2 * This program is free software; you can redistribute it and/or
3 * modify it under the terms of the GNU General Public License
4 * as published by the Free Software Foundation; either version
5 * 2 of the License, or (at your option) any later version.
7 * Robert Olsson <robert.olsson@its.uu.se> Uppsala Universitet
8 * & Swedish University of Agricultural Sciences.
10 * Jens Laas <jens.laas@data.slu.se> Swedish University of
11 * Agricultural Sciences.
13 * Hans Liss <hans.liss@its.uu.se> Uppsala Universitet
15 * This work is based on the LPC-trie which is originally described in:
17 * An experimental study of compression methods for dynamic tries
18 * Stefan Nilsson and Matti Tikkanen. Algorithmica, 33(1):19-33, 2002.
19 * http://www.csc.kth.se/~snilsson/software/dyntrie2/
22 * IP-address lookup using LC-tries. Stefan Nilsson and Gunnar Karlsson
23 * IEEE Journal on Selected Areas in Communications, 17(6):1083-1092, June 1999
26 * Code from fib_hash has been reused which includes the following header:
29 * INET An implementation of the TCP/IP protocol suite for the LINUX
30 * operating system. INET is implemented using the BSD Socket
31 * interface as the means of communication with the user level.
33 * IPv4 FIB: lookup engine and maintenance routines.
36 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
38 * This program is free software; you can redistribute it and/or
39 * modify it under the terms of the GNU General Public License
40 * as published by the Free Software Foundation; either version
41 * 2 of the License, or (at your option) any later version.
43 * Substantial contributions to this work comes from:
45 * David S. Miller, <davem@davemloft.net>
46 * Stephen Hemminger <shemminger@osdl.org>
47 * Paul E. McKenney <paulmck@us.ibm.com>
48 * Patrick McHardy <kaber@trash.net>
51 #define VERSION "0.409"
53 #include <asm/uaccess.h>
54 #include <linux/bitops.h>
55 #include <linux/types.h>
56 #include <linux/kernel.h>
58 #include <linux/string.h>
59 #include <linux/socket.h>
60 #include <linux/sockios.h>
61 #include <linux/errno.h>
63 #include <linux/inet.h>
64 #include <linux/inetdevice.h>
65 #include <linux/netdevice.h>
66 #include <linux/if_arp.h>
67 #include <linux/proc_fs.h>
68 #include <linux/rcupdate.h>
69 #include <linux/skbuff.h>
70 #include <linux/netlink.h>
71 #include <linux/init.h>
72 #include <linux/list.h>
73 #include <linux/slab.h>
74 #include <linux/export.h>
75 #include <linux/vmalloc.h>
76 #include <linux/notifier.h>
77 #include <net/net_namespace.h>
79 #include <net/protocol.h>
80 #include <net/route.h>
83 #include <net/ip_fib.h>
84 #include <trace/events/fib.h>
85 #include "fib_lookup.h"
87 static unsigned int fib_seq_sum(void)
89 unsigned int fib_seq = 0;
94 fib_seq += net->ipv4.fib_seq;
100 static ATOMIC_NOTIFIER_HEAD(fib_chain);
102 static int call_fib_notifier(struct notifier_block *nb, struct net *net,
103 enum fib_event_type event_type,
104 struct fib_notifier_info *info)
107 return nb->notifier_call(nb, event_type, info);
110 static void fib_rules_notify(struct net *net, struct notifier_block *nb,
111 enum fib_event_type event_type)
113 #ifdef CONFIG_IP_MULTIPLE_TABLES
114 struct fib_notifier_info info;
116 if (net->ipv4.fib_has_custom_rules)
117 call_fib_notifier(nb, net, event_type, &info);
121 static void fib_notify(struct net *net, struct notifier_block *nb,
122 enum fib_event_type event_type);
124 static int call_fib_entry_notifier(struct notifier_block *nb, struct net *net,
125 enum fib_event_type event_type, u32 dst,
126 int dst_len, struct fib_info *fi,
127 u8 tos, u8 type, u32 tb_id, u32 nlflags)
129 struct fib_entry_notifier_info info = {
138 return call_fib_notifier(nb, net, event_type, &info.info);
141 static bool fib_dump_is_consistent(struct notifier_block *nb,
142 void (*cb)(struct notifier_block *nb),
143 unsigned int fib_seq)
145 atomic_notifier_chain_register(&fib_chain, nb);
146 if (fib_seq == fib_seq_sum())
148 atomic_notifier_chain_unregister(&fib_chain, nb);
154 #define FIB_DUMP_MAX_RETRIES 5
155 int register_fib_notifier(struct notifier_block *nb,
156 void (*cb)(struct notifier_block *nb))
161 unsigned int fib_seq = fib_seq_sum();
164 /* Mutex semantics guarantee that every change done to
165 * FIB tries before we read the change sequence counter
166 * is now visible to us.
169 for_each_net_rcu(net) {
170 fib_rules_notify(net, nb, FIB_EVENT_RULE_ADD);
171 fib_notify(net, nb, FIB_EVENT_ENTRY_ADD);
175 if (fib_dump_is_consistent(nb, cb, fib_seq))
177 } while (++retries < FIB_DUMP_MAX_RETRIES);
181 EXPORT_SYMBOL(register_fib_notifier);
183 int unregister_fib_notifier(struct notifier_block *nb)
185 return atomic_notifier_chain_unregister(&fib_chain, nb);
187 EXPORT_SYMBOL(unregister_fib_notifier);
189 int call_fib_notifiers(struct net *net, enum fib_event_type event_type,
190 struct fib_notifier_info *info)
194 return atomic_notifier_call_chain(&fib_chain, event_type, info);
197 static int call_fib_entry_notifiers(struct net *net,
198 enum fib_event_type event_type, u32 dst,
199 int dst_len, struct fib_info *fi,
200 u8 tos, u8 type, u32 tb_id, u32 nlflags)
202 struct fib_entry_notifier_info info = {
211 return call_fib_notifiers(net, event_type, &info.info);
214 #define MAX_STAT_DEPTH 32
216 #define KEYLENGTH (8*sizeof(t_key))
217 #define KEY_MAX ((t_key)~0)
219 typedef unsigned int t_key;
221 #define IS_TRIE(n) ((n)->pos >= KEYLENGTH)
222 #define IS_TNODE(n) ((n)->bits)
223 #define IS_LEAF(n) (!(n)->bits)
227 unsigned char pos; /* 2log(KEYLENGTH) bits needed */
228 unsigned char bits; /* 2log(KEYLENGTH) bits needed */
231 /* This list pointer if valid if (pos | bits) == 0 (LEAF) */
232 struct hlist_head leaf;
233 /* This array is valid if (pos | bits) > 0 (TNODE) */
234 struct key_vector __rcu *tnode[0];
240 t_key empty_children; /* KEYLENGTH bits needed */
241 t_key full_children; /* KEYLENGTH bits needed */
242 struct key_vector __rcu *parent;
243 struct key_vector kv[1];
244 #define tn_bits kv[0].bits
247 #define TNODE_SIZE(n) offsetof(struct tnode, kv[0].tnode[n])
248 #define LEAF_SIZE TNODE_SIZE(1)
250 #ifdef CONFIG_IP_FIB_TRIE_STATS
251 struct trie_use_stats {
253 unsigned int backtrack;
254 unsigned int semantic_match_passed;
255 unsigned int semantic_match_miss;
256 unsigned int null_node_hit;
257 unsigned int resize_node_skipped;
262 unsigned int totdepth;
263 unsigned int maxdepth;
266 unsigned int nullpointers;
267 unsigned int prefixes;
268 unsigned int nodesizes[MAX_STAT_DEPTH];
272 struct key_vector kv[1];
273 #ifdef CONFIG_IP_FIB_TRIE_STATS
274 struct trie_use_stats __percpu *stats;
278 static struct key_vector *resize(struct trie *t, struct key_vector *tn);
279 static size_t tnode_free_size;
282 * synchronize_rcu after call_rcu for that many pages; it should be especially
283 * useful before resizing the root node with PREEMPT_NONE configs; the value was
284 * obtained experimentally, aiming to avoid visible slowdown.
286 static const int sync_pages = 128;
288 static struct kmem_cache *fn_alias_kmem __read_mostly;
289 static struct kmem_cache *trie_leaf_kmem __read_mostly;
291 static inline struct tnode *tn_info(struct key_vector *kv)
293 return container_of(kv, struct tnode, kv[0]);
296 /* caller must hold RTNL */
297 #define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
298 #define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
300 /* caller must hold RCU read lock or RTNL */
301 #define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
302 #define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
304 /* wrapper for rcu_assign_pointer */
305 static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
308 rcu_assign_pointer(tn_info(n)->parent, tp);
311 #define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
313 /* This provides us with the number of children in this node, in the case of a
314 * leaf this will return 0 meaning none of the children are accessible.
316 static inline unsigned long child_length(const struct key_vector *tn)
318 return (1ul << tn->bits) & ~(1ul);
321 #define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
323 static inline unsigned long get_index(t_key key, struct key_vector *kv)
325 unsigned long index = key ^ kv->key;
327 if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
330 return index >> kv->pos;
333 /* To understand this stuff, an understanding of keys and all their bits is
334 * necessary. Every node in the trie has a key associated with it, but not
335 * all of the bits in that key are significant.
337 * Consider a node 'n' and its parent 'tp'.
339 * If n is a leaf, every bit in its key is significant. Its presence is
340 * necessitated by path compression, since during a tree traversal (when
341 * searching for a leaf - unless we are doing an insertion) we will completely
342 * ignore all skipped bits we encounter. Thus we need to verify, at the end of
343 * a potentially successful search, that we have indeed been walking the
346 * Note that we can never "miss" the correct key in the tree if present by
347 * following the wrong path. Path compression ensures that segments of the key
348 * that are the same for all keys with a given prefix are skipped, but the
349 * skipped part *is* identical for each node in the subtrie below the skipped
350 * bit! trie_insert() in this implementation takes care of that.
352 * if n is an internal node - a 'tnode' here, the various parts of its key
353 * have many different meanings.
356 * _________________________________________________________________
357 * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
358 * -----------------------------------------------------------------
359 * 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16
361 * _________________________________________________________________
362 * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
363 * -----------------------------------------------------------------
364 * 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
371 * First, let's just ignore the bits that come before the parent tp, that is
372 * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
373 * point we do not use them for anything.
375 * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
376 * index into the parent's child array. That is, they will be used to find
377 * 'n' among tp's children.
379 * The bits from (n->pos + n->bits) to (tp->pos - 1) - "S" - are skipped bits
382 * All the bits we have seen so far are significant to the node n. The rest
383 * of the bits are really not needed or indeed known in n->key.
385 * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
386 * n's child array, and will of course be different for each child.
388 * The rest of the bits, from 0 to (n->pos -1) - "u" - are completely unknown
392 static const int halve_threshold = 25;
393 static const int inflate_threshold = 50;
394 static const int halve_threshold_root = 15;
395 static const int inflate_threshold_root = 30;
397 static void __alias_free_mem(struct rcu_head *head)
399 struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
400 kmem_cache_free(fn_alias_kmem, fa);
403 static inline void alias_free_mem_rcu(struct fib_alias *fa)
405 call_rcu(&fa->rcu, __alias_free_mem);
408 #define TNODE_KMALLOC_MAX \
409 ilog2((PAGE_SIZE - TNODE_SIZE(0)) / sizeof(struct key_vector *))
410 #define TNODE_VMALLOC_MAX \
411 ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
413 static void __node_free_rcu(struct rcu_head *head)
415 struct tnode *n = container_of(head, struct tnode, rcu);
418 kmem_cache_free(trie_leaf_kmem, n);
423 #define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
425 static struct tnode *tnode_alloc(int bits)
429 /* verify bits is within bounds */
430 if (bits > TNODE_VMALLOC_MAX)
433 /* determine size and verify it is non-zero and didn't overflow */
434 size = TNODE_SIZE(1ul << bits);
436 if (size <= PAGE_SIZE)
437 return kzalloc(size, GFP_KERNEL);
439 return vzalloc(size);
442 static inline void empty_child_inc(struct key_vector *n)
444 ++tn_info(n)->empty_children ? : ++tn_info(n)->full_children;
447 static inline void empty_child_dec(struct key_vector *n)
449 tn_info(n)->empty_children-- ? : tn_info(n)->full_children--;
452 static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
454 struct key_vector *l;
457 kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
461 /* initialize key vector */
466 l->slen = fa->fa_slen;
468 /* link leaf to fib alias */
469 INIT_HLIST_HEAD(&l->leaf);
470 hlist_add_head(&fa->fa_list, &l->leaf);
475 static struct key_vector *tnode_new(t_key key, int pos, int bits)
477 unsigned int shift = pos + bits;
478 struct key_vector *tn;
481 /* verify bits and pos their msb bits clear and values are valid */
482 BUG_ON(!bits || (shift > KEYLENGTH));
484 tnode = tnode_alloc(bits);
488 pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
489 sizeof(struct key_vector *) << bits);
491 if (bits == KEYLENGTH)
492 tnode->full_children = 1;
494 tnode->empty_children = 1ul << bits;
497 tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
505 /* Check whether a tnode 'n' is "full", i.e. it is an internal node
506 * and no bits are skipped. See discussion in dyntree paper p. 6
508 static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
510 return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
513 /* Add a child at position i overwriting the old value.
514 * Update the value of full_children and empty_children.
516 static void put_child(struct key_vector *tn, unsigned long i,
517 struct key_vector *n)
519 struct key_vector *chi = get_child(tn, i);
522 BUG_ON(i >= child_length(tn));
524 /* update emptyChildren, overflow into fullChildren */
530 /* update fullChildren */
531 wasfull = tnode_full(tn, chi);
532 isfull = tnode_full(tn, n);
534 if (wasfull && !isfull)
535 tn_info(tn)->full_children--;
536 else if (!wasfull && isfull)
537 tn_info(tn)->full_children++;
539 if (n && (tn->slen < n->slen))
542 rcu_assign_pointer(tn->tnode[i], n);
545 static void update_children(struct key_vector *tn)
549 /* update all of the child parent pointers */
550 for (i = child_length(tn); i;) {
551 struct key_vector *inode = get_child(tn, --i);
556 /* Either update the children of a tnode that
557 * already belongs to us or update the child
558 * to point to ourselves.
560 if (node_parent(inode) == tn)
561 update_children(inode);
563 node_set_parent(inode, tn);
567 static inline void put_child_root(struct key_vector *tp, t_key key,
568 struct key_vector *n)
571 rcu_assign_pointer(tp->tnode[0], n);
573 put_child(tp, get_index(key, tp), n);
576 static inline void tnode_free_init(struct key_vector *tn)
578 tn_info(tn)->rcu.next = NULL;
581 static inline void tnode_free_append(struct key_vector *tn,
582 struct key_vector *n)
584 tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
585 tn_info(tn)->rcu.next = &tn_info(n)->rcu;
588 static void tnode_free(struct key_vector *tn)
590 struct callback_head *head = &tn_info(tn)->rcu;
594 tnode_free_size += TNODE_SIZE(1ul << tn->bits);
597 tn = container_of(head, struct tnode, rcu)->kv;
600 if (tnode_free_size >= PAGE_SIZE * sync_pages) {
606 static struct key_vector *replace(struct trie *t,
607 struct key_vector *oldtnode,
608 struct key_vector *tn)
610 struct key_vector *tp = node_parent(oldtnode);
613 /* setup the parent pointer out of and back into this node */
614 NODE_INIT_PARENT(tn, tp);
615 put_child_root(tp, tn->key, tn);
617 /* update all of the child parent pointers */
620 /* all pointers should be clean so we are done */
621 tnode_free(oldtnode);
623 /* resize children now that oldtnode is freed */
624 for (i = child_length(tn); i;) {
625 struct key_vector *inode = get_child(tn, --i);
627 /* resize child node */
628 if (tnode_full(tn, inode))
629 tn = resize(t, inode);
635 static struct key_vector *inflate(struct trie *t,
636 struct key_vector *oldtnode)
638 struct key_vector *tn;
642 pr_debug("In inflate\n");
644 tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
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
656 for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
657 struct key_vector *inode = get_child(oldtnode, --i);
658 struct key_vector *node0, *node1;
665 /* A leaf or an internal node with skipped bits */
666 if (!tnode_full(oldtnode, inode)) {
667 put_child(tn, get_index(inode->key, tn), inode);
671 /* drop the node in the old tnode free list */
672 tnode_free_append(oldtnode, inode);
674 /* An internal node with two children */
675 if (inode->bits == 1) {
676 put_child(tn, 2 * i + 1, get_child(inode, 1));
677 put_child(tn, 2 * i, get_child(inode, 0));
681 /* We will replace this node 'inode' with two new
682 * ones, 'node0' and 'node1', each with half of the
683 * original children. The two new nodes will have
684 * a position one bit further down the key and this
685 * means that the "significant" part of their keys
686 * (see the discussion near the top of this file)
687 * will differ by one bit, which will be "0" in
688 * node0's key and "1" in node1's key. Since we are
689 * moving the key position by one step, the bit that
690 * we are moving away from - the bit at position
691 * (tn->pos) - is the one that will differ between
692 * node0 and node1. So... we synthesize that bit in the
695 node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
698 node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
700 tnode_free_append(tn, node1);
703 tnode_free_append(tn, node0);
705 /* populate child pointers in new nodes */
706 for (k = child_length(inode), j = k / 2; j;) {
707 put_child(node1, --j, get_child(inode, --k));
708 put_child(node0, j, get_child(inode, j));
709 put_child(node1, --j, get_child(inode, --k));
710 put_child(node0, j, get_child(inode, j));
713 /* link new nodes to parent */
714 NODE_INIT_PARENT(node1, tn);
715 NODE_INIT_PARENT(node0, tn);
717 /* link parent to nodes */
718 put_child(tn, 2 * i + 1, node1);
719 put_child(tn, 2 * i, node0);
722 /* setup the parent pointers into and out of this node */
723 return replace(t, oldtnode, tn);
725 /* all pointers should be clean so we are done */
731 static struct key_vector *halve(struct trie *t,
732 struct key_vector *oldtnode)
734 struct key_vector *tn;
737 pr_debug("In halve\n");
739 tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
743 /* prepare oldtnode to be freed */
744 tnode_free_init(oldtnode);
746 /* Assemble all of the pointers in our cluster, in this case that
747 * represents all of the pointers out of our allocated nodes that
748 * point to existing tnodes and the links between our allocated
751 for (i = child_length(oldtnode); i;) {
752 struct key_vector *node1 = get_child(oldtnode, --i);
753 struct key_vector *node0 = get_child(oldtnode, --i);
754 struct key_vector *inode;
756 /* At least one of the children is empty */
757 if (!node1 || !node0) {
758 put_child(tn, i / 2, node1 ? : node0);
762 /* Two nonempty children */
763 inode = tnode_new(node0->key, oldtnode->pos, 1);
766 tnode_free_append(tn, inode);
768 /* initialize pointers out of node */
769 put_child(inode, 1, node1);
770 put_child(inode, 0, node0);
771 NODE_INIT_PARENT(inode, tn);
773 /* link parent to node */
774 put_child(tn, i / 2, inode);
777 /* setup the parent pointers into and out of this node */
778 return replace(t, oldtnode, tn);
780 /* all pointers should be clean so we are done */
786 static struct key_vector *collapse(struct trie *t,
787 struct key_vector *oldtnode)
789 struct key_vector *n, *tp;
792 /* scan the tnode looking for that one child that might still exist */
793 for (n = NULL, i = child_length(oldtnode); !n && i;)
794 n = get_child(oldtnode, --i);
796 /* compress one level */
797 tp = node_parent(oldtnode);
798 put_child_root(tp, oldtnode->key, n);
799 node_set_parent(n, tp);
807 static unsigned char update_suffix(struct key_vector *tn)
809 unsigned char slen = tn->pos;
810 unsigned long stride, i;
812 /* search though the list of children looking for nodes that might
813 * have a suffix greater than the one we currently have. This is
814 * why we start with a stride of 2 since a stride of 1 would
815 * represent the nodes with suffix length equal to tn->pos
817 for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
818 struct key_vector *n = get_child(tn, i);
820 if (!n || (n->slen <= slen))
823 /* update stride and slen based on new value */
824 stride <<= (n->slen - slen);
828 /* if slen covers all but the last bit we can stop here
829 * there will be nothing longer than that since only node
830 * 0 and 1 << (bits - 1) could have that as their suffix
833 if ((slen + 1) >= (tn->pos + tn->bits))
842 /* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
843 * the Helsinki University of Technology and Matti Tikkanen of Nokia
844 * Telecommunications, page 6:
845 * "A node is doubled if the ratio of non-empty children to all
846 * children in the *doubled* node is at least 'high'."
848 * 'high' in this instance is the variable 'inflate_threshold'. It
849 * is expressed as a percentage, so we multiply it with
850 * child_length() and instead of multiplying by 2 (since the
851 * child array will be doubled by inflate()) and multiplying
852 * the left-hand side by 100 (to handle the percentage thing) we
853 * multiply the left-hand side by 50.
855 * The left-hand side may look a bit weird: child_length(tn)
856 * - tn->empty_children is of course the number of non-null children
857 * in the current node. tn->full_children is the number of "full"
858 * children, that is non-null tnodes with a skip value of 0.
859 * All of those will be doubled in the resulting inflated tnode, so
860 * we just count them one extra time here.
862 * A clearer way to write this would be:
864 * to_be_doubled = tn->full_children;
865 * not_to_be_doubled = child_length(tn) - tn->empty_children -
868 * new_child_length = child_length(tn) * 2;
870 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
872 * if (new_fill_factor >= inflate_threshold)
874 * ...and so on, tho it would mess up the while () loop.
877 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
881 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
882 * inflate_threshold * new_child_length
884 * expand not_to_be_doubled and to_be_doubled, and shorten:
885 * 100 * (child_length(tn) - tn->empty_children +
886 * tn->full_children) >= inflate_threshold * new_child_length
888 * expand new_child_length:
889 * 100 * (child_length(tn) - tn->empty_children +
890 * tn->full_children) >=
891 * inflate_threshold * child_length(tn) * 2
894 * 50 * (tn->full_children + child_length(tn) -
895 * tn->empty_children) >= inflate_threshold *
899 static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
901 unsigned long used = child_length(tn);
902 unsigned long threshold = used;
904 /* Keep root node larger */
905 threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
906 used -= tn_info(tn)->empty_children;
907 used += tn_info(tn)->full_children;
909 /* if bits == KEYLENGTH then pos = 0, and will fail below */
911 return (used > 1) && tn->pos && ((50 * used) >= threshold);
914 static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
916 unsigned long used = child_length(tn);
917 unsigned long threshold = used;
919 /* Keep root node larger */
920 threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
921 used -= tn_info(tn)->empty_children;
923 /* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
925 return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
928 static inline bool should_collapse(struct key_vector *tn)
930 unsigned long used = child_length(tn);
932 used -= tn_info(tn)->empty_children;
934 /* account for bits == KEYLENGTH case */
935 if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
938 /* One child or none, time to drop us from the trie */
943 static struct key_vector *resize(struct trie *t, struct key_vector *tn)
945 #ifdef CONFIG_IP_FIB_TRIE_STATS
946 struct trie_use_stats __percpu *stats = t->stats;
948 struct key_vector *tp = node_parent(tn);
949 unsigned long cindex = get_index(tn->key, tp);
950 int max_work = MAX_WORK;
952 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
953 tn, inflate_threshold, halve_threshold);
955 /* track the tnode via the pointer from the parent instead of
956 * doing it ourselves. This way we can let RCU fully do its
957 * thing without us interfering
959 BUG_ON(tn != get_child(tp, cindex));
961 /* Double as long as the resulting node has a number of
962 * nonempty nodes that are above the threshold.
964 while (should_inflate(tp, tn) && max_work) {
967 #ifdef CONFIG_IP_FIB_TRIE_STATS
968 this_cpu_inc(stats->resize_node_skipped);
974 tn = get_child(tp, cindex);
977 /* update parent in case inflate failed */
978 tp = node_parent(tn);
980 /* Return if at least one inflate is run */
981 if (max_work != MAX_WORK)
984 /* Halve as long as the number of empty children in this
985 * node is above threshold.
987 while (should_halve(tp, tn) && max_work) {
990 #ifdef CONFIG_IP_FIB_TRIE_STATS
991 this_cpu_inc(stats->resize_node_skipped);
997 tn = get_child(tp, cindex);
1000 /* Only one child remains */
1001 if (should_collapse(tn))
1002 return collapse(t, tn);
1004 /* update parent in case halve failed */
1005 tp = node_parent(tn);
1007 /* Return if at least one deflate was run */
1008 if (max_work != MAX_WORK)
1011 /* push the suffix length to the parent node */
1012 if (tn->slen > tn->pos) {
1013 unsigned char slen = update_suffix(tn);
1015 if (slen > tp->slen)
1022 static void leaf_pull_suffix(struct key_vector *tp, struct key_vector *l)
1024 while ((tp->slen > tp->pos) && (tp->slen > l->slen)) {
1025 if (update_suffix(tp) > l->slen)
1027 tp = node_parent(tp);
1031 static void leaf_push_suffix(struct key_vector *tn, struct key_vector *l)
1033 /* if this is a new leaf then tn will be NULL and we can sort
1034 * out parent suffix lengths as a part of trie_rebalance
1036 while (tn->slen < l->slen) {
1038 tn = node_parent(tn);
1042 /* rcu_read_lock needs to be hold by caller from readside */
1043 static struct key_vector *fib_find_node(struct trie *t,
1044 struct key_vector **tp, u32 key)
1046 struct key_vector *pn, *n = t->kv;
1047 unsigned long index = 0;
1051 n = get_child_rcu(n, index);
1056 index = get_cindex(key, n);
1058 /* This bit of code is a bit tricky but it combines multiple
1059 * checks into a single check. The prefix consists of the
1060 * prefix plus zeros for the bits in the cindex. The index
1061 * is the difference between the key and this value. From
1062 * this we can actually derive several pieces of data.
1063 * if (index >= (1ul << bits))
1064 * we have a mismatch in skip bits and failed
1066 * we know the value is cindex
1068 * This check is safe even if bits == KEYLENGTH due to the
1069 * fact that we can only allocate a node with 32 bits if a
1070 * long is greater than 32 bits.
1072 if (index >= (1ul << n->bits)) {
1077 /* keep searching until we find a perfect match leaf or NULL */
1078 } while (IS_TNODE(n));
1085 /* Return the first fib alias matching TOS with
1086 * priority less than or equal to PRIO.
1088 static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
1089 u8 tos, u32 prio, u32 tb_id)
1091 struct fib_alias *fa;
1096 hlist_for_each_entry(fa, fah, fa_list) {
1097 if (fa->fa_slen < slen)
1099 if (fa->fa_slen != slen)
1101 if (fa->tb_id > tb_id)
1103 if (fa->tb_id != tb_id)
1105 if (fa->fa_tos > tos)
1107 if (fa->fa_info->fib_priority >= prio || fa->fa_tos < tos)
1114 static void trie_rebalance(struct trie *t, struct key_vector *tn)
1116 while (!IS_TRIE(tn))
1120 static int fib_insert_node(struct trie *t, struct key_vector *tp,
1121 struct fib_alias *new, t_key key)
1123 struct key_vector *n, *l;
1125 l = leaf_new(key, new);
1129 /* retrieve child from parent node */
1130 n = get_child(tp, get_index(key, tp));
1132 /* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1134 * Add a new tnode here
1135 * first tnode need some special handling
1136 * leaves us in position for handling as case 3
1139 struct key_vector *tn;
1141 tn = tnode_new(key, __fls(key ^ n->key), 1);
1145 /* initialize routes out of node */
1146 NODE_INIT_PARENT(tn, tp);
1147 put_child(tn, get_index(key, tn) ^ 1, n);
1149 /* start adding routes into the node */
1150 put_child_root(tp, key, tn);
1151 node_set_parent(n, tn);
1153 /* parent now has a NULL spot where the leaf can go */
1157 /* Case 3: n is NULL, and will just insert a new leaf */
1158 NODE_INIT_PARENT(l, tp);
1159 put_child_root(tp, key, l);
1160 trie_rebalance(t, tp);
1169 static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1170 struct key_vector *l, struct fib_alias *new,
1171 struct fib_alias *fa, t_key key)
1174 return fib_insert_node(t, tp, new, key);
1177 hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1179 struct fib_alias *last;
1181 hlist_for_each_entry(last, &l->leaf, fa_list) {
1182 if (new->fa_slen < last->fa_slen)
1184 if ((new->fa_slen == last->fa_slen) &&
1185 (new->tb_id > last->tb_id))
1191 hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1193 hlist_add_head_rcu(&new->fa_list, &l->leaf);
1196 /* if we added to the tail node then we need to update slen */
1197 if (l->slen < new->fa_slen) {
1198 l->slen = new->fa_slen;
1199 leaf_push_suffix(tp, l);
1205 /* Caller must hold RTNL. */
1206 int fib_table_insert(struct net *net, struct fib_table *tb,
1207 struct fib_config *cfg)
1209 struct trie *t = (struct trie *)tb->tb_data;
1210 struct fib_alias *fa, *new_fa;
1211 struct key_vector *l, *tp;
1212 u16 nlflags = NLM_F_EXCL;
1213 struct fib_info *fi;
1214 u8 plen = cfg->fc_dst_len;
1215 u8 slen = KEYLENGTH - plen;
1216 u8 tos = cfg->fc_tos;
1220 if (plen > KEYLENGTH)
1223 key = ntohl(cfg->fc_dst);
1225 pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1227 if ((plen < KEYLENGTH) && (key << plen))
1230 fi = fib_create_info(cfg);
1236 l = fib_find_node(t, &tp, key);
1237 fa = l ? fib_find_alias(&l->leaf, slen, tos, fi->fib_priority,
1240 /* Now fa, if non-NULL, points to the first fib alias
1241 * with the same keys [prefix,tos,priority], if such key already
1242 * exists or to the node before which we will insert new one.
1244 * If fa is NULL, we will need to allocate a new one and
1245 * insert to the tail of the section matching the suffix length
1249 if (fa && fa->fa_tos == tos &&
1250 fa->fa_info->fib_priority == fi->fib_priority) {
1251 struct fib_alias *fa_first, *fa_match;
1254 if (cfg->fc_nlflags & NLM_F_EXCL)
1257 nlflags &= ~NLM_F_EXCL;
1260 * 1. Find exact match for type, scope, fib_info to avoid
1262 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1266 hlist_for_each_entry_from(fa, fa_list) {
1267 if ((fa->fa_slen != slen) ||
1268 (fa->tb_id != tb->tb_id) ||
1269 (fa->fa_tos != tos))
1271 if (fa->fa_info->fib_priority != fi->fib_priority)
1273 if (fa->fa_type == cfg->fc_type &&
1274 fa->fa_info == fi) {
1280 if (cfg->fc_nlflags & NLM_F_REPLACE) {
1281 struct fib_info *fi_drop;
1284 nlflags |= NLM_F_REPLACE;
1292 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1296 fi_drop = fa->fa_info;
1297 new_fa->fa_tos = fa->fa_tos;
1298 new_fa->fa_info = fi;
1299 new_fa->fa_type = cfg->fc_type;
1300 state = fa->fa_state;
1301 new_fa->fa_state = state & ~FA_S_ACCESSED;
1302 new_fa->fa_slen = fa->fa_slen;
1303 new_fa->tb_id = tb->tb_id;
1304 new_fa->fa_default = -1;
1306 hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1308 alias_free_mem_rcu(fa);
1310 fib_release_info(fi_drop);
1311 if (state & FA_S_ACCESSED)
1312 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1314 call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_ADD,
1316 new_fa->fa_tos, cfg->fc_type,
1317 tb->tb_id, cfg->fc_nlflags);
1318 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1319 tb->tb_id, &cfg->fc_nlinfo, nlflags);
1323 /* Error if we find a perfect match which
1324 * uses the same scope, type, and nexthop
1330 if (cfg->fc_nlflags & NLM_F_APPEND)
1331 nlflags |= NLM_F_APPEND;
1336 if (!(cfg->fc_nlflags & NLM_F_CREATE))
1339 nlflags |= NLM_F_CREATE;
1341 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1345 new_fa->fa_info = fi;
1346 new_fa->fa_tos = tos;
1347 new_fa->fa_type = cfg->fc_type;
1348 new_fa->fa_state = 0;
1349 new_fa->fa_slen = slen;
1350 new_fa->tb_id = tb->tb_id;
1351 new_fa->fa_default = -1;
1353 /* Insert new entry to the list. */
1354 err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1356 goto out_free_new_fa;
1359 tb->tb_num_default++;
1361 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1362 call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_ADD, key, plen, fi, tos,
1363 cfg->fc_type, tb->tb_id, cfg->fc_nlflags);
1364 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1365 &cfg->fc_nlinfo, nlflags);
1370 kmem_cache_free(fn_alias_kmem, new_fa);
1372 fib_release_info(fi);
1377 static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1379 t_key prefix = n->key;
1381 return (key ^ prefix) & (prefix | -prefix);
1384 /* should be called with rcu_read_lock */
1385 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1386 struct fib_result *res, int fib_flags)
1388 struct trie *t = (struct trie *) tb->tb_data;
1389 #ifdef CONFIG_IP_FIB_TRIE_STATS
1390 struct trie_use_stats __percpu *stats = t->stats;
1392 const t_key key = ntohl(flp->daddr);
1393 struct key_vector *n, *pn;
1394 struct fib_alias *fa;
1395 unsigned long index;
1398 trace_fib_table_lookup(tb->tb_id, flp);
1403 n = get_child_rcu(pn, cindex);
1407 #ifdef CONFIG_IP_FIB_TRIE_STATS
1408 this_cpu_inc(stats->gets);
1411 /* Step 1: Travel to the longest prefix match in the trie */
1413 index = get_cindex(key, n);
1415 /* This bit of code is a bit tricky but it combines multiple
1416 * checks into a single check. The prefix consists of the
1417 * prefix plus zeros for the "bits" in the prefix. The index
1418 * is the difference between the key and this value. From
1419 * this we can actually derive several pieces of data.
1420 * if (index >= (1ul << bits))
1421 * we have a mismatch in skip bits and failed
1423 * we know the value is cindex
1425 * This check is safe even if bits == KEYLENGTH due to the
1426 * fact that we can only allocate a node with 32 bits if a
1427 * long is greater than 32 bits.
1429 if (index >= (1ul << n->bits))
1432 /* we have found a leaf. Prefixes have already been compared */
1436 /* only record pn and cindex if we are going to be chopping
1437 * bits later. Otherwise we are just wasting cycles.
1439 if (n->slen > n->pos) {
1444 n = get_child_rcu(n, index);
1449 /* Step 2: Sort out leaves and begin backtracing for longest prefix */
1451 /* record the pointer where our next node pointer is stored */
1452 struct key_vector __rcu **cptr = n->tnode;
1454 /* This test verifies that none of the bits that differ
1455 * between the key and the prefix exist in the region of
1456 * the lsb and higher in the prefix.
1458 if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1461 /* exit out and process leaf */
1462 if (unlikely(IS_LEAF(n)))
1465 /* Don't bother recording parent info. Since we are in
1466 * prefix match mode we will have to come back to wherever
1467 * we started this traversal anyway
1470 while ((n = rcu_dereference(*cptr)) == NULL) {
1472 #ifdef CONFIG_IP_FIB_TRIE_STATS
1474 this_cpu_inc(stats->null_node_hit);
1476 /* If we are at cindex 0 there are no more bits for
1477 * us to strip at this level so we must ascend back
1478 * up one level to see if there are any more bits to
1479 * be stripped there.
1482 t_key pkey = pn->key;
1484 /* If we don't have a parent then there is
1485 * nothing for us to do as we do not have any
1486 * further nodes to parse.
1490 #ifdef CONFIG_IP_FIB_TRIE_STATS
1491 this_cpu_inc(stats->backtrack);
1493 /* Get Child's index */
1494 pn = node_parent_rcu(pn);
1495 cindex = get_index(pkey, pn);
1498 /* strip the least significant bit from the cindex */
1499 cindex &= cindex - 1;
1501 /* grab pointer for next child node */
1502 cptr = &pn->tnode[cindex];
1507 /* this line carries forward the xor from earlier in the function */
1508 index = key ^ n->key;
1510 /* Step 3: Process the leaf, if that fails fall back to backtracing */
1511 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1512 struct fib_info *fi = fa->fa_info;
1515 if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1516 if (index >= (1ul << fa->fa_slen))
1519 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1523 if (fa->fa_info->fib_scope < flp->flowi4_scope)
1525 fib_alias_accessed(fa);
1526 err = fib_props[fa->fa_type].error;
1527 if (unlikely(err < 0)) {
1528 #ifdef CONFIG_IP_FIB_TRIE_STATS
1529 this_cpu_inc(stats->semantic_match_passed);
1533 if (fi->fib_flags & RTNH_F_DEAD)
1535 for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
1536 const struct fib_nh *nh = &fi->fib_nh[nhsel];
1537 struct in_device *in_dev = __in_dev_get_rcu(nh->nh_dev);
1539 if (nh->nh_flags & RTNH_F_DEAD)
1542 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1543 nh->nh_flags & RTNH_F_LINKDOWN &&
1544 !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1546 if (!(flp->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF)) {
1547 if (flp->flowi4_oif &&
1548 flp->flowi4_oif != nh->nh_oif)
1552 if (!(fib_flags & FIB_LOOKUP_NOREF))
1553 atomic_inc(&fi->fib_clntref);
1555 res->prefixlen = KEYLENGTH - fa->fa_slen;
1556 res->nh_sel = nhsel;
1557 res->type = fa->fa_type;
1558 res->scope = fi->fib_scope;
1561 res->fa_head = &n->leaf;
1562 #ifdef CONFIG_IP_FIB_TRIE_STATS
1563 this_cpu_inc(stats->semantic_match_passed);
1565 trace_fib_table_lookup_nh(nh);
1570 #ifdef CONFIG_IP_FIB_TRIE_STATS
1571 this_cpu_inc(stats->semantic_match_miss);
1575 EXPORT_SYMBOL_GPL(fib_table_lookup);
1577 static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1578 struct key_vector *l, struct fib_alias *old)
1580 /* record the location of the previous list_info entry */
1581 struct hlist_node **pprev = old->fa_list.pprev;
1582 struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1584 /* remove the fib_alias from the list */
1585 hlist_del_rcu(&old->fa_list);
1587 /* if we emptied the list this leaf will be freed and we can sort
1588 * out parent suffix lengths as a part of trie_rebalance
1590 if (hlist_empty(&l->leaf)) {
1591 put_child_root(tp, l->key, NULL);
1593 trie_rebalance(t, tp);
1597 /* only access fa if it is pointing at the last valid hlist_node */
1601 /* update the trie with the latest suffix length */
1602 l->slen = fa->fa_slen;
1603 leaf_pull_suffix(tp, l);
1606 /* Caller must hold RTNL. */
1607 int fib_table_delete(struct net *net, struct fib_table *tb,
1608 struct fib_config *cfg)
1610 struct trie *t = (struct trie *) tb->tb_data;
1611 struct fib_alias *fa, *fa_to_delete;
1612 struct key_vector *l, *tp;
1613 u8 plen = cfg->fc_dst_len;
1614 u8 slen = KEYLENGTH - plen;
1615 u8 tos = cfg->fc_tos;
1618 if (plen > KEYLENGTH)
1621 key = ntohl(cfg->fc_dst);
1623 if ((plen < KEYLENGTH) && (key << plen))
1626 l = fib_find_node(t, &tp, key);
1630 fa = fib_find_alias(&l->leaf, slen, tos, 0, tb->tb_id);
1634 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1636 fa_to_delete = NULL;
1637 hlist_for_each_entry_from(fa, fa_list) {
1638 struct fib_info *fi = fa->fa_info;
1640 if ((fa->fa_slen != slen) ||
1641 (fa->tb_id != tb->tb_id) ||
1642 (fa->fa_tos != tos))
1645 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1646 (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1647 fa->fa_info->fib_scope == cfg->fc_scope) &&
1648 (!cfg->fc_prefsrc ||
1649 fi->fib_prefsrc == cfg->fc_prefsrc) &&
1650 (!cfg->fc_protocol ||
1651 fi->fib_protocol == cfg->fc_protocol) &&
1652 fib_nh_match(cfg, fi) == 0) {
1661 call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, key, plen,
1662 fa_to_delete->fa_info, tos, cfg->fc_type,
1664 rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1665 &cfg->fc_nlinfo, 0);
1668 tb->tb_num_default--;
1670 fib_remove_alias(t, tp, l, fa_to_delete);
1672 if (fa_to_delete->fa_state & FA_S_ACCESSED)
1673 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1675 fib_release_info(fa_to_delete->fa_info);
1676 alias_free_mem_rcu(fa_to_delete);
1680 /* Scan for the next leaf starting at the provided key value */
1681 static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1683 struct key_vector *pn, *n = *tn;
1684 unsigned long cindex;
1686 /* this loop is meant to try and find the key in the trie */
1688 /* record parent and next child index */
1690 cindex = (key > pn->key) ? get_index(key, pn) : 0;
1692 if (cindex >> pn->bits)
1695 /* descend into the next child */
1696 n = get_child_rcu(pn, cindex++);
1700 /* guarantee forward progress on the keys */
1701 if (IS_LEAF(n) && (n->key >= key))
1703 } while (IS_TNODE(n));
1705 /* this loop will search for the next leaf with a greater key */
1706 while (!IS_TRIE(pn)) {
1707 /* if we exhausted the parent node we will need to climb */
1708 if (cindex >= (1ul << pn->bits)) {
1709 t_key pkey = pn->key;
1711 pn = node_parent_rcu(pn);
1712 cindex = get_index(pkey, pn) + 1;
1716 /* grab the next available node */
1717 n = get_child_rcu(pn, cindex++);
1721 /* no need to compare keys since we bumped the index */
1725 /* Rescan start scanning in new node */
1731 return NULL; /* Root of trie */
1733 /* if we are at the limit for keys just return NULL for the tnode */
1738 static void fib_trie_free(struct fib_table *tb)
1740 struct trie *t = (struct trie *)tb->tb_data;
1741 struct key_vector *pn = t->kv;
1742 unsigned long cindex = 1;
1743 struct hlist_node *tmp;
1744 struct fib_alias *fa;
1746 /* walk trie in reverse order and free everything */
1748 struct key_vector *n;
1751 t_key pkey = pn->key;
1757 pn = node_parent(pn);
1759 /* drop emptied tnode */
1760 put_child_root(pn, n->key, NULL);
1763 cindex = get_index(pkey, pn);
1768 /* grab the next available node */
1769 n = get_child(pn, cindex);
1774 /* record pn and cindex for leaf walking */
1776 cindex = 1ul << n->bits;
1781 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1782 hlist_del_rcu(&fa->fa_list);
1783 alias_free_mem_rcu(fa);
1786 put_child_root(pn, n->key, NULL);
1790 #ifdef CONFIG_IP_FIB_TRIE_STATS
1791 free_percpu(t->stats);
1796 struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1798 struct trie *ot = (struct trie *)oldtb->tb_data;
1799 struct key_vector *l, *tp = ot->kv;
1800 struct fib_table *local_tb;
1801 struct fib_alias *fa;
1805 if (oldtb->tb_data == oldtb->__data)
1808 local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1812 lt = (struct trie *)local_tb->tb_data;
1814 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1815 struct key_vector *local_l = NULL, *local_tp;
1817 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1818 struct fib_alias *new_fa;
1820 if (local_tb->tb_id != fa->tb_id)
1823 /* clone fa for new local table */
1824 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1828 memcpy(new_fa, fa, sizeof(*fa));
1830 /* insert clone into table */
1832 local_l = fib_find_node(lt, &local_tp, l->key);
1834 if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1836 kmem_cache_free(fn_alias_kmem, new_fa);
1841 /* stop loop if key wrapped back to 0 */
1849 fib_trie_free(local_tb);
1854 /* Caller must hold RTNL */
1855 void fib_table_flush_external(struct fib_table *tb)
1857 struct trie *t = (struct trie *)tb->tb_data;
1858 struct key_vector *pn = t->kv;
1859 unsigned long cindex = 1;
1860 struct hlist_node *tmp;
1861 struct fib_alias *fa;
1863 /* walk trie in reverse order */
1865 unsigned char slen = 0;
1866 struct key_vector *n;
1869 t_key pkey = pn->key;
1871 /* cannot resize the trie vector */
1875 /* resize completed node */
1877 cindex = get_index(pkey, pn);
1882 /* grab the next available node */
1883 n = get_child(pn, cindex);
1888 /* record pn and cindex for leaf walking */
1890 cindex = 1ul << n->bits;
1895 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1896 /* if alias was cloned to local then we just
1897 * need to remove the local copy from main
1899 if (tb->tb_id != fa->tb_id) {
1900 hlist_del_rcu(&fa->fa_list);
1901 alias_free_mem_rcu(fa);
1905 /* record local slen */
1909 /* update leaf slen */
1912 if (hlist_empty(&n->leaf)) {
1913 put_child_root(pn, n->key, NULL);
1919 /* Caller must hold RTNL. */
1920 int fib_table_flush(struct net *net, struct fib_table *tb)
1922 struct trie *t = (struct trie *)tb->tb_data;
1923 struct key_vector *pn = t->kv;
1924 unsigned long cindex = 1;
1925 struct hlist_node *tmp;
1926 struct fib_alias *fa;
1929 /* walk trie in reverse order */
1931 unsigned char slen = 0;
1932 struct key_vector *n;
1935 t_key pkey = pn->key;
1937 /* cannot resize the trie vector */
1941 /* resize completed node */
1943 cindex = get_index(pkey, pn);
1948 /* grab the next available node */
1949 n = get_child(pn, cindex);
1954 /* record pn and cindex for leaf walking */
1956 cindex = 1ul << n->bits;
1961 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1962 struct fib_info *fi = fa->fa_info;
1964 if (!fi || !(fi->fib_flags & RTNH_F_DEAD)) {
1969 call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL,
1971 KEYLENGTH - fa->fa_slen,
1972 fi, fa->fa_tos, fa->fa_type,
1974 hlist_del_rcu(&fa->fa_list);
1975 fib_release_info(fa->fa_info);
1976 alias_free_mem_rcu(fa);
1980 /* update leaf slen */
1983 if (hlist_empty(&n->leaf)) {
1984 put_child_root(pn, n->key, NULL);
1989 pr_debug("trie_flush found=%d\n", found);
1993 static void fib_leaf_notify(struct net *net, struct key_vector *l,
1994 struct fib_table *tb, struct notifier_block *nb,
1995 enum fib_event_type event_type)
1997 struct fib_alias *fa;
1999 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2000 struct fib_info *fi = fa->fa_info;
2005 /* local and main table can share the same trie,
2006 * so don't notify twice for the same entry.
2008 if (tb->tb_id != fa->tb_id)
2011 call_fib_entry_notifier(nb, net, event_type, l->key,
2012 KEYLENGTH - fa->fa_slen, fi, fa->fa_tos,
2013 fa->fa_type, fa->tb_id, 0);
2017 static void fib_table_notify(struct net *net, struct fib_table *tb,
2018 struct notifier_block *nb,
2019 enum fib_event_type event_type)
2021 struct trie *t = (struct trie *)tb->tb_data;
2022 struct key_vector *l, *tp = t->kv;
2025 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2026 fib_leaf_notify(net, l, tb, nb, event_type);
2029 /* stop in case of wrap around */
2035 static void fib_notify(struct net *net, struct notifier_block *nb,
2036 enum fib_event_type event_type)
2040 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2041 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2042 struct fib_table *tb;
2044 hlist_for_each_entry_rcu(tb, head, tb_hlist)
2045 fib_table_notify(net, tb, nb, event_type);
2049 static void __trie_free_rcu(struct rcu_head *head)
2051 struct fib_table *tb = container_of(head, struct fib_table, rcu);
2052 #ifdef CONFIG_IP_FIB_TRIE_STATS
2053 struct trie *t = (struct trie *)tb->tb_data;
2055 if (tb->tb_data == tb->__data)
2056 free_percpu(t->stats);
2057 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2061 void fib_free_table(struct fib_table *tb)
2063 call_rcu(&tb->rcu, __trie_free_rcu);
2066 static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
2067 struct sk_buff *skb, struct netlink_callback *cb)
2069 __be32 xkey = htonl(l->key);
2070 struct fib_alias *fa;
2076 /* rcu_read_lock is hold by caller */
2077 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2083 if (tb->tb_id != fa->tb_id) {
2088 if (fib_dump_info(skb, NETLINK_CB(cb->skb).portid,
2094 KEYLENGTH - fa->fa_slen,
2096 fa->fa_info, NLM_F_MULTI) < 0) {
2107 /* rcu_read_lock needs to be hold by caller from readside */
2108 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
2109 struct netlink_callback *cb)
2111 struct trie *t = (struct trie *)tb->tb_data;
2112 struct key_vector *l, *tp = t->kv;
2113 /* Dump starting at last key.
2114 * Note: 0.0.0.0/0 (ie default) is first key.
2116 int count = cb->args[2];
2117 t_key key = cb->args[3];
2119 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2120 if (fn_trie_dump_leaf(l, tb, skb, cb) < 0) {
2122 cb->args[2] = count;
2129 memset(&cb->args[4], 0,
2130 sizeof(cb->args) - 4*sizeof(cb->args[0]));
2132 /* stop loop if key wrapped back to 0 */
2138 cb->args[2] = count;
2143 void __init fib_trie_init(void)
2145 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
2146 sizeof(struct fib_alias),
2147 0, SLAB_PANIC, NULL);
2149 trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
2151 0, SLAB_PANIC, NULL);
2154 struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
2156 struct fib_table *tb;
2158 size_t sz = sizeof(*tb);
2161 sz += sizeof(struct trie);
2163 tb = kzalloc(sz, GFP_KERNEL);
2168 tb->tb_num_default = 0;
2169 tb->tb_data = (alias ? alias->__data : tb->__data);
2174 t = (struct trie *) tb->tb_data;
2175 t->kv[0].pos = KEYLENGTH;
2176 t->kv[0].slen = KEYLENGTH;
2177 #ifdef CONFIG_IP_FIB_TRIE_STATS
2178 t->stats = alloc_percpu(struct trie_use_stats);
2188 #ifdef CONFIG_PROC_FS
2189 /* Depth first Trie walk iterator */
2190 struct fib_trie_iter {
2191 struct seq_net_private p;
2192 struct fib_table *tb;
2193 struct key_vector *tnode;
2198 static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2200 unsigned long cindex = iter->index;
2201 struct key_vector *pn = iter->tnode;
2204 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2205 iter->tnode, iter->index, iter->depth);
2207 while (!IS_TRIE(pn)) {
2208 while (cindex < child_length(pn)) {
2209 struct key_vector *n = get_child_rcu(pn, cindex++);
2216 iter->index = cindex;
2218 /* push down one level */
2227 /* Current node exhausted, pop back up */
2229 pn = node_parent_rcu(pn);
2230 cindex = get_index(pkey, pn) + 1;
2234 /* record root node so further searches know we are done */
2241 static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2244 struct key_vector *n, *pn;
2250 n = rcu_dereference(pn->tnode[0]);
2267 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2269 struct key_vector *n;
2270 struct fib_trie_iter iter;
2272 memset(s, 0, sizeof(*s));
2275 for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2277 struct fib_alias *fa;
2280 s->totdepth += iter.depth;
2281 if (iter.depth > s->maxdepth)
2282 s->maxdepth = iter.depth;
2284 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2288 if (n->bits < MAX_STAT_DEPTH)
2289 s->nodesizes[n->bits]++;
2290 s->nullpointers += tn_info(n)->empty_children;
2297 * This outputs /proc/net/fib_triestats
2299 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2301 unsigned int i, max, pointers, bytes, avdepth;
2304 avdepth = stat->totdepth*100 / stat->leaves;
2308 seq_printf(seq, "\tAver depth: %u.%02d\n",
2309 avdepth / 100, avdepth % 100);
2310 seq_printf(seq, "\tMax depth: %u\n", stat->maxdepth);
2312 seq_printf(seq, "\tLeaves: %u\n", stat->leaves);
2313 bytes = LEAF_SIZE * stat->leaves;
2315 seq_printf(seq, "\tPrefixes: %u\n", stat->prefixes);
2316 bytes += sizeof(struct fib_alias) * stat->prefixes;
2318 seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2319 bytes += TNODE_SIZE(0) * stat->tnodes;
2321 max = MAX_STAT_DEPTH;
2322 while (max > 0 && stat->nodesizes[max-1] == 0)
2326 for (i = 1; i < max; i++)
2327 if (stat->nodesizes[i] != 0) {
2328 seq_printf(seq, " %u: %u", i, stat->nodesizes[i]);
2329 pointers += (1<<i) * stat->nodesizes[i];
2331 seq_putc(seq, '\n');
2332 seq_printf(seq, "\tPointers: %u\n", pointers);
2334 bytes += sizeof(struct key_vector *) * pointers;
2335 seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2336 seq_printf(seq, "Total size: %u kB\n", (bytes + 1023) / 1024);
2339 #ifdef CONFIG_IP_FIB_TRIE_STATS
2340 static void trie_show_usage(struct seq_file *seq,
2341 const struct trie_use_stats __percpu *stats)
2343 struct trie_use_stats s = { 0 };
2346 /* loop through all of the CPUs and gather up the stats */
2347 for_each_possible_cpu(cpu) {
2348 const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2350 s.gets += pcpu->gets;
2351 s.backtrack += pcpu->backtrack;
2352 s.semantic_match_passed += pcpu->semantic_match_passed;
2353 s.semantic_match_miss += pcpu->semantic_match_miss;
2354 s.null_node_hit += pcpu->null_node_hit;
2355 s.resize_node_skipped += pcpu->resize_node_skipped;
2358 seq_printf(seq, "\nCounters:\n---------\n");
2359 seq_printf(seq, "gets = %u\n", s.gets);
2360 seq_printf(seq, "backtracks = %u\n", s.backtrack);
2361 seq_printf(seq, "semantic match passed = %u\n",
2362 s.semantic_match_passed);
2363 seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2364 seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2365 seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2367 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2369 static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2371 if (tb->tb_id == RT_TABLE_LOCAL)
2372 seq_puts(seq, "Local:\n");
2373 else if (tb->tb_id == RT_TABLE_MAIN)
2374 seq_puts(seq, "Main:\n");
2376 seq_printf(seq, "Id %d:\n", tb->tb_id);
2380 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2382 struct net *net = (struct net *)seq->private;
2386 "Basic info: size of leaf:"
2387 " %Zd bytes, size of tnode: %Zd bytes.\n",
2388 LEAF_SIZE, TNODE_SIZE(0));
2390 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2391 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2392 struct fib_table *tb;
2394 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2395 struct trie *t = (struct trie *) tb->tb_data;
2396 struct trie_stat stat;
2401 fib_table_print(seq, tb);
2403 trie_collect_stats(t, &stat);
2404 trie_show_stats(seq, &stat);
2405 #ifdef CONFIG_IP_FIB_TRIE_STATS
2406 trie_show_usage(seq, t->stats);
2414 static int fib_triestat_seq_open(struct inode *inode, struct file *file)
2416 return single_open_net(inode, file, fib_triestat_seq_show);
2419 static const struct file_operations fib_triestat_fops = {
2420 .owner = THIS_MODULE,
2421 .open = fib_triestat_seq_open,
2423 .llseek = seq_lseek,
2424 .release = single_release_net,
2427 static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2429 struct fib_trie_iter *iter = seq->private;
2430 struct net *net = seq_file_net(seq);
2434 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2435 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2436 struct fib_table *tb;
2438 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2439 struct key_vector *n;
2441 for (n = fib_trie_get_first(iter,
2442 (struct trie *) tb->tb_data);
2443 n; n = fib_trie_get_next(iter))
2454 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2458 return fib_trie_get_idx(seq, *pos);
2461 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2463 struct fib_trie_iter *iter = seq->private;
2464 struct net *net = seq_file_net(seq);
2465 struct fib_table *tb = iter->tb;
2466 struct hlist_node *tb_node;
2468 struct key_vector *n;
2471 /* next node in same table */
2472 n = fib_trie_get_next(iter);
2476 /* walk rest of this hash chain */
2477 h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2478 while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2479 tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2480 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2485 /* new hash chain */
2486 while (++h < FIB_TABLE_HASHSZ) {
2487 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2488 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2489 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2501 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2507 static void seq_indent(struct seq_file *seq, int n)
2513 static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2516 case RT_SCOPE_UNIVERSE: return "universe";
2517 case RT_SCOPE_SITE: return "site";
2518 case RT_SCOPE_LINK: return "link";
2519 case RT_SCOPE_HOST: return "host";
2520 case RT_SCOPE_NOWHERE: return "nowhere";
2522 snprintf(buf, len, "scope=%d", s);
2527 static const char *const rtn_type_names[__RTN_MAX] = {
2528 [RTN_UNSPEC] = "UNSPEC",
2529 [RTN_UNICAST] = "UNICAST",
2530 [RTN_LOCAL] = "LOCAL",
2531 [RTN_BROADCAST] = "BROADCAST",
2532 [RTN_ANYCAST] = "ANYCAST",
2533 [RTN_MULTICAST] = "MULTICAST",
2534 [RTN_BLACKHOLE] = "BLACKHOLE",
2535 [RTN_UNREACHABLE] = "UNREACHABLE",
2536 [RTN_PROHIBIT] = "PROHIBIT",
2537 [RTN_THROW] = "THROW",
2539 [RTN_XRESOLVE] = "XRESOLVE",
2542 static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2544 if (t < __RTN_MAX && rtn_type_names[t])
2545 return rtn_type_names[t];
2546 snprintf(buf, len, "type %u", t);
2550 /* Pretty print the trie */
2551 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2553 const struct fib_trie_iter *iter = seq->private;
2554 struct key_vector *n = v;
2556 if (IS_TRIE(node_parent_rcu(n)))
2557 fib_table_print(seq, iter->tb);
2560 __be32 prf = htonl(n->key);
2562 seq_indent(seq, iter->depth-1);
2563 seq_printf(seq, " +-- %pI4/%zu %u %u %u\n",
2564 &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2565 tn_info(n)->full_children,
2566 tn_info(n)->empty_children);
2568 __be32 val = htonl(n->key);
2569 struct fib_alias *fa;
2571 seq_indent(seq, iter->depth);
2572 seq_printf(seq, " |-- %pI4\n", &val);
2574 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2575 char buf1[32], buf2[32];
2577 seq_indent(seq, iter->depth + 1);
2578 seq_printf(seq, " /%zu %s %s",
2579 KEYLENGTH - fa->fa_slen,
2580 rtn_scope(buf1, sizeof(buf1),
2581 fa->fa_info->fib_scope),
2582 rtn_type(buf2, sizeof(buf2),
2585 seq_printf(seq, " tos=%d", fa->fa_tos);
2586 seq_putc(seq, '\n');
2593 static const struct seq_operations fib_trie_seq_ops = {
2594 .start = fib_trie_seq_start,
2595 .next = fib_trie_seq_next,
2596 .stop = fib_trie_seq_stop,
2597 .show = fib_trie_seq_show,
2600 static int fib_trie_seq_open(struct inode *inode, struct file *file)
2602 return seq_open_net(inode, file, &fib_trie_seq_ops,
2603 sizeof(struct fib_trie_iter));
2606 static const struct file_operations fib_trie_fops = {
2607 .owner = THIS_MODULE,
2608 .open = fib_trie_seq_open,
2610 .llseek = seq_lseek,
2611 .release = seq_release_net,
2614 struct fib_route_iter {
2615 struct seq_net_private p;
2616 struct fib_table *main_tb;
2617 struct key_vector *tnode;
2622 static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2625 struct key_vector *l, **tp = &iter->tnode;
2628 /* use cached location of previously found key */
2629 if (iter->pos > 0 && pos >= iter->pos) {
2638 while ((l = leaf_walk_rcu(tp, key)) && (pos-- > 0)) {
2643 /* handle unlikely case of a key wrap */
2649 iter->key = l->key; /* remember it */
2651 iter->pos = 0; /* forget it */
2656 static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2659 struct fib_route_iter *iter = seq->private;
2660 struct fib_table *tb;
2665 tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2670 t = (struct trie *)tb->tb_data;
2671 iter->tnode = t->kv;
2674 return fib_route_get_idx(iter, *pos);
2677 iter->key = KEY_MAX;
2679 return SEQ_START_TOKEN;
2682 static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2684 struct fib_route_iter *iter = seq->private;
2685 struct key_vector *l = NULL;
2686 t_key key = iter->key + 1;
2690 /* only allow key of 0 for start of sequence */
2691 if ((v == SEQ_START_TOKEN) || key)
2692 l = leaf_walk_rcu(&iter->tnode, key);
2704 static void fib_route_seq_stop(struct seq_file *seq, void *v)
2710 static unsigned int fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2712 unsigned int flags = 0;
2714 if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2716 if (fi && fi->fib_nh->nh_gw)
2717 flags |= RTF_GATEWAY;
2718 if (mask == htonl(0xFFFFFFFF))
2725 * This outputs /proc/net/route.
2726 * The format of the file is not supposed to be changed
2727 * and needs to be same as fib_hash output to avoid breaking
2730 static int fib_route_seq_show(struct seq_file *seq, void *v)
2732 struct fib_route_iter *iter = seq->private;
2733 struct fib_table *tb = iter->main_tb;
2734 struct fib_alias *fa;
2735 struct key_vector *l = v;
2738 if (v == SEQ_START_TOKEN) {
2739 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2740 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2745 prefix = htonl(l->key);
2747 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2748 const struct fib_info *fi = fa->fa_info;
2749 __be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2750 unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2752 if ((fa->fa_type == RTN_BROADCAST) ||
2753 (fa->fa_type == RTN_MULTICAST))
2756 if (fa->tb_id != tb->tb_id)
2759 seq_setwidth(seq, 127);
2763 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2764 "%d\t%08X\t%d\t%u\t%u",
2765 fi->fib_dev ? fi->fib_dev->name : "*",
2767 fi->fib_nh->nh_gw, flags, 0, 0,
2771 fi->fib_advmss + 40 : 0),
2776 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2777 "%d\t%08X\t%d\t%u\t%u",
2778 prefix, 0, flags, 0, 0, 0,
2787 static const struct seq_operations fib_route_seq_ops = {
2788 .start = fib_route_seq_start,
2789 .next = fib_route_seq_next,
2790 .stop = fib_route_seq_stop,
2791 .show = fib_route_seq_show,
2794 static int fib_route_seq_open(struct inode *inode, struct file *file)
2796 return seq_open_net(inode, file, &fib_route_seq_ops,
2797 sizeof(struct fib_route_iter));
2800 static const struct file_operations fib_route_fops = {
2801 .owner = THIS_MODULE,
2802 .open = fib_route_seq_open,
2804 .llseek = seq_lseek,
2805 .release = seq_release_net,
2808 int __net_init fib_proc_init(struct net *net)
2810 if (!proc_create("fib_trie", S_IRUGO, net->proc_net, &fib_trie_fops))
2813 if (!proc_create("fib_triestat", S_IRUGO, net->proc_net,
2814 &fib_triestat_fops))
2817 if (!proc_create("route", S_IRUGO, net->proc_net, &fib_route_fops))
2823 remove_proc_entry("fib_triestat", net->proc_net);
2825 remove_proc_entry("fib_trie", net->proc_net);
2830 void __net_exit fib_proc_exit(struct net *net)
2832 remove_proc_entry("fib_trie", net->proc_net);
2833 remove_proc_entry("fib_triestat", net->proc_net);
2834 remove_proc_entry("route", net->proc_net);
2837 #endif /* CONFIG_PROC_FS */