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 <linux/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 <linux/vmalloc.h>
76 #include <linux/notifier.h>
77 #include <net/net_namespace.h>
79 #include <net/protocol.h>
80 #include <net/route.h>
83 #include <net/ip_fib.h>
84 #include <trace/events/fib.h>
85 #include "fib_lookup.h"
87 static unsigned int fib_seq_sum(void)
89 unsigned int fib_seq
= 0;
94 fib_seq
+= net
->ipv4
.fib_seq
;
100 static ATOMIC_NOTIFIER_HEAD(fib_chain
);
102 static int call_fib_notifier(struct notifier_block
*nb
, struct net
*net
,
103 enum fib_event_type event_type
,
104 struct fib_notifier_info
*info
)
107 return nb
->notifier_call(nb
, event_type
, info
);
110 static void fib_rules_notify(struct net
*net
, struct notifier_block
*nb
,
111 enum fib_event_type event_type
)
113 #ifdef CONFIG_IP_MULTIPLE_TABLES
114 struct fib_notifier_info info
;
116 if (net
->ipv4
.fib_has_custom_rules
)
117 call_fib_notifier(nb
, net
, event_type
, &info
);
121 static void fib_notify(struct net
*net
, struct notifier_block
*nb
,
122 enum fib_event_type event_type
);
124 static int call_fib_entry_notifier(struct notifier_block
*nb
, struct net
*net
,
125 enum fib_event_type event_type
, u32 dst
,
126 int dst_len
, struct fib_info
*fi
,
127 u8 tos
, u8 type
, u32 tb_id
)
129 struct fib_entry_notifier_info info
= {
137 return call_fib_notifier(nb
, net
, event_type
, &info
.info
);
140 static bool fib_dump_is_consistent(struct notifier_block
*nb
,
141 void (*cb
)(struct notifier_block
*nb
),
142 unsigned int fib_seq
)
144 atomic_notifier_chain_register(&fib_chain
, nb
);
145 if (fib_seq
== fib_seq_sum())
147 atomic_notifier_chain_unregister(&fib_chain
, nb
);
153 #define FIB_DUMP_MAX_RETRIES 5
154 int register_fib_notifier(struct notifier_block
*nb
,
155 void (*cb
)(struct notifier_block
*nb
))
160 unsigned int fib_seq
= fib_seq_sum();
163 /* Mutex semantics guarantee that every change done to
164 * FIB tries before we read the change sequence counter
165 * is now visible to us.
168 for_each_net_rcu(net
) {
169 fib_rules_notify(net
, nb
, FIB_EVENT_RULE_ADD
);
170 fib_notify(net
, nb
, FIB_EVENT_ENTRY_ADD
);
174 if (fib_dump_is_consistent(nb
, cb
, fib_seq
))
176 } while (++retries
< FIB_DUMP_MAX_RETRIES
);
180 EXPORT_SYMBOL(register_fib_notifier
);
182 int unregister_fib_notifier(struct notifier_block
*nb
)
184 return atomic_notifier_chain_unregister(&fib_chain
, nb
);
186 EXPORT_SYMBOL(unregister_fib_notifier
);
188 int call_fib_notifiers(struct net
*net
, enum fib_event_type event_type
,
189 struct fib_notifier_info
*info
)
193 return atomic_notifier_call_chain(&fib_chain
, event_type
, info
);
196 static int call_fib_entry_notifiers(struct net
*net
,
197 enum fib_event_type event_type
, u32 dst
,
198 int dst_len
, struct fib_info
*fi
,
199 u8 tos
, u8 type
, u32 tb_id
)
201 struct fib_entry_notifier_info info
= {
209 return call_fib_notifiers(net
, event_type
, &info
.info
);
212 #define MAX_STAT_DEPTH 32
214 #define KEYLENGTH (8*sizeof(t_key))
215 #define KEY_MAX ((t_key)~0)
217 typedef unsigned int t_key
;
219 #define IS_TRIE(n) ((n)->pos >= KEYLENGTH)
220 #define IS_TNODE(n) ((n)->bits)
221 #define IS_LEAF(n) (!(n)->bits)
225 unsigned char pos
; /* 2log(KEYLENGTH) bits needed */
226 unsigned char bits
; /* 2log(KEYLENGTH) bits needed */
229 /* This list pointer if valid if (pos | bits) == 0 (LEAF) */
230 struct hlist_head leaf
;
231 /* This array is valid if (pos | bits) > 0 (TNODE) */
232 struct key_vector __rcu
*tnode
[0];
238 t_key empty_children
; /* KEYLENGTH bits needed */
239 t_key full_children
; /* KEYLENGTH bits needed */
240 struct key_vector __rcu
*parent
;
241 struct key_vector kv
[1];
242 #define tn_bits kv[0].bits
245 #define TNODE_SIZE(n) offsetof(struct tnode, kv[0].tnode[n])
246 #define LEAF_SIZE TNODE_SIZE(1)
248 #ifdef CONFIG_IP_FIB_TRIE_STATS
249 struct trie_use_stats
{
251 unsigned int backtrack
;
252 unsigned int semantic_match_passed
;
253 unsigned int semantic_match_miss
;
254 unsigned int null_node_hit
;
255 unsigned int resize_node_skipped
;
260 unsigned int totdepth
;
261 unsigned int maxdepth
;
264 unsigned int nullpointers
;
265 unsigned int prefixes
;
266 unsigned int nodesizes
[MAX_STAT_DEPTH
];
270 struct key_vector kv
[1];
271 #ifdef CONFIG_IP_FIB_TRIE_STATS
272 struct trie_use_stats __percpu
*stats
;
276 static struct key_vector
*resize(struct trie
*t
, struct key_vector
*tn
);
277 static size_t tnode_free_size
;
280 * synchronize_rcu after call_rcu for that many pages; it should be especially
281 * useful before resizing the root node with PREEMPT_NONE configs; the value was
282 * obtained experimentally, aiming to avoid visible slowdown.
284 static const int sync_pages
= 128;
286 static struct kmem_cache
*fn_alias_kmem __read_mostly
;
287 static struct kmem_cache
*trie_leaf_kmem __read_mostly
;
289 static inline struct tnode
*tn_info(struct key_vector
*kv
)
291 return container_of(kv
, struct tnode
, kv
[0]);
294 /* caller must hold RTNL */
295 #define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
296 #define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
298 /* caller must hold RCU read lock or RTNL */
299 #define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
300 #define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
302 /* wrapper for rcu_assign_pointer */
303 static inline void node_set_parent(struct key_vector
*n
, struct key_vector
*tp
)
306 rcu_assign_pointer(tn_info(n
)->parent
, tp
);
309 #define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
311 /* This provides us with the number of children in this node, in the case of a
312 * leaf this will return 0 meaning none of the children are accessible.
314 static inline unsigned long child_length(const struct key_vector
*tn
)
316 return (1ul << tn
->bits
) & ~(1ul);
319 #define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
321 static inline unsigned long get_index(t_key key
, struct key_vector
*kv
)
323 unsigned long index
= key
^ kv
->key
;
325 if ((BITS_PER_LONG
<= KEYLENGTH
) && (KEYLENGTH
== kv
->pos
))
328 return index
>> kv
->pos
;
331 /* To understand this stuff, an understanding of keys and all their bits is
332 * necessary. Every node in the trie has a key associated with it, but not
333 * all of the bits in that key are significant.
335 * Consider a node 'n' and its parent 'tp'.
337 * If n is a leaf, every bit in its key is significant. Its presence is
338 * necessitated by path compression, since during a tree traversal (when
339 * searching for a leaf - unless we are doing an insertion) we will completely
340 * ignore all skipped bits we encounter. Thus we need to verify, at the end of
341 * a potentially successful search, that we have indeed been walking the
344 * Note that we can never "miss" the correct key in the tree if present by
345 * following the wrong path. Path compression ensures that segments of the key
346 * that are the same for all keys with a given prefix are skipped, but the
347 * skipped part *is* identical for each node in the subtrie below the skipped
348 * bit! trie_insert() in this implementation takes care of that.
350 * if n is an internal node - a 'tnode' here, the various parts of its key
351 * have many different meanings.
354 * _________________________________________________________________
355 * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
356 * -----------------------------------------------------------------
357 * 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16
359 * _________________________________________________________________
360 * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
361 * -----------------------------------------------------------------
362 * 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
369 * First, let's just ignore the bits that come before the parent tp, that is
370 * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
371 * point we do not use them for anything.
373 * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
374 * index into the parent's child array. That is, they will be used to find
375 * 'n' among tp's children.
377 * The bits from (n->pos + n->bits) to (tp->pos - 1) - "S" - are skipped bits
380 * All the bits we have seen so far are significant to the node n. The rest
381 * of the bits are really not needed or indeed known in n->key.
383 * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
384 * n's child array, and will of course be different for each child.
386 * The rest of the bits, from 0 to (n->pos -1) - "u" - are completely unknown
390 static const int halve_threshold
= 25;
391 static const int inflate_threshold
= 50;
392 static const int halve_threshold_root
= 15;
393 static const int inflate_threshold_root
= 30;
395 static void __alias_free_mem(struct rcu_head
*head
)
397 struct fib_alias
*fa
= container_of(head
, struct fib_alias
, rcu
);
398 kmem_cache_free(fn_alias_kmem
, fa
);
401 static inline void alias_free_mem_rcu(struct fib_alias
*fa
)
403 call_rcu(&fa
->rcu
, __alias_free_mem
);
406 #define TNODE_KMALLOC_MAX \
407 ilog2((PAGE_SIZE - TNODE_SIZE(0)) / sizeof(struct key_vector *))
408 #define TNODE_VMALLOC_MAX \
409 ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
411 static void __node_free_rcu(struct rcu_head
*head
)
413 struct tnode
*n
= container_of(head
, struct tnode
, rcu
);
416 kmem_cache_free(trie_leaf_kmem
, n
);
421 #define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
423 static struct tnode
*tnode_alloc(int bits
)
427 /* verify bits is within bounds */
428 if (bits
> TNODE_VMALLOC_MAX
)
431 /* determine size and verify it is non-zero and didn't overflow */
432 size
= TNODE_SIZE(1ul << bits
);
434 if (size
<= PAGE_SIZE
)
435 return kzalloc(size
, GFP_KERNEL
);
437 return vzalloc(size
);
440 static inline void empty_child_inc(struct key_vector
*n
)
442 ++tn_info(n
)->empty_children
? : ++tn_info(n
)->full_children
;
445 static inline void empty_child_dec(struct key_vector
*n
)
447 tn_info(n
)->empty_children
-- ? : tn_info(n
)->full_children
--;
450 static struct key_vector
*leaf_new(t_key key
, struct fib_alias
*fa
)
452 struct key_vector
*l
;
455 kv
= kmem_cache_alloc(trie_leaf_kmem
, GFP_KERNEL
);
459 /* initialize key vector */
464 l
->slen
= fa
->fa_slen
;
466 /* link leaf to fib alias */
467 INIT_HLIST_HEAD(&l
->leaf
);
468 hlist_add_head(&fa
->fa_list
, &l
->leaf
);
473 static struct key_vector
*tnode_new(t_key key
, int pos
, int bits
)
475 unsigned int shift
= pos
+ bits
;
476 struct key_vector
*tn
;
479 /* verify bits and pos their msb bits clear and values are valid */
480 BUG_ON(!bits
|| (shift
> KEYLENGTH
));
482 tnode
= tnode_alloc(bits
);
486 pr_debug("AT %p s=%zu %zu\n", tnode
, TNODE_SIZE(0),
487 sizeof(struct key_vector
*) << bits
);
489 if (bits
== KEYLENGTH
)
490 tnode
->full_children
= 1;
492 tnode
->empty_children
= 1ul << bits
;
495 tn
->key
= (shift
< KEYLENGTH
) ? (key
>> shift
) << shift
: 0;
503 /* Check whether a tnode 'n' is "full", i.e. it is an internal node
504 * and no bits are skipped. See discussion in dyntree paper p. 6
506 static inline int tnode_full(struct key_vector
*tn
, struct key_vector
*n
)
508 return n
&& ((n
->pos
+ n
->bits
) == tn
->pos
) && IS_TNODE(n
);
511 /* Add a child at position i overwriting the old value.
512 * Update the value of full_children and empty_children.
514 static void put_child(struct key_vector
*tn
, unsigned long i
,
515 struct key_vector
*n
)
517 struct key_vector
*chi
= get_child(tn
, i
);
520 BUG_ON(i
>= child_length(tn
));
522 /* update emptyChildren, overflow into fullChildren */
528 /* update fullChildren */
529 wasfull
= tnode_full(tn
, chi
);
530 isfull
= tnode_full(tn
, n
);
532 if (wasfull
&& !isfull
)
533 tn_info(tn
)->full_children
--;
534 else if (!wasfull
&& isfull
)
535 tn_info(tn
)->full_children
++;
537 if (n
&& (tn
->slen
< n
->slen
))
540 rcu_assign_pointer(tn
->tnode
[i
], n
);
543 static void update_children(struct key_vector
*tn
)
547 /* update all of the child parent pointers */
548 for (i
= child_length(tn
); i
;) {
549 struct key_vector
*inode
= get_child(tn
, --i
);
554 /* Either update the children of a tnode that
555 * already belongs to us or update the child
556 * to point to ourselves.
558 if (node_parent(inode
) == tn
)
559 update_children(inode
);
561 node_set_parent(inode
, tn
);
565 static inline void put_child_root(struct key_vector
*tp
, t_key key
,
566 struct key_vector
*n
)
569 rcu_assign_pointer(tp
->tnode
[0], n
);
571 put_child(tp
, get_index(key
, tp
), n
);
574 static inline void tnode_free_init(struct key_vector
*tn
)
576 tn_info(tn
)->rcu
.next
= NULL
;
579 static inline void tnode_free_append(struct key_vector
*tn
,
580 struct key_vector
*n
)
582 tn_info(n
)->rcu
.next
= tn_info(tn
)->rcu
.next
;
583 tn_info(tn
)->rcu
.next
= &tn_info(n
)->rcu
;
586 static void tnode_free(struct key_vector
*tn
)
588 struct callback_head
*head
= &tn_info(tn
)->rcu
;
592 tnode_free_size
+= TNODE_SIZE(1ul << tn
->bits
);
595 tn
= container_of(head
, struct tnode
, rcu
)->kv
;
598 if (tnode_free_size
>= PAGE_SIZE
* sync_pages
) {
604 static struct key_vector
*replace(struct trie
*t
,
605 struct key_vector
*oldtnode
,
606 struct key_vector
*tn
)
608 struct key_vector
*tp
= node_parent(oldtnode
);
611 /* setup the parent pointer out of and back into this node */
612 NODE_INIT_PARENT(tn
, tp
);
613 put_child_root(tp
, tn
->key
, tn
);
615 /* update all of the child parent pointers */
618 /* all pointers should be clean so we are done */
619 tnode_free(oldtnode
);
621 /* resize children now that oldtnode is freed */
622 for (i
= child_length(tn
); i
;) {
623 struct key_vector
*inode
= get_child(tn
, --i
);
625 /* resize child node */
626 if (tnode_full(tn
, inode
))
627 tn
= resize(t
, inode
);
633 static struct key_vector
*inflate(struct trie
*t
,
634 struct key_vector
*oldtnode
)
636 struct key_vector
*tn
;
640 pr_debug("In inflate\n");
642 tn
= tnode_new(oldtnode
->key
, oldtnode
->pos
- 1, oldtnode
->bits
+ 1);
646 /* prepare oldtnode to be freed */
647 tnode_free_init(oldtnode
);
649 /* Assemble all of the pointers in our cluster, in this case that
650 * represents all of the pointers out of our allocated nodes that
651 * point to existing tnodes and the links between our allocated
654 for (i
= child_length(oldtnode
), m
= 1u << tn
->pos
; i
;) {
655 struct key_vector
*inode
= get_child(oldtnode
, --i
);
656 struct key_vector
*node0
, *node1
;
663 /* A leaf or an internal node with skipped bits */
664 if (!tnode_full(oldtnode
, inode
)) {
665 put_child(tn
, get_index(inode
->key
, tn
), inode
);
669 /* drop the node in the old tnode free list */
670 tnode_free_append(oldtnode
, inode
);
672 /* An internal node with two children */
673 if (inode
->bits
== 1) {
674 put_child(tn
, 2 * i
+ 1, get_child(inode
, 1));
675 put_child(tn
, 2 * i
, get_child(inode
, 0));
679 /* We will replace this node 'inode' with two new
680 * ones, 'node0' and 'node1', each with half of the
681 * original children. The two new nodes will have
682 * a position one bit further down the key and this
683 * means that the "significant" part of their keys
684 * (see the discussion near the top of this file)
685 * will differ by one bit, which will be "0" in
686 * node0's key and "1" in node1's key. Since we are
687 * moving the key position by one step, the bit that
688 * we are moving away from - the bit at position
689 * (tn->pos) - is the one that will differ between
690 * node0 and node1. So... we synthesize that bit in the
693 node1
= tnode_new(inode
->key
| m
, inode
->pos
, inode
->bits
- 1);
696 node0
= tnode_new(inode
->key
, inode
->pos
, inode
->bits
- 1);
698 tnode_free_append(tn
, node1
);
701 tnode_free_append(tn
, node0
);
703 /* populate child pointers in new nodes */
704 for (k
= child_length(inode
), j
= k
/ 2; j
;) {
705 put_child(node1
, --j
, get_child(inode
, --k
));
706 put_child(node0
, j
, get_child(inode
, j
));
707 put_child(node1
, --j
, get_child(inode
, --k
));
708 put_child(node0
, j
, get_child(inode
, j
));
711 /* link new nodes to parent */
712 NODE_INIT_PARENT(node1
, tn
);
713 NODE_INIT_PARENT(node0
, tn
);
715 /* link parent to nodes */
716 put_child(tn
, 2 * i
+ 1, node1
);
717 put_child(tn
, 2 * i
, node0
);
720 /* setup the parent pointers into and out of this node */
721 return replace(t
, oldtnode
, tn
);
723 /* all pointers should be clean so we are done */
729 static struct key_vector
*halve(struct trie
*t
,
730 struct key_vector
*oldtnode
)
732 struct key_vector
*tn
;
735 pr_debug("In halve\n");
737 tn
= tnode_new(oldtnode
->key
, oldtnode
->pos
+ 1, oldtnode
->bits
- 1);
741 /* prepare oldtnode to be freed */
742 tnode_free_init(oldtnode
);
744 /* Assemble all of the pointers in our cluster, in this case that
745 * represents all of the pointers out of our allocated nodes that
746 * point to existing tnodes and the links between our allocated
749 for (i
= child_length(oldtnode
); i
;) {
750 struct key_vector
*node1
= get_child(oldtnode
, --i
);
751 struct key_vector
*node0
= get_child(oldtnode
, --i
);
752 struct key_vector
*inode
;
754 /* At least one of the children is empty */
755 if (!node1
|| !node0
) {
756 put_child(tn
, i
/ 2, node1
? : node0
);
760 /* Two nonempty children */
761 inode
= tnode_new(node0
->key
, oldtnode
->pos
, 1);
764 tnode_free_append(tn
, inode
);
766 /* initialize pointers out of node */
767 put_child(inode
, 1, node1
);
768 put_child(inode
, 0, node0
);
769 NODE_INIT_PARENT(inode
, tn
);
771 /* link parent to node */
772 put_child(tn
, i
/ 2, inode
);
775 /* setup the parent pointers into and out of this node */
776 return replace(t
, oldtnode
, tn
);
778 /* all pointers should be clean so we are done */
784 static struct key_vector
*collapse(struct trie
*t
,
785 struct key_vector
*oldtnode
)
787 struct key_vector
*n
, *tp
;
790 /* scan the tnode looking for that one child that might still exist */
791 for (n
= NULL
, i
= child_length(oldtnode
); !n
&& i
;)
792 n
= get_child(oldtnode
, --i
);
794 /* compress one level */
795 tp
= node_parent(oldtnode
);
796 put_child_root(tp
, oldtnode
->key
, n
);
797 node_set_parent(n
, tp
);
805 static unsigned char update_suffix(struct key_vector
*tn
)
807 unsigned char slen
= tn
->pos
;
808 unsigned long stride
, i
;
809 unsigned char slen_max
;
811 /* only vector 0 can have a suffix length greater than or equal to
812 * tn->pos + tn->bits, the second highest node will have a suffix
813 * length at most of tn->pos + tn->bits - 1
815 slen_max
= min_t(unsigned char, tn
->pos
+ tn
->bits
- 1, tn
->slen
);
817 /* search though the list of children looking for nodes that might
818 * have a suffix greater than the one we currently have. This is
819 * why we start with a stride of 2 since a stride of 1 would
820 * represent the nodes with suffix length equal to tn->pos
822 for (i
= 0, stride
= 0x2ul
; i
< child_length(tn
); i
+= stride
) {
823 struct key_vector
*n
= get_child(tn
, i
);
825 if (!n
|| (n
->slen
<= slen
))
828 /* update stride and slen based on new value */
829 stride
<<= (n
->slen
- slen
);
833 /* stop searching if we have hit the maximum possible value */
834 if (slen
>= slen_max
)
843 /* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
844 * the Helsinki University of Technology and Matti Tikkanen of Nokia
845 * Telecommunications, page 6:
846 * "A node is doubled if the ratio of non-empty children to all
847 * children in the *doubled* node is at least 'high'."
849 * 'high' in this instance is the variable 'inflate_threshold'. It
850 * is expressed as a percentage, so we multiply it with
851 * child_length() and instead of multiplying by 2 (since the
852 * child array will be doubled by inflate()) and multiplying
853 * the left-hand side by 100 (to handle the percentage thing) we
854 * multiply the left-hand side by 50.
856 * The left-hand side may look a bit weird: child_length(tn)
857 * - tn->empty_children is of course the number of non-null children
858 * in the current node. tn->full_children is the number of "full"
859 * children, that is non-null tnodes with a skip value of 0.
860 * All of those will be doubled in the resulting inflated tnode, so
861 * we just count them one extra time here.
863 * A clearer way to write this would be:
865 * to_be_doubled = tn->full_children;
866 * not_to_be_doubled = child_length(tn) - tn->empty_children -
869 * new_child_length = child_length(tn) * 2;
871 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
873 * if (new_fill_factor >= inflate_threshold)
875 * ...and so on, tho it would mess up the while () loop.
878 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
882 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
883 * inflate_threshold * new_child_length
885 * expand not_to_be_doubled and to_be_doubled, and shorten:
886 * 100 * (child_length(tn) - tn->empty_children +
887 * tn->full_children) >= inflate_threshold * new_child_length
889 * expand new_child_length:
890 * 100 * (child_length(tn) - tn->empty_children +
891 * tn->full_children) >=
892 * inflate_threshold * child_length(tn) * 2
895 * 50 * (tn->full_children + child_length(tn) -
896 * tn->empty_children) >= inflate_threshold *
900 static inline bool should_inflate(struct key_vector
*tp
, struct key_vector
*tn
)
902 unsigned long used
= child_length(tn
);
903 unsigned long threshold
= used
;
905 /* Keep root node larger */
906 threshold
*= IS_TRIE(tp
) ? inflate_threshold_root
: inflate_threshold
;
907 used
-= tn_info(tn
)->empty_children
;
908 used
+= tn_info(tn
)->full_children
;
910 /* if bits == KEYLENGTH then pos = 0, and will fail below */
912 return (used
> 1) && tn
->pos
&& ((50 * used
) >= threshold
);
915 static inline bool should_halve(struct key_vector
*tp
, struct key_vector
*tn
)
917 unsigned long used
= child_length(tn
);
918 unsigned long threshold
= used
;
920 /* Keep root node larger */
921 threshold
*= IS_TRIE(tp
) ? halve_threshold_root
: halve_threshold
;
922 used
-= tn_info(tn
)->empty_children
;
924 /* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
926 return (used
> 1) && (tn
->bits
> 1) && ((100 * used
) < threshold
);
929 static inline bool should_collapse(struct key_vector
*tn
)
931 unsigned long used
= child_length(tn
);
933 used
-= tn_info(tn
)->empty_children
;
935 /* account for bits == KEYLENGTH case */
936 if ((tn
->bits
== KEYLENGTH
) && tn_info(tn
)->full_children
)
939 /* One child or none, time to drop us from the trie */
944 static struct key_vector
*resize(struct trie
*t
, struct key_vector
*tn
)
946 #ifdef CONFIG_IP_FIB_TRIE_STATS
947 struct trie_use_stats __percpu
*stats
= t
->stats
;
949 struct key_vector
*tp
= node_parent(tn
);
950 unsigned long cindex
= get_index(tn
->key
, tp
);
951 int max_work
= MAX_WORK
;
953 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
954 tn
, inflate_threshold
, halve_threshold
);
956 /* track the tnode via the pointer from the parent instead of
957 * doing it ourselves. This way we can let RCU fully do its
958 * thing without us interfering
960 BUG_ON(tn
!= get_child(tp
, cindex
));
962 /* Double as long as the resulting node has a number of
963 * nonempty nodes that are above the threshold.
965 while (should_inflate(tp
, tn
) && max_work
) {
968 #ifdef CONFIG_IP_FIB_TRIE_STATS
969 this_cpu_inc(stats
->resize_node_skipped
);
975 tn
= get_child(tp
, cindex
);
978 /* update parent in case inflate failed */
979 tp
= node_parent(tn
);
981 /* Return if at least one inflate is run */
982 if (max_work
!= MAX_WORK
)
985 /* Halve as long as the number of empty children in this
986 * node is above threshold.
988 while (should_halve(tp
, tn
) && max_work
) {
991 #ifdef CONFIG_IP_FIB_TRIE_STATS
992 this_cpu_inc(stats
->resize_node_skipped
);
998 tn
= get_child(tp
, cindex
);
1001 /* Only one child remains */
1002 if (should_collapse(tn
))
1003 return collapse(t
, tn
);
1005 /* update parent in case halve failed */
1006 return node_parent(tn
);
1009 static void node_pull_suffix(struct key_vector
*tn
, unsigned char slen
)
1011 unsigned char node_slen
= tn
->slen
;
1013 while ((node_slen
> tn
->pos
) && (node_slen
> slen
)) {
1014 slen
= update_suffix(tn
);
1015 if (node_slen
== slen
)
1018 tn
= node_parent(tn
);
1019 node_slen
= tn
->slen
;
1023 static void node_push_suffix(struct key_vector
*tn
, unsigned char slen
)
1025 while (tn
->slen
< slen
) {
1027 tn
= node_parent(tn
);
1031 /* rcu_read_lock needs to be hold by caller from readside */
1032 static struct key_vector
*fib_find_node(struct trie
*t
,
1033 struct key_vector
**tp
, u32 key
)
1035 struct key_vector
*pn
, *n
= t
->kv
;
1036 unsigned long index
= 0;
1040 n
= get_child_rcu(n
, index
);
1045 index
= get_cindex(key
, n
);
1047 /* This bit of code is a bit tricky but it combines multiple
1048 * checks into a single check. The prefix consists of the
1049 * prefix plus zeros for the bits in the cindex. The index
1050 * is the difference between the key and this value. From
1051 * this we can actually derive several pieces of data.
1052 * if (index >= (1ul << bits))
1053 * we have a mismatch in skip bits and failed
1055 * we know the value is cindex
1057 * This check is safe even if bits == KEYLENGTH due to the
1058 * fact that we can only allocate a node with 32 bits if a
1059 * long is greater than 32 bits.
1061 if (index
>= (1ul << n
->bits
)) {
1066 /* keep searching until we find a perfect match leaf or NULL */
1067 } while (IS_TNODE(n
));
1074 /* Return the first fib alias matching TOS with
1075 * priority less than or equal to PRIO.
1077 static struct fib_alias
*fib_find_alias(struct hlist_head
*fah
, u8 slen
,
1078 u8 tos
, u32 prio
, u32 tb_id
)
1080 struct fib_alias
*fa
;
1085 hlist_for_each_entry(fa
, fah
, fa_list
) {
1086 if (fa
->fa_slen
< slen
)
1088 if (fa
->fa_slen
!= slen
)
1090 if (fa
->tb_id
> tb_id
)
1092 if (fa
->tb_id
!= tb_id
)
1094 if (fa
->fa_tos
> tos
)
1096 if (fa
->fa_info
->fib_priority
>= prio
|| fa
->fa_tos
< tos
)
1103 static void trie_rebalance(struct trie
*t
, struct key_vector
*tn
)
1105 while (!IS_TRIE(tn
))
1109 static int fib_insert_node(struct trie
*t
, struct key_vector
*tp
,
1110 struct fib_alias
*new, t_key key
)
1112 struct key_vector
*n
, *l
;
1114 l
= leaf_new(key
, new);
1118 /* retrieve child from parent node */
1119 n
= get_child(tp
, get_index(key
, tp
));
1121 /* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1123 * Add a new tnode here
1124 * first tnode need some special handling
1125 * leaves us in position for handling as case 3
1128 struct key_vector
*tn
;
1130 tn
= tnode_new(key
, __fls(key
^ n
->key
), 1);
1134 /* initialize routes out of node */
1135 NODE_INIT_PARENT(tn
, tp
);
1136 put_child(tn
, get_index(key
, tn
) ^ 1, n
);
1138 /* start adding routes into the node */
1139 put_child_root(tp
, key
, tn
);
1140 node_set_parent(n
, tn
);
1142 /* parent now has a NULL spot where the leaf can go */
1146 /* Case 3: n is NULL, and will just insert a new leaf */
1147 node_push_suffix(tp
, new->fa_slen
);
1148 NODE_INIT_PARENT(l
, tp
);
1149 put_child_root(tp
, key
, l
);
1150 trie_rebalance(t
, tp
);
1159 static int fib_insert_alias(struct trie
*t
, struct key_vector
*tp
,
1160 struct key_vector
*l
, struct fib_alias
*new,
1161 struct fib_alias
*fa
, t_key key
)
1164 return fib_insert_node(t
, tp
, new, key
);
1167 hlist_add_before_rcu(&new->fa_list
, &fa
->fa_list
);
1169 struct fib_alias
*last
;
1171 hlist_for_each_entry(last
, &l
->leaf
, fa_list
) {
1172 if (new->fa_slen
< last
->fa_slen
)
1174 if ((new->fa_slen
== last
->fa_slen
) &&
1175 (new->tb_id
> last
->tb_id
))
1181 hlist_add_behind_rcu(&new->fa_list
, &fa
->fa_list
);
1183 hlist_add_head_rcu(&new->fa_list
, &l
->leaf
);
1186 /* if we added to the tail node then we need to update slen */
1187 if (l
->slen
< new->fa_slen
) {
1188 l
->slen
= new->fa_slen
;
1189 node_push_suffix(tp
, new->fa_slen
);
1195 /* Caller must hold RTNL. */
1196 int fib_table_insert(struct net
*net
, struct fib_table
*tb
,
1197 struct fib_config
*cfg
)
1199 enum fib_event_type event
= FIB_EVENT_ENTRY_ADD
;
1200 struct trie
*t
= (struct trie
*)tb
->tb_data
;
1201 struct fib_alias
*fa
, *new_fa
;
1202 struct key_vector
*l
, *tp
;
1203 u16 nlflags
= NLM_F_EXCL
;
1204 struct fib_info
*fi
;
1205 u8 plen
= cfg
->fc_dst_len
;
1206 u8 slen
= KEYLENGTH
- plen
;
1207 u8 tos
= cfg
->fc_tos
;
1211 if (plen
> KEYLENGTH
)
1214 key
= ntohl(cfg
->fc_dst
);
1216 pr_debug("Insert table=%u %08x/%d\n", tb
->tb_id
, key
, plen
);
1218 if ((plen
< KEYLENGTH
) && (key
<< plen
))
1221 fi
= fib_create_info(cfg
);
1227 l
= fib_find_node(t
, &tp
, key
);
1228 fa
= l
? fib_find_alias(&l
->leaf
, slen
, tos
, fi
->fib_priority
,
1231 /* Now fa, if non-NULL, points to the first fib alias
1232 * with the same keys [prefix,tos,priority], if such key already
1233 * exists or to the node before which we will insert new one.
1235 * If fa is NULL, we will need to allocate a new one and
1236 * insert to the tail of the section matching the suffix length
1240 if (fa
&& fa
->fa_tos
== tos
&&
1241 fa
->fa_info
->fib_priority
== fi
->fib_priority
) {
1242 struct fib_alias
*fa_first
, *fa_match
;
1245 if (cfg
->fc_nlflags
& NLM_F_EXCL
)
1248 nlflags
&= ~NLM_F_EXCL
;
1251 * 1. Find exact match for type, scope, fib_info to avoid
1253 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1257 hlist_for_each_entry_from(fa
, fa_list
) {
1258 if ((fa
->fa_slen
!= slen
) ||
1259 (fa
->tb_id
!= tb
->tb_id
) ||
1260 (fa
->fa_tos
!= tos
))
1262 if (fa
->fa_info
->fib_priority
!= fi
->fib_priority
)
1264 if (fa
->fa_type
== cfg
->fc_type
&&
1265 fa
->fa_info
== fi
) {
1271 if (cfg
->fc_nlflags
& NLM_F_REPLACE
) {
1272 struct fib_info
*fi_drop
;
1275 nlflags
|= NLM_F_REPLACE
;
1283 new_fa
= kmem_cache_alloc(fn_alias_kmem
, GFP_KERNEL
);
1287 fi_drop
= fa
->fa_info
;
1288 new_fa
->fa_tos
= fa
->fa_tos
;
1289 new_fa
->fa_info
= fi
;
1290 new_fa
->fa_type
= cfg
->fc_type
;
1291 state
= fa
->fa_state
;
1292 new_fa
->fa_state
= state
& ~FA_S_ACCESSED
;
1293 new_fa
->fa_slen
= fa
->fa_slen
;
1294 new_fa
->tb_id
= tb
->tb_id
;
1295 new_fa
->fa_default
= -1;
1297 call_fib_entry_notifiers(net
, FIB_EVENT_ENTRY_REPLACE
,
1299 new_fa
->fa_tos
, cfg
->fc_type
,
1301 rtmsg_fib(RTM_NEWROUTE
, htonl(key
), new_fa
, plen
,
1302 tb
->tb_id
, &cfg
->fc_nlinfo
, nlflags
);
1304 hlist_replace_rcu(&fa
->fa_list
, &new_fa
->fa_list
);
1306 alias_free_mem_rcu(fa
);
1308 fib_release_info(fi_drop
);
1309 if (state
& FA_S_ACCESSED
)
1310 rt_cache_flush(cfg
->fc_nlinfo
.nl_net
);
1314 /* Error if we find a perfect match which
1315 * uses the same scope, type, and nexthop
1321 if (cfg
->fc_nlflags
& NLM_F_APPEND
) {
1322 event
= FIB_EVENT_ENTRY_APPEND
;
1323 nlflags
|= NLM_F_APPEND
;
1329 if (!(cfg
->fc_nlflags
& NLM_F_CREATE
))
1332 nlflags
|= NLM_F_CREATE
;
1334 new_fa
= kmem_cache_alloc(fn_alias_kmem
, GFP_KERNEL
);
1338 new_fa
->fa_info
= fi
;
1339 new_fa
->fa_tos
= tos
;
1340 new_fa
->fa_type
= cfg
->fc_type
;
1341 new_fa
->fa_state
= 0;
1342 new_fa
->fa_slen
= slen
;
1343 new_fa
->tb_id
= tb
->tb_id
;
1344 new_fa
->fa_default
= -1;
1346 /* Insert new entry to the list. */
1347 err
= fib_insert_alias(t
, tp
, l
, new_fa
, fa
, key
);
1349 goto out_free_new_fa
;
1352 tb
->tb_num_default
++;
1354 rt_cache_flush(cfg
->fc_nlinfo
.nl_net
);
1355 call_fib_entry_notifiers(net
, event
, key
, plen
, fi
, tos
, cfg
->fc_type
,
1357 rtmsg_fib(RTM_NEWROUTE
, htonl(key
), new_fa
, plen
, new_fa
->tb_id
,
1358 &cfg
->fc_nlinfo
, nlflags
);
1363 kmem_cache_free(fn_alias_kmem
, new_fa
);
1365 fib_release_info(fi
);
1370 static inline t_key
prefix_mismatch(t_key key
, struct key_vector
*n
)
1372 t_key prefix
= n
->key
;
1374 return (key
^ prefix
) & (prefix
| -prefix
);
1377 /* should be called with rcu_read_lock */
1378 int fib_table_lookup(struct fib_table
*tb
, const struct flowi4
*flp
,
1379 struct fib_result
*res
, int fib_flags
)
1381 struct trie
*t
= (struct trie
*) tb
->tb_data
;
1382 #ifdef CONFIG_IP_FIB_TRIE_STATS
1383 struct trie_use_stats __percpu
*stats
= t
->stats
;
1385 const t_key key
= ntohl(flp
->daddr
);
1386 struct key_vector
*n
, *pn
;
1387 struct fib_alias
*fa
;
1388 unsigned long index
;
1391 trace_fib_table_lookup(tb
->tb_id
, flp
);
1396 n
= get_child_rcu(pn
, cindex
);
1400 #ifdef CONFIG_IP_FIB_TRIE_STATS
1401 this_cpu_inc(stats
->gets
);
1404 /* Step 1: Travel to the longest prefix match in the trie */
1406 index
= get_cindex(key
, n
);
1408 /* This bit of code is a bit tricky but it combines multiple
1409 * checks into a single check. The prefix consists of the
1410 * prefix plus zeros for the "bits" in the prefix. The index
1411 * is the difference between the key and this value. From
1412 * this we can actually derive several pieces of data.
1413 * if (index >= (1ul << bits))
1414 * we have a mismatch in skip bits and failed
1416 * we know the value is cindex
1418 * This check is safe even if bits == KEYLENGTH due to the
1419 * fact that we can only allocate a node with 32 bits if a
1420 * long is greater than 32 bits.
1422 if (index
>= (1ul << n
->bits
))
1425 /* we have found a leaf. Prefixes have already been compared */
1429 /* only record pn and cindex if we are going to be chopping
1430 * bits later. Otherwise we are just wasting cycles.
1432 if (n
->slen
> n
->pos
) {
1437 n
= get_child_rcu(n
, index
);
1442 /* Step 2: Sort out leaves and begin backtracing for longest prefix */
1444 /* record the pointer where our next node pointer is stored */
1445 struct key_vector __rcu
**cptr
= n
->tnode
;
1447 /* This test verifies that none of the bits that differ
1448 * between the key and the prefix exist in the region of
1449 * the lsb and higher in the prefix.
1451 if (unlikely(prefix_mismatch(key
, n
)) || (n
->slen
== n
->pos
))
1454 /* exit out and process leaf */
1455 if (unlikely(IS_LEAF(n
)))
1458 /* Don't bother recording parent info. Since we are in
1459 * prefix match mode we will have to come back to wherever
1460 * we started this traversal anyway
1463 while ((n
= rcu_dereference(*cptr
)) == NULL
) {
1465 #ifdef CONFIG_IP_FIB_TRIE_STATS
1467 this_cpu_inc(stats
->null_node_hit
);
1469 /* If we are at cindex 0 there are no more bits for
1470 * us to strip at this level so we must ascend back
1471 * up one level to see if there are any more bits to
1472 * be stripped there.
1475 t_key pkey
= pn
->key
;
1477 /* If we don't have a parent then there is
1478 * nothing for us to do as we do not have any
1479 * further nodes to parse.
1483 #ifdef CONFIG_IP_FIB_TRIE_STATS
1484 this_cpu_inc(stats
->backtrack
);
1486 /* Get Child's index */
1487 pn
= node_parent_rcu(pn
);
1488 cindex
= get_index(pkey
, pn
);
1491 /* strip the least significant bit from the cindex */
1492 cindex
&= cindex
- 1;
1494 /* grab pointer for next child node */
1495 cptr
= &pn
->tnode
[cindex
];
1500 /* this line carries forward the xor from earlier in the function */
1501 index
= key
^ n
->key
;
1503 /* Step 3: Process the leaf, if that fails fall back to backtracing */
1504 hlist_for_each_entry_rcu(fa
, &n
->leaf
, fa_list
) {
1505 struct fib_info
*fi
= fa
->fa_info
;
1508 if ((BITS_PER_LONG
> KEYLENGTH
) || (fa
->fa_slen
< KEYLENGTH
)) {
1509 if (index
>= (1ul << fa
->fa_slen
))
1512 if (fa
->fa_tos
&& fa
->fa_tos
!= flp
->flowi4_tos
)
1516 if (fa
->fa_info
->fib_scope
< flp
->flowi4_scope
)
1518 fib_alias_accessed(fa
);
1519 err
= fib_props
[fa
->fa_type
].error
;
1520 if (unlikely(err
< 0)) {
1521 #ifdef CONFIG_IP_FIB_TRIE_STATS
1522 this_cpu_inc(stats
->semantic_match_passed
);
1526 if (fi
->fib_flags
& RTNH_F_DEAD
)
1528 for (nhsel
= 0; nhsel
< fi
->fib_nhs
; nhsel
++) {
1529 const struct fib_nh
*nh
= &fi
->fib_nh
[nhsel
];
1530 struct in_device
*in_dev
= __in_dev_get_rcu(nh
->nh_dev
);
1532 if (nh
->nh_flags
& RTNH_F_DEAD
)
1535 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev
) &&
1536 nh
->nh_flags
& RTNH_F_LINKDOWN
&&
1537 !(fib_flags
& FIB_LOOKUP_IGNORE_LINKSTATE
))
1539 if (!(flp
->flowi4_flags
& FLOWI_FLAG_SKIP_NH_OIF
)) {
1540 if (flp
->flowi4_oif
&&
1541 flp
->flowi4_oif
!= nh
->nh_oif
)
1545 if (!(fib_flags
& FIB_LOOKUP_NOREF
))
1546 atomic_inc(&fi
->fib_clntref
);
1548 res
->prefixlen
= KEYLENGTH
- fa
->fa_slen
;
1549 res
->nh_sel
= nhsel
;
1550 res
->type
= fa
->fa_type
;
1551 res
->scope
= fi
->fib_scope
;
1554 res
->fa_head
= &n
->leaf
;
1555 #ifdef CONFIG_IP_FIB_TRIE_STATS
1556 this_cpu_inc(stats
->semantic_match_passed
);
1558 trace_fib_table_lookup_nh(nh
);
1563 #ifdef CONFIG_IP_FIB_TRIE_STATS
1564 this_cpu_inc(stats
->semantic_match_miss
);
1568 EXPORT_SYMBOL_GPL(fib_table_lookup
);
1570 static void fib_remove_alias(struct trie
*t
, struct key_vector
*tp
,
1571 struct key_vector
*l
, struct fib_alias
*old
)
1573 /* record the location of the previous list_info entry */
1574 struct hlist_node
**pprev
= old
->fa_list
.pprev
;
1575 struct fib_alias
*fa
= hlist_entry(pprev
, typeof(*fa
), fa_list
.next
);
1577 /* remove the fib_alias from the list */
1578 hlist_del_rcu(&old
->fa_list
);
1580 /* if we emptied the list this leaf will be freed and we can sort
1581 * out parent suffix lengths as a part of trie_rebalance
1583 if (hlist_empty(&l
->leaf
)) {
1584 if (tp
->slen
== l
->slen
)
1585 node_pull_suffix(tp
, tp
->pos
);
1586 put_child_root(tp
, l
->key
, NULL
);
1588 trie_rebalance(t
, tp
);
1592 /* only access fa if it is pointing at the last valid hlist_node */
1596 /* update the trie with the latest suffix length */
1597 l
->slen
= fa
->fa_slen
;
1598 node_pull_suffix(tp
, fa
->fa_slen
);
1601 /* Caller must hold RTNL. */
1602 int fib_table_delete(struct net
*net
, struct fib_table
*tb
,
1603 struct fib_config
*cfg
)
1605 struct trie
*t
= (struct trie
*) tb
->tb_data
;
1606 struct fib_alias
*fa
, *fa_to_delete
;
1607 struct key_vector
*l
, *tp
;
1608 u8 plen
= cfg
->fc_dst_len
;
1609 u8 slen
= KEYLENGTH
- plen
;
1610 u8 tos
= cfg
->fc_tos
;
1613 if (plen
> KEYLENGTH
)
1616 key
= ntohl(cfg
->fc_dst
);
1618 if ((plen
< KEYLENGTH
) && (key
<< plen
))
1621 l
= fib_find_node(t
, &tp
, key
);
1625 fa
= fib_find_alias(&l
->leaf
, slen
, tos
, 0, tb
->tb_id
);
1629 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key
, plen
, tos
, t
);
1631 fa_to_delete
= NULL
;
1632 hlist_for_each_entry_from(fa
, fa_list
) {
1633 struct fib_info
*fi
= fa
->fa_info
;
1635 if ((fa
->fa_slen
!= slen
) ||
1636 (fa
->tb_id
!= tb
->tb_id
) ||
1637 (fa
->fa_tos
!= tos
))
1640 if ((!cfg
->fc_type
|| fa
->fa_type
== cfg
->fc_type
) &&
1641 (cfg
->fc_scope
== RT_SCOPE_NOWHERE
||
1642 fa
->fa_info
->fib_scope
== cfg
->fc_scope
) &&
1643 (!cfg
->fc_prefsrc
||
1644 fi
->fib_prefsrc
== cfg
->fc_prefsrc
) &&
1645 (!cfg
->fc_protocol
||
1646 fi
->fib_protocol
== cfg
->fc_protocol
) &&
1647 fib_nh_match(cfg
, fi
) == 0) {
1656 call_fib_entry_notifiers(net
, FIB_EVENT_ENTRY_DEL
, key
, plen
,
1657 fa_to_delete
->fa_info
, tos
,
1658 fa_to_delete
->fa_type
, tb
->tb_id
);
1659 rtmsg_fib(RTM_DELROUTE
, htonl(key
), fa_to_delete
, plen
, tb
->tb_id
,
1660 &cfg
->fc_nlinfo
, 0);
1663 tb
->tb_num_default
--;
1665 fib_remove_alias(t
, tp
, l
, fa_to_delete
);
1667 if (fa_to_delete
->fa_state
& FA_S_ACCESSED
)
1668 rt_cache_flush(cfg
->fc_nlinfo
.nl_net
);
1670 fib_release_info(fa_to_delete
->fa_info
);
1671 alias_free_mem_rcu(fa_to_delete
);
1675 /* Scan for the next leaf starting at the provided key value */
1676 static struct key_vector
*leaf_walk_rcu(struct key_vector
**tn
, t_key key
)
1678 struct key_vector
*pn
, *n
= *tn
;
1679 unsigned long cindex
;
1681 /* this loop is meant to try and find the key in the trie */
1683 /* record parent and next child index */
1685 cindex
= (key
> pn
->key
) ? get_index(key
, pn
) : 0;
1687 if (cindex
>> pn
->bits
)
1690 /* descend into the next child */
1691 n
= get_child_rcu(pn
, cindex
++);
1695 /* guarantee forward progress on the keys */
1696 if (IS_LEAF(n
) && (n
->key
>= key
))
1698 } while (IS_TNODE(n
));
1700 /* this loop will search for the next leaf with a greater key */
1701 while (!IS_TRIE(pn
)) {
1702 /* if we exhausted the parent node we will need to climb */
1703 if (cindex
>= (1ul << pn
->bits
)) {
1704 t_key pkey
= pn
->key
;
1706 pn
= node_parent_rcu(pn
);
1707 cindex
= get_index(pkey
, pn
) + 1;
1711 /* grab the next available node */
1712 n
= get_child_rcu(pn
, cindex
++);
1716 /* no need to compare keys since we bumped the index */
1720 /* Rescan start scanning in new node */
1726 return NULL
; /* Root of trie */
1728 /* if we are at the limit for keys just return NULL for the tnode */
1733 static void fib_trie_free(struct fib_table
*tb
)
1735 struct trie
*t
= (struct trie
*)tb
->tb_data
;
1736 struct key_vector
*pn
= t
->kv
;
1737 unsigned long cindex
= 1;
1738 struct hlist_node
*tmp
;
1739 struct fib_alias
*fa
;
1741 /* walk trie in reverse order and free everything */
1743 struct key_vector
*n
;
1746 t_key pkey
= pn
->key
;
1752 pn
= node_parent(pn
);
1754 /* drop emptied tnode */
1755 put_child_root(pn
, n
->key
, NULL
);
1758 cindex
= get_index(pkey
, pn
);
1763 /* grab the next available node */
1764 n
= get_child(pn
, cindex
);
1769 /* record pn and cindex for leaf walking */
1771 cindex
= 1ul << n
->bits
;
1776 hlist_for_each_entry_safe(fa
, tmp
, &n
->leaf
, fa_list
) {
1777 hlist_del_rcu(&fa
->fa_list
);
1778 alias_free_mem_rcu(fa
);
1781 put_child_root(pn
, n
->key
, NULL
);
1785 #ifdef CONFIG_IP_FIB_TRIE_STATS
1786 free_percpu(t
->stats
);
1791 struct fib_table
*fib_trie_unmerge(struct fib_table
*oldtb
)
1793 struct trie
*ot
= (struct trie
*)oldtb
->tb_data
;
1794 struct key_vector
*l
, *tp
= ot
->kv
;
1795 struct fib_table
*local_tb
;
1796 struct fib_alias
*fa
;
1800 if (oldtb
->tb_data
== oldtb
->__data
)
1803 local_tb
= fib_trie_table(RT_TABLE_LOCAL
, NULL
);
1807 lt
= (struct trie
*)local_tb
->tb_data
;
1809 while ((l
= leaf_walk_rcu(&tp
, key
)) != NULL
) {
1810 struct key_vector
*local_l
= NULL
, *local_tp
;
1812 hlist_for_each_entry_rcu(fa
, &l
->leaf
, fa_list
) {
1813 struct fib_alias
*new_fa
;
1815 if (local_tb
->tb_id
!= fa
->tb_id
)
1818 /* clone fa for new local table */
1819 new_fa
= kmem_cache_alloc(fn_alias_kmem
, GFP_KERNEL
);
1823 memcpy(new_fa
, fa
, sizeof(*fa
));
1825 /* insert clone into table */
1827 local_l
= fib_find_node(lt
, &local_tp
, l
->key
);
1829 if (fib_insert_alias(lt
, local_tp
, local_l
, new_fa
,
1831 kmem_cache_free(fn_alias_kmem
, new_fa
);
1836 /* stop loop if key wrapped back to 0 */
1844 fib_trie_free(local_tb
);
1849 /* Caller must hold RTNL */
1850 void fib_table_flush_external(struct fib_table
*tb
)
1852 struct trie
*t
= (struct trie
*)tb
->tb_data
;
1853 struct key_vector
*pn
= t
->kv
;
1854 unsigned long cindex
= 1;
1855 struct hlist_node
*tmp
;
1856 struct fib_alias
*fa
;
1858 /* walk trie in reverse order */
1860 unsigned char slen
= 0;
1861 struct key_vector
*n
;
1864 t_key pkey
= pn
->key
;
1866 /* cannot resize the trie vector */
1870 /* update the suffix to address pulled leaves */
1871 if (pn
->slen
> pn
->pos
)
1874 /* resize completed node */
1876 cindex
= get_index(pkey
, pn
);
1881 /* grab the next available node */
1882 n
= get_child(pn
, cindex
);
1887 /* record pn and cindex for leaf walking */
1889 cindex
= 1ul << n
->bits
;
1894 hlist_for_each_entry_safe(fa
, tmp
, &n
->leaf
, fa_list
) {
1895 /* if alias was cloned to local then we just
1896 * need to remove the local copy from main
1898 if (tb
->tb_id
!= fa
->tb_id
) {
1899 hlist_del_rcu(&fa
->fa_list
);
1900 alias_free_mem_rcu(fa
);
1904 /* record local slen */
1908 /* update leaf slen */
1911 if (hlist_empty(&n
->leaf
)) {
1912 put_child_root(pn
, n
->key
, NULL
);
1918 /* Caller must hold RTNL. */
1919 int fib_table_flush(struct net
*net
, struct fib_table
*tb
)
1921 struct trie
*t
= (struct trie
*)tb
->tb_data
;
1922 struct key_vector
*pn
= t
->kv
;
1923 unsigned long cindex
= 1;
1924 struct hlist_node
*tmp
;
1925 struct fib_alias
*fa
;
1928 /* walk trie in reverse order */
1930 unsigned char slen
= 0;
1931 struct key_vector
*n
;
1934 t_key pkey
= pn
->key
;
1936 /* cannot resize the trie vector */
1940 /* update the suffix to address pulled leaves */
1941 if (pn
->slen
> pn
->pos
)
1944 /* resize completed node */
1946 cindex
= get_index(pkey
, pn
);
1951 /* grab the next available node */
1952 n
= get_child(pn
, cindex
);
1957 /* record pn and cindex for leaf walking */
1959 cindex
= 1ul << n
->bits
;
1964 hlist_for_each_entry_safe(fa
, tmp
, &n
->leaf
, fa_list
) {
1965 struct fib_info
*fi
= fa
->fa_info
;
1967 if (!fi
|| !(fi
->fib_flags
& RTNH_F_DEAD
) ||
1968 tb
->tb_id
!= fa
->tb_id
) {
1973 call_fib_entry_notifiers(net
, FIB_EVENT_ENTRY_DEL
,
1975 KEYLENGTH
- fa
->fa_slen
,
1976 fi
, fa
->fa_tos
, fa
->fa_type
,
1978 hlist_del_rcu(&fa
->fa_list
);
1979 fib_release_info(fa
->fa_info
);
1980 alias_free_mem_rcu(fa
);
1984 /* update leaf slen */
1987 if (hlist_empty(&n
->leaf
)) {
1988 put_child_root(pn
, n
->key
, NULL
);
1993 pr_debug("trie_flush found=%d\n", found
);
1997 static void fib_leaf_notify(struct net
*net
, struct key_vector
*l
,
1998 struct fib_table
*tb
, struct notifier_block
*nb
,
1999 enum fib_event_type event_type
)
2001 struct fib_alias
*fa
;
2003 hlist_for_each_entry_rcu(fa
, &l
->leaf
, fa_list
) {
2004 struct fib_info
*fi
= fa
->fa_info
;
2009 /* local and main table can share the same trie,
2010 * so don't notify twice for the same entry.
2012 if (tb
->tb_id
!= fa
->tb_id
)
2015 call_fib_entry_notifier(nb
, net
, event_type
, l
->key
,
2016 KEYLENGTH
- fa
->fa_slen
, fi
, fa
->fa_tos
,
2017 fa
->fa_type
, fa
->tb_id
);
2021 static void fib_table_notify(struct net
*net
, struct fib_table
*tb
,
2022 struct notifier_block
*nb
,
2023 enum fib_event_type event_type
)
2025 struct trie
*t
= (struct trie
*)tb
->tb_data
;
2026 struct key_vector
*l
, *tp
= t
->kv
;
2029 while ((l
= leaf_walk_rcu(&tp
, key
)) != NULL
) {
2030 fib_leaf_notify(net
, l
, tb
, nb
, event_type
);
2033 /* stop in case of wrap around */
2039 static void fib_notify(struct net
*net
, struct notifier_block
*nb
,
2040 enum fib_event_type event_type
)
2044 for (h
= 0; h
< FIB_TABLE_HASHSZ
; h
++) {
2045 struct hlist_head
*head
= &net
->ipv4
.fib_table_hash
[h
];
2046 struct fib_table
*tb
;
2048 hlist_for_each_entry_rcu(tb
, head
, tb_hlist
)
2049 fib_table_notify(net
, tb
, nb
, event_type
);
2053 static void __trie_free_rcu(struct rcu_head
*head
)
2055 struct fib_table
*tb
= container_of(head
, struct fib_table
, rcu
);
2056 #ifdef CONFIG_IP_FIB_TRIE_STATS
2057 struct trie
*t
= (struct trie
*)tb
->tb_data
;
2059 if (tb
->tb_data
== tb
->__data
)
2060 free_percpu(t
->stats
);
2061 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2065 void fib_free_table(struct fib_table
*tb
)
2067 call_rcu(&tb
->rcu
, __trie_free_rcu
);
2070 static int fn_trie_dump_leaf(struct key_vector
*l
, struct fib_table
*tb
,
2071 struct sk_buff
*skb
, struct netlink_callback
*cb
)
2073 __be32 xkey
= htonl(l
->key
);
2074 struct fib_alias
*fa
;
2080 /* rcu_read_lock is hold by caller */
2081 hlist_for_each_entry_rcu(fa
, &l
->leaf
, fa_list
) {
2087 if (tb
->tb_id
!= fa
->tb_id
) {
2092 if (fib_dump_info(skb
, NETLINK_CB(cb
->skb
).portid
,
2098 KEYLENGTH
- fa
->fa_slen
,
2100 fa
->fa_info
, NLM_F_MULTI
) < 0) {
2111 /* rcu_read_lock needs to be hold by caller from readside */
2112 int fib_table_dump(struct fib_table
*tb
, struct sk_buff
*skb
,
2113 struct netlink_callback
*cb
)
2115 struct trie
*t
= (struct trie
*)tb
->tb_data
;
2116 struct key_vector
*l
, *tp
= t
->kv
;
2117 /* Dump starting at last key.
2118 * Note: 0.0.0.0/0 (ie default) is first key.
2120 int count
= cb
->args
[2];
2121 t_key key
= cb
->args
[3];
2123 while ((l
= leaf_walk_rcu(&tp
, key
)) != NULL
) {
2124 if (fn_trie_dump_leaf(l
, tb
, skb
, cb
) < 0) {
2126 cb
->args
[2] = count
;
2133 memset(&cb
->args
[4], 0,
2134 sizeof(cb
->args
) - 4*sizeof(cb
->args
[0]));
2136 /* stop loop if key wrapped back to 0 */
2142 cb
->args
[2] = count
;
2147 void __init
fib_trie_init(void)
2149 fn_alias_kmem
= kmem_cache_create("ip_fib_alias",
2150 sizeof(struct fib_alias
),
2151 0, SLAB_PANIC
, NULL
);
2153 trie_leaf_kmem
= kmem_cache_create("ip_fib_trie",
2155 0, SLAB_PANIC
, NULL
);
2158 struct fib_table
*fib_trie_table(u32 id
, struct fib_table
*alias
)
2160 struct fib_table
*tb
;
2162 size_t sz
= sizeof(*tb
);
2165 sz
+= sizeof(struct trie
);
2167 tb
= kzalloc(sz
, GFP_KERNEL
);
2172 tb
->tb_num_default
= 0;
2173 tb
->tb_data
= (alias
? alias
->__data
: tb
->__data
);
2178 t
= (struct trie
*) tb
->tb_data
;
2179 t
->kv
[0].pos
= KEYLENGTH
;
2180 t
->kv
[0].slen
= KEYLENGTH
;
2181 #ifdef CONFIG_IP_FIB_TRIE_STATS
2182 t
->stats
= alloc_percpu(struct trie_use_stats
);
2192 #ifdef CONFIG_PROC_FS
2193 /* Depth first Trie walk iterator */
2194 struct fib_trie_iter
{
2195 struct seq_net_private p
;
2196 struct fib_table
*tb
;
2197 struct key_vector
*tnode
;
2202 static struct key_vector
*fib_trie_get_next(struct fib_trie_iter
*iter
)
2204 unsigned long cindex
= iter
->index
;
2205 struct key_vector
*pn
= iter
->tnode
;
2208 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2209 iter
->tnode
, iter
->index
, iter
->depth
);
2211 while (!IS_TRIE(pn
)) {
2212 while (cindex
< child_length(pn
)) {
2213 struct key_vector
*n
= get_child_rcu(pn
, cindex
++);
2220 iter
->index
= cindex
;
2222 /* push down one level */
2231 /* Current node exhausted, pop back up */
2233 pn
= node_parent_rcu(pn
);
2234 cindex
= get_index(pkey
, pn
) + 1;
2238 /* record root node so further searches know we are done */
2245 static struct key_vector
*fib_trie_get_first(struct fib_trie_iter
*iter
,
2248 struct key_vector
*n
, *pn
;
2254 n
= rcu_dereference(pn
->tnode
[0]);
2271 static void trie_collect_stats(struct trie
*t
, struct trie_stat
*s
)
2273 struct key_vector
*n
;
2274 struct fib_trie_iter iter
;
2276 memset(s
, 0, sizeof(*s
));
2279 for (n
= fib_trie_get_first(&iter
, t
); n
; n
= fib_trie_get_next(&iter
)) {
2281 struct fib_alias
*fa
;
2284 s
->totdepth
+= iter
.depth
;
2285 if (iter
.depth
> s
->maxdepth
)
2286 s
->maxdepth
= iter
.depth
;
2288 hlist_for_each_entry_rcu(fa
, &n
->leaf
, fa_list
)
2292 if (n
->bits
< MAX_STAT_DEPTH
)
2293 s
->nodesizes
[n
->bits
]++;
2294 s
->nullpointers
+= tn_info(n
)->empty_children
;
2301 * This outputs /proc/net/fib_triestats
2303 static void trie_show_stats(struct seq_file
*seq
, struct trie_stat
*stat
)
2305 unsigned int i
, max
, pointers
, bytes
, avdepth
;
2308 avdepth
= stat
->totdepth
*100 / stat
->leaves
;
2312 seq_printf(seq
, "\tAver depth: %u.%02d\n",
2313 avdepth
/ 100, avdepth
% 100);
2314 seq_printf(seq
, "\tMax depth: %u\n", stat
->maxdepth
);
2316 seq_printf(seq
, "\tLeaves: %u\n", stat
->leaves
);
2317 bytes
= LEAF_SIZE
* stat
->leaves
;
2319 seq_printf(seq
, "\tPrefixes: %u\n", stat
->prefixes
);
2320 bytes
+= sizeof(struct fib_alias
) * stat
->prefixes
;
2322 seq_printf(seq
, "\tInternal nodes: %u\n\t", stat
->tnodes
);
2323 bytes
+= TNODE_SIZE(0) * stat
->tnodes
;
2325 max
= MAX_STAT_DEPTH
;
2326 while (max
> 0 && stat
->nodesizes
[max
-1] == 0)
2330 for (i
= 1; i
< max
; i
++)
2331 if (stat
->nodesizes
[i
] != 0) {
2332 seq_printf(seq
, " %u: %u", i
, stat
->nodesizes
[i
]);
2333 pointers
+= (1<<i
) * stat
->nodesizes
[i
];
2335 seq_putc(seq
, '\n');
2336 seq_printf(seq
, "\tPointers: %u\n", pointers
);
2338 bytes
+= sizeof(struct key_vector
*) * pointers
;
2339 seq_printf(seq
, "Null ptrs: %u\n", stat
->nullpointers
);
2340 seq_printf(seq
, "Total size: %u kB\n", (bytes
+ 1023) / 1024);
2343 #ifdef CONFIG_IP_FIB_TRIE_STATS
2344 static void trie_show_usage(struct seq_file
*seq
,
2345 const struct trie_use_stats __percpu
*stats
)
2347 struct trie_use_stats s
= { 0 };
2350 /* loop through all of the CPUs and gather up the stats */
2351 for_each_possible_cpu(cpu
) {
2352 const struct trie_use_stats
*pcpu
= per_cpu_ptr(stats
, cpu
);
2354 s
.gets
+= pcpu
->gets
;
2355 s
.backtrack
+= pcpu
->backtrack
;
2356 s
.semantic_match_passed
+= pcpu
->semantic_match_passed
;
2357 s
.semantic_match_miss
+= pcpu
->semantic_match_miss
;
2358 s
.null_node_hit
+= pcpu
->null_node_hit
;
2359 s
.resize_node_skipped
+= pcpu
->resize_node_skipped
;
2362 seq_printf(seq
, "\nCounters:\n---------\n");
2363 seq_printf(seq
, "gets = %u\n", s
.gets
);
2364 seq_printf(seq
, "backtracks = %u\n", s
.backtrack
);
2365 seq_printf(seq
, "semantic match passed = %u\n",
2366 s
.semantic_match_passed
);
2367 seq_printf(seq
, "semantic match miss = %u\n", s
.semantic_match_miss
);
2368 seq_printf(seq
, "null node hit= %u\n", s
.null_node_hit
);
2369 seq_printf(seq
, "skipped node resize = %u\n\n", s
.resize_node_skipped
);
2371 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2373 static void fib_table_print(struct seq_file
*seq
, struct fib_table
*tb
)
2375 if (tb
->tb_id
== RT_TABLE_LOCAL
)
2376 seq_puts(seq
, "Local:\n");
2377 else if (tb
->tb_id
== RT_TABLE_MAIN
)
2378 seq_puts(seq
, "Main:\n");
2380 seq_printf(seq
, "Id %d:\n", tb
->tb_id
);
2384 static int fib_triestat_seq_show(struct seq_file
*seq
, void *v
)
2386 struct net
*net
= (struct net
*)seq
->private;
2390 "Basic info: size of leaf:"
2391 " %zd bytes, size of tnode: %zd bytes.\n",
2392 LEAF_SIZE
, TNODE_SIZE(0));
2394 for (h
= 0; h
< FIB_TABLE_HASHSZ
; h
++) {
2395 struct hlist_head
*head
= &net
->ipv4
.fib_table_hash
[h
];
2396 struct fib_table
*tb
;
2398 hlist_for_each_entry_rcu(tb
, head
, tb_hlist
) {
2399 struct trie
*t
= (struct trie
*) tb
->tb_data
;
2400 struct trie_stat stat
;
2405 fib_table_print(seq
, tb
);
2407 trie_collect_stats(t
, &stat
);
2408 trie_show_stats(seq
, &stat
);
2409 #ifdef CONFIG_IP_FIB_TRIE_STATS
2410 trie_show_usage(seq
, t
->stats
);
2418 static int fib_triestat_seq_open(struct inode
*inode
, struct file
*file
)
2420 return single_open_net(inode
, file
, fib_triestat_seq_show
);
2423 static const struct file_operations fib_triestat_fops
= {
2424 .owner
= THIS_MODULE
,
2425 .open
= fib_triestat_seq_open
,
2427 .llseek
= seq_lseek
,
2428 .release
= single_release_net
,
2431 static struct key_vector
*fib_trie_get_idx(struct seq_file
*seq
, loff_t pos
)
2433 struct fib_trie_iter
*iter
= seq
->private;
2434 struct net
*net
= seq_file_net(seq
);
2438 for (h
= 0; h
< FIB_TABLE_HASHSZ
; h
++) {
2439 struct hlist_head
*head
= &net
->ipv4
.fib_table_hash
[h
];
2440 struct fib_table
*tb
;
2442 hlist_for_each_entry_rcu(tb
, head
, tb_hlist
) {
2443 struct key_vector
*n
;
2445 for (n
= fib_trie_get_first(iter
,
2446 (struct trie
*) tb
->tb_data
);
2447 n
; n
= fib_trie_get_next(iter
))
2458 static void *fib_trie_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2462 return fib_trie_get_idx(seq
, *pos
);
2465 static void *fib_trie_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2467 struct fib_trie_iter
*iter
= seq
->private;
2468 struct net
*net
= seq_file_net(seq
);
2469 struct fib_table
*tb
= iter
->tb
;
2470 struct hlist_node
*tb_node
;
2472 struct key_vector
*n
;
2475 /* next node in same table */
2476 n
= fib_trie_get_next(iter
);
2480 /* walk rest of this hash chain */
2481 h
= tb
->tb_id
& (FIB_TABLE_HASHSZ
- 1);
2482 while ((tb_node
= rcu_dereference(hlist_next_rcu(&tb
->tb_hlist
)))) {
2483 tb
= hlist_entry(tb_node
, struct fib_table
, tb_hlist
);
2484 n
= fib_trie_get_first(iter
, (struct trie
*) tb
->tb_data
);
2489 /* new hash chain */
2490 while (++h
< FIB_TABLE_HASHSZ
) {
2491 struct hlist_head
*head
= &net
->ipv4
.fib_table_hash
[h
];
2492 hlist_for_each_entry_rcu(tb
, head
, tb_hlist
) {
2493 n
= fib_trie_get_first(iter
, (struct trie
*) tb
->tb_data
);
2505 static void fib_trie_seq_stop(struct seq_file
*seq
, void *v
)
2511 static void seq_indent(struct seq_file
*seq
, int n
)
2517 static inline const char *rtn_scope(char *buf
, size_t len
, enum rt_scope_t s
)
2520 case RT_SCOPE_UNIVERSE
: return "universe";
2521 case RT_SCOPE_SITE
: return "site";
2522 case RT_SCOPE_LINK
: return "link";
2523 case RT_SCOPE_HOST
: return "host";
2524 case RT_SCOPE_NOWHERE
: return "nowhere";
2526 snprintf(buf
, len
, "scope=%d", s
);
2531 static const char *const rtn_type_names
[__RTN_MAX
] = {
2532 [RTN_UNSPEC
] = "UNSPEC",
2533 [RTN_UNICAST
] = "UNICAST",
2534 [RTN_LOCAL
] = "LOCAL",
2535 [RTN_BROADCAST
] = "BROADCAST",
2536 [RTN_ANYCAST
] = "ANYCAST",
2537 [RTN_MULTICAST
] = "MULTICAST",
2538 [RTN_BLACKHOLE
] = "BLACKHOLE",
2539 [RTN_UNREACHABLE
] = "UNREACHABLE",
2540 [RTN_PROHIBIT
] = "PROHIBIT",
2541 [RTN_THROW
] = "THROW",
2543 [RTN_XRESOLVE
] = "XRESOLVE",
2546 static inline const char *rtn_type(char *buf
, size_t len
, unsigned int t
)
2548 if (t
< __RTN_MAX
&& rtn_type_names
[t
])
2549 return rtn_type_names
[t
];
2550 snprintf(buf
, len
, "type %u", t
);
2554 /* Pretty print the trie */
2555 static int fib_trie_seq_show(struct seq_file
*seq
, void *v
)
2557 const struct fib_trie_iter
*iter
= seq
->private;
2558 struct key_vector
*n
= v
;
2560 if (IS_TRIE(node_parent_rcu(n
)))
2561 fib_table_print(seq
, iter
->tb
);
2564 __be32 prf
= htonl(n
->key
);
2566 seq_indent(seq
, iter
->depth
-1);
2567 seq_printf(seq
, " +-- %pI4/%zu %u %u %u\n",
2568 &prf
, KEYLENGTH
- n
->pos
- n
->bits
, n
->bits
,
2569 tn_info(n
)->full_children
,
2570 tn_info(n
)->empty_children
);
2572 __be32 val
= htonl(n
->key
);
2573 struct fib_alias
*fa
;
2575 seq_indent(seq
, iter
->depth
);
2576 seq_printf(seq
, " |-- %pI4\n", &val
);
2578 hlist_for_each_entry_rcu(fa
, &n
->leaf
, fa_list
) {
2579 char buf1
[32], buf2
[32];
2581 seq_indent(seq
, iter
->depth
+ 1);
2582 seq_printf(seq
, " /%zu %s %s",
2583 KEYLENGTH
- fa
->fa_slen
,
2584 rtn_scope(buf1
, sizeof(buf1
),
2585 fa
->fa_info
->fib_scope
),
2586 rtn_type(buf2
, sizeof(buf2
),
2589 seq_printf(seq
, " tos=%d", fa
->fa_tos
);
2590 seq_putc(seq
, '\n');
2597 static const struct seq_operations fib_trie_seq_ops
= {
2598 .start
= fib_trie_seq_start
,
2599 .next
= fib_trie_seq_next
,
2600 .stop
= fib_trie_seq_stop
,
2601 .show
= fib_trie_seq_show
,
2604 static int fib_trie_seq_open(struct inode
*inode
, struct file
*file
)
2606 return seq_open_net(inode
, file
, &fib_trie_seq_ops
,
2607 sizeof(struct fib_trie_iter
));
2610 static const struct file_operations fib_trie_fops
= {
2611 .owner
= THIS_MODULE
,
2612 .open
= fib_trie_seq_open
,
2614 .llseek
= seq_lseek
,
2615 .release
= seq_release_net
,
2618 struct fib_route_iter
{
2619 struct seq_net_private p
;
2620 struct fib_table
*main_tb
;
2621 struct key_vector
*tnode
;
2626 static struct key_vector
*fib_route_get_idx(struct fib_route_iter
*iter
,
2629 struct key_vector
*l
, **tp
= &iter
->tnode
;
2632 /* use cached location of previously found key */
2633 if (iter
->pos
> 0 && pos
>= iter
->pos
) {
2642 while ((l
= leaf_walk_rcu(tp
, key
)) && (pos
-- > 0)) {
2647 /* handle unlikely case of a key wrap */
2653 iter
->key
= l
->key
; /* remember it */
2655 iter
->pos
= 0; /* forget it */
2660 static void *fib_route_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2663 struct fib_route_iter
*iter
= seq
->private;
2664 struct fib_table
*tb
;
2669 tb
= fib_get_table(seq_file_net(seq
), RT_TABLE_MAIN
);
2674 t
= (struct trie
*)tb
->tb_data
;
2675 iter
->tnode
= t
->kv
;
2678 return fib_route_get_idx(iter
, *pos
);
2681 iter
->key
= KEY_MAX
;
2683 return SEQ_START_TOKEN
;
2686 static void *fib_route_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2688 struct fib_route_iter
*iter
= seq
->private;
2689 struct key_vector
*l
= NULL
;
2690 t_key key
= iter
->key
+ 1;
2694 /* only allow key of 0 for start of sequence */
2695 if ((v
== SEQ_START_TOKEN
) || key
)
2696 l
= leaf_walk_rcu(&iter
->tnode
, key
);
2708 static void fib_route_seq_stop(struct seq_file
*seq
, void *v
)
2714 static unsigned int fib_flag_trans(int type
, __be32 mask
, const struct fib_info
*fi
)
2716 unsigned int flags
= 0;
2718 if (type
== RTN_UNREACHABLE
|| type
== RTN_PROHIBIT
)
2720 if (fi
&& fi
->fib_nh
->nh_gw
)
2721 flags
|= RTF_GATEWAY
;
2722 if (mask
== htonl(0xFFFFFFFF))
2729 * This outputs /proc/net/route.
2730 * The format of the file is not supposed to be changed
2731 * and needs to be same as fib_hash output to avoid breaking
2734 static int fib_route_seq_show(struct seq_file
*seq
, void *v
)
2736 struct fib_route_iter
*iter
= seq
->private;
2737 struct fib_table
*tb
= iter
->main_tb
;
2738 struct fib_alias
*fa
;
2739 struct key_vector
*l
= v
;
2742 if (v
== SEQ_START_TOKEN
) {
2743 seq_printf(seq
, "%-127s\n", "Iface\tDestination\tGateway "
2744 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2749 prefix
= htonl(l
->key
);
2751 hlist_for_each_entry_rcu(fa
, &l
->leaf
, fa_list
) {
2752 const struct fib_info
*fi
= fa
->fa_info
;
2753 __be32 mask
= inet_make_mask(KEYLENGTH
- fa
->fa_slen
);
2754 unsigned int flags
= fib_flag_trans(fa
->fa_type
, mask
, fi
);
2756 if ((fa
->fa_type
== RTN_BROADCAST
) ||
2757 (fa
->fa_type
== RTN_MULTICAST
))
2760 if (fa
->tb_id
!= tb
->tb_id
)
2763 seq_setwidth(seq
, 127);
2767 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2768 "%d\t%08X\t%d\t%u\t%u",
2769 fi
->fib_dev
? fi
->fib_dev
->name
: "*",
2771 fi
->fib_nh
->nh_gw
, flags
, 0, 0,
2775 fi
->fib_advmss
+ 40 : 0),
2780 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2781 "%d\t%08X\t%d\t%u\t%u",
2782 prefix
, 0, flags
, 0, 0, 0,
2791 static const struct seq_operations fib_route_seq_ops
= {
2792 .start
= fib_route_seq_start
,
2793 .next
= fib_route_seq_next
,
2794 .stop
= fib_route_seq_stop
,
2795 .show
= fib_route_seq_show
,
2798 static int fib_route_seq_open(struct inode
*inode
, struct file
*file
)
2800 return seq_open_net(inode
, file
, &fib_route_seq_ops
,
2801 sizeof(struct fib_route_iter
));
2804 static const struct file_operations fib_route_fops
= {
2805 .owner
= THIS_MODULE
,
2806 .open
= fib_route_seq_open
,
2808 .llseek
= seq_lseek
,
2809 .release
= seq_release_net
,
2812 int __net_init
fib_proc_init(struct net
*net
)
2814 if (!proc_create("fib_trie", S_IRUGO
, net
->proc_net
, &fib_trie_fops
))
2817 if (!proc_create("fib_triestat", S_IRUGO
, net
->proc_net
,
2818 &fib_triestat_fops
))
2821 if (!proc_create("route", S_IRUGO
, net
->proc_net
, &fib_route_fops
))
2827 remove_proc_entry("fib_triestat", net
->proc_net
);
2829 remove_proc_entry("fib_trie", net
->proc_net
);
2834 void __net_exit
fib_proc_exit(struct net
*net
)
2836 remove_proc_entry("fib_trie", net
->proc_net
);
2837 remove_proc_entry("fib_triestat", net
->proc_net
);
2838 remove_proc_entry("route", net
->proc_net
);
2841 #endif /* CONFIG_PROC_FS */