush_all && !(fi->fib_flags & RTNH_F_DEAD)) {
2066 				slen = fa->fa_slen;
2067 				continue;
2068 			}
2069 
2070 			fib_notify_alias_delete(net, n->key, &n->leaf, fa,
2071 						NULL);
2072 			if (fi->pfsrc_removed)
2073 				rtmsg_fib(RTM_DELROUTE, htonl(n->key), fa,
2074 					  KEYLENGTH - fa->fa_slen, tb->tb_id, &info, 0);
2075 			hlist_del_rcu(&fa->fa_list);
2076 			fib_release_info(fa->fa_info);
2077 			alias_free_mem_rcu(fa);
2078 			found++;
2079 		}
2080 
2081 		/* update leaf slen */
2082 		n->slen = slen;
2083 
2084 		if (hlist_empty(&n->leaf)) {
2085 			put_child_root(pn, n->key, NULL);
2086 			node_free(n);
2087 		}
2088 	}
2089 
2090 	pr_debug("trie_flush found=%d\n", found);
2091 	return found;
2092 }
2093 
2094 /* derived from fib_trie_free */
2095 static void __fib_info_notify_update(struct net *net, struct fib_table *tb,
2096 				     struct nl_info *info)
2097 {
2098 	struct trie *t = (struct trie *)tb->tb_data;
2099 	struct key_vector *pn = t->kv;
2100 	unsigned long cindex = 1;
2101 	struct fib_alias *fa;
2102 
2103 	for (;;) {
2104 		struct key_vector *n;
2105 
2106 		if (!(cindex--)) {
2107 			t_key pkey = pn->key;
2108 
2109 			if (IS_TRIE(pn))
2110 				break;
2111 
2112 			pn = node_parent(pn);
2113 			cindex = get_index(pkey, pn);
2114 			continue;
2115 		}
2116 
2117 		/* grab the next available node */
2118 		n = get_child(pn, cindex);
2119 		if (!n)
2120 			continue;
2121 
2122 		if (IS_TNODE(n)) {
2123 			/* record pn and cindex for leaf walking */
2124 			pn = n;
2125 			cindex = 1ul << n->bits;
2126 
2127 			continue;
2128 		}
2129 
2130 		hlist_for_each_entry(fa, &n->leaf, fa_list) {
2131 			struct fib_info *fi = fa->fa_info;
2132 
2133 			if (!fi || !fi->nh_updated || fa->tb_id != tb->tb_id)
2134 				continue;
2135 
2136 			rtmsg_fib(RTM_NEWROUTE, htonl(n->key), fa,
2137 				  KEYLENGTH - fa->fa_slen, tb->tb_id,
2138 				  info, NLM_F_REPLACE);
2139 		}
2140 	}
2141 }
2142 
2143 void fib_info_notify_update(struct net *net, struct nl_info *info)
2144 {
2145 	unsigned int h;
2146 
2147 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2148 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2149 		struct fib_table *tb;
2150 
2151 		hlist_for_each_entry_rcu(tb, head, tb_hlist, true)
2152 			__fib_info_notify_update(net, tb, info);
2153 	}
2154 }
2155 
2156 static int fib_leaf_notify(struct key_vector *l, struct fib_table *tb,
2157 			   struct notifier_block *nb,
2158 			   struct netlink_ext_ack *extack)
2159 {
2160 	struct fib_alias *fa;
2161 	int last_slen = -1;
2162 	int err;
2163 
2164 	hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2165 		struct fib_info *fi = fa->fa_info;
2166 
2167 		if (!fi)
2168 			continue;
2169 
2170 		/* local and main table can share the same trie,
2171 		 * so don't notify twice for the same entry.
2172 		 */
2173 		if (tb->tb_id != fa->tb_id)
2174 			continue;
2175 
2176 		if (fa->fa_slen == last_slen)
2177 			continue;
2178 
2179 		if (!fib_info_hold_safe(fa->fa_info))
2180 			continue;
2181 
2182 		last_slen = fa->fa_slen;
2183 		err = call_fib_entry_notifier(nb, FIB_EVENT_ENTRY_REPLACE,
2184 					      l->key, KEYLENGTH - fa->fa_slen,
2185 					      fa, extack);
2186 		fib_info_put(fa->fa_info);
2187 		if (err)
2188 			return err;
2189 	}
2190 	return 0;
2191 }
2192 
2193 static int fib_table_notify(struct fib_table *tb, struct notifier_block *nb,
2194 			    struct netlink_ext_ack *extack)
2195 {
2196 	struct trie *t = (struct trie *)tb->tb_data;
2197 	struct key_vector *l, *tp = t->kv;
2198 	t_key key = 0;
2199 	int err;
2200 
2201 	while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2202 		err = fib_leaf_notify(l, tb, nb, extack);
2203 		if (err)
2204 			return err;
2205 
2206 		key = l->key + 1;
2207 		/* stop in case of wrap around */
2208 		if (key < l->key)
2209 			break;
2210 	}
2211 	return 0;
2212 }
2213 
2214 int fib_notify(struct net *net, struct notifier_block *nb,
2215 	       struct netlink_ext_ack *extack)
2216 {
2217 	unsigned int h;
2218 	int err;
2219 
2220 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2221 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2222 		struct fib_table *tb;
2223 
2224 		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2225 			err = fib_table_notify(tb, nb, extack);
2226 			if (err)
2227 				return err;
2228 		}
2229 	}
2230 	return 0;
2231 }
2232 
2233 static void __trie_free_rcu(struct rcu_head *head)
2234 {
2235 	struct fib_table *tb = container_of(head, struct fib_table, rcu);
2236 #ifdef CONFIG_IP_FIB_TRIE_STATS
2237 	struct trie *t = (struct trie *)tb->tb_data;
2238 
2239 	if (tb->tb_data == tb->__data)
2240 		free_percpu(t->stats);
2241 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2242 	kfree(tb);
2243 }
2244 
2245 void fib_free_table(struct fib_table *tb)
2246 {
2247 	call_rcu(&tb->rcu, __trie_free_rcu);
2248 }
2249 
2250 static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
2251 			     struct sk_buff *skb, struct netlink_callback *cb,
2252 			     struct fib_dump_filter *filter)
2253 {
2254 	unsigned int flags = NLM_F_MULTI;
2255 	__be32 xkey = htonl(l->key);
2256 	int i, s_i, i_fa, s_fa, err;
2257 	struct fib_alias *fa;
2258 
2259 	if (filter->filter_set ||
2260 	    !filter->dump_exceptions || !filter->dump_routes)
2261 		flags |= NLM_F_DUMP_FILTERED;
2262 
2263 	s_i = cb->args[4];
2264 	s_fa = cb->args[5];
2265 	i = 0;
2266 
2267 	/* rcu_read_lock is hold by caller */
2268 	hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2269 		struct fib_info *fi = fa->fa_info;
2270 
2271 		if (i < s_i)
2272 			goto next;
2273 
2274 		i_fa = 0;
2275 
2276 		if (tb->tb_id != fa->tb_id)
2277 			goto next;
2278 
2279 		if (filter->filter_set) {
2280 			if (filter->rt_type && fa->fa_type != filter->rt_type)
2281 				goto next;
2282 
2283 			if ((filter->protocol &&
2284 			     fi->fib_protocol != filter->protocol))
2285 				goto next;
2286 
2287 			if (filter->dev &&
2288 			    !fib_info_nh_uses_dev(fi, filter->dev))
2289 				goto next;
2290 		}
2291 
2292 		if (filter->dump_routes) {
2293 			if (!s_fa) {
2294 				struct fib_rt_info fri;
2295 
2296 				fri.fi = fi;
2297 				fri.tb_id = tb->tb_id;
2298 				fri.dst = xkey;
2299 				fri.dst_len = KEYLENGTH - fa->fa_slen;
2300 				fri.dscp = fa->fa_dscp;
2301 				fri.type = fa->fa_type;
2302 				fri.offload = READ_ONCE(fa->offload);
2303 				fri.trap = READ_ONCE(fa->trap);
2304 				fri.offload_failed = READ_ONCE(fa->offload_failed);
2305 				err = fib_dump_info(skb,
2306 						    NETLINK_CB(cb->skb).portid,
2307 						    cb->nlh->nlmsg_seq,
2308 						    RTM_NEWROUTE, &fri, flags);
2309 				if (err < 0)
2310 					goto stop;
2311 			}
2312 
2313 			i_fa++;
2314 		}
2315 
2316 		if (filter->dump_exceptions) {
2317 			err = fib_dump_info_fnhe(skb, cb, tb->tb_id, fi,
2318 						 &i_fa, s_fa, flags);
2319 			if (err < 0)
2320 				goto stop;
2321 		}
2322 
2323 next:
2324 		i++;
2325 	}
2326 
2327 	cb->args[4] = i;
2328 	return skb->len;
2329 
2330 stop:
2331 	cb->args[4] = i;
2332 	cb->args[5] = i_fa;
2333 	return err;
2334 }
2335 
2336 /* rcu_read_lock needs to be hold by caller from readside */
2337 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
2338 		   struct netlink_callback *cb, struct fib_dump_filter *filter)
2339 {
2340 	struct trie *t = (struct trie *)tb->tb_data;
2341 	struct key_vector *l, *tp = t->kv;
2342 	/* Dump starting at last key.
2343 	 * Note: 0.0.0.0/0 (ie default) is first key.
2344 	 */
2345 	int count = cb->args[2];
2346 	t_key key = cb->args[3];
2347 
2348 	/* First time here, count and key are both always 0. Count > 0
2349 	 * and key == 0 means the dump has wrapped around and we are done.
2350 	 */
2351 	if (count && !key)
2352 		return 0;
2353 
2354 	while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2355 		int err;
2356 
2357 		err = fn_trie_dump_leaf(l, tb, skb, cb, filter);
2358 		if (err < 0) {
2359 			cb->args[3] = key;
2360 			cb->args[2] = count;
2361 			return err;
2362 		}
2363 
2364 		++count;
2365 		key = l->key + 1;
2366 
2367 		memset(&cb->args[4], 0,
2368 		       sizeof(cb->args) - 4*sizeof(cb->args[0]));
2369 
2370 		/* stop loop if key wrapped back to 0 */
2371 		if (key < l->key)
2372 			break;
2373 	}
2374 
2375 	cb->args[3] = key;
2376 	cb->args[2] = count;
2377 
2378 	return 0;
2379 }
2380 
2381 void __init fib_trie_init(void)
2382 {
2383 	fn_alias_kmem = kmem_cache_create("ip_fib_alias",
2384 					  sizeof(struct fib_alias),
2385 					  0, SLAB_PANIC | SLAB_ACCOUNT, NULL);
2386 
2387 	trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
2388 					   LEAF_SIZE,
2389 					   0, SLAB_PANIC | SLAB_ACCOUNT, NULL);
2390 }
2391 
2392 struct fib_table *fib_trie_table(struct net *net, u32 id,
2393 				 struct fib_table *alias)
2394 {
2395 	struct fib_table *tb;
2396 	struct trie *t;
2397 	size_t sz = sizeof(*tb);
2398 
2399 	if (!alias)
2400 		sz += sizeof(struct trie);
2401 
2402 	tb = kzalloc(sz, GFP_KERNEL);
2403 	if (!tb)
2404 		return NULL;
2405 
2406 	tb->tb_net = net;
2407 	tb->tb_id = id;
2408 	tb->tb_num_default = 0;
2409 	tb->tb_data = (alias ? alias->__data : tb->__data);
2410 
2411 	if (alias)
2412 		return tb;
2413 
2414 	t = (struct trie *) tb->tb_data;
2415 	t->kv[0].pos = KEYLENGTH;
2416 	t->kv[0].slen = KEYLENGTH;
2417 #ifdef CONFIG_IP_FIB_TRIE_STATS
2418 	t->stats = alloc_percpu(struct trie_use_stats);
2419 	if (!t->stats) {
2420 		kfree(tb);
2421 		tb = NULL;
2422 	}
2423 #endif
2424 
2425 	return tb;
2426 }
2427 
2428 #ifdef CONFIG_PROC_FS
2429 /* Depth first Trie walk iterator */
2430 struct fib_trie_iter {
2431 	struct seq_net_private p;
2432 	struct fib_table *tb;
2433 	struct key_vector *tnode;
2434 	unsigned int index;
2435 	unsigned int depth;
2436 };
2437 
2438 static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2439 {
2440 	unsigned long cindex = iter->index;
2441 	struct key_vector *pn = iter->tnode;
2442 	t_key pkey;
2443 
2444 	pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2445 		 iter->tnode, iter->index, iter->depth);
2446 
2447 	while (!IS_TRIE(pn)) {
2448 		while (cindex < child_length(pn)) {
2449 			struct key_vector *n;
2450 
2451 			cindex = array_index_nospec(cindex, child_length(pn));
2452 			n = get_child_rcu(pn, cindex++);
2453 
2454 			if (!n)
2455 				continue;
2456 
2457 			if (IS_LEAF(n)) {
2458 				iter->tnode = pn;
2459 				iter->index = cindex;
2460 			} else {
2461 				/* push down one level */
2462 				iter->tnode = n;
2463 				iter->index = 0;
2464 				++iter->depth;
2465 			}
2466 
2467 			return n;
2468 		}
2469 
2470 		/* Current node exhausted, pop back up */
2471 		pkey = pn->key;
2472 		pn = node_parent_rcu(pn);
2473 		cindex = get_index(pkey, pn) + 1;
2474 		--iter->depth;
2475 	}
2476 
2477 	/* record root node so further searches know we are done */
2478 	iter->tnode = pn;
2479 	iter->index = 0;
2480 
2481 	return NULL;
2482 }
2483 
2484 static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2485 					     struct trie *t)
2486 {
2487 	struct key_vector *n, *pn;
2488 
2489 	if (!t)
2490 		return NULL;
2491 
2492 	pn = t->kv;
2493 	n = rcu_dereference(pn->tnode[0]);
2494 	if (!n)
2495 		return NULL;
2496 
2497 	if (IS_TNODE(n)) {
2498 		iter->tnode = n;
2499 		iter->index = 0;
2500 		iter->depth = 1;
2501 	} else {
2502 		iter->tnode = pn;
2503 		iter->index = 0;
2504 		iter->depth = 0;
2505 	}
2506 
2507 	return n;
2508 }
2509 
2510 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2511 {
2512 	struct key_vector *n;
2513 	struct fib_trie_iter iter;
2514 
2515 	memset(s, 0, sizeof(*s));
2516 
2517 	rcu_read_lock();
2518 	for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2519 		if (IS_LEAF(n)) {
2520 			struct fib_alias *fa;
2521 
2522 			s->leaves++;
2523 			s->totdepth += iter.depth;
2524 			if (iter.depth > s->maxdepth)
2525 				s->maxdepth = iter.depth;
2526 
2527 			hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2528 				++s->prefixes;
2529 		} else {
2530 			s->tnodes++;
2531 			if (n->bits < MAX_STAT_DEPTH)
2532 				s->nodesizes[n->bits]++;
2533 			s->nullpointers += tn_info(n)->empty_children;
2534 		}
2535 	}
2536 	rcu_read_unlock();
2537 }
2538 
2539 /*
2540  *	This outputs /proc/net/fib_triestats
2541  */
2542 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2543 {
2544 	unsigned int i, max, pointers, bytes, avdepth;
2545 
2546 	if (stat->leaves)
2547 		avdepth = stat->totdepth*100 / stat->leaves;
2548 	else
2549 		avdepth = 0;
2550 
2551 	seq_printf(seq, "\tAver depth:     %u.%02d\n",
2552 		   avdepth / 100, avdepth % 100);
2553 	seq_printf(seq, "\tMax depth:      %u\n", stat->maxdepth);
2554 
2555 	seq_printf(seq, "\tLeaves:         %u\n", stat->leaves);
2556 	bytes = LEAF_SIZE * stat->leaves;
2557 
2558 	seq_printf(seq, "\tPrefixes:       %u\n", stat->prefixes);
2559 	bytes += sizeof(struct fib_alias) * stat->prefixes;
2560 
2561 	seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2562 	bytes += TNODE_SIZE(0) * stat->tnodes;
2563 
2564 	max = MAX_STAT_DEPTH;
2565 	while (max > 0 && stat->nodesizes[max-1] == 0)
2566 		max--;
2567 
2568 	pointers = 0;
2569 	for (i = 1; i < max; i++)
2570 		if (stat->nodesizes[i] != 0) {
2571 			seq_printf(seq, "  %u: %u",  i, stat->nodesizes[i]);
2572 			pointers += (1<<i) * stat->nodesizes[i];
2573 		}
2574 	seq_putc(seq, '\n');
2575 	seq_printf(seq, "\tPointers: %u\n", pointers);
2576 
2577 	bytes += sizeof(struct key_vector *) * pointers;
2578 	seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2579 	seq_printf(seq, "Total size: %u  kB\n", (bytes + 1023) / 1024);
2580 }
2581 
2582 #ifdef CONFIG_IP_FIB_TRIE_STATS
2583 static void trie_show_usage(struct seq_file *seq,
2584 			    const struct trie_use_stats __percpu *stats)
2585 {
2586 	struct trie_use_stats s = { 0 };
2587 	int cpu;
2588 
2589 	/* loop through all of the CPUs and gather up the stats */
2590 	for_each_possible_cpu(cpu) {
2591 		const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2592 
2593 		s.gets += pcpu->gets;
2594 		s.backtrack += pcpu->backtrack;
2595 		s.semantic_match_passed += pcpu->semantic_match_passed;
2596 		s.semantic_match_miss += pcpu->semantic_match_miss;
2597 		s.null_node_hit += pcpu->null_node_hit;
2598 		s.resize_node_skipped += pcpu->resize_node_skipped;
2599 	}
2600 
2601 	seq_printf(seq, "\nCounters:\n---------\n");
2602 	seq_printf(seq, "gets = %u\n", s.gets);
2603 	seq_printf(seq, "backtracks = %u\n", s.backtrack);
2604 	seq_printf(seq, "semantic match passed = %u\n",
2605 		   s.semantic_match_passed);
2606 	seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2607 	seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2608 	seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2609 }
2610 #endif /*  CONFIG_IP_FIB_TRIE_STATS */
2611 
2612 static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2613 {
2614 	if (tb->tb_id == RT_TABLE_LOCAL)
2615 		seq_puts(seq, "Local:\n");
2616 	else if (tb->tb_id == RT_TABLE_MAIN)
2617 		seq_puts(seq, "Main:\n");
2618 	else
2619 		seq_printf(seq, "Id %d:\n", tb->tb_id);
2620 }
2621 
2622 
2623 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2624 {
2625 	struct net *net = seq->private;
2626 	unsigned int h;
2627 
2628 	seq_printf(seq,
2629 		   "Basic info: size of leaf:"
2630 		   " %zd bytes, size of tnode: %zd bytes.\n",
2631 		   LEAF_SIZE, TNODE_SIZE(0));
2632 
2633 	rcu_read_lock();
2634 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2635 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2636 		struct fib_table *tb;
2637 
2638 		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2639 			struct trie *t = (struct trie *) tb->tb_data;
2640 			struct trie_stat stat;
2641 
2642 			if (!t)
2643 				continue;
2644 
2645 			fib_table_print(seq, tb);
2646 
2647 			trie_collect_stats(t, &stat);
2648 			trie_show_stats(seq, &stat);
2649 #ifdef CONFIG_IP_FIB_TRIE_STATS
2650 			trie_show_usage(seq, t->stats);
2651 #endif
2652 		}
2653 		cond_resched_rcu();
2654 	}
2655 	rcu_read_unlock();
2656 
2657 	return 0;
2658 }
2659 
2660 static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2661 {
2662 	struct fib_trie_iter *iter = seq->private;
2663 	struct net *net = seq_file_net(seq);
2664 	loff_t idx = 0;
2665 	unsigned int h;
2666 
2667 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2668 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2669 		struct fib_table *tb;
2670 
2671 		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2672 			struct key_vector *n;
2673 
2674 			for (n = fib_trie_get_first(iter,
2675 						    (struct trie *) tb->tb_data);
2676 			     n; n = fib_trie_get_next(iter))
2677 				if (pos == idx++) {
2678 					iter->tb = tb;
2679 					return n;
2680 				}
2681 		}
2682 	}
2683 
2684 	return NULL;
2685 }
2686 
2687 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2688 	__acquires(RCU)
2689 {
2690 	rcu_read_lock();
2691 	return fib_trie_get_idx(seq, *pos);
2692 }
2693 
2694 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2695 {
2696 	struct fib_trie_iter *iter = seq->private;
2697 	struct net *net = seq_file_net(seq);
2698 	struct fib_table *tb = iter->tb;
2699 	struct hlist_node *tb_node;
2700 	unsigned int h;
2701 	struct key_vector *n;
2702 
2703 	++*pos;
2704 	/* next node in same table */
2705 	n = fib_trie_get_next(iter);
2706 	if (n)
2707 		return n;
2708 
2709 	/* walk rest of this hash chain */
2710 	h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2711 	while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2712 		tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2713 		n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2714 		if (n)
2715 			goto found;
2716 	}
2717 
2718 	/* new hash chain */
2719 	while (++h < FIB_TABLE_HASHSZ) {
2720 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2721 		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2722 			n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2723 			if (n)
2724 				goto found;
2725 		}
2726 	}
2727 	return NULL;
2728 
2729 found:
2730 	iter->tb = tb;
2731 	return n;
2732 }
2733 
2734 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2735 	__releases(RCU)
2736 {
2737 	rcu_read_unlock();
2738 }
2739 
2740 static void seq_indent(struct seq_file *seq, int n)
2741 {
2742 	while (n-- > 0)
2743 		seq_puts(seq, "   ");
2744 }
2745 
2746 static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2747 {
2748 	switch (s) {
2749 	case RT_SCOPE_UNIVERSE: return "universe";
2750 	case RT_SCOPE_SITE:	return "site";
2751 	case RT_SCOPE_LINK:	return "link";
2752 	case RT_SCOPE_HOST:	return "host";
2753 	case RT_SCOPE_NOWHERE:	return "nowhere";
2754 	default:
2755 		snprintf(buf, len, "scope=%d", s);
2756 		return buf;
2757 	}
2758 }
2759 
2760 static const char *const rtn_type_names[__RTN_MAX] = {
2761 	[RTN_UNSPEC] = "UNSPEC",
2762 	[RTN_UNICAST] = "UNICAST",
2763 	[RTN_LOCAL] = "LOCAL",
2764 	[RTN_BROADCAST] = "BROADCAST",
2765 	[RTN_ANYCAST] = "ANYCAST",
2766 	[RTN_MULTICAST] = "MULTICAST",
2767 	[RTN_BLACKHOLE] = "BLACKHOLE",
2768 	[RTN_UNREACHABLE] = "UNREACHABLE",
2769 	[RTN_PROHIBIT] = "PROHIBIT",
2770 	[RTN_THROW] = "THROW",
2771 	[RTN_NAT] = "NAT",
2772 	[RTN_XRESOLVE] = "XRESOLVE",
2773 };
2774 
2775 static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2776 {
2777 	if (t < __RTN_MAX && rtn_type_names[t])
2778 		return rtn_type_names[t];
2779 	snprintf(buf, len, "type %u", t);
2780 	return buf;
2781 }
2782 
2783 /* Pretty print the trie */
2784 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2785 {
2786 	const struct fib_trie_iter *iter = seq->private;
2787 	struct key_vector *n = v;
2788 
2789 	if (IS_TRIE(node_parent_rcu(n)))
2790 		fib_table_print(seq, iter->tb);
2791 
2792 	if (IS_TNODE(n)) {
2793 		__be32 prf = htonl(n->key);
2794 
2795 		seq_indent(seq, iter->depth-1);
2796 		seq_printf(seq, "  +-- %pI4/%zu %u %u %u\n",
2797 			   &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2798 			   tn_info(n)->full_children,
2799 			   tn_info(n)->empty_children);
2800 	} else {
2801 		__be32 val = htonl(n->key);
2802 		struct fib_alias *fa;
2803 
2804 		seq_indent(seq, iter->depth);
2805 		seq_printf(seq, "  |-- %pI4\n", &val);
2806 
2807 		hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2808 			char buf1[32], buf2[32];
2809 
2810 			seq_indent(seq, iter->depth + 1);
2811 			seq_printf(seq, "  /%zu %s %s",
2812 				   KEYLENGTH - fa->fa_slen,
2813 				   rtn_scope(buf1, sizeof(buf1),
2814 					     fa->fa_info->fib_scope),
2815 				   rtn_type(buf2, sizeof(buf2),
2816 					    fa->fa_type));
2817 			if (fa->fa_dscp)
2818 				seq_printf(seq, " tos=%d",
2819 					   inet_dscp_to_dsfield(fa->fa_dscp));
2820 			seq_putc(seq, '\n');
2821 		}
2822 	}
2823 
2824 	return 0;
2825 }
2826 
2827 static const struct seq_operations fib_trie_seq_ops = {
2828 	.start  = fib_trie_seq_start,
2829 	.next   = fib_trie_seq_next,
2830 	.stop   = fib_trie_seq_stop,
2831 	.show   = fib_trie_seq_show,
2832 };
2833 
2834 struct fib_route_iter {
2835 	struct seq_net_private p;
2836 	struct fib_table *main_tb;
2837 	struct key_vector *tnode;
2838 	loff_t	pos;
2839 	t_key	key;
2840 };
2841 
2842 static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2843 					    loff_t pos)
2844 {
2845 	struct key_vector *l, **tp = &iter->tnode;
2846 	t_key key;
2847 
2848 	/* use cached location of previously found key */
2849 	if (iter->pos > 0 && pos >= iter->pos) {
2850 		key = iter->key;
2851 	} else {
2852 		iter->pos = 1;
2853 		key = 0;
2854 	}
2855 
2856 	pos -= iter->pos;
2857 
2858 	while ((l = leaf_walk_rcu(tp, key)) && (pos-- > 0)) {
2859 		key = l->key + 1;
2860 		iter->pos++;
2861 		l = NULL;
2862 
2863 		/* handle unlikely case of a key wrap */
2864 		if (!key)
2865 			break;
2866 	}
2867 
2868 	if (l)
2869 		iter->key = l->key;	/* remember it */
2870 	else
2871 		iter->pos = 0;		/* forget it */
2872 
2873 	return l;
2874 }
2875 
2876 static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2877 	__acquires(RCU)
2878 {
2879 	struct fib_route_iter *iter = seq->private;
2880 	struct fib_table *tb;
2881 	struct trie *t;
2882 
2883 	rcu_read_lock();
2884 
2885 	tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2886 	if (!tb)
2887 		return NULL;
2888 
2889 	iter->main_tb = tb;
2890 	t = (struct trie *)tb->tb_data;
2891 	iter->tnode = t->kv;
2892 
2893 	if (*pos != 0)
2894 		return fib_route_get_idx(iter, *pos);
2895 
2896 	iter->pos = 0;
2897 	iter->key = KEY_MAX;
2898 
2899 	return SEQ_START_TOKEN;
2900 }
2901 
2902 static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2903 {
2904 	struct fib_route_iter *iter = seq->private;
2905 	struct key_vector *l = NULL;
2906 	t_key key = iter->key + 1;
2907 
2908 	++*pos;
2909 
2910 	/* only allow key of 0 for start of sequence */
2911 	if ((v == SEQ_START_TOKEN) || key)
2912 		l = leaf_walk_rcu(&iter->tnode, key);
2913 
2914 	if (l) {
2915 		iter->key = l->key;
2916 		iter->pos++;
2917 	} else {
2918 		iter->pos = 0;
2919 	}
2920 
2921 	return l;
2922 }
2923 
2924 static void fib_route_seq_stop(struct seq_file *seq, void *v)
2925 	__releases(RCU)
2926 {
2927 	rcu_read_unlock();
2928 }
2929 
2930 static unsigned int fib_flag_trans(int type, __be32 mask, struct fib_info *fi)
2931 {
2932 	unsigned int flags = 0;
2933 
2934 	if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2935 		flags = RTF_REJECT;
2936 	if (fi) {
2937 		const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2938 
2939 		if (nhc->nhc_gw.ipv4)
2940 			flags |= RTF_GATEWAY;
2941 	}
2942 	if (mask == htonl(0xFFFFFFFF))
2943 		flags |= RTF_HOST;
2944 	flags |= RTF_UP;
2945 	return flags;
2946 }
2947 
2948 /*
2949  *	This outputs /proc/net/route.
2950  *	The format of the file is not supposed to be changed
2951  *	and needs to be same as fib_hash output to avoid breaking
2952  *	legacy utilities
2953  */
2954 static int fib_route_seq_show(struct seq_file *seq, void *v)
2955 {
2956 	struct fib_route_iter *iter = seq->private;
2957 	struct fib_table *tb = iter->main_tb;
2958 	struct fib_alias *fa;
2959 	struct key_vector *l = v;
2960 	__be32 prefix;
2961 
2962 	if (v == SEQ_START_TOKEN) {
2963 		seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2964 			   "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2965 			   "\tWindow\tIRTT");
2966 		return 0;
2967 	}
2968 
2969 	prefix = htonl(l->key);
2970 
2971 	hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2972 		struct fib_info *fi = fa->fa_info;
2973 		__be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2974 		unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2975 
2976 		if ((fa->fa_type == RTN_BROADCAST) ||
2977 		    (fa->fa_type == RTN_MULTICAST))
2978 			continue;
2979 
2980 		if (fa->tb_id != tb->tb_id)
2981 			continue;
2982 
2983 		seq_setwidth(seq, 127);
2984 
2985 		if (fi) {
2986 			struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2987 			__be32 gw = 0;
2988 
2989 			if (nhc->nhc_gw_family == AF_INET)
2990 				gw = nhc->nhc_gw.ipv4;
2991 
2992 			seq_printf(seq,
2993 				   "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2994 				   "%u\t%08X\t%d\t%u\t%u",
2995 				   nhc->nhc_dev ? nhc->nhc_dev->name : "*",
2996 				   prefix, gw, flags, 0, 0,
2997 				   fi->fib_priority,
2998 				   mask,
2999 				   (fi->fib_advmss ?
3000 				    fi->fib_advmss + 40 : 0),
3001 				   fi->fib_window,
3002 				   fi->fib_rtt >> 3);
3003 		} else {
3004 			seq_printf(seq,
3005 				   "*\t%08X\t%08X\t%04X\t%d\t%u\t"
3006 				   "%u\t%08X\t%d\t%u\t%u",
3007 				   prefix, 0, flags, 0, 0, 0,
3008 				   mask, 0, 0, 0);
3009 		}
3010 		seq_pad(seq, '\n');
3011 	}
3012 
3013 	return 0;
3014 }
3015 
3016 static const struct seq_operations fib_route_seq_ops = {
3017 	.start  = fib_route_seq_start,
3018 	.next   = fib_route_seq_next,
3019 	.stop   = fib_route_seq_stop,
3020 	.show   = fib_route_seq_show,
3021 };
3022 
3023 int __net_init fib_proc_init(struct net *net)
3024 {
3025 	if (!proc_create_net("fib_trie", 0444, net->proc_net, &fib_trie_seq_ops,
3026 			sizeof(struct fib_trie_iter)))
3027 		goto out1;
3028 
3029 	if (!proc_create_net_single("fib_triestat", 0444, net->proc_net,
3030 			fib_triestat_seq_show, NULL))
3031 		goto out2;
3032 
3033 	if (!proc_create_net("route", 0444, net->proc_net, &fib_route_seq_ops,
3034 			sizeof(struct fib_route_iter)))
3035 		goto out3;
3036 
3037 	return 0;
3038 
3039 out3:
3040 	remove_proc_entry("fib_triestat", net->proc_net);
3041 out2:
3042 	remove_proc_entry("fib_trie", net->proc_net);
3043 out1:
3044 	return -ENOMEM;
3045 }
3046 
3047 void __net_exit fib_proc_exit(struct net *net)
3048 {
3049 	remove_proc_entry("fib_trie", net->proc_net);
3050 	remove_proc_entry("fib_triestat", net->proc_net);
3051 	remove_proc_entry("route", net->proc_net);
3052 }
3053 
3054 #endif /* CONFIG_PROC_FS */
3055