bnx2x: properly initialize FW stats
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / net / ipv4 / fib_trie.c
blob271c89bdf049706dbc8288457c308f437a427205
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.
7 * Robert Olsson <robert.olsson@its.uu.se> Uppsala Universitet
8 * & Swedish University of Agricultural Sciences.
10 * Jens Laas <jens.laas@data.slu.se> Swedish University of
11 * Agricultural Sciences.
13 * Hans Liss <hans.liss@its.uu.se> Uppsala Universitet
15 * This work is based on the LPC-trie which is originally descibed in:
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/
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
26 * Code from fib_hash has been reused which includes the following header:
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.
33 * IPv4 FIB: lookup engine and maintenance routines.
36 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
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.
43 * Substantial contributions to this work comes from:
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>
51 #define VERSION "0.409"
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"
84 #define MAX_STAT_DEPTH 32
86 #define KEYLENGTH (8*sizeof(t_key))
88 typedef unsigned int t_key;
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)
95 #define IS_TNODE(n) (!(n->parent & T_LEAF))
96 #define IS_LEAF(n) (n->parent & T_LEAF)
98 struct node {
99 unsigned long parent;
100 t_key key;
103 struct leaf {
104 unsigned long parent;
105 t_key key;
106 struct hlist_head list;
107 struct rcu_head rcu;
110 struct leaf_info {
111 struct hlist_node hlist;
112 struct rcu_head rcu;
113 int plen;
114 struct list_head falh;
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;
129 struct node *child[0];
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;
141 #endif
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];
153 struct trie {
154 struct node *trie;
155 #ifdef CONFIG_IP_FIB_TRIE_STATS
156 struct trie_use_stats stats;
157 #endif
160 static void put_child(struct trie *t, struct tnode *tn, int i, struct node *n);
161 static void tnode_put_child_reorg(struct tnode *tn, int i, struct node *n,
162 int wasfull);
163 static struct 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;
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.
175 static const int sync_pages = 128;
177 static struct kmem_cache *fn_alias_kmem __read_mostly;
178 static struct kmem_cache *trie_leaf_kmem __read_mostly;
180 static inline struct tnode *node_parent(struct node *node)
182 return (struct tnode *)(node->parent & ~NODE_TYPE_MASK);
185 static inline struct tnode *node_parent_rcu(struct node *node)
187 struct tnode *ret = node_parent(node);
189 return rcu_dereference_rtnl(ret);
192 /* Same as rcu_assign_pointer
193 * but that macro() assumes that value is a pointer.
195 static inline void node_set_parent(struct node *node, struct tnode *ptr)
197 smp_wmb();
198 node->parent = (unsigned long)ptr | NODE_TYPE(node);
201 static inline struct node *tnode_get_child(struct tnode *tn, unsigned int i)
203 BUG_ON(i >= 1U << tn->bits);
205 return tn->child[i];
208 static inline struct node *tnode_get_child_rcu(struct tnode *tn, unsigned int i)
210 struct node *ret = tnode_get_child(tn, i);
212 return rcu_dereference_rtnl(ret);
215 static inline int tnode_child_length(const struct tnode *tn)
217 return 1 << tn->bits;
220 static inline t_key mask_pfx(t_key k, unsigned short l)
222 return (l == 0) ? 0 : k >> (KEYLENGTH-l) << (KEYLENGTH-l);
225 static inline t_key tkey_extract_bits(t_key a, int offset, int bits)
227 if (offset < KEYLENGTH)
228 return ((t_key)(a << offset)) >> (KEYLENGTH - bits);
229 else
230 return 0;
233 static inline int tkey_equals(t_key a, t_key b)
235 return a == b;
238 static inline int tkey_sub_equals(t_key a, int offset, int bits, t_key b)
240 if (bits == 0 || offset >= KEYLENGTH)
241 return 1;
242 bits = bits > KEYLENGTH ? KEYLENGTH : bits;
243 return ((a ^ b) << offset) >> (KEYLENGTH - bits) == 0;
246 static inline int tkey_mismatch(t_key a, int offset, t_key b)
248 t_key diff = a ^ b;
249 int i = offset;
251 if (!diff)
252 return 0;
253 while ((diff << i) >> (KEYLENGTH-1) == 0)
254 i++;
255 return i;
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.
263 Consider a node 'n' and its parent 'tp'.
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.
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().
279 if n is an internal node - a 'tnode' here, the various parts of its key
280 have many different meanings.
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
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
293 tp->pos = 7
294 tp->bits = 3
295 n->pos = 15
296 n->bits = 4
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.
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.
306 The bits from (tp->pos + tp->bits) to (n->pos - 1) - "S" - are skipped bits
307 for the node n.
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.
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.
316 The rest of the bits, from (n->pos + n->bits) onward, are completely unknown
317 at this point.
321 static inline void check_tnode(const struct tnode *tn)
323 WARN_ON(tn && tn->pos+tn->bits > 32);
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;
331 static void __alias_free_mem(struct rcu_head *head)
333 struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
334 kmem_cache_free(fn_alias_kmem, fa);
337 static inline void alias_free_mem_rcu(struct fib_alias *fa)
339 call_rcu(&fa->rcu, __alias_free_mem);
342 static void __leaf_free_rcu(struct rcu_head *head)
344 struct leaf *l = container_of(head, struct leaf, rcu);
345 kmem_cache_free(trie_leaf_kmem, l);
348 static inline void free_leaf(struct leaf *l)
350 call_rcu_bh(&l->rcu, __leaf_free_rcu);
353 static void __leaf_info_free_rcu(struct rcu_head *head)
355 kfree(container_of(head, struct leaf_info, rcu));
358 static inline void free_leaf_info(struct leaf_info *leaf)
360 call_rcu(&leaf->rcu, __leaf_info_free_rcu);
363 static struct tnode *tnode_alloc(size_t size)
365 if (size <= PAGE_SIZE)
366 return kzalloc(size, GFP_KERNEL);
367 else
368 return __vmalloc(size, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL);
371 static void __tnode_vfree(struct work_struct *arg)
373 struct tnode *tn = container_of(arg, struct tnode, work);
374 vfree(tn);
377 static void __tnode_free_rcu(struct rcu_head *head)
379 struct tnode *tn = container_of(head, struct tnode, rcu);
380 size_t size = sizeof(struct tnode) +
381 (sizeof(struct node *) << tn->bits);
383 if (size <= PAGE_SIZE)
384 kfree(tn);
385 else {
386 INIT_WORK(&tn->work, __tnode_vfree);
387 schedule_work(&tn->work);
391 static inline void tnode_free(struct tnode *tn)
393 if (IS_LEAF(tn))
394 free_leaf((struct leaf *) tn);
395 else
396 call_rcu(&tn->rcu, __tnode_free_rcu);
399 static void tnode_free_safe(struct tnode *tn)
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 node *) << tn->bits);
408 static void tnode_free_flush(void)
410 struct tnode *tn;
412 while ((tn = tnode_free_head)) {
413 tnode_free_head = tn->tnode_free;
414 tn->tnode_free = NULL;
415 tnode_free(tn);
418 if (tnode_free_size >= PAGE_SIZE * sync_pages) {
419 tnode_free_size = 0;
420 synchronize_rcu();
424 static struct leaf *leaf_new(void)
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);
431 return l;
434 static struct leaf_info *leaf_info_new(int plen)
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);
441 return li;
444 static struct tnode *tnode_new(t_key key, int pos, int bits)
446 size_t sz = sizeof(struct tnode) + (sizeof(struct node *) << bits);
447 struct tnode *tn = tnode_alloc(sz);
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;
458 pr_debug("AT %p s=%zu %zu\n", tn, sizeof(struct tnode),
459 sizeof(struct node) << bits);
460 return tn;
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
468 static inline int tnode_full(const struct tnode *tn, const struct node *n)
470 if (n == NULL || IS_LEAF(n))
471 return 0;
473 return ((struct tnode *) n)->pos == tn->pos + tn->bits;
476 static inline void put_child(struct trie *t, struct tnode *tn, int i,
477 struct node *n)
479 tnode_put_child_reorg(tn, i, n, -1);
483 * Add a child at position i overwriting the old value.
484 * Update the value of full_children and empty_children.
487 static void tnode_put_child_reorg(struct tnode *tn, int i, struct node *n,
488 int wasfull)
490 struct node *chi = tn->child[i];
491 int isfull;
493 BUG_ON(i >= 1<<tn->bits);
495 /* update emptyChildren */
496 if (n == NULL && chi != NULL)
497 tn->empty_children++;
498 else if (n != NULL && chi == NULL)
499 tn->empty_children--;
501 /* update fullChildren */
502 if (wasfull == -1)
503 wasfull = tnode_full(tn, chi);
505 isfull = tnode_full(tn, n);
506 if (wasfull && !isfull)
507 tn->full_children--;
508 else if (!wasfull && isfull)
509 tn->full_children++;
511 if (n)
512 node_set_parent(n, tn);
514 rcu_assign_pointer(tn->child[i], n);
517 #define MAX_WORK 10
518 static struct node *resize(struct trie *t, struct tnode *tn)
520 int i;
521 struct tnode *old_tn;
522 int inflate_threshold_use;
523 int halve_threshold_use;
524 int max_work;
526 if (!tn)
527 return NULL;
529 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
530 tn, inflate_threshold, halve_threshold);
532 /* No children */
533 if (tn->empty_children == tnode_child_length(tn)) {
534 tnode_free_safe(tn);
535 return NULL;
537 /* One child */
538 if (tn->empty_children == tnode_child_length(tn) - 1)
539 goto one_child;
541 * Double as long as the resulting node has a number of
542 * nonempty nodes that are above the threshold.
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'."
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.
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.
566 * A clearer way to write this would be:
568 * to_be_doubled = tn->full_children;
569 * not_to_be_doubled = tnode_child_length(tn) - tn->empty_children -
570 * tn->full_children;
572 * new_child_length = tnode_child_length(tn) * 2;
574 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
575 * new_child_length;
576 * if (new_fill_factor >= inflate_threshold)
578 * ...and so on, tho it would mess up the while () loop.
580 * anyway,
581 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
582 * inflate_threshold
584 * avoid a division:
585 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
586 * inflate_threshold * new_child_length
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
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
597 * shorten again:
598 * 50 * (tn->full_children + tnode_child_length(tn) -
599 * tn->empty_children) >= inflate_threshold *
600 * tnode_child_length(tn)
604 check_tnode(tn);
606 /* Keep root node larger */
608 if (!node_parent((struct 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;
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))) {
622 old_tn = tn;
623 tn = inflate(t, tn);
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;
634 check_tnode(tn);
636 /* Return if at least one inflate is run */
637 if (max_work != MAX_WORK)
638 return (struct node *) tn;
641 * Halve as long as the number of empty children in this
642 * node is above threshold.
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)) {
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;
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 node *n;
668 n = tn->child[i];
669 if (!n)
670 continue;
672 /* compress one level */
674 node_set_parent(n, NULL);
675 tnode_free_safe(tn);
676 return n;
679 return (struct node *) tn;
682 static struct tnode *inflate(struct trie *t, struct tnode *tn)
684 struct tnode *oldtnode = tn;
685 int olen = tnode_child_length(tn);
686 int i;
688 pr_debug("In inflate\n");
690 tn = tnode_new(oldtnode->key, oldtnode->pos, oldtnode->bits + 1);
692 if (!tn)
693 return ERR_PTR(-ENOMEM);
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.
702 for (i = 0; i < olen; i++) {
703 struct tnode *inode;
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;
713 left = tnode_new(inode->key&(~m), inode->pos + 1,
714 inode->bits - 1);
715 if (!left)
716 goto nomem;
718 right = tnode_new(inode->key|m, inode->pos + 1,
719 inode->bits - 1);
721 if (!right) {
722 tnode_free(left);
723 goto nomem;
726 put_child(t, tn, 2*i, (struct node *) left);
727 put_child(t, tn, 2*i+1, (struct node *) right);
731 for (i = 0; i < olen; i++) {
732 struct tnode *inode;
733 struct node *node = tnode_get_child(oldtnode, i);
734 struct tnode *left, *right;
735 int size, j;
737 /* An empty child */
738 if (node == NULL)
739 continue;
741 /* A leaf or an internal node with skipped bits */
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;
754 /* An internal node with two children */
755 inode = (struct tnode *) node;
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]);
761 tnode_free_safe(inode);
762 continue;
765 /* An internal node with more than two children */
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)
784 /* Use the old key, but set the new significant
785 * bit to zero.
788 left = (struct tnode *) tnode_get_child(tn, 2*i);
789 put_child(t, tn, 2*i, NULL);
791 BUG_ON(!left);
793 right = (struct tnode *) tnode_get_child(tn, 2*i+1);
794 put_child(t, tn, 2*i+1, NULL);
796 BUG_ON(!right);
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]);
803 put_child(t, tn, 2*i, resize(t, left));
804 put_child(t, tn, 2*i+1, resize(t, right));
806 tnode_free_safe(inode);
808 tnode_free_safe(oldtnode);
809 return tn;
810 nomem:
812 int size = tnode_child_length(tn);
813 int j;
815 for (j = 0; j < size; j++)
816 if (tn->child[j])
817 tnode_free((struct tnode *)tn->child[j]);
819 tnode_free(tn);
821 return ERR_PTR(-ENOMEM);
825 static struct tnode *halve(struct trie *t, struct tnode *tn)
827 struct tnode *oldtnode = tn;
828 struct node *left, *right;
829 int i;
830 int olen = tnode_child_length(tn);
832 pr_debug("In halve\n");
834 tn = tnode_new(oldtnode->key, oldtnode->pos, oldtnode->bits - 1);
836 if (!tn)
837 return ERR_PTR(-ENOMEM);
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.
846 for (i = 0; i < olen; i += 2) {
847 left = tnode_get_child(oldtnode, i);
848 right = tnode_get_child(oldtnode, i+1);
850 /* Two nonempty children */
851 if (left && right) {
852 struct tnode *newn;
854 newn = tnode_new(left->key, tn->pos + tn->bits, 1);
856 if (!newn)
857 goto nomem;
859 put_child(t, tn, i/2, (struct node *)newn);
864 for (i = 0; i < olen; i += 2) {
865 struct tnode *newBinNode;
867 left = tnode_get_child(oldtnode, i);
868 right = tnode_get_child(oldtnode, i+1);
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;
878 if (right == NULL) {
879 put_child(t, tn, i/2, left);
880 continue;
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));
890 tnode_free_safe(oldtnode);
891 return tn;
892 nomem:
894 int size = tnode_child_length(tn);
895 int j;
897 for (j = 0; j < size; j++)
898 if (tn->child[j])
899 tnode_free((struct tnode *)tn->child[j]);
901 tnode_free(tn);
903 return ERR_PTR(-ENOMEM);
907 /* readside must use rcu_read_lock currently dump routines
908 via get_fa_head and dump */
910 static struct leaf_info *find_leaf_info(struct leaf *l, int plen)
912 struct hlist_head *head = &l->list;
913 struct hlist_node *node;
914 struct leaf_info *li;
916 hlist_for_each_entry_rcu(li, node, head, hlist)
917 if (li->plen == plen)
918 return li;
920 return NULL;
923 static inline struct list_head *get_fa_head(struct leaf *l, int plen)
925 struct leaf_info *li = find_leaf_info(l, plen);
927 if (!li)
928 return NULL;
930 return &li->falh;
933 static void insert_leaf_info(struct hlist_head *head, struct leaf_info *new)
935 struct leaf_info *li = NULL, *last = NULL;
936 struct hlist_node *node;
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;
945 last = li;
947 if (last)
948 hlist_add_after_rcu(&last->hlist, &new->hlist);
949 else
950 hlist_add_before_rcu(&new->hlist, &li->hlist);
954 /* rcu_read_lock needs to be hold by caller from readside */
956 static struct leaf *
957 fib_find_node(struct trie *t, u32 key)
959 int pos;
960 struct tnode *tn;
961 struct node *n;
963 pos = 0;
964 n = rcu_dereference_rtnl(t->trie);
966 while (n != NULL && NODE_TYPE(n) == T_TNODE) {
967 tn = (struct tnode *) n;
969 check_tnode(tn);
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;
980 /* Case we have found a leaf. Compare prefixes */
982 if (n != NULL && IS_LEAF(n) && tkey_equals(key, n->key))
983 return (struct leaf *)n;
985 return NULL;
988 static void trie_rebalance(struct trie *t, struct tnode *tn)
990 int wasfull;
991 t_key cindex, key;
992 struct tnode *tp;
994 key = tn->key;
996 while (tn != NULL && (tp = node_parent((struct 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);
1001 tnode_put_child_reorg((struct tnode *)tp, cindex,
1002 (struct node *)tn, wasfull);
1004 tp = node_parent((struct node *) tn);
1005 if (!tp)
1006 rcu_assign_pointer(t->trie, (struct node *)tn);
1008 tnode_free_flush();
1009 if (!tp)
1010 break;
1011 tn = tp;
1014 /* Handle last (top) tnode */
1015 if (IS_TNODE(tn))
1016 tn = (struct tnode *)resize(t, (struct tnode *)tn);
1018 rcu_assign_pointer(t->trie, (struct node *)tn);
1019 tnode_free_flush();
1022 /* only used from updater-side */
1024 static struct list_head *fib_insert_node(struct trie *t, u32 key, int plen)
1026 int pos, newpos;
1027 struct tnode *tp = NULL, *tn = NULL;
1028 struct 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;
1035 pos = 0;
1036 n = t->trie;
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'!
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.
1048 * If it doesn't, we have to replace it with a new T_TNODE.
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.
1056 while (n != NULL && NODE_TYPE(n) == T_TNODE) {
1057 tn = (struct tnode *) n;
1059 check_tnode(tn);
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));
1069 BUG_ON(n && node_parent(n) != tn);
1070 } else
1071 break;
1075 * n ----> NULL, LEAF or TNODE
1077 * tp is n's (parent) ----> NULL or TNODE
1080 BUG_ON(tp && IS_LEAF(tp));
1082 /* Case 1: n is a leaf. Compare prefixes */
1084 if (n != NULL && IS_LEAF(n) && tkey_equals(key, n->key)) {
1085 l = (struct leaf *) n;
1086 li = leaf_info_new(plen);
1088 if (!li)
1089 return NULL;
1091 fa_head = &li->falh;
1092 insert_leaf_info(&l->list, li);
1093 goto done;
1095 l = leaf_new();
1097 if (!l)
1098 return NULL;
1100 l->key = key;
1101 li = leaf_info_new(plen);
1103 if (!li) {
1104 free_leaf(l);
1105 return NULL;
1108 fa_head = &li->falh;
1109 insert_leaf_info(&l->list, li);
1111 if (t->trie && n == NULL) {
1112 /* Case 2: n is NULL, and will just insert a new leaf */
1114 node_set_parent((struct node *)l, tp);
1116 cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1117 put_child(t, (struct tnode *)tp, cindex, (struct node *)l);
1118 } else {
1119 /* Case 3: n is a LEAF or a TNODE and the key doesn't match. */
1121 * Add a new tnode here
1122 * first tnode need some special handling
1125 if (tp)
1126 pos = tp->pos+tp->bits;
1127 else
1128 pos = 0;
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 */
1138 if (!tn) {
1139 free_leaf_info(li);
1140 free_leaf(l);
1141 return NULL;
1144 node_set_parent((struct node *)tn, tp);
1146 missbit = tkey_extract_bits(key, newpos, 1);
1147 put_child(t, tn, missbit, (struct node *)l);
1148 put_child(t, tn, 1-missbit, n);
1150 if (tp) {
1151 cindex = tkey_extract_bits(key, tp->pos, tp->bits);
1152 put_child(t, (struct tnode *)tp, cindex,
1153 (struct node *)tn);
1154 } else {
1155 rcu_assign_pointer(t->trie, (struct node *)tn);
1156 tp = tn;
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);
1165 /* Rebalance the trie */
1167 trie_rebalance(t, tp);
1168 done:
1169 return fa_head;
1173 * Caller must hold RTNL.
1175 int fib_table_insert(struct fib_table *tb, struct fib_config *cfg)
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;
1187 if (plen > 32)
1188 return -EINVAL;
1190 key = ntohl(cfg->fc_dst);
1192 pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1194 mask = ntohl(inet_make_mask(plen));
1196 if (key & ~mask)
1197 return -EINVAL;
1199 key = key & mask;
1201 fi = fib_create_info(cfg);
1202 if (IS_ERR(fi)) {
1203 err = PTR_ERR(fi);
1204 goto err;
1207 l = fib_find_node(t, key);
1208 fa = NULL;
1210 if (l) {
1211 fa_head = get_fa_head(l, plen);
1212 fa = fib_find_alias(fa_head, tos, fi->fib_priority);
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.
1219 * If fa is NULL, we will need to allocate a new one and
1220 * insert to the head of f.
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.
1226 if (fa && fa->fa_tos == tos &&
1227 fa->fa_info->fib_priority == fi->fib_priority) {
1228 struct fib_alias *fa_first, *fa_match;
1230 err = -EEXIST;
1231 if (cfg->fc_nlflags & NLM_F_EXCL)
1232 goto out;
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
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_scope == cfg->fc_scope &&
1249 fa->fa_info == fi) {
1250 fa_match = fa;
1251 break;
1255 if (cfg->fc_nlflags & NLM_F_REPLACE) {
1256 struct fib_info *fi_drop;
1257 u8 state;
1259 fa = fa_first;
1260 if (fa_match) {
1261 if (fa == fa_match)
1262 err = 0;
1263 goto out;
1265 err = -ENOBUFS;
1266 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1267 if (new_fa == NULL)
1268 goto out;
1270 fi_drop = fa->fa_info;
1271 new_fa->fa_tos = fa->fa_tos;
1272 new_fa->fa_info = fi;
1273 new_fa->fa_type = cfg->fc_type;
1274 new_fa->fa_scope = cfg->fc_scope;
1275 state = fa->fa_state;
1276 new_fa->fa_state = state & ~FA_S_ACCESSED;
1278 list_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1279 alias_free_mem_rcu(fa);
1281 fib_release_info(fi_drop);
1282 if (state & FA_S_ACCESSED)
1283 rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1284 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1285 tb->tb_id, &cfg->fc_nlinfo, NLM_F_REPLACE);
1287 goto succeeded;
1289 /* Error if we find a perfect match which
1290 * uses the same scope, type, and nexthop
1291 * information.
1293 if (fa_match)
1294 goto out;
1296 if (!(cfg->fc_nlflags & NLM_F_APPEND))
1297 fa = fa_first;
1299 err = -ENOENT;
1300 if (!(cfg->fc_nlflags & NLM_F_CREATE))
1301 goto out;
1303 err = -ENOBUFS;
1304 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1305 if (new_fa == NULL)
1306 goto out;
1308 new_fa->fa_info = fi;
1309 new_fa->fa_tos = tos;
1310 new_fa->fa_type = cfg->fc_type;
1311 new_fa->fa_scope = cfg->fc_scope;
1312 new_fa->fa_state = 0;
1314 * Insert new entry to the list.
1317 if (!fa_head) {
1318 fa_head = fib_insert_node(t, key, plen);
1319 if (unlikely(!fa_head)) {
1320 err = -ENOMEM;
1321 goto out_free_new_fa;
1325 list_add_tail_rcu(&new_fa->fa_list,
1326 (fa ? &fa->fa_list : fa_head));
1328 rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1329 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, tb->tb_id,
1330 &cfg->fc_nlinfo, 0);
1331 succeeded:
1332 return 0;
1334 out_free_new_fa:
1335 kmem_cache_free(fn_alias_kmem, new_fa);
1336 out:
1337 fib_release_info(fi);
1338 err:
1339 return err;
1342 /* should be called with rcu_read_lock */
1343 static int check_leaf(struct trie *t, struct leaf *l,
1344 t_key key, const struct flowi *flp,
1345 struct fib_result *res, int fib_flags)
1347 struct leaf_info *li;
1348 struct hlist_head *hhead = &l->list;
1349 struct hlist_node *node;
1351 hlist_for_each_entry_rcu(li, node, hhead, hlist) {
1352 int err;
1353 int plen = li->plen;
1354 __be32 mask = inet_make_mask(plen);
1356 if (l->key != (key & ntohl(mask)))
1357 continue;
1359 err = fib_semantic_match(&li->falh, flp, res, plen, fib_flags);
1361 #ifdef CONFIG_IP_FIB_TRIE_STATS
1362 if (err <= 0)
1363 t->stats.semantic_match_passed++;
1364 else
1365 t->stats.semantic_match_miss++;
1366 #endif
1367 if (err <= 0)
1368 return err;
1371 return 1;
1374 int fib_table_lookup(struct fib_table *tb, const struct flowi *flp,
1375 struct fib_result *res, int fib_flags)
1377 struct trie *t = (struct trie *) tb->tb_data;
1378 int ret;
1379 struct node *n;
1380 struct tnode *pn;
1381 int pos, bits;
1382 t_key key = ntohl(flp->fl4_dst);
1383 int chopped_off;
1384 t_key cindex = 0;
1385 int current_prefix_length = KEYLENGTH;
1386 struct tnode *cn;
1387 t_key node_prefix, key_prefix, pref_mismatch;
1388 int mp;
1390 rcu_read_lock();
1392 n = rcu_dereference(t->trie);
1393 if (!n)
1394 goto failed;
1396 #ifdef CONFIG_IP_FIB_TRIE_STATS
1397 t->stats.gets++;
1398 #endif
1400 /* Just a leaf? */
1401 if (IS_LEAF(n)) {
1402 ret = check_leaf(t, (struct leaf *)n, key, flp, res, fib_flags);
1403 goto found;
1406 pn = (struct tnode *) n;
1407 chopped_off = 0;
1409 while (pn) {
1410 pos = pn->pos;
1411 bits = pn->bits;
1413 if (!chopped_off)
1414 cindex = tkey_extract_bits(mask_pfx(key, current_prefix_length),
1415 pos, bits);
1417 n = tnode_get_child_rcu(pn, cindex);
1419 if (n == NULL) {
1420 #ifdef CONFIG_IP_FIB_TRIE_STATS
1421 t->stats.null_node_hit++;
1422 #endif
1423 goto backtrace;
1426 if (IS_LEAF(n)) {
1427 ret = check_leaf(t, (struct leaf *)n, key, flp, res, fib_flags);
1428 if (ret > 0)
1429 goto backtrace;
1430 goto found;
1433 cn = (struct tnode *)n;
1436 * It's a tnode, and we can do some extra checks here if we
1437 * like, to avoid descending into a dead-end branch.
1438 * This tnode is in the parent's child array at index
1439 * key[p_pos..p_pos+p_bits] but potentially with some bits
1440 * chopped off, so in reality the index may be just a
1441 * subprefix, padded with zero at the end.
1442 * We can also take a look at any skipped bits in this
1443 * tnode - everything up to p_pos is supposed to be ok,
1444 * and the non-chopped bits of the index (se previous
1445 * paragraph) are also guaranteed ok, but the rest is
1446 * considered unknown.
1448 * The skipped bits are key[pos+bits..cn->pos].
1451 /* If current_prefix_length < pos+bits, we are already doing
1452 * actual prefix matching, which means everything from
1453 * pos+(bits-chopped_off) onward must be zero along some
1454 * branch of this subtree - otherwise there is *no* valid
1455 * prefix present. Here we can only check the skipped
1456 * bits. Remember, since we have already indexed into the
1457 * parent's child array, we know that the bits we chopped of
1458 * *are* zero.
1461 /* NOTA BENE: Checking only skipped bits
1462 for the new node here */
1464 if (current_prefix_length < pos+bits) {
1465 if (tkey_extract_bits(cn->key, current_prefix_length,
1466 cn->pos - current_prefix_length)
1467 || !(cn->child[0]))
1468 goto backtrace;
1472 * If chopped_off=0, the index is fully validated and we
1473 * only need to look at the skipped bits for this, the new,
1474 * tnode. What we actually want to do is to find out if
1475 * these skipped bits match our key perfectly, or if we will
1476 * have to count on finding a matching prefix further down,
1477 * because if we do, we would like to have some way of
1478 * verifying the existence of such a prefix at this point.
1481 /* The only thing we can do at this point is to verify that
1482 * any such matching prefix can indeed be a prefix to our
1483 * key, and if the bits in the node we are inspecting that
1484 * do not match our key are not ZERO, this cannot be true.
1485 * Thus, find out where there is a mismatch (before cn->pos)
1486 * and verify that all the mismatching bits are zero in the
1487 * new tnode's key.
1491 * Note: We aren't very concerned about the piece of
1492 * the key that precede pn->pos+pn->bits, since these
1493 * have already been checked. The bits after cn->pos
1494 * aren't checked since these are by definition
1495 * "unknown" at this point. Thus, what we want to see
1496 * is if we are about to enter the "prefix matching"
1497 * state, and in that case verify that the skipped
1498 * bits that will prevail throughout this subtree are
1499 * zero, as they have to be if we are to find a
1500 * matching prefix.
1503 node_prefix = mask_pfx(cn->key, cn->pos);
1504 key_prefix = mask_pfx(key, cn->pos);
1505 pref_mismatch = key_prefix^node_prefix;
1506 mp = 0;
1509 * In short: If skipped bits in this node do not match
1510 * the search key, enter the "prefix matching"
1511 * state.directly.
1513 if (pref_mismatch) {
1514 while (!(pref_mismatch & (1<<(KEYLENGTH-1)))) {
1515 mp++;
1516 pref_mismatch = pref_mismatch << 1;
1518 key_prefix = tkey_extract_bits(cn->key, mp, cn->pos-mp);
1520 if (key_prefix != 0)
1521 goto backtrace;
1523 if (current_prefix_length >= cn->pos)
1524 current_prefix_length = mp;
1527 pn = (struct tnode *)n; /* Descend */
1528 chopped_off = 0;
1529 continue;
1531 backtrace:
1532 chopped_off++;
1534 /* As zero don't change the child key (cindex) */
1535 while ((chopped_off <= pn->bits)
1536 && !(cindex & (1<<(chopped_off-1))))
1537 chopped_off++;
1539 /* Decrease current_... with bits chopped off */
1540 if (current_prefix_length > pn->pos + pn->bits - chopped_off)
1541 current_prefix_length = pn->pos + pn->bits
1542 - chopped_off;
1545 * Either we do the actual chop off according or if we have
1546 * chopped off all bits in this tnode walk up to our parent.
1549 if (chopped_off <= pn->bits) {
1550 cindex &= ~(1 << (chopped_off-1));
1551 } else {
1552 struct tnode *parent = node_parent_rcu((struct node *) pn);
1553 if (!parent)
1554 goto failed;
1556 /* Get Child's index */
1557 cindex = tkey_extract_bits(pn->key, parent->pos, parent->bits);
1558 pn = parent;
1559 chopped_off = 0;
1561 #ifdef CONFIG_IP_FIB_TRIE_STATS
1562 t->stats.backtrack++;
1563 #endif
1564 goto backtrace;
1567 failed:
1568 ret = 1;
1569 found:
1570 rcu_read_unlock();
1571 return ret;
1575 * Remove the leaf and return parent.
1577 static void trie_leaf_remove(struct trie *t, struct leaf *l)
1579 struct tnode *tp = node_parent((struct node *) l);
1581 pr_debug("entering trie_leaf_remove(%p)\n", l);
1583 if (tp) {
1584 t_key cindex = tkey_extract_bits(l->key, tp->pos, tp->bits);
1585 put_child(t, (struct tnode *)tp, cindex, NULL);
1586 trie_rebalance(t, tp);
1587 } else
1588 rcu_assign_pointer(t->trie, NULL);
1590 free_leaf(l);
1594 * Caller must hold RTNL.
1596 int fib_table_delete(struct fib_table *tb, struct fib_config *cfg)
1598 struct trie *t = (struct trie *) tb->tb_data;
1599 u32 key, mask;
1600 int plen = cfg->fc_dst_len;
1601 u8 tos = cfg->fc_tos;
1602 struct fib_alias *fa, *fa_to_delete;
1603 struct list_head *fa_head;
1604 struct leaf *l;
1605 struct leaf_info *li;
1607 if (plen > 32)
1608 return -EINVAL;
1610 key = ntohl(cfg->fc_dst);
1611 mask = ntohl(inet_make_mask(plen));
1613 if (key & ~mask)
1614 return -EINVAL;
1616 key = key & mask;
1617 l = fib_find_node(t, key);
1619 if (!l)
1620 return -ESRCH;
1622 fa_head = get_fa_head(l, plen);
1623 fa = fib_find_alias(fa_head, tos, 0);
1625 if (!fa)
1626 return -ESRCH;
1628 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1630 fa_to_delete = NULL;
1631 fa = list_entry(fa->fa_list.prev, struct fib_alias, fa_list);
1632 list_for_each_entry_continue(fa, fa_head, fa_list) {
1633 struct fib_info *fi = fa->fa_info;
1635 if (fa->fa_tos != tos)
1636 break;
1638 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1639 (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1640 fa->fa_scope == cfg->fc_scope) &&
1641 (!cfg->fc_protocol ||
1642 fi->fib_protocol == cfg->fc_protocol) &&
1643 fib_nh_match(cfg, fi) == 0) {
1644 fa_to_delete = fa;
1645 break;
1649 if (!fa_to_delete)
1650 return -ESRCH;
1652 fa = fa_to_delete;
1653 rtmsg_fib(RTM_DELROUTE, htonl(key), fa, plen, tb->tb_id,
1654 &cfg->fc_nlinfo, 0);
1656 l = fib_find_node(t, key);
1657 li = find_leaf_info(l, plen);
1659 list_del_rcu(&fa->fa_list);
1661 if (list_empty(fa_head)) {
1662 hlist_del_rcu(&li->hlist);
1663 free_leaf_info(li);
1666 if (hlist_empty(&l->list))
1667 trie_leaf_remove(t, l);
1669 if (fa->fa_state & FA_S_ACCESSED)
1670 rt_cache_flush(cfg->fc_nlinfo.nl_net, -1);
1672 fib_release_info(fa->fa_info);
1673 alias_free_mem_rcu(fa);
1674 return 0;
1677 static int trie_flush_list(struct list_head *head)
1679 struct fib_alias *fa, *fa_node;
1680 int found = 0;
1682 list_for_each_entry_safe(fa, fa_node, head, fa_list) {
1683 struct fib_info *fi = fa->fa_info;
1685 if (fi && (fi->fib_flags & RTNH_F_DEAD)) {
1686 list_del_rcu(&fa->fa_list);
1687 fib_release_info(fa->fa_info);
1688 alias_free_mem_rcu(fa);
1689 found++;
1692 return found;
1695 static int trie_flush_leaf(struct leaf *l)
1697 int found = 0;
1698 struct hlist_head *lih = &l->list;
1699 struct hlist_node *node, *tmp;
1700 struct leaf_info *li = NULL;
1702 hlist_for_each_entry_safe(li, node, tmp, lih, hlist) {
1703 found += trie_flush_list(&li->falh);
1705 if (list_empty(&li->falh)) {
1706 hlist_del_rcu(&li->hlist);
1707 free_leaf_info(li);
1710 return found;
1714 * Scan for the next right leaf starting at node p->child[idx]
1715 * Since we have back pointer, no recursion necessary.
1717 static struct leaf *leaf_walk_rcu(struct tnode *p, struct node *c)
1719 do {
1720 t_key idx;
1722 if (c)
1723 idx = tkey_extract_bits(c->key, p->pos, p->bits) + 1;
1724 else
1725 idx = 0;
1727 while (idx < 1u << p->bits) {
1728 c = tnode_get_child_rcu(p, idx++);
1729 if (!c)
1730 continue;
1732 if (IS_LEAF(c)) {
1733 prefetch(p->child[idx]);
1734 return (struct leaf *) c;
1737 /* Rescan start scanning in new node */
1738 p = (struct tnode *) c;
1739 idx = 0;
1742 /* Node empty, walk back up to parent */
1743 c = (struct node *) p;
1744 } while ((p = node_parent_rcu(c)) != NULL);
1746 return NULL; /* Root of trie */
1749 static struct leaf *trie_firstleaf(struct trie *t)
1751 struct tnode *n = (struct tnode *)rcu_dereference_rtnl(t->trie);
1753 if (!n)
1754 return NULL;
1756 if (IS_LEAF(n)) /* trie is just a leaf */
1757 return (struct leaf *) n;
1759 return leaf_walk_rcu(n, NULL);
1762 static struct leaf *trie_nextleaf(struct leaf *l)
1764 struct node *c = (struct node *) l;
1765 struct tnode *p = node_parent_rcu(c);
1767 if (!p)
1768 return NULL; /* trie with just one leaf */
1770 return leaf_walk_rcu(p, c);
1773 static struct leaf *trie_leafindex(struct trie *t, int index)
1775 struct leaf *l = trie_firstleaf(t);
1777 while (l && index-- > 0)
1778 l = trie_nextleaf(l);
1780 return l;
1785 * Caller must hold RTNL.
1787 int fib_table_flush(struct fib_table *tb)
1789 struct trie *t = (struct trie *) tb->tb_data;
1790 struct leaf *l, *ll = NULL;
1791 int found = 0;
1793 for (l = trie_firstleaf(t); l; l = trie_nextleaf(l)) {
1794 found += trie_flush_leaf(l);
1796 if (ll && hlist_empty(&ll->list))
1797 trie_leaf_remove(t, ll);
1798 ll = l;
1801 if (ll && hlist_empty(&ll->list))
1802 trie_leaf_remove(t, ll);
1804 pr_debug("trie_flush found=%d\n", found);
1805 return found;
1808 void fib_table_select_default(struct fib_table *tb,
1809 const struct flowi *flp,
1810 struct fib_result *res)
1812 struct trie *t = (struct trie *) tb->tb_data;
1813 int order, last_idx;
1814 struct fib_info *fi = NULL;
1815 struct fib_info *last_resort;
1816 struct fib_alias *fa = NULL;
1817 struct list_head *fa_head;
1818 struct leaf *l;
1820 last_idx = -1;
1821 last_resort = NULL;
1822 order = -1;
1824 rcu_read_lock();
1826 l = fib_find_node(t, 0);
1827 if (!l)
1828 goto out;
1830 fa_head = get_fa_head(l, 0);
1831 if (!fa_head)
1832 goto out;
1834 if (list_empty(fa_head))
1835 goto out;
1837 list_for_each_entry_rcu(fa, fa_head, fa_list) {
1838 struct fib_info *next_fi = fa->fa_info;
1840 if (fa->fa_scope != res->scope ||
1841 fa->fa_type != RTN_UNICAST)
1842 continue;
1844 if (next_fi->fib_priority > res->fi->fib_priority)
1845 break;
1846 if (!next_fi->fib_nh[0].nh_gw ||
1847 next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1848 continue;
1849 fa->fa_state |= FA_S_ACCESSED;
1851 if (fi == NULL) {
1852 if (next_fi != res->fi)
1853 break;
1854 } else if (!fib_detect_death(fi, order, &last_resort,
1855 &last_idx, tb->tb_default)) {
1856 fib_result_assign(res, fi);
1857 tb->tb_default = order;
1858 goto out;
1860 fi = next_fi;
1861 order++;
1863 if (order <= 0 || fi == NULL) {
1864 tb->tb_default = -1;
1865 goto out;
1868 if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1869 tb->tb_default)) {
1870 fib_result_assign(res, fi);
1871 tb->tb_default = order;
1872 goto out;
1874 if (last_idx >= 0)
1875 fib_result_assign(res, last_resort);
1876 tb->tb_default = last_idx;
1877 out:
1878 rcu_read_unlock();
1881 static int fn_trie_dump_fa(t_key key, int plen, struct list_head *fah,
1882 struct fib_table *tb,
1883 struct sk_buff *skb, struct netlink_callback *cb)
1885 int i, s_i;
1886 struct fib_alias *fa;
1887 __be32 xkey = htonl(key);
1889 s_i = cb->args[5];
1890 i = 0;
1892 /* rcu_read_lock is hold by caller */
1894 list_for_each_entry_rcu(fa, fah, fa_list) {
1895 if (i < s_i) {
1896 i++;
1897 continue;
1900 if (fib_dump_info(skb, NETLINK_CB(cb->skb).pid,
1901 cb->nlh->nlmsg_seq,
1902 RTM_NEWROUTE,
1903 tb->tb_id,
1904 fa->fa_type,
1905 fa->fa_scope,
1906 xkey,
1907 plen,
1908 fa->fa_tos,
1909 fa->fa_info, NLM_F_MULTI) < 0) {
1910 cb->args[5] = i;
1911 return -1;
1913 i++;
1915 cb->args[5] = i;
1916 return skb->len;
1919 static int fn_trie_dump_leaf(struct leaf *l, struct fib_table *tb,
1920 struct sk_buff *skb, struct netlink_callback *cb)
1922 struct leaf_info *li;
1923 struct hlist_node *node;
1924 int i, s_i;
1926 s_i = cb->args[4];
1927 i = 0;
1929 /* rcu_read_lock is hold by caller */
1930 hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
1931 if (i < s_i) {
1932 i++;
1933 continue;
1936 if (i > s_i)
1937 cb->args[5] = 0;
1939 if (list_empty(&li->falh))
1940 continue;
1942 if (fn_trie_dump_fa(l->key, li->plen, &li->falh, tb, skb, cb) < 0) {
1943 cb->args[4] = i;
1944 return -1;
1946 i++;
1949 cb->args[4] = i;
1950 return skb->len;
1953 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
1954 struct netlink_callback *cb)
1956 struct leaf *l;
1957 struct trie *t = (struct trie *) tb->tb_data;
1958 t_key key = cb->args[2];
1959 int count = cb->args[3];
1961 rcu_read_lock();
1962 /* Dump starting at last key.
1963 * Note: 0.0.0.0/0 (ie default) is first key.
1965 if (count == 0)
1966 l = trie_firstleaf(t);
1967 else {
1968 /* Normally, continue from last key, but if that is missing
1969 * fallback to using slow rescan
1971 l = fib_find_node(t, key);
1972 if (!l)
1973 l = trie_leafindex(t, count);
1976 while (l) {
1977 cb->args[2] = l->key;
1978 if (fn_trie_dump_leaf(l, tb, skb, cb) < 0) {
1979 cb->args[3] = count;
1980 rcu_read_unlock();
1981 return -1;
1984 ++count;
1985 l = trie_nextleaf(l);
1986 memset(&cb->args[4], 0,
1987 sizeof(cb->args) - 4*sizeof(cb->args[0]));
1989 cb->args[3] = count;
1990 rcu_read_unlock();
1992 return skb->len;
1995 void __init fib_hash_init(void)
1997 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
1998 sizeof(struct fib_alias),
1999 0, SLAB_PANIC, NULL);
2001 trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
2002 max(sizeof(struct leaf),
2003 sizeof(struct leaf_info)),
2004 0, SLAB_PANIC, NULL);
2008 /* Fix more generic FIB names for init later */
2009 struct fib_table *fib_hash_table(u32 id)
2011 struct fib_table *tb;
2012 struct trie *t;
2014 tb = kmalloc(sizeof(struct fib_table) + sizeof(struct trie),
2015 GFP_KERNEL);
2016 if (tb == NULL)
2017 return NULL;
2019 tb->tb_id = id;
2020 tb->tb_default = -1;
2022 t = (struct trie *) tb->tb_data;
2023 memset(t, 0, sizeof(*t));
2025 if (id == RT_TABLE_LOCAL)
2026 pr_info("IPv4 FIB: Using LC-trie version %s\n", VERSION);
2028 return tb;
2031 #ifdef CONFIG_PROC_FS
2032 /* Depth first Trie walk iterator */
2033 struct fib_trie_iter {
2034 struct seq_net_private p;
2035 struct fib_table *tb;
2036 struct tnode *tnode;
2037 unsigned int index;
2038 unsigned int depth;
2041 static struct node *fib_trie_get_next(struct fib_trie_iter *iter)
2043 struct tnode *tn = iter->tnode;
2044 unsigned int cindex = iter->index;
2045 struct tnode *p;
2047 /* A single entry routing table */
2048 if (!tn)
2049 return NULL;
2051 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2052 iter->tnode, iter->index, iter->depth);
2053 rescan:
2054 while (cindex < (1<<tn->bits)) {
2055 struct node *n = tnode_get_child_rcu(tn, cindex);
2057 if (n) {
2058 if (IS_LEAF(n)) {
2059 iter->tnode = tn;
2060 iter->index = cindex + 1;
2061 } else {
2062 /* push down one level */
2063 iter->tnode = (struct tnode *) n;
2064 iter->index = 0;
2065 ++iter->depth;
2067 return n;
2070 ++cindex;
2073 /* Current node exhausted, pop back up */
2074 p = node_parent_rcu((struct node *)tn);
2075 if (p) {
2076 cindex = tkey_extract_bits(tn->key, p->pos, p->bits)+1;
2077 tn = p;
2078 --iter->depth;
2079 goto rescan;
2082 /* got root? */
2083 return NULL;
2086 static struct node *fib_trie_get_first(struct fib_trie_iter *iter,
2087 struct trie *t)
2089 struct node *n;
2091 if (!t)
2092 return NULL;
2094 n = rcu_dereference(t->trie);
2095 if (!n)
2096 return NULL;
2098 if (IS_TNODE(n)) {
2099 iter->tnode = (struct tnode *) n;
2100 iter->index = 0;
2101 iter->depth = 1;
2102 } else {
2103 iter->tnode = NULL;
2104 iter->index = 0;
2105 iter->depth = 0;
2108 return n;
2111 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2113 struct node *n;
2114 struct fib_trie_iter iter;
2116 memset(s, 0, sizeof(*s));
2118 rcu_read_lock();
2119 for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2120 if (IS_LEAF(n)) {
2121 struct leaf *l = (struct leaf *)n;
2122 struct leaf_info *li;
2123 struct hlist_node *tmp;
2125 s->leaves++;
2126 s->totdepth += iter.depth;
2127 if (iter.depth > s->maxdepth)
2128 s->maxdepth = iter.depth;
2130 hlist_for_each_entry_rcu(li, tmp, &l->list, hlist)
2131 ++s->prefixes;
2132 } else {
2133 const struct tnode *tn = (const struct tnode *) n;
2134 int i;
2136 s->tnodes++;
2137 if (tn->bits < MAX_STAT_DEPTH)
2138 s->nodesizes[tn->bits]++;
2140 for (i = 0; i < (1<<tn->bits); i++)
2141 if (!tn->child[i])
2142 s->nullpointers++;
2145 rcu_read_unlock();
2149 * This outputs /proc/net/fib_triestats
2151 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2153 unsigned int i, max, pointers, bytes, avdepth;
2155 if (stat->leaves)
2156 avdepth = stat->totdepth*100 / stat->leaves;
2157 else
2158 avdepth = 0;
2160 seq_printf(seq, "\tAver depth: %u.%02d\n",
2161 avdepth / 100, avdepth % 100);
2162 seq_printf(seq, "\tMax depth: %u\n", stat->maxdepth);
2164 seq_printf(seq, "\tLeaves: %u\n", stat->leaves);
2165 bytes = sizeof(struct leaf) * stat->leaves;
2167 seq_printf(seq, "\tPrefixes: %u\n", stat->prefixes);
2168 bytes += sizeof(struct leaf_info) * stat->prefixes;
2170 seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2171 bytes += sizeof(struct tnode) * stat->tnodes;
2173 max = MAX_STAT_DEPTH;
2174 while (max > 0 && stat->nodesizes[max-1] == 0)
2175 max--;
2177 pointers = 0;
2178 for (i = 1; i <= max; i++)
2179 if (stat->nodesizes[i] != 0) {
2180 seq_printf(seq, " %u: %u", i, stat->nodesizes[i]);
2181 pointers += (1<<i) * stat->nodesizes[i];
2183 seq_putc(seq, '\n');
2184 seq_printf(seq, "\tPointers: %u\n", pointers);
2186 bytes += sizeof(struct node *) * pointers;
2187 seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2188 seq_printf(seq, "Total size: %u kB\n", (bytes + 1023) / 1024);
2191 #ifdef CONFIG_IP_FIB_TRIE_STATS
2192 static void trie_show_usage(struct seq_file *seq,
2193 const struct trie_use_stats *stats)
2195 seq_printf(seq, "\nCounters:\n---------\n");
2196 seq_printf(seq, "gets = %u\n", stats->gets);
2197 seq_printf(seq, "backtracks = %u\n", stats->backtrack);
2198 seq_printf(seq, "semantic match passed = %u\n",
2199 stats->semantic_match_passed);
2200 seq_printf(seq, "semantic match miss = %u\n",
2201 stats->semantic_match_miss);
2202 seq_printf(seq, "null node hit= %u\n", stats->null_node_hit);
2203 seq_printf(seq, "skipped node resize = %u\n\n",
2204 stats->resize_node_skipped);
2206 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2208 static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2210 if (tb->tb_id == RT_TABLE_LOCAL)
2211 seq_puts(seq, "Local:\n");
2212 else if (tb->tb_id == RT_TABLE_MAIN)
2213 seq_puts(seq, "Main:\n");
2214 else
2215 seq_printf(seq, "Id %d:\n", tb->tb_id);
2219 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2221 struct net *net = (struct net *)seq->private;
2222 unsigned int h;
2224 seq_printf(seq,
2225 "Basic info: size of leaf:"
2226 " %Zd bytes, size of tnode: %Zd bytes.\n",
2227 sizeof(struct leaf), sizeof(struct tnode));
2229 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2230 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2231 struct hlist_node *node;
2232 struct fib_table *tb;
2234 hlist_for_each_entry_rcu(tb, node, head, tb_hlist) {
2235 struct trie *t = (struct trie *) tb->tb_data;
2236 struct trie_stat stat;
2238 if (!t)
2239 continue;
2241 fib_table_print(seq, tb);
2243 trie_collect_stats(t, &stat);
2244 trie_show_stats(seq, &stat);
2245 #ifdef CONFIG_IP_FIB_TRIE_STATS
2246 trie_show_usage(seq, &t->stats);
2247 #endif
2251 return 0;
2254 static int fib_triestat_seq_open(struct inode *inode, struct file *file)
2256 return single_open_net(inode, file, fib_triestat_seq_show);
2259 static const struct file_operations fib_triestat_fops = {
2260 .owner = THIS_MODULE,
2261 .open = fib_triestat_seq_open,
2262 .read = seq_read,
2263 .llseek = seq_lseek,
2264 .release = single_release_net,
2267 static struct node *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2269 struct fib_trie_iter *iter = seq->private;
2270 struct net *net = seq_file_net(seq);
2271 loff_t idx = 0;
2272 unsigned int h;
2274 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2275 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2276 struct hlist_node *node;
2277 struct fib_table *tb;
2279 hlist_for_each_entry_rcu(tb, node, head, tb_hlist) {
2280 struct node *n;
2282 for (n = fib_trie_get_first(iter,
2283 (struct trie *) tb->tb_data);
2284 n; n = fib_trie_get_next(iter))
2285 if (pos == idx++) {
2286 iter->tb = tb;
2287 return n;
2292 return NULL;
2295 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2296 __acquires(RCU)
2298 rcu_read_lock();
2299 return fib_trie_get_idx(seq, *pos);
2302 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2304 struct fib_trie_iter *iter = seq->private;
2305 struct net *net = seq_file_net(seq);
2306 struct fib_table *tb = iter->tb;
2307 struct hlist_node *tb_node;
2308 unsigned int h;
2309 struct node *n;
2311 ++*pos;
2312 /* next node in same table */
2313 n = fib_trie_get_next(iter);
2314 if (n)
2315 return n;
2317 /* walk rest of this hash chain */
2318 h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2319 while ( (tb_node = rcu_dereference(tb->tb_hlist.next)) ) {
2320 tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2321 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2322 if (n)
2323 goto found;
2326 /* new hash chain */
2327 while (++h < FIB_TABLE_HASHSZ) {
2328 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2329 hlist_for_each_entry_rcu(tb, tb_node, head, tb_hlist) {
2330 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2331 if (n)
2332 goto found;
2335 return NULL;
2337 found:
2338 iter->tb = tb;
2339 return n;
2342 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2343 __releases(RCU)
2345 rcu_read_unlock();
2348 static void seq_indent(struct seq_file *seq, int n)
2350 while (n-- > 0)
2351 seq_puts(seq, " ");
2354 static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2356 switch (s) {
2357 case RT_SCOPE_UNIVERSE: return "universe";
2358 case RT_SCOPE_SITE: return "site";
2359 case RT_SCOPE_LINK: return "link";
2360 case RT_SCOPE_HOST: return "host";
2361 case RT_SCOPE_NOWHERE: return "nowhere";
2362 default:
2363 snprintf(buf, len, "scope=%d", s);
2364 return buf;
2368 static const char *const rtn_type_names[__RTN_MAX] = {
2369 [RTN_UNSPEC] = "UNSPEC",
2370 [RTN_UNICAST] = "UNICAST",
2371 [RTN_LOCAL] = "LOCAL",
2372 [RTN_BROADCAST] = "BROADCAST",
2373 [RTN_ANYCAST] = "ANYCAST",
2374 [RTN_MULTICAST] = "MULTICAST",
2375 [RTN_BLACKHOLE] = "BLACKHOLE",
2376 [RTN_UNREACHABLE] = "UNREACHABLE",
2377 [RTN_PROHIBIT] = "PROHIBIT",
2378 [RTN_THROW] = "THROW",
2379 [RTN_NAT] = "NAT",
2380 [RTN_XRESOLVE] = "XRESOLVE",
2383 static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2385 if (t < __RTN_MAX && rtn_type_names[t])
2386 return rtn_type_names[t];
2387 snprintf(buf, len, "type %u", t);
2388 return buf;
2391 /* Pretty print the trie */
2392 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2394 const struct fib_trie_iter *iter = seq->private;
2395 struct node *n = v;
2397 if (!node_parent_rcu(n))
2398 fib_table_print(seq, iter->tb);
2400 if (IS_TNODE(n)) {
2401 struct tnode *tn = (struct tnode *) n;
2402 __be32 prf = htonl(mask_pfx(tn->key, tn->pos));
2404 seq_indent(seq, iter->depth-1);
2405 seq_printf(seq, " +-- %pI4/%d %d %d %d\n",
2406 &prf, tn->pos, tn->bits, tn->full_children,
2407 tn->empty_children);
2409 } else {
2410 struct leaf *l = (struct leaf *) n;
2411 struct leaf_info *li;
2412 struct hlist_node *node;
2413 __be32 val = htonl(l->key);
2415 seq_indent(seq, iter->depth);
2416 seq_printf(seq, " |-- %pI4\n", &val);
2418 hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
2419 struct fib_alias *fa;
2421 list_for_each_entry_rcu(fa, &li->falh, fa_list) {
2422 char buf1[32], buf2[32];
2424 seq_indent(seq, iter->depth+1);
2425 seq_printf(seq, " /%d %s %s", li->plen,
2426 rtn_scope(buf1, sizeof(buf1),
2427 fa->fa_scope),
2428 rtn_type(buf2, sizeof(buf2),
2429 fa->fa_type));
2430 if (fa->fa_tos)
2431 seq_printf(seq, " tos=%d", fa->fa_tos);
2432 seq_putc(seq, '\n');
2437 return 0;
2440 static const struct seq_operations fib_trie_seq_ops = {
2441 .start = fib_trie_seq_start,
2442 .next = fib_trie_seq_next,
2443 .stop = fib_trie_seq_stop,
2444 .show = fib_trie_seq_show,
2447 static int fib_trie_seq_open(struct inode *inode, struct file *file)
2449 return seq_open_net(inode, file, &fib_trie_seq_ops,
2450 sizeof(struct fib_trie_iter));
2453 static const struct file_operations fib_trie_fops = {
2454 .owner = THIS_MODULE,
2455 .open = fib_trie_seq_open,
2456 .read = seq_read,
2457 .llseek = seq_lseek,
2458 .release = seq_release_net,
2461 struct fib_route_iter {
2462 struct seq_net_private p;
2463 struct trie *main_trie;
2464 loff_t pos;
2465 t_key key;
2468 static struct leaf *fib_route_get_idx(struct fib_route_iter *iter, loff_t pos)
2470 struct leaf *l = NULL;
2471 struct trie *t = iter->main_trie;
2473 /* use cache location of last found key */
2474 if (iter->pos > 0 && pos >= iter->pos && (l = fib_find_node(t, iter->key)))
2475 pos -= iter->pos;
2476 else {
2477 iter->pos = 0;
2478 l = trie_firstleaf(t);
2481 while (l && pos-- > 0) {
2482 iter->pos++;
2483 l = trie_nextleaf(l);
2486 if (l)
2487 iter->key = pos; /* remember it */
2488 else
2489 iter->pos = 0; /* forget it */
2491 return l;
2494 static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2495 __acquires(RCU)
2497 struct fib_route_iter *iter = seq->private;
2498 struct fib_table *tb;
2500 rcu_read_lock();
2501 tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2502 if (!tb)
2503 return NULL;
2505 iter->main_trie = (struct trie *) tb->tb_data;
2506 if (*pos == 0)
2507 return SEQ_START_TOKEN;
2508 else
2509 return fib_route_get_idx(iter, *pos - 1);
2512 static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2514 struct fib_route_iter *iter = seq->private;
2515 struct leaf *l = v;
2517 ++*pos;
2518 if (v == SEQ_START_TOKEN) {
2519 iter->pos = 0;
2520 l = trie_firstleaf(iter->main_trie);
2521 } else {
2522 iter->pos++;
2523 l = trie_nextleaf(l);
2526 if (l)
2527 iter->key = l->key;
2528 else
2529 iter->pos = 0;
2530 return l;
2533 static void fib_route_seq_stop(struct seq_file *seq, void *v)
2534 __releases(RCU)
2536 rcu_read_unlock();
2539 static unsigned int fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2541 unsigned int flags = 0;
2543 if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2544 flags = RTF_REJECT;
2545 if (fi && fi->fib_nh->nh_gw)
2546 flags |= RTF_GATEWAY;
2547 if (mask == htonl(0xFFFFFFFF))
2548 flags |= RTF_HOST;
2549 flags |= RTF_UP;
2550 return flags;
2554 * This outputs /proc/net/route.
2555 * The format of the file is not supposed to be changed
2556 * and needs to be same as fib_hash output to avoid breaking
2557 * legacy utilities
2559 static int fib_route_seq_show(struct seq_file *seq, void *v)
2561 struct leaf *l = v;
2562 struct leaf_info *li;
2563 struct hlist_node *node;
2565 if (v == SEQ_START_TOKEN) {
2566 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2567 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2568 "\tWindow\tIRTT");
2569 return 0;
2572 hlist_for_each_entry_rcu(li, node, &l->list, hlist) {
2573 struct fib_alias *fa;
2574 __be32 mask, prefix;
2576 mask = inet_make_mask(li->plen);
2577 prefix = htonl(l->key);
2579 list_for_each_entry_rcu(fa, &li->falh, fa_list) {
2580 const struct fib_info *fi = fa->fa_info;
2581 unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2582 int len;
2584 if (fa->fa_type == RTN_BROADCAST
2585 || fa->fa_type == RTN_MULTICAST)
2586 continue;
2588 if (fi)
2589 seq_printf(seq,
2590 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2591 "%d\t%08X\t%d\t%u\t%u%n",
2592 fi->fib_dev ? fi->fib_dev->name : "*",
2593 prefix,
2594 fi->fib_nh->nh_gw, flags, 0, 0,
2595 fi->fib_priority,
2596 mask,
2597 (fi->fib_advmss ?
2598 fi->fib_advmss + 40 : 0),
2599 fi->fib_window,
2600 fi->fib_rtt >> 3, &len);
2601 else
2602 seq_printf(seq,
2603 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2604 "%d\t%08X\t%d\t%u\t%u%n",
2605 prefix, 0, flags, 0, 0, 0,
2606 mask, 0, 0, 0, &len);
2608 seq_printf(seq, "%*s\n", 127 - len, "");
2612 return 0;
2615 static const struct seq_operations fib_route_seq_ops = {
2616 .start = fib_route_seq_start,
2617 .next = fib_route_seq_next,
2618 .stop = fib_route_seq_stop,
2619 .show = fib_route_seq_show,
2622 static int fib_route_seq_open(struct inode *inode, struct file *file)
2624 return seq_open_net(inode, file, &fib_route_seq_ops,
2625 sizeof(struct fib_route_iter));
2628 static const struct file_operations fib_route_fops = {
2629 .owner = THIS_MODULE,
2630 .open = fib_route_seq_open,
2631 .read = seq_read,
2632 .llseek = seq_lseek,
2633 .release = seq_release_net,
2636 int __net_init fib_proc_init(struct net *net)
2638 if (!proc_net_fops_create(net, "fib_trie", S_IRUGO, &fib_trie_fops))
2639 goto out1;
2641 if (!proc_net_fops_create(net, "fib_triestat", S_IRUGO,
2642 &fib_triestat_fops))
2643 goto out2;
2645 if (!proc_net_fops_create(net, "route", S_IRUGO, &fib_route_fops))
2646 goto out3;
2648 return 0;
2650 out3:
2651 proc_net_remove(net, "fib_triestat");
2652 out2:
2653 proc_net_remove(net, "fib_trie");
2654 out1:
2655 return -ENOMEM;
2658 void __net_exit fib_proc_exit(struct net *net)
2660 proc_net_remove(net, "fib_trie");
2661 proc_net_remove(net, "fib_triestat");
2662 proc_net_remove(net, "route");
2665 #endif /* CONFIG_PROC_FS */