xref: /openbmc/linux/net/ipv4/fib_trie.c (revision 94ab3170)
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/inet_dscp.h>
65 #include <net/ip.h>
66 #include <net/protocol.h>
67 #include <net/route.h>
68 #include <net/tcp.h>
69 #include <net/sock.h>
70 #include <net/ip_fib.h>
71 #include <net/fib_notifier.h>
72 #include <trace/events/fib.h>
73 #include "fib_lookup.h"
74 
75 static int call_fib_entry_notifier(struct notifier_block *nb,
76 				   enum fib_event_type event_type, u32 dst,
77 				   int dst_len, struct fib_alias *fa,
78 				   struct netlink_ext_ack *extack)
79 {
80 	struct fib_entry_notifier_info info = {
81 		.info.extack = extack,
82 		.dst = dst,
83 		.dst_len = dst_len,
84 		.fi = fa->fa_info,
85 		.dscp = fa->fa_dscp,
86 		.type = fa->fa_type,
87 		.tb_id = fa->tb_id,
88 	};
89 	return call_fib4_notifier(nb, event_type, &info.info);
90 }
91 
92 static int call_fib_entry_notifiers(struct net *net,
93 				    enum fib_event_type event_type, u32 dst,
94 				    int dst_len, struct fib_alias *fa,
95 				    struct netlink_ext_ack *extack)
96 {
97 	struct fib_entry_notifier_info info = {
98 		.info.extack = extack,
99 		.dst = dst,
100 		.dst_len = dst_len,
101 		.fi = fa->fa_info,
102 		.dscp = fa->fa_dscp,
103 		.type = fa->fa_type,
104 		.tb_id = fa->tb_id,
105 	};
106 	return call_fib4_notifiers(net, event_type, &info.info);
107 }
108 
109 #define MAX_STAT_DEPTH 32
110 
111 #define KEYLENGTH	(8*sizeof(t_key))
112 #define KEY_MAX		((t_key)~0)
113 
114 typedef unsigned int t_key;
115 
116 #define IS_TRIE(n)	((n)->pos >= KEYLENGTH)
117 #define IS_TNODE(n)	((n)->bits)
118 #define IS_LEAF(n)	(!(n)->bits)
119 
120 struct key_vector {
121 	t_key key;
122 	unsigned char pos;		/* 2log(KEYLENGTH) bits needed */
123 	unsigned char bits;		/* 2log(KEYLENGTH) bits needed */
124 	unsigned char slen;
125 	union {
126 		/* This list pointer if valid if (pos | bits) == 0 (LEAF) */
127 		struct hlist_head leaf;
128 		/* This array is valid if (pos | bits) > 0 (TNODE) */
129 		DECLARE_FLEX_ARRAY(struct key_vector __rcu *, tnode);
130 	};
131 };
132 
133 struct tnode {
134 	struct rcu_head rcu;
135 	t_key empty_children;		/* KEYLENGTH bits needed */
136 	t_key full_children;		/* KEYLENGTH bits needed */
137 	struct key_vector __rcu *parent;
138 	struct key_vector kv[1];
139 #define tn_bits kv[0].bits
140 };
141 
142 #define TNODE_SIZE(n)	offsetof(struct tnode, kv[0].tnode[n])
143 #define LEAF_SIZE	TNODE_SIZE(1)
144 
145 #ifdef CONFIG_IP_FIB_TRIE_STATS
146 struct trie_use_stats {
147 	unsigned int gets;
148 	unsigned int backtrack;
149 	unsigned int semantic_match_passed;
150 	unsigned int semantic_match_miss;
151 	unsigned int null_node_hit;
152 	unsigned int resize_node_skipped;
153 };
154 #endif
155 
156 struct trie_stat {
157 	unsigned int totdepth;
158 	unsigned int maxdepth;
159 	unsigned int tnodes;
160 	unsigned int leaves;
161 	unsigned int nullpointers;
162 	unsigned int prefixes;
163 	unsigned int nodesizes[MAX_STAT_DEPTH];
164 };
165 
166 struct trie {
167 	struct key_vector kv[1];
168 #ifdef CONFIG_IP_FIB_TRIE_STATS
169 	struct trie_use_stats __percpu *stats;
170 #endif
171 };
172 
173 static struct key_vector *resize(struct trie *t, struct key_vector *tn);
174 static unsigned int tnode_free_size;
175 
176 /*
177  * synchronize_rcu after call_rcu for outstanding dirty memory; it should be
178  * especially useful before resizing the root node with PREEMPT_NONE configs;
179  * the value was obtained experimentally, aiming to avoid visible slowdown.
180  */
181 unsigned int sysctl_fib_sync_mem = 512 * 1024;
182 unsigned int sysctl_fib_sync_mem_min = 64 * 1024;
183 unsigned int sysctl_fib_sync_mem_max = 64 * 1024 * 1024;
184 
185 static struct kmem_cache *fn_alias_kmem __ro_after_init;
186 static struct kmem_cache *trie_leaf_kmem __ro_after_init;
187 
188 static inline struct tnode *tn_info(struct key_vector *kv)
189 {
190 	return container_of(kv, struct tnode, kv[0]);
191 }
192 
193 /* caller must hold RTNL */
194 #define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
195 #define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
196 
197 /* caller must hold RCU read lock or RTNL */
198 #define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
199 #define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
200 
201 /* wrapper for rcu_assign_pointer */
202 static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
203 {
204 	if (n)
205 		rcu_assign_pointer(tn_info(n)->parent, tp);
206 }
207 
208 #define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
209 
210 /* This provides us with the number of children in this node, in the case of a
211  * leaf this will return 0 meaning none of the children are accessible.
212  */
213 static inline unsigned long child_length(const struct key_vector *tn)
214 {
215 	return (1ul << tn->bits) & ~(1ul);
216 }
217 
218 #define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
219 
220 static inline unsigned long get_index(t_key key, struct key_vector *kv)
221 {
222 	unsigned long index = key ^ kv->key;
223 
224 	if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
225 		return 0;
226 
227 	return index >> kv->pos;
228 }
229 
230 /* To understand this stuff, an understanding of keys and all their bits is
231  * necessary. Every node in the trie has a key associated with it, but not
232  * all of the bits in that key are significant.
233  *
234  * Consider a node 'n' and its parent 'tp'.
235  *
236  * If n is a leaf, every bit in its key is significant. Its presence is
237  * necessitated by path compression, since during a tree traversal (when
238  * searching for a leaf - unless we are doing an insertion) we will completely
239  * ignore all skipped bits we encounter. Thus we need to verify, at the end of
240  * a potentially successful search, that we have indeed been walking the
241  * correct key path.
242  *
243  * Note that we can never "miss" the correct key in the tree if present by
244  * following the wrong path. Path compression ensures that segments of the key
245  * that are the same for all keys with a given prefix are skipped, but the
246  * skipped part *is* identical for each node in the subtrie below the skipped
247  * bit! trie_insert() in this implementation takes care of that.
248  *
249  * if n is an internal node - a 'tnode' here, the various parts of its key
250  * have many different meanings.
251  *
252  * Example:
253  * _________________________________________________________________
254  * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
255  * -----------------------------------------------------------------
256  *  31  30  29  28  27  26  25  24  23  22  21  20  19  18  17  16
257  *
258  * _________________________________________________________________
259  * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
260  * -----------------------------------------------------------------
261  *  15  14  13  12  11  10   9   8   7   6   5   4   3   2   1   0
262  *
263  * tp->pos = 22
264  * tp->bits = 3
265  * n->pos = 13
266  * n->bits = 4
267  *
268  * First, let's just ignore the bits that come before the parent tp, that is
269  * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
270  * point we do not use them for anything.
271  *
272  * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
273  * index into the parent's child array. That is, they will be used to find
274  * 'n' among tp's children.
275  *
276  * The bits from (n->pos + n->bits) to (tp->pos - 1) - "S" - are skipped bits
277  * for the node n.
278  *
279  * All the bits we have seen so far are significant to the node n. The rest
280  * of the bits are really not needed or indeed known in n->key.
281  *
282  * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
283  * n's child array, and will of course be different for each child.
284  *
285  * The rest of the bits, from 0 to (n->pos -1) - "u" - are completely unknown
286  * at this point.
287  */
288 
289 static const int halve_threshold = 25;
290 static const int inflate_threshold = 50;
291 static const int halve_threshold_root = 15;
292 static const int inflate_threshold_root = 30;
293 
294 static void __alias_free_mem(struct rcu_head *head)
295 {
296 	struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
297 	kmem_cache_free(fn_alias_kmem, fa);
298 }
299 
300 static inline void alias_free_mem_rcu(struct fib_alias *fa)
301 {
302 	call_rcu(&fa->rcu, __alias_free_mem);
303 }
304 
305 #define TNODE_VMALLOC_MAX \
306 	ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
307 
308 static void __node_free_rcu(struct rcu_head *head)
309 {
310 	struct tnode *n = container_of(head, struct tnode, rcu);
311 
312 	if (!n->tn_bits)
313 		kmem_cache_free(trie_leaf_kmem, n);
314 	else
315 		kvfree(n);
316 }
317 
318 #define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
319 
320 static struct tnode *tnode_alloc(int bits)
321 {
322 	size_t size;
323 
324 	/* verify bits is within bounds */
325 	if (bits > TNODE_VMALLOC_MAX)
326 		return NULL;
327 
328 	/* determine size and verify it is non-zero and didn't overflow */
329 	size = TNODE_SIZE(1ul << bits);
330 
331 	if (size <= PAGE_SIZE)
332 		return kzalloc(size, GFP_KERNEL);
333 	else
334 		return vzalloc(size);
335 }
336 
337 static inline void empty_child_inc(struct key_vector *n)
338 {
339 	tn_info(n)->empty_children++;
340 
341 	if (!tn_info(n)->empty_children)
342 		tn_info(n)->full_children++;
343 }
344 
345 static inline void empty_child_dec(struct key_vector *n)
346 {
347 	if (!tn_info(n)->empty_children)
348 		tn_info(n)->full_children--;
349 
350 	tn_info(n)->empty_children--;
351 }
352 
353 static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
354 {
355 	struct key_vector *l;
356 	struct tnode *kv;
357 
358 	kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
359 	if (!kv)
360 		return NULL;
361 
362 	/* initialize key vector */
363 	l = kv->kv;
364 	l->key = key;
365 	l->pos = 0;
366 	l->bits = 0;
367 	l->slen = fa->fa_slen;
368 
369 	/* link leaf to fib alias */
370 	INIT_HLIST_HEAD(&l->leaf);
371 	hlist_add_head(&fa->fa_list, &l->leaf);
372 
373 	return l;
374 }
375 
376 static struct key_vector *tnode_new(t_key key, int pos, int bits)
377 {
378 	unsigned int shift = pos + bits;
379 	struct key_vector *tn;
380 	struct tnode *tnode;
381 
382 	/* verify bits and pos their msb bits clear and values are valid */
383 	BUG_ON(!bits || (shift > KEYLENGTH));
384 
385 	tnode = tnode_alloc(bits);
386 	if (!tnode)
387 		return NULL;
388 
389 	pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
390 		 sizeof(struct key_vector *) << bits);
391 
392 	if (bits == KEYLENGTH)
393 		tnode->full_children = 1;
394 	else
395 		tnode->empty_children = 1ul << bits;
396 
397 	tn = tnode->kv;
398 	tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
399 	tn->pos = pos;
400 	tn->bits = bits;
401 	tn->slen = pos;
402 
403 	return tn;
404 }
405 
406 /* Check whether a tnode 'n' is "full", i.e. it is an internal node
407  * and no bits are skipped. See discussion in dyntree paper p. 6
408  */
409 static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
410 {
411 	return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
412 }
413 
414 /* Add a child at position i overwriting the old value.
415  * Update the value of full_children and empty_children.
416  */
417 static void put_child(struct key_vector *tn, unsigned long i,
418 		      struct key_vector *n)
419 {
420 	struct key_vector *chi = get_child(tn, i);
421 	int isfull, wasfull;
422 
423 	BUG_ON(i >= child_length(tn));
424 
425 	/* update emptyChildren, overflow into fullChildren */
426 	if (!n && chi)
427 		empty_child_inc(tn);
428 	if (n && !chi)
429 		empty_child_dec(tn);
430 
431 	/* update fullChildren */
432 	wasfull = tnode_full(tn, chi);
433 	isfull = tnode_full(tn, n);
434 
435 	if (wasfull && !isfull)
436 		tn_info(tn)->full_children--;
437 	else if (!wasfull && isfull)
438 		tn_info(tn)->full_children++;
439 
440 	if (n && (tn->slen < n->slen))
441 		tn->slen = n->slen;
442 
443 	rcu_assign_pointer(tn->tnode[i], n);
444 }
445 
446 static void update_children(struct key_vector *tn)
447 {
448 	unsigned long i;
449 
450 	/* update all of the child parent pointers */
451 	for (i = child_length(tn); i;) {
452 		struct key_vector *inode = get_child(tn, --i);
453 
454 		if (!inode)
455 			continue;
456 
457 		/* Either update the children of a tnode that
458 		 * already belongs to us or update the child
459 		 * to point to ourselves.
460 		 */
461 		if (node_parent(inode) == tn)
462 			update_children(inode);
463 		else
464 			node_set_parent(inode, tn);
465 	}
466 }
467 
468 static inline void put_child_root(struct key_vector *tp, t_key key,
469 				  struct key_vector *n)
470 {
471 	if (IS_TRIE(tp))
472 		rcu_assign_pointer(tp->tnode[0], n);
473 	else
474 		put_child(tp, get_index(key, tp), n);
475 }
476 
477 static inline void tnode_free_init(struct key_vector *tn)
478 {
479 	tn_info(tn)->rcu.next = NULL;
480 }
481 
482 static inline void tnode_free_append(struct key_vector *tn,
483 				     struct key_vector *n)
484 {
485 	tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
486 	tn_info(tn)->rcu.next = &tn_info(n)->rcu;
487 }
488 
489 static void tnode_free(struct key_vector *tn)
490 {
491 	struct callback_head *head = &tn_info(tn)->rcu;
492 
493 	while (head) {
494 		head = head->next;
495 		tnode_free_size += TNODE_SIZE(1ul << tn->bits);
496 		node_free(tn);
497 
498 		tn = container_of(head, struct tnode, rcu)->kv;
499 	}
500 
501 	if (tnode_free_size >= READ_ONCE(sysctl_fib_sync_mem)) {
502 		tnode_free_size = 0;
503 		synchronize_rcu();
504 	}
505 }
506 
507 static struct key_vector *replace(struct trie *t,
508 				  struct key_vector *oldtnode,
509 				  struct key_vector *tn)
510 {
511 	struct key_vector *tp = node_parent(oldtnode);
512 	unsigned long i;
513 
514 	/* setup the parent pointer out of and back into this node */
515 	NODE_INIT_PARENT(tn, tp);
516 	put_child_root(tp, tn->key, tn);
517 
518 	/* update all of the child parent pointers */
519 	update_children(tn);
520 
521 	/* all pointers should be clean so we are done */
522 	tnode_free(oldtnode);
523 
524 	/* resize children now that oldtnode is freed */
525 	for (i = child_length(tn); i;) {
526 		struct key_vector *inode = get_child(tn, --i);
527 
528 		/* resize child node */
529 		if (tnode_full(tn, inode))
530 			tn = resize(t, inode);
531 	}
532 
533 	return tp;
534 }
535 
536 static struct key_vector *inflate(struct trie *t,
537 				  struct key_vector *oldtnode)
538 {
539 	struct key_vector *tn;
540 	unsigned long i;
541 	t_key m;
542 
543 	pr_debug("In inflate\n");
544 
545 	tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
546 	if (!tn)
547 		goto notnode;
548 
549 	/* prepare oldtnode to be freed */
550 	tnode_free_init(oldtnode);
551 
552 	/* Assemble all of the pointers in our cluster, in this case that
553 	 * represents all of the pointers out of our allocated nodes that
554 	 * point to existing tnodes and the links between our allocated
555 	 * nodes.
556 	 */
557 	for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
558 		struct key_vector *inode = get_child(oldtnode, --i);
559 		struct key_vector *node0, *node1;
560 		unsigned long j, k;
561 
562 		/* An empty child */
563 		if (!inode)
564 			continue;
565 
566 		/* A leaf or an internal node with skipped bits */
567 		if (!tnode_full(oldtnode, inode)) {
568 			put_child(tn, get_index(inode->key, tn), inode);
569 			continue;
570 		}
571 
572 		/* drop the node in the old tnode free list */
573 		tnode_free_append(oldtnode, inode);
574 
575 		/* An internal node with two children */
576 		if (inode->bits == 1) {
577 			put_child(tn, 2 * i + 1, get_child(inode, 1));
578 			put_child(tn, 2 * i, get_child(inode, 0));
579 			continue;
580 		}
581 
582 		/* We will replace this node 'inode' with two new
583 		 * ones, 'node0' and 'node1', each with half of the
584 		 * original children. The two new nodes will have
585 		 * a position one bit further down the key and this
586 		 * means that the "significant" part of their keys
587 		 * (see the discussion near the top of this file)
588 		 * will differ by one bit, which will be "0" in
589 		 * node0's key and "1" in node1's key. Since we are
590 		 * moving the key position by one step, the bit that
591 		 * we are moving away from - the bit at position
592 		 * (tn->pos) - is the one that will differ between
593 		 * node0 and node1. So... we synthesize that bit in the
594 		 * two new keys.
595 		 */
596 		node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
597 		if (!node1)
598 			goto nomem;
599 		node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
600 
601 		tnode_free_append(tn, node1);
602 		if (!node0)
603 			goto nomem;
604 		tnode_free_append(tn, node0);
605 
606 		/* populate child pointers in new nodes */
607 		for (k = child_length(inode), j = k / 2; j;) {
608 			put_child(node1, --j, get_child(inode, --k));
609 			put_child(node0, j, get_child(inode, j));
610 			put_child(node1, --j, get_child(inode, --k));
611 			put_child(node0, j, get_child(inode, j));
612 		}
613 
614 		/* link new nodes to parent */
615 		NODE_INIT_PARENT(node1, tn);
616 		NODE_INIT_PARENT(node0, tn);
617 
618 		/* link parent to nodes */
619 		put_child(tn, 2 * i + 1, node1);
620 		put_child(tn, 2 * i, node0);
621 	}
622 
623 	/* setup the parent pointers into and out of this node */
624 	return replace(t, oldtnode, tn);
625 nomem:
626 	/* all pointers should be clean so we are done */
627 	tnode_free(tn);
628 notnode:
629 	return NULL;
630 }
631 
632 static struct key_vector *halve(struct trie *t,
633 				struct key_vector *oldtnode)
634 {
635 	struct key_vector *tn;
636 	unsigned long i;
637 
638 	pr_debug("In halve\n");
639 
640 	tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
641 	if (!tn)
642 		goto notnode;
643 
644 	/* prepare oldtnode to be freed */
645 	tnode_free_init(oldtnode);
646 
647 	/* Assemble all of the pointers in our cluster, in this case that
648 	 * represents all of the pointers out of our allocated nodes that
649 	 * point to existing tnodes and the links between our allocated
650 	 * nodes.
651 	 */
652 	for (i = child_length(oldtnode); i;) {
653 		struct key_vector *node1 = get_child(oldtnode, --i);
654 		struct key_vector *node0 = get_child(oldtnode, --i);
655 		struct key_vector *inode;
656 
657 		/* At least one of the children is empty */
658 		if (!node1 || !node0) {
659 			put_child(tn, i / 2, node1 ? : node0);
660 			continue;
661 		}
662 
663 		/* Two nonempty children */
664 		inode = tnode_new(node0->key, oldtnode->pos, 1);
665 		if (!inode)
666 			goto nomem;
667 		tnode_free_append(tn, inode);
668 
669 		/* initialize pointers out of node */
670 		put_child(inode, 1, node1);
671 		put_child(inode, 0, node0);
672 		NODE_INIT_PARENT(inode, tn);
673 
674 		/* link parent to node */
675 		put_child(tn, i / 2, inode);
676 	}
677 
678 	/* setup the parent pointers into and out of this node */
679 	return replace(t, oldtnode, tn);
680 nomem:
681 	/* all pointers should be clean so we are done */
682 	tnode_free(tn);
683 notnode:
684 	return NULL;
685 }
686 
687 static struct key_vector *collapse(struct trie *t,
688 				   struct key_vector *oldtnode)
689 {
690 	struct key_vector *n, *tp;
691 	unsigned long i;
692 
693 	/* scan the tnode looking for that one child that might still exist */
694 	for (n = NULL, i = child_length(oldtnode); !n && i;)
695 		n = get_child(oldtnode, --i);
696 
697 	/* compress one level */
698 	tp = node_parent(oldtnode);
699 	put_child_root(tp, oldtnode->key, n);
700 	node_set_parent(n, tp);
701 
702 	/* drop dead node */
703 	node_free(oldtnode);
704 
705 	return tp;
706 }
707 
708 static unsigned char update_suffix(struct key_vector *tn)
709 {
710 	unsigned char slen = tn->pos;
711 	unsigned long stride, i;
712 	unsigned char slen_max;
713 
714 	/* only vector 0 can have a suffix length greater than or equal to
715 	 * tn->pos + tn->bits, the second highest node will have a suffix
716 	 * length at most of tn->pos + tn->bits - 1
717 	 */
718 	slen_max = min_t(unsigned char, tn->pos + tn->bits - 1, tn->slen);
719 
720 	/* search though the list of children looking for nodes that might
721 	 * have a suffix greater than the one we currently have.  This is
722 	 * why we start with a stride of 2 since a stride of 1 would
723 	 * represent the nodes with suffix length equal to tn->pos
724 	 */
725 	for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
726 		struct key_vector *n = get_child(tn, i);
727 
728 		if (!n || (n->slen <= slen))
729 			continue;
730 
731 		/* update stride and slen based on new value */
732 		stride <<= (n->slen - slen);
733 		slen = n->slen;
734 		i &= ~(stride - 1);
735 
736 		/* stop searching if we have hit the maximum possible value */
737 		if (slen >= slen_max)
738 			break;
739 	}
740 
741 	tn->slen = slen;
742 
743 	return slen;
744 }
745 
746 /* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
747  * the Helsinki University of Technology and Matti Tikkanen of Nokia
748  * Telecommunications, page 6:
749  * "A node is doubled if the ratio of non-empty children to all
750  * children in the *doubled* node is at least 'high'."
751  *
752  * 'high' in this instance is the variable 'inflate_threshold'. It
753  * is expressed as a percentage, so we multiply it with
754  * child_length() and instead of multiplying by 2 (since the
755  * child array will be doubled by inflate()) and multiplying
756  * the left-hand side by 100 (to handle the percentage thing) we
757  * multiply the left-hand side by 50.
758  *
759  * The left-hand side may look a bit weird: child_length(tn)
760  * - tn->empty_children is of course the number of non-null children
761  * in the current node. tn->full_children is the number of "full"
762  * children, that is non-null tnodes with a skip value of 0.
763  * All of those will be doubled in the resulting inflated tnode, so
764  * we just count them one extra time here.
765  *
766  * A clearer way to write this would be:
767  *
768  * to_be_doubled = tn->full_children;
769  * not_to_be_doubled = child_length(tn) - tn->empty_children -
770  *     tn->full_children;
771  *
772  * new_child_length = child_length(tn) * 2;
773  *
774  * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
775  *      new_child_length;
776  * if (new_fill_factor >= inflate_threshold)
777  *
778  * ...and so on, tho it would mess up the while () loop.
779  *
780  * anyway,
781  * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
782  *      inflate_threshold
783  *
784  * avoid a division:
785  * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
786  *      inflate_threshold * new_child_length
787  *
788  * expand not_to_be_doubled and to_be_doubled, and shorten:
789  * 100 * (child_length(tn) - tn->empty_children +
790  *    tn->full_children) >= inflate_threshold * new_child_length
791  *
792  * expand new_child_length:
793  * 100 * (child_length(tn) - tn->empty_children +
794  *    tn->full_children) >=
795  *      inflate_threshold * child_length(tn) * 2
796  *
797  * shorten again:
798  * 50 * (tn->full_children + child_length(tn) -
799  *    tn->empty_children) >= inflate_threshold *
800  *    child_length(tn)
801  *
802  */
803 static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
804 {
805 	unsigned long used = child_length(tn);
806 	unsigned long threshold = used;
807 
808 	/* Keep root node larger */
809 	threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
810 	used -= tn_info(tn)->empty_children;
811 	used += tn_info(tn)->full_children;
812 
813 	/* if bits == KEYLENGTH then pos = 0, and will fail below */
814 
815 	return (used > 1) && tn->pos && ((50 * used) >= threshold);
816 }
817 
818 static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
819 {
820 	unsigned long used = child_length(tn);
821 	unsigned long threshold = used;
822 
823 	/* Keep root node larger */
824 	threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
825 	used -= tn_info(tn)->empty_children;
826 
827 	/* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
828 
829 	return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
830 }
831 
832 static inline bool should_collapse(struct key_vector *tn)
833 {
834 	unsigned long used = child_length(tn);
835 
836 	used -= tn_info(tn)->empty_children;
837 
838 	/* account for bits == KEYLENGTH case */
839 	if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
840 		used -= KEY_MAX;
841 
842 	/* One child or none, time to drop us from the trie */
843 	return used < 2;
844 }
845 
846 #define MAX_WORK 10
847 static struct key_vector *resize(struct trie *t, struct key_vector *tn)
848 {
849 #ifdef CONFIG_IP_FIB_TRIE_STATS
850 	struct trie_use_stats __percpu *stats = t->stats;
851 #endif
852 	struct key_vector *tp = node_parent(tn);
853 	unsigned long cindex = get_index(tn->key, tp);
854 	int max_work = MAX_WORK;
855 
856 	pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
857 		 tn, inflate_threshold, halve_threshold);
858 
859 	/* track the tnode via the pointer from the parent instead of
860 	 * doing it ourselves.  This way we can let RCU fully do its
861 	 * thing without us interfering
862 	 */
863 	BUG_ON(tn != get_child(tp, cindex));
864 
865 	/* Double as long as the resulting node has a number of
866 	 * nonempty nodes that are above the threshold.
867 	 */
868 	while (should_inflate(tp, tn) && max_work) {
869 		tp = inflate(t, tn);
870 		if (!tp) {
871 #ifdef CONFIG_IP_FIB_TRIE_STATS
872 			this_cpu_inc(stats->resize_node_skipped);
873 #endif
874 			break;
875 		}
876 
877 		max_work--;
878 		tn = get_child(tp, cindex);
879 	}
880 
881 	/* update parent in case inflate failed */
882 	tp = node_parent(tn);
883 
884 	/* Return if at least one inflate is run */
885 	if (max_work != MAX_WORK)
886 		return tp;
887 
888 	/* Halve as long as the number of empty children in this
889 	 * node is above threshold.
890 	 */
891 	while (should_halve(tp, tn) && max_work) {
892 		tp = halve(t, tn);
893 		if (!tp) {
894 #ifdef CONFIG_IP_FIB_TRIE_STATS
895 			this_cpu_inc(stats->resize_node_skipped);
896 #endif
897 			break;
898 		}
899 
900 		max_work--;
901 		tn = get_child(tp, cindex);
902 	}
903 
904 	/* Only one child remains */
905 	if (should_collapse(tn))
906 		return collapse(t, tn);
907 
908 	/* update parent in case halve failed */
909 	return node_parent(tn);
910 }
911 
912 static void node_pull_suffix(struct key_vector *tn, unsigned char slen)
913 {
914 	unsigned char node_slen = tn->slen;
915 
916 	while ((node_slen > tn->pos) && (node_slen > slen)) {
917 		slen = update_suffix(tn);
918 		if (node_slen == slen)
919 			break;
920 
921 		tn = node_parent(tn);
922 		node_slen = tn->slen;
923 	}
924 }
925 
926 static void node_push_suffix(struct key_vector *tn, unsigned char slen)
927 {
928 	while (tn->slen < slen) {
929 		tn->slen = slen;
930 		tn = node_parent(tn);
931 	}
932 }
933 
934 /* rcu_read_lock needs to be hold by caller from readside */
935 static struct key_vector *fib_find_node(struct trie *t,
936 					struct key_vector **tp, u32 key)
937 {
938 	struct key_vector *pn, *n = t->kv;
939 	unsigned long index = 0;
940 
941 	do {
942 		pn = n;
943 		n = get_child_rcu(n, index);
944 
945 		if (!n)
946 			break;
947 
948 		index = get_cindex(key, n);
949 
950 		/* This bit of code is a bit tricky but it combines multiple
951 		 * checks into a single check.  The prefix consists of the
952 		 * prefix plus zeros for the bits in the cindex. The index
953 		 * is the difference between the key and this value.  From
954 		 * this we can actually derive several pieces of data.
955 		 *   if (index >= (1ul << bits))
956 		 *     we have a mismatch in skip bits and failed
957 		 *   else
958 		 *     we know the value is cindex
959 		 *
960 		 * This check is safe even if bits == KEYLENGTH due to the
961 		 * fact that we can only allocate a node with 32 bits if a
962 		 * long is greater than 32 bits.
963 		 */
964 		if (index >= (1ul << n->bits)) {
965 			n = NULL;
966 			break;
967 		}
968 
969 		/* keep searching until we find a perfect match leaf or NULL */
970 	} while (IS_TNODE(n));
971 
972 	*tp = pn;
973 
974 	return n;
975 }
976 
977 /* Return the first fib alias matching DSCP with
978  * priority less than or equal to PRIO.
979  * If 'find_first' is set, return the first matching
980  * fib alias, regardless of DSCP and priority.
981  */
982 static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
983 					dscp_t dscp, u32 prio, u32 tb_id,
984 					bool find_first)
985 {
986 	struct fib_alias *fa;
987 
988 	if (!fah)
989 		return NULL;
990 
991 	hlist_for_each_entry(fa, fah, fa_list) {
992 		/* Avoid Sparse warning when using dscp_t in inequalities */
993 		u8 __fa_dscp = inet_dscp_to_dsfield(fa->fa_dscp);
994 		u8 __dscp = inet_dscp_to_dsfield(dscp);
995 
996 		if (fa->fa_slen < slen)
997 			continue;
998 		if (fa->fa_slen != slen)
999 			break;
1000 		if (fa->tb_id > tb_id)
1001 			continue;
1002 		if (fa->tb_id != tb_id)
1003 			break;
1004 		if (find_first)
1005 			return fa;
1006 		if (__fa_dscp > __dscp)
1007 			continue;
1008 		if (fa->fa_info->fib_priority >= prio || __fa_dscp < __dscp)
1009 			return fa;
1010 	}
1011 
1012 	return NULL;
1013 }
1014 
1015 static struct fib_alias *
1016 fib_find_matching_alias(struct net *net, const struct fib_rt_info *fri)
1017 {
1018 	u8 slen = KEYLENGTH - fri->dst_len;
1019 	struct key_vector *l, *tp;
1020 	struct fib_table *tb;
1021 	struct fib_alias *fa;
1022 	struct trie *t;
1023 
1024 	tb = fib_get_table(net, fri->tb_id);
1025 	if (!tb)
1026 		return NULL;
1027 
1028 	t = (struct trie *)tb->tb_data;
1029 	l = fib_find_node(t, &tp, be32_to_cpu(fri->dst));
1030 	if (!l)
1031 		return NULL;
1032 
1033 	hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1034 		if (fa->fa_slen == slen && fa->tb_id == fri->tb_id &&
1035 		    fa->fa_dscp == fri->dscp && fa->fa_info == fri->fi &&
1036 		    fa->fa_type == fri->type)
1037 			return fa;
1038 	}
1039 
1040 	return NULL;
1041 }
1042 
1043 void fib_alias_hw_flags_set(struct net *net, const struct fib_rt_info *fri)
1044 {
1045 	u8 fib_notify_on_flag_change;
1046 	struct fib_alias *fa_match;
1047 	struct sk_buff *skb;
1048 	int err;
1049 
1050 	rcu_read_lock();
1051 
1052 	fa_match = fib_find_matching_alias(net, fri);
1053 	if (!fa_match)
1054 		goto out;
1055 
1056 	/* These are paired with the WRITE_ONCE() happening in this function.
1057 	 * The reason is that we are only protected by RCU at this point.
1058 	 */
1059 	if (READ_ONCE(fa_match->offload) == fri->offload &&
1060 	    READ_ONCE(fa_match->trap) == fri->trap &&
1061 	    READ_ONCE(fa_match->offload_failed) == fri->offload_failed)
1062 		goto out;
1063 
1064 	WRITE_ONCE(fa_match->offload, fri->offload);
1065 	WRITE_ONCE(fa_match->trap, fri->trap);
1066 
1067 	fib_notify_on_flag_change = READ_ONCE(net->ipv4.sysctl_fib_notify_on_flag_change);
1068 
1069 	/* 2 means send notifications only if offload_failed was changed. */
1070 	if (fib_notify_on_flag_change == 2 &&
1071 	    READ_ONCE(fa_match->offload_failed) == fri->offload_failed)
1072 		goto out;
1073 
1074 	WRITE_ONCE(fa_match->offload_failed, fri->offload_failed);
1075 
1076 	if (!fib_notify_on_flag_change)
1077 		goto out;
1078 
1079 	skb = nlmsg_new(fib_nlmsg_size(fa_match->fa_info), GFP_ATOMIC);
1080 	if (!skb) {
1081 		err = -ENOBUFS;
1082 		goto errout;
1083 	}
1084 
1085 	err = fib_dump_info(skb, 0, 0, RTM_NEWROUTE, fri, 0);
1086 	if (err < 0) {
1087 		/* -EMSGSIZE implies BUG in fib_nlmsg_size() */
1088 		WARN_ON(err == -EMSGSIZE);
1089 		kfree_skb(skb);
1090 		goto errout;
1091 	}
1092 
1093 	rtnl_notify(skb, net, 0, RTNLGRP_IPV4_ROUTE, NULL, GFP_ATOMIC);
1094 	goto out;
1095 
1096 errout:
1097 	rtnl_set_sk_err(net, RTNLGRP_IPV4_ROUTE, err);
1098 out:
1099 	rcu_read_unlock();
1100 }
1101 EXPORT_SYMBOL_GPL(fib_alias_hw_flags_set);
1102 
1103 static void trie_rebalance(struct trie *t, struct key_vector *tn)
1104 {
1105 	while (!IS_TRIE(tn))
1106 		tn = resize(t, tn);
1107 }
1108 
1109 static int fib_insert_node(struct trie *t, struct key_vector *tp,
1110 			   struct fib_alias *new, t_key key)
1111 {
1112 	struct key_vector *n, *l;
1113 
1114 	l = leaf_new(key, new);
1115 	if (!l)
1116 		goto noleaf;
1117 
1118 	/* retrieve child from parent node */
1119 	n = get_child(tp, get_index(key, tp));
1120 
1121 	/* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1122 	 *
1123 	 *  Add a new tnode here
1124 	 *  first tnode need some special handling
1125 	 *  leaves us in position for handling as case 3
1126 	 */
1127 	if (n) {
1128 		struct key_vector *tn;
1129 
1130 		tn = tnode_new(key, __fls(key ^ n->key), 1);
1131 		if (!tn)
1132 			goto notnode;
1133 
1134 		/* initialize routes out of node */
1135 		NODE_INIT_PARENT(tn, tp);
1136 		put_child(tn, get_index(key, tn) ^ 1, n);
1137 
1138 		/* start adding routes into the node */
1139 		put_child_root(tp, key, tn);
1140 		node_set_parent(n, tn);
1141 
1142 		/* parent now has a NULL spot where the leaf can go */
1143 		tp = tn;
1144 	}
1145 
1146 	/* Case 3: n is NULL, and will just insert a new leaf */
1147 	node_push_suffix(tp, new->fa_slen);
1148 	NODE_INIT_PARENT(l, tp);
1149 	put_child_root(tp, key, l);
1150 	trie_rebalance(t, tp);
1151 
1152 	return 0;
1153 notnode:
1154 	node_free(l);
1155 noleaf:
1156 	return -ENOMEM;
1157 }
1158 
1159 static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1160 			    struct key_vector *l, struct fib_alias *new,
1161 			    struct fib_alias *fa, t_key key)
1162 {
1163 	if (!l)
1164 		return fib_insert_node(t, tp, new, key);
1165 
1166 	if (fa) {
1167 		hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1168 	} else {
1169 		struct fib_alias *last;
1170 
1171 		hlist_for_each_entry(last, &l->leaf, fa_list) {
1172 			if (new->fa_slen < last->fa_slen)
1173 				break;
1174 			if ((new->fa_slen == last->fa_slen) &&
1175 			    (new->tb_id > last->tb_id))
1176 				break;
1177 			fa = last;
1178 		}
1179 
1180 		if (fa)
1181 			hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1182 		else
1183 			hlist_add_head_rcu(&new->fa_list, &l->leaf);
1184 	}
1185 
1186 	/* if we added to the tail node then we need to update slen */
1187 	if (l->slen < new->fa_slen) {
1188 		l->slen = new->fa_slen;
1189 		node_push_suffix(tp, new->fa_slen);
1190 	}
1191 
1192 	return 0;
1193 }
1194 
1195 static bool fib_valid_key_len(u32 key, u8 plen, struct netlink_ext_ack *extack)
1196 {
1197 	if (plen > KEYLENGTH) {
1198 		NL_SET_ERR_MSG(extack, "Invalid prefix length");
1199 		return false;
1200 	}
1201 
1202 	if ((plen < KEYLENGTH) && (key << plen)) {
1203 		NL_SET_ERR_MSG(extack,
1204 			       "Invalid prefix for given prefix length");
1205 		return false;
1206 	}
1207 
1208 	return true;
1209 }
1210 
1211 static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1212 			     struct key_vector *l, struct fib_alias *old);
1213 
1214 /* Caller must hold RTNL. */
1215 int fib_table_insert(struct net *net, struct fib_table *tb,
1216 		     struct fib_config *cfg, struct netlink_ext_ack *extack)
1217 {
1218 	struct trie *t = (struct trie *)tb->tb_data;
1219 	struct fib_alias *fa, *new_fa;
1220 	struct key_vector *l, *tp;
1221 	u16 nlflags = NLM_F_EXCL;
1222 	struct fib_info *fi;
1223 	u8 plen = cfg->fc_dst_len;
1224 	u8 slen = KEYLENGTH - plen;
1225 	dscp_t dscp;
1226 	u32 key;
1227 	int err;
1228 
1229 	key = ntohl(cfg->fc_dst);
1230 
1231 	if (!fib_valid_key_len(key, plen, extack))
1232 		return -EINVAL;
1233 
1234 	pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1235 
1236 	fi = fib_create_info(cfg, extack);
1237 	if (IS_ERR(fi)) {
1238 		err = PTR_ERR(fi);
1239 		goto err;
1240 	}
1241 
1242 	dscp = cfg->fc_dscp;
1243 	l = fib_find_node(t, &tp, key);
1244 	fa = l ? fib_find_alias(&l->leaf, slen, dscp, fi->fib_priority,
1245 				tb->tb_id, false) : NULL;
1246 
1247 	/* Now fa, if non-NULL, points to the first fib alias
1248 	 * with the same keys [prefix,dscp,priority], if such key already
1249 	 * exists or to the node before which we will insert new one.
1250 	 *
1251 	 * If fa is NULL, we will need to allocate a new one and
1252 	 * insert to the tail of the section matching the suffix length
1253 	 * of the new alias.
1254 	 */
1255 
1256 	if (fa && fa->fa_dscp == dscp &&
1257 	    fa->fa_info->fib_priority == fi->fib_priority) {
1258 		struct fib_alias *fa_first, *fa_match;
1259 
1260 		err = -EEXIST;
1261 		if (cfg->fc_nlflags & NLM_F_EXCL)
1262 			goto out;
1263 
1264 		nlflags &= ~NLM_F_EXCL;
1265 
1266 		/* We have 2 goals:
1267 		 * 1. Find exact match for type, scope, fib_info to avoid
1268 		 * duplicate routes
1269 		 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1270 		 */
1271 		fa_match = NULL;
1272 		fa_first = fa;
1273 		hlist_for_each_entry_from(fa, fa_list) {
1274 			if ((fa->fa_slen != slen) ||
1275 			    (fa->tb_id != tb->tb_id) ||
1276 			    (fa->fa_dscp != dscp))
1277 				break;
1278 			if (fa->fa_info->fib_priority != fi->fib_priority)
1279 				break;
1280 			if (fa->fa_type == cfg->fc_type &&
1281 			    fa->fa_info == fi) {
1282 				fa_match = fa;
1283 				break;
1284 			}
1285 		}
1286 
1287 		if (cfg->fc_nlflags & NLM_F_REPLACE) {
1288 			struct fib_info *fi_drop;
1289 			u8 state;
1290 
1291 			nlflags |= NLM_F_REPLACE;
1292 			fa = fa_first;
1293 			if (fa_match) {
1294 				if (fa == fa_match)
1295 					err = 0;
1296 				goto out;
1297 			}
1298 			err = -ENOBUFS;
1299 			new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1300 			if (!new_fa)
1301 				goto out;
1302 
1303 			fi_drop = fa->fa_info;
1304 			new_fa->fa_dscp = fa->fa_dscp;
1305 			new_fa->fa_info = fi;
1306 			new_fa->fa_type = cfg->fc_type;
1307 			state = fa->fa_state;
1308 			new_fa->fa_state = state & ~FA_S_ACCESSED;
1309 			new_fa->fa_slen = fa->fa_slen;
1310 			new_fa->tb_id = tb->tb_id;
1311 			new_fa->fa_default = -1;
1312 			new_fa->offload = 0;
1313 			new_fa->trap = 0;
1314 			new_fa->offload_failed = 0;
1315 
1316 			hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1317 
1318 			if (fib_find_alias(&l->leaf, fa->fa_slen, 0, 0,
1319 					   tb->tb_id, true) == new_fa) {
1320 				enum fib_event_type fib_event;
1321 
1322 				fib_event = FIB_EVENT_ENTRY_REPLACE;
1323 				err = call_fib_entry_notifiers(net, fib_event,
1324 							       key, plen,
1325 							       new_fa, extack);
1326 				if (err) {
1327 					hlist_replace_rcu(&new_fa->fa_list,
1328 							  &fa->fa_list);
1329 					goto out_free_new_fa;
1330 				}
1331 			}
1332 
1333 			rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1334 				  tb->tb_id, &cfg->fc_nlinfo, nlflags);
1335 
1336 			alias_free_mem_rcu(fa);
1337 
1338 			fib_release_info(fi_drop);
1339 			if (state & FA_S_ACCESSED)
1340 				rt_cache_flush(cfg->fc_nlinfo.nl_net);
1341 
1342 			goto succeeded;
1343 		}
1344 		/* Error if we find a perfect match which
1345 		 * uses the same scope, type, and nexthop
1346 		 * information.
1347 		 */
1348 		if (fa_match)
1349 			goto out;
1350 
1351 		if (cfg->fc_nlflags & NLM_F_APPEND)
1352 			nlflags |= NLM_F_APPEND;
1353 		else
1354 			fa = fa_first;
1355 	}
1356 	err = -ENOENT;
1357 	if (!(cfg->fc_nlflags & NLM_F_CREATE))
1358 		goto out;
1359 
1360 	nlflags |= NLM_F_CREATE;
1361 	err = -ENOBUFS;
1362 	new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1363 	if (!new_fa)
1364 		goto out;
1365 
1366 	new_fa->fa_info = fi;
1367 	new_fa->fa_dscp = dscp;
1368 	new_fa->fa_type = cfg->fc_type;
1369 	new_fa->fa_state = 0;
1370 	new_fa->fa_slen = slen;
1371 	new_fa->tb_id = tb->tb_id;
1372 	new_fa->fa_default = -1;
1373 	new_fa->offload = 0;
1374 	new_fa->trap = 0;
1375 	new_fa->offload_failed = 0;
1376 
1377 	/* Insert new entry to the list. */
1378 	err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1379 	if (err)
1380 		goto out_free_new_fa;
1381 
1382 	/* The alias was already inserted, so the node must exist. */
1383 	l = l ? l : fib_find_node(t, &tp, key);
1384 	if (WARN_ON_ONCE(!l)) {
1385 		err = -ENOENT;
1386 		goto out_free_new_fa;
1387 	}
1388 
1389 	if (fib_find_alias(&l->leaf, new_fa->fa_slen, 0, 0, tb->tb_id, true) ==
1390 	    new_fa) {
1391 		enum fib_event_type fib_event;
1392 
1393 		fib_event = FIB_EVENT_ENTRY_REPLACE;
1394 		err = call_fib_entry_notifiers(net, fib_event, key, plen,
1395 					       new_fa, extack);
1396 		if (err)
1397 			goto out_remove_new_fa;
1398 	}
1399 
1400 	if (!plen)
1401 		tb->tb_num_default++;
1402 
1403 	rt_cache_flush(cfg->fc_nlinfo.nl_net);
1404 	rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1405 		  &cfg->fc_nlinfo, nlflags);
1406 succeeded:
1407 	return 0;
1408 
1409 out_remove_new_fa:
1410 	fib_remove_alias(t, tp, l, new_fa);
1411 out_free_new_fa:
1412 	kmem_cache_free(fn_alias_kmem, new_fa);
1413 out:
1414 	fib_release_info(fi);
1415 err:
1416 	return err;
1417 }
1418 
1419 static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1420 {
1421 	t_key prefix = n->key;
1422 
1423 	return (key ^ prefix) & (prefix | -prefix);
1424 }
1425 
1426 bool fib_lookup_good_nhc(const struct fib_nh_common *nhc, int fib_flags,
1427 			 const struct flowi4 *flp)
1428 {
1429 	if (nhc->nhc_flags & RTNH_F_DEAD)
1430 		return false;
1431 
1432 	if (ip_ignore_linkdown(nhc->nhc_dev) &&
1433 	    nhc->nhc_flags & RTNH_F_LINKDOWN &&
1434 	    !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1435 		return false;
1436 
1437 	if (flp->flowi4_oif && flp->flowi4_oif != nhc->nhc_oif)
1438 		return false;
1439 
1440 	return true;
1441 }
1442 
1443 /* should be called with rcu_read_lock */
1444 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1445 		     struct fib_result *res, int fib_flags)
1446 {
1447 	struct trie *t = (struct trie *) tb->tb_data;
1448 #ifdef CONFIG_IP_FIB_TRIE_STATS
1449 	struct trie_use_stats __percpu *stats = t->stats;
1450 #endif
1451 	const t_key key = ntohl(flp->daddr);
1452 	struct key_vector *n, *pn;
1453 	struct fib_alias *fa;
1454 	unsigned long index;
1455 	t_key cindex;
1456 
1457 	pn = t->kv;
1458 	cindex = 0;
1459 
1460 	n = get_child_rcu(pn, cindex);
1461 	if (!n) {
1462 		trace_fib_table_lookup(tb->tb_id, flp, NULL, -EAGAIN);
1463 		return -EAGAIN;
1464 	}
1465 
1466 #ifdef CONFIG_IP_FIB_TRIE_STATS
1467 	this_cpu_inc(stats->gets);
1468 #endif
1469 
1470 	/* Step 1: Travel to the longest prefix match in the trie */
1471 	for (;;) {
1472 		index = get_cindex(key, n);
1473 
1474 		/* This bit of code is a bit tricky but it combines multiple
1475 		 * checks into a single check.  The prefix consists of the
1476 		 * prefix plus zeros for the "bits" in the prefix. The index
1477 		 * is the difference between the key and this value.  From
1478 		 * this we can actually derive several pieces of data.
1479 		 *   if (index >= (1ul << bits))
1480 		 *     we have a mismatch in skip bits and failed
1481 		 *   else
1482 		 *     we know the value is cindex
1483 		 *
1484 		 * This check is safe even if bits == KEYLENGTH due to the
1485 		 * fact that we can only allocate a node with 32 bits if a
1486 		 * long is greater than 32 bits.
1487 		 */
1488 		if (index >= (1ul << n->bits))
1489 			break;
1490 
1491 		/* we have found a leaf. Prefixes have already been compared */
1492 		if (IS_LEAF(n))
1493 			goto found;
1494 
1495 		/* only record pn and cindex if we are going to be chopping
1496 		 * bits later.  Otherwise we are just wasting cycles.
1497 		 */
1498 		if (n->slen > n->pos) {
1499 			pn = n;
1500 			cindex = index;
1501 		}
1502 
1503 		n = get_child_rcu(n, index);
1504 		if (unlikely(!n))
1505 			goto backtrace;
1506 	}
1507 
1508 	/* Step 2: Sort out leaves and begin backtracing for longest prefix */
1509 	for (;;) {
1510 		/* record the pointer where our next node pointer is stored */
1511 		struct key_vector __rcu **cptr = n->tnode;
1512 
1513 		/* This test verifies that none of the bits that differ
1514 		 * between the key and the prefix exist in the region of
1515 		 * the lsb and higher in the prefix.
1516 		 */
1517 		if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1518 			goto backtrace;
1519 
1520 		/* exit out and process leaf */
1521 		if (unlikely(IS_LEAF(n)))
1522 			break;
1523 
1524 		/* Don't bother recording parent info.  Since we are in
1525 		 * prefix match mode we will have to come back to wherever
1526 		 * we started this traversal anyway
1527 		 */
1528 
1529 		while ((n = rcu_dereference(*cptr)) == NULL) {
1530 backtrace:
1531 #ifdef CONFIG_IP_FIB_TRIE_STATS
1532 			if (!n)
1533 				this_cpu_inc(stats->null_node_hit);
1534 #endif
1535 			/* If we are at cindex 0 there are no more bits for
1536 			 * us to strip at this level so we must ascend back
1537 			 * up one level to see if there are any more bits to
1538 			 * be stripped there.
1539 			 */
1540 			while (!cindex) {
1541 				t_key pkey = pn->key;
1542 
1543 				/* If we don't have a parent then there is
1544 				 * nothing for us to do as we do not have any
1545 				 * further nodes to parse.
1546 				 */
1547 				if (IS_TRIE(pn)) {
1548 					trace_fib_table_lookup(tb->tb_id, flp,
1549 							       NULL, -EAGAIN);
1550 					return -EAGAIN;
1551 				}
1552 #ifdef CONFIG_IP_FIB_TRIE_STATS
1553 				this_cpu_inc(stats->backtrack);
1554 #endif
1555 				/* Get Child's index */
1556 				pn = node_parent_rcu(pn);
1557 				cindex = get_index(pkey, pn);
1558 			}
1559 
1560 			/* strip the least significant bit from the cindex */
1561 			cindex &= cindex - 1;
1562 
1563 			/* grab pointer for next child node */
1564 			cptr = &pn->tnode[cindex];
1565 		}
1566 	}
1567 
1568 found:
1569 	/* this line carries forward the xor from earlier in the function */
1570 	index = key ^ n->key;
1571 
1572 	/* Step 3: Process the leaf, if that fails fall back to backtracing */
1573 	hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1574 		struct fib_info *fi = fa->fa_info;
1575 		struct fib_nh_common *nhc;
1576 		int nhsel, err;
1577 
1578 		if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1579 			if (index >= (1ul << fa->fa_slen))
1580 				continue;
1581 		}
1582 		if (fa->fa_dscp &&
1583 		    inet_dscp_to_dsfield(fa->fa_dscp) != flp->flowi4_tos)
1584 			continue;
1585 		/* Paired with WRITE_ONCE() in fib_release_info() */
1586 		if (READ_ONCE(fi->fib_dead))
1587 			continue;
1588 		if (fa->fa_info->fib_scope < flp->flowi4_scope)
1589 			continue;
1590 		fib_alias_accessed(fa);
1591 		err = fib_props[fa->fa_type].error;
1592 		if (unlikely(err < 0)) {
1593 out_reject:
1594 #ifdef CONFIG_IP_FIB_TRIE_STATS
1595 			this_cpu_inc(stats->semantic_match_passed);
1596 #endif
1597 			trace_fib_table_lookup(tb->tb_id, flp, NULL, err);
1598 			return err;
1599 		}
1600 		if (fi->fib_flags & RTNH_F_DEAD)
1601 			continue;
1602 
1603 		if (unlikely(fi->nh)) {
1604 			if (nexthop_is_blackhole(fi->nh)) {
1605 				err = fib_props[RTN_BLACKHOLE].error;
1606 				goto out_reject;
1607 			}
1608 
1609 			nhc = nexthop_get_nhc_lookup(fi->nh, fib_flags, flp,
1610 						     &nhsel);
1611 			if (nhc)
1612 				goto set_result;
1613 			goto miss;
1614 		}
1615 
1616 		for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
1617 			nhc = fib_info_nhc(fi, nhsel);
1618 
1619 			if (!fib_lookup_good_nhc(nhc, fib_flags, flp))
1620 				continue;
1621 set_result:
1622 			if (!(fib_flags & FIB_LOOKUP_NOREF))
1623 				refcount_inc(&fi->fib_clntref);
1624 
1625 			res->prefix = htonl(n->key);
1626 			res->prefixlen = KEYLENGTH - fa->fa_slen;
1627 			res->nh_sel = nhsel;
1628 			res->nhc = nhc;
1629 			res->type = fa->fa_type;
1630 			res->scope = fi->fib_scope;
1631 			res->dscp = fa->fa_dscp;
1632 			res->fi = fi;
1633 			res->table = tb;
1634 			res->fa_head = &n->leaf;
1635 #ifdef CONFIG_IP_FIB_TRIE_STATS
1636 			this_cpu_inc(stats->semantic_match_passed);
1637 #endif
1638 			trace_fib_table_lookup(tb->tb_id, flp, nhc, err);
1639 
1640 			return err;
1641 		}
1642 	}
1643 miss:
1644 #ifdef CONFIG_IP_FIB_TRIE_STATS
1645 	this_cpu_inc(stats->semantic_match_miss);
1646 #endif
1647 	goto backtrace;
1648 }
1649 EXPORT_SYMBOL_GPL(fib_table_lookup);
1650 
1651 static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1652 			     struct key_vector *l, struct fib_alias *old)
1653 {
1654 	/* record the location of the previous list_info entry */
1655 	struct hlist_node **pprev = old->fa_list.pprev;
1656 	struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1657 
1658 	/* remove the fib_alias from the list */
1659 	hlist_del_rcu(&old->fa_list);
1660 
1661 	/* if we emptied the list this leaf will be freed and we can sort
1662 	 * out parent suffix lengths as a part of trie_rebalance
1663 	 */
1664 	if (hlist_empty(&l->leaf)) {
1665 		if (tp->slen == l->slen)
1666 			node_pull_suffix(tp, tp->pos);
1667 		put_child_root(tp, l->key, NULL);
1668 		node_free(l);
1669 		trie_rebalance(t, tp);
1670 		return;
1671 	}
1672 
1673 	/* only access fa if it is pointing at the last valid hlist_node */
1674 	if (*pprev)
1675 		return;
1676 
1677 	/* update the trie with the latest suffix length */
1678 	l->slen = fa->fa_slen;
1679 	node_pull_suffix(tp, fa->fa_slen);
1680 }
1681 
1682 static void fib_notify_alias_delete(struct net *net, u32 key,
1683 				    struct hlist_head *fah,
1684 				    struct fib_alias *fa_to_delete,
1685 				    struct netlink_ext_ack *extack)
1686 {
1687 	struct fib_alias *fa_next, *fa_to_notify;
1688 	u32 tb_id = fa_to_delete->tb_id;
1689 	u8 slen = fa_to_delete->fa_slen;
1690 	enum fib_event_type fib_event;
1691 
1692 	/* Do not notify if we do not care about the route. */
1693 	if (fib_find_alias(fah, slen, 0, 0, tb_id, true) != fa_to_delete)
1694 		return;
1695 
1696 	/* Determine if the route should be replaced by the next route in the
1697 	 * list.
1698 	 */
1699 	fa_next = hlist_entry_safe(fa_to_delete->fa_list.next,
1700 				   struct fib_alias, fa_list);
1701 	if (fa_next && fa_next->fa_slen == slen && fa_next->tb_id == tb_id) {
1702 		fib_event = FIB_EVENT_ENTRY_REPLACE;
1703 		fa_to_notify = fa_next;
1704 	} else {
1705 		fib_event = FIB_EVENT_ENTRY_DEL;
1706 		fa_to_notify = fa_to_delete;
1707 	}
1708 	call_fib_entry_notifiers(net, fib_event, key, KEYLENGTH - slen,
1709 				 fa_to_notify, extack);
1710 }
1711 
1712 /* Caller must hold RTNL. */
1713 int fib_table_delete(struct net *net, struct fib_table *tb,
1714 		     struct fib_config *cfg, struct netlink_ext_ack *extack)
1715 {
1716 	struct trie *t = (struct trie *) tb->tb_data;
1717 	struct fib_alias *fa, *fa_to_delete;
1718 	struct key_vector *l, *tp;
1719 	u8 plen = cfg->fc_dst_len;
1720 	u8 slen = KEYLENGTH - plen;
1721 	dscp_t dscp;
1722 	u32 key;
1723 
1724 	key = ntohl(cfg->fc_dst);
1725 
1726 	if (!fib_valid_key_len(key, plen, extack))
1727 		return -EINVAL;
1728 
1729 	l = fib_find_node(t, &tp, key);
1730 	if (!l)
1731 		return -ESRCH;
1732 
1733 	dscp = cfg->fc_dscp;
1734 	fa = fib_find_alias(&l->leaf, slen, dscp, 0, tb->tb_id, false);
1735 	if (!fa)
1736 		return -ESRCH;
1737 
1738 	pr_debug("Deleting %08x/%d dsfield=0x%02x t=%p\n", key, plen,
1739 		 inet_dscp_to_dsfield(dscp), t);
1740 
1741 	fa_to_delete = NULL;
1742 	hlist_for_each_entry_from(fa, fa_list) {
1743 		struct fib_info *fi = fa->fa_info;
1744 
1745 		if ((fa->fa_slen != slen) ||
1746 		    (fa->tb_id != tb->tb_id) ||
1747 		    (fa->fa_dscp != dscp))
1748 			break;
1749 
1750 		if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1751 		    (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1752 		     fa->fa_info->fib_scope == cfg->fc_scope) &&
1753 		    (!cfg->fc_prefsrc ||
1754 		     fi->fib_prefsrc == cfg->fc_prefsrc) &&
1755 		    (!cfg->fc_protocol ||
1756 		     fi->fib_protocol == cfg->fc_protocol) &&
1757 		    fib_nh_match(net, cfg, fi, extack) == 0 &&
1758 		    fib_metrics_match(cfg, fi)) {
1759 			fa_to_delete = fa;
1760 			break;
1761 		}
1762 	}
1763 
1764 	if (!fa_to_delete)
1765 		return -ESRCH;
1766 
1767 	fib_notify_alias_delete(net, key, &l->leaf, fa_to_delete, extack);
1768 	rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1769 		  &cfg->fc_nlinfo, 0);
1770 
1771 	if (!plen)
1772 		tb->tb_num_default--;
1773 
1774 	fib_remove_alias(t, tp, l, fa_to_delete);
1775 
1776 	if (fa_to_delete->fa_state & FA_S_ACCESSED)
1777 		rt_cache_flush(cfg->fc_nlinfo.nl_net);
1778 
1779 	fib_release_info(fa_to_delete->fa_info);
1780 	alias_free_mem_rcu(fa_to_delete);
1781 	return 0;
1782 }
1783 
1784 /* Scan for the next leaf starting at the provided key value */
1785 static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1786 {
1787 	struct key_vector *pn, *n = *tn;
1788 	unsigned long cindex;
1789 
1790 	/* this loop is meant to try and find the key in the trie */
1791 	do {
1792 		/* record parent and next child index */
1793 		pn = n;
1794 		cindex = (key > pn->key) ? get_index(key, pn) : 0;
1795 
1796 		if (cindex >> pn->bits)
1797 			break;
1798 
1799 		/* descend into the next child */
1800 		n = get_child_rcu(pn, cindex++);
1801 		if (!n)
1802 			break;
1803 
1804 		/* guarantee forward progress on the keys */
1805 		if (IS_LEAF(n) && (n->key >= key))
1806 			goto found;
1807 	} while (IS_TNODE(n));
1808 
1809 	/* this loop will search for the next leaf with a greater key */
1810 	while (!IS_TRIE(pn)) {
1811 		/* if we exhausted the parent node we will need to climb */
1812 		if (cindex >= (1ul << pn->bits)) {
1813 			t_key pkey = pn->key;
1814 
1815 			pn = node_parent_rcu(pn);
1816 			cindex = get_index(pkey, pn) + 1;
1817 			continue;
1818 		}
1819 
1820 		/* grab the next available node */
1821 		n = get_child_rcu(pn, cindex++);
1822 		if (!n)
1823 			continue;
1824 
1825 		/* no need to compare keys since we bumped the index */
1826 		if (IS_LEAF(n))
1827 			goto found;
1828 
1829 		/* Rescan start scanning in new node */
1830 		pn = n;
1831 		cindex = 0;
1832 	}
1833 
1834 	*tn = pn;
1835 	return NULL; /* Root of trie */
1836 found:
1837 	/* if we are at the limit for keys just return NULL for the tnode */
1838 	*tn = pn;
1839 	return n;
1840 }
1841 
1842 static void fib_trie_free(struct fib_table *tb)
1843 {
1844 	struct trie *t = (struct trie *)tb->tb_data;
1845 	struct key_vector *pn = t->kv;
1846 	unsigned long cindex = 1;
1847 	struct hlist_node *tmp;
1848 	struct fib_alias *fa;
1849 
1850 	/* walk trie in reverse order and free everything */
1851 	for (;;) {
1852 		struct key_vector *n;
1853 
1854 		if (!(cindex--)) {
1855 			t_key pkey = pn->key;
1856 
1857 			if (IS_TRIE(pn))
1858 				break;
1859 
1860 			n = pn;
1861 			pn = node_parent(pn);
1862 
1863 			/* drop emptied tnode */
1864 			put_child_root(pn, n->key, NULL);
1865 			node_free(n);
1866 
1867 			cindex = get_index(pkey, pn);
1868 
1869 			continue;
1870 		}
1871 
1872 		/* grab the next available node */
1873 		n = get_child(pn, cindex);
1874 		if (!n)
1875 			continue;
1876 
1877 		if (IS_TNODE(n)) {
1878 			/* record pn and cindex for leaf walking */
1879 			pn = n;
1880 			cindex = 1ul << n->bits;
1881 
1882 			continue;
1883 		}
1884 
1885 		hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1886 			hlist_del_rcu(&fa->fa_list);
1887 			alias_free_mem_rcu(fa);
1888 		}
1889 
1890 		put_child_root(pn, n->key, NULL);
1891 		node_free(n);
1892 	}
1893 
1894 #ifdef CONFIG_IP_FIB_TRIE_STATS
1895 	free_percpu(t->stats);
1896 #endif
1897 	kfree(tb);
1898 }
1899 
1900 struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1901 {
1902 	struct trie *ot = (struct trie *)oldtb->tb_data;
1903 	struct key_vector *l, *tp = ot->kv;
1904 	struct fib_table *local_tb;
1905 	struct fib_alias *fa;
1906 	struct trie *lt;
1907 	t_key key = 0;
1908 
1909 	if (oldtb->tb_data == oldtb->__data)
1910 		return oldtb;
1911 
1912 	local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1913 	if (!local_tb)
1914 		return NULL;
1915 
1916 	lt = (struct trie *)local_tb->tb_data;
1917 
1918 	while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1919 		struct key_vector *local_l = NULL, *local_tp;
1920 
1921 		hlist_for_each_entry(fa, &l->leaf, fa_list) {
1922 			struct fib_alias *new_fa;
1923 
1924 			if (local_tb->tb_id != fa->tb_id)
1925 				continue;
1926 
1927 			/* clone fa for new local table */
1928 			new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1929 			if (!new_fa)
1930 				goto out;
1931 
1932 			memcpy(new_fa, fa, sizeof(*fa));
1933 
1934 			/* insert clone into table */
1935 			if (!local_l)
1936 				local_l = fib_find_node(lt, &local_tp, l->key);
1937 
1938 			if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1939 					     NULL, l->key)) {
1940 				kmem_cache_free(fn_alias_kmem, new_fa);
1941 				goto out;
1942 			}
1943 		}
1944 
1945 		/* stop loop if key wrapped back to 0 */
1946 		key = l->key + 1;
1947 		if (key < l->key)
1948 			break;
1949 	}
1950 
1951 	return local_tb;
1952 out:
1953 	fib_trie_free(local_tb);
1954 
1955 	return NULL;
1956 }
1957 
1958 /* Caller must hold RTNL */
1959 void fib_table_flush_external(struct fib_table *tb)
1960 {
1961 	struct trie *t = (struct trie *)tb->tb_data;
1962 	struct key_vector *pn = t->kv;
1963 	unsigned long cindex = 1;
1964 	struct hlist_node *tmp;
1965 	struct fib_alias *fa;
1966 
1967 	/* walk trie in reverse order */
1968 	for (;;) {
1969 		unsigned char slen = 0;
1970 		struct key_vector *n;
1971 
1972 		if (!(cindex--)) {
1973 			t_key pkey = pn->key;
1974 
1975 			/* cannot resize the trie vector */
1976 			if (IS_TRIE(pn))
1977 				break;
1978 
1979 			/* update the suffix to address pulled leaves */
1980 			if (pn->slen > pn->pos)
1981 				update_suffix(pn);
1982 
1983 			/* resize completed node */
1984 			pn = resize(t, pn);
1985 			cindex = get_index(pkey, pn);
1986 
1987 			continue;
1988 		}
1989 
1990 		/* grab the next available node */
1991 		n = get_child(pn, cindex);
1992 		if (!n)
1993 			continue;
1994 
1995 		if (IS_TNODE(n)) {
1996 			/* record pn and cindex for leaf walking */
1997 			pn = n;
1998 			cindex = 1ul << n->bits;
1999 
2000 			continue;
2001 		}
2002 
2003 		hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
2004 			/* if alias was cloned to local then we just
2005 			 * need to remove the local copy from main
2006 			 */
2007 			if (tb->tb_id != fa->tb_id) {
2008 				hlist_del_rcu(&fa->fa_list);
2009 				alias_free_mem_rcu(fa);
2010 				continue;
2011 			}
2012 
2013 			/* record local slen */
2014 			slen = fa->fa_slen;
2015 		}
2016 
2017 		/* update leaf slen */
2018 		n->slen = slen;
2019 
2020 		if (hlist_empty(&n->leaf)) {
2021 			put_child_root(pn, n->key, NULL);
2022 			node_free(n);
2023 		}
2024 	}
2025 }
2026 
2027 /* Caller must hold RTNL. */
2028 int fib_table_flush(struct net *net, struct fib_table *tb, bool flush_all)
2029 {
2030 	struct trie *t = (struct trie *)tb->tb_data;
2031 	struct nl_info info = { .nl_net = net };
2032 	struct key_vector *pn = t->kv;
2033 	unsigned long cindex = 1;
2034 	struct hlist_node *tmp;
2035 	struct fib_alias *fa;
2036 	int found = 0;
2037 
2038 	/* walk trie in reverse order */
2039 	for (;;) {
2040 		unsigned char slen = 0;
2041 		struct key_vector *n;
2042 
2043 		if (!(cindex--)) {
2044 			t_key pkey = pn->key;
2045 
2046 			/* cannot resize the trie vector */
2047 			if (IS_TRIE(pn))
2048 				break;
2049 
2050 			/* update the suffix to address pulled leaves */
2051 			if (pn->slen > pn->pos)
2052 				update_suffix(pn);
2053 
2054 			/* resize completed node */
2055 			pn = resize(t, pn);
2056 			cindex = get_index(pkey, pn);
2057 
2058 			continue;
2059 		}
2060 
2061 		/* grab the next available node */
2062 		n = get_child(pn, cindex);
2063 		if (!n)
2064 			continue;
2065 
2066 		if (IS_TNODE(n)) {
2067 			/* record pn and cindex for leaf walking */
2068 			pn = n;
2069 			cindex = 1ul << n->bits;
2070 
2071 			continue;
2072 		}
2073 
2074 		hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
2075 			struct fib_info *fi = fa->fa_info;
2076 
2077 			if (!fi || tb->tb_id != fa->tb_id ||
2078 			    (!(fi->fib_flags & RTNH_F_DEAD) &&
2079 			     !fib_props[fa->fa_type].error)) {
2080 				slen = fa->fa_slen;
2081 				continue;
2082 			}
2083 
2084 			/* Do not flush error routes if network namespace is
2085 			 * not being dismantled
2086 			 */
2087 			if (!flush_all && fib_props[fa->fa_type].error) {
2088 				slen = fa->fa_slen;
2089 				continue;
2090 			}
2091 
2092 			fib_notify_alias_delete(net, n->key, &n->leaf, fa,
2093 						NULL);
2094 			if (fi->pfsrc_removed)
2095 				rtmsg_fib(RTM_DELROUTE, htonl(n->key), fa,
2096 					  KEYLENGTH - fa->fa_slen, tb->tb_id, &info, 0);
2097 			hlist_del_rcu(&fa->fa_list);
2098 			fib_release_info(fa->fa_info);
2099 			alias_free_mem_rcu(fa);
2100 			found++;
2101 		}
2102 
2103 		/* update leaf slen */
2104 		n->slen = slen;
2105 
2106 		if (hlist_empty(&n->leaf)) {
2107 			put_child_root(pn, n->key, NULL);
2108 			node_free(n);
2109 		}
2110 	}
2111 
2112 	pr_debug("trie_flush found=%d\n", found);
2113 	return found;
2114 }
2115 
2116 /* derived from fib_trie_free */
2117 static void __fib_info_notify_update(struct net *net, struct fib_table *tb,
2118 				     struct nl_info *info)
2119 {
2120 	struct trie *t = (struct trie *)tb->tb_data;
2121 	struct key_vector *pn = t->kv;
2122 	unsigned long cindex = 1;
2123 	struct fib_alias *fa;
2124 
2125 	for (;;) {
2126 		struct key_vector *n;
2127 
2128 		if (!(cindex--)) {
2129 			t_key pkey = pn->key;
2130 
2131 			if (IS_TRIE(pn))
2132 				break;
2133 
2134 			pn = node_parent(pn);
2135 			cindex = get_index(pkey, pn);
2136 			continue;
2137 		}
2138 
2139 		/* grab the next available node */
2140 		n = get_child(pn, cindex);
2141 		if (!n)
2142 			continue;
2143 
2144 		if (IS_TNODE(n)) {
2145 			/* record pn and cindex for leaf walking */
2146 			pn = n;
2147 			cindex = 1ul << n->bits;
2148 
2149 			continue;
2150 		}
2151 
2152 		hlist_for_each_entry(fa, &n->leaf, fa_list) {
2153 			struct fib_info *fi = fa->fa_info;
2154 
2155 			if (!fi || !fi->nh_updated || fa->tb_id != tb->tb_id)
2156 				continue;
2157 
2158 			rtmsg_fib(RTM_NEWROUTE, htonl(n->key), fa,
2159 				  KEYLENGTH - fa->fa_slen, tb->tb_id,
2160 				  info, NLM_F_REPLACE);
2161 		}
2162 	}
2163 }
2164 
2165 void fib_info_notify_update(struct net *net, struct nl_info *info)
2166 {
2167 	unsigned int h;
2168 
2169 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2170 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2171 		struct fib_table *tb;
2172 
2173 		hlist_for_each_entry_rcu(tb, head, tb_hlist,
2174 					 lockdep_rtnl_is_held())
2175 			__fib_info_notify_update(net, tb, info);
2176 	}
2177 }
2178 
2179 static int fib_leaf_notify(struct key_vector *l, struct fib_table *tb,
2180 			   struct notifier_block *nb,
2181 			   struct netlink_ext_ack *extack)
2182 {
2183 	struct fib_alias *fa;
2184 	int last_slen = -1;
2185 	int err;
2186 
2187 	hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2188 		struct fib_info *fi = fa->fa_info;
2189 
2190 		if (!fi)
2191 			continue;
2192 
2193 		/* local and main table can share the same trie,
2194 		 * so don't notify twice for the same entry.
2195 		 */
2196 		if (tb->tb_id != fa->tb_id)
2197 			continue;
2198 
2199 		if (fa->fa_slen == last_slen)
2200 			continue;
2201 
2202 		last_slen = fa->fa_slen;
2203 		err = call_fib_entry_notifier(nb, FIB_EVENT_ENTRY_REPLACE,
2204 					      l->key, KEYLENGTH - fa->fa_slen,
2205 					      fa, extack);
2206 		if (err)
2207 			return err;
2208 	}
2209 	return 0;
2210 }
2211 
2212 static int fib_table_notify(struct fib_table *tb, struct notifier_block *nb,
2213 			    struct netlink_ext_ack *extack)
2214 {
2215 	struct trie *t = (struct trie *)tb->tb_data;
2216 	struct key_vector *l, *tp = t->kv;
2217 	t_key key = 0;
2218 	int err;
2219 
2220 	while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2221 		err = fib_leaf_notify(l, tb, nb, extack);
2222 		if (err)
2223 			return err;
2224 
2225 		key = l->key + 1;
2226 		/* stop in case of wrap around */
2227 		if (key < l->key)
2228 			break;
2229 	}
2230 	return 0;
2231 }
2232 
2233 int fib_notify(struct net *net, struct notifier_block *nb,
2234 	       struct netlink_ext_ack *extack)
2235 {
2236 	unsigned int h;
2237 	int err;
2238 
2239 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2240 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2241 		struct fib_table *tb;
2242 
2243 		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2244 			err = fib_table_notify(tb, nb, extack);
2245 			if (err)
2246 				return err;
2247 		}
2248 	}
2249 	return 0;
2250 }
2251 
2252 static void __trie_free_rcu(struct rcu_head *head)
2253 {
2254 	struct fib_table *tb = container_of(head, struct fib_table, rcu);
2255 #ifdef CONFIG_IP_FIB_TRIE_STATS
2256 	struct trie *t = (struct trie *)tb->tb_data;
2257 
2258 	if (tb->tb_data == tb->__data)
2259 		free_percpu(t->stats);
2260 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2261 	kfree(tb);
2262 }
2263 
2264 void fib_free_table(struct fib_table *tb)
2265 {
2266 	call_rcu(&tb->rcu, __trie_free_rcu);
2267 }
2268 
2269 static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
2270 			     struct sk_buff *skb, struct netlink_callback *cb,
2271 			     struct fib_dump_filter *filter)
2272 {
2273 	unsigned int flags = NLM_F_MULTI;
2274 	__be32 xkey = htonl(l->key);
2275 	int i, s_i, i_fa, s_fa, err;
2276 	struct fib_alias *fa;
2277 
2278 	if (filter->filter_set ||
2279 	    !filter->dump_exceptions || !filter->dump_routes)
2280 		flags |= NLM_F_DUMP_FILTERED;
2281 
2282 	s_i = cb->args[4];
2283 	s_fa = cb->args[5];
2284 	i = 0;
2285 
2286 	/* rcu_read_lock is hold by caller */
2287 	hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2288 		struct fib_info *fi = fa->fa_info;
2289 
2290 		if (i < s_i)
2291 			goto next;
2292 
2293 		i_fa = 0;
2294 
2295 		if (tb->tb_id != fa->tb_id)
2296 			goto next;
2297 
2298 		if (filter->filter_set) {
2299 			if (filter->rt_type && fa->fa_type != filter->rt_type)
2300 				goto next;
2301 
2302 			if ((filter->protocol &&
2303 			     fi->fib_protocol != filter->protocol))
2304 				goto next;
2305 
2306 			if (filter->dev &&
2307 			    !fib_info_nh_uses_dev(fi, filter->dev))
2308 				goto next;
2309 		}
2310 
2311 		if (filter->dump_routes) {
2312 			if (!s_fa) {
2313 				struct fib_rt_info fri;
2314 
2315 				fri.fi = fi;
2316 				fri.tb_id = tb->tb_id;
2317 				fri.dst = xkey;
2318 				fri.dst_len = KEYLENGTH - fa->fa_slen;
2319 				fri.dscp = fa->fa_dscp;
2320 				fri.type = fa->fa_type;
2321 				fri.offload = READ_ONCE(fa->offload);
2322 				fri.trap = READ_ONCE(fa->trap);
2323 				fri.offload_failed = READ_ONCE(fa->offload_failed);
2324 				err = fib_dump_info(skb,
2325 						    NETLINK_CB(cb->skb).portid,
2326 						    cb->nlh->nlmsg_seq,
2327 						    RTM_NEWROUTE, &fri, flags);
2328 				if (err < 0)
2329 					goto stop;
2330 			}
2331 
2332 			i_fa++;
2333 		}
2334 
2335 		if (filter->dump_exceptions) {
2336 			err = fib_dump_info_fnhe(skb, cb, tb->tb_id, fi,
2337 						 &i_fa, s_fa, flags);
2338 			if (err < 0)
2339 				goto stop;
2340 		}
2341 
2342 next:
2343 		i++;
2344 	}
2345 
2346 	cb->args[4] = i;
2347 	return skb->len;
2348 
2349 stop:
2350 	cb->args[4] = i;
2351 	cb->args[5] = i_fa;
2352 	return err;
2353 }
2354 
2355 /* rcu_read_lock needs to be hold by caller from readside */
2356 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
2357 		   struct netlink_callback *cb, struct fib_dump_filter *filter)
2358 {
2359 	struct trie *t = (struct trie *)tb->tb_data;
2360 	struct key_vector *l, *tp = t->kv;
2361 	/* Dump starting at last key.
2362 	 * Note: 0.0.0.0/0 (ie default) is first key.
2363 	 */
2364 	int count = cb->args[2];
2365 	t_key key = cb->args[3];
2366 
2367 	/* First time here, count and key are both always 0. Count > 0
2368 	 * and key == 0 means the dump has wrapped around and we are done.
2369 	 */
2370 	if (count && !key)
2371 		return skb->len;
2372 
2373 	while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2374 		int err;
2375 
2376 		err = fn_trie_dump_leaf(l, tb, skb, cb, filter);
2377 		if (err < 0) {
2378 			cb->args[3] = key;
2379 			cb->args[2] = count;
2380 			return err;
2381 		}
2382 
2383 		++count;
2384 		key = l->key + 1;
2385 
2386 		memset(&cb->args[4], 0,
2387 		       sizeof(cb->args) - 4*sizeof(cb->args[0]));
2388 
2389 		/* stop loop if key wrapped back to 0 */
2390 		if (key < l->key)
2391 			break;
2392 	}
2393 
2394 	cb->args[3] = key;
2395 	cb->args[2] = count;
2396 
2397 	return skb->len;
2398 }
2399 
2400 void __init fib_trie_init(void)
2401 {
2402 	fn_alias_kmem = kmem_cache_create("ip_fib_alias",
2403 					  sizeof(struct fib_alias),
2404 					  0, SLAB_PANIC | SLAB_ACCOUNT, NULL);
2405 
2406 	trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
2407 					   LEAF_SIZE,
2408 					   0, SLAB_PANIC | SLAB_ACCOUNT, NULL);
2409 }
2410 
2411 struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
2412 {
2413 	struct fib_table *tb;
2414 	struct trie *t;
2415 	size_t sz = sizeof(*tb);
2416 
2417 	if (!alias)
2418 		sz += sizeof(struct trie);
2419 
2420 	tb = kzalloc(sz, GFP_KERNEL);
2421 	if (!tb)
2422 		return NULL;
2423 
2424 	tb->tb_id = id;
2425 	tb->tb_num_default = 0;
2426 	tb->tb_data = (alias ? alias->__data : tb->__data);
2427 
2428 	if (alias)
2429 		return tb;
2430 
2431 	t = (struct trie *) tb->tb_data;
2432 	t->kv[0].pos = KEYLENGTH;
2433 	t->kv[0].slen = KEYLENGTH;
2434 #ifdef CONFIG_IP_FIB_TRIE_STATS
2435 	t->stats = alloc_percpu(struct trie_use_stats);
2436 	if (!t->stats) {
2437 		kfree(tb);
2438 		tb = NULL;
2439 	}
2440 #endif
2441 
2442 	return tb;
2443 }
2444 
2445 #ifdef CONFIG_PROC_FS
2446 /* Depth first Trie walk iterator */
2447 struct fib_trie_iter {
2448 	struct seq_net_private p;
2449 	struct fib_table *tb;
2450 	struct key_vector *tnode;
2451 	unsigned int index;
2452 	unsigned int depth;
2453 };
2454 
2455 static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2456 {
2457 	unsigned long cindex = iter->index;
2458 	struct key_vector *pn = iter->tnode;
2459 	t_key pkey;
2460 
2461 	pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2462 		 iter->tnode, iter->index, iter->depth);
2463 
2464 	while (!IS_TRIE(pn)) {
2465 		while (cindex < child_length(pn)) {
2466 			struct key_vector *n = get_child_rcu(pn, cindex++);
2467 
2468 			if (!n)
2469 				continue;
2470 
2471 			if (IS_LEAF(n)) {
2472 				iter->tnode = pn;
2473 				iter->index = cindex;
2474 			} else {
2475 				/* push down one level */
2476 				iter->tnode = n;
2477 				iter->index = 0;
2478 				++iter->depth;
2479 			}
2480 
2481 			return n;
2482 		}
2483 
2484 		/* Current node exhausted, pop back up */
2485 		pkey = pn->key;
2486 		pn = node_parent_rcu(pn);
2487 		cindex = get_index(pkey, pn) + 1;
2488 		--iter->depth;
2489 	}
2490 
2491 	/* record root node so further searches know we are done */
2492 	iter->tnode = pn;
2493 	iter->index = 0;
2494 
2495 	return NULL;
2496 }
2497 
2498 static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2499 					     struct trie *t)
2500 {
2501 	struct key_vector *n, *pn;
2502 
2503 	if (!t)
2504 		return NULL;
2505 
2506 	pn = t->kv;
2507 	n = rcu_dereference(pn->tnode[0]);
2508 	if (!n)
2509 		return NULL;
2510 
2511 	if (IS_TNODE(n)) {
2512 		iter->tnode = n;
2513 		iter->index = 0;
2514 		iter->depth = 1;
2515 	} else {
2516 		iter->tnode = pn;
2517 		iter->index = 0;
2518 		iter->depth = 0;
2519 	}
2520 
2521 	return n;
2522 }
2523 
2524 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2525 {
2526 	struct key_vector *n;
2527 	struct fib_trie_iter iter;
2528 
2529 	memset(s, 0, sizeof(*s));
2530 
2531 	rcu_read_lock();
2532 	for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2533 		if (IS_LEAF(n)) {
2534 			struct fib_alias *fa;
2535 
2536 			s->leaves++;
2537 			s->totdepth += iter.depth;
2538 			if (iter.depth > s->maxdepth)
2539 				s->maxdepth = iter.depth;
2540 
2541 			hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2542 				++s->prefixes;
2543 		} else {
2544 			s->tnodes++;
2545 			if (n->bits < MAX_STAT_DEPTH)
2546 				s->nodesizes[n->bits]++;
2547 			s->nullpointers += tn_info(n)->empty_children;
2548 		}
2549 	}
2550 	rcu_read_unlock();
2551 }
2552 
2553 /*
2554  *	This outputs /proc/net/fib_triestats
2555  */
2556 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2557 {
2558 	unsigned int i, max, pointers, bytes, avdepth;
2559 
2560 	if (stat->leaves)
2561 		avdepth = stat->totdepth*100 / stat->leaves;
2562 	else
2563 		avdepth = 0;
2564 
2565 	seq_printf(seq, "\tAver depth:     %u.%02d\n",
2566 		   avdepth / 100, avdepth % 100);
2567 	seq_printf(seq, "\tMax depth:      %u\n", stat->maxdepth);
2568 
2569 	seq_printf(seq, "\tLeaves:         %u\n", stat->leaves);
2570 	bytes = LEAF_SIZE * stat->leaves;
2571 
2572 	seq_printf(seq, "\tPrefixes:       %u\n", stat->prefixes);
2573 	bytes += sizeof(struct fib_alias) * stat->prefixes;
2574 
2575 	seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2576 	bytes += TNODE_SIZE(0) * stat->tnodes;
2577 
2578 	max = MAX_STAT_DEPTH;
2579 	while (max > 0 && stat->nodesizes[max-1] == 0)
2580 		max--;
2581 
2582 	pointers = 0;
2583 	for (i = 1; i < max; i++)
2584 		if (stat->nodesizes[i] != 0) {
2585 			seq_printf(seq, "  %u: %u",  i, stat->nodesizes[i]);
2586 			pointers += (1<<i) * stat->nodesizes[i];
2587 		}
2588 	seq_putc(seq, '\n');
2589 	seq_printf(seq, "\tPointers: %u\n", pointers);
2590 
2591 	bytes += sizeof(struct key_vector *) * pointers;
2592 	seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2593 	seq_printf(seq, "Total size: %u  kB\n", (bytes + 1023) / 1024);
2594 }
2595 
2596 #ifdef CONFIG_IP_FIB_TRIE_STATS
2597 static void trie_show_usage(struct seq_file *seq,
2598 			    const struct trie_use_stats __percpu *stats)
2599 {
2600 	struct trie_use_stats s = { 0 };
2601 	int cpu;
2602 
2603 	/* loop through all of the CPUs and gather up the stats */
2604 	for_each_possible_cpu(cpu) {
2605 		const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2606 
2607 		s.gets += pcpu->gets;
2608 		s.backtrack += pcpu->backtrack;
2609 		s.semantic_match_passed += pcpu->semantic_match_passed;
2610 		s.semantic_match_miss += pcpu->semantic_match_miss;
2611 		s.null_node_hit += pcpu->null_node_hit;
2612 		s.resize_node_skipped += pcpu->resize_node_skipped;
2613 	}
2614 
2615 	seq_printf(seq, "\nCounters:\n---------\n");
2616 	seq_printf(seq, "gets = %u\n", s.gets);
2617 	seq_printf(seq, "backtracks = %u\n", s.backtrack);
2618 	seq_printf(seq, "semantic match passed = %u\n",
2619 		   s.semantic_match_passed);
2620 	seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2621 	seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2622 	seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2623 }
2624 #endif /*  CONFIG_IP_FIB_TRIE_STATS */
2625 
2626 static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2627 {
2628 	if (tb->tb_id == RT_TABLE_LOCAL)
2629 		seq_puts(seq, "Local:\n");
2630 	else if (tb->tb_id == RT_TABLE_MAIN)
2631 		seq_puts(seq, "Main:\n");
2632 	else
2633 		seq_printf(seq, "Id %d:\n", tb->tb_id);
2634 }
2635 
2636 
2637 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2638 {
2639 	struct net *net = seq->private;
2640 	unsigned int h;
2641 
2642 	seq_printf(seq,
2643 		   "Basic info: size of leaf:"
2644 		   " %zd bytes, size of tnode: %zd bytes.\n",
2645 		   LEAF_SIZE, TNODE_SIZE(0));
2646 
2647 	rcu_read_lock();
2648 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2649 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2650 		struct fib_table *tb;
2651 
2652 		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2653 			struct trie *t = (struct trie *) tb->tb_data;
2654 			struct trie_stat stat;
2655 
2656 			if (!t)
2657 				continue;
2658 
2659 			fib_table_print(seq, tb);
2660 
2661 			trie_collect_stats(t, &stat);
2662 			trie_show_stats(seq, &stat);
2663 #ifdef CONFIG_IP_FIB_TRIE_STATS
2664 			trie_show_usage(seq, t->stats);
2665 #endif
2666 		}
2667 		cond_resched_rcu();
2668 	}
2669 	rcu_read_unlock();
2670 
2671 	return 0;
2672 }
2673 
2674 static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2675 {
2676 	struct fib_trie_iter *iter = seq->private;
2677 	struct net *net = seq_file_net(seq);
2678 	loff_t idx = 0;
2679 	unsigned int h;
2680 
2681 	for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2682 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2683 		struct fib_table *tb;
2684 
2685 		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2686 			struct key_vector *n;
2687 
2688 			for (n = fib_trie_get_first(iter,
2689 						    (struct trie *) tb->tb_data);
2690 			     n; n = fib_trie_get_next(iter))
2691 				if (pos == idx++) {
2692 					iter->tb = tb;
2693 					return n;
2694 				}
2695 		}
2696 	}
2697 
2698 	return NULL;
2699 }
2700 
2701 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2702 	__acquires(RCU)
2703 {
2704 	rcu_read_lock();
2705 	return fib_trie_get_idx(seq, *pos);
2706 }
2707 
2708 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2709 {
2710 	struct fib_trie_iter *iter = seq->private;
2711 	struct net *net = seq_file_net(seq);
2712 	struct fib_table *tb = iter->tb;
2713 	struct hlist_node *tb_node;
2714 	unsigned int h;
2715 	struct key_vector *n;
2716 
2717 	++*pos;
2718 	/* next node in same table */
2719 	n = fib_trie_get_next(iter);
2720 	if (n)
2721 		return n;
2722 
2723 	/* walk rest of this hash chain */
2724 	h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2725 	while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2726 		tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2727 		n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2728 		if (n)
2729 			goto found;
2730 	}
2731 
2732 	/* new hash chain */
2733 	while (++h < FIB_TABLE_HASHSZ) {
2734 		struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2735 		hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2736 			n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2737 			if (n)
2738 				goto found;
2739 		}
2740 	}
2741 	return NULL;
2742 
2743 found:
2744 	iter->tb = tb;
2745 	return n;
2746 }
2747 
2748 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2749 	__releases(RCU)
2750 {
2751 	rcu_read_unlock();
2752 }
2753 
2754 static void seq_indent(struct seq_file *seq, int n)
2755 {
2756 	while (n-- > 0)
2757 		seq_puts(seq, "   ");
2758 }
2759 
2760 static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2761 {
2762 	switch (s) {
2763 	case RT_SCOPE_UNIVERSE: return "universe";
2764 	case RT_SCOPE_SITE:	return "site";
2765 	case RT_SCOPE_LINK:	return "link";
2766 	case RT_SCOPE_HOST:	return "host";
2767 	case RT_SCOPE_NOWHERE:	return "nowhere";
2768 	default:
2769 		snprintf(buf, len, "scope=%d", s);
2770 		return buf;
2771 	}
2772 }
2773 
2774 static const char *const rtn_type_names[__RTN_MAX] = {
2775 	[RTN_UNSPEC] = "UNSPEC",
2776 	[RTN_UNICAST] = "UNICAST",
2777 	[RTN_LOCAL] = "LOCAL",
2778 	[RTN_BROADCAST] = "BROADCAST",
2779 	[RTN_ANYCAST] = "ANYCAST",
2780 	[RTN_MULTICAST] = "MULTICAST",
2781 	[RTN_BLACKHOLE] = "BLACKHOLE",
2782 	[RTN_UNREACHABLE] = "UNREACHABLE",
2783 	[RTN_PROHIBIT] = "PROHIBIT",
2784 	[RTN_THROW] = "THROW",
2785 	[RTN_NAT] = "NAT",
2786 	[RTN_XRESOLVE] = "XRESOLVE",
2787 };
2788 
2789 static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2790 {
2791 	if (t < __RTN_MAX && rtn_type_names[t])
2792 		return rtn_type_names[t];
2793 	snprintf(buf, len, "type %u", t);
2794 	return buf;
2795 }
2796 
2797 /* Pretty print the trie */
2798 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2799 {
2800 	const struct fib_trie_iter *iter = seq->private;
2801 	struct key_vector *n = v;
2802 
2803 	if (IS_TRIE(node_parent_rcu(n)))
2804 		fib_table_print(seq, iter->tb);
2805 
2806 	if (IS_TNODE(n)) {
2807 		__be32 prf = htonl(n->key);
2808 
2809 		seq_indent(seq, iter->depth-1);
2810 		seq_printf(seq, "  +-- %pI4/%zu %u %u %u\n",
2811 			   &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2812 			   tn_info(n)->full_children,
2813 			   tn_info(n)->empty_children);
2814 	} else {
2815 		__be32 val = htonl(n->key);
2816 		struct fib_alias *fa;
2817 
2818 		seq_indent(seq, iter->depth);
2819 		seq_printf(seq, "  |-- %pI4\n", &val);
2820 
2821 		hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2822 			char buf1[32], buf2[32];
2823 
2824 			seq_indent(seq, iter->depth + 1);
2825 			seq_printf(seq, "  /%zu %s %s",
2826 				   KEYLENGTH - fa->fa_slen,
2827 				   rtn_scope(buf1, sizeof(buf1),
2828 					     fa->fa_info->fib_scope),
2829 				   rtn_type(buf2, sizeof(buf2),
2830 					    fa->fa_type));
2831 			if (fa->fa_dscp)
2832 				seq_printf(seq, " tos=%d",
2833 					   inet_dscp_to_dsfield(fa->fa_dscp));
2834 			seq_putc(seq, '\n');
2835 		}
2836 	}
2837 
2838 	return 0;
2839 }
2840 
2841 static const struct seq_operations fib_trie_seq_ops = {
2842 	.start  = fib_trie_seq_start,
2843 	.next   = fib_trie_seq_next,
2844 	.stop   = fib_trie_seq_stop,
2845 	.show   = fib_trie_seq_show,
2846 };
2847 
2848 struct fib_route_iter {
2849 	struct seq_net_private p;
2850 	struct fib_table *main_tb;
2851 	struct key_vector *tnode;
2852 	loff_t	pos;
2853 	t_key	key;
2854 };
2855 
2856 static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2857 					    loff_t pos)
2858 {
2859 	struct key_vector *l, **tp = &iter->tnode;
2860 	t_key key;
2861 
2862 	/* use cached location of previously found key */
2863 	if (iter->pos > 0 && pos >= iter->pos) {
2864 		key = iter->key;
2865 	} else {
2866 		iter->pos = 1;
2867 		key = 0;
2868 	}
2869 
2870 	pos -= iter->pos;
2871 
2872 	while ((l = leaf_walk_rcu(tp, key)) && (pos-- > 0)) {
2873 		key = l->key + 1;
2874 		iter->pos++;
2875 		l = NULL;
2876 
2877 		/* handle unlikely case of a key wrap */
2878 		if (!key)
2879 			break;
2880 	}
2881 
2882 	if (l)
2883 		iter->key = l->key;	/* remember it */
2884 	else
2885 		iter->pos = 0;		/* forget it */
2886 
2887 	return l;
2888 }
2889 
2890 static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2891 	__acquires(RCU)
2892 {
2893 	struct fib_route_iter *iter = seq->private;
2894 	struct fib_table *tb;
2895 	struct trie *t;
2896 
2897 	rcu_read_lock();
2898 
2899 	tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2900 	if (!tb)
2901 		return NULL;
2902 
2903 	iter->main_tb = tb;
2904 	t = (struct trie *)tb->tb_data;
2905 	iter->tnode = t->kv;
2906 
2907 	if (*pos != 0)
2908 		return fib_route_get_idx(iter, *pos);
2909 
2910 	iter->pos = 0;
2911 	iter->key = KEY_MAX;
2912 
2913 	return SEQ_START_TOKEN;
2914 }
2915 
2916 static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2917 {
2918 	struct fib_route_iter *iter = seq->private;
2919 	struct key_vector *l = NULL;
2920 	t_key key = iter->key + 1;
2921 
2922 	++*pos;
2923 
2924 	/* only allow key of 0 for start of sequence */
2925 	if ((v == SEQ_START_TOKEN) || key)
2926 		l = leaf_walk_rcu(&iter->tnode, key);
2927 
2928 	if (l) {
2929 		iter->key = l->key;
2930 		iter->pos++;
2931 	} else {
2932 		iter->pos = 0;
2933 	}
2934 
2935 	return l;
2936 }
2937 
2938 static void fib_route_seq_stop(struct seq_file *seq, void *v)
2939 	__releases(RCU)
2940 {
2941 	rcu_read_unlock();
2942 }
2943 
2944 static unsigned int fib_flag_trans(int type, __be32 mask, struct fib_info *fi)
2945 {
2946 	unsigned int flags = 0;
2947 
2948 	if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2949 		flags = RTF_REJECT;
2950 	if (fi) {
2951 		const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
2952 
2953 		if (nhc->nhc_gw.ipv4)
2954 			flags |= RTF_GATEWAY;
2955 	}
2956 	if (mask == htonl(0xFFFFFFFF))
2957 		flags |= RTF_HOST;
2958 	flags |= RTF_UP;
2959 	return flags;
2960 }
2961 
2962 /*
2963  *	This outputs /proc/net/route.
2964  *	The format of the file is not supposed to be changed
2965  *	and needs to be same as fib_hash output to avoid breaking
2966  *	legacy utilities
2967  */
2968 static int fib_route_seq_show(struct seq_file *seq, void *v)
2969 {
2970 	struct fib_route_iter *iter = seq->private;
2971 	struct fib_table *tb = iter->main_tb;
2972 	struct fib_alias *fa;
2973 	struct key_vector *l = v;
2974 	__be32 prefix;
2975 
2976 	if (v == SEQ_START_TOKEN) {
2977 		seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2978 			   "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2979 			   "\tWindow\tIRTT");
2980 		return 0;
2981 	}
2982 
2983 	prefix = htonl(l->key);
2984 
2985 	hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2986 		struct fib_info *fi = fa->fa_info;
2987 		__be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2988 		unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2989 
2990 		if ((fa->fa_type == RTN_BROADCAST) ||
2991 		    (fa->fa_type == RTN_MULTICAST))
2992 			continue;
2993 
2994 		if (fa->tb_id != tb->tb_id)
2995 			continue;
2996 
2997 		seq_setwidth(seq, 127);
2998 
2999 		if (fi) {
3000 			struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
3001 			__be32 gw = 0;
3002 
3003 			if (nhc->nhc_gw_family == AF_INET)
3004 				gw = nhc->nhc_gw.ipv4;
3005 
3006 			seq_printf(seq,
3007 				   "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
3008 				   "%d\t%08X\t%d\t%u\t%u",
3009 				   nhc->nhc_dev ? nhc->nhc_dev->name : "*",
3010 				   prefix, gw, flags, 0, 0,
3011 				   fi->fib_priority,
3012 				   mask,
3013 				   (fi->fib_advmss ?
3014 				    fi->fib_advmss + 40 : 0),
3015 				   fi->fib_window,
3016 				   fi->fib_rtt >> 3);
3017 		} else {
3018 			seq_printf(seq,
3019 				   "*\t%08X\t%08X\t%04X\t%d\t%u\t"
3020 				   "%d\t%08X\t%d\t%u\t%u",
3021 				   prefix, 0, flags, 0, 0, 0,
3022 				   mask, 0, 0, 0);
3023 		}
3024 		seq_pad(seq, '\n');
3025 	}
3026 
3027 	return 0;
3028 }
3029 
3030 static const struct seq_operations fib_route_seq_ops = {
3031 	.start  = fib_route_seq_start,
3032 	.next   = fib_route_seq_next,
3033 	.stop   = fib_route_seq_stop,
3034 	.show   = fib_route_seq_show,
3035 };
3036 
3037 int __net_init fib_proc_init(struct net *net)
3038 {
3039 	if (!proc_create_net("fib_trie", 0444, net->proc_net, &fib_trie_seq_ops,
3040 			sizeof(struct fib_trie_iter)))
3041 		goto out1;
3042 
3043 	if (!proc_create_net_single("fib_triestat", 0444, net->proc_net,
3044 			fib_triestat_seq_show, NULL))
3045 		goto out2;
3046 
3047 	if (!proc_create_net("route", 0444, net->proc_net, &fib_route_seq_ops,
3048 			sizeof(struct fib_route_iter)))
3049 		goto out3;
3050 
3051 	return 0;
3052 
3053 out3:
3054 	remove_proc_entry("fib_triestat", net->proc_net);
3055 out2:
3056 	remove_proc_entry("fib_trie", net->proc_net);
3057 out1:
3058 	return -ENOMEM;
3059 }
3060 
3061 void __net_exit fib_proc_exit(struct net *net)
3062 {
3063 	remove_proc_entry("fib_trie", net->proc_net);
3064 	remove_proc_entry("fib_triestat", net->proc_net);
3065 	remove_proc_entry("route", net->proc_net);
3066 }
3067 
3068 #endif /* CONFIG_PROC_FS */
3069