2 * Copyright (c) 1988, 1989, 1993
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgment:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * @(#)radix.c 8.4 (Berkeley) 11/2/94
35 * $FreeBSD: src/sbin/routed/radix.c,v 1.5 1999/09/05 17:49:11 peter Exp $
36 * $DragonFly: src/sbin/routed/radix.c,v 1.3 2004/02/04 17:40:01 joerg Exp $
40 * Routines to build and maintain radix trees for routing lookups.
45 #if !defined(sgi) && !defined(__NetBSD__)
46 static char sccsid[] __attribute__((unused)) = "@(#)rdisc.c 8.1 (Berkeley) x/y/95";
47 #elif defined(__NetBSD__)
51 #define log(x, msg) syslog(x, msg)
52 #define panic(s) {log(LOG_ERR,s); exit(1);}
53 #define min(a,b) (((a)<(b))?(a):(b))
56 struct radix_mask *rn_mkfreelist;
57 struct radix_node_head *mask_rnhead;
58 static char *addmask_key;
59 static char normal_chars[] = {0, 0x80, 0xc0, 0xe0, 0xf0, 0xf8, 0xfc, 0xfe, -1};
60 static char *rn_zeros, *rn_ones;
62 #define rn_masktop (mask_rnhead->rnh_treetop)
64 #define Bcmp(a, b, l) (l == 0 ? 0 \
65 : memcmp((caddr_t)(a), (caddr_t)(b), (size_t)l))
67 static int rn_satisfies_leaf(char *, struct radix_node *, int);
70 * The data structure for the keys is a radix tree with one way
71 * branching removed. The index rn_b at an internal node n represents a bit
72 * position to be tested. The tree is arranged so that all descendants
73 * of a node n have keys whose bits all agree up to position rn_b - 1.
74 * (We say the index of n is rn_b.)
76 * There is at least one descendant which has a one bit at position rn_b,
77 * and at least one with a zero there.
79 * A route is determined by a pair of key and mask. We require that the
80 * bit-wise logical and of the key and mask to be the key.
81 * We define the index of a route to associated with the mask to be
82 * the first bit number in the mask where 0 occurs (with bit number 0
83 * representing the highest order bit).
85 * We say a mask is normal if every bit is 0, past the index of the mask.
86 * If a node n has a descendant (k, m) with index(m) == index(n) == rn_b,
87 * and m is a normal mask, then the route applies to every descendant of n.
88 * If the index(m) < rn_b, this implies the trailing last few bits of k
89 * before bit b are all 0, (and hence consequently true of every descendant
90 * of n), so the route applies to all descendants of the node as well.
92 * Similar logic shows that a non-normal mask m such that
93 * index(m) <= index(n) could potentially apply to many children of n.
94 * Thus, for each non-host route, we attach its mask to a list at an internal
95 * node as high in the tree as we can go.
97 * The present version of the code makes use of normal routes in short-
98 * circuiting an explict mask and compare operation when testing whether
99 * a key satisfies a normal route, and also in remembering the unique leaf
100 * that governs a subtree.
104 rn_search(void *v_arg,
105 struct radix_node *head)
107 struct radix_node *x;
110 for (x = head, v = v_arg; x->rn_b >= 0;) {
111 if (x->rn_bmask & v[x->rn_off])
120 rn_search_m(void *v_arg,
121 struct radix_node *head,
124 struct radix_node *x;
125 caddr_t v = v_arg, m = m_arg;
127 for (x = head; x->rn_b >= 0;) {
128 if ((x->rn_bmask & m[x->rn_off]) &&
129 (x->rn_bmask & v[x->rn_off]))
138 rn_refines(void* m_arg, void *n_arg)
140 caddr_t m = m_arg, n = n_arg;
141 caddr_t lim, lim2 = lim = n + *(u_char *)n;
142 int longer = (*(u_char *)n++) - (int)(*(u_char *)m++);
143 int masks_are_equal = 1;
156 if (masks_are_equal && (longer < 0))
157 for (lim2 = m - longer; m < lim2; )
160 return (!masks_are_equal);
164 rn_lookup(void *v_arg, void *m_arg, struct radix_node_head *head)
166 struct radix_node *x;
170 if ((x = rn_addmask(m_arg, 1, head->rnh_treetop->rn_off)) == 0)
174 x = rn_match(v_arg, head);
176 while (x && x->rn_mask != netmask)
183 rn_satisfies_leaf(char *trial,
184 struct radix_node *leaf,
187 char *cp = trial, *cp2 = leaf->rn_key, *cp3 = leaf->rn_mask;
189 int length = min(*(u_char *)cp, *(u_char *)cp2);
194 length = min(length, *(u_char *)cp3);
195 cplim = cp + length; cp3 += skip; cp2 += skip;
196 for (cp += skip; cp < cplim; cp++, cp2++, cp3++)
197 if ((*cp ^ *cp2) & *cp3)
203 rn_match(void *v_arg,
204 struct radix_node_head *head)
207 struct radix_node *t = head->rnh_treetop, *x;
210 struct radix_node *saved_t, *top = t;
211 int off = t->rn_off, vlen = *(u_char *)cp, matched_off;
215 * Open code rn_search(v, top) to avoid overhead of extra
218 for (; t->rn_b >= 0; ) {
219 if (t->rn_bmask & cp[t->rn_off])
225 * See if we match exactly as a host destination
226 * or at least learn how many bits match, for normal mask finesse.
228 * It doesn't hurt us to limit how many bytes to check
229 * to the length of the mask, since if it matches we had a genuine
230 * match and the leaf we have is the most specific one anyway;
231 * if it didn't match with a shorter length it would fail
232 * with a long one. This wins big for class B&C netmasks which
233 * are probably the most common case...
236 vlen = *(u_char *)t->rn_mask;
237 cp += off; cp2 = t->rn_key + off; cplim = v + vlen;
238 for (; cp < cplim; cp++, cp2++)
242 * This extra grot is in case we are explicitly asked
243 * to look up the default. Ugh!
246 * In this case, we have a complete match of the key. Unless
247 * the node is one of the roots, we are finished.
248 * If it is the zeros root, then take what we have, prefering
250 * If it is the ones root, then pretend the target key was followed
251 * by a byte of zeros.
253 if (!(t->rn_flags & RNF_ROOT))
254 return t; /* not a root */
255 if (t->rn_dupedkey) {
257 return t; /* have some real data */
260 return t; /* not the ones root */
261 b = 0; /* fake a zero after 255.255.255.255 */
264 test = (*cp ^ *cp2) & 0xff; /* find first bit that differs */
265 for (b = 7; (test >>= 1) > 0;)
268 matched_off = cp - v;
269 b += matched_off << 3;
272 * If there is a host route in a duped-key chain, it will be first.
274 if ((saved_t = t)->rn_mask == 0)
276 for (; t; t = t->rn_dupedkey) {
278 * Even if we don't match exactly as a host,
279 * we may match if the leaf we wound up at is
282 if (t->rn_flags & RNF_NORMAL) {
285 } else if (rn_satisfies_leaf(v, t, matched_off)) {
290 /* start searching up the tree */
292 struct radix_mask *m;
294 if ((m = t->rn_mklist)) {
296 * If non-contiguous masks ever become important
297 * we can restore the masking and open coding of
298 * the search and satisfaction test and put the
299 * calculation of "off" back before the "do".
302 if (m->rm_flags & RNF_NORMAL) {
306 off = min(t->rn_off, matched_off);
307 x = rn_search_m(v, t, m->rm_mask);
308 while (x && x->rn_mask != m->rm_mask)
310 if (x && rn_satisfies_leaf(v, x, off))
313 } while ((m = m->rm_mklist));
321 struct radix_node *rn_clist;
327 rn_newpair(void *v, int b, struct radix_node nodes[2])
329 struct radix_node *tt = nodes, *t = tt + 1;
330 t->rn_b = b; t->rn_bmask = 0x80 >> (b & 7);
331 t->rn_l = tt; t->rn_off = b >> 3;
332 tt->rn_b = -1; tt->rn_key = (caddr_t)v; tt->rn_p = t;
333 tt->rn_flags = t->rn_flags = RNF_ACTIVE;
335 tt->rn_info = rn_nodenum++; t->rn_info = rn_nodenum++;
336 tt->rn_twin = t; tt->rn_ybro = rn_clist; rn_clist = tt;
342 rn_insert(void* v_arg,
343 struct radix_node_head *head,
345 struct radix_node nodes[2])
348 struct radix_node *top = head->rnh_treetop;
349 int head_off = top->rn_off, vlen = (int)*((u_char *)v);
350 struct radix_node *t = rn_search(v_arg, top);
351 caddr_t cp = v + head_off;
353 struct radix_node *tt;
356 * Find first bit at which v and t->rn_key differ
359 caddr_t cp2 = t->rn_key + head_off;
361 caddr_t cplim = v + vlen;
366 /* handle adding 255.255.255.255 */
367 if (!(t->rn_flags & RNF_ROOT) || *(cp2-1) == 0) {
373 cmp_res = (cp[-1] ^ cp2[-1]) & 0xff;
374 for (b = (cp - v) << 3; cmp_res; b--)
378 struct radix_node *p, *x = top;
382 if (cp[x->rn_off] & x->rn_bmask)
385 } while ((unsigned)b > (unsigned)x->rn_b);
388 log(LOG_DEBUG, "rn_insert: Going In:\n"), traverse(p);
390 t = rn_newpair(v_arg, b, nodes); tt = t->rn_l;
391 if ((cp[p->rn_off] & p->rn_bmask) == 0)
395 x->rn_p = t; t->rn_p = p; /* frees x, p as temp vars below */
396 if ((cp[t->rn_off] & t->rn_bmask) == 0) {
399 t->rn_r = tt; t->rn_l = x;
403 log(LOG_DEBUG, "rn_insert: Coming Out:\n"), traverse(p);
410 rn_addmask(void *n_arg, int search, int skip)
412 caddr_t netmask = (caddr_t)n_arg;
413 struct radix_node *x;
416 int maskduplicated, m0, isnormal;
417 struct radix_node *saved_x;
418 static int last_zeroed = 0;
420 if ((mlen = *(u_char *)netmask) > max_keylen)
425 return (mask_rnhead->rnh_nodes);
427 Bcopy(rn_ones + 1, addmask_key + 1, skip - 1);
428 if ((m0 = mlen) > skip)
429 Bcopy(netmask + skip, addmask_key + skip, mlen - skip);
431 * Trim trailing zeroes.
433 for (cp = addmask_key + mlen; (cp > addmask_key) && cp[-1] == 0;)
435 mlen = cp - addmask_key;
437 if (m0 >= last_zeroed)
439 return (mask_rnhead->rnh_nodes);
441 if (m0 < last_zeroed)
442 Bzero(addmask_key + m0, last_zeroed - m0);
443 *addmask_key = last_zeroed = mlen;
444 x = rn_search(addmask_key, rn_masktop);
445 if (Bcmp(addmask_key, x->rn_key, mlen) != 0)
449 x = (struct radix_node *)rtmalloc(max_keylen + 2*sizeof(*x),
452 Bzero(x, max_keylen + 2 * sizeof (*x));
453 netmask = cp = (caddr_t)(x + 2);
454 Bcopy(addmask_key, cp, mlen);
455 x = rn_insert(cp, mask_rnhead, &maskduplicated, x);
456 if (maskduplicated) {
457 log(LOG_ERR, "rn_addmask: mask impossibly already in tree");
462 * Calculate index of mask, and check for normalcy.
464 cplim = netmask + mlen; isnormal = 1;
465 for (cp = netmask + skip; (cp < cplim) && *(u_char *)cp == 0xff;)
468 for (j = 0x80; (j & *cp) != 0; j >>= 1)
470 if (*cp != normal_chars[b] || cp != (cplim - 1))
473 b += (cp - netmask) << 3;
476 x->rn_flags |= RNF_NORMAL;
480 static int /* XXX: arbitrary ordering for non-contiguous masks */
481 rn_lexobetter(void *m_arg, void *n_arg)
483 u_char *mp = m_arg, *np = n_arg, *lim;
486 return 1; /* not really, but need to check longer one first */
488 for (lim = mp + *mp; mp < lim;)
494 static struct radix_mask *
495 rn_new_radix_mask(struct radix_node *tt,
496 struct radix_mask *next)
498 struct radix_mask *m;
502 log(LOG_ERR, "Mask for route not entered\n");
507 m->rm_flags = tt->rn_flags;
508 if (tt->rn_flags & RNF_NORMAL)
511 m->rm_mask = tt->rn_mask;
518 rn_addroute(void *v_arg,
520 struct radix_node_head *head,
521 struct radix_node treenodes[2])
523 caddr_t v = (caddr_t)v_arg, netmask = (caddr_t)n_arg;
524 struct radix_node *t, *x = 0, *tt;
525 struct radix_node *saved_tt, *top = head->rnh_treetop;
526 short b = 0, b_leaf = 0;
529 struct radix_mask *m, **mp;
532 * In dealing with non-contiguous masks, there may be
533 * many different routes which have the same mask.
534 * We will find it useful to have a unique pointer to
535 * the mask to speed avoiding duplicate references at
536 * nodes and possibly save time in calculating indices.
539 if ((x = rn_addmask(netmask, 0, top->rn_off)) == 0)
546 * Deal with duplicated keys: attach node to previous instance
548 saved_tt = tt = rn_insert(v, head, &keyduplicated, treenodes);
550 for (t = tt; tt; t = tt, tt = tt->rn_dupedkey) {
551 if (tt->rn_mask == netmask)
555 ((b_leaf < tt->rn_b) || /* index(netmask) > node */
556 rn_refines(netmask, tt->rn_mask) ||
557 rn_lexobetter(netmask, tt->rn_mask))))
561 * If the mask is not duplicated, we wouldn't
562 * find it among possible duplicate key entries
563 * anyway, so the above test doesn't hurt.
565 * We sort the masks for a duplicated key the same way as
566 * in a masklist -- most specific to least specific.
567 * This may require the unfortunate nuisance of relocating
568 * the head of the list.
570 if (tt == saved_tt) {
571 struct radix_node *xx = x;
572 /* link in at head of list */
573 (tt = treenodes)->rn_dupedkey = t;
574 tt->rn_flags = t->rn_flags;
575 tt->rn_p = x = t->rn_p;
576 if (x->rn_l == t) x->rn_l = tt; else x->rn_r = tt;
577 saved_tt = tt; x = xx;
579 (tt = treenodes)->rn_dupedkey = t->rn_dupedkey;
583 t=tt+1; tt->rn_info = rn_nodenum++; t->rn_info = rn_nodenum++;
584 tt->rn_twin = t; tt->rn_ybro = rn_clist; rn_clist = tt;
586 tt->rn_key = (caddr_t) v;
588 tt->rn_flags = RNF_ACTIVE;
594 tt->rn_mask = netmask;
596 tt->rn_flags |= x->rn_flags & RNF_NORMAL;
601 b_leaf = -1 - t->rn_b;
602 if (t->rn_r == saved_tt) x = t->rn_l; else x = t->rn_r;
603 /* Promote general routes from below */
605 for (mp = &t->rn_mklist; x; x = x->rn_dupedkey)
606 if (x->rn_mask && (x->rn_b >= b_leaf) && x->rn_mklist == 0) {
607 if ((*mp = m = rn_new_radix_mask(x, 0)))
610 } else if (x->rn_mklist) {
612 * Skip over masks whose index is > that of new node
614 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist)
615 if (m->rm_b >= b_leaf)
617 t->rn_mklist = m; *mp = 0;
620 /* Add new route to highest possible ancestor's list */
621 if ((netmask == 0) || (b > t->rn_b ))
622 return tt; /* can't lift at all */
627 } while (b <= t->rn_b && x != top);
629 * Search through routes associated with node to
630 * insert new route according to index.
631 * Need same criteria as when sorting dupedkeys to avoid
632 * double loop on deletion.
634 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist) {
635 if (m->rm_b < b_leaf)
637 if (m->rm_b > b_leaf)
639 if (m->rm_flags & RNF_NORMAL) {
640 mmask = m->rm_leaf->rn_mask;
641 if (tt->rn_flags & RNF_NORMAL) {
643 "Non-unique normal route, mask not entered");
648 if (mmask == netmask) {
653 if (rn_refines(netmask, mmask) || rn_lexobetter(netmask, mmask))
656 *mp = rn_new_radix_mask(tt, *mp);
661 rn_delete(void *v_arg,
663 struct radix_node_head *head)
665 struct radix_node *t, *p, *x, *tt;
666 struct radix_mask *m, *saved_m, **mp;
667 struct radix_node *dupedkey, *saved_tt, *top;
669 int b, head_off, vlen;
672 netmask = netmask_arg;
673 x = head->rnh_treetop;
674 tt = rn_search(v, x);
675 head_off = x->rn_off;
680 Bcmp(v + head_off, tt->rn_key + head_off, vlen - head_off))
683 * Delete our route from mask lists.
686 if ((x = rn_addmask(netmask, 1, head_off)) == 0)
689 while (tt->rn_mask != netmask)
690 if ((tt = tt->rn_dupedkey) == 0)
693 if (tt->rn_mask == 0 || (saved_m = m = tt->rn_mklist) == 0)
695 if (tt->rn_flags & RNF_NORMAL) {
696 if (m->rm_leaf != tt || m->rm_refs > 0) {
697 log(LOG_ERR, "rn_delete: inconsistent annotation\n");
698 return 0; /* dangling ref could cause disaster */
701 if (m->rm_mask != tt->rn_mask) {
702 log(LOG_ERR, "rn_delete: inconsistent annotation\n");
705 if (--m->rm_refs >= 0)
711 goto on1; /* Wasn't lifted at all */
715 } while (b <= t->rn_b && x != top);
716 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist)
723 log(LOG_ERR, "rn_delete: couldn't find our annotation\n");
724 if (tt->rn_flags & RNF_NORMAL)
725 return (0); /* Dangling ref to us */
729 * Eliminate us from tree
731 if (tt->rn_flags & RNF_ROOT)
734 /* Get us out of the creation list */
735 for (t = rn_clist; t && t->rn_ybro != tt; t = t->rn_ybro) {}
736 if (t) t->rn_ybro = tt->rn_ybro;
739 if ((dupedkey = saved_tt->rn_dupedkey)) {
740 if (tt == saved_tt) {
741 x = dupedkey; x->rn_p = t;
742 if (t->rn_l == tt) t->rn_l = x; else t->rn_r = x;
744 for (x = p = saved_tt; p && p->rn_dupedkey != tt;)
746 if (p) p->rn_dupedkey = tt->rn_dupedkey;
747 else log(LOG_ERR, "rn_delete: couldn't find us\n");
750 if (t->rn_flags & RNF_ACTIVE) {
752 *++x = *t; p = t->rn_p;
754 b = t->rn_info; *++x = *t; t->rn_info = b; p = t->rn_p;
756 if (p->rn_l == t) p->rn_l = x; else p->rn_r = x;
757 x->rn_l->rn_p = x; x->rn_r->rn_p = x;
761 if (t->rn_l == tt) x = t->rn_r; else x = t->rn_l;
763 if (p->rn_r == t) p->rn_r = x; else p->rn_l = x;
766 * Demote routes attached to us.
770 for (mp = &x->rn_mklist; (m = *mp);)
774 /* If there are any key,mask pairs in a sibling
775 duped-key chain, some subset will appear sorted
776 in the same order attached to our mklist */
777 for (m = t->rn_mklist; m && x; x = x->rn_dupedkey)
778 if (m == x->rn_mklist) {
779 struct radix_mask *mm = m->rm_mklist;
781 if (--(m->rm_refs) < 0)
786 syslog(LOG_ERR, "%s 0x%lx at 0x%lx\n",
787 "rn_delete: Orphaned Mask",
793 * We may be holding an active internal node in the tree.
800 b = t->rn_info; *t = *x; t->rn_info = b;
802 t->rn_l->rn_p = t; t->rn_r->rn_p = t;
804 if (p->rn_l == x) p->rn_l = t; else p->rn_r = t;
807 tt->rn_flags &= ~RNF_ACTIVE;
808 tt[1].rn_flags &= ~RNF_ACTIVE;
813 rn_walktree(struct radix_node_head *h,
814 int (*f)(struct radix_node *, struct walkarg *),
818 struct radix_node *base, *next;
819 struct radix_node *rn = h->rnh_treetop;
821 * This gets complicated because we may delete the node
822 * while applying the function f to it, so we need to calculate
823 * the successor node in advance.
825 /* First time through node, go left */
826 while (rn->rn_b >= 0)
830 /* If at right child go back up, otherwise, go right */
831 while (rn->rn_p->rn_r == rn && (rn->rn_flags & RNF_ROOT) == 0)
833 /* Find the next *leaf* since next node might vanish, too */
834 for (rn = rn->rn_p->rn_r; rn->rn_b >= 0;)
838 while ((rn = base)) {
839 base = rn->rn_dupedkey;
840 if (!(rn->rn_flags & RNF_ROOT) && (error = (*f)(rn, w)))
844 if (rn->rn_flags & RNF_ROOT)
851 rn_inithead(struct radix_node_head **head, int off)
853 struct radix_node_head *rnh;
854 struct radix_node *t, *tt, *ttt;
857 rnh = (struct radix_node_head *)rtmalloc(sizeof(*rnh), "rn_inithead");
858 Bzero(rnh, sizeof (*rnh));
860 t = rn_newpair(rn_zeros, off, rnh->rnh_nodes);
861 ttt = rnh->rnh_nodes + 2;
865 tt->rn_flags = t->rn_flags = RNF_ROOT | RNF_ACTIVE;
868 ttt->rn_key = rn_ones;
869 rnh->rnh_addaddr = rn_addroute;
870 rnh->rnh_deladdr = rn_delete;
871 rnh->rnh_matchaddr = rn_match;
872 rnh->rnh_lookup = rn_lookup;
873 rnh->rnh_walktree = rn_walktree;
874 rnh->rnh_treetop = t;
882 if (max_keylen == 0) {
883 printf("rn_init: radix functions require max_keylen be set\n");
886 rn_zeros = (char *)rtmalloc(3 * max_keylen, "rn_init");
887 Bzero(rn_zeros, 3 * max_keylen);
888 rn_ones = cp = rn_zeros + max_keylen;
889 addmask_key = cplim = rn_ones + max_keylen;
892 if (rn_inithead(&mask_rnhead, 0) == 0)