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 <net/net_namespace.h>
77 #include <net/protocol.h>
78 #include <net/route.h>
81 #include <net/ip_fib.h>
82 #include <net/switchdev.h>
83 #include "fib_lookup.h"
85 #define MAX_STAT_DEPTH 32
87 #define KEYLENGTH (8*sizeof(t_key))
88 #define KEY_MAX ((t_key)~0)
90 typedef unsigned int t_key;
92 #define IS_TRIE(n) ((n)->pos >= KEYLENGTH)
93 #define IS_TNODE(n) ((n)->bits)
94 #define IS_LEAF(n) (!(n)->bits)
98 unsigned char pos; /* 2log(KEYLENGTH) bits needed */
99 unsigned char bits; /* 2log(KEYLENGTH) bits needed */
102 /* This list pointer if valid if (pos | bits) == 0 (LEAF) */
103 struct hlist_head leaf;
104 /* This array is valid if (pos | bits) > 0 (TNODE) */
105 struct key_vector __rcu *tnode[0];
111 t_key empty_children; /* KEYLENGTH bits needed */
112 t_key full_children; /* KEYLENGTH bits needed */
113 struct key_vector __rcu *parent;
114 struct key_vector kv[1];
115 #define tn_bits kv[0].bits
118 #define TNODE_SIZE(n) offsetof(struct tnode, kv[0].tnode[n])
119 #define LEAF_SIZE TNODE_SIZE(1)
121 #ifdef CONFIG_IP_FIB_TRIE_STATS
122 struct trie_use_stats {
124 unsigned int backtrack;
125 unsigned int semantic_match_passed;
126 unsigned int semantic_match_miss;
127 unsigned int null_node_hit;
128 unsigned int resize_node_skipped;
133 unsigned int totdepth;
134 unsigned int maxdepth;
137 unsigned int nullpointers;
138 unsigned int prefixes;
139 unsigned int nodesizes[MAX_STAT_DEPTH];
143 struct key_vector kv[1];
144 #ifdef CONFIG_IP_FIB_TRIE_STATS
145 struct trie_use_stats __percpu *stats;
149 static struct key_vector *resize(struct trie *t, struct key_vector *tn);
150 static size_t tnode_free_size;
153 * synchronize_rcu after call_rcu for that many pages; it should be especially
154 * useful before resizing the root node with PREEMPT_NONE configs; the value was
155 * obtained experimentally, aiming to avoid visible slowdown.
157 static const int sync_pages = 128;
159 static struct kmem_cache *fn_alias_kmem __read_mostly;
160 static struct kmem_cache *trie_leaf_kmem __read_mostly;
162 static inline struct tnode *tn_info(struct key_vector *kv)
164 return container_of(kv, struct tnode, kv[0]);
167 /* caller must hold RTNL */
168 #define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
169 #define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
171 /* caller must hold RCU read lock or RTNL */
172 #define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
173 #define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
175 /* wrapper for rcu_assign_pointer */
176 static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
179 rcu_assign_pointer(tn_info(n)->parent, tp);
182 #define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
184 /* This provides us with the number of children in this node, in the case of a
185 * leaf this will return 0 meaning none of the children are accessible.
187 static inline unsigned long child_length(const struct key_vector *tn)
189 return (1ul << tn->bits) & ~(1ul);
192 #define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
194 static inline unsigned long get_index(t_key key, struct key_vector *kv)
196 unsigned long index = key ^ kv->key;
198 if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
201 return index >> kv->pos;
204 /* To understand this stuff, an understanding of keys and all their bits is
205 * necessary. Every node in the trie has a key associated with it, but not
206 * all of the bits in that key are significant.
208 * Consider a node 'n' and its parent 'tp'.
210 * If n is a leaf, every bit in its key is significant. Its presence is
211 * necessitated by path compression, since during a tree traversal (when
212 * searching for a leaf - unless we are doing an insertion) we will completely
213 * ignore all skipped bits we encounter. Thus we need to verify, at the end of
214 * a potentially successful search, that we have indeed been walking the
217 * Note that we can never "miss" the correct key in the tree if present by
218 * following the wrong path. Path compression ensures that segments of the key
219 * that are the same for all keys with a given prefix are skipped, but the
220 * skipped part *is* identical for each node in the subtrie below the skipped
221 * bit! trie_insert() in this implementation takes care of that.
223 * if n is an internal node - a 'tnode' here, the various parts of its key
224 * have many different meanings.
227 * _________________________________________________________________
228 * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
229 * -----------------------------------------------------------------
230 * 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16
232 * _________________________________________________________________
233 * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
234 * -----------------------------------------------------------------
235 * 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
242 * First, let's just ignore the bits that come before the parent tp, that is
243 * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
244 * point we do not use them for anything.
246 * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
247 * index into the parent's child array. That is, they will be used to find
248 * 'n' among tp's children.
250 * The bits from (n->pos + n->bits) to (tn->pos - 1) - "S" - are skipped bits
253 * All the bits we have seen so far are significant to the node n. The rest
254 * of the bits are really not needed or indeed known in n->key.
256 * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
257 * n's child array, and will of course be different for each child.
259 * The rest of the bits, from 0 to (n->pos + n->bits), are completely unknown
263 static const int halve_threshold = 25;
264 static const int inflate_threshold = 50;
265 static const int halve_threshold_root = 15;
266 static const int inflate_threshold_root = 30;
268 static void __alias_free_mem(struct rcu_head *head)
270 struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
271 kmem_cache_free(fn_alias_kmem, fa);
274 static inline void alias_free_mem_rcu(struct fib_alias *fa)
276 call_rcu(&fa->rcu, __alias_free_mem);
279 #define TNODE_KMALLOC_MAX \
280 ilog2((PAGE_SIZE - TNODE_SIZE(0)) / sizeof(struct key_vector *))
281 #define TNODE_VMALLOC_MAX \
282 ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
284 static void __node_free_rcu(struct rcu_head *head)
286 struct tnode *n = container_of(head, struct tnode, rcu);
289 kmem_cache_free(trie_leaf_kmem, n);
290 else if (n->tn_bits <= TNODE_KMALLOC_MAX)
296 #define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
298 static struct tnode *tnode_alloc(int bits)
302 /* verify bits is within bounds */
303 if (bits > TNODE_VMALLOC_MAX)
306 /* determine size and verify it is non-zero and didn't overflow */
307 size = TNODE_SIZE(1ul << bits);
309 if (size <= PAGE_SIZE)
310 return kzalloc(size, GFP_KERNEL);
312 return vzalloc(size);
315 static inline void empty_child_inc(struct key_vector *n)
317 ++tn_info(n)->empty_children ? : ++tn_info(n)->full_children;
320 static inline void empty_child_dec(struct key_vector *n)
322 tn_info(n)->empty_children-- ? : tn_info(n)->full_children--;
325 static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
327 struct tnode *kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
328 struct key_vector *l = kv->kv;
333 /* initialize key vector */
337 l->slen = fa->fa_slen;
339 /* link leaf to fib alias */
340 INIT_HLIST_HEAD(&l->leaf);
341 hlist_add_head(&fa->fa_list, &l->leaf);
346 static struct key_vector *tnode_new(t_key key, int pos, int bits)
348 struct tnode *tnode = tnode_alloc(bits);
349 unsigned int shift = pos + bits;
350 struct key_vector *tn = tnode->kv;
352 /* verify bits and pos their msb bits clear and values are valid */
353 BUG_ON(!bits || (shift > KEYLENGTH));
355 pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
356 sizeof(struct key_vector *) << bits);
361 if (bits == KEYLENGTH)
362 tnode->full_children = 1;
364 tnode->empty_children = 1ul << bits;
366 tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
374 /* Check whether a tnode 'n' is "full", i.e. it is an internal node
375 * and no bits are skipped. See discussion in dyntree paper p. 6
377 static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
379 return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
382 /* Add a child at position i overwriting the old value.
383 * Update the value of full_children and empty_children.
385 static void put_child(struct key_vector *tn, unsigned long i,
386 struct key_vector *n)
388 struct key_vector *chi = get_child(tn, i);
391 BUG_ON(i >= child_length(tn));
393 /* update emptyChildren, overflow into fullChildren */
399 /* update fullChildren */
400 wasfull = tnode_full(tn, chi);
401 isfull = tnode_full(tn, n);
403 if (wasfull && !isfull)
404 tn_info(tn)->full_children--;
405 else if (!wasfull && isfull)
406 tn_info(tn)->full_children++;
408 if (n && (tn->slen < n->slen))
411 rcu_assign_pointer(tn->tnode[i], n);
414 static void update_children(struct key_vector *tn)
418 /* update all of the child parent pointers */
419 for (i = child_length(tn); i;) {
420 struct key_vector *inode = get_child(tn, --i);
425 /* Either update the children of a tnode that
426 * already belongs to us or update the child
427 * to point to ourselves.
429 if (node_parent(inode) == tn)
430 update_children(inode);
432 node_set_parent(inode, tn);
436 static inline void put_child_root(struct key_vector *tp, t_key key,
437 struct key_vector *n)
440 rcu_assign_pointer(tp->tnode[0], n);
442 put_child(tp, get_index(key, tp), n);
445 static inline void tnode_free_init(struct key_vector *tn)
447 tn_info(tn)->rcu.next = NULL;
450 static inline void tnode_free_append(struct key_vector *tn,
451 struct key_vector *n)
453 tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
454 tn_info(tn)->rcu.next = &tn_info(n)->rcu;
457 static void tnode_free(struct key_vector *tn)
459 struct callback_head *head = &tn_info(tn)->rcu;
463 tnode_free_size += TNODE_SIZE(1ul << tn->bits);
466 tn = container_of(head, struct tnode, rcu)->kv;
469 if (tnode_free_size >= PAGE_SIZE * sync_pages) {
475 static struct key_vector *replace(struct trie *t,
476 struct key_vector *oldtnode,
477 struct key_vector *tn)
479 struct key_vector *tp = node_parent(oldtnode);
482 /* setup the parent pointer out of and back into this node */
483 NODE_INIT_PARENT(tn, tp);
484 put_child_root(tp, tn->key, tn);
486 /* update all of the child parent pointers */
489 /* all pointers should be clean so we are done */
490 tnode_free(oldtnode);
492 /* resize children now that oldtnode is freed */
493 for (i = child_length(tn); i;) {
494 struct key_vector *inode = get_child(tn, --i);
496 /* resize child node */
497 if (tnode_full(tn, inode))
498 tn = resize(t, inode);
504 static struct key_vector *inflate(struct trie *t,
505 struct key_vector *oldtnode)
507 struct key_vector *tn;
511 pr_debug("In inflate\n");
513 tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
517 /* prepare oldtnode to be freed */
518 tnode_free_init(oldtnode);
520 /* Assemble all of the pointers in our cluster, in this case that
521 * represents all of the pointers out of our allocated nodes that
522 * point to existing tnodes and the links between our allocated
525 for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
526 struct key_vector *inode = get_child(oldtnode, --i);
527 struct key_vector *node0, *node1;
534 /* A leaf or an internal node with skipped bits */
535 if (!tnode_full(oldtnode, inode)) {
536 put_child(tn, get_index(inode->key, tn), inode);
540 /* drop the node in the old tnode free list */
541 tnode_free_append(oldtnode, inode);
543 /* An internal node with two children */
544 if (inode->bits == 1) {
545 put_child(tn, 2 * i + 1, get_child(inode, 1));
546 put_child(tn, 2 * i, get_child(inode, 0));
550 /* We will replace this node 'inode' with two new
551 * ones, 'node0' and 'node1', each with half of the
552 * original children. The two new nodes will have
553 * a position one bit further down the key and this
554 * means that the "significant" part of their keys
555 * (see the discussion near the top of this file)
556 * will differ by one bit, which will be "0" in
557 * node0's key and "1" in node1's key. Since we are
558 * moving the key position by one step, the bit that
559 * we are moving away from - the bit at position
560 * (tn->pos) - is the one that will differ between
561 * node0 and node1. So... we synthesize that bit in the
564 node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
567 node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
569 tnode_free_append(tn, node1);
572 tnode_free_append(tn, node0);
574 /* populate child pointers in new nodes */
575 for (k = child_length(inode), j = k / 2; j;) {
576 put_child(node1, --j, get_child(inode, --k));
577 put_child(node0, j, get_child(inode, j));
578 put_child(node1, --j, get_child(inode, --k));
579 put_child(node0, j, get_child(inode, j));
582 /* link new nodes to parent */
583 NODE_INIT_PARENT(node1, tn);
584 NODE_INIT_PARENT(node0, tn);
586 /* link parent to nodes */
587 put_child(tn, 2 * i + 1, node1);
588 put_child(tn, 2 * i, node0);
591 /* setup the parent pointers into and out of this node */
592 return replace(t, oldtnode, tn);
594 /* all pointers should be clean so we are done */
600 static struct key_vector *halve(struct trie *t,
601 struct key_vector *oldtnode)
603 struct key_vector *tn;
606 pr_debug("In halve\n");
608 tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
612 /* prepare oldtnode to be freed */
613 tnode_free_init(oldtnode);
615 /* Assemble all of the pointers in our cluster, in this case that
616 * represents all of the pointers out of our allocated nodes that
617 * point to existing tnodes and the links between our allocated
620 for (i = child_length(oldtnode); i;) {
621 struct key_vector *node1 = get_child(oldtnode, --i);
622 struct key_vector *node0 = get_child(oldtnode, --i);
623 struct key_vector *inode;
625 /* At least one of the children is empty */
626 if (!node1 || !node0) {
627 put_child(tn, i / 2, node1 ? : node0);
631 /* Two nonempty children */
632 inode = tnode_new(node0->key, oldtnode->pos, 1);
635 tnode_free_append(tn, inode);
637 /* initialize pointers out of node */
638 put_child(inode, 1, node1);
639 put_child(inode, 0, node0);
640 NODE_INIT_PARENT(inode, tn);
642 /* link parent to node */
643 put_child(tn, i / 2, inode);
646 /* setup the parent pointers into and out of this node */
647 return replace(t, oldtnode, tn);
649 /* all pointers should be clean so we are done */
655 static struct key_vector *collapse(struct trie *t,
656 struct key_vector *oldtnode)
658 struct key_vector *n, *tp;
661 /* scan the tnode looking for that one child that might still exist */
662 for (n = NULL, i = child_length(oldtnode); !n && i;)
663 n = get_child(oldtnode, --i);
665 /* compress one level */
666 tp = node_parent(oldtnode);
667 put_child_root(tp, oldtnode->key, n);
668 node_set_parent(n, tp);
676 static unsigned char update_suffix(struct key_vector *tn)
678 unsigned char slen = tn->pos;
679 unsigned long stride, i;
681 /* search though the list of children looking for nodes that might
682 * have a suffix greater than the one we currently have. This is
683 * why we start with a stride of 2 since a stride of 1 would
684 * represent the nodes with suffix length equal to tn->pos
686 for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
687 struct key_vector *n = get_child(tn, i);
689 if (!n || (n->slen <= slen))
692 /* update stride and slen based on new value */
693 stride <<= (n->slen - slen);
697 /* if slen covers all but the last bit we can stop here
698 * there will be nothing longer than that since only node
699 * 0 and 1 << (bits - 1) could have that as their suffix
702 if ((slen + 1) >= (tn->pos + tn->bits))
711 /* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
712 * the Helsinki University of Technology and Matti Tikkanen of Nokia
713 * Telecommunications, page 6:
714 * "A node is doubled if the ratio of non-empty children to all
715 * children in the *doubled* node is at least 'high'."
717 * 'high' in this instance is the variable 'inflate_threshold'. It
718 * is expressed as a percentage, so we multiply it with
719 * child_length() and instead of multiplying by 2 (since the
720 * child array will be doubled by inflate()) and multiplying
721 * the left-hand side by 100 (to handle the percentage thing) we
722 * multiply the left-hand side by 50.
724 * The left-hand side may look a bit weird: child_length(tn)
725 * - tn->empty_children is of course the number of non-null children
726 * in the current node. tn->full_children is the number of "full"
727 * children, that is non-null tnodes with a skip value of 0.
728 * All of those will be doubled in the resulting inflated tnode, so
729 * we just count them one extra time here.
731 * A clearer way to write this would be:
733 * to_be_doubled = tn->full_children;
734 * not_to_be_doubled = child_length(tn) - tn->empty_children -
737 * new_child_length = child_length(tn) * 2;
739 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
741 * if (new_fill_factor >= inflate_threshold)
743 * ...and so on, tho it would mess up the while () loop.
746 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
750 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
751 * inflate_threshold * new_child_length
753 * expand not_to_be_doubled and to_be_doubled, and shorten:
754 * 100 * (child_length(tn) - tn->empty_children +
755 * tn->full_children) >= inflate_threshold * new_child_length
757 * expand new_child_length:
758 * 100 * (child_length(tn) - tn->empty_children +
759 * tn->full_children) >=
760 * inflate_threshold * child_length(tn) * 2
763 * 50 * (tn->full_children + child_length(tn) -
764 * tn->empty_children) >= inflate_threshold *
768 static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
770 unsigned long used = child_length(tn);
771 unsigned long threshold = used;
773 /* Keep root node larger */
774 threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
775 used -= tn_info(tn)->empty_children;
776 used += tn_info(tn)->full_children;
778 /* if bits == KEYLENGTH then pos = 0, and will fail below */
780 return (used > 1) && tn->pos && ((50 * used) >= threshold);
783 static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
785 unsigned long used = child_length(tn);
786 unsigned long threshold = used;
788 /* Keep root node larger */
789 threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
790 used -= tn_info(tn)->empty_children;
792 /* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
794 return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
797 static inline bool should_collapse(struct key_vector *tn)
799 unsigned long used = child_length(tn);
801 used -= tn_info(tn)->empty_children;
803 /* account for bits == KEYLENGTH case */
804 if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
807 /* One child or none, time to drop us from the trie */
812 static struct key_vector *resize(struct trie *t, struct key_vector *tn)
814 #ifdef CONFIG_IP_FIB_TRIE_STATS
815 struct trie_use_stats __percpu *stats = t->stats;
817 struct key_vector *tp = node_parent(tn);
818 unsigned long cindex = get_index(tn->key, tp);
819 int max_work = MAX_WORK;
821 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
822 tn, inflate_threshold, halve_threshold);
824 /* track the tnode via the pointer from the parent instead of
825 * doing it ourselves. This way we can let RCU fully do its
826 * thing without us interfering
828 BUG_ON(tn != get_child(tp, cindex));
830 /* Double as long as the resulting node has a number of
831 * nonempty nodes that are above the threshold.
833 while (should_inflate(tp, tn) && max_work) {
836 #ifdef CONFIG_IP_FIB_TRIE_STATS
837 this_cpu_inc(stats->resize_node_skipped);
843 tn = get_child(tp, cindex);
846 /* update parent in case inflate failed */
847 tp = node_parent(tn);
849 /* Return if at least one inflate is run */
850 if (max_work != MAX_WORK)
853 /* Halve as long as the number of empty children in this
854 * node is above threshold.
856 while (should_halve(tp, tn) && max_work) {
859 #ifdef CONFIG_IP_FIB_TRIE_STATS
860 this_cpu_inc(stats->resize_node_skipped);
866 tn = get_child(tp, cindex);
869 /* Only one child remains */
870 if (should_collapse(tn))
871 return collapse(t, tn);
873 /* update parent in case halve failed */
874 tp = node_parent(tn);
876 /* Return if at least one deflate was run */
877 if (max_work != MAX_WORK)
880 /* push the suffix length to the parent node */
881 if (tn->slen > tn->pos) {
882 unsigned char slen = update_suffix(tn);
891 static void leaf_pull_suffix(struct key_vector *tp, struct key_vector *l)
893 while ((tp->slen > tp->pos) && (tp->slen > l->slen)) {
894 if (update_suffix(tp) > l->slen)
896 tp = node_parent(tp);
900 static void leaf_push_suffix(struct key_vector *tn, struct key_vector *l)
902 /* if this is a new leaf then tn will be NULL and we can sort
903 * out parent suffix lengths as a part of trie_rebalance
905 while (tn->slen < l->slen) {
907 tn = node_parent(tn);
911 /* rcu_read_lock needs to be hold by caller from readside */
912 static struct key_vector *fib_find_node(struct trie *t,
913 struct key_vector **tp, u32 key)
915 struct key_vector *pn, *n = t->kv;
916 unsigned long index = 0;
920 n = get_child_rcu(n, index);
925 index = get_cindex(key, n);
927 /* This bit of code is a bit tricky but it combines multiple
928 * checks into a single check. The prefix consists of the
929 * prefix plus zeros for the bits in the cindex. The index
930 * is the difference between the key and this value. From
931 * this we can actually derive several pieces of data.
932 * if (index >= (1ul << bits))
933 * we have a mismatch in skip bits and failed
935 * we know the value is cindex
937 * This check is safe even if bits == KEYLENGTH due to the
938 * fact that we can only allocate a node with 32 bits if a
939 * long is greater than 32 bits.
941 if (index >= (1ul << n->bits)) {
946 /* keep searching until we find a perfect match leaf or NULL */
947 } while (IS_TNODE(n));
954 /* Return the first fib alias matching TOS with
955 * priority less than or equal to PRIO.
957 static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
958 u8 tos, u32 prio, u32 tb_id)
960 struct fib_alias *fa;
965 hlist_for_each_entry(fa, fah, fa_list) {
966 if (fa->fa_slen < slen)
968 if (fa->fa_slen != slen)
970 if (fa->tb_id > tb_id)
972 if (fa->tb_id != tb_id)
974 if (fa->fa_tos > tos)
976 if (fa->fa_info->fib_priority >= prio || fa->fa_tos < tos)
983 static void trie_rebalance(struct trie *t, struct key_vector *tn)
989 static int fib_insert_node(struct trie *t, struct key_vector *tp,
990 struct fib_alias *new, t_key key)
992 struct key_vector *n, *l;
994 l = leaf_new(key, new);
998 /* retrieve child from parent node */
999 n = get_child(tp, get_index(key, tp));
1001 /* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1003 * Add a new tnode here
1004 * first tnode need some special handling
1005 * leaves us in position for handling as case 3
1008 struct key_vector *tn;
1010 tn = tnode_new(key, __fls(key ^ n->key), 1);
1014 /* initialize routes out of node */
1015 NODE_INIT_PARENT(tn, tp);
1016 put_child(tn, get_index(key, tn) ^ 1, n);
1018 /* start adding routes into the node */
1019 put_child_root(tp, key, tn);
1020 node_set_parent(n, tn);
1022 /* parent now has a NULL spot where the leaf can go */
1026 /* Case 3: n is NULL, and will just insert a new leaf */
1027 NODE_INIT_PARENT(l, tp);
1028 put_child_root(tp, key, l);
1029 trie_rebalance(t, tp);
1038 static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1039 struct key_vector *l, struct fib_alias *new,
1040 struct fib_alias *fa, t_key key)
1043 return fib_insert_node(t, tp, new, key);
1046 hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1048 struct fib_alias *last;
1050 hlist_for_each_entry(last, &l->leaf, fa_list) {
1051 if (new->fa_slen < last->fa_slen)
1053 if ((new->fa_slen == last->fa_slen) &&
1054 (new->tb_id > last->tb_id))
1060 hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1062 hlist_add_head_rcu(&new->fa_list, &l->leaf);
1065 /* if we added to the tail node then we need to update slen */
1066 if (l->slen < new->fa_slen) {
1067 l->slen = new->fa_slen;
1068 leaf_push_suffix(tp, l);
1074 /* Caller must hold RTNL. */
1075 int fib_table_insert(struct fib_table *tb, struct fib_config *cfg)
1077 struct trie *t = (struct trie *)tb->tb_data;
1078 struct fib_alias *fa, *new_fa;
1079 struct key_vector *l, *tp;
1080 struct fib_info *fi;
1081 u8 plen = cfg->fc_dst_len;
1082 u8 slen = KEYLENGTH - plen;
1083 u8 tos = cfg->fc_tos;
1087 if (plen > KEYLENGTH)
1090 key = ntohl(cfg->fc_dst);
1092 pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1094 if ((plen < KEYLENGTH) && (key << plen))
1097 fi = fib_create_info(cfg);
1103 l = fib_find_node(t, &tp, key);
1104 fa = l ? fib_find_alias(&l->leaf, slen, tos, fi->fib_priority,
1107 /* Now fa, if non-NULL, points to the first fib alias
1108 * with the same keys [prefix,tos,priority], if such key already
1109 * exists or to the node before which we will insert new one.
1111 * If fa is NULL, we will need to allocate a new one and
1112 * insert to the tail of the section matching the suffix length
1116 if (fa && fa->fa_tos == tos &&
1117 fa->fa_info->fib_priority == fi->fib_priority) {
1118 struct fib_alias *fa_first, *fa_match;
1121 if (cfg->fc_nlflags & NLM_F_EXCL)
1125 * 1. Find exact match for type, scope, fib_info to avoid
1127 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1131 hlist_for_each_entry_from(fa, fa_list) {
1132 if ((fa->fa_slen != slen) ||
1133 (fa->tb_id != tb->tb_id) ||
1134 (fa->fa_tos != tos))
1136 if (fa->fa_info->fib_priority != fi->fib_priority)
1138 if (fa->fa_type == cfg->fc_type &&
1139 fa->fa_info == fi) {
1145 if (cfg->fc_nlflags & NLM_F_REPLACE) {
1146 struct fib_info *fi_drop;
1156 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1160 fi_drop = fa->fa_info;
1161 new_fa->fa_tos = fa->fa_tos;
1162 new_fa->fa_info = fi;
1163 new_fa->fa_type = cfg->fc_type;
1164 state = fa->fa_state;
1165 new_fa->fa_state = state & ~FA_S_ACCESSED;
1166 new_fa->fa_slen = fa->fa_slen;
1167 new_fa->tb_id = tb->tb_id;
1169 err = netdev_switch_fib_ipv4_add(key, plen, fi,
1175 netdev_switch_fib_ipv4_abort(fi);
1176 kmem_cache_free(fn_alias_kmem, new_fa);
1180 hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1182 alias_free_mem_rcu(fa);
1184 fib_release_info(fi_drop);
1185 if (state & FA_S_ACCESSED)
1186 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1187 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1188 tb->tb_id, &cfg->fc_nlinfo, NLM_F_REPLACE);
1192 /* Error if we find a perfect match which
1193 * uses the same scope, type, and nexthop
1199 if (!(cfg->fc_nlflags & NLM_F_APPEND))
1203 if (!(cfg->fc_nlflags & NLM_F_CREATE))
1207 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1211 new_fa->fa_info = fi;
1212 new_fa->fa_tos = tos;
1213 new_fa->fa_type = cfg->fc_type;
1214 new_fa->fa_state = 0;
1215 new_fa->fa_slen = slen;
1216 new_fa->tb_id = tb->tb_id;
1218 /* (Optionally) offload fib entry to switch hardware. */
1219 err = netdev_switch_fib_ipv4_add(key, plen, fi, tos,
1224 netdev_switch_fib_ipv4_abort(fi);
1225 goto out_free_new_fa;
1228 /* Insert new entry to the list. */
1229 err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1231 goto out_sw_fib_del;
1234 tb->tb_num_default++;
1236 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1237 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1238 &cfg->fc_nlinfo, 0);
1243 netdev_switch_fib_ipv4_del(key, plen, fi, tos, cfg->fc_type, tb->tb_id);
1245 kmem_cache_free(fn_alias_kmem, new_fa);
1247 fib_release_info(fi);
1252 static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1254 t_key prefix = n->key;
1256 return (key ^ prefix) & (prefix | -prefix);
1259 /* should be called with rcu_read_lock */
1260 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1261 struct fib_result *res, int fib_flags)
1263 struct trie *t = (struct trie *) tb->tb_data;
1264 #ifdef CONFIG_IP_FIB_TRIE_STATS
1265 struct trie_use_stats __percpu *stats = t->stats;
1267 const t_key key = ntohl(flp->daddr);
1268 struct key_vector *n, *pn;
1269 struct fib_alias *fa;
1270 unsigned long index;
1276 n = get_child_rcu(pn, cindex);
1280 #ifdef CONFIG_IP_FIB_TRIE_STATS
1281 this_cpu_inc(stats->gets);
1284 /* Step 1: Travel to the longest prefix match in the trie */
1286 index = get_cindex(key, n);
1288 /* This bit of code is a bit tricky but it combines multiple
1289 * checks into a single check. The prefix consists of the
1290 * prefix plus zeros for the "bits" in the prefix. The index
1291 * is the difference between the key and this value. From
1292 * this we can actually derive several pieces of data.
1293 * if (index >= (1ul << bits))
1294 * we have a mismatch in skip bits and failed
1296 * we know the value is cindex
1298 * This check is safe even if bits == KEYLENGTH due to the
1299 * fact that we can only allocate a node with 32 bits if a
1300 * long is greater than 32 bits.
1302 if (index >= (1ul << n->bits))
1305 /* we have found a leaf. Prefixes have already been compared */
1309 /* only record pn and cindex if we are going to be chopping
1310 * bits later. Otherwise we are just wasting cycles.
1312 if (n->slen > n->pos) {
1317 n = get_child_rcu(n, index);
1322 /* Step 2: Sort out leaves and begin backtracing for longest prefix */
1324 /* record the pointer where our next node pointer is stored */
1325 struct key_vector __rcu **cptr = n->tnode;
1327 /* This test verifies that none of the bits that differ
1328 * between the key and the prefix exist in the region of
1329 * the lsb and higher in the prefix.
1331 if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1334 /* exit out and process leaf */
1335 if (unlikely(IS_LEAF(n)))
1338 /* Don't bother recording parent info. Since we are in
1339 * prefix match mode we will have to come back to wherever
1340 * we started this traversal anyway
1343 while ((n = rcu_dereference(*cptr)) == NULL) {
1345 #ifdef CONFIG_IP_FIB_TRIE_STATS
1347 this_cpu_inc(stats->null_node_hit);
1349 /* If we are at cindex 0 there are no more bits for
1350 * us to strip at this level so we must ascend back
1351 * up one level to see if there are any more bits to
1352 * be stripped there.
1355 t_key pkey = pn->key;
1357 /* If we don't have a parent then there is
1358 * nothing for us to do as we do not have any
1359 * further nodes to parse.
1363 #ifdef CONFIG_IP_FIB_TRIE_STATS
1364 this_cpu_inc(stats->backtrack);
1366 /* Get Child's index */
1367 pn = node_parent_rcu(pn);
1368 cindex = get_index(pkey, pn);
1371 /* strip the least significant bit from the cindex */
1372 cindex &= cindex - 1;
1374 /* grab pointer for next child node */
1375 cptr = &pn->tnode[cindex];
1380 /* this line carries forward the xor from earlier in the function */
1381 index = key ^ n->key;
1383 /* Step 3: Process the leaf, if that fails fall back to backtracing */
1384 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1385 struct fib_info *fi = fa->fa_info;
1388 if ((index >= (1ul << fa->fa_slen)) &&
1389 ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen != KEYLENGTH)))
1391 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1395 if (fa->fa_info->fib_scope < flp->flowi4_scope)
1397 fib_alias_accessed(fa);
1398 err = fib_props[fa->fa_type].error;
1399 if (unlikely(err < 0)) {
1400 #ifdef CONFIG_IP_FIB_TRIE_STATS
1401 this_cpu_inc(stats->semantic_match_passed);
1405 if (fi->fib_flags & RTNH_F_DEAD)
1407 for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
1408 const struct fib_nh *nh = &fi->fib_nh[nhsel];
1410 if (nh->nh_flags & RTNH_F_DEAD)
1412 if (flp->flowi4_oif && flp->flowi4_oif != nh->nh_oif)
1415 if (!(fib_flags & FIB_LOOKUP_NOREF))
1416 atomic_inc(&fi->fib_clntref);
1418 res->prefixlen = KEYLENGTH - fa->fa_slen;
1419 res->nh_sel = nhsel;
1420 res->type = fa->fa_type;
1421 res->scope = fi->fib_scope;
1424 res->fa_head = &n->leaf;
1425 #ifdef CONFIG_IP_FIB_TRIE_STATS
1426 this_cpu_inc(stats->semantic_match_passed);
1431 #ifdef CONFIG_IP_FIB_TRIE_STATS
1432 this_cpu_inc(stats->semantic_match_miss);
1436 EXPORT_SYMBOL_GPL(fib_table_lookup);
1438 static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1439 struct key_vector *l, struct fib_alias *old)
1441 /* record the location of the previous list_info entry */
1442 struct hlist_node **pprev = old->fa_list.pprev;
1443 struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1445 /* remove the fib_alias from the list */
1446 hlist_del_rcu(&old->fa_list);
1448 /* if we emptied the list this leaf will be freed and we can sort
1449 * out parent suffix lengths as a part of trie_rebalance
1451 if (hlist_empty(&l->leaf)) {
1452 put_child_root(tp, l->key, NULL);
1454 trie_rebalance(t, tp);
1458 /* only access fa if it is pointing at the last valid hlist_node */
1462 /* update the trie with the latest suffix length */
1463 l->slen = fa->fa_slen;
1464 leaf_pull_suffix(tp, l);
1467 /* Caller must hold RTNL. */
1468 int fib_table_delete(struct fib_table *tb, struct fib_config *cfg)
1470 struct trie *t = (struct trie *) tb->tb_data;
1471 struct fib_alias *fa, *fa_to_delete;
1472 struct key_vector *l, *tp;
1473 u8 plen = cfg->fc_dst_len;
1474 u8 slen = KEYLENGTH - plen;
1475 u8 tos = cfg->fc_tos;
1478 if (plen > KEYLENGTH)
1481 key = ntohl(cfg->fc_dst);
1483 if ((plen < KEYLENGTH) && (key << plen))
1486 l = fib_find_node(t, &tp, key);
1490 fa = fib_find_alias(&l->leaf, slen, tos, 0, tb->tb_id);
1494 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1496 fa_to_delete = NULL;
1497 hlist_for_each_entry_from(fa, fa_list) {
1498 struct fib_info *fi = fa->fa_info;
1500 if ((fa->fa_slen != slen) ||
1501 (fa->tb_id != tb->tb_id) ||
1502 (fa->fa_tos != tos))
1505 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1506 (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1507 fa->fa_info->fib_scope == cfg->fc_scope) &&
1508 (!cfg->fc_prefsrc ||
1509 fi->fib_prefsrc == cfg->fc_prefsrc) &&
1510 (!cfg->fc_protocol ||
1511 fi->fib_protocol == cfg->fc_protocol) &&
1512 fib_nh_match(cfg, fi) == 0) {
1521 netdev_switch_fib_ipv4_del(key, plen, fa_to_delete->fa_info, tos,
1522 cfg->fc_type, tb->tb_id);
1524 rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1525 &cfg->fc_nlinfo, 0);
1528 tb->tb_num_default--;
1530 fib_remove_alias(t, tp, l, fa_to_delete);
1532 if (fa_to_delete->fa_state & FA_S_ACCESSED)
1533 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1535 fib_release_info(fa_to_delete->fa_info);
1536 alias_free_mem_rcu(fa_to_delete);
1540 /* Scan for the next leaf starting at the provided key value */
1541 static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1543 struct key_vector *pn, *n = *tn;
1544 unsigned long cindex;
1546 /* this loop is meant to try and find the key in the trie */
1548 /* record parent and next child index */
1550 cindex = key ? get_index(key, pn) : 0;
1552 if (cindex >> pn->bits)
1555 /* descend into the next child */
1556 n = get_child_rcu(pn, cindex++);
1560 /* guarantee forward progress on the keys */
1561 if (IS_LEAF(n) && (n->key >= key))
1563 } while (IS_TNODE(n));
1565 /* this loop will search for the next leaf with a greater key */
1566 while (!IS_TRIE(pn)) {
1567 /* if we exhausted the parent node we will need to climb */
1568 if (cindex >= (1ul << pn->bits)) {
1569 t_key pkey = pn->key;
1571 pn = node_parent_rcu(pn);
1572 cindex = get_index(pkey, pn) + 1;
1576 /* grab the next available node */
1577 n = get_child_rcu(pn, cindex++);
1581 /* no need to compare keys since we bumped the index */
1585 /* Rescan start scanning in new node */
1591 return NULL; /* Root of trie */
1593 /* if we are at the limit for keys just return NULL for the tnode */
1598 static void fib_trie_free(struct fib_table *tb)
1600 struct trie *t = (struct trie *)tb->tb_data;
1601 struct key_vector *pn = t->kv;
1602 unsigned long cindex = 1;
1603 struct hlist_node *tmp;
1604 struct fib_alias *fa;
1606 /* walk trie in reverse order and free everything */
1608 struct key_vector *n;
1611 t_key pkey = pn->key;
1617 pn = node_parent(pn);
1619 /* drop emptied tnode */
1620 put_child_root(pn, n->key, NULL);
1623 cindex = get_index(pkey, pn);
1628 /* grab the next available node */
1629 n = get_child(pn, cindex);
1634 /* record pn and cindex for leaf walking */
1636 cindex = 1ul << n->bits;
1641 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1642 hlist_del_rcu(&fa->fa_list);
1643 alias_free_mem_rcu(fa);
1646 put_child_root(pn, n->key, NULL);
1650 #ifdef CONFIG_IP_FIB_TRIE_STATS
1651 free_percpu(t->stats);
1656 struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1658 struct trie *ot = (struct trie *)oldtb->tb_data;
1659 struct key_vector *l, *tp = ot->kv;
1660 struct fib_table *local_tb;
1661 struct fib_alias *fa;
1665 if (oldtb->tb_data == oldtb->__data)
1668 local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1672 lt = (struct trie *)local_tb->tb_data;
1674 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1675 struct key_vector *local_l = NULL, *local_tp;
1677 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1678 struct fib_alias *new_fa;
1680 if (local_tb->tb_id != fa->tb_id)
1683 /* clone fa for new local table */
1684 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1688 memcpy(new_fa, fa, sizeof(*fa));
1690 /* insert clone into table */
1692 local_l = fib_find_node(lt, &local_tp, l->key);
1694 if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1699 /* stop loop if key wrapped back to 0 */
1707 fib_trie_free(local_tb);
1712 /* Caller must hold RTNL */
1713 void fib_table_flush_external(struct fib_table *tb)
1715 struct trie *t = (struct trie *)tb->tb_data;
1716 struct key_vector *pn = t->kv;
1717 unsigned long cindex = 1;
1718 struct hlist_node *tmp;
1719 struct fib_alias *fa;
1721 /* walk trie in reverse order */
1723 unsigned char slen = 0;
1724 struct key_vector *n;
1727 t_key pkey = pn->key;
1729 /* cannot resize the trie vector */
1733 /* resize completed node */
1735 cindex = get_index(pkey, pn);
1740 /* grab the next available node */
1741 n = get_child(pn, cindex);
1746 /* record pn and cindex for leaf walking */
1748 cindex = 1ul << n->bits;
1753 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1754 struct fib_info *fi = fa->fa_info;
1756 /* if alias was cloned to local then we just
1757 * need to remove the local copy from main
1759 if (tb->tb_id != fa->tb_id) {
1760 hlist_del_rcu(&fa->fa_list);
1761 alias_free_mem_rcu(fa);
1765 /* record local slen */
1768 if (!fi || !(fi->fib_flags & RTNH_F_OFFLOAD))
1771 netdev_switch_fib_ipv4_del(n->key,
1772 KEYLENGTH - fa->fa_slen,
1774 fa->fa_type, tb->tb_id);
1777 /* update leaf slen */
1780 if (hlist_empty(&n->leaf)) {
1781 put_child_root(pn, n->key, NULL);
1784 leaf_pull_suffix(pn, n);
1789 /* Caller must hold RTNL. */
1790 int fib_table_flush(struct fib_table *tb)
1792 struct trie *t = (struct trie *)tb->tb_data;
1793 struct key_vector *pn = t->kv;
1794 unsigned long cindex = 1;
1795 struct hlist_node *tmp;
1796 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 /* resize completed node */
1813 cindex = get_index(pkey, pn);
1818 /* grab the next available node */
1819 n = get_child(pn, cindex);
1824 /* record pn and cindex for leaf walking */
1826 cindex = 1ul << n->bits;
1831 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1832 struct fib_info *fi = fa->fa_info;
1834 if (!fi || !(fi->fib_flags & RTNH_F_DEAD)) {
1839 netdev_switch_fib_ipv4_del(n->key,
1840 KEYLENGTH - fa->fa_slen,
1842 fa->fa_type, tb->tb_id);
1843 hlist_del_rcu(&fa->fa_list);
1844 fib_release_info(fa->fa_info);
1845 alias_free_mem_rcu(fa);
1849 /* update leaf slen */
1852 if (hlist_empty(&n->leaf)) {
1853 put_child_root(pn, n->key, NULL);
1856 leaf_pull_suffix(pn, n);
1860 pr_debug("trie_flush found=%d\n", found);
1864 static void __trie_free_rcu(struct rcu_head *head)
1866 struct fib_table *tb = container_of(head, struct fib_table, rcu);
1867 #ifdef CONFIG_IP_FIB_TRIE_STATS
1868 struct trie *t = (struct trie *)tb->tb_data;
1870 if (tb->tb_data == tb->__data)
1871 free_percpu(t->stats);
1872 #endif /* CONFIG_IP_FIB_TRIE_STATS */
1876 void fib_free_table(struct fib_table *tb)
1878 call_rcu(&tb->rcu, __trie_free_rcu);
1881 static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
1882 struct sk_buff *skb, struct netlink_callback *cb)
1884 __be32 xkey = htonl(l->key);
1885 struct fib_alias *fa;
1891 /* rcu_read_lock is hold by caller */
1892 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1898 if (tb->tb_id != fa->tb_id) {
1903 if (fib_dump_info(skb, NETLINK_CB(cb->skb).portid,
1909 KEYLENGTH - fa->fa_slen,
1911 fa->fa_info, NLM_F_MULTI) < 0) {
1922 /* rcu_read_lock needs to be hold by caller from readside */
1923 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
1924 struct netlink_callback *cb)
1926 struct trie *t = (struct trie *)tb->tb_data;
1927 struct key_vector *l, *tp = t->kv;
1928 /* Dump starting at last key.
1929 * Note: 0.0.0.0/0 (ie default) is first key.
1931 int count = cb->args[2];
1932 t_key key = cb->args[3];
1934 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1935 if (fn_trie_dump_leaf(l, tb, skb, cb) < 0) {
1937 cb->args[2] = count;
1944 memset(&cb->args[4], 0,
1945 sizeof(cb->args) - 4*sizeof(cb->args[0]));
1947 /* stop loop if key wrapped back to 0 */
1953 cb->args[2] = count;
1958 void __init fib_trie_init(void)
1960 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
1961 sizeof(struct fib_alias),
1962 0, SLAB_PANIC, NULL);
1964 trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
1966 0, SLAB_PANIC, NULL);
1969 struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
1971 struct fib_table *tb;
1973 size_t sz = sizeof(*tb);
1976 sz += sizeof(struct trie);
1978 tb = kzalloc(sz, GFP_KERNEL);
1983 tb->tb_default = -1;
1984 tb->tb_num_default = 0;
1985 tb->tb_data = (alias ? alias->__data : tb->__data);
1990 t = (struct trie *) tb->tb_data;
1991 t->kv[0].pos = KEYLENGTH;
1992 t->kv[0].slen = KEYLENGTH;
1993 #ifdef CONFIG_IP_FIB_TRIE_STATS
1994 t->stats = alloc_percpu(struct trie_use_stats);
2004 #ifdef CONFIG_PROC_FS
2005 /* Depth first Trie walk iterator */
2006 struct fib_trie_iter {
2007 struct seq_net_private p;
2008 struct fib_table *tb;
2009 struct key_vector *tnode;
2014 static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2016 unsigned long cindex = iter->index;
2017 struct key_vector *pn = iter->tnode;
2020 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2021 iter->tnode, iter->index, iter->depth);
2023 while (!IS_TRIE(pn)) {
2024 while (cindex < child_length(pn)) {
2025 struct key_vector *n = get_child_rcu(pn, cindex++);
2032 iter->index = cindex;
2034 /* push down one level */
2043 /* Current node exhausted, pop back up */
2045 pn = node_parent_rcu(pn);
2046 cindex = get_index(pkey, pn) + 1;
2050 /* record root node so further searches know we are done */
2057 static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2060 struct key_vector *n, *pn = t->kv;
2065 n = rcu_dereference(pn->tnode[0]);
2082 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2084 struct key_vector *n;
2085 struct fib_trie_iter iter;
2087 memset(s, 0, sizeof(*s));
2090 for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2092 struct fib_alias *fa;
2095 s->totdepth += iter.depth;
2096 if (iter.depth > s->maxdepth)
2097 s->maxdepth = iter.depth;
2099 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2103 if (n->bits < MAX_STAT_DEPTH)
2104 s->nodesizes[n->bits]++;
2105 s->nullpointers += tn_info(n)->empty_children;
2112 * This outputs /proc/net/fib_triestats
2114 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2116 unsigned int i, max, pointers, bytes, avdepth;
2119 avdepth = stat->totdepth*100 / stat->leaves;
2123 seq_printf(seq, "\tAver depth: %u.%02d\n",
2124 avdepth / 100, avdepth % 100);
2125 seq_printf(seq, "\tMax depth: %u\n", stat->maxdepth);
2127 seq_printf(seq, "\tLeaves: %u\n", stat->leaves);
2128 bytes = LEAF_SIZE * stat->leaves;
2130 seq_printf(seq, "\tPrefixes: %u\n", stat->prefixes);
2131 bytes += sizeof(struct fib_alias) * stat->prefixes;
2133 seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2134 bytes += TNODE_SIZE(0) * stat->tnodes;
2136 max = MAX_STAT_DEPTH;
2137 while (max > 0 && stat->nodesizes[max-1] == 0)
2141 for (i = 1; i < max; i++)
2142 if (stat->nodesizes[i] != 0) {
2143 seq_printf(seq, " %u: %u", i, stat->nodesizes[i]);
2144 pointers += (1<<i) * stat->nodesizes[i];
2146 seq_putc(seq, '\n');
2147 seq_printf(seq, "\tPointers: %u\n", pointers);
2149 bytes += sizeof(struct key_vector *) * pointers;
2150 seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2151 seq_printf(seq, "Total size: %u kB\n", (bytes + 1023) / 1024);
2154 #ifdef CONFIG_IP_FIB_TRIE_STATS
2155 static void trie_show_usage(struct seq_file *seq,
2156 const struct trie_use_stats __percpu *stats)
2158 struct trie_use_stats s = { 0 };
2161 /* loop through all of the CPUs and gather up the stats */
2162 for_each_possible_cpu(cpu) {
2163 const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2165 s.gets += pcpu->gets;
2166 s.backtrack += pcpu->backtrack;
2167 s.semantic_match_passed += pcpu->semantic_match_passed;
2168 s.semantic_match_miss += pcpu->semantic_match_miss;
2169 s.null_node_hit += pcpu->null_node_hit;
2170 s.resize_node_skipped += pcpu->resize_node_skipped;
2173 seq_printf(seq, "\nCounters:\n---------\n");
2174 seq_printf(seq, "gets = %u\n", s.gets);
2175 seq_printf(seq, "backtracks = %u\n", s.backtrack);
2176 seq_printf(seq, "semantic match passed = %u\n",
2177 s.semantic_match_passed);
2178 seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2179 seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2180 seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2182 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2184 static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2186 if (tb->tb_id == RT_TABLE_LOCAL)
2187 seq_puts(seq, "Local:\n");
2188 else if (tb->tb_id == RT_TABLE_MAIN)
2189 seq_puts(seq, "Main:\n");
2191 seq_printf(seq, "Id %d:\n", tb->tb_id);
2195 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2197 struct net *net = (struct net *)seq->private;
2201 "Basic info: size of leaf:"
2202 " %Zd bytes, size of tnode: %Zd bytes.\n",
2203 LEAF_SIZE, TNODE_SIZE(0));
2205 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2206 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2207 struct fib_table *tb;
2209 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2210 struct trie *t = (struct trie *) tb->tb_data;
2211 struct trie_stat stat;
2216 fib_table_print(seq, tb);
2218 trie_collect_stats(t, &stat);
2219 trie_show_stats(seq, &stat);
2220 #ifdef CONFIG_IP_FIB_TRIE_STATS
2221 trie_show_usage(seq, t->stats);
2229 static int fib_triestat_seq_open(struct inode *inode, struct file *file)
2231 return single_open_net(inode, file, fib_triestat_seq_show);
2234 static const struct file_operations fib_triestat_fops = {
2235 .owner = THIS_MODULE,
2236 .open = fib_triestat_seq_open,
2238 .llseek = seq_lseek,
2239 .release = single_release_net,
2242 static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2244 struct fib_trie_iter *iter = seq->private;
2245 struct net *net = seq_file_net(seq);
2249 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2250 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2251 struct fib_table *tb;
2253 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2254 struct key_vector *n;
2256 for (n = fib_trie_get_first(iter,
2257 (struct trie *) tb->tb_data);
2258 n; n = fib_trie_get_next(iter))
2269 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2273 return fib_trie_get_idx(seq, *pos);
2276 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2278 struct fib_trie_iter *iter = seq->private;
2279 struct net *net = seq_file_net(seq);
2280 struct fib_table *tb = iter->tb;
2281 struct hlist_node *tb_node;
2283 struct key_vector *n;
2286 /* next node in same table */
2287 n = fib_trie_get_next(iter);
2291 /* walk rest of this hash chain */
2292 h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2293 while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2294 tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2295 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2300 /* new hash chain */
2301 while (++h < FIB_TABLE_HASHSZ) {
2302 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2303 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2304 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2316 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2322 static void seq_indent(struct seq_file *seq, int n)
2328 static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2331 case RT_SCOPE_UNIVERSE: return "universe";
2332 case RT_SCOPE_SITE: return "site";
2333 case RT_SCOPE_LINK: return "link";
2334 case RT_SCOPE_HOST: return "host";
2335 case RT_SCOPE_NOWHERE: return "nowhere";
2337 snprintf(buf, len, "scope=%d", s);
2342 static const char *const rtn_type_names[__RTN_MAX] = {
2343 [RTN_UNSPEC] = "UNSPEC",
2344 [RTN_UNICAST] = "UNICAST",
2345 [RTN_LOCAL] = "LOCAL",
2346 [RTN_BROADCAST] = "BROADCAST",
2347 [RTN_ANYCAST] = "ANYCAST",
2348 [RTN_MULTICAST] = "MULTICAST",
2349 [RTN_BLACKHOLE] = "BLACKHOLE",
2350 [RTN_UNREACHABLE] = "UNREACHABLE",
2351 [RTN_PROHIBIT] = "PROHIBIT",
2352 [RTN_THROW] = "THROW",
2354 [RTN_XRESOLVE] = "XRESOLVE",
2357 static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2359 if (t < __RTN_MAX && rtn_type_names[t])
2360 return rtn_type_names[t];
2361 snprintf(buf, len, "type %u", t);
2365 /* Pretty print the trie */
2366 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2368 const struct fib_trie_iter *iter = seq->private;
2369 struct key_vector *n = v;
2371 if (IS_TRIE(node_parent_rcu(n)))
2372 fib_table_print(seq, iter->tb);
2375 __be32 prf = htonl(n->key);
2377 seq_indent(seq, iter->depth-1);
2378 seq_printf(seq, " +-- %pI4/%zu %u %u %u\n",
2379 &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2380 tn_info(n)->full_children,
2381 tn_info(n)->empty_children);
2383 __be32 val = htonl(n->key);
2384 struct fib_alias *fa;
2386 seq_indent(seq, iter->depth);
2387 seq_printf(seq, " |-- %pI4\n", &val);
2389 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2390 char buf1[32], buf2[32];
2392 seq_indent(seq, iter->depth + 1);
2393 seq_printf(seq, " /%zu %s %s",
2394 KEYLENGTH - fa->fa_slen,
2395 rtn_scope(buf1, sizeof(buf1),
2396 fa->fa_info->fib_scope),
2397 rtn_type(buf2, sizeof(buf2),
2400 seq_printf(seq, " tos=%d", fa->fa_tos);
2401 seq_putc(seq, '\n');
2408 static const struct seq_operations fib_trie_seq_ops = {
2409 .start = fib_trie_seq_start,
2410 .next = fib_trie_seq_next,
2411 .stop = fib_trie_seq_stop,
2412 .show = fib_trie_seq_show,
2415 static int fib_trie_seq_open(struct inode *inode, struct file *file)
2417 return seq_open_net(inode, file, &fib_trie_seq_ops,
2418 sizeof(struct fib_trie_iter));
2421 static const struct file_operations fib_trie_fops = {
2422 .owner = THIS_MODULE,
2423 .open = fib_trie_seq_open,
2425 .llseek = seq_lseek,
2426 .release = seq_release_net,
2429 struct fib_route_iter {
2430 struct seq_net_private p;
2431 struct fib_table *main_tb;
2432 struct key_vector *tnode;
2437 static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2440 struct fib_table *tb = iter->main_tb;
2441 struct key_vector *l, **tp = &iter->tnode;
2445 /* use cache location of next-to-find key */
2446 if (iter->pos > 0 && pos >= iter->pos) {
2450 t = (struct trie *)tb->tb_data;
2451 iter->tnode = t->kv;
2456 while ((l = leaf_walk_rcu(tp, key)) != NULL) {
2465 /* handle unlikely case of a key wrap */
2471 iter->key = key; /* remember it */
2473 iter->pos = 0; /* forget it */
2478 static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2481 struct fib_route_iter *iter = seq->private;
2482 struct fib_table *tb;
2487 tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2494 return fib_route_get_idx(iter, *pos);
2496 t = (struct trie *)tb->tb_data;
2497 iter->tnode = t->kv;
2501 return SEQ_START_TOKEN;
2504 static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2506 struct fib_route_iter *iter = seq->private;
2507 struct key_vector *l = NULL;
2508 t_key key = iter->key;
2512 /* only allow key of 0 for start of sequence */
2513 if ((v == SEQ_START_TOKEN) || key)
2514 l = leaf_walk_rcu(&iter->tnode, key);
2517 iter->key = l->key + 1;
2526 static void fib_route_seq_stop(struct seq_file *seq, void *v)
2532 static unsigned int fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2534 unsigned int flags = 0;
2536 if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2538 if (fi && fi->fib_nh->nh_gw)
2539 flags |= RTF_GATEWAY;
2540 if (mask == htonl(0xFFFFFFFF))
2547 * This outputs /proc/net/route.
2548 * The format of the file is not supposed to be changed
2549 * and needs to be same as fib_hash output to avoid breaking
2552 static int fib_route_seq_show(struct seq_file *seq, void *v)
2554 struct fib_route_iter *iter = seq->private;
2555 struct fib_table *tb = iter->main_tb;
2556 struct fib_alias *fa;
2557 struct key_vector *l = v;
2560 if (v == SEQ_START_TOKEN) {
2561 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2562 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2567 prefix = htonl(l->key);
2569 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2570 const struct fib_info *fi = fa->fa_info;
2571 __be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2572 unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2574 if ((fa->fa_type == RTN_BROADCAST) ||
2575 (fa->fa_type == RTN_MULTICAST))
2578 if (fa->tb_id != tb->tb_id)
2581 seq_setwidth(seq, 127);
2585 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2586 "%d\t%08X\t%d\t%u\t%u",
2587 fi->fib_dev ? fi->fib_dev->name : "*",
2589 fi->fib_nh->nh_gw, flags, 0, 0,
2593 fi->fib_advmss + 40 : 0),
2598 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2599 "%d\t%08X\t%d\t%u\t%u",
2600 prefix, 0, flags, 0, 0, 0,
2609 static const struct seq_operations fib_route_seq_ops = {
2610 .start = fib_route_seq_start,
2611 .next = fib_route_seq_next,
2612 .stop = fib_route_seq_stop,
2613 .show = fib_route_seq_show,
2616 static int fib_route_seq_open(struct inode *inode, struct file *file)
2618 return seq_open_net(inode, file, &fib_route_seq_ops,
2619 sizeof(struct fib_route_iter));
2622 static const struct file_operations fib_route_fops = {
2623 .owner = THIS_MODULE,
2624 .open = fib_route_seq_open,
2626 .llseek = seq_lseek,
2627 .release = seq_release_net,
2630 int __net_init fib_proc_init(struct net *net)
2632 if (!proc_create("fib_trie", S_IRUGO, net->proc_net, &fib_trie_fops))
2635 if (!proc_create("fib_triestat", S_IRUGO, net->proc_net,
2636 &fib_triestat_fops))
2639 if (!proc_create("route", S_IRUGO, net->proc_net, &fib_route_fops))
2645 remove_proc_entry("fib_triestat", net->proc_net);
2647 remove_proc_entry("fib_trie", net->proc_net);
2652 void __net_exit fib_proc_exit(struct net *net)
2654 remove_proc_entry("fib_trie", net->proc_net);
2655 remove_proc_entry("fib_triestat", net->proc_net);
2656 remove_proc_entry("route", net->proc_net);
2659 #endif /* CONFIG_PROC_FS */