1/*	$FreeBSD$	*/
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 *
16 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 *
28 *	@(#)radix.c	8.6 (Berkeley) 10/17/95
29 */
30
31/*
32 * Routines to build and maintain radix trees for routing lookups.
33 */
34#if defined(KERNEL) || defined(_KERNEL)
35# undef KERNEL
36# undef _KERNEL
37# define        KERNEL  1
38# define        _KERNEL 1
39#endif
40#define __SYS_ATOMIC_OPS_H__
41#if !defined(__svr4__) && !defined(__SVR4) && !defined(__osf__) && \
42    !defined(__hpux) && !defined(__sgi)
43#include <sys/cdefs.h>
44#endif
45#ifndef __P
46# ifdef __STDC__
47#  define       __P(x)  x
48# else
49#  define       __P(x)  ()
50# endif
51#endif
52#ifdef __osf__
53# define CONST
54# define _IPV6_SWTAB_H
55# define _PROTO_NET_H_
56# define _PROTO_IPV6_H
57# include <sys/malloc.h>
58#endif
59
60#include <sys/param.h>
61#ifdef	_KERNEL
62#include <sys/systm.h>
63#else
64void panic __P((char *str));
65#include <stdlib.h>
66#include <stdio.h>
67#include <stdarg.h>
68#include <string.h>
69#endif
70#ifdef __hpux
71#include <syslog.h>
72#else
73#include <sys/syslog.h>
74#endif
75#include <sys/time.h>
76#include <netinet/in.h>
77#include <sys/socket.h>
78#include <net/if.h>
79#ifdef SOLARIS2
80# define _RADIX_H_
81#endif
82#include "netinet/ip_compat.h"
83#include "netinet/ip_fil.h"
84#ifdef SOLARIS2
85# undef _RADIX_H_
86#endif
87/* END OF INCLUDES */
88#include "radix_ipf.h"
89#ifndef min
90# define	min	MIN
91#endif
92#ifndef max
93# define	max	MAX
94#endif
95
96int	max_keylen = 16;
97static struct radix_mask *rn_mkfreelist;
98static struct radix_node_head *mask_rnhead;
99static char *addmask_key;
100static u_char normal_chars[] = {0, 0x80, 0xc0, 0xe0, 0xf0, 0xf8, 0xfc, 0xfe, 0xff};
101static char *rn_zeros = NULL, *rn_ones = NULL;
102
103#define rn_masktop (mask_rnhead->rnh_treetop)
104#undef Bcmp
105#define Bcmp(a, b, l) (l == 0 ? 0 : bcmp((caddr_t)(a), (caddr_t)(b), (u_long)l))
106
107static int rn_satisfies_leaf __P((char *, struct radix_node *, int));
108static int rn_lexobetter __P((void *, void *));
109static struct radix_mask *rn_new_radix_mask __P((struct radix_node *,
110    struct radix_mask *));
111static int rn_freenode __P((struct radix_node *, void *));
112#if defined(AIX) && !defined(_KERNEL)
113struct radix_node *rn_match __P((void *, struct radix_node_head *));
114struct radix_node *rn_addmask __P((int, int, void *));
115#define	FreeS(x, y)	KFREES(x, y)
116#define	Bcopy(x, y, z)	bcopy(x, y, z)
117#endif
118
119/*
120 * The data structure for the keys is a radix tree with one way
121 * branching removed.  The index rn_b at an internal node n represents a bit
122 * position to be tested.  The tree is arranged so that all descendants
123 * of a node n have keys whose bits all agree up to position rn_b - 1.
124 * (We say the index of n is rn_b.)
125 *
126 * There is at least one descendant which has a one bit at position rn_b,
127 * and at least one with a zero there.
128 *
129 * A route is determined by a pair of key and mask.  We require that the
130 * bit-wise logical and of the key and mask to be the key.
131 * We define the index of a route to associated with the mask to be
132 * the first bit number in the mask where 0 occurs (with bit number 0
133 * representing the highest order bit).
134 *
135 * We say a mask is normal if every bit is 0, past the index of the mask.
136 * If a node n has a descendant (k, m) with index(m) == index(n) == rn_b,
137 * and m is a normal mask, then the route applies to every descendant of n.
138 * If the index(m) < rn_b, this implies the trailing last few bits of k
139 * before bit b are all 0, (and hence consequently true of every descendant
140 * of n), so the route applies to all descendants of the node as well.
141 *
142 * Similar logic shows that a non-normal mask m such that
143 * index(m) <= index(n) could potentially apply to many children of n.
144 * Thus, for each non-host route, we attach its mask to a list at an internal
145 * node as high in the tree as we can go.
146 *
147 * The present version of the code makes use of normal routes in short-
148 * circuiting an explicit mask and compare operation when testing whether
149 * a key satisfies a normal route, and also in remembering the unique leaf
150 * that governs a subtree.
151 */
152
153struct radix_node *
154rn_search(v_arg, head)
155	void *v_arg;
156	struct radix_node *head;
157{
158	struct radix_node *x;
159	caddr_t v;
160
161	for (x = head, v = v_arg; x->rn_b >= 0;) {
162		if (x->rn_bmask & v[x->rn_off])
163			x = x->rn_r;
164		else
165			x = x->rn_l;
166	}
167	return (x);
168}
169
170struct radix_node *
171rn_search_m(v_arg, head, m_arg)
172	struct radix_node *head;
173	void *v_arg, *m_arg;
174{
175	struct radix_node *x;
176	caddr_t v = v_arg, m = m_arg;
177
178	for (x = head; x->rn_b >= 0;) {
179		if ((x->rn_bmask & m[x->rn_off]) &&
180		    (x->rn_bmask & v[x->rn_off]))
181			x = x->rn_r;
182		else
183			x = x->rn_l;
184	}
185	return x;
186}
187
188int
189rn_refines(m_arg, n_arg)
190	void *m_arg, *n_arg;
191{
192	caddr_t m = m_arg, n = n_arg;
193	caddr_t lim, lim2 = lim = n + *(u_char *)n;
194	int longer = (*(u_char *)n++) - (int)(*(u_char *)m++);
195	int masks_are_equal = 1;
196
197	if (longer > 0)
198		lim -= longer;
199	while (n < lim) {
200		if (*n & ~(*m))
201			return 0;
202		if (*n++ != *m++)
203			masks_are_equal = 0;
204	}
205	while (n < lim2)
206		if (*n++)
207			return 0;
208	if (masks_are_equal && (longer < 0))
209		for (lim2 = m - longer; m < lim2; )
210			if (*m++)
211				return 1;
212	return (!masks_are_equal);
213}
214
215struct radix_node *
216rn_lookup(v_arg, m_arg, head)
217	void *v_arg, *m_arg;
218	struct radix_node_head *head;
219{
220	struct radix_node *x;
221	caddr_t netmask = 0;
222
223	if (m_arg) {
224		if ((x = rn_addmask(m_arg, 1, head->rnh_treetop->rn_off)) == 0)
225			return (0);
226		netmask = x->rn_key;
227	}
228	x = rn_match(v_arg, head);
229	if (x && netmask) {
230		while (x && x->rn_mask != netmask)
231			x = x->rn_dupedkey;
232	}
233	return x;
234}
235
236static int
237rn_satisfies_leaf(trial, leaf, skip)
238	char *trial;
239	struct radix_node *leaf;
240	int skip;
241{
242	char *cp = trial, *cp2 = leaf->rn_key, *cp3 = leaf->rn_mask;
243	char *cplim;
244	int length = min(*(u_char *)cp, *(u_char *)cp2);
245
246	if (cp3 == 0)
247		cp3 = rn_ones;
248	else
249		length = min(length, *(u_char *)cp3);
250	cplim = cp + length;
251	cp3 += skip;
252	cp2 += skip;
253	for (cp += skip; cp < cplim; cp++, cp2++, cp3++)
254		if ((*cp ^ *cp2) & *cp3)
255			return 0;
256	return 1;
257}
258
259struct radix_node *
260rn_match(v_arg, head)
261	void *v_arg;
262	struct radix_node_head *head;
263{
264	caddr_t v = v_arg;
265	struct radix_node *t = head->rnh_treetop, *x;
266	caddr_t cp = v, cp2;
267	caddr_t cplim;
268	struct radix_node *saved_t, *top = t;
269	int off = t->rn_off, vlen = *(u_char *)cp, matched_off;
270	int test, b, rn_b;
271
272	/*
273	 * Open code rn_search(v, top) to avoid overhead of extra
274	 * subroutine call.
275	 */
276	for (; t->rn_b >= 0; ) {
277		if (t->rn_bmask & cp[t->rn_off])
278			t = t->rn_r;
279		else
280			t = t->rn_l;
281	}
282	/*
283	 * See if we match exactly as a host destination
284	 * or at least learn how many bits match, for normal mask finesse.
285	 *
286	 * It doesn't hurt us to limit how many bytes to check
287	 * to the length of the mask, since if it matches we had a genuine
288	 * match and the leaf we have is the most specific one anyway;
289	 * if it didn't match with a shorter length it would fail
290	 * with a long one.  This wins big for class B&C netmasks which
291	 * are probably the most common case...
292	 */
293	if (t->rn_mask)
294		vlen = *(u_char *)t->rn_mask;
295	cp += off;
296	cp2 = t->rn_key + off;
297	cplim = v + vlen;
298	for (; cp < cplim; cp++, cp2++)
299		if (*cp != *cp2)
300			goto on1;
301	/*
302	 * This extra grot is in case we are explicitly asked
303	 * to look up the default.  Ugh!
304	 */
305	if ((t->rn_flags & RNF_ROOT) && t->rn_dupedkey)
306		t = t->rn_dupedkey;
307	return t;
308on1:
309	test = (*cp ^ *cp2) & 0xff; /* find first bit that differs */
310	for (b = 7; (test >>= 1) > 0;)
311		b--;
312	matched_off = cp - v;
313	b += matched_off << 3;
314	rn_b = -1 - b;
315	/*
316	 * If there is a host route in a duped-key chain, it will be first.
317	 */
318	if ((saved_t = t)->rn_mask == 0)
319		t = t->rn_dupedkey;
320	for (; t; t = t->rn_dupedkey)
321		/*
322		 * Even if we don't match exactly as a host,
323		 * we may match if the leaf we wound up at is
324		 * a route to a net.
325		 */
326		if (t->rn_flags & RNF_NORMAL) {
327			if (rn_b <= t->rn_b)
328				return t;
329		} else if (rn_satisfies_leaf(v, t, matched_off))
330				return t;
331	t = saved_t;
332	/* start searching up the tree */
333	do {
334		struct radix_mask *m;
335		t = t->rn_p;
336		m = t->rn_mklist;
337		if (m) {
338			/*
339			 * If non-contiguous masks ever become important
340			 * we can restore the masking and open coding of
341			 * the search and satisfaction test and put the
342			 * calculation of "off" back before the "do".
343			 */
344			do {
345				if (m->rm_flags & RNF_NORMAL) {
346					if (rn_b <= m->rm_b)
347						return (m->rm_leaf);
348				} else {
349					off = min(t->rn_off, matched_off);
350					x = rn_search_m(v, t, m->rm_mask);
351					while (x && x->rn_mask != m->rm_mask)
352						x = x->rn_dupedkey;
353					if (x && rn_satisfies_leaf(v, x, off))
354						return x;
355				}
356				m = m->rm_mklist;
357			} while (m);
358		}
359	} while (t != top);
360	return 0;
361}
362
363#ifdef RN_DEBUG
364int	rn_nodenum;
365struct	radix_node *rn_clist;
366int	rn_saveinfo;
367int	rn_debug =  1;
368#endif
369
370struct radix_node *
371rn_newpair(v, b, nodes)
372	void *v;
373	int b;
374	struct radix_node nodes[2];
375{
376	struct radix_node *tt = nodes, *t = tt + 1;
377	t->rn_b = b;
378	t->rn_bmask = 0x80 >> (b & 7);
379	t->rn_l = tt;
380	t->rn_off = b >> 3;
381	tt->rn_b = -1;
382	tt->rn_key = (caddr_t)v;
383	tt->rn_p = t;
384	tt->rn_flags = t->rn_flags = RNF_ACTIVE;
385#ifdef RN_DEBUG
386	tt->rn_info = rn_nodenum++;
387	t->rn_info = rn_nodenum++;
388	tt->rn_twin = t;
389	tt->rn_ybro = rn_clist;
390	rn_clist = tt;
391#endif
392	return t;
393}
394
395struct radix_node *
396rn_insert(v_arg, head, dupentry, nodes)
397	void *v_arg;
398	struct radix_node_head *head;
399	int *dupentry;
400	struct radix_node nodes[2];
401{
402	caddr_t v = v_arg;
403	struct radix_node *top = head->rnh_treetop;
404	int head_off = top->rn_off, vlen = (int)*((u_char *)v);
405	struct radix_node *t = rn_search(v_arg, top);
406	caddr_t cp = v + head_off;
407	int b;
408	struct radix_node *tt;
409
410#ifdef	RN_DEBUG
411	if (rn_debug)
412		log(LOG_DEBUG, "rn_insert(%p,%p,%p,%p)\n", v_arg, head, dupentry, nodes);
413#endif
414    	/*
415	 * Find first bit at which v and t->rn_key differ
416	 */
417    {
418	caddr_t cp2 = t->rn_key + head_off;
419	int cmp_res;
420	caddr_t cplim = v + vlen;
421
422	while (cp < cplim)
423		if (*cp2++ != *cp++)
424			goto on1;
425	*dupentry = 1;
426	return t;
427on1:
428	*dupentry = 0;
429	cmp_res = (cp[-1] ^ cp2[-1]) & 0xff;
430	for (b = (cp - v) << 3; cmp_res; b--)
431		cmp_res >>= 1;
432    }
433    {
434	struct radix_node *p, *x = top;
435	cp = v;
436	do {
437		p = x;
438		if (cp[x->rn_off] & x->rn_bmask)
439			x = x->rn_r;
440		else
441			x = x->rn_l;
442	} while (b > (unsigned) x->rn_b); /* x->rn_b < b && x->rn_b >= 0 */
443#ifdef RN_DEBUG
444	if (rn_debug)
445		log(LOG_DEBUG, "rn_insert: Going In:\n"); // traverse(p);
446#endif
447	t = rn_newpair(v_arg, b, nodes);
448	tt = t->rn_l;
449	if ((cp[p->rn_off] & p->rn_bmask) == 0)
450		p->rn_l = t;
451	else
452		p->rn_r = t;
453	x->rn_p = t;
454	t->rn_p = p; /* frees x, p as temp vars below */
455	if ((cp[t->rn_off] & t->rn_bmask) == 0) {
456		t->rn_r = x;
457	} else {
458		t->rn_r = tt;
459		t->rn_l = x;
460	}
461#ifdef RN_DEBUG
462	if (rn_debug)
463		log(LOG_DEBUG, "rn_insert: Coming Out:\n"); // traverse(p);
464#endif
465    }
466	return (tt);
467}
468
469struct radix_node *
470rn_addmask(n_arg, search, skip)
471	int search, skip;
472	void *n_arg;
473{
474	caddr_t netmask = (caddr_t)n_arg;
475	struct radix_node *x;
476	caddr_t cp, cplim;
477	int b = 0, mlen, j;
478	int maskduplicated, m0, isnormal;
479	struct radix_node *saved_x;
480	static int last_zeroed = 0;
481
482#ifdef	RN_DEBUG
483	if (rn_debug)
484		log(LOG_DEBUG, "rn_addmask(%p,%d,%d)\n", n_arg, search, skip);
485#endif
486	mlen = *(u_char *)netmask;
487	if ((mlen = *(u_char *)netmask) > max_keylen)
488		mlen = max_keylen;
489	if (skip == 0)
490		skip = 1;
491	if (mlen <= skip)
492		return (mask_rnhead->rnh_nodes);
493	if (skip > 1)
494		Bcopy(rn_ones + 1, addmask_key + 1, skip - 1);
495	if ((m0 = mlen) > skip)
496		Bcopy(netmask + skip, addmask_key + skip, mlen - skip);
497	/*
498	 * Trim trailing zeroes.
499	 */
500	for (cp = addmask_key + mlen; (cp > addmask_key) && cp[-1] == 0;)
501		cp--;
502	mlen = cp - addmask_key;
503	if (mlen <= skip) {
504		if (m0 >= last_zeroed)
505			last_zeroed = mlen;
506		return (mask_rnhead->rnh_nodes);
507	}
508	if (m0 < last_zeroed)
509		Bzero(addmask_key + m0, last_zeroed - m0);
510	*addmask_key = last_zeroed = mlen;
511	x = rn_search(addmask_key, rn_masktop);
512	if (Bcmp(addmask_key, x->rn_key, mlen) != 0)
513		x = 0;
514	if (x || search)
515		return (x);
516	R_Malloc(x, struct radix_node *, max_keylen + 2 * sizeof (*x));
517	if ((saved_x = x) == 0)
518		return (0);
519	Bzero(x, max_keylen + 2 * sizeof (*x));
520	netmask = cp = (caddr_t)(x + 2);
521	Bcopy(addmask_key, cp, mlen);
522	x = rn_insert(cp, mask_rnhead, &maskduplicated, x);
523	if (maskduplicated) {
524#if 0
525		log(LOG_ERR, "rn_addmask: mask impossibly already in tree\n");
526#endif
527		Free(saved_x);
528		return (x);
529	}
530	/*
531	 * Calculate index of mask, and check for normalcy.
532	 */
533	cplim = netmask + mlen;
534	isnormal = 1;
535	for (cp = netmask + skip; (cp < cplim) && *(u_char *)cp == 0xff;)
536		cp++;
537	if (cp != cplim) {
538		for (j = 0x80; (j & *cp) != 0; j >>= 1)
539			b++;
540		if (*cp != normal_chars[b] || cp != (cplim - 1))
541			isnormal = 0;
542	}
543	b += (cp - netmask) << 3;
544	x->rn_b = -1 - b;
545	if (isnormal)
546		x->rn_flags |= RNF_NORMAL;
547	return (x);
548}
549
550static int	/* XXX: arbitrary ordering for non-contiguous masks */
551rn_lexobetter(m_arg, n_arg)
552	void *m_arg, *n_arg;
553{
554	u_char *mp = m_arg, *np = n_arg, *lim;
555
556	if (*mp > *np)
557		return 1;  /* not really, but need to check longer one first */
558	if (*mp == *np)
559		for (lim = mp + *mp; mp < lim;)
560			if (*mp++ > *np++)
561				return 1;
562	return 0;
563}
564
565static struct radix_mask *
566rn_new_radix_mask(tt, next)
567	struct radix_node *tt;
568	struct radix_mask *next;
569{
570	struct radix_mask *m;
571
572	MKGet(m);
573	if (m == 0) {
574#if 0
575		log(LOG_ERR, "Mask for route not entered\n");
576#endif
577		return (0);
578	}
579	Bzero(m, sizeof *m);
580	m->rm_b = tt->rn_b;
581	m->rm_flags = tt->rn_flags;
582	if (tt->rn_flags & RNF_NORMAL)
583		m->rm_leaf = tt;
584	else
585		m->rm_mask = tt->rn_mask;
586	m->rm_mklist = next;
587	tt->rn_mklist = m;
588	return m;
589}
590
591struct radix_node *
592rn_addroute(v_arg, n_arg, head, treenodes)
593	void *v_arg, *n_arg;
594	struct radix_node_head *head;
595	struct radix_node treenodes[2];
596{
597	caddr_t v = (caddr_t)v_arg, netmask = (caddr_t)n_arg;
598	struct radix_node *t, *x = NULL, *tt;
599	struct radix_node *saved_tt, *top = head->rnh_treetop;
600	short b = 0, b_leaf = 0;
601	int keyduplicated;
602	caddr_t mmask;
603	struct radix_mask *m, **mp;
604
605#ifdef	RN_DEBUG
606	if (rn_debug)
607		log(LOG_DEBUG, "rn_addroute(%p,%p,%p,%p)\n", v_arg, n_arg, head, treenodes);
608#endif
609	/*
610	 * In dealing with non-contiguous masks, there may be
611	 * many different routes which have the same mask.
612	 * We will find it useful to have a unique pointer to
613	 * the mask to speed avoiding duplicate references at
614	 * nodes and possibly save time in calculating indices.
615	 */
616	if (netmask) {
617		if ((x = rn_addmask(netmask, 0, top->rn_off)) == 0)
618			return (0);
619		b_leaf = x->rn_b;
620		b = -1 - x->rn_b;
621		netmask = x->rn_key;
622	}
623	/*
624	 * Deal with duplicated keys: attach node to previous instance
625	 */
626	saved_tt = tt = rn_insert(v, head, &keyduplicated, treenodes);
627	if (keyduplicated) {
628		for (t = tt; tt; t = tt, tt = tt->rn_dupedkey) {
629			if (tt->rn_mask == netmask)
630				return (0);
631			if (netmask == 0 ||
632			    (tt->rn_mask &&
633			     ((b_leaf < tt->rn_b) || /* index(netmask) > node */
634			       rn_refines(netmask, tt->rn_mask) ||
635			       rn_lexobetter(netmask, tt->rn_mask))))
636				break;
637		}
638		/*
639		 * If the mask is not duplicated, we wouldn't
640		 * find it among possible duplicate key entries
641		 * anyway, so the above test doesn't hurt.
642		 *
643		 * We sort the masks for a duplicated key the same way as
644		 * in a masklist -- most specific to least specific.
645		 * This may require the unfortunate nuisance of relocating
646		 * the head of the list.
647		 *
648		 * We also reverse, or doubly link the list through the
649		 * parent pointer.
650		 */
651		if (tt == saved_tt) {
652			struct	radix_node *xx = x;
653			/* link in at head of list */
654			(tt = treenodes)->rn_dupedkey = t;
655			tt->rn_flags = t->rn_flags;
656			tt->rn_p = x = t->rn_p;
657			t->rn_p = tt;
658			if (x->rn_l == t)
659				x->rn_l = tt;
660			else
661				x->rn_r = tt;
662			saved_tt = tt;
663			x = xx;
664		} else {
665			(tt = treenodes)->rn_dupedkey = t->rn_dupedkey;
666			t->rn_dupedkey = tt;
667			tt->rn_p = t;
668			if (tt->rn_dupedkey)
669				tt->rn_dupedkey->rn_p = tt;
670		}
671#ifdef RN_DEBUG
672		t=tt+1;
673		tt->rn_info = rn_nodenum++;
674		t->rn_info = rn_nodenum++;
675		tt->rn_twin = t;
676		tt->rn_ybro = rn_clist;
677		rn_clist = tt;
678#endif
679		tt->rn_key = (caddr_t) v;
680		tt->rn_b = -1;
681		tt->rn_flags = RNF_ACTIVE;
682	}
683	/*
684	 * Put mask in tree.
685	 */
686	if (netmask) {
687		tt->rn_mask = netmask;
688		tt->rn_b = x->rn_b;
689		tt->rn_flags |= x->rn_flags & RNF_NORMAL;
690	}
691	t = saved_tt->rn_p;
692	if (keyduplicated)
693		goto on2;
694	b_leaf = -1 - t->rn_b;
695	if (t->rn_r == saved_tt)
696		x = t->rn_l;
697	else
698		x = t->rn_r;
699	/* Promote general routes from below */
700	if (x->rn_b < 0) {
701	    for (mp = &t->rn_mklist; x; x = x->rn_dupedkey)
702		if (x->rn_mask && (x->rn_b >= b_leaf) && x->rn_mklist == 0) {
703			*mp = m = rn_new_radix_mask(x, 0);
704			if (m)
705				mp = &m->rm_mklist;
706		}
707	} else if (x->rn_mklist) {
708		/*
709		 * Skip over masks whose index is > that of new node
710		 */
711		for (mp = &x->rn_mklist; (m = *mp) != NULL; mp = &m->rm_mklist)
712			if (m->rm_b >= b_leaf)
713				break;
714		t->rn_mklist = m;
715		*mp = 0;
716	}
717on2:
718	/* Add new route to highest possible ancestor's list */
719	if ((netmask == 0) || (b > t->rn_b ))
720		return tt; /* can't lift at all */
721	b_leaf = tt->rn_b;
722	do {
723		x = t;
724		t = t->rn_p;
725	} while (b <= t->rn_b && x != top);
726	/*
727	 * Search through routes associated with node to
728	 * insert new route according to index.
729	 * Need same criteria as when sorting dupedkeys to avoid
730	 * double loop on deletion.
731	 */
732	for (mp = &x->rn_mklist; (m = *mp) != NULL; mp = &m->rm_mklist) {
733		if (m->rm_b < b_leaf)
734			continue;
735		if (m->rm_b > b_leaf)
736			break;
737		if (m->rm_flags & RNF_NORMAL) {
738			mmask = m->rm_leaf->rn_mask;
739			if (tt->rn_flags & RNF_NORMAL) {
740#if 0
741				log(LOG_ERR, "Non-unique normal route,"
742				    " mask not entered\n");
743#endif
744				return tt;
745			}
746		} else
747			mmask = m->rm_mask;
748		if (mmask == netmask) {
749			m->rm_refs++;
750			tt->rn_mklist = m;
751			return tt;
752		}
753		if (rn_refines(netmask, mmask)
754		    || rn_lexobetter(netmask, mmask))
755			break;
756	}
757	*mp = rn_new_radix_mask(tt, *mp);
758	return tt;
759}
760
761struct radix_node *
762rn_delete(v_arg, netmask_arg, head)
763	void *v_arg, *netmask_arg;
764	struct radix_node_head *head;
765{
766	struct radix_node *t, *p, *x, *tt;
767	struct radix_mask *m, *saved_m, **mp;
768	struct radix_node *dupedkey, *saved_tt, *top;
769	caddr_t v, netmask;
770	int b, head_off, vlen;
771
772	v = v_arg;
773	netmask = netmask_arg;
774	x = head->rnh_treetop;
775	tt = rn_search(v, x);
776	head_off = x->rn_off;
777	vlen =  *(u_char *)v;
778	saved_tt = tt;
779	top = x;
780	if (tt == 0 ||
781	    Bcmp(v + head_off, tt->rn_key + head_off, vlen - head_off))
782		return (0);
783	/*
784	 * Delete our route from mask lists.
785	 */
786	if (netmask) {
787		if ((x = rn_addmask(netmask, 1, head_off)) == 0)
788			return (0);
789		netmask = x->rn_key;
790		while (tt->rn_mask != netmask)
791			if ((tt = tt->rn_dupedkey) == 0)
792				return (0);
793	}
794	if (tt->rn_mask == 0 || (saved_m = m = tt->rn_mklist) == 0)
795		goto on1;
796	if (tt->rn_flags & RNF_NORMAL) {
797		if (m->rm_leaf != tt || m->rm_refs > 0) {
798#if 0
799			log(LOG_ERR, "rn_delete: inconsistent annotation\n");
800#endif
801			return 0;  /* dangling ref could cause disaster */
802		}
803	} else {
804		if (m->rm_mask != tt->rn_mask) {
805#if 0
806			log(LOG_ERR, "rn_delete: inconsistent annotation\n");
807#endif
808			goto on1;
809		}
810		if (--m->rm_refs >= 0)
811			goto on1;
812	}
813	b = -1 - tt->rn_b;
814	t = saved_tt->rn_p;
815	if (b > t->rn_b)
816		goto on1; /* Wasn't lifted at all */
817	do {
818		x = t;
819		t = t->rn_p;
820	} while (b <= t->rn_b && x != top);
821	for (mp = &x->rn_mklist; (m = *mp) != NULL; mp = &m->rm_mklist)
822		if (m == saved_m) {
823			*mp = m->rm_mklist;
824			MKFree(m);
825			break;
826		}
827	if (m == 0) {
828#if 0
829		log(LOG_ERR, "rn_delete: couldn't find our annotation\n");
830#endif
831		if (tt->rn_flags & RNF_NORMAL)
832			return (0); /* Dangling ref to us */
833	}
834on1:
835	/*
836	 * Eliminate us from tree
837	 */
838	if (tt->rn_flags & RNF_ROOT)
839		return (0);
840#ifdef RN_DEBUG
841	/* Get us out of the creation list */
842	for (t = rn_clist; t && t->rn_ybro != tt; t = t->rn_ybro)
843		;
844	if (t) t->rn_ybro = tt->rn_ybro;
845#endif
846	t = tt->rn_p;
847	dupedkey = saved_tt->rn_dupedkey;
848	if (dupedkey) {
849		/*
850		 * Here, tt is the deletion target and
851		 * saved_tt is the head of the dupedkey chain.
852		 */
853		if (tt == saved_tt) {
854			x = dupedkey;
855			x->rn_p = t;
856			if (t->rn_l == tt)
857				t->rn_l = x;
858			else
859				t->rn_r = x;
860		} else {
861			/* find node in front of tt on the chain */
862			for (x = p = saved_tt; p && p->rn_dupedkey != tt;)
863				p = p->rn_dupedkey;
864			if (p) {
865				p->rn_dupedkey = tt->rn_dupedkey;
866				if (tt->rn_dupedkey)
867					tt->rn_dupedkey->rn_p = p;
868			}
869#if 0
870			else
871				log(LOG_ERR, "rn_delete: couldn't find us\n");
872#endif
873		}
874		t = tt + 1;
875		if  (t->rn_flags & RNF_ACTIVE) {
876#ifndef RN_DEBUG
877			*++x = *t;
878			p = t->rn_p;
879#else
880			b = t->rn_info;
881			*++x = *t;
882			t->rn_info = b;
883			p = t->rn_p;
884#endif
885			if (p->rn_l == t)
886				p->rn_l = x;
887			else
888				p->rn_r = x;
889			x->rn_l->rn_p = x;
890			x->rn_r->rn_p = x;
891		}
892		goto out;
893	}
894	if (t->rn_l == tt)
895		x = t->rn_r;
896	else
897		x = t->rn_l;
898	p = t->rn_p;
899	if (p->rn_r == t)
900		p->rn_r = x;
901	else
902		p->rn_l = x;
903	x->rn_p = p;
904	/*
905	 * Demote routes attached to us.
906	 */
907	if (t->rn_mklist) {
908		if (x->rn_b >= 0) {
909			for (mp = &x->rn_mklist; (m = *mp) != NULL;)
910				mp = &m->rm_mklist;
911			*mp = t->rn_mklist;
912		} else {
913			/* If there are any key,mask pairs in a sibling
914			   duped-key chain, some subset will appear sorted
915			   in the same order attached to our mklist */
916			for (m = t->rn_mklist; m && x; x = x->rn_dupedkey)
917				if (m == x->rn_mklist) {
918					struct radix_mask *mm = m->rm_mklist;
919					x->rn_mklist = 0;
920					if (--(m->rm_refs) < 0)
921						MKFree(m);
922					m = mm;
923				}
924#if 0
925			if (m)
926				log(LOG_ERR, "%s %p at %p\n",
927				    "rn_delete: Orphaned Mask", m, x);
928#endif
929		}
930	}
931	/*
932	 * We may be holding an active internal node in the tree.
933	 */
934	x = tt + 1;
935	if (t != x) {
936#ifndef RN_DEBUG
937		*t = *x;
938#else
939		b = t->rn_info;
940		*t = *x;
941		t->rn_info = b;
942#endif
943		t->rn_l->rn_p = t;
944		t->rn_r->rn_p = t;
945		p = x->rn_p;
946		if (p->rn_l == x)
947			p->rn_l = t;
948		else
949			p->rn_r = t;
950	}
951out:
952	tt->rn_flags &= ~RNF_ACTIVE;
953	tt[1].rn_flags &= ~RNF_ACTIVE;
954	return (tt);
955}
956
957int
958rn_walktree(h, f, w)
959	struct radix_node_head *h;
960	int (*f) __P((struct radix_node *, void *));
961	void *w;
962{
963	int error;
964	struct radix_node *base, *next;
965	struct radix_node *rn = h->rnh_treetop;
966	/*
967	 * This gets complicated because we may delete the node
968	 * while applying the function f to it, so we need to calculate
969	 * the successor node in advance.
970	 */
971	/* First time through node, go left */
972	while (rn->rn_b >= 0)
973		rn = rn->rn_l;
974	for (;;) {
975		base = rn;
976		/* If at right child go back up, otherwise, go right */
977		while (rn->rn_p->rn_r == rn && (rn->rn_flags & RNF_ROOT) == 0)
978			rn = rn->rn_p;
979		/* Find the next *leaf* since next node might vanish, too */
980		for (rn = rn->rn_p->rn_r; rn->rn_b >= 0;)
981			rn = rn->rn_l;
982		next = rn;
983		/* Process leaves */
984		while ((rn = base) != NULL) {
985			base = rn->rn_dupedkey;
986			if (!(rn->rn_flags & RNF_ROOT)
987			    && (error = (*f)(rn, w)))
988				return (error);
989		}
990		rn = next;
991		if (rn->rn_flags & RNF_ROOT)
992			return (0);
993	}
994	/* NOTREACHED */
995}
996
997int
998rn_inithead(head, off)
999	void **head;
1000	int off;
1001{
1002	struct radix_node_head *rnh;
1003
1004	if (*head)
1005		return (1);
1006	R_Malloc(rnh, struct radix_node_head *, sizeof (*rnh));
1007	if (rnh == 0)
1008		return (0);
1009	*head = rnh;
1010	return rn_inithead0(rnh, off);
1011}
1012
1013int
1014rn_inithead0(rnh, off)
1015	struct radix_node_head *rnh;
1016	int off;
1017{
1018	struct radix_node *t, *tt, *ttt;
1019
1020	Bzero(rnh, sizeof (*rnh));
1021	t = rn_newpair(rn_zeros, off, rnh->rnh_nodes);
1022	ttt = rnh->rnh_nodes + 2;
1023	t->rn_r = ttt;
1024	t->rn_p = t;
1025	tt = t->rn_l;
1026	tt->rn_flags = t->rn_flags = RNF_ROOT | RNF_ACTIVE;
1027	tt->rn_b = -1 - off;
1028	*ttt = *tt;
1029	ttt->rn_key = rn_ones;
1030	rnh->rnh_addaddr = rn_addroute;
1031	rnh->rnh_deladdr = rn_delete;
1032	rnh->rnh_matchaddr = rn_match;
1033	rnh->rnh_lookup = rn_lookup;
1034	rnh->rnh_walktree = rn_walktree;
1035	rnh->rnh_treetop = t;
1036	return (1);
1037}
1038
1039void
1040rn_init()
1041{
1042	char *cp, *cplim;
1043
1044	if (max_keylen == 0) {
1045#if 0
1046		log(LOG_ERR,
1047		    "rn_init: radix functions require max_keylen be set\n");
1048#endif
1049		return;
1050	}
1051	if (rn_zeros == NULL) {
1052		R_Malloc(rn_zeros, char *, 3 * max_keylen);
1053	}
1054	if (rn_zeros == NULL)
1055		panic("rn_init");
1056	Bzero(rn_zeros, 3 * max_keylen);
1057	rn_ones = cp = rn_zeros + max_keylen;
1058	addmask_key = cplim = rn_ones + max_keylen;
1059	while (cp < cplim)
1060		*cp++ = -1;
1061	if (rn_inithead((void *)&mask_rnhead, 0) == 0)
1062		panic("rn_init 2");
1063}
1064
1065
1066static int
1067rn_freenode(struct radix_node *n, void *p)
1068{
1069	struct radix_node_head *rnh = p;
1070	struct radix_node *d;
1071
1072	d = rnh->rnh_deladdr(n->rn_key, NULL, rnh);
1073	if (d != NULL) {
1074		FreeS(d, max_keylen + 2 * sizeof (*d));
1075	}
1076	return 0;
1077}
1078
1079
1080void
1081rn_freehead(rnh)
1082      struct radix_node_head *rnh;
1083{
1084
1085	(void)rn_walktree(rnh, rn_freenode, rnh);
1086
1087	rnh->rnh_addaddr = NULL;
1088	rnh->rnh_deladdr = NULL;
1089	rnh->rnh_matchaddr = NULL;
1090	rnh->rnh_lookup = NULL;
1091	rnh->rnh_walktree = NULL;
1092
1093	Free(rnh);
1094}
1095
1096
1097void
1098rn_fini()
1099{
1100	struct radix_mask *m;
1101
1102	if (rn_zeros != NULL) {
1103		FreeS(rn_zeros, 3 * max_keylen);
1104		rn_zeros = NULL;
1105	}
1106
1107	if (mask_rnhead != NULL) {
1108		rn_freehead(mask_rnhead);
1109		mask_rnhead = NULL;
1110	}
1111
1112	while ((m = rn_mkfreelist) != NULL) {
1113		rn_mkfreelist = m->rm_mklist;
1114		KFREE(m);
1115	}
1116}
1117
1118
1119#ifdef	USE_MAIN
1120
1121typedef struct myst {
1122	addrfamily_t	dst;
1123	addrfamily_t	mask;
1124	struct radix_node nodes[2];
1125} myst_t;
1126
1127int
1128main(int argc, char *argv[])
1129{
1130	struct radix_node_head *rnh;
1131	struct radix_node *rn;
1132	addrfamily_t af, mf;
1133	myst_t st1, st2, *stp;
1134
1135	memset(&st1, 0, sizeof(st1));
1136	memset(&st2, 0, sizeof(st2));
1137	memset(&af, 0, sizeof(af));
1138
1139	rn_init();
1140
1141	rnh = NULL;
1142	rn_inithead(&rnh, offsetof(addrfamily_t, adf_addr) << 3);
1143
1144	st1.dst.adf_len = sizeof(st1);
1145	st1.mask.adf_len = sizeof(st1);
1146	st1.dst.adf_addr.in4.s_addr = inet_addr("127.0.0.0");
1147	st1.mask.adf_addr.in4.s_addr = inet_addr("255.0.0.0");
1148	rn = rnh->rnh_addaddr(&st1.dst, &st1.mask, rnh, st1.nodes);
1149	printf("add.1 %p\n", rn);
1150
1151	st2.dst.adf_len = sizeof(st2);
1152	st2.mask.adf_len = sizeof(st2);
1153	st2.dst.adf_addr.in4.s_addr = inet_addr("127.0.1.0");
1154	st2.mask.adf_addr.in4.s_addr = inet_addr("255.255.255.0");
1155	rn = rnh->rnh_addaddr(&st2.dst, &st2.mask, rnh, st2.nodes);
1156	printf("add.2 %p\n", rn);
1157
1158	af.adf_len = sizeof(af);
1159	af.adf_addr.in4.s_addr = inet_addr("127.0.1.0");
1160	rn = rnh->rnh_matchaddr(&af, rnh);
1161	if (rn != NULL) {
1162		printf("1.lookup = %p key %p mask %p\n", rn, rn->rn_key, rn->rn_mask);
1163		stp = rn->rn_key;
1164		printf("%s/", inet_ntoa(stp->dst.adf_addr.in4));
1165		stp = rn->rn_mask;
1166		printf("%s\n", inet_ntoa(stp->dst.adf_addr.in4));
1167	}
1168
1169	mf.adf_len = sizeof(mf);
1170	mf.adf_addr.in4.s_addr = inet_addr("255.255.255.0");
1171	rn = rnh->rnh_lookup(&af, &mf, rnh);
1172	if (rn != NULL) {
1173		printf("2.lookup = %p key %p mask %p\n", rn, rn->rn_key, rn->rn_mask);
1174		stp = rn->rn_key;
1175		printf("%s/", inet_ntoa(stp->dst.adf_addr.in4));
1176		stp = rn->rn_mask;
1177		printf("%s\n", inet_ntoa(stp->dst.adf_addr.in4));
1178	}
1179
1180	af.adf_len = sizeof(af);
1181	af.adf_addr.in4.s_addr = inet_addr("126.0.0.1");
1182	rn = rnh->rnh_matchaddr(&af, rnh);
1183	if (rn != NULL) {
1184		printf("3.lookup = %p key %p mask %p\n", rn, rn->rn_key, rn->rn_mask);
1185		stp = rn->rn_key;
1186		printf("%s/", inet_ntoa(stp->dst.adf_addr.in4));
1187		stp = rn->rn_mask;
1188		printf("%s\n", inet_ntoa(stp->dst.adf_addr.in4));
1189	}
1190
1191	return 0;
1192}
1193
1194
1195void
1196log(int level, char *format, ...)
1197{
1198	va_list ap;
1199
1200	va_start(ap, format);
1201	vfprintf(stderr, format, ap);
1202	va_end(ap);
1203}
1204#endif
1205
1206
1207#ifndef	_KERNEL
1208void
1209panic(char *str)
1210{
1211	fputs(str, stderr);
1212	abort();
1213}
1214#endif
1215