ipv4: fix debug info in tnode_new
[deliverable/linux.git] / net / ipv4 / fib_trie.c
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
89 typedef 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
99 struct rt_trie_node {
100 unsigned long parent;
101 t_key key;
102 };
103
104 struct leaf {
105 unsigned long parent;
106 t_key key;
107 struct hlist_head list;
108 struct rcu_head rcu;
109 };
110
111 struct 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
119 struct 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
135 struct 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
145 struct 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
155 struct 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
162 static void put_child(struct trie *t, struct tnode *tn, int i, struct rt_trie_node *n);
163 static void tnode_put_child_reorg(struct tnode *tn, int i, struct rt_trie_node *n,
164 int wasfull);
165 static struct rt_trie_node *resize(struct trie *t, struct tnode *tn);
166 static struct tnode *inflate(struct trie *t, struct tnode *tn);
167 static struct tnode *halve(struct trie *t, struct tnode *tn);
168 /* tnodes to free after resize(); protected by RTNL */
169 static struct tnode *tnode_free_head;
170 static 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 */
177 static const int sync_pages = 128;
178
179 static struct kmem_cache *fn_alias_kmem __read_mostly;
180 static struct kmem_cache *trie_leaf_kmem __read_mostly;
181
182 /*
183 * caller must hold RTNL
184 */
185 static 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 */
197 static 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 */
210 static 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 */
219 static 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 */
229 static 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
236 static inline int tnode_child_length(const struct tnode *tn)
237 {
238 return 1 << tn->bits;
239 }
240
241 static inline t_key mask_pfx(t_key k, unsigned int l)
242 {
243 return (l == 0) ? 0 : k >> (KEYLENGTH-l) << (KEYLENGTH-l);
244 }
245
246 static 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
254 static inline int tkey_equals(t_key a, t_key b)
255 {
256 return a == b;
257 }
258
259 static 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
267 static 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
342 static inline void check_tnode(const struct tnode *tn)
343 {
344 WARN_ON(tn && tn->pos+tn->bits > 32);
345 }
346
347 static const int halve_threshold = 25;
348 static const int inflate_threshold = 50;
349 static const int halve_threshold_root = 15;
350 static const int inflate_threshold_root = 30;
351
352 static 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
358 static inline void alias_free_mem_rcu(struct fib_alias *fa)
359 {
360 call_rcu(&fa->rcu, __alias_free_mem);
361 }
362
363 static 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
369 static inline void free_leaf(struct leaf *l)
370 {
371 call_rcu_bh(&l->rcu, __leaf_free_rcu);
372 }
373
374 static inline void free_leaf_info(struct leaf_info *leaf)
375 {
376 kfree_rcu(leaf, rcu);
377 }
378
379 static 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
387 static void __tnode_vfree(struct work_struct *arg)
388 {
389 struct tnode *tn = container_of(arg, struct tnode, work);
390 vfree(tn);
391 }
392
393 static 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
407 static 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
415 static 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
424 static 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
440 static 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
450 static 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
461 static 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
485 static 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
493 static 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
504 static 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
535 static 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) {
681 one_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
700 static 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
713 static 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;
841 nomem:
842 tnode_clean_free(tn);
843 return ERR_PTR(-ENOMEM);
844 }
845
846 static 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;
913 nomem:
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
921 static 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
934 static 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
944 static 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
967 static struct leaf *
968 fib_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
999 static 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, tn);
1011
1012 tnode_put_child_reorg(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, 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
1035 static 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, 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, tp, cindex, (struct rt_trie_node *)tn);
1164 } else {
1165 rcu_assign_pointer(t->trie, (struct rt_trie_node *)tn);
1166 tp = tn;
1167 }
1168 }
1169
1170 if (tp && tp->pos + tp->bits > 32)
1171 pr_warn("fib_trie tp=%p pos=%d, bits=%d, key=%0x plen=%d\n",
1172 tp, tp->pos, tp->bits, key, plen);
1173
1174 /* Rebalance the trie */
1175
1176 trie_rebalance(t, tp);
1177 done:
1178 return fa_head;
1179 }
1180
1181 /*
1182 * Caller must hold RTNL.
1183 */
1184 int fib_table_insert(struct fib_table *tb, struct fib_config *cfg)
1185 {
1186 struct trie *t = (struct trie *) tb->tb_data;
1187 struct fib_alias *fa, *new_fa;
1188 struct list_head *fa_head = NULL;
1189 struct fib_info *fi;
1190 int plen = cfg->fc_dst_len;
1191 u8 tos = cfg->fc_tos;
1192 u32 key, mask;
1193 int err;
1194 struct leaf *l;
1195
1196 if (plen > 32)
1197 return -EINVAL;
1198
1199 key = ntohl(cfg->fc_dst);
1200
1201 pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1202
1203 mask = ntohl(inet_make_mask(plen));
1204
1205 if (key & ~mask)
1206 return -EINVAL;
1207
1208 key = key & mask;
1209
1210 fi = fib_create_info(cfg);
1211 if (IS_ERR(fi)) {
1212 err = PTR_ERR(fi);
1213 goto err;
1214 }
1215
1216 l = fib_find_node(t, key);
1217 fa = NULL;
1218
1219 if (l) {
1220 fa_head = get_fa_head(l, plen);
1221 fa = fib_find_alias(fa_head, tos, fi->fib_priority);
1222 }
1223
1224 /* Now fa, if non-NULL, points to the first fib alias
1225 * with the same keys [prefix,tos,priority], if such key already
1226 * exists or to the node before which we will insert new one.
1227 *
1228 * If fa is NULL, we will need to allocate a new one and
1229 * insert to the head of f.
1230 *
1231 * If f is NULL, no fib node matched the destination key
1232 * and we need to allocate a new one of those as well.
1233 */
1234
1235 if (fa && fa->fa_tos == tos &&
1236 fa->fa_info->fib_priority == fi->fib_priority) {
1237 struct fib_alias *fa_first, *fa_match;
1238
1239 err = -EEXIST;
1240 if (cfg->fc_nlflags & NLM_F_EXCL)
1241 goto out;
1242
1243 /* We have 2 goals:
1244 * 1. Find exact match for type, scope, fib_info to avoid
1245 * duplicate routes
1246 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1247 */
1248 fa_match = NULL;
1249 fa_first = fa;
1250 fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
1251 list_for_each_entry_continue(fa, fa_head, fa_list) {
1252 if (fa->fa_tos != tos)
1253 break;
1254 if (fa->fa_info->fib_priority != fi->fib_priority)
1255 break;
1256 if (fa->fa_type == cfg->fc_type &&
1257 fa->fa_info == fi) {
1258 fa_match = fa;
1259 break;
1260 }
1261 }
1262
1263 if (cfg->fc_nlflags & NLM_F_REPLACE) {
1264 struct fib_info *fi_drop;
1265 u8 state;
1266
1267 fa = fa_first;
1268 if (fa_match) {
1269 if (fa == fa_match)
1270 err = 0;
1271 goto out;
1272 }
1273 err = -ENOBUFS;
1274 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1275 if (new_fa == NULL)
1276 goto out;
1277
1278 fi_drop = fa->fa_info;
1279 new_fa->fa_tos = fa->fa_tos;
1280 new_fa->fa_info = fi;
1281 new_fa->fa_type = cfg->fc_type;
1282 state = fa->fa_state;
1283 new_fa->fa_state = state & ~FA_S_ACCESSED;
1284
1285 list_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1286 alias_free_mem_rcu(fa);
1287
1288 fib_release_info(fi_drop);
1289 if (state & FA_S_ACCESSED)
1290 rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1291 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1292 tb->tb_id, &cfg->fc_nlinfo, NLM_F_REPLACE);
1293
1294 goto succeeded;
1295 }
1296 /* Error if we find a perfect match which
1297 * uses the same scope, type, and nexthop
1298 * information.
1299 */
1300 if (fa_match)
1301 goto out;
1302
1303 if (!(cfg->fc_nlflags & NLM_F_APPEND))
1304 fa = fa_first;
1305 }
1306 err = -ENOENT;
1307 if (!(cfg->fc_nlflags & NLM_F_CREATE))
1308 goto out;
1309
1310 err = -ENOBUFS;
1311 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1312 if (new_fa == NULL)
1313 goto out;
1314
1315 new_fa->fa_info = fi;
1316 new_fa->fa_tos = tos;
1317 new_fa->fa_type = cfg->fc_type;
1318 new_fa->fa_state = 0;
1319 /*
1320 * Insert new entry to the list.
1321 */
1322
1323 if (!fa_head) {
1324 fa_head = fib_insert_node(t, key, plen);
1325 if (unlikely(!fa_head)) {
1326 err = -ENOMEM;
1327 goto out_free_new_fa;
1328 }
1329 }
1330
1331 if (!plen)
1332 tb->tb_num_default++;
1333
1334 list_add_tail_rcu(&new_fa->fa_list,
1335 (fa ? &fa->fa_list : fa_head));
1336
1337 rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1338 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, tb->tb_id,
1339 &cfg->fc_nlinfo, 0);
1340 succeeded:
1341 return 0;
1342
1343 out_free_new_fa:
1344 kmem_cache_free(fn_alias_kmem, new_fa);
1345 out:
1346 fib_release_info(fi);
1347 err:
1348 return err;
1349 }
1350
1351 /* should be called with rcu_read_lock */
1352 static int check_leaf(struct fib_table *tb, struct trie *t, struct leaf *l,
1353 t_key key, const struct flowi4 *flp,
1354 struct fib_result *res, int fib_flags)
1355 {
1356 struct leaf_info *li;
1357 struct hlist_head *hhead = &l->list;
1358 struct hlist_node *node;
1359
1360 hlist_for_each_entry_rcu(li, node, hhead, hlist) {
1361 struct fib_alias *fa;
1362
1363 if (l->key != (key & li->mask_plen))
1364 continue;
1365
1366 list_for_each_entry_rcu(fa, &li->falh, fa_list) {
1367 struct fib_info *fi = fa->fa_info;
1368 int nhsel, err;
1369
1370 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1371 continue;
1372 if (fi->fib_dead)
1373 continue;
1374 if (fa->fa_info->fib_scope < flp->flowi4_scope)
1375 continue;
1376 fib_alias_accessed(fa);
1377 err = fib_props[fa->fa_type].error;
1378 if (err) {
1379 #ifdef CONFIG_IP_FIB_TRIE_STATS
1380 t->stats.semantic_match_passed++;
1381 #endif
1382 return err;
1383 }
1384 if (fi->fib_flags & RTNH_F_DEAD)
1385 continue;
1386 for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
1387 const struct fib_nh *nh = &fi->fib_nh[nhsel];
1388
1389 if (nh->nh_flags & RTNH_F_DEAD)
1390 continue;
1391 if (flp->flowi4_oif && flp->flowi4_oif != nh->nh_oif)
1392 continue;
1393
1394 #ifdef CONFIG_IP_FIB_TRIE_STATS
1395 t->stats.semantic_match_passed++;
1396 #endif
1397 res->prefixlen = li->plen;
1398 res->nh_sel = nhsel;
1399 res->type = fa->fa_type;
1400 res->scope = fa->fa_info->fib_scope;
1401 res->fi = fi;
1402 res->table = tb;
1403 res->fa_head = &li->falh;
1404 if (!(fib_flags & FIB_LOOKUP_NOREF))
1405 atomic_inc(&fi->fib_clntref);
1406 return 0;
1407 }
1408 }
1409
1410 #ifdef CONFIG_IP_FIB_TRIE_STATS
1411 t->stats.semantic_match_miss++;
1412 #endif
1413 }
1414
1415 return 1;
1416 }
1417
1418 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1419 struct fib_result *res, int fib_flags)
1420 {
1421 struct trie *t = (struct trie *) tb->tb_data;
1422 int ret;
1423 struct rt_trie_node *n;
1424 struct tnode *pn;
1425 unsigned int pos, bits;
1426 t_key key = ntohl(flp->daddr);
1427 unsigned int chopped_off;
1428 t_key cindex = 0;
1429 unsigned int current_prefix_length = KEYLENGTH;
1430 struct tnode *cn;
1431 t_key pref_mismatch;
1432
1433 rcu_read_lock();
1434
1435 n = rcu_dereference(t->trie);
1436 if (!n)
1437 goto failed;
1438
1439 #ifdef CONFIG_IP_FIB_TRIE_STATS
1440 t->stats.gets++;
1441 #endif
1442
1443 /* Just a leaf? */
1444 if (IS_LEAF(n)) {
1445 ret = check_leaf(tb, t, (struct leaf *)n, key, flp, res, fib_flags);
1446 goto found;
1447 }
1448
1449 pn = (struct tnode *) n;
1450 chopped_off = 0;
1451
1452 while (pn) {
1453 pos = pn->pos;
1454 bits = pn->bits;
1455
1456 if (!chopped_off)
1457 cindex = tkey_extract_bits(mask_pfx(key, current_prefix_length),
1458 pos, bits);
1459
1460 n = tnode_get_child_rcu(pn, cindex);
1461
1462 if (n == NULL) {
1463 #ifdef CONFIG_IP_FIB_TRIE_STATS
1464 t->stats.null_node_hit++;
1465 #endif
1466 goto backtrace;
1467 }
1468
1469 if (IS_LEAF(n)) {
1470 ret = check_leaf(tb, t, (struct leaf *)n, key, flp, res, fib_flags);
1471 if (ret > 0)
1472 goto backtrace;
1473 goto found;
1474 }
1475
1476 cn = (struct tnode *)n;
1477
1478 /*
1479 * It's a tnode, and we can do some extra checks here if we
1480 * like, to avoid descending into a dead-end branch.
1481 * This tnode is in the parent's child array at index
1482 * key[p_pos..p_pos+p_bits] but potentially with some bits
1483 * chopped off, so in reality the index may be just a
1484 * subprefix, padded with zero at the end.
1485 * We can also take a look at any skipped bits in this
1486 * tnode - everything up to p_pos is supposed to be ok,
1487 * and the non-chopped bits of the index (se previous
1488 * paragraph) are also guaranteed ok, but the rest is
1489 * considered unknown.
1490 *
1491 * The skipped bits are key[pos+bits..cn->pos].
1492 */
1493
1494 /* If current_prefix_length < pos+bits, we are already doing
1495 * actual prefix matching, which means everything from
1496 * pos+(bits-chopped_off) onward must be zero along some
1497 * branch of this subtree - otherwise there is *no* valid
1498 * prefix present. Here we can only check the skipped
1499 * bits. Remember, since we have already indexed into the
1500 * parent's child array, we know that the bits we chopped of
1501 * *are* zero.
1502 */
1503
1504 /* NOTA BENE: Checking only skipped bits
1505 for the new node here */
1506
1507 if (current_prefix_length < pos+bits) {
1508 if (tkey_extract_bits(cn->key, current_prefix_length,
1509 cn->pos - current_prefix_length)
1510 || !(cn->child[0]))
1511 goto backtrace;
1512 }
1513
1514 /*
1515 * If chopped_off=0, the index is fully validated and we
1516 * only need to look at the skipped bits for this, the new,
1517 * tnode. What we actually want to do is to find out if
1518 * these skipped bits match our key perfectly, or if we will
1519 * have to count on finding a matching prefix further down,
1520 * because if we do, we would like to have some way of
1521 * verifying the existence of such a prefix at this point.
1522 */
1523
1524 /* The only thing we can do at this point is to verify that
1525 * any such matching prefix can indeed be a prefix to our
1526 * key, and if the bits in the node we are inspecting that
1527 * do not match our key are not ZERO, this cannot be true.
1528 * Thus, find out where there is a mismatch (before cn->pos)
1529 * and verify that all the mismatching bits are zero in the
1530 * new tnode's key.
1531 */
1532
1533 /*
1534 * Note: We aren't very concerned about the piece of
1535 * the key that precede pn->pos+pn->bits, since these
1536 * have already been checked. The bits after cn->pos
1537 * aren't checked since these are by definition
1538 * "unknown" at this point. Thus, what we want to see
1539 * is if we are about to enter the "prefix matching"
1540 * state, and in that case verify that the skipped
1541 * bits that will prevail throughout this subtree are
1542 * zero, as they have to be if we are to find a
1543 * matching prefix.
1544 */
1545
1546 pref_mismatch = mask_pfx(cn->key ^ key, cn->pos);
1547
1548 /*
1549 * In short: If skipped bits in this node do not match
1550 * the search key, enter the "prefix matching"
1551 * state.directly.
1552 */
1553 if (pref_mismatch) {
1554 int mp = KEYLENGTH - fls(pref_mismatch);
1555
1556 if (tkey_extract_bits(cn->key, mp, cn->pos - mp) != 0)
1557 goto backtrace;
1558
1559 if (current_prefix_length >= cn->pos)
1560 current_prefix_length = mp;
1561 }
1562
1563 pn = (struct tnode *)n; /* Descend */
1564 chopped_off = 0;
1565 continue;
1566
1567 backtrace:
1568 chopped_off++;
1569
1570 /* As zero don't change the child key (cindex) */
1571 while ((chopped_off <= pn->bits)
1572 && !(cindex & (1<<(chopped_off-1))))
1573 chopped_off++;
1574
1575 /* Decrease current_... with bits chopped off */
1576 if (current_prefix_length > pn->pos + pn->bits - chopped_off)
1577 current_prefix_length = pn->pos + pn->bits
1578 - chopped_off;
1579
1580 /*
1581 * Either we do the actual chop off according or if we have
1582 * chopped off all bits in this tnode walk up to our parent.
1583 */
1584
1585 if (chopped_off <= pn->bits) {
1586 cindex &= ~(1 << (chopped_off-1));
1587 } else {
1588 struct tnode *parent = node_parent_rcu((struct rt_trie_node *) pn);
1589 if (!parent)
1590 goto failed;
1591
1592 /* Get Child's index */
1593 cindex = tkey_extract_bits(pn->key, parent->pos, parent->bits);
1594 pn = parent;
1595 chopped_off = 0;
1596
1597 #ifdef CONFIG_IP_FIB_TRIE_STATS
1598 t->stats.backtrack++;
1599 #endif
1600 goto backtrace;
1601 }
1602 }
1603 failed:
1604 ret = 1;
1605 found:
1606 rcu_read_unlock();
1607 return ret;
1608 }
1609 EXPORT_SYMBOL_GPL(fib_table_lookup);
1610
1611 /*
1612 * Remove the leaf and return parent.
1613 */
1614 static void trie_leaf_remove(struct trie *t, struct leaf *l)
1615 {
1616 struct tnode *tp = node_parent((struct rt_trie_node *) l);
1617
1618 pr_debug("entering trie_leaf_remove(%p)\n", l);
1619
1620 if (tp) {
1621 t_key cindex = tkey_extract_bits(l->key, tp->pos, tp->bits);
1622 put_child(t, tp, cindex, NULL);
1623 trie_rebalance(t, tp);
1624 } else
1625 RCU_INIT_POINTER(t->trie, NULL);
1626
1627 free_leaf(l);
1628 }
1629
1630 /*
1631 * Caller must hold RTNL.
1632 */
1633 int fib_table_delete(struct fib_table *tb, struct fib_config *cfg)
1634 {
1635 struct trie *t = (struct trie *) tb->tb_data;
1636 u32 key, mask;
1637 int plen = cfg->fc_dst_len;
1638 u8 tos = cfg->fc_tos;
1639 struct fib_alias *fa, *fa_to_delete;
1640 struct list_head *fa_head;
1641 struct leaf *l;
1642 struct leaf_info *li;
1643
1644 if (plen > 32)
1645 return -EINVAL;
1646
1647 key = ntohl(cfg->fc_dst);
1648 mask = ntohl(inet_make_mask(plen));
1649
1650 if (key & ~mask)
1651 return -EINVAL;
1652
1653 key = key & mask;
1654 l = fib_find_node(t, key);
1655
1656 if (!l)
1657 return -ESRCH;
1658
1659 fa_head = get_fa_head(l, plen);
1660 fa = fib_find_alias(fa_head, tos, 0);
1661
1662 if (!fa)
1663 return -ESRCH;
1664
1665 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1666
1667 fa_to_delete = NULL;
1668 fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
1669 list_for_each_entry_continue(fa, fa_head, fa_list) {
1670 struct fib_info *fi = fa->fa_info;
1671
1672 if (fa->fa_tos != tos)
1673 break;
1674
1675 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1676 (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1677 fa->fa_info->fib_scope == cfg->fc_scope) &&
1678 (!cfg->fc_prefsrc ||
1679 fi->fib_prefsrc == cfg->fc_prefsrc) &&
1680 (!cfg->fc_protocol ||
1681 fi->fib_protocol == cfg->fc_protocol) &&
1682 fib_nh_match(cfg, fi) == 0) {
1683 fa_to_delete = fa;
1684 break;
1685 }
1686 }
1687
1688 if (!fa_to_delete)
1689 return -ESRCH;
1690
1691 fa = fa_to_delete;
1692 rtmsg_fib(RTM_DELROUTE, htonl(key), fa, plen, tb->tb_id,
1693 &cfg->fc_nlinfo, 0);
1694
1695 l = fib_find_node(t, key);
1696 li = find_leaf_info(l, plen);
1697
1698 list_del_rcu(&fa->fa_list);
1699
1700 if (!plen)
1701 tb->tb_num_default--;
1702
1703 if (list_empty(fa_head)) {
1704 hlist_del_rcu(&li->hlist);
1705 free_leaf_info(li);
1706 }
1707
1708 if (hlist_empty(&l->list))
1709 trie_leaf_remove(t, l);
1710
1711 if (fa->fa_state & FA_S_ACCESSED)
1712 rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1713
1714 fib_release_info(fa->fa_info);
1715 alias_free_mem_rcu(fa);
1716 return 0;
1717 }
1718
1719 static int trie_flush_list(struct list_head *head)
1720 {
1721 struct fib_alias *fa, *fa_node;
1722 int found = 0;
1723
1724 list_for_each_entry_safe(fa, fa_node, head, fa_list) {
1725 struct fib_info *fi = fa->fa_info;
1726
1727 if (fi && (fi->fib_flags & RTNH_F_DEAD)) {
1728 list_del_rcu(&fa->fa_list);
1729 fib_release_info(fa->fa_info);
1730 alias_free_mem_rcu(fa);
1731 found++;
1732 }
1733 }
1734 return found;
1735 }
1736
1737 static int trie_flush_leaf(struct leaf *l)
1738 {
1739 int found = 0;
1740 struct hlist_head *lih = &l->list;
1741 struct hlist_node *node, *tmp;
1742 struct leaf_info *li = NULL;
1743
1744 hlist_for_each_entry_safe(li, node, tmp, lih, hlist) {
1745 found += trie_flush_list(&li->falh);
1746
1747 if (list_empty(&li->falh)) {
1748 hlist_del_rcu(&li->hlist);
1749 free_leaf_info(li);
1750 }
1751 }
1752 return found;
1753 }
1754
1755 /*
1756 * Scan for the next right leaf starting at node p->child[idx]
1757 * Since we have back pointer, no recursion necessary.
1758 */
1759 static struct leaf *leaf_walk_rcu(struct tnode *p, struct rt_trie_node *c)
1760 {
1761 do {
1762 t_key idx;
1763
1764 if (c)
1765 idx = tkey_extract_bits(c->key, p->pos, p->bits) + 1;
1766 else
1767 idx = 0;
1768
1769 while (idx < 1u << p->bits) {
1770 c = tnode_get_child_rcu(p, idx++);
1771 if (!c)
1772 continue;
1773
1774 if (IS_LEAF(c)) {
1775 prefetch(rcu_dereference_rtnl(p->child[idx]));
1776 return (struct leaf *) c;
1777 }
1778
1779 /* Rescan start scanning in new node */
1780 p = (struct tnode *) c;
1781 idx = 0;
1782 }
1783
1784 /* Node empty, walk back up to parent */
1785 c = (struct rt_trie_node *) p;
1786 } while ((p = node_parent_rcu(c)) != NULL);
1787
1788 return NULL; /* Root of trie */
1789 }
1790
1791 static struct leaf *trie_firstleaf(struct trie *t)
1792 {
1793 struct tnode *n = (struct tnode *)rcu_dereference_rtnl(t->trie);
1794
1795 if (!n)
1796 return NULL;
1797
1798 if (IS_LEAF(n)) /* trie is just a leaf */
1799 return (struct leaf *) n;
1800
1801 return leaf_walk_rcu(n, NULL);
1802 }
1803
1804 static struct leaf *trie_nextleaf(struct leaf *l)
1805 {
1806 struct rt_trie_node *c = (struct rt_trie_node *) l;
1807 struct tnode *p = node_parent_rcu(c);
1808
1809 if (!p)
1810 return NULL; /* trie with just one leaf */
1811
1812 return leaf_walk_rcu(p, c);
1813 }
1814
1815 static struct leaf *trie_leafindex(struct trie *t, int index)
1816 {
1817 struct leaf *l = trie_firstleaf(t);
1818
1819 while (l && index-- > 0)
1820 l = trie_nextleaf(l);
1821
1822 return l;
1823 }
1824
1825
1826 /*
1827 * Caller must hold RTNL.
1828 */
1829 int fib_table_flush(struct fib_table *tb)
1830 {
1831 struct trie *t = (struct trie *) tb->tb_data;
1832 struct leaf *l, *ll = NULL;
1833 int found = 0;
1834
1835 for (l = trie_firstleaf(t); l; l = trie_nextleaf(l)) {
1836 found += trie_flush_leaf(l);
1837
1838 if (ll && hlist_empty(&ll->list))
1839 trie_leaf_remove(t, ll);
1840 ll = l;
1841 }
1842
1843 if (ll && hlist_empty(&ll->list))
1844 trie_leaf_remove(t, ll);
1845
1846 pr_debug("trie_flush found=%d\n", found);
1847 return found;
1848 }
1849
1850 void fib_free_table(struct fib_table *tb)
1851 {
1852 kfree(tb);
1853 }
1854
1855 static int fn_trie_dump_fa(t_key key, int plen, struct list_head *fah,
1856 struct fib_table *tb,
1857 struct sk_buff *skb, struct netlink_callback *cb)
1858 {
1859 int i, s_i;
1860 struct fib_alias *fa;
1861 __be32 xkey = htonl(key);
1862
1863 s_i = cb->args[5];
1864 i = 0;
1865
1866 /* rcu_read_lock is hold by caller */
1867
1868 list_for_each_entry_rcu(fa, fah, fa_list) {
1869 if (i < s_i) {
1870 i++;
1871 continue;
1872 }
1873
1874 if (fib_dump_info(skb, NETLINK_CB(cb->skb).pid,
1875 cb->nlh->nlmsg_seq,
1876 RTM_NEWROUTE,
1877 tb->tb_id,
1878 fa->fa_type,
1879 xkey,
1880 plen,
1881 fa->fa_tos,
1882 fa->fa_info, NLM_F_MULTI) < 0) {
1883 cb->args[5] = i;
1884 return -1;
1885 }
1886 i++;
1887 }
1888 cb->args[5] = i;
1889 return skb->len;
1890 }
1891
1892 static int fn_trie_dump_leaf(struct leaf *l, struct fib_table *tb,
1893 struct sk_buff *skb, struct netlink_callback *cb)
1894 {
1895 struct leaf_info *li;
1896 struct hlist_node *node;
1897 int i, s_i;
1898
1899 s_i = cb->args[4];
1900 i = 0;
1901
1902 /* rcu_read_lock is hold by caller */
1903 hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
1904 if (i < s_i) {
1905 i++;
1906 continue;
1907 }
1908
1909 if (i > s_i)
1910 cb->args[5] = 0;
1911
1912 if (list_empty(&li->falh))
1913 continue;
1914
1915 if (fn_trie_dump_fa(l->key, li->plen, &li->falh, tb, skb, cb) < 0) {
1916 cb->args[4] = i;
1917 return -1;
1918 }
1919 i++;
1920 }
1921
1922 cb->args[4] = i;
1923 return skb->len;
1924 }
1925
1926 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
1927 struct netlink_callback *cb)
1928 {
1929 struct leaf *l;
1930 struct trie *t = (struct trie *) tb->tb_data;
1931 t_key key = cb->args[2];
1932 int count = cb->args[3];
1933
1934 rcu_read_lock();
1935 /* Dump starting at last key.
1936 * Note: 0.0.0.0/0 (ie default) is first key.
1937 */
1938 if (count == 0)
1939 l = trie_firstleaf(t);
1940 else {
1941 /* Normally, continue from last key, but if that is missing
1942 * fallback to using slow rescan
1943 */
1944 l = fib_find_node(t, key);
1945 if (!l)
1946 l = trie_leafindex(t, count);
1947 }
1948
1949 while (l) {
1950 cb->args[2] = l->key;
1951 if (fn_trie_dump_leaf(l, tb, skb, cb) < 0) {
1952 cb->args[3] = count;
1953 rcu_read_unlock();
1954 return -1;
1955 }
1956
1957 ++count;
1958 l = trie_nextleaf(l);
1959 memset(&cb->args[4], 0,
1960 sizeof(cb->args) - 4*sizeof(cb->args[0]));
1961 }
1962 cb->args[3] = count;
1963 rcu_read_unlock();
1964
1965 return skb->len;
1966 }
1967
1968 void __init fib_trie_init(void)
1969 {
1970 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
1971 sizeof(struct fib_alias),
1972 0, SLAB_PANIC, NULL);
1973
1974 trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
1975 max(sizeof(struct leaf),
1976 sizeof(struct leaf_info)),
1977 0, SLAB_PANIC, NULL);
1978 }
1979
1980
1981 struct fib_table *fib_trie_table(u32 id)
1982 {
1983 struct fib_table *tb;
1984 struct trie *t;
1985
1986 tb = kmalloc(sizeof(struct fib_table) + sizeof(struct trie),
1987 GFP_KERNEL);
1988 if (tb == NULL)
1989 return NULL;
1990
1991 tb->tb_id = id;
1992 tb->tb_default = -1;
1993 tb->tb_num_default = 0;
1994
1995 t = (struct trie *) tb->tb_data;
1996 memset(t, 0, sizeof(*t));
1997
1998 return tb;
1999 }
2000
2001 #ifdef CONFIG_PROC_FS
2002 /* Depth first Trie walk iterator */
2003 struct fib_trie_iter {
2004 struct seq_net_private p;
2005 struct fib_table *tb;
2006 struct tnode *tnode;
2007 unsigned int index;
2008 unsigned int depth;
2009 };
2010
2011 static struct rt_trie_node *fib_trie_get_next(struct fib_trie_iter *iter)
2012 {
2013 struct tnode *tn = iter->tnode;
2014 unsigned int cindex = iter->index;
2015 struct tnode *p;
2016
2017 /* A single entry routing table */
2018 if (!tn)
2019 return NULL;
2020
2021 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2022 iter->tnode, iter->index, iter->depth);
2023 rescan:
2024 while (cindex < (1<<tn->bits)) {
2025 struct rt_trie_node *n = tnode_get_child_rcu(tn, cindex);
2026
2027 if (n) {
2028 if (IS_LEAF(n)) {
2029 iter->tnode = tn;
2030 iter->index = cindex + 1;
2031 } else {
2032 /* push down one level */
2033 iter->tnode = (struct tnode *) n;
2034 iter->index = 0;
2035 ++iter->depth;
2036 }
2037 return n;
2038 }
2039
2040 ++cindex;
2041 }
2042
2043 /* Current node exhausted, pop back up */
2044 p = node_parent_rcu((struct rt_trie_node *)tn);
2045 if (p) {
2046 cindex = tkey_extract_bits(tn->key, p->pos, p->bits)+1;
2047 tn = p;
2048 --iter->depth;
2049 goto rescan;
2050 }
2051
2052 /* got root? */
2053 return NULL;
2054 }
2055
2056 static struct rt_trie_node *fib_trie_get_first(struct fib_trie_iter *iter,
2057 struct trie *t)
2058 {
2059 struct rt_trie_node *n;
2060
2061 if (!t)
2062 return NULL;
2063
2064 n = rcu_dereference(t->trie);
2065 if (!n)
2066 return NULL;
2067
2068 if (IS_TNODE(n)) {
2069 iter->tnode = (struct tnode *) n;
2070 iter->index = 0;
2071 iter->depth = 1;
2072 } else {
2073 iter->tnode = NULL;
2074 iter->index = 0;
2075 iter->depth = 0;
2076 }
2077
2078 return n;
2079 }
2080
2081 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2082 {
2083 struct rt_trie_node *n;
2084 struct fib_trie_iter iter;
2085
2086 memset(s, 0, sizeof(*s));
2087
2088 rcu_read_lock();
2089 for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2090 if (IS_LEAF(n)) {
2091 struct leaf *l = (struct leaf *)n;
2092 struct leaf_info *li;
2093 struct hlist_node *tmp;
2094
2095 s->leaves++;
2096 s->totdepth += iter.depth;
2097 if (iter.depth > s->maxdepth)
2098 s->maxdepth = iter.depth;
2099
2100 hlist_for_each_entry_rcu(li, tmp, &l->list, hlist)
2101 ++s->prefixes;
2102 } else {
2103 const struct tnode *tn = (const struct tnode *) n;
2104 int i;
2105
2106 s->tnodes++;
2107 if (tn->bits < MAX_STAT_DEPTH)
2108 s->nodesizes[tn->bits]++;
2109
2110 for (i = 0; i < (1<<tn->bits); i++)
2111 if (!tn->child[i])
2112 s->nullpointers++;
2113 }
2114 }
2115 rcu_read_unlock();
2116 }
2117
2118 /*
2119 * This outputs /proc/net/fib_triestats
2120 */
2121 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2122 {
2123 unsigned int i, max, pointers, bytes, avdepth;
2124
2125 if (stat->leaves)
2126 avdepth = stat->totdepth*100 / stat->leaves;
2127 else
2128 avdepth = 0;
2129
2130 seq_printf(seq, "\tAver depth: %u.%02d\n",
2131 avdepth / 100, avdepth % 100);
2132 seq_printf(seq, "\tMax depth: %u\n", stat->maxdepth);
2133
2134 seq_printf(seq, "\tLeaves: %u\n", stat->leaves);
2135 bytes = sizeof(struct leaf) * stat->leaves;
2136
2137 seq_printf(seq, "\tPrefixes: %u\n", stat->prefixes);
2138 bytes += sizeof(struct leaf_info) * stat->prefixes;
2139
2140 seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2141 bytes += sizeof(struct tnode) * stat->tnodes;
2142
2143 max = MAX_STAT_DEPTH;
2144 while (max > 0 && stat->nodesizes[max-1] == 0)
2145 max--;
2146
2147 pointers = 0;
2148 for (i = 1; i <= max; i++)
2149 if (stat->nodesizes[i] != 0) {
2150 seq_printf(seq, " %u: %u", i, stat->nodesizes[i]);
2151 pointers += (1<<i) * stat->nodesizes[i];
2152 }
2153 seq_putc(seq, '\n');
2154 seq_printf(seq, "\tPointers: %u\n", pointers);
2155
2156 bytes += sizeof(struct rt_trie_node *) * pointers;
2157 seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2158 seq_printf(seq, "Total size: %u kB\n", (bytes + 1023) / 1024);
2159 }
2160
2161 #ifdef CONFIG_IP_FIB_TRIE_STATS
2162 static void trie_show_usage(struct seq_file *seq,
2163 const struct trie_use_stats *stats)
2164 {
2165 seq_printf(seq, "\nCounters:\n---------\n");
2166 seq_printf(seq, "gets = %u\n", stats->gets);
2167 seq_printf(seq, "backtracks = %u\n", stats->backtrack);
2168 seq_printf(seq, "semantic match passed = %u\n",
2169 stats->semantic_match_passed);
2170 seq_printf(seq, "semantic match miss = %u\n",
2171 stats->semantic_match_miss);
2172 seq_printf(seq, "null node hit= %u\n", stats->null_node_hit);
2173 seq_printf(seq, "skipped node resize = %u\n\n",
2174 stats->resize_node_skipped);
2175 }
2176 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2177
2178 static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2179 {
2180 if (tb->tb_id == RT_TABLE_LOCAL)
2181 seq_puts(seq, "Local:\n");
2182 else if (tb->tb_id == RT_TABLE_MAIN)
2183 seq_puts(seq, "Main:\n");
2184 else
2185 seq_printf(seq, "Id %d:\n", tb->tb_id);
2186 }
2187
2188
2189 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2190 {
2191 struct net *net = (struct net *)seq->private;
2192 unsigned int h;
2193
2194 seq_printf(seq,
2195 "Basic info: size of leaf:"
2196 " %Zd bytes, size of tnode: %Zd bytes.\n",
2197 sizeof(struct leaf), sizeof(struct tnode));
2198
2199 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2200 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2201 struct hlist_node *node;
2202 struct fib_table *tb;
2203
2204 hlist_for_each_entry_rcu(tb, node, head, tb_hlist) {
2205 struct trie *t = (struct trie *) tb->tb_data;
2206 struct trie_stat stat;
2207
2208 if (!t)
2209 continue;
2210
2211 fib_table_print(seq, tb);
2212
2213 trie_collect_stats(t, &stat);
2214 trie_show_stats(seq, &stat);
2215 #ifdef CONFIG_IP_FIB_TRIE_STATS
2216 trie_show_usage(seq, &t->stats);
2217 #endif
2218 }
2219 }
2220
2221 return 0;
2222 }
2223
2224 static int fib_triestat_seq_open(struct inode *inode, struct file *file)
2225 {
2226 return single_open_net(inode, file, fib_triestat_seq_show);
2227 }
2228
2229 static const struct file_operations fib_triestat_fops = {
2230 .owner = THIS_MODULE,
2231 .open = fib_triestat_seq_open,
2232 .read = seq_read,
2233 .llseek = seq_lseek,
2234 .release = single_release_net,
2235 };
2236
2237 static struct rt_trie_node *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2238 {
2239 struct fib_trie_iter *iter = seq->private;
2240 struct net *net = seq_file_net(seq);
2241 loff_t idx = 0;
2242 unsigned int h;
2243
2244 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2245 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2246 struct hlist_node *node;
2247 struct fib_table *tb;
2248
2249 hlist_for_each_entry_rcu(tb, node, head, tb_hlist) {
2250 struct rt_trie_node *n;
2251
2252 for (n = fib_trie_get_first(iter,
2253 (struct trie *) tb->tb_data);
2254 n; n = fib_trie_get_next(iter))
2255 if (pos == idx++) {
2256 iter->tb = tb;
2257 return n;
2258 }
2259 }
2260 }
2261
2262 return NULL;
2263 }
2264
2265 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2266 __acquires(RCU)
2267 {
2268 rcu_read_lock();
2269 return fib_trie_get_idx(seq, *pos);
2270 }
2271
2272 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2273 {
2274 struct fib_trie_iter *iter = seq->private;
2275 struct net *net = seq_file_net(seq);
2276 struct fib_table *tb = iter->tb;
2277 struct hlist_node *tb_node;
2278 unsigned int h;
2279 struct rt_trie_node *n;
2280
2281 ++*pos;
2282 /* next node in same table */
2283 n = fib_trie_get_next(iter);
2284 if (n)
2285 return n;
2286
2287 /* walk rest of this hash chain */
2288 h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2289 while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2290 tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2291 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2292 if (n)
2293 goto found;
2294 }
2295
2296 /* new hash chain */
2297 while (++h < FIB_TABLE_HASHSZ) {
2298 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2299 hlist_for_each_entry_rcu(tb, tb_node, head, tb_hlist) {
2300 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2301 if (n)
2302 goto found;
2303 }
2304 }
2305 return NULL;
2306
2307 found:
2308 iter->tb = tb;
2309 return n;
2310 }
2311
2312 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2313 __releases(RCU)
2314 {
2315 rcu_read_unlock();
2316 }
2317
2318 static void seq_indent(struct seq_file *seq, int n)
2319 {
2320 while (n-- > 0)
2321 seq_puts(seq, " ");
2322 }
2323
2324 static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2325 {
2326 switch (s) {
2327 case RT_SCOPE_UNIVERSE: return "universe";
2328 case RT_SCOPE_SITE: return "site";
2329 case RT_SCOPE_LINK: return "link";
2330 case RT_SCOPE_HOST: return "host";
2331 case RT_SCOPE_NOWHERE: return "nowhere";
2332 default:
2333 snprintf(buf, len, "scope=%d", s);
2334 return buf;
2335 }
2336 }
2337
2338 static const char *const rtn_type_names[__RTN_MAX] = {
2339 [RTN_UNSPEC] = "UNSPEC",
2340 [RTN_UNICAST] = "UNICAST",
2341 [RTN_LOCAL] = "LOCAL",
2342 [RTN_BROADCAST] = "BROADCAST",
2343 [RTN_ANYCAST] = "ANYCAST",
2344 [RTN_MULTICAST] = "MULTICAST",
2345 [RTN_BLACKHOLE] = "BLACKHOLE",
2346 [RTN_UNREACHABLE] = "UNREACHABLE",
2347 [RTN_PROHIBIT] = "PROHIBIT",
2348 [RTN_THROW] = "THROW",
2349 [RTN_NAT] = "NAT",
2350 [RTN_XRESOLVE] = "XRESOLVE",
2351 };
2352
2353 static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2354 {
2355 if (t < __RTN_MAX && rtn_type_names[t])
2356 return rtn_type_names[t];
2357 snprintf(buf, len, "type %u", t);
2358 return buf;
2359 }
2360
2361 /* Pretty print the trie */
2362 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2363 {
2364 const struct fib_trie_iter *iter = seq->private;
2365 struct rt_trie_node *n = v;
2366
2367 if (!node_parent_rcu(n))
2368 fib_table_print(seq, iter->tb);
2369
2370 if (IS_TNODE(n)) {
2371 struct tnode *tn = (struct tnode *) n;
2372 __be32 prf = htonl(mask_pfx(tn->key, tn->pos));
2373
2374 seq_indent(seq, iter->depth-1);
2375 seq_printf(seq, " +-- %pI4/%d %d %d %d\n",
2376 &prf, tn->pos, tn->bits, tn->full_children,
2377 tn->empty_children);
2378
2379 } else {
2380 struct leaf *l = (struct leaf *) n;
2381 struct leaf_info *li;
2382 struct hlist_node *node;
2383 __be32 val = htonl(l->key);
2384
2385 seq_indent(seq, iter->depth);
2386 seq_printf(seq, " |-- %pI4\n", &val);
2387
2388 hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
2389 struct fib_alias *fa;
2390
2391 list_for_each_entry_rcu(fa, &li->falh, fa_list) {
2392 char buf1[32], buf2[32];
2393
2394 seq_indent(seq, iter->depth+1);
2395 seq_printf(seq, " /%d %s %s", li->plen,
2396 rtn_scope(buf1, sizeof(buf1),
2397 fa->fa_info->fib_scope),
2398 rtn_type(buf2, sizeof(buf2),
2399 fa->fa_type));
2400 if (fa->fa_tos)
2401 seq_printf(seq, " tos=%d", fa->fa_tos);
2402 seq_putc(seq, '\n');
2403 }
2404 }
2405 }
2406
2407 return 0;
2408 }
2409
2410 static const struct seq_operations fib_trie_seq_ops = {
2411 .start = fib_trie_seq_start,
2412 .next = fib_trie_seq_next,
2413 .stop = fib_trie_seq_stop,
2414 .show = fib_trie_seq_show,
2415 };
2416
2417 static int fib_trie_seq_open(struct inode *inode, struct file *file)
2418 {
2419 return seq_open_net(inode, file, &fib_trie_seq_ops,
2420 sizeof(struct fib_trie_iter));
2421 }
2422
2423 static const struct file_operations fib_trie_fops = {
2424 .owner = THIS_MODULE,
2425 .open = fib_trie_seq_open,
2426 .read = seq_read,
2427 .llseek = seq_lseek,
2428 .release = seq_release_net,
2429 };
2430
2431 struct fib_route_iter {
2432 struct seq_net_private p;
2433 struct trie *main_trie;
2434 loff_t pos;
2435 t_key key;
2436 };
2437
2438 static struct leaf *fib_route_get_idx(struct fib_route_iter *iter, loff_t pos)
2439 {
2440 struct leaf *l = NULL;
2441 struct trie *t = iter->main_trie;
2442
2443 /* use cache location of last found key */
2444 if (iter->pos > 0 && pos >= iter->pos && (l = fib_find_node(t, iter->key)))
2445 pos -= iter->pos;
2446 else {
2447 iter->pos = 0;
2448 l = trie_firstleaf(t);
2449 }
2450
2451 while (l && pos-- > 0) {
2452 iter->pos++;
2453 l = trie_nextleaf(l);
2454 }
2455
2456 if (l)
2457 iter->key = pos; /* remember it */
2458 else
2459 iter->pos = 0; /* forget it */
2460
2461 return l;
2462 }
2463
2464 static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2465 __acquires(RCU)
2466 {
2467 struct fib_route_iter *iter = seq->private;
2468 struct fib_table *tb;
2469
2470 rcu_read_lock();
2471 tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2472 if (!tb)
2473 return NULL;
2474
2475 iter->main_trie = (struct trie *) tb->tb_data;
2476 if (*pos == 0)
2477 return SEQ_START_TOKEN;
2478 else
2479 return fib_route_get_idx(iter, *pos - 1);
2480 }
2481
2482 static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2483 {
2484 struct fib_route_iter *iter = seq->private;
2485 struct leaf *l = v;
2486
2487 ++*pos;
2488 if (v == SEQ_START_TOKEN) {
2489 iter->pos = 0;
2490 l = trie_firstleaf(iter->main_trie);
2491 } else {
2492 iter->pos++;
2493 l = trie_nextleaf(l);
2494 }
2495
2496 if (l)
2497 iter->key = l->key;
2498 else
2499 iter->pos = 0;
2500 return l;
2501 }
2502
2503 static void fib_route_seq_stop(struct seq_file *seq, void *v)
2504 __releases(RCU)
2505 {
2506 rcu_read_unlock();
2507 }
2508
2509 static unsigned int fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2510 {
2511 unsigned int flags = 0;
2512
2513 if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2514 flags = RTF_REJECT;
2515 if (fi && fi->fib_nh->nh_gw)
2516 flags |= RTF_GATEWAY;
2517 if (mask == htonl(0xFFFFFFFF))
2518 flags |= RTF_HOST;
2519 flags |= RTF_UP;
2520 return flags;
2521 }
2522
2523 /*
2524 * This outputs /proc/net/route.
2525 * The format of the file is not supposed to be changed
2526 * and needs to be same as fib_hash output to avoid breaking
2527 * legacy utilities
2528 */
2529 static int fib_route_seq_show(struct seq_file *seq, void *v)
2530 {
2531 struct leaf *l = v;
2532 struct leaf_info *li;
2533 struct hlist_node *node;
2534
2535 if (v == SEQ_START_TOKEN) {
2536 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2537 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2538 "\tWindow\tIRTT");
2539 return 0;
2540 }
2541
2542 hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
2543 struct fib_alias *fa;
2544 __be32 mask, prefix;
2545
2546 mask = inet_make_mask(li->plen);
2547 prefix = htonl(l->key);
2548
2549 list_for_each_entry_rcu(fa, &li->falh, fa_list) {
2550 const struct fib_info *fi = fa->fa_info;
2551 unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2552 int len;
2553
2554 if (fa->fa_type == RTN_BROADCAST
2555 || fa->fa_type == RTN_MULTICAST)
2556 continue;
2557
2558 if (fi)
2559 seq_printf(seq,
2560 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2561 "%d\t%08X\t%d\t%u\t%u%n",
2562 fi->fib_dev ? fi->fib_dev->name : "*",
2563 prefix,
2564 fi->fib_nh->nh_gw, flags, 0, 0,
2565 fi->fib_priority,
2566 mask,
2567 (fi->fib_advmss ?
2568 fi->fib_advmss + 40 : 0),
2569 fi->fib_window,
2570 fi->fib_rtt >> 3, &len);
2571 else
2572 seq_printf(seq,
2573 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2574 "%d\t%08X\t%d\t%u\t%u%n",
2575 prefix, 0, flags, 0, 0, 0,
2576 mask, 0, 0, 0, &len);
2577
2578 seq_printf(seq, "%*s\n", 127 - len, "");
2579 }
2580 }
2581
2582 return 0;
2583 }
2584
2585 static const struct seq_operations fib_route_seq_ops = {
2586 .start = fib_route_seq_start,
2587 .next = fib_route_seq_next,
2588 .stop = fib_route_seq_stop,
2589 .show = fib_route_seq_show,
2590 };
2591
2592 static int fib_route_seq_open(struct inode *inode, struct file *file)
2593 {
2594 return seq_open_net(inode, file, &fib_route_seq_ops,
2595 sizeof(struct fib_route_iter));
2596 }
2597
2598 static const struct file_operations fib_route_fops = {
2599 .owner = THIS_MODULE,
2600 .open = fib_route_seq_open,
2601 .read = seq_read,
2602 .llseek = seq_lseek,
2603 .release = seq_release_net,
2604 };
2605
2606 int __net_init fib_proc_init(struct net *net)
2607 {
2608 if (!proc_net_fops_create(net, "fib_trie", S_IRUGO, &fib_trie_fops))
2609 goto out1;
2610
2611 if (!proc_net_fops_create(net, "fib_triestat", S_IRUGO,
2612 &fib_triestat_fops))
2613 goto out2;
2614
2615 if (!proc_net_fops_create(net, "route", S_IRUGO, &fib_route_fops))
2616 goto out3;
2617
2618 return 0;
2619
2620 out3:
2621 proc_net_remove(net, "fib_triestat");
2622 out2:
2623 proc_net_remove(net, "fib_trie");
2624 out1:
2625 return -ENOMEM;
2626 }
2627
2628 void __net_exit fib_proc_exit(struct net *net)
2629 {
2630 proc_net_remove(net, "fib_trie");
2631 proc_net_remove(net, "fib_triestat");
2632 proc_net_remove(net, "route");
2633 }
2634
2635 #endif /* CONFIG_PROC_FS */
This page took 0.083531 seconds and 5 git commands to generate.