2 * This program is free software; you can redistribute it and/or
3 * modify it under the terms of the GNU General Public License
4 * as published by the Free Software Foundation; either version
5 * 2 of the License, or (at your option) any later version.
7 * Robert Olsson <robert.olsson@its.uu.se> Uppsala Universitet
8 * & Swedish University of Agricultural Sciences.
10 * Jens Laas <jens.laas@data.slu.se> Swedish University of
11 * Agricultural Sciences.
13 * Hans Liss <hans.liss@its.uu.se> Uppsala Universitet
15 * This work is based on the LPC-trie which is originally described in:
17 * An experimental study of compression methods for dynamic tries
18 * Stefan Nilsson and Matti Tikkanen. Algorithmica, 33(1):19-33, 2002.
19 * http://www.csc.kth.se/~snilsson/software/dyntrie2/
22 * IP-address lookup using LC-tries. Stefan Nilsson and Gunnar Karlsson
23 * IEEE Journal on Selected Areas in Communications, 17(6):1083-1092, June 1999
26 * Code from fib_hash has been reused which includes the following header:
29 * INET An implementation of the TCP/IP protocol suite for the LINUX
30 * operating system. INET is implemented using the BSD Socket
31 * interface as the means of communication with the user level.
33 * IPv4 FIB: lookup engine and maintenance routines.
36 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
38 * This program is free software; you can redistribute it and/or
39 * modify it under the terms of the GNU General Public License
40 * as published by the Free Software Foundation; either version
41 * 2 of the License, or (at your option) any later version.
43 * Substantial contributions to this work comes from:
45 * David S. Miller, <davem@davemloft.net>
46 * Stephen Hemminger <shemminger@osdl.org>
47 * Paul E. McKenney <paulmck@us.ibm.com>
48 * Patrick McHardy <kaber@trash.net>
51 #define VERSION "0.409"
53 #include <asm/uaccess.h>
54 #include <linux/bitops.h>
55 #include <linux/types.h>
56 #include <linux/kernel.h>
58 #include <linux/string.h>
59 #include <linux/socket.h>
60 #include <linux/sockios.h>
61 #include <linux/errno.h>
63 #include <linux/inet.h>
64 #include <linux/inetdevice.h>
65 #include <linux/netdevice.h>
66 #include <linux/if_arp.h>
67 #include <linux/proc_fs.h>
68 #include <linux/rcupdate.h>
69 #include <linux/skbuff.h>
70 #include <linux/netlink.h>
71 #include <linux/init.h>
72 #include <linux/list.h>
73 #include <linux/slab.h>
74 #include <linux/export.h>
75 #include <linux/vmalloc.h>
76 #include <net/net_namespace.h>
78 #include <net/protocol.h>
79 #include <net/route.h>
82 #include <net/ip_fib.h>
83 #include <net/switchdev.h>
84 #include <trace/events/fib.h>
85 #include "fib_lookup.h"
87 #define MAX_STAT_DEPTH 32
89 #define KEYLENGTH (8*sizeof(t_key))
90 #define KEY_MAX ((t_key)~0)
92 typedef unsigned int t_key;
94 #define IS_TRIE(n) ((n)->pos >= KEYLENGTH)
95 #define IS_TNODE(n) ((n)->bits)
96 #define IS_LEAF(n) (!(n)->bits)
100 unsigned char pos; /* 2log(KEYLENGTH) bits needed */
101 unsigned char bits; /* 2log(KEYLENGTH) bits needed */
104 /* This list pointer if valid if (pos | bits) == 0 (LEAF) */
105 struct hlist_head leaf;
106 /* This array is valid if (pos | bits) > 0 (TNODE) */
107 struct key_vector __rcu *tnode[0];
113 t_key empty_children; /* KEYLENGTH bits needed */
114 t_key full_children; /* KEYLENGTH bits needed */
115 struct key_vector __rcu *parent;
116 struct key_vector kv[1];
117 #define tn_bits kv[0].bits
120 #define TNODE_SIZE(n) offsetof(struct tnode, kv[0].tnode[n])
121 #define LEAF_SIZE TNODE_SIZE(1)
123 #ifdef CONFIG_IP_FIB_TRIE_STATS
124 struct trie_use_stats {
126 unsigned int backtrack;
127 unsigned int semantic_match_passed;
128 unsigned int semantic_match_miss;
129 unsigned int null_node_hit;
130 unsigned int resize_node_skipped;
135 unsigned int totdepth;
136 unsigned int maxdepth;
139 unsigned int nullpointers;
140 unsigned int prefixes;
141 unsigned int nodesizes[MAX_STAT_DEPTH];
145 struct key_vector kv[1];
146 #ifdef CONFIG_IP_FIB_TRIE_STATS
147 struct trie_use_stats __percpu *stats;
151 static struct key_vector *resize(struct trie *t, struct key_vector *tn);
152 static size_t tnode_free_size;
155 * synchronize_rcu after call_rcu for that many pages; it should be especially
156 * useful before resizing the root node with PREEMPT_NONE configs; the value was
157 * obtained experimentally, aiming to avoid visible slowdown.
159 static const int sync_pages = 128;
161 static struct kmem_cache *fn_alias_kmem __read_mostly;
162 static struct kmem_cache *trie_leaf_kmem __read_mostly;
164 static inline struct tnode *tn_info(struct key_vector *kv)
166 return container_of(kv, struct tnode, kv[0]);
169 /* caller must hold RTNL */
170 #define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
171 #define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
173 /* caller must hold RCU read lock or RTNL */
174 #define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
175 #define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
177 /* wrapper for rcu_assign_pointer */
178 static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
181 rcu_assign_pointer(tn_info(n)->parent, tp);
184 #define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
186 /* This provides us with the number of children in this node, in the case of a
187 * leaf this will return 0 meaning none of the children are accessible.
189 static inline unsigned long child_length(const struct key_vector *tn)
191 return (1ul << tn->bits) & ~(1ul);
194 #define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
196 static inline unsigned long get_index(t_key key, struct key_vector *kv)
198 unsigned long index = key ^ kv->key;
200 if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
203 return index >> kv->pos;
206 /* To understand this stuff, an understanding of keys and all their bits is
207 * necessary. Every node in the trie has a key associated with it, but not
208 * all of the bits in that key are significant.
210 * Consider a node 'n' and its parent 'tp'.
212 * If n is a leaf, every bit in its key is significant. Its presence is
213 * necessitated by path compression, since during a tree traversal (when
214 * searching for a leaf - unless we are doing an insertion) we will completely
215 * ignore all skipped bits we encounter. Thus we need to verify, at the end of
216 * a potentially successful search, that we have indeed been walking the
219 * Note that we can never "miss" the correct key in the tree if present by
220 * following the wrong path. Path compression ensures that segments of the key
221 * that are the same for all keys with a given prefix are skipped, but the
222 * skipped part *is* identical for each node in the subtrie below the skipped
223 * bit! trie_insert() in this implementation takes care of that.
225 * if n is an internal node - a 'tnode' here, the various parts of its key
226 * have many different meanings.
229 * _________________________________________________________________
230 * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
231 * -----------------------------------------------------------------
232 * 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16
234 * _________________________________________________________________
235 * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
236 * -----------------------------------------------------------------
237 * 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
244 * First, let's just ignore the bits that come before the parent tp, that is
245 * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
246 * point we do not use them for anything.
248 * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
249 * index into the parent's child array. That is, they will be used to find
250 * 'n' among tp's children.
252 * The bits from (n->pos + n->bits) to (tn->pos - 1) - "S" - are skipped bits
255 * All the bits we have seen so far are significant to the node n. The rest
256 * of the bits are really not needed or indeed known in n->key.
258 * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
259 * n's child array, and will of course be different for each child.
261 * The rest of the bits, from 0 to (n->pos + n->bits), are completely unknown
265 static const int halve_threshold = 25;
266 static const int inflate_threshold = 50;
267 static const int halve_threshold_root = 15;
268 static const int inflate_threshold_root = 30;
270 static void __alias_free_mem(struct rcu_head *head)
272 struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
273 kmem_cache_free(fn_alias_kmem, fa);
276 static inline void alias_free_mem_rcu(struct fib_alias *fa)
278 call_rcu(&fa->rcu, __alias_free_mem);
281 #define TNODE_KMALLOC_MAX \
282 ilog2((PAGE_SIZE - TNODE_SIZE(0)) / sizeof(struct key_vector *))
283 #define TNODE_VMALLOC_MAX \
284 ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
286 static void __node_free_rcu(struct rcu_head *head)
288 struct tnode *n = container_of(head, struct tnode, rcu);
291 kmem_cache_free(trie_leaf_kmem, n);
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 key_vector *l;
330 kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
334 /* initialize key vector */
339 l->slen = fa->fa_slen;
341 /* link leaf to fib alias */
342 INIT_HLIST_HEAD(&l->leaf);
343 hlist_add_head(&fa->fa_list, &l->leaf);
348 static struct key_vector *tnode_new(t_key key, int pos, int bits)
350 unsigned int shift = pos + bits;
351 struct key_vector *tn;
354 /* verify bits and pos their msb bits clear and values are valid */
355 BUG_ON(!bits || (shift > KEYLENGTH));
357 tnode = tnode_alloc(bits);
361 pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
362 sizeof(struct key_vector *) << bits);
364 if (bits == KEYLENGTH)
365 tnode->full_children = 1;
367 tnode->empty_children = 1ul << bits;
370 tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
378 /* Check whether a tnode 'n' is "full", i.e. it is an internal node
379 * and no bits are skipped. See discussion in dyntree paper p. 6
381 static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
383 return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
386 /* Add a child at position i overwriting the old value.
387 * Update the value of full_children and empty_children.
389 static void put_child(struct key_vector *tn, unsigned long i,
390 struct key_vector *n)
392 struct key_vector *chi = get_child(tn, i);
395 BUG_ON(i >= child_length(tn));
397 /* update emptyChildren, overflow into fullChildren */
403 /* update fullChildren */
404 wasfull = tnode_full(tn, chi);
405 isfull = tnode_full(tn, n);
407 if (wasfull && !isfull)
408 tn_info(tn)->full_children--;
409 else if (!wasfull && isfull)
410 tn_info(tn)->full_children++;
412 if (n && (tn->slen < n->slen))
415 rcu_assign_pointer(tn->tnode[i], n);
418 static void update_children(struct key_vector *tn)
422 /* update all of the child parent pointers */
423 for (i = child_length(tn); i;) {
424 struct key_vector *inode = get_child(tn, --i);
429 /* Either update the children of a tnode that
430 * already belongs to us or update the child
431 * to point to ourselves.
433 if (node_parent(inode) == tn)
434 update_children(inode);
436 node_set_parent(inode, tn);
440 static inline void put_child_root(struct key_vector *tp, t_key key,
441 struct key_vector *n)
444 rcu_assign_pointer(tp->tnode[0], n);
446 put_child(tp, get_index(key, tp), n);
449 static inline void tnode_free_init(struct key_vector *tn)
451 tn_info(tn)->rcu.next = NULL;
454 static inline void tnode_free_append(struct key_vector *tn,
455 struct key_vector *n)
457 tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
458 tn_info(tn)->rcu.next = &tn_info(n)->rcu;
461 static void tnode_free(struct key_vector *tn)
463 struct callback_head *head = &tn_info(tn)->rcu;
467 tnode_free_size += TNODE_SIZE(1ul << tn->bits);
470 tn = container_of(head, struct tnode, rcu)->kv;
473 if (tnode_free_size >= PAGE_SIZE * sync_pages) {
479 static struct key_vector *replace(struct trie *t,
480 struct key_vector *oldtnode,
481 struct key_vector *tn)
483 struct key_vector *tp = node_parent(oldtnode);
486 /* setup the parent pointer out of and back into this node */
487 NODE_INIT_PARENT(tn, tp);
488 put_child_root(tp, tn->key, tn);
490 /* update all of the child parent pointers */
493 /* all pointers should be clean so we are done */
494 tnode_free(oldtnode);
496 /* resize children now that oldtnode is freed */
497 for (i = child_length(tn); i;) {
498 struct key_vector *inode = get_child(tn, --i);
500 /* resize child node */
501 if (tnode_full(tn, inode))
502 tn = resize(t, inode);
508 static struct key_vector *inflate(struct trie *t,
509 struct key_vector *oldtnode)
511 struct key_vector *tn;
515 pr_debug("In inflate\n");
517 tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
521 /* prepare oldtnode to be freed */
522 tnode_free_init(oldtnode);
524 /* Assemble all of the pointers in our cluster, in this case that
525 * represents all of the pointers out of our allocated nodes that
526 * point to existing tnodes and the links between our allocated
529 for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
530 struct key_vector *inode = get_child(oldtnode, --i);
531 struct key_vector *node0, *node1;
538 /* A leaf or an internal node with skipped bits */
539 if (!tnode_full(oldtnode, inode)) {
540 put_child(tn, get_index(inode->key, tn), inode);
544 /* drop the node in the old tnode free list */
545 tnode_free_append(oldtnode, inode);
547 /* An internal node with two children */
548 if (inode->bits == 1) {
549 put_child(tn, 2 * i + 1, get_child(inode, 1));
550 put_child(tn, 2 * i, get_child(inode, 0));
554 /* We will replace this node 'inode' with two new
555 * ones, 'node0' and 'node1', each with half of the
556 * original children. The two new nodes will have
557 * a position one bit further down the key and this
558 * means that the "significant" part of their keys
559 * (see the discussion near the top of this file)
560 * will differ by one bit, which will be "0" in
561 * node0's key and "1" in node1's key. Since we are
562 * moving the key position by one step, the bit that
563 * we are moving away from - the bit at position
564 * (tn->pos) - is the one that will differ between
565 * node0 and node1. So... we synthesize that bit in the
568 node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
571 node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
573 tnode_free_append(tn, node1);
576 tnode_free_append(tn, node0);
578 /* populate child pointers in new nodes */
579 for (k = child_length(inode), j = k / 2; j;) {
580 put_child(node1, --j, get_child(inode, --k));
581 put_child(node0, j, get_child(inode, j));
582 put_child(node1, --j, get_child(inode, --k));
583 put_child(node0, j, get_child(inode, j));
586 /* link new nodes to parent */
587 NODE_INIT_PARENT(node1, tn);
588 NODE_INIT_PARENT(node0, tn);
590 /* link parent to nodes */
591 put_child(tn, 2 * i + 1, node1);
592 put_child(tn, 2 * i, node0);
595 /* setup the parent pointers into and out of this node */
596 return replace(t, oldtnode, tn);
598 /* all pointers should be clean so we are done */
604 static struct key_vector *halve(struct trie *t,
605 struct key_vector *oldtnode)
607 struct key_vector *tn;
610 pr_debug("In halve\n");
612 tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
616 /* prepare oldtnode to be freed */
617 tnode_free_init(oldtnode);
619 /* Assemble all of the pointers in our cluster, in this case that
620 * represents all of the pointers out of our allocated nodes that
621 * point to existing tnodes and the links between our allocated
624 for (i = child_length(oldtnode); i;) {
625 struct key_vector *node1 = get_child(oldtnode, --i);
626 struct key_vector *node0 = get_child(oldtnode, --i);
627 struct key_vector *inode;
629 /* At least one of the children is empty */
630 if (!node1 || !node0) {
631 put_child(tn, i / 2, node1 ? : node0);
635 /* Two nonempty children */
636 inode = tnode_new(node0->key, oldtnode->pos, 1);
639 tnode_free_append(tn, inode);
641 /* initialize pointers out of node */
642 put_child(inode, 1, node1);
643 put_child(inode, 0, node0);
644 NODE_INIT_PARENT(inode, tn);
646 /* link parent to node */
647 put_child(tn, i / 2, inode);
650 /* setup the parent pointers into and out of this node */
651 return replace(t, oldtnode, tn);
653 /* all pointers should be clean so we are done */
659 static struct key_vector *collapse(struct trie *t,
660 struct key_vector *oldtnode)
662 struct key_vector *n, *tp;
665 /* scan the tnode looking for that one child that might still exist */
666 for (n = NULL, i = child_length(oldtnode); !n && i;)
667 n = get_child(oldtnode, --i);
669 /* compress one level */
670 tp = node_parent(oldtnode);
671 put_child_root(tp, oldtnode->key, n);
672 node_set_parent(n, tp);
680 static unsigned char update_suffix(struct key_vector *tn)
682 unsigned char slen = tn->pos;
683 unsigned long stride, i;
685 /* search though the list of children looking for nodes that might
686 * have a suffix greater than the one we currently have. This is
687 * why we start with a stride of 2 since a stride of 1 would
688 * represent the nodes with suffix length equal to tn->pos
690 for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
691 struct key_vector *n = get_child(tn, i);
693 if (!n || (n->slen <= slen))
696 /* update stride and slen based on new value */
697 stride <<= (n->slen - slen);
701 /* if slen covers all but the last bit we can stop here
702 * there will be nothing longer than that since only node
703 * 0 and 1 << (bits - 1) could have that as their suffix
706 if ((slen + 1) >= (tn->pos + tn->bits))
715 /* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
716 * the Helsinki University of Technology and Matti Tikkanen of Nokia
717 * Telecommunications, page 6:
718 * "A node is doubled if the ratio of non-empty children to all
719 * children in the *doubled* node is at least 'high'."
721 * 'high' in this instance is the variable 'inflate_threshold'. It
722 * is expressed as a percentage, so we multiply it with
723 * child_length() and instead of multiplying by 2 (since the
724 * child array will be doubled by inflate()) and multiplying
725 * the left-hand side by 100 (to handle the percentage thing) we
726 * multiply the left-hand side by 50.
728 * The left-hand side may look a bit weird: child_length(tn)
729 * - tn->empty_children is of course the number of non-null children
730 * in the current node. tn->full_children is the number of "full"
731 * children, that is non-null tnodes with a skip value of 0.
732 * All of those will be doubled in the resulting inflated tnode, so
733 * we just count them one extra time here.
735 * A clearer way to write this would be:
737 * to_be_doubled = tn->full_children;
738 * not_to_be_doubled = child_length(tn) - tn->empty_children -
741 * new_child_length = child_length(tn) * 2;
743 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
745 * if (new_fill_factor >= inflate_threshold)
747 * ...and so on, tho it would mess up the while () loop.
750 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
754 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
755 * inflate_threshold * new_child_length
757 * expand not_to_be_doubled and to_be_doubled, and shorten:
758 * 100 * (child_length(tn) - tn->empty_children +
759 * tn->full_children) >= inflate_threshold * new_child_length
761 * expand new_child_length:
762 * 100 * (child_length(tn) - tn->empty_children +
763 * tn->full_children) >=
764 * inflate_threshold * child_length(tn) * 2
767 * 50 * (tn->full_children + child_length(tn) -
768 * tn->empty_children) >= inflate_threshold *
772 static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
774 unsigned long used = child_length(tn);
775 unsigned long threshold = used;
777 /* Keep root node larger */
778 threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
779 used -= tn_info(tn)->empty_children;
780 used += tn_info(tn)->full_children;
782 /* if bits == KEYLENGTH then pos = 0, and will fail below */
784 return (used > 1) && tn->pos && ((50 * used) >= threshold);
787 static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
789 unsigned long used = child_length(tn);
790 unsigned long threshold = used;
792 /* Keep root node larger */
793 threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
794 used -= tn_info(tn)->empty_children;
796 /* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
798 return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
801 static inline bool should_collapse(struct key_vector *tn)
803 unsigned long used = child_length(tn);
805 used -= tn_info(tn)->empty_children;
807 /* account for bits == KEYLENGTH case */
808 if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
811 /* One child or none, time to drop us from the trie */
816 static struct key_vector *resize(struct trie *t, struct key_vector *tn)
818 #ifdef CONFIG_IP_FIB_TRIE_STATS
819 struct trie_use_stats __percpu *stats = t->stats;
821 struct key_vector *tp = node_parent(tn);
822 unsigned long cindex = get_index(tn->key, tp);
823 int max_work = MAX_WORK;
825 pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
826 tn, inflate_threshold, halve_threshold);
828 /* track the tnode via the pointer from the parent instead of
829 * doing it ourselves. This way we can let RCU fully do its
830 * thing without us interfering
832 BUG_ON(tn != get_child(tp, cindex));
834 /* Double as long as the resulting node has a number of
835 * nonempty nodes that are above the threshold.
837 while (should_inflate(tp, tn) && max_work) {
840 #ifdef CONFIG_IP_FIB_TRIE_STATS
841 this_cpu_inc(stats->resize_node_skipped);
847 tn = get_child(tp, cindex);
850 /* update parent in case inflate failed */
851 tp = node_parent(tn);
853 /* Return if at least one inflate is run */
854 if (max_work != MAX_WORK)
857 /* Halve as long as the number of empty children in this
858 * node is above threshold.
860 while (should_halve(tp, tn) && max_work) {
863 #ifdef CONFIG_IP_FIB_TRIE_STATS
864 this_cpu_inc(stats->resize_node_skipped);
870 tn = get_child(tp, cindex);
873 /* Only one child remains */
874 if (should_collapse(tn))
875 return collapse(t, tn);
877 /* update parent in case halve failed */
878 tp = node_parent(tn);
880 /* Return if at least one deflate was run */
881 if (max_work != MAX_WORK)
884 /* push the suffix length to the parent node */
885 if (tn->slen > tn->pos) {
886 unsigned char slen = update_suffix(tn);
895 static void leaf_pull_suffix(struct key_vector *tp, struct key_vector *l)
897 while ((tp->slen > tp->pos) && (tp->slen > l->slen)) {
898 if (update_suffix(tp) > l->slen)
900 tp = node_parent(tp);
904 static void leaf_push_suffix(struct key_vector *tn, struct key_vector *l)
906 /* if this is a new leaf then tn will be NULL and we can sort
907 * out parent suffix lengths as a part of trie_rebalance
909 while (tn->slen < l->slen) {
911 tn = node_parent(tn);
915 /* rcu_read_lock needs to be hold by caller from readside */
916 static struct key_vector *fib_find_node(struct trie *t,
917 struct key_vector **tp, u32 key)
919 struct key_vector *pn, *n = t->kv;
920 unsigned long index = 0;
924 n = get_child_rcu(n, index);
929 index = get_cindex(key, n);
931 /* This bit of code is a bit tricky but it combines multiple
932 * checks into a single check. The prefix consists of the
933 * prefix plus zeros for the bits in the cindex. The index
934 * is the difference between the key and this value. From
935 * this we can actually derive several pieces of data.
936 * if (index >= (1ul << bits))
937 * we have a mismatch in skip bits and failed
939 * we know the value is cindex
941 * This check is safe even if bits == KEYLENGTH due to the
942 * fact that we can only allocate a node with 32 bits if a
943 * long is greater than 32 bits.
945 if (index >= (1ul << n->bits)) {
950 /* keep searching until we find a perfect match leaf or NULL */
951 } while (IS_TNODE(n));
958 /* Return the first fib alias matching TOS with
959 * priority less than or equal to PRIO.
961 static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
962 u8 tos, u32 prio, u32 tb_id)
964 struct fib_alias *fa;
969 hlist_for_each_entry(fa, fah, fa_list) {
970 if (fa->fa_slen < slen)
972 if (fa->fa_slen != slen)
974 if (fa->tb_id > tb_id)
976 if (fa->tb_id != tb_id)
978 if (fa->fa_tos > tos)
980 if (fa->fa_info->fib_priority >= prio || fa->fa_tos < tos)
987 static void trie_rebalance(struct trie *t, struct key_vector *tn)
993 static int fib_insert_node(struct trie *t, struct key_vector *tp,
994 struct fib_alias *new, t_key key)
996 struct key_vector *n, *l;
998 l = leaf_new(key, new);
1002 /* retrieve child from parent node */
1003 n = get_child(tp, get_index(key, tp));
1005 /* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1007 * Add a new tnode here
1008 * first tnode need some special handling
1009 * leaves us in position for handling as case 3
1012 struct key_vector *tn;
1014 tn = tnode_new(key, __fls(key ^ n->key), 1);
1018 /* initialize routes out of node */
1019 NODE_INIT_PARENT(tn, tp);
1020 put_child(tn, get_index(key, tn) ^ 1, n);
1022 /* start adding routes into the node */
1023 put_child_root(tp, key, tn);
1024 node_set_parent(n, tn);
1026 /* parent now has a NULL spot where the leaf can go */
1030 /* Case 3: n is NULL, and will just insert a new leaf */
1031 NODE_INIT_PARENT(l, tp);
1032 put_child_root(tp, key, l);
1033 trie_rebalance(t, tp);
1042 static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1043 struct key_vector *l, struct fib_alias *new,
1044 struct fib_alias *fa, t_key key)
1047 return fib_insert_node(t, tp, new, key);
1050 hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1052 struct fib_alias *last;
1054 hlist_for_each_entry(last, &l->leaf, fa_list) {
1055 if (new->fa_slen < last->fa_slen)
1057 if ((new->fa_slen == last->fa_slen) &&
1058 (new->tb_id > last->tb_id))
1064 hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1066 hlist_add_head_rcu(&new->fa_list, &l->leaf);
1069 /* if we added to the tail node then we need to update slen */
1070 if (l->slen < new->fa_slen) {
1071 l->slen = new->fa_slen;
1072 leaf_push_suffix(tp, l);
1078 /* Caller must hold RTNL. */
1079 int fib_table_insert(struct fib_table *tb, struct fib_config *cfg)
1081 struct trie *t = (struct trie *)tb->tb_data;
1082 struct fib_alias *fa, *new_fa;
1083 struct key_vector *l, *tp;
1084 unsigned int nlflags = 0;
1085 struct fib_info *fi;
1086 u8 plen = cfg->fc_dst_len;
1087 u8 slen = KEYLENGTH - plen;
1088 u8 tos = cfg->fc_tos;
1092 if (plen > KEYLENGTH)
1095 key = ntohl(cfg->fc_dst);
1097 pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1099 if ((plen < KEYLENGTH) && (key << plen))
1102 fi = fib_create_info(cfg);
1108 l = fib_find_node(t, &tp, key);
1109 fa = l ? fib_find_alias(&l->leaf, slen, tos, fi->fib_priority,
1112 /* Now fa, if non-NULL, points to the first fib alias
1113 * with the same keys [prefix,tos,priority], if such key already
1114 * exists or to the node before which we will insert new one.
1116 * If fa is NULL, we will need to allocate a new one and
1117 * insert to the tail of the section matching the suffix length
1121 if (fa && fa->fa_tos == tos &&
1122 fa->fa_info->fib_priority == fi->fib_priority) {
1123 struct fib_alias *fa_first, *fa_match;
1126 if (cfg->fc_nlflags & NLM_F_EXCL)
1130 * 1. Find exact match for type, scope, fib_info to avoid
1132 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1136 hlist_for_each_entry_from(fa, fa_list) {
1137 if ((fa->fa_slen != slen) ||
1138 (fa->tb_id != tb->tb_id) ||
1139 (fa->fa_tos != tos))
1141 if (fa->fa_info->fib_priority != fi->fib_priority)
1143 if (fa->fa_type == cfg->fc_type &&
1144 fa->fa_info == fi) {
1150 if (cfg->fc_nlflags & NLM_F_REPLACE) {
1151 struct fib_info *fi_drop;
1161 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1165 fi_drop = fa->fa_info;
1166 new_fa->fa_tos = fa->fa_tos;
1167 new_fa->fa_info = fi;
1168 new_fa->fa_type = cfg->fc_type;
1169 state = fa->fa_state;
1170 new_fa->fa_state = state & ~FA_S_ACCESSED;
1171 new_fa->fa_slen = fa->fa_slen;
1172 new_fa->tb_id = tb->tb_id;
1173 new_fa->fa_default = -1;
1175 err = switchdev_fib_ipv4_add(key, plen, fi,
1181 switchdev_fib_ipv4_abort(fi);
1182 kmem_cache_free(fn_alias_kmem, new_fa);
1186 hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1188 alias_free_mem_rcu(fa);
1190 fib_release_info(fi_drop);
1191 if (state & FA_S_ACCESSED)
1192 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1193 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1194 tb->tb_id, &cfg->fc_nlinfo, NLM_F_REPLACE);
1198 /* Error if we find a perfect match which
1199 * uses the same scope, type, and nexthop
1205 if (cfg->fc_nlflags & NLM_F_APPEND)
1206 nlflags = NLM_F_APPEND;
1211 if (!(cfg->fc_nlflags & NLM_F_CREATE))
1215 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1219 new_fa->fa_info = fi;
1220 new_fa->fa_tos = tos;
1221 new_fa->fa_type = cfg->fc_type;
1222 new_fa->fa_state = 0;
1223 new_fa->fa_slen = slen;
1224 new_fa->tb_id = tb->tb_id;
1225 new_fa->fa_default = -1;
1227 /* (Optionally) offload fib entry to switch hardware. */
1228 err = switchdev_fib_ipv4_add(key, plen, fi, tos, cfg->fc_type,
1229 cfg->fc_nlflags, tb->tb_id);
1231 switchdev_fib_ipv4_abort(fi);
1232 goto out_free_new_fa;
1235 /* Insert new entry to the list. */
1236 err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1238 goto out_sw_fib_del;
1241 tb->tb_num_default++;
1243 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1244 rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1245 &cfg->fc_nlinfo, nlflags);
1250 switchdev_fib_ipv4_del(key, plen, fi, tos, cfg->fc_type, tb->tb_id);
1252 kmem_cache_free(fn_alias_kmem, new_fa);
1254 fib_release_info(fi);
1259 static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1261 t_key prefix = n->key;
1263 return (key ^ prefix) & (prefix | -prefix);
1266 /* should be called with rcu_read_lock */
1267 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1268 struct fib_result *res, int fib_flags)
1270 struct trie *t = (struct trie *) tb->tb_data;
1271 #ifdef CONFIG_IP_FIB_TRIE_STATS
1272 struct trie_use_stats __percpu *stats = t->stats;
1274 const t_key key = ntohl(flp->daddr);
1275 struct key_vector *n, *pn;
1276 struct fib_alias *fa;
1277 unsigned long index;
1280 trace_fib_table_lookup(tb->tb_id, flp);
1285 n = get_child_rcu(pn, cindex);
1289 #ifdef CONFIG_IP_FIB_TRIE_STATS
1290 this_cpu_inc(stats->gets);
1293 /* Step 1: Travel to the longest prefix match in the trie */
1295 index = get_cindex(key, n);
1297 /* This bit of code is a bit tricky but it combines multiple
1298 * checks into a single check. The prefix consists of the
1299 * prefix plus zeros for the "bits" in the prefix. The index
1300 * is the difference between the key and this value. From
1301 * this we can actually derive several pieces of data.
1302 * if (index >= (1ul << bits))
1303 * we have a mismatch in skip bits and failed
1305 * we know the value is cindex
1307 * This check is safe even if bits == KEYLENGTH due to the
1308 * fact that we can only allocate a node with 32 bits if a
1309 * long is greater than 32 bits.
1311 if (index >= (1ul << n->bits))
1314 /* we have found a leaf. Prefixes have already been compared */
1318 /* only record pn and cindex if we are going to be chopping
1319 * bits later. Otherwise we are just wasting cycles.
1321 if (n->slen > n->pos) {
1326 n = get_child_rcu(n, index);
1331 /* Step 2: Sort out leaves and begin backtracing for longest prefix */
1333 /* record the pointer where our next node pointer is stored */
1334 struct key_vector __rcu **cptr = n->tnode;
1336 /* This test verifies that none of the bits that differ
1337 * between the key and the prefix exist in the region of
1338 * the lsb and higher in the prefix.
1340 if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1343 /* exit out and process leaf */
1344 if (unlikely(IS_LEAF(n)))
1347 /* Don't bother recording parent info. Since we are in
1348 * prefix match mode we will have to come back to wherever
1349 * we started this traversal anyway
1352 while ((n = rcu_dereference(*cptr)) == NULL) {
1354 #ifdef CONFIG_IP_FIB_TRIE_STATS
1356 this_cpu_inc(stats->null_node_hit);
1358 /* If we are at cindex 0 there are no more bits for
1359 * us to strip at this level so we must ascend back
1360 * up one level to see if there are any more bits to
1361 * be stripped there.
1364 t_key pkey = pn->key;
1366 /* If we don't have a parent then there is
1367 * nothing for us to do as we do not have any
1368 * further nodes to parse.
1372 #ifdef CONFIG_IP_FIB_TRIE_STATS
1373 this_cpu_inc(stats->backtrack);
1375 /* Get Child's index */
1376 pn = node_parent_rcu(pn);
1377 cindex = get_index(pkey, pn);
1380 /* strip the least significant bit from the cindex */
1381 cindex &= cindex - 1;
1383 /* grab pointer for next child node */
1384 cptr = &pn->tnode[cindex];
1389 /* this line carries forward the xor from earlier in the function */
1390 index = key ^ n->key;
1392 /* Step 3: Process the leaf, if that fails fall back to backtracing */
1393 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1394 struct fib_info *fi = fa->fa_info;
1397 if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1398 if (index >= (1ul << fa->fa_slen))
1401 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1405 if (fa->fa_info->fib_scope < flp->flowi4_scope)
1407 fib_alias_accessed(fa);
1408 err = fib_props[fa->fa_type].error;
1409 if (unlikely(err < 0)) {
1410 #ifdef CONFIG_IP_FIB_TRIE_STATS
1411 this_cpu_inc(stats->semantic_match_passed);
1415 if (fi->fib_flags & RTNH_F_DEAD)
1417 for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
1418 const struct fib_nh *nh = &fi->fib_nh[nhsel];
1419 struct in_device *in_dev = __in_dev_get_rcu(nh->nh_dev);
1421 if (nh->nh_flags & RTNH_F_DEAD)
1424 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1425 nh->nh_flags & RTNH_F_LINKDOWN &&
1426 !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1428 if (!(flp->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF)) {
1429 if (flp->flowi4_oif &&
1430 flp->flowi4_oif != nh->nh_oif)
1434 if (!(fib_flags & FIB_LOOKUP_NOREF))
1435 atomic_inc(&fi->fib_clntref);
1437 res->prefixlen = KEYLENGTH - fa->fa_slen;
1438 res->nh_sel = nhsel;
1439 res->type = fa->fa_type;
1440 res->scope = fi->fib_scope;
1443 res->fa_head = &n->leaf;
1444 #ifdef CONFIG_IP_FIB_TRIE_STATS
1445 this_cpu_inc(stats->semantic_match_passed);
1447 trace_fib_table_lookup_nh(nh);
1452 #ifdef CONFIG_IP_FIB_TRIE_STATS
1453 this_cpu_inc(stats->semantic_match_miss);
1457 EXPORT_SYMBOL_GPL(fib_table_lookup);
1459 static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1460 struct key_vector *l, struct fib_alias *old)
1462 /* record the location of the previous list_info entry */
1463 struct hlist_node **pprev = old->fa_list.pprev;
1464 struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1466 /* remove the fib_alias from the list */
1467 hlist_del_rcu(&old->fa_list);
1469 /* if we emptied the list this leaf will be freed and we can sort
1470 * out parent suffix lengths as a part of trie_rebalance
1472 if (hlist_empty(&l->leaf)) {
1473 put_child_root(tp, l->key, NULL);
1475 trie_rebalance(t, tp);
1479 /* only access fa if it is pointing at the last valid hlist_node */
1483 /* update the trie with the latest suffix length */
1484 l->slen = fa->fa_slen;
1485 leaf_pull_suffix(tp, l);
1488 /* Caller must hold RTNL. */
1489 int fib_table_delete(struct fib_table *tb, struct fib_config *cfg)
1491 struct trie *t = (struct trie *) tb->tb_data;
1492 struct fib_alias *fa, *fa_to_delete;
1493 struct key_vector *l, *tp;
1494 u8 plen = cfg->fc_dst_len;
1495 u8 slen = KEYLENGTH - plen;
1496 u8 tos = cfg->fc_tos;
1499 if (plen > KEYLENGTH)
1502 key = ntohl(cfg->fc_dst);
1504 if ((plen < KEYLENGTH) && (key << plen))
1507 l = fib_find_node(t, &tp, key);
1511 fa = fib_find_alias(&l->leaf, slen, tos, 0, tb->tb_id);
1515 pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1517 fa_to_delete = NULL;
1518 hlist_for_each_entry_from(fa, fa_list) {
1519 struct fib_info *fi = fa->fa_info;
1521 if ((fa->fa_slen != slen) ||
1522 (fa->tb_id != tb->tb_id) ||
1523 (fa->fa_tos != tos))
1526 if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1527 (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1528 fa->fa_info->fib_scope == cfg->fc_scope) &&
1529 (!cfg->fc_prefsrc ||
1530 fi->fib_prefsrc == cfg->fc_prefsrc) &&
1531 (!cfg->fc_protocol ||
1532 fi->fib_protocol == cfg->fc_protocol) &&
1533 fib_nh_match(cfg, fi) == 0) {
1542 switchdev_fib_ipv4_del(key, plen, fa_to_delete->fa_info, tos,
1543 cfg->fc_type, tb->tb_id);
1545 rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1546 &cfg->fc_nlinfo, 0);
1549 tb->tb_num_default--;
1551 fib_remove_alias(t, tp, l, fa_to_delete);
1553 if (fa_to_delete->fa_state & FA_S_ACCESSED)
1554 rt_cache_flush(cfg->fc_nlinfo.nl_net);
1556 fib_release_info(fa_to_delete->fa_info);
1557 alias_free_mem_rcu(fa_to_delete);
1561 /* Scan for the next leaf starting at the provided key value */
1562 static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1564 struct key_vector *pn, *n = *tn;
1565 unsigned long cindex;
1567 /* this loop is meant to try and find the key in the trie */
1569 /* record parent and next child index */
1571 cindex = (key > pn->key) ? get_index(key, pn) : 0;
1573 if (cindex >> pn->bits)
1576 /* descend into the next child */
1577 n = get_child_rcu(pn, cindex++);
1581 /* guarantee forward progress on the keys */
1582 if (IS_LEAF(n) && (n->key >= key))
1584 } while (IS_TNODE(n));
1586 /* this loop will search for the next leaf with a greater key */
1587 while (!IS_TRIE(pn)) {
1588 /* if we exhausted the parent node we will need to climb */
1589 if (cindex >= (1ul << pn->bits)) {
1590 t_key pkey = pn->key;
1592 pn = node_parent_rcu(pn);
1593 cindex = get_index(pkey, pn) + 1;
1597 /* grab the next available node */
1598 n = get_child_rcu(pn, cindex++);
1602 /* no need to compare keys since we bumped the index */
1606 /* Rescan start scanning in new node */
1612 return NULL; /* Root of trie */
1614 /* if we are at the limit for keys just return NULL for the tnode */
1619 static void fib_trie_free(struct fib_table *tb)
1621 struct trie *t = (struct trie *)tb->tb_data;
1622 struct key_vector *pn = t->kv;
1623 unsigned long cindex = 1;
1624 struct hlist_node *tmp;
1625 struct fib_alias *fa;
1627 /* walk trie in reverse order and free everything */
1629 struct key_vector *n;
1632 t_key pkey = pn->key;
1638 pn = node_parent(pn);
1640 /* drop emptied tnode */
1641 put_child_root(pn, n->key, NULL);
1644 cindex = get_index(pkey, pn);
1649 /* grab the next available node */
1650 n = get_child(pn, cindex);
1655 /* record pn and cindex for leaf walking */
1657 cindex = 1ul << n->bits;
1662 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1663 hlist_del_rcu(&fa->fa_list);
1664 alias_free_mem_rcu(fa);
1667 put_child_root(pn, n->key, NULL);
1671 #ifdef CONFIG_IP_FIB_TRIE_STATS
1672 free_percpu(t->stats);
1677 struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1679 struct trie *ot = (struct trie *)oldtb->tb_data;
1680 struct key_vector *l, *tp = ot->kv;
1681 struct fib_table *local_tb;
1682 struct fib_alias *fa;
1686 if (oldtb->tb_data == oldtb->__data)
1689 local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1693 lt = (struct trie *)local_tb->tb_data;
1695 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1696 struct key_vector *local_l = NULL, *local_tp;
1698 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1699 struct fib_alias *new_fa;
1701 if (local_tb->tb_id != fa->tb_id)
1704 /* clone fa for new local table */
1705 new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1709 memcpy(new_fa, fa, sizeof(*fa));
1711 /* insert clone into table */
1713 local_l = fib_find_node(lt, &local_tp, l->key);
1715 if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1720 /* stop loop if key wrapped back to 0 */
1728 fib_trie_free(local_tb);
1733 /* Caller must hold RTNL */
1734 void fib_table_flush_external(struct fib_table *tb)
1736 struct trie *t = (struct trie *)tb->tb_data;
1737 struct key_vector *pn = t->kv;
1738 unsigned long cindex = 1;
1739 struct hlist_node *tmp;
1740 struct fib_alias *fa;
1742 /* walk trie in reverse order */
1744 unsigned char slen = 0;
1745 struct key_vector *n;
1748 t_key pkey = pn->key;
1750 /* cannot resize the trie vector */
1754 /* resize completed node */
1756 cindex = get_index(pkey, pn);
1761 /* grab the next available node */
1762 n = get_child(pn, cindex);
1767 /* record pn and cindex for leaf walking */
1769 cindex = 1ul << n->bits;
1774 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1775 struct fib_info *fi = fa->fa_info;
1777 /* if alias was cloned to local then we just
1778 * need to remove the local copy from main
1780 if (tb->tb_id != fa->tb_id) {
1781 hlist_del_rcu(&fa->fa_list);
1782 alias_free_mem_rcu(fa);
1786 /* record local slen */
1789 if (!fi || !(fi->fib_flags & RTNH_F_OFFLOAD))
1792 switchdev_fib_ipv4_del(n->key, KEYLENGTH - fa->fa_slen,
1793 fi, fa->fa_tos, fa->fa_type,
1797 /* update leaf slen */
1800 if (hlist_empty(&n->leaf)) {
1801 put_child_root(pn, n->key, NULL);
1807 /* Caller must hold RTNL. */
1808 int fib_table_flush(struct fib_table *tb)
1810 struct trie *t = (struct trie *)tb->tb_data;
1811 struct key_vector *pn = t->kv;
1812 unsigned long cindex = 1;
1813 struct hlist_node *tmp;
1814 struct fib_alias *fa;
1817 /* walk trie in reverse order */
1819 unsigned char slen = 0;
1820 struct key_vector *n;
1823 t_key pkey = pn->key;
1825 /* cannot resize the trie vector */
1829 /* resize completed node */
1831 cindex = get_index(pkey, pn);
1836 /* grab the next available node */
1837 n = get_child(pn, cindex);
1842 /* record pn and cindex for leaf walking */
1844 cindex = 1ul << n->bits;
1849 hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1850 struct fib_info *fi = fa->fa_info;
1852 if (!fi || !(fi->fib_flags & RTNH_F_DEAD)) {
1857 switchdev_fib_ipv4_del(n->key, KEYLENGTH - fa->fa_slen,
1858 fi, fa->fa_tos, fa->fa_type,
1860 hlist_del_rcu(&fa->fa_list);
1861 fib_release_info(fa->fa_info);
1862 alias_free_mem_rcu(fa);
1866 /* update leaf slen */
1869 if (hlist_empty(&n->leaf)) {
1870 put_child_root(pn, n->key, NULL);
1875 pr_debug("trie_flush found=%d\n", found);
1879 static void __trie_free_rcu(struct rcu_head *head)
1881 struct fib_table *tb = container_of(head, struct fib_table, rcu);
1882 #ifdef CONFIG_IP_FIB_TRIE_STATS
1883 struct trie *t = (struct trie *)tb->tb_data;
1885 if (tb->tb_data == tb->__data)
1886 free_percpu(t->stats);
1887 #endif /* CONFIG_IP_FIB_TRIE_STATS */
1891 void fib_free_table(struct fib_table *tb)
1893 call_rcu(&tb->rcu, __trie_free_rcu);
1896 static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
1897 struct sk_buff *skb, struct netlink_callback *cb)
1899 __be32 xkey = htonl(l->key);
1900 struct fib_alias *fa;
1906 /* rcu_read_lock is hold by caller */
1907 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1913 if (tb->tb_id != fa->tb_id) {
1918 if (fib_dump_info(skb, NETLINK_CB(cb->skb).portid,
1924 KEYLENGTH - fa->fa_slen,
1926 fa->fa_info, NLM_F_MULTI) < 0) {
1937 /* rcu_read_lock needs to be hold by caller from readside */
1938 int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
1939 struct netlink_callback *cb)
1941 struct trie *t = (struct trie *)tb->tb_data;
1942 struct key_vector *l, *tp = t->kv;
1943 /* Dump starting at last key.
1944 * Note: 0.0.0.0/0 (ie default) is first key.
1946 int count = cb->args[2];
1947 t_key key = cb->args[3];
1949 while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1950 if (fn_trie_dump_leaf(l, tb, skb, cb) < 0) {
1952 cb->args[2] = count;
1959 memset(&cb->args[4], 0,
1960 sizeof(cb->args) - 4*sizeof(cb->args[0]));
1962 /* stop loop if key wrapped back to 0 */
1968 cb->args[2] = count;
1973 void __init fib_trie_init(void)
1975 fn_alias_kmem = kmem_cache_create("ip_fib_alias",
1976 sizeof(struct fib_alias),
1977 0, SLAB_PANIC, NULL);
1979 trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
1981 0, SLAB_PANIC, NULL);
1984 struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
1986 struct fib_table *tb;
1988 size_t sz = sizeof(*tb);
1991 sz += sizeof(struct trie);
1993 tb = kzalloc(sz, GFP_KERNEL);
1998 tb->tb_num_default = 0;
1999 tb->tb_data = (alias ? alias->__data : tb->__data);
2004 t = (struct trie *) tb->tb_data;
2005 t->kv[0].pos = KEYLENGTH;
2006 t->kv[0].slen = KEYLENGTH;
2007 #ifdef CONFIG_IP_FIB_TRIE_STATS
2008 t->stats = alloc_percpu(struct trie_use_stats);
2018 #ifdef CONFIG_PROC_FS
2019 /* Depth first Trie walk iterator */
2020 struct fib_trie_iter {
2021 struct seq_net_private p;
2022 struct fib_table *tb;
2023 struct key_vector *tnode;
2028 static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2030 unsigned long cindex = iter->index;
2031 struct key_vector *pn = iter->tnode;
2034 pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2035 iter->tnode, iter->index, iter->depth);
2037 while (!IS_TRIE(pn)) {
2038 while (cindex < child_length(pn)) {
2039 struct key_vector *n = get_child_rcu(pn, cindex++);
2046 iter->index = cindex;
2048 /* push down one level */
2057 /* Current node exhausted, pop back up */
2059 pn = node_parent_rcu(pn);
2060 cindex = get_index(pkey, pn) + 1;
2064 /* record root node so further searches know we are done */
2071 static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2074 struct key_vector *n, *pn;
2080 n = rcu_dereference(pn->tnode[0]);
2097 static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2099 struct key_vector *n;
2100 struct fib_trie_iter iter;
2102 memset(s, 0, sizeof(*s));
2105 for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2107 struct fib_alias *fa;
2110 s->totdepth += iter.depth;
2111 if (iter.depth > s->maxdepth)
2112 s->maxdepth = iter.depth;
2114 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2118 if (n->bits < MAX_STAT_DEPTH)
2119 s->nodesizes[n->bits]++;
2120 s->nullpointers += tn_info(n)->empty_children;
2127 * This outputs /proc/net/fib_triestats
2129 static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2131 unsigned int i, max, pointers, bytes, avdepth;
2134 avdepth = stat->totdepth*100 / stat->leaves;
2138 seq_printf(seq, "\tAver depth: %u.%02d\n",
2139 avdepth / 100, avdepth % 100);
2140 seq_printf(seq, "\tMax depth: %u\n", stat->maxdepth);
2142 seq_printf(seq, "\tLeaves: %u\n", stat->leaves);
2143 bytes = LEAF_SIZE * stat->leaves;
2145 seq_printf(seq, "\tPrefixes: %u\n", stat->prefixes);
2146 bytes += sizeof(struct fib_alias) * stat->prefixes;
2148 seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2149 bytes += TNODE_SIZE(0) * stat->tnodes;
2151 max = MAX_STAT_DEPTH;
2152 while (max > 0 && stat->nodesizes[max-1] == 0)
2156 for (i = 1; i < max; i++)
2157 if (stat->nodesizes[i] != 0) {
2158 seq_printf(seq, " %u: %u", i, stat->nodesizes[i]);
2159 pointers += (1<<i) * stat->nodesizes[i];
2161 seq_putc(seq, '\n');
2162 seq_printf(seq, "\tPointers: %u\n", pointers);
2164 bytes += sizeof(struct key_vector *) * pointers;
2165 seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2166 seq_printf(seq, "Total size: %u kB\n", (bytes + 1023) / 1024);
2169 #ifdef CONFIG_IP_FIB_TRIE_STATS
2170 static void trie_show_usage(struct seq_file *seq,
2171 const struct trie_use_stats __percpu *stats)
2173 struct trie_use_stats s = { 0 };
2176 /* loop through all of the CPUs and gather up the stats */
2177 for_each_possible_cpu(cpu) {
2178 const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2180 s.gets += pcpu->gets;
2181 s.backtrack += pcpu->backtrack;
2182 s.semantic_match_passed += pcpu->semantic_match_passed;
2183 s.semantic_match_miss += pcpu->semantic_match_miss;
2184 s.null_node_hit += pcpu->null_node_hit;
2185 s.resize_node_skipped += pcpu->resize_node_skipped;
2188 seq_printf(seq, "\nCounters:\n---------\n");
2189 seq_printf(seq, "gets = %u\n", s.gets);
2190 seq_printf(seq, "backtracks = %u\n", s.backtrack);
2191 seq_printf(seq, "semantic match passed = %u\n",
2192 s.semantic_match_passed);
2193 seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2194 seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2195 seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2197 #endif /* CONFIG_IP_FIB_TRIE_STATS */
2199 static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2201 if (tb->tb_id == RT_TABLE_LOCAL)
2202 seq_puts(seq, "Local:\n");
2203 else if (tb->tb_id == RT_TABLE_MAIN)
2204 seq_puts(seq, "Main:\n");
2206 seq_printf(seq, "Id %d:\n", tb->tb_id);
2210 static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2212 struct net *net = (struct net *)seq->private;
2216 "Basic info: size of leaf:"
2217 " %Zd bytes, size of tnode: %Zd bytes.\n",
2218 LEAF_SIZE, TNODE_SIZE(0));
2220 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2221 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2222 struct fib_table *tb;
2224 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2225 struct trie *t = (struct trie *) tb->tb_data;
2226 struct trie_stat stat;
2231 fib_table_print(seq, tb);
2233 trie_collect_stats(t, &stat);
2234 trie_show_stats(seq, &stat);
2235 #ifdef CONFIG_IP_FIB_TRIE_STATS
2236 trie_show_usage(seq, t->stats);
2244 static int fib_triestat_seq_open(struct inode *inode, struct file *file)
2246 return single_open_net(inode, file, fib_triestat_seq_show);
2249 static const struct file_operations fib_triestat_fops = {
2250 .owner = THIS_MODULE,
2251 .open = fib_triestat_seq_open,
2253 .llseek = seq_lseek,
2254 .release = single_release_net,
2257 static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2259 struct fib_trie_iter *iter = seq->private;
2260 struct net *net = seq_file_net(seq);
2264 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2265 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2266 struct fib_table *tb;
2268 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2269 struct key_vector *n;
2271 for (n = fib_trie_get_first(iter,
2272 (struct trie *) tb->tb_data);
2273 n; n = fib_trie_get_next(iter))
2284 static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2288 return fib_trie_get_idx(seq, *pos);
2291 static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2293 struct fib_trie_iter *iter = seq->private;
2294 struct net *net = seq_file_net(seq);
2295 struct fib_table *tb = iter->tb;
2296 struct hlist_node *tb_node;
2298 struct key_vector *n;
2301 /* next node in same table */
2302 n = fib_trie_get_next(iter);
2306 /* walk rest of this hash chain */
2307 h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2308 while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2309 tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2310 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2315 /* new hash chain */
2316 while (++h < FIB_TABLE_HASHSZ) {
2317 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2318 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2319 n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2331 static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2337 static void seq_indent(struct seq_file *seq, int n)
2343 static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2346 case RT_SCOPE_UNIVERSE: return "universe";
2347 case RT_SCOPE_SITE: return "site";
2348 case RT_SCOPE_LINK: return "link";
2349 case RT_SCOPE_HOST: return "host";
2350 case RT_SCOPE_NOWHERE: return "nowhere";
2352 snprintf(buf, len, "scope=%d", s);
2357 static const char *const rtn_type_names[__RTN_MAX] = {
2358 [RTN_UNSPEC] = "UNSPEC",
2359 [RTN_UNICAST] = "UNICAST",
2360 [RTN_LOCAL] = "LOCAL",
2361 [RTN_BROADCAST] = "BROADCAST",
2362 [RTN_ANYCAST] = "ANYCAST",
2363 [RTN_MULTICAST] = "MULTICAST",
2364 [RTN_BLACKHOLE] = "BLACKHOLE",
2365 [RTN_UNREACHABLE] = "UNREACHABLE",
2366 [RTN_PROHIBIT] = "PROHIBIT",
2367 [RTN_THROW] = "THROW",
2369 [RTN_XRESOLVE] = "XRESOLVE",
2372 static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2374 if (t < __RTN_MAX && rtn_type_names[t])
2375 return rtn_type_names[t];
2376 snprintf(buf, len, "type %u", t);
2380 /* Pretty print the trie */
2381 static int fib_trie_seq_show(struct seq_file *seq, void *v)
2383 const struct fib_trie_iter *iter = seq->private;
2384 struct key_vector *n = v;
2386 if (IS_TRIE(node_parent_rcu(n)))
2387 fib_table_print(seq, iter->tb);
2390 __be32 prf = htonl(n->key);
2392 seq_indent(seq, iter->depth-1);
2393 seq_printf(seq, " +-- %pI4/%zu %u %u %u\n",
2394 &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2395 tn_info(n)->full_children,
2396 tn_info(n)->empty_children);
2398 __be32 val = htonl(n->key);
2399 struct fib_alias *fa;
2401 seq_indent(seq, iter->depth);
2402 seq_printf(seq, " |-- %pI4\n", &val);
2404 hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2405 char buf1[32], buf2[32];
2407 seq_indent(seq, iter->depth + 1);
2408 seq_printf(seq, " /%zu %s %s",
2409 KEYLENGTH - fa->fa_slen,
2410 rtn_scope(buf1, sizeof(buf1),
2411 fa->fa_info->fib_scope),
2412 rtn_type(buf2, sizeof(buf2),
2415 seq_printf(seq, " tos=%d", fa->fa_tos);
2416 seq_putc(seq, '\n');
2423 static const struct seq_operations fib_trie_seq_ops = {
2424 .start = fib_trie_seq_start,
2425 .next = fib_trie_seq_next,
2426 .stop = fib_trie_seq_stop,
2427 .show = fib_trie_seq_show,
2430 static int fib_trie_seq_open(struct inode *inode, struct file *file)
2432 return seq_open_net(inode, file, &fib_trie_seq_ops,
2433 sizeof(struct fib_trie_iter));
2436 static const struct file_operations fib_trie_fops = {
2437 .owner = THIS_MODULE,
2438 .open = fib_trie_seq_open,
2440 .llseek = seq_lseek,
2441 .release = seq_release_net,
2444 struct fib_route_iter {
2445 struct seq_net_private p;
2446 struct fib_table *main_tb;
2447 struct key_vector *tnode;
2452 static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2455 struct fib_table *tb = iter->main_tb;
2456 struct key_vector *l, **tp = &iter->tnode;
2460 /* use cache location of next-to-find key */
2461 if (iter->pos > 0 && pos >= iter->pos) {
2465 t = (struct trie *)tb->tb_data;
2466 iter->tnode = t->kv;
2471 while ((l = leaf_walk_rcu(tp, key)) != NULL) {
2480 /* handle unlikely case of a key wrap */
2486 iter->key = key; /* remember it */
2488 iter->pos = 0; /* forget it */
2493 static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2496 struct fib_route_iter *iter = seq->private;
2497 struct fib_table *tb;
2502 tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2509 return fib_route_get_idx(iter, *pos);
2511 t = (struct trie *)tb->tb_data;
2512 iter->tnode = t->kv;
2516 return SEQ_START_TOKEN;
2519 static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2521 struct fib_route_iter *iter = seq->private;
2522 struct key_vector *l = NULL;
2523 t_key key = iter->key;
2527 /* only allow key of 0 for start of sequence */
2528 if ((v == SEQ_START_TOKEN) || key)
2529 l = leaf_walk_rcu(&iter->tnode, key);
2532 iter->key = l->key + 1;
2541 static void fib_route_seq_stop(struct seq_file *seq, void *v)
2547 static unsigned int fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2549 unsigned int flags = 0;
2551 if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2553 if (fi && fi->fib_nh->nh_gw)
2554 flags |= RTF_GATEWAY;
2555 if (mask == htonl(0xFFFFFFFF))
2562 * This outputs /proc/net/route.
2563 * The format of the file is not supposed to be changed
2564 * and needs to be same as fib_hash output to avoid breaking
2567 static int fib_route_seq_show(struct seq_file *seq, void *v)
2569 struct fib_route_iter *iter = seq->private;
2570 struct fib_table *tb = iter->main_tb;
2571 struct fib_alias *fa;
2572 struct key_vector *l = v;
2575 if (v == SEQ_START_TOKEN) {
2576 seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2577 "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2582 prefix = htonl(l->key);
2584 hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2585 const struct fib_info *fi = fa->fa_info;
2586 __be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2587 unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2589 if ((fa->fa_type == RTN_BROADCAST) ||
2590 (fa->fa_type == RTN_MULTICAST))
2593 if (fa->tb_id != tb->tb_id)
2596 seq_setwidth(seq, 127);
2600 "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2601 "%d\t%08X\t%d\t%u\t%u",
2602 fi->fib_dev ? fi->fib_dev->name : "*",
2604 fi->fib_nh->nh_gw, flags, 0, 0,
2608 fi->fib_advmss + 40 : 0),
2613 "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2614 "%d\t%08X\t%d\t%u\t%u",
2615 prefix, 0, flags, 0, 0, 0,
2624 static const struct seq_operations fib_route_seq_ops = {
2625 .start = fib_route_seq_start,
2626 .next = fib_route_seq_next,
2627 .stop = fib_route_seq_stop,
2628 .show = fib_route_seq_show,
2631 static int fib_route_seq_open(struct inode *inode, struct file *file)
2633 return seq_open_net(inode, file, &fib_route_seq_ops,
2634 sizeof(struct fib_route_iter));
2637 static const struct file_operations fib_route_fops = {
2638 .owner = THIS_MODULE,
2639 .open = fib_route_seq_open,
2641 .llseek = seq_lseek,
2642 .release = seq_release_net,
2645 int __net_init fib_proc_init(struct net *net)
2647 if (!proc_create("fib_trie", S_IRUGO, net->proc_net, &fib_trie_fops))
2650 if (!proc_create("fib_triestat", S_IRUGO, net->proc_net,
2651 &fib_triestat_fops))
2654 if (!proc_create("route", S_IRUGO, net->proc_net, &fib_route_fops))
2660 remove_proc_entry("fib_triestat", net->proc_net);
2662 remove_proc_entry("fib_trie", net->proc_net);
2667 void __net_exit fib_proc_exit(struct net *net)
2669 remove_proc_entry("fib_trie", net->proc_net);
2670 remove_proc_entry("fib_triestat", net->proc_net);
2671 remove_proc_entry("route", net->proc_net);
2674 #endif /* CONFIG_PROC_FS */