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 <net/switchdev.h>
83 #include "fib_lookup.h"
85 #define MAX_STAT_DEPTH 32
87 #define KEYLENGTH (8*sizeof(t_key))
88 #define KEY_MAX ((t_key)~0)
90 typedef unsigned int t_key
;
92 #define IS_TNODE(n) ((n)->bits)
93 #define IS_LEAF(n) (!(n)->bits)
96 t_key empty_children
; /* KEYLENGTH bits needed */
97 t_key full_children
; /* KEYLENGTH bits needed */
98 struct key_vector __rcu
*parent
;
101 unsigned char pos
; /* 2log(KEYLENGTH) bits needed */
102 unsigned char bits
; /* 2log(KEYLENGTH) bits needed */
105 /* This list pointer if valid if (pos | bits) == 0 (LEAF) */
106 struct hlist_head leaf
;
107 /* This array is valid if (pos | bits) > 0 (TNODE) */
108 struct key_vector __rcu
*tnode
[0];
114 struct key_vector kv
[1];
115 #define tn_bits kv[0].bits
118 #define TNODE_SIZE(n) offsetof(struct tnode, kv[0].tnode[n])
119 #define LEAF_SIZE TNODE_SIZE(1)
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 key_vector __rcu
*tnode
[1];
144 #ifdef CONFIG_IP_FIB_TRIE_STATS
145 struct trie_use_stats __percpu
*stats
;
149 static struct key_vector
**resize(struct trie
*t
, struct key_vector
*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 static inline struct tnode
*tn_info(struct key_vector
*kv
)
164 return container_of(kv
, struct tnode
, kv
[0]);
167 /* caller must hold RTNL */
168 #define node_parent(n) rtnl_dereference((n)->parent)
169 #define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
171 /* caller must hold RCU read lock or RTNL */
172 #define node_parent_rcu(n) rcu_dereference_rtnl((n)->parent)
173 #define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
175 /* wrapper for rcu_assign_pointer */
176 static inline void node_set_parent(struct key_vector
*n
, struct key_vector
*tp
)
179 rcu_assign_pointer(n
->parent
, tp
);
182 #define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER((n)->parent, p)
184 /* This provides us with the number of children in this node, in the case of a
185 * leaf this will return 0 meaning none of the children are accessible.
187 static inline unsigned long child_length(const struct key_vector
*tn
)
189 return (1ul << tn
->bits
) & ~(1ul);
192 static inline unsigned long get_index(t_key key
, struct key_vector
*kv
)
194 unsigned long index
= key
^ kv
->key
;
196 return index
>> kv
->pos
;
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 - TNODE_SIZE(0)) / sizeof(struct key_vector *))
276 #define TNODE_VMALLOC_MAX \
277 ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
279 static void __node_free_rcu(struct rcu_head
*head
)
281 struct tnode
*n
= container_of(head
, struct tnode
, rcu
);
284 kmem_cache_free(trie_leaf_kmem
, n
);
285 else if (n
->tn_bits
<= TNODE_KMALLOC_MAX
)
291 #define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
293 static struct tnode
*tnode_alloc(int bits
)
297 /* verify bits is within bounds */
298 if (bits
> TNODE_VMALLOC_MAX
)
301 /* determine size and verify it is non-zero and didn't overflow */
302 size
= TNODE_SIZE(1ul << bits
);
304 if (size
<= PAGE_SIZE
)
305 return kzalloc(size
, GFP_KERNEL
);
307 return vzalloc(size
);
310 static inline void empty_child_inc(struct key_vector
*n
)
312 ++n
->empty_children
? : ++n
->full_children
;
315 static inline void empty_child_dec(struct key_vector
*n
)
317 n
->empty_children
-- ? : n
->full_children
--;
320 static struct key_vector
*leaf_new(t_key key
, struct fib_alias
*fa
)
322 struct tnode
*kv
= kmem_cache_alloc(trie_leaf_kmem
, GFP_KERNEL
);
323 struct key_vector
*l
= kv
->kv
;
328 /* initialize key vector */
332 l
->slen
= fa
->fa_slen
;
334 /* link leaf to fib alias */
335 INIT_HLIST_HEAD(&l
->leaf
);
336 hlist_add_head(&fa
->fa_list
, &l
->leaf
);
341 static struct key_vector
*tnode_new(t_key key
, int pos
, int bits
)
343 struct tnode
*tnode
= tnode_alloc(bits
);
344 unsigned int shift
= pos
+ bits
;
345 struct key_vector
*tn
= tnode
->kv
;
347 /* verify bits and pos their msb bits clear and values are valid */
348 BUG_ON(!bits
|| (shift
> KEYLENGTH
));
350 pr_debug("AT %p s=%zu %zu\n", tnode
, TNODE_SIZE(0),
351 sizeof(struct key_vector
*) << bits
);
356 if (bits
== KEYLENGTH
)
357 tn
->full_children
= 1;
359 tn
->empty_children
= 1ul << bits
;
361 tn
->key
= (shift
< KEYLENGTH
) ? (key
>> shift
) << shift
: 0;
369 /* Check whether a tnode 'n' is "full", i.e. it is an internal node
370 * and no bits are skipped. See discussion in dyntree paper p. 6
372 static inline int tnode_full(struct key_vector
*tn
, struct key_vector
*n
)
374 return n
&& ((n
->pos
+ n
->bits
) == tn
->pos
) && IS_TNODE(n
);
377 /* Add a child at position i overwriting the old value.
378 * Update the value of full_children and empty_children.
380 static void put_child(struct key_vector
*tn
, unsigned long i
,
381 struct key_vector
*n
)
383 struct key_vector
*chi
= get_child(tn
, i
);
386 BUG_ON(i
>= child_length(tn
));
388 /* update emptyChildren, overflow into fullChildren */
389 if (n
== NULL
&& chi
!= NULL
)
391 if (n
!= NULL
&& chi
== NULL
)
394 /* update fullChildren */
395 wasfull
= tnode_full(tn
, chi
);
396 isfull
= tnode_full(tn
, n
);
398 if (wasfull
&& !isfull
)
400 else if (!wasfull
&& isfull
)
403 if (n
&& (tn
->slen
< n
->slen
))
406 rcu_assign_pointer(tn
->tnode
[i
], n
);
409 static void update_children(struct key_vector
*tn
)
413 /* update all of the child parent pointers */
414 for (i
= child_length(tn
); i
;) {
415 struct key_vector
*inode
= get_child(tn
, --i
);
420 /* Either update the children of a tnode that
421 * already belongs to us or update the child
422 * to point to ourselves.
424 if (node_parent(inode
) == tn
)
425 update_children(inode
);
427 node_set_parent(inode
, tn
);
431 static inline void put_child_root(struct key_vector
*tp
, struct trie
*t
,
432 t_key key
, struct key_vector
*n
)
435 put_child(tp
, get_index(key
, tp
), n
);
437 rcu_assign_pointer(t
->tnode
[0], n
);
440 static inline void tnode_free_init(struct key_vector
*tn
)
442 tn_info(tn
)->rcu
.next
= NULL
;
445 static inline void tnode_free_append(struct key_vector
*tn
,
446 struct key_vector
*n
)
448 tn_info(n
)->rcu
.next
= tn_info(tn
)->rcu
.next
;
449 tn_info(tn
)->rcu
.next
= &tn_info(n
)->rcu
;
452 static void tnode_free(struct key_vector
*tn
)
454 struct callback_head
*head
= &tn_info(tn
)->rcu
;
458 tnode_free_size
+= TNODE_SIZE(1ul << tn
->bits
);
461 tn
= container_of(head
, struct tnode
, rcu
)->kv
;
464 if (tnode_free_size
>= PAGE_SIZE
* sync_pages
) {
470 static struct key_vector __rcu
**replace(struct trie
*t
,
471 struct key_vector
*oldtnode
,
472 struct key_vector
*tn
)
474 struct key_vector
*tp
= node_parent(oldtnode
);
475 struct key_vector
**cptr
;
478 /* setup the parent pointer out of and back into this node */
479 NODE_INIT_PARENT(tn
, tp
);
480 put_child_root(tp
, t
, tn
->key
, tn
);
482 /* update all of the child parent pointers */
485 /* all pointers should be clean so we are done */
486 tnode_free(oldtnode
);
488 /* record the pointer that is pointing to this node */
489 cptr
= tp
? tp
->tnode
: t
->tnode
;
491 /* resize children now that oldtnode is freed */
492 for (i
= child_length(tn
); i
;) {
493 struct key_vector
*inode
= get_child(tn
, --i
);
495 /* resize child node */
496 if (tnode_full(tn
, inode
))
503 static struct key_vector __rcu
**inflate(struct trie
*t
,
504 struct key_vector
*oldtnode
)
506 struct key_vector
*tn
;
510 pr_debug("In inflate\n");
512 tn
= tnode_new(oldtnode
->key
, oldtnode
->pos
- 1, oldtnode
->bits
+ 1);
516 /* prepare oldtnode to be freed */
517 tnode_free_init(oldtnode
);
519 /* Assemble all of the pointers in our cluster, in this case that
520 * represents all of the pointers out of our allocated nodes that
521 * point to existing tnodes and the links between our allocated
524 for (i
= child_length(oldtnode
), m
= 1u << tn
->pos
; i
;) {
525 struct key_vector
*inode
= get_child(oldtnode
, --i
);
526 struct key_vector
*node0
, *node1
;
533 /* A leaf or an internal node with skipped bits */
534 if (!tnode_full(oldtnode
, inode
)) {
535 put_child(tn
, get_index(inode
->key
, tn
), inode
);
539 /* drop the node in the old tnode free list */
540 tnode_free_append(oldtnode
, inode
);
542 /* An internal node with two children */
543 if (inode
->bits
== 1) {
544 put_child(tn
, 2 * i
+ 1, get_child(inode
, 1));
545 put_child(tn
, 2 * i
, get_child(inode
, 0));
549 /* We will replace this node 'inode' with two new
550 * ones, 'node0' and 'node1', each with half of the
551 * original children. The two new nodes will have
552 * a position one bit further down the key and this
553 * means that the "significant" part of their keys
554 * (see the discussion near the top of this file)
555 * will differ by one bit, which will be "0" in
556 * node0's key and "1" in node1's key. Since we are
557 * moving the key position by one step, the bit that
558 * we are moving away from - the bit at position
559 * (tn->pos) - is the one that will differ between
560 * node0 and node1. So... we synthesize that bit in the
563 node1
= tnode_new(inode
->key
| m
, inode
->pos
, inode
->bits
- 1);
566 node0
= tnode_new(inode
->key
, inode
->pos
, inode
->bits
- 1);
568 tnode_free_append(tn
, node1
);
571 tnode_free_append(tn
, node0
);
573 /* populate child pointers in new nodes */
574 for (k
= child_length(inode
), j
= k
/ 2; j
;) {
575 put_child(node1
, --j
, get_child(inode
, --k
));
576 put_child(node0
, j
, get_child(inode
, j
));
577 put_child(node1
, --j
, get_child(inode
, --k
));
578 put_child(node0
, j
, get_child(inode
, j
));
581 /* link new nodes to parent */
582 NODE_INIT_PARENT(node1
, tn
);
583 NODE_INIT_PARENT(node0
, tn
);
585 /* link parent to nodes */
586 put_child(tn
, 2 * i
+ 1, node1
);
587 put_child(tn
, 2 * i
, node0
);
590 /* setup the parent pointers into and out of this node */
591 return replace(t
, oldtnode
, tn
);
593 /* all pointers should be clean so we are done */
599 static struct key_vector __rcu
**halve(struct trie
*t
,
600 struct key_vector
*oldtnode
)
602 struct key_vector
*tn
;
605 pr_debug("In halve\n");
607 tn
= tnode_new(oldtnode
->key
, oldtnode
->pos
+ 1, oldtnode
->bits
- 1);
611 /* prepare oldtnode to be freed */
612 tnode_free_init(oldtnode
);
614 /* Assemble all of the pointers in our cluster, in this case that
615 * represents all of the pointers out of our allocated nodes that
616 * point to existing tnodes and the links between our allocated
619 for (i
= child_length(oldtnode
); i
;) {
620 struct key_vector
*node1
= get_child(oldtnode
, --i
);
621 struct key_vector
*node0
= get_child(oldtnode
, --i
);
622 struct key_vector
*inode
;
624 /* At least one of the children is empty */
625 if (!node1
|| !node0
) {
626 put_child(tn
, i
/ 2, node1
? : node0
);
630 /* Two nonempty children */
631 inode
= tnode_new(node0
->key
, oldtnode
->pos
, 1);
634 tnode_free_append(tn
, inode
);
636 /* initialize pointers out of node */
637 put_child(inode
, 1, node1
);
638 put_child(inode
, 0, node0
);
639 NODE_INIT_PARENT(inode
, tn
);
641 /* link parent to node */
642 put_child(tn
, i
/ 2, inode
);
645 /* setup the parent pointers into and out of this node */
646 return replace(t
, oldtnode
, tn
);
648 /* all pointers should be clean so we are done */
654 static void collapse(struct trie
*t
, struct key_vector
*oldtnode
)
656 struct key_vector
*n
, *tp
;
659 /* scan the tnode looking for that one child that might still exist */
660 for (n
= NULL
, i
= child_length(oldtnode
); !n
&& i
;)
661 n
= get_child(oldtnode
, --i
);
663 /* compress one level */
664 tp
= node_parent(oldtnode
);
665 put_child_root(tp
, t
, oldtnode
->key
, n
);
666 node_set_parent(n
, tp
);
672 static unsigned char update_suffix(struct key_vector
*tn
)
674 unsigned char slen
= tn
->pos
;
675 unsigned long stride
, i
;
677 /* search though the list of children looking for nodes that might
678 * have a suffix greater than the one we currently have. This is
679 * why we start with a stride of 2 since a stride of 1 would
680 * represent the nodes with suffix length equal to tn->pos
682 for (i
= 0, stride
= 0x2ul
; i
< child_length(tn
); i
+= stride
) {
683 struct key_vector
*n
= get_child(tn
, i
);
685 if (!n
|| (n
->slen
<= slen
))
688 /* update stride and slen based on new value */
689 stride
<<= (n
->slen
- slen
);
693 /* if slen covers all but the last bit we can stop here
694 * there will be nothing longer than that since only node
695 * 0 and 1 << (bits - 1) could have that as their suffix
698 if ((slen
+ 1) >= (tn
->pos
+ tn
->bits
))
707 /* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
708 * the Helsinki University of Technology and Matti Tikkanen of Nokia
709 * Telecommunications, page 6:
710 * "A node is doubled if the ratio of non-empty children to all
711 * children in the *doubled* node is at least 'high'."
713 * 'high' in this instance is the variable 'inflate_threshold'. It
714 * is expressed as a percentage, so we multiply it with
715 * child_length() and instead of multiplying by 2 (since the
716 * child array will be doubled by inflate()) and multiplying
717 * the left-hand side by 100 (to handle the percentage thing) we
718 * multiply the left-hand side by 50.
720 * The left-hand side may look a bit weird: child_length(tn)
721 * - tn->empty_children is of course the number of non-null children
722 * in the current node. tn->full_children is the number of "full"
723 * children, that is non-null tnodes with a skip value of 0.
724 * All of those will be doubled in the resulting inflated tnode, so
725 * we just count them one extra time here.
727 * A clearer way to write this would be:
729 * to_be_doubled = tn->full_children;
730 * not_to_be_doubled = child_length(tn) - tn->empty_children -
733 * new_child_length = child_length(tn) * 2;
735 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
737 * if (new_fill_factor >= inflate_threshold)
739 * ...and so on, tho it would mess up the while () loop.
742 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
746 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
747 * inflate_threshold * new_child_length
749 * expand not_to_be_doubled and to_be_doubled, and shorten:
750 * 100 * (child_length(tn) - tn->empty_children +
751 * tn->full_children) >= inflate_threshold * new_child_length
753 * expand new_child_length:
754 * 100 * (child_length(tn) - tn->empty_children +
755 * tn->full_children) >=
756 * inflate_threshold * child_length(tn) * 2
759 * 50 * (tn->full_children + child_length(tn) -
760 * tn->empty_children) >= inflate_threshold *
764 static inline bool should_inflate(struct key_vector
*tp
, struct key_vector
*tn
)
766 unsigned long used
= child_length(tn
);
767 unsigned long threshold
= used
;
769 /* Keep root node larger */
770 threshold
*= tp
? inflate_threshold
: inflate_threshold_root
;
771 used
-= tn
->empty_children
;
772 used
+= tn
->full_children
;
774 /* if bits == KEYLENGTH then pos = 0, and will fail below */
776 return (used
> 1) && tn
->pos
&& ((50 * used
) >= threshold
);
779 static inline bool should_halve(struct key_vector
*tp
, struct key_vector
*tn
)
781 unsigned long used
= child_length(tn
);
782 unsigned long threshold
= used
;
784 /* Keep root node larger */
785 threshold
*= tp
? halve_threshold
: halve_threshold_root
;
786 used
-= tn
->empty_children
;
788 /* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
790 return (used
> 1) && (tn
->bits
> 1) && ((100 * used
) < threshold
);
793 static inline bool should_collapse(struct key_vector
*tn
)
795 unsigned long used
= child_length(tn
);
797 used
-= tn
->empty_children
;
799 /* account for bits == KEYLENGTH case */
800 if ((tn
->bits
== KEYLENGTH
) && tn
->full_children
)
803 /* One child or none, time to drop us from the trie */
808 static struct key_vector __rcu
**resize(struct trie
*t
,
809 struct key_vector
*tn
)
811 #ifdef CONFIG_IP_FIB_TRIE_STATS
812 struct trie_use_stats __percpu
*stats
= t
->stats
;
814 struct key_vector
*tp
= node_parent(tn
);
815 unsigned long cindex
= tp
? get_index(tn
->key
, tp
) : 0;
816 struct key_vector __rcu
**cptr
= tp
? tp
->tnode
: t
->tnode
;
817 int max_work
= MAX_WORK
;
819 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
820 tn
, inflate_threshold
, halve_threshold
);
822 /* track the tnode via the pointer from the parent instead of
823 * doing it ourselves. This way we can let RCU fully do its
824 * thing without us interfering
826 BUG_ON(tn
!= rtnl_dereference(cptr
[cindex
]));
828 /* Double as long as the resulting node has a number of
829 * nonempty nodes that are above the threshold.
831 while (should_inflate(tp
, tn
) && max_work
) {
832 struct key_vector __rcu
**tcptr
= inflate(t
, tn
);
835 #ifdef CONFIG_IP_FIB_TRIE_STATS
836 this_cpu_inc(stats
->resize_node_skipped
);
843 tn
= rtnl_dereference(cptr
[cindex
]);
846 /* Return if at least one inflate is run */
847 if (max_work
!= MAX_WORK
)
850 /* Halve as long as the number of empty children in this
851 * node is above threshold.
853 while (should_halve(tp
, tn
) && max_work
) {
854 struct key_vector __rcu
**tcptr
= halve(t
, tn
);
857 #ifdef CONFIG_IP_FIB_TRIE_STATS
858 this_cpu_inc(stats
->resize_node_skipped
);
865 tn
= rtnl_dereference(cptr
[cindex
]);
868 /* Only one child remains */
869 if (should_collapse(tn
)) {
874 /* Return if at least one deflate was run */
875 if (max_work
!= MAX_WORK
)
878 /* push the suffix length to the parent node */
879 if (tn
->slen
> tn
->pos
) {
880 unsigned char slen
= update_suffix(tn
);
882 if (tp
&& (slen
> tp
->slen
))
889 static void leaf_pull_suffix(struct key_vector
*tp
, struct key_vector
*l
)
891 while (tp
&& (tp
->slen
> tp
->pos
) && (tp
->slen
> l
->slen
)) {
892 if (update_suffix(tp
) > l
->slen
)
894 tp
= node_parent(tp
);
898 static void leaf_push_suffix(struct key_vector
*tn
, struct key_vector
*l
)
900 /* if this is a new leaf then tn will be NULL and we can sort
901 * out parent suffix lengths as a part of trie_rebalance
903 while (tn
&& (tn
->slen
< l
->slen
)) {
905 tn
= node_parent(tn
);
909 /* rcu_read_lock needs to be hold by caller from readside */
910 static struct key_vector
*fib_find_node(struct trie
*t
,
911 struct key_vector
**tp
, u32 key
)
913 struct key_vector
*pn
= NULL
, *n
= rcu_dereference_rtnl(t
->tnode
[0]);
916 unsigned long index
= get_index(key
, n
);
918 /* This bit of code is a bit tricky but it combines multiple
919 * checks into a single check. The prefix consists of the
920 * prefix plus zeros for the bits in the cindex. The index
921 * is the difference between the key and this value. From
922 * this we can actually derive several pieces of data.
923 * if (index >= (1ul << bits))
924 * we have a mismatch in skip bits and failed
926 * we know the value is cindex
928 * This check is safe even if bits == KEYLENGTH due to the
929 * fact that we can only allocate a node with 32 bits if a
930 * long is greater than 32 bits.
932 if (index
>= (1ul << n
->bits
)) {
937 /* we have found a leaf. Prefixes have already been compared */
942 n
= get_child_rcu(n
, index
);
950 /* Return the first fib alias matching TOS with
951 * priority less than or equal to PRIO.
953 static struct fib_alias
*fib_find_alias(struct hlist_head
*fah
, u8 slen
,
956 struct fib_alias
*fa
;
961 hlist_for_each_entry(fa
, fah
, fa_list
) {
962 if (fa
->fa_slen
< slen
)
964 if (fa
->fa_slen
!= slen
)
966 if (fa
->fa_tos
> tos
)
968 if (fa
->fa_info
->fib_priority
>= prio
|| fa
->fa_tos
< tos
)
975 static void trie_rebalance(struct trie
*t
, struct key_vector
*tn
)
977 struct key_vector __rcu
**cptr
= t
->tnode
;
980 struct key_vector
*tp
= node_parent(tn
);
982 cptr
= resize(t
, tn
);
985 tn
= container_of(cptr
, struct key_vector
, tnode
[0]);
989 static int fib_insert_node(struct trie
*t
, struct key_vector
*tp
,
990 struct fib_alias
*new, t_key key
)
992 struct key_vector
*n
, *l
;
994 l
= leaf_new(key
, new);
998 /* retrieve child from parent node */
1000 n
= get_child(tp
, get_index(key
, tp
));
1002 n
= rcu_dereference_rtnl(t
->tnode
[0]);
1004 /* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1006 * Add a new tnode here
1007 * first tnode need some special handling
1008 * leaves us in position for handling as case 3
1011 struct key_vector
*tn
;
1013 tn
= tnode_new(key
, __fls(key
^ n
->key
), 1);
1017 /* initialize routes out of node */
1018 NODE_INIT_PARENT(tn
, tp
);
1019 put_child(tn
, get_index(key
, tn
) ^ 1, n
);
1021 /* start adding routes into the node */
1022 put_child_root(tp
, t
, key
, tn
);
1023 node_set_parent(n
, tn
);
1025 /* parent now has a NULL spot where the leaf can go */
1029 /* Case 3: n is NULL, and will just insert a new leaf */
1030 NODE_INIT_PARENT(l
, tp
);
1031 put_child_root(tp
, t
, key
, l
);
1032 trie_rebalance(t
, tp
);
1041 static int fib_insert_alias(struct trie
*t
, struct key_vector
*tp
,
1042 struct key_vector
*l
, struct fib_alias
*new,
1043 struct fib_alias
*fa
, t_key key
)
1046 return fib_insert_node(t
, tp
, new, key
);
1049 hlist_add_before_rcu(&new->fa_list
, &fa
->fa_list
);
1051 struct fib_alias
*last
;
1053 hlist_for_each_entry(last
, &l
->leaf
, fa_list
) {
1054 if (new->fa_slen
< last
->fa_slen
)
1060 hlist_add_behind_rcu(&new->fa_list
, &fa
->fa_list
);
1062 hlist_add_head_rcu(&new->fa_list
, &l
->leaf
);
1065 /* if we added to the tail node then we need to update slen */
1066 if (l
->slen
< new->fa_slen
) {
1067 l
->slen
= new->fa_slen
;
1068 leaf_push_suffix(tp
, l
);
1074 /* Caller must hold RTNL. */
1075 int fib_table_insert(struct fib_table
*tb
, struct fib_config
*cfg
)
1077 struct trie
*t
= (struct trie
*)tb
->tb_data
;
1078 struct fib_alias
*fa
, *new_fa
;
1079 struct key_vector
*l
, *tp
;
1080 struct fib_info
*fi
;
1081 u8 plen
= cfg
->fc_dst_len
;
1082 u8 slen
= KEYLENGTH
- plen
;
1083 u8 tos
= cfg
->fc_tos
;
1087 if (plen
> KEYLENGTH
)
1090 key
= ntohl(cfg
->fc_dst
);
1092 pr_debug("Insert table=%u %08x/%d\n", tb
->tb_id
, key
, plen
);
1094 if ((plen
< KEYLENGTH
) && (key
<< plen
))
1097 fi
= fib_create_info(cfg
);
1103 l
= fib_find_node(t
, &tp
, key
);
1104 fa
= l
? fib_find_alias(&l
->leaf
, slen
, tos
, fi
->fib_priority
) : NULL
;
1106 /* Now fa, if non-NULL, points to the first fib alias
1107 * with the same keys [prefix,tos,priority], if such key already
1108 * exists or to the node before which we will insert new one.
1110 * If fa is NULL, we will need to allocate a new one and
1111 * insert to the tail of the section matching the suffix length
1115 if (fa
&& fa
->fa_tos
== tos
&&
1116 fa
->fa_info
->fib_priority
== fi
->fib_priority
) {
1117 struct fib_alias
*fa_first
, *fa_match
;
1120 if (cfg
->fc_nlflags
& NLM_F_EXCL
)
1124 * 1. Find exact match for type, scope, fib_info to avoid
1126 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1130 hlist_for_each_entry_from(fa
, fa_list
) {
1131 if ((fa
->fa_slen
!= slen
) || (fa
->fa_tos
!= tos
))
1133 if (fa
->fa_info
->fib_priority
!= fi
->fib_priority
)
1135 if (fa
->fa_type
== cfg
->fc_type
&&
1136 fa
->fa_info
== fi
) {
1142 if (cfg
->fc_nlflags
& NLM_F_REPLACE
) {
1143 struct fib_info
*fi_drop
;
1153 new_fa
= kmem_cache_alloc(fn_alias_kmem
, GFP_KERNEL
);
1157 fi_drop
= fa
->fa_info
;
1158 new_fa
->fa_tos
= fa
->fa_tos
;
1159 new_fa
->fa_info
= fi
;
1160 new_fa
->fa_type
= cfg
->fc_type
;
1161 state
= fa
->fa_state
;
1162 new_fa
->fa_state
= state
& ~FA_S_ACCESSED
;
1163 new_fa
->fa_slen
= fa
->fa_slen
;
1165 err
= netdev_switch_fib_ipv4_add(key
, plen
, fi
,
1170 netdev_switch_fib_ipv4_abort(fi
);
1171 kmem_cache_free(fn_alias_kmem
, new_fa
);
1175 hlist_replace_rcu(&fa
->fa_list
, &new_fa
->fa_list
);
1177 alias_free_mem_rcu(fa
);
1179 fib_release_info(fi_drop
);
1180 if (state
& FA_S_ACCESSED
)
1181 rt_cache_flush(cfg
->fc_nlinfo
.nl_net
);
1182 rtmsg_fib(RTM_NEWROUTE
, htonl(key
), new_fa
, plen
,
1183 tb
->tb_id
, &cfg
->fc_nlinfo
, NLM_F_REPLACE
);
1187 /* Error if we find a perfect match which
1188 * uses the same scope, type, and nexthop
1194 if (!(cfg
->fc_nlflags
& NLM_F_APPEND
))
1198 if (!(cfg
->fc_nlflags
& NLM_F_CREATE
))
1202 new_fa
= kmem_cache_alloc(fn_alias_kmem
, GFP_KERNEL
);
1206 new_fa
->fa_info
= fi
;
1207 new_fa
->fa_tos
= tos
;
1208 new_fa
->fa_type
= cfg
->fc_type
;
1209 new_fa
->fa_state
= 0;
1210 new_fa
->fa_slen
= slen
;
1212 /* (Optionally) offload fib entry to switch hardware. */
1213 err
= netdev_switch_fib_ipv4_add(key
, plen
, fi
, tos
,
1214 cfg
->fc_type
, tb
->tb_id
);
1216 netdev_switch_fib_ipv4_abort(fi
);
1217 goto out_free_new_fa
;
1220 /* Insert new entry to the list. */
1221 err
= fib_insert_alias(t
, tp
, l
, new_fa
, fa
, key
);
1223 goto out_sw_fib_del
;
1226 tb
->tb_num_default
++;
1228 rt_cache_flush(cfg
->fc_nlinfo
.nl_net
);
1229 rtmsg_fib(RTM_NEWROUTE
, htonl(key
), new_fa
, plen
, tb
->tb_id
,
1230 &cfg
->fc_nlinfo
, 0);
1235 netdev_switch_fib_ipv4_del(key
, plen
, fi
, tos
, cfg
->fc_type
, tb
->tb_id
);
1237 kmem_cache_free(fn_alias_kmem
, new_fa
);
1239 fib_release_info(fi
);
1244 static inline t_key
prefix_mismatch(t_key key
, struct key_vector
*n
)
1246 t_key prefix
= n
->key
;
1248 return (key
^ prefix
) & (prefix
| -prefix
);
1251 /* should be called with rcu_read_lock */
1252 int fib_table_lookup(struct fib_table
*tb
, const struct flowi4
*flp
,
1253 struct fib_result
*res
, int fib_flags
)
1255 struct trie
*t
= (struct trie
*)tb
->tb_data
;
1256 #ifdef CONFIG_IP_FIB_TRIE_STATS
1257 struct trie_use_stats __percpu
*stats
= t
->stats
;
1259 const t_key key
= ntohl(flp
->daddr
);
1260 struct key_vector
*n
, *pn
;
1261 struct fib_alias
*fa
;
1262 unsigned long index
;
1265 n
= rcu_dereference(t
->tnode
[0]);
1269 #ifdef CONFIG_IP_FIB_TRIE_STATS
1270 this_cpu_inc(stats
->gets
);
1276 /* Step 1: Travel to the longest prefix match in the trie */
1278 index
= get_index(key
, n
);
1280 /* This bit of code is a bit tricky but it combines multiple
1281 * checks into a single check. The prefix consists of the
1282 * prefix plus zeros for the "bits" in the prefix. The index
1283 * is the difference between the key and this value. From
1284 * this we can actually derive several pieces of data.
1285 * if (index >= (1ul << bits))
1286 * we have a mismatch in skip bits and failed
1288 * we know the value is cindex
1290 * This check is safe even if bits == KEYLENGTH due to the
1291 * fact that we can only allocate a node with 32 bits if a
1292 * long is greater than 32 bits.
1294 if (index
>= (1ul << n
->bits
))
1297 /* we have found a leaf. Prefixes have already been compared */
1301 /* only record pn and cindex if we are going to be chopping
1302 * bits later. Otherwise we are just wasting cycles.
1304 if (n
->slen
> n
->pos
) {
1309 n
= get_child_rcu(n
, index
);
1314 /* Step 2: Sort out leaves and begin backtracing for longest prefix */
1316 /* record the pointer where our next node pointer is stored */
1317 struct key_vector __rcu
**cptr
= n
->tnode
;
1319 /* This test verifies that none of the bits that differ
1320 * between the key and the prefix exist in the region of
1321 * the lsb and higher in the prefix.
1323 if (unlikely(prefix_mismatch(key
, n
)) || (n
->slen
== n
->pos
))
1326 /* exit out and process leaf */
1327 if (unlikely(IS_LEAF(n
)))
1330 /* Don't bother recording parent info. Since we are in
1331 * prefix match mode we will have to come back to wherever
1332 * we started this traversal anyway
1335 while ((n
= rcu_dereference(*cptr
)) == NULL
) {
1337 #ifdef CONFIG_IP_FIB_TRIE_STATS
1339 this_cpu_inc(stats
->null_node_hit
);
1341 /* If we are at cindex 0 there are no more bits for
1342 * us to strip at this level so we must ascend back
1343 * up one level to see if there are any more bits to
1344 * be stripped there.
1347 t_key pkey
= pn
->key
;
1349 pn
= node_parent_rcu(pn
);
1352 #ifdef CONFIG_IP_FIB_TRIE_STATS
1353 this_cpu_inc(stats
->backtrack
);
1355 /* Get Child's index */
1356 cindex
= get_index(pkey
, pn
);
1359 /* strip the least significant bit from the cindex */
1360 cindex
&= cindex
- 1;
1362 /* grab pointer for next child node */
1363 cptr
= &pn
->tnode
[cindex
];
1368 /* this line carries forward the xor from earlier in the function */
1369 index
= key
^ n
->key
;
1371 /* Step 3: Process the leaf, if that fails fall back to backtracing */
1372 hlist_for_each_entry_rcu(fa
, &n
->leaf
, fa_list
) {
1373 struct fib_info
*fi
= fa
->fa_info
;
1376 if ((index
>= (1ul << fa
->fa_slen
)) &&
1377 ((BITS_PER_LONG
> KEYLENGTH
) || (fa
->fa_slen
!= KEYLENGTH
)))
1379 if (fa
->fa_tos
&& fa
->fa_tos
!= flp
->flowi4_tos
)
1383 if (fa
->fa_info
->fib_scope
< flp
->flowi4_scope
)
1385 fib_alias_accessed(fa
);
1386 err
= fib_props
[fa
->fa_type
].error
;
1387 if (unlikely(err
< 0)) {
1388 #ifdef CONFIG_IP_FIB_TRIE_STATS
1389 this_cpu_inc(stats
->semantic_match_passed
);
1393 if (fi
->fib_flags
& RTNH_F_DEAD
)
1395 for (nhsel
= 0; nhsel
< fi
->fib_nhs
; nhsel
++) {
1396 const struct fib_nh
*nh
= &fi
->fib_nh
[nhsel
];
1398 if (nh
->nh_flags
& RTNH_F_DEAD
)
1400 if (flp
->flowi4_oif
&& flp
->flowi4_oif
!= nh
->nh_oif
)
1403 if (!(fib_flags
& FIB_LOOKUP_NOREF
))
1404 atomic_inc(&fi
->fib_clntref
);
1406 res
->prefixlen
= KEYLENGTH
- fa
->fa_slen
;
1407 res
->nh_sel
= nhsel
;
1408 res
->type
= fa
->fa_type
;
1409 res
->scope
= fi
->fib_scope
;
1412 res
->fa_head
= &n
->leaf
;
1413 #ifdef CONFIG_IP_FIB_TRIE_STATS
1414 this_cpu_inc(stats
->semantic_match_passed
);
1419 #ifdef CONFIG_IP_FIB_TRIE_STATS
1420 this_cpu_inc(stats
->semantic_match_miss
);
1424 EXPORT_SYMBOL_GPL(fib_table_lookup
);
1426 static void fib_remove_alias(struct trie
*t
, struct key_vector
*tp
,
1427 struct key_vector
*l
, struct fib_alias
*old
)
1429 /* record the location of the previous list_info entry */
1430 struct hlist_node
**pprev
= old
->fa_list
.pprev
;
1431 struct fib_alias
*fa
= hlist_entry(pprev
, typeof(*fa
), fa_list
.next
);
1433 /* remove the fib_alias from the list */
1434 hlist_del_rcu(&old
->fa_list
);
1436 /* if we emptied the list this leaf will be freed and we can sort
1437 * out parent suffix lengths as a part of trie_rebalance
1439 if (hlist_empty(&l
->leaf
)) {
1440 put_child_root(tp
, t
, l
->key
, NULL
);
1442 trie_rebalance(t
, tp
);
1446 /* only access fa if it is pointing at the last valid hlist_node */
1450 /* update the trie with the latest suffix length */
1451 l
->slen
= fa
->fa_slen
;
1452 leaf_pull_suffix(tp
, l
);
1455 /* Caller must hold RTNL. */
1456 int fib_table_delete(struct fib_table
*tb
, struct fib_config
*cfg
)
1458 struct trie
*t
= (struct trie
*) tb
->tb_data
;
1459 struct fib_alias
*fa
, *fa_to_delete
;
1460 struct key_vector
*l
, *tp
;
1461 u8 plen
= cfg
->fc_dst_len
;
1462 u8 slen
= KEYLENGTH
- plen
;
1463 u8 tos
= cfg
->fc_tos
;
1466 if (plen
> KEYLENGTH
)
1469 key
= ntohl(cfg
->fc_dst
);
1471 if ((plen
< KEYLENGTH
) && (key
<< plen
))
1474 l
= fib_find_node(t
, &tp
, key
);
1478 fa
= fib_find_alias(&l
->leaf
, slen
, tos
, 0);
1482 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key
, plen
, tos
, t
);
1484 fa_to_delete
= NULL
;
1485 hlist_for_each_entry_from(fa
, fa_list
) {
1486 struct fib_info
*fi
= fa
->fa_info
;
1488 if ((fa
->fa_slen
!= slen
) || (fa
->fa_tos
!= tos
))
1491 if ((!cfg
->fc_type
|| fa
->fa_type
== cfg
->fc_type
) &&
1492 (cfg
->fc_scope
== RT_SCOPE_NOWHERE
||
1493 fa
->fa_info
->fib_scope
== cfg
->fc_scope
) &&
1494 (!cfg
->fc_prefsrc
||
1495 fi
->fib_prefsrc
== cfg
->fc_prefsrc
) &&
1496 (!cfg
->fc_protocol
||
1497 fi
->fib_protocol
== cfg
->fc_protocol
) &&
1498 fib_nh_match(cfg
, fi
) == 0) {
1507 netdev_switch_fib_ipv4_del(key
, plen
, fa_to_delete
->fa_info
, tos
,
1508 cfg
->fc_type
, tb
->tb_id
);
1510 rtmsg_fib(RTM_DELROUTE
, htonl(key
), fa_to_delete
, plen
, tb
->tb_id
,
1511 &cfg
->fc_nlinfo
, 0);
1514 tb
->tb_num_default
--;
1516 fib_remove_alias(t
, tp
, l
, fa_to_delete
);
1518 if (fa_to_delete
->fa_state
& FA_S_ACCESSED
)
1519 rt_cache_flush(cfg
->fc_nlinfo
.nl_net
);
1521 fib_release_info(fa_to_delete
->fa_info
);
1522 alias_free_mem_rcu(fa_to_delete
);
1526 /* Scan for the next leaf starting at the provided key value */
1527 static struct key_vector
*leaf_walk_rcu(struct key_vector
**tn
, t_key key
)
1529 struct key_vector
*pn
, *n
= *tn
;
1530 unsigned long cindex
;
1532 /* record parent node for backtracing */
1534 cindex
= n
? get_index(key
, n
) : 0;
1536 /* this loop is meant to try and find the key in the trie */
1538 unsigned long idx
= get_index(key
, n
);
1540 /* guarantee forward progress on the keys */
1541 if (IS_LEAF(n
) && (n
->key
>= key
))
1543 if (idx
>= (1ul << n
->bits
))
1546 /* record parent and next child index */
1550 /* descend into the next child */
1551 n
= get_child_rcu(pn
, cindex
++);
1554 /* this loop will search for the next leaf with a greater key */
1556 /* if we exhausted the parent node we will need to climb */
1557 if (cindex
>= (1ul << pn
->bits
)) {
1558 t_key pkey
= pn
->key
;
1560 pn
= node_parent_rcu(pn
);
1564 cindex
= get_index(pkey
, pn
) + 1;
1568 /* grab the next available node */
1569 n
= get_child_rcu(pn
, cindex
++);
1573 /* no need to compare keys since we bumped the index */
1577 /* Rescan start scanning in new node */
1583 return NULL
; /* Root of trie */
1585 /* if we are at the limit for keys just return NULL for the tnode */
1586 *tn
= (n
->key
== KEY_MAX
) ? NULL
: pn
;
1590 /* Caller must hold RTNL */
1591 void fib_table_flush_external(struct fib_table
*tb
)
1593 struct trie
*t
= (struct trie
*)tb
->tb_data
;
1594 struct fib_alias
*fa
;
1595 struct key_vector
*n
, *pn
;
1596 unsigned long cindex
;
1598 n
= rcu_dereference(t
->tnode
[0]);
1605 while (IS_TNODE(n
)) {
1606 /* record pn and cindex for leaf walking */
1608 cindex
= 1ul << n
->bits
;
1610 /* walk trie in reverse order */
1612 while (!(cindex
--)) {
1613 t_key pkey
= pn
->key
;
1615 /* if we got the root we are done */
1616 pn
= node_parent(pn
);
1620 cindex
= get_index(pkey
, pn
);
1623 /* grab the next available node */
1624 n
= get_child(pn
, cindex
);
1628 hlist_for_each_entry(fa
, &n
->leaf
, fa_list
) {
1629 struct fib_info
*fi
= fa
->fa_info
;
1631 if (!fi
|| !(fi
->fib_flags
& RTNH_F_EXTERNAL
))
1634 netdev_switch_fib_ipv4_del(n
->key
,
1635 KEYLENGTH
- fa
->fa_slen
,
1637 fa
->fa_type
, tb
->tb_id
);
1640 /* if trie is leaf only loop is completed */
1645 /* Caller must hold RTNL. */
1646 int fib_table_flush(struct fib_table
*tb
)
1648 struct trie
*t
= (struct trie
*)tb
->tb_data
;
1649 struct key_vector
*n
, *pn
;
1650 struct hlist_node
*tmp
;
1651 struct fib_alias
*fa
;
1652 unsigned long cindex
;
1656 n
= rcu_dereference(t
->tnode
[0]);
1658 goto flush_complete
;
1663 while (IS_TNODE(n
)) {
1664 /* record pn and cindex for leaf walking */
1666 cindex
= 1ul << n
->bits
;
1668 /* walk trie in reverse order */
1670 while (!(cindex
--)) {
1671 struct key_vector __rcu
**cptr
;
1672 t_key pkey
= pn
->key
;
1675 pn
= node_parent(n
);
1677 /* resize completed node */
1678 cptr
= resize(t
, n
);
1680 /* if we got the root we are done */
1682 goto flush_complete
;
1684 pn
= container_of(cptr
, struct key_vector
,
1686 cindex
= get_index(pkey
, pn
);
1689 /* grab the next available node */
1690 n
= get_child(pn
, cindex
);
1694 /* track slen in case any prefixes survive */
1697 hlist_for_each_entry_safe(fa
, tmp
, &n
->leaf
, fa_list
) {
1698 struct fib_info
*fi
= fa
->fa_info
;
1700 if (fi
&& (fi
->fib_flags
& RTNH_F_DEAD
)) {
1701 netdev_switch_fib_ipv4_del(n
->key
,
1702 KEYLENGTH
- fa
->fa_slen
,
1704 fa
->fa_type
, tb
->tb_id
);
1705 hlist_del_rcu(&fa
->fa_list
);
1706 fib_release_info(fa
->fa_info
);
1707 alias_free_mem_rcu(fa
);
1716 /* update leaf slen */
1719 if (hlist_empty(&n
->leaf
)) {
1720 put_child_root(pn
, t
, n
->key
, NULL
);
1723 leaf_pull_suffix(pn
, n
);
1726 /* if trie is leaf only loop is completed */
1730 pr_debug("trie_flush found=%d\n", found
);
1734 static void __trie_free_rcu(struct rcu_head
*head
)
1736 struct fib_table
*tb
= container_of(head
, struct fib_table
, rcu
);
1737 #ifdef CONFIG_IP_FIB_TRIE_STATS
1738 struct trie
*t
= (struct trie
*)tb
->tb_data
;
1740 free_percpu(t
->stats
);
1741 #endif /* CONFIG_IP_FIB_TRIE_STATS */
1745 void fib_free_table(struct fib_table
*tb
)
1747 call_rcu(&tb
->rcu
, __trie_free_rcu
);
1750 static int fn_trie_dump_leaf(struct key_vector
*l
, struct fib_table
*tb
,
1751 struct sk_buff
*skb
, struct netlink_callback
*cb
)
1753 __be32 xkey
= htonl(l
->key
);
1754 struct fib_alias
*fa
;
1760 /* rcu_read_lock is hold by caller */
1761 hlist_for_each_entry_rcu(fa
, &l
->leaf
, fa_list
) {
1767 if (fib_dump_info(skb
, NETLINK_CB(cb
->skb
).portid
,
1773 KEYLENGTH
- fa
->fa_slen
,
1775 fa
->fa_info
, NLM_F_MULTI
) < 0) {
1786 /* rcu_read_lock needs to be hold by caller from readside */
1787 int fib_table_dump(struct fib_table
*tb
, struct sk_buff
*skb
,
1788 struct netlink_callback
*cb
)
1790 struct trie
*t
= (struct trie
*)tb
->tb_data
;
1791 struct key_vector
*l
, *tp
;
1792 /* Dump starting at last key.
1793 * Note: 0.0.0.0/0 (ie default) is first key.
1795 int count
= cb
->args
[2];
1796 t_key key
= cb
->args
[3];
1798 tp
= rcu_dereference_rtnl(t
->tnode
[0]);
1800 while ((l
= leaf_walk_rcu(&tp
, key
)) != NULL
) {
1801 if (fn_trie_dump_leaf(l
, tb
, skb
, cb
) < 0) {
1803 cb
->args
[2] = count
;
1810 memset(&cb
->args
[4], 0,
1811 sizeof(cb
->args
) - 4*sizeof(cb
->args
[0]));
1813 /* stop loop if key wrapped back to 0 */
1819 cb
->args
[2] = count
;
1824 void __init
fib_trie_init(void)
1826 fn_alias_kmem
= kmem_cache_create("ip_fib_alias",
1827 sizeof(struct fib_alias
),
1828 0, SLAB_PANIC
, NULL
);
1830 trie_leaf_kmem
= kmem_cache_create("ip_fib_trie",
1832 0, SLAB_PANIC
, NULL
);
1836 struct fib_table
*fib_trie_table(u32 id
)
1838 struct fib_table
*tb
;
1841 tb
= kmalloc(sizeof(struct fib_table
) + sizeof(struct trie
),
1847 tb
->tb_default
= -1;
1848 tb
->tb_num_default
= 0;
1850 t
= (struct trie
*) tb
->tb_data
;
1851 RCU_INIT_POINTER(t
->tnode
[0], NULL
);
1852 #ifdef CONFIG_IP_FIB_TRIE_STATS
1853 t
->stats
= alloc_percpu(struct trie_use_stats
);
1863 #ifdef CONFIG_PROC_FS
1864 /* Depth first Trie walk iterator */
1865 struct fib_trie_iter
{
1866 struct seq_net_private p
;
1867 struct fib_table
*tb
;
1868 struct key_vector
*tnode
;
1873 static struct key_vector
*fib_trie_get_next(struct fib_trie_iter
*iter
)
1875 unsigned long cindex
= iter
->index
;
1876 struct key_vector
*tn
= iter
->tnode
;
1877 struct key_vector
*p
;
1879 /* A single entry routing table */
1883 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
1884 iter
->tnode
, iter
->index
, iter
->depth
);
1886 while (cindex
< child_length(tn
)) {
1887 struct key_vector
*n
= get_child_rcu(tn
, cindex
);
1892 iter
->index
= cindex
+ 1;
1894 /* push down one level */
1905 /* Current node exhausted, pop back up */
1906 p
= node_parent_rcu(tn
);
1908 cindex
= get_index(tn
->key
, p
) + 1;
1918 static struct key_vector
*fib_trie_get_first(struct fib_trie_iter
*iter
,
1921 struct key_vector
*n
;
1926 n
= rcu_dereference(t
->tnode
[0]);
1943 static void trie_collect_stats(struct trie
*t
, struct trie_stat
*s
)
1945 struct key_vector
*n
;
1946 struct fib_trie_iter iter
;
1948 memset(s
, 0, sizeof(*s
));
1951 for (n
= fib_trie_get_first(&iter
, t
); n
; n
= fib_trie_get_next(&iter
)) {
1953 struct fib_alias
*fa
;
1956 s
->totdepth
+= iter
.depth
;
1957 if (iter
.depth
> s
->maxdepth
)
1958 s
->maxdepth
= iter
.depth
;
1960 hlist_for_each_entry_rcu(fa
, &n
->leaf
, fa_list
)
1964 if (n
->bits
< MAX_STAT_DEPTH
)
1965 s
->nodesizes
[n
->bits
]++;
1966 s
->nullpointers
+= n
->empty_children
;
1973 * This outputs /proc/net/fib_triestats
1975 static void trie_show_stats(struct seq_file
*seq
, struct trie_stat
*stat
)
1977 unsigned int i
, max
, pointers
, bytes
, avdepth
;
1980 avdepth
= stat
->totdepth
*100 / stat
->leaves
;
1984 seq_printf(seq
, "\tAver depth: %u.%02d\n",
1985 avdepth
/ 100, avdepth
% 100);
1986 seq_printf(seq
, "\tMax depth: %u\n", stat
->maxdepth
);
1988 seq_printf(seq
, "\tLeaves: %u\n", stat
->leaves
);
1989 bytes
= LEAF_SIZE
* stat
->leaves
;
1991 seq_printf(seq
, "\tPrefixes: %u\n", stat
->prefixes
);
1992 bytes
+= sizeof(struct fib_alias
) * stat
->prefixes
;
1994 seq_printf(seq
, "\tInternal nodes: %u\n\t", stat
->tnodes
);
1995 bytes
+= TNODE_SIZE(0) * stat
->tnodes
;
1997 max
= MAX_STAT_DEPTH
;
1998 while (max
> 0 && stat
->nodesizes
[max
-1] == 0)
2002 for (i
= 1; i
< max
; i
++)
2003 if (stat
->nodesizes
[i
] != 0) {
2004 seq_printf(seq
, " %u: %u", i
, stat
->nodesizes
[i
]);
2005 pointers
+= (1<<i
) * stat
->nodesizes
[i
];
2007 seq_putc(seq
, '\n');
2008 seq_printf(seq
, "\tPointers: %u\n", pointers
);
2010 bytes
+= sizeof(struct key_vector
*) * pointers
;
2011 seq_printf(seq
, "Null ptrs: %u\n", stat
->nullpointers
);
2012 seq_printf(seq
, "Total size: %u kB\n", (bytes
+ 1023) / 1024);
2015 #ifdef CONFIG_IP_FIB_TRIE_STATS
2016 static void trie_show_usage(struct seq_file
*seq
,
2017 const struct trie_use_stats __percpu
*stats
)
2019 struct trie_use_stats s
= { 0 };
2022 /* loop through all of the CPUs and gather up the stats */
2023 for_each_possible_cpu(cpu
) {
2024 const struct trie_use_stats
*pcpu
= per_cpu_ptr(stats
, cpu
);
2026 s
.gets
+= pcpu
->gets
;
2027 s
.backtrack
+= pcpu
->backtrack
;
2028 s
.semantic_match_passed
+= pcpu
->semantic_match_passed
;
2029 s
.semantic_match_miss
+= pcpu
->semantic_match_miss
;
2030 s
.null_node_hit
+= pcpu
->null_node_hit
;
2031 s
.resize_node_skipped
+= pcpu
->resize_node_skipped
;
2034 seq_printf(seq
, "\nCounters:\n---------\n");
2035 seq_printf(seq
, "gets = %u\n", s
.gets
);
2036 seq_printf(seq
, "backtracks = %u\n", s
.backtrack
);
2037 seq_printf(seq
, "semantic match passed = %u\n",
2038 s
.semantic_match_passed
);
2039 seq_printf(seq
, "semantic match miss = %u\n", s
.semantic_match_miss
);
2040 seq_printf(seq
, "null node hit= %u\n", s
.null_node_hit
);
2041 seq_printf(seq
, "skipped node resize = %u\n\n", s
.resize_node_skipped
);
2043 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2045 static void fib_table_print(struct seq_file
*seq
, struct fib_table
*tb
)
2047 if (tb
->tb_id
== RT_TABLE_LOCAL
)
2048 seq_puts(seq
, "Local:\n");
2049 else if (tb
->tb_id
== RT_TABLE_MAIN
)
2050 seq_puts(seq
, "Main:\n");
2052 seq_printf(seq
, "Id %d:\n", tb
->tb_id
);
2056 static int fib_triestat_seq_show(struct seq_file
*seq
, void *v
)
2058 struct net
*net
= (struct net
*)seq
->private;
2062 "Basic info: size of leaf:"
2063 " %Zd bytes, size of tnode: %Zd bytes.\n",
2064 LEAF_SIZE
, TNODE_SIZE(0));
2066 for (h
= 0; h
< FIB_TABLE_HASHSZ
; h
++) {
2067 struct hlist_head
*head
= &net
->ipv4
.fib_table_hash
[h
];
2068 struct fib_table
*tb
;
2070 hlist_for_each_entry_rcu(tb
, head
, tb_hlist
) {
2071 struct trie
*t
= (struct trie
*) tb
->tb_data
;
2072 struct trie_stat stat
;
2077 fib_table_print(seq
, tb
);
2079 trie_collect_stats(t
, &stat
);
2080 trie_show_stats(seq
, &stat
);
2081 #ifdef CONFIG_IP_FIB_TRIE_STATS
2082 trie_show_usage(seq
, t
->stats
);
2090 static int fib_triestat_seq_open(struct inode
*inode
, struct file
*file
)
2092 return single_open_net(inode
, file
, fib_triestat_seq_show
);
2095 static const struct file_operations fib_triestat_fops
= {
2096 .owner
= THIS_MODULE
,
2097 .open
= fib_triestat_seq_open
,
2099 .llseek
= seq_lseek
,
2100 .release
= single_release_net
,
2103 static struct key_vector
*fib_trie_get_idx(struct seq_file
*seq
, loff_t pos
)
2105 struct fib_trie_iter
*iter
= seq
->private;
2106 struct net
*net
= seq_file_net(seq
);
2110 for (h
= 0; h
< FIB_TABLE_HASHSZ
; h
++) {
2111 struct hlist_head
*head
= &net
->ipv4
.fib_table_hash
[h
];
2112 struct fib_table
*tb
;
2114 hlist_for_each_entry_rcu(tb
, head
, tb_hlist
) {
2115 struct key_vector
*n
;
2117 for (n
= fib_trie_get_first(iter
,
2118 (struct trie
*) tb
->tb_data
);
2119 n
; n
= fib_trie_get_next(iter
))
2130 static void *fib_trie_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2134 return fib_trie_get_idx(seq
, *pos
);
2137 static void *fib_trie_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2139 struct fib_trie_iter
*iter
= seq
->private;
2140 struct net
*net
= seq_file_net(seq
);
2141 struct fib_table
*tb
= iter
->tb
;
2142 struct hlist_node
*tb_node
;
2144 struct key_vector
*n
;
2147 /* next node in same table */
2148 n
= fib_trie_get_next(iter
);
2152 /* walk rest of this hash chain */
2153 h
= tb
->tb_id
& (FIB_TABLE_HASHSZ
- 1);
2154 while ((tb_node
= rcu_dereference(hlist_next_rcu(&tb
->tb_hlist
)))) {
2155 tb
= hlist_entry(tb_node
, struct fib_table
, tb_hlist
);
2156 n
= fib_trie_get_first(iter
, (struct trie
*) tb
->tb_data
);
2161 /* new hash chain */
2162 while (++h
< FIB_TABLE_HASHSZ
) {
2163 struct hlist_head
*head
= &net
->ipv4
.fib_table_hash
[h
];
2164 hlist_for_each_entry_rcu(tb
, head
, tb_hlist
) {
2165 n
= fib_trie_get_first(iter
, (struct trie
*) tb
->tb_data
);
2177 static void fib_trie_seq_stop(struct seq_file
*seq
, void *v
)
2183 static void seq_indent(struct seq_file
*seq
, int n
)
2189 static inline const char *rtn_scope(char *buf
, size_t len
, enum rt_scope_t s
)
2192 case RT_SCOPE_UNIVERSE
: return "universe";
2193 case RT_SCOPE_SITE
: return "site";
2194 case RT_SCOPE_LINK
: return "link";
2195 case RT_SCOPE_HOST
: return "host";
2196 case RT_SCOPE_NOWHERE
: return "nowhere";
2198 snprintf(buf
, len
, "scope=%d", s
);
2203 static const char *const rtn_type_names
[__RTN_MAX
] = {
2204 [RTN_UNSPEC
] = "UNSPEC",
2205 [RTN_UNICAST
] = "UNICAST",
2206 [RTN_LOCAL
] = "LOCAL",
2207 [RTN_BROADCAST
] = "BROADCAST",
2208 [RTN_ANYCAST
] = "ANYCAST",
2209 [RTN_MULTICAST
] = "MULTICAST",
2210 [RTN_BLACKHOLE
] = "BLACKHOLE",
2211 [RTN_UNREACHABLE
] = "UNREACHABLE",
2212 [RTN_PROHIBIT
] = "PROHIBIT",
2213 [RTN_THROW
] = "THROW",
2215 [RTN_XRESOLVE
] = "XRESOLVE",
2218 static inline const char *rtn_type(char *buf
, size_t len
, unsigned int t
)
2220 if (t
< __RTN_MAX
&& rtn_type_names
[t
])
2221 return rtn_type_names
[t
];
2222 snprintf(buf
, len
, "type %u", t
);
2226 /* Pretty print the trie */
2227 static int fib_trie_seq_show(struct seq_file
*seq
, void *v
)
2229 const struct fib_trie_iter
*iter
= seq
->private;
2230 struct key_vector
*n
= v
;
2232 if (!node_parent_rcu(n
))
2233 fib_table_print(seq
, iter
->tb
);
2236 __be32 prf
= htonl(n
->key
);
2238 seq_indent(seq
, iter
->depth
-1);
2239 seq_printf(seq
, " +-- %pI4/%zu %u %u %u\n",
2240 &prf
, KEYLENGTH
- n
->pos
- n
->bits
, n
->bits
,
2241 n
->full_children
, n
->empty_children
);
2243 __be32 val
= htonl(n
->key
);
2244 struct fib_alias
*fa
;
2246 seq_indent(seq
, iter
->depth
);
2247 seq_printf(seq
, " |-- %pI4\n", &val
);
2249 hlist_for_each_entry_rcu(fa
, &n
->leaf
, fa_list
) {
2250 char buf1
[32], buf2
[32];
2252 seq_indent(seq
, iter
->depth
+ 1);
2253 seq_printf(seq
, " /%zu %s %s",
2254 KEYLENGTH
- fa
->fa_slen
,
2255 rtn_scope(buf1
, sizeof(buf1
),
2256 fa
->fa_info
->fib_scope
),
2257 rtn_type(buf2
, sizeof(buf2
),
2260 seq_printf(seq
, " tos=%d", fa
->fa_tos
);
2261 seq_putc(seq
, '\n');
2268 static const struct seq_operations fib_trie_seq_ops
= {
2269 .start
= fib_trie_seq_start
,
2270 .next
= fib_trie_seq_next
,
2271 .stop
= fib_trie_seq_stop
,
2272 .show
= fib_trie_seq_show
,
2275 static int fib_trie_seq_open(struct inode
*inode
, struct file
*file
)
2277 return seq_open_net(inode
, file
, &fib_trie_seq_ops
,
2278 sizeof(struct fib_trie_iter
));
2281 static const struct file_operations fib_trie_fops
= {
2282 .owner
= THIS_MODULE
,
2283 .open
= fib_trie_seq_open
,
2285 .llseek
= seq_lseek
,
2286 .release
= seq_release_net
,
2289 struct fib_route_iter
{
2290 struct seq_net_private p
;
2291 struct fib_table
*main_tb
;
2292 struct key_vector
*tnode
;
2297 static struct key_vector
*fib_route_get_idx(struct fib_route_iter
*iter
,
2300 struct fib_table
*tb
= iter
->main_tb
;
2301 struct key_vector
*l
, **tp
= &iter
->tnode
;
2305 /* use cache location of next-to-find key */
2306 if (iter
->pos
> 0 && pos
>= iter
->pos
) {
2310 t
= (struct trie
*)tb
->tb_data
;
2311 iter
->tnode
= rcu_dereference_rtnl(t
->tnode
[0]);
2316 while ((l
= leaf_walk_rcu(tp
, key
)) != NULL
) {
2325 /* handle unlikely case of a key wrap */
2331 iter
->key
= key
; /* remember it */
2333 iter
->pos
= 0; /* forget it */
2338 static void *fib_route_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2341 struct fib_route_iter
*iter
= seq
->private;
2342 struct fib_table
*tb
;
2347 tb
= fib_get_table(seq_file_net(seq
), RT_TABLE_MAIN
);
2354 return fib_route_get_idx(iter
, *pos
);
2356 t
= (struct trie
*)tb
->tb_data
;
2357 iter
->tnode
= rcu_dereference_rtnl(t
->tnode
[0]);
2361 return SEQ_START_TOKEN
;
2364 static void *fib_route_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2366 struct fib_route_iter
*iter
= seq
->private;
2367 struct key_vector
*l
= NULL
;
2368 t_key key
= iter
->key
;
2372 /* only allow key of 0 for start of sequence */
2373 if ((v
== SEQ_START_TOKEN
) || key
)
2374 l
= leaf_walk_rcu(&iter
->tnode
, key
);
2377 iter
->key
= l
->key
+ 1;
2386 static void fib_route_seq_stop(struct seq_file
*seq
, void *v
)
2392 static unsigned int fib_flag_trans(int type
, __be32 mask
, const struct fib_info
*fi
)
2394 unsigned int flags
= 0;
2396 if (type
== RTN_UNREACHABLE
|| type
== RTN_PROHIBIT
)
2398 if (fi
&& fi
->fib_nh
->nh_gw
)
2399 flags
|= RTF_GATEWAY
;
2400 if (mask
== htonl(0xFFFFFFFF))
2407 * This outputs /proc/net/route.
2408 * The format of the file is not supposed to be changed
2409 * and needs to be same as fib_hash output to avoid breaking
2412 static int fib_route_seq_show(struct seq_file
*seq
, void *v
)
2414 struct fib_alias
*fa
;
2415 struct key_vector
*l
= v
;
2418 if (v
== SEQ_START_TOKEN
) {
2419 seq_printf(seq
, "%-127s\n", "Iface\tDestination\tGateway "
2420 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2425 prefix
= htonl(l
->key
);
2427 hlist_for_each_entry_rcu(fa
, &l
->leaf
, fa_list
) {
2428 const struct fib_info
*fi
= fa
->fa_info
;
2429 __be32 mask
= inet_make_mask(KEYLENGTH
- fa
->fa_slen
);
2430 unsigned int flags
= fib_flag_trans(fa
->fa_type
, mask
, fi
);
2432 if ((fa
->fa_type
== RTN_BROADCAST
) ||
2433 (fa
->fa_type
== RTN_MULTICAST
))
2436 seq_setwidth(seq
, 127);
2440 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2441 "%d\t%08X\t%d\t%u\t%u",
2442 fi
->fib_dev
? fi
->fib_dev
->name
: "*",
2444 fi
->fib_nh
->nh_gw
, flags
, 0, 0,
2448 fi
->fib_advmss
+ 40 : 0),
2453 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2454 "%d\t%08X\t%d\t%u\t%u",
2455 prefix
, 0, flags
, 0, 0, 0,
2464 static const struct seq_operations fib_route_seq_ops
= {
2465 .start
= fib_route_seq_start
,
2466 .next
= fib_route_seq_next
,
2467 .stop
= fib_route_seq_stop
,
2468 .show
= fib_route_seq_show
,
2471 static int fib_route_seq_open(struct inode
*inode
, struct file
*file
)
2473 return seq_open_net(inode
, file
, &fib_route_seq_ops
,
2474 sizeof(struct fib_route_iter
));
2477 static const struct file_operations fib_route_fops
= {
2478 .owner
= THIS_MODULE
,
2479 .open
= fib_route_seq_open
,
2481 .llseek
= seq_lseek
,
2482 .release
= seq_release_net
,
2485 int __net_init
fib_proc_init(struct net
*net
)
2487 if (!proc_create("fib_trie", S_IRUGO
, net
->proc_net
, &fib_trie_fops
))
2490 if (!proc_create("fib_triestat", S_IRUGO
, net
->proc_net
,
2491 &fib_triestat_fops
))
2494 if (!proc_create("route", S_IRUGO
, net
->proc_net
, &fib_route_fops
))
2500 remove_proc_entry("fib_triestat", net
->proc_net
);
2502 remove_proc_entry("fib_trie", net
->proc_net
);
2507 void __net_exit
fib_proc_exit(struct net
*net
)
2509 remove_proc_entry("fib_trie", net
->proc_net
);
2510 remove_proc_entry("fib_triestat", net
->proc_net
);
2511 remove_proc_entry("route", net
->proc_net
);
2514 #endif /* CONFIG_PROC_FS */