2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1988, 1989, 1993
5 * The Regents of the University of California. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * @(#)radix.c 8.4 (Berkeley) 11/2/94
37 * Routines to build and maintain radix trees for routing lookups.
44 #elif defined(__FreeBSD__)
47 __RCSID("$Revision: 2.23 $");
48 #ident "$Revision: 2.23 $"
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 static struct radix_mask *rn_mkfreelist;
57 static struct radix_node_head *mask_rnhead;
58 static char *addmask_key;
59 static const uint8_t normal_chars[] =
60 { 0, 0x80, 0xc0, 0xe0, 0xf0, 0xf8, 0xfc, 0xfe, 0xff};
61 static char *rn_zeros, *rn_ones;
63 #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);
68 static struct radix_node *rn_addmask(void *n_arg, int search, int skip);
69 static struct radix_node *rn_addroute(void *v_arg, void *n_arg,
70 struct radix_node_head *head, struct radix_node treenodes[2]);
71 static struct radix_node *rn_match(void *v_arg, struct radix_node_head *head);
74 * The data structure for the keys is a radix tree with one way
75 * branching removed. The index rn_b at an internal node n represents a bit
76 * position to be tested. The tree is arranged so that all descendants
77 * of a node n have keys whose bits all agree up to position rn_b - 1.
78 * (We say the index of n is rn_b.)
80 * There is at least one descendant which has a one bit at position rn_b,
81 * and at least one with a zero there.
83 * A route is determined by a pair of key and mask. We require that the
84 * bit-wise logical and of the key and mask to be the key.
85 * We define the index of a route to associated with the mask to be
86 * the first bit number in the mask where 0 occurs (with bit number 0
87 * representing the highest order bit).
89 * We say a mask is normal if every bit is 0, past the index of the mask.
90 * If a node n has a descendant (k, m) with index(m) == index(n) == rn_b,
91 * and m is a normal mask, then the route applies to every descendant of n.
92 * If the index(m) < rn_b, this implies the trailing last few bits of k
93 * before bit b are all 0, (and hence consequently true of every descendant
94 * of n), so the route applies to all descendants of the node as well.
96 * Similar logic shows that a non-normal mask m such that
97 * index(m) <= index(n) could potentially apply to many children of n.
98 * Thus, for each non-host route, we attach its mask to a list at an internal
99 * node as high in the tree as we can go.
101 * The present version of the code makes use of normal routes in short-
102 * circuiting an explicit mask and compare operation when testing whether
103 * a key satisfies a normal route, and also in remembering the unique leaf
104 * that governs a subtree.
107 static struct radix_node *
108 rn_search(void *v_arg,
109 struct radix_node *head)
111 struct radix_node *x;
114 for (x = head, v = v_arg; x->rn_b >= 0;) {
115 if (x->rn_bmask & v[x->rn_off])
123 static struct radix_node *
124 rn_search_m(void *v_arg,
125 struct radix_node *head,
128 struct radix_node *x;
129 caddr_t v = v_arg, m = m_arg;
131 for (x = head; x->rn_b >= 0;) {
132 if ((x->rn_bmask & m[x->rn_off]) &&
133 (x->rn_bmask & v[x->rn_off]))
142 rn_refines(void* m_arg, void *n_arg)
144 caddr_t m = m_arg, n = n_arg;
145 caddr_t lim, lim2 = lim = n + *(u_char *)n;
146 int longer = (*(u_char *)n++) - (int)(*(u_char *)m++);
147 int masks_are_equal = 1;
160 if (masks_are_equal && (longer < 0))
161 for (lim2 = m - longer; m < lim2; )
164 return (!masks_are_equal);
167 static struct radix_node *
168 rn_lookup(void *v_arg, void *m_arg, struct radix_node_head *head)
170 struct radix_node *x;
174 if ((x = rn_addmask(m_arg, 1,
175 head->rnh_treetop->rn_off)) == NULL)
179 x = rn_match(v_arg, head);
181 while (x && x->rn_mask != netmask)
188 rn_satisfies_leaf(char *trial,
189 struct radix_node *leaf,
192 char *cp = trial, *cp2 = leaf->rn_key, *cp3 = leaf->rn_mask;
194 int length = min(*(u_char *)cp, *(u_char *)cp2);
199 length = min(length, *(u_char *)cp3);
200 cplim = cp + length; cp3 += skip; cp2 += skip;
201 for (cp += skip; cp < cplim; cp++, cp2++, cp3++)
202 if ((*cp ^ *cp2) & *cp3)
207 static struct radix_node *
208 rn_match(void *v_arg,
209 struct radix_node_head *head)
212 struct radix_node *t = head->rnh_treetop, *x;
215 struct radix_node *saved_t, *top = t;
216 int off = t->rn_off, vlen = *(u_char *)cp, matched_off;
220 * Open code rn_search(v, top) to avoid overhead of extra
223 for (; t->rn_b >= 0; ) {
224 if (t->rn_bmask & cp[t->rn_off])
230 * See if we match exactly as a host destination
231 * or at least learn how many bits match, for normal mask finesse.
233 * It doesn't hurt us to limit how many bytes to check
234 * to the length of the mask, since if it matches we had a genuine
235 * match and the leaf we have is the most specific one anyway;
236 * if it didn't match with a shorter length it would fail
237 * with a long one. This wins big for class B&C netmasks which
238 * are probably the most common case...
241 vlen = *(u_char *)t->rn_mask;
242 cp += off; cp2 = t->rn_key + off; cplim = v + vlen;
243 for (; cp < cplim; cp++, cp2++)
247 * This extra grot is in case we are explicitly asked
248 * to look up the default. Ugh!
251 * In this case, we have a complete match of the key. Unless
252 * the node is one of the roots, we are finished.
253 * If it is the zeros root, then take what we have, preferring
255 * If it is the ones root, then pretend the target key was followed
256 * by a byte of zeros.
258 if (!(t->rn_flags & RNF_ROOT))
259 return t; /* not a root */
260 if (t->rn_dupedkey) {
262 return t; /* have some real data */
265 return t; /* not the ones root */
266 b = 0; /* fake a zero after 255.255.255.255 */
269 test = (*cp ^ *cp2) & 0xff; /* find first bit that differs */
270 for (b = 7; (test >>= 1) > 0;)
273 matched_off = cp - v;
274 b += matched_off << 3;
277 * If there is a host route in a duped-key chain, it will be first.
279 if ((saved_t = t)->rn_mask == 0)
281 for (; t; t = t->rn_dupedkey) {
283 * Even if we don't match exactly as a host,
284 * we may match if the leaf we wound up at is
287 if (t->rn_flags & RNF_NORMAL) {
290 } else if (rn_satisfies_leaf(v, t, matched_off)) {
295 /* start searching up the tree */
297 struct radix_mask *m;
299 if ((m = t->rn_mklist)) {
301 * If non-contiguous masks ever become important
302 * we can restore the masking and open coding of
303 * the search and satisfaction test and put the
304 * calculation of "off" back before the "do".
307 if (m->rm_flags & RNF_NORMAL) {
311 off = min(t->rn_off, matched_off);
312 x = rn_search_m(v, t, m->rm_mask);
313 while (x && x->rn_mask != m->rm_mask)
315 if (x && rn_satisfies_leaf(v, x, off))
318 } while ((m = m->rm_mklist));
326 struct radix_node *rn_clist;
331 static struct radix_node *
332 rn_newpair(void *v, int b, struct radix_node nodes[2])
334 struct radix_node *tt = nodes, *t = tt + 1;
335 t->rn_b = b; t->rn_bmask = 0x80 >> (b & 7);
336 t->rn_l = tt; t->rn_off = b >> 3;
337 tt->rn_b = -1; tt->rn_key = (caddr_t)v; tt->rn_p = t;
338 tt->rn_flags = t->rn_flags = RNF_ACTIVE;
340 tt->rn_info = rn_nodenum++; t->rn_info = rn_nodenum++;
341 tt->rn_twin = t; tt->rn_ybro = rn_clist; rn_clist = tt;
346 static struct radix_node *
347 rn_insert(void* v_arg,
348 struct radix_node_head *head,
350 struct radix_node nodes[2])
353 struct radix_node *top = head->rnh_treetop;
354 int head_off = top->rn_off, vlen = (int)*((u_char *)v);
355 struct radix_node *t = rn_search(v_arg, top);
356 caddr_t cp = v + head_off;
358 struct radix_node *tt;
361 * Find first bit at which v and t->rn_key differ
364 caddr_t cp2 = t->rn_key + head_off;
366 caddr_t cplim = v + vlen;
371 /* handle adding 255.255.255.255 */
372 if (!(t->rn_flags & RNF_ROOT) || *(cp2-1) == 0) {
378 cmp_res = (cp[-1] ^ cp2[-1]) & 0xff;
379 for (b = (cp - v) << 3; cmp_res; b--)
383 struct radix_node *p, *x = top;
387 if (cp[x->rn_off] & x->rn_bmask)
390 } while ((unsigned)b > (unsigned)x->rn_b);
393 log(LOG_DEBUG, "rn_insert: Going In:\n"), traverse(p);
395 t = rn_newpair(v_arg, b, nodes); tt = t->rn_l;
396 if ((cp[p->rn_off] & p->rn_bmask) == 0)
400 x->rn_p = t; t->rn_p = p; /* frees x, p as temp vars below */
401 if ((cp[t->rn_off] & t->rn_bmask) == 0) {
404 t->rn_r = tt; t->rn_l = x;
408 log(LOG_DEBUG, "rn_insert: Coming Out:\n"), traverse(p);
414 static struct radix_node *
415 rn_addmask(void *n_arg, int search, int skip)
417 caddr_t netmask = (caddr_t)n_arg;
418 struct radix_node *x;
421 int maskduplicated, m0, isnormal;
422 struct radix_node *saved_x;
423 static int last_zeroed = 0;
425 if ((mlen = *(u_char *)netmask) > max_keylen)
430 return (mask_rnhead->rnh_nodes);
432 Bcopy(rn_ones + 1, addmask_key + 1, skip - 1);
433 if ((m0 = mlen) > skip)
434 Bcopy(netmask + skip, addmask_key + skip, mlen - skip);
436 * Trim trailing zeroes.
438 for (cp = addmask_key + mlen; (cp > addmask_key) && cp[-1] == 0;)
440 mlen = cp - addmask_key;
442 if (m0 >= last_zeroed)
444 return (mask_rnhead->rnh_nodes);
446 if (m0 < last_zeroed)
447 Bzero(addmask_key + m0, last_zeroed - m0);
448 *addmask_key = last_zeroed = mlen;
449 x = rn_search(addmask_key, rn_masktop);
450 if (Bcmp(addmask_key, x->rn_key, mlen) != 0)
454 x = (struct radix_node *)rtmalloc(max_keylen + 2*sizeof(*x),
457 Bzero(x, max_keylen + 2 * sizeof (*x));
458 netmask = cp = (caddr_t)(x + 2);
459 Bcopy(addmask_key, cp, mlen);
460 x = rn_insert(cp, mask_rnhead, &maskduplicated, x);
461 if (maskduplicated) {
462 log(LOG_ERR, "rn_addmask: mask impossibly already in tree");
467 * Calculate index of mask, and check for normalcy.
469 cplim = netmask + mlen; isnormal = 1;
470 for (cp = netmask + skip; (cp < cplim) && *(u_char *)cp == 0xff;)
473 for (j = 0x80; (j & *cp) != 0; j >>= 1)
475 if (*cp != normal_chars[b] || cp != (cplim - 1))
478 b += (cp - netmask) << 3;
481 x->rn_flags |= RNF_NORMAL;
485 static int /* XXX: arbitrary ordering for non-contiguous masks */
486 rn_lexobetter(void *m_arg, void *n_arg)
488 u_char *mp = m_arg, *np = n_arg, *lim;
491 return 1; /* not really, but need to check longer one first */
493 for (lim = mp + *mp; mp < lim;)
499 static struct radix_mask *
500 rn_new_radix_mask(struct radix_node *tt,
501 struct radix_mask *next)
503 struct radix_mask *m;
507 log(LOG_ERR, "Mask for route not entered\n");
512 m->rm_flags = tt->rn_flags;
513 if (tt->rn_flags & RNF_NORMAL)
516 m->rm_mask = tt->rn_mask;
522 static struct radix_node *
523 rn_addroute(void *v_arg,
525 struct radix_node_head *head,
526 struct radix_node treenodes[2])
528 caddr_t v = (caddr_t)v_arg, netmask = (caddr_t)n_arg;
529 struct radix_node *t, *x = NULL, *tt;
530 struct radix_node *saved_tt, *top = head->rnh_treetop;
531 short b = 0, b_leaf = 0;
534 struct radix_mask *m, **mp;
537 * In dealing with non-contiguous masks, there may be
538 * many different routes which have the same mask.
539 * We will find it useful to have a unique pointer to
540 * the mask to speed avoiding duplicate references at
541 * nodes and possibly save time in calculating indices.
544 if ((x = rn_addmask(netmask, 0, top->rn_off)) == NULL)
551 * Deal with duplicated keys: attach node to previous instance
553 saved_tt = tt = rn_insert(v, head, &keyduplicated, treenodes);
555 for (t = tt; tt; t = tt, tt = tt->rn_dupedkey) {
556 if (tt->rn_mask == netmask)
560 ((b_leaf < tt->rn_b) || /* index(netmask) > node */
561 rn_refines(netmask, tt->rn_mask) ||
562 rn_lexobetter(netmask, tt->rn_mask))))
566 * If the mask is not duplicated, we wouldn't
567 * find it among possible duplicate key entries
568 * anyway, so the above test doesn't hurt.
570 * We sort the masks for a duplicated key the same way as
571 * in a masklist -- most specific to least specific.
572 * This may require the unfortunate nuisance of relocating
573 * the head of the list.
575 if (tt == saved_tt) {
576 struct radix_node *xx = x;
577 /* link in at head of list */
578 (tt = treenodes)->rn_dupedkey = t;
579 tt->rn_flags = t->rn_flags;
580 tt->rn_p = x = t->rn_p;
581 if (x->rn_l == t) x->rn_l = tt; else x->rn_r = tt;
582 saved_tt = tt; x = xx;
584 (tt = treenodes)->rn_dupedkey = t->rn_dupedkey;
588 t=tt+1; tt->rn_info = rn_nodenum++; t->rn_info = rn_nodenum++;
589 tt->rn_twin = t; tt->rn_ybro = rn_clist; rn_clist = tt;
591 tt->rn_key = (caddr_t) v;
593 tt->rn_flags = RNF_ACTIVE;
599 tt->rn_mask = netmask;
601 tt->rn_flags |= x->rn_flags & RNF_NORMAL;
606 b_leaf = -1 - t->rn_b;
607 if (t->rn_r == saved_tt) x = t->rn_l; else x = t->rn_r;
608 /* Promote general routes from below */
610 for (mp = &t->rn_mklist; x; x = x->rn_dupedkey)
611 if (x->rn_mask && (x->rn_b >= b_leaf) && x->rn_mklist == 0) {
612 if ((*mp = m = rn_new_radix_mask(x, 0)))
615 } else if (x->rn_mklist) {
617 * Skip over masks whose index is > that of new node
619 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist)
620 if (m->rm_b >= b_leaf)
622 t->rn_mklist = m; *mp = NULL;
625 /* Add new route to highest possible ancestor's list */
626 if ((netmask == 0) || (b > t->rn_b ))
627 return tt; /* can't lift at all */
632 } while (b <= t->rn_b && x != top);
634 * Search through routes associated with node to
635 * insert new route according to index.
636 * Need same criteria as when sorting dupedkeys to avoid
637 * double loop on deletion.
639 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist) {
640 if (m->rm_b < b_leaf)
642 if (m->rm_b > b_leaf)
644 if (m->rm_flags & RNF_NORMAL) {
645 mmask = m->rm_leaf->rn_mask;
646 if (tt->rn_flags & RNF_NORMAL) {
648 "Non-unique normal route, mask not entered");
653 if (mmask == netmask) {
658 if (rn_refines(netmask, mmask) || rn_lexobetter(netmask, mmask))
661 *mp = rn_new_radix_mask(tt, *mp);
665 static struct radix_node *
666 rn_delete(void *v_arg,
668 struct radix_node_head *head)
670 struct radix_node *t, *p, *x, *tt;
671 struct radix_mask *m, *saved_m, **mp;
672 struct radix_node *dupedkey, *saved_tt, *top;
674 int b, head_off, vlen;
677 netmask = netmask_arg;
678 x = head->rnh_treetop;
679 tt = rn_search(v, x);
680 head_off = x->rn_off;
685 Bcmp(v + head_off, tt->rn_key + head_off, vlen - head_off))
688 * Delete our route from mask lists.
691 if ((x = rn_addmask(netmask, 1, head_off)) == NULL)
694 while (tt->rn_mask != netmask)
695 if ((tt = tt->rn_dupedkey) == NULL)
698 if (tt->rn_mask == 0 || (saved_m = m = tt->rn_mklist) == NULL)
700 if (tt->rn_flags & RNF_NORMAL) {
701 if (m->rm_leaf != tt || m->rm_refs > 0) {
702 log(LOG_ERR, "rn_delete: inconsistent annotation\n");
703 return 0; /* dangling ref could cause disaster */
706 if (m->rm_mask != tt->rn_mask) {
707 log(LOG_ERR, "rn_delete: inconsistent annotation\n");
710 if (--m->rm_refs >= 0)
716 goto on1; /* Wasn't lifted at all */
720 } while (b <= t->rn_b && x != top);
721 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist)
728 log(LOG_ERR, "rn_delete: couldn't find our annotation\n");
729 if (tt->rn_flags & RNF_NORMAL)
730 return (0); /* Dangling ref to us */
734 * Eliminate us from tree
736 if (tt->rn_flags & RNF_ROOT)
739 /* Get us out of the creation list */
740 for (t = rn_clist; t && t->rn_ybro != tt; t = t->rn_ybro) {}
741 if (t) t->rn_ybro = tt->rn_ybro;
744 if ((dupedkey = saved_tt->rn_dupedkey)) {
745 if (tt == saved_tt) {
746 x = dupedkey; x->rn_p = t;
747 if (t->rn_l == tt) t->rn_l = x; else t->rn_r = x;
749 for (x = p = saved_tt; p && p->rn_dupedkey != tt;)
751 if (p) p->rn_dupedkey = tt->rn_dupedkey;
752 else log(LOG_ERR, "rn_delete: couldn't find us\n");
755 if (t->rn_flags & RNF_ACTIVE) {
757 *++x = *t; p = t->rn_p;
759 b = t->rn_info; *++x = *t; t->rn_info = b; p = t->rn_p;
761 if (p->rn_l == t) p->rn_l = x; else p->rn_r = x;
762 x->rn_l->rn_p = x; x->rn_r->rn_p = x;
766 if (t->rn_l == tt) x = t->rn_r; else x = t->rn_l;
768 if (p->rn_r == t) p->rn_r = x; else p->rn_l = x;
771 * Demote routes attached to us.
775 for (mp = &x->rn_mklist; (m = *mp);)
779 /* If there are any key,mask pairs in a sibling
780 duped-key chain, some subset will appear sorted
781 in the same order attached to our mklist */
782 for (m = t->rn_mklist; m && x; x = x->rn_dupedkey)
783 if (m == x->rn_mklist) {
784 struct radix_mask *mm = m->rm_mklist;
786 if (--(m->rm_refs) < 0)
791 syslog(LOG_ERR, "%s 0x%lx at 0x%lx\n",
792 "rn_delete: Orphaned Mask",
798 * We may be holding an active internal node in the tree.
805 b = t->rn_info; *t = *x; t->rn_info = b;
807 t->rn_l->rn_p = t; t->rn_r->rn_p = t;
809 if (p->rn_l == x) p->rn_l = t; else p->rn_r = t;
812 tt->rn_flags &= ~RNF_ACTIVE;
813 tt[1].rn_flags &= ~RNF_ACTIVE;
818 rn_walktree(struct radix_node_head *h,
819 int (*f)(struct radix_node *, struct walkarg *),
823 struct radix_node *base, *next;
824 struct radix_node *rn = h->rnh_treetop;
826 * This gets complicated because we may delete the node
827 * while applying the function f to it, so we need to calculate
828 * the successor node in advance.
830 /* First time through node, go left */
831 while (rn->rn_b >= 0)
835 /* If at right child go back up, otherwise, go right */
836 while (rn->rn_p->rn_r == rn && (rn->rn_flags & RNF_ROOT) == 0)
838 /* Find the next *leaf* since next node might vanish, too */
839 for (rn = rn->rn_p->rn_r; rn->rn_b >= 0;)
843 while ((rn = base)) {
844 base = rn->rn_dupedkey;
845 if (!(rn->rn_flags & RNF_ROOT) && (error = (*f)(rn, w)))
849 if (rn->rn_flags & RNF_ROOT)
856 rn_inithead(struct radix_node_head **head, int off)
858 struct radix_node_head *rnh;
859 struct radix_node *t, *tt, *ttt;
862 rnh = (struct radix_node_head *)rtmalloc(sizeof(*rnh), "rn_inithead");
863 Bzero(rnh, sizeof (*rnh));
865 t = rn_newpair(rn_zeros, off, rnh->rnh_nodes);
866 ttt = rnh->rnh_nodes + 2;
870 tt->rn_flags = t->rn_flags = RNF_ROOT | RNF_ACTIVE;
873 ttt->rn_key = rn_ones;
874 rnh->rnh_addaddr = rn_addroute;
875 rnh->rnh_deladdr = rn_delete;
876 rnh->rnh_matchaddr = rn_match;
877 rnh->rnh_lookup = rn_lookup;
878 rnh->rnh_walktree = rn_walktree;
879 rnh->rnh_treetop = t;
887 if (max_keylen == 0) {
888 printf("rn_init: radix functions require max_keylen be set\n");
891 rn_zeros = (char *)rtmalloc(3 * max_keylen, "rn_init");
892 Bzero(rn_zeros, 3 * max_keylen);
893 rn_ones = cp = rn_zeros + max_keylen;
894 addmask_key = cplim = rn_ones + max_keylen;
897 if (rn_inithead(&mask_rnhead, 0) == 0)