ping: implement proper locking
[linux/fpc-iii.git] / net / ipv4 / fib_trie.c
blobe3665bf7a7f373cfb2cea7a57a9ba478b9cf11df
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 <asm/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 BLOCKING_NOTIFIER_HEAD(fib_chain);
89 int register_fib_notifier(struct notifier_block *nb)
91 return blocking_notifier_chain_register(&fib_chain, nb);
93 EXPORT_SYMBOL(register_fib_notifier);
95 int unregister_fib_notifier(struct notifier_block *nb)
97 return blocking_notifier_chain_unregister(&fib_chain, nb);
99 EXPORT_SYMBOL(unregister_fib_notifier);
101 int call_fib_notifiers(struct net *net, enum fib_event_type event_type,
102 struct fib_notifier_info *info)
104 info->net = net;
105 return blocking_notifier_call_chain(&fib_chain, event_type, info);
108 static int call_fib_entry_notifiers(struct net *net,
109 enum fib_event_type event_type, u32 dst,
110 int dst_len, struct fib_info *fi,
111 u8 tos, u8 type, u32 tb_id, u32 nlflags)
113 struct fib_entry_notifier_info info = {
114 .dst = dst,
115 .dst_len = dst_len,
116 .fi = fi,
117 .tos = tos,
118 .type = type,
119 .tb_id = tb_id,
120 .nlflags = nlflags,
122 return call_fib_notifiers(net, event_type, &info.info);
125 #define MAX_STAT_DEPTH 32
127 #define KEYLENGTH (8*sizeof(t_key))
128 #define KEY_MAX ((t_key)~0)
130 typedef unsigned int t_key;
132 #define IS_TRIE(n) ((n)->pos >= KEYLENGTH)
133 #define IS_TNODE(n) ((n)->bits)
134 #define IS_LEAF(n) (!(n)->bits)
136 struct key_vector {
137 t_key key;
138 unsigned char pos; /* 2log(KEYLENGTH) bits needed */
139 unsigned char bits; /* 2log(KEYLENGTH) bits needed */
140 unsigned char slen;
141 union {
142 /* This list pointer if valid if (pos | bits) == 0 (LEAF) */
143 struct hlist_head leaf;
144 /* This array is valid if (pos | bits) > 0 (TNODE) */
145 struct key_vector __rcu *tnode[0];
149 struct tnode {
150 struct rcu_head rcu;
151 t_key empty_children; /* KEYLENGTH bits needed */
152 t_key full_children; /* KEYLENGTH bits needed */
153 struct key_vector __rcu *parent;
154 struct key_vector kv[1];
155 #define tn_bits kv[0].bits
158 #define TNODE_SIZE(n) offsetof(struct tnode, kv[0].tnode[n])
159 #define LEAF_SIZE TNODE_SIZE(1)
161 #ifdef CONFIG_IP_FIB_TRIE_STATS
162 struct trie_use_stats {
163 unsigned int gets;
164 unsigned int backtrack;
165 unsigned int semantic_match_passed;
166 unsigned int semantic_match_miss;
167 unsigned int null_node_hit;
168 unsigned int resize_node_skipped;
170 #endif
172 struct trie_stat {
173 unsigned int totdepth;
174 unsigned int maxdepth;
175 unsigned int tnodes;
176 unsigned int leaves;
177 unsigned int nullpointers;
178 unsigned int prefixes;
179 unsigned int nodesizes[MAX_STAT_DEPTH];
182 struct trie {
183 struct key_vector kv[1];
184 #ifdef CONFIG_IP_FIB_TRIE_STATS
185 struct trie_use_stats __percpu *stats;
186 #endif
189 static struct key_vector *resize(struct trie *t, struct key_vector *tn);
190 static size_t tnode_free_size;
193 * synchronize_rcu after call_rcu for that many pages; it should be especially
194 * useful before resizing the root node with PREEMPT_NONE configs; the value was
195 * obtained experimentally, aiming to avoid visible slowdown.
197 static const int sync_pages = 128;
199 static struct kmem_cache *fn_alias_kmem __read_mostly;
200 static struct kmem_cache *trie_leaf_kmem __read_mostly;
202 static inline struct tnode *tn_info(struct key_vector *kv)
204 return container_of(kv, struct tnode, kv[0]);
207 /* caller must hold RTNL */
208 #define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
209 #define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
211 /* caller must hold RCU read lock or RTNL */
212 #define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
213 #define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
215 /* wrapper for rcu_assign_pointer */
216 static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
218 if (n)
219 rcu_assign_pointer(tn_info(n)->parent, tp);
222 #define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
224 /* This provides us with the number of children in this node, in the case of a
225 * leaf this will return 0 meaning none of the children are accessible.
227 static inline unsigned long child_length(const struct key_vector *tn)
229 return (1ul << tn->bits) & ~(1ul);
232 #define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
234 static inline unsigned long get_index(t_key key, struct key_vector *kv)
236 unsigned long index = key ^ kv->key;
238 if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
239 return 0;
241 return index >> kv->pos;
244 /* To understand this stuff, an understanding of keys and all their bits is
245 * necessary. Every node in the trie has a key associated with it, but not
246 * all of the bits in that key are significant.
248 * Consider a node 'n' and its parent 'tp'.
250 * If n is a leaf, every bit in its key is significant. Its presence is
251 * necessitated by path compression, since during a tree traversal (when
252 * searching for a leaf - unless we are doing an insertion) we will completely
253 * ignore all skipped bits we encounter. Thus we need to verify, at the end of
254 * a potentially successful search, that we have indeed been walking the
255 * correct key path.
257 * Note that we can never "miss" the correct key in the tree if present by
258 * following the wrong path. Path compression ensures that segments of the key
259 * that are the same for all keys with a given prefix are skipped, but the
260 * skipped part *is* identical for each node in the subtrie below the skipped
261 * bit! trie_insert() in this implementation takes care of that.
263 * if n is an internal node - a 'tnode' here, the various parts of its key
264 * have many different meanings.
266 * Example:
267 * _________________________________________________________________
268 * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
269 * -----------------------------------------------------------------
270 * 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16
272 * _________________________________________________________________
273 * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
274 * -----------------------------------------------------------------
275 * 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
277 * tp->pos = 22
278 * tp->bits = 3
279 * n->pos = 13
280 * n->bits = 4
282 * First, let's just ignore the bits that come before the parent tp, that is
283 * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
284 * point we do not use them for anything.
286 * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
287 * index into the parent's child array. That is, they will be used to find
288 * 'n' among tp's children.
290 * The bits from (n->pos + n->bits) to (tp->pos - 1) - "S" - are skipped bits
291 * for the node n.
293 * All the bits we have seen so far are significant to the node n. The rest
294 * of the bits are really not needed or indeed known in n->key.
296 * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
297 * n's child array, and will of course be different for each child.
299 * The rest of the bits, from 0 to (n->pos -1) - "u" - are completely unknown
300 * at this point.
303 static const int halve_threshold = 25;
304 static const int inflate_threshold = 50;
305 static const int halve_threshold_root = 15;
306 static const int inflate_threshold_root = 30;
308 static void __alias_free_mem(struct rcu_head *head)
310 struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
311 kmem_cache_free(fn_alias_kmem, fa);
314 static inline void alias_free_mem_rcu(struct fib_alias *fa)
316 call_rcu(&fa->rcu, __alias_free_mem);
319 #define TNODE_KMALLOC_MAX \
320 ilog2((PAGE_SIZE - TNODE_SIZE(0)) / sizeof(struct key_vector *))
321 #define TNODE_VMALLOC_MAX \
322 ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
324 static void __node_free_rcu(struct rcu_head *head)
326 struct tnode *n = container_of(head, struct tnode, rcu);
328 if (!n->tn_bits)
329 kmem_cache_free(trie_leaf_kmem, n);
330 else
331 kvfree(n);
334 #define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
336 static struct tnode *tnode_alloc(int bits)
338 size_t size;
340 /* verify bits is within bounds */
341 if (bits > TNODE_VMALLOC_MAX)
342 return NULL;
344 /* determine size and verify it is non-zero and didn't overflow */
345 size = TNODE_SIZE(1ul << bits);
347 if (size <= PAGE_SIZE)
348 return kzalloc(size, GFP_KERNEL);
349 else
350 return vzalloc(size);
353 static inline void empty_child_inc(struct key_vector *n)
355 ++tn_info(n)->empty_children ? : ++tn_info(n)->full_children;
358 static inline void empty_child_dec(struct key_vector *n)
360 tn_info(n)->empty_children-- ? : tn_info(n)->full_children--;
363 static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
365 struct key_vector *l;
366 struct tnode *kv;
368 kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
369 if (!kv)
370 return NULL;
372 /* initialize key vector */
373 l = kv->kv;
374 l->key = key;
375 l->pos = 0;
376 l->bits = 0;
377 l->slen = fa->fa_slen;
379 /* link leaf to fib alias */
380 INIT_HLIST_HEAD(&l->leaf);
381 hlist_add_head(&fa->fa_list, &l->leaf);
383 return l;
386 static struct key_vector *tnode_new(t_key key, int pos, int bits)
388 unsigned int shift = pos + bits;
389 struct key_vector *tn;
390 struct tnode *tnode;
392 /* verify bits and pos their msb bits clear and values are valid */
393 BUG_ON(!bits || (shift > KEYLENGTH));
395 tnode = tnode_alloc(bits);
396 if (!tnode)
397 return NULL;
399 pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
400 sizeof(struct key_vector *) << bits);
402 if (bits == KEYLENGTH)
403 tnode->full_children = 1;
404 else
405 tnode->empty_children = 1ul << bits;
407 tn = tnode->kv;
408 tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
409 tn->pos = pos;
410 tn->bits = bits;
411 tn->slen = pos;
413 return tn;
416 /* Check whether a tnode 'n' is "full", i.e. it is an internal node
417 * and no bits are skipped. See discussion in dyntree paper p. 6
419 static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
421 return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
424 /* Add a child at position i overwriting the old value.
425 * Update the value of full_children and empty_children.
427 static void put_child(struct key_vector *tn, unsigned long i,
428 struct key_vector *n)
430 struct key_vector *chi = get_child(tn, i);
431 int isfull, wasfull;
433 BUG_ON(i >= child_length(tn));
435 /* update emptyChildren, overflow into fullChildren */
436 if (!n && chi)
437 empty_child_inc(tn);
438 if (n && !chi)
439 empty_child_dec(tn);
441 /* update fullChildren */
442 wasfull = tnode_full(tn, chi);
443 isfull = tnode_full(tn, n);
445 if (wasfull && !isfull)
446 tn_info(tn)->full_children--;
447 else if (!wasfull && isfull)
448 tn_info(tn)->full_children++;
450 if (n && (tn->slen < n->slen))
451 tn->slen = n->slen;
453 rcu_assign_pointer(tn->tnode[i], n);
456 static void update_children(struct key_vector *tn)
458 unsigned long i;
460 /* update all of the child parent pointers */
461 for (i = child_length(tn); i;) {
462 struct key_vector *inode = get_child(tn, --i);
464 if (!inode)
465 continue;
467 /* Either update the children of a tnode that
468 * already belongs to us or update the child
469 * to point to ourselves.
471 if (node_parent(inode) == tn)
472 update_children(inode);
473 else
474 node_set_parent(inode, tn);
478 static inline void put_child_root(struct key_vector *tp, t_key key,
479 struct key_vector *n)
481 if (IS_TRIE(tp))
482 rcu_assign_pointer(tp->tnode[0], n);
483 else
484 put_child(tp, get_index(key, tp), n);
487 static inline void tnode_free_init(struct key_vector *tn)
489 tn_info(tn)->rcu.next = NULL;
492 static inline void tnode_free_append(struct key_vector *tn,
493 struct key_vector *n)
495 tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
496 tn_info(tn)->rcu.next = &tn_info(n)->rcu;
499 static void tnode_free(struct key_vector *tn)
501 struct callback_head *head = &tn_info(tn)->rcu;
503 while (head) {
504 head = head->next;
505 tnode_free_size += TNODE_SIZE(1ul << tn->bits);
506 node_free(tn);
508 tn = container_of(head, struct tnode, rcu)->kv;
511 if (tnode_free_size >= PAGE_SIZE * sync_pages) {
512 tnode_free_size = 0;
513 synchronize_rcu();
517 static struct key_vector *replace(struct trie *t,
518 struct key_vector *oldtnode,
519 struct key_vector *tn)
521 struct key_vector *tp = node_parent(oldtnode);
522 unsigned long i;
524 /* setup the parent pointer out of and back into this node */
525 NODE_INIT_PARENT(tn, tp);
526 put_child_root(tp, tn->key, tn);
528 /* update all of the child parent pointers */
529 update_children(tn);
531 /* all pointers should be clean so we are done */
532 tnode_free(oldtnode);
534 /* resize children now that oldtnode is freed */
535 for (i = child_length(tn); i;) {
536 struct key_vector *inode = get_child(tn, --i);
538 /* resize child node */
539 if (tnode_full(tn, inode))
540 tn = resize(t, inode);
543 return tp;
546 static struct key_vector *inflate(struct trie *t,
547 struct key_vector *oldtnode)
549 struct key_vector *tn;
550 unsigned long i;
551 t_key m;
553 pr_debug("In inflate\n");
555 tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
556 if (!tn)
557 goto notnode;
559 /* prepare oldtnode to be freed */
560 tnode_free_init(oldtnode);
562 /* Assemble all of the pointers in our cluster, in this case that
563 * represents all of the pointers out of our allocated nodes that
564 * point to existing tnodes and the links between our allocated
565 * nodes.
567 for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
568 struct key_vector *inode = get_child(oldtnode, --i);
569 struct key_vector *node0, *node1;
570 unsigned long j, k;
572 /* An empty child */
573 if (!inode)
574 continue;
576 /* A leaf or an internal node with skipped bits */
577 if (!tnode_full(oldtnode, inode)) {
578 put_child(tn, get_index(inode->key, tn), inode);
579 continue;
582 /* drop the node in the old tnode free list */
583 tnode_free_append(oldtnode, inode);
585 /* An internal node with two children */
586 if (inode->bits == 1) {
587 put_child(tn, 2 * i + 1, get_child(inode, 1));
588 put_child(tn, 2 * i, get_child(inode, 0));
589 continue;
592 /* We will replace this node 'inode' with two new
593 * ones, 'node0' and 'node1', each with half of the
594 * original children. The two new nodes will have
595 * a position one bit further down the key and this
596 * means that the "significant" part of their keys
597 * (see the discussion near the top of this file)
598 * will differ by one bit, which will be "0" in
599 * node0's key and "1" in node1's key. Since we are
600 * moving the key position by one step, the bit that
601 * we are moving away from - the bit at position
602 * (tn->pos) - is the one that will differ between
603 * node0 and node1. So... we synthesize that bit in the
604 * two new keys.
606 node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
607 if (!node1)
608 goto nomem;
609 node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
611 tnode_free_append(tn, node1);
612 if (!node0)
613 goto nomem;
614 tnode_free_append(tn, node0);
616 /* populate child pointers in new nodes */
617 for (k = child_length(inode), j = k / 2; j;) {
618 put_child(node1, --j, get_child(inode, --k));
619 put_child(node0, j, get_child(inode, j));
620 put_child(node1, --j, get_child(inode, --k));
621 put_child(node0, j, get_child(inode, j));
624 /* link new nodes to parent */
625 NODE_INIT_PARENT(node1, tn);
626 NODE_INIT_PARENT(node0, tn);
628 /* link parent to nodes */
629 put_child(tn, 2 * i + 1, node1);
630 put_child(tn, 2 * i, node0);
633 /* setup the parent pointers into and out of this node */
634 return replace(t, oldtnode, tn);
635 nomem:
636 /* all pointers should be clean so we are done */
637 tnode_free(tn);
638 notnode:
639 return NULL;
642 static struct key_vector *halve(struct trie *t,
643 struct key_vector *oldtnode)
645 struct key_vector *tn;
646 unsigned long i;
648 pr_debug("In halve\n");
650 tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
651 if (!tn)
652 goto notnode;
654 /* prepare oldtnode to be freed */
655 tnode_free_init(oldtnode);
657 /* Assemble all of the pointers in our cluster, in this case that
658 * represents all of the pointers out of our allocated nodes that
659 * point to existing tnodes and the links between our allocated
660 * nodes.
662 for (i = child_length(oldtnode); i;) {
663 struct key_vector *node1 = get_child(oldtnode, --i);
664 struct key_vector *node0 = get_child(oldtnode, --i);
665 struct key_vector *inode;
667 /* At least one of the children is empty */
668 if (!node1 || !node0) {
669 put_child(tn, i / 2, node1 ? : node0);
670 continue;
673 /* Two nonempty children */
674 inode = tnode_new(node0->key, oldtnode->pos, 1);
675 if (!inode)
676 goto nomem;
677 tnode_free_append(tn, inode);
679 /* initialize pointers out of node */
680 put_child(inode, 1, node1);
681 put_child(inode, 0, node0);
682 NODE_INIT_PARENT(inode, tn);
684 /* link parent to node */
685 put_child(tn, i / 2, inode);
688 /* setup the parent pointers into and out of this node */
689 return replace(t, oldtnode, tn);
690 nomem:
691 /* all pointers should be clean so we are done */
692 tnode_free(tn);
693 notnode:
694 return NULL;
697 static struct key_vector *collapse(struct trie *t,
698 struct key_vector *oldtnode)
700 struct key_vector *n, *tp;
701 unsigned long i;
703 /* scan the tnode looking for that one child that might still exist */
704 for (n = NULL, i = child_length(oldtnode); !n && i;)
705 n = get_child(oldtnode, --i);
707 /* compress one level */
708 tp = node_parent(oldtnode);
709 put_child_root(tp, oldtnode->key, n);
710 node_set_parent(n, tp);
712 /* drop dead node */
713 node_free(oldtnode);
715 return tp;
718 static unsigned char update_suffix(struct key_vector *tn)
720 unsigned char slen = tn->pos;
721 unsigned long stride, i;
722 unsigned char slen_max;
724 /* only vector 0 can have a suffix length greater than or equal to
725 * tn->pos + tn->bits, the second highest node will have a suffix
726 * length at most of tn->pos + tn->bits - 1
728 slen_max = min_t(unsigned char, tn->pos + tn->bits - 1, tn->slen);
730 /* search though the list of children looking for nodes that might
731 * have a suffix greater than the one we currently have. This is
732 * why we start with a stride of 2 since a stride of 1 would
733 * represent the nodes with suffix length equal to tn->pos
735 for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
736 struct key_vector *n = get_child(tn, i);
738 if (!n || (n->slen <= slen))
739 continue;
741 /* update stride and slen based on new value */
742 stride <<= (n->slen - slen);
743 slen = n->slen;
744 i &= ~(stride - 1);
746 /* stop searching if we have hit the maximum possible value */
747 if (slen >= slen_max)
748 break;
751 tn->slen = slen;
753 return slen;
756 /* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
757 * the Helsinki University of Technology and Matti Tikkanen of Nokia
758 * Telecommunications, page 6:
759 * "A node is doubled if the ratio of non-empty children to all
760 * children in the *doubled* node is at least 'high'."
762 * 'high' in this instance is the variable 'inflate_threshold'. It
763 * is expressed as a percentage, so we multiply it with
764 * child_length() and instead of multiplying by 2 (since the
765 * child array will be doubled by inflate()) and multiplying
766 * the left-hand side by 100 (to handle the percentage thing) we
767 * multiply the left-hand side by 50.
769 * The left-hand side may look a bit weird: child_length(tn)
770 * - tn->empty_children is of course the number of non-null children
771 * in the current node. tn->full_children is the number of "full"
772 * children, that is non-null tnodes with a skip value of 0.
773 * All of those will be doubled in the resulting inflated tnode, so
774 * we just count them one extra time here.
776 * A clearer way to write this would be:
778 * to_be_doubled = tn->full_children;
779 * not_to_be_doubled = child_length(tn) - tn->empty_children -
780 * tn->full_children;
782 * new_child_length = child_length(tn) * 2;
784 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
785 * new_child_length;
786 * if (new_fill_factor >= inflate_threshold)
788 * ...and so on, tho it would mess up the while () loop.
790 * anyway,
791 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
792 * inflate_threshold
794 * avoid a division:
795 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
796 * inflate_threshold * new_child_length
798 * expand not_to_be_doubled and to_be_doubled, and shorten:
799 * 100 * (child_length(tn) - tn->empty_children +
800 * tn->full_children) >= inflate_threshold * new_child_length
802 * expand new_child_length:
803 * 100 * (child_length(tn) - tn->empty_children +
804 * tn->full_children) >=
805 * inflate_threshold * child_length(tn) * 2
807 * shorten again:
808 * 50 * (tn->full_children + child_length(tn) -
809 * tn->empty_children) >= inflate_threshold *
810 * child_length(tn)
813 static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
815 unsigned long used = child_length(tn);
816 unsigned long threshold = used;
818 /* Keep root node larger */
819 threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
820 used -= tn_info(tn)->empty_children;
821 used += tn_info(tn)->full_children;
823 /* if bits == KEYLENGTH then pos = 0, and will fail below */
825 return (used > 1) && tn->pos && ((50 * used) >= threshold);
828 static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
830 unsigned long used = child_length(tn);
831 unsigned long threshold = used;
833 /* Keep root node larger */
834 threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
835 used -= tn_info(tn)->empty_children;
837 /* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
839 return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
842 static inline bool should_collapse(struct key_vector *tn)
844 unsigned long used = child_length(tn);
846 used -= tn_info(tn)->empty_children;
848 /* account for bits == KEYLENGTH case */
849 if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
850 used -= KEY_MAX;
852 /* One child or none, time to drop us from the trie */
853 return used < 2;
856 #define MAX_WORK 10
857 static struct key_vector *resize(struct trie *t, struct key_vector *tn)
859 #ifdef CONFIG_IP_FIB_TRIE_STATS
860 struct trie_use_stats __percpu *stats = t->stats;
861 #endif
862 struct key_vector *tp = node_parent(tn);
863 unsigned long cindex = get_index(tn->key, tp);
864 int max_work = MAX_WORK;
866 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
867 tn, inflate_threshold, halve_threshold);
869 /* track the tnode via the pointer from the parent instead of
870 * doing it ourselves. This way we can let RCU fully do its
871 * thing without us interfering
873 BUG_ON(tn != get_child(tp, cindex));
875 /* Double as long as the resulting node has a number of
876 * nonempty nodes that are above the threshold.
878 while (should_inflate(tp, tn) && max_work) {
879 tp = inflate(t, tn);
880 if (!tp) {
881 #ifdef CONFIG_IP_FIB_TRIE_STATS
882 this_cpu_inc(stats->resize_node_skipped);
883 #endif
884 break;
887 max_work--;
888 tn = get_child(tp, cindex);
891 /* update parent in case inflate failed */
892 tp = node_parent(tn);
894 /* Return if at least one inflate is run */
895 if (max_work != MAX_WORK)
896 return tp;
898 /* Halve as long as the number of empty children in this
899 * node is above threshold.
901 while (should_halve(tp, tn) && max_work) {
902 tp = halve(t, tn);
903 if (!tp) {
904 #ifdef CONFIG_IP_FIB_TRIE_STATS
905 this_cpu_inc(stats->resize_node_skipped);
906 #endif
907 break;
910 max_work--;
911 tn = get_child(tp, cindex);
914 /* Only one child remains */
915 if (should_collapse(tn))
916 return collapse(t, tn);
918 /* update parent in case halve failed */
919 return node_parent(tn);
922 static void node_pull_suffix(struct key_vector *tn, unsigned char slen)
924 unsigned char node_slen = tn->slen;
926 while ((node_slen > tn->pos) && (node_slen > slen)) {
927 slen = update_suffix(tn);
928 if (node_slen == slen)
929 break;
931 tn = node_parent(tn);
932 node_slen = tn->slen;
936 static void node_push_suffix(struct key_vector *tn, unsigned char slen)
938 while (tn->slen < slen) {
939 tn->slen = slen;
940 tn = node_parent(tn);
944 /* rcu_read_lock needs to be hold by caller from readside */
945 static struct key_vector *fib_find_node(struct trie *t,
946 struct key_vector **tp, u32 key)
948 struct key_vector *pn, *n = t->kv;
949 unsigned long index = 0;
951 do {
952 pn = n;
953 n = get_child_rcu(n, index);
955 if (!n)
956 break;
958 index = get_cindex(key, n);
960 /* This bit of code is a bit tricky but it combines multiple
961 * checks into a single check. The prefix consists of the
962 * prefix plus zeros for the bits in the cindex. The index
963 * is the difference between the key and this value. From
964 * this we can actually derive several pieces of data.
965 * if (index >= (1ul << bits))
966 * we have a mismatch in skip bits and failed
967 * else
968 * we know the value is cindex
970 * This check is safe even if bits == KEYLENGTH due to the
971 * fact that we can only allocate a node with 32 bits if a
972 * long is greater than 32 bits.
974 if (index >= (1ul << n->bits)) {
975 n = NULL;
976 break;
979 /* keep searching until we find a perfect match leaf or NULL */
980 } while (IS_TNODE(n));
982 *tp = pn;
984 return n;
987 /* Return the first fib alias matching TOS with
988 * priority less than or equal to PRIO.
990 static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
991 u8 tos, u32 prio, u32 tb_id)
993 struct fib_alias *fa;
995 if (!fah)
996 return NULL;
998 hlist_for_each_entry(fa, fah, fa_list) {
999 if (fa->fa_slen < slen)
1000 continue;
1001 if (fa->fa_slen != slen)
1002 break;
1003 if (fa->tb_id > tb_id)
1004 continue;
1005 if (fa->tb_id != tb_id)
1006 break;
1007 if (fa->fa_tos > tos)
1008 continue;
1009 if (fa->fa_info->fib_priority >= prio || fa->fa_tos < tos)
1010 return fa;
1013 return NULL;
1016 static void trie_rebalance(struct trie *t, struct key_vector *tn)
1018 while (!IS_TRIE(tn))
1019 tn = resize(t, tn);
1022 static int fib_insert_node(struct trie *t, struct key_vector *tp,
1023 struct fib_alias *new, t_key key)
1025 struct key_vector *n, *l;
1027 l = leaf_new(key, new);
1028 if (!l)
1029 goto noleaf;
1031 /* retrieve child from parent node */
1032 n = get_child(tp, get_index(key, tp));
1034 /* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1036 * Add a new tnode here
1037 * first tnode need some special handling
1038 * leaves us in position for handling as case 3
1040 if (n) {
1041 struct key_vector *tn;
1043 tn = tnode_new(key, __fls(key ^ n->key), 1);
1044 if (!tn)
1045 goto notnode;
1047 /* initialize routes out of node */
1048 NODE_INIT_PARENT(tn, tp);
1049 put_child(tn, get_index(key, tn) ^ 1, n);
1051 /* start adding routes into the node */
1052 put_child_root(tp, key, tn);
1053 node_set_parent(n, tn);
1055 /* parent now has a NULL spot where the leaf can go */
1056 tp = tn;
1059 /* Case 3: n is NULL, and will just insert a new leaf */
1060 node_push_suffix(tp, new->fa_slen);
1061 NODE_INIT_PARENT(l, tp);
1062 put_child_root(tp, key, l);
1063 trie_rebalance(t, tp);
1065 return 0;
1066 notnode:
1067 node_free(l);
1068 noleaf:
1069 return -ENOMEM;
1072 static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1073 struct key_vector *l, struct fib_alias *new,
1074 struct fib_alias *fa, t_key key)
1076 if (!l)
1077 return fib_insert_node(t, tp, new, key);
1079 if (fa) {
1080 hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1081 } else {
1082 struct fib_alias *last;
1084 hlist_for_each_entry(last, &l->leaf, fa_list) {
1085 if (new->fa_slen < last->fa_slen)
1086 break;
1087 if ((new->fa_slen == last->fa_slen) &&
1088 (new->tb_id > last->tb_id))
1089 break;
1090 fa = last;
1093 if (fa)
1094 hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1095 else
1096 hlist_add_head_rcu(&new->fa_list, &l->leaf);
1099 /* if we added to the tail node then we need to update slen */
1100 if (l->slen < new->fa_slen) {
1101 l->slen = new->fa_slen;
1102 node_push_suffix(tp, new->fa_slen);
1105 return 0;
1108 /* Caller must hold RTNL. */
1109 int fib_table_insert(struct net *net, struct fib_table *tb,
1110 struct fib_config *cfg)
1112 struct trie *t = (struct trie *)tb->tb_data;
1113 struct fib_alias *fa, *new_fa;
1114 struct key_vector *l, *tp;
1115 u16 nlflags = NLM_F_EXCL;
1116 struct fib_info *fi;
1117 u8 plen = cfg->fc_dst_len;
1118 u8 slen = KEYLENGTH - plen;
1119 u8 tos = cfg->fc_tos;
1120 u32 key;
1121 int err;
1123 if (plen > KEYLENGTH)
1124 return -EINVAL;
1126 key = ntohl(cfg->fc_dst);
1128 pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1130 if ((plen < KEYLENGTH) && (key << plen))
1131 return -EINVAL;
1133 fi = fib_create_info(cfg);
1134 if (IS_ERR(fi)) {
1135 err = PTR_ERR(fi);
1136 goto err;
1139 l = fib_find_node(t, &tp, key);
1140 fa = l ? fib_find_alias(&l->leaf, slen, tos, fi->fib_priority,
1141 tb->tb_id) : NULL;
1143 /* Now fa, if non-NULL, points to the first fib alias
1144 * with the same keys [prefix,tos,priority], if such key already
1145 * exists or to the node before which we will insert new one.
1147 * If fa is NULL, we will need to allocate a new one and
1148 * insert to the tail of the section matching the suffix length
1149 * of the new alias.
1152 if (fa && fa->fa_tos == tos &&
1153 fa->fa_info->fib_priority == fi->fib_priority) {
1154 struct fib_alias *fa_first, *fa_match;
1156 err = -EEXIST;
1157 if (cfg->fc_nlflags & NLM_F_EXCL)
1158 goto out;
1160 nlflags &= ~NLM_F_EXCL;
1162 /* We have 2 goals:
1163 * 1. Find exact match for type, scope, fib_info to avoid
1164 * duplicate routes
1165 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1167 fa_match = NULL;
1168 fa_first = fa;
1169 hlist_for_each_entry_from(fa, fa_list) {
1170 if ((fa->fa_slen != slen) ||
1171 (fa->tb_id != tb->tb_id) ||
1172 (fa->fa_tos != tos))
1173 break;
1174 if (fa->fa_info->fib_priority != fi->fib_priority)
1175 break;
1176 if (fa->fa_type == cfg->fc_type &&
1177 fa->fa_info == fi) {
1178 fa_match = fa;
1179 break;
1183 if (cfg->fc_nlflags & NLM_F_REPLACE) {
1184 struct fib_info *fi_drop;
1185 u8 state;
1187 nlflags |= NLM_F_REPLACE;
1188 fa = fa_first;
1189 if (fa_match) {
1190 if (fa == fa_match)
1191 err = 0;
1192 goto out;
1194 err = -ENOBUFS;
1195 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1196 if (!new_fa)
1197 goto out;
1199 fi_drop = fa->fa_info;
1200 new_fa->fa_tos = fa->fa_tos;
1201 new_fa->fa_info = fi;
1202 new_fa->fa_type = cfg->fc_type;
1203 state = fa->fa_state;
1204 new_fa->fa_state = state & ~FA_S_ACCESSED;
1205 new_fa->fa_slen = fa->fa_slen;
1206 new_fa->tb_id = tb->tb_id;
1207 new_fa->fa_default = -1;
1209 hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1211 alias_free_mem_rcu(fa);
1213 fib_release_info(fi_drop);
1214 if (state & FA_S_ACCESSED)
1215 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1217 call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_ADD,
1218 key, plen, fi,
1219 new_fa->fa_tos, cfg->fc_type,
1220 tb->tb_id, cfg->fc_nlflags);
1221 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1222 tb->tb_id, &cfg->fc_nlinfo, nlflags);
1224 goto succeeded;
1226 /* Error if we find a perfect match which
1227 * uses the same scope, type, and nexthop
1228 * information.
1230 if (fa_match)
1231 goto out;
1233 if (cfg->fc_nlflags & NLM_F_APPEND)
1234 nlflags |= NLM_F_APPEND;
1235 else
1236 fa = fa_first;
1238 err = -ENOENT;
1239 if (!(cfg->fc_nlflags & NLM_F_CREATE))
1240 goto out;
1242 nlflags |= NLM_F_CREATE;
1243 err = -ENOBUFS;
1244 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1245 if (!new_fa)
1246 goto out;
1248 new_fa->fa_info = fi;
1249 new_fa->fa_tos = tos;
1250 new_fa->fa_type = cfg->fc_type;
1251 new_fa->fa_state = 0;
1252 new_fa->fa_slen = slen;
1253 new_fa->tb_id = tb->tb_id;
1254 new_fa->fa_default = -1;
1256 /* Insert new entry to the list. */
1257 err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1258 if (err)
1259 goto out_free_new_fa;
1261 if (!plen)
1262 tb->tb_num_default++;
1264 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1265 call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_ADD, key, plen, fi, tos,
1266 cfg->fc_type, tb->tb_id, cfg->fc_nlflags);
1267 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1268 &cfg->fc_nlinfo, nlflags);
1269 succeeded:
1270 return 0;
1272 out_free_new_fa:
1273 kmem_cache_free(fn_alias_kmem, new_fa);
1274 out:
1275 fib_release_info(fi);
1276 err:
1277 return err;
1280 static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1282 t_key prefix = n->key;
1284 return (key ^ prefix) & (prefix | -prefix);
1287 /* should be called with rcu_read_lock */
1288 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1289 struct fib_result *res, int fib_flags)
1291 struct trie *t = (struct trie *) tb->tb_data;
1292 #ifdef CONFIG_IP_FIB_TRIE_STATS
1293 struct trie_use_stats __percpu *stats = t->stats;
1294 #endif
1295 const t_key key = ntohl(flp->daddr);
1296 struct key_vector *n, *pn;
1297 struct fib_alias *fa;
1298 unsigned long index;
1299 t_key cindex;
1301 trace_fib_table_lookup(tb->tb_id, flp);
1303 pn = t->kv;
1304 cindex = 0;
1306 n = get_child_rcu(pn, cindex);
1307 if (!n)
1308 return -EAGAIN;
1310 #ifdef CONFIG_IP_FIB_TRIE_STATS
1311 this_cpu_inc(stats->gets);
1312 #endif
1314 /* Step 1: Travel to the longest prefix match in the trie */
1315 for (;;) {
1316 index = get_cindex(key, n);
1318 /* This bit of code is a bit tricky but it combines multiple
1319 * checks into a single check. The prefix consists of the
1320 * prefix plus zeros for the "bits" in the prefix. The index
1321 * is the difference between the key and this value. From
1322 * this we can actually derive several pieces of data.
1323 * if (index >= (1ul << bits))
1324 * we have a mismatch in skip bits and failed
1325 * else
1326 * we know the value is cindex
1328 * This check is safe even if bits == KEYLENGTH due to the
1329 * fact that we can only allocate a node with 32 bits if a
1330 * long is greater than 32 bits.
1332 if (index >= (1ul << n->bits))
1333 break;
1335 /* we have found a leaf. Prefixes have already been compared */
1336 if (IS_LEAF(n))
1337 goto found;
1339 /* only record pn and cindex if we are going to be chopping
1340 * bits later. Otherwise we are just wasting cycles.
1342 if (n->slen > n->pos) {
1343 pn = n;
1344 cindex = index;
1347 n = get_child_rcu(n, index);
1348 if (unlikely(!n))
1349 goto backtrace;
1352 /* Step 2: Sort out leaves and begin backtracing for longest prefix */
1353 for (;;) {
1354 /* record the pointer where our next node pointer is stored */
1355 struct key_vector __rcu **cptr = n->tnode;
1357 /* This test verifies that none of the bits that differ
1358 * between the key and the prefix exist in the region of
1359 * the lsb and higher in the prefix.
1361 if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1362 goto backtrace;
1364 /* exit out and process leaf */
1365 if (unlikely(IS_LEAF(n)))
1366 break;
1368 /* Don't bother recording parent info. Since we are in
1369 * prefix match mode we will have to come back to wherever
1370 * we started this traversal anyway
1373 while ((n = rcu_dereference(*cptr)) == NULL) {
1374 backtrace:
1375 #ifdef CONFIG_IP_FIB_TRIE_STATS
1376 if (!n)
1377 this_cpu_inc(stats->null_node_hit);
1378 #endif
1379 /* If we are at cindex 0 there are no more bits for
1380 * us to strip at this level so we must ascend back
1381 * up one level to see if there are any more bits to
1382 * be stripped there.
1384 while (!cindex) {
1385 t_key pkey = pn->key;
1387 /* If we don't have a parent then there is
1388 * nothing for us to do as we do not have any
1389 * further nodes to parse.
1391 if (IS_TRIE(pn))
1392 return -EAGAIN;
1393 #ifdef CONFIG_IP_FIB_TRIE_STATS
1394 this_cpu_inc(stats->backtrack);
1395 #endif
1396 /* Get Child's index */
1397 pn = node_parent_rcu(pn);
1398 cindex = get_index(pkey, pn);
1401 /* strip the least significant bit from the cindex */
1402 cindex &= cindex - 1;
1404 /* grab pointer for next child node */
1405 cptr = &pn->tnode[cindex];
1409 found:
1410 /* this line carries forward the xor from earlier in the function */
1411 index = key ^ n->key;
1413 /* Step 3: Process the leaf, if that fails fall back to backtracing */
1414 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1415 struct fib_info *fi = fa->fa_info;
1416 int nhsel, err;
1418 if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1419 if (index >= (1ul << fa->fa_slen))
1420 continue;
1422 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1423 continue;
1424 if (fi->fib_dead)
1425 continue;
1426 if (fa->fa_info->fib_scope < flp->flowi4_scope)
1427 continue;
1428 fib_alias_accessed(fa);
1429 err = fib_props[fa->fa_type].error;
1430 if (unlikely(err < 0)) {
1431 #ifdef CONFIG_IP_FIB_TRIE_STATS
1432 this_cpu_inc(stats->semantic_match_passed);
1433 #endif
1434 return err;
1436 if (fi->fib_flags & RTNH_F_DEAD)
1437 continue;
1438 for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
1439 const struct fib_nh *nh = &fi->fib_nh[nhsel];
1440 struct in_device *in_dev = __in_dev_get_rcu(nh->nh_dev);
1442 if (nh->nh_flags & RTNH_F_DEAD)
1443 continue;
1444 if (in_dev &&
1445 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1446 nh->nh_flags & RTNH_F_LINKDOWN &&
1447 !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1448 continue;
1449 if (!(flp->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF)) {
1450 if (flp->flowi4_oif &&
1451 flp->flowi4_oif != nh->nh_oif)
1452 continue;
1455 if (!(fib_flags & FIB_LOOKUP_NOREF))
1456 atomic_inc(&fi->fib_clntref);
1458 res->prefixlen = KEYLENGTH - fa->fa_slen;
1459 res->nh_sel = nhsel;
1460 res->type = fa->fa_type;
1461 res->scope = fi->fib_scope;
1462 res->fi = fi;
1463 res->table = tb;
1464 res->fa_head = &n->leaf;
1465 #ifdef CONFIG_IP_FIB_TRIE_STATS
1466 this_cpu_inc(stats->semantic_match_passed);
1467 #endif
1468 trace_fib_table_lookup_nh(nh);
1470 return err;
1473 #ifdef CONFIG_IP_FIB_TRIE_STATS
1474 this_cpu_inc(stats->semantic_match_miss);
1475 #endif
1476 goto backtrace;
1478 EXPORT_SYMBOL_GPL(fib_table_lookup);
1480 static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1481 struct key_vector *l, struct fib_alias *old)
1483 /* record the location of the previous list_info entry */
1484 struct hlist_node **pprev = old->fa_list.pprev;
1485 struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1487 /* remove the fib_alias from the list */
1488 hlist_del_rcu(&old->fa_list);
1490 /* if we emptied the list this leaf will be freed and we can sort
1491 * out parent suffix lengths as a part of trie_rebalance
1493 if (hlist_empty(&l->leaf)) {
1494 if (tp->slen == l->slen)
1495 node_pull_suffix(tp, tp->pos);
1496 put_child_root(tp, l->key, NULL);
1497 node_free(l);
1498 trie_rebalance(t, tp);
1499 return;
1502 /* only access fa if it is pointing at the last valid hlist_node */
1503 if (*pprev)
1504 return;
1506 /* update the trie with the latest suffix length */
1507 l->slen = fa->fa_slen;
1508 node_pull_suffix(tp, fa->fa_slen);
1511 /* Caller must hold RTNL. */
1512 int fib_table_delete(struct net *net, struct fib_table *tb,
1513 struct fib_config *cfg)
1515 struct trie *t = (struct trie *) tb->tb_data;
1516 struct fib_alias *fa, *fa_to_delete;
1517 struct key_vector *l, *tp;
1518 u8 plen = cfg->fc_dst_len;
1519 u8 slen = KEYLENGTH - plen;
1520 u8 tos = cfg->fc_tos;
1521 u32 key;
1523 if (plen > KEYLENGTH)
1524 return -EINVAL;
1526 key = ntohl(cfg->fc_dst);
1528 if ((plen < KEYLENGTH) && (key << plen))
1529 return -EINVAL;
1531 l = fib_find_node(t, &tp, key);
1532 if (!l)
1533 return -ESRCH;
1535 fa = fib_find_alias(&l->leaf, slen, tos, 0, tb->tb_id);
1536 if (!fa)
1537 return -ESRCH;
1539 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1541 fa_to_delete = NULL;
1542 hlist_for_each_entry_from(fa, fa_list) {
1543 struct fib_info *fi = fa->fa_info;
1545 if ((fa->fa_slen != slen) ||
1546 (fa->tb_id != tb->tb_id) ||
1547 (fa->fa_tos != tos))
1548 break;
1550 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1551 (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1552 fa->fa_info->fib_scope == cfg->fc_scope) &&
1553 (!cfg->fc_prefsrc ||
1554 fi->fib_prefsrc == cfg->fc_prefsrc) &&
1555 (!cfg->fc_protocol ||
1556 fi->fib_protocol == cfg->fc_protocol) &&
1557 fib_nh_match(cfg, fi) == 0) {
1558 fa_to_delete = fa;
1559 break;
1563 if (!fa_to_delete)
1564 return -ESRCH;
1566 call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, key, plen,
1567 fa_to_delete->fa_info, tos, cfg->fc_type,
1568 tb->tb_id, 0);
1569 rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1570 &cfg->fc_nlinfo, 0);
1572 if (!plen)
1573 tb->tb_num_default--;
1575 fib_remove_alias(t, tp, l, fa_to_delete);
1577 if (fa_to_delete->fa_state & FA_S_ACCESSED)
1578 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1580 fib_release_info(fa_to_delete->fa_info);
1581 alias_free_mem_rcu(fa_to_delete);
1582 return 0;
1585 /* Scan for the next leaf starting at the provided key value */
1586 static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1588 struct key_vector *pn, *n = *tn;
1589 unsigned long cindex;
1591 /* this loop is meant to try and find the key in the trie */
1592 do {
1593 /* record parent and next child index */
1594 pn = n;
1595 cindex = (key > pn->key) ? get_index(key, pn) : 0;
1597 if (cindex >> pn->bits)
1598 break;
1600 /* descend into the next child */
1601 n = get_child_rcu(pn, cindex++);
1602 if (!n)
1603 break;
1605 /* guarantee forward progress on the keys */
1606 if (IS_LEAF(n) && (n->key >= key))
1607 goto found;
1608 } while (IS_TNODE(n));
1610 /* this loop will search for the next leaf with a greater key */
1611 while (!IS_TRIE(pn)) {
1612 /* if we exhausted the parent node we will need to climb */
1613 if (cindex >= (1ul << pn->bits)) {
1614 t_key pkey = pn->key;
1616 pn = node_parent_rcu(pn);
1617 cindex = get_index(pkey, pn) + 1;
1618 continue;
1621 /* grab the next available node */
1622 n = get_child_rcu(pn, cindex++);
1623 if (!n)
1624 continue;
1626 /* no need to compare keys since we bumped the index */
1627 if (IS_LEAF(n))
1628 goto found;
1630 /* Rescan start scanning in new node */
1631 pn = n;
1632 cindex = 0;
1635 *tn = pn;
1636 return NULL; /* Root of trie */
1637 found:
1638 /* if we are at the limit for keys just return NULL for the tnode */
1639 *tn = pn;
1640 return n;
1643 static void fib_trie_free(struct fib_table *tb)
1645 struct trie *t = (struct trie *)tb->tb_data;
1646 struct key_vector *pn = t->kv;
1647 unsigned long cindex = 1;
1648 struct hlist_node *tmp;
1649 struct fib_alias *fa;
1651 /* walk trie in reverse order and free everything */
1652 for (;;) {
1653 struct key_vector *n;
1655 if (!(cindex--)) {
1656 t_key pkey = pn->key;
1658 if (IS_TRIE(pn))
1659 break;
1661 n = pn;
1662 pn = node_parent(pn);
1664 /* drop emptied tnode */
1665 put_child_root(pn, n->key, NULL);
1666 node_free(n);
1668 cindex = get_index(pkey, pn);
1670 continue;
1673 /* grab the next available node */
1674 n = get_child(pn, cindex);
1675 if (!n)
1676 continue;
1678 if (IS_TNODE(n)) {
1679 /* record pn and cindex for leaf walking */
1680 pn = n;
1681 cindex = 1ul << n->bits;
1683 continue;
1686 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1687 hlist_del_rcu(&fa->fa_list);
1688 alias_free_mem_rcu(fa);
1691 put_child_root(pn, n->key, NULL);
1692 node_free(n);
1695 #ifdef CONFIG_IP_FIB_TRIE_STATS
1696 free_percpu(t->stats);
1697 #endif
1698 kfree(tb);
1701 struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1703 struct trie *ot = (struct trie *)oldtb->tb_data;
1704 struct key_vector *l, *tp = ot->kv;
1705 struct fib_table *local_tb;
1706 struct fib_alias *fa;
1707 struct trie *lt;
1708 t_key key = 0;
1710 if (oldtb->tb_data == oldtb->__data)
1711 return oldtb;
1713 local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1714 if (!local_tb)
1715 return NULL;
1717 lt = (struct trie *)local_tb->tb_data;
1719 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1720 struct key_vector *local_l = NULL, *local_tp;
1722 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1723 struct fib_alias *new_fa;
1725 if (local_tb->tb_id != fa->tb_id)
1726 continue;
1728 /* clone fa for new local table */
1729 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1730 if (!new_fa)
1731 goto out;
1733 memcpy(new_fa, fa, sizeof(*fa));
1735 /* insert clone into table */
1736 if (!local_l)
1737 local_l = fib_find_node(lt, &local_tp, l->key);
1739 if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1740 NULL, l->key)) {
1741 kmem_cache_free(fn_alias_kmem, new_fa);
1742 goto out;
1746 /* stop loop if key wrapped back to 0 */
1747 key = l->key + 1;
1748 if (key < l->key)
1749 break;
1752 return local_tb;
1753 out:
1754 fib_trie_free(local_tb);
1756 return NULL;
1759 /* Caller must hold RTNL */
1760 void fib_table_flush_external(struct fib_table *tb)
1762 struct trie *t = (struct trie *)tb->tb_data;
1763 struct key_vector *pn = t->kv;
1764 unsigned long cindex = 1;
1765 struct hlist_node *tmp;
1766 struct fib_alias *fa;
1768 /* walk trie in reverse order */
1769 for (;;) {
1770 unsigned char slen = 0;
1771 struct key_vector *n;
1773 if (!(cindex--)) {
1774 t_key pkey = pn->key;
1776 /* cannot resize the trie vector */
1777 if (IS_TRIE(pn))
1778 break;
1780 /* update the suffix to address pulled leaves */
1781 if (pn->slen > pn->pos)
1782 update_suffix(pn);
1784 /* resize completed node */
1785 pn = resize(t, pn);
1786 cindex = get_index(pkey, pn);
1788 continue;
1791 /* grab the next available node */
1792 n = get_child(pn, cindex);
1793 if (!n)
1794 continue;
1796 if (IS_TNODE(n)) {
1797 /* record pn and cindex for leaf walking */
1798 pn = n;
1799 cindex = 1ul << n->bits;
1801 continue;
1804 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1805 /* if alias was cloned to local then we just
1806 * need to remove the local copy from main
1808 if (tb->tb_id != fa->tb_id) {
1809 hlist_del_rcu(&fa->fa_list);
1810 alias_free_mem_rcu(fa);
1811 continue;
1814 /* record local slen */
1815 slen = fa->fa_slen;
1818 /* update leaf slen */
1819 n->slen = slen;
1821 if (hlist_empty(&n->leaf)) {
1822 put_child_root(pn, n->key, NULL);
1823 node_free(n);
1828 /* Caller must hold RTNL. */
1829 int fib_table_flush(struct net *net, struct fib_table *tb)
1831 struct trie *t = (struct trie *)tb->tb_data;
1832 struct key_vector *pn = t->kv;
1833 unsigned long cindex = 1;
1834 struct hlist_node *tmp;
1835 struct fib_alias *fa;
1836 int found = 0;
1838 /* walk trie in reverse order */
1839 for (;;) {
1840 unsigned char slen = 0;
1841 struct key_vector *n;
1843 if (!(cindex--)) {
1844 t_key pkey = pn->key;
1846 /* cannot resize the trie vector */
1847 if (IS_TRIE(pn))
1848 break;
1850 /* update the suffix to address pulled leaves */
1851 if (pn->slen > pn->pos)
1852 update_suffix(pn);
1854 /* resize completed node */
1855 pn = resize(t, pn);
1856 cindex = get_index(pkey, pn);
1858 continue;
1861 /* grab the next available node */
1862 n = get_child(pn, cindex);
1863 if (!n)
1864 continue;
1866 if (IS_TNODE(n)) {
1867 /* record pn and cindex for leaf walking */
1868 pn = n;
1869 cindex = 1ul << n->bits;
1871 continue;
1874 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1875 struct fib_info *fi = fa->fa_info;
1877 if (!fi || !(fi->fib_flags & RTNH_F_DEAD)) {
1878 slen = fa->fa_slen;
1879 continue;
1882 call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL,
1883 n->key,
1884 KEYLENGTH - fa->fa_slen,
1885 fi, fa->fa_tos, fa->fa_type,
1886 tb->tb_id, 0);
1887 hlist_del_rcu(&fa->fa_list);
1888 fib_release_info(fa->fa_info);
1889 alias_free_mem_rcu(fa);
1890 found++;
1893 /* update leaf slen */
1894 n->slen = slen;
1896 if (hlist_empty(&n->leaf)) {
1897 put_child_root(pn, n->key, NULL);
1898 node_free(n);
1902 pr_debug("trie_flush found=%d\n", found);
1903 return found;
1906 static void __trie_free_rcu(struct rcu_head *head)
1908 struct fib_table *tb = container_of(head, struct fib_table, rcu);
1909 #ifdef CONFIG_IP_FIB_TRIE_STATS
1910 struct trie *t = (struct trie *)tb->tb_data;
1912 if (tb->tb_data == tb->__data)
1913 free_percpu(t->stats);
1914 #endif /* CONFIG_IP_FIB_TRIE_STATS */
1915 kfree(tb);
1918 void fib_free_table(struct fib_table *tb)
1920 call_rcu(&tb->rcu, __trie_free_rcu);
1923 static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
1924 struct sk_buff *skb, struct netlink_callback *cb)
1926 __be32 xkey = htonl(l->key);
1927 struct fib_alias *fa;
1928 int i, s_i;
1930 s_i = cb->args[4];
1931 i = 0;
1933 /* rcu_read_lock is hold by caller */
1934 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1935 if (i < s_i) {
1936 i++;
1937 continue;
1940 if (tb->tb_id != fa->tb_id) {
1941 i++;
1942 continue;
1945 if (fib_dump_info(skb, NETLINK_CB(cb->skb).portid,
1946 cb->nlh->nlmsg_seq,
1947 RTM_NEWROUTE,
1948 tb->tb_id,
1949 fa->fa_type,
1950 xkey,
1951 KEYLENGTH - fa->fa_slen,
1952 fa->fa_tos,
1953 fa->fa_info, NLM_F_MULTI) < 0) {
1954 cb->args[4] = i;
1955 return -1;
1957 i++;
1960 cb->args[4] = i;
1961 return skb->len;
1964 /* rcu_read_lock needs to be hold by caller from readside */
1965 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
1966 struct netlink_callback *cb)
1968 struct trie *t = (struct trie *)tb->tb_data;
1969 struct key_vector *l, *tp = t->kv;
1970 /* Dump starting at last key.
1971 * Note: 0.0.0.0/0 (ie default) is first key.
1973 int count = cb->args[2];
1974 t_key key = cb->args[3];
1976 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1977 if (fn_trie_dump_leaf(l, tb, skb, cb) < 0) {
1978 cb->args[3] = key;
1979 cb->args[2] = count;
1980 return -1;
1983 ++count;
1984 key = l->key + 1;
1986 memset(&cb->args[4], 0,
1987 sizeof(cb->args) - 4*sizeof(cb->args[0]));
1989 /* stop loop if key wrapped back to 0 */
1990 if (key < l->key)
1991 break;
1994 cb->args[3] = key;
1995 cb->args[2] = count;
1997 return skb->len;
2000 void __init fib_trie_init(void)
2002 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
2003 sizeof(struct fib_alias),
2004 0, SLAB_PANIC, NULL);
2006 trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
2007 LEAF_SIZE,
2008 0, SLAB_PANIC, NULL);
2011 struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
2013 struct fib_table *tb;
2014 struct trie *t;
2015 size_t sz = sizeof(*tb);
2017 if (!alias)
2018 sz += sizeof(struct trie);
2020 tb = kzalloc(sz, GFP_KERNEL);
2021 if (!tb)
2022 return NULL;
2024 tb->tb_id = id;
2025 tb->tb_num_default = 0;
2026 tb->tb_data = (alias ? alias->__data : tb->__data);
2028 if (alias)
2029 return tb;
2031 t = (struct trie *) tb->tb_data;
2032 t->kv[0].pos = KEYLENGTH;
2033 t->kv[0].slen = KEYLENGTH;
2034 #ifdef CONFIG_IP_FIB_TRIE_STATS
2035 t->stats = alloc_percpu(struct trie_use_stats);
2036 if (!t->stats) {
2037 kfree(tb);
2038 tb = NULL;
2040 #endif
2042 return tb;
2045 #ifdef CONFIG_PROC_FS
2046 /* Depth first Trie walk iterator */
2047 struct fib_trie_iter {
2048 struct seq_net_private p;
2049 struct fib_table *tb;
2050 struct key_vector *tnode;
2051 unsigned int index;
2052 unsigned int depth;
2055 static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2057 unsigned long cindex = iter->index;
2058 struct key_vector *pn = iter->tnode;
2059 t_key pkey;
2061 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2062 iter->tnode, iter->index, iter->depth);
2064 while (!IS_TRIE(pn)) {
2065 while (cindex < child_length(pn)) {
2066 struct key_vector *n = get_child_rcu(pn, cindex++);
2068 if (!n)
2069 continue;
2071 if (IS_LEAF(n)) {
2072 iter->tnode = pn;
2073 iter->index = cindex;
2074 } else {
2075 /* push down one level */
2076 iter->tnode = n;
2077 iter->index = 0;
2078 ++iter->depth;
2081 return n;
2084 /* Current node exhausted, pop back up */
2085 pkey = pn->key;
2086 pn = node_parent_rcu(pn);
2087 cindex = get_index(pkey, pn) + 1;
2088 --iter->depth;
2091 /* record root node so further searches know we are done */
2092 iter->tnode = pn;
2093 iter->index = 0;
2095 return NULL;
2098 static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2099 struct trie *t)
2101 struct key_vector *n, *pn;
2103 if (!t)
2104 return NULL;
2106 pn = t->kv;
2107 n = rcu_dereference(pn->tnode[0]);
2108 if (!n)
2109 return NULL;
2111 if (IS_TNODE(n)) {
2112 iter->tnode = n;
2113 iter->index = 0;
2114 iter->depth = 1;
2115 } else {
2116 iter->tnode = pn;
2117 iter->index = 0;
2118 iter->depth = 0;
2121 return n;
2124 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2126 struct key_vector *n;
2127 struct fib_trie_iter iter;
2129 memset(s, 0, sizeof(*s));
2131 rcu_read_lock();
2132 for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2133 if (IS_LEAF(n)) {
2134 struct fib_alias *fa;
2136 s->leaves++;
2137 s->totdepth += iter.depth;
2138 if (iter.depth > s->maxdepth)
2139 s->maxdepth = iter.depth;
2141 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2142 ++s->prefixes;
2143 } else {
2144 s->tnodes++;
2145 if (n->bits < MAX_STAT_DEPTH)
2146 s->nodesizes[n->bits]++;
2147 s->nullpointers += tn_info(n)->empty_children;
2150 rcu_read_unlock();
2154 * This outputs /proc/net/fib_triestats
2156 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2158 unsigned int i, max, pointers, bytes, avdepth;
2160 if (stat->leaves)
2161 avdepth = stat->totdepth*100 / stat->leaves;
2162 else
2163 avdepth = 0;
2165 seq_printf(seq, "\tAver depth: %u.%02d\n",
2166 avdepth / 100, avdepth % 100);
2167 seq_printf(seq, "\tMax depth: %u\n", stat->maxdepth);
2169 seq_printf(seq, "\tLeaves: %u\n", stat->leaves);
2170 bytes = LEAF_SIZE * stat->leaves;
2172 seq_printf(seq, "\tPrefixes: %u\n", stat->prefixes);
2173 bytes += sizeof(struct fib_alias) * stat->prefixes;
2175 seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2176 bytes += TNODE_SIZE(0) * stat->tnodes;
2178 max = MAX_STAT_DEPTH;
2179 while (max > 0 && stat->nodesizes[max-1] == 0)
2180 max--;
2182 pointers = 0;
2183 for (i = 1; i < max; i++)
2184 if (stat->nodesizes[i] != 0) {
2185 seq_printf(seq, " %u: %u", i, stat->nodesizes[i]);
2186 pointers += (1<<i) * stat->nodesizes[i];
2188 seq_putc(seq, '\n');
2189 seq_printf(seq, "\tPointers: %u\n", pointers);
2191 bytes += sizeof(struct key_vector *) * pointers;
2192 seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2193 seq_printf(seq, "Total size: %u kB\n", (bytes + 1023) / 1024);
2196 #ifdef CONFIG_IP_FIB_TRIE_STATS
2197 static void trie_show_usage(struct seq_file *seq,
2198 const struct trie_use_stats __percpu *stats)
2200 struct trie_use_stats s = { 0 };
2201 int cpu;
2203 /* loop through all of the CPUs and gather up the stats */
2204 for_each_possible_cpu(cpu) {
2205 const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2207 s.gets += pcpu->gets;
2208 s.backtrack += pcpu->backtrack;
2209 s.semantic_match_passed += pcpu->semantic_match_passed;
2210 s.semantic_match_miss += pcpu->semantic_match_miss;
2211 s.null_node_hit += pcpu->null_node_hit;
2212 s.resize_node_skipped += pcpu->resize_node_skipped;
2215 seq_printf(seq, "\nCounters:\n---------\n");
2216 seq_printf(seq, "gets = %u\n", s.gets);
2217 seq_printf(seq, "backtracks = %u\n", s.backtrack);
2218 seq_printf(seq, "semantic match passed = %u\n",
2219 s.semantic_match_passed);
2220 seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2221 seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2222 seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2224 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2226 static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2228 if (tb->tb_id == RT_TABLE_LOCAL)
2229 seq_puts(seq, "Local:\n");
2230 else if (tb->tb_id == RT_TABLE_MAIN)
2231 seq_puts(seq, "Main:\n");
2232 else
2233 seq_printf(seq, "Id %d:\n", tb->tb_id);
2237 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2239 struct net *net = (struct net *)seq->private;
2240 unsigned int h;
2242 seq_printf(seq,
2243 "Basic info: size of leaf:"
2244 " %Zd bytes, size of tnode: %Zd bytes.\n",
2245 LEAF_SIZE, TNODE_SIZE(0));
2247 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2248 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2249 struct fib_table *tb;
2251 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2252 struct trie *t = (struct trie *) tb->tb_data;
2253 struct trie_stat stat;
2255 if (!t)
2256 continue;
2258 fib_table_print(seq, tb);
2260 trie_collect_stats(t, &stat);
2261 trie_show_stats(seq, &stat);
2262 #ifdef CONFIG_IP_FIB_TRIE_STATS
2263 trie_show_usage(seq, t->stats);
2264 #endif
2268 return 0;
2271 static int fib_triestat_seq_open(struct inode *inode, struct file *file)
2273 return single_open_net(inode, file, fib_triestat_seq_show);
2276 static const struct file_operations fib_triestat_fops = {
2277 .owner = THIS_MODULE,
2278 .open = fib_triestat_seq_open,
2279 .read = seq_read,
2280 .llseek = seq_lseek,
2281 .release = single_release_net,
2284 static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2286 struct fib_trie_iter *iter = seq->private;
2287 struct net *net = seq_file_net(seq);
2288 loff_t idx = 0;
2289 unsigned int h;
2291 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2292 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2293 struct fib_table *tb;
2295 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2296 struct key_vector *n;
2298 for (n = fib_trie_get_first(iter,
2299 (struct trie *) tb->tb_data);
2300 n; n = fib_trie_get_next(iter))
2301 if (pos == idx++) {
2302 iter->tb = tb;
2303 return n;
2308 return NULL;
2311 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2312 __acquires(RCU)
2314 rcu_read_lock();
2315 return fib_trie_get_idx(seq, *pos);
2318 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2320 struct fib_trie_iter *iter = seq->private;
2321 struct net *net = seq_file_net(seq);
2322 struct fib_table *tb = iter->tb;
2323 struct hlist_node *tb_node;
2324 unsigned int h;
2325 struct key_vector *n;
2327 ++*pos;
2328 /* next node in same table */
2329 n = fib_trie_get_next(iter);
2330 if (n)
2331 return n;
2333 /* walk rest of this hash chain */
2334 h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2335 while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2336 tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2337 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2338 if (n)
2339 goto found;
2342 /* new hash chain */
2343 while (++h < FIB_TABLE_HASHSZ) {
2344 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2345 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2346 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2347 if (n)
2348 goto found;
2351 return NULL;
2353 found:
2354 iter->tb = tb;
2355 return n;
2358 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2359 __releases(RCU)
2361 rcu_read_unlock();
2364 static void seq_indent(struct seq_file *seq, int n)
2366 while (n-- > 0)
2367 seq_puts(seq, " ");
2370 static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2372 switch (s) {
2373 case RT_SCOPE_UNIVERSE: return "universe";
2374 case RT_SCOPE_SITE: return "site";
2375 case RT_SCOPE_LINK: return "link";
2376 case RT_SCOPE_HOST: return "host";
2377 case RT_SCOPE_NOWHERE: return "nowhere";
2378 default:
2379 snprintf(buf, len, "scope=%d", s);
2380 return buf;
2384 static const char *const rtn_type_names[__RTN_MAX] = {
2385 [RTN_UNSPEC] = "UNSPEC",
2386 [RTN_UNICAST] = "UNICAST",
2387 [RTN_LOCAL] = "LOCAL",
2388 [RTN_BROADCAST] = "BROADCAST",
2389 [RTN_ANYCAST] = "ANYCAST",
2390 [RTN_MULTICAST] = "MULTICAST",
2391 [RTN_BLACKHOLE] = "BLACKHOLE",
2392 [RTN_UNREACHABLE] = "UNREACHABLE",
2393 [RTN_PROHIBIT] = "PROHIBIT",
2394 [RTN_THROW] = "THROW",
2395 [RTN_NAT] = "NAT",
2396 [RTN_XRESOLVE] = "XRESOLVE",
2399 static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2401 if (t < __RTN_MAX && rtn_type_names[t])
2402 return rtn_type_names[t];
2403 snprintf(buf, len, "type %u", t);
2404 return buf;
2407 /* Pretty print the trie */
2408 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2410 const struct fib_trie_iter *iter = seq->private;
2411 struct key_vector *n = v;
2413 if (IS_TRIE(node_parent_rcu(n)))
2414 fib_table_print(seq, iter->tb);
2416 if (IS_TNODE(n)) {
2417 __be32 prf = htonl(n->key);
2419 seq_indent(seq, iter->depth-1);
2420 seq_printf(seq, " +-- %pI4/%zu %u %u %u\n",
2421 &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2422 tn_info(n)->full_children,
2423 tn_info(n)->empty_children);
2424 } else {
2425 __be32 val = htonl(n->key);
2426 struct fib_alias *fa;
2428 seq_indent(seq, iter->depth);
2429 seq_printf(seq, " |-- %pI4\n", &val);
2431 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2432 char buf1[32], buf2[32];
2434 seq_indent(seq, iter->depth + 1);
2435 seq_printf(seq, " /%zu %s %s",
2436 KEYLENGTH - fa->fa_slen,
2437 rtn_scope(buf1, sizeof(buf1),
2438 fa->fa_info->fib_scope),
2439 rtn_type(buf2, sizeof(buf2),
2440 fa->fa_type));
2441 if (fa->fa_tos)
2442 seq_printf(seq, " tos=%d", fa->fa_tos);
2443 seq_putc(seq, '\n');
2447 return 0;
2450 static const struct seq_operations fib_trie_seq_ops = {
2451 .start = fib_trie_seq_start,
2452 .next = fib_trie_seq_next,
2453 .stop = fib_trie_seq_stop,
2454 .show = fib_trie_seq_show,
2457 static int fib_trie_seq_open(struct inode *inode, struct file *file)
2459 return seq_open_net(inode, file, &fib_trie_seq_ops,
2460 sizeof(struct fib_trie_iter));
2463 static const struct file_operations fib_trie_fops = {
2464 .owner = THIS_MODULE,
2465 .open = fib_trie_seq_open,
2466 .read = seq_read,
2467 .llseek = seq_lseek,
2468 .release = seq_release_net,
2471 struct fib_route_iter {
2472 struct seq_net_private p;
2473 struct fib_table *main_tb;
2474 struct key_vector *tnode;
2475 loff_t pos;
2476 t_key key;
2479 static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2480 loff_t pos)
2482 struct key_vector *l, **tp = &iter->tnode;
2483 t_key key;
2485 /* use cached location of previously found key */
2486 if (iter->pos > 0 && pos >= iter->pos) {
2487 key = iter->key;
2488 } else {
2489 iter->pos = 1;
2490 key = 0;
2493 pos -= iter->pos;
2495 while ((l = leaf_walk_rcu(tp, key)) && (pos-- > 0)) {
2496 key = l->key + 1;
2497 iter->pos++;
2498 l = NULL;
2500 /* handle unlikely case of a key wrap */
2501 if (!key)
2502 break;
2505 if (l)
2506 iter->key = l->key; /* remember it */
2507 else
2508 iter->pos = 0; /* forget it */
2510 return l;
2513 static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2514 __acquires(RCU)
2516 struct fib_route_iter *iter = seq->private;
2517 struct fib_table *tb;
2518 struct trie *t;
2520 rcu_read_lock();
2522 tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2523 if (!tb)
2524 return NULL;
2526 iter->main_tb = tb;
2527 t = (struct trie *)tb->tb_data;
2528 iter->tnode = t->kv;
2530 if (*pos != 0)
2531 return fib_route_get_idx(iter, *pos);
2533 iter->pos = 0;
2534 iter->key = KEY_MAX;
2536 return SEQ_START_TOKEN;
2539 static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2541 struct fib_route_iter *iter = seq->private;
2542 struct key_vector *l = NULL;
2543 t_key key = iter->key + 1;
2545 ++*pos;
2547 /* only allow key of 0 for start of sequence */
2548 if ((v == SEQ_START_TOKEN) || key)
2549 l = leaf_walk_rcu(&iter->tnode, key);
2551 if (l) {
2552 iter->key = l->key;
2553 iter->pos++;
2554 } else {
2555 iter->pos = 0;
2558 return l;
2561 static void fib_route_seq_stop(struct seq_file *seq, void *v)
2562 __releases(RCU)
2564 rcu_read_unlock();
2567 static unsigned int fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2569 unsigned int flags = 0;
2571 if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2572 flags = RTF_REJECT;
2573 if (fi && fi->fib_nh->nh_gw)
2574 flags |= RTF_GATEWAY;
2575 if (mask == htonl(0xFFFFFFFF))
2576 flags |= RTF_HOST;
2577 flags |= RTF_UP;
2578 return flags;
2582 * This outputs /proc/net/route.
2583 * The format of the file is not supposed to be changed
2584 * and needs to be same as fib_hash output to avoid breaking
2585 * legacy utilities
2587 static int fib_route_seq_show(struct seq_file *seq, void *v)
2589 struct fib_route_iter *iter = seq->private;
2590 struct fib_table *tb = iter->main_tb;
2591 struct fib_alias *fa;
2592 struct key_vector *l = v;
2593 __be32 prefix;
2595 if (v == SEQ_START_TOKEN) {
2596 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2597 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2598 "\tWindow\tIRTT");
2599 return 0;
2602 prefix = htonl(l->key);
2604 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2605 const struct fib_info *fi = fa->fa_info;
2606 __be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2607 unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2609 if ((fa->fa_type == RTN_BROADCAST) ||
2610 (fa->fa_type == RTN_MULTICAST))
2611 continue;
2613 if (fa->tb_id != tb->tb_id)
2614 continue;
2616 seq_setwidth(seq, 127);
2618 if (fi)
2619 seq_printf(seq,
2620 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2621 "%d\t%08X\t%d\t%u\t%u",
2622 fi->fib_dev ? fi->fib_dev->name : "*",
2623 prefix,
2624 fi->fib_nh->nh_gw, flags, 0, 0,
2625 fi->fib_priority,
2626 mask,
2627 (fi->fib_advmss ?
2628 fi->fib_advmss + 40 : 0),
2629 fi->fib_window,
2630 fi->fib_rtt >> 3);
2631 else
2632 seq_printf(seq,
2633 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2634 "%d\t%08X\t%d\t%u\t%u",
2635 prefix, 0, flags, 0, 0, 0,
2636 mask, 0, 0, 0);
2638 seq_pad(seq, '\n');
2641 return 0;
2644 static const struct seq_operations fib_route_seq_ops = {
2645 .start = fib_route_seq_start,
2646 .next = fib_route_seq_next,
2647 .stop = fib_route_seq_stop,
2648 .show = fib_route_seq_show,
2651 static int fib_route_seq_open(struct inode *inode, struct file *file)
2653 return seq_open_net(inode, file, &fib_route_seq_ops,
2654 sizeof(struct fib_route_iter));
2657 static const struct file_operations fib_route_fops = {
2658 .owner = THIS_MODULE,
2659 .open = fib_route_seq_open,
2660 .read = seq_read,
2661 .llseek = seq_lseek,
2662 .release = seq_release_net,
2665 int __net_init fib_proc_init(struct net *net)
2667 if (!proc_create("fib_trie", S_IRUGO, net->proc_net, &fib_trie_fops))
2668 goto out1;
2670 if (!proc_create("fib_triestat", S_IRUGO, net->proc_net,
2671 &fib_triestat_fops))
2672 goto out2;
2674 if (!proc_create("route", S_IRUGO, net->proc_net, &fib_route_fops))
2675 goto out3;
2677 return 0;
2679 out3:
2680 remove_proc_entry("fib_triestat", net->proc_net);
2681 out2:
2682 remove_proc_entry("fib_trie", net->proc_net);
2683 out1:
2684 return -ENOMEM;
2687 void __net_exit fib_proc_exit(struct net *net)
2689 remove_proc_entry("fib_trie", net->proc_net);
2690 remove_proc_entry("fib_triestat", net->proc_net);
2691 remove_proc_entry("route", net->proc_net);
2694 #endif /* CONFIG_PROC_FS */