2 * This program is free software; you can redistribute it and/or
3 * modify it under the terms of the GNU General Public License
4 * as published by the Free Software Foundation; either version
5 * 2 of the License, or (at your option) any later version.
7 * Robert Olsson <robert.olsson@its.uu.se> Uppsala Universitet
8 * & Swedish University of Agricultural Sciences.
10 * Jens Laas <jens.laas@data.slu.se> Swedish University of
11 * Agricultural Sciences.
13 * Hans Liss <hans.liss@its.uu.se> Uppsala Universitet
15 * This work is based on the LPC-trie which is originally described in:
17 * An experimental study of compression methods for dynamic tries
18 * Stefan Nilsson and Matti Tikkanen. Algorithmica, 33(1):19-33, 2002.
19 * http://www.csc.kth.se/~snilsson/software/dyntrie2/
22 * IP-address lookup using LC-tries. Stefan Nilsson and Gunnar Karlsson
23 * IEEE Journal on Selected Areas in Communications, 17(6):1083-1092, June 1999
26 * Code from fib_hash has been reused which includes the following header:
29 * INET An implementation of the TCP/IP protocol suite for the LINUX
30 * operating system. INET is implemented using the BSD Socket
31 * interface as the means of communication with the user level.
33 * IPv4 FIB: lookup engine and maintenance routines.
36 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
38 * This program is free software; you can redistribute it and/or
39 * modify it under the terms of the GNU General Public License
40 * as published by the Free Software Foundation; either version
41 * 2 of the License, or (at your option) any later version.
43 * Substantial contributions to this work comes from:
45 * David S. Miller, <davem@davemloft.net>
46 * Stephen Hemminger <shemminger@osdl.org>
47 * Paul E. McKenney <paulmck@us.ibm.com>
48 * Patrick McHardy <kaber@trash.net>
51 #define VERSION "0.409"
53 #include <linux/cache.h>
54 #include <linux/uaccess.h>
55 #include <linux/bitops.h>
56 #include <linux/types.h>
57 #include <linux/kernel.h>
59 #include <linux/string.h>
60 #include <linux/socket.h>
61 #include <linux/sockios.h>
62 #include <linux/errno.h>
64 #include <linux/inet.h>
65 #include <linux/inetdevice.h>
66 #include <linux/netdevice.h>
67 #include <linux/if_arp.h>
68 #include <linux/proc_fs.h>
69 #include <linux/rcupdate.h>
70 #include <linux/skbuff.h>
71 #include <linux/netlink.h>
72 #include <linux/init.h>
73 #include <linux/list.h>
74 #include <linux/slab.h>
75 #include <linux/export.h>
76 #include <linux/vmalloc.h>
77 #include <linux/notifier.h>
78 #include <net/net_namespace.h>
80 #include <net/protocol.h>
81 #include <net/route.h>
84 #include <net/ip_fib.h>
85 #include <net/fib_notifier.h>
86 #include <trace/events/fib.h>
87 #include "fib_lookup.h"
89 static int call_fib_entry_notifier(struct notifier_block *nb, struct net *net,
90 enum fib_event_type event_type, u32 dst,
91 int dst_len, struct fib_alias *fa)
93 struct fib_entry_notifier_info info = {
101 return call_fib4_notifier(nb, net, event_type, &info.info);
104 static int call_fib_entry_notifiers(struct net *net,
105 enum fib_event_type event_type, u32 dst,
106 int dst_len, struct fib_alias *fa,
107 struct netlink_ext_ack *extack)
109 struct fib_entry_notifier_info info = {
110 .info.extack = extack,
118 return call_fib4_notifiers(net, event_type, &info.info);
121 #define MAX_STAT_DEPTH 32
123 #define KEYLENGTH (8*sizeof(t_key))
124 #define KEY_MAX ((t_key)~0)
126 typedef unsigned int t_key;
128 #define IS_TRIE(n) ((n)->pos >= KEYLENGTH)
129 #define IS_TNODE(n) ((n)->bits)
130 #define IS_LEAF(n) (!(n)->bits)
134 unsigned char pos; /* 2log(KEYLENGTH) bits needed */
135 unsigned char bits; /* 2log(KEYLENGTH) bits needed */
138 /* This list pointer if valid if (pos | bits) == 0 (LEAF) */
139 struct hlist_head leaf;
140 /* This array is valid if (pos | bits) > 0 (TNODE) */
141 struct key_vector __rcu *tnode[0];
147 t_key empty_children; /* KEYLENGTH bits needed */
148 t_key full_children; /* KEYLENGTH bits needed */
149 struct key_vector __rcu *parent;
150 struct key_vector kv[1];
151 #define tn_bits kv[0].bits
154 #define TNODE_SIZE(n) offsetof(struct tnode, kv[0].tnode[n])
155 #define LEAF_SIZE TNODE_SIZE(1)
157 #ifdef CONFIG_IP_FIB_TRIE_STATS
158 struct trie_use_stats {
160 unsigned int backtrack;
161 unsigned int semantic_match_passed;
162 unsigned int semantic_match_miss;
163 unsigned int null_node_hit;
164 unsigned int resize_node_skipped;
169 unsigned int totdepth;
170 unsigned int maxdepth;
173 unsigned int nullpointers;
174 unsigned int prefixes;
175 unsigned int nodesizes[MAX_STAT_DEPTH];
179 struct key_vector kv[1];
180 #ifdef CONFIG_IP_FIB_TRIE_STATS
181 struct trie_use_stats __percpu *stats;
185 static struct key_vector *resize(struct trie *t, struct key_vector *tn);
186 static unsigned int tnode_free_size;
189 * synchronize_rcu after call_rcu for outstanding dirty memory; it should be
190 * especially useful before resizing the root node with PREEMPT_NONE configs;
191 * the value was obtained experimentally, aiming to avoid visible slowdown.
193 unsigned int sysctl_fib_sync_mem = 512 * 1024;
194 unsigned int sysctl_fib_sync_mem_min = 64 * 1024;
195 unsigned int sysctl_fib_sync_mem_max = 64 * 1024 * 1024;
197 static struct kmem_cache *fn_alias_kmem __ro_after_init;
198 static struct kmem_cache *trie_leaf_kmem __ro_after_init;
200 static inline struct tnode *tn_info(struct key_vector *kv)
202 return container_of(kv, struct tnode, kv[0]);
205 /* caller must hold RTNL */
206 #define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
207 #define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
209 /* caller must hold RCU read lock or RTNL */
210 #define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
211 #define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
213 /* wrapper for rcu_assign_pointer */
214 static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
217 rcu_assign_pointer(tn_info(n)->parent, tp);
220 #define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
222 /* This provides us with the number of children in this node, in the case of a
223 * leaf this will return 0 meaning none of the children are accessible.
225 static inline unsigned long child_length(const struct key_vector *tn)
227 return (1ul << tn->bits) & ~(1ul);
230 #define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
232 static inline unsigned long get_index(t_key key, struct key_vector *kv)
234 unsigned long index = key ^ kv->key;
236 if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
239 return index >> kv->pos;
242 /* To understand this stuff, an understanding of keys and all their bits is
243 * necessary. Every node in the trie has a key associated with it, but not
244 * all of the bits in that key are significant.
246 * Consider a node 'n' and its parent 'tp'.
248 * If n is a leaf, every bit in its key is significant. Its presence is
249 * necessitated by path compression, since during a tree traversal (when
250 * searching for a leaf - unless we are doing an insertion) we will completely
251 * ignore all skipped bits we encounter. Thus we need to verify, at the end of
252 * a potentially successful search, that we have indeed been walking the
255 * Note that we can never "miss" the correct key in the tree if present by
256 * following the wrong path. Path compression ensures that segments of the key
257 * that are the same for all keys with a given prefix are skipped, but the
258 * skipped part *is* identical for each node in the subtrie below the skipped
259 * bit! trie_insert() in this implementation takes care of that.
261 * if n is an internal node - a 'tnode' here, the various parts of its key
262 * have many different meanings.
265 * _________________________________________________________________
266 * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
267 * -----------------------------------------------------------------
268 * 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16
270 * _________________________________________________________________
271 * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
272 * -----------------------------------------------------------------
273 * 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
280 * First, let's just ignore the bits that come before the parent tp, that is
281 * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
282 * point we do not use them for anything.
284 * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
285 * index into the parent's child array. That is, they will be used to find
286 * 'n' among tp's children.
288 * The bits from (n->pos + n->bits) to (tp->pos - 1) - "S" - are skipped bits
291 * All the bits we have seen so far are significant to the node n. The rest
292 * of the bits are really not needed or indeed known in n->key.
294 * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
295 * n's child array, and will of course be different for each child.
297 * The rest of the bits, from 0 to (n->pos -1) - "u" - are completely unknown
301 static const int halve_threshold = 25;
302 static const int inflate_threshold = 50;
303 static const int halve_threshold_root = 15;
304 static const int inflate_threshold_root = 30;
306 static void __alias_free_mem(struct rcu_head *head)
308 struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
309 kmem_cache_free(fn_alias_kmem, fa);
312 static inline void alias_free_mem_rcu(struct fib_alias *fa)
314 call_rcu(&fa->rcu, __alias_free_mem);
317 #define TNODE_KMALLOC_MAX \
318 ilog2((PAGE_SIZE - TNODE_SIZE(0)) / sizeof(struct key_vector *))
319 #define TNODE_VMALLOC_MAX \
320 ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
322 static void __node_free_rcu(struct rcu_head *head)
324 struct tnode *n = container_of(head, struct tnode, rcu);
327 kmem_cache_free(trie_leaf_kmem, n);
332 #define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
334 static struct tnode *tnode_alloc(int bits)
338 /* verify bits is within bounds */
339 if (bits > TNODE_VMALLOC_MAX)
342 /* determine size and verify it is non-zero and didn't overflow */
343 size = TNODE_SIZE(1ul << bits);
345 if (size <= PAGE_SIZE)
346 return kzalloc(size, GFP_KERNEL);
348 return vzalloc(size);
351 static inline void empty_child_inc(struct key_vector *n)
353 ++tn_info(n)->empty_children ? : ++tn_info(n)->full_children;
356 static inline void empty_child_dec(struct key_vector *n)
358 tn_info(n)->empty_children-- ? : tn_info(n)->full_children--;
361 static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
363 struct key_vector *l;
366 kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
370 /* initialize key vector */
375 l->slen = fa->fa_slen;
377 /* link leaf to fib alias */
378 INIT_HLIST_HEAD(&l->leaf);
379 hlist_add_head(&fa->fa_list, &l->leaf);
384 static struct key_vector *tnode_new(t_key key, int pos, int bits)
386 unsigned int shift = pos + bits;
387 struct key_vector *tn;
390 /* verify bits and pos their msb bits clear and values are valid */
391 BUG_ON(!bits || (shift > KEYLENGTH));
393 tnode = tnode_alloc(bits);
397 pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
398 sizeof(struct key_vector *) << bits);
400 if (bits == KEYLENGTH)
401 tnode->full_children = 1;
403 tnode->empty_children = 1ul << bits;
406 tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
414 /* Check whether a tnode 'n' is "full", i.e. it is an internal node
415 * and no bits are skipped. See discussion in dyntree paper p. 6
417 static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
419 return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
422 /* Add a child at position i overwriting the old value.
423 * Update the value of full_children and empty_children.
425 static void put_child(struct key_vector *tn, unsigned long i,
426 struct key_vector *n)
428 struct key_vector *chi = get_child(tn, i);
431 BUG_ON(i >= child_length(tn));
433 /* update emptyChildren, overflow into fullChildren */
439 /* update fullChildren */
440 wasfull = tnode_full(tn, chi);
441 isfull = tnode_full(tn, n);
443 if (wasfull && !isfull)
444 tn_info(tn)->full_children--;
445 else if (!wasfull && isfull)
446 tn_info(tn)->full_children++;
448 if (n && (tn->slen < n->slen))
451 rcu_assign_pointer(tn->tnode[i], n);
454 static void update_children(struct key_vector *tn)
458 /* update all of the child parent pointers */
459 for (i = child_length(tn); i;) {
460 struct key_vector *inode = get_child(tn, --i);
465 /* Either update the children of a tnode that
466 * already belongs to us or update the child
467 * to point to ourselves.
469 if (node_parent(inode) == tn)
470 update_children(inode);
472 node_set_parent(inode, tn);
476 static inline void put_child_root(struct key_vector *tp, t_key key,
477 struct key_vector *n)
480 rcu_assign_pointer(tp->tnode[0], n);
482 put_child(tp, get_index(key, tp), n);
485 static inline void tnode_free_init(struct key_vector *tn)
487 tn_info(tn)->rcu.next = NULL;
490 static inline void tnode_free_append(struct key_vector *tn,
491 struct key_vector *n)
493 tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
494 tn_info(tn)->rcu.next = &tn_info(n)->rcu;
497 static void tnode_free(struct key_vector *tn)
499 struct callback_head *head = &tn_info(tn)->rcu;
503 tnode_free_size += TNODE_SIZE(1ul << tn->bits);
506 tn = container_of(head, struct tnode, rcu)->kv;
509 if (tnode_free_size >= sysctl_fib_sync_mem) {
515 static struct key_vector *replace(struct trie *t,
516 struct key_vector *oldtnode,
517 struct key_vector *tn)
519 struct key_vector *tp = node_parent(oldtnode);
522 /* setup the parent pointer out of and back into this node */
523 NODE_INIT_PARENT(tn, tp);
524 put_child_root(tp, tn->key, tn);
526 /* update all of the child parent pointers */
529 /* all pointers should be clean so we are done */
530 tnode_free(oldtnode);
532 /* resize children now that oldtnode is freed */
533 for (i = child_length(tn); i;) {
534 struct key_vector *inode = get_child(tn, --i);
536 /* resize child node */
537 if (tnode_full(tn, inode))
538 tn = resize(t, inode);
544 static struct key_vector *inflate(struct trie *t,
545 struct key_vector *oldtnode)
547 struct key_vector *tn;
551 pr_debug("In inflate\n");
553 tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
557 /* prepare oldtnode to be freed */
558 tnode_free_init(oldtnode);
560 /* Assemble all of the pointers in our cluster, in this case that
561 * represents all of the pointers out of our allocated nodes that
562 * point to existing tnodes and the links between our allocated
565 for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
566 struct key_vector *inode = get_child(oldtnode, --i);
567 struct key_vector *node0, *node1;
574 /* A leaf or an internal node with skipped bits */
575 if (!tnode_full(oldtnode, inode)) {
576 put_child(tn, get_index(inode->key, tn), inode);
580 /* drop the node in the old tnode free list */
581 tnode_free_append(oldtnode, inode);
583 /* An internal node with two children */
584 if (inode->bits == 1) {
585 put_child(tn, 2 * i + 1, get_child(inode, 1));
586 put_child(tn, 2 * i, get_child(inode, 0));
590 /* We will replace this node 'inode' with two new
591 * ones, 'node0' and 'node1', each with half of the
592 * original children. The two new nodes will have
593 * a position one bit further down the key and this
594 * means that the "significant" part of their keys
595 * (see the discussion near the top of this file)
596 * will differ by one bit, which will be "0" in
597 * node0's key and "1" in node1's key. Since we are
598 * moving the key position by one step, the bit that
599 * we are moving away from - the bit at position
600 * (tn->pos) - is the one that will differ between
601 * node0 and node1. So... we synthesize that bit in the
604 node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
607 node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
609 tnode_free_append(tn, node1);
612 tnode_free_append(tn, node0);
614 /* populate child pointers in new nodes */
615 for (k = child_length(inode), j = k / 2; j;) {
616 put_child(node1, --j, get_child(inode, --k));
617 put_child(node0, j, get_child(inode, j));
618 put_child(node1, --j, get_child(inode, --k));
619 put_child(node0, j, get_child(inode, j));
622 /* link new nodes to parent */
623 NODE_INIT_PARENT(node1, tn);
624 NODE_INIT_PARENT(node0, tn);
626 /* link parent to nodes */
627 put_child(tn, 2 * i + 1, node1);
628 put_child(tn, 2 * i, node0);
631 /* setup the parent pointers into and out of this node */
632 return replace(t, oldtnode, tn);
634 /* all pointers should be clean so we are done */
640 static struct key_vector *halve(struct trie *t,
641 struct key_vector *oldtnode)
643 struct key_vector *tn;
646 pr_debug("In halve\n");
648 tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
652 /* prepare oldtnode to be freed */
653 tnode_free_init(oldtnode);
655 /* Assemble all of the pointers in our cluster, in this case that
656 * represents all of the pointers out of our allocated nodes that
657 * point to existing tnodes and the links between our allocated
660 for (i = child_length(oldtnode); i;) {
661 struct key_vector *node1 = get_child(oldtnode, --i);
662 struct key_vector *node0 = get_child(oldtnode, --i);
663 struct key_vector *inode;
665 /* At least one of the children is empty */
666 if (!node1 || !node0) {
667 put_child(tn, i / 2, node1 ? : node0);
671 /* Two nonempty children */
672 inode = tnode_new(node0->key, oldtnode->pos, 1);
675 tnode_free_append(tn, inode);
677 /* initialize pointers out of node */
678 put_child(inode, 1, node1);
679 put_child(inode, 0, node0);
680 NODE_INIT_PARENT(inode, tn);
682 /* link parent to node */
683 put_child(tn, i / 2, inode);
686 /* setup the parent pointers into and out of this node */
687 return replace(t, oldtnode, tn);
689 /* all pointers should be clean so we are done */
695 static struct key_vector *collapse(struct trie *t,
696 struct key_vector *oldtnode)
698 struct key_vector *n, *tp;
701 /* scan the tnode looking for that one child that might still exist */
702 for (n = NULL, i = child_length(oldtnode); !n && i;)
703 n = get_child(oldtnode, --i);
705 /* compress one level */
706 tp = node_parent(oldtnode);
707 put_child_root(tp, oldtnode->key, n);
708 node_set_parent(n, tp);
716 static unsigned char update_suffix(struct key_vector *tn)
718 unsigned char slen = tn->pos;
719 unsigned long stride, i;
720 unsigned char slen_max;
722 /* only vector 0 can have a suffix length greater than or equal to
723 * tn->pos + tn->bits, the second highest node will have a suffix
724 * length at most of tn->pos + tn->bits - 1
726 slen_max = min_t(unsigned char, tn->pos + tn->bits - 1, tn->slen);
728 /* search though the list of children looking for nodes that might
729 * have a suffix greater than the one we currently have. This is
730 * why we start with a stride of 2 since a stride of 1 would
731 * represent the nodes with suffix length equal to tn->pos
733 for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
734 struct key_vector *n = get_child(tn, i);
736 if (!n || (n->slen <= slen))
739 /* update stride and slen based on new value */
740 stride <<= (n->slen - slen);
744 /* stop searching if we have hit the maximum possible value */
745 if (slen >= slen_max)
754 /* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
755 * the Helsinki University of Technology and Matti Tikkanen of Nokia
756 * Telecommunications, page 6:
757 * "A node is doubled if the ratio of non-empty children to all
758 * children in the *doubled* node is at least 'high'."
760 * 'high' in this instance is the variable 'inflate_threshold'. It
761 * is expressed as a percentage, so we multiply it with
762 * child_length() and instead of multiplying by 2 (since the
763 * child array will be doubled by inflate()) and multiplying
764 * the left-hand side by 100 (to handle the percentage thing) we
765 * multiply the left-hand side by 50.
767 * The left-hand side may look a bit weird: child_length(tn)
768 * - tn->empty_children is of course the number of non-null children
769 * in the current node. tn->full_children is the number of "full"
770 * children, that is non-null tnodes with a skip value of 0.
771 * All of those will be doubled in the resulting inflated tnode, so
772 * we just count them one extra time here.
774 * A clearer way to write this would be:
776 * to_be_doubled = tn->full_children;
777 * not_to_be_doubled = child_length(tn) - tn->empty_children -
780 * new_child_length = child_length(tn) * 2;
782 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
784 * if (new_fill_factor >= inflate_threshold)
786 * ...and so on, tho it would mess up the while () loop.
789 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
793 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
794 * inflate_threshold * new_child_length
796 * expand not_to_be_doubled and to_be_doubled, and shorten:
797 * 100 * (child_length(tn) - tn->empty_children +
798 * tn->full_children) >= inflate_threshold * new_child_length
800 * expand new_child_length:
801 * 100 * (child_length(tn) - tn->empty_children +
802 * tn->full_children) >=
803 * inflate_threshold * child_length(tn) * 2
806 * 50 * (tn->full_children + child_length(tn) -
807 * tn->empty_children) >= inflate_threshold *
811 static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
813 unsigned long used = child_length(tn);
814 unsigned long threshold = used;
816 /* Keep root node larger */
817 threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
818 used -= tn_info(tn)->empty_children;
819 used += tn_info(tn)->full_children;
821 /* if bits == KEYLENGTH then pos = 0, and will fail below */
823 return (used > 1) && tn->pos && ((50 * used) >= threshold);
826 static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
828 unsigned long used = child_length(tn);
829 unsigned long threshold = used;
831 /* Keep root node larger */
832 threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
833 used -= tn_info(tn)->empty_children;
835 /* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
837 return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
840 static inline bool should_collapse(struct key_vector *tn)
842 unsigned long used = child_length(tn);
844 used -= tn_info(tn)->empty_children;
846 /* account for bits == KEYLENGTH case */
847 if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
850 /* One child or none, time to drop us from the trie */
855 static struct key_vector *resize(struct trie *t, struct key_vector *tn)
857 #ifdef CONFIG_IP_FIB_TRIE_STATS
858 struct trie_use_stats __percpu *stats = t->stats;
860 struct key_vector *tp = node_parent(tn);
861 unsigned long cindex = get_index(tn->key, tp);
862 int max_work = MAX_WORK;
864 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
865 tn, inflate_threshold, halve_threshold);
867 /* track the tnode via the pointer from the parent instead of
868 * doing it ourselves. This way we can let RCU fully do its
869 * thing without us interfering
871 BUG_ON(tn != get_child(tp, cindex));
873 /* Double as long as the resulting node has a number of
874 * nonempty nodes that are above the threshold.
876 while (should_inflate(tp, tn) && max_work) {
879 #ifdef CONFIG_IP_FIB_TRIE_STATS
880 this_cpu_inc(stats->resize_node_skipped);
886 tn = get_child(tp, cindex);
889 /* update parent in case inflate failed */
890 tp = node_parent(tn);
892 /* Return if at least one inflate is run */
893 if (max_work != MAX_WORK)
896 /* Halve as long as the number of empty children in this
897 * node is above threshold.
899 while (should_halve(tp, tn) && max_work) {
902 #ifdef CONFIG_IP_FIB_TRIE_STATS
903 this_cpu_inc(stats->resize_node_skipped);
909 tn = get_child(tp, cindex);
912 /* Only one child remains */
913 if (should_collapse(tn))
914 return collapse(t, tn);
916 /* update parent in case halve failed */
917 return node_parent(tn);
920 static void node_pull_suffix(struct key_vector *tn, unsigned char slen)
922 unsigned char node_slen = tn->slen;
924 while ((node_slen > tn->pos) && (node_slen > slen)) {
925 slen = update_suffix(tn);
926 if (node_slen == slen)
929 tn = node_parent(tn);
930 node_slen = tn->slen;
934 static void node_push_suffix(struct key_vector *tn, unsigned char slen)
936 while (tn->slen < slen) {
938 tn = node_parent(tn);
942 /* rcu_read_lock needs to be hold by caller from readside */
943 static struct key_vector *fib_find_node(struct trie *t,
944 struct key_vector **tp, u32 key)
946 struct key_vector *pn, *n = t->kv;
947 unsigned long index = 0;
951 n = get_child_rcu(n, index);
956 index = get_cindex(key, n);
958 /* This bit of code is a bit tricky but it combines multiple
959 * checks into a single check. The prefix consists of the
960 * prefix plus zeros for the bits in the cindex. The index
961 * is the difference between the key and this value. From
962 * this we can actually derive several pieces of data.
963 * if (index >= (1ul << bits))
964 * we have a mismatch in skip bits and failed
966 * we know the value is cindex
968 * This check is safe even if bits == KEYLENGTH due to the
969 * fact that we can only allocate a node with 32 bits if a
970 * long is greater than 32 bits.
972 if (index >= (1ul << n->bits)) {
977 /* keep searching until we find a perfect match leaf or NULL */
978 } while (IS_TNODE(n));
985 /* Return the first fib alias matching TOS with
986 * priority less than or equal to PRIO.
988 static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
989 u8 tos, u32 prio, u32 tb_id)
991 struct fib_alias *fa;
996 hlist_for_each_entry(fa, fah, fa_list) {
997 if (fa->fa_slen < slen)
999 if (fa->fa_slen != slen)
1001 if (fa->tb_id > tb_id)
1003 if (fa->tb_id != tb_id)
1005 if (fa->fa_tos > tos)
1007 if (fa->fa_info->fib_priority >= prio || fa->fa_tos < tos)
1014 static void trie_rebalance(struct trie *t, struct key_vector *tn)
1016 while (!IS_TRIE(tn))
1020 static int fib_insert_node(struct trie *t, struct key_vector *tp,
1021 struct fib_alias *new, t_key key)
1023 struct key_vector *n, *l;
1025 l = leaf_new(key, new);
1029 /* retrieve child from parent node */
1030 n = get_child(tp, get_index(key, tp));
1032 /* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1034 * Add a new tnode here
1035 * first tnode need some special handling
1036 * leaves us in position for handling as case 3
1039 struct key_vector *tn;
1041 tn = tnode_new(key, __fls(key ^ n->key), 1);
1045 /* initialize routes out of node */
1046 NODE_INIT_PARENT(tn, tp);
1047 put_child(tn, get_index(key, tn) ^ 1, n);
1049 /* start adding routes into the node */
1050 put_child_root(tp, key, tn);
1051 node_set_parent(n, tn);
1053 /* parent now has a NULL spot where the leaf can go */
1057 /* Case 3: n is NULL, and will just insert a new leaf */
1058 node_push_suffix(tp, new->fa_slen);
1059 NODE_INIT_PARENT(l, tp);
1060 put_child_root(tp, key, l);
1061 trie_rebalance(t, tp);
1070 /* fib notifier for ADD is sent before calling fib_insert_alias with
1071 * the expectation that the only possible failure ENOMEM
1073 static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1074 struct key_vector *l, struct fib_alias *new,
1075 struct fib_alias *fa, t_key key)
1078 return fib_insert_node(t, tp, new, key);
1081 hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1083 struct fib_alias *last;
1085 hlist_for_each_entry(last, &l->leaf, fa_list) {
1086 if (new->fa_slen < last->fa_slen)
1088 if ((new->fa_slen == last->fa_slen) &&
1089 (new->tb_id > last->tb_id))
1095 hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1097 hlist_add_head_rcu(&new->fa_list, &l->leaf);
1100 /* if we added to the tail node then we need to update slen */
1101 if (l->slen < new->fa_slen) {
1102 l->slen = new->fa_slen;
1103 node_push_suffix(tp, new->fa_slen);
1109 static bool fib_valid_key_len(u32 key, u8 plen, struct netlink_ext_ack *extack)
1111 if (plen > KEYLENGTH) {
1112 NL_SET_ERR_MSG(extack, "Invalid prefix length");
1116 if ((plen < KEYLENGTH) && (key << plen)) {
1117 NL_SET_ERR_MSG(extack,
1118 "Invalid prefix for given prefix length");
1125 /* Caller must hold RTNL. */
1126 int fib_table_insert(struct net *net, struct fib_table *tb,
1127 struct fib_config *cfg, struct netlink_ext_ack *extack)
1129 enum fib_event_type event = FIB_EVENT_ENTRY_ADD;
1130 struct trie *t = (struct trie *)tb->tb_data;
1131 struct fib_alias *fa, *new_fa;
1132 struct key_vector *l, *tp;
1133 u16 nlflags = NLM_F_EXCL;
1134 struct fib_info *fi;
1135 u8 plen = cfg->fc_dst_len;
1136 u8 slen = KEYLENGTH - plen;
1137 u8 tos = cfg->fc_tos;
1141 key = ntohl(cfg->fc_dst);
1143 if (!fib_valid_key_len(key, plen, extack))
1146 pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1148 fi = fib_create_info(cfg, extack);
1154 l = fib_find_node(t, &tp, key);
1155 fa = l ? fib_find_alias(&l->leaf, slen, tos, fi->fib_priority,
1158 /* Now fa, if non-NULL, points to the first fib alias
1159 * with the same keys [prefix,tos,priority], if such key already
1160 * exists or to the node before which we will insert new one.
1162 * If fa is NULL, we will need to allocate a new one and
1163 * insert to the tail of the section matching the suffix length
1167 if (fa && fa->fa_tos == tos &&
1168 fa->fa_info->fib_priority == fi->fib_priority) {
1169 struct fib_alias *fa_first, *fa_match;
1172 if (cfg->fc_nlflags & NLM_F_EXCL)
1175 nlflags &= ~NLM_F_EXCL;
1178 * 1. Find exact match for type, scope, fib_info to avoid
1180 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1184 hlist_for_each_entry_from(fa, fa_list) {
1185 if ((fa->fa_slen != slen) ||
1186 (fa->tb_id != tb->tb_id) ||
1187 (fa->fa_tos != tos))
1189 if (fa->fa_info->fib_priority != fi->fib_priority)
1191 if (fa->fa_type == cfg->fc_type &&
1192 fa->fa_info == fi) {
1198 if (cfg->fc_nlflags & NLM_F_REPLACE) {
1199 struct fib_info *fi_drop;
1202 nlflags |= NLM_F_REPLACE;
1210 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1214 fi_drop = fa->fa_info;
1215 new_fa->fa_tos = fa->fa_tos;
1216 new_fa->fa_info = fi;
1217 new_fa->fa_type = cfg->fc_type;
1218 state = fa->fa_state;
1219 new_fa->fa_state = state & ~FA_S_ACCESSED;
1220 new_fa->fa_slen = fa->fa_slen;
1221 new_fa->tb_id = tb->tb_id;
1222 new_fa->fa_default = -1;
1224 err = call_fib_entry_notifiers(net,
1225 FIB_EVENT_ENTRY_REPLACE,
1229 goto out_free_new_fa;
1231 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1232 tb->tb_id, &cfg->fc_nlinfo, nlflags);
1234 hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1236 alias_free_mem_rcu(fa);
1238 fib_release_info(fi_drop);
1239 if (state & FA_S_ACCESSED)
1240 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1244 /* Error if we find a perfect match which
1245 * uses the same scope, type, and nexthop
1251 if (cfg->fc_nlflags & NLM_F_APPEND) {
1252 event = FIB_EVENT_ENTRY_APPEND;
1253 nlflags |= NLM_F_APPEND;
1259 if (!(cfg->fc_nlflags & NLM_F_CREATE))
1262 nlflags |= NLM_F_CREATE;
1264 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1268 new_fa->fa_info = fi;
1269 new_fa->fa_tos = tos;
1270 new_fa->fa_type = cfg->fc_type;
1271 new_fa->fa_state = 0;
1272 new_fa->fa_slen = slen;
1273 new_fa->tb_id = tb->tb_id;
1274 new_fa->fa_default = -1;
1276 err = call_fib_entry_notifiers(net, event, key, plen, new_fa, extack);
1278 goto out_free_new_fa;
1280 /* Insert new entry to the list. */
1281 err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1286 tb->tb_num_default++;
1288 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1289 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1290 &cfg->fc_nlinfo, nlflags);
1295 /* notifier was sent that entry would be added to trie, but
1296 * the add failed and need to recover. Only failure for
1297 * fib_insert_alias is ENOMEM.
1299 NL_SET_ERR_MSG(extack, "Failed to insert route into trie");
1300 call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, key,
1301 plen, new_fa, NULL);
1303 kmem_cache_free(fn_alias_kmem, new_fa);
1305 fib_release_info(fi);
1310 static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1312 t_key prefix = n->key;
1314 return (key ^ prefix) & (prefix | -prefix);
1317 /* should be called with rcu_read_lock */
1318 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1319 struct fib_result *res, int fib_flags)
1321 struct trie *t = (struct trie *) tb->tb_data;
1322 #ifdef CONFIG_IP_FIB_TRIE_STATS
1323 struct trie_use_stats __percpu *stats = t->stats;
1325 const t_key key = ntohl(flp->daddr);
1326 struct key_vector *n, *pn;
1327 struct fib_alias *fa;
1328 unsigned long index;
1334 n = get_child_rcu(pn, cindex);
1336 trace_fib_table_lookup(tb->tb_id, flp, NULL, -EAGAIN);
1340 #ifdef CONFIG_IP_FIB_TRIE_STATS
1341 this_cpu_inc(stats->gets);
1344 /* Step 1: Travel to the longest prefix match in the trie */
1346 index = get_cindex(key, n);
1348 /* This bit of code is a bit tricky but it combines multiple
1349 * checks into a single check. The prefix consists of the
1350 * prefix plus zeros for the "bits" in the prefix. The index
1351 * is the difference between the key and this value. From
1352 * this we can actually derive several pieces of data.
1353 * if (index >= (1ul << bits))
1354 * we have a mismatch in skip bits and failed
1356 * we know the value is cindex
1358 * This check is safe even if bits == KEYLENGTH due to the
1359 * fact that we can only allocate a node with 32 bits if a
1360 * long is greater than 32 bits.
1362 if (index >= (1ul << n->bits))
1365 /* we have found a leaf. Prefixes have already been compared */
1369 /* only record pn and cindex if we are going to be chopping
1370 * bits later. Otherwise we are just wasting cycles.
1372 if (n->slen > n->pos) {
1377 n = get_child_rcu(n, index);
1382 /* Step 2: Sort out leaves and begin backtracing for longest prefix */
1384 /* record the pointer where our next node pointer is stored */
1385 struct key_vector __rcu **cptr = n->tnode;
1387 /* This test verifies that none of the bits that differ
1388 * between the key and the prefix exist in the region of
1389 * the lsb and higher in the prefix.
1391 if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1394 /* exit out and process leaf */
1395 if (unlikely(IS_LEAF(n)))
1398 /* Don't bother recording parent info. Since we are in
1399 * prefix match mode we will have to come back to wherever
1400 * we started this traversal anyway
1403 while ((n = rcu_dereference(*cptr)) == NULL) {
1405 #ifdef CONFIG_IP_FIB_TRIE_STATS
1407 this_cpu_inc(stats->null_node_hit);
1409 /* If we are at cindex 0 there are no more bits for
1410 * us to strip at this level so we must ascend back
1411 * up one level to see if there are any more bits to
1412 * be stripped there.
1415 t_key pkey = pn->key;
1417 /* If we don't have a parent then there is
1418 * nothing for us to do as we do not have any
1419 * further nodes to parse.
1422 trace_fib_table_lookup(tb->tb_id, flp,
1426 #ifdef CONFIG_IP_FIB_TRIE_STATS
1427 this_cpu_inc(stats->backtrack);
1429 /* Get Child's index */
1430 pn = node_parent_rcu(pn);
1431 cindex = get_index(pkey, pn);
1434 /* strip the least significant bit from the cindex */
1435 cindex &= cindex - 1;
1437 /* grab pointer for next child node */
1438 cptr = &pn->tnode[cindex];
1443 /* this line carries forward the xor from earlier in the function */
1444 index = key ^ n->key;
1446 /* Step 3: Process the leaf, if that fails fall back to backtracing */
1447 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1448 struct fib_info *fi = fa->fa_info;
1451 if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1452 if (index >= (1ul << fa->fa_slen))
1455 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1459 if (fa->fa_info->fib_scope < flp->flowi4_scope)
1461 fib_alias_accessed(fa);
1462 err = fib_props[fa->fa_type].error;
1463 if (unlikely(err < 0)) {
1464 #ifdef CONFIG_IP_FIB_TRIE_STATS
1465 this_cpu_inc(stats->semantic_match_passed);
1467 trace_fib_table_lookup(tb->tb_id, flp, NULL, err);
1470 if (fi->fib_flags & RTNH_F_DEAD)
1472 for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
1473 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
1475 if (nhc->nhc_flags & RTNH_F_DEAD)
1477 if (ip_ignore_linkdown(nhc->nhc_dev) &&
1478 nhc->nhc_flags & RTNH_F_LINKDOWN &&
1479 !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1481 if (!(flp->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF)) {
1482 if (flp->flowi4_oif &&
1483 flp->flowi4_oif != nhc->nhc_oif)
1487 if (!(fib_flags & FIB_LOOKUP_NOREF))
1488 refcount_inc(&fi->fib_clntref);
1490 res->prefix = htonl(n->key);
1491 res->prefixlen = KEYLENGTH - fa->fa_slen;
1492 res->nh_sel = nhsel;
1494 res->type = fa->fa_type;
1495 res->scope = fi->fib_scope;
1498 res->fa_head = &n->leaf;
1499 #ifdef CONFIG_IP_FIB_TRIE_STATS
1500 this_cpu_inc(stats->semantic_match_passed);
1502 trace_fib_table_lookup(tb->tb_id, flp, nhc, err);
1507 #ifdef CONFIG_IP_FIB_TRIE_STATS
1508 this_cpu_inc(stats->semantic_match_miss);
1512 EXPORT_SYMBOL_GPL(fib_table_lookup);
1514 static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1515 struct key_vector *l, struct fib_alias *old)
1517 /* record the location of the previous list_info entry */
1518 struct hlist_node **pprev = old->fa_list.pprev;
1519 struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1521 /* remove the fib_alias from the list */
1522 hlist_del_rcu(&old->fa_list);
1524 /* if we emptied the list this leaf will be freed and we can sort
1525 * out parent suffix lengths as a part of trie_rebalance
1527 if (hlist_empty(&l->leaf)) {
1528 if (tp->slen == l->slen)
1529 node_pull_suffix(tp, tp->pos);
1530 put_child_root(tp, l->key, NULL);
1532 trie_rebalance(t, tp);
1536 /* only access fa if it is pointing at the last valid hlist_node */
1540 /* update the trie with the latest suffix length */
1541 l->slen = fa->fa_slen;
1542 node_pull_suffix(tp, fa->fa_slen);
1545 /* Caller must hold RTNL. */
1546 int fib_table_delete(struct net *net, struct fib_table *tb,
1547 struct fib_config *cfg, struct netlink_ext_ack *extack)
1549 struct trie *t = (struct trie *) tb->tb_data;
1550 struct fib_alias *fa, *fa_to_delete;
1551 struct key_vector *l, *tp;
1552 u8 plen = cfg->fc_dst_len;
1553 u8 slen = KEYLENGTH - plen;
1554 u8 tos = cfg->fc_tos;
1557 key = ntohl(cfg->fc_dst);
1559 if (!fib_valid_key_len(key, plen, extack))
1562 l = fib_find_node(t, &tp, key);
1566 fa = fib_find_alias(&l->leaf, slen, tos, 0, tb->tb_id);
1570 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1572 fa_to_delete = NULL;
1573 hlist_for_each_entry_from(fa, fa_list) {
1574 struct fib_info *fi = fa->fa_info;
1576 if ((fa->fa_slen != slen) ||
1577 (fa->tb_id != tb->tb_id) ||
1578 (fa->fa_tos != tos))
1581 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1582 (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1583 fa->fa_info->fib_scope == cfg->fc_scope) &&
1584 (!cfg->fc_prefsrc ||
1585 fi->fib_prefsrc == cfg->fc_prefsrc) &&
1586 (!cfg->fc_protocol ||
1587 fi->fib_protocol == cfg->fc_protocol) &&
1588 fib_nh_match(cfg, fi, extack) == 0 &&
1589 fib_metrics_match(cfg, fi)) {
1598 call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, key, plen,
1599 fa_to_delete, extack);
1600 rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1601 &cfg->fc_nlinfo, 0);
1604 tb->tb_num_default--;
1606 fib_remove_alias(t, tp, l, fa_to_delete);
1608 if (fa_to_delete->fa_state & FA_S_ACCESSED)
1609 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1611 fib_release_info(fa_to_delete->fa_info);
1612 alias_free_mem_rcu(fa_to_delete);
1616 /* Scan for the next leaf starting at the provided key value */
1617 static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1619 struct key_vector *pn, *n = *tn;
1620 unsigned long cindex;
1622 /* this loop is meant to try and find the key in the trie */
1624 /* record parent and next child index */
1626 cindex = (key > pn->key) ? get_index(key, pn) : 0;
1628 if (cindex >> pn->bits)
1631 /* descend into the next child */
1632 n = get_child_rcu(pn, cindex++);
1636 /* guarantee forward progress on the keys */
1637 if (IS_LEAF(n) && (n->key >= key))
1639 } while (IS_TNODE(n));
1641 /* this loop will search for the next leaf with a greater key */
1642 while (!IS_TRIE(pn)) {
1643 /* if we exhausted the parent node we will need to climb */
1644 if (cindex >= (1ul << pn->bits)) {
1645 t_key pkey = pn->key;
1647 pn = node_parent_rcu(pn);
1648 cindex = get_index(pkey, pn) + 1;
1652 /* grab the next available node */
1653 n = get_child_rcu(pn, cindex++);
1657 /* no need to compare keys since we bumped the index */
1661 /* Rescan start scanning in new node */
1667 return NULL; /* Root of trie */
1669 /* if we are at the limit for keys just return NULL for the tnode */
1674 static void fib_trie_free(struct fib_table *tb)
1676 struct trie *t = (struct trie *)tb->tb_data;
1677 struct key_vector *pn = t->kv;
1678 unsigned long cindex = 1;
1679 struct hlist_node *tmp;
1680 struct fib_alias *fa;
1682 /* walk trie in reverse order and free everything */
1684 struct key_vector *n;
1687 t_key pkey = pn->key;
1693 pn = node_parent(pn);
1695 /* drop emptied tnode */
1696 put_child_root(pn, n->key, NULL);
1699 cindex = get_index(pkey, pn);
1704 /* grab the next available node */
1705 n = get_child(pn, cindex);
1710 /* record pn and cindex for leaf walking */
1712 cindex = 1ul << n->bits;
1717 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1718 hlist_del_rcu(&fa->fa_list);
1719 alias_free_mem_rcu(fa);
1722 put_child_root(pn, n->key, NULL);
1726 #ifdef CONFIG_IP_FIB_TRIE_STATS
1727 free_percpu(t->stats);
1732 struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1734 struct trie *ot = (struct trie *)oldtb->tb_data;
1735 struct key_vector *l, *tp = ot->kv;
1736 struct fib_table *local_tb;
1737 struct fib_alias *fa;
1741 if (oldtb->tb_data == oldtb->__data)
1744 local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1748 lt = (struct trie *)local_tb->tb_data;
1750 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1751 struct key_vector *local_l = NULL, *local_tp;
1753 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1754 struct fib_alias *new_fa;
1756 if (local_tb->tb_id != fa->tb_id)
1759 /* clone fa for new local table */
1760 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1764 memcpy(new_fa, fa, sizeof(*fa));
1766 /* insert clone into table */
1768 local_l = fib_find_node(lt, &local_tp, l->key);
1770 if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1772 kmem_cache_free(fn_alias_kmem, new_fa);
1777 /* stop loop if key wrapped back to 0 */
1785 fib_trie_free(local_tb);
1790 /* Caller must hold RTNL */
1791 void fib_table_flush_external(struct fib_table *tb)
1793 struct trie *t = (struct trie *)tb->tb_data;
1794 struct key_vector *pn = t->kv;
1795 unsigned long cindex = 1;
1796 struct hlist_node *tmp;
1797 struct fib_alias *fa;
1799 /* walk trie in reverse order */
1801 unsigned char slen = 0;
1802 struct key_vector *n;
1805 t_key pkey = pn->key;
1807 /* cannot resize the trie vector */
1811 /* update the suffix to address pulled leaves */
1812 if (pn->slen > pn->pos)
1815 /* resize completed node */
1817 cindex = get_index(pkey, pn);
1822 /* grab the next available node */
1823 n = get_child(pn, cindex);
1828 /* record pn and cindex for leaf walking */
1830 cindex = 1ul << n->bits;
1835 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1836 /* if alias was cloned to local then we just
1837 * need to remove the local copy from main
1839 if (tb->tb_id != fa->tb_id) {
1840 hlist_del_rcu(&fa->fa_list);
1841 alias_free_mem_rcu(fa);
1845 /* record local slen */
1849 /* update leaf slen */
1852 if (hlist_empty(&n->leaf)) {
1853 put_child_root(pn, n->key, NULL);
1859 /* Caller must hold RTNL. */
1860 int fib_table_flush(struct net *net, struct fib_table *tb, bool flush_all)
1862 struct trie *t = (struct trie *)tb->tb_data;
1863 struct key_vector *pn = t->kv;
1864 unsigned long cindex = 1;
1865 struct hlist_node *tmp;
1866 struct fib_alias *fa;
1869 /* walk trie in reverse order */
1871 unsigned char slen = 0;
1872 struct key_vector *n;
1875 t_key pkey = pn->key;
1877 /* cannot resize the trie vector */
1881 /* update the suffix to address pulled leaves */
1882 if (pn->slen > pn->pos)
1885 /* resize completed node */
1887 cindex = get_index(pkey, pn);
1892 /* grab the next available node */
1893 n = get_child(pn, cindex);
1898 /* record pn and cindex for leaf walking */
1900 cindex = 1ul << n->bits;
1905 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1906 struct fib_info *fi = fa->fa_info;
1908 if (!fi || tb->tb_id != fa->tb_id ||
1909 (!(fi->fib_flags & RTNH_F_DEAD) &&
1910 !fib_props[fa->fa_type].error)) {
1915 /* Do not flush error routes if network namespace is
1916 * not being dismantled
1918 if (!flush_all && fib_props[fa->fa_type].error) {
1923 call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL,
1925 KEYLENGTH - fa->fa_slen, fa,
1927 hlist_del_rcu(&fa->fa_list);
1928 fib_release_info(fa->fa_info);
1929 alias_free_mem_rcu(fa);
1933 /* update leaf slen */
1936 if (hlist_empty(&n->leaf)) {
1937 put_child_root(pn, n->key, NULL);
1942 pr_debug("trie_flush found=%d\n", found);
1946 /* derived from fib_trie_free */
1947 static void __fib_info_notify_update(struct net *net, struct fib_table *tb,
1948 struct nl_info *info)
1950 struct trie *t = (struct trie *)tb->tb_data;
1951 struct key_vector *pn = t->kv;
1952 unsigned long cindex = 1;
1953 struct fib_alias *fa;
1956 struct key_vector *n;
1959 t_key pkey = pn->key;
1965 pn = node_parent(pn);
1966 cindex = get_index(pkey, pn);
1970 /* grab the next available node */
1971 n = get_child(pn, cindex);
1976 /* record pn and cindex for leaf walking */
1978 cindex = 1ul << n->bits;
1983 hlist_for_each_entry(fa, &n->leaf, fa_list) {
1984 struct fib_info *fi = fa->fa_info;
1986 if (!fi || !fi->nh_updated || fa->tb_id != tb->tb_id)
1989 rtmsg_fib(RTM_NEWROUTE, htonl(n->key), fa,
1990 KEYLENGTH - fa->fa_slen, tb->tb_id,
1991 info, NLM_F_REPLACE);
1993 /* call_fib_entry_notifiers will be removed when
1994 * in-kernel notifier is implemented and supported
1995 * for nexthop objects
1997 call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_REPLACE,
1999 KEYLENGTH - fa->fa_slen, fa,
2005 void fib_info_notify_update(struct net *net, struct nl_info *info)
2009 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2010 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2011 struct fib_table *tb;
2013 hlist_for_each_entry_rcu(tb, head, tb_hlist)
2014 __fib_info_notify_update(net, tb, info);
2018 static void fib_leaf_notify(struct net *net, struct key_vector *l,
2019 struct fib_table *tb, struct notifier_block *nb)
2021 struct fib_alias *fa;
2023 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2024 struct fib_info *fi = fa->fa_info;
2029 /* local and main table can share the same trie,
2030 * so don't notify twice for the same entry.
2032 if (tb->tb_id != fa->tb_id)
2035 call_fib_entry_notifier(nb, net, FIB_EVENT_ENTRY_ADD, l->key,
2036 KEYLENGTH - fa->fa_slen, fa);
2040 static void fib_table_notify(struct net *net, struct fib_table *tb,
2041 struct notifier_block *nb)
2043 struct trie *t = (struct trie *)tb->tb_data;
2044 struct key_vector *l, *tp = t->kv;
2047 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2048 fib_leaf_notify(net, l, tb, nb);
2051 /* stop in case of wrap around */
2057 void fib_notify(struct net *net, struct notifier_block *nb)
2061 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2062 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2063 struct fib_table *tb;
2065 hlist_for_each_entry_rcu(tb, head, tb_hlist)
2066 fib_table_notify(net, tb, nb);
2070 static void __trie_free_rcu(struct rcu_head *head)
2072 struct fib_table *tb = container_of(head, struct fib_table, rcu);
2073 #ifdef CONFIG_IP_FIB_TRIE_STATS
2074 struct trie *t = (struct trie *)tb->tb_data;
2076 if (tb->tb_data == tb->__data)
2077 free_percpu(t->stats);
2078 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2082 void fib_free_table(struct fib_table *tb)
2084 call_rcu(&tb->rcu, __trie_free_rcu);
2087 static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
2088 struct sk_buff *skb, struct netlink_callback *cb,
2089 struct fib_dump_filter *filter)
2091 unsigned int flags = NLM_F_MULTI;
2092 __be32 xkey = htonl(l->key);
2093 struct fib_alias *fa;
2096 if (filter->filter_set)
2097 flags |= NLM_F_DUMP_FILTERED;
2102 /* rcu_read_lock is hold by caller */
2103 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2109 if (tb->tb_id != fa->tb_id)
2112 if (filter->filter_set) {
2113 if (filter->rt_type && fa->fa_type != filter->rt_type)
2116 if ((filter->protocol &&
2117 fa->fa_info->fib_protocol != filter->protocol))
2121 !fib_info_nh_uses_dev(fa->fa_info, filter->dev))
2125 err = fib_dump_info(skb, NETLINK_CB(cb->skb).portid,
2126 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2127 tb->tb_id, fa->fa_type,
2128 xkey, KEYLENGTH - fa->fa_slen,
2129 fa->fa_tos, fa->fa_info, flags);
2142 /* rcu_read_lock needs to be hold by caller from readside */
2143 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
2144 struct netlink_callback *cb, struct fib_dump_filter *filter)
2146 struct trie *t = (struct trie *)tb->tb_data;
2147 struct key_vector *l, *tp = t->kv;
2148 /* Dump starting at last key.
2149 * Note: 0.0.0.0/0 (ie default) is first key.
2151 int count = cb->args[2];
2152 t_key key = cb->args[3];
2154 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2157 err = fn_trie_dump_leaf(l, tb, skb, cb, filter);
2160 cb->args[2] = count;
2167 memset(&cb->args[4], 0,
2168 sizeof(cb->args) - 4*sizeof(cb->args[0]));
2170 /* stop loop if key wrapped back to 0 */
2176 cb->args[2] = count;
2181 void __init fib_trie_init(void)
2183 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
2184 sizeof(struct fib_alias),
2185 0, SLAB_PANIC, NULL);
2187 trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
2189 0, SLAB_PANIC, NULL);
2192 struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
2194 struct fib_table *tb;
2196 size_t sz = sizeof(*tb);
2199 sz += sizeof(struct trie);
2201 tb = kzalloc(sz, GFP_KERNEL);
2206 tb->tb_num_default = 0;
2207 tb->tb_data = (alias ? alias->__data : tb->__data);
2212 t = (struct trie *) tb->tb_data;
2213 t->kv[0].pos = KEYLENGTH;
2214 t->kv[0].slen = KEYLENGTH;
2215 #ifdef CONFIG_IP_FIB_TRIE_STATS
2216 t->stats = alloc_percpu(struct trie_use_stats);
2226 #ifdef CONFIG_PROC_FS
2227 /* Depth first Trie walk iterator */
2228 struct fib_trie_iter {
2229 struct seq_net_private p;
2230 struct fib_table *tb;
2231 struct key_vector *tnode;
2236 static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2238 unsigned long cindex = iter->index;
2239 struct key_vector *pn = iter->tnode;
2242 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2243 iter->tnode, iter->index, iter->depth);
2245 while (!IS_TRIE(pn)) {
2246 while (cindex < child_length(pn)) {
2247 struct key_vector *n = get_child_rcu(pn, cindex++);
2254 iter->index = cindex;
2256 /* push down one level */
2265 /* Current node exhausted, pop back up */
2267 pn = node_parent_rcu(pn);
2268 cindex = get_index(pkey, pn) + 1;
2272 /* record root node so further searches know we are done */
2279 static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2282 struct key_vector *n, *pn;
2288 n = rcu_dereference(pn->tnode[0]);
2305 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2307 struct key_vector *n;
2308 struct fib_trie_iter iter;
2310 memset(s, 0, sizeof(*s));
2313 for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2315 struct fib_alias *fa;
2318 s->totdepth += iter.depth;
2319 if (iter.depth > s->maxdepth)
2320 s->maxdepth = iter.depth;
2322 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2326 if (n->bits < MAX_STAT_DEPTH)
2327 s->nodesizes[n->bits]++;
2328 s->nullpointers += tn_info(n)->empty_children;
2335 * This outputs /proc/net/fib_triestats
2337 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2339 unsigned int i, max, pointers, bytes, avdepth;
2342 avdepth = stat->totdepth*100 / stat->leaves;
2346 seq_printf(seq, "\tAver depth: %u.%02d\n",
2347 avdepth / 100, avdepth % 100);
2348 seq_printf(seq, "\tMax depth: %u\n", stat->maxdepth);
2350 seq_printf(seq, "\tLeaves: %u\n", stat->leaves);
2351 bytes = LEAF_SIZE * stat->leaves;
2353 seq_printf(seq, "\tPrefixes: %u\n", stat->prefixes);
2354 bytes += sizeof(struct fib_alias) * stat->prefixes;
2356 seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2357 bytes += TNODE_SIZE(0) * stat->tnodes;
2359 max = MAX_STAT_DEPTH;
2360 while (max > 0 && stat->nodesizes[max-1] == 0)
2364 for (i = 1; i < max; i++)
2365 if (stat->nodesizes[i] != 0) {
2366 seq_printf(seq, " %u: %u", i, stat->nodesizes[i]);
2367 pointers += (1<<i) * stat->nodesizes[i];
2369 seq_putc(seq, '\n');
2370 seq_printf(seq, "\tPointers: %u\n", pointers);
2372 bytes += sizeof(struct key_vector *) * pointers;
2373 seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2374 seq_printf(seq, "Total size: %u kB\n", (bytes + 1023) / 1024);
2377 #ifdef CONFIG_IP_FIB_TRIE_STATS
2378 static void trie_show_usage(struct seq_file *seq,
2379 const struct trie_use_stats __percpu *stats)
2381 struct trie_use_stats s = { 0 };
2384 /* loop through all of the CPUs and gather up the stats */
2385 for_each_possible_cpu(cpu) {
2386 const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2388 s.gets += pcpu->gets;
2389 s.backtrack += pcpu->backtrack;
2390 s.semantic_match_passed += pcpu->semantic_match_passed;
2391 s.semantic_match_miss += pcpu->semantic_match_miss;
2392 s.null_node_hit += pcpu->null_node_hit;
2393 s.resize_node_skipped += pcpu->resize_node_skipped;
2396 seq_printf(seq, "\nCounters:\n---------\n");
2397 seq_printf(seq, "gets = %u\n", s.gets);
2398 seq_printf(seq, "backtracks = %u\n", s.backtrack);
2399 seq_printf(seq, "semantic match passed = %u\n",
2400 s.semantic_match_passed);
2401 seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2402 seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2403 seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2405 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2407 static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2409 if (tb->tb_id == RT_TABLE_LOCAL)
2410 seq_puts(seq, "Local:\n");
2411 else if (tb->tb_id == RT_TABLE_MAIN)
2412 seq_puts(seq, "Main:\n");
2414 seq_printf(seq, "Id %d:\n", tb->tb_id);
2418 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2420 struct net *net = (struct net *)seq->private;
2424 "Basic info: size of leaf:"
2425 " %zd bytes, size of tnode: %zd bytes.\n",
2426 LEAF_SIZE, TNODE_SIZE(0));
2428 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2429 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2430 struct fib_table *tb;
2432 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2433 struct trie *t = (struct trie *) tb->tb_data;
2434 struct trie_stat stat;
2439 fib_table_print(seq, tb);
2441 trie_collect_stats(t, &stat);
2442 trie_show_stats(seq, &stat);
2443 #ifdef CONFIG_IP_FIB_TRIE_STATS
2444 trie_show_usage(seq, t->stats);
2452 static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2454 struct fib_trie_iter *iter = seq->private;
2455 struct net *net = seq_file_net(seq);
2459 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2460 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2461 struct fib_table *tb;
2463 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2464 struct key_vector *n;
2466 for (n = fib_trie_get_first(iter,
2467 (struct trie *) tb->tb_data);
2468 n; n = fib_trie_get_next(iter))
2479 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2483 return fib_trie_get_idx(seq, *pos);
2486 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2488 struct fib_trie_iter *iter = seq->private;
2489 struct net *net = seq_file_net(seq);
2490 struct fib_table *tb = iter->tb;
2491 struct hlist_node *tb_node;
2493 struct key_vector *n;
2496 /* next node in same table */
2497 n = fib_trie_get_next(iter);
2501 /* walk rest of this hash chain */
2502 h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2503 while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2504 tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2505 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2510 /* new hash chain */
2511 while (++h < FIB_TABLE_HASHSZ) {
2512 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2513 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2514 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2526 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2532 static void seq_indent(struct seq_file *seq, int n)
2538 static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2541 case RT_SCOPE_UNIVERSE: return "universe";
2542 case RT_SCOPE_SITE: return "site";
2543 case RT_SCOPE_LINK: return "link";
2544 case RT_SCOPE_HOST: return "host";
2545 case RT_SCOPE_NOWHERE: return "nowhere";
2547 snprintf(buf, len, "scope=%d", s);
2552 static const char *const rtn_type_names[__RTN_MAX] = {
2553 [RTN_UNSPEC] = "UNSPEC",
2554 [RTN_UNICAST] = "UNICAST",
2555 [RTN_LOCAL] = "LOCAL",
2556 [RTN_BROADCAST] = "BROADCAST",
2557 [RTN_ANYCAST] = "ANYCAST",
2558 [RTN_MULTICAST] = "MULTICAST",
2559 [RTN_BLACKHOLE] = "BLACKHOLE",
2560 [RTN_UNREACHABLE] = "UNREACHABLE",
2561 [RTN_PROHIBIT] = "PROHIBIT",
2562 [RTN_THROW] = "THROW",
2564 [RTN_XRESOLVE] = "XRESOLVE",
2567 static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2569 if (t < __RTN_MAX && rtn_type_names[t])
2570 return rtn_type_names[t];
2571 snprintf(buf, len, "type %u", t);
2575 /* Pretty print the trie */
2576 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2578 const struct fib_trie_iter *iter = seq->private;
2579 struct key_vector *n = v;
2581 if (IS_TRIE(node_parent_rcu(n)))
2582 fib_table_print(seq, iter->tb);
2585 __be32 prf = htonl(n->key);
2587 seq_indent(seq, iter->depth-1);
2588 seq_printf(seq, " +-- %pI4/%zu %u %u %u\n",
2589 &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2590 tn_info(n)->full_children,
2591 tn_info(n)->empty_children);
2593 __be32 val = htonl(n->key);
2594 struct fib_alias *fa;
2596 seq_indent(seq, iter->depth);
2597 seq_printf(seq, " |-- %pI4\n", &val);
2599 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2600 char buf1[32], buf2[32];
2602 seq_indent(seq, iter->depth + 1);
2603 seq_printf(seq, " /%zu %s %s",
2604 KEYLENGTH - fa->fa_slen,
2605 rtn_scope(buf1, sizeof(buf1),
2606 fa->fa_info->fib_scope),
2607 rtn_type(buf2, sizeof(buf2),
2610 seq_printf(seq, " tos=%d", fa->fa_tos);
2611 seq_putc(seq, '\n');
2618 static const struct seq_operations fib_trie_seq_ops = {
2619 .start = fib_trie_seq_start,
2620 .next = fib_trie_seq_next,
2621 .stop = fib_trie_seq_stop,
2622 .show = fib_trie_seq_show,
2625 struct fib_route_iter {
2626 struct seq_net_private p;
2627 struct fib_table *main_tb;
2628 struct key_vector *tnode;
2633 static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2636 struct key_vector *l, **tp = &iter->tnode;
2639 /* use cached location of previously found key */
2640 if (iter->pos > 0 && pos >= iter->pos) {
2649 while ((l = leaf_walk_rcu(tp, key)) && (pos-- > 0)) {
2654 /* handle unlikely case of a key wrap */
2660 iter->key = l->key; /* remember it */
2662 iter->pos = 0; /* forget it */
2667 static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2670 struct fib_route_iter *iter = seq->private;
2671 struct fib_table *tb;
2676 tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2681 t = (struct trie *)tb->tb_data;
2682 iter->tnode = t->kv;
2685 return fib_route_get_idx(iter, *pos);
2688 iter->key = KEY_MAX;
2690 return SEQ_START_TOKEN;
2693 static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2695 struct fib_route_iter *iter = seq->private;
2696 struct key_vector *l = NULL;
2697 t_key key = iter->key + 1;
2701 /* only allow key of 0 for start of sequence */
2702 if ((v == SEQ_START_TOKEN) || key)
2703 l = leaf_walk_rcu(&iter->tnode, key);
2715 static void fib_route_seq_stop(struct seq_file *seq, void *v)
2721 static unsigned int fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2723 unsigned int flags = 0;
2725 if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2727 if (fi && fi->fib_nh->fib_nh_gw4)
2728 flags |= RTF_GATEWAY;
2729 if (mask == htonl(0xFFFFFFFF))
2736 * This outputs /proc/net/route.
2737 * The format of the file is not supposed to be changed
2738 * and needs to be same as fib_hash output to avoid breaking
2741 static int fib_route_seq_show(struct seq_file *seq, void *v)
2743 struct fib_route_iter *iter = seq->private;
2744 struct fib_table *tb = iter->main_tb;
2745 struct fib_alias *fa;
2746 struct key_vector *l = v;
2749 if (v == SEQ_START_TOKEN) {
2750 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2751 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2756 prefix = htonl(l->key);
2758 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2759 const struct fib_info *fi = fa->fa_info;
2760 __be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2761 unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2763 if ((fa->fa_type == RTN_BROADCAST) ||
2764 (fa->fa_type == RTN_MULTICAST))
2767 if (fa->tb_id != tb->tb_id)
2770 seq_setwidth(seq, 127);
2774 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2775 "%d\t%08X\t%d\t%u\t%u",
2776 fi->fib_dev ? fi->fib_dev->name : "*",
2778 fi->fib_nh->fib_nh_gw4, flags, 0, 0,
2782 fi->fib_advmss + 40 : 0),
2787 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2788 "%d\t%08X\t%d\t%u\t%u",
2789 prefix, 0, flags, 0, 0, 0,
2798 static const struct seq_operations fib_route_seq_ops = {
2799 .start = fib_route_seq_start,
2800 .next = fib_route_seq_next,
2801 .stop = fib_route_seq_stop,
2802 .show = fib_route_seq_show,
2805 int __net_init fib_proc_init(struct net *net)
2807 if (!proc_create_net("fib_trie", 0444, net->proc_net, &fib_trie_seq_ops,
2808 sizeof(struct fib_trie_iter)))
2811 if (!proc_create_net_single("fib_triestat", 0444, net->proc_net,
2812 fib_triestat_seq_show, NULL))
2815 if (!proc_create_net("route", 0444, net->proc_net, &fib_route_seq_ops,
2816 sizeof(struct fib_route_iter)))
2822 remove_proc_entry("fib_triestat", net->proc_net);
2824 remove_proc_entry("fib_trie", net->proc_net);
2829 void __net_exit fib_proc_exit(struct net *net)
2831 remove_proc_entry("fib_trie", net->proc_net);
2832 remove_proc_entry("fib_triestat", net->proc_net);
2833 remove_proc_entry("route", net->proc_net);
2836 #endif /* CONFIG_PROC_FS */