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 described 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.csc.kth.se/~snilsson/software/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 <linux/bitops.h>
55 #include <linux/types.h>
56 #include <linux/kernel.h>
58 #include <linux/string.h>
59 #include <linux/socket.h>
60 #include <linux/sockios.h>
61 #include <linux/errno.h>
63 #include <linux/inet.h>
64 #include <linux/inetdevice.h>
65 #include <linux/netdevice.h>
66 #include <linux/if_arp.h>
67 #include <linux/proc_fs.h>
68 #include <linux/rcupdate.h>
69 #include <linux/skbuff.h>
70 #include <linux/netlink.h>
71 #include <linux/init.h>
72 #include <linux/list.h>
73 #include <linux/slab.h>
74 #include <linux/export.h>
75 #include <net/net_namespace.h>
77 #include <net/protocol.h>
78 #include <net/route.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 IS_TNODE(n) ((n)->bits)
91 #define IS_LEAF(n) (!(n)->bits)
93 #define get_index(_key, _kv) (((_key) ^ (_kv)->key) >> (_kv)->pos)
97 unsigned char bits
; /* 2log(KEYLENGTH) bits needed */
98 unsigned char pos
; /* 2log(KEYLENGTH) bits needed */
99 struct tnode __rcu
*parent
;
102 /* The fields in this struct are valid if bits > 0 (TNODE) */
104 unsigned int full_children
; /* KEYLENGTH bits needed */
105 unsigned int empty_children
; /* KEYLENGTH bits needed */
106 struct tnode __rcu
*child
[0];
108 /* This list pointer if valid if bits == 0 (LEAF) */
109 struct hlist_head list
;
114 struct hlist_node hlist
;
116 u32 mask_plen
; /* ntohl(inet_make_mask(plen)) */
117 struct list_head falh
;
121 #ifdef CONFIG_IP_FIB_TRIE_STATS
122 struct trie_use_stats
{
124 unsigned int backtrack
;
125 unsigned int semantic_match_passed
;
126 unsigned int semantic_match_miss
;
127 unsigned int null_node_hit
;
128 unsigned int resize_node_skipped
;
133 unsigned int totdepth
;
134 unsigned int maxdepth
;
137 unsigned int nullpointers
;
138 unsigned int prefixes
;
139 unsigned int nodesizes
[MAX_STAT_DEPTH
];
143 struct tnode __rcu
*trie
;
144 #ifdef CONFIG_IP_FIB_TRIE_STATS
145 struct trie_use_stats __percpu
*stats
;
149 static void resize(struct trie
*t
, struct tnode
*tn
);
150 static size_t tnode_free_size
;
153 * synchronize_rcu after call_rcu for that many pages; it should be especially
154 * useful before resizing the root node with PREEMPT_NONE configs; the value was
155 * obtained experimentally, aiming to avoid visible slowdown.
157 static const int sync_pages
= 128;
159 static struct kmem_cache
*fn_alias_kmem __read_mostly
;
160 static struct kmem_cache
*trie_leaf_kmem __read_mostly
;
162 /* caller must hold RTNL */
163 #define node_parent(n) rtnl_dereference((n)->parent)
165 /* caller must hold RCU read lock or RTNL */
166 #define node_parent_rcu(n) rcu_dereference_rtnl((n)->parent)
168 /* wrapper for rcu_assign_pointer */
169 static inline void node_set_parent(struct tnode
*n
, struct tnode
*tp
)
172 rcu_assign_pointer(n
->parent
, tp
);
175 #define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER((n)->parent, p)
177 /* This provides us with the number of children in this node, in the case of a
178 * leaf this will return 0 meaning none of the children are accessible.
180 static inline unsigned long tnode_child_length(const struct tnode
*tn
)
182 return (1ul << tn
->bits
) & ~(1ul);
185 /* caller must hold RTNL */
186 static inline struct tnode
*tnode_get_child(const struct tnode
*tn
,
189 return rtnl_dereference(tn
->child
[i
]);
192 /* caller must hold RCU read lock or RTNL */
193 static inline struct tnode
*tnode_get_child_rcu(const struct tnode
*tn
,
196 return rcu_dereference_rtnl(tn
->child
[i
]);
199 /* To understand this stuff, an understanding of keys and all their bits is
200 * necessary. Every node in the trie has a key associated with it, but not
201 * all of the bits in that key are significant.
203 * Consider a node 'n' and its parent 'tp'.
205 * If n is a leaf, every bit in its key is significant. Its presence is
206 * necessitated by path compression, since during a tree traversal (when
207 * searching for a leaf - unless we are doing an insertion) we will completely
208 * ignore all skipped bits we encounter. Thus we need to verify, at the end of
209 * a potentially successful search, that we have indeed been walking the
212 * Note that we can never "miss" the correct key in the tree if present by
213 * following the wrong path. Path compression ensures that segments of the key
214 * that are the same for all keys with a given prefix are skipped, but the
215 * skipped part *is* identical for each node in the subtrie below the skipped
216 * bit! trie_insert() in this implementation takes care of that.
218 * if n is an internal node - a 'tnode' here, the various parts of its key
219 * have many different meanings.
222 * _________________________________________________________________
223 * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
224 * -----------------------------------------------------------------
225 * 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16
227 * _________________________________________________________________
228 * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
229 * -----------------------------------------------------------------
230 * 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
237 * First, let's just ignore the bits that come before the parent tp, that is
238 * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
239 * point we do not use them for anything.
241 * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
242 * index into the parent's child array. That is, they will be used to find
243 * 'n' among tp's children.
245 * The bits from (n->pos + n->bits) to (tn->pos - 1) - "S" - are skipped bits
248 * All the bits we have seen so far are significant to the node n. The rest
249 * of the bits are really not needed or indeed known in n->key.
251 * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
252 * n's child array, and will of course be different for each child.
254 * The rest of the bits, from 0 to (n->pos + n->bits), are completely unknown
258 static const int halve_threshold
= 25;
259 static const int inflate_threshold
= 50;
260 static const int halve_threshold_root
= 15;
261 static const int inflate_threshold_root
= 30;
263 static void __alias_free_mem(struct rcu_head
*head
)
265 struct fib_alias
*fa
= container_of(head
, struct fib_alias
, rcu
);
266 kmem_cache_free(fn_alias_kmem
, fa
);
269 static inline void alias_free_mem_rcu(struct fib_alias
*fa
)
271 call_rcu(&fa
->rcu
, __alias_free_mem
);
274 #define TNODE_KMALLOC_MAX \
275 ilog2((PAGE_SIZE - sizeof(struct tnode)) / sizeof(struct tnode *))
277 static void __node_free_rcu(struct rcu_head
*head
)
279 struct tnode
*n
= container_of(head
, struct tnode
, rcu
);
282 kmem_cache_free(trie_leaf_kmem
, n
);
283 else if (n
->bits
<= TNODE_KMALLOC_MAX
)
289 #define node_free(n) call_rcu(&n->rcu, __node_free_rcu)
291 static inline void free_leaf_info(struct leaf_info
*leaf
)
293 kfree_rcu(leaf
, rcu
);
296 static struct tnode
*tnode_alloc(size_t size
)
298 if (size
<= PAGE_SIZE
)
299 return kzalloc(size
, GFP_KERNEL
);
301 return vzalloc(size
);
304 static struct tnode
*leaf_new(t_key key
)
306 struct tnode
*l
= kmem_cache_alloc(trie_leaf_kmem
, GFP_KERNEL
);
309 /* set key and pos to reflect full key value
310 * any trailing zeros in the key should be ignored
311 * as the nodes are searched
315 /* set bits to 0 indicating we are not a tnode */
318 INIT_HLIST_HEAD(&l
->list
);
323 static struct leaf_info
*leaf_info_new(int plen
)
325 struct leaf_info
*li
= kmalloc(sizeof(struct leaf_info
), GFP_KERNEL
);
328 li
->mask_plen
= ntohl(inet_make_mask(plen
));
329 INIT_LIST_HEAD(&li
->falh
);
334 static struct tnode
*tnode_new(t_key key
, int pos
, int bits
)
336 size_t sz
= offsetof(struct tnode
, child
[1 << bits
]);
337 struct tnode
*tn
= tnode_alloc(sz
);
338 unsigned int shift
= pos
+ bits
;
340 /* verify bits and pos their msb bits clear and values are valid */
341 BUG_ON(!bits
|| (shift
> KEYLENGTH
));
347 tn
->key
= (shift
< KEYLENGTH
) ? (key
>> shift
) << shift
: 0;
348 tn
->full_children
= 0;
349 tn
->empty_children
= 1<<bits
;
352 pr_debug("AT %p s=%zu %zu\n", tn
, sizeof(struct tnode
),
353 sizeof(struct tnode
*) << bits
);
357 /* Check whether a tnode 'n' is "full", i.e. it is an internal node
358 * and no bits are skipped. See discussion in dyntree paper p. 6
360 static inline int tnode_full(const struct tnode
*tn
, const struct tnode
*n
)
362 return n
&& ((n
->pos
+ n
->bits
) == tn
->pos
) && IS_TNODE(n
);
365 /* Add a child at position i overwriting the old value.
366 * Update the value of full_children and empty_children.
368 static void put_child(struct tnode
*tn
, unsigned long i
, struct tnode
*n
)
370 struct tnode
*chi
= tnode_get_child(tn
, i
);
373 BUG_ON(i
>= tnode_child_length(tn
));
375 /* update emptyChildren */
376 if (n
== NULL
&& chi
!= NULL
)
377 tn
->empty_children
++;
378 else if (n
!= NULL
&& chi
== NULL
)
379 tn
->empty_children
--;
381 /* update fullChildren */
382 wasfull
= tnode_full(tn
, chi
);
383 isfull
= tnode_full(tn
, n
);
385 if (wasfull
&& !isfull
)
387 else if (!wasfull
&& isfull
)
390 rcu_assign_pointer(tn
->child
[i
], n
);
393 static void put_child_root(struct tnode
*tp
, struct trie
*t
,
394 t_key key
, struct tnode
*n
)
397 put_child(tp
, get_index(key
, tp
), n
);
399 rcu_assign_pointer(t
->trie
, n
);
402 static inline void tnode_free_init(struct tnode
*tn
)
407 static inline void tnode_free_append(struct tnode
*tn
, struct tnode
*n
)
409 n
->rcu
.next
= tn
->rcu
.next
;
410 tn
->rcu
.next
= &n
->rcu
;
413 static void tnode_free(struct tnode
*tn
)
415 struct callback_head
*head
= &tn
->rcu
;
419 tnode_free_size
+= offsetof(struct tnode
, child
[1 << tn
->bits
]);
422 tn
= container_of(head
, struct tnode
, rcu
);
425 if (tnode_free_size
>= PAGE_SIZE
* sync_pages
) {
431 static int inflate(struct trie
*t
, struct tnode
*oldtnode
)
433 struct tnode
*inode
, *node0
, *node1
, *tn
, *tp
;
434 unsigned long i
, j
, k
;
437 pr_debug("In inflate\n");
439 tn
= tnode_new(oldtnode
->key
, oldtnode
->pos
- 1, oldtnode
->bits
+ 1);
443 /* Assemble all of the pointers in our cluster, in this case that
444 * represents all of the pointers out of our allocated nodes that
445 * point to existing tnodes and the links between our allocated
448 for (i
= tnode_child_length(oldtnode
), m
= 1u << tn
->pos
; i
;) {
449 inode
= tnode_get_child(oldtnode
, --i
);
455 /* A leaf or an internal node with skipped bits */
456 if (!tnode_full(oldtnode
, inode
)) {
457 put_child(tn
, get_index(inode
->key
, tn
), inode
);
461 /* An internal node with two children */
462 if (inode
->bits
== 1) {
463 put_child(tn
, 2 * i
+ 1, tnode_get_child(inode
, 1));
464 put_child(tn
, 2 * i
, tnode_get_child(inode
, 0));
468 /* We will replace this node 'inode' with two new
469 * ones, 'node0' and 'node1', each with half of the
470 * original children. The two new nodes will have
471 * a position one bit further down the key and this
472 * means that the "significant" part of their keys
473 * (see the discussion near the top of this file)
474 * will differ by one bit, which will be "0" in
475 * node0's key and "1" in node1's key. Since we are
476 * moving the key position by one step, the bit that
477 * we are moving away from - the bit at position
478 * (tn->pos) - is the one that will differ between
479 * node0 and node1. So... we synthesize that bit in the
482 node1
= tnode_new(inode
->key
| m
, inode
->pos
, inode
->bits
- 1);
485 tnode_free_append(tn
, node1
);
487 node0
= tnode_new(inode
->key
& ~m
, inode
->pos
, inode
->bits
- 1);
490 tnode_free_append(tn
, node0
);
492 /* populate child pointers in new nodes */
493 for (k
= tnode_child_length(inode
), j
= k
/ 2; j
;) {
494 put_child(node1
, --j
, tnode_get_child(inode
, --k
));
495 put_child(node0
, j
, tnode_get_child(inode
, j
));
496 put_child(node1
, --j
, tnode_get_child(inode
, --k
));
497 put_child(node0
, j
, tnode_get_child(inode
, j
));
500 /* link new nodes to parent */
501 NODE_INIT_PARENT(node1
, tn
);
502 NODE_INIT_PARENT(node0
, tn
);
504 /* link parent to nodes */
505 put_child(tn
, 2 * i
+ 1, node1
);
506 put_child(tn
, 2 * i
, node0
);
509 /* setup the parent pointer into and out of this node */
510 tp
= node_parent(oldtnode
);
511 NODE_INIT_PARENT(tn
, tp
);
512 put_child_root(tp
, t
, tn
->key
, tn
);
514 /* prepare oldtnode to be freed */
515 tnode_free_init(oldtnode
);
517 /* update all child nodes parent pointers to route to us */
518 for (i
= tnode_child_length(oldtnode
); i
;) {
519 inode
= tnode_get_child(oldtnode
, --i
);
521 /* A leaf or an internal node with skipped bits */
522 if (!tnode_full(oldtnode
, inode
)) {
523 node_set_parent(inode
, tn
);
527 /* drop the node in the old tnode free list */
528 tnode_free_append(oldtnode
, inode
);
530 /* fetch new nodes */
531 node1
= tnode_get_child(tn
, 2 * i
+ 1);
532 node0
= tnode_get_child(tn
, 2 * i
);
534 /* bits == 1 then node0 and node1 represent inode's children */
535 if (inode
->bits
== 1) {
536 node_set_parent(node1
, tn
);
537 node_set_parent(node0
, tn
);
541 /* update parent pointers in child node's children */
542 for (k
= tnode_child_length(inode
), j
= k
/ 2; j
;) {
543 node_set_parent(tnode_get_child(inode
, --k
), node1
);
544 node_set_parent(tnode_get_child(inode
, --j
), node0
);
545 node_set_parent(tnode_get_child(inode
, --k
), node1
);
546 node_set_parent(tnode_get_child(inode
, --j
), node0
);
549 /* resize child nodes */
554 /* we completed without error, prepare to free old node */
555 tnode_free(oldtnode
);
558 /* all pointers should be clean so we are done */
563 static int halve(struct trie
*t
, struct tnode
*oldtnode
)
565 struct tnode
*tn
, *tp
, *inode
, *node0
, *node1
;
568 pr_debug("In halve\n");
570 tn
= tnode_new(oldtnode
->key
, oldtnode
->pos
+ 1, oldtnode
->bits
- 1);
574 /* Assemble all of the pointers in our cluster, in this case that
575 * represents all of the pointers out of our allocated nodes that
576 * point to existing tnodes and the links between our allocated
579 for (i
= tnode_child_length(oldtnode
); i
;) {
580 node1
= tnode_get_child(oldtnode
, --i
);
581 node0
= tnode_get_child(oldtnode
, --i
);
583 /* At least one of the children is empty */
584 if (!node1
|| !node0
) {
585 put_child(tn
, i
/ 2, node1
? : node0
);
589 /* Two nonempty children */
590 inode
= tnode_new(node0
->key
, oldtnode
->pos
, 1);
595 tnode_free_append(tn
, inode
);
597 /* initialize pointers out of node */
598 put_child(inode
, 1, node1
);
599 put_child(inode
, 0, node0
);
600 NODE_INIT_PARENT(inode
, tn
);
602 /* link parent to node */
603 put_child(tn
, i
/ 2, inode
);
606 /* setup the parent pointer out of and back into this node */
607 tp
= node_parent(oldtnode
);
608 NODE_INIT_PARENT(tn
, tp
);
609 put_child_root(tp
, t
, tn
->key
, tn
);
611 /* prepare oldtnode to be freed */
612 tnode_free_init(oldtnode
);
614 /* update all of the child parent pointers */
615 for (i
= tnode_child_length(tn
); i
;) {
616 inode
= tnode_get_child(tn
, --i
);
618 /* only new tnodes will be considered "full" nodes */
619 if (!tnode_full(tn
, inode
)) {
620 node_set_parent(inode
, tn
);
624 /* Two nonempty children */
625 node_set_parent(tnode_get_child(inode
, 1), inode
);
626 node_set_parent(tnode_get_child(inode
, 0), inode
);
628 /* resize child node */
632 /* all pointers should be clean so we are done */
633 tnode_free(oldtnode
);
638 /* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
639 * the Helsinki University of Technology and Matti Tikkanen of Nokia
640 * Telecommunications, page 6:
641 * "A node is doubled if the ratio of non-empty children to all
642 * children in the *doubled* node is at least 'high'."
644 * 'high' in this instance is the variable 'inflate_threshold'. It
645 * is expressed as a percentage, so we multiply it with
646 * tnode_child_length() and instead of multiplying by 2 (since the
647 * child array will be doubled by inflate()) and multiplying
648 * the left-hand side by 100 (to handle the percentage thing) we
649 * multiply the left-hand side by 50.
651 * The left-hand side may look a bit weird: tnode_child_length(tn)
652 * - tn->empty_children is of course the number of non-null children
653 * in the current node. tn->full_children is the number of "full"
654 * children, that is non-null tnodes with a skip value of 0.
655 * All of those will be doubled in the resulting inflated tnode, so
656 * we just count them one extra time here.
658 * A clearer way to write this would be:
660 * to_be_doubled = tn->full_children;
661 * not_to_be_doubled = tnode_child_length(tn) - tn->empty_children -
664 * new_child_length = tnode_child_length(tn) * 2;
666 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
668 * if (new_fill_factor >= inflate_threshold)
670 * ...and so on, tho it would mess up the while () loop.
673 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
677 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
678 * inflate_threshold * new_child_length
680 * expand not_to_be_doubled and to_be_doubled, and shorten:
681 * 100 * (tnode_child_length(tn) - tn->empty_children +
682 * tn->full_children) >= inflate_threshold * new_child_length
684 * expand new_child_length:
685 * 100 * (tnode_child_length(tn) - tn->empty_children +
686 * tn->full_children) >=
687 * inflate_threshold * tnode_child_length(tn) * 2
690 * 50 * (tn->full_children + tnode_child_length(tn) -
691 * tn->empty_children) >= inflate_threshold *
692 * tnode_child_length(tn)
695 static bool should_inflate(const struct tnode
*tp
, const struct tnode
*tn
)
697 unsigned long used
= tnode_child_length(tn
);
698 unsigned long threshold
= used
;
700 /* Keep root node larger */
701 threshold
*= tp
? inflate_threshold
: inflate_threshold_root
;
702 used
+= tn
->full_children
;
703 used
-= tn
->empty_children
;
705 return tn
->pos
&& ((50 * used
) >= threshold
);
708 static bool should_halve(const struct tnode
*tp
, const struct tnode
*tn
)
710 unsigned long used
= tnode_child_length(tn
);
711 unsigned long threshold
= used
;
713 /* Keep root node larger */
714 threshold
*= tp
? halve_threshold
: halve_threshold_root
;
715 used
-= tn
->empty_children
;
717 return (tn
->bits
> 1) && ((100 * used
) < threshold
);
721 static void resize(struct trie
*t
, struct tnode
*tn
)
723 struct tnode
*tp
= node_parent(tn
), *n
= NULL
;
724 struct tnode __rcu
**cptr
;
727 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
728 tn
, inflate_threshold
, halve_threshold
);
730 /* track the tnode via the pointer from the parent instead of
731 * doing it ourselves. This way we can let RCU fully do its
732 * thing without us interfering
734 cptr
= tp
? &tp
->child
[get_index(tn
->key
, tp
)] : &t
->trie
;
735 BUG_ON(tn
!= rtnl_dereference(*cptr
));
738 if (tn
->empty_children
> (tnode_child_length(tn
) - 1))
742 if (tn
->empty_children
== (tnode_child_length(tn
) - 1))
745 /* Double as long as the resulting node has a number of
746 * nonempty nodes that are above the threshold.
749 while (should_inflate(tp
, tn
) && max_work
--) {
750 if (inflate(t
, tn
)) {
751 #ifdef CONFIG_IP_FIB_TRIE_STATS
752 this_cpu_inc(t
->stats
->resize_node_skipped
);
757 tn
= rtnl_dereference(*cptr
);
760 /* Return if at least one inflate is run */
761 if (max_work
!= MAX_WORK
)
764 /* Halve as long as the number of empty children in this
765 * node is above threshold.
768 while (should_halve(tp
, tn
) && max_work
--) {
770 #ifdef CONFIG_IP_FIB_TRIE_STATS
771 this_cpu_inc(t
->stats
->resize_node_skipped
);
776 tn
= rtnl_dereference(*cptr
);
779 /* Only one child remains */
780 if (tn
->empty_children
== (tnode_child_length(tn
) - 1)) {
783 for (i
= tnode_child_length(tn
); !n
&& i
;)
784 n
= tnode_get_child(tn
, --i
);
786 /* compress one level */
787 put_child_root(tp
, t
, tn
->key
, n
);
788 node_set_parent(n
, tp
);
796 /* readside must use rcu_read_lock currently dump routines
797 via get_fa_head and dump */
799 static struct leaf_info
*find_leaf_info(struct tnode
*l
, int plen
)
801 struct hlist_head
*head
= &l
->list
;
802 struct leaf_info
*li
;
804 hlist_for_each_entry_rcu(li
, head
, hlist
)
805 if (li
->plen
== plen
)
811 static inline struct list_head
*get_fa_head(struct tnode
*l
, int plen
)
813 struct leaf_info
*li
= find_leaf_info(l
, plen
);
821 static void insert_leaf_info(struct hlist_head
*head
, struct leaf_info
*new)
823 struct leaf_info
*li
= NULL
, *last
= NULL
;
825 if (hlist_empty(head
)) {
826 hlist_add_head_rcu(&new->hlist
, head
);
828 hlist_for_each_entry(li
, head
, hlist
) {
829 if (new->plen
> li
->plen
)
835 hlist_add_behind_rcu(&new->hlist
, &last
->hlist
);
837 hlist_add_before_rcu(&new->hlist
, &li
->hlist
);
841 /* rcu_read_lock needs to be hold by caller from readside */
842 static struct tnode
*fib_find_node(struct trie
*t
, u32 key
)
844 struct tnode
*n
= rcu_dereference_rtnl(t
->trie
);
847 unsigned long index
= get_index(key
, n
);
849 /* This bit of code is a bit tricky but it combines multiple
850 * checks into a single check. The prefix consists of the
851 * prefix plus zeros for the bits in the cindex. The index
852 * is the difference between the key and this value. From
853 * this we can actually derive several pieces of data.
854 * if !(index >> bits)
855 * we know the value is cindex
857 * we have a mismatch in skip bits and failed
859 if (index
>> n
->bits
)
862 /* we have found a leaf. Prefixes have already been compared */
866 n
= tnode_get_child_rcu(n
, index
);
872 static void trie_rebalance(struct trie
*t
, struct tnode
*tn
)
876 while ((tp
= node_parent(tn
)) != NULL
) {
881 /* Handle last (top) tnode */
886 /* only used from updater-side */
888 static struct list_head
*fib_insert_node(struct trie
*t
, u32 key
, int plen
)
890 struct list_head
*fa_head
= NULL
;
891 struct tnode
*l
, *n
, *tp
= NULL
;
892 struct leaf_info
*li
;
894 li
= leaf_info_new(plen
);
899 n
= rtnl_dereference(t
->trie
);
901 /* If we point to NULL, stop. Either the tree is empty and we should
902 * just put a new leaf in if, or we have reached an empty child slot,
903 * and we should just put our new leaf in that.
905 * If we hit a node with a key that does't match then we should stop
906 * and create a new tnode to replace that node and insert ourselves
907 * and the other node into the new tnode.
910 unsigned long index
= get_index(key
, n
);
912 /* This bit of code is a bit tricky but it combines multiple
913 * checks into a single check. The prefix consists of the
914 * prefix plus zeros for the "bits" in the prefix. The index
915 * is the difference between the key and this value. From
916 * this we can actually derive several pieces of data.
917 * if !(index >> bits)
918 * we know the value is child index
920 * we have a mismatch in skip bits and failed
922 if (index
>> n
->bits
)
925 /* we have found a leaf. Prefixes have already been compared */
927 /* Case 1: n is a leaf, and prefixes match*/
928 insert_leaf_info(&n
->list
, li
);
933 n
= tnode_get_child_rcu(n
, index
);
942 insert_leaf_info(&l
->list
, li
);
944 /* Case 2: n is a LEAF or a TNODE and the key doesn't match.
946 * Add a new tnode here
947 * first tnode need some special handling
948 * leaves us in position for handling as case 3
953 tn
= tnode_new(key
, __fls(key
^ n
->key
), 1);
960 /* initialize routes out of node */
961 NODE_INIT_PARENT(tn
, tp
);
962 put_child(tn
, get_index(key
, tn
) ^ 1, n
);
964 /* start adding routes into the node */
965 put_child_root(tp
, t
, key
, tn
);
966 node_set_parent(n
, tn
);
968 /* parent now has a NULL spot where the leaf can go */
972 /* Case 3: n is NULL, and will just insert a new leaf */
974 NODE_INIT_PARENT(l
, tp
);
975 put_child(tp
, get_index(key
, tp
), l
);
976 trie_rebalance(t
, tp
);
978 rcu_assign_pointer(t
->trie
, l
);
985 * Caller must hold RTNL.
987 int fib_table_insert(struct fib_table
*tb
, struct fib_config
*cfg
)
989 struct trie
*t
= (struct trie
*) tb
->tb_data
;
990 struct fib_alias
*fa
, *new_fa
;
991 struct list_head
*fa_head
= NULL
;
993 int plen
= cfg
->fc_dst_len
;
994 u8 tos
= cfg
->fc_tos
;
1002 key
= ntohl(cfg
->fc_dst
);
1004 pr_debug("Insert table=%u %08x/%d\n", tb
->tb_id
, key
, plen
);
1006 mask
= ntohl(inet_make_mask(plen
));
1013 fi
= fib_create_info(cfg
);
1019 l
= fib_find_node(t
, key
);
1023 fa_head
= get_fa_head(l
, plen
);
1024 fa
= fib_find_alias(fa_head
, tos
, fi
->fib_priority
);
1027 /* Now fa, if non-NULL, points to the first fib alias
1028 * with the same keys [prefix,tos,priority], if such key already
1029 * exists or to the node before which we will insert new one.
1031 * If fa is NULL, we will need to allocate a new one and
1032 * insert to the head of f.
1034 * If f is NULL, no fib node matched the destination key
1035 * and we need to allocate a new one of those as well.
1038 if (fa
&& fa
->fa_tos
== tos
&&
1039 fa
->fa_info
->fib_priority
== fi
->fib_priority
) {
1040 struct fib_alias
*fa_first
, *fa_match
;
1043 if (cfg
->fc_nlflags
& NLM_F_EXCL
)
1047 * 1. Find exact match for type, scope, fib_info to avoid
1049 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1053 fa
= list_entry(fa
->fa_list
.prev
, struct fib_alias
, fa_list
);
1054 list_for_each_entry_continue(fa
, fa_head
, fa_list
) {
1055 if (fa
->fa_tos
!= tos
)
1057 if (fa
->fa_info
->fib_priority
!= fi
->fib_priority
)
1059 if (fa
->fa_type
== cfg
->fc_type
&&
1060 fa
->fa_info
== fi
) {
1066 if (cfg
->fc_nlflags
& NLM_F_REPLACE
) {
1067 struct fib_info
*fi_drop
;
1077 new_fa
= kmem_cache_alloc(fn_alias_kmem
, GFP_KERNEL
);
1081 fi_drop
= fa
->fa_info
;
1082 new_fa
->fa_tos
= fa
->fa_tos
;
1083 new_fa
->fa_info
= fi
;
1084 new_fa
->fa_type
= cfg
->fc_type
;
1085 state
= fa
->fa_state
;
1086 new_fa
->fa_state
= state
& ~FA_S_ACCESSED
;
1088 list_replace_rcu(&fa
->fa_list
, &new_fa
->fa_list
);
1089 alias_free_mem_rcu(fa
);
1091 fib_release_info(fi_drop
);
1092 if (state
& FA_S_ACCESSED
)
1093 rt_cache_flush(cfg
->fc_nlinfo
.nl_net
);
1094 rtmsg_fib(RTM_NEWROUTE
, htonl(key
), new_fa
, plen
,
1095 tb
->tb_id
, &cfg
->fc_nlinfo
, NLM_F_REPLACE
);
1099 /* Error if we find a perfect match which
1100 * uses the same scope, type, and nexthop
1106 if (!(cfg
->fc_nlflags
& NLM_F_APPEND
))
1110 if (!(cfg
->fc_nlflags
& NLM_F_CREATE
))
1114 new_fa
= kmem_cache_alloc(fn_alias_kmem
, GFP_KERNEL
);
1118 new_fa
->fa_info
= fi
;
1119 new_fa
->fa_tos
= tos
;
1120 new_fa
->fa_type
= cfg
->fc_type
;
1121 new_fa
->fa_state
= 0;
1123 * Insert new entry to the list.
1127 fa_head
= fib_insert_node(t
, key
, plen
);
1128 if (unlikely(!fa_head
)) {
1130 goto out_free_new_fa
;
1135 tb
->tb_num_default
++;
1137 list_add_tail_rcu(&new_fa
->fa_list
,
1138 (fa
? &fa
->fa_list
: fa_head
));
1140 rt_cache_flush(cfg
->fc_nlinfo
.nl_net
);
1141 rtmsg_fib(RTM_NEWROUTE
, htonl(key
), new_fa
, plen
, tb
->tb_id
,
1142 &cfg
->fc_nlinfo
, 0);
1147 kmem_cache_free(fn_alias_kmem
, new_fa
);
1149 fib_release_info(fi
);
1154 static inline t_key
prefix_mismatch(t_key key
, struct tnode
*n
)
1156 t_key prefix
= n
->key
;
1158 return (key
^ prefix
) & (prefix
| -prefix
);
1161 /* should be called with rcu_read_lock */
1162 int fib_table_lookup(struct fib_table
*tb
, const struct flowi4
*flp
,
1163 struct fib_result
*res
, int fib_flags
)
1165 struct trie
*t
= (struct trie
*)tb
->tb_data
;
1166 #ifdef CONFIG_IP_FIB_TRIE_STATS
1167 struct trie_use_stats __percpu
*stats
= t
->stats
;
1169 const t_key key
= ntohl(flp
->daddr
);
1170 struct tnode
*n
, *pn
;
1171 struct leaf_info
*li
;
1174 n
= rcu_dereference(t
->trie
);
1178 #ifdef CONFIG_IP_FIB_TRIE_STATS
1179 this_cpu_inc(stats
->gets
);
1185 /* Step 1: Travel to the longest prefix match in the trie */
1187 unsigned long index
= get_index(key
, n
);
1189 /* This bit of code is a bit tricky but it combines multiple
1190 * checks into a single check. The prefix consists of the
1191 * prefix plus zeros for the "bits" in the prefix. The index
1192 * is the difference between the key and this value. From
1193 * this we can actually derive several pieces of data.
1194 * if !(index >> bits)
1195 * we know the value is child index
1197 * we have a mismatch in skip bits and failed
1199 if (index
>> n
->bits
)
1202 /* we have found a leaf. Prefixes have already been compared */
1206 /* only record pn and cindex if we are going to be chopping
1207 * bits later. Otherwise we are just wasting cycles.
1214 n
= tnode_get_child_rcu(n
, index
);
1219 /* Step 2: Sort out leaves and begin backtracing for longest prefix */
1221 /* record the pointer where our next node pointer is stored */
1222 struct tnode __rcu
**cptr
= n
->child
;
1224 /* This test verifies that none of the bits that differ
1225 * between the key and the prefix exist in the region of
1226 * the lsb and higher in the prefix.
1228 if (unlikely(prefix_mismatch(key
, n
)))
1231 /* exit out and process leaf */
1232 if (unlikely(IS_LEAF(n
)))
1235 /* Don't bother recording parent info. Since we are in
1236 * prefix match mode we will have to come back to wherever
1237 * we started this traversal anyway
1240 while ((n
= rcu_dereference(*cptr
)) == NULL
) {
1242 #ifdef CONFIG_IP_FIB_TRIE_STATS
1244 this_cpu_inc(stats
->null_node_hit
);
1246 /* If we are at cindex 0 there are no more bits for
1247 * us to strip at this level so we must ascend back
1248 * up one level to see if there are any more bits to
1249 * be stripped there.
1252 t_key pkey
= pn
->key
;
1254 pn
= node_parent_rcu(pn
);
1257 #ifdef CONFIG_IP_FIB_TRIE_STATS
1258 this_cpu_inc(stats
->backtrack
);
1260 /* Get Child's index */
1261 cindex
= get_index(pkey
, pn
);
1264 /* strip the least significant bit from the cindex */
1265 cindex
&= cindex
- 1;
1267 /* grab pointer for next child node */
1268 cptr
= &pn
->child
[cindex
];
1273 /* Step 3: Process the leaf, if that fails fall back to backtracing */
1274 hlist_for_each_entry_rcu(li
, &n
->list
, hlist
) {
1275 struct fib_alias
*fa
;
1277 if ((key
^ n
->key
) & li
->mask_plen
)
1280 list_for_each_entry_rcu(fa
, &li
->falh
, fa_list
) {
1281 struct fib_info
*fi
= fa
->fa_info
;
1284 if (fa
->fa_tos
&& fa
->fa_tos
!= flp
->flowi4_tos
)
1288 if (fa
->fa_info
->fib_scope
< flp
->flowi4_scope
)
1290 fib_alias_accessed(fa
);
1291 err
= fib_props
[fa
->fa_type
].error
;
1292 if (unlikely(err
< 0)) {
1293 #ifdef CONFIG_IP_FIB_TRIE_STATS
1294 this_cpu_inc(stats
->semantic_match_passed
);
1298 if (fi
->fib_flags
& RTNH_F_DEAD
)
1300 for (nhsel
= 0; nhsel
< fi
->fib_nhs
; nhsel
++) {
1301 const struct fib_nh
*nh
= &fi
->fib_nh
[nhsel
];
1303 if (nh
->nh_flags
& RTNH_F_DEAD
)
1305 if (flp
->flowi4_oif
&& flp
->flowi4_oif
!= nh
->nh_oif
)
1308 if (!(fib_flags
& FIB_LOOKUP_NOREF
))
1309 atomic_inc(&fi
->fib_clntref
);
1311 res
->prefixlen
= li
->plen
;
1312 res
->nh_sel
= nhsel
;
1313 res
->type
= fa
->fa_type
;
1314 res
->scope
= fi
->fib_scope
;
1317 res
->fa_head
= &li
->falh
;
1318 #ifdef CONFIG_IP_FIB_TRIE_STATS
1319 this_cpu_inc(stats
->semantic_match_passed
);
1325 #ifdef CONFIG_IP_FIB_TRIE_STATS
1326 this_cpu_inc(stats
->semantic_match_miss
);
1331 EXPORT_SYMBOL_GPL(fib_table_lookup
);
1334 * Remove the leaf and return parent.
1336 static void trie_leaf_remove(struct trie
*t
, struct tnode
*l
)
1338 struct tnode
*tp
= node_parent(l
);
1340 pr_debug("entering trie_leaf_remove(%p)\n", l
);
1343 put_child(tp
, get_index(l
->key
, tp
), NULL
);
1344 trie_rebalance(t
, tp
);
1346 RCU_INIT_POINTER(t
->trie
, NULL
);
1353 * Caller must hold RTNL.
1355 int fib_table_delete(struct fib_table
*tb
, struct fib_config
*cfg
)
1357 struct trie
*t
= (struct trie
*) tb
->tb_data
;
1359 int plen
= cfg
->fc_dst_len
;
1360 u8 tos
= cfg
->fc_tos
;
1361 struct fib_alias
*fa
, *fa_to_delete
;
1362 struct list_head
*fa_head
;
1364 struct leaf_info
*li
;
1369 key
= ntohl(cfg
->fc_dst
);
1370 mask
= ntohl(inet_make_mask(plen
));
1376 l
= fib_find_node(t
, key
);
1381 li
= find_leaf_info(l
, plen
);
1386 fa_head
= &li
->falh
;
1387 fa
= fib_find_alias(fa_head
, tos
, 0);
1392 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key
, plen
, tos
, t
);
1394 fa_to_delete
= NULL
;
1395 fa
= list_entry(fa
->fa_list
.prev
, struct fib_alias
, fa_list
);
1396 list_for_each_entry_continue(fa
, fa_head
, fa_list
) {
1397 struct fib_info
*fi
= fa
->fa_info
;
1399 if (fa
->fa_tos
!= tos
)
1402 if ((!cfg
->fc_type
|| fa
->fa_type
== cfg
->fc_type
) &&
1403 (cfg
->fc_scope
== RT_SCOPE_NOWHERE
||
1404 fa
->fa_info
->fib_scope
== cfg
->fc_scope
) &&
1405 (!cfg
->fc_prefsrc
||
1406 fi
->fib_prefsrc
== cfg
->fc_prefsrc
) &&
1407 (!cfg
->fc_protocol
||
1408 fi
->fib_protocol
== cfg
->fc_protocol
) &&
1409 fib_nh_match(cfg
, fi
) == 0) {
1419 rtmsg_fib(RTM_DELROUTE
, htonl(key
), fa
, plen
, tb
->tb_id
,
1420 &cfg
->fc_nlinfo
, 0);
1422 list_del_rcu(&fa
->fa_list
);
1425 tb
->tb_num_default
--;
1427 if (list_empty(fa_head
)) {
1428 hlist_del_rcu(&li
->hlist
);
1432 if (hlist_empty(&l
->list
))
1433 trie_leaf_remove(t
, l
);
1435 if (fa
->fa_state
& FA_S_ACCESSED
)
1436 rt_cache_flush(cfg
->fc_nlinfo
.nl_net
);
1438 fib_release_info(fa
->fa_info
);
1439 alias_free_mem_rcu(fa
);
1443 static int trie_flush_list(struct list_head
*head
)
1445 struct fib_alias
*fa
, *fa_node
;
1448 list_for_each_entry_safe(fa
, fa_node
, head
, fa_list
) {
1449 struct fib_info
*fi
= fa
->fa_info
;
1451 if (fi
&& (fi
->fib_flags
& RTNH_F_DEAD
)) {
1452 list_del_rcu(&fa
->fa_list
);
1453 fib_release_info(fa
->fa_info
);
1454 alias_free_mem_rcu(fa
);
1461 static int trie_flush_leaf(struct tnode
*l
)
1464 struct hlist_head
*lih
= &l
->list
;
1465 struct hlist_node
*tmp
;
1466 struct leaf_info
*li
= NULL
;
1468 hlist_for_each_entry_safe(li
, tmp
, lih
, hlist
) {
1469 found
+= trie_flush_list(&li
->falh
);
1471 if (list_empty(&li
->falh
)) {
1472 hlist_del_rcu(&li
->hlist
);
1480 * Scan for the next right leaf starting at node p->child[idx]
1481 * Since we have back pointer, no recursion necessary.
1483 static struct tnode
*leaf_walk_rcu(struct tnode
*p
, struct tnode
*c
)
1486 unsigned long idx
= c
? idx
= get_index(c
->key
, p
) + 1 : 0;
1488 while (idx
< tnode_child_length(p
)) {
1489 c
= tnode_get_child_rcu(p
, idx
++);
1496 /* Rescan start scanning in new node */
1501 /* Node empty, walk back up to parent */
1503 } while ((p
= node_parent_rcu(c
)) != NULL
);
1505 return NULL
; /* Root of trie */
1508 static struct tnode
*trie_firstleaf(struct trie
*t
)
1510 struct tnode
*n
= rcu_dereference_rtnl(t
->trie
);
1515 if (IS_LEAF(n
)) /* trie is just a leaf */
1518 return leaf_walk_rcu(n
, NULL
);
1521 static struct tnode
*trie_nextleaf(struct tnode
*l
)
1523 struct tnode
*p
= node_parent_rcu(l
);
1526 return NULL
; /* trie with just one leaf */
1528 return leaf_walk_rcu(p
, l
);
1531 static struct tnode
*trie_leafindex(struct trie
*t
, int index
)
1533 struct tnode
*l
= trie_firstleaf(t
);
1535 while (l
&& index
-- > 0)
1536 l
= trie_nextleaf(l
);
1543 * Caller must hold RTNL.
1545 int fib_table_flush(struct fib_table
*tb
)
1547 struct trie
*t
= (struct trie
*) tb
->tb_data
;
1548 struct tnode
*l
, *ll
= NULL
;
1551 for (l
= trie_firstleaf(t
); l
; l
= trie_nextleaf(l
)) {
1552 found
+= trie_flush_leaf(l
);
1554 if (ll
&& hlist_empty(&ll
->list
))
1555 trie_leaf_remove(t
, ll
);
1559 if (ll
&& hlist_empty(&ll
->list
))
1560 trie_leaf_remove(t
, ll
);
1562 pr_debug("trie_flush found=%d\n", found
);
1566 void fib_free_table(struct fib_table
*tb
)
1568 #ifdef CONFIG_IP_FIB_TRIE_STATS
1569 struct trie
*t
= (struct trie
*)tb
->tb_data
;
1571 free_percpu(t
->stats
);
1572 #endif /* CONFIG_IP_FIB_TRIE_STATS */
1576 static int fn_trie_dump_fa(t_key key
, int plen
, struct list_head
*fah
,
1577 struct fib_table
*tb
,
1578 struct sk_buff
*skb
, struct netlink_callback
*cb
)
1581 struct fib_alias
*fa
;
1582 __be32 xkey
= htonl(key
);
1587 /* rcu_read_lock is hold by caller */
1589 list_for_each_entry_rcu(fa
, fah
, fa_list
) {
1595 if (fib_dump_info(skb
, NETLINK_CB(cb
->skb
).portid
,
1603 fa
->fa_info
, NLM_F_MULTI
) < 0) {
1613 static int fn_trie_dump_leaf(struct tnode
*l
, struct fib_table
*tb
,
1614 struct sk_buff
*skb
, struct netlink_callback
*cb
)
1616 struct leaf_info
*li
;
1622 /* rcu_read_lock is hold by caller */
1623 hlist_for_each_entry_rcu(li
, &l
->list
, hlist
) {
1632 if (list_empty(&li
->falh
))
1635 if (fn_trie_dump_fa(l
->key
, li
->plen
, &li
->falh
, tb
, skb
, cb
) < 0) {
1646 int fib_table_dump(struct fib_table
*tb
, struct sk_buff
*skb
,
1647 struct netlink_callback
*cb
)
1650 struct trie
*t
= (struct trie
*) tb
->tb_data
;
1651 t_key key
= cb
->args
[2];
1652 int count
= cb
->args
[3];
1655 /* Dump starting at last key.
1656 * Note: 0.0.0.0/0 (ie default) is first key.
1659 l
= trie_firstleaf(t
);
1661 /* Normally, continue from last key, but if that is missing
1662 * fallback to using slow rescan
1664 l
= fib_find_node(t
, key
);
1666 l
= trie_leafindex(t
, count
);
1670 cb
->args
[2] = l
->key
;
1671 if (fn_trie_dump_leaf(l
, tb
, skb
, cb
) < 0) {
1672 cb
->args
[3] = count
;
1678 l
= trie_nextleaf(l
);
1679 memset(&cb
->args
[4], 0,
1680 sizeof(cb
->args
) - 4*sizeof(cb
->args
[0]));
1682 cb
->args
[3] = count
;
1688 void __init
fib_trie_init(void)
1690 fn_alias_kmem
= kmem_cache_create("ip_fib_alias",
1691 sizeof(struct fib_alias
),
1692 0, SLAB_PANIC
, NULL
);
1694 trie_leaf_kmem
= kmem_cache_create("ip_fib_trie",
1695 max(sizeof(struct tnode
),
1696 sizeof(struct leaf_info
)),
1697 0, SLAB_PANIC
, NULL
);
1701 struct fib_table
*fib_trie_table(u32 id
)
1703 struct fib_table
*tb
;
1706 tb
= kmalloc(sizeof(struct fib_table
) + sizeof(struct trie
),
1712 tb
->tb_default
= -1;
1713 tb
->tb_num_default
= 0;
1715 t
= (struct trie
*) tb
->tb_data
;
1716 RCU_INIT_POINTER(t
->trie
, NULL
);
1717 #ifdef CONFIG_IP_FIB_TRIE_STATS
1718 t
->stats
= alloc_percpu(struct trie_use_stats
);
1728 #ifdef CONFIG_PROC_FS
1729 /* Depth first Trie walk iterator */
1730 struct fib_trie_iter
{
1731 struct seq_net_private p
;
1732 struct fib_table
*tb
;
1733 struct tnode
*tnode
;
1738 static struct tnode
*fib_trie_get_next(struct fib_trie_iter
*iter
)
1740 unsigned long cindex
= iter
->index
;
1741 struct tnode
*tn
= iter
->tnode
;
1744 /* A single entry routing table */
1748 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
1749 iter
->tnode
, iter
->index
, iter
->depth
);
1751 while (cindex
< tnode_child_length(tn
)) {
1752 struct tnode
*n
= tnode_get_child_rcu(tn
, cindex
);
1757 iter
->index
= cindex
+ 1;
1759 /* push down one level */
1770 /* Current node exhausted, pop back up */
1771 p
= node_parent_rcu(tn
);
1773 cindex
= get_index(tn
->key
, p
) + 1;
1783 static struct tnode
*fib_trie_get_first(struct fib_trie_iter
*iter
,
1791 n
= rcu_dereference(t
->trie
);
1808 static void trie_collect_stats(struct trie
*t
, struct trie_stat
*s
)
1811 struct fib_trie_iter iter
;
1813 memset(s
, 0, sizeof(*s
));
1816 for (n
= fib_trie_get_first(&iter
, t
); n
; n
= fib_trie_get_next(&iter
)) {
1818 struct leaf_info
*li
;
1821 s
->totdepth
+= iter
.depth
;
1822 if (iter
.depth
> s
->maxdepth
)
1823 s
->maxdepth
= iter
.depth
;
1825 hlist_for_each_entry_rcu(li
, &n
->list
, hlist
)
1831 if (n
->bits
< MAX_STAT_DEPTH
)
1832 s
->nodesizes
[n
->bits
]++;
1834 for (i
= tnode_child_length(n
); i
--;) {
1835 if (!rcu_access_pointer(n
->child
[i
]))
1844 * This outputs /proc/net/fib_triestats
1846 static void trie_show_stats(struct seq_file
*seq
, struct trie_stat
*stat
)
1848 unsigned int i
, max
, pointers
, bytes
, avdepth
;
1851 avdepth
= stat
->totdepth
*100 / stat
->leaves
;
1855 seq_printf(seq
, "\tAver depth: %u.%02d\n",
1856 avdepth
/ 100, avdepth
% 100);
1857 seq_printf(seq
, "\tMax depth: %u\n", stat
->maxdepth
);
1859 seq_printf(seq
, "\tLeaves: %u\n", stat
->leaves
);
1860 bytes
= sizeof(struct tnode
) * stat
->leaves
;
1862 seq_printf(seq
, "\tPrefixes: %u\n", stat
->prefixes
);
1863 bytes
+= sizeof(struct leaf_info
) * stat
->prefixes
;
1865 seq_printf(seq
, "\tInternal nodes: %u\n\t", stat
->tnodes
);
1866 bytes
+= sizeof(struct tnode
) * stat
->tnodes
;
1868 max
= MAX_STAT_DEPTH
;
1869 while (max
> 0 && stat
->nodesizes
[max
-1] == 0)
1873 for (i
= 1; i
< max
; i
++)
1874 if (stat
->nodesizes
[i
] != 0) {
1875 seq_printf(seq
, " %u: %u", i
, stat
->nodesizes
[i
]);
1876 pointers
+= (1<<i
) * stat
->nodesizes
[i
];
1878 seq_putc(seq
, '\n');
1879 seq_printf(seq
, "\tPointers: %u\n", pointers
);
1881 bytes
+= sizeof(struct tnode
*) * pointers
;
1882 seq_printf(seq
, "Null ptrs: %u\n", stat
->nullpointers
);
1883 seq_printf(seq
, "Total size: %u kB\n", (bytes
+ 1023) / 1024);
1886 #ifdef CONFIG_IP_FIB_TRIE_STATS
1887 static void trie_show_usage(struct seq_file
*seq
,
1888 const struct trie_use_stats __percpu
*stats
)
1890 struct trie_use_stats s
= { 0 };
1893 /* loop through all of the CPUs and gather up the stats */
1894 for_each_possible_cpu(cpu
) {
1895 const struct trie_use_stats
*pcpu
= per_cpu_ptr(stats
, cpu
);
1897 s
.gets
+= pcpu
->gets
;
1898 s
.backtrack
+= pcpu
->backtrack
;
1899 s
.semantic_match_passed
+= pcpu
->semantic_match_passed
;
1900 s
.semantic_match_miss
+= pcpu
->semantic_match_miss
;
1901 s
.null_node_hit
+= pcpu
->null_node_hit
;
1902 s
.resize_node_skipped
+= pcpu
->resize_node_skipped
;
1905 seq_printf(seq
, "\nCounters:\n---------\n");
1906 seq_printf(seq
, "gets = %u\n", s
.gets
);
1907 seq_printf(seq
, "backtracks = %u\n", s
.backtrack
);
1908 seq_printf(seq
, "semantic match passed = %u\n",
1909 s
.semantic_match_passed
);
1910 seq_printf(seq
, "semantic match miss = %u\n", s
.semantic_match_miss
);
1911 seq_printf(seq
, "null node hit= %u\n", s
.null_node_hit
);
1912 seq_printf(seq
, "skipped node resize = %u\n\n", s
.resize_node_skipped
);
1914 #endif /* CONFIG_IP_FIB_TRIE_STATS */
1916 static void fib_table_print(struct seq_file
*seq
, struct fib_table
*tb
)
1918 if (tb
->tb_id
== RT_TABLE_LOCAL
)
1919 seq_puts(seq
, "Local:\n");
1920 else if (tb
->tb_id
== RT_TABLE_MAIN
)
1921 seq_puts(seq
, "Main:\n");
1923 seq_printf(seq
, "Id %d:\n", tb
->tb_id
);
1927 static int fib_triestat_seq_show(struct seq_file
*seq
, void *v
)
1929 struct net
*net
= (struct net
*)seq
->private;
1933 "Basic info: size of leaf:"
1934 " %Zd bytes, size of tnode: %Zd bytes.\n",
1935 sizeof(struct tnode
), sizeof(struct tnode
));
1937 for (h
= 0; h
< FIB_TABLE_HASHSZ
; h
++) {
1938 struct hlist_head
*head
= &net
->ipv4
.fib_table_hash
[h
];
1939 struct fib_table
*tb
;
1941 hlist_for_each_entry_rcu(tb
, head
, tb_hlist
) {
1942 struct trie
*t
= (struct trie
*) tb
->tb_data
;
1943 struct trie_stat stat
;
1948 fib_table_print(seq
, tb
);
1950 trie_collect_stats(t
, &stat
);
1951 trie_show_stats(seq
, &stat
);
1952 #ifdef CONFIG_IP_FIB_TRIE_STATS
1953 trie_show_usage(seq
, t
->stats
);
1961 static int fib_triestat_seq_open(struct inode
*inode
, struct file
*file
)
1963 return single_open_net(inode
, file
, fib_triestat_seq_show
);
1966 static const struct file_operations fib_triestat_fops
= {
1967 .owner
= THIS_MODULE
,
1968 .open
= fib_triestat_seq_open
,
1970 .llseek
= seq_lseek
,
1971 .release
= single_release_net
,
1974 static struct tnode
*fib_trie_get_idx(struct seq_file
*seq
, loff_t pos
)
1976 struct fib_trie_iter
*iter
= seq
->private;
1977 struct net
*net
= seq_file_net(seq
);
1981 for (h
= 0; h
< FIB_TABLE_HASHSZ
; h
++) {
1982 struct hlist_head
*head
= &net
->ipv4
.fib_table_hash
[h
];
1983 struct fib_table
*tb
;
1985 hlist_for_each_entry_rcu(tb
, head
, tb_hlist
) {
1988 for (n
= fib_trie_get_first(iter
,
1989 (struct trie
*) tb
->tb_data
);
1990 n
; n
= fib_trie_get_next(iter
))
2001 static void *fib_trie_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2005 return fib_trie_get_idx(seq
, *pos
);
2008 static void *fib_trie_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2010 struct fib_trie_iter
*iter
= seq
->private;
2011 struct net
*net
= seq_file_net(seq
);
2012 struct fib_table
*tb
= iter
->tb
;
2013 struct hlist_node
*tb_node
;
2018 /* next node in same table */
2019 n
= fib_trie_get_next(iter
);
2023 /* walk rest of this hash chain */
2024 h
= tb
->tb_id
& (FIB_TABLE_HASHSZ
- 1);
2025 while ((tb_node
= rcu_dereference(hlist_next_rcu(&tb
->tb_hlist
)))) {
2026 tb
= hlist_entry(tb_node
, struct fib_table
, tb_hlist
);
2027 n
= fib_trie_get_first(iter
, (struct trie
*) tb
->tb_data
);
2032 /* new hash chain */
2033 while (++h
< FIB_TABLE_HASHSZ
) {
2034 struct hlist_head
*head
= &net
->ipv4
.fib_table_hash
[h
];
2035 hlist_for_each_entry_rcu(tb
, head
, tb_hlist
) {
2036 n
= fib_trie_get_first(iter
, (struct trie
*) tb
->tb_data
);
2048 static void fib_trie_seq_stop(struct seq_file
*seq
, void *v
)
2054 static void seq_indent(struct seq_file
*seq
, int n
)
2060 static inline const char *rtn_scope(char *buf
, size_t len
, enum rt_scope_t s
)
2063 case RT_SCOPE_UNIVERSE
: return "universe";
2064 case RT_SCOPE_SITE
: return "site";
2065 case RT_SCOPE_LINK
: return "link";
2066 case RT_SCOPE_HOST
: return "host";
2067 case RT_SCOPE_NOWHERE
: return "nowhere";
2069 snprintf(buf
, len
, "scope=%d", s
);
2074 static const char *const rtn_type_names
[__RTN_MAX
] = {
2075 [RTN_UNSPEC
] = "UNSPEC",
2076 [RTN_UNICAST
] = "UNICAST",
2077 [RTN_LOCAL
] = "LOCAL",
2078 [RTN_BROADCAST
] = "BROADCAST",
2079 [RTN_ANYCAST
] = "ANYCAST",
2080 [RTN_MULTICAST
] = "MULTICAST",
2081 [RTN_BLACKHOLE
] = "BLACKHOLE",
2082 [RTN_UNREACHABLE
] = "UNREACHABLE",
2083 [RTN_PROHIBIT
] = "PROHIBIT",
2084 [RTN_THROW
] = "THROW",
2086 [RTN_XRESOLVE
] = "XRESOLVE",
2089 static inline const char *rtn_type(char *buf
, size_t len
, unsigned int t
)
2091 if (t
< __RTN_MAX
&& rtn_type_names
[t
])
2092 return rtn_type_names
[t
];
2093 snprintf(buf
, len
, "type %u", t
);
2097 /* Pretty print the trie */
2098 static int fib_trie_seq_show(struct seq_file
*seq
, void *v
)
2100 const struct fib_trie_iter
*iter
= seq
->private;
2101 struct tnode
*n
= v
;
2103 if (!node_parent_rcu(n
))
2104 fib_table_print(seq
, iter
->tb
);
2107 __be32 prf
= htonl(n
->key
);
2109 seq_indent(seq
, iter
->depth
-1);
2110 seq_printf(seq
, " +-- %pI4/%zu %u %u %u\n",
2111 &prf
, KEYLENGTH
- n
->pos
- n
->bits
, n
->bits
,
2112 n
->full_children
, n
->empty_children
);
2114 struct leaf_info
*li
;
2115 __be32 val
= htonl(n
->key
);
2117 seq_indent(seq
, iter
->depth
);
2118 seq_printf(seq
, " |-- %pI4\n", &val
);
2120 hlist_for_each_entry_rcu(li
, &n
->list
, hlist
) {
2121 struct fib_alias
*fa
;
2123 list_for_each_entry_rcu(fa
, &li
->falh
, fa_list
) {
2124 char buf1
[32], buf2
[32];
2126 seq_indent(seq
, iter
->depth
+1);
2127 seq_printf(seq
, " /%d %s %s", li
->plen
,
2128 rtn_scope(buf1
, sizeof(buf1
),
2129 fa
->fa_info
->fib_scope
),
2130 rtn_type(buf2
, sizeof(buf2
),
2133 seq_printf(seq
, " tos=%d", fa
->fa_tos
);
2134 seq_putc(seq
, '\n');
2142 static const struct seq_operations fib_trie_seq_ops
= {
2143 .start
= fib_trie_seq_start
,
2144 .next
= fib_trie_seq_next
,
2145 .stop
= fib_trie_seq_stop
,
2146 .show
= fib_trie_seq_show
,
2149 static int fib_trie_seq_open(struct inode
*inode
, struct file
*file
)
2151 return seq_open_net(inode
, file
, &fib_trie_seq_ops
,
2152 sizeof(struct fib_trie_iter
));
2155 static const struct file_operations fib_trie_fops
= {
2156 .owner
= THIS_MODULE
,
2157 .open
= fib_trie_seq_open
,
2159 .llseek
= seq_lseek
,
2160 .release
= seq_release_net
,
2163 struct fib_route_iter
{
2164 struct seq_net_private p
;
2165 struct trie
*main_trie
;
2170 static struct tnode
*fib_route_get_idx(struct fib_route_iter
*iter
, loff_t pos
)
2172 struct tnode
*l
= NULL
;
2173 struct trie
*t
= iter
->main_trie
;
2175 /* use cache location of last found key */
2176 if (iter
->pos
> 0 && pos
>= iter
->pos
&& (l
= fib_find_node(t
, iter
->key
)))
2180 l
= trie_firstleaf(t
);
2183 while (l
&& pos
-- > 0) {
2185 l
= trie_nextleaf(l
);
2189 iter
->key
= pos
; /* remember it */
2191 iter
->pos
= 0; /* forget it */
2196 static void *fib_route_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2199 struct fib_route_iter
*iter
= seq
->private;
2200 struct fib_table
*tb
;
2203 tb
= fib_get_table(seq_file_net(seq
), RT_TABLE_MAIN
);
2207 iter
->main_trie
= (struct trie
*) tb
->tb_data
;
2209 return SEQ_START_TOKEN
;
2211 return fib_route_get_idx(iter
, *pos
- 1);
2214 static void *fib_route_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2216 struct fib_route_iter
*iter
= seq
->private;
2217 struct tnode
*l
= v
;
2220 if (v
== SEQ_START_TOKEN
) {
2222 l
= trie_firstleaf(iter
->main_trie
);
2225 l
= trie_nextleaf(l
);
2235 static void fib_route_seq_stop(struct seq_file
*seq
, void *v
)
2241 static unsigned int fib_flag_trans(int type
, __be32 mask
, const struct fib_info
*fi
)
2243 unsigned int flags
= 0;
2245 if (type
== RTN_UNREACHABLE
|| type
== RTN_PROHIBIT
)
2247 if (fi
&& fi
->fib_nh
->nh_gw
)
2248 flags
|= RTF_GATEWAY
;
2249 if (mask
== htonl(0xFFFFFFFF))
2256 * This outputs /proc/net/route.
2257 * The format of the file is not supposed to be changed
2258 * and needs to be same as fib_hash output to avoid breaking
2261 static int fib_route_seq_show(struct seq_file
*seq
, void *v
)
2263 struct tnode
*l
= v
;
2264 struct leaf_info
*li
;
2266 if (v
== SEQ_START_TOKEN
) {
2267 seq_printf(seq
, "%-127s\n", "Iface\tDestination\tGateway "
2268 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2273 hlist_for_each_entry_rcu(li
, &l
->list
, hlist
) {
2274 struct fib_alias
*fa
;
2275 __be32 mask
, prefix
;
2277 mask
= inet_make_mask(li
->plen
);
2278 prefix
= htonl(l
->key
);
2280 list_for_each_entry_rcu(fa
, &li
->falh
, fa_list
) {
2281 const struct fib_info
*fi
= fa
->fa_info
;
2282 unsigned int flags
= fib_flag_trans(fa
->fa_type
, mask
, fi
);
2284 if (fa
->fa_type
== RTN_BROADCAST
2285 || fa
->fa_type
== RTN_MULTICAST
)
2288 seq_setwidth(seq
, 127);
2292 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2293 "%d\t%08X\t%d\t%u\t%u",
2294 fi
->fib_dev
? fi
->fib_dev
->name
: "*",
2296 fi
->fib_nh
->nh_gw
, flags
, 0, 0,
2300 fi
->fib_advmss
+ 40 : 0),
2305 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2306 "%d\t%08X\t%d\t%u\t%u",
2307 prefix
, 0, flags
, 0, 0, 0,
2317 static const struct seq_operations fib_route_seq_ops
= {
2318 .start
= fib_route_seq_start
,
2319 .next
= fib_route_seq_next
,
2320 .stop
= fib_route_seq_stop
,
2321 .show
= fib_route_seq_show
,
2324 static int fib_route_seq_open(struct inode
*inode
, struct file
*file
)
2326 return seq_open_net(inode
, file
, &fib_route_seq_ops
,
2327 sizeof(struct fib_route_iter
));
2330 static const struct file_operations fib_route_fops
= {
2331 .owner
= THIS_MODULE
,
2332 .open
= fib_route_seq_open
,
2334 .llseek
= seq_lseek
,
2335 .release
= seq_release_net
,
2338 int __net_init
fib_proc_init(struct net
*net
)
2340 if (!proc_create("fib_trie", S_IRUGO
, net
->proc_net
, &fib_trie_fops
))
2343 if (!proc_create("fib_triestat", S_IRUGO
, net
->proc_net
,
2344 &fib_triestat_fops
))
2347 if (!proc_create("route", S_IRUGO
, net
->proc_net
, &fib_route_fops
))
2353 remove_proc_entry("fib_triestat", net
->proc_net
);
2355 remove_proc_entry("fib_trie", net
->proc_net
);
2360 void __net_exit
fib_proc_exit(struct net
*net
)
2362 remove_proc_entry("fib_trie", net
->proc_net
);
2363 remove_proc_entry("fib_triestat", net
->proc_net
);
2364 remove_proc_entry("route", net
->proc_net
);
2367 #endif /* CONFIG_PROC_FS */