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