radix.c revision 37908
1/*
2 * Copyright (c) 1988, 1989, 1993
3 *	The Regents of the University of California.  All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
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 acknowledgement:
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.
20 *
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
31 * SUCH DAMAGE.
32 *
33 */
34
35#ifndef lint
36#if 0
37static char sccsid[] = "@(#)radix.c	8.4 (Berkeley) 11/2/94";
38#endif
39static const char rcsid[] =
40	"$Id$";
41#endif /* not lint */
42
43/*
44 * Routines to build and maintain radix trees for routing lookups.
45 */
46
47#include "defs.h"
48
49#define log(x, msg) syslog(x, msg)
50#define panic(s) {log(LOG_ERR,s); exit(1);}
51#define min(a,b) (((a)<(b))?(a):(b))
52
53int	max_keylen;
54struct radix_mask *rn_mkfreelist;
55struct radix_node_head *mask_rnhead;
56static char *addmask_key;
57static char normal_chars[] = {0, 0x80, 0xc0, 0xe0, 0xf0, 0xf8, 0xfc, 0xfe, -1};
58static char *rn_zeros, *rn_ones;
59
60#define rn_masktop (mask_rnhead->rnh_treetop)
61#undef Bcmp
62#define Bcmp(a, b, l) (l == 0 ? 0 : bcmp((caddr_t)(a), (caddr_t)(b), (u_long)l))
63
64static int rn_satsifies_leaf(char *, struct radix_node *, int);
65
66/*
67 * The data structure for the keys is a radix tree with one way
68 * branching removed.  The index rn_b at an internal node n represents a bit
69 * position to be tested.  The tree is arranged so that all descendants
70 * of a node n have keys whose bits all agree up to position rn_b - 1.
71 * (We say the index of n is rn_b.)
72 *
73 * There is at least one descendant which has a one bit at position rn_b,
74 * and at least one with a zero there.
75 *
76 * A route is determined by a pair of key and mask.  We require that the
77 * bit-wise logical and of the key and mask to be the key.
78 * We define the index of a route to associated with the mask to be
79 * the first bit number in the mask where 0 occurs (with bit number 0
80 * representing the highest order bit).
81 *
82 * We say a mask is normal if every bit is 0, past the index of the mask.
83 * If a node n has a descendant (k, m) with index(m) == index(n) == rn_b,
84 * and m is a normal mask, then the route applies to every descendant of n.
85 * If the index(m) < rn_b, this implies the trailing last few bits of k
86 * before bit b are all 0, (and hence consequently true of every descendant
87 * of n), so the route applies to all descendants of the node as well.
88 *
89 * Similar logic shows that a non-normal mask m such that
90 * index(m) <= index(n) could potentially apply to many children of n.
91 * Thus, for each non-host route, we attach its mask to a list at an internal
92 * node as high in the tree as we can go.
93 *
94 * The present version of the code makes use of normal routes in short-
95 * circuiting an explict mask and compare operation when testing whether
96 * a key satisfies a normal route, and also in remembering the unique leaf
97 * that governs a subtree.
98 */
99
100struct radix_node *
101rn_search(void *v_arg,
102	  struct radix_node *head)
103{
104	register struct radix_node *x;
105	register caddr_t v;
106
107	for (x = head, v = v_arg; x->rn_b >= 0;) {
108		if (x->rn_bmask & v[x->rn_off])
109			x = x->rn_r;
110		else
111			x = x->rn_l;
112	}
113	return (x);
114}
115
116struct radix_node *
117rn_search_m(void *v_arg,
118	    struct radix_node *head,
119	    void *m_arg)
120{
121	register struct radix_node *x;
122	register caddr_t v = v_arg, m = m_arg;
123
124	for (x = head; x->rn_b >= 0;) {
125		if ((x->rn_bmask & m[x->rn_off]) &&
126		    (x->rn_bmask & v[x->rn_off]))
127			x = x->rn_r;
128		else
129			x = x->rn_l;
130	}
131	return x;
132}
133
134int
135rn_refines(void* m_arg, void *n_arg)
136{
137	register caddr_t m = m_arg, n = n_arg;
138	register caddr_t lim, lim2 = lim = n + *(u_char *)n;
139	int longer = (*(u_char *)n++) - (int)(*(u_char *)m++);
140	int masks_are_equal = 1;
141
142	if (longer > 0)
143		lim -= longer;
144	while (n < lim) {
145		if (*n & ~(*m))
146			return 0;
147		if (*n++ != *m++)
148			masks_are_equal = 0;
149	}
150	while (n < lim2)
151		if (*n++)
152			return 0;
153	if (masks_are_equal && (longer < 0))
154		for (lim2 = m - longer; m < lim2; )
155			if (*m++)
156				return 1;
157	return (!masks_are_equal);
158}
159
160struct radix_node *
161rn_lookup(v_arg, m_arg, head)
162	void *v_arg, *m_arg;
163	struct radix_node_head *head;
164{
165	register struct radix_node *x;
166	caddr_t netmask = 0;
167
168	if (m_arg) {
169		if ((x = rn_addmask(m_arg, 1, head->rnh_treetop->rn_off)) == 0)
170			return (0);
171		netmask = x->rn_key;
172	}
173	x = rn_match(v_arg, head);
174	if (x && netmask) {
175		while (x && x->rn_mask != netmask)
176			x = x->rn_dupedkey;
177	}
178	return x;
179}
180
181static int
182rn_satsifies_leaf(char *trial,
183		  register struct radix_node *leaf,
184		  int skip)
185{
186	register char *cp = trial, *cp2 = leaf->rn_key, *cp3 = leaf->rn_mask;
187	char *cplim;
188	int length = min(*(u_char *)cp, *(u_char *)cp2);
189
190	if (cp3 == 0)
191		cp3 = rn_ones;
192	else
193		length = min(length, *(u_char *)cp3);
194	cplim = cp + length; cp3 += skip; cp2 += skip;
195	for (cp += skip; cp < cplim; cp++, cp2++, cp3++)
196		if ((*cp ^ *cp2) & *cp3)
197			return 0;
198	return 1;
199}
200
201struct radix_node *
202rn_match(void *v_arg,
203	 struct radix_node_head *head)
204{
205	caddr_t v = v_arg;
206	register struct radix_node *t = head->rnh_treetop, *x;
207	register caddr_t cp = v, cp2;
208	caddr_t cplim;
209	struct radix_node *saved_t, *top = t;
210	int off = t->rn_off, vlen = *(u_char *)cp, matched_off;
211	register int test, b, rn_b;
212
213	/*
214	 * Open code rn_search(v, top) to avoid overhead of extra
215	 * subroutine call.
216	 */
217	for (; t->rn_b >= 0; ) {
218		if (t->rn_bmask & cp[t->rn_off])
219			t = t->rn_r;
220		else
221			t = t->rn_l;
222	}
223	/*
224	 * See if we match exactly as a host destination
225	 * or at least learn how many bits match, for normal mask finesse.
226	 *
227	 * It doesn't hurt us to limit how many bytes to check
228	 * to the length of the mask, since if it matches we had a genuine
229	 * match and the leaf we have is the most specific one anyway;
230	 * if it didn't match with a shorter length it would fail
231	 * with a long one.  This wins big for class B&C netmasks which
232	 * are probably the most common case...
233	 */
234	if (t->rn_mask)
235		vlen = *(u_char *)t->rn_mask;
236	cp += off; cp2 = t->rn_key + off; cplim = v + vlen;
237	for (; cp < cplim; cp++, cp2++)
238		if (*cp != *cp2)
239			goto on1;
240	/*
241	 * This extra grot is in case we are explicitly asked
242	 * to look up the default.  Ugh!
243	 * Or 255.255.255.255
244	 *
245	 * In this case, we have a complete match of the key.  Unless
246	 * the node is one of the roots, we are finished.
247	 * If it is the zeros root, then take what we have, prefering
248	 * any real data.
249	 * If it is the ones root, then pretend the target key was followed
250	 * by a byte of zeros.
251	 */
252	if (!(t->rn_flags & RNF_ROOT))
253		return t;		/* not a root */
254	if (t->rn_dupedkey) {
255		t = t->rn_dupedkey;
256		return t;		/* have some real data */
257	}
258	if (*(cp-1) == 0)
259		return t;		/* not the ones root */
260	b = 0;				/* fake a zero after 255.255.255.255 */
261	goto on2;
262on1:
263	test = (*cp ^ *cp2) & 0xff; /* find first bit that differs */
264	for (b = 7; (test >>= 1) > 0;)
265		b--;
266on2:
267	matched_off = cp - v;
268	b += matched_off << 3;
269	rn_b = -1 - b;
270	/*
271	 * If there is a host route in a duped-key chain, it will be first.
272	 */
273	if ((saved_t = t)->rn_mask == 0)
274		t = t->rn_dupedkey;
275	for (; t; t = t->rn_dupedkey)
276		/*
277		 * Even if we don't match exactly as a host,
278		 * we may match if the leaf we wound up at is
279		 * a route to a net.
280		 */
281		if (t->rn_flags & RNF_NORMAL) {
282			if (rn_b <= t->rn_b)
283				return t;
284		} else if (rn_satsifies_leaf(v, t, matched_off))
285				return t;
286	t = saved_t;
287	/* start searching up the tree */
288	do {
289		register struct radix_mask *m;
290		t = t->rn_p;
291		if ((m = t->rn_mklist)) {
292			/*
293			 * If non-contiguous masks ever become important
294			 * we can restore the masking and open coding of
295			 * the search and satisfaction test and put the
296			 * calculation of "off" back before the "do".
297			 */
298			do {
299				if (m->rm_flags & RNF_NORMAL) {
300					if (rn_b <= m->rm_b)
301						return (m->rm_leaf);
302				} else {
303					off = min(t->rn_off, matched_off);
304					x = rn_search_m(v, t, m->rm_mask);
305					while (x && x->rn_mask != m->rm_mask)
306						x = x->rn_dupedkey;
307					if (x && rn_satsifies_leaf(v, x, off))
308						    return x;
309				}
310			} while ((m = m->rm_mklist));
311		}
312	} while (t != top);
313	return 0;
314}
315
316#ifdef RN_DEBUG
317int	rn_nodenum;
318struct	radix_node *rn_clist;
319int	rn_saveinfo;
320int	rn_debug =  1;
321#endif
322
323struct radix_node *
324rn_newpair(void *v, int b, struct radix_node nodes[2])
325{
326	register struct radix_node *tt = nodes, *t = tt + 1;
327	t->rn_b = b; t->rn_bmask = 0x80 >> (b & 7);
328	t->rn_l = tt; t->rn_off = b >> 3;
329	tt->rn_b = -1; tt->rn_key = (caddr_t)v; tt->rn_p = t;
330	tt->rn_flags = t->rn_flags = RNF_ACTIVE;
331#ifdef RN_DEBUG
332	tt->rn_info = rn_nodenum++; t->rn_info = rn_nodenum++;
333	tt->rn_twin = t; tt->rn_ybro = rn_clist; rn_clist = tt;
334#endif
335	return t;
336}
337
338struct radix_node *
339rn_insert(void* v_arg,
340	  struct radix_node_head *head,
341	  int *dupentry,
342	  struct radix_node nodes[2])
343{
344	caddr_t v = v_arg;
345	struct radix_node *top = head->rnh_treetop;
346	int head_off = top->rn_off, vlen = (int)*((u_char *)v);
347	register struct radix_node *t = rn_search(v_arg, top);
348	register caddr_t cp = v + head_off;
349	register int b;
350	struct radix_node *tt;
351
352	/*
353	 * Find first bit at which v and t->rn_key differ
354	 */
355    {
356	register caddr_t cp2 = t->rn_key + head_off;
357	register int cmp_res;
358	caddr_t cplim = v + vlen;
359
360	while (cp < cplim)
361		if (*cp2++ != *cp++)
362			goto on1;
363	/* handle adding 255.255.255.255 */
364	if (!(t->rn_flags & RNF_ROOT) || *(cp2-1) == 0) {
365		*dupentry = 1;
366		return t;
367	}
368on1:
369	*dupentry = 0;
370	cmp_res = (cp[-1] ^ cp2[-1]) & 0xff;
371	for (b = (cp - v) << 3; cmp_res; b--)
372		cmp_res >>= 1;
373    }
374    {
375	register struct radix_node *p, *x = top;
376	cp = v;
377	do {
378		p = x;
379		if (cp[x->rn_off] & x->rn_bmask)
380			x = x->rn_r;
381		else x = x->rn_l;
382	} while (b > (unsigned) x->rn_b); /* x->rn_b < b && x->rn_b >= 0 */
383#ifdef RN_DEBUG
384	if (rn_debug)
385		log(LOG_DEBUG, "rn_insert: Going In:\n"), traverse(p);
386#endif
387	t = rn_newpair(v_arg, b, nodes); tt = t->rn_l;
388	if ((cp[p->rn_off] & p->rn_bmask) == 0)
389		p->rn_l = t;
390	else
391		p->rn_r = t;
392	x->rn_p = t; t->rn_p = p; /* frees x, p as temp vars below */
393	if ((cp[t->rn_off] & t->rn_bmask) == 0) {
394		t->rn_r = x;
395	} else {
396		t->rn_r = tt; t->rn_l = x;
397	}
398#ifdef RN_DEBUG
399	if (rn_debug)
400		log(LOG_DEBUG, "rn_insert: Coming Out:\n"), traverse(p);
401#endif
402    }
403	return (tt);
404}
405
406struct radix_node *
407rn_addmask(void *n_arg, int search, int skip)
408{
409	caddr_t netmask = (caddr_t)n_arg;
410	register struct radix_node *x;
411	register caddr_t cp, cplim;
412	register int b = 0, mlen, j;
413	int maskduplicated, m0, isnormal;
414	struct radix_node *saved_x;
415	static int last_zeroed = 0;
416
417	if ((mlen = *(u_char *)netmask) > max_keylen)
418		mlen = max_keylen;
419	if (skip == 0)
420		skip = 1;
421	if (mlen <= skip)
422		return (mask_rnhead->rnh_nodes);
423	if (skip > 1)
424		Bcopy(rn_ones + 1, addmask_key + 1, skip - 1);
425	if ((m0 = mlen) > skip)
426		Bcopy(netmask + skip, addmask_key + skip, mlen - skip);
427	/*
428	 * Trim trailing zeroes.
429	 */
430	for (cp = addmask_key + mlen; (cp > addmask_key) && cp[-1] == 0;)
431		cp--;
432	mlen = cp - addmask_key;
433	if (mlen <= skip) {
434		if (m0 >= last_zeroed)
435			last_zeroed = mlen;
436		return (mask_rnhead->rnh_nodes);
437	}
438	if (m0 < last_zeroed)
439		Bzero(addmask_key + m0, last_zeroed - m0);
440	*addmask_key = last_zeroed = mlen;
441	x = rn_search(addmask_key, rn_masktop);
442	if (Bcmp(addmask_key, x->rn_key, mlen) != 0)
443		x = 0;
444	if (x || search)
445		return (x);
446	x = (struct radix_node *)
447			rtmalloc(max_keylen + 2 * sizeof (*x), "rn_addmask");
448	if ((saved_x = x) == 0)
449		return (0);
450	Bzero(x, max_keylen + 2 * sizeof (*x));
451	netmask = cp = (caddr_t)(x + 2);
452	Bcopy(addmask_key, cp, mlen);
453	x = rn_insert(cp, mask_rnhead, &maskduplicated, x);
454	if (maskduplicated) {
455		log(LOG_ERR, "rn_addmask: mask impossibly already in tree");
456		Free(saved_x);
457		return (x);
458	}
459	/*
460	 * Calculate index of mask, and check for normalcy.
461	 */
462	cplim = netmask + mlen; isnormal = 1;
463	for (cp = netmask + skip; (cp < cplim) && *(u_char *)cp == 0xff;)
464		cp++;
465	if (cp != cplim) {
466		for (j = 0x80; (j & *cp) != 0; j >>= 1)
467			b++;
468		if (*cp != normal_chars[b] || cp != (cplim - 1))
469			isnormal = 0;
470	}
471	b += (cp - netmask) << 3;
472	x->rn_b = -1 - b;
473	if (isnormal)
474		x->rn_flags |= RNF_NORMAL;
475	return (x);
476}
477
478static int	/* XXX: arbitrary ordering for non-contiguous masks */
479rn_lexobetter(void *m_arg, void *n_arg)
480{
481	register u_char *mp = m_arg, *np = n_arg, *lim;
482
483	if (*mp > *np)
484		return 1;  /* not really, but need to check longer one first */
485	if (*mp == *np)
486		for (lim = mp + *mp; mp < lim;)
487			if (*mp++ > *np++)
488				return 1;
489	return 0;
490}
491
492static struct radix_mask *
493rn_new_radix_mask(register struct radix_node *tt,
494		  register struct radix_mask *next)
495{
496	register struct radix_mask *m;
497
498	MKGet(m);
499	if (m == 0) {
500		log(LOG_ERR, "Mask for route not entered\n");
501		return (0);
502	}
503	Bzero(m, sizeof *m);
504	m->rm_b = tt->rn_b;
505	m->rm_flags = tt->rn_flags;
506	if (tt->rn_flags & RNF_NORMAL)
507		m->rm_leaf = tt;
508	else
509		m->rm_mask = tt->rn_mask;
510	m->rm_mklist = next;
511	tt->rn_mklist = m;
512	return m;
513}
514
515struct radix_node *
516rn_addroute(void *v_arg,
517	    void *n_arg,
518	    struct radix_node_head *head,
519	    struct radix_node treenodes[2])
520{
521	caddr_t v = (caddr_t)v_arg, netmask = (caddr_t)n_arg;
522	register struct radix_node *t, *x = 0, *tt;
523	struct radix_node *saved_tt, *top = head->rnh_treetop;
524	short b = 0, b_leaf = 0;
525	int keyduplicated;
526	caddr_t mmask;
527	struct radix_mask *m, **mp;
528
529	/*
530	 * In dealing with non-contiguous masks, there may be
531	 * many different routes which have the same mask.
532	 * We will find it useful to have a unique pointer to
533	 * the mask to speed avoiding duplicate references at
534	 * nodes and possibly save time in calculating indices.
535	 */
536	if (netmask)  {
537		if ((x = rn_addmask(netmask, 0, top->rn_off)) == 0)
538			return (0);
539		b_leaf = x->rn_b;
540		b = -1 - x->rn_b;
541		netmask = x->rn_key;
542	}
543	/*
544	 * Deal with duplicated keys: attach node to previous instance
545	 */
546	saved_tt = tt = rn_insert(v, head, &keyduplicated, treenodes);
547	if (keyduplicated) {
548		for (t = tt; tt; t = tt, tt = tt->rn_dupedkey) {
549			if (tt->rn_mask == netmask)
550				return (0);
551			if (netmask == 0 ||
552			    (tt->rn_mask &&
553			     ((b_leaf < tt->rn_b) || /* index(netmask) > node */
554			       rn_refines(netmask, tt->rn_mask) ||
555			       rn_lexobetter(netmask, tt->rn_mask))))
556				break;
557		}
558		/*
559		 * If the mask is not duplicated, we wouldn't
560		 * find it among possible duplicate key entries
561		 * anyway, so the above test doesn't hurt.
562		 *
563		 * We sort the masks for a duplicated key the same way as
564		 * in a masklist -- most specific to least specific.
565		 * This may require the unfortunate nuisance of relocating
566		 * the head of the list.
567		 */
568		if (tt == saved_tt) {
569			struct	radix_node *xx = x;
570			/* link in at head of list */
571			(tt = treenodes)->rn_dupedkey = t;
572			tt->rn_flags = t->rn_flags;
573			tt->rn_p = x = t->rn_p;
574			if (x->rn_l == t) x->rn_l = tt; else x->rn_r = tt;
575			saved_tt = tt; x = xx;
576		} else {
577			(tt = treenodes)->rn_dupedkey = t->rn_dupedkey;
578			t->rn_dupedkey = tt;
579		}
580#ifdef RN_DEBUG
581		t=tt+1; tt->rn_info = rn_nodenum++; t->rn_info = rn_nodenum++;
582		tt->rn_twin = t; tt->rn_ybro = rn_clist; rn_clist = tt;
583#endif
584		tt->rn_key = (caddr_t) v;
585		tt->rn_b = -1;
586		tt->rn_flags = RNF_ACTIVE;
587	}
588	/*
589	 * Put mask in tree.
590	 */
591	if (netmask) {
592		tt->rn_mask = netmask;
593		tt->rn_b = x->rn_b;
594		tt->rn_flags |= x->rn_flags & RNF_NORMAL;
595	}
596	t = saved_tt->rn_p;
597	if (keyduplicated)
598		goto on2;
599	b_leaf = -1 - t->rn_b;
600	if (t->rn_r == saved_tt) x = t->rn_l; else x = t->rn_r;
601	/* Promote general routes from below */
602	if (x->rn_b < 0) {
603	    for (mp = &t->rn_mklist; x; x = x->rn_dupedkey)
604		if (x->rn_mask && (x->rn_b >= b_leaf) && x->rn_mklist == 0) {
605			if ((*mp = m = rn_new_radix_mask(x, 0)))
606				mp = &m->rm_mklist;
607		}
608	} else if (x->rn_mklist) {
609		/*
610		 * Skip over masks whose index is > that of new node
611		 */
612		for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist)
613			if (m->rm_b >= b_leaf)
614				break;
615		t->rn_mklist = m; *mp = 0;
616	}
617on2:
618	/* Add new route to highest possible ancestor's list */
619	if ((netmask == 0) || (b > t->rn_b ))
620		return tt; /* can't lift at all */
621	b_leaf = tt->rn_b;
622	do {
623		x = t;
624		t = t->rn_p;
625	} while (b <= t->rn_b && x != top);
626	/*
627	 * Search through routes associated with node to
628	 * insert new route according to index.
629	 * Need same criteria as when sorting dupedkeys to avoid
630	 * double loop on deletion.
631	 */
632	for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist) {
633		if (m->rm_b < b_leaf)
634			continue;
635		if (m->rm_b > b_leaf)
636			break;
637		if (m->rm_flags & RNF_NORMAL) {
638			mmask = m->rm_leaf->rn_mask;
639			if (tt->rn_flags & RNF_NORMAL) {
640				log(LOG_ERR,
641				   "Non-unique normal route, mask not entered");
642				return tt;
643			}
644		} else
645			mmask = m->rm_mask;
646		if (mmask == netmask) {
647			m->rm_refs++;
648			tt->rn_mklist = m;
649			return tt;
650		}
651		if (rn_refines(netmask, mmask) || rn_lexobetter(netmask, mmask))
652			break;
653	}
654	*mp = rn_new_radix_mask(tt, *mp);
655	return tt;
656}
657
658struct radix_node *
659rn_delete(void *v_arg,
660	  void *netmask_arg,
661	  struct radix_node_head *head)
662{
663	register struct radix_node *t, *p, *x, *tt;
664	struct radix_mask *m, *saved_m, **mp;
665	struct radix_node *dupedkey, *saved_tt, *top;
666	caddr_t v, netmask;
667	int b, head_off, vlen;
668
669	v = v_arg;
670	netmask = netmask_arg;
671	x = head->rnh_treetop;
672	tt = rn_search(v, x);
673	head_off = x->rn_off;
674	vlen =  *(u_char *)v;
675	saved_tt = tt;
676	top = x;
677	if (tt == 0 ||
678	    Bcmp(v + head_off, tt->rn_key + head_off, vlen - head_off))
679		return (0);
680	/*
681	 * Delete our route from mask lists.
682	 */
683	if (netmask) {
684		if ((x = rn_addmask(netmask, 1, head_off)) == 0)
685			return (0);
686		netmask = x->rn_key;
687		while (tt->rn_mask != netmask)
688			if ((tt = tt->rn_dupedkey) == 0)
689				return (0);
690	}
691	if (tt->rn_mask == 0 || (saved_m = m = tt->rn_mklist) == 0)
692		goto on1;
693	if (tt->rn_flags & RNF_NORMAL) {
694		if (m->rm_leaf != tt || m->rm_refs > 0) {
695			log(LOG_ERR, "rn_delete: inconsistent annotation\n");
696			return 0;  /* dangling ref could cause disaster */
697		}
698	} else {
699		if (m->rm_mask != tt->rn_mask) {
700			log(LOG_ERR, "rn_delete: inconsistent annotation\n");
701			goto on1;
702		}
703		if (--m->rm_refs >= 0)
704			goto on1;
705	}
706	b = -1 - tt->rn_b;
707	t = saved_tt->rn_p;
708	if (b > t->rn_b)
709		goto on1; /* Wasn't lifted at all */
710	do {
711		x = t;
712		t = t->rn_p;
713	} while (b <= t->rn_b && x != top);
714	for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist)
715		if (m == saved_m) {
716			*mp = m->rm_mklist;
717			MKFree(m);
718			break;
719		}
720	if (m == 0) {
721		log(LOG_ERR, "rn_delete: couldn't find our annotation\n");
722		if (tt->rn_flags & RNF_NORMAL)
723			return (0); /* Dangling ref to us */
724	}
725on1:
726	/*
727	 * Eliminate us from tree
728	 */
729	if (tt->rn_flags & RNF_ROOT)
730		return (0);
731#ifdef RN_DEBUG
732	/* Get us out of the creation list */
733	for (t = rn_clist; t && t->rn_ybro != tt; t = t->rn_ybro) {}
734	if (t) t->rn_ybro = tt->rn_ybro;
735#endif
736	t = tt->rn_p;
737	if ((dupedkey = saved_tt->rn_dupedkey)) {
738		if (tt == saved_tt) {
739			x = dupedkey; x->rn_p = t;
740			if (t->rn_l == tt) t->rn_l = x; else t->rn_r = x;
741		} else {
742			for (x = p = saved_tt; p && p->rn_dupedkey != tt;)
743				p = p->rn_dupedkey;
744			if (p) p->rn_dupedkey = tt->rn_dupedkey;
745			else log(LOG_ERR, "rn_delete: couldn't find us\n");
746		}
747		t = tt + 1;
748		if  (t->rn_flags & RNF_ACTIVE) {
749#ifndef RN_DEBUG
750			*++x = *t; p = t->rn_p;
751#else
752			b = t->rn_info; *++x = *t; t->rn_info = b; p = t->rn_p;
753#endif
754			if (p->rn_l == t) p->rn_l = x; else p->rn_r = x;
755			x->rn_l->rn_p = x; x->rn_r->rn_p = x;
756		}
757		goto out;
758	}
759	if (t->rn_l == tt) x = t->rn_r; else x = t->rn_l;
760	p = t->rn_p;
761	if (p->rn_r == t) p->rn_r = x; else p->rn_l = x;
762	x->rn_p = p;
763	/*
764	 * Demote routes attached to us.
765	 */
766	if (t->rn_mklist) {
767		if (x->rn_b >= 0) {
768			for (mp = &x->rn_mklist; (m = *mp);)
769				mp = &m->rm_mklist;
770			*mp = t->rn_mklist;
771		} else {
772			/* If there are any key,mask pairs in a sibling
773			   duped-key chain, some subset will appear sorted
774			   in the same order attached to our mklist */
775			for (m = t->rn_mklist; m && x; x = x->rn_dupedkey)
776				if (m == x->rn_mklist) {
777					struct radix_mask *mm = m->rm_mklist;
778					x->rn_mklist = 0;
779					if (--(m->rm_refs) < 0)
780						MKFree(m);
781					m = mm;
782				}
783			if (m)
784				syslog(LOG_ERR, "%s %lx at %lx\n",
785				       "rn_delete: Orphaned Mask",
786				       (unsigned long)m,
787				       (unsigned long)x);
788		}
789	}
790	/*
791	 * We may be holding an active internal node in the tree.
792	 */
793	x = tt + 1;
794	if (t != x) {
795#ifndef RN_DEBUG
796		*t = *x;
797#else
798		b = t->rn_info; *t = *x; t->rn_info = b;
799#endif
800		t->rn_l->rn_p = t; t->rn_r->rn_p = t;
801		p = x->rn_p;
802		if (p->rn_l == x) p->rn_l = t; else p->rn_r = t;
803	}
804out:
805	tt->rn_flags &= ~RNF_ACTIVE;
806	tt[1].rn_flags &= ~RNF_ACTIVE;
807	return (tt);
808}
809
810int
811rn_walktree(struct radix_node_head *h,
812	    register int (*f)(struct radix_node *, struct walkarg*),
813	    struct walkarg *w)
814{
815	int error;
816	struct radix_node *base, *next;
817	register struct radix_node *rn = h->rnh_treetop;
818	/*
819	 * This gets complicated because we may delete the node
820	 * while applying the function f to it, so we need to calculate
821	 * the successor node in advance.
822	 */
823	/* First time through node, go left */
824	while (rn->rn_b >= 0)
825		rn = rn->rn_l;
826	for (;;) {
827		base = rn;
828		/* If at right child go back up, otherwise, go right */
829		while (rn->rn_p->rn_r == rn && (rn->rn_flags & RNF_ROOT) == 0)
830			rn = rn->rn_p;
831		/* Find the next *leaf* since next node might vanish, too */
832		for (rn = rn->rn_p->rn_r; rn->rn_b >= 0;)
833			rn = rn->rn_l;
834		next = rn;
835		/* Process leaves */
836		while ((rn = base)) {
837			base = rn->rn_dupedkey;
838			if (!(rn->rn_flags & RNF_ROOT) && (error = (*f)(rn, w)))
839				return (error);
840		}
841		rn = next;
842		if (rn->rn_flags & RNF_ROOT)
843			return (0);
844	}
845	/* NOTREACHED */
846}
847
848int
849rn_inithead(void **head, int off)
850{
851	register struct radix_node_head *rnh;
852	register struct radix_node *t, *tt, *ttt;
853	if (*head)
854		return (1);
855	rnh = (struct radix_node_head *)rtmalloc(sizeof (*rnh), "rn_inithead");
856	if (rnh == 0)
857		return (0);
858	Bzero(rnh, sizeof (*rnh));
859	*head = rnh;
860	t = rn_newpair(rn_zeros, off, rnh->rnh_nodes);
861	ttt = rnh->rnh_nodes + 2;
862	t->rn_r = ttt;
863	t->rn_p = t;
864	tt = t->rn_l;
865	tt->rn_flags = t->rn_flags = RNF_ROOT | RNF_ACTIVE;
866	tt->rn_b = -1 - off;
867	*ttt = *tt;
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;
875	return (1);
876}
877
878void
879rn_init(void)
880{
881	char *cp, *cplim;
882	if (max_keylen == 0) {
883		printf("rn_init: radix functions require max_keylen be set\n");
884		return;
885	}
886	rn_zeros = (char *)rtmalloc(3 * max_keylen, "rn_init");
887	if (rn_zeros == NULL)
888		panic("rn_init");
889	Bzero(rn_zeros, 3 * max_keylen);
890	rn_ones = cp = rn_zeros + max_keylen;
891	addmask_key = cplim = rn_ones + max_keylen;
892	while (cp < cplim)
893		*cp++ = -1;
894	if (rn_inithead((void **)&mask_rnhead, 0) == 0)
895		panic("rn_init 2");
896}
897
898