Merge branch 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux-2.6
[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 descibed 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.nada.kth.se/~snilsson/public/papers/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 * Version: $Id: fib_trie.c,v 1.3 2005/06/08 14:20:01 robert Exp $
26 *
27 *
28 * Code from fib_hash has been reused which includes the following header:
29 *
30 *
31 * INET An implementation of the TCP/IP protocol suite for the LINUX
32 * operating system. INET is implemented using the BSD Socket
33 * interface as the means of communication with the user level.
34 *
35 * IPv4 FIB: lookup engine and maintenance routines.
36 *
37 *
38 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
39 *
40 * This program is free software; you can redistribute it and/or
41 * modify it under the terms of the GNU General Public License
42 * as published by the Free Software Foundation; either version
43 * 2 of the License, or (at your option) any later version.
44 *
45 * Substantial contributions to this work comes from:
46 *
47 * David S. Miller, <davem@davemloft.net>
48 * Stephen Hemminger <shemminger@osdl.org>
49 * Paul E. McKenney <paulmck@us.ibm.com>
50 * Patrick McHardy <kaber@trash.net>
51 */
52
53 #define VERSION "0.407"
54
55 #include <asm/uaccess.h>
56 #include <asm/system.h>
57 #include <asm/bitops.h>
58 #include <linux/types.h>
59 #include <linux/kernel.h>
60 #include <linux/mm.h>
61 #include <linux/string.h>
62 #include <linux/socket.h>
63 #include <linux/sockios.h>
64 #include <linux/errno.h>
65 #include <linux/in.h>
66 #include <linux/inet.h>
67 #include <linux/inetdevice.h>
68 #include <linux/netdevice.h>
69 #include <linux/if_arp.h>
70 #include <linux/proc_fs.h>
71 #include <linux/rcupdate.h>
72 #include <linux/skbuff.h>
73 #include <linux/netlink.h>
74 #include <linux/init.h>
75 #include <linux/list.h>
76 #include <net/ip.h>
77 #include <net/protocol.h>
78 #include <net/route.h>
79 #include <net/tcp.h>
80 #include <net/sock.h>
81 #include <net/ip_fib.h>
82 #include "fib_lookup.h"
83
84 #undef CONFIG_IP_FIB_TRIE_STATS
85 #define MAX_STAT_DEPTH 32
86
87 #define KEYLENGTH (8*sizeof(t_key))
88 #define MASK_PFX(k, l) (((l)==0)?0:(k >> (KEYLENGTH-l)) << (KEYLENGTH-l))
89 #define TKEY_GET_MASK(offset, bits) (((bits)==0)?0:((t_key)(-1) << (KEYLENGTH - bits) >> offset))
90
91 typedef unsigned int t_key;
92
93 #define T_TNODE 0
94 #define T_LEAF 1
95 #define NODE_TYPE_MASK 0x1UL
96 #define NODE_PARENT(node) \
97 ((struct tnode *)rcu_dereference(((node)->parent & ~NODE_TYPE_MASK)))
98
99 #define NODE_TYPE(node) ((node)->parent & NODE_TYPE_MASK)
100
101 #define NODE_SET_PARENT(node, ptr) \
102 rcu_assign_pointer((node)->parent, \
103 ((unsigned long)(ptr)) | NODE_TYPE(node))
104
105 #define IS_TNODE(n) (!(n->parent & T_LEAF))
106 #define IS_LEAF(n) (n->parent & T_LEAF)
107
108 struct node {
109 t_key key;
110 unsigned long parent;
111 };
112
113 struct leaf {
114 t_key key;
115 unsigned long parent;
116 struct hlist_head list;
117 struct rcu_head rcu;
118 };
119
120 struct leaf_info {
121 struct hlist_node hlist;
122 struct rcu_head rcu;
123 int plen;
124 struct list_head falh;
125 };
126
127 struct tnode {
128 t_key key;
129 unsigned long parent;
130 unsigned short pos:5; /* 2log(KEYLENGTH) bits needed */
131 unsigned short bits:5; /* 2log(KEYLENGTH) bits needed */
132 unsigned short full_children; /* KEYLENGTH bits needed */
133 unsigned short empty_children; /* KEYLENGTH bits needed */
134 struct rcu_head rcu;
135 struct node *child[0];
136 };
137
138 #ifdef CONFIG_IP_FIB_TRIE_STATS
139 struct trie_use_stats {
140 unsigned int gets;
141 unsigned int backtrack;
142 unsigned int semantic_match_passed;
143 unsigned int semantic_match_miss;
144 unsigned int null_node_hit;
145 unsigned int resize_node_skipped;
146 };
147 #endif
148
149 struct trie_stat {
150 unsigned int totdepth;
151 unsigned int maxdepth;
152 unsigned int tnodes;
153 unsigned int leaves;
154 unsigned int nullpointers;
155 unsigned int nodesizes[MAX_STAT_DEPTH];
156 };
157
158 struct trie {
159 struct node *trie;
160 #ifdef CONFIG_IP_FIB_TRIE_STATS
161 struct trie_use_stats stats;
162 #endif
163 int size;
164 unsigned int revision;
165 };
166
167 static void put_child(struct trie *t, struct tnode *tn, int i, struct node *n);
168 static void tnode_put_child_reorg(struct tnode *tn, int i, struct node *n, int wasfull);
169 static struct node *resize(struct trie *t, struct tnode *tn);
170 static struct tnode *inflate(struct trie *t, struct tnode *tn);
171 static struct tnode *halve(struct trie *t, struct tnode *tn);
172 static void tnode_free(struct tnode *tn);
173
174 static struct kmem_cache *fn_alias_kmem __read_mostly;
175 static struct trie *trie_local = NULL, *trie_main = NULL;
176
177
178 /* rcu_read_lock needs to be hold by caller from readside */
179
180 static inline struct node *tnode_get_child(struct tnode *tn, int i)
181 {
182 BUG_ON(i >= 1 << tn->bits);
183
184 return rcu_dereference(tn->child[i]);
185 }
186
187 static inline int tnode_child_length(const struct tnode *tn)
188 {
189 return 1 << tn->bits;
190 }
191
192 static inline t_key tkey_extract_bits(t_key a, int offset, int bits)
193 {
194 if (offset < KEYLENGTH)
195 return ((t_key)(a << offset)) >> (KEYLENGTH - bits);
196 else
197 return 0;
198 }
199
200 static inline int tkey_equals(t_key a, t_key b)
201 {
202 return a == b;
203 }
204
205 static inline int tkey_sub_equals(t_key a, int offset, int bits, t_key b)
206 {
207 if (bits == 0 || offset >= KEYLENGTH)
208 return 1;
209 bits = bits > KEYLENGTH ? KEYLENGTH : bits;
210 return ((a ^ b) << offset) >> (KEYLENGTH - bits) == 0;
211 }
212
213 static inline int tkey_mismatch(t_key a, int offset, t_key b)
214 {
215 t_key diff = a ^ b;
216 int i = offset;
217
218 if (!diff)
219 return 0;
220 while ((diff << i) >> (KEYLENGTH-1) == 0)
221 i++;
222 return i;
223 }
224
225 /*
226 To understand this stuff, an understanding of keys and all their bits is
227 necessary. Every node in the trie has a key associated with it, but not
228 all of the bits in that key are significant.
229
230 Consider a node 'n' and its parent 'tp'.
231
232 If n is a leaf, every bit in its key is significant. Its presence is
233 necessitated by path compression, since during a tree traversal (when
234 searching for a leaf - unless we are doing an insertion) we will completely
235 ignore all skipped bits we encounter. Thus we need to verify, at the end of
236 a potentially successful search, that we have indeed been walking the
237 correct key path.
238
239 Note that we can never "miss" the correct key in the tree if present by
240 following the wrong path. Path compression ensures that segments of the key
241 that are the same for all keys with a given prefix are skipped, but the
242 skipped part *is* identical for each node in the subtrie below the skipped
243 bit! trie_insert() in this implementation takes care of that - note the
244 call to tkey_sub_equals() in trie_insert().
245
246 if n is an internal node - a 'tnode' here, the various parts of its key
247 have many different meanings.
248
249 Example:
250 _________________________________________________________________
251 | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
252 -----------------------------------------------------------------
253 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
254
255 _________________________________________________________________
256 | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
257 -----------------------------------------------------------------
258 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
259
260 tp->pos = 7
261 tp->bits = 3
262 n->pos = 15
263 n->bits = 4
264
265 First, let's just ignore the bits that come before the parent tp, that is
266 the bits from 0 to (tp->pos-1). They are *known* but at this point we do
267 not use them for anything.
268
269 The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
270 index into the parent's child array. That is, they will be used to find
271 'n' among tp's children.
272
273 The bits from (tp->pos + tp->bits) to (n->pos - 1) - "S" - are skipped bits
274 for the node n.
275
276 All the bits we have seen so far are significant to the node n. The rest
277 of the bits are really not needed or indeed known in n->key.
278
279 The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
280 n's child array, and will of course be different for each child.
281
282
283 The rest of the bits, from (n->pos + n->bits) onward, are completely unknown
284 at this point.
285
286 */
287
288 static inline void check_tnode(const struct tnode *tn)
289 {
290 WARN_ON(tn && tn->pos+tn->bits > 32);
291 }
292
293 static int halve_threshold = 25;
294 static int inflate_threshold = 50;
295 static int halve_threshold_root = 15;
296 static int inflate_threshold_root = 25;
297
298
299 static void __alias_free_mem(struct rcu_head *head)
300 {
301 struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
302 kmem_cache_free(fn_alias_kmem, fa);
303 }
304
305 static inline void alias_free_mem_rcu(struct fib_alias *fa)
306 {
307 call_rcu(&fa->rcu, __alias_free_mem);
308 }
309
310 static void __leaf_free_rcu(struct rcu_head *head)
311 {
312 kfree(container_of(head, struct leaf, rcu));
313 }
314
315 static void __leaf_info_free_rcu(struct rcu_head *head)
316 {
317 kfree(container_of(head, struct leaf_info, rcu));
318 }
319
320 static inline void free_leaf_info(struct leaf_info *leaf)
321 {
322 call_rcu(&leaf->rcu, __leaf_info_free_rcu);
323 }
324
325 static struct tnode *tnode_alloc(unsigned int size)
326 {
327 struct page *pages;
328
329 if (size <= PAGE_SIZE)
330 return kcalloc(size, 1, GFP_KERNEL);
331
332 pages = alloc_pages(GFP_KERNEL|__GFP_ZERO, get_order(size));
333 if (!pages)
334 return NULL;
335
336 return page_address(pages);
337 }
338
339 static void __tnode_free_rcu(struct rcu_head *head)
340 {
341 struct tnode *tn = container_of(head, struct tnode, rcu);
342 unsigned int size = sizeof(struct tnode) +
343 (1 << tn->bits) * sizeof(struct node *);
344
345 if (size <= PAGE_SIZE)
346 kfree(tn);
347 else
348 free_pages((unsigned long)tn, get_order(size));
349 }
350
351 static inline void tnode_free(struct tnode *tn)
352 {
353 if(IS_LEAF(tn)) {
354 struct leaf *l = (struct leaf *) tn;
355 call_rcu_bh(&l->rcu, __leaf_free_rcu);
356 }
357 else
358 call_rcu(&tn->rcu, __tnode_free_rcu);
359 }
360
361 static struct leaf *leaf_new(void)
362 {
363 struct leaf *l = kmalloc(sizeof(struct leaf), GFP_KERNEL);
364 if (l) {
365 l->parent = T_LEAF;
366 INIT_HLIST_HEAD(&l->list);
367 }
368 return l;
369 }
370
371 static struct leaf_info *leaf_info_new(int plen)
372 {
373 struct leaf_info *li = kmalloc(sizeof(struct leaf_info), GFP_KERNEL);
374 if (li) {
375 li->plen = plen;
376 INIT_LIST_HEAD(&li->falh);
377 }
378 return li;
379 }
380
381 static struct tnode* tnode_new(t_key key, int pos, int bits)
382 {
383 int nchildren = 1<<bits;
384 int sz = sizeof(struct tnode) + nchildren * sizeof(struct node *);
385 struct tnode *tn = tnode_alloc(sz);
386
387 if (tn) {
388 memset(tn, 0, sz);
389 tn->parent = T_TNODE;
390 tn->pos = pos;
391 tn->bits = bits;
392 tn->key = key;
393 tn->full_children = 0;
394 tn->empty_children = 1<<bits;
395 }
396
397 pr_debug("AT %p s=%u %u\n", tn, (unsigned int) sizeof(struct tnode),
398 (unsigned int) (sizeof(struct node) * 1<<bits));
399 return tn;
400 }
401
402 /*
403 * Check whether a tnode 'n' is "full", i.e. it is an internal node
404 * and no bits are skipped. See discussion in dyntree paper p. 6
405 */
406
407 static inline int tnode_full(const struct tnode *tn, const struct node *n)
408 {
409 if (n == NULL || IS_LEAF(n))
410 return 0;
411
412 return ((struct tnode *) n)->pos == tn->pos + tn->bits;
413 }
414
415 static inline void put_child(struct trie *t, struct tnode *tn, int i, struct node *n)
416 {
417 tnode_put_child_reorg(tn, i, n, -1);
418 }
419
420 /*
421 * Add a child at position i overwriting the old value.
422 * Update the value of full_children and empty_children.
423 */
424
425 static void tnode_put_child_reorg(struct tnode *tn, int i, struct node *n, int wasfull)
426 {
427 struct node *chi = tn->child[i];
428 int isfull;
429
430 BUG_ON(i >= 1<<tn->bits);
431
432
433 /* update emptyChildren */
434 if (n == NULL && chi != NULL)
435 tn->empty_children++;
436 else if (n != NULL && chi == NULL)
437 tn->empty_children--;
438
439 /* update fullChildren */
440 if (wasfull == -1)
441 wasfull = tnode_full(tn, chi);
442
443 isfull = tnode_full(tn, n);
444 if (wasfull && !isfull)
445 tn->full_children--;
446 else if (!wasfull && isfull)
447 tn->full_children++;
448
449 if (n)
450 NODE_SET_PARENT(n, tn);
451
452 rcu_assign_pointer(tn->child[i], n);
453 }
454
455 static struct node *resize(struct trie *t, struct tnode *tn)
456 {
457 int i;
458 int err = 0;
459 struct tnode *old_tn;
460 int inflate_threshold_use;
461 int halve_threshold_use;
462
463 if (!tn)
464 return NULL;
465
466 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
467 tn, inflate_threshold, halve_threshold);
468
469 /* No children */
470 if (tn->empty_children == tnode_child_length(tn)) {
471 tnode_free(tn);
472 return NULL;
473 }
474 /* One child */
475 if (tn->empty_children == tnode_child_length(tn) - 1)
476 for (i = 0; i < tnode_child_length(tn); i++) {
477 struct node *n;
478
479 n = tn->child[i];
480 if (!n)
481 continue;
482
483 /* compress one level */
484 NODE_SET_PARENT(n, NULL);
485 tnode_free(tn);
486 return n;
487 }
488 /*
489 * Double as long as the resulting node has a number of
490 * nonempty nodes that are above the threshold.
491 */
492
493 /*
494 * From "Implementing a dynamic compressed trie" by Stefan Nilsson of
495 * the Helsinki University of Technology and Matti Tikkanen of Nokia
496 * Telecommunications, page 6:
497 * "A node is doubled if the ratio of non-empty children to all
498 * children in the *doubled* node is at least 'high'."
499 *
500 * 'high' in this instance is the variable 'inflate_threshold'. It
501 * is expressed as a percentage, so we multiply it with
502 * tnode_child_length() and instead of multiplying by 2 (since the
503 * child array will be doubled by inflate()) and multiplying
504 * the left-hand side by 100 (to handle the percentage thing) we
505 * multiply the left-hand side by 50.
506 *
507 * The left-hand side may look a bit weird: tnode_child_length(tn)
508 * - tn->empty_children is of course the number of non-null children
509 * in the current node. tn->full_children is the number of "full"
510 * children, that is non-null tnodes with a skip value of 0.
511 * All of those will be doubled in the resulting inflated tnode, so
512 * we just count them one extra time here.
513 *
514 * A clearer way to write this would be:
515 *
516 * to_be_doubled = tn->full_children;
517 * not_to_be_doubled = tnode_child_length(tn) - tn->empty_children -
518 * tn->full_children;
519 *
520 * new_child_length = tnode_child_length(tn) * 2;
521 *
522 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
523 * new_child_length;
524 * if (new_fill_factor >= inflate_threshold)
525 *
526 * ...and so on, tho it would mess up the while () loop.
527 *
528 * anyway,
529 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
530 * inflate_threshold
531 *
532 * avoid a division:
533 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
534 * inflate_threshold * new_child_length
535 *
536 * expand not_to_be_doubled and to_be_doubled, and shorten:
537 * 100 * (tnode_child_length(tn) - tn->empty_children +
538 * tn->full_children) >= inflate_threshold * new_child_length
539 *
540 * expand new_child_length:
541 * 100 * (tnode_child_length(tn) - tn->empty_children +
542 * tn->full_children) >=
543 * inflate_threshold * tnode_child_length(tn) * 2
544 *
545 * shorten again:
546 * 50 * (tn->full_children + tnode_child_length(tn) -
547 * tn->empty_children) >= inflate_threshold *
548 * tnode_child_length(tn)
549 *
550 */
551
552 check_tnode(tn);
553
554 /* Keep root node larger */
555
556 if(!tn->parent)
557 inflate_threshold_use = inflate_threshold_root;
558 else
559 inflate_threshold_use = inflate_threshold;
560
561 err = 0;
562 while ((tn->full_children > 0 &&
563 50 * (tn->full_children + tnode_child_length(tn) - tn->empty_children) >=
564 inflate_threshold_use * tnode_child_length(tn))) {
565
566 old_tn = tn;
567 tn = inflate(t, tn);
568 if (IS_ERR(tn)) {
569 tn = old_tn;
570 #ifdef CONFIG_IP_FIB_TRIE_STATS
571 t->stats.resize_node_skipped++;
572 #endif
573 break;
574 }
575 }
576
577 check_tnode(tn);
578
579 /*
580 * Halve as long as the number of empty children in this
581 * node is above threshold.
582 */
583
584
585 /* Keep root node larger */
586
587 if(!tn->parent)
588 halve_threshold_use = halve_threshold_root;
589 else
590 halve_threshold_use = halve_threshold;
591
592 err = 0;
593 while (tn->bits > 1 &&
594 100 * (tnode_child_length(tn) - tn->empty_children) <
595 halve_threshold_use * tnode_child_length(tn)) {
596
597 old_tn = tn;
598 tn = halve(t, tn);
599 if (IS_ERR(tn)) {
600 tn = old_tn;
601 #ifdef CONFIG_IP_FIB_TRIE_STATS
602 t->stats.resize_node_skipped++;
603 #endif
604 break;
605 }
606 }
607
608
609 /* Only one child remains */
610 if (tn->empty_children == tnode_child_length(tn) - 1)
611 for (i = 0; i < tnode_child_length(tn); i++) {
612 struct node *n;
613
614 n = tn->child[i];
615 if (!n)
616 continue;
617
618 /* compress one level */
619
620 NODE_SET_PARENT(n, NULL);
621 tnode_free(tn);
622 return n;
623 }
624
625 return (struct node *) tn;
626 }
627
628 static struct tnode *inflate(struct trie *t, struct tnode *tn)
629 {
630 struct tnode *inode;
631 struct tnode *oldtnode = tn;
632 int olen = tnode_child_length(tn);
633 int i;
634
635 pr_debug("In inflate\n");
636
637 tn = tnode_new(oldtnode->key, oldtnode->pos, oldtnode->bits + 1);
638
639 if (!tn)
640 return ERR_PTR(-ENOMEM);
641
642 /*
643 * Preallocate and store tnodes before the actual work so we
644 * don't get into an inconsistent state if memory allocation
645 * fails. In case of failure we return the oldnode and inflate
646 * of tnode is ignored.
647 */
648
649 for (i = 0; i < olen; i++) {
650 struct tnode *inode = (struct tnode *) tnode_get_child(oldtnode, i);
651
652 if (inode &&
653 IS_TNODE(inode) &&
654 inode->pos == oldtnode->pos + oldtnode->bits &&
655 inode->bits > 1) {
656 struct tnode *left, *right;
657 t_key m = TKEY_GET_MASK(inode->pos, 1);
658
659 left = tnode_new(inode->key&(~m), inode->pos + 1,
660 inode->bits - 1);
661 if (!left)
662 goto nomem;
663
664 right = tnode_new(inode->key|m, inode->pos + 1,
665 inode->bits - 1);
666
667 if (!right) {
668 tnode_free(left);
669 goto nomem;
670 }
671
672 put_child(t, tn, 2*i, (struct node *) left);
673 put_child(t, tn, 2*i+1, (struct node *) right);
674 }
675 }
676
677 for (i = 0; i < olen; i++) {
678 struct node *node = tnode_get_child(oldtnode, i);
679 struct tnode *left, *right;
680 int size, j;
681
682 /* An empty child */
683 if (node == NULL)
684 continue;
685
686 /* A leaf or an internal node with skipped bits */
687
688 if (IS_LEAF(node) || ((struct tnode *) node)->pos >
689 tn->pos + tn->bits - 1) {
690 if (tkey_extract_bits(node->key, oldtnode->pos + oldtnode->bits,
691 1) == 0)
692 put_child(t, tn, 2*i, node);
693 else
694 put_child(t, tn, 2*i+1, node);
695 continue;
696 }
697
698 /* An internal node with two children */
699 inode = (struct tnode *) node;
700
701 if (inode->bits == 1) {
702 put_child(t, tn, 2*i, inode->child[0]);
703 put_child(t, tn, 2*i+1, inode->child[1]);
704
705 tnode_free(inode);
706 continue;
707 }
708
709 /* An internal node with more than two children */
710
711 /* We will replace this node 'inode' with two new
712 * ones, 'left' and 'right', each with half of the
713 * original children. The two new nodes will have
714 * a position one bit further down the key and this
715 * means that the "significant" part of their keys
716 * (see the discussion near the top of this file)
717 * will differ by one bit, which will be "0" in
718 * left's key and "1" in right's key. Since we are
719 * moving the key position by one step, the bit that
720 * we are moving away from - the bit at position
721 * (inode->pos) - is the one that will differ between
722 * left and right. So... we synthesize that bit in the
723 * two new keys.
724 * The mask 'm' below will be a single "one" bit at
725 * the position (inode->pos)
726 */
727
728 /* Use the old key, but set the new significant
729 * bit to zero.
730 */
731
732 left = (struct tnode *) tnode_get_child(tn, 2*i);
733 put_child(t, tn, 2*i, NULL);
734
735 BUG_ON(!left);
736
737 right = (struct tnode *) tnode_get_child(tn, 2*i+1);
738 put_child(t, tn, 2*i+1, NULL);
739
740 BUG_ON(!right);
741
742 size = tnode_child_length(left);
743 for (j = 0; j < size; j++) {
744 put_child(t, left, j, inode->child[j]);
745 put_child(t, right, j, inode->child[j + size]);
746 }
747 put_child(t, tn, 2*i, resize(t, left));
748 put_child(t, tn, 2*i+1, resize(t, right));
749
750 tnode_free(inode);
751 }
752 tnode_free(oldtnode);
753 return tn;
754 nomem:
755 {
756 int size = tnode_child_length(tn);
757 int j;
758
759 for (j = 0; j < size; j++)
760 if (tn->child[j])
761 tnode_free((struct tnode *)tn->child[j]);
762
763 tnode_free(tn);
764
765 return ERR_PTR(-ENOMEM);
766 }
767 }
768
769 static struct tnode *halve(struct trie *t, struct tnode *tn)
770 {
771 struct tnode *oldtnode = tn;
772 struct node *left, *right;
773 int i;
774 int olen = tnode_child_length(tn);
775
776 pr_debug("In halve\n");
777
778 tn = tnode_new(oldtnode->key, oldtnode->pos, oldtnode->bits - 1);
779
780 if (!tn)
781 return ERR_PTR(-ENOMEM);
782
783 /*
784 * Preallocate and store tnodes before the actual work so we
785 * don't get into an inconsistent state if memory allocation
786 * fails. In case of failure we return the oldnode and halve
787 * of tnode is ignored.
788 */
789
790 for (i = 0; i < olen; i += 2) {
791 left = tnode_get_child(oldtnode, i);
792 right = tnode_get_child(oldtnode, i+1);
793
794 /* Two nonempty children */
795 if (left && right) {
796 struct tnode *newn;
797
798 newn = tnode_new(left->key, tn->pos + tn->bits, 1);
799
800 if (!newn)
801 goto nomem;
802
803 put_child(t, tn, i/2, (struct node *)newn);
804 }
805
806 }
807
808 for (i = 0; i < olen; i += 2) {
809 struct tnode *newBinNode;
810
811 left = tnode_get_child(oldtnode, i);
812 right = tnode_get_child(oldtnode, i+1);
813
814 /* At least one of the children is empty */
815 if (left == NULL) {
816 if (right == NULL) /* Both are empty */
817 continue;
818 put_child(t, tn, i/2, right);
819 continue;
820 }
821
822 if (right == NULL) {
823 put_child(t, tn, i/2, left);
824 continue;
825 }
826
827 /* Two nonempty children */
828 newBinNode = (struct tnode *) tnode_get_child(tn, i/2);
829 put_child(t, tn, i/2, NULL);
830 put_child(t, newBinNode, 0, left);
831 put_child(t, newBinNode, 1, right);
832 put_child(t, tn, i/2, resize(t, newBinNode));
833 }
834 tnode_free(oldtnode);
835 return tn;
836 nomem:
837 {
838 int size = tnode_child_length(tn);
839 int j;
840
841 for (j = 0; j < size; j++)
842 if (tn->child[j])
843 tnode_free((struct tnode *)tn->child[j]);
844
845 tnode_free(tn);
846
847 return ERR_PTR(-ENOMEM);
848 }
849 }
850
851 static void trie_init(struct trie *t)
852 {
853 if (!t)
854 return;
855
856 t->size = 0;
857 rcu_assign_pointer(t->trie, NULL);
858 t->revision = 0;
859 #ifdef CONFIG_IP_FIB_TRIE_STATS
860 memset(&t->stats, 0, sizeof(struct trie_use_stats));
861 #endif
862 }
863
864 /* readside must use rcu_read_lock currently dump routines
865 via get_fa_head and dump */
866
867 static struct leaf_info *find_leaf_info(struct leaf *l, int plen)
868 {
869 struct hlist_head *head = &l->list;
870 struct hlist_node *node;
871 struct leaf_info *li;
872
873 hlist_for_each_entry_rcu(li, node, head, hlist)
874 if (li->plen == plen)
875 return li;
876
877 return NULL;
878 }
879
880 static inline struct list_head * get_fa_head(struct leaf *l, int plen)
881 {
882 struct leaf_info *li = find_leaf_info(l, plen);
883
884 if (!li)
885 return NULL;
886
887 return &li->falh;
888 }
889
890 static void insert_leaf_info(struct hlist_head *head, struct leaf_info *new)
891 {
892 struct leaf_info *li = NULL, *last = NULL;
893 struct hlist_node *node;
894
895 if (hlist_empty(head)) {
896 hlist_add_head_rcu(&new->hlist, head);
897 } else {
898 hlist_for_each_entry(li, node, head, hlist) {
899 if (new->plen > li->plen)
900 break;
901
902 last = li;
903 }
904 if (last)
905 hlist_add_after_rcu(&last->hlist, &new->hlist);
906 else
907 hlist_add_before_rcu(&new->hlist, &li->hlist);
908 }
909 }
910
911 /* rcu_read_lock needs to be hold by caller from readside */
912
913 static struct leaf *
914 fib_find_node(struct trie *t, u32 key)
915 {
916 int pos;
917 struct tnode *tn;
918 struct node *n;
919
920 pos = 0;
921 n = rcu_dereference(t->trie);
922
923 while (n != NULL && NODE_TYPE(n) == T_TNODE) {
924 tn = (struct tnode *) n;
925
926 check_tnode(tn);
927
928 if (tkey_sub_equals(tn->key, pos, tn->pos-pos, key)) {
929 pos = tn->pos + tn->bits;
930 n = tnode_get_child(tn, tkey_extract_bits(key, tn->pos, tn->bits));
931 } else
932 break;
933 }
934 /* Case we have found a leaf. Compare prefixes */
935
936 if (n != NULL && IS_LEAF(n) && tkey_equals(key, n->key))
937 return (struct leaf *)n;
938
939 return NULL;
940 }
941
942 static struct node *trie_rebalance(struct trie *t, struct tnode *tn)
943 {
944 int wasfull;
945 t_key cindex, key;
946 struct tnode *tp = NULL;
947
948 key = tn->key;
949
950 while (tn != NULL && NODE_PARENT(tn) != NULL) {
951
952 tp = NODE_PARENT(tn);
953 cindex = tkey_extract_bits(key, tp->pos, tp->bits);
954 wasfull = tnode_full(tp, tnode_get_child(tp, cindex));
955 tn = (struct tnode *) resize (t, (struct tnode *)tn);
956 tnode_put_child_reorg((struct tnode *)tp, cindex,(struct node*)tn, wasfull);
957
958 if (!NODE_PARENT(tn))
959 break;
960
961 tn = NODE_PARENT(tn);
962 }
963 /* Handle last (top) tnode */
964 if (IS_TNODE(tn))
965 tn = (struct tnode*) resize(t, (struct tnode *)tn);
966
967 return (struct node*) tn;
968 }
969
970 /* only used from updater-side */
971
972 static struct list_head *
973 fib_insert_node(struct trie *t, int *err, u32 key, int plen)
974 {
975 int pos, newpos;
976 struct tnode *tp = NULL, *tn = NULL;
977 struct node *n;
978 struct leaf *l;
979 int missbit;
980 struct list_head *fa_head = NULL;
981 struct leaf_info *li;
982 t_key cindex;
983
984 pos = 0;
985 n = t->trie;
986
987 /* If we point to NULL, stop. Either the tree is empty and we should
988 * just put a new leaf in if, or we have reached an empty child slot,
989 * and we should just put our new leaf in that.
990 * If we point to a T_TNODE, check if it matches our key. Note that
991 * a T_TNODE might be skipping any number of bits - its 'pos' need
992 * not be the parent's 'pos'+'bits'!
993 *
994 * If it does match the current key, get pos/bits from it, extract
995 * the index from our key, push the T_TNODE and walk the tree.
996 *
997 * If it doesn't, we have to replace it with a new T_TNODE.
998 *
999 * If we point to a T_LEAF, it might or might not have the same key
1000 * as we do. If it does, just change the value, update the T_LEAF's
1001 * value, and return it.
1002 * If it doesn't, we need to replace it with a T_TNODE.
1003 */
1004
1005 while (n != NULL && NODE_TYPE(n) == T_TNODE) {
1006 tn = (struct tnode *) n;
1007
1008 check_tnode(tn);
1009
1010 if (tkey_sub_equals(tn->key, pos, tn->pos-pos, key)) {
1011 tp = tn;
1012 pos = tn->pos + tn->bits;
1013 n = tnode_get_child(tn, tkey_extract_bits(key, tn->pos, tn->bits));
1014
1015 BUG_ON(n && NODE_PARENT(n) != tn);
1016 } else
1017 break;
1018 }
1019
1020 /*
1021 * n ----> NULL, LEAF or TNODE
1022 *
1023 * tp is n's (parent) ----> NULL or TNODE
1024 */
1025
1026 BUG_ON(tp && IS_LEAF(tp));
1027
1028 /* Case 1: n is a leaf. Compare prefixes */
1029
1030 if (n != NULL && IS_LEAF(n) && tkey_equals(key, n->key)) {
1031 struct leaf *l = (struct leaf *) n;
1032
1033 li = leaf_info_new(plen);
1034
1035 if (!li) {
1036 *err = -ENOMEM;
1037 goto err;
1038 }
1039
1040 fa_head = &li->falh;
1041 insert_leaf_info(&l->list, li);
1042 goto done;
1043 }
1044 t->size++;
1045 l = leaf_new();
1046
1047 if (!l) {
1048 *err = -ENOMEM;
1049 goto err;
1050 }
1051
1052 l->key = key;
1053 li = leaf_info_new(plen);
1054
1055 if (!li) {
1056 tnode_free((struct tnode *) l);
1057 *err = -ENOMEM;
1058 goto err;
1059 }
1060
1061 fa_head = &li->falh;
1062 insert_leaf_info(&l->list, li);
1063
1064 if (t->trie && n == NULL) {
1065 /* Case 2: n is NULL, and will just insert a new leaf */
1066
1067 NODE_SET_PARENT(l, tp);
1068
1069 cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1070 put_child(t, (struct tnode *)tp, cindex, (struct node *)l);
1071 } else {
1072 /* Case 3: n is a LEAF or a TNODE and the key doesn't match. */
1073 /*
1074 * Add a new tnode here
1075 * first tnode need some special handling
1076 */
1077
1078 if (tp)
1079 pos = tp->pos+tp->bits;
1080 else
1081 pos = 0;
1082
1083 if (n) {
1084 newpos = tkey_mismatch(key, pos, n->key);
1085 tn = tnode_new(n->key, newpos, 1);
1086 } else {
1087 newpos = 0;
1088 tn = tnode_new(key, newpos, 1); /* First tnode */
1089 }
1090
1091 if (!tn) {
1092 free_leaf_info(li);
1093 tnode_free((struct tnode *) l);
1094 *err = -ENOMEM;
1095 goto err;
1096 }
1097
1098 NODE_SET_PARENT(tn, tp);
1099
1100 missbit = tkey_extract_bits(key, newpos, 1);
1101 put_child(t, tn, missbit, (struct node *)l);
1102 put_child(t, tn, 1-missbit, n);
1103
1104 if (tp) {
1105 cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1106 put_child(t, (struct tnode *)tp, cindex, (struct node *)tn);
1107 } else {
1108 rcu_assign_pointer(t->trie, (struct node *)tn); /* First tnode */
1109 tp = tn;
1110 }
1111 }
1112
1113 if (tp && tp->pos + tp->bits > 32)
1114 printk(KERN_WARNING "fib_trie tp=%p pos=%d, bits=%d, key=%0x plen=%d\n",
1115 tp, tp->pos, tp->bits, key, plen);
1116
1117 /* Rebalance the trie */
1118
1119 rcu_assign_pointer(t->trie, trie_rebalance(t, tp));
1120 done:
1121 t->revision++;
1122 err:
1123 return fa_head;
1124 }
1125
1126 static int fn_trie_insert(struct fib_table *tb, struct fib_config *cfg)
1127 {
1128 struct trie *t = (struct trie *) tb->tb_data;
1129 struct fib_alias *fa, *new_fa;
1130 struct list_head *fa_head = NULL;
1131 struct fib_info *fi;
1132 int plen = cfg->fc_dst_len;
1133 u8 tos = cfg->fc_tos;
1134 u32 key, mask;
1135 int err;
1136 struct leaf *l;
1137
1138 if (plen > 32)
1139 return -EINVAL;
1140
1141 key = ntohl(cfg->fc_dst);
1142
1143 pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1144
1145 mask = ntohl(inet_make_mask(plen));
1146
1147 if (key & ~mask)
1148 return -EINVAL;
1149
1150 key = key & mask;
1151
1152 fi = fib_create_info(cfg);
1153 if (IS_ERR(fi)) {
1154 err = PTR_ERR(fi);
1155 goto err;
1156 }
1157
1158 l = fib_find_node(t, key);
1159 fa = NULL;
1160
1161 if (l) {
1162 fa_head = get_fa_head(l, plen);
1163 fa = fib_find_alias(fa_head, tos, fi->fib_priority);
1164 }
1165
1166 /* Now fa, if non-NULL, points to the first fib alias
1167 * with the same keys [prefix,tos,priority], if such key already
1168 * exists or to the node before which we will insert new one.
1169 *
1170 * If fa is NULL, we will need to allocate a new one and
1171 * insert to the head of f.
1172 *
1173 * If f is NULL, no fib node matched the destination key
1174 * and we need to allocate a new one of those as well.
1175 */
1176
1177 if (fa && fa->fa_info->fib_priority == fi->fib_priority) {
1178 struct fib_alias *fa_orig;
1179
1180 err = -EEXIST;
1181 if (cfg->fc_nlflags & NLM_F_EXCL)
1182 goto out;
1183
1184 if (cfg->fc_nlflags & NLM_F_REPLACE) {
1185 struct fib_info *fi_drop;
1186 u8 state;
1187
1188 err = -ENOBUFS;
1189 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1190 if (new_fa == NULL)
1191 goto out;
1192
1193 fi_drop = fa->fa_info;
1194 new_fa->fa_tos = fa->fa_tos;
1195 new_fa->fa_info = fi;
1196 new_fa->fa_type = cfg->fc_type;
1197 new_fa->fa_scope = cfg->fc_scope;
1198 state = fa->fa_state;
1199 new_fa->fa_state &= ~FA_S_ACCESSED;
1200
1201 list_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1202 alias_free_mem_rcu(fa);
1203
1204 fib_release_info(fi_drop);
1205 if (state & FA_S_ACCESSED)
1206 rt_cache_flush(-1);
1207
1208 goto succeeded;
1209 }
1210 /* Error if we find a perfect match which
1211 * uses the same scope, type, and nexthop
1212 * information.
1213 */
1214 fa_orig = fa;
1215 list_for_each_entry(fa, fa_orig->fa_list.prev, fa_list) {
1216 if (fa->fa_tos != tos)
1217 break;
1218 if (fa->fa_info->fib_priority != fi->fib_priority)
1219 break;
1220 if (fa->fa_type == cfg->fc_type &&
1221 fa->fa_scope == cfg->fc_scope &&
1222 fa->fa_info == fi) {
1223 goto out;
1224 }
1225 }
1226 if (!(cfg->fc_nlflags & NLM_F_APPEND))
1227 fa = fa_orig;
1228 }
1229 err = -ENOENT;
1230 if (!(cfg->fc_nlflags & NLM_F_CREATE))
1231 goto out;
1232
1233 err = -ENOBUFS;
1234 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1235 if (new_fa == NULL)
1236 goto out;
1237
1238 new_fa->fa_info = fi;
1239 new_fa->fa_tos = tos;
1240 new_fa->fa_type = cfg->fc_type;
1241 new_fa->fa_scope = cfg->fc_scope;
1242 new_fa->fa_state = 0;
1243 /*
1244 * Insert new entry to the list.
1245 */
1246
1247 if (!fa_head) {
1248 err = 0;
1249 fa_head = fib_insert_node(t, &err, key, plen);
1250 if (err)
1251 goto out_free_new_fa;
1252 }
1253
1254 list_add_tail_rcu(&new_fa->fa_list,
1255 (fa ? &fa->fa_list : fa_head));
1256
1257 rt_cache_flush(-1);
1258 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, tb->tb_id,
1259 &cfg->fc_nlinfo);
1260 succeeded:
1261 return 0;
1262
1263 out_free_new_fa:
1264 kmem_cache_free(fn_alias_kmem, new_fa);
1265 out:
1266 fib_release_info(fi);
1267 err:
1268 return err;
1269 }
1270
1271
1272 /* should be called with rcu_read_lock */
1273 static inline int check_leaf(struct trie *t, struct leaf *l,
1274 t_key key, int *plen, const struct flowi *flp,
1275 struct fib_result *res)
1276 {
1277 int err, i;
1278 __be32 mask;
1279 struct leaf_info *li;
1280 struct hlist_head *hhead = &l->list;
1281 struct hlist_node *node;
1282
1283 hlist_for_each_entry_rcu(li, node, hhead, hlist) {
1284 i = li->plen;
1285 mask = inet_make_mask(i);
1286 if (l->key != (key & ntohl(mask)))
1287 continue;
1288
1289 if ((err = fib_semantic_match(&li->falh, flp, res, htonl(l->key), mask, i)) <= 0) {
1290 *plen = i;
1291 #ifdef CONFIG_IP_FIB_TRIE_STATS
1292 t->stats.semantic_match_passed++;
1293 #endif
1294 return err;
1295 }
1296 #ifdef CONFIG_IP_FIB_TRIE_STATS
1297 t->stats.semantic_match_miss++;
1298 #endif
1299 }
1300 return 1;
1301 }
1302
1303 static int
1304 fn_trie_lookup(struct fib_table *tb, const struct flowi *flp, struct fib_result *res)
1305 {
1306 struct trie *t = (struct trie *) tb->tb_data;
1307 int plen, ret = 0;
1308 struct node *n;
1309 struct tnode *pn;
1310 int pos, bits;
1311 t_key key = ntohl(flp->fl4_dst);
1312 int chopped_off;
1313 t_key cindex = 0;
1314 int current_prefix_length = KEYLENGTH;
1315 struct tnode *cn;
1316 t_key node_prefix, key_prefix, pref_mismatch;
1317 int mp;
1318
1319 rcu_read_lock();
1320
1321 n = rcu_dereference(t->trie);
1322 if (!n)
1323 goto failed;
1324
1325 #ifdef CONFIG_IP_FIB_TRIE_STATS
1326 t->stats.gets++;
1327 #endif
1328
1329 /* Just a leaf? */
1330 if (IS_LEAF(n)) {
1331 if ((ret = check_leaf(t, (struct leaf *)n, key, &plen, flp, res)) <= 0)
1332 goto found;
1333 goto failed;
1334 }
1335 pn = (struct tnode *) n;
1336 chopped_off = 0;
1337
1338 while (pn) {
1339 pos = pn->pos;
1340 bits = pn->bits;
1341
1342 if (!chopped_off)
1343 cindex = tkey_extract_bits(MASK_PFX(key, current_prefix_length), pos, bits);
1344
1345 n = tnode_get_child(pn, cindex);
1346
1347 if (n == NULL) {
1348 #ifdef CONFIG_IP_FIB_TRIE_STATS
1349 t->stats.null_node_hit++;
1350 #endif
1351 goto backtrace;
1352 }
1353
1354 if (IS_LEAF(n)) {
1355 if ((ret = check_leaf(t, (struct leaf *)n, key, &plen, flp, res)) <= 0)
1356 goto found;
1357 else
1358 goto backtrace;
1359 }
1360
1361 #define HL_OPTIMIZE
1362 #ifdef HL_OPTIMIZE
1363 cn = (struct tnode *)n;
1364
1365 /*
1366 * It's a tnode, and we can do some extra checks here if we
1367 * like, to avoid descending into a dead-end branch.
1368 * This tnode is in the parent's child array at index
1369 * key[p_pos..p_pos+p_bits] but potentially with some bits
1370 * chopped off, so in reality the index may be just a
1371 * subprefix, padded with zero at the end.
1372 * We can also take a look at any skipped bits in this
1373 * tnode - everything up to p_pos is supposed to be ok,
1374 * and the non-chopped bits of the index (se previous
1375 * paragraph) are also guaranteed ok, but the rest is
1376 * considered unknown.
1377 *
1378 * The skipped bits are key[pos+bits..cn->pos].
1379 */
1380
1381 /* If current_prefix_length < pos+bits, we are already doing
1382 * actual prefix matching, which means everything from
1383 * pos+(bits-chopped_off) onward must be zero along some
1384 * branch of this subtree - otherwise there is *no* valid
1385 * prefix present. Here we can only check the skipped
1386 * bits. Remember, since we have already indexed into the
1387 * parent's child array, we know that the bits we chopped of
1388 * *are* zero.
1389 */
1390
1391 /* NOTA BENE: CHECKING ONLY SKIPPED BITS FOR THE NEW NODE HERE */
1392
1393 if (current_prefix_length < pos+bits) {
1394 if (tkey_extract_bits(cn->key, current_prefix_length,
1395 cn->pos - current_prefix_length) != 0 ||
1396 !(cn->child[0]))
1397 goto backtrace;
1398 }
1399
1400 /*
1401 * If chopped_off=0, the index is fully validated and we
1402 * only need to look at the skipped bits for this, the new,
1403 * tnode. What we actually want to do is to find out if
1404 * these skipped bits match our key perfectly, or if we will
1405 * have to count on finding a matching prefix further down,
1406 * because if we do, we would like to have some way of
1407 * verifying the existence of such a prefix at this point.
1408 */
1409
1410 /* The only thing we can do at this point is to verify that
1411 * any such matching prefix can indeed be a prefix to our
1412 * key, and if the bits in the node we are inspecting that
1413 * do not match our key are not ZERO, this cannot be true.
1414 * Thus, find out where there is a mismatch (before cn->pos)
1415 * and verify that all the mismatching bits are zero in the
1416 * new tnode's key.
1417 */
1418
1419 /* Note: We aren't very concerned about the piece of the key
1420 * that precede pn->pos+pn->bits, since these have already been
1421 * checked. The bits after cn->pos aren't checked since these are
1422 * by definition "unknown" at this point. Thus, what we want to
1423 * see is if we are about to enter the "prefix matching" state,
1424 * and in that case verify that the skipped bits that will prevail
1425 * throughout this subtree are zero, as they have to be if we are
1426 * to find a matching prefix.
1427 */
1428
1429 node_prefix = MASK_PFX(cn->key, cn->pos);
1430 key_prefix = MASK_PFX(key, cn->pos);
1431 pref_mismatch = key_prefix^node_prefix;
1432 mp = 0;
1433
1434 /* In short: If skipped bits in this node do not match the search
1435 * key, enter the "prefix matching" state.directly.
1436 */
1437 if (pref_mismatch) {
1438 while (!(pref_mismatch & (1<<(KEYLENGTH-1)))) {
1439 mp++;
1440 pref_mismatch = pref_mismatch <<1;
1441 }
1442 key_prefix = tkey_extract_bits(cn->key, mp, cn->pos-mp);
1443
1444 if (key_prefix != 0)
1445 goto backtrace;
1446
1447 if (current_prefix_length >= cn->pos)
1448 current_prefix_length = mp;
1449 }
1450 #endif
1451 pn = (struct tnode *)n; /* Descend */
1452 chopped_off = 0;
1453 continue;
1454
1455 backtrace:
1456 chopped_off++;
1457
1458 /* As zero don't change the child key (cindex) */
1459 while ((chopped_off <= pn->bits) && !(cindex & (1<<(chopped_off-1))))
1460 chopped_off++;
1461
1462 /* Decrease current_... with bits chopped off */
1463 if (current_prefix_length > pn->pos + pn->bits - chopped_off)
1464 current_prefix_length = pn->pos + pn->bits - chopped_off;
1465
1466 /*
1467 * Either we do the actual chop off according or if we have
1468 * chopped off all bits in this tnode walk up to our parent.
1469 */
1470
1471 if (chopped_off <= pn->bits) {
1472 cindex &= ~(1 << (chopped_off-1));
1473 } else {
1474 if (NODE_PARENT(pn) == NULL)
1475 goto failed;
1476
1477 /* Get Child's index */
1478 cindex = tkey_extract_bits(pn->key, NODE_PARENT(pn)->pos, NODE_PARENT(pn)->bits);
1479 pn = NODE_PARENT(pn);
1480 chopped_off = 0;
1481
1482 #ifdef CONFIG_IP_FIB_TRIE_STATS
1483 t->stats.backtrack++;
1484 #endif
1485 goto backtrace;
1486 }
1487 }
1488 failed:
1489 ret = 1;
1490 found:
1491 rcu_read_unlock();
1492 return ret;
1493 }
1494
1495 /* only called from updater side */
1496 static int trie_leaf_remove(struct trie *t, t_key key)
1497 {
1498 t_key cindex;
1499 struct tnode *tp = NULL;
1500 struct node *n = t->trie;
1501 struct leaf *l;
1502
1503 pr_debug("entering trie_leaf_remove(%p)\n", n);
1504
1505 /* Note that in the case skipped bits, those bits are *not* checked!
1506 * When we finish this, we will have NULL or a T_LEAF, and the
1507 * T_LEAF may or may not match our key.
1508 */
1509
1510 while (n != NULL && IS_TNODE(n)) {
1511 struct tnode *tn = (struct tnode *) n;
1512 check_tnode(tn);
1513 n = tnode_get_child(tn ,tkey_extract_bits(key, tn->pos, tn->bits));
1514
1515 BUG_ON(n && NODE_PARENT(n) != tn);
1516 }
1517 l = (struct leaf *) n;
1518
1519 if (!n || !tkey_equals(l->key, key))
1520 return 0;
1521
1522 /*
1523 * Key found.
1524 * Remove the leaf and rebalance the tree
1525 */
1526
1527 t->revision++;
1528 t->size--;
1529
1530 tp = NODE_PARENT(n);
1531 tnode_free((struct tnode *) n);
1532
1533 if (tp) {
1534 cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1535 put_child(t, (struct tnode *)tp, cindex, NULL);
1536 rcu_assign_pointer(t->trie, trie_rebalance(t, tp));
1537 } else
1538 rcu_assign_pointer(t->trie, NULL);
1539
1540 return 1;
1541 }
1542
1543 static int fn_trie_delete(struct fib_table *tb, struct fib_config *cfg)
1544 {
1545 struct trie *t = (struct trie *) tb->tb_data;
1546 u32 key, mask;
1547 int plen = cfg->fc_dst_len;
1548 u8 tos = cfg->fc_tos;
1549 struct fib_alias *fa, *fa_to_delete;
1550 struct list_head *fa_head;
1551 struct leaf *l;
1552 struct leaf_info *li;
1553
1554 if (plen > 32)
1555 return -EINVAL;
1556
1557 key = ntohl(cfg->fc_dst);
1558 mask = ntohl(inet_make_mask(plen));
1559
1560 if (key & ~mask)
1561 return -EINVAL;
1562
1563 key = key & mask;
1564 l = fib_find_node(t, key);
1565
1566 if (!l)
1567 return -ESRCH;
1568
1569 fa_head = get_fa_head(l, plen);
1570 fa = fib_find_alias(fa_head, tos, 0);
1571
1572 if (!fa)
1573 return -ESRCH;
1574
1575 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1576
1577 fa_to_delete = NULL;
1578 fa_head = fa->fa_list.prev;
1579
1580 list_for_each_entry(fa, fa_head, fa_list) {
1581 struct fib_info *fi = fa->fa_info;
1582
1583 if (fa->fa_tos != tos)
1584 break;
1585
1586 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1587 (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1588 fa->fa_scope == cfg->fc_scope) &&
1589 (!cfg->fc_protocol ||
1590 fi->fib_protocol == cfg->fc_protocol) &&
1591 fib_nh_match(cfg, fi) == 0) {
1592 fa_to_delete = fa;
1593 break;
1594 }
1595 }
1596
1597 if (!fa_to_delete)
1598 return -ESRCH;
1599
1600 fa = fa_to_delete;
1601 rtmsg_fib(RTM_DELROUTE, htonl(key), fa, plen, tb->tb_id,
1602 &cfg->fc_nlinfo);
1603
1604 l = fib_find_node(t, key);
1605 li = find_leaf_info(l, plen);
1606
1607 list_del_rcu(&fa->fa_list);
1608
1609 if (list_empty(fa_head)) {
1610 hlist_del_rcu(&li->hlist);
1611 free_leaf_info(li);
1612 }
1613
1614 if (hlist_empty(&l->list))
1615 trie_leaf_remove(t, key);
1616
1617 if (fa->fa_state & FA_S_ACCESSED)
1618 rt_cache_flush(-1);
1619
1620 fib_release_info(fa->fa_info);
1621 alias_free_mem_rcu(fa);
1622 return 0;
1623 }
1624
1625 static int trie_flush_list(struct trie *t, struct list_head *head)
1626 {
1627 struct fib_alias *fa, *fa_node;
1628 int found = 0;
1629
1630 list_for_each_entry_safe(fa, fa_node, head, fa_list) {
1631 struct fib_info *fi = fa->fa_info;
1632
1633 if (fi && (fi->fib_flags & RTNH_F_DEAD)) {
1634 list_del_rcu(&fa->fa_list);
1635 fib_release_info(fa->fa_info);
1636 alias_free_mem_rcu(fa);
1637 found++;
1638 }
1639 }
1640 return found;
1641 }
1642
1643 static int trie_flush_leaf(struct trie *t, struct leaf *l)
1644 {
1645 int found = 0;
1646 struct hlist_head *lih = &l->list;
1647 struct hlist_node *node, *tmp;
1648 struct leaf_info *li = NULL;
1649
1650 hlist_for_each_entry_safe(li, node, tmp, lih, hlist) {
1651 found += trie_flush_list(t, &li->falh);
1652
1653 if (list_empty(&li->falh)) {
1654 hlist_del_rcu(&li->hlist);
1655 free_leaf_info(li);
1656 }
1657 }
1658 return found;
1659 }
1660
1661 /* rcu_read_lock needs to be hold by caller from readside */
1662
1663 static struct leaf *nextleaf(struct trie *t, struct leaf *thisleaf)
1664 {
1665 struct node *c = (struct node *) thisleaf;
1666 struct tnode *p;
1667 int idx;
1668 struct node *trie = rcu_dereference(t->trie);
1669
1670 if (c == NULL) {
1671 if (trie == NULL)
1672 return NULL;
1673
1674 if (IS_LEAF(trie)) /* trie w. just a leaf */
1675 return (struct leaf *) trie;
1676
1677 p = (struct tnode*) trie; /* Start */
1678 } else
1679 p = (struct tnode *) NODE_PARENT(c);
1680
1681 while (p) {
1682 int pos, last;
1683
1684 /* Find the next child of the parent */
1685 if (c)
1686 pos = 1 + tkey_extract_bits(c->key, p->pos, p->bits);
1687 else
1688 pos = 0;
1689
1690 last = 1 << p->bits;
1691 for (idx = pos; idx < last ; idx++) {
1692 c = rcu_dereference(p->child[idx]);
1693
1694 if (!c)
1695 continue;
1696
1697 /* Decend if tnode */
1698 while (IS_TNODE(c)) {
1699 p = (struct tnode *) c;
1700 idx = 0;
1701
1702 /* Rightmost non-NULL branch */
1703 if (p && IS_TNODE(p))
1704 while (!(c = rcu_dereference(p->child[idx]))
1705 && idx < (1<<p->bits)) idx++;
1706
1707 /* Done with this tnode? */
1708 if (idx >= (1 << p->bits) || !c)
1709 goto up;
1710 }
1711 return (struct leaf *) c;
1712 }
1713 up:
1714 /* No more children go up one step */
1715 c = (struct node *) p;
1716 p = (struct tnode *) NODE_PARENT(p);
1717 }
1718 return NULL; /* Ready. Root of trie */
1719 }
1720
1721 static int fn_trie_flush(struct fib_table *tb)
1722 {
1723 struct trie *t = (struct trie *) tb->tb_data;
1724 struct leaf *ll = NULL, *l = NULL;
1725 int found = 0, h;
1726
1727 t->revision++;
1728
1729 for (h = 0; (l = nextleaf(t, l)) != NULL; h++) {
1730 found += trie_flush_leaf(t, l);
1731
1732 if (ll && hlist_empty(&ll->list))
1733 trie_leaf_remove(t, ll->key);
1734 ll = l;
1735 }
1736
1737 if (ll && hlist_empty(&ll->list))
1738 trie_leaf_remove(t, ll->key);
1739
1740 pr_debug("trie_flush found=%d\n", found);
1741 return found;
1742 }
1743
1744 static int trie_last_dflt = -1;
1745
1746 static void
1747 fn_trie_select_default(struct fib_table *tb, const struct flowi *flp, struct fib_result *res)
1748 {
1749 struct trie *t = (struct trie *) tb->tb_data;
1750 int order, last_idx;
1751 struct fib_info *fi = NULL;
1752 struct fib_info *last_resort;
1753 struct fib_alias *fa = NULL;
1754 struct list_head *fa_head;
1755 struct leaf *l;
1756
1757 last_idx = -1;
1758 last_resort = NULL;
1759 order = -1;
1760
1761 rcu_read_lock();
1762
1763 l = fib_find_node(t, 0);
1764 if (!l)
1765 goto out;
1766
1767 fa_head = get_fa_head(l, 0);
1768 if (!fa_head)
1769 goto out;
1770
1771 if (list_empty(fa_head))
1772 goto out;
1773
1774 list_for_each_entry_rcu(fa, fa_head, fa_list) {
1775 struct fib_info *next_fi = fa->fa_info;
1776
1777 if (fa->fa_scope != res->scope ||
1778 fa->fa_type != RTN_UNICAST)
1779 continue;
1780
1781 if (next_fi->fib_priority > res->fi->fib_priority)
1782 break;
1783 if (!next_fi->fib_nh[0].nh_gw ||
1784 next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1785 continue;
1786 fa->fa_state |= FA_S_ACCESSED;
1787
1788 if (fi == NULL) {
1789 if (next_fi != res->fi)
1790 break;
1791 } else if (!fib_detect_death(fi, order, &last_resort,
1792 &last_idx, &trie_last_dflt)) {
1793 if (res->fi)
1794 fib_info_put(res->fi);
1795 res->fi = fi;
1796 atomic_inc(&fi->fib_clntref);
1797 trie_last_dflt = order;
1798 goto out;
1799 }
1800 fi = next_fi;
1801 order++;
1802 }
1803 if (order <= 0 || fi == NULL) {
1804 trie_last_dflt = -1;
1805 goto out;
1806 }
1807
1808 if (!fib_detect_death(fi, order, &last_resort, &last_idx, &trie_last_dflt)) {
1809 if (res->fi)
1810 fib_info_put(res->fi);
1811 res->fi = fi;
1812 atomic_inc(&fi->fib_clntref);
1813 trie_last_dflt = order;
1814 goto out;
1815 }
1816 if (last_idx >= 0) {
1817 if (res->fi)
1818 fib_info_put(res->fi);
1819 res->fi = last_resort;
1820 if (last_resort)
1821 atomic_inc(&last_resort->fib_clntref);
1822 }
1823 trie_last_dflt = last_idx;
1824 out:;
1825 rcu_read_unlock();
1826 }
1827
1828 static int fn_trie_dump_fa(t_key key, int plen, struct list_head *fah, struct fib_table *tb,
1829 struct sk_buff *skb, struct netlink_callback *cb)
1830 {
1831 int i, s_i;
1832 struct fib_alias *fa;
1833
1834 __be32 xkey = htonl(key);
1835
1836 s_i = cb->args[4];
1837 i = 0;
1838
1839 /* rcu_read_lock is hold by caller */
1840
1841 list_for_each_entry_rcu(fa, fah, fa_list) {
1842 if (i < s_i) {
1843 i++;
1844 continue;
1845 }
1846 BUG_ON(!fa->fa_info);
1847
1848 if (fib_dump_info(skb, NETLINK_CB(cb->skb).pid,
1849 cb->nlh->nlmsg_seq,
1850 RTM_NEWROUTE,
1851 tb->tb_id,
1852 fa->fa_type,
1853 fa->fa_scope,
1854 xkey,
1855 plen,
1856 fa->fa_tos,
1857 fa->fa_info, 0) < 0) {
1858 cb->args[4] = i;
1859 return -1;
1860 }
1861 i++;
1862 }
1863 cb->args[4] = i;
1864 return skb->len;
1865 }
1866
1867 static int fn_trie_dump_plen(struct trie *t, int plen, struct fib_table *tb, struct sk_buff *skb,
1868 struct netlink_callback *cb)
1869 {
1870 int h, s_h;
1871 struct list_head *fa_head;
1872 struct leaf *l = NULL;
1873
1874 s_h = cb->args[3];
1875
1876 for (h = 0; (l = nextleaf(t, l)) != NULL; h++) {
1877 if (h < s_h)
1878 continue;
1879 if (h > s_h)
1880 memset(&cb->args[4], 0,
1881 sizeof(cb->args) - 4*sizeof(cb->args[0]));
1882
1883 fa_head = get_fa_head(l, plen);
1884
1885 if (!fa_head)
1886 continue;
1887
1888 if (list_empty(fa_head))
1889 continue;
1890
1891 if (fn_trie_dump_fa(l->key, plen, fa_head, tb, skb, cb)<0) {
1892 cb->args[3] = h;
1893 return -1;
1894 }
1895 }
1896 cb->args[3] = h;
1897 return skb->len;
1898 }
1899
1900 static int fn_trie_dump(struct fib_table *tb, struct sk_buff *skb, struct netlink_callback *cb)
1901 {
1902 int m, s_m;
1903 struct trie *t = (struct trie *) tb->tb_data;
1904
1905 s_m = cb->args[2];
1906
1907 rcu_read_lock();
1908 for (m = 0; m <= 32; m++) {
1909 if (m < s_m)
1910 continue;
1911 if (m > s_m)
1912 memset(&cb->args[3], 0,
1913 sizeof(cb->args) - 3*sizeof(cb->args[0]));
1914
1915 if (fn_trie_dump_plen(t, 32-m, tb, skb, cb)<0) {
1916 cb->args[2] = m;
1917 goto out;
1918 }
1919 }
1920 rcu_read_unlock();
1921 cb->args[2] = m;
1922 return skb->len;
1923 out:
1924 rcu_read_unlock();
1925 return -1;
1926 }
1927
1928 /* Fix more generic FIB names for init later */
1929
1930 #ifdef CONFIG_IP_MULTIPLE_TABLES
1931 struct fib_table * fib_hash_init(u32 id)
1932 #else
1933 struct fib_table * __init fib_hash_init(u32 id)
1934 #endif
1935 {
1936 struct fib_table *tb;
1937 struct trie *t;
1938
1939 if (fn_alias_kmem == NULL)
1940 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
1941 sizeof(struct fib_alias),
1942 0, SLAB_HWCACHE_ALIGN,
1943 NULL, NULL);
1944
1945 tb = kmalloc(sizeof(struct fib_table) + sizeof(struct trie),
1946 GFP_KERNEL);
1947 if (tb == NULL)
1948 return NULL;
1949
1950 tb->tb_id = id;
1951 tb->tb_lookup = fn_trie_lookup;
1952 tb->tb_insert = fn_trie_insert;
1953 tb->tb_delete = fn_trie_delete;
1954 tb->tb_flush = fn_trie_flush;
1955 tb->tb_select_default = fn_trie_select_default;
1956 tb->tb_dump = fn_trie_dump;
1957 memset(tb->tb_data, 0, sizeof(struct trie));
1958
1959 t = (struct trie *) tb->tb_data;
1960
1961 trie_init(t);
1962
1963 if (id == RT_TABLE_LOCAL)
1964 trie_local = t;
1965 else if (id == RT_TABLE_MAIN)
1966 trie_main = t;
1967
1968 if (id == RT_TABLE_LOCAL)
1969 printk(KERN_INFO "IPv4 FIB: Using LC-trie version %s\n", VERSION);
1970
1971 return tb;
1972 }
1973
1974 #ifdef CONFIG_PROC_FS
1975 /* Depth first Trie walk iterator */
1976 struct fib_trie_iter {
1977 struct tnode *tnode;
1978 struct trie *trie;
1979 unsigned index;
1980 unsigned depth;
1981 };
1982
1983 static struct node *fib_trie_get_next(struct fib_trie_iter *iter)
1984 {
1985 struct tnode *tn = iter->tnode;
1986 unsigned cindex = iter->index;
1987 struct tnode *p;
1988
1989 /* A single entry routing table */
1990 if (!tn)
1991 return NULL;
1992
1993 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
1994 iter->tnode, iter->index, iter->depth);
1995 rescan:
1996 while (cindex < (1<<tn->bits)) {
1997 struct node *n = tnode_get_child(tn, cindex);
1998
1999 if (n) {
2000 if (IS_LEAF(n)) {
2001 iter->tnode = tn;
2002 iter->index = cindex + 1;
2003 } else {
2004 /* push down one level */
2005 iter->tnode = (struct tnode *) n;
2006 iter->index = 0;
2007 ++iter->depth;
2008 }
2009 return n;
2010 }
2011
2012 ++cindex;
2013 }
2014
2015 /* Current node exhausted, pop back up */
2016 p = NODE_PARENT(tn);
2017 if (p) {
2018 cindex = tkey_extract_bits(tn->key, p->pos, p->bits)+1;
2019 tn = p;
2020 --iter->depth;
2021 goto rescan;
2022 }
2023
2024 /* got root? */
2025 return NULL;
2026 }
2027
2028 static struct node *fib_trie_get_first(struct fib_trie_iter *iter,
2029 struct trie *t)
2030 {
2031 struct node *n ;
2032
2033 if(!t)
2034 return NULL;
2035
2036 n = rcu_dereference(t->trie);
2037
2038 if(!iter)
2039 return NULL;
2040
2041 if (n) {
2042 if (IS_TNODE(n)) {
2043 iter->tnode = (struct tnode *) n;
2044 iter->trie = t;
2045 iter->index = 0;
2046 iter->depth = 1;
2047 } else {
2048 iter->tnode = NULL;
2049 iter->trie = t;
2050 iter->index = 0;
2051 iter->depth = 0;
2052 }
2053 return n;
2054 }
2055 return NULL;
2056 }
2057
2058 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2059 {
2060 struct node *n;
2061 struct fib_trie_iter iter;
2062
2063 memset(s, 0, sizeof(*s));
2064
2065 rcu_read_lock();
2066 for (n = fib_trie_get_first(&iter, t); n;
2067 n = fib_trie_get_next(&iter)) {
2068 if (IS_LEAF(n)) {
2069 s->leaves++;
2070 s->totdepth += iter.depth;
2071 if (iter.depth > s->maxdepth)
2072 s->maxdepth = iter.depth;
2073 } else {
2074 const struct tnode *tn = (const struct tnode *) n;
2075 int i;
2076
2077 s->tnodes++;
2078 if(tn->bits < MAX_STAT_DEPTH)
2079 s->nodesizes[tn->bits]++;
2080
2081 for (i = 0; i < (1<<tn->bits); i++)
2082 if (!tn->child[i])
2083 s->nullpointers++;
2084 }
2085 }
2086 rcu_read_unlock();
2087 }
2088
2089 /*
2090 * This outputs /proc/net/fib_triestats
2091 */
2092 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2093 {
2094 unsigned i, max, pointers, bytes, avdepth;
2095
2096 if (stat->leaves)
2097 avdepth = stat->totdepth*100 / stat->leaves;
2098 else
2099 avdepth = 0;
2100
2101 seq_printf(seq, "\tAver depth: %d.%02d\n", avdepth / 100, avdepth % 100 );
2102 seq_printf(seq, "\tMax depth: %u\n", stat->maxdepth);
2103
2104 seq_printf(seq, "\tLeaves: %u\n", stat->leaves);
2105
2106 bytes = sizeof(struct leaf) * stat->leaves;
2107 seq_printf(seq, "\tInternal nodes: %d\n\t", stat->tnodes);
2108 bytes += sizeof(struct tnode) * stat->tnodes;
2109
2110 max = MAX_STAT_DEPTH;
2111 while (max > 0 && stat->nodesizes[max-1] == 0)
2112 max--;
2113
2114 pointers = 0;
2115 for (i = 1; i <= max; i++)
2116 if (stat->nodesizes[i] != 0) {
2117 seq_printf(seq, " %d: %d", i, stat->nodesizes[i]);
2118 pointers += (1<<i) * stat->nodesizes[i];
2119 }
2120 seq_putc(seq, '\n');
2121 seq_printf(seq, "\tPointers: %d\n", pointers);
2122
2123 bytes += sizeof(struct node *) * pointers;
2124 seq_printf(seq, "Null ptrs: %d\n", stat->nullpointers);
2125 seq_printf(seq, "Total size: %d kB\n", (bytes + 1023) / 1024);
2126
2127 #ifdef CONFIG_IP_FIB_TRIE_STATS
2128 seq_printf(seq, "Counters:\n---------\n");
2129 seq_printf(seq,"gets = %d\n", t->stats.gets);
2130 seq_printf(seq,"backtracks = %d\n", t->stats.backtrack);
2131 seq_printf(seq,"semantic match passed = %d\n", t->stats.semantic_match_passed);
2132 seq_printf(seq,"semantic match miss = %d\n", t->stats.semantic_match_miss);
2133 seq_printf(seq,"null node hit= %d\n", t->stats.null_node_hit);
2134 seq_printf(seq,"skipped node resize = %d\n", t->stats.resize_node_skipped);
2135 #ifdef CLEAR_STATS
2136 memset(&(t->stats), 0, sizeof(t->stats));
2137 #endif
2138 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2139 }
2140
2141 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2142 {
2143 struct trie_stat *stat;
2144
2145 stat = kmalloc(sizeof(*stat), GFP_KERNEL);
2146 if (!stat)
2147 return -ENOMEM;
2148
2149 seq_printf(seq, "Basic info: size of leaf: %Zd bytes, size of tnode: %Zd bytes.\n",
2150 sizeof(struct leaf), sizeof(struct tnode));
2151
2152 if (trie_local) {
2153 seq_printf(seq, "Local:\n");
2154 trie_collect_stats(trie_local, stat);
2155 trie_show_stats(seq, stat);
2156 }
2157
2158 if (trie_main) {
2159 seq_printf(seq, "Main:\n");
2160 trie_collect_stats(trie_main, stat);
2161 trie_show_stats(seq, stat);
2162 }
2163 kfree(stat);
2164
2165 return 0;
2166 }
2167
2168 static int fib_triestat_seq_open(struct inode *inode, struct file *file)
2169 {
2170 return single_open(file, fib_triestat_seq_show, NULL);
2171 }
2172
2173 static const struct file_operations fib_triestat_fops = {
2174 .owner = THIS_MODULE,
2175 .open = fib_triestat_seq_open,
2176 .read = seq_read,
2177 .llseek = seq_lseek,
2178 .release = single_release,
2179 };
2180
2181 static struct node *fib_trie_get_idx(struct fib_trie_iter *iter,
2182 loff_t pos)
2183 {
2184 loff_t idx = 0;
2185 struct node *n;
2186
2187 for (n = fib_trie_get_first(iter, trie_local);
2188 n; ++idx, n = fib_trie_get_next(iter)) {
2189 if (pos == idx)
2190 return n;
2191 }
2192
2193 for (n = fib_trie_get_first(iter, trie_main);
2194 n; ++idx, n = fib_trie_get_next(iter)) {
2195 if (pos == idx)
2196 return n;
2197 }
2198 return NULL;
2199 }
2200
2201 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2202 {
2203 rcu_read_lock();
2204 if (*pos == 0)
2205 return SEQ_START_TOKEN;
2206 return fib_trie_get_idx(seq->private, *pos - 1);
2207 }
2208
2209 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2210 {
2211 struct fib_trie_iter *iter = seq->private;
2212 void *l = v;
2213
2214 ++*pos;
2215 if (v == SEQ_START_TOKEN)
2216 return fib_trie_get_idx(iter, 0);
2217
2218 v = fib_trie_get_next(iter);
2219 BUG_ON(v == l);
2220 if (v)
2221 return v;
2222
2223 /* continue scan in next trie */
2224 if (iter->trie == trie_local)
2225 return fib_trie_get_first(iter, trie_main);
2226
2227 return NULL;
2228 }
2229
2230 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2231 {
2232 rcu_read_unlock();
2233 }
2234
2235 static void seq_indent(struct seq_file *seq, int n)
2236 {
2237 while (n-- > 0) seq_puts(seq, " ");
2238 }
2239
2240 static inline const char *rtn_scope(enum rt_scope_t s)
2241 {
2242 static char buf[32];
2243
2244 switch(s) {
2245 case RT_SCOPE_UNIVERSE: return "universe";
2246 case RT_SCOPE_SITE: return "site";
2247 case RT_SCOPE_LINK: return "link";
2248 case RT_SCOPE_HOST: return "host";
2249 case RT_SCOPE_NOWHERE: return "nowhere";
2250 default:
2251 snprintf(buf, sizeof(buf), "scope=%d", s);
2252 return buf;
2253 }
2254 }
2255
2256 static const char *rtn_type_names[__RTN_MAX] = {
2257 [RTN_UNSPEC] = "UNSPEC",
2258 [RTN_UNICAST] = "UNICAST",
2259 [RTN_LOCAL] = "LOCAL",
2260 [RTN_BROADCAST] = "BROADCAST",
2261 [RTN_ANYCAST] = "ANYCAST",
2262 [RTN_MULTICAST] = "MULTICAST",
2263 [RTN_BLACKHOLE] = "BLACKHOLE",
2264 [RTN_UNREACHABLE] = "UNREACHABLE",
2265 [RTN_PROHIBIT] = "PROHIBIT",
2266 [RTN_THROW] = "THROW",
2267 [RTN_NAT] = "NAT",
2268 [RTN_XRESOLVE] = "XRESOLVE",
2269 };
2270
2271 static inline const char *rtn_type(unsigned t)
2272 {
2273 static char buf[32];
2274
2275 if (t < __RTN_MAX && rtn_type_names[t])
2276 return rtn_type_names[t];
2277 snprintf(buf, sizeof(buf), "type %d", t);
2278 return buf;
2279 }
2280
2281 /* Pretty print the trie */
2282 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2283 {
2284 const struct fib_trie_iter *iter = seq->private;
2285 struct node *n = v;
2286
2287 if (v == SEQ_START_TOKEN)
2288 return 0;
2289
2290 if (!NODE_PARENT(n)) {
2291 if (iter->trie == trie_local)
2292 seq_puts(seq, "<local>:\n");
2293 else
2294 seq_puts(seq, "<main>:\n");
2295 }
2296
2297 if (IS_TNODE(n)) {
2298 struct tnode *tn = (struct tnode *) n;
2299 __be32 prf = htonl(MASK_PFX(tn->key, tn->pos));
2300
2301 seq_indent(seq, iter->depth-1);
2302 seq_printf(seq, " +-- %d.%d.%d.%d/%d %d %d %d\n",
2303 NIPQUAD(prf), tn->pos, tn->bits, tn->full_children,
2304 tn->empty_children);
2305
2306 } else {
2307 struct leaf *l = (struct leaf *) n;
2308 int i;
2309 __be32 val = htonl(l->key);
2310
2311 seq_indent(seq, iter->depth);
2312 seq_printf(seq, " |-- %d.%d.%d.%d\n", NIPQUAD(val));
2313 for (i = 32; i >= 0; i--) {
2314 struct leaf_info *li = find_leaf_info(l, i);
2315 if (li) {
2316 struct fib_alias *fa;
2317 list_for_each_entry_rcu(fa, &li->falh, fa_list) {
2318 seq_indent(seq, iter->depth+1);
2319 seq_printf(seq, " /%d %s %s", i,
2320 rtn_scope(fa->fa_scope),
2321 rtn_type(fa->fa_type));
2322 if (fa->fa_tos)
2323 seq_printf(seq, "tos =%d\n",
2324 fa->fa_tos);
2325 seq_putc(seq, '\n');
2326 }
2327 }
2328 }
2329 }
2330
2331 return 0;
2332 }
2333
2334 static struct seq_operations fib_trie_seq_ops = {
2335 .start = fib_trie_seq_start,
2336 .next = fib_trie_seq_next,
2337 .stop = fib_trie_seq_stop,
2338 .show = fib_trie_seq_show,
2339 };
2340
2341 static int fib_trie_seq_open(struct inode *inode, struct file *file)
2342 {
2343 struct seq_file *seq;
2344 int rc = -ENOMEM;
2345 struct fib_trie_iter *s = kmalloc(sizeof(*s), GFP_KERNEL);
2346
2347 if (!s)
2348 goto out;
2349
2350 rc = seq_open(file, &fib_trie_seq_ops);
2351 if (rc)
2352 goto out_kfree;
2353
2354 seq = file->private_data;
2355 seq->private = s;
2356 memset(s, 0, sizeof(*s));
2357 out:
2358 return rc;
2359 out_kfree:
2360 kfree(s);
2361 goto out;
2362 }
2363
2364 static const struct file_operations fib_trie_fops = {
2365 .owner = THIS_MODULE,
2366 .open = fib_trie_seq_open,
2367 .read = seq_read,
2368 .llseek = seq_lseek,
2369 .release = seq_release_private,
2370 };
2371
2372 static unsigned fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2373 {
2374 static unsigned type2flags[RTN_MAX + 1] = {
2375 [7] = RTF_REJECT, [8] = RTF_REJECT,
2376 };
2377 unsigned flags = type2flags[type];
2378
2379 if (fi && fi->fib_nh->nh_gw)
2380 flags |= RTF_GATEWAY;
2381 if (mask == htonl(0xFFFFFFFF))
2382 flags |= RTF_HOST;
2383 flags |= RTF_UP;
2384 return flags;
2385 }
2386
2387 /*
2388 * This outputs /proc/net/route.
2389 * The format of the file is not supposed to be changed
2390 * and needs to be same as fib_hash output to avoid breaking
2391 * legacy utilities
2392 */
2393 static int fib_route_seq_show(struct seq_file *seq, void *v)
2394 {
2395 const struct fib_trie_iter *iter = seq->private;
2396 struct leaf *l = v;
2397 int i;
2398 char bf[128];
2399
2400 if (v == SEQ_START_TOKEN) {
2401 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2402 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2403 "\tWindow\tIRTT");
2404 return 0;
2405 }
2406
2407 if (iter->trie == trie_local)
2408 return 0;
2409 if (IS_TNODE(l))
2410 return 0;
2411
2412 for (i=32; i>=0; i--) {
2413 struct leaf_info *li = find_leaf_info(l, i);
2414 struct fib_alias *fa;
2415 __be32 mask, prefix;
2416
2417 if (!li)
2418 continue;
2419
2420 mask = inet_make_mask(li->plen);
2421 prefix = htonl(l->key);
2422
2423 list_for_each_entry_rcu(fa, &li->falh, fa_list) {
2424 const struct fib_info *fi = fa->fa_info;
2425 unsigned flags = fib_flag_trans(fa->fa_type, mask, fi);
2426
2427 if (fa->fa_type == RTN_BROADCAST
2428 || fa->fa_type == RTN_MULTICAST)
2429 continue;
2430
2431 if (fi)
2432 snprintf(bf, sizeof(bf),
2433 "%s\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u",
2434 fi->fib_dev ? fi->fib_dev->name : "*",
2435 prefix,
2436 fi->fib_nh->nh_gw, flags, 0, 0,
2437 fi->fib_priority,
2438 mask,
2439 (fi->fib_advmss ? fi->fib_advmss + 40 : 0),
2440 fi->fib_window,
2441 fi->fib_rtt >> 3);
2442 else
2443 snprintf(bf, sizeof(bf),
2444 "*\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u",
2445 prefix, 0, flags, 0, 0, 0,
2446 mask, 0, 0, 0);
2447
2448 seq_printf(seq, "%-127s\n", bf);
2449 }
2450 }
2451
2452 return 0;
2453 }
2454
2455 static struct seq_operations fib_route_seq_ops = {
2456 .start = fib_trie_seq_start,
2457 .next = fib_trie_seq_next,
2458 .stop = fib_trie_seq_stop,
2459 .show = fib_route_seq_show,
2460 };
2461
2462 static int fib_route_seq_open(struct inode *inode, struct file *file)
2463 {
2464 struct seq_file *seq;
2465 int rc = -ENOMEM;
2466 struct fib_trie_iter *s = kmalloc(sizeof(*s), GFP_KERNEL);
2467
2468 if (!s)
2469 goto out;
2470
2471 rc = seq_open(file, &fib_route_seq_ops);
2472 if (rc)
2473 goto out_kfree;
2474
2475 seq = file->private_data;
2476 seq->private = s;
2477 memset(s, 0, sizeof(*s));
2478 out:
2479 return rc;
2480 out_kfree:
2481 kfree(s);
2482 goto out;
2483 }
2484
2485 static const struct file_operations fib_route_fops = {
2486 .owner = THIS_MODULE,
2487 .open = fib_route_seq_open,
2488 .read = seq_read,
2489 .llseek = seq_lseek,
2490 .release = seq_release_private,
2491 };
2492
2493 int __init fib_proc_init(void)
2494 {
2495 if (!proc_net_fops_create("fib_trie", S_IRUGO, &fib_trie_fops))
2496 goto out1;
2497
2498 if (!proc_net_fops_create("fib_triestat", S_IRUGO, &fib_triestat_fops))
2499 goto out2;
2500
2501 if (!proc_net_fops_create("route", S_IRUGO, &fib_route_fops))
2502 goto out3;
2503
2504 return 0;
2505
2506 out3:
2507 proc_net_remove("fib_triestat");
2508 out2:
2509 proc_net_remove("fib_trie");
2510 out1:
2511 return -ENOMEM;
2512 }
2513
2514 void __init fib_proc_exit(void)
2515 {
2516 proc_net_remove("fib_trie");
2517 proc_net_remove("fib_triestat");
2518 proc_net_remove("route");
2519 }
2520
2521 #endif /* CONFIG_PROC_FS */
This page took 0.081745 seconds and 6 git commands to generate.