ASoC: rt5670: move set_sysclk to codec level
[linux-2.6/btrfs-unstable.git] / net / ipv4 / fib_trie.c
blob1201409ba1dcb18ee028003b065410b87bf4a602
1 /*
2 * This program is free software; you can redistribute it and/or
3 * modify it under the terms of the GNU General Public License
4 * as published by the Free Software Foundation; either version
5 * 2 of the License, or (at your option) any later version.
7 * Robert Olsson <robert.olsson@its.uu.se> Uppsala Universitet
8 * & Swedish University of Agricultural Sciences.
10 * Jens Laas <jens.laas@data.slu.se> Swedish University of
11 * Agricultural Sciences.
13 * Hans Liss <hans.liss@its.uu.se> Uppsala Universitet
15 * This work is based on the LPC-trie which is originally 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>
57 #include <linux/mm.h>
58 #include <linux/string.h>
59 #include <linux/socket.h>
60 #include <linux/sockios.h>
61 #include <linux/errno.h>
62 #include <linux/in.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>
78 #include <net/ip.h>
79 #include <net/protocol.h>
80 #include <net/route.h>
81 #include <net/tcp.h>
82 #include <net/sock.h>
83 #include <net/ip_fib.h>
84 #include <trace/events/fib.h>
85 #include "fib_lookup.h"
87 static int call_fib_entry_notifier(struct notifier_block *nb, struct net *net,
88 enum fib_event_type event_type, u32 dst,
89 int dst_len, struct fib_info *fi,
90 u8 tos, u8 type, u32 tb_id)
92 struct fib_entry_notifier_info info = {
93 .dst = dst,
94 .dst_len = dst_len,
95 .fi = fi,
96 .tos = tos,
97 .type = type,
98 .tb_id = tb_id,
100 return call_fib_notifier(nb, net, event_type, &info.info);
103 static int call_fib_entry_notifiers(struct net *net,
104 enum fib_event_type event_type, u32 dst,
105 int dst_len, struct fib_info *fi,
106 u8 tos, u8 type, u32 tb_id)
108 struct fib_entry_notifier_info info = {
109 .dst = dst,
110 .dst_len = dst_len,
111 .fi = fi,
112 .tos = tos,
113 .type = type,
114 .tb_id = tb_id,
116 return call_fib_notifiers(net, event_type, &info.info);
119 #define MAX_STAT_DEPTH 32
121 #define KEYLENGTH (8*sizeof(t_key))
122 #define KEY_MAX ((t_key)~0)
124 typedef unsigned int t_key;
126 #define IS_TRIE(n) ((n)->pos >= KEYLENGTH)
127 #define IS_TNODE(n) ((n)->bits)
128 #define IS_LEAF(n) (!(n)->bits)
130 struct key_vector {
131 t_key key;
132 unsigned char pos; /* 2log(KEYLENGTH) bits needed */
133 unsigned char bits; /* 2log(KEYLENGTH) bits needed */
134 unsigned char slen;
135 union {
136 /* This list pointer if valid if (pos | bits) == 0 (LEAF) */
137 struct hlist_head leaf;
138 /* This array is valid if (pos | bits) > 0 (TNODE) */
139 struct key_vector __rcu *tnode[0];
143 struct tnode {
144 struct rcu_head rcu;
145 t_key empty_children; /* KEYLENGTH bits needed */
146 t_key full_children; /* KEYLENGTH bits needed */
147 struct key_vector __rcu *parent;
148 struct key_vector kv[1];
149 #define tn_bits kv[0].bits
152 #define TNODE_SIZE(n) offsetof(struct tnode, kv[0].tnode[n])
153 #define LEAF_SIZE TNODE_SIZE(1)
155 #ifdef CONFIG_IP_FIB_TRIE_STATS
156 struct trie_use_stats {
157 unsigned int gets;
158 unsigned int backtrack;
159 unsigned int semantic_match_passed;
160 unsigned int semantic_match_miss;
161 unsigned int null_node_hit;
162 unsigned int resize_node_skipped;
164 #endif
166 struct trie_stat {
167 unsigned int totdepth;
168 unsigned int maxdepth;
169 unsigned int tnodes;
170 unsigned int leaves;
171 unsigned int nullpointers;
172 unsigned int prefixes;
173 unsigned int nodesizes[MAX_STAT_DEPTH];
176 struct trie {
177 struct key_vector kv[1];
178 #ifdef CONFIG_IP_FIB_TRIE_STATS
179 struct trie_use_stats __percpu *stats;
180 #endif
183 static struct key_vector *resize(struct trie *t, struct key_vector *tn);
184 static size_t tnode_free_size;
187 * synchronize_rcu after call_rcu for that many pages; it should be especially
188 * useful before resizing the root node with PREEMPT_NONE configs; the value was
189 * obtained experimentally, aiming to avoid visible slowdown.
191 static const int sync_pages = 128;
193 static struct kmem_cache *fn_alias_kmem __read_mostly;
194 static struct kmem_cache *trie_leaf_kmem __read_mostly;
196 static inline struct tnode *tn_info(struct key_vector *kv)
198 return container_of(kv, struct tnode, kv[0]);
201 /* caller must hold RTNL */
202 #define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
203 #define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
205 /* caller must hold RCU read lock or RTNL */
206 #define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
207 #define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
209 /* wrapper for rcu_assign_pointer */
210 static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
212 if (n)
213 rcu_assign_pointer(tn_info(n)->parent, tp);
216 #define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
218 /* This provides us with the number of children in this node, in the case of a
219 * leaf this will return 0 meaning none of the children are accessible.
221 static inline unsigned long child_length(const struct key_vector *tn)
223 return (1ul << tn->bits) & ~(1ul);
226 #define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
228 static inline unsigned long get_index(t_key key, struct key_vector *kv)
230 unsigned long index = key ^ kv->key;
232 if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
233 return 0;
235 return index >> kv->pos;
238 /* To understand this stuff, an understanding of keys and all their bits is
239 * necessary. Every node in the trie has a key associated with it, but not
240 * all of the bits in that key are significant.
242 * Consider a node 'n' and its parent 'tp'.
244 * If n is a leaf, every bit in its key is significant. Its presence is
245 * necessitated by path compression, since during a tree traversal (when
246 * searching for a leaf - unless we are doing an insertion) we will completely
247 * ignore all skipped bits we encounter. Thus we need to verify, at the end of
248 * a potentially successful search, that we have indeed been walking the
249 * correct key path.
251 * Note that we can never "miss" the correct key in the tree if present by
252 * following the wrong path. Path compression ensures that segments of the key
253 * that are the same for all keys with a given prefix are skipped, but the
254 * skipped part *is* identical for each node in the subtrie below the skipped
255 * bit! trie_insert() in this implementation takes care of that.
257 * if n is an internal node - a 'tnode' here, the various parts of its key
258 * have many different meanings.
260 * Example:
261 * _________________________________________________________________
262 * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
263 * -----------------------------------------------------------------
264 * 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16
266 * _________________________________________________________________
267 * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
268 * -----------------------------------------------------------------
269 * 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
271 * tp->pos = 22
272 * tp->bits = 3
273 * n->pos = 13
274 * n->bits = 4
276 * First, let's just ignore the bits that come before the parent tp, that is
277 * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
278 * point we do not use them for anything.
280 * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
281 * index into the parent's child array. That is, they will be used to find
282 * 'n' among tp's children.
284 * The bits from (n->pos + n->bits) to (tp->pos - 1) - "S" - are skipped bits
285 * for the node n.
287 * All the bits we have seen so far are significant to the node n. The rest
288 * of the bits are really not needed or indeed known in n->key.
290 * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
291 * n's child array, and will of course be different for each child.
293 * The rest of the bits, from 0 to (n->pos -1) - "u" - are completely unknown
294 * at this point.
297 static const int halve_threshold = 25;
298 static const int inflate_threshold = 50;
299 static const int halve_threshold_root = 15;
300 static const int inflate_threshold_root = 30;
302 static void __alias_free_mem(struct rcu_head *head)
304 struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
305 kmem_cache_free(fn_alias_kmem, fa);
308 static inline void alias_free_mem_rcu(struct fib_alias *fa)
310 call_rcu(&fa->rcu, __alias_free_mem);
313 #define TNODE_KMALLOC_MAX \
314 ilog2((PAGE_SIZE - TNODE_SIZE(0)) / sizeof(struct key_vector *))
315 #define TNODE_VMALLOC_MAX \
316 ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
318 static void __node_free_rcu(struct rcu_head *head)
320 struct tnode *n = container_of(head, struct tnode, rcu);
322 if (!n->tn_bits)
323 kmem_cache_free(trie_leaf_kmem, n);
324 else
325 kvfree(n);
328 #define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
330 static struct tnode *tnode_alloc(int bits)
332 size_t size;
334 /* verify bits is within bounds */
335 if (bits > TNODE_VMALLOC_MAX)
336 return NULL;
338 /* determine size and verify it is non-zero and didn't overflow */
339 size = TNODE_SIZE(1ul << bits);
341 if (size <= PAGE_SIZE)
342 return kzalloc(size, GFP_KERNEL);
343 else
344 return vzalloc(size);
347 static inline void empty_child_inc(struct key_vector *n)
349 ++tn_info(n)->empty_children ? : ++tn_info(n)->full_children;
352 static inline void empty_child_dec(struct key_vector *n)
354 tn_info(n)->empty_children-- ? : tn_info(n)->full_children--;
357 static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
359 struct key_vector *l;
360 struct tnode *kv;
362 kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
363 if (!kv)
364 return NULL;
366 /* initialize key vector */
367 l = kv->kv;
368 l->key = key;
369 l->pos = 0;
370 l->bits = 0;
371 l->slen = fa->fa_slen;
373 /* link leaf to fib alias */
374 INIT_HLIST_HEAD(&l->leaf);
375 hlist_add_head(&fa->fa_list, &l->leaf);
377 return l;
380 static struct key_vector *tnode_new(t_key key, int pos, int bits)
382 unsigned int shift = pos + bits;
383 struct key_vector *tn;
384 struct tnode *tnode;
386 /* verify bits and pos their msb bits clear and values are valid */
387 BUG_ON(!bits || (shift > KEYLENGTH));
389 tnode = tnode_alloc(bits);
390 if (!tnode)
391 return NULL;
393 pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
394 sizeof(struct key_vector *) << bits);
396 if (bits == KEYLENGTH)
397 tnode->full_children = 1;
398 else
399 tnode->empty_children = 1ul << bits;
401 tn = tnode->kv;
402 tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
403 tn->pos = pos;
404 tn->bits = bits;
405 tn->slen = pos;
407 return tn;
410 /* Check whether a tnode 'n' is "full", i.e. it is an internal node
411 * and no bits are skipped. See discussion in dyntree paper p. 6
413 static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
415 return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
418 /* Add a child at position i overwriting the old value.
419 * Update the value of full_children and empty_children.
421 static void put_child(struct key_vector *tn, unsigned long i,
422 struct key_vector *n)
424 struct key_vector *chi = get_child(tn, i);
425 int isfull, wasfull;
427 BUG_ON(i >= child_length(tn));
429 /* update emptyChildren, overflow into fullChildren */
430 if (!n && chi)
431 empty_child_inc(tn);
432 if (n && !chi)
433 empty_child_dec(tn);
435 /* update fullChildren */
436 wasfull = tnode_full(tn, chi);
437 isfull = tnode_full(tn, n);
439 if (wasfull && !isfull)
440 tn_info(tn)->full_children--;
441 else if (!wasfull && isfull)
442 tn_info(tn)->full_children++;
444 if (n && (tn->slen < n->slen))
445 tn->slen = n->slen;
447 rcu_assign_pointer(tn->tnode[i], n);
450 static void update_children(struct key_vector *tn)
452 unsigned long i;
454 /* update all of the child parent pointers */
455 for (i = child_length(tn); i;) {
456 struct key_vector *inode = get_child(tn, --i);
458 if (!inode)
459 continue;
461 /* Either update the children of a tnode that
462 * already belongs to us or update the child
463 * to point to ourselves.
465 if (node_parent(inode) == tn)
466 update_children(inode);
467 else
468 node_set_parent(inode, tn);
472 static inline void put_child_root(struct key_vector *tp, t_key key,
473 struct key_vector *n)
475 if (IS_TRIE(tp))
476 rcu_assign_pointer(tp->tnode[0], n);
477 else
478 put_child(tp, get_index(key, tp), n);
481 static inline void tnode_free_init(struct key_vector *tn)
483 tn_info(tn)->rcu.next = NULL;
486 static inline void tnode_free_append(struct key_vector *tn,
487 struct key_vector *n)
489 tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
490 tn_info(tn)->rcu.next = &tn_info(n)->rcu;
493 static void tnode_free(struct key_vector *tn)
495 struct callback_head *head = &tn_info(tn)->rcu;
497 while (head) {
498 head = head->next;
499 tnode_free_size += TNODE_SIZE(1ul << tn->bits);
500 node_free(tn);
502 tn = container_of(head, struct tnode, rcu)->kv;
505 if (tnode_free_size >= PAGE_SIZE * sync_pages) {
506 tnode_free_size = 0;
507 synchronize_rcu();
511 static struct key_vector *replace(struct trie *t,
512 struct key_vector *oldtnode,
513 struct key_vector *tn)
515 struct key_vector *tp = node_parent(oldtnode);
516 unsigned long i;
518 /* setup the parent pointer out of and back into this node */
519 NODE_INIT_PARENT(tn, tp);
520 put_child_root(tp, tn->key, tn);
522 /* update all of the child parent pointers */
523 update_children(tn);
525 /* all pointers should be clean so we are done */
526 tnode_free(oldtnode);
528 /* resize children now that oldtnode is freed */
529 for (i = child_length(tn); i;) {
530 struct key_vector *inode = get_child(tn, --i);
532 /* resize child node */
533 if (tnode_full(tn, inode))
534 tn = resize(t, inode);
537 return tp;
540 static struct key_vector *inflate(struct trie *t,
541 struct key_vector *oldtnode)
543 struct key_vector *tn;
544 unsigned long i;
545 t_key m;
547 pr_debug("In inflate\n");
549 tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
550 if (!tn)
551 goto notnode;
553 /* prepare oldtnode to be freed */
554 tnode_free_init(oldtnode);
556 /* Assemble all of the pointers in our cluster, in this case that
557 * represents all of the pointers out of our allocated nodes that
558 * point to existing tnodes and the links between our allocated
559 * nodes.
561 for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
562 struct key_vector *inode = get_child(oldtnode, --i);
563 struct key_vector *node0, *node1;
564 unsigned long j, k;
566 /* An empty child */
567 if (!inode)
568 continue;
570 /* A leaf or an internal node with skipped bits */
571 if (!tnode_full(oldtnode, inode)) {
572 put_child(tn, get_index(inode->key, tn), inode);
573 continue;
576 /* drop the node in the old tnode free list */
577 tnode_free_append(oldtnode, inode);
579 /* An internal node with two children */
580 if (inode->bits == 1) {
581 put_child(tn, 2 * i + 1, get_child(inode, 1));
582 put_child(tn, 2 * i, get_child(inode, 0));
583 continue;
586 /* We will replace this node 'inode' with two new
587 * ones, 'node0' and 'node1', each with half of the
588 * original children. The two new nodes will have
589 * a position one bit further down the key and this
590 * means that the "significant" part of their keys
591 * (see the discussion near the top of this file)
592 * will differ by one bit, which will be "0" in
593 * node0's key and "1" in node1's key. Since we are
594 * moving the key position by one step, the bit that
595 * we are moving away from - the bit at position
596 * (tn->pos) - is the one that will differ between
597 * node0 and node1. So... we synthesize that bit in the
598 * two new keys.
600 node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
601 if (!node1)
602 goto nomem;
603 node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
605 tnode_free_append(tn, node1);
606 if (!node0)
607 goto nomem;
608 tnode_free_append(tn, node0);
610 /* populate child pointers in new nodes */
611 for (k = child_length(inode), j = k / 2; j;) {
612 put_child(node1, --j, get_child(inode, --k));
613 put_child(node0, j, get_child(inode, j));
614 put_child(node1, --j, get_child(inode, --k));
615 put_child(node0, j, get_child(inode, j));
618 /* link new nodes to parent */
619 NODE_INIT_PARENT(node1, tn);
620 NODE_INIT_PARENT(node0, tn);
622 /* link parent to nodes */
623 put_child(tn, 2 * i + 1, node1);
624 put_child(tn, 2 * i, node0);
627 /* setup the parent pointers into and out of this node */
628 return replace(t, oldtnode, tn);
629 nomem:
630 /* all pointers should be clean so we are done */
631 tnode_free(tn);
632 notnode:
633 return NULL;
636 static struct key_vector *halve(struct trie *t,
637 struct key_vector *oldtnode)
639 struct key_vector *tn;
640 unsigned long i;
642 pr_debug("In halve\n");
644 tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
645 if (!tn)
646 goto notnode;
648 /* prepare oldtnode to be freed */
649 tnode_free_init(oldtnode);
651 /* Assemble all of the pointers in our cluster, in this case that
652 * represents all of the pointers out of our allocated nodes that
653 * point to existing tnodes and the links between our allocated
654 * nodes.
656 for (i = child_length(oldtnode); i;) {
657 struct key_vector *node1 = get_child(oldtnode, --i);
658 struct key_vector *node0 = get_child(oldtnode, --i);
659 struct key_vector *inode;
661 /* At least one of the children is empty */
662 if (!node1 || !node0) {
663 put_child(tn, i / 2, node1 ? : node0);
664 continue;
667 /* Two nonempty children */
668 inode = tnode_new(node0->key, oldtnode->pos, 1);
669 if (!inode)
670 goto nomem;
671 tnode_free_append(tn, inode);
673 /* initialize pointers out of node */
674 put_child(inode, 1, node1);
675 put_child(inode, 0, node0);
676 NODE_INIT_PARENT(inode, tn);
678 /* link parent to node */
679 put_child(tn, i / 2, inode);
682 /* setup the parent pointers into and out of this node */
683 return replace(t, oldtnode, tn);
684 nomem:
685 /* all pointers should be clean so we are done */
686 tnode_free(tn);
687 notnode:
688 return NULL;
691 static struct key_vector *collapse(struct trie *t,
692 struct key_vector *oldtnode)
694 struct key_vector *n, *tp;
695 unsigned long i;
697 /* scan the tnode looking for that one child that might still exist */
698 for (n = NULL, i = child_length(oldtnode); !n && i;)
699 n = get_child(oldtnode, --i);
701 /* compress one level */
702 tp = node_parent(oldtnode);
703 put_child_root(tp, oldtnode->key, n);
704 node_set_parent(n, tp);
706 /* drop dead node */
707 node_free(oldtnode);
709 return tp;
712 static unsigned char update_suffix(struct key_vector *tn)
714 unsigned char slen = tn->pos;
715 unsigned long stride, i;
716 unsigned char slen_max;
718 /* only vector 0 can have a suffix length greater than or equal to
719 * tn->pos + tn->bits, the second highest node will have a suffix
720 * length at most of tn->pos + tn->bits - 1
722 slen_max = min_t(unsigned char, tn->pos + tn->bits - 1, tn->slen);
724 /* search though the list of children looking for nodes that might
725 * have a suffix greater than the one we currently have. This is
726 * why we start with a stride of 2 since a stride of 1 would
727 * represent the nodes with suffix length equal to tn->pos
729 for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
730 struct key_vector *n = get_child(tn, i);
732 if (!n || (n->slen <= slen))
733 continue;
735 /* update stride and slen based on new value */
736 stride <<= (n->slen - slen);
737 slen = n->slen;
738 i &= ~(stride - 1);
740 /* stop searching if we have hit the maximum possible value */
741 if (slen >= slen_max)
742 break;
745 tn->slen = slen;
747 return slen;
750 /* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
751 * the Helsinki University of Technology and Matti Tikkanen of Nokia
752 * Telecommunications, page 6:
753 * "A node is doubled if the ratio of non-empty children to all
754 * children in the *doubled* node is at least 'high'."
756 * 'high' in this instance is the variable 'inflate_threshold'. It
757 * is expressed as a percentage, so we multiply it with
758 * child_length() and instead of multiplying by 2 (since the
759 * child array will be doubled by inflate()) and multiplying
760 * the left-hand side by 100 (to handle the percentage thing) we
761 * multiply the left-hand side by 50.
763 * The left-hand side may look a bit weird: child_length(tn)
764 * - tn->empty_children is of course the number of non-null children
765 * in the current node. tn->full_children is the number of "full"
766 * children, that is non-null tnodes with a skip value of 0.
767 * All of those will be doubled in the resulting inflated tnode, so
768 * we just count them one extra time here.
770 * A clearer way to write this would be:
772 * to_be_doubled = tn->full_children;
773 * not_to_be_doubled = child_length(tn) - tn->empty_children -
774 * tn->full_children;
776 * new_child_length = child_length(tn) * 2;
778 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
779 * new_child_length;
780 * if (new_fill_factor >= inflate_threshold)
782 * ...and so on, tho it would mess up the while () loop.
784 * anyway,
785 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
786 * inflate_threshold
788 * avoid a division:
789 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
790 * inflate_threshold * new_child_length
792 * expand not_to_be_doubled and to_be_doubled, and shorten:
793 * 100 * (child_length(tn) - tn->empty_children +
794 * tn->full_children) >= inflate_threshold * new_child_length
796 * expand new_child_length:
797 * 100 * (child_length(tn) - tn->empty_children +
798 * tn->full_children) >=
799 * inflate_threshold * child_length(tn) * 2
801 * shorten again:
802 * 50 * (tn->full_children + child_length(tn) -
803 * tn->empty_children) >= inflate_threshold *
804 * child_length(tn)
807 static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
809 unsigned long used = child_length(tn);
810 unsigned long threshold = used;
812 /* Keep root node larger */
813 threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
814 used -= tn_info(tn)->empty_children;
815 used += tn_info(tn)->full_children;
817 /* if bits == KEYLENGTH then pos = 0, and will fail below */
819 return (used > 1) && tn->pos && ((50 * used) >= threshold);
822 static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
824 unsigned long used = child_length(tn);
825 unsigned long threshold = used;
827 /* Keep root node larger */
828 threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
829 used -= tn_info(tn)->empty_children;
831 /* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
833 return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
836 static inline bool should_collapse(struct key_vector *tn)
838 unsigned long used = child_length(tn);
840 used -= tn_info(tn)->empty_children;
842 /* account for bits == KEYLENGTH case */
843 if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
844 used -= KEY_MAX;
846 /* One child or none, time to drop us from the trie */
847 return used < 2;
850 #define MAX_WORK 10
851 static struct key_vector *resize(struct trie *t, struct key_vector *tn)
853 #ifdef CONFIG_IP_FIB_TRIE_STATS
854 struct trie_use_stats __percpu *stats = t->stats;
855 #endif
856 struct key_vector *tp = node_parent(tn);
857 unsigned long cindex = get_index(tn->key, tp);
858 int max_work = MAX_WORK;
860 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
861 tn, inflate_threshold, halve_threshold);
863 /* track the tnode via the pointer from the parent instead of
864 * doing it ourselves. This way we can let RCU fully do its
865 * thing without us interfering
867 BUG_ON(tn != get_child(tp, cindex));
869 /* Double as long as the resulting node has a number of
870 * nonempty nodes that are above the threshold.
872 while (should_inflate(tp, tn) && max_work) {
873 tp = inflate(t, tn);
874 if (!tp) {
875 #ifdef CONFIG_IP_FIB_TRIE_STATS
876 this_cpu_inc(stats->resize_node_skipped);
877 #endif
878 break;
881 max_work--;
882 tn = get_child(tp, cindex);
885 /* update parent in case inflate failed */
886 tp = node_parent(tn);
888 /* Return if at least one inflate is run */
889 if (max_work != MAX_WORK)
890 return tp;
892 /* Halve as long as the number of empty children in this
893 * node is above threshold.
895 while (should_halve(tp, tn) && max_work) {
896 tp = halve(t, tn);
897 if (!tp) {
898 #ifdef CONFIG_IP_FIB_TRIE_STATS
899 this_cpu_inc(stats->resize_node_skipped);
900 #endif
901 break;
904 max_work--;
905 tn = get_child(tp, cindex);
908 /* Only one child remains */
909 if (should_collapse(tn))
910 return collapse(t, tn);
912 /* update parent in case halve failed */
913 return node_parent(tn);
916 static void node_pull_suffix(struct key_vector *tn, unsigned char slen)
918 unsigned char node_slen = tn->slen;
920 while ((node_slen > tn->pos) && (node_slen > slen)) {
921 slen = update_suffix(tn);
922 if (node_slen == slen)
923 break;
925 tn = node_parent(tn);
926 node_slen = tn->slen;
930 static void node_push_suffix(struct key_vector *tn, unsigned char slen)
932 while (tn->slen < slen) {
933 tn->slen = slen;
934 tn = node_parent(tn);
938 /* rcu_read_lock needs to be hold by caller from readside */
939 static struct key_vector *fib_find_node(struct trie *t,
940 struct key_vector **tp, u32 key)
942 struct key_vector *pn, *n = t->kv;
943 unsigned long index = 0;
945 do {
946 pn = n;
947 n = get_child_rcu(n, index);
949 if (!n)
950 break;
952 index = get_cindex(key, n);
954 /* This bit of code is a bit tricky but it combines multiple
955 * checks into a single check. The prefix consists of the
956 * prefix plus zeros for the bits in the cindex. The index
957 * is the difference between the key and this value. From
958 * this we can actually derive several pieces of data.
959 * if (index >= (1ul << bits))
960 * we have a mismatch in skip bits and failed
961 * else
962 * we know the value is cindex
964 * This check is safe even if bits == KEYLENGTH due to the
965 * fact that we can only allocate a node with 32 bits if a
966 * long is greater than 32 bits.
968 if (index >= (1ul << n->bits)) {
969 n = NULL;
970 break;
973 /* keep searching until we find a perfect match leaf or NULL */
974 } while (IS_TNODE(n));
976 *tp = pn;
978 return n;
981 /* Return the first fib alias matching TOS with
982 * priority less than or equal to PRIO.
984 static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
985 u8 tos, u32 prio, u32 tb_id)
987 struct fib_alias *fa;
989 if (!fah)
990 return NULL;
992 hlist_for_each_entry(fa, fah, fa_list) {
993 if (fa->fa_slen < slen)
994 continue;
995 if (fa->fa_slen != slen)
996 break;
997 if (fa->tb_id > tb_id)
998 continue;
999 if (fa->tb_id != tb_id)
1000 break;
1001 if (fa->fa_tos > tos)
1002 continue;
1003 if (fa->fa_info->fib_priority >= prio || fa->fa_tos < tos)
1004 return fa;
1007 return NULL;
1010 static void trie_rebalance(struct trie *t, struct key_vector *tn)
1012 while (!IS_TRIE(tn))
1013 tn = resize(t, tn);
1016 static int fib_insert_node(struct trie *t, struct key_vector *tp,
1017 struct fib_alias *new, t_key key)
1019 struct key_vector *n, *l;
1021 l = leaf_new(key, new);
1022 if (!l)
1023 goto noleaf;
1025 /* retrieve child from parent node */
1026 n = get_child(tp, get_index(key, tp));
1028 /* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1030 * Add a new tnode here
1031 * first tnode need some special handling
1032 * leaves us in position for handling as case 3
1034 if (n) {
1035 struct key_vector *tn;
1037 tn = tnode_new(key, __fls(key ^ n->key), 1);
1038 if (!tn)
1039 goto notnode;
1041 /* initialize routes out of node */
1042 NODE_INIT_PARENT(tn, tp);
1043 put_child(tn, get_index(key, tn) ^ 1, n);
1045 /* start adding routes into the node */
1046 put_child_root(tp, key, tn);
1047 node_set_parent(n, tn);
1049 /* parent now has a NULL spot where the leaf can go */
1050 tp = tn;
1053 /* Case 3: n is NULL, and will just insert a new leaf */
1054 node_push_suffix(tp, new->fa_slen);
1055 NODE_INIT_PARENT(l, tp);
1056 put_child_root(tp, key, l);
1057 trie_rebalance(t, tp);
1059 return 0;
1060 notnode:
1061 node_free(l);
1062 noleaf:
1063 return -ENOMEM;
1066 static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1067 struct key_vector *l, struct fib_alias *new,
1068 struct fib_alias *fa, t_key key)
1070 if (!l)
1071 return fib_insert_node(t, tp, new, key);
1073 if (fa) {
1074 hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1075 } else {
1076 struct fib_alias *last;
1078 hlist_for_each_entry(last, &l->leaf, fa_list) {
1079 if (new->fa_slen < last->fa_slen)
1080 break;
1081 if ((new->fa_slen == last->fa_slen) &&
1082 (new->tb_id > last->tb_id))
1083 break;
1084 fa = last;
1087 if (fa)
1088 hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1089 else
1090 hlist_add_head_rcu(&new->fa_list, &l->leaf);
1093 /* if we added to the tail node then we need to update slen */
1094 if (l->slen < new->fa_slen) {
1095 l->slen = new->fa_slen;
1096 node_push_suffix(tp, new->fa_slen);
1099 return 0;
1102 /* Caller must hold RTNL. */
1103 int fib_table_insert(struct net *net, struct fib_table *tb,
1104 struct fib_config *cfg)
1106 enum fib_event_type event = FIB_EVENT_ENTRY_ADD;
1107 struct trie *t = (struct trie *)tb->tb_data;
1108 struct fib_alias *fa, *new_fa;
1109 struct key_vector *l, *tp;
1110 u16 nlflags = NLM_F_EXCL;
1111 struct fib_info *fi;
1112 u8 plen = cfg->fc_dst_len;
1113 u8 slen = KEYLENGTH - plen;
1114 u8 tos = cfg->fc_tos;
1115 u32 key;
1116 int err;
1118 if (plen > KEYLENGTH)
1119 return -EINVAL;
1121 key = ntohl(cfg->fc_dst);
1123 pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1125 if ((plen < KEYLENGTH) && (key << plen))
1126 return -EINVAL;
1128 fi = fib_create_info(cfg);
1129 if (IS_ERR(fi)) {
1130 err = PTR_ERR(fi);
1131 goto err;
1134 l = fib_find_node(t, &tp, key);
1135 fa = l ? fib_find_alias(&l->leaf, slen, tos, fi->fib_priority,
1136 tb->tb_id) : NULL;
1138 /* Now fa, if non-NULL, points to the first fib alias
1139 * with the same keys [prefix,tos,priority], if such key already
1140 * exists or to the node before which we will insert new one.
1142 * If fa is NULL, we will need to allocate a new one and
1143 * insert to the tail of the section matching the suffix length
1144 * of the new alias.
1147 if (fa && fa->fa_tos == tos &&
1148 fa->fa_info->fib_priority == fi->fib_priority) {
1149 struct fib_alias *fa_first, *fa_match;
1151 err = -EEXIST;
1152 if (cfg->fc_nlflags & NLM_F_EXCL)
1153 goto out;
1155 nlflags &= ~NLM_F_EXCL;
1157 /* We have 2 goals:
1158 * 1. Find exact match for type, scope, fib_info to avoid
1159 * duplicate routes
1160 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1162 fa_match = NULL;
1163 fa_first = fa;
1164 hlist_for_each_entry_from(fa, fa_list) {
1165 if ((fa->fa_slen != slen) ||
1166 (fa->tb_id != tb->tb_id) ||
1167 (fa->fa_tos != tos))
1168 break;
1169 if (fa->fa_info->fib_priority != fi->fib_priority)
1170 break;
1171 if (fa->fa_type == cfg->fc_type &&
1172 fa->fa_info == fi) {
1173 fa_match = fa;
1174 break;
1178 if (cfg->fc_nlflags & NLM_F_REPLACE) {
1179 struct fib_info *fi_drop;
1180 u8 state;
1182 nlflags |= NLM_F_REPLACE;
1183 fa = fa_first;
1184 if (fa_match) {
1185 if (fa == fa_match)
1186 err = 0;
1187 goto out;
1189 err = -ENOBUFS;
1190 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1191 if (!new_fa)
1192 goto out;
1194 fi_drop = fa->fa_info;
1195 new_fa->fa_tos = fa->fa_tos;
1196 new_fa->fa_info = fi;
1197 new_fa->fa_type = cfg->fc_type;
1198 state = fa->fa_state;
1199 new_fa->fa_state = state & ~FA_S_ACCESSED;
1200 new_fa->fa_slen = fa->fa_slen;
1201 new_fa->tb_id = tb->tb_id;
1202 new_fa->fa_default = -1;
1204 call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_REPLACE,
1205 key, plen, fi,
1206 new_fa->fa_tos, cfg->fc_type,
1207 tb->tb_id);
1208 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1209 tb->tb_id, &cfg->fc_nlinfo, nlflags);
1211 hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1213 alias_free_mem_rcu(fa);
1215 fib_release_info(fi_drop);
1216 if (state & FA_S_ACCESSED)
1217 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1219 goto succeeded;
1221 /* Error if we find a perfect match which
1222 * uses the same scope, type, and nexthop
1223 * information.
1225 if (fa_match)
1226 goto out;
1228 if (cfg->fc_nlflags & NLM_F_APPEND) {
1229 event = FIB_EVENT_ENTRY_APPEND;
1230 nlflags |= NLM_F_APPEND;
1231 } else {
1232 fa = fa_first;
1235 err = -ENOENT;
1236 if (!(cfg->fc_nlflags & NLM_F_CREATE))
1237 goto out;
1239 nlflags |= NLM_F_CREATE;
1240 err = -ENOBUFS;
1241 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1242 if (!new_fa)
1243 goto out;
1245 new_fa->fa_info = fi;
1246 new_fa->fa_tos = tos;
1247 new_fa->fa_type = cfg->fc_type;
1248 new_fa->fa_state = 0;
1249 new_fa->fa_slen = slen;
1250 new_fa->tb_id = tb->tb_id;
1251 new_fa->fa_default = -1;
1253 /* Insert new entry to the list. */
1254 err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1255 if (err)
1256 goto out_free_new_fa;
1258 if (!plen)
1259 tb->tb_num_default++;
1261 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1262 call_fib_entry_notifiers(net, event, key, plen, fi, tos, cfg->fc_type,
1263 tb->tb_id);
1264 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1265 &cfg->fc_nlinfo, nlflags);
1266 succeeded:
1267 return 0;
1269 out_free_new_fa:
1270 kmem_cache_free(fn_alias_kmem, new_fa);
1271 out:
1272 fib_release_info(fi);
1273 err:
1274 return err;
1277 static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1279 t_key prefix = n->key;
1281 return (key ^ prefix) & (prefix | -prefix);
1284 /* should be called with rcu_read_lock */
1285 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1286 struct fib_result *res, int fib_flags)
1288 struct trie *t = (struct trie *) tb->tb_data;
1289 #ifdef CONFIG_IP_FIB_TRIE_STATS
1290 struct trie_use_stats __percpu *stats = t->stats;
1291 #endif
1292 const t_key key = ntohl(flp->daddr);
1293 struct key_vector *n, *pn;
1294 struct fib_alias *fa;
1295 unsigned long index;
1296 t_key cindex;
1298 trace_fib_table_lookup(tb->tb_id, flp);
1300 pn = t->kv;
1301 cindex = 0;
1303 n = get_child_rcu(pn, cindex);
1304 if (!n)
1305 return -EAGAIN;
1307 #ifdef CONFIG_IP_FIB_TRIE_STATS
1308 this_cpu_inc(stats->gets);
1309 #endif
1311 /* Step 1: Travel to the longest prefix match in the trie */
1312 for (;;) {
1313 index = get_cindex(key, n);
1315 /* This bit of code is a bit tricky but it combines multiple
1316 * checks into a single check. The prefix consists of the
1317 * prefix plus zeros for the "bits" in the prefix. The index
1318 * is the difference between the key and this value. From
1319 * this we can actually derive several pieces of data.
1320 * if (index >= (1ul << bits))
1321 * we have a mismatch in skip bits and failed
1322 * else
1323 * we know the value is cindex
1325 * This check is safe even if bits == KEYLENGTH due to the
1326 * fact that we can only allocate a node with 32 bits if a
1327 * long is greater than 32 bits.
1329 if (index >= (1ul << n->bits))
1330 break;
1332 /* we have found a leaf. Prefixes have already been compared */
1333 if (IS_LEAF(n))
1334 goto found;
1336 /* only record pn and cindex if we are going to be chopping
1337 * bits later. Otherwise we are just wasting cycles.
1339 if (n->slen > n->pos) {
1340 pn = n;
1341 cindex = index;
1344 n = get_child_rcu(n, index);
1345 if (unlikely(!n))
1346 goto backtrace;
1349 /* Step 2: Sort out leaves and begin backtracing for longest prefix */
1350 for (;;) {
1351 /* record the pointer where our next node pointer is stored */
1352 struct key_vector __rcu **cptr = n->tnode;
1354 /* This test verifies that none of the bits that differ
1355 * between the key and the prefix exist in the region of
1356 * the lsb and higher in the prefix.
1358 if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1359 goto backtrace;
1361 /* exit out and process leaf */
1362 if (unlikely(IS_LEAF(n)))
1363 break;
1365 /* Don't bother recording parent info. Since we are in
1366 * prefix match mode we will have to come back to wherever
1367 * we started this traversal anyway
1370 while ((n = rcu_dereference(*cptr)) == NULL) {
1371 backtrace:
1372 #ifdef CONFIG_IP_FIB_TRIE_STATS
1373 if (!n)
1374 this_cpu_inc(stats->null_node_hit);
1375 #endif
1376 /* If we are at cindex 0 there are no more bits for
1377 * us to strip at this level so we must ascend back
1378 * up one level to see if there are any more bits to
1379 * be stripped there.
1381 while (!cindex) {
1382 t_key pkey = pn->key;
1384 /* If we don't have a parent then there is
1385 * nothing for us to do as we do not have any
1386 * further nodes to parse.
1388 if (IS_TRIE(pn))
1389 return -EAGAIN;
1390 #ifdef CONFIG_IP_FIB_TRIE_STATS
1391 this_cpu_inc(stats->backtrack);
1392 #endif
1393 /* Get Child's index */
1394 pn = node_parent_rcu(pn);
1395 cindex = get_index(pkey, pn);
1398 /* strip the least significant bit from the cindex */
1399 cindex &= cindex - 1;
1401 /* grab pointer for next child node */
1402 cptr = &pn->tnode[cindex];
1406 found:
1407 /* this line carries forward the xor from earlier in the function */
1408 index = key ^ n->key;
1410 /* Step 3: Process the leaf, if that fails fall back to backtracing */
1411 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1412 struct fib_info *fi = fa->fa_info;
1413 int nhsel, err;
1415 if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1416 if (index >= (1ul << fa->fa_slen))
1417 continue;
1419 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1420 continue;
1421 if (fi->fib_dead)
1422 continue;
1423 if (fa->fa_info->fib_scope < flp->flowi4_scope)
1424 continue;
1425 fib_alias_accessed(fa);
1426 err = fib_props[fa->fa_type].error;
1427 if (unlikely(err < 0)) {
1428 #ifdef CONFIG_IP_FIB_TRIE_STATS
1429 this_cpu_inc(stats->semantic_match_passed);
1430 #endif
1431 return err;
1433 if (fi->fib_flags & RTNH_F_DEAD)
1434 continue;
1435 for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
1436 const struct fib_nh *nh = &fi->fib_nh[nhsel];
1437 struct in_device *in_dev = __in_dev_get_rcu(nh->nh_dev);
1439 if (nh->nh_flags & RTNH_F_DEAD)
1440 continue;
1441 if (in_dev &&
1442 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1443 nh->nh_flags & RTNH_F_LINKDOWN &&
1444 !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1445 continue;
1446 if (!(flp->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF)) {
1447 if (flp->flowi4_oif &&
1448 flp->flowi4_oif != nh->nh_oif)
1449 continue;
1452 if (!(fib_flags & FIB_LOOKUP_NOREF))
1453 atomic_inc(&fi->fib_clntref);
1455 res->prefixlen = KEYLENGTH - fa->fa_slen;
1456 res->nh_sel = nhsel;
1457 res->type = fa->fa_type;
1458 res->scope = fi->fib_scope;
1459 res->fi = fi;
1460 res->table = tb;
1461 res->fa_head = &n->leaf;
1462 #ifdef CONFIG_IP_FIB_TRIE_STATS
1463 this_cpu_inc(stats->semantic_match_passed);
1464 #endif
1465 trace_fib_table_lookup_nh(nh);
1467 return err;
1470 #ifdef CONFIG_IP_FIB_TRIE_STATS
1471 this_cpu_inc(stats->semantic_match_miss);
1472 #endif
1473 goto backtrace;
1475 EXPORT_SYMBOL_GPL(fib_table_lookup);
1477 static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1478 struct key_vector *l, struct fib_alias *old)
1480 /* record the location of the previous list_info entry */
1481 struct hlist_node **pprev = old->fa_list.pprev;
1482 struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1484 /* remove the fib_alias from the list */
1485 hlist_del_rcu(&old->fa_list);
1487 /* if we emptied the list this leaf will be freed and we can sort
1488 * out parent suffix lengths as a part of trie_rebalance
1490 if (hlist_empty(&l->leaf)) {
1491 if (tp->slen == l->slen)
1492 node_pull_suffix(tp, tp->pos);
1493 put_child_root(tp, l->key, NULL);
1494 node_free(l);
1495 trie_rebalance(t, tp);
1496 return;
1499 /* only access fa if it is pointing at the last valid hlist_node */
1500 if (*pprev)
1501 return;
1503 /* update the trie with the latest suffix length */
1504 l->slen = fa->fa_slen;
1505 node_pull_suffix(tp, fa->fa_slen);
1508 /* Caller must hold RTNL. */
1509 int fib_table_delete(struct net *net, struct fib_table *tb,
1510 struct fib_config *cfg)
1512 struct trie *t = (struct trie *) tb->tb_data;
1513 struct fib_alias *fa, *fa_to_delete;
1514 struct key_vector *l, *tp;
1515 u8 plen = cfg->fc_dst_len;
1516 u8 slen = KEYLENGTH - plen;
1517 u8 tos = cfg->fc_tos;
1518 u32 key;
1520 if (plen > KEYLENGTH)
1521 return -EINVAL;
1523 key = ntohl(cfg->fc_dst);
1525 if ((plen < KEYLENGTH) && (key << plen))
1526 return -EINVAL;
1528 l = fib_find_node(t, &tp, key);
1529 if (!l)
1530 return -ESRCH;
1532 fa = fib_find_alias(&l->leaf, slen, tos, 0, tb->tb_id);
1533 if (!fa)
1534 return -ESRCH;
1536 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1538 fa_to_delete = NULL;
1539 hlist_for_each_entry_from(fa, fa_list) {
1540 struct fib_info *fi = fa->fa_info;
1542 if ((fa->fa_slen != slen) ||
1543 (fa->tb_id != tb->tb_id) ||
1544 (fa->fa_tos != tos))
1545 break;
1547 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1548 (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1549 fa->fa_info->fib_scope == cfg->fc_scope) &&
1550 (!cfg->fc_prefsrc ||
1551 fi->fib_prefsrc == cfg->fc_prefsrc) &&
1552 (!cfg->fc_protocol ||
1553 fi->fib_protocol == cfg->fc_protocol) &&
1554 fib_nh_match(cfg, fi) == 0) {
1555 fa_to_delete = fa;
1556 break;
1560 if (!fa_to_delete)
1561 return -ESRCH;
1563 call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, key, plen,
1564 fa_to_delete->fa_info, tos,
1565 fa_to_delete->fa_type, tb->tb_id);
1566 rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1567 &cfg->fc_nlinfo, 0);
1569 if (!plen)
1570 tb->tb_num_default--;
1572 fib_remove_alias(t, tp, l, fa_to_delete);
1574 if (fa_to_delete->fa_state & FA_S_ACCESSED)
1575 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1577 fib_release_info(fa_to_delete->fa_info);
1578 alias_free_mem_rcu(fa_to_delete);
1579 return 0;
1582 /* Scan for the next leaf starting at the provided key value */
1583 static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1585 struct key_vector *pn, *n = *tn;
1586 unsigned long cindex;
1588 /* this loop is meant to try and find the key in the trie */
1589 do {
1590 /* record parent and next child index */
1591 pn = n;
1592 cindex = (key > pn->key) ? get_index(key, pn) : 0;
1594 if (cindex >> pn->bits)
1595 break;
1597 /* descend into the next child */
1598 n = get_child_rcu(pn, cindex++);
1599 if (!n)
1600 break;
1602 /* guarantee forward progress on the keys */
1603 if (IS_LEAF(n) && (n->key >= key))
1604 goto found;
1605 } while (IS_TNODE(n));
1607 /* this loop will search for the next leaf with a greater key */
1608 while (!IS_TRIE(pn)) {
1609 /* if we exhausted the parent node we will need to climb */
1610 if (cindex >= (1ul << pn->bits)) {
1611 t_key pkey = pn->key;
1613 pn = node_parent_rcu(pn);
1614 cindex = get_index(pkey, pn) + 1;
1615 continue;
1618 /* grab the next available node */
1619 n = get_child_rcu(pn, cindex++);
1620 if (!n)
1621 continue;
1623 /* no need to compare keys since we bumped the index */
1624 if (IS_LEAF(n))
1625 goto found;
1627 /* Rescan start scanning in new node */
1628 pn = n;
1629 cindex = 0;
1632 *tn = pn;
1633 return NULL; /* Root of trie */
1634 found:
1635 /* if we are at the limit for keys just return NULL for the tnode */
1636 *tn = pn;
1637 return n;
1640 static void fib_trie_free(struct fib_table *tb)
1642 struct trie *t = (struct trie *)tb->tb_data;
1643 struct key_vector *pn = t->kv;
1644 unsigned long cindex = 1;
1645 struct hlist_node *tmp;
1646 struct fib_alias *fa;
1648 /* walk trie in reverse order and free everything */
1649 for (;;) {
1650 struct key_vector *n;
1652 if (!(cindex--)) {
1653 t_key pkey = pn->key;
1655 if (IS_TRIE(pn))
1656 break;
1658 n = pn;
1659 pn = node_parent(pn);
1661 /* drop emptied tnode */
1662 put_child_root(pn, n->key, NULL);
1663 node_free(n);
1665 cindex = get_index(pkey, pn);
1667 continue;
1670 /* grab the next available node */
1671 n = get_child(pn, cindex);
1672 if (!n)
1673 continue;
1675 if (IS_TNODE(n)) {
1676 /* record pn and cindex for leaf walking */
1677 pn = n;
1678 cindex = 1ul << n->bits;
1680 continue;
1683 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1684 hlist_del_rcu(&fa->fa_list);
1685 alias_free_mem_rcu(fa);
1688 put_child_root(pn, n->key, NULL);
1689 node_free(n);
1692 #ifdef CONFIG_IP_FIB_TRIE_STATS
1693 free_percpu(t->stats);
1694 #endif
1695 kfree(tb);
1698 struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1700 struct trie *ot = (struct trie *)oldtb->tb_data;
1701 struct key_vector *l, *tp = ot->kv;
1702 struct fib_table *local_tb;
1703 struct fib_alias *fa;
1704 struct trie *lt;
1705 t_key key = 0;
1707 if (oldtb->tb_data == oldtb->__data)
1708 return oldtb;
1710 local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1711 if (!local_tb)
1712 return NULL;
1714 lt = (struct trie *)local_tb->tb_data;
1716 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1717 struct key_vector *local_l = NULL, *local_tp;
1719 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1720 struct fib_alias *new_fa;
1722 if (local_tb->tb_id != fa->tb_id)
1723 continue;
1725 /* clone fa for new local table */
1726 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1727 if (!new_fa)
1728 goto out;
1730 memcpy(new_fa, fa, sizeof(*fa));
1732 /* insert clone into table */
1733 if (!local_l)
1734 local_l = fib_find_node(lt, &local_tp, l->key);
1736 if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1737 NULL, l->key)) {
1738 kmem_cache_free(fn_alias_kmem, new_fa);
1739 goto out;
1743 /* stop loop if key wrapped back to 0 */
1744 key = l->key + 1;
1745 if (key < l->key)
1746 break;
1749 return local_tb;
1750 out:
1751 fib_trie_free(local_tb);
1753 return NULL;
1756 /* Caller must hold RTNL */
1757 void fib_table_flush_external(struct fib_table *tb)
1759 struct trie *t = (struct trie *)tb->tb_data;
1760 struct key_vector *pn = t->kv;
1761 unsigned long cindex = 1;
1762 struct hlist_node *tmp;
1763 struct fib_alias *fa;
1765 /* walk trie in reverse order */
1766 for (;;) {
1767 unsigned char slen = 0;
1768 struct key_vector *n;
1770 if (!(cindex--)) {
1771 t_key pkey = pn->key;
1773 /* cannot resize the trie vector */
1774 if (IS_TRIE(pn))
1775 break;
1777 /* update the suffix to address pulled leaves */
1778 if (pn->slen > pn->pos)
1779 update_suffix(pn);
1781 /* resize completed node */
1782 pn = resize(t, pn);
1783 cindex = get_index(pkey, pn);
1785 continue;
1788 /* grab the next available node */
1789 n = get_child(pn, cindex);
1790 if (!n)
1791 continue;
1793 if (IS_TNODE(n)) {
1794 /* record pn and cindex for leaf walking */
1795 pn = n;
1796 cindex = 1ul << n->bits;
1798 continue;
1801 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1802 /* if alias was cloned to local then we just
1803 * need to remove the local copy from main
1805 if (tb->tb_id != fa->tb_id) {
1806 hlist_del_rcu(&fa->fa_list);
1807 alias_free_mem_rcu(fa);
1808 continue;
1811 /* record local slen */
1812 slen = fa->fa_slen;
1815 /* update leaf slen */
1816 n->slen = slen;
1818 if (hlist_empty(&n->leaf)) {
1819 put_child_root(pn, n->key, NULL);
1820 node_free(n);
1825 /* Caller must hold RTNL. */
1826 int fib_table_flush(struct net *net, struct fib_table *tb)
1828 struct trie *t = (struct trie *)tb->tb_data;
1829 struct key_vector *pn = t->kv;
1830 unsigned long cindex = 1;
1831 struct hlist_node *tmp;
1832 struct fib_alias *fa;
1833 int found = 0;
1835 /* walk trie in reverse order */
1836 for (;;) {
1837 unsigned char slen = 0;
1838 struct key_vector *n;
1840 if (!(cindex--)) {
1841 t_key pkey = pn->key;
1843 /* cannot resize the trie vector */
1844 if (IS_TRIE(pn))
1845 break;
1847 /* update the suffix to address pulled leaves */
1848 if (pn->slen > pn->pos)
1849 update_suffix(pn);
1851 /* resize completed node */
1852 pn = resize(t, pn);
1853 cindex = get_index(pkey, pn);
1855 continue;
1858 /* grab the next available node */
1859 n = get_child(pn, cindex);
1860 if (!n)
1861 continue;
1863 if (IS_TNODE(n)) {
1864 /* record pn and cindex for leaf walking */
1865 pn = n;
1866 cindex = 1ul << n->bits;
1868 continue;
1871 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1872 struct fib_info *fi = fa->fa_info;
1874 if (!fi || !(fi->fib_flags & RTNH_F_DEAD) ||
1875 tb->tb_id != fa->tb_id) {
1876 slen = fa->fa_slen;
1877 continue;
1880 call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL,
1881 n->key,
1882 KEYLENGTH - fa->fa_slen,
1883 fi, fa->fa_tos, fa->fa_type,
1884 tb->tb_id);
1885 hlist_del_rcu(&fa->fa_list);
1886 fib_release_info(fa->fa_info);
1887 alias_free_mem_rcu(fa);
1888 found++;
1891 /* update leaf slen */
1892 n->slen = slen;
1894 if (hlist_empty(&n->leaf)) {
1895 put_child_root(pn, n->key, NULL);
1896 node_free(n);
1900 pr_debug("trie_flush found=%d\n", found);
1901 return found;
1904 static void fib_leaf_notify(struct net *net, struct key_vector *l,
1905 struct fib_table *tb, struct notifier_block *nb)
1907 struct fib_alias *fa;
1909 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1910 struct fib_info *fi = fa->fa_info;
1912 if (!fi)
1913 continue;
1915 /* local and main table can share the same trie,
1916 * so don't notify twice for the same entry.
1918 if (tb->tb_id != fa->tb_id)
1919 continue;
1921 call_fib_entry_notifier(nb, net, FIB_EVENT_ENTRY_ADD, l->key,
1922 KEYLENGTH - fa->fa_slen, fi, fa->fa_tos,
1923 fa->fa_type, fa->tb_id);
1927 static void fib_table_notify(struct net *net, struct fib_table *tb,
1928 struct notifier_block *nb)
1930 struct trie *t = (struct trie *)tb->tb_data;
1931 struct key_vector *l, *tp = t->kv;
1932 t_key key = 0;
1934 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1935 fib_leaf_notify(net, l, tb, nb);
1937 key = l->key + 1;
1938 /* stop in case of wrap around */
1939 if (key < l->key)
1940 break;
1944 void fib_notify(struct net *net, struct notifier_block *nb)
1946 unsigned int h;
1948 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
1949 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
1950 struct fib_table *tb;
1952 hlist_for_each_entry_rcu(tb, head, tb_hlist)
1953 fib_table_notify(net, tb, nb);
1957 static void __trie_free_rcu(struct rcu_head *head)
1959 struct fib_table *tb = container_of(head, struct fib_table, rcu);
1960 #ifdef CONFIG_IP_FIB_TRIE_STATS
1961 struct trie *t = (struct trie *)tb->tb_data;
1963 if (tb->tb_data == tb->__data)
1964 free_percpu(t->stats);
1965 #endif /* CONFIG_IP_FIB_TRIE_STATS */
1966 kfree(tb);
1969 void fib_free_table(struct fib_table *tb)
1971 call_rcu(&tb->rcu, __trie_free_rcu);
1974 static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
1975 struct sk_buff *skb, struct netlink_callback *cb)
1977 __be32 xkey = htonl(l->key);
1978 struct fib_alias *fa;
1979 int i, s_i;
1981 s_i = cb->args[4];
1982 i = 0;
1984 /* rcu_read_lock is hold by caller */
1985 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1986 if (i < s_i) {
1987 i++;
1988 continue;
1991 if (tb->tb_id != fa->tb_id) {
1992 i++;
1993 continue;
1996 if (fib_dump_info(skb, NETLINK_CB(cb->skb).portid,
1997 cb->nlh->nlmsg_seq,
1998 RTM_NEWROUTE,
1999 tb->tb_id,
2000 fa->fa_type,
2001 xkey,
2002 KEYLENGTH - fa->fa_slen,
2003 fa->fa_tos,
2004 fa->fa_info, NLM_F_MULTI) < 0) {
2005 cb->args[4] = i;
2006 return -1;
2008 i++;
2011 cb->args[4] = i;
2012 return skb->len;
2015 /* rcu_read_lock needs to be hold by caller from readside */
2016 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
2017 struct netlink_callback *cb)
2019 struct trie *t = (struct trie *)tb->tb_data;
2020 struct key_vector *l, *tp = t->kv;
2021 /* Dump starting at last key.
2022 * Note: 0.0.0.0/0 (ie default) is first key.
2024 int count = cb->args[2];
2025 t_key key = cb->args[3];
2027 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2028 if (fn_trie_dump_leaf(l, tb, skb, cb) < 0) {
2029 cb->args[3] = key;
2030 cb->args[2] = count;
2031 return -1;
2034 ++count;
2035 key = l->key + 1;
2037 memset(&cb->args[4], 0,
2038 sizeof(cb->args) - 4*sizeof(cb->args[0]));
2040 /* stop loop if key wrapped back to 0 */
2041 if (key < l->key)
2042 break;
2045 cb->args[3] = key;
2046 cb->args[2] = count;
2048 return skb->len;
2051 void __init fib_trie_init(void)
2053 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
2054 sizeof(struct fib_alias),
2055 0, SLAB_PANIC, NULL);
2057 trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
2058 LEAF_SIZE,
2059 0, SLAB_PANIC, NULL);
2062 struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
2064 struct fib_table *tb;
2065 struct trie *t;
2066 size_t sz = sizeof(*tb);
2068 if (!alias)
2069 sz += sizeof(struct trie);
2071 tb = kzalloc(sz, GFP_KERNEL);
2072 if (!tb)
2073 return NULL;
2075 tb->tb_id = id;
2076 tb->tb_num_default = 0;
2077 tb->tb_data = (alias ? alias->__data : tb->__data);
2079 if (alias)
2080 return tb;
2082 t = (struct trie *) tb->tb_data;
2083 t->kv[0].pos = KEYLENGTH;
2084 t->kv[0].slen = KEYLENGTH;
2085 #ifdef CONFIG_IP_FIB_TRIE_STATS
2086 t->stats = alloc_percpu(struct trie_use_stats);
2087 if (!t->stats) {
2088 kfree(tb);
2089 tb = NULL;
2091 #endif
2093 return tb;
2096 #ifdef CONFIG_PROC_FS
2097 /* Depth first Trie walk iterator */
2098 struct fib_trie_iter {
2099 struct seq_net_private p;
2100 struct fib_table *tb;
2101 struct key_vector *tnode;
2102 unsigned int index;
2103 unsigned int depth;
2106 static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2108 unsigned long cindex = iter->index;
2109 struct key_vector *pn = iter->tnode;
2110 t_key pkey;
2112 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2113 iter->tnode, iter->index, iter->depth);
2115 while (!IS_TRIE(pn)) {
2116 while (cindex < child_length(pn)) {
2117 struct key_vector *n = get_child_rcu(pn, cindex++);
2119 if (!n)
2120 continue;
2122 if (IS_LEAF(n)) {
2123 iter->tnode = pn;
2124 iter->index = cindex;
2125 } else {
2126 /* push down one level */
2127 iter->tnode = n;
2128 iter->index = 0;
2129 ++iter->depth;
2132 return n;
2135 /* Current node exhausted, pop back up */
2136 pkey = pn->key;
2137 pn = node_parent_rcu(pn);
2138 cindex = get_index(pkey, pn) + 1;
2139 --iter->depth;
2142 /* record root node so further searches know we are done */
2143 iter->tnode = pn;
2144 iter->index = 0;
2146 return NULL;
2149 static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2150 struct trie *t)
2152 struct key_vector *n, *pn;
2154 if (!t)
2155 return NULL;
2157 pn = t->kv;
2158 n = rcu_dereference(pn->tnode[0]);
2159 if (!n)
2160 return NULL;
2162 if (IS_TNODE(n)) {
2163 iter->tnode = n;
2164 iter->index = 0;
2165 iter->depth = 1;
2166 } else {
2167 iter->tnode = pn;
2168 iter->index = 0;
2169 iter->depth = 0;
2172 return n;
2175 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2177 struct key_vector *n;
2178 struct fib_trie_iter iter;
2180 memset(s, 0, sizeof(*s));
2182 rcu_read_lock();
2183 for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2184 if (IS_LEAF(n)) {
2185 struct fib_alias *fa;
2187 s->leaves++;
2188 s->totdepth += iter.depth;
2189 if (iter.depth > s->maxdepth)
2190 s->maxdepth = iter.depth;
2192 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2193 ++s->prefixes;
2194 } else {
2195 s->tnodes++;
2196 if (n->bits < MAX_STAT_DEPTH)
2197 s->nodesizes[n->bits]++;
2198 s->nullpointers += tn_info(n)->empty_children;
2201 rcu_read_unlock();
2205 * This outputs /proc/net/fib_triestats
2207 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2209 unsigned int i, max, pointers, bytes, avdepth;
2211 if (stat->leaves)
2212 avdepth = stat->totdepth*100 / stat->leaves;
2213 else
2214 avdepth = 0;
2216 seq_printf(seq, "\tAver depth: %u.%02d\n",
2217 avdepth / 100, avdepth % 100);
2218 seq_printf(seq, "\tMax depth: %u\n", stat->maxdepth);
2220 seq_printf(seq, "\tLeaves: %u\n", stat->leaves);
2221 bytes = LEAF_SIZE * stat->leaves;
2223 seq_printf(seq, "\tPrefixes: %u\n", stat->prefixes);
2224 bytes += sizeof(struct fib_alias) * stat->prefixes;
2226 seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2227 bytes += TNODE_SIZE(0) * stat->tnodes;
2229 max = MAX_STAT_DEPTH;
2230 while (max > 0 && stat->nodesizes[max-1] == 0)
2231 max--;
2233 pointers = 0;
2234 for (i = 1; i < max; i++)
2235 if (stat->nodesizes[i] != 0) {
2236 seq_printf(seq, " %u: %u", i, stat->nodesizes[i]);
2237 pointers += (1<<i) * stat->nodesizes[i];
2239 seq_putc(seq, '\n');
2240 seq_printf(seq, "\tPointers: %u\n", pointers);
2242 bytes += sizeof(struct key_vector *) * pointers;
2243 seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2244 seq_printf(seq, "Total size: %u kB\n", (bytes + 1023) / 1024);
2247 #ifdef CONFIG_IP_FIB_TRIE_STATS
2248 static void trie_show_usage(struct seq_file *seq,
2249 const struct trie_use_stats __percpu *stats)
2251 struct trie_use_stats s = { 0 };
2252 int cpu;
2254 /* loop through all of the CPUs and gather up the stats */
2255 for_each_possible_cpu(cpu) {
2256 const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2258 s.gets += pcpu->gets;
2259 s.backtrack += pcpu->backtrack;
2260 s.semantic_match_passed += pcpu->semantic_match_passed;
2261 s.semantic_match_miss += pcpu->semantic_match_miss;
2262 s.null_node_hit += pcpu->null_node_hit;
2263 s.resize_node_skipped += pcpu->resize_node_skipped;
2266 seq_printf(seq, "\nCounters:\n---------\n");
2267 seq_printf(seq, "gets = %u\n", s.gets);
2268 seq_printf(seq, "backtracks = %u\n", s.backtrack);
2269 seq_printf(seq, "semantic match passed = %u\n",
2270 s.semantic_match_passed);
2271 seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2272 seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2273 seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2275 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2277 static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2279 if (tb->tb_id == RT_TABLE_LOCAL)
2280 seq_puts(seq, "Local:\n");
2281 else if (tb->tb_id == RT_TABLE_MAIN)
2282 seq_puts(seq, "Main:\n");
2283 else
2284 seq_printf(seq, "Id %d:\n", tb->tb_id);
2288 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2290 struct net *net = (struct net *)seq->private;
2291 unsigned int h;
2293 seq_printf(seq,
2294 "Basic info: size of leaf:"
2295 " %zd bytes, size of tnode: %zd bytes.\n",
2296 LEAF_SIZE, TNODE_SIZE(0));
2298 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2299 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2300 struct fib_table *tb;
2302 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2303 struct trie *t = (struct trie *) tb->tb_data;
2304 struct trie_stat stat;
2306 if (!t)
2307 continue;
2309 fib_table_print(seq, tb);
2311 trie_collect_stats(t, &stat);
2312 trie_show_stats(seq, &stat);
2313 #ifdef CONFIG_IP_FIB_TRIE_STATS
2314 trie_show_usage(seq, t->stats);
2315 #endif
2319 return 0;
2322 static int fib_triestat_seq_open(struct inode *inode, struct file *file)
2324 return single_open_net(inode, file, fib_triestat_seq_show);
2327 static const struct file_operations fib_triestat_fops = {
2328 .owner = THIS_MODULE,
2329 .open = fib_triestat_seq_open,
2330 .read = seq_read,
2331 .llseek = seq_lseek,
2332 .release = single_release_net,
2335 static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2337 struct fib_trie_iter *iter = seq->private;
2338 struct net *net = seq_file_net(seq);
2339 loff_t idx = 0;
2340 unsigned int h;
2342 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2343 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2344 struct fib_table *tb;
2346 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2347 struct key_vector *n;
2349 for (n = fib_trie_get_first(iter,
2350 (struct trie *) tb->tb_data);
2351 n; n = fib_trie_get_next(iter))
2352 if (pos == idx++) {
2353 iter->tb = tb;
2354 return n;
2359 return NULL;
2362 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2363 __acquires(RCU)
2365 rcu_read_lock();
2366 return fib_trie_get_idx(seq, *pos);
2369 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2371 struct fib_trie_iter *iter = seq->private;
2372 struct net *net = seq_file_net(seq);
2373 struct fib_table *tb = iter->tb;
2374 struct hlist_node *tb_node;
2375 unsigned int h;
2376 struct key_vector *n;
2378 ++*pos;
2379 /* next node in same table */
2380 n = fib_trie_get_next(iter);
2381 if (n)
2382 return n;
2384 /* walk rest of this hash chain */
2385 h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2386 while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2387 tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2388 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2389 if (n)
2390 goto found;
2393 /* new hash chain */
2394 while (++h < FIB_TABLE_HASHSZ) {
2395 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2396 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2397 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2398 if (n)
2399 goto found;
2402 return NULL;
2404 found:
2405 iter->tb = tb;
2406 return n;
2409 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2410 __releases(RCU)
2412 rcu_read_unlock();
2415 static void seq_indent(struct seq_file *seq, int n)
2417 while (n-- > 0)
2418 seq_puts(seq, " ");
2421 static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2423 switch (s) {
2424 case RT_SCOPE_UNIVERSE: return "universe";
2425 case RT_SCOPE_SITE: return "site";
2426 case RT_SCOPE_LINK: return "link";
2427 case RT_SCOPE_HOST: return "host";
2428 case RT_SCOPE_NOWHERE: return "nowhere";
2429 default:
2430 snprintf(buf, len, "scope=%d", s);
2431 return buf;
2435 static const char *const rtn_type_names[__RTN_MAX] = {
2436 [RTN_UNSPEC] = "UNSPEC",
2437 [RTN_UNICAST] = "UNICAST",
2438 [RTN_LOCAL] = "LOCAL",
2439 [RTN_BROADCAST] = "BROADCAST",
2440 [RTN_ANYCAST] = "ANYCAST",
2441 [RTN_MULTICAST] = "MULTICAST",
2442 [RTN_BLACKHOLE] = "BLACKHOLE",
2443 [RTN_UNREACHABLE] = "UNREACHABLE",
2444 [RTN_PROHIBIT] = "PROHIBIT",
2445 [RTN_THROW] = "THROW",
2446 [RTN_NAT] = "NAT",
2447 [RTN_XRESOLVE] = "XRESOLVE",
2450 static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2452 if (t < __RTN_MAX && rtn_type_names[t])
2453 return rtn_type_names[t];
2454 snprintf(buf, len, "type %u", t);
2455 return buf;
2458 /* Pretty print the trie */
2459 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2461 const struct fib_trie_iter *iter = seq->private;
2462 struct key_vector *n = v;
2464 if (IS_TRIE(node_parent_rcu(n)))
2465 fib_table_print(seq, iter->tb);
2467 if (IS_TNODE(n)) {
2468 __be32 prf = htonl(n->key);
2470 seq_indent(seq, iter->depth-1);
2471 seq_printf(seq, " +-- %pI4/%zu %u %u %u\n",
2472 &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2473 tn_info(n)->full_children,
2474 tn_info(n)->empty_children);
2475 } else {
2476 __be32 val = htonl(n->key);
2477 struct fib_alias *fa;
2479 seq_indent(seq, iter->depth);
2480 seq_printf(seq, " |-- %pI4\n", &val);
2482 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2483 char buf1[32], buf2[32];
2485 seq_indent(seq, iter->depth + 1);
2486 seq_printf(seq, " /%zu %s %s",
2487 KEYLENGTH - fa->fa_slen,
2488 rtn_scope(buf1, sizeof(buf1),
2489 fa->fa_info->fib_scope),
2490 rtn_type(buf2, sizeof(buf2),
2491 fa->fa_type));
2492 if (fa->fa_tos)
2493 seq_printf(seq, " tos=%d", fa->fa_tos);
2494 seq_putc(seq, '\n');
2498 return 0;
2501 static const struct seq_operations fib_trie_seq_ops = {
2502 .start = fib_trie_seq_start,
2503 .next = fib_trie_seq_next,
2504 .stop = fib_trie_seq_stop,
2505 .show = fib_trie_seq_show,
2508 static int fib_trie_seq_open(struct inode *inode, struct file *file)
2510 return seq_open_net(inode, file, &fib_trie_seq_ops,
2511 sizeof(struct fib_trie_iter));
2514 static const struct file_operations fib_trie_fops = {
2515 .owner = THIS_MODULE,
2516 .open = fib_trie_seq_open,
2517 .read = seq_read,
2518 .llseek = seq_lseek,
2519 .release = seq_release_net,
2522 struct fib_route_iter {
2523 struct seq_net_private p;
2524 struct fib_table *main_tb;
2525 struct key_vector *tnode;
2526 loff_t pos;
2527 t_key key;
2530 static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2531 loff_t pos)
2533 struct key_vector *l, **tp = &iter->tnode;
2534 t_key key;
2536 /* use cached location of previously found key */
2537 if (iter->pos > 0 && pos >= iter->pos) {
2538 key = iter->key;
2539 } else {
2540 iter->pos = 1;
2541 key = 0;
2544 pos -= iter->pos;
2546 while ((l = leaf_walk_rcu(tp, key)) && (pos-- > 0)) {
2547 key = l->key + 1;
2548 iter->pos++;
2549 l = NULL;
2551 /* handle unlikely case of a key wrap */
2552 if (!key)
2553 break;
2556 if (l)
2557 iter->key = l->key; /* remember it */
2558 else
2559 iter->pos = 0; /* forget it */
2561 return l;
2564 static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2565 __acquires(RCU)
2567 struct fib_route_iter *iter = seq->private;
2568 struct fib_table *tb;
2569 struct trie *t;
2571 rcu_read_lock();
2573 tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2574 if (!tb)
2575 return NULL;
2577 iter->main_tb = tb;
2578 t = (struct trie *)tb->tb_data;
2579 iter->tnode = t->kv;
2581 if (*pos != 0)
2582 return fib_route_get_idx(iter, *pos);
2584 iter->pos = 0;
2585 iter->key = KEY_MAX;
2587 return SEQ_START_TOKEN;
2590 static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2592 struct fib_route_iter *iter = seq->private;
2593 struct key_vector *l = NULL;
2594 t_key key = iter->key + 1;
2596 ++*pos;
2598 /* only allow key of 0 for start of sequence */
2599 if ((v == SEQ_START_TOKEN) || key)
2600 l = leaf_walk_rcu(&iter->tnode, key);
2602 if (l) {
2603 iter->key = l->key;
2604 iter->pos++;
2605 } else {
2606 iter->pos = 0;
2609 return l;
2612 static void fib_route_seq_stop(struct seq_file *seq, void *v)
2613 __releases(RCU)
2615 rcu_read_unlock();
2618 static unsigned int fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2620 unsigned int flags = 0;
2622 if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2623 flags = RTF_REJECT;
2624 if (fi && fi->fib_nh->nh_gw)
2625 flags |= RTF_GATEWAY;
2626 if (mask == htonl(0xFFFFFFFF))
2627 flags |= RTF_HOST;
2628 flags |= RTF_UP;
2629 return flags;
2633 * This outputs /proc/net/route.
2634 * The format of the file is not supposed to be changed
2635 * and needs to be same as fib_hash output to avoid breaking
2636 * legacy utilities
2638 static int fib_route_seq_show(struct seq_file *seq, void *v)
2640 struct fib_route_iter *iter = seq->private;
2641 struct fib_table *tb = iter->main_tb;
2642 struct fib_alias *fa;
2643 struct key_vector *l = v;
2644 __be32 prefix;
2646 if (v == SEQ_START_TOKEN) {
2647 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2648 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2649 "\tWindow\tIRTT");
2650 return 0;
2653 prefix = htonl(l->key);
2655 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2656 const struct fib_info *fi = fa->fa_info;
2657 __be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2658 unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2660 if ((fa->fa_type == RTN_BROADCAST) ||
2661 (fa->fa_type == RTN_MULTICAST))
2662 continue;
2664 if (fa->tb_id != tb->tb_id)
2665 continue;
2667 seq_setwidth(seq, 127);
2669 if (fi)
2670 seq_printf(seq,
2671 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2672 "%d\t%08X\t%d\t%u\t%u",
2673 fi->fib_dev ? fi->fib_dev->name : "*",
2674 prefix,
2675 fi->fib_nh->nh_gw, flags, 0, 0,
2676 fi->fib_priority,
2677 mask,
2678 (fi->fib_advmss ?
2679 fi->fib_advmss + 40 : 0),
2680 fi->fib_window,
2681 fi->fib_rtt >> 3);
2682 else
2683 seq_printf(seq,
2684 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2685 "%d\t%08X\t%d\t%u\t%u",
2686 prefix, 0, flags, 0, 0, 0,
2687 mask, 0, 0, 0);
2689 seq_pad(seq, '\n');
2692 return 0;
2695 static const struct seq_operations fib_route_seq_ops = {
2696 .start = fib_route_seq_start,
2697 .next = fib_route_seq_next,
2698 .stop = fib_route_seq_stop,
2699 .show = fib_route_seq_show,
2702 static int fib_route_seq_open(struct inode *inode, struct file *file)
2704 return seq_open_net(inode, file, &fib_route_seq_ops,
2705 sizeof(struct fib_route_iter));
2708 static const struct file_operations fib_route_fops = {
2709 .owner = THIS_MODULE,
2710 .open = fib_route_seq_open,
2711 .read = seq_read,
2712 .llseek = seq_lseek,
2713 .release = seq_release_net,
2716 int __net_init fib_proc_init(struct net *net)
2718 if (!proc_create("fib_trie", S_IRUGO, net->proc_net, &fib_trie_fops))
2719 goto out1;
2721 if (!proc_create("fib_triestat", S_IRUGO, net->proc_net,
2722 &fib_triestat_fops))
2723 goto out2;
2725 if (!proc_create("route", S_IRUGO, net->proc_net, &fib_route_fops))
2726 goto out3;
2728 return 0;
2730 out3:
2731 remove_proc_entry("fib_triestat", net->proc_net);
2732 out2:
2733 remove_proc_entry("fib_trie", net->proc_net);
2734 out1:
2735 return -ENOMEM;
2738 void __net_exit fib_proc_exit(struct net *net)
2740 remove_proc_entry("fib_trie", net->proc_net);
2741 remove_proc_entry("fib_triestat", net->proc_net);
2742 remove_proc_entry("route", net->proc_net);
2745 #endif /* CONFIG_PROC_FS */