Loading...
1/*
2 * This program is free software; you can redistribute it and/or
3 * modify it under the terms of the GNU General Public License
4 * as published by the Free Software Foundation; either version
5 * 2 of the License, or (at your option) any later version.
6 *
7 * Robert Olsson <robert.olsson@its.uu.se> Uppsala Universitet
8 * & Swedish University of Agricultural Sciences.
9 *
10 * Jens Laas <jens.laas@data.slu.se> Swedish University of
11 * Agricultural Sciences.
12 *
13 * Hans Liss <hans.liss@its.uu.se> Uppsala Universitet
14 *
15 * This work is based on the LPC-trie which is originally described in:
16 *
17 * An experimental study of compression methods for dynamic tries
18 * Stefan Nilsson and Matti Tikkanen. Algorithmica, 33(1):19-33, 2002.
19 * http://www.csc.kth.se/~snilsson/software/dyntrie2/
20 *
21 *
22 * IP-address lookup using LC-tries. Stefan Nilsson and Gunnar Karlsson
23 * IEEE Journal on Selected Areas in Communications, 17(6):1083-1092, June 1999
24 *
25 *
26 * Code from fib_hash has been reused which includes the following header:
27 *
28 *
29 * INET An implementation of the TCP/IP protocol suite for the LINUX
30 * operating system. INET is implemented using the BSD Socket
31 * interface as the means of communication with the user level.
32 *
33 * IPv4 FIB: lookup engine and maintenance routines.
34 *
35 *
36 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
37 *
38 * This program is free software; you can redistribute it and/or
39 * modify it under the terms of the GNU General Public License
40 * as published by the Free Software Foundation; either version
41 * 2 of the License, or (at your option) any later version.
42 *
43 * Substantial contributions to this work comes from:
44 *
45 * David S. Miller, <davem@davemloft.net>
46 * Stephen Hemminger <shemminger@osdl.org>
47 * Paul E. McKenney <paulmck@us.ibm.com>
48 * Patrick McHardy <kaber@trash.net>
49 */
50
51#define VERSION "0.409"
52
53#include <asm/uaccess.h>
54#include <linux/bitops.h>
55#include <linux/types.h>
56#include <linux/kernel.h>
57#include <linux/mm.h>
58#include <linux/string.h>
59#include <linux/socket.h>
60#include <linux/sockios.h>
61#include <linux/errno.h>
62#include <linux/in.h>
63#include <linux/inet.h>
64#include <linux/inetdevice.h>
65#include <linux/netdevice.h>
66#include <linux/if_arp.h>
67#include <linux/proc_fs.h>
68#include <linux/rcupdate.h>
69#include <linux/skbuff.h>
70#include <linux/netlink.h>
71#include <linux/init.h>
72#include <linux/list.h>
73#include <linux/slab.h>
74#include <linux/prefetch.h>
75#include <linux/export.h>
76#include <net/net_namespace.h>
77#include <net/ip.h>
78#include <net/protocol.h>
79#include <net/route.h>
80#include <net/tcp.h>
81#include <net/sock.h>
82#include <net/ip_fib.h>
83#include "fib_lookup.h"
84
85#define MAX_STAT_DEPTH 32
86
87#define KEYLENGTH (8*sizeof(t_key))
88
89typedef unsigned int t_key;
90
91#define T_TNODE 0
92#define T_LEAF 1
93#define NODE_TYPE_MASK 0x1UL
94#define NODE_TYPE(node) ((node)->parent & NODE_TYPE_MASK)
95
96#define IS_TNODE(n) (!(n->parent & T_LEAF))
97#define IS_LEAF(n) (n->parent & T_LEAF)
98
99struct rt_trie_node {
100 unsigned long parent;
101 t_key key;
102};
103
104struct leaf {
105 unsigned long parent;
106 t_key key;
107 struct hlist_head list;
108 struct rcu_head rcu;
109};
110
111struct leaf_info {
112 struct hlist_node hlist;
113 int plen;
114 u32 mask_plen; /* ntohl(inet_make_mask(plen)) */
115 struct list_head falh;
116 struct rcu_head rcu;
117};
118
119struct tnode {
120 unsigned long parent;
121 t_key key;
122 unsigned char pos; /* 2log(KEYLENGTH) bits needed */
123 unsigned char bits; /* 2log(KEYLENGTH) bits needed */
124 unsigned int full_children; /* KEYLENGTH bits needed */
125 unsigned int empty_children; /* KEYLENGTH bits needed */
126 union {
127 struct rcu_head rcu;
128 struct work_struct work;
129 struct tnode *tnode_free;
130 };
131 struct rt_trie_node __rcu *child[0];
132};
133
134#ifdef CONFIG_IP_FIB_TRIE_STATS
135struct trie_use_stats {
136 unsigned int gets;
137 unsigned int backtrack;
138 unsigned int semantic_match_passed;
139 unsigned int semantic_match_miss;
140 unsigned int null_node_hit;
141 unsigned int resize_node_skipped;
142};
143#endif
144
145struct trie_stat {
146 unsigned int totdepth;
147 unsigned int maxdepth;
148 unsigned int tnodes;
149 unsigned int leaves;
150 unsigned int nullpointers;
151 unsigned int prefixes;
152 unsigned int nodesizes[MAX_STAT_DEPTH];
153};
154
155struct trie {
156 struct rt_trie_node __rcu *trie;
157#ifdef CONFIG_IP_FIB_TRIE_STATS
158 struct trie_use_stats stats;
159#endif
160};
161
162static void put_child(struct trie *t, struct tnode *tn, int i, struct rt_trie_node *n);
163static void tnode_put_child_reorg(struct tnode *tn, int i, struct rt_trie_node *n,
164 int wasfull);
165static struct rt_trie_node *resize(struct trie *t, struct tnode *tn);
166static struct tnode *inflate(struct trie *t, struct tnode *tn);
167static struct tnode *halve(struct trie *t, struct tnode *tn);
168/* tnodes to free after resize(); protected by RTNL */
169static struct tnode *tnode_free_head;
170static size_t tnode_free_size;
171
172/*
173 * synchronize_rcu after call_rcu for that many pages; it should be especially
174 * useful before resizing the root node with PREEMPT_NONE configs; the value was
175 * obtained experimentally, aiming to avoid visible slowdown.
176 */
177static const int sync_pages = 128;
178
179static struct kmem_cache *fn_alias_kmem __read_mostly;
180static struct kmem_cache *trie_leaf_kmem __read_mostly;
181
182/*
183 * caller must hold RTNL
184 */
185static inline struct tnode *node_parent(const struct rt_trie_node *node)
186{
187 unsigned long parent;
188
189 parent = rcu_dereference_index_check(node->parent, lockdep_rtnl_is_held());
190
191 return (struct tnode *)(parent & ~NODE_TYPE_MASK);
192}
193
194/*
195 * caller must hold RCU read lock or RTNL
196 */
197static inline struct tnode *node_parent_rcu(const struct rt_trie_node *node)
198{
199 unsigned long parent;
200
201 parent = rcu_dereference_index_check(node->parent, rcu_read_lock_held() ||
202 lockdep_rtnl_is_held());
203
204 return (struct tnode *)(parent & ~NODE_TYPE_MASK);
205}
206
207/* Same as rcu_assign_pointer
208 * but that macro() assumes that value is a pointer.
209 */
210static inline void node_set_parent(struct rt_trie_node *node, struct tnode *ptr)
211{
212 smp_wmb();
213 node->parent = (unsigned long)ptr | NODE_TYPE(node);
214}
215
216/*
217 * caller must hold RTNL
218 */
219static inline struct rt_trie_node *tnode_get_child(const struct tnode *tn, unsigned int i)
220{
221 BUG_ON(i >= 1U << tn->bits);
222
223 return rtnl_dereference(tn->child[i]);
224}
225
226/*
227 * caller must hold RCU read lock or RTNL
228 */
229static inline struct rt_trie_node *tnode_get_child_rcu(const struct tnode *tn, unsigned int i)
230{
231 BUG_ON(i >= 1U << tn->bits);
232
233 return rcu_dereference_rtnl(tn->child[i]);
234}
235
236static inline int tnode_child_length(const struct tnode *tn)
237{
238 return 1 << tn->bits;
239}
240
241static inline t_key mask_pfx(t_key k, unsigned int l)
242{
243 return (l == 0) ? 0 : k >> (KEYLENGTH-l) << (KEYLENGTH-l);
244}
245
246static inline t_key tkey_extract_bits(t_key a, unsigned int offset, unsigned int bits)
247{
248 if (offset < KEYLENGTH)
249 return ((t_key)(a << offset)) >> (KEYLENGTH - bits);
250 else
251 return 0;
252}
253
254static inline int tkey_equals(t_key a, t_key b)
255{
256 return a == b;
257}
258
259static inline int tkey_sub_equals(t_key a, int offset, int bits, t_key b)
260{
261 if (bits == 0 || offset >= KEYLENGTH)
262 return 1;
263 bits = bits > KEYLENGTH ? KEYLENGTH : bits;
264 return ((a ^ b) << offset) >> (KEYLENGTH - bits) == 0;
265}
266
267static inline int tkey_mismatch(t_key a, int offset, t_key b)
268{
269 t_key diff = a ^ b;
270 int i = offset;
271
272 if (!diff)
273 return 0;
274 while ((diff << i) >> (KEYLENGTH-1) == 0)
275 i++;
276 return i;
277}
278
279/*
280 To understand this stuff, an understanding of keys and all their bits is
281 necessary. Every node in the trie has a key associated with it, but not
282 all of the bits in that key are significant.
283
284 Consider a node 'n' and its parent 'tp'.
285
286 If n is a leaf, every bit in its key is significant. Its presence is
287 necessitated by path compression, since during a tree traversal (when
288 searching for a leaf - unless we are doing an insertion) we will completely
289 ignore all skipped bits we encounter. Thus we need to verify, at the end of
290 a potentially successful search, that we have indeed been walking the
291 correct key path.
292
293 Note that we can never "miss" the correct key in the tree if present by
294 following the wrong path. Path compression ensures that segments of the key
295 that are the same for all keys with a given prefix are skipped, but the
296 skipped part *is* identical for each node in the subtrie below the skipped
297 bit! trie_insert() in this implementation takes care of that - note the
298 call to tkey_sub_equals() in trie_insert().
299
300 if n is an internal node - a 'tnode' here, the various parts of its key
301 have many different meanings.
302
303 Example:
304 _________________________________________________________________
305 | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
306 -----------------------------------------------------------------
307 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
308
309 _________________________________________________________________
310 | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
311 -----------------------------------------------------------------
312 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
313
314 tp->pos = 7
315 tp->bits = 3
316 n->pos = 15
317 n->bits = 4
318
319 First, let's just ignore the bits that come before the parent tp, that is
320 the bits from 0 to (tp->pos-1). They are *known* but at this point we do
321 not use them for anything.
322
323 The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
324 index into the parent's child array. That is, they will be used to find
325 'n' among tp's children.
326
327 The bits from (tp->pos + tp->bits) to (n->pos - 1) - "S" - are skipped bits
328 for the node n.
329
330 All the bits we have seen so far are significant to the node n. The rest
331 of the bits are really not needed or indeed known in n->key.
332
333 The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
334 n's child array, and will of course be different for each child.
335
336
337 The rest of the bits, from (n->pos + n->bits) onward, are completely unknown
338 at this point.
339
340*/
341
342static inline void check_tnode(const struct tnode *tn)
343{
344 WARN_ON(tn && tn->pos+tn->bits > 32);
345}
346
347static const int halve_threshold = 25;
348static const int inflate_threshold = 50;
349static const int halve_threshold_root = 15;
350static const int inflate_threshold_root = 30;
351
352static void __alias_free_mem(struct rcu_head *head)
353{
354 struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
355 kmem_cache_free(fn_alias_kmem, fa);
356}
357
358static inline void alias_free_mem_rcu(struct fib_alias *fa)
359{
360 call_rcu(&fa->rcu, __alias_free_mem);
361}
362
363static void __leaf_free_rcu(struct rcu_head *head)
364{
365 struct leaf *l = container_of(head, struct leaf, rcu);
366 kmem_cache_free(trie_leaf_kmem, l);
367}
368
369static inline void free_leaf(struct leaf *l)
370{
371 call_rcu_bh(&l->rcu, __leaf_free_rcu);
372}
373
374static inline void free_leaf_info(struct leaf_info *leaf)
375{
376 kfree_rcu(leaf, rcu);
377}
378
379static struct tnode *tnode_alloc(size_t size)
380{
381 if (size <= PAGE_SIZE)
382 return kzalloc(size, GFP_KERNEL);
383 else
384 return vzalloc(size);
385}
386
387static void __tnode_vfree(struct work_struct *arg)
388{
389 struct tnode *tn = container_of(arg, struct tnode, work);
390 vfree(tn);
391}
392
393static void __tnode_free_rcu(struct rcu_head *head)
394{
395 struct tnode *tn = container_of(head, struct tnode, rcu);
396 size_t size = sizeof(struct tnode) +
397 (sizeof(struct rt_trie_node *) << tn->bits);
398
399 if (size <= PAGE_SIZE)
400 kfree(tn);
401 else {
402 INIT_WORK(&tn->work, __tnode_vfree);
403 schedule_work(&tn->work);
404 }
405}
406
407static inline void tnode_free(struct tnode *tn)
408{
409 if (IS_LEAF(tn))
410 free_leaf((struct leaf *) tn);
411 else
412 call_rcu(&tn->rcu, __tnode_free_rcu);
413}
414
415static void tnode_free_safe(struct tnode *tn)
416{
417 BUG_ON(IS_LEAF(tn));
418 tn->tnode_free = tnode_free_head;
419 tnode_free_head = tn;
420 tnode_free_size += sizeof(struct tnode) +
421 (sizeof(struct rt_trie_node *) << tn->bits);
422}
423
424static void tnode_free_flush(void)
425{
426 struct tnode *tn;
427
428 while ((tn = tnode_free_head)) {
429 tnode_free_head = tn->tnode_free;
430 tn->tnode_free = NULL;
431 tnode_free(tn);
432 }
433
434 if (tnode_free_size >= PAGE_SIZE * sync_pages) {
435 tnode_free_size = 0;
436 synchronize_rcu();
437 }
438}
439
440static struct leaf *leaf_new(void)
441{
442 struct leaf *l = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
443 if (l) {
444 l->parent = T_LEAF;
445 INIT_HLIST_HEAD(&l->list);
446 }
447 return l;
448}
449
450static struct leaf_info *leaf_info_new(int plen)
451{
452 struct leaf_info *li = kmalloc(sizeof(struct leaf_info), GFP_KERNEL);
453 if (li) {
454 li->plen = plen;
455 li->mask_plen = ntohl(inet_make_mask(plen));
456 INIT_LIST_HEAD(&li->falh);
457 }
458 return li;
459}
460
461static struct tnode *tnode_new(t_key key, int pos, int bits)
462{
463 size_t sz = sizeof(struct tnode) + (sizeof(struct rt_trie_node *) << bits);
464 struct tnode *tn = tnode_alloc(sz);
465
466 if (tn) {
467 tn->parent = T_TNODE;
468 tn->pos = pos;
469 tn->bits = bits;
470 tn->key = key;
471 tn->full_children = 0;
472 tn->empty_children = 1<<bits;
473 }
474
475 pr_debug("AT %p s=%zu %zu\n", tn, sizeof(struct tnode),
476 sizeof(struct rt_trie_node) << bits);
477 return tn;
478}
479
480/*
481 * Check whether a tnode 'n' is "full", i.e. it is an internal node
482 * and no bits are skipped. See discussion in dyntree paper p. 6
483 */
484
485static inline int tnode_full(const struct tnode *tn, const struct rt_trie_node *n)
486{
487 if (n == NULL || IS_LEAF(n))
488 return 0;
489
490 return ((struct tnode *) n)->pos == tn->pos + tn->bits;
491}
492
493static inline void put_child(struct trie *t, struct tnode *tn, int i,
494 struct rt_trie_node *n)
495{
496 tnode_put_child_reorg(tn, i, n, -1);
497}
498
499 /*
500 * Add a child at position i overwriting the old value.
501 * Update the value of full_children and empty_children.
502 */
503
504static void tnode_put_child_reorg(struct tnode *tn, int i, struct rt_trie_node *n,
505 int wasfull)
506{
507 struct rt_trie_node *chi = rtnl_dereference(tn->child[i]);
508 int isfull;
509
510 BUG_ON(i >= 1<<tn->bits);
511
512 /* update emptyChildren */
513 if (n == NULL && chi != NULL)
514 tn->empty_children++;
515 else if (n != NULL && chi == NULL)
516 tn->empty_children--;
517
518 /* update fullChildren */
519 if (wasfull == -1)
520 wasfull = tnode_full(tn, chi);
521
522 isfull = tnode_full(tn, n);
523 if (wasfull && !isfull)
524 tn->full_children--;
525 else if (!wasfull && isfull)
526 tn->full_children++;
527
528 if (n)
529 node_set_parent(n, tn);
530
531 rcu_assign_pointer(tn->child[i], n);
532}
533
534#define MAX_WORK 10
535static struct rt_trie_node *resize(struct trie *t, struct tnode *tn)
536{
537 int i;
538 struct tnode *old_tn;
539 int inflate_threshold_use;
540 int halve_threshold_use;
541 int max_work;
542
543 if (!tn)
544 return NULL;
545
546 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
547 tn, inflate_threshold, halve_threshold);
548
549 /* No children */
550 if (tn->empty_children == tnode_child_length(tn)) {
551 tnode_free_safe(tn);
552 return NULL;
553 }
554 /* One child */
555 if (tn->empty_children == tnode_child_length(tn) - 1)
556 goto one_child;
557 /*
558 * Double as long as the resulting node has a number of
559 * nonempty nodes that are above the threshold.
560 */
561
562 /*
563 * From "Implementing a dynamic compressed trie" by Stefan Nilsson of
564 * the Helsinki University of Technology and Matti Tikkanen of Nokia
565 * Telecommunications, page 6:
566 * "A node is doubled if the ratio of non-empty children to all
567 * children in the *doubled* node is at least 'high'."
568 *
569 * 'high' in this instance is the variable 'inflate_threshold'. It
570 * is expressed as a percentage, so we multiply it with
571 * tnode_child_length() and instead of multiplying by 2 (since the
572 * child array will be doubled by inflate()) and multiplying
573 * the left-hand side by 100 (to handle the percentage thing) we
574 * multiply the left-hand side by 50.
575 *
576 * The left-hand side may look a bit weird: tnode_child_length(tn)
577 * - tn->empty_children is of course the number of non-null children
578 * in the current node. tn->full_children is the number of "full"
579 * children, that is non-null tnodes with a skip value of 0.
580 * All of those will be doubled in the resulting inflated tnode, so
581 * we just count them one extra time here.
582 *
583 * A clearer way to write this would be:
584 *
585 * to_be_doubled = tn->full_children;
586 * not_to_be_doubled = tnode_child_length(tn) - tn->empty_children -
587 * tn->full_children;
588 *
589 * new_child_length = tnode_child_length(tn) * 2;
590 *
591 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
592 * new_child_length;
593 * if (new_fill_factor >= inflate_threshold)
594 *
595 * ...and so on, tho it would mess up the while () loop.
596 *
597 * anyway,
598 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
599 * inflate_threshold
600 *
601 * avoid a division:
602 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
603 * inflate_threshold * new_child_length
604 *
605 * expand not_to_be_doubled and to_be_doubled, and shorten:
606 * 100 * (tnode_child_length(tn) - tn->empty_children +
607 * tn->full_children) >= inflate_threshold * new_child_length
608 *
609 * expand new_child_length:
610 * 100 * (tnode_child_length(tn) - tn->empty_children +
611 * tn->full_children) >=
612 * inflate_threshold * tnode_child_length(tn) * 2
613 *
614 * shorten again:
615 * 50 * (tn->full_children + tnode_child_length(tn) -
616 * tn->empty_children) >= inflate_threshold *
617 * tnode_child_length(tn)
618 *
619 */
620
621 check_tnode(tn);
622
623 /* Keep root node larger */
624
625 if (!node_parent((struct rt_trie_node *)tn)) {
626 inflate_threshold_use = inflate_threshold_root;
627 halve_threshold_use = halve_threshold_root;
628 } else {
629 inflate_threshold_use = inflate_threshold;
630 halve_threshold_use = halve_threshold;
631 }
632
633 max_work = MAX_WORK;
634 while ((tn->full_children > 0 && max_work-- &&
635 50 * (tn->full_children + tnode_child_length(tn)
636 - tn->empty_children)
637 >= inflate_threshold_use * tnode_child_length(tn))) {
638
639 old_tn = tn;
640 tn = inflate(t, tn);
641
642 if (IS_ERR(tn)) {
643 tn = old_tn;
644#ifdef CONFIG_IP_FIB_TRIE_STATS
645 t->stats.resize_node_skipped++;
646#endif
647 break;
648 }
649 }
650
651 check_tnode(tn);
652
653 /* Return if at least one inflate is run */
654 if (max_work != MAX_WORK)
655 return (struct rt_trie_node *) tn;
656
657 /*
658 * Halve as long as the number of empty children in this
659 * node is above threshold.
660 */
661
662 max_work = MAX_WORK;
663 while (tn->bits > 1 && max_work-- &&
664 100 * (tnode_child_length(tn) - tn->empty_children) <
665 halve_threshold_use * tnode_child_length(tn)) {
666
667 old_tn = tn;
668 tn = halve(t, tn);
669 if (IS_ERR(tn)) {
670 tn = old_tn;
671#ifdef CONFIG_IP_FIB_TRIE_STATS
672 t->stats.resize_node_skipped++;
673#endif
674 break;
675 }
676 }
677
678
679 /* Only one child remains */
680 if (tn->empty_children == tnode_child_length(tn) - 1) {
681one_child:
682 for (i = 0; i < tnode_child_length(tn); i++) {
683 struct rt_trie_node *n;
684
685 n = rtnl_dereference(tn->child[i]);
686 if (!n)
687 continue;
688
689 /* compress one level */
690
691 node_set_parent(n, NULL);
692 tnode_free_safe(tn);
693 return n;
694 }
695 }
696 return (struct rt_trie_node *) tn;
697}
698
699
700static void tnode_clean_free(struct tnode *tn)
701{
702 int i;
703 struct tnode *tofree;
704
705 for (i = 0; i < tnode_child_length(tn); i++) {
706 tofree = (struct tnode *)rtnl_dereference(tn->child[i]);
707 if (tofree)
708 tnode_free(tofree);
709 }
710 tnode_free(tn);
711}
712
713static struct tnode *inflate(struct trie *t, struct tnode *tn)
714{
715 struct tnode *oldtnode = tn;
716 int olen = tnode_child_length(tn);
717 int i;
718
719 pr_debug("In inflate\n");
720
721 tn = tnode_new(oldtnode->key, oldtnode->pos, oldtnode->bits + 1);
722
723 if (!tn)
724 return ERR_PTR(-ENOMEM);
725
726 /*
727 * Preallocate and store tnodes before the actual work so we
728 * don't get into an inconsistent state if memory allocation
729 * fails. In case of failure we return the oldnode and inflate
730 * of tnode is ignored.
731 */
732
733 for (i = 0; i < olen; i++) {
734 struct tnode *inode;
735
736 inode = (struct tnode *) tnode_get_child(oldtnode, i);
737 if (inode &&
738 IS_TNODE(inode) &&
739 inode->pos == oldtnode->pos + oldtnode->bits &&
740 inode->bits > 1) {
741 struct tnode *left, *right;
742 t_key m = ~0U << (KEYLENGTH - 1) >> inode->pos;
743
744 left = tnode_new(inode->key&(~m), inode->pos + 1,
745 inode->bits - 1);
746 if (!left)
747 goto nomem;
748
749 right = tnode_new(inode->key|m, inode->pos + 1,
750 inode->bits - 1);
751
752 if (!right) {
753 tnode_free(left);
754 goto nomem;
755 }
756
757 put_child(t, tn, 2*i, (struct rt_trie_node *) left);
758 put_child(t, tn, 2*i+1, (struct rt_trie_node *) right);
759 }
760 }
761
762 for (i = 0; i < olen; i++) {
763 struct tnode *inode;
764 struct rt_trie_node *node = tnode_get_child(oldtnode, i);
765 struct tnode *left, *right;
766 int size, j;
767
768 /* An empty child */
769 if (node == NULL)
770 continue;
771
772 /* A leaf or an internal node with skipped bits */
773
774 if (IS_LEAF(node) || ((struct tnode *) node)->pos >
775 tn->pos + tn->bits - 1) {
776 if (tkey_extract_bits(node->key,
777 oldtnode->pos + oldtnode->bits,
778 1) == 0)
779 put_child(t, tn, 2*i, node);
780 else
781 put_child(t, tn, 2*i+1, node);
782 continue;
783 }
784
785 /* An internal node with two children */
786 inode = (struct tnode *) node;
787
788 if (inode->bits == 1) {
789 put_child(t, tn, 2*i, rtnl_dereference(inode->child[0]));
790 put_child(t, tn, 2*i+1, rtnl_dereference(inode->child[1]));
791
792 tnode_free_safe(inode);
793 continue;
794 }
795
796 /* An internal node with more than two children */
797
798 /* We will replace this node 'inode' with two new
799 * ones, 'left' and 'right', each with half of the
800 * original children. The two new nodes will have
801 * a position one bit further down the key and this
802 * means that the "significant" part of their keys
803 * (see the discussion near the top of this file)
804 * will differ by one bit, which will be "0" in
805 * left's key and "1" in right's key. Since we are
806 * moving the key position by one step, the bit that
807 * we are moving away from - the bit at position
808 * (inode->pos) - is the one that will differ between
809 * left and right. So... we synthesize that bit in the
810 * two new keys.
811 * The mask 'm' below will be a single "one" bit at
812 * the position (inode->pos)
813 */
814
815 /* Use the old key, but set the new significant
816 * bit to zero.
817 */
818
819 left = (struct tnode *) tnode_get_child(tn, 2*i);
820 put_child(t, tn, 2*i, NULL);
821
822 BUG_ON(!left);
823
824 right = (struct tnode *) tnode_get_child(tn, 2*i+1);
825 put_child(t, tn, 2*i+1, NULL);
826
827 BUG_ON(!right);
828
829 size = tnode_child_length(left);
830 for (j = 0; j < size; j++) {
831 put_child(t, left, j, rtnl_dereference(inode->child[j]));
832 put_child(t, right, j, rtnl_dereference(inode->child[j + size]));
833 }
834 put_child(t, tn, 2*i, resize(t, left));
835 put_child(t, tn, 2*i+1, resize(t, right));
836
837 tnode_free_safe(inode);
838 }
839 tnode_free_safe(oldtnode);
840 return tn;
841nomem:
842 tnode_clean_free(tn);
843 return ERR_PTR(-ENOMEM);
844}
845
846static struct tnode *halve(struct trie *t, struct tnode *tn)
847{
848 struct tnode *oldtnode = tn;
849 struct rt_trie_node *left, *right;
850 int i;
851 int olen = tnode_child_length(tn);
852
853 pr_debug("In halve\n");
854
855 tn = tnode_new(oldtnode->key, oldtnode->pos, oldtnode->bits - 1);
856
857 if (!tn)
858 return ERR_PTR(-ENOMEM);
859
860 /*
861 * Preallocate and store tnodes before the actual work so we
862 * don't get into an inconsistent state if memory allocation
863 * fails. In case of failure we return the oldnode and halve
864 * of tnode is ignored.
865 */
866
867 for (i = 0; i < olen; i += 2) {
868 left = tnode_get_child(oldtnode, i);
869 right = tnode_get_child(oldtnode, i+1);
870
871 /* Two nonempty children */
872 if (left && right) {
873 struct tnode *newn;
874
875 newn = tnode_new(left->key, tn->pos + tn->bits, 1);
876
877 if (!newn)
878 goto nomem;
879
880 put_child(t, tn, i/2, (struct rt_trie_node *)newn);
881 }
882
883 }
884
885 for (i = 0; i < olen; i += 2) {
886 struct tnode *newBinNode;
887
888 left = tnode_get_child(oldtnode, i);
889 right = tnode_get_child(oldtnode, i+1);
890
891 /* At least one of the children is empty */
892 if (left == NULL) {
893 if (right == NULL) /* Both are empty */
894 continue;
895 put_child(t, tn, i/2, right);
896 continue;
897 }
898
899 if (right == NULL) {
900 put_child(t, tn, i/2, left);
901 continue;
902 }
903
904 /* Two nonempty children */
905 newBinNode = (struct tnode *) tnode_get_child(tn, i/2);
906 put_child(t, tn, i/2, NULL);
907 put_child(t, newBinNode, 0, left);
908 put_child(t, newBinNode, 1, right);
909 put_child(t, tn, i/2, resize(t, newBinNode));
910 }
911 tnode_free_safe(oldtnode);
912 return tn;
913nomem:
914 tnode_clean_free(tn);
915 return ERR_PTR(-ENOMEM);
916}
917
918/* readside must use rcu_read_lock currently dump routines
919 via get_fa_head and dump */
920
921static struct leaf_info *find_leaf_info(struct leaf *l, int plen)
922{
923 struct hlist_head *head = &l->list;
924 struct hlist_node *node;
925 struct leaf_info *li;
926
927 hlist_for_each_entry_rcu(li, node, head, hlist)
928 if (li->plen == plen)
929 return li;
930
931 return NULL;
932}
933
934static inline struct list_head *get_fa_head(struct leaf *l, int plen)
935{
936 struct leaf_info *li = find_leaf_info(l, plen);
937
938 if (!li)
939 return NULL;
940
941 return &li->falh;
942}
943
944static void insert_leaf_info(struct hlist_head *head, struct leaf_info *new)
945{
946 struct leaf_info *li = NULL, *last = NULL;
947 struct hlist_node *node;
948
949 if (hlist_empty(head)) {
950 hlist_add_head_rcu(&new->hlist, head);
951 } else {
952 hlist_for_each_entry(li, node, head, hlist) {
953 if (new->plen > li->plen)
954 break;
955
956 last = li;
957 }
958 if (last)
959 hlist_add_after_rcu(&last->hlist, &new->hlist);
960 else
961 hlist_add_before_rcu(&new->hlist, &li->hlist);
962 }
963}
964
965/* rcu_read_lock needs to be hold by caller from readside */
966
967static struct leaf *
968fib_find_node(struct trie *t, u32 key)
969{
970 int pos;
971 struct tnode *tn;
972 struct rt_trie_node *n;
973
974 pos = 0;
975 n = rcu_dereference_rtnl(t->trie);
976
977 while (n != NULL && NODE_TYPE(n) == T_TNODE) {
978 tn = (struct tnode *) n;
979
980 check_tnode(tn);
981
982 if (tkey_sub_equals(tn->key, pos, tn->pos-pos, key)) {
983 pos = tn->pos + tn->bits;
984 n = tnode_get_child_rcu(tn,
985 tkey_extract_bits(key,
986 tn->pos,
987 tn->bits));
988 } else
989 break;
990 }
991 /* Case we have found a leaf. Compare prefixes */
992
993 if (n != NULL && IS_LEAF(n) && tkey_equals(key, n->key))
994 return (struct leaf *)n;
995
996 return NULL;
997}
998
999static void trie_rebalance(struct trie *t, struct tnode *tn)
1000{
1001 int wasfull;
1002 t_key cindex, key;
1003 struct tnode *tp;
1004
1005 key = tn->key;
1006
1007 while (tn != NULL && (tp = node_parent((struct rt_trie_node *)tn)) != NULL) {
1008 cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1009 wasfull = tnode_full(tp, tnode_get_child(tp, cindex));
1010 tn = (struct tnode *) resize(t, (struct tnode *)tn);
1011
1012 tnode_put_child_reorg((struct tnode *)tp, cindex,
1013 (struct rt_trie_node *)tn, wasfull);
1014
1015 tp = node_parent((struct rt_trie_node *) tn);
1016 if (!tp)
1017 rcu_assign_pointer(t->trie, (struct rt_trie_node *)tn);
1018
1019 tnode_free_flush();
1020 if (!tp)
1021 break;
1022 tn = tp;
1023 }
1024
1025 /* Handle last (top) tnode */
1026 if (IS_TNODE(tn))
1027 tn = (struct tnode *)resize(t, (struct tnode *)tn);
1028
1029 rcu_assign_pointer(t->trie, (struct rt_trie_node *)tn);
1030 tnode_free_flush();
1031}
1032
1033/* only used from updater-side */
1034
1035static struct list_head *fib_insert_node(struct trie *t, u32 key, int plen)
1036{
1037 int pos, newpos;
1038 struct tnode *tp = NULL, *tn = NULL;
1039 struct rt_trie_node *n;
1040 struct leaf *l;
1041 int missbit;
1042 struct list_head *fa_head = NULL;
1043 struct leaf_info *li;
1044 t_key cindex;
1045
1046 pos = 0;
1047 n = rtnl_dereference(t->trie);
1048
1049 /* If we point to NULL, stop. Either the tree is empty and we should
1050 * just put a new leaf in if, or we have reached an empty child slot,
1051 * and we should just put our new leaf in that.
1052 * If we point to a T_TNODE, check if it matches our key. Note that
1053 * a T_TNODE might be skipping any number of bits - its 'pos' need
1054 * not be the parent's 'pos'+'bits'!
1055 *
1056 * If it does match the current key, get pos/bits from it, extract
1057 * the index from our key, push the T_TNODE and walk the tree.
1058 *
1059 * If it doesn't, we have to replace it with a new T_TNODE.
1060 *
1061 * If we point to a T_LEAF, it might or might not have the same key
1062 * as we do. If it does, just change the value, update the T_LEAF's
1063 * value, and return it.
1064 * If it doesn't, we need to replace it with a T_TNODE.
1065 */
1066
1067 while (n != NULL && NODE_TYPE(n) == T_TNODE) {
1068 tn = (struct tnode *) n;
1069
1070 check_tnode(tn);
1071
1072 if (tkey_sub_equals(tn->key, pos, tn->pos-pos, key)) {
1073 tp = tn;
1074 pos = tn->pos + tn->bits;
1075 n = tnode_get_child(tn,
1076 tkey_extract_bits(key,
1077 tn->pos,
1078 tn->bits));
1079
1080 BUG_ON(n && node_parent(n) != tn);
1081 } else
1082 break;
1083 }
1084
1085 /*
1086 * n ----> NULL, LEAF or TNODE
1087 *
1088 * tp is n's (parent) ----> NULL or TNODE
1089 */
1090
1091 BUG_ON(tp && IS_LEAF(tp));
1092
1093 /* Case 1: n is a leaf. Compare prefixes */
1094
1095 if (n != NULL && IS_LEAF(n) && tkey_equals(key, n->key)) {
1096 l = (struct leaf *) n;
1097 li = leaf_info_new(plen);
1098
1099 if (!li)
1100 return NULL;
1101
1102 fa_head = &li->falh;
1103 insert_leaf_info(&l->list, li);
1104 goto done;
1105 }
1106 l = leaf_new();
1107
1108 if (!l)
1109 return NULL;
1110
1111 l->key = key;
1112 li = leaf_info_new(plen);
1113
1114 if (!li) {
1115 free_leaf(l);
1116 return NULL;
1117 }
1118
1119 fa_head = &li->falh;
1120 insert_leaf_info(&l->list, li);
1121
1122 if (t->trie && n == NULL) {
1123 /* Case 2: n is NULL, and will just insert a new leaf */
1124
1125 node_set_parent((struct rt_trie_node *)l, tp);
1126
1127 cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1128 put_child(t, (struct tnode *)tp, cindex, (struct rt_trie_node *)l);
1129 } else {
1130 /* Case 3: n is a LEAF or a TNODE and the key doesn't match. */
1131 /*
1132 * Add a new tnode here
1133 * first tnode need some special handling
1134 */
1135
1136 if (tp)
1137 pos = tp->pos+tp->bits;
1138 else
1139 pos = 0;
1140
1141 if (n) {
1142 newpos = tkey_mismatch(key, pos, n->key);
1143 tn = tnode_new(n->key, newpos, 1);
1144 } else {
1145 newpos = 0;
1146 tn = tnode_new(key, newpos, 1); /* First tnode */
1147 }
1148
1149 if (!tn) {
1150 free_leaf_info(li);
1151 free_leaf(l);
1152 return NULL;
1153 }
1154
1155 node_set_parent((struct rt_trie_node *)tn, tp);
1156
1157 missbit = tkey_extract_bits(key, newpos, 1);
1158 put_child(t, tn, missbit, (struct rt_trie_node *)l);
1159 put_child(t, tn, 1-missbit, n);
1160
1161 if (tp) {
1162 cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1163 put_child(t, (struct tnode *)tp, cindex,
1164 (struct rt_trie_node *)tn);
1165 } else {
1166 rcu_assign_pointer(t->trie, (struct rt_trie_node *)tn);
1167 tp = tn;
1168 }
1169 }
1170
1171 if (tp && tp->pos + tp->bits > 32)
1172 pr_warn("fib_trie tp=%p pos=%d, bits=%d, key=%0x plen=%d\n",
1173 tp, tp->pos, tp->bits, key, plen);
1174
1175 /* Rebalance the trie */
1176
1177 trie_rebalance(t, tp);
1178done:
1179 return fa_head;
1180}
1181
1182/*
1183 * Caller must hold RTNL.
1184 */
1185int fib_table_insert(struct fib_table *tb, struct fib_config *cfg)
1186{
1187 struct trie *t = (struct trie *) tb->tb_data;
1188 struct fib_alias *fa, *new_fa;
1189 struct list_head *fa_head = NULL;
1190 struct fib_info *fi;
1191 int plen = cfg->fc_dst_len;
1192 u8 tos = cfg->fc_tos;
1193 u32 key, mask;
1194 int err;
1195 struct leaf *l;
1196
1197 if (plen > 32)
1198 return -EINVAL;
1199
1200 key = ntohl(cfg->fc_dst);
1201
1202 pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1203
1204 mask = ntohl(inet_make_mask(plen));
1205
1206 if (key & ~mask)
1207 return -EINVAL;
1208
1209 key = key & mask;
1210
1211 fi = fib_create_info(cfg);
1212 if (IS_ERR(fi)) {
1213 err = PTR_ERR(fi);
1214 goto err;
1215 }
1216
1217 l = fib_find_node(t, key);
1218 fa = NULL;
1219
1220 if (l) {
1221 fa_head = get_fa_head(l, plen);
1222 fa = fib_find_alias(fa_head, tos, fi->fib_priority);
1223 }
1224
1225 /* Now fa, if non-NULL, points to the first fib alias
1226 * with the same keys [prefix,tos,priority], if such key already
1227 * exists or to the node before which we will insert new one.
1228 *
1229 * If fa is NULL, we will need to allocate a new one and
1230 * insert to the head of f.
1231 *
1232 * If f is NULL, no fib node matched the destination key
1233 * and we need to allocate a new one of those as well.
1234 */
1235
1236 if (fa && fa->fa_tos == tos &&
1237 fa->fa_info->fib_priority == fi->fib_priority) {
1238 struct fib_alias *fa_first, *fa_match;
1239
1240 err = -EEXIST;
1241 if (cfg->fc_nlflags & NLM_F_EXCL)
1242 goto out;
1243
1244 /* We have 2 goals:
1245 * 1. Find exact match for type, scope, fib_info to avoid
1246 * duplicate routes
1247 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1248 */
1249 fa_match = NULL;
1250 fa_first = fa;
1251 fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
1252 list_for_each_entry_continue(fa, fa_head, fa_list) {
1253 if (fa->fa_tos != tos)
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 fa = fa_first;
1269 if (fa_match) {
1270 if (fa == fa_match)
1271 err = 0;
1272 goto out;
1273 }
1274 err = -ENOBUFS;
1275 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1276 if (new_fa == NULL)
1277 goto out;
1278
1279 fi_drop = fa->fa_info;
1280 new_fa->fa_tos = fa->fa_tos;
1281 new_fa->fa_info = fi;
1282 new_fa->fa_type = cfg->fc_type;
1283 state = fa->fa_state;
1284 new_fa->fa_state = state & ~FA_S_ACCESSED;
1285
1286 list_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1287 alias_free_mem_rcu(fa);
1288
1289 fib_release_info(fi_drop);
1290 if (state & FA_S_ACCESSED)
1291 rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1292 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1293 tb->tb_id, &cfg->fc_nlinfo, NLM_F_REPLACE);
1294
1295 goto succeeded;
1296 }
1297 /* Error if we find a perfect match which
1298 * uses the same scope, type, and nexthop
1299 * information.
1300 */
1301 if (fa_match)
1302 goto out;
1303
1304 if (!(cfg->fc_nlflags & NLM_F_APPEND))
1305 fa = fa_first;
1306 }
1307 err = -ENOENT;
1308 if (!(cfg->fc_nlflags & NLM_F_CREATE))
1309 goto out;
1310
1311 err = -ENOBUFS;
1312 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1313 if (new_fa == NULL)
1314 goto out;
1315
1316 new_fa->fa_info = fi;
1317 new_fa->fa_tos = tos;
1318 new_fa->fa_type = cfg->fc_type;
1319 new_fa->fa_state = 0;
1320 /*
1321 * Insert new entry to the list.
1322 */
1323
1324 if (!fa_head) {
1325 fa_head = fib_insert_node(t, key, plen);
1326 if (unlikely(!fa_head)) {
1327 err = -ENOMEM;
1328 goto out_free_new_fa;
1329 }
1330 }
1331
1332 if (!plen)
1333 tb->tb_num_default++;
1334
1335 list_add_tail_rcu(&new_fa->fa_list,
1336 (fa ? &fa->fa_list : fa_head));
1337
1338 rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1339 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, tb->tb_id,
1340 &cfg->fc_nlinfo, 0);
1341succeeded:
1342 return 0;
1343
1344out_free_new_fa:
1345 kmem_cache_free(fn_alias_kmem, new_fa);
1346out:
1347 fib_release_info(fi);
1348err:
1349 return err;
1350}
1351
1352/* should be called with rcu_read_lock */
1353static int check_leaf(struct fib_table *tb, struct trie *t, struct leaf *l,
1354 t_key key, const struct flowi4 *flp,
1355 struct fib_result *res, int fib_flags)
1356{
1357 struct leaf_info *li;
1358 struct hlist_head *hhead = &l->list;
1359 struct hlist_node *node;
1360
1361 hlist_for_each_entry_rcu(li, node, hhead, hlist) {
1362 struct fib_alias *fa;
1363
1364 if (l->key != (key & li->mask_plen))
1365 continue;
1366
1367 list_for_each_entry_rcu(fa, &li->falh, fa_list) {
1368 struct fib_info *fi = fa->fa_info;
1369 int nhsel, err;
1370
1371 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1372 continue;
1373 if (fi->fib_dead)
1374 continue;
1375 if (fa->fa_info->fib_scope < flp->flowi4_scope)
1376 continue;
1377 fib_alias_accessed(fa);
1378 err = fib_props[fa->fa_type].error;
1379 if (err) {
1380#ifdef CONFIG_IP_FIB_TRIE_STATS
1381 t->stats.semantic_match_passed++;
1382#endif
1383 return err;
1384 }
1385 if (fi->fib_flags & RTNH_F_DEAD)
1386 continue;
1387 for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
1388 const struct fib_nh *nh = &fi->fib_nh[nhsel];
1389
1390 if (nh->nh_flags & RTNH_F_DEAD)
1391 continue;
1392 if (flp->flowi4_oif && flp->flowi4_oif != nh->nh_oif)
1393 continue;
1394
1395#ifdef CONFIG_IP_FIB_TRIE_STATS
1396 t->stats.semantic_match_passed++;
1397#endif
1398 res->prefixlen = li->plen;
1399 res->nh_sel = nhsel;
1400 res->type = fa->fa_type;
1401 res->scope = fa->fa_info->fib_scope;
1402 res->fi = fi;
1403 res->table = tb;
1404 res->fa_head = &li->falh;
1405 if (!(fib_flags & FIB_LOOKUP_NOREF))
1406 atomic_inc(&fi->fib_clntref);
1407 return 0;
1408 }
1409 }
1410
1411#ifdef CONFIG_IP_FIB_TRIE_STATS
1412 t->stats.semantic_match_miss++;
1413#endif
1414 }
1415
1416 return 1;
1417}
1418
1419int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1420 struct fib_result *res, int fib_flags)
1421{
1422 struct trie *t = (struct trie *) tb->tb_data;
1423 int ret;
1424 struct rt_trie_node *n;
1425 struct tnode *pn;
1426 unsigned int pos, bits;
1427 t_key key = ntohl(flp->daddr);
1428 unsigned int chopped_off;
1429 t_key cindex = 0;
1430 unsigned int current_prefix_length = KEYLENGTH;
1431 struct tnode *cn;
1432 t_key pref_mismatch;
1433
1434 rcu_read_lock();
1435
1436 n = rcu_dereference(t->trie);
1437 if (!n)
1438 goto failed;
1439
1440#ifdef CONFIG_IP_FIB_TRIE_STATS
1441 t->stats.gets++;
1442#endif
1443
1444 /* Just a leaf? */
1445 if (IS_LEAF(n)) {
1446 ret = check_leaf(tb, t, (struct leaf *)n, key, flp, res, fib_flags);
1447 goto found;
1448 }
1449
1450 pn = (struct tnode *) n;
1451 chopped_off = 0;
1452
1453 while (pn) {
1454 pos = pn->pos;
1455 bits = pn->bits;
1456
1457 if (!chopped_off)
1458 cindex = tkey_extract_bits(mask_pfx(key, current_prefix_length),
1459 pos, bits);
1460
1461 n = tnode_get_child_rcu(pn, cindex);
1462
1463 if (n == NULL) {
1464#ifdef CONFIG_IP_FIB_TRIE_STATS
1465 t->stats.null_node_hit++;
1466#endif
1467 goto backtrace;
1468 }
1469
1470 if (IS_LEAF(n)) {
1471 ret = check_leaf(tb, t, (struct leaf *)n, key, flp, res, fib_flags);
1472 if (ret > 0)
1473 goto backtrace;
1474 goto found;
1475 }
1476
1477 cn = (struct tnode *)n;
1478
1479 /*
1480 * It's a tnode, and we can do some extra checks here if we
1481 * like, to avoid descending into a dead-end branch.
1482 * This tnode is in the parent's child array at index
1483 * key[p_pos..p_pos+p_bits] but potentially with some bits
1484 * chopped off, so in reality the index may be just a
1485 * subprefix, padded with zero at the end.
1486 * We can also take a look at any skipped bits in this
1487 * tnode - everything up to p_pos is supposed to be ok,
1488 * and the non-chopped bits of the index (se previous
1489 * paragraph) are also guaranteed ok, but the rest is
1490 * considered unknown.
1491 *
1492 * The skipped bits are key[pos+bits..cn->pos].
1493 */
1494
1495 /* If current_prefix_length < pos+bits, we are already doing
1496 * actual prefix matching, which means everything from
1497 * pos+(bits-chopped_off) onward must be zero along some
1498 * branch of this subtree - otherwise there is *no* valid
1499 * prefix present. Here we can only check the skipped
1500 * bits. Remember, since we have already indexed into the
1501 * parent's child array, we know that the bits we chopped of
1502 * *are* zero.
1503 */
1504
1505 /* NOTA BENE: Checking only skipped bits
1506 for the new node here */
1507
1508 if (current_prefix_length < pos+bits) {
1509 if (tkey_extract_bits(cn->key, current_prefix_length,
1510 cn->pos - current_prefix_length)
1511 || !(cn->child[0]))
1512 goto backtrace;
1513 }
1514
1515 /*
1516 * If chopped_off=0, the index is fully validated and we
1517 * only need to look at the skipped bits for this, the new,
1518 * tnode. What we actually want to do is to find out if
1519 * these skipped bits match our key perfectly, or if we will
1520 * have to count on finding a matching prefix further down,
1521 * because if we do, we would like to have some way of
1522 * verifying the existence of such a prefix at this point.
1523 */
1524
1525 /* The only thing we can do at this point is to verify that
1526 * any such matching prefix can indeed be a prefix to our
1527 * key, and if the bits in the node we are inspecting that
1528 * do not match our key are not ZERO, this cannot be true.
1529 * Thus, find out where there is a mismatch (before cn->pos)
1530 * and verify that all the mismatching bits are zero in the
1531 * new tnode's key.
1532 */
1533
1534 /*
1535 * Note: We aren't very concerned about the piece of
1536 * the key that precede pn->pos+pn->bits, since these
1537 * have already been checked. The bits after cn->pos
1538 * aren't checked since these are by definition
1539 * "unknown" at this point. Thus, what we want to see
1540 * is if we are about to enter the "prefix matching"
1541 * state, and in that case verify that the skipped
1542 * bits that will prevail throughout this subtree are
1543 * zero, as they have to be if we are to find a
1544 * matching prefix.
1545 */
1546
1547 pref_mismatch = mask_pfx(cn->key ^ key, cn->pos);
1548
1549 /*
1550 * In short: If skipped bits in this node do not match
1551 * the search key, enter the "prefix matching"
1552 * state.directly.
1553 */
1554 if (pref_mismatch) {
1555 int mp = KEYLENGTH - fls(pref_mismatch);
1556
1557 if (tkey_extract_bits(cn->key, mp, cn->pos - mp) != 0)
1558 goto backtrace;
1559
1560 if (current_prefix_length >= cn->pos)
1561 current_prefix_length = mp;
1562 }
1563
1564 pn = (struct tnode *)n; /* Descend */
1565 chopped_off = 0;
1566 continue;
1567
1568backtrace:
1569 chopped_off++;
1570
1571 /* As zero don't change the child key (cindex) */
1572 while ((chopped_off <= pn->bits)
1573 && !(cindex & (1<<(chopped_off-1))))
1574 chopped_off++;
1575
1576 /* Decrease current_... with bits chopped off */
1577 if (current_prefix_length > pn->pos + pn->bits - chopped_off)
1578 current_prefix_length = pn->pos + pn->bits
1579 - chopped_off;
1580
1581 /*
1582 * Either we do the actual chop off according or if we have
1583 * chopped off all bits in this tnode walk up to our parent.
1584 */
1585
1586 if (chopped_off <= pn->bits) {
1587 cindex &= ~(1 << (chopped_off-1));
1588 } else {
1589 struct tnode *parent = node_parent_rcu((struct rt_trie_node *) pn);
1590 if (!parent)
1591 goto failed;
1592
1593 /* Get Child's index */
1594 cindex = tkey_extract_bits(pn->key, parent->pos, parent->bits);
1595 pn = parent;
1596 chopped_off = 0;
1597
1598#ifdef CONFIG_IP_FIB_TRIE_STATS
1599 t->stats.backtrack++;
1600#endif
1601 goto backtrace;
1602 }
1603 }
1604failed:
1605 ret = 1;
1606found:
1607 rcu_read_unlock();
1608 return ret;
1609}
1610EXPORT_SYMBOL_GPL(fib_table_lookup);
1611
1612/*
1613 * Remove the leaf and return parent.
1614 */
1615static void trie_leaf_remove(struct trie *t, struct leaf *l)
1616{
1617 struct tnode *tp = node_parent((struct rt_trie_node *) l);
1618
1619 pr_debug("entering trie_leaf_remove(%p)\n", l);
1620
1621 if (tp) {
1622 t_key cindex = tkey_extract_bits(l->key, tp->pos, tp->bits);
1623 put_child(t, (struct tnode *)tp, cindex, NULL);
1624 trie_rebalance(t, tp);
1625 } else
1626 RCU_INIT_POINTER(t->trie, NULL);
1627
1628 free_leaf(l);
1629}
1630
1631/*
1632 * Caller must hold RTNL.
1633 */
1634int fib_table_delete(struct fib_table *tb, struct fib_config *cfg)
1635{
1636 struct trie *t = (struct trie *) tb->tb_data;
1637 u32 key, mask;
1638 int plen = cfg->fc_dst_len;
1639 u8 tos = cfg->fc_tos;
1640 struct fib_alias *fa, *fa_to_delete;
1641 struct list_head *fa_head;
1642 struct leaf *l;
1643 struct leaf_info *li;
1644
1645 if (plen > 32)
1646 return -EINVAL;
1647
1648 key = ntohl(cfg->fc_dst);
1649 mask = ntohl(inet_make_mask(plen));
1650
1651 if (key & ~mask)
1652 return -EINVAL;
1653
1654 key = key & mask;
1655 l = fib_find_node(t, key);
1656
1657 if (!l)
1658 return -ESRCH;
1659
1660 fa_head = get_fa_head(l, plen);
1661 fa = fib_find_alias(fa_head, tos, 0);
1662
1663 if (!fa)
1664 return -ESRCH;
1665
1666 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1667
1668 fa_to_delete = NULL;
1669 fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
1670 list_for_each_entry_continue(fa, fa_head, fa_list) {
1671 struct fib_info *fi = fa->fa_info;
1672
1673 if (fa->fa_tos != tos)
1674 break;
1675
1676 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1677 (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1678 fa->fa_info->fib_scope == cfg->fc_scope) &&
1679 (!cfg->fc_prefsrc ||
1680 fi->fib_prefsrc == cfg->fc_prefsrc) &&
1681 (!cfg->fc_protocol ||
1682 fi->fib_protocol == cfg->fc_protocol) &&
1683 fib_nh_match(cfg, fi) == 0) {
1684 fa_to_delete = fa;
1685 break;
1686 }
1687 }
1688
1689 if (!fa_to_delete)
1690 return -ESRCH;
1691
1692 fa = fa_to_delete;
1693 rtmsg_fib(RTM_DELROUTE, htonl(key), fa, plen, tb->tb_id,
1694 &cfg->fc_nlinfo, 0);
1695
1696 l = fib_find_node(t, key);
1697 li = find_leaf_info(l, plen);
1698
1699 list_del_rcu(&fa->fa_list);
1700
1701 if (!plen)
1702 tb->tb_num_default--;
1703
1704 if (list_empty(fa_head)) {
1705 hlist_del_rcu(&li->hlist);
1706 free_leaf_info(li);
1707 }
1708
1709 if (hlist_empty(&l->list))
1710 trie_leaf_remove(t, l);
1711
1712 if (fa->fa_state & FA_S_ACCESSED)
1713 rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1714
1715 fib_release_info(fa->fa_info);
1716 alias_free_mem_rcu(fa);
1717 return 0;
1718}
1719
1720static int trie_flush_list(struct list_head *head)
1721{
1722 struct fib_alias *fa, *fa_node;
1723 int found = 0;
1724
1725 list_for_each_entry_safe(fa, fa_node, head, fa_list) {
1726 struct fib_info *fi = fa->fa_info;
1727
1728 if (fi && (fi->fib_flags & RTNH_F_DEAD)) {
1729 list_del_rcu(&fa->fa_list);
1730 fib_release_info(fa->fa_info);
1731 alias_free_mem_rcu(fa);
1732 found++;
1733 }
1734 }
1735 return found;
1736}
1737
1738static int trie_flush_leaf(struct leaf *l)
1739{
1740 int found = 0;
1741 struct hlist_head *lih = &l->list;
1742 struct hlist_node *node, *tmp;
1743 struct leaf_info *li = NULL;
1744
1745 hlist_for_each_entry_safe(li, node, tmp, lih, hlist) {
1746 found += trie_flush_list(&li->falh);
1747
1748 if (list_empty(&li->falh)) {
1749 hlist_del_rcu(&li->hlist);
1750 free_leaf_info(li);
1751 }
1752 }
1753 return found;
1754}
1755
1756/*
1757 * Scan for the next right leaf starting at node p->child[idx]
1758 * Since we have back pointer, no recursion necessary.
1759 */
1760static struct leaf *leaf_walk_rcu(struct tnode *p, struct rt_trie_node *c)
1761{
1762 do {
1763 t_key idx;
1764
1765 if (c)
1766 idx = tkey_extract_bits(c->key, p->pos, p->bits) + 1;
1767 else
1768 idx = 0;
1769
1770 while (idx < 1u << p->bits) {
1771 c = tnode_get_child_rcu(p, idx++);
1772 if (!c)
1773 continue;
1774
1775 if (IS_LEAF(c)) {
1776 prefetch(rcu_dereference_rtnl(p->child[idx]));
1777 return (struct leaf *) c;
1778 }
1779
1780 /* Rescan start scanning in new node */
1781 p = (struct tnode *) c;
1782 idx = 0;
1783 }
1784
1785 /* Node empty, walk back up to parent */
1786 c = (struct rt_trie_node *) p;
1787 } while ((p = node_parent_rcu(c)) != NULL);
1788
1789 return NULL; /* Root of trie */
1790}
1791
1792static struct leaf *trie_firstleaf(struct trie *t)
1793{
1794 struct tnode *n = (struct tnode *)rcu_dereference_rtnl(t->trie);
1795
1796 if (!n)
1797 return NULL;
1798
1799 if (IS_LEAF(n)) /* trie is just a leaf */
1800 return (struct leaf *) n;
1801
1802 return leaf_walk_rcu(n, NULL);
1803}
1804
1805static struct leaf *trie_nextleaf(struct leaf *l)
1806{
1807 struct rt_trie_node *c = (struct rt_trie_node *) l;
1808 struct tnode *p = node_parent_rcu(c);
1809
1810 if (!p)
1811 return NULL; /* trie with just one leaf */
1812
1813 return leaf_walk_rcu(p, c);
1814}
1815
1816static struct leaf *trie_leafindex(struct trie *t, int index)
1817{
1818 struct leaf *l = trie_firstleaf(t);
1819
1820 while (l && index-- > 0)
1821 l = trie_nextleaf(l);
1822
1823 return l;
1824}
1825
1826
1827/*
1828 * Caller must hold RTNL.
1829 */
1830int fib_table_flush(struct fib_table *tb)
1831{
1832 struct trie *t = (struct trie *) tb->tb_data;
1833 struct leaf *l, *ll = NULL;
1834 int found = 0;
1835
1836 for (l = trie_firstleaf(t); l; l = trie_nextleaf(l)) {
1837 found += trie_flush_leaf(l);
1838
1839 if (ll && hlist_empty(&ll->list))
1840 trie_leaf_remove(t, ll);
1841 ll = l;
1842 }
1843
1844 if (ll && hlist_empty(&ll->list))
1845 trie_leaf_remove(t, ll);
1846
1847 pr_debug("trie_flush found=%d\n", found);
1848 return found;
1849}
1850
1851void fib_free_table(struct fib_table *tb)
1852{
1853 kfree(tb);
1854}
1855
1856static int fn_trie_dump_fa(t_key key, int plen, struct list_head *fah,
1857 struct fib_table *tb,
1858 struct sk_buff *skb, struct netlink_callback *cb)
1859{
1860 int i, s_i;
1861 struct fib_alias *fa;
1862 __be32 xkey = htonl(key);
1863
1864 s_i = cb->args[5];
1865 i = 0;
1866
1867 /* rcu_read_lock is hold by caller */
1868
1869 list_for_each_entry_rcu(fa, fah, fa_list) {
1870 if (i < s_i) {
1871 i++;
1872 continue;
1873 }
1874
1875 if (fib_dump_info(skb, NETLINK_CB(cb->skb).pid,
1876 cb->nlh->nlmsg_seq,
1877 RTM_NEWROUTE,
1878 tb->tb_id,
1879 fa->fa_type,
1880 xkey,
1881 plen,
1882 fa->fa_tos,
1883 fa->fa_info, NLM_F_MULTI) < 0) {
1884 cb->args[5] = i;
1885 return -1;
1886 }
1887 i++;
1888 }
1889 cb->args[5] = i;
1890 return skb->len;
1891}
1892
1893static int fn_trie_dump_leaf(struct leaf *l, struct fib_table *tb,
1894 struct sk_buff *skb, struct netlink_callback *cb)
1895{
1896 struct leaf_info *li;
1897 struct hlist_node *node;
1898 int i, s_i;
1899
1900 s_i = cb->args[4];
1901 i = 0;
1902
1903 /* rcu_read_lock is hold by caller */
1904 hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
1905 if (i < s_i) {
1906 i++;
1907 continue;
1908 }
1909
1910 if (i > s_i)
1911 cb->args[5] = 0;
1912
1913 if (list_empty(&li->falh))
1914 continue;
1915
1916 if (fn_trie_dump_fa(l->key, li->plen, &li->falh, tb, skb, cb) < 0) {
1917 cb->args[4] = i;
1918 return -1;
1919 }
1920 i++;
1921 }
1922
1923 cb->args[4] = i;
1924 return skb->len;
1925}
1926
1927int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
1928 struct netlink_callback *cb)
1929{
1930 struct leaf *l;
1931 struct trie *t = (struct trie *) tb->tb_data;
1932 t_key key = cb->args[2];
1933 int count = cb->args[3];
1934
1935 rcu_read_lock();
1936 /* Dump starting at last key.
1937 * Note: 0.0.0.0/0 (ie default) is first key.
1938 */
1939 if (count == 0)
1940 l = trie_firstleaf(t);
1941 else {
1942 /* Normally, continue from last key, but if that is missing
1943 * fallback to using slow rescan
1944 */
1945 l = fib_find_node(t, key);
1946 if (!l)
1947 l = trie_leafindex(t, count);
1948 }
1949
1950 while (l) {
1951 cb->args[2] = l->key;
1952 if (fn_trie_dump_leaf(l, tb, skb, cb) < 0) {
1953 cb->args[3] = count;
1954 rcu_read_unlock();
1955 return -1;
1956 }
1957
1958 ++count;
1959 l = trie_nextleaf(l);
1960 memset(&cb->args[4], 0,
1961 sizeof(cb->args) - 4*sizeof(cb->args[0]));
1962 }
1963 cb->args[3] = count;
1964 rcu_read_unlock();
1965
1966 return skb->len;
1967}
1968
1969void __init fib_trie_init(void)
1970{
1971 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
1972 sizeof(struct fib_alias),
1973 0, SLAB_PANIC, NULL);
1974
1975 trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
1976 max(sizeof(struct leaf),
1977 sizeof(struct leaf_info)),
1978 0, SLAB_PANIC, NULL);
1979}
1980
1981
1982struct fib_table *fib_trie_table(u32 id)
1983{
1984 struct fib_table *tb;
1985 struct trie *t;
1986
1987 tb = kmalloc(sizeof(struct fib_table) + sizeof(struct trie),
1988 GFP_KERNEL);
1989 if (tb == NULL)
1990 return NULL;
1991
1992 tb->tb_id = id;
1993 tb->tb_default = -1;
1994 tb->tb_num_default = 0;
1995
1996 t = (struct trie *) tb->tb_data;
1997 memset(t, 0, sizeof(*t));
1998
1999 return tb;
2000}
2001
2002#ifdef CONFIG_PROC_FS
2003/* Depth first Trie walk iterator */
2004struct fib_trie_iter {
2005 struct seq_net_private p;
2006 struct fib_table *tb;
2007 struct tnode *tnode;
2008 unsigned int index;
2009 unsigned int depth;
2010};
2011
2012static struct rt_trie_node *fib_trie_get_next(struct fib_trie_iter *iter)
2013{
2014 struct tnode *tn = iter->tnode;
2015 unsigned int cindex = iter->index;
2016 struct tnode *p;
2017
2018 /* A single entry routing table */
2019 if (!tn)
2020 return NULL;
2021
2022 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2023 iter->tnode, iter->index, iter->depth);
2024rescan:
2025 while (cindex < (1<<tn->bits)) {
2026 struct rt_trie_node *n = tnode_get_child_rcu(tn, cindex);
2027
2028 if (n) {
2029 if (IS_LEAF(n)) {
2030 iter->tnode = tn;
2031 iter->index = cindex + 1;
2032 } else {
2033 /* push down one level */
2034 iter->tnode = (struct tnode *) n;
2035 iter->index = 0;
2036 ++iter->depth;
2037 }
2038 return n;
2039 }
2040
2041 ++cindex;
2042 }
2043
2044 /* Current node exhausted, pop back up */
2045 p = node_parent_rcu((struct rt_trie_node *)tn);
2046 if (p) {
2047 cindex = tkey_extract_bits(tn->key, p->pos, p->bits)+1;
2048 tn = p;
2049 --iter->depth;
2050 goto rescan;
2051 }
2052
2053 /* got root? */
2054 return NULL;
2055}
2056
2057static struct rt_trie_node *fib_trie_get_first(struct fib_trie_iter *iter,
2058 struct trie *t)
2059{
2060 struct rt_trie_node *n;
2061
2062 if (!t)
2063 return NULL;
2064
2065 n = rcu_dereference(t->trie);
2066 if (!n)
2067 return NULL;
2068
2069 if (IS_TNODE(n)) {
2070 iter->tnode = (struct tnode *) n;
2071 iter->index = 0;
2072 iter->depth = 1;
2073 } else {
2074 iter->tnode = NULL;
2075 iter->index = 0;
2076 iter->depth = 0;
2077 }
2078
2079 return n;
2080}
2081
2082static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2083{
2084 struct rt_trie_node *n;
2085 struct fib_trie_iter iter;
2086
2087 memset(s, 0, sizeof(*s));
2088
2089 rcu_read_lock();
2090 for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2091 if (IS_LEAF(n)) {
2092 struct leaf *l = (struct leaf *)n;
2093 struct leaf_info *li;
2094 struct hlist_node *tmp;
2095
2096 s->leaves++;
2097 s->totdepth += iter.depth;
2098 if (iter.depth > s->maxdepth)
2099 s->maxdepth = iter.depth;
2100
2101 hlist_for_each_entry_rcu(li, tmp, &l->list, hlist)
2102 ++s->prefixes;
2103 } else {
2104 const struct tnode *tn = (const struct tnode *) n;
2105 int i;
2106
2107 s->tnodes++;
2108 if (tn->bits < MAX_STAT_DEPTH)
2109 s->nodesizes[tn->bits]++;
2110
2111 for (i = 0; i < (1<<tn->bits); i++)
2112 if (!tn->child[i])
2113 s->nullpointers++;
2114 }
2115 }
2116 rcu_read_unlock();
2117}
2118
2119/*
2120 * This outputs /proc/net/fib_triestats
2121 */
2122static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2123{
2124 unsigned int i, max, pointers, bytes, avdepth;
2125
2126 if (stat->leaves)
2127 avdepth = stat->totdepth*100 / stat->leaves;
2128 else
2129 avdepth = 0;
2130
2131 seq_printf(seq, "\tAver depth: %u.%02d\n",
2132 avdepth / 100, avdepth % 100);
2133 seq_printf(seq, "\tMax depth: %u\n", stat->maxdepth);
2134
2135 seq_printf(seq, "\tLeaves: %u\n", stat->leaves);
2136 bytes = sizeof(struct leaf) * stat->leaves;
2137
2138 seq_printf(seq, "\tPrefixes: %u\n", stat->prefixes);
2139 bytes += sizeof(struct leaf_info) * stat->prefixes;
2140
2141 seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2142 bytes += sizeof(struct tnode) * stat->tnodes;
2143
2144 max = MAX_STAT_DEPTH;
2145 while (max > 0 && stat->nodesizes[max-1] == 0)
2146 max--;
2147
2148 pointers = 0;
2149 for (i = 1; i <= max; i++)
2150 if (stat->nodesizes[i] != 0) {
2151 seq_printf(seq, " %u: %u", i, stat->nodesizes[i]);
2152 pointers += (1<<i) * stat->nodesizes[i];
2153 }
2154 seq_putc(seq, '\n');
2155 seq_printf(seq, "\tPointers: %u\n", pointers);
2156
2157 bytes += sizeof(struct rt_trie_node *) * pointers;
2158 seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2159 seq_printf(seq, "Total size: %u kB\n", (bytes + 1023) / 1024);
2160}
2161
2162#ifdef CONFIG_IP_FIB_TRIE_STATS
2163static void trie_show_usage(struct seq_file *seq,
2164 const struct trie_use_stats *stats)
2165{
2166 seq_printf(seq, "\nCounters:\n---------\n");
2167 seq_printf(seq, "gets = %u\n", stats->gets);
2168 seq_printf(seq, "backtracks = %u\n", stats->backtrack);
2169 seq_printf(seq, "semantic match passed = %u\n",
2170 stats->semantic_match_passed);
2171 seq_printf(seq, "semantic match miss = %u\n",
2172 stats->semantic_match_miss);
2173 seq_printf(seq, "null node hit= %u\n", stats->null_node_hit);
2174 seq_printf(seq, "skipped node resize = %u\n\n",
2175 stats->resize_node_skipped);
2176}
2177#endif /* CONFIG_IP_FIB_TRIE_STATS */
2178
2179static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2180{
2181 if (tb->tb_id == RT_TABLE_LOCAL)
2182 seq_puts(seq, "Local:\n");
2183 else if (tb->tb_id == RT_TABLE_MAIN)
2184 seq_puts(seq, "Main:\n");
2185 else
2186 seq_printf(seq, "Id %d:\n", tb->tb_id);
2187}
2188
2189
2190static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2191{
2192 struct net *net = (struct net *)seq->private;
2193 unsigned int h;
2194
2195 seq_printf(seq,
2196 "Basic info: size of leaf:"
2197 " %Zd bytes, size of tnode: %Zd bytes.\n",
2198 sizeof(struct leaf), sizeof(struct tnode));
2199
2200 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2201 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2202 struct hlist_node *node;
2203 struct fib_table *tb;
2204
2205 hlist_for_each_entry_rcu(tb, node, head, tb_hlist) {
2206 struct trie *t = (struct trie *) tb->tb_data;
2207 struct trie_stat stat;
2208
2209 if (!t)
2210 continue;
2211
2212 fib_table_print(seq, tb);
2213
2214 trie_collect_stats(t, &stat);
2215 trie_show_stats(seq, &stat);
2216#ifdef CONFIG_IP_FIB_TRIE_STATS
2217 trie_show_usage(seq, &t->stats);
2218#endif
2219 }
2220 }
2221
2222 return 0;
2223}
2224
2225static int fib_triestat_seq_open(struct inode *inode, struct file *file)
2226{
2227 return single_open_net(inode, file, fib_triestat_seq_show);
2228}
2229
2230static const struct file_operations fib_triestat_fops = {
2231 .owner = THIS_MODULE,
2232 .open = fib_triestat_seq_open,
2233 .read = seq_read,
2234 .llseek = seq_lseek,
2235 .release = single_release_net,
2236};
2237
2238static struct rt_trie_node *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2239{
2240 struct fib_trie_iter *iter = seq->private;
2241 struct net *net = seq_file_net(seq);
2242 loff_t idx = 0;
2243 unsigned int h;
2244
2245 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2246 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2247 struct hlist_node *node;
2248 struct fib_table *tb;
2249
2250 hlist_for_each_entry_rcu(tb, node, head, tb_hlist) {
2251 struct rt_trie_node *n;
2252
2253 for (n = fib_trie_get_first(iter,
2254 (struct trie *) tb->tb_data);
2255 n; n = fib_trie_get_next(iter))
2256 if (pos == idx++) {
2257 iter->tb = tb;
2258 return n;
2259 }
2260 }
2261 }
2262
2263 return NULL;
2264}
2265
2266static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2267 __acquires(RCU)
2268{
2269 rcu_read_lock();
2270 return fib_trie_get_idx(seq, *pos);
2271}
2272
2273static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2274{
2275 struct fib_trie_iter *iter = seq->private;
2276 struct net *net = seq_file_net(seq);
2277 struct fib_table *tb = iter->tb;
2278 struct hlist_node *tb_node;
2279 unsigned int h;
2280 struct rt_trie_node *n;
2281
2282 ++*pos;
2283 /* next node in same table */
2284 n = fib_trie_get_next(iter);
2285 if (n)
2286 return n;
2287
2288 /* walk rest of this hash chain */
2289 h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2290 while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2291 tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2292 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2293 if (n)
2294 goto found;
2295 }
2296
2297 /* new hash chain */
2298 while (++h < FIB_TABLE_HASHSZ) {
2299 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2300 hlist_for_each_entry_rcu(tb, tb_node, head, tb_hlist) {
2301 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2302 if (n)
2303 goto found;
2304 }
2305 }
2306 return NULL;
2307
2308found:
2309 iter->tb = tb;
2310 return n;
2311}
2312
2313static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2314 __releases(RCU)
2315{
2316 rcu_read_unlock();
2317}
2318
2319static void seq_indent(struct seq_file *seq, int n)
2320{
2321 while (n-- > 0)
2322 seq_puts(seq, " ");
2323}
2324
2325static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2326{
2327 switch (s) {
2328 case RT_SCOPE_UNIVERSE: return "universe";
2329 case RT_SCOPE_SITE: return "site";
2330 case RT_SCOPE_LINK: return "link";
2331 case RT_SCOPE_HOST: return "host";
2332 case RT_SCOPE_NOWHERE: return "nowhere";
2333 default:
2334 snprintf(buf, len, "scope=%d", s);
2335 return buf;
2336 }
2337}
2338
2339static const char *const rtn_type_names[__RTN_MAX] = {
2340 [RTN_UNSPEC] = "UNSPEC",
2341 [RTN_UNICAST] = "UNICAST",
2342 [RTN_LOCAL] = "LOCAL",
2343 [RTN_BROADCAST] = "BROADCAST",
2344 [RTN_ANYCAST] = "ANYCAST",
2345 [RTN_MULTICAST] = "MULTICAST",
2346 [RTN_BLACKHOLE] = "BLACKHOLE",
2347 [RTN_UNREACHABLE] = "UNREACHABLE",
2348 [RTN_PROHIBIT] = "PROHIBIT",
2349 [RTN_THROW] = "THROW",
2350 [RTN_NAT] = "NAT",
2351 [RTN_XRESOLVE] = "XRESOLVE",
2352};
2353
2354static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2355{
2356 if (t < __RTN_MAX && rtn_type_names[t])
2357 return rtn_type_names[t];
2358 snprintf(buf, len, "type %u", t);
2359 return buf;
2360}
2361
2362/* Pretty print the trie */
2363static int fib_trie_seq_show(struct seq_file *seq, void *v)
2364{
2365 const struct fib_trie_iter *iter = seq->private;
2366 struct rt_trie_node *n = v;
2367
2368 if (!node_parent_rcu(n))
2369 fib_table_print(seq, iter->tb);
2370
2371 if (IS_TNODE(n)) {
2372 struct tnode *tn = (struct tnode *) n;
2373 __be32 prf = htonl(mask_pfx(tn->key, tn->pos));
2374
2375 seq_indent(seq, iter->depth-1);
2376 seq_printf(seq, " +-- %pI4/%d %d %d %d\n",
2377 &prf, tn->pos, tn->bits, tn->full_children,
2378 tn->empty_children);
2379
2380 } else {
2381 struct leaf *l = (struct leaf *) n;
2382 struct leaf_info *li;
2383 struct hlist_node *node;
2384 __be32 val = htonl(l->key);
2385
2386 seq_indent(seq, iter->depth);
2387 seq_printf(seq, " |-- %pI4\n", &val);
2388
2389 hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
2390 struct fib_alias *fa;
2391
2392 list_for_each_entry_rcu(fa, &li->falh, fa_list) {
2393 char buf1[32], buf2[32];
2394
2395 seq_indent(seq, iter->depth+1);
2396 seq_printf(seq, " /%d %s %s", li->plen,
2397 rtn_scope(buf1, sizeof(buf1),
2398 fa->fa_info->fib_scope),
2399 rtn_type(buf2, sizeof(buf2),
2400 fa->fa_type));
2401 if (fa->fa_tos)
2402 seq_printf(seq, " tos=%d", fa->fa_tos);
2403 seq_putc(seq, '\n');
2404 }
2405 }
2406 }
2407
2408 return 0;
2409}
2410
2411static const struct seq_operations fib_trie_seq_ops = {
2412 .start = fib_trie_seq_start,
2413 .next = fib_trie_seq_next,
2414 .stop = fib_trie_seq_stop,
2415 .show = fib_trie_seq_show,
2416};
2417
2418static int fib_trie_seq_open(struct inode *inode, struct file *file)
2419{
2420 return seq_open_net(inode, file, &fib_trie_seq_ops,
2421 sizeof(struct fib_trie_iter));
2422}
2423
2424static const struct file_operations fib_trie_fops = {
2425 .owner = THIS_MODULE,
2426 .open = fib_trie_seq_open,
2427 .read = seq_read,
2428 .llseek = seq_lseek,
2429 .release = seq_release_net,
2430};
2431
2432struct fib_route_iter {
2433 struct seq_net_private p;
2434 struct trie *main_trie;
2435 loff_t pos;
2436 t_key key;
2437};
2438
2439static struct leaf *fib_route_get_idx(struct fib_route_iter *iter, loff_t pos)
2440{
2441 struct leaf *l = NULL;
2442 struct trie *t = iter->main_trie;
2443
2444 /* use cache location of last found key */
2445 if (iter->pos > 0 && pos >= iter->pos && (l = fib_find_node(t, iter->key)))
2446 pos -= iter->pos;
2447 else {
2448 iter->pos = 0;
2449 l = trie_firstleaf(t);
2450 }
2451
2452 while (l && pos-- > 0) {
2453 iter->pos++;
2454 l = trie_nextleaf(l);
2455 }
2456
2457 if (l)
2458 iter->key = pos; /* remember it */
2459 else
2460 iter->pos = 0; /* forget it */
2461
2462 return l;
2463}
2464
2465static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2466 __acquires(RCU)
2467{
2468 struct fib_route_iter *iter = seq->private;
2469 struct fib_table *tb;
2470
2471 rcu_read_lock();
2472 tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2473 if (!tb)
2474 return NULL;
2475
2476 iter->main_trie = (struct trie *) tb->tb_data;
2477 if (*pos == 0)
2478 return SEQ_START_TOKEN;
2479 else
2480 return fib_route_get_idx(iter, *pos - 1);
2481}
2482
2483static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2484{
2485 struct fib_route_iter *iter = seq->private;
2486 struct leaf *l = v;
2487
2488 ++*pos;
2489 if (v == SEQ_START_TOKEN) {
2490 iter->pos = 0;
2491 l = trie_firstleaf(iter->main_trie);
2492 } else {
2493 iter->pos++;
2494 l = trie_nextleaf(l);
2495 }
2496
2497 if (l)
2498 iter->key = l->key;
2499 else
2500 iter->pos = 0;
2501 return l;
2502}
2503
2504static void fib_route_seq_stop(struct seq_file *seq, void *v)
2505 __releases(RCU)
2506{
2507 rcu_read_unlock();
2508}
2509
2510static unsigned int fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2511{
2512 unsigned int flags = 0;
2513
2514 if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2515 flags = RTF_REJECT;
2516 if (fi && fi->fib_nh->nh_gw)
2517 flags |= RTF_GATEWAY;
2518 if (mask == htonl(0xFFFFFFFF))
2519 flags |= RTF_HOST;
2520 flags |= RTF_UP;
2521 return flags;
2522}
2523
2524/*
2525 * This outputs /proc/net/route.
2526 * The format of the file is not supposed to be changed
2527 * and needs to be same as fib_hash output to avoid breaking
2528 * legacy utilities
2529 */
2530static int fib_route_seq_show(struct seq_file *seq, void *v)
2531{
2532 struct leaf *l = v;
2533 struct leaf_info *li;
2534 struct hlist_node *node;
2535
2536 if (v == SEQ_START_TOKEN) {
2537 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2538 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2539 "\tWindow\tIRTT");
2540 return 0;
2541 }
2542
2543 hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
2544 struct fib_alias *fa;
2545 __be32 mask, prefix;
2546
2547 mask = inet_make_mask(li->plen);
2548 prefix = htonl(l->key);
2549
2550 list_for_each_entry_rcu(fa, &li->falh, fa_list) {
2551 const struct fib_info *fi = fa->fa_info;
2552 unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2553 int len;
2554
2555 if (fa->fa_type == RTN_BROADCAST
2556 || fa->fa_type == RTN_MULTICAST)
2557 continue;
2558
2559 if (fi)
2560 seq_printf(seq,
2561 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2562 "%d\t%08X\t%d\t%u\t%u%n",
2563 fi->fib_dev ? fi->fib_dev->name : "*",
2564 prefix,
2565 fi->fib_nh->nh_gw, flags, 0, 0,
2566 fi->fib_priority,
2567 mask,
2568 (fi->fib_advmss ?
2569 fi->fib_advmss + 40 : 0),
2570 fi->fib_window,
2571 fi->fib_rtt >> 3, &len);
2572 else
2573 seq_printf(seq,
2574 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2575 "%d\t%08X\t%d\t%u\t%u%n",
2576 prefix, 0, flags, 0, 0, 0,
2577 mask, 0, 0, 0, &len);
2578
2579 seq_printf(seq, "%*s\n", 127 - len, "");
2580 }
2581 }
2582
2583 return 0;
2584}
2585
2586static const struct seq_operations fib_route_seq_ops = {
2587 .start = fib_route_seq_start,
2588 .next = fib_route_seq_next,
2589 .stop = fib_route_seq_stop,
2590 .show = fib_route_seq_show,
2591};
2592
2593static int fib_route_seq_open(struct inode *inode, struct file *file)
2594{
2595 return seq_open_net(inode, file, &fib_route_seq_ops,
2596 sizeof(struct fib_route_iter));
2597}
2598
2599static const struct file_operations fib_route_fops = {
2600 .owner = THIS_MODULE,
2601 .open = fib_route_seq_open,
2602 .read = seq_read,
2603 .llseek = seq_lseek,
2604 .release = seq_release_net,
2605};
2606
2607int __net_init fib_proc_init(struct net *net)
2608{
2609 if (!proc_net_fops_create(net, "fib_trie", S_IRUGO, &fib_trie_fops))
2610 goto out1;
2611
2612 if (!proc_net_fops_create(net, "fib_triestat", S_IRUGO,
2613 &fib_triestat_fops))
2614 goto out2;
2615
2616 if (!proc_net_fops_create(net, "route", S_IRUGO, &fib_route_fops))
2617 goto out3;
2618
2619 return 0;
2620
2621out3:
2622 proc_net_remove(net, "fib_triestat");
2623out2:
2624 proc_net_remove(net, "fib_trie");
2625out1:
2626 return -ENOMEM;
2627}
2628
2629void __net_exit fib_proc_exit(struct net *net)
2630{
2631 proc_net_remove(net, "fib_trie");
2632 proc_net_remove(net, "fib_triestat");
2633 proc_net_remove(net, "route");
2634}
2635
2636#endif /* CONFIG_PROC_FS */
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 <net/net_namespace.h>
65#include <net/inet_dscp.h>
66#include <net/ip.h>
67#include <net/protocol.h>
68#include <net/route.h>
69#include <net/tcp.h>
70#include <net/sock.h>
71#include <net/ip_fib.h>
72#include <net/fib_notifier.h>
73#include <trace/events/fib.h>
74#include "fib_lookup.h"
75
76static int call_fib_entry_notifier(struct notifier_block *nb,
77 enum fib_event_type event_type, u32 dst,
78 int dst_len, struct fib_alias *fa,
79 struct netlink_ext_ack *extack)
80{
81 struct fib_entry_notifier_info info = {
82 .info.extack = extack,
83 .dst = dst,
84 .dst_len = dst_len,
85 .fi = fa->fa_info,
86 .dscp = fa->fa_dscp,
87 .type = fa->fa_type,
88 .tb_id = fa->tb_id,
89 };
90 return call_fib4_notifier(nb, event_type, &info.info);
91}
92
93static int call_fib_entry_notifiers(struct net *net,
94 enum fib_event_type event_type, u32 dst,
95 int dst_len, struct fib_alias *fa,
96 struct netlink_ext_ack *extack)
97{
98 struct fib_entry_notifier_info info = {
99 .info.extack = extack,
100 .dst = dst,
101 .dst_len = dst_len,
102 .fi = fa->fa_info,
103 .dscp = fa->fa_dscp,
104 .type = fa->fa_type,
105 .tb_id = fa->tb_id,
106 };
107 return call_fib4_notifiers(net, event_type, &info.info);
108}
109
110#define MAX_STAT_DEPTH 32
111
112#define KEYLENGTH (8*sizeof(t_key))
113#define KEY_MAX ((t_key)~0)
114
115typedef unsigned int t_key;
116
117#define IS_TRIE(n) ((n)->pos >= KEYLENGTH)
118#define IS_TNODE(n) ((n)->bits)
119#define IS_LEAF(n) (!(n)->bits)
120
121struct key_vector {
122 t_key key;
123 unsigned char pos; /* 2log(KEYLENGTH) bits needed */
124 unsigned char bits; /* 2log(KEYLENGTH) bits needed */
125 unsigned char slen;
126 union {
127 /* This list pointer if valid if (pos | bits) == 0 (LEAF) */
128 struct hlist_head leaf;
129 /* This array is valid if (pos | bits) > 0 (TNODE) */
130 DECLARE_FLEX_ARRAY(struct key_vector __rcu *, tnode);
131 };
132};
133
134struct tnode {
135 struct rcu_head rcu;
136 t_key empty_children; /* KEYLENGTH bits needed */
137 t_key full_children; /* KEYLENGTH bits needed */
138 struct key_vector __rcu *parent;
139 struct key_vector kv[1];
140#define tn_bits kv[0].bits
141};
142
143#define TNODE_SIZE(n) offsetof(struct tnode, kv[0].tnode[n])
144#define LEAF_SIZE TNODE_SIZE(1)
145
146#ifdef CONFIG_IP_FIB_TRIE_STATS
147struct trie_use_stats {
148 unsigned int gets;
149 unsigned int backtrack;
150 unsigned int semantic_match_passed;
151 unsigned int semantic_match_miss;
152 unsigned int null_node_hit;
153 unsigned int resize_node_skipped;
154};
155#endif
156
157struct trie_stat {
158 unsigned int totdepth;
159 unsigned int maxdepth;
160 unsigned int tnodes;
161 unsigned int leaves;
162 unsigned int nullpointers;
163 unsigned int prefixes;
164 unsigned int nodesizes[MAX_STAT_DEPTH];
165};
166
167struct trie {
168 struct key_vector kv[1];
169#ifdef CONFIG_IP_FIB_TRIE_STATS
170 struct trie_use_stats __percpu *stats;
171#endif
172};
173
174static struct key_vector *resize(struct trie *t, struct key_vector *tn);
175static unsigned int tnode_free_size;
176
177/*
178 * synchronize_rcu after call_rcu for outstanding dirty memory; it should be
179 * especially useful before resizing the root node with PREEMPT_NONE configs;
180 * the value was obtained experimentally, aiming to avoid visible slowdown.
181 */
182unsigned int sysctl_fib_sync_mem = 512 * 1024;
183unsigned int sysctl_fib_sync_mem_min = 64 * 1024;
184unsigned int sysctl_fib_sync_mem_max = 64 * 1024 * 1024;
185
186static struct kmem_cache *fn_alias_kmem __ro_after_init;
187static struct kmem_cache *trie_leaf_kmem __ro_after_init;
188
189static inline struct tnode *tn_info(struct key_vector *kv)
190{
191 return container_of(kv, struct tnode, kv[0]);
192}
193
194/* caller must hold RTNL */
195#define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
196#define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
197
198/* caller must hold RCU read lock or RTNL */
199#define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
200#define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
201
202/* wrapper for rcu_assign_pointer */
203static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
204{
205 if (n)
206 rcu_assign_pointer(tn_info(n)->parent, tp);
207}
208
209#define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
210
211/* This provides us with the number of children in this node, in the case of a
212 * leaf this will return 0 meaning none of the children are accessible.
213 */
214static inline unsigned long child_length(const struct key_vector *tn)
215{
216 return (1ul << tn->bits) & ~(1ul);
217}
218
219#define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
220
221static inline unsigned long get_index(t_key key, struct key_vector *kv)
222{
223 unsigned long index = key ^ kv->key;
224
225 if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
226 return 0;
227
228 return index >> kv->pos;
229}
230
231/* To understand this stuff, an understanding of keys and all their bits is
232 * necessary. Every node in the trie has a key associated with it, but not
233 * all of the bits in that key are significant.
234 *
235 * Consider a node 'n' and its parent 'tp'.
236 *
237 * If n is a leaf, every bit in its key is significant. Its presence is
238 * necessitated by path compression, since during a tree traversal (when
239 * searching for a leaf - unless we are doing an insertion) we will completely
240 * ignore all skipped bits we encounter. Thus we need to verify, at the end of
241 * a potentially successful search, that we have indeed been walking the
242 * correct key path.
243 *
244 * Note that we can never "miss" the correct key in the tree if present by
245 * following the wrong path. Path compression ensures that segments of the key
246 * that are the same for all keys with a given prefix are skipped, but the
247 * skipped part *is* identical for each node in the subtrie below the skipped
248 * bit! trie_insert() in this implementation takes care of that.
249 *
250 * if n is an internal node - a 'tnode' here, the various parts of its key
251 * have many different meanings.
252 *
253 * Example:
254 * _________________________________________________________________
255 * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
256 * -----------------------------------------------------------------
257 * 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16
258 *
259 * _________________________________________________________________
260 * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
261 * -----------------------------------------------------------------
262 * 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
263 *
264 * tp->pos = 22
265 * tp->bits = 3
266 * n->pos = 13
267 * n->bits = 4
268 *
269 * First, let's just ignore the bits that come before the parent tp, that is
270 * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
271 * point we do not use them for anything.
272 *
273 * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
274 * index into the parent's child array. That is, they will be used to find
275 * 'n' among tp's children.
276 *
277 * The bits from (n->pos + n->bits) to (tp->pos - 1) - "S" - are skipped bits
278 * for the node n.
279 *
280 * All the bits we have seen so far are significant to the node n. The rest
281 * of the bits are really not needed or indeed known in n->key.
282 *
283 * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
284 * n's child array, and will of course be different for each child.
285 *
286 * The rest of the bits, from 0 to (n->pos -1) - "u" - are completely unknown
287 * at this point.
288 */
289
290static const int halve_threshold = 25;
291static const int inflate_threshold = 50;
292static const int halve_threshold_root = 15;
293static const int inflate_threshold_root = 30;
294
295static void __alias_free_mem(struct rcu_head *head)
296{
297 struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
298 kmem_cache_free(fn_alias_kmem, fa);
299}
300
301static inline void alias_free_mem_rcu(struct fib_alias *fa)
302{
303 call_rcu(&fa->rcu, __alias_free_mem);
304}
305
306#define TNODE_VMALLOC_MAX \
307 ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
308
309static void __node_free_rcu(struct rcu_head *head)
310{
311 struct tnode *n = container_of(head, struct tnode, rcu);
312
313 if (!n->tn_bits)
314 kmem_cache_free(trie_leaf_kmem, n);
315 else
316 kvfree(n);
317}
318
319#define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
320
321static struct tnode *tnode_alloc(int bits)
322{
323 size_t size;
324
325 /* verify bits is within bounds */
326 if (bits > TNODE_VMALLOC_MAX)
327 return NULL;
328
329 /* determine size and verify it is non-zero and didn't overflow */
330 size = TNODE_SIZE(1ul << bits);
331
332 if (size <= PAGE_SIZE)
333 return kzalloc(size, GFP_KERNEL);
334 else
335 return vzalloc(size);
336}
337
338static inline void empty_child_inc(struct key_vector *n)
339{
340 tn_info(n)->empty_children++;
341
342 if (!tn_info(n)->empty_children)
343 tn_info(n)->full_children++;
344}
345
346static inline void empty_child_dec(struct key_vector *n)
347{
348 if (!tn_info(n)->empty_children)
349 tn_info(n)->full_children--;
350
351 tn_info(n)->empty_children--;
352}
353
354static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
355{
356 struct key_vector *l;
357 struct tnode *kv;
358
359 kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
360 if (!kv)
361 return NULL;
362
363 /* initialize key vector */
364 l = kv->kv;
365 l->key = key;
366 l->pos = 0;
367 l->bits = 0;
368 l->slen = fa->fa_slen;
369
370 /* link leaf to fib alias */
371 INIT_HLIST_HEAD(&l->leaf);
372 hlist_add_head(&fa->fa_list, &l->leaf);
373
374 return l;
375}
376
377static struct key_vector *tnode_new(t_key key, int pos, int bits)
378{
379 unsigned int shift = pos + bits;
380 struct key_vector *tn;
381 struct tnode *tnode;
382
383 /* verify bits and pos their msb bits clear and values are valid */
384 BUG_ON(!bits || (shift > KEYLENGTH));
385
386 tnode = tnode_alloc(bits);
387 if (!tnode)
388 return NULL;
389
390 pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
391 sizeof(struct key_vector *) << bits);
392
393 if (bits == KEYLENGTH)
394 tnode->full_children = 1;
395 else
396 tnode->empty_children = 1ul << bits;
397
398 tn = tnode->kv;
399 tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
400 tn->pos = pos;
401 tn->bits = bits;
402 tn->slen = pos;
403
404 return tn;
405}
406
407/* Check whether a tnode 'n' is "full", i.e. it is an internal node
408 * and no bits are skipped. See discussion in dyntree paper p. 6
409 */
410static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
411{
412 return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
413}
414
415/* Add a child at position i overwriting the old value.
416 * Update the value of full_children and empty_children.
417 */
418static void put_child(struct key_vector *tn, unsigned long i,
419 struct key_vector *n)
420{
421 struct key_vector *chi = get_child(tn, i);
422 int isfull, wasfull;
423
424 BUG_ON(i >= child_length(tn));
425
426 /* update emptyChildren, overflow into fullChildren */
427 if (!n && chi)
428 empty_child_inc(tn);
429 if (n && !chi)
430 empty_child_dec(tn);
431
432 /* update fullChildren */
433 wasfull = tnode_full(tn, chi);
434 isfull = tnode_full(tn, n);
435
436 if (wasfull && !isfull)
437 tn_info(tn)->full_children--;
438 else if (!wasfull && isfull)
439 tn_info(tn)->full_children++;
440
441 if (n && (tn->slen < n->slen))
442 tn->slen = n->slen;
443
444 rcu_assign_pointer(tn->tnode[i], n);
445}
446
447static void update_children(struct key_vector *tn)
448{
449 unsigned long i;
450
451 /* update all of the child parent pointers */
452 for (i = child_length(tn); i;) {
453 struct key_vector *inode = get_child(tn, --i);
454
455 if (!inode)
456 continue;
457
458 /* Either update the children of a tnode that
459 * already belongs to us or update the child
460 * to point to ourselves.
461 */
462 if (node_parent(inode) == tn)
463 update_children(inode);
464 else
465 node_set_parent(inode, tn);
466 }
467}
468
469static inline void put_child_root(struct key_vector *tp, t_key key,
470 struct key_vector *n)
471{
472 if (IS_TRIE(tp))
473 rcu_assign_pointer(tp->tnode[0], n);
474 else
475 put_child(tp, get_index(key, tp), n);
476}
477
478static inline void tnode_free_init(struct key_vector *tn)
479{
480 tn_info(tn)->rcu.next = NULL;
481}
482
483static inline void tnode_free_append(struct key_vector *tn,
484 struct key_vector *n)
485{
486 tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
487 tn_info(tn)->rcu.next = &tn_info(n)->rcu;
488}
489
490static void tnode_free(struct key_vector *tn)
491{
492 struct callback_head *head = &tn_info(tn)->rcu;
493
494 while (head) {
495 head = head->next;
496 tnode_free_size += TNODE_SIZE(1ul << tn->bits);
497 node_free(tn);
498
499 tn = container_of(head, struct tnode, rcu)->kv;
500 }
501
502 if (tnode_free_size >= READ_ONCE(sysctl_fib_sync_mem)) {
503 tnode_free_size = 0;
504 synchronize_rcu();
505 }
506}
507
508static struct key_vector *replace(struct trie *t,
509 struct key_vector *oldtnode,
510 struct key_vector *tn)
511{
512 struct key_vector *tp = node_parent(oldtnode);
513 unsigned long i;
514
515 /* setup the parent pointer out of and back into this node */
516 NODE_INIT_PARENT(tn, tp);
517 put_child_root(tp, tn->key, tn);
518
519 /* update all of the child parent pointers */
520 update_children(tn);
521
522 /* all pointers should be clean so we are done */
523 tnode_free(oldtnode);
524
525 /* resize children now that oldtnode is freed */
526 for (i = child_length(tn); i;) {
527 struct key_vector *inode = get_child(tn, --i);
528
529 /* resize child node */
530 if (tnode_full(tn, inode))
531 tn = resize(t, inode);
532 }
533
534 return tp;
535}
536
537static struct key_vector *inflate(struct trie *t,
538 struct key_vector *oldtnode)
539{
540 struct key_vector *tn;
541 unsigned long i;
542 t_key m;
543
544 pr_debug("In inflate\n");
545
546 tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
547 if (!tn)
548 goto notnode;
549
550 /* prepare oldtnode to be freed */
551 tnode_free_init(oldtnode);
552
553 /* Assemble all of the pointers in our cluster, in this case that
554 * represents all of the pointers out of our allocated nodes that
555 * point to existing tnodes and the links between our allocated
556 * nodes.
557 */
558 for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
559 struct key_vector *inode = get_child(oldtnode, --i);
560 struct key_vector *node0, *node1;
561 unsigned long j, k;
562
563 /* An empty child */
564 if (!inode)
565 continue;
566
567 /* A leaf or an internal node with skipped bits */
568 if (!tnode_full(oldtnode, inode)) {
569 put_child(tn, get_index(inode->key, tn), inode);
570 continue;
571 }
572
573 /* drop the node in the old tnode free list */
574 tnode_free_append(oldtnode, inode);
575
576 /* An internal node with two children */
577 if (inode->bits == 1) {
578 put_child(tn, 2 * i + 1, get_child(inode, 1));
579 put_child(tn, 2 * i, get_child(inode, 0));
580 continue;
581 }
582
583 /* We will replace this node 'inode' with two new
584 * ones, 'node0' and 'node1', each with half of the
585 * original children. The two new nodes will have
586 * a position one bit further down the key and this
587 * means that the "significant" part of their keys
588 * (see the discussion near the top of this file)
589 * will differ by one bit, which will be "0" in
590 * node0's key and "1" in node1's key. Since we are
591 * moving the key position by one step, the bit that
592 * we are moving away from - the bit at position
593 * (tn->pos) - is the one that will differ between
594 * node0 and node1. So... we synthesize that bit in the
595 * two new keys.
596 */
597 node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
598 if (!node1)
599 goto nomem;
600 node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
601
602 tnode_free_append(tn, node1);
603 if (!node0)
604 goto nomem;
605 tnode_free_append(tn, node0);
606
607 /* populate child pointers in new nodes */
608 for (k = child_length(inode), j = k / 2; j;) {
609 put_child(node1, --j, get_child(inode, --k));
610 put_child(node0, j, get_child(inode, j));
611 put_child(node1, --j, get_child(inode, --k));
612 put_child(node0, j, get_child(inode, j));
613 }
614
615 /* link new nodes to parent */
616 NODE_INIT_PARENT(node1, tn);
617 NODE_INIT_PARENT(node0, tn);
618
619 /* link parent to nodes */
620 put_child(tn, 2 * i + 1, node1);
621 put_child(tn, 2 * i, node0);
622 }
623
624 /* setup the parent pointers into and out of this node */
625 return replace(t, oldtnode, tn);
626nomem:
627 /* all pointers should be clean so we are done */
628 tnode_free(tn);
629notnode:
630 return NULL;
631}
632
633static struct key_vector *halve(struct trie *t,
634 struct key_vector *oldtnode)
635{
636 struct key_vector *tn;
637 unsigned long i;
638
639 pr_debug("In halve\n");
640
641 tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
642 if (!tn)
643 goto notnode;
644
645 /* prepare oldtnode to be freed */
646 tnode_free_init(oldtnode);
647
648 /* Assemble all of the pointers in our cluster, in this case that
649 * represents all of the pointers out of our allocated nodes that
650 * point to existing tnodes and the links between our allocated
651 * nodes.
652 */
653 for (i = child_length(oldtnode); i;) {
654 struct key_vector *node1 = get_child(oldtnode, --i);
655 struct key_vector *node0 = get_child(oldtnode, --i);
656 struct key_vector *inode;
657
658 /* At least one of the children is empty */
659 if (!node1 || !node0) {
660 put_child(tn, i / 2, node1 ? : node0);
661 continue;
662 }
663
664 /* Two nonempty children */
665 inode = tnode_new(node0->key, oldtnode->pos, 1);
666 if (!inode)
667 goto nomem;
668 tnode_free_append(tn, inode);
669
670 /* initialize pointers out of node */
671 put_child(inode, 1, node1);
672 put_child(inode, 0, node0);
673 NODE_INIT_PARENT(inode, tn);
674
675 /* link parent to node */
676 put_child(tn, i / 2, inode);
677 }
678
679 /* setup the parent pointers into and out of this node */
680 return replace(t, oldtnode, tn);
681nomem:
682 /* all pointers should be clean so we are done */
683 tnode_free(tn);
684notnode:
685 return NULL;
686}
687
688static struct key_vector *collapse(struct trie *t,
689 struct key_vector *oldtnode)
690{
691 struct key_vector *n, *tp;
692 unsigned long i;
693
694 /* scan the tnode looking for that one child that might still exist */
695 for (n = NULL, i = child_length(oldtnode); !n && i;)
696 n = get_child(oldtnode, --i);
697
698 /* compress one level */
699 tp = node_parent(oldtnode);
700 put_child_root(tp, oldtnode->key, n);
701 node_set_parent(n, tp);
702
703 /* drop dead node */
704 node_free(oldtnode);
705
706 return tp;
707}
708
709static unsigned char update_suffix(struct key_vector *tn)
710{
711 unsigned char slen = tn->pos;
712 unsigned long stride, i;
713 unsigned char slen_max;
714
715 /* only vector 0 can have a suffix length greater than or equal to
716 * tn->pos + tn->bits, the second highest node will have a suffix
717 * length at most of tn->pos + tn->bits - 1
718 */
719 slen_max = min_t(unsigned char, tn->pos + tn->bits - 1, tn->slen);
720
721 /* search though the list of children looking for nodes that might
722 * have a suffix greater than the one we currently have. This is
723 * why we start with a stride of 2 since a stride of 1 would
724 * represent the nodes with suffix length equal to tn->pos
725 */
726 for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
727 struct key_vector *n = get_child(tn, i);
728
729 if (!n || (n->slen <= slen))
730 continue;
731
732 /* update stride and slen based on new value */
733 stride <<= (n->slen - slen);
734 slen = n->slen;
735 i &= ~(stride - 1);
736
737 /* stop searching if we have hit the maximum possible value */
738 if (slen >= slen_max)
739 break;
740 }
741
742 tn->slen = slen;
743
744 return slen;
745}
746
747/* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
748 * the Helsinki University of Technology and Matti Tikkanen of Nokia
749 * Telecommunications, page 6:
750 * "A node is doubled if the ratio of non-empty children to all
751 * children in the *doubled* node is at least 'high'."
752 *
753 * 'high' in this instance is the variable 'inflate_threshold'. It
754 * is expressed as a percentage, so we multiply it with
755 * child_length() and instead of multiplying by 2 (since the
756 * child array will be doubled by inflate()) and multiplying
757 * the left-hand side by 100 (to handle the percentage thing) we
758 * multiply the left-hand side by 50.
759 *
760 * The left-hand side may look a bit weird: child_length(tn)
761 * - tn->empty_children is of course the number of non-null children
762 * in the current node. tn->full_children is the number of "full"
763 * children, that is non-null tnodes with a skip value of 0.
764 * All of those will be doubled in the resulting inflated tnode, so
765 * we just count them one extra time here.
766 *
767 * A clearer way to write this would be:
768 *
769 * to_be_doubled = tn->full_children;
770 * not_to_be_doubled = child_length(tn) - tn->empty_children -
771 * tn->full_children;
772 *
773 * new_child_length = child_length(tn) * 2;
774 *
775 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
776 * new_child_length;
777 * if (new_fill_factor >= inflate_threshold)
778 *
779 * ...and so on, tho it would mess up the while () loop.
780 *
781 * anyway,
782 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
783 * inflate_threshold
784 *
785 * avoid a division:
786 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
787 * inflate_threshold * new_child_length
788 *
789 * expand not_to_be_doubled and to_be_doubled, and shorten:
790 * 100 * (child_length(tn) - tn->empty_children +
791 * tn->full_children) >= inflate_threshold * new_child_length
792 *
793 * expand new_child_length:
794 * 100 * (child_length(tn) - tn->empty_children +
795 * tn->full_children) >=
796 * inflate_threshold * child_length(tn) * 2
797 *
798 * shorten again:
799 * 50 * (tn->full_children + child_length(tn) -
800 * tn->empty_children) >= inflate_threshold *
801 * child_length(tn)
802 *
803 */
804static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
805{
806 unsigned long used = child_length(tn);
807 unsigned long threshold = used;
808
809 /* Keep root node larger */
810 threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
811 used -= tn_info(tn)->empty_children;
812 used += tn_info(tn)->full_children;
813
814 /* if bits == KEYLENGTH then pos = 0, and will fail below */
815
816 return (used > 1) && tn->pos && ((50 * used) >= threshold);
817}
818
819static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
820{
821 unsigned long used = child_length(tn);
822 unsigned long threshold = used;
823
824 /* Keep root node larger */
825 threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
826 used -= tn_info(tn)->empty_children;
827
828 /* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
829
830 return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
831}
832
833static inline bool should_collapse(struct key_vector *tn)
834{
835 unsigned long used = child_length(tn);
836
837 used -= tn_info(tn)->empty_children;
838
839 /* account for bits == KEYLENGTH case */
840 if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
841 used -= KEY_MAX;
842
843 /* One child or none, time to drop us from the trie */
844 return used < 2;
845}
846
847#define MAX_WORK 10
848static struct key_vector *resize(struct trie *t, struct key_vector *tn)
849{
850#ifdef CONFIG_IP_FIB_TRIE_STATS
851 struct trie_use_stats __percpu *stats = t->stats;
852#endif
853 struct key_vector *tp = node_parent(tn);
854 unsigned long cindex = get_index(tn->key, tp);
855 int max_work = MAX_WORK;
856
857 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
858 tn, inflate_threshold, halve_threshold);
859
860 /* track the tnode via the pointer from the parent instead of
861 * doing it ourselves. This way we can let RCU fully do its
862 * thing without us interfering
863 */
864 BUG_ON(tn != get_child(tp, cindex));
865
866 /* Double as long as the resulting node has a number of
867 * nonempty nodes that are above the threshold.
868 */
869 while (should_inflate(tp, tn) && max_work) {
870 tp = inflate(t, tn);
871 if (!tp) {
872#ifdef CONFIG_IP_FIB_TRIE_STATS
873 this_cpu_inc(stats->resize_node_skipped);
874#endif
875 break;
876 }
877
878 max_work--;
879 tn = get_child(tp, cindex);
880 }
881
882 /* update parent in case inflate failed */
883 tp = node_parent(tn);
884
885 /* Return if at least one inflate is run */
886 if (max_work != MAX_WORK)
887 return tp;
888
889 /* Halve as long as the number of empty children in this
890 * node is above threshold.
891 */
892 while (should_halve(tp, tn) && max_work) {
893 tp = halve(t, tn);
894 if (!tp) {
895#ifdef CONFIG_IP_FIB_TRIE_STATS
896 this_cpu_inc(stats->resize_node_skipped);
897#endif
898 break;
899 }
900
901 max_work--;
902 tn = get_child(tp, cindex);
903 }
904
905 /* Only one child remains */
906 if (should_collapse(tn))
907 return collapse(t, tn);
908
909 /* update parent in case halve failed */
910 return node_parent(tn);
911}
912
913static void node_pull_suffix(struct key_vector *tn, unsigned char slen)
914{
915 unsigned char node_slen = tn->slen;
916
917 while ((node_slen > tn->pos) && (node_slen > slen)) {
918 slen = update_suffix(tn);
919 if (node_slen == slen)
920 break;
921
922 tn = node_parent(tn);
923 node_slen = tn->slen;
924 }
925}
926
927static void node_push_suffix(struct key_vector *tn, unsigned char slen)
928{
929 while (tn->slen < slen) {
930 tn->slen = slen;
931 tn = node_parent(tn);
932 }
933}
934
935/* rcu_read_lock needs to be hold by caller from readside */
936static struct key_vector *fib_find_node(struct trie *t,
937 struct key_vector **tp, u32 key)
938{
939 struct key_vector *pn, *n = t->kv;
940 unsigned long index = 0;
941
942 do {
943 pn = n;
944 n = get_child_rcu(n, index);
945
946 if (!n)
947 break;
948
949 index = get_cindex(key, n);
950
951 /* This bit of code is a bit tricky but it combines multiple
952 * checks into a single check. The prefix consists of the
953 * prefix plus zeros for the bits in the cindex. The index
954 * is the difference between the key and this value. From
955 * this we can actually derive several pieces of data.
956 * if (index >= (1ul << bits))
957 * we have a mismatch in skip bits and failed
958 * else
959 * we know the value is cindex
960 *
961 * This check is safe even if bits == KEYLENGTH due to the
962 * fact that we can only allocate a node with 32 bits if a
963 * long is greater than 32 bits.
964 */
965 if (index >= (1ul << n->bits)) {
966 n = NULL;
967 break;
968 }
969
970 /* keep searching until we find a perfect match leaf or NULL */
971 } while (IS_TNODE(n));
972
973 *tp = pn;
974
975 return n;
976}
977
978/* Return the first fib alias matching DSCP with
979 * priority less than or equal to PRIO.
980 * If 'find_first' is set, return the first matching
981 * fib alias, regardless of DSCP and priority.
982 */
983static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
984 dscp_t dscp, u32 prio, u32 tb_id,
985 bool find_first)
986{
987 struct fib_alias *fa;
988
989 if (!fah)
990 return NULL;
991
992 hlist_for_each_entry(fa, fah, fa_list) {
993 /* Avoid Sparse warning when using dscp_t in inequalities */
994 u8 __fa_dscp = inet_dscp_to_dsfield(fa->fa_dscp);
995 u8 __dscp = inet_dscp_to_dsfield(dscp);
996
997 if (fa->fa_slen < slen)
998 continue;
999 if (fa->fa_slen != slen)
1000 break;
1001 if (fa->tb_id > tb_id)
1002 continue;
1003 if (fa->tb_id != tb_id)
1004 break;
1005 if (find_first)
1006 return fa;
1007 if (__fa_dscp > __dscp)
1008 continue;
1009 if (fa->fa_info->fib_priority >= prio || __fa_dscp < __dscp)
1010 return fa;
1011 }
1012
1013 return NULL;
1014}
1015
1016static struct fib_alias *
1017fib_find_matching_alias(struct net *net, const struct fib_rt_info *fri)
1018{
1019 u8 slen = KEYLENGTH - fri->dst_len;
1020 struct key_vector *l, *tp;
1021 struct fib_table *tb;
1022 struct fib_alias *fa;
1023 struct trie *t;
1024
1025 tb = fib_get_table(net, fri->tb_id);
1026 if (!tb)
1027 return NULL;
1028
1029 t = (struct trie *)tb->tb_data;
1030 l = fib_find_node(t, &tp, be32_to_cpu(fri->dst));
1031 if (!l)
1032 return NULL;
1033
1034 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1035 if (fa->fa_slen == slen && fa->tb_id == fri->tb_id &&
1036 fa->fa_dscp == fri->dscp && fa->fa_info == fri->fi &&
1037 fa->fa_type == fri->type)
1038 return fa;
1039 }
1040
1041 return NULL;
1042}
1043
1044void fib_alias_hw_flags_set(struct net *net, const struct fib_rt_info *fri)
1045{
1046 u8 fib_notify_on_flag_change;
1047 struct fib_alias *fa_match;
1048 struct sk_buff *skb;
1049 int err;
1050
1051 rcu_read_lock();
1052
1053 fa_match = fib_find_matching_alias(net, fri);
1054 if (!fa_match)
1055 goto out;
1056
1057 /* These are paired with the WRITE_ONCE() happening in this function.
1058 * The reason is that we are only protected by RCU at this point.
1059 */
1060 if (READ_ONCE(fa_match->offload) == fri->offload &&
1061 READ_ONCE(fa_match->trap) == fri->trap &&
1062 READ_ONCE(fa_match->offload_failed) == fri->offload_failed)
1063 goto out;
1064
1065 WRITE_ONCE(fa_match->offload, fri->offload);
1066 WRITE_ONCE(fa_match->trap, fri->trap);
1067
1068 fib_notify_on_flag_change = READ_ONCE(net->ipv4.sysctl_fib_notify_on_flag_change);
1069
1070 /* 2 means send notifications only if offload_failed was changed. */
1071 if (fib_notify_on_flag_change == 2 &&
1072 READ_ONCE(fa_match->offload_failed) == fri->offload_failed)
1073 goto out;
1074
1075 WRITE_ONCE(fa_match->offload_failed, fri->offload_failed);
1076
1077 if (!fib_notify_on_flag_change)
1078 goto out;
1079
1080 skb = nlmsg_new(fib_nlmsg_size(fa_match->fa_info), GFP_ATOMIC);
1081 if (!skb) {
1082 err = -ENOBUFS;
1083 goto errout;
1084 }
1085
1086 err = fib_dump_info(skb, 0, 0, RTM_NEWROUTE, fri, 0);
1087 if (err < 0) {
1088 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
1089 WARN_ON(err == -EMSGSIZE);
1090 kfree_skb(skb);
1091 goto errout;
1092 }
1093
1094 rtnl_notify(skb, net, 0, RTNLGRP_IPV4_ROUTE, NULL, GFP_ATOMIC);
1095 goto out;
1096
1097errout:
1098 rtnl_set_sk_err(net, RTNLGRP_IPV4_ROUTE, err);
1099out:
1100 rcu_read_unlock();
1101}
1102EXPORT_SYMBOL_GPL(fib_alias_hw_flags_set);
1103
1104static void trie_rebalance(struct trie *t, struct key_vector *tn)
1105{
1106 while (!IS_TRIE(tn))
1107 tn = resize(t, tn);
1108}
1109
1110static int fib_insert_node(struct trie *t, struct key_vector *tp,
1111 struct fib_alias *new, t_key key)
1112{
1113 struct key_vector *n, *l;
1114
1115 l = leaf_new(key, new);
1116 if (!l)
1117 goto noleaf;
1118
1119 /* retrieve child from parent node */
1120 n = get_child(tp, get_index(key, tp));
1121
1122 /* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1123 *
1124 * Add a new tnode here
1125 * first tnode need some special handling
1126 * leaves us in position for handling as case 3
1127 */
1128 if (n) {
1129 struct key_vector *tn;
1130
1131 tn = tnode_new(key, __fls(key ^ n->key), 1);
1132 if (!tn)
1133 goto notnode;
1134
1135 /* initialize routes out of node */
1136 NODE_INIT_PARENT(tn, tp);
1137 put_child(tn, get_index(key, tn) ^ 1, n);
1138
1139 /* start adding routes into the node */
1140 put_child_root(tp, key, tn);
1141 node_set_parent(n, tn);
1142
1143 /* parent now has a NULL spot where the leaf can go */
1144 tp = tn;
1145 }
1146
1147 /* Case 3: n is NULL, and will just insert a new leaf */
1148 node_push_suffix(tp, new->fa_slen);
1149 NODE_INIT_PARENT(l, tp);
1150 put_child_root(tp, key, l);
1151 trie_rebalance(t, tp);
1152
1153 return 0;
1154notnode:
1155 node_free(l);
1156noleaf:
1157 return -ENOMEM;
1158}
1159
1160static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1161 struct key_vector *l, struct fib_alias *new,
1162 struct fib_alias *fa, t_key key)
1163{
1164 if (!l)
1165 return fib_insert_node(t, tp, new, key);
1166
1167 if (fa) {
1168 hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1169 } else {
1170 struct fib_alias *last;
1171
1172 hlist_for_each_entry(last, &l->leaf, fa_list) {
1173 if (new->fa_slen < last->fa_slen)
1174 break;
1175 if ((new->fa_slen == last->fa_slen) &&
1176 (new->tb_id > last->tb_id))
1177 break;
1178 fa = last;
1179 }
1180
1181 if (fa)
1182 hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1183 else
1184 hlist_add_head_rcu(&new->fa_list, &l->leaf);
1185 }
1186
1187 /* if we added to the tail node then we need to update slen */
1188 if (l->slen < new->fa_slen) {
1189 l->slen = new->fa_slen;
1190 node_push_suffix(tp, new->fa_slen);
1191 }
1192
1193 return 0;
1194}
1195
1196static bool fib_valid_key_len(u32 key, u8 plen, struct netlink_ext_ack *extack)
1197{
1198 if (plen > KEYLENGTH) {
1199 NL_SET_ERR_MSG(extack, "Invalid prefix length");
1200 return false;
1201 }
1202
1203 if ((plen < KEYLENGTH) && (key << plen)) {
1204 NL_SET_ERR_MSG(extack,
1205 "Invalid prefix for given prefix length");
1206 return false;
1207 }
1208
1209 return true;
1210}
1211
1212static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1213 struct key_vector *l, struct fib_alias *old);
1214
1215/* Caller must hold RTNL. */
1216int fib_table_insert(struct net *net, struct fib_table *tb,
1217 struct fib_config *cfg, struct netlink_ext_ack *extack)
1218{
1219 struct trie *t = (struct trie *)tb->tb_data;
1220 struct fib_alias *fa, *new_fa;
1221 struct key_vector *l, *tp;
1222 u16 nlflags = NLM_F_EXCL;
1223 struct fib_info *fi;
1224 u8 plen = cfg->fc_dst_len;
1225 u8 slen = KEYLENGTH - plen;
1226 dscp_t dscp;
1227 u32 key;
1228 int err;
1229
1230 key = ntohl(cfg->fc_dst);
1231
1232 if (!fib_valid_key_len(key, plen, extack))
1233 return -EINVAL;
1234
1235 pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1236
1237 fi = fib_create_info(cfg, extack);
1238 if (IS_ERR(fi)) {
1239 err = PTR_ERR(fi);
1240 goto err;
1241 }
1242
1243 dscp = cfg->fc_dscp;
1244 l = fib_find_node(t, &tp, key);
1245 fa = l ? fib_find_alias(&l->leaf, slen, dscp, fi->fib_priority,
1246 tb->tb_id, false) : NULL;
1247
1248 /* Now fa, if non-NULL, points to the first fib alias
1249 * with the same keys [prefix,dscp,priority], if such key already
1250 * exists or to the node before which we will insert new one.
1251 *
1252 * If fa is NULL, we will need to allocate a new one and
1253 * insert to the tail of the section matching the suffix length
1254 * of the new alias.
1255 */
1256
1257 if (fa && fa->fa_dscp == dscp &&
1258 fa->fa_info->fib_priority == fi->fib_priority) {
1259 struct fib_alias *fa_first, *fa_match;
1260
1261 err = -EEXIST;
1262 if (cfg->fc_nlflags & NLM_F_EXCL)
1263 goto out;
1264
1265 nlflags &= ~NLM_F_EXCL;
1266
1267 /* We have 2 goals:
1268 * 1. Find exact match for type, scope, fib_info to avoid
1269 * duplicate routes
1270 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1271 */
1272 fa_match = NULL;
1273 fa_first = fa;
1274 hlist_for_each_entry_from(fa, fa_list) {
1275 if ((fa->fa_slen != slen) ||
1276 (fa->tb_id != tb->tb_id) ||
1277 (fa->fa_dscp != dscp))
1278 break;
1279 if (fa->fa_info->fib_priority != fi->fib_priority)
1280 break;
1281 if (fa->fa_type == cfg->fc_type &&
1282 fa->fa_info == fi) {
1283 fa_match = fa;
1284 break;
1285 }
1286 }
1287
1288 if (cfg->fc_nlflags & NLM_F_REPLACE) {
1289 struct fib_info *fi_drop;
1290 u8 state;
1291
1292 nlflags |= NLM_F_REPLACE;
1293 fa = fa_first;
1294 if (fa_match) {
1295 if (fa == fa_match)
1296 err = 0;
1297 goto out;
1298 }
1299 err = -ENOBUFS;
1300 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1301 if (!new_fa)
1302 goto out;
1303
1304 fi_drop = fa->fa_info;
1305 new_fa->fa_dscp = fa->fa_dscp;
1306 new_fa->fa_info = fi;
1307 new_fa->fa_type = cfg->fc_type;
1308 state = fa->fa_state;
1309 new_fa->fa_state = state & ~FA_S_ACCESSED;
1310 new_fa->fa_slen = fa->fa_slen;
1311 new_fa->tb_id = tb->tb_id;
1312 new_fa->fa_default = -1;
1313 new_fa->offload = 0;
1314 new_fa->trap = 0;
1315 new_fa->offload_failed = 0;
1316
1317 hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1318
1319 if (fib_find_alias(&l->leaf, fa->fa_slen, 0, 0,
1320 tb->tb_id, true) == new_fa) {
1321 enum fib_event_type fib_event;
1322
1323 fib_event = FIB_EVENT_ENTRY_REPLACE;
1324 err = call_fib_entry_notifiers(net, fib_event,
1325 key, plen,
1326 new_fa, extack);
1327 if (err) {
1328 hlist_replace_rcu(&new_fa->fa_list,
1329 &fa->fa_list);
1330 goto out_free_new_fa;
1331 }
1332 }
1333
1334 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1335 tb->tb_id, &cfg->fc_nlinfo, nlflags);
1336
1337 alias_free_mem_rcu(fa);
1338
1339 fib_release_info(fi_drop);
1340 if (state & FA_S_ACCESSED)
1341 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1342
1343 goto succeeded;
1344 }
1345 /* Error if we find a perfect match which
1346 * uses the same scope, type, and nexthop
1347 * information.
1348 */
1349 if (fa_match)
1350 goto out;
1351
1352 if (cfg->fc_nlflags & NLM_F_APPEND)
1353 nlflags |= NLM_F_APPEND;
1354 else
1355 fa = fa_first;
1356 }
1357 err = -ENOENT;
1358 if (!(cfg->fc_nlflags & NLM_F_CREATE))
1359 goto out;
1360
1361 nlflags |= NLM_F_CREATE;
1362 err = -ENOBUFS;
1363 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1364 if (!new_fa)
1365 goto out;
1366
1367 new_fa->fa_info = fi;
1368 new_fa->fa_dscp = dscp;
1369 new_fa->fa_type = cfg->fc_type;
1370 new_fa->fa_state = 0;
1371 new_fa->fa_slen = slen;
1372 new_fa->tb_id = tb->tb_id;
1373 new_fa->fa_default = -1;
1374 new_fa->offload = 0;
1375 new_fa->trap = 0;
1376 new_fa->offload_failed = 0;
1377
1378 /* Insert new entry to the list. */
1379 err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1380 if (err)
1381 goto out_free_new_fa;
1382
1383 /* The alias was already inserted, so the node must exist. */
1384 l = l ? l : fib_find_node(t, &tp, key);
1385 if (WARN_ON_ONCE(!l)) {
1386 err = -ENOENT;
1387 goto out_free_new_fa;
1388 }
1389
1390 if (fib_find_alias(&l->leaf, new_fa->fa_slen, 0, 0, tb->tb_id, true) ==
1391 new_fa) {
1392 enum fib_event_type fib_event;
1393
1394 fib_event = FIB_EVENT_ENTRY_REPLACE;
1395 err = call_fib_entry_notifiers(net, fib_event, key, plen,
1396 new_fa, extack);
1397 if (err)
1398 goto out_remove_new_fa;
1399 }
1400
1401 if (!plen)
1402 tb->tb_num_default++;
1403
1404 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1405 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1406 &cfg->fc_nlinfo, nlflags);
1407succeeded:
1408 return 0;
1409
1410out_remove_new_fa:
1411 fib_remove_alias(t, tp, l, new_fa);
1412out_free_new_fa:
1413 kmem_cache_free(fn_alias_kmem, new_fa);
1414out:
1415 fib_release_info(fi);
1416err:
1417 return err;
1418}
1419
1420static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1421{
1422 t_key prefix = n->key;
1423
1424 return (key ^ prefix) & (prefix | -prefix);
1425}
1426
1427bool fib_lookup_good_nhc(const struct fib_nh_common *nhc, int fib_flags,
1428 const struct flowi4 *flp)
1429{
1430 if (nhc->nhc_flags & RTNH_F_DEAD)
1431 return false;
1432
1433 if (ip_ignore_linkdown(nhc->nhc_dev) &&
1434 nhc->nhc_flags & RTNH_F_LINKDOWN &&
1435 !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1436 return false;
1437
1438 if (flp->flowi4_oif && flp->flowi4_oif != nhc->nhc_oif)
1439 return false;
1440
1441 return true;
1442}
1443
1444/* should be called with rcu_read_lock */
1445int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1446 struct fib_result *res, int fib_flags)
1447{
1448 struct trie *t = (struct trie *) tb->tb_data;
1449#ifdef CONFIG_IP_FIB_TRIE_STATS
1450 struct trie_use_stats __percpu *stats = t->stats;
1451#endif
1452 const t_key key = ntohl(flp->daddr);
1453 struct key_vector *n, *pn;
1454 struct fib_alias *fa;
1455 unsigned long index;
1456 t_key cindex;
1457
1458 pn = t->kv;
1459 cindex = 0;
1460
1461 n = get_child_rcu(pn, cindex);
1462 if (!n) {
1463 trace_fib_table_lookup(tb->tb_id, flp, NULL, -EAGAIN);
1464 return -EAGAIN;
1465 }
1466
1467#ifdef CONFIG_IP_FIB_TRIE_STATS
1468 this_cpu_inc(stats->gets);
1469#endif
1470
1471 /* Step 1: Travel to the longest prefix match in the trie */
1472 for (;;) {
1473 index = get_cindex(key, n);
1474
1475 /* This bit of code is a bit tricky but it combines multiple
1476 * checks into a single check. The prefix consists of the
1477 * prefix plus zeros for the "bits" in the prefix. The index
1478 * is the difference between the key and this value. From
1479 * this we can actually derive several pieces of data.
1480 * if (index >= (1ul << bits))
1481 * we have a mismatch in skip bits and failed
1482 * else
1483 * we know the value is cindex
1484 *
1485 * This check is safe even if bits == KEYLENGTH due to the
1486 * fact that we can only allocate a node with 32 bits if a
1487 * long is greater than 32 bits.
1488 */
1489 if (index >= (1ul << n->bits))
1490 break;
1491
1492 /* we have found a leaf. Prefixes have already been compared */
1493 if (IS_LEAF(n))
1494 goto found;
1495
1496 /* only record pn and cindex if we are going to be chopping
1497 * bits later. Otherwise we are just wasting cycles.
1498 */
1499 if (n->slen > n->pos) {
1500 pn = n;
1501 cindex = index;
1502 }
1503
1504 n = get_child_rcu(n, index);
1505 if (unlikely(!n))
1506 goto backtrace;
1507 }
1508
1509 /* Step 2: Sort out leaves and begin backtracing for longest prefix */
1510 for (;;) {
1511 /* record the pointer where our next node pointer is stored */
1512 struct key_vector __rcu **cptr = n->tnode;
1513
1514 /* This test verifies that none of the bits that differ
1515 * between the key and the prefix exist in the region of
1516 * the lsb and higher in the prefix.
1517 */
1518 if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1519 goto backtrace;
1520
1521 /* exit out and process leaf */
1522 if (unlikely(IS_LEAF(n)))
1523 break;
1524
1525 /* Don't bother recording parent info. Since we are in
1526 * prefix match mode we will have to come back to wherever
1527 * we started this traversal anyway
1528 */
1529
1530 while ((n = rcu_dereference(*cptr)) == NULL) {
1531backtrace:
1532#ifdef CONFIG_IP_FIB_TRIE_STATS
1533 if (!n)
1534 this_cpu_inc(stats->null_node_hit);
1535#endif
1536 /* If we are at cindex 0 there are no more bits for
1537 * us to strip at this level so we must ascend back
1538 * up one level to see if there are any more bits to
1539 * be stripped there.
1540 */
1541 while (!cindex) {
1542 t_key pkey = pn->key;
1543
1544 /* If we don't have a parent then there is
1545 * nothing for us to do as we do not have any
1546 * further nodes to parse.
1547 */
1548 if (IS_TRIE(pn)) {
1549 trace_fib_table_lookup(tb->tb_id, flp,
1550 NULL, -EAGAIN);
1551 return -EAGAIN;
1552 }
1553#ifdef CONFIG_IP_FIB_TRIE_STATS
1554 this_cpu_inc(stats->backtrack);
1555#endif
1556 /* Get Child's index */
1557 pn = node_parent_rcu(pn);
1558 cindex = get_index(pkey, pn);
1559 }
1560
1561 /* strip the least significant bit from the cindex */
1562 cindex &= cindex - 1;
1563
1564 /* grab pointer for next child node */
1565 cptr = &pn->tnode[cindex];
1566 }
1567 }
1568
1569found:
1570 /* this line carries forward the xor from earlier in the function */
1571 index = key ^ n->key;
1572
1573 /* Step 3: Process the leaf, if that fails fall back to backtracing */
1574 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1575 struct fib_info *fi = fa->fa_info;
1576 struct fib_nh_common *nhc;
1577 int nhsel, err;
1578
1579 if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1580 if (index >= (1ul << fa->fa_slen))
1581 continue;
1582 }
1583 if (fa->fa_dscp &&
1584 inet_dscp_to_dsfield(fa->fa_dscp) != flp->flowi4_tos)
1585 continue;
1586 /* Paired with WRITE_ONCE() in fib_release_info() */
1587 if (READ_ONCE(fi->fib_dead))
1588 continue;
1589 if (fa->fa_info->fib_scope < flp->flowi4_scope)
1590 continue;
1591 fib_alias_accessed(fa);
1592 err = fib_props[fa->fa_type].error;
1593 if (unlikely(err < 0)) {
1594out_reject:
1595#ifdef CONFIG_IP_FIB_TRIE_STATS
1596 this_cpu_inc(stats->semantic_match_passed);
1597#endif
1598 trace_fib_table_lookup(tb->tb_id, flp, NULL, err);
1599 return err;
1600 }
1601 if (fi->fib_flags & RTNH_F_DEAD)
1602 continue;
1603
1604 if (unlikely(fi->nh)) {
1605 if (nexthop_is_blackhole(fi->nh)) {
1606 err = fib_props[RTN_BLACKHOLE].error;
1607 goto out_reject;
1608 }
1609
1610 nhc = nexthop_get_nhc_lookup(fi->nh, fib_flags, flp,
1611 &nhsel);
1612 if (nhc)
1613 goto set_result;
1614 goto miss;
1615 }
1616
1617 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) {
1618 nhc = fib_info_nhc(fi, nhsel);
1619
1620 if (!fib_lookup_good_nhc(nhc, fib_flags, flp))
1621 continue;
1622set_result:
1623 if (!(fib_flags & FIB_LOOKUP_NOREF))
1624 refcount_inc(&fi->fib_clntref);
1625
1626 res->prefix = htonl(n->key);
1627 res->prefixlen = KEYLENGTH - fa->fa_slen;
1628 res->nh_sel = nhsel;
1629 res->nhc = nhc;
1630 res->type = fa->fa_type;
1631 res->scope = fi->fib_scope;
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 }
1643miss:
1644#ifdef CONFIG_IP_FIB_TRIE_STATS
1645 this_cpu_inc(stats->semantic_match_miss);
1646#endif
1647 goto backtrace;
1648}
1649EXPORT_SYMBOL_GPL(fib_table_lookup);
1650
1651static 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
1682static 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. */
1713int 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 */
1785static 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 */
1836found:
1837 /* if we are at the limit for keys just return NULL for the tnode */
1838 *tn = pn;
1839 return n;
1840}
1841
1842static 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
1900struct 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;
1952out:
1953 fib_trie_free(local_tb);
1954
1955 return NULL;
1956}
1957
1958/* Caller must hold RTNL */
1959void 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. */
2028int 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 */
2117static 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
2165void 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
2179static 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
2212static 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
2233int 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
2252static 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
2264void fib_free_table(struct fib_table *tb)
2265{
2266 call_rcu(&tb->rcu, __trie_free_rcu);
2267}
2268
2269static 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
2342next:
2343 i++;
2344 }
2345
2346 cb->args[4] = i;
2347 return skb->len;
2348
2349stop:
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 */
2356int 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
2400void __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
2411struct 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 */
2447struct 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
2455static 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
2498static 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
2524static 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 */
2556static 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
2597static 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
2626static 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
2637static 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
2674static 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
2701static 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
2708static 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
2743found:
2744 iter->tb = tb;
2745 return n;
2746}
2747
2748static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2749 __releases(RCU)
2750{
2751 rcu_read_unlock();
2752}
2753
2754static void seq_indent(struct seq_file *seq, int n)
2755{
2756 while (n-- > 0)
2757 seq_puts(seq, " ");
2758}
2759
2760static 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
2774static 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
2789static 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 */
2798static 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
2841static 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
2848struct 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
2856static 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
2890static 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
2916static 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
2938static void fib_route_seq_stop(struct seq_file *seq, void *v)
2939 __releases(RCU)
2940{
2941 rcu_read_unlock();
2942}
2943
2944static 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 */
2968static 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
3030static 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
3037int __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
3053out3:
3054 remove_proc_entry("fib_triestat", net->proc_net);
3055out2:
3056 remove_proc_entry("fib_trie", net->proc_net);
3057out1:
3058 return -ENOMEM;
3059}
3060
3061void __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 */