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. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * @(#)radix.c 8.4 (Berkeley) 11/2/94
31 * $FreeBSD: src/sbin/routed/radix.c,v 1.5 1999/09/05 17:49:11 peter Exp $
35 * Routines to build and maintain radix trees for routing lookups.
40 #define log(x, msg) syslog(x, msg)
41 #define panic(s) {log(LOG_ERR,s); exit(1);}
42 #define min(a,b) (((a)<(b))?(a):(b))
45 struct radix_mask *rn_mkfreelist;
46 struct radix_node_head *mask_rnhead;
47 static char *addmask_key;
48 static char normal_chars[] = {0, 0x80, 0xc0, 0xe0, 0xf0, 0xf8, 0xfc, 0xfe, -1};
49 static char *rn_zeros, *rn_ones;
51 #define rn_masktop (mask_rnhead->rnh_treetop)
53 #define Bcmp(a, b, l) (l == 0 ? 0 \
54 : memcmp((caddr_t)(a), (caddr_t)(b), (size_t)l))
56 static int rn_satisfies_leaf(char *, struct radix_node *, int);
59 * The data structure for the keys is a radix tree with one way
60 * branching removed. The index rn_b at an internal node n represents a bit
61 * position to be tested. The tree is arranged so that all descendants
62 * of a node n have keys whose bits all agree up to position rn_b - 1.
63 * (We say the index of n is rn_b.)
65 * There is at least one descendant which has a one bit at position rn_b,
66 * and at least one with a zero there.
68 * A route is determined by a pair of key and mask. We require that the
69 * bit-wise logical and of the key and mask to be the key.
70 * We define the index of a route to associated with the mask to be
71 * the first bit number in the mask where 0 occurs (with bit number 0
72 * representing the highest order bit).
74 * We say a mask is normal if every bit is 0, past the index of the mask.
75 * If a node n has a descendant (k, m) with index(m) == index(n) == rn_b,
76 * and m is a normal mask, then the route applies to every descendant of n.
77 * If the index(m) < rn_b, this implies the trailing last few bits of k
78 * before bit b are all 0, (and hence consequently true of every descendant
79 * of n), so the route applies to all descendants of the node as well.
81 * Similar logic shows that a non-normal mask m such that
82 * index(m) <= index(n) could potentially apply to many children of n.
83 * Thus, for each non-host route, we attach its mask to a list at an internal
84 * node as high in the tree as we can go.
86 * The present version of the code makes use of normal routes in short-
87 * circuiting an explict mask and compare operation when testing whether
88 * a key satisfies a normal route, and also in remembering the unique leaf
89 * that governs a subtree.
93 rn_search(void *v_arg,
94 struct radix_node *head)
99 for (x = head, v = v_arg; x->rn_b >= 0;) {
100 if (x->rn_bmask & v[x->rn_off])
109 rn_search_m(void *v_arg,
110 struct radix_node *head,
113 struct radix_node *x;
114 caddr_t v = v_arg, m = m_arg;
116 for (x = head; x->rn_b >= 0;) {
117 if ((x->rn_bmask & m[x->rn_off]) &&
118 (x->rn_bmask & v[x->rn_off]))
127 rn_refines(void* m_arg, void *n_arg)
129 caddr_t m = m_arg, n = n_arg;
130 caddr_t lim, lim2 = lim = n + *(u_char *)n;
131 int longer = (*(u_char *)n++) - (int)(*(u_char *)m++);
132 int masks_are_equal = 1;
145 if (masks_are_equal && (longer < 0))
146 for (lim2 = m - longer; m < lim2; )
149 return (!masks_are_equal);
153 rn_lookup(void *v_arg, void *m_arg, struct radix_node_head *head)
155 struct radix_node *x;
159 if ((x = rn_addmask(m_arg, 1, head->rnh_treetop->rn_off)) == NULL)
163 x = rn_match(v_arg, head);
165 while (x && x->rn_mask != netmask)
172 rn_satisfies_leaf(char *trial,
173 struct radix_node *leaf,
176 char *cp = trial, *cp2 = leaf->rn_key, *cp3 = leaf->rn_mask;
178 int length = min(*(u_char *)cp, *(u_char *)cp2);
183 length = min(length, *(u_char *)cp3);
184 cplim = cp + length; cp3 += skip; cp2 += skip;
185 for (cp += skip; cp < cplim; cp++, cp2++, cp3++)
186 if ((*cp ^ *cp2) & *cp3)
192 rn_match(void *v_arg,
193 struct radix_node_head *head)
196 struct radix_node *t = head->rnh_treetop, *x;
199 struct radix_node *saved_t, *top = t;
200 int off = t->rn_off, vlen = *(u_char *)cp, matched_off;
204 * Open code rn_search(v, top) to avoid overhead of extra
207 for (; t->rn_b >= 0; ) {
208 if (t->rn_bmask & cp[t->rn_off])
214 * See if we match exactly as a host destination
215 * or at least learn how many bits match, for normal mask finesse.
217 * It doesn't hurt us to limit how many bytes to check
218 * to the length of the mask, since if it matches we had a genuine
219 * match and the leaf we have is the most specific one anyway;
220 * if it didn't match with a shorter length it would fail
221 * with a long one. This wins big for class B&C netmasks which
222 * are probably the most common case...
225 vlen = *(u_char *)t->rn_mask;
226 cp += off; cp2 = t->rn_key + off; cplim = v + vlen;
227 for (; cp < cplim; cp++, cp2++)
231 * This extra grot is in case we are explicitly asked
232 * to look up the default. Ugh!
235 * In this case, we have a complete match of the key. Unless
236 * the node is one of the roots, we are finished.
237 * If it is the zeros root, then take what we have, prefering
239 * If it is the ones root, then pretend the target key was followed
240 * by a byte of zeros.
242 if (!(t->rn_flags & RNF_ROOT))
243 return t; /* not a root */
244 if (t->rn_dupedkey) {
246 return t; /* have some real data */
249 return t; /* not the ones root */
250 b = 0; /* fake a zero after 255.255.255.255 */
253 test = (*cp ^ *cp2) & 0xff; /* find first bit that differs */
254 for (b = 7; (test >>= 1) > 0;)
257 matched_off = cp - v;
258 b += matched_off << 3;
261 * If there is a host route in a duped-key chain, it will be first.
263 if ((saved_t = t)->rn_mask == 0)
265 for (; t; t = t->rn_dupedkey) {
267 * Even if we don't match exactly as a host,
268 * we may match if the leaf we wound up at is
271 if (t->rn_flags & RNF_NORMAL) {
274 } else if (rn_satisfies_leaf(v, t, matched_off)) {
279 /* start searching up the tree */
281 struct radix_mask *m;
283 if ((m = t->rn_mklist)) {
285 * If non-contiguous masks ever become important
286 * we can restore the masking and open coding of
287 * the search and satisfaction test and put the
288 * calculation of "off" back before the "do".
291 if (m->rm_flags & RNF_NORMAL) {
295 off = min(t->rn_off, matched_off);
296 x = rn_search_m(v, t, m->rm_mask);
297 while (x && x->rn_mask != m->rm_mask)
299 if (x && rn_satisfies_leaf(v, x, off))
302 } while ((m = m->rm_mklist));
310 struct radix_node *rn_clist;
316 rn_newpair(void *v, int b, struct radix_node nodes[2])
318 struct radix_node *tt = nodes, *t = tt + 1;
319 t->rn_b = b; t->rn_bmask = 0x80 >> (b & 7);
320 t->rn_l = tt; t->rn_off = b >> 3;
321 tt->rn_b = -1; tt->rn_key = (caddr_t)v; tt->rn_p = t;
322 tt->rn_flags = t->rn_flags = RNF_ACTIVE;
324 tt->rn_info = rn_nodenum++; t->rn_info = rn_nodenum++;
325 tt->rn_twin = t; tt->rn_ybro = rn_clist; rn_clist = tt;
331 rn_insert(void* v_arg,
332 struct radix_node_head *head,
334 struct radix_node nodes[2])
337 struct radix_node *top = head->rnh_treetop;
338 int head_off = top->rn_off, vlen = (int)*((u_char *)v);
339 struct radix_node *t = rn_search(v_arg, top);
340 caddr_t cp = v + head_off;
342 struct radix_node *tt;
345 * Find first bit at which v and t->rn_key differ
348 caddr_t cp2 = t->rn_key + head_off;
350 caddr_t cplim = v + vlen;
355 /* handle adding 255.255.255.255 */
356 if (!(t->rn_flags & RNF_ROOT) || *(cp2-1) == 0) {
362 cmp_res = (cp[-1] ^ cp2[-1]) & 0xff;
363 for (b = (cp - v) << 3; cmp_res; b--)
367 struct radix_node *p, *x = top;
371 if (cp[x->rn_off] & x->rn_bmask)
374 } while ((unsigned)b > (unsigned)x->rn_b);
377 log(LOG_DEBUG, "rn_insert: Going In:\n"), traverse(p);
379 t = rn_newpair(v_arg, b, nodes); tt = t->rn_l;
380 if ((cp[p->rn_off] & p->rn_bmask) == 0)
384 x->rn_p = t; t->rn_p = p; /* frees x, p as temp vars below */
385 if ((cp[t->rn_off] & t->rn_bmask) == 0) {
388 t->rn_r = tt; t->rn_l = x;
392 log(LOG_DEBUG, "rn_insert: Coming Out:\n"), traverse(p);
399 rn_addmask(void *n_arg, int search, int skip)
401 caddr_t netmask = (caddr_t)n_arg;
402 struct radix_node *x;
405 int maskduplicated, m0, isnormal;
406 struct radix_node *saved_x;
407 static int last_zeroed = 0;
409 if ((mlen = *(u_char *)netmask) > max_keylen)
414 return (mask_rnhead->rnh_nodes);
416 Bcopy(rn_ones + 1, addmask_key + 1, skip - 1);
417 if ((m0 = mlen) > skip)
418 Bcopy(netmask + skip, addmask_key + skip, mlen - skip);
420 * Trim trailing zeroes.
422 for (cp = addmask_key + mlen; (cp > addmask_key) && cp[-1] == 0;)
424 mlen = cp - addmask_key;
426 if (m0 >= last_zeroed)
428 return (mask_rnhead->rnh_nodes);
430 if (m0 < last_zeroed)
431 Bzero(addmask_key + m0, last_zeroed - m0);
432 *addmask_key = last_zeroed = mlen;
433 x = rn_search(addmask_key, rn_masktop);
434 if (Bcmp(addmask_key, x->rn_key, mlen) != 0)
438 x = (struct radix_node *)rtmalloc(max_keylen + 2*sizeof(*x),
441 Bzero(x, max_keylen + 2 * sizeof (*x));
442 netmask = cp = (caddr_t)(x + 2);
443 Bcopy(addmask_key, cp, mlen);
444 x = rn_insert(cp, mask_rnhead, &maskduplicated, x);
445 if (maskduplicated) {
446 log(LOG_ERR, "rn_addmask: mask impossibly already in tree");
451 * Calculate index of mask, and check for normalcy.
453 cplim = netmask + mlen; isnormal = 1;
454 for (cp = netmask + skip; (cp < cplim) && *(u_char *)cp == 0xff;)
457 for (j = 0x80; (j & *cp) != 0; j >>= 1)
459 if (*cp != normal_chars[b] || cp != (cplim - 1))
462 b += (cp - netmask) << 3;
465 x->rn_flags |= RNF_NORMAL;
469 static int /* XXX: arbitrary ordering for non-contiguous masks */
470 rn_lexobetter(void *m_arg, void *n_arg)
472 u_char *mp = m_arg, *np = n_arg, *lim;
475 return 1; /* not really, but need to check longer one first */
477 for (lim = mp + *mp; mp < lim;)
483 static struct radix_mask *
484 rn_new_radix_mask(struct radix_node *tt,
485 struct radix_mask *next)
487 struct radix_mask *m;
491 log(LOG_ERR, "Mask for route not entered\n");
496 m->rm_flags = tt->rn_flags;
497 if (tt->rn_flags & RNF_NORMAL)
500 m->rm_mask = tt->rn_mask;
507 rn_addroute(void *v_arg,
509 struct radix_node_head *head,
510 struct radix_node treenodes[2])
512 caddr_t v = (caddr_t)v_arg, netmask = (caddr_t)n_arg;
513 struct radix_node *t, *x = NULL, *tt;
514 struct radix_node *saved_tt, *top = head->rnh_treetop;
515 short b = 0, b_leaf = 0;
518 struct radix_mask *m, **mp;
521 * In dealing with non-contiguous masks, there may be
522 * many different routes which have the same mask.
523 * We will find it useful to have a unique pointer to
524 * the mask to speed avoiding duplicate references at
525 * nodes and possibly save time in calculating indices.
528 if ((x = rn_addmask(netmask, 0, top->rn_off)) == NULL)
535 * Deal with duplicated keys: attach node to previous instance
537 saved_tt = tt = rn_insert(v, head, &keyduplicated, treenodes);
539 for (t = tt; tt; t = tt, tt = tt->rn_dupedkey) {
540 if (tt->rn_mask == netmask)
544 ((b_leaf < tt->rn_b) || /* index(netmask) > node */
545 rn_refines(netmask, tt->rn_mask) ||
546 rn_lexobetter(netmask, tt->rn_mask))))
550 * If the mask is not duplicated, we wouldn't
551 * find it among possible duplicate key entries
552 * anyway, so the above test doesn't hurt.
554 * We sort the masks for a duplicated key the same way as
555 * in a masklist -- most specific to least specific.
556 * This may require the unfortunate nuisance of relocating
557 * the head of the list.
559 if (tt == saved_tt) {
560 struct radix_node *xx = x;
561 /* link in at head of list */
562 (tt = treenodes)->rn_dupedkey = t;
563 tt->rn_flags = t->rn_flags;
564 tt->rn_p = x = t->rn_p;
565 if (x->rn_l == t) x->rn_l = tt; else x->rn_r = tt;
566 saved_tt = tt; x = xx;
568 (tt = treenodes)->rn_dupedkey = t->rn_dupedkey;
572 t=tt+1; tt->rn_info = rn_nodenum++; t->rn_info = rn_nodenum++;
573 tt->rn_twin = t; tt->rn_ybro = rn_clist; rn_clist = tt;
575 tt->rn_key = (caddr_t) v;
577 tt->rn_flags = RNF_ACTIVE;
583 tt->rn_mask = netmask;
585 tt->rn_flags |= x->rn_flags & RNF_NORMAL;
590 b_leaf = -1 - t->rn_b;
591 if (t->rn_r == saved_tt) x = t->rn_l; else x = t->rn_r;
592 /* Promote general routes from below */
594 for (mp = &t->rn_mklist; x; x = x->rn_dupedkey)
595 if (x->rn_mask && (x->rn_b >= b_leaf) && x->rn_mklist == 0) {
596 if ((*mp = m = rn_new_radix_mask(x, 0)))
599 } else if (x->rn_mklist) {
601 * Skip over masks whose index is > that of new node
603 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist)
604 if (m->rm_b >= b_leaf)
606 t->rn_mklist = m; *mp = NULL;
609 /* Add new route to highest possible ancestor's list */
610 if ((netmask == 0) || (b > t->rn_b ))
611 return tt; /* can't lift at all */
616 } while (b <= t->rn_b && x != top);
618 * Search through routes associated with node to
619 * insert new route according to index.
620 * Need same criteria as when sorting dupedkeys to avoid
621 * double loop on deletion.
623 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist) {
624 if (m->rm_b < b_leaf)
626 if (m->rm_b > b_leaf)
628 if (m->rm_flags & RNF_NORMAL) {
629 mmask = m->rm_leaf->rn_mask;
630 if (tt->rn_flags & RNF_NORMAL) {
632 "Non-unique normal route, mask not entered");
637 if (mmask == netmask) {
642 if (rn_refines(netmask, mmask) || rn_lexobetter(netmask, mmask))
645 *mp = rn_new_radix_mask(tt, *mp);
650 rn_delete(void *v_arg,
652 struct radix_node_head *head)
654 struct radix_node *t, *p, *x, *tt;
655 struct radix_mask *m, *saved_m, **mp;
656 struct radix_node *dupedkey, *saved_tt, *top;
658 int b, head_off, vlen;
661 netmask = netmask_arg;
662 x = head->rnh_treetop;
663 tt = rn_search(v, x);
664 head_off = x->rn_off;
669 Bcmp(v + head_off, tt->rn_key + head_off, vlen - head_off))
672 * Delete our route from mask lists.
675 if ((x = rn_addmask(netmask, 1, head_off)) == NULL)
678 while (tt->rn_mask != netmask)
679 if ((tt = tt->rn_dupedkey) == NULL)
682 if (tt->rn_mask == 0 || (saved_m = m = tt->rn_mklist) == NULL)
684 if (tt->rn_flags & RNF_NORMAL) {
685 if (m->rm_leaf != tt || m->rm_refs > 0) {
686 log(LOG_ERR, "rn_delete: inconsistent annotation\n");
687 return 0; /* dangling ref could cause disaster */
690 if (m->rm_mask != tt->rn_mask) {
691 log(LOG_ERR, "rn_delete: inconsistent annotation\n");
694 if (--m->rm_refs >= 0)
700 goto on1; /* Wasn't lifted at all */
704 } while (b <= t->rn_b && x != top);
705 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist)
712 log(LOG_ERR, "rn_delete: couldn't find our annotation\n");
713 if (tt->rn_flags & RNF_NORMAL)
714 return (0); /* Dangling ref to us */
718 * Eliminate us from tree
720 if (tt->rn_flags & RNF_ROOT)
723 /* Get us out of the creation list */
724 for (t = rn_clist; t && t->rn_ybro != tt; t = t->rn_ybro) {}
725 if (t) t->rn_ybro = tt->rn_ybro;
728 if ((dupedkey = saved_tt->rn_dupedkey)) {
729 if (tt == saved_tt) {
730 x = dupedkey; x->rn_p = t;
731 if (t->rn_l == tt) t->rn_l = x; else t->rn_r = x;
733 for (x = p = saved_tt; p && p->rn_dupedkey != tt;)
735 if (p) p->rn_dupedkey = tt->rn_dupedkey;
736 else log(LOG_ERR, "rn_delete: couldn't find us\n");
739 if (t->rn_flags & RNF_ACTIVE) {
741 *++x = *t; p = t->rn_p;
743 b = t->rn_info; *++x = *t; t->rn_info = b; p = t->rn_p;
745 if (p->rn_l == t) p->rn_l = x; else p->rn_r = x;
746 x->rn_l->rn_p = x; x->rn_r->rn_p = x;
750 if (t->rn_l == tt) x = t->rn_r; else x = t->rn_l;
752 if (p->rn_r == t) p->rn_r = x; else p->rn_l = x;
755 * Demote routes attached to us.
759 for (mp = &x->rn_mklist; (m = *mp);)
763 /* If there are any key,mask pairs in a sibling
764 duped-key chain, some subset will appear sorted
765 in the same order attached to our mklist */
766 for (m = t->rn_mklist; m && x; x = x->rn_dupedkey)
767 if (m == x->rn_mklist) {
768 struct radix_mask *mm = m->rm_mklist;
770 if (--(m->rm_refs) < 0)
775 syslog(LOG_ERR, "%s 0x%lx at 0x%lx\n",
776 "rn_delete: Orphaned Mask",
782 * We may be holding an active internal node in the tree.
789 b = t->rn_info; *t = *x; t->rn_info = b;
791 t->rn_l->rn_p = t; t->rn_r->rn_p = t;
793 if (p->rn_l == x) p->rn_l = t; else p->rn_r = t;
796 tt->rn_flags &= ~RNF_ACTIVE;
797 tt[1].rn_flags &= ~RNF_ACTIVE;
802 rn_walktree(struct radix_node_head *h,
803 int (*f)(struct radix_node *, struct walkarg *),
807 struct radix_node *base, *next;
808 struct radix_node *rn = h->rnh_treetop;
810 * This gets complicated because we may delete the node
811 * while applying the function f to it, so we need to calculate
812 * the successor node in advance.
814 /* First time through node, go left */
815 while (rn->rn_b >= 0)
819 /* If at right child go back up, otherwise, go right */
820 while (rn->rn_p->rn_r == rn && (rn->rn_flags & RNF_ROOT) == 0)
822 /* Find the next *leaf* since next node might vanish, too */
823 for (rn = rn->rn_p->rn_r; rn->rn_b >= 0;)
827 while ((rn = base)) {
828 base = rn->rn_dupedkey;
829 if (!(rn->rn_flags & RNF_ROOT) && (error = (*f)(rn, w)))
833 if (rn->rn_flags & RNF_ROOT)
840 rn_inithead(struct radix_node_head **head, int off)
842 struct radix_node_head *rnh;
843 struct radix_node *t, *tt, *ttt;
846 rnh = (struct radix_node_head *)rtmalloc(sizeof(*rnh), "rn_inithead");
847 Bzero(rnh, sizeof (*rnh));
849 t = rn_newpair(rn_zeros, off, rnh->rnh_nodes);
850 ttt = rnh->rnh_nodes + 2;
854 tt->rn_flags = t->rn_flags = RNF_ROOT | RNF_ACTIVE;
857 ttt->rn_key = rn_ones;
858 rnh->rnh_addaddr = rn_addroute;
859 rnh->rnh_deladdr = rn_delete;
860 rnh->rnh_matchaddr = rn_match;
861 rnh->rnh_lookup = rn_lookup;
862 rnh->rnh_walktree = rn_walktree;
863 rnh->rnh_treetop = t;
871 if (max_keylen == 0) {
872 printf("rn_init: radix functions require max_keylen be set\n");
875 rn_zeros = (char *)rtmalloc(3 * max_keylen, "rn_init");
876 Bzero(rn_zeros, 3 * max_keylen);
877 rn_ones = cp = rn_zeros + max_keylen;
878 addmask_key = cplim = rn_ones + max_keylen;
881 if (rn_inithead(&mask_rnhead, 0) == 0)