radix.c revision 190712
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 * 4. 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.
16 *
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
27 * SUCH DAMAGE.
28 *
29 *	@(#)radix.c	8.4 (Berkeley) 11/2/94
30 *
31 * $FreeBSD: head/sbin/routed/radix.c 190712 2009-04-05 14:15:05Z phk $
32 */
33
34/*
35 * Routines to build and maintain radix trees for routing lookups.
36 */
37
38#include "defs.h"
39
40#ifdef __NetBSD__
41__RCSID("$NetBSD$");
42#elif defined(__FreeBSD__)
43__RCSID("$FreeBSD: head/sbin/routed/radix.c 190712 2009-04-05 14:15:05Z phk $");
44#else
45__RCSID("$Revision: 2.23 $");
46#ident "$Revision: 2.23 $"
47#endif
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 \
63		       : memcmp((caddr_t)(a), (caddr_t)(b), (size_t)l))
64
65static int rn_satisfies_leaf(char *, struct radix_node *, int);
66
67/*
68 * The data structure for the keys is a radix tree with one way
69 * branching removed.  The index rn_b at an internal node n represents a bit
70 * position to be tested.  The tree is arranged so that all descendants
71 * of a node n have keys whose bits all agree up to position rn_b - 1.
72 * (We say the index of n is rn_b.)
73 *
74 * There is at least one descendant which has a one bit at position rn_b,
75 * and at least one with a zero there.
76 *
77 * A route is determined by a pair of key and mask.  We require that the
78 * bit-wise logical and of the key and mask to be the key.
79 * We define the index of a route to associated with the mask to be
80 * the first bit number in the mask where 0 occurs (with bit number 0
81 * representing the highest order bit).
82 *
83 * We say a mask is normal if every bit is 0, past the index of the mask.
84 * If a node n has a descendant (k, m) with index(m) == index(n) == rn_b,
85 * and m is a normal mask, then the route applies to every descendant of n.
86 * If the index(m) < rn_b, this implies the trailing last few bits of k
87 * before bit b are all 0, (and hence consequently true of every descendant
88 * of n), so the route applies to all descendants of the node as well.
89 *
90 * Similar logic shows that a non-normal mask m such that
91 * index(m) <= index(n) could potentially apply to many children of n.
92 * Thus, for each non-host route, we attach its mask to a list at an internal
93 * node as high in the tree as we can go.
94 *
95 * The present version of the code makes use of normal routes in short-
96 * circuiting an explict mask and compare operation when testing whether
97 * a key satisfies a normal route, and also in remembering the unique leaf
98 * that governs a subtree.
99 */
100
101struct radix_node *
102rn_search(void *v_arg,
103	  struct radix_node *head)
104{
105	struct radix_node *x;
106	caddr_t v;
107
108	for (x = head, v = v_arg; x->rn_b >= 0;) {
109		if (x->rn_bmask & v[x->rn_off])
110			x = x->rn_r;
111		else
112			x = x->rn_l;
113	}
114	return (x);
115}
116
117struct radix_node *
118rn_search_m(void *v_arg,
119	    struct radix_node *head,
120	    void *m_arg)
121{
122	struct radix_node *x;
123	caddr_t v = v_arg, m = m_arg;
124
125	for (x = head; x->rn_b >= 0;) {
126		if ((x->rn_bmask & m[x->rn_off]) &&
127		    (x->rn_bmask & v[x->rn_off]))
128			x = x->rn_r;
129		else
130			x = x->rn_l;
131	}
132	return x;
133}
134
135int
136rn_refines(void* m_arg, void *n_arg)
137{
138	caddr_t m = m_arg, n = n_arg;
139	caddr_t lim, lim2 = lim = n + *(u_char *)n;
140	int longer = (*(u_char *)n++) - (int)(*(u_char *)m++);
141	int masks_are_equal = 1;
142
143	if (longer > 0)
144		lim -= longer;
145	while (n < lim) {
146		if (*n & ~(*m))
147			return 0;
148		if (*n++ != *m++)
149			masks_are_equal = 0;
150	}
151	while (n < lim2)
152		if (*n++)
153			return 0;
154	if (masks_are_equal && (longer < 0))
155		for (lim2 = m - longer; m < lim2; )
156			if (*m++)
157				return 1;
158	return (!masks_are_equal);
159}
160
161struct radix_node *
162rn_lookup(void *v_arg, void *m_arg, struct radix_node_head *head)
163{
164	struct radix_node *x;
165	caddr_t netmask = 0;
166
167	if (m_arg) {
168		if ((x = rn_addmask(m_arg, 1, head->rnh_treetop->rn_off)) == 0)
169			return (0);
170		netmask = x->rn_key;
171	}
172	x = rn_match(v_arg, head);
173	if (x && netmask) {
174		while (x && x->rn_mask != netmask)
175			x = x->rn_dupedkey;
176	}
177	return x;
178}
179
180static int
181rn_satisfies_leaf(char *trial,
182		  struct radix_node *leaf,
183		  int skip)
184{
185	char *cp = trial, *cp2 = leaf->rn_key, *cp3 = leaf->rn_mask;
186	char *cplim;
187	int length = min(*(u_char *)cp, *(u_char *)cp2);
188
189	if (cp3 == 0)
190		cp3 = rn_ones;
191	else
192		length = min(length, *(u_char *)cp3);
193	cplim = cp + length; cp3 += skip; cp2 += skip;
194	for (cp += skip; cp < cplim; cp++, cp2++, cp3++)
195		if ((*cp ^ *cp2) & *cp3)
196			return 0;
197	return 1;
198}
199
200struct radix_node *
201rn_match(void *v_arg,
202	 struct radix_node_head *head)
203{
204	caddr_t v = v_arg;
205	struct radix_node *t = head->rnh_treetop, *x;
206	caddr_t cp = v, cp2;
207	caddr_t cplim;
208	struct radix_node *saved_t, *top = t;
209	int off = t->rn_off, vlen = *(u_char *)cp, matched_off;
210	int test, b, rn_b;
211
212	/*
213	 * Open code rn_search(v, top) to avoid overhead of extra
214	 * subroutine call.
215	 */
216	for (; t->rn_b >= 0; ) {
217		if (t->rn_bmask & cp[t->rn_off])
218			t = t->rn_r;
219		else
220			t = t->rn_l;
221	}
222	/*
223	 * See if we match exactly as a host destination
224	 * or at least learn how many bits match, for normal mask finesse.
225	 *
226	 * It doesn't hurt us to limit how many bytes to check
227	 * to the length of the mask, since if it matches we had a genuine
228	 * match and the leaf we have is the most specific one anyway;
229	 * if it didn't match with a shorter length it would fail
230	 * with a long one.  This wins big for class B&C netmasks which
231	 * are probably the most common case...
232	 */
233	if (t->rn_mask)
234		vlen = *(u_char *)t->rn_mask;
235	cp += off; cp2 = t->rn_key + off; cplim = v + vlen;
236	for (; cp < cplim; cp++, cp2++)
237		if (*cp != *cp2)
238			goto on1;
239	/*
240	 * This extra grot is in case we are explicitly asked
241	 * to look up the default.  Ugh!
242	 * Or 255.255.255.255
243	 *
244	 * In this case, we have a complete match of the key.  Unless
245	 * the node is one of the roots, we are finished.
246	 * If it is the zeros root, then take what we have, prefering
247	 * any real data.
248	 * If it is the ones root, then pretend the target key was followed
249	 * by a byte of zeros.
250	 */
251	if (!(t->rn_flags & RNF_ROOT))
252		return t;		/* not a root */
253	if (t->rn_dupedkey) {
254		t = t->rn_dupedkey;
255		return t;		/* have some real data */
256	}
257	if (*(cp-1) == 0)
258		return t;		/* not the ones root */
259	b = 0;				/* fake a zero after 255.255.255.255 */
260	goto on2;
261on1:
262	test = (*cp ^ *cp2) & 0xff; /* find first bit that differs */
263	for (b = 7; (test >>= 1) > 0;)
264		b--;
265on2:
266	matched_off = cp - v;
267	b += matched_off << 3;
268	rn_b = -1 - b;
269	/*
270	 * If there is a host route in a duped-key chain, it will be first.
271	 */
272	if ((saved_t = t)->rn_mask == 0)
273		t = t->rn_dupedkey;
274	for (; t; t = t->rn_dupedkey) {
275		/*
276		 * Even if we don't match exactly as a host,
277		 * we may match if the leaf we wound up at is
278		 * a route to a net.
279		 */
280		if (t->rn_flags & RNF_NORMAL) {
281			if (rn_b <= t->rn_b)
282				return t;
283		} else if (rn_satisfies_leaf(v, t, matched_off)) {
284			return t;
285		}
286	}
287	t = saved_t;
288	/* start searching up the tree */
289	do {
290		struct radix_mask *m;
291		t = t->rn_p;
292		if ((m = t->rn_mklist)) {
293			/*
294			 * If non-contiguous masks ever become important
295			 * we can restore the masking and open coding of
296			 * the search and satisfaction test and put the
297			 * calculation of "off" back before the "do".
298			 */
299			do {
300				if (m->rm_flags & RNF_NORMAL) {
301					if (rn_b <= m->rm_b)
302						return (m->rm_leaf);
303				} else {
304					off = min(t->rn_off, matched_off);
305					x = rn_search_m(v, t, m->rm_mask);
306					while (x && x->rn_mask != m->rm_mask)
307						x = x->rn_dupedkey;
308					if (x && rn_satisfies_leaf(v, x, off))
309						    return x;
310				}
311			} while ((m = m->rm_mklist));
312		}
313	} while (t != top);
314	return 0;
315}
316
317#ifdef RN_DEBUG
318int	rn_nodenum;
319struct	radix_node *rn_clist;
320int	rn_saveinfo;
321int	rn_debug =  1;
322#endif
323
324struct radix_node *
325rn_newpair(void *v, int b, struct radix_node nodes[2])
326{
327	struct radix_node *tt = nodes, *t = tt + 1;
328	t->rn_b = b; t->rn_bmask = 0x80 >> (b & 7);
329	t->rn_l = tt; t->rn_off = b >> 3;
330	tt->rn_b = -1; tt->rn_key = (caddr_t)v; tt->rn_p = t;
331	tt->rn_flags = t->rn_flags = RNF_ACTIVE;
332#ifdef RN_DEBUG
333	tt->rn_info = rn_nodenum++; t->rn_info = rn_nodenum++;
334	tt->rn_twin = t; tt->rn_ybro = rn_clist; rn_clist = tt;
335#endif
336	return t;
337}
338
339struct radix_node *
340rn_insert(void* v_arg,
341	  struct radix_node_head *head,
342	  int *dupentry,
343	  struct radix_node nodes[2])
344{
345	caddr_t v = v_arg;
346	struct radix_node *top = head->rnh_treetop;
347	int head_off = top->rn_off, vlen = (int)*((u_char *)v);
348	struct radix_node *t = rn_search(v_arg, top);
349	caddr_t cp = v + head_off;
350	int b;
351	struct radix_node *tt;
352
353	/*
354	 * Find first bit at which v and t->rn_key differ
355	 */
356    {
357		caddr_t cp2 = t->rn_key + head_off;
358		int cmp_res;
359	caddr_t cplim = v + vlen;
360
361	while (cp < cplim)
362		if (*cp2++ != *cp++)
363			goto on1;
364	/* handle adding 255.255.255.255 */
365	if (!(t->rn_flags & RNF_ROOT) || *(cp2-1) == 0) {
366		*dupentry = 1;
367		return t;
368	}
369on1:
370	*dupentry = 0;
371	cmp_res = (cp[-1] ^ cp2[-1]) & 0xff;
372	for (b = (cp - v) << 3; cmp_res; b--)
373		cmp_res >>= 1;
374    }
375    {
376	    struct radix_node *p, *x = top;
377	cp = v;
378	do {
379		p = x;
380		if (cp[x->rn_off] & x->rn_bmask)
381			x = x->rn_r;
382		else x = x->rn_l;
383	} while ((unsigned)b > (unsigned)x->rn_b);
384#ifdef RN_DEBUG
385	if (rn_debug)
386		log(LOG_DEBUG, "rn_insert: Going In:\n"), traverse(p);
387#endif
388	t = rn_newpair(v_arg, b, nodes); tt = t->rn_l;
389	if ((cp[p->rn_off] & p->rn_bmask) == 0)
390		p->rn_l = t;
391	else
392		p->rn_r = t;
393	x->rn_p = t; t->rn_p = p; /* frees x, p as temp vars below */
394	if ((cp[t->rn_off] & t->rn_bmask) == 0) {
395		t->rn_r = x;
396	} else {
397		t->rn_r = tt; t->rn_l = x;
398	}
399#ifdef RN_DEBUG
400	if (rn_debug)
401		log(LOG_DEBUG, "rn_insert: Coming Out:\n"), traverse(p);
402#endif
403    }
404	return (tt);
405}
406
407struct radix_node *
408rn_addmask(void *n_arg, int search, int skip)
409{
410	caddr_t netmask = (caddr_t)n_arg;
411	struct radix_node *x;
412	caddr_t cp, cplim;
413	int b = 0, mlen, j;
414	int maskduplicated, m0, isnormal;
415	struct radix_node *saved_x;
416	static int last_zeroed = 0;
417
418	if ((mlen = *(u_char *)netmask) > max_keylen)
419		mlen = max_keylen;
420	if (skip == 0)
421		skip = 1;
422	if (mlen <= skip)
423		return (mask_rnhead->rnh_nodes);
424	if (skip > 1)
425		Bcopy(rn_ones + 1, addmask_key + 1, skip - 1);
426	if ((m0 = mlen) > skip)
427		Bcopy(netmask + skip, addmask_key + skip, mlen - skip);
428	/*
429	 * Trim trailing zeroes.
430	 */
431	for (cp = addmask_key + mlen; (cp > addmask_key) && cp[-1] == 0;)
432		cp--;
433	mlen = cp - addmask_key;
434	if (mlen <= skip) {
435		if (m0 >= last_zeroed)
436			last_zeroed = mlen;
437		return (mask_rnhead->rnh_nodes);
438	}
439	if (m0 < last_zeroed)
440		Bzero(addmask_key + m0, last_zeroed - m0);
441	*addmask_key = last_zeroed = mlen;
442	x = rn_search(addmask_key, rn_masktop);
443	if (Bcmp(addmask_key, x->rn_key, mlen) != 0)
444		x = 0;
445	if (x || search)
446		return (x);
447	x = (struct radix_node *)rtmalloc(max_keylen + 2*sizeof(*x),
448					  "rn_addmask");
449	saved_x = x;
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	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(struct radix_node *tt,
494		  struct radix_mask *next)
495{
496	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	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	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 0x%lx at 0x%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	    int (*f)(struct radix_node *, struct walkarg *),
813	    struct walkarg *w)
814{
815	int error;
816	struct radix_node *base, *next;
817	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(struct radix_node_head **head, int off)
850{
851	struct radix_node_head *rnh;
852	struct radix_node *t, *tt, *ttt;
853	if (*head)
854		return (1);
855	rnh = (struct radix_node_head *)rtmalloc(sizeof(*rnh), "rn_inithead");
856	Bzero(rnh, sizeof (*rnh));
857	*head = rnh;
858	t = rn_newpair(rn_zeros, off, rnh->rnh_nodes);
859	ttt = rnh->rnh_nodes + 2;
860	t->rn_r = ttt;
861	t->rn_p = t;
862	tt = t->rn_l;
863	tt->rn_flags = t->rn_flags = RNF_ROOT | RNF_ACTIVE;
864	tt->rn_b = -1 - off;
865	*ttt = *tt;
866	ttt->rn_key = rn_ones;
867	rnh->rnh_addaddr = rn_addroute;
868	rnh->rnh_deladdr = rn_delete;
869	rnh->rnh_matchaddr = rn_match;
870	rnh->rnh_lookup = rn_lookup;
871	rnh->rnh_walktree = rn_walktree;
872	rnh->rnh_treetop = t;
873	return (1);
874}
875
876void
877rn_init(void)
878{
879	char *cp, *cplim;
880	if (max_keylen == 0) {
881		printf("rn_init: radix functions require max_keylen be set\n");
882		return;
883	}
884	rn_zeros = (char *)rtmalloc(3 * max_keylen, "rn_init");
885	Bzero(rn_zeros, 3 * max_keylen);
886	rn_ones = cp = rn_zeros + max_keylen;
887	addmask_key = cplim = rn_ones + max_keylen;
888	while (cp < cplim)
889		*cp++ = -1;
890	if (rn_inithead(&mask_rnhead, 0) == 0)
891		panic("rn_init 2");
892}
893
894