radix.c revision 1.16
1/*	$NetBSD: radix.c,v 1.16 2001/01/04 00:17:52 enami Exp $	*/
2
3/*
4 * Copyright (c) 1988, 1989, 1993
5 *	The Regents of the University of California.  All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
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. All advertising materials mentioning features or use of this software
16 *    must display the following acknowledgement:
17 *	This product includes software developed by the University of
18 *	California, Berkeley and its contributors.
19 * 4. Neither the name of the University nor the names of its contributors
20 *    may be used to endorse or promote products derived from this software
21 *    without specific prior written permission.
22 *
23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * SUCH DAMAGE.
34 *
35 *	@(#)radix.c	8.6 (Berkeley) 10/17/95
36 */
37
38/*
39 * Routines to build and maintain radix trees for routing lookups.
40 */
41#ifndef _NET_RADIX_H_
42#include <sys/param.h>
43#ifdef	_KERNEL
44#include <sys/systm.h>
45#include <sys/malloc.h>
46#define	M_DONTWAIT M_NOWAIT
47#include <sys/domain.h>
48#else
49#include <stdlib.h>
50#endif
51#include <sys/syslog.h>
52#include <net/radix.h>
53#endif
54
55int	max_keylen;
56struct radix_mask *rn_mkfreelist;
57struct radix_node_head *mask_rnhead;
58static char *addmask_key;
59static char normal_chars[] = {0, 0x80, 0xc0, 0xe0, 0xf0, 0xf8, 0xfc, 0xfe, -1};
60static char *rn_zeros, *rn_ones;
61
62#define rn_masktop (mask_rnhead->rnh_treetop)
63#undef Bcmp
64#define Bcmp(a, b, l) (l == 0 ? 0 : bcmp((caddr_t)(a), (caddr_t)(b), (u_long)l))
65
66static int rn_satisfies_leaf __P((char *, struct radix_node *, int));
67static int rn_lexobetter __P((void *, void *));
68static struct radix_mask *rn_new_radix_mask __P((struct radix_node *,
69    struct radix_mask *));
70
71/*
72 * The data structure for the keys is a radix tree with one way
73 * branching removed.  The index rn_b at an internal node n represents a bit
74 * position to be tested.  The tree is arranged so that all descendants
75 * of a node n have keys whose bits all agree up to position rn_b - 1.
76 * (We say the index of n is rn_b.)
77 *
78 * There is at least one descendant which has a one bit at position rn_b,
79 * and at least one with a zero there.
80 *
81 * A route is determined by a pair of key and mask.  We require that the
82 * bit-wise logical and of the key and mask to be the key.
83 * We define the index of a route to associated with the mask to be
84 * the first bit number in the mask where 0 occurs (with bit number 0
85 * representing the highest order bit).
86 *
87 * We say a mask is normal if every bit is 0, past the index of the mask.
88 * If a node n has a descendant (k, m) with index(m) == index(n) == rn_b,
89 * and m is a normal mask, then the route applies to every descendant of n.
90 * If the index(m) < rn_b, this implies the trailing last few bits of k
91 * before bit b are all 0, (and hence consequently true of every descendant
92 * of n), so the route applies to all descendants of the node as well.
93 *
94 * Similar logic shows that a non-normal mask m such that
95 * index(m) <= index(n) could potentially apply to many children of n.
96 * Thus, for each non-host route, we attach its mask to a list at an internal
97 * node as high in the tree as we can go.
98 *
99 * The present version of the code makes use of normal routes in short-
100 * circuiting an explict mask and compare operation when testing whether
101 * a key satisfies a normal route, and also in remembering the unique leaf
102 * that governs a subtree.
103 */
104
105struct radix_node *
106rn_search(v_arg, head)
107	void *v_arg;
108	struct radix_node *head;
109{
110	struct radix_node *x;
111	caddr_t v;
112
113	for (x = head, v = v_arg; x->rn_b >= 0;) {
114		if (x->rn_bmask & v[x->rn_off])
115			x = x->rn_r;
116		else
117			x = x->rn_l;
118	}
119	return (x);
120}
121
122struct radix_node *
123rn_search_m(v_arg, head, m_arg)
124	struct radix_node *head;
125	void *v_arg, *m_arg;
126{
127	struct radix_node *x;
128	caddr_t v = v_arg, m = m_arg;
129
130	for (x = head; x->rn_b >= 0;) {
131		if ((x->rn_bmask & m[x->rn_off]) &&
132		    (x->rn_bmask & v[x->rn_off]))
133			x = x->rn_r;
134		else
135			x = x->rn_l;
136	}
137	return x;
138}
139
140int
141rn_refines(m_arg, n_arg)
142	void *m_arg, *n_arg;
143{
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;
148
149	if (longer > 0)
150		lim -= longer;
151	while (n < lim) {
152		if (*n & ~(*m))
153			return 0;
154		if (*n++ != *m++)
155			masks_are_equal = 0;
156	}
157	while (n < lim2)
158		if (*n++)
159			return 0;
160	if (masks_are_equal && (longer < 0))
161		for (lim2 = m - longer; m < lim2; )
162			if (*m++)
163				return 1;
164	return (!masks_are_equal);
165}
166
167struct radix_node *
168rn_lookup(v_arg, m_arg, head)
169	void *v_arg, *m_arg;
170	struct radix_node_head *head;
171{
172	struct radix_node *x;
173	caddr_t netmask = 0;
174
175	if (m_arg) {
176		if ((x = rn_addmask(m_arg, 1, head->rnh_treetop->rn_off)) == 0)
177			return (0);
178		netmask = x->rn_key;
179	}
180	x = rn_match(v_arg, head);
181	if (x && netmask) {
182		while (x && x->rn_mask != netmask)
183			x = x->rn_dupedkey;
184	}
185	return x;
186}
187
188static int
189rn_satisfies_leaf(trial, leaf, skip)
190	char *trial;
191	struct radix_node *leaf;
192	int skip;
193{
194	char *cp = trial, *cp2 = leaf->rn_key, *cp3 = leaf->rn_mask;
195	char *cplim;
196	int length = min(*(u_char *)cp, *(u_char *)cp2);
197
198	if (cp3 == 0)
199		cp3 = rn_ones;
200	else
201		length = min(length, *(u_char *)cp3);
202	cplim = cp + length; cp3 += skip; cp2 += skip;
203	for (cp += skip; cp < cplim; cp++, cp2++, cp3++)
204		if ((*cp ^ *cp2) & *cp3)
205			return 0;
206	return 1;
207}
208
209struct radix_node *
210rn_match(v_arg, head)
211	void *v_arg;
212	struct radix_node_head *head;
213{
214	caddr_t v = v_arg;
215	struct radix_node *t = head->rnh_treetop, *x;
216	caddr_t cp = v, cp2;
217	caddr_t cplim;
218	struct radix_node *saved_t, *top = t;
219	int off = t->rn_off, vlen = *(u_char *)cp, matched_off;
220	int test, b, rn_b;
221
222	/*
223	 * Open code rn_search(v, top) to avoid overhead of extra
224	 * subroutine call.
225	 */
226	for (; t->rn_b >= 0; ) {
227		if (t->rn_bmask & cp[t->rn_off])
228			t = t->rn_r;
229		else
230			t = t->rn_l;
231	}
232	/*
233	 * See if we match exactly as a host destination
234	 * or at least learn how many bits match, for normal mask finesse.
235	 *
236	 * It doesn't hurt us to limit how many bytes to check
237	 * to the length of the mask, since if it matches we had a genuine
238	 * match and the leaf we have is the most specific one anyway;
239	 * if it didn't match with a shorter length it would fail
240	 * with a long one.  This wins big for class B&C netmasks which
241	 * are probably the most common case...
242	 */
243	if (t->rn_mask)
244		vlen = *(u_char *)t->rn_mask;
245	cp += off; cp2 = t->rn_key + off; cplim = v + vlen;
246	for (; cp < cplim; cp++, cp2++)
247		if (*cp != *cp2)
248			goto on1;
249	/*
250	 * This extra grot is in case we are explicitly asked
251	 * to look up the default.  Ugh!
252	 */
253	if ((t->rn_flags & RNF_ROOT) && t->rn_dupedkey)
254		t = t->rn_dupedkey;
255	return t;
256on1:
257	test = (*cp ^ *cp2) & 0xff; /* find first bit that differs */
258	for (b = 7; (test >>= 1) > 0;)
259		b--;
260	matched_off = cp - v;
261	b += matched_off << 3;
262	rn_b = -1 - b;
263	/*
264	 * If there is a host route in a duped-key chain, it will be first.
265	 */
266	if ((saved_t = t)->rn_mask == 0)
267		t = t->rn_dupedkey;
268	for (; t; t = t->rn_dupedkey)
269		/*
270		 * Even if we don't match exactly as a host,
271		 * we may match if the leaf we wound up at is
272		 * a route to a net.
273		 */
274		if (t->rn_flags & RNF_NORMAL) {
275			if (rn_b <= t->rn_b)
276				return t;
277		} else if (rn_satisfies_leaf(v, t, matched_off))
278				return t;
279	t = saved_t;
280	/* start searching up the tree */
281	do {
282		struct radix_mask *m;
283		t = t->rn_p;
284		m = t->rn_mklist;
285		if (m) {
286			/*
287			 * If non-contiguous masks ever become important
288			 * we can restore the masking and open coding of
289			 * the search and satisfaction test and put the
290			 * calculation of "off" back before the "do".
291			 */
292			do {
293				if (m->rm_flags & RNF_NORMAL) {
294					if (rn_b <= m->rm_b)
295						return (m->rm_leaf);
296				} else {
297					off = min(t->rn_off, matched_off);
298					x = rn_search_m(v, t, m->rm_mask);
299					while (x && x->rn_mask != m->rm_mask)
300						x = x->rn_dupedkey;
301					if (x && rn_satisfies_leaf(v, x, off))
302						    return x;
303				}
304				m = m->rm_mklist;
305			} while (m);
306		}
307	} while (t != top);
308	return 0;
309}
310
311#ifdef RN_DEBUG
312int	rn_nodenum;
313struct	radix_node *rn_clist;
314int	rn_saveinfo;
315int	rn_debug =  1;
316#endif
317
318struct radix_node *
319rn_newpair(v, b, nodes)
320	void *v;
321	int b;
322	struct radix_node nodes[2];
323{
324	struct radix_node *tt = nodes, *t = tt + 1;
325	t->rn_b = b; t->rn_bmask = 0x80 >> (b & 7);
326	t->rn_l = tt; t->rn_off = b >> 3;
327	tt->rn_b = -1; tt->rn_key = (caddr_t)v; tt->rn_p = t;
328	tt->rn_flags = t->rn_flags = RNF_ACTIVE;
329#ifdef RN_DEBUG
330	tt->rn_info = rn_nodenum++; t->rn_info = rn_nodenum++;
331	tt->rn_twin = t; tt->rn_ybro = rn_clist; rn_clist = tt;
332#endif
333	return t;
334}
335
336struct radix_node *
337rn_insert(v_arg, head, dupentry, nodes)
338	void *v_arg;
339	struct radix_node_head *head;
340	int *dupentry;
341	struct radix_node nodes[2];
342{
343	caddr_t v = v_arg;
344	struct radix_node *top = head->rnh_treetop;
345	int head_off = top->rn_off, vlen = (int)*((u_char *)v);
346	struct radix_node *t = rn_search(v_arg, top);
347	caddr_t cp = v + head_off;
348	int b;
349	struct radix_node *tt;
350    	/*
351	 * Find first bit at which v and t->rn_key differ
352	 */
353    {
354	caddr_t cp2 = t->rn_key + head_off;
355	int cmp_res;
356	caddr_t cplim = v + vlen;
357
358	while (cp < cplim)
359		if (*cp2++ != *cp++)
360			goto on1;
361	*dupentry = 1;
362	return t;
363on1:
364	*dupentry = 0;
365	cmp_res = (cp[-1] ^ cp2[-1]) & 0xff;
366	for (b = (cp - v) << 3; cmp_res; b--)
367		cmp_res >>= 1;
368    }
369    {
370	struct radix_node *p, *x = top;
371	cp = v;
372	do {
373		p = x;
374		if (cp[x->rn_off] & x->rn_bmask)
375			x = x->rn_r;
376		else x = x->rn_l;
377	} while (b > (unsigned) x->rn_b); /* x->rn_b < b && x->rn_b >= 0 */
378#ifdef RN_DEBUG
379	if (rn_debug)
380		log(LOG_DEBUG, "rn_insert: Going In:\n"), traverse(p);
381#endif
382	t = rn_newpair(v_arg, b, nodes); tt = t->rn_l;
383	if ((cp[p->rn_off] & p->rn_bmask) == 0)
384		p->rn_l = t;
385	else
386		p->rn_r = t;
387	x->rn_p = t; t->rn_p = p; /* frees x, p as temp vars below */
388	if ((cp[t->rn_off] & t->rn_bmask) == 0) {
389		t->rn_r = x;
390	} else {
391		t->rn_r = tt; t->rn_l = x;
392	}
393#ifdef RN_DEBUG
394	if (rn_debug)
395		log(LOG_DEBUG, "rn_insert: Coming Out:\n"), traverse(p);
396#endif
397    }
398	return (tt);
399}
400
401struct radix_node *
402rn_addmask(n_arg, search, skip)
403	int search, skip;
404	void *n_arg;
405{
406	caddr_t netmask = (caddr_t)n_arg;
407	struct radix_node *x;
408	caddr_t cp, cplim;
409	int b = 0, mlen, j;
410	int maskduplicated, m0, isnormal;
411	struct radix_node *saved_x;
412	static int last_zeroed = 0;
413
414	if ((mlen = *(u_char *)netmask) > max_keylen)
415		mlen = max_keylen;
416	if (skip == 0)
417		skip = 1;
418	if (mlen <= skip)
419		return (mask_rnhead->rnh_nodes);
420	if (skip > 1)
421		Bcopy(rn_ones + 1, addmask_key + 1, skip - 1);
422	if ((m0 = mlen) > skip)
423		Bcopy(netmask + skip, addmask_key + skip, mlen - skip);
424	/*
425	 * Trim trailing zeroes.
426	 */
427	for (cp = addmask_key + mlen; (cp > addmask_key) && cp[-1] == 0;)
428		cp--;
429	mlen = cp - addmask_key;
430	if (mlen <= skip) {
431		if (m0 >= last_zeroed)
432			last_zeroed = mlen;
433		return (mask_rnhead->rnh_nodes);
434	}
435	if (m0 < last_zeroed)
436		Bzero(addmask_key + m0, last_zeroed - m0);
437	*addmask_key = last_zeroed = mlen;
438	x = rn_search(addmask_key, rn_masktop);
439	if (Bcmp(addmask_key, x->rn_key, mlen) != 0)
440		x = 0;
441	if (x || search)
442		return (x);
443	R_Malloc(x, struct radix_node *, max_keylen + 2 * sizeof (*x));
444	if ((saved_x = x) == 0)
445		return (0);
446	Bzero(x, max_keylen + 2 * sizeof (*x));
447	netmask = cp = (caddr_t)(x + 2);
448	Bcopy(addmask_key, cp, mlen);
449	x = rn_insert(cp, mask_rnhead, &maskduplicated, x);
450	if (maskduplicated) {
451		log(LOG_ERR, "rn_addmask: mask impossibly already in tree\n");
452		Free(saved_x);
453		return (x);
454	}
455	/*
456	 * Calculate index of mask, and check for normalcy.
457	 */
458	cplim = netmask + mlen; isnormal = 1;
459	for (cp = netmask + skip; (cp < cplim) && *(u_char *)cp == 0xff;)
460		cp++;
461	if (cp != cplim) {
462		for (j = 0x80; (j & *cp) != 0; j >>= 1)
463			b++;
464		if (*cp != normal_chars[b] || cp != (cplim - 1))
465			isnormal = 0;
466	}
467	b += (cp - netmask) << 3;
468	x->rn_b = -1 - b;
469	if (isnormal)
470		x->rn_flags |= RNF_NORMAL;
471	return (x);
472}
473
474static int	/* XXX: arbitrary ordering for non-contiguous masks */
475rn_lexobetter(m_arg, n_arg)
476	void *m_arg, *n_arg;
477{
478	u_char *mp = m_arg, *np = n_arg, *lim;
479
480	if (*mp > *np)
481		return 1;  /* not really, but need to check longer one first */
482	if (*mp == *np)
483		for (lim = mp + *mp; mp < lim;)
484			if (*mp++ > *np++)
485				return 1;
486	return 0;
487}
488
489static struct radix_mask *
490rn_new_radix_mask(tt, next)
491	struct radix_node *tt;
492	struct radix_mask *next;
493{
494	struct radix_mask *m;
495
496	MKGet(m);
497	if (m == 0) {
498		log(LOG_ERR, "Mask for route not entered\n");
499		return (0);
500	}
501	Bzero(m, sizeof *m);
502	m->rm_b = tt->rn_b;
503	m->rm_flags = tt->rn_flags;
504	if (tt->rn_flags & RNF_NORMAL)
505		m->rm_leaf = tt;
506	else
507		m->rm_mask = tt->rn_mask;
508	m->rm_mklist = next;
509	tt->rn_mklist = m;
510	return m;
511}
512
513struct radix_node *
514rn_addroute(v_arg, n_arg, head, treenodes)
515	void *v_arg, *n_arg;
516	struct radix_node_head *head;
517	struct radix_node treenodes[2];
518{
519	caddr_t v = (caddr_t)v_arg, netmask = (caddr_t)n_arg;
520	struct radix_node *t, *x = 0, *tt;
521	struct radix_node *saved_tt, *top = head->rnh_treetop;
522	short b = 0, b_leaf = 0;
523	int keyduplicated;
524	caddr_t mmask;
525	struct radix_mask *m, **mp;
526
527	/*
528	 * In dealing with non-contiguous masks, there may be
529	 * many different routes which have the same mask.
530	 * We will find it useful to have a unique pointer to
531	 * the mask to speed avoiding duplicate references at
532	 * nodes and possibly save time in calculating indices.
533	 */
534	if (netmask)  {
535		if ((x = rn_addmask(netmask, 0, top->rn_off)) == 0)
536			return (0);
537		b_leaf = x->rn_b;
538		b = -1 - x->rn_b;
539		netmask = x->rn_key;
540	}
541	/*
542	 * Deal with duplicated keys: attach node to previous instance
543	 */
544	saved_tt = tt = rn_insert(v, head, &keyduplicated, treenodes);
545	if (keyduplicated) {
546		for (t = tt; tt; t = tt, tt = tt->rn_dupedkey) {
547			if (tt->rn_mask == netmask)
548				return (0);
549			if (netmask == 0 ||
550			    (tt->rn_mask &&
551			     ((b_leaf < tt->rn_b) || /* index(netmask) > node */
552			       rn_refines(netmask, tt->rn_mask) ||
553			       rn_lexobetter(netmask, tt->rn_mask))))
554				break;
555		}
556		/*
557		 * If the mask is not duplicated, we wouldn't
558		 * find it among possible duplicate key entries
559		 * anyway, so the above test doesn't hurt.
560		 *
561		 * We sort the masks for a duplicated key the same way as
562		 * in a masklist -- most specific to least specific.
563		 * This may require the unfortunate nuisance of relocating
564		 * the head of the list.
565		 *
566		 * We also reverse, or doubly link the list through the
567		 * parent pointer.
568		 */
569		if (tt == saved_tt) {
570			struct	radix_node *xx = x;
571			/* link in at head of list */
572			(tt = treenodes)->rn_dupedkey = t;
573			tt->rn_flags = t->rn_flags;
574			tt->rn_p = x = t->rn_p;
575			t->rn_p = tt;
576			if (x->rn_l == t) x->rn_l = tt; else x->rn_r = tt;
577			saved_tt = tt; x = xx;
578		} else {
579			(tt = treenodes)->rn_dupedkey = t->rn_dupedkey;
580			t->rn_dupedkey = tt;
581			tt->rn_p = t;
582			if (tt->rn_dupedkey)
583				tt->rn_dupedkey->rn_p = tt;
584		}
585#ifdef RN_DEBUG
586		t=tt+1; tt->rn_info = rn_nodenum++; t->rn_info = rn_nodenum++;
587		tt->rn_twin = t; tt->rn_ybro = rn_clist; rn_clist = tt;
588#endif
589		tt->rn_key = (caddr_t) v;
590		tt->rn_b = -1;
591		tt->rn_flags = RNF_ACTIVE;
592	}
593	/*
594	 * Put mask in tree.
595	 */
596	if (netmask) {
597		tt->rn_mask = netmask;
598		tt->rn_b = x->rn_b;
599		tt->rn_flags |= x->rn_flags & RNF_NORMAL;
600	}
601	t = saved_tt->rn_p;
602	if (keyduplicated)
603		goto on2;
604	b_leaf = -1 - t->rn_b;
605	if (t->rn_r == saved_tt) x = t->rn_l; else x = t->rn_r;
606	/* Promote general routes from below */
607	if (x->rn_b < 0) {
608	    for (mp = &t->rn_mklist; x; x = x->rn_dupedkey)
609		if (x->rn_mask && (x->rn_b >= b_leaf) && x->rn_mklist == 0) {
610			*mp = m = rn_new_radix_mask(x, 0);
611			if (m)
612				mp = &m->rm_mklist;
613		}
614	} else if (x->rn_mklist) {
615		/*
616		 * Skip over masks whose index is > that of new node
617		 */
618		for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist)
619			if (m->rm_b >= b_leaf)
620				break;
621		t->rn_mklist = m; *mp = 0;
622	}
623on2:
624	/* Add new route to highest possible ancestor's list */
625	if ((netmask == 0) || (b > t->rn_b ))
626		return tt; /* can't lift at all */
627	b_leaf = tt->rn_b;
628	do {
629		x = t;
630		t = t->rn_p;
631	} while (b <= t->rn_b && x != top);
632	/*
633	 * Search through routes associated with node to
634	 * insert new route according to index.
635	 * Need same criteria as when sorting dupedkeys to avoid
636	 * double loop on deletion.
637	 */
638	for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist) {
639		if (m->rm_b < b_leaf)
640			continue;
641		if (m->rm_b > b_leaf)
642			break;
643		if (m->rm_flags & RNF_NORMAL) {
644			mmask = m->rm_leaf->rn_mask;
645			if (tt->rn_flags & RNF_NORMAL) {
646				log(LOG_ERR, "Non-unique normal route,"
647				    " mask not entered\n");
648				return tt;
649			}
650		} else
651			mmask = m->rm_mask;
652		if (mmask == netmask) {
653			m->rm_refs++;
654			tt->rn_mklist = m;
655			return tt;
656		}
657		if (rn_refines(netmask, mmask) || rn_lexobetter(netmask, mmask))
658			break;
659	}
660	*mp = rn_new_radix_mask(tt, *mp);
661	return tt;
662}
663
664struct radix_node *
665rn_delete(v_arg, netmask_arg, head)
666	void *v_arg, *netmask_arg;
667	struct radix_node_head *head;
668{
669	struct radix_node *t, *p, *x, *tt;
670	struct radix_mask *m, *saved_m, **mp;
671	struct radix_node *dupedkey, *saved_tt, *top;
672	caddr_t v, netmask;
673	int b, head_off, vlen;
674
675	v = v_arg;
676	netmask = netmask_arg;
677	x = head->rnh_treetop;
678	tt = rn_search(v, x);
679	head_off = x->rn_off;
680	vlen =  *(u_char *)v;
681	saved_tt = tt;
682	top = x;
683	if (tt == 0 ||
684	    Bcmp(v + head_off, tt->rn_key + head_off, vlen - head_off))
685		return (0);
686	/*
687	 * Delete our route from mask lists.
688	 */
689	if (netmask) {
690		if ((x = rn_addmask(netmask, 1, head_off)) == 0)
691			return (0);
692		netmask = x->rn_key;
693		while (tt->rn_mask != netmask)
694			if ((tt = tt->rn_dupedkey) == 0)
695				return (0);
696	}
697	if (tt->rn_mask == 0 || (saved_m = m = tt->rn_mklist) == 0)
698		goto on1;
699	if (tt->rn_flags & RNF_NORMAL) {
700		if (m->rm_leaf != tt || m->rm_refs > 0) {
701			log(LOG_ERR, "rn_delete: inconsistent annotation\n");
702			return 0;  /* dangling ref could cause disaster */
703		}
704	} else {
705		if (m->rm_mask != tt->rn_mask) {
706			log(LOG_ERR, "rn_delete: inconsistent annotation\n");
707			goto on1;
708		}
709		if (--m->rm_refs >= 0)
710			goto on1;
711	}
712	b = -1 - tt->rn_b;
713	t = saved_tt->rn_p;
714	if (b > t->rn_b)
715		goto on1; /* Wasn't lifted at all */
716	do {
717		x = t;
718		t = t->rn_p;
719	} while (b <= t->rn_b && x != top);
720	for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist)
721		if (m == saved_m) {
722			*mp = m->rm_mklist;
723			MKFree(m);
724			break;
725		}
726	if (m == 0) {
727		log(LOG_ERR, "rn_delete: couldn't find our annotation\n");
728		if (tt->rn_flags & RNF_NORMAL)
729			return (0); /* Dangling ref to us */
730	}
731on1:
732	/*
733	 * Eliminate us from tree
734	 */
735	if (tt->rn_flags & RNF_ROOT)
736		return (0);
737#ifdef RN_DEBUG
738	/* Get us out of the creation list */
739	for (t = rn_clist; t && t->rn_ybro != tt; t = t->rn_ybro) {}
740	if (t) t->rn_ybro = tt->rn_ybro;
741#endif
742	t = tt->rn_p;
743	dupedkey = saved_tt->rn_dupedkey;
744	if (dupedkey) {
745		/*
746		 * Here, tt is the deletion target, and
747		 * saved_tt is the head of the dupedkey chain.
748		 */
749		if (tt == saved_tt) {
750			x = dupedkey; x->rn_p = t;
751			if (t->rn_l == tt) t->rn_l = x; else t->rn_r = x;
752		} else {
753			/* find node in front of tt on the chain */
754			for (x = p = saved_tt; p && p->rn_dupedkey != tt;)
755				p = p->rn_dupedkey;
756			if (p) {
757				p->rn_dupedkey = tt->rn_dupedkey;
758				if (tt->rn_dupedkey)
759					tt->rn_dupedkey->rn_p = p;
760			} else log(LOG_ERR, "rn_delete: couldn't find us\n");
761		}
762		t = tt + 1;
763		if  (t->rn_flags & RNF_ACTIVE) {
764#ifndef RN_DEBUG
765			*++x = *t; p = t->rn_p;
766#else
767			b = t->rn_info; *++x = *t; t->rn_info = b; p = t->rn_p;
768#endif
769			if (p->rn_l == t) p->rn_l = x; else p->rn_r = x;
770			x->rn_l->rn_p = x; x->rn_r->rn_p = x;
771		}
772		goto out;
773	}
774	if (t->rn_l == tt) x = t->rn_r; else x = t->rn_l;
775	p = t->rn_p;
776	if (p->rn_r == t) p->rn_r = x; else p->rn_l = x;
777	x->rn_p = p;
778	/*
779	 * Demote routes attached to us.
780	 */
781	if (t->rn_mklist) {
782		if (x->rn_b >= 0) {
783			for (mp = &x->rn_mklist; (m = *mp);)
784				mp = &m->rm_mklist;
785			*mp = t->rn_mklist;
786		} else {
787			/* If there are any key,mask pairs in a sibling
788			   duped-key chain, some subset will appear sorted
789			   in the same order attached to our mklist */
790			for (m = t->rn_mklist; m && x; x = x->rn_dupedkey)
791				if (m == x->rn_mklist) {
792					struct radix_mask *mm = m->rm_mklist;
793					x->rn_mklist = 0;
794					if (--(m->rm_refs) < 0)
795						MKFree(m);
796					m = mm;
797				}
798			if (m)
799				log(LOG_ERR, "%s %p at %p\n",
800					    "rn_delete: Orphaned Mask", m, x);
801		}
802	}
803	/*
804	 * We may be holding an active internal node in the tree.
805	 */
806	x = tt + 1;
807	if (t != x) {
808#ifndef RN_DEBUG
809		*t = *x;
810#else
811		b = t->rn_info; *t = *x; t->rn_info = b;
812#endif
813		t->rn_l->rn_p = t; t->rn_r->rn_p = t;
814		p = x->rn_p;
815		if (p->rn_l == x) p->rn_l = t; else p->rn_r = t;
816	}
817out:
818	tt->rn_flags &= ~RNF_ACTIVE;
819	tt[1].rn_flags &= ~RNF_ACTIVE;
820	return (tt);
821}
822
823int
824rn_walktree(h, f, w)
825	struct radix_node_head *h;
826	int (*f) __P((struct radix_node *, void *));
827	void *w;
828{
829	int error;
830	struct radix_node *base, *next;
831	struct radix_node *rn = h->rnh_treetop;
832	/*
833	 * This gets complicated because we may delete the node
834	 * while applying the function f to it, so we need to calculate
835	 * the successor node in advance.
836	 */
837	/* First time through node, go left */
838	while (rn->rn_b >= 0)
839		rn = rn->rn_l;
840	for (;;) {
841		base = rn;
842		/* If at right child go back up, otherwise, go right */
843		while (rn->rn_p->rn_r == rn && (rn->rn_flags & RNF_ROOT) == 0)
844			rn = rn->rn_p;
845		/* Find the next *leaf* since next node might vanish, too */
846		for (rn = rn->rn_p->rn_r; rn->rn_b >= 0;)
847			rn = rn->rn_l;
848		next = rn;
849		/* Process leaves */
850		while ((rn = base) != NULL) {
851			base = rn->rn_dupedkey;
852			if (!(rn->rn_flags & RNF_ROOT) && (error = (*f)(rn, w)))
853				return (error);
854		}
855		rn = next;
856		if (rn->rn_flags & RNF_ROOT)
857			return (0);
858	}
859	/* NOTREACHED */
860}
861
862int
863rn_inithead(head, off)
864	void **head;
865	int off;
866{
867	struct radix_node_head *rnh;
868
869	if (*head)
870		return (1);
871	R_Malloc(rnh, struct radix_node_head *, sizeof (*rnh));
872	if (rnh == 0)
873		return (0);
874	*head = rnh;
875	return rn_inithead0(rnh, off);
876}
877
878int
879rn_inithead0(rnh, off)
880	struct radix_node_head *rnh;
881	int off;
882{
883	struct radix_node *t, *tt, *ttt;
884
885	Bzero(rnh, sizeof (*rnh));
886	t = rn_newpair(rn_zeros, off, rnh->rnh_nodes);
887	ttt = rnh->rnh_nodes + 2;
888	t->rn_r = ttt;
889	t->rn_p = t;
890	tt = t->rn_l;
891	tt->rn_flags = t->rn_flags = RNF_ROOT | RNF_ACTIVE;
892	tt->rn_b = -1 - off;
893	*ttt = *tt;
894	ttt->rn_key = rn_ones;
895	rnh->rnh_addaddr = rn_addroute;
896	rnh->rnh_deladdr = rn_delete;
897	rnh->rnh_matchaddr = rn_match;
898	rnh->rnh_lookup = rn_lookup;
899	rnh->rnh_walktree = rn_walktree;
900	rnh->rnh_treetop = t;
901	return (1);
902}
903
904void
905rn_init()
906{
907	char *cp, *cplim;
908#ifdef _KERNEL
909	struct domain *dom;
910
911	for (dom = domains; dom; dom = dom->dom_next)
912		if (dom->dom_maxrtkey > max_keylen)
913			max_keylen = dom->dom_maxrtkey;
914#endif
915	if (max_keylen == 0) {
916		log(LOG_ERR,
917		    "rn_init: radix functions require max_keylen be set\n");
918		return;
919	}
920	R_Malloc(rn_zeros, char *, 3 * max_keylen);
921	if (rn_zeros == NULL)
922		panic("rn_init");
923	Bzero(rn_zeros, 3 * max_keylen);
924	rn_ones = cp = rn_zeros + max_keylen;
925	addmask_key = cplim = rn_ones + max_keylen;
926	while (cp < cplim)
927		*cp++ = -1;
928	if (rn_inithead((void **)&mask_rnhead, 0) == 0)
929		panic("rn_init 2");
930}
931