xref: /netbsd-src/sys/net/radix.c (revision 404fbe5fb94ca1e054339640cabb2801ce52dd30)
1 /*	$NetBSD: radix.c,v 1.40 2008/11/25 18:28:06 pooka 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. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	@(#)radix.c	8.6 (Berkeley) 10/17/95
32  */
33 
34 /*
35  * Routines to build and maintain radix trees for routing lookups.
36  */
37 
38 #include <sys/cdefs.h>
39 __KERNEL_RCSID(0, "$NetBSD: radix.c,v 1.40 2008/11/25 18:28:06 pooka Exp $");
40 
41 #ifndef _NET_RADIX_H_
42 #include <sys/param.h>
43 #ifdef	_KERNEL
44 #include "opt_inet.h"
45 
46 #include <sys/systm.h>
47 #include <sys/malloc.h>
48 #define	M_DONTWAIT M_NOWAIT
49 #include <sys/domain.h>
50 #else
51 #include <stdlib.h>
52 #endif
53 #include <machine/stdarg.h>
54 #include <sys/syslog.h>
55 #include <net/radix.h>
56 #endif
57 
58 typedef void (*rn_printer_t)(void *, const char *fmt, ...);
59 
60 int	max_keylen;
61 struct radix_mask *rn_mkfreelist;
62 struct radix_node_head *mask_rnhead;
63 static char *addmask_key;
64 static const char normal_chars[] =
65     {0, 0x80, 0xc0, 0xe0, 0xf0, 0xf8, 0xfc, 0xfe, -1};
66 static char *rn_zeros, *rn_ones;
67 
68 #define rn_masktop (mask_rnhead->rnh_treetop)
69 
70 static int rn_satisfies_leaf(const char *, struct radix_node *, int);
71 static int rn_lexobetter(const void *, const void *);
72 static struct radix_mask *rn_new_radix_mask(struct radix_node *,
73     struct radix_mask *);
74 static struct radix_node *rn_walknext(struct radix_node *, rn_printer_t,
75     void *);
76 static struct radix_node *rn_walkfirst(struct radix_node *, rn_printer_t,
77     void *);
78 static void rn_nodeprint(struct radix_node *, rn_printer_t, void *,
79     const char *);
80 
81 #define	SUBTREE_OPEN	"[ "
82 #define	SUBTREE_CLOSE	" ]"
83 
84 #ifdef RN_DEBUG
85 static void rn_treeprint(struct radix_node_head *, rn_printer_t, void *);
86 #endif /* RN_DEBUG */
87 
88 /*
89  * The data structure for the keys is a radix tree with one way
90  * branching removed.  The index rn_b at an internal node n represents a bit
91  * position to be tested.  The tree is arranged so that all descendants
92  * of a node n have keys whose bits all agree up to position rn_b - 1.
93  * (We say the index of n is rn_b.)
94  *
95  * There is at least one descendant which has a one bit at position rn_b,
96  * and at least one with a zero there.
97  *
98  * A route is determined by a pair of key and mask.  We require that the
99  * bit-wise logical and of the key and mask to be the key.
100  * We define the index of a route to associated with the mask to be
101  * the first bit number in the mask where 0 occurs (with bit number 0
102  * representing the highest order bit).
103  *
104  * We say a mask is normal if every bit is 0, past the index of the mask.
105  * If a node n has a descendant (k, m) with index(m) == index(n) == rn_b,
106  * and m is a normal mask, then the route applies to every descendant of n.
107  * If the index(m) < rn_b, this implies the trailing last few bits of k
108  * before bit b are all 0, (and hence consequently true of every descendant
109  * of n), so the route applies to all descendants of the node as well.
110  *
111  * Similar logic shows that a non-normal mask m such that
112  * index(m) <= index(n) could potentially apply to many children of n.
113  * Thus, for each non-host route, we attach its mask to a list at an internal
114  * node as high in the tree as we can go.
115  *
116  * The present version of the code makes use of normal routes in short-
117  * circuiting an explicit mask and compare operation when testing whether
118  * a key satisfies a normal route, and also in remembering the unique leaf
119  * that governs a subtree.
120  */
121 
122 struct radix_node *
123 rn_search(
124 	const void *v_arg,
125 	struct radix_node *head)
126 {
127 	const u_char * const v = v_arg;
128 	struct radix_node *x;
129 
130 	for (x = head; x->rn_b >= 0;) {
131 		if (x->rn_bmask & v[x->rn_off])
132 			x = x->rn_r;
133 		else
134 			x = x->rn_l;
135 	}
136 	return x;
137 }
138 
139 struct radix_node *
140 rn_search_m(
141 	const void *v_arg,
142 	struct radix_node *head,
143 	const void *m_arg)
144 {
145 	struct radix_node *x;
146 	const u_char * const v = v_arg;
147 	const u_char * const m = m_arg;
148 
149 	for (x = head; x->rn_b >= 0;) {
150 		if ((x->rn_bmask & m[x->rn_off]) &&
151 		    (x->rn_bmask & v[x->rn_off]))
152 			x = x->rn_r;
153 		else
154 			x = x->rn_l;
155 	}
156 	return x;
157 }
158 
159 int
160 rn_refines(
161 	const void *m_arg,
162 	const void *n_arg)
163 {
164 	const char *m = m_arg;
165 	const char *n = n_arg;
166 	const char *lim = n + *(const u_char *)n;
167 	const char *lim2 = lim;
168 	int longer = (*(const u_char *)n++) - (int)(*(const u_char *)m++);
169 	int masks_are_equal = 1;
170 
171 	if (longer > 0)
172 		lim -= longer;
173 	while (n < lim) {
174 		if (*n & ~(*m))
175 			return 0;
176 		if (*n++ != *m++)
177 			masks_are_equal = 0;
178 	}
179 	while (n < lim2)
180 		if (*n++)
181 			return 0;
182 	if (masks_are_equal && (longer < 0))
183 		for (lim2 = m - longer; m < lim2; )
184 			if (*m++)
185 				return 1;
186 	return !masks_are_equal;
187 }
188 
189 struct radix_node *
190 rn_lookup(
191 	const void *v_arg,
192 	const void *m_arg,
193 	struct radix_node_head *head)
194 {
195 	struct radix_node *x;
196 	const char *netmask = NULL;
197 
198 	if (m_arg) {
199 		if ((x = rn_addmask(m_arg, 1, head->rnh_treetop->rn_off)) == 0)
200 			return NULL;
201 		netmask = x->rn_key;
202 	}
203 	x = rn_match(v_arg, head);
204 	if (x != NULL && netmask != NULL) {
205 		while (x != NULL && x->rn_mask != netmask)
206 			x = x->rn_dupedkey;
207 	}
208 	return x;
209 }
210 
211 static int
212 rn_satisfies_leaf(
213 	const char *trial,
214 	struct radix_node *leaf,
215 	int skip)
216 {
217 	const char *cp = trial;
218 	const char *cp2 = leaf->rn_key;
219 	const char *cp3 = leaf->rn_mask;
220 	const char *cplim;
221 	int length = min(*(const u_char *)cp, *(const u_char *)cp2);
222 
223 	if (cp3 == 0)
224 		cp3 = rn_ones;
225 	else
226 		length = min(length, *(const u_char *)cp3);
227 	cplim = cp + length; cp3 += skip; cp2 += skip;
228 	for (cp += skip; cp < cplim; cp++, cp2++, cp3++)
229 		if ((*cp ^ *cp2) & *cp3)
230 			return 0;
231 	return 1;
232 }
233 
234 struct radix_node *
235 rn_match(
236 	const void *v_arg,
237 	struct radix_node_head *head)
238 {
239 	const char * const v = v_arg;
240 	struct radix_node *t = head->rnh_treetop;
241 	struct radix_node *top = t;
242 	struct radix_node *x;
243 	struct radix_node *saved_t;
244 	const char *cp = v;
245 	const char *cp2;
246 	const char *cplim;
247 	int off = t->rn_off;
248 	int vlen = *(const u_char *)cp;
249 	int matched_off;
250 	int test, b, rn_b;
251 
252 	/*
253 	 * Open code rn_search(v, top) to avoid overhead of extra
254 	 * subroutine call.
255 	 */
256 	for (; t->rn_b >= 0; ) {
257 		if (t->rn_bmask & cp[t->rn_off])
258 			t = t->rn_r;
259 		else
260 			t = t->rn_l;
261 	}
262 	/*
263 	 * See if we match exactly as a host destination
264 	 * or at least learn how many bits match, for normal mask finesse.
265 	 *
266 	 * It doesn't hurt us to limit how many bytes to check
267 	 * to the length of the mask, since if it matches we had a genuine
268 	 * match and the leaf we have is the most specific one anyway;
269 	 * if it didn't match with a shorter length it would fail
270 	 * with a long one.  This wins big for class B&C netmasks which
271 	 * are probably the most common case...
272 	 */
273 	if (t->rn_mask)
274 		vlen = *(const u_char *)t->rn_mask;
275 	cp += off; cp2 = t->rn_key + off; cplim = v + vlen;
276 	for (; cp < cplim; cp++, cp2++)
277 		if (*cp != *cp2)
278 			goto on1;
279 	/*
280 	 * This extra grot is in case we are explicitly asked
281 	 * to look up the default.  Ugh!
282 	 */
283 	if ((t->rn_flags & RNF_ROOT) && t->rn_dupedkey)
284 		t = t->rn_dupedkey;
285 	return t;
286 on1:
287 	test = (*cp ^ *cp2) & 0xff; /* find first bit that differs */
288 	for (b = 7; (test >>= 1) > 0;)
289 		b--;
290 	matched_off = cp - v;
291 	b += matched_off << 3;
292 	rn_b = -1 - b;
293 	/*
294 	 * If there is a host route in a duped-key chain, it will be first.
295 	 */
296 	if ((saved_t = t)->rn_mask == 0)
297 		t = t->rn_dupedkey;
298 	for (; t; t = t->rn_dupedkey)
299 		/*
300 		 * Even if we don't match exactly as a host,
301 		 * we may match if the leaf we wound up at is
302 		 * a route to a net.
303 		 */
304 		if (t->rn_flags & RNF_NORMAL) {
305 			if (rn_b <= t->rn_b)
306 				return t;
307 		} else if (rn_satisfies_leaf(v, t, matched_off))
308 				return t;
309 	t = saved_t;
310 	/* start searching up the tree */
311 	do {
312 		struct radix_mask *m;
313 		t = t->rn_p;
314 		m = t->rn_mklist;
315 		if (m) {
316 			/*
317 			 * If non-contiguous masks ever become important
318 			 * we can restore the masking and open coding of
319 			 * the search and satisfaction test and put the
320 			 * calculation of "off" back before the "do".
321 			 */
322 			do {
323 				if (m->rm_flags & RNF_NORMAL) {
324 					if (rn_b <= m->rm_b)
325 						return m->rm_leaf;
326 				} else {
327 					off = min(t->rn_off, matched_off);
328 					x = rn_search_m(v, t, m->rm_mask);
329 					while (x && x->rn_mask != m->rm_mask)
330 						x = x->rn_dupedkey;
331 					if (x && rn_satisfies_leaf(v, x, off))
332 						return x;
333 				}
334 				m = m->rm_mklist;
335 			} while (m);
336 		}
337 	} while (t != top);
338 	return NULL;
339 }
340 
341 static void
342 rn_nodeprint(struct radix_node *rn, rn_printer_t printer, void *arg,
343     const char *delim)
344 {
345 	(*printer)(arg, "%s(%s%p: p<%p> l<%p> r<%p>)",
346 	    delim, ((void *)rn == arg) ? "*" : "", rn, rn->rn_p,
347 	    rn->rn_l, rn->rn_r);
348 }
349 
350 #ifdef RN_DEBUG
351 int	rn_debug =  1;
352 
353 static void
354 rn_dbg_print(void *arg, const char *fmt, ...)
355 {
356 	va_list ap;
357 
358 	va_start(ap, fmt);
359 	vlog(LOG_DEBUG, fmt, ap);
360 	va_end(ap);
361 }
362 
363 static void
364 rn_treeprint(struct radix_node_head *h, rn_printer_t printer, void *arg)
365 {
366 	struct radix_node *dup, *rn;
367 	const char *delim;
368 
369 	if (printer == NULL)
370 		return;
371 
372 	rn = rn_walkfirst(h->rnh_treetop, printer, arg);
373 	for (;;) {
374 		/* Process leaves */
375 		delim = "";
376 		for (dup = rn; dup != NULL; dup = dup->rn_dupedkey) {
377 			if ((dup->rn_flags & RNF_ROOT) != 0)
378 				continue;
379 			rn_nodeprint(dup, printer, arg, delim);
380 			delim = ", ";
381 		}
382 		rn = rn_walknext(rn, printer, arg);
383 		if (rn->rn_flags & RNF_ROOT)
384 			return;
385 	}
386 	/* NOTREACHED */
387 }
388 
389 #define	traverse(__head, __rn)	rn_treeprint((__head), rn_dbg_print, (__rn))
390 #endif /* RN_DEBUG */
391 
392 struct radix_node *
393 rn_newpair(
394 	const void *v,
395 	int b,
396 	struct radix_node nodes[2])
397 {
398 	struct radix_node *tt = nodes;
399 	struct radix_node *t = tt + 1;
400 	t->rn_b = b; t->rn_bmask = 0x80 >> (b & 7);
401 	t->rn_l = tt; t->rn_off = b >> 3;
402 	tt->rn_b = -1; tt->rn_key = v; tt->rn_p = t;
403 	tt->rn_flags = t->rn_flags = RNF_ACTIVE;
404 	return t;
405 }
406 
407 struct radix_node *
408 rn_insert(
409 	const void *v_arg,
410 	struct radix_node_head *head,
411 	int *dupentry,
412 	struct radix_node nodes[2])
413 {
414 	struct radix_node *top = head->rnh_treetop;
415 	struct radix_node *t = rn_search(v_arg, top);
416 	struct radix_node *tt;
417 	const char *v = v_arg;
418 	int head_off = top->rn_off;
419 	int vlen = *((const u_char *)v);
420 	const char *cp = v + head_off;
421 	int b;
422     	/*
423 	 * Find first bit at which v and t->rn_key differ
424 	 */
425     {
426 	const char *cp2 = t->rn_key + head_off;
427 	const char *cplim = v + vlen;
428 	int cmp_res;
429 
430 	while (cp < cplim)
431 		if (*cp2++ != *cp++)
432 			goto on1;
433 	*dupentry = 1;
434 	return t;
435 on1:
436 	*dupentry = 0;
437 	cmp_res = (cp[-1] ^ cp2[-1]) & 0xff;
438 	for (b = (cp - v) << 3; cmp_res; b--)
439 		cmp_res >>= 1;
440     }
441     {
442 	struct radix_node *p, *x = top;
443 	cp = v;
444 	do {
445 		p = x;
446 		if (cp[x->rn_off] & x->rn_bmask)
447 			x = x->rn_r;
448 		else x = x->rn_l;
449 	} while (b > (unsigned) x->rn_b); /* x->rn_b < b && x->rn_b >= 0 */
450 #ifdef RN_DEBUG
451 	if (rn_debug)
452 		log(LOG_DEBUG, "%s: Going In:\n", __func__), traverse(head, p);
453 #endif
454 	t = rn_newpair(v_arg, b, nodes); tt = t->rn_l;
455 	if ((cp[p->rn_off] & p->rn_bmask) == 0)
456 		p->rn_l = t;
457 	else
458 		p->rn_r = t;
459 	x->rn_p = t; t->rn_p = p; /* frees x, p as temp vars below */
460 	if ((cp[t->rn_off] & t->rn_bmask) == 0) {
461 		t->rn_r = x;
462 	} else {
463 		t->rn_r = tt; t->rn_l = x;
464 	}
465 #ifdef RN_DEBUG
466 	if (rn_debug) {
467 		log(LOG_DEBUG, "%s: Coming Out:\n", __func__),
468 		    traverse(head, p);
469 	}
470 #endif /* RN_DEBUG */
471     }
472 	return tt;
473 }
474 
475 struct radix_node *
476 rn_addmask(
477 	const void *n_arg,
478 	int search,
479 	int skip)
480 {
481 	const char *netmask = n_arg;
482 	const char *cp;
483 	const char *cplim;
484 	struct radix_node *x;
485 	struct radix_node *saved_x;
486 	int b = 0, mlen, j;
487 	int maskduplicated, m0, isnormal;
488 	static int last_zeroed = 0;
489 
490 	if ((mlen = *(const u_char *)netmask) > max_keylen)
491 		mlen = max_keylen;
492 	if (skip == 0)
493 		skip = 1;
494 	if (mlen <= skip)
495 		return mask_rnhead->rnh_nodes;
496 	if (skip > 1)
497 		memmove(addmask_key + 1, rn_ones + 1, skip - 1);
498 	if ((m0 = mlen) > skip)
499 		memmove(addmask_key + skip, netmask + skip, mlen - skip);
500 	/*
501 	 * Trim trailing zeroes.
502 	 */
503 	for (cp = addmask_key + mlen; (cp > addmask_key) && cp[-1] == 0;)
504 		cp--;
505 	mlen = cp - addmask_key;
506 	if (mlen <= skip) {
507 		if (m0 >= last_zeroed)
508 			last_zeroed = mlen;
509 		return mask_rnhead->rnh_nodes;
510 	}
511 	if (m0 < last_zeroed)
512 		memset(addmask_key + m0, 0, last_zeroed - m0);
513 	*addmask_key = last_zeroed = mlen;
514 	x = rn_search(addmask_key, rn_masktop);
515 	if (memcmp(addmask_key, x->rn_key, mlen) != 0)
516 		x = 0;
517 	if (x || search)
518 		return x;
519 	R_Malloc(x, struct radix_node *, max_keylen + 2 * sizeof (*x));
520 	if ((saved_x = x) == NULL)
521 		return NULL;
522 	memset(x, 0, max_keylen + 2 * sizeof (*x));
523 	cp = netmask = (void *)(x + 2);
524 	memmove(x + 2, addmask_key, mlen);
525 	x = rn_insert(cp, mask_rnhead, &maskduplicated, x);
526 	if (maskduplicated) {
527 		log(LOG_ERR, "rn_addmask: mask impossibly already in tree\n");
528 		Free(saved_x);
529 		return x;
530 	}
531 	/*
532 	 * Calculate index of mask, and check for normalcy.
533 	 */
534 	cplim = netmask + mlen; isnormal = 1;
535 	for (cp = netmask + skip; (cp < cplim) && *(const 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 
550 static int	/* XXX: arbitrary ordering for non-contiguous masks */
551 rn_lexobetter(
552 	const void *m_arg,
553 	const void *n_arg)
554 {
555 	const u_char *mp = m_arg;
556 	const u_char *np = n_arg;
557 	const u_char *lim;
558 
559 	if (*mp > *np)
560 		return 1;  /* not really, but need to check longer one first */
561 	if (*mp == *np)
562 		for (lim = mp + *mp; mp < lim;)
563 			if (*mp++ > *np++)
564 				return 1;
565 	return 0;
566 }
567 
568 static struct radix_mask *
569 rn_new_radix_mask(
570 	struct radix_node *tt,
571 	struct radix_mask *next)
572 {
573 	struct radix_mask *m;
574 
575 	MKGet(m);
576 	if (m == NULL) {
577 		log(LOG_ERR, "Mask for route not entered\n");
578 		return NULL;
579 	}
580 	memset(m, 0, sizeof(*m));
581 	m->rm_b = tt->rn_b;
582 	m->rm_flags = tt->rn_flags;
583 	if (tt->rn_flags & RNF_NORMAL)
584 		m->rm_leaf = tt;
585 	else
586 		m->rm_mask = tt->rn_mask;
587 	m->rm_mklist = next;
588 	tt->rn_mklist = m;
589 	return m;
590 }
591 
592 struct radix_node *
593 rn_addroute(
594 	const void *v_arg,
595 	const void *n_arg,
596 	struct radix_node_head *head,
597 	struct radix_node treenodes[2])
598 {
599 	const char *v = v_arg, *netmask = n_arg;
600 	struct radix_node *t, *x = NULL, *tt;
601 	struct radix_node *saved_tt, *top = head->rnh_treetop;
602 	short b = 0, b_leaf = 0;
603 	int keyduplicated;
604 	const char *mmask;
605 	struct radix_mask *m, **mp;
606 
607 	/*
608 	 * In dealing with non-contiguous masks, there may be
609 	 * many different routes which have the same mask.
610 	 * We will find it useful to have a unique pointer to
611 	 * the mask to speed avoiding duplicate references at
612 	 * nodes and possibly save time in calculating indices.
613 	 */
614 	if (netmask != NULL) {
615 		if ((x = rn_addmask(netmask, 0, top->rn_off)) == NULL)
616 			return NULL;
617 		b_leaf = x->rn_b;
618 		b = -1 - x->rn_b;
619 		netmask = x->rn_key;
620 	}
621 	/*
622 	 * Deal with duplicated keys: attach node to previous instance
623 	 */
624 	saved_tt = tt = rn_insert(v, head, &keyduplicated, treenodes);
625 	if (keyduplicated) {
626 		for (t = tt; tt != NULL; t = tt, tt = tt->rn_dupedkey) {
627 			if (tt->rn_mask == netmask)
628 				return NULL;
629 			if (netmask == NULL ||
630 			    (tt->rn_mask != NULL &&
631 			     (b_leaf < tt->rn_b || /* index(netmask) > node */
632 			       rn_refines(netmask, tt->rn_mask) ||
633 			       rn_lexobetter(netmask, tt->rn_mask))))
634 				break;
635 		}
636 		/*
637 		 * If the mask is not duplicated, we wouldn't
638 		 * find it among possible duplicate key entries
639 		 * anyway, so the above test doesn't hurt.
640 		 *
641 		 * We sort the masks for a duplicated key the same way as
642 		 * in a masklist -- most specific to least specific.
643 		 * This may require the unfortunate nuisance of relocating
644 		 * the head of the list.
645 		 *
646 		 * We also reverse, or doubly link the list through the
647 		 * parent pointer.
648 		 */
649 		if (tt == saved_tt) {
650 			struct	radix_node *xx = x;
651 			/* link in at head of list */
652 			(tt = treenodes)->rn_dupedkey = t;
653 			tt->rn_flags = t->rn_flags;
654 			tt->rn_p = x = t->rn_p;
655 			t->rn_p = tt;
656 			if (x->rn_l == t)
657 				x->rn_l = tt;
658 			else
659 				x->rn_r = tt;
660 			saved_tt = tt;
661 			x = xx;
662 		} else {
663 			(tt = treenodes)->rn_dupedkey = t->rn_dupedkey;
664 			t->rn_dupedkey = tt;
665 			tt->rn_p = t;
666 			if (tt->rn_dupedkey)
667 				tt->rn_dupedkey->rn_p = tt;
668 		}
669 		tt->rn_key = v;
670 		tt->rn_b = -1;
671 		tt->rn_flags = RNF_ACTIVE;
672 	}
673 	/*
674 	 * Put mask in tree.
675 	 */
676 	if (netmask != NULL) {
677 		tt->rn_mask = netmask;
678 		tt->rn_b = x->rn_b;
679 		tt->rn_flags |= x->rn_flags & RNF_NORMAL;
680 	}
681 	t = saved_tt->rn_p;
682 	if (keyduplicated)
683 		goto on2;
684 	b_leaf = -1 - t->rn_b;
685 	if (t->rn_r == saved_tt)
686 		x = t->rn_l;
687 	else
688 		x = t->rn_r;
689 	/* Promote general routes from below */
690 	if (x->rn_b < 0) {
691 		for (mp = &t->rn_mklist; x != NULL; x = x->rn_dupedkey) {
692 			if (x->rn_mask != NULL && x->rn_b >= b_leaf &&
693 			    x->rn_mklist == NULL) {
694 				*mp = m = rn_new_radix_mask(x, NULL);
695 				if (m != NULL)
696 					mp = &m->rm_mklist;
697 			}
698 		}
699 	} else if (x->rn_mklist != NULL) {
700 		/*
701 		 * Skip over masks whose index is > that of new node
702 		 */
703 		for (mp = &x->rn_mklist; (m = *mp) != NULL; mp = &m->rm_mklist)
704 			if (m->rm_b >= b_leaf)
705 				break;
706 		t->rn_mklist = m;
707 		*mp = NULL;
708 	}
709 on2:
710 	/* Add new route to highest possible ancestor's list */
711 	if (netmask == NULL || b > t->rn_b)
712 		return tt; /* can't lift at all */
713 	b_leaf = tt->rn_b;
714 	do {
715 		x = t;
716 		t = t->rn_p;
717 	} while (b <= t->rn_b && x != top);
718 	/*
719 	 * Search through routes associated with node to
720 	 * insert new route according to index.
721 	 * Need same criteria as when sorting dupedkeys to avoid
722 	 * double loop on deletion.
723 	 */
724 	for (mp = &x->rn_mklist; (m = *mp) != NULL; mp = &m->rm_mklist) {
725 		if (m->rm_b < b_leaf)
726 			continue;
727 		if (m->rm_b > b_leaf)
728 			break;
729 		if (m->rm_flags & RNF_NORMAL) {
730 			mmask = m->rm_leaf->rn_mask;
731 			if (tt->rn_flags & RNF_NORMAL) {
732 				log(LOG_ERR, "Non-unique normal route,"
733 				    " mask not entered\n");
734 				return tt;
735 			}
736 		} else
737 			mmask = m->rm_mask;
738 		if (mmask == netmask) {
739 			m->rm_refs++;
740 			tt->rn_mklist = m;
741 			return tt;
742 		}
743 		if (rn_refines(netmask, mmask) || rn_lexobetter(netmask, mmask))
744 			break;
745 	}
746 	*mp = rn_new_radix_mask(tt, *mp);
747 	return tt;
748 }
749 
750 struct radix_node *
751 rn_delete1(
752 	const void *v_arg,
753 	const void *netmask_arg,
754 	struct radix_node_head *head,
755 	struct radix_node *rn)
756 {
757 	struct radix_node *t, *p, *x, *tt;
758 	struct radix_mask *m, *saved_m, **mp;
759 	struct radix_node *dupedkey, *saved_tt, *top;
760 	const char *v, *netmask;
761 	int b, head_off, vlen;
762 
763 	v = v_arg;
764 	netmask = netmask_arg;
765 	x = head->rnh_treetop;
766 	tt = rn_search(v, x);
767 	head_off = x->rn_off;
768 	vlen =  *(const u_char *)v;
769 	saved_tt = tt;
770 	top = x;
771 	if (tt == NULL ||
772 	    memcmp(v + head_off, tt->rn_key + head_off, vlen - head_off) != 0)
773 		return NULL;
774 	/*
775 	 * Delete our route from mask lists.
776 	 */
777 	if (netmask != NULL) {
778 		if ((x = rn_addmask(netmask, 1, head_off)) == NULL)
779 			return NULL;
780 		netmask = x->rn_key;
781 		while (tt->rn_mask != netmask)
782 			if ((tt = tt->rn_dupedkey) == NULL)
783 				return NULL;
784 	}
785 	if (tt->rn_mask == NULL || (saved_m = m = tt->rn_mklist) == NULL)
786 		goto on1;
787 	if (tt->rn_flags & RNF_NORMAL) {
788 		if (m->rm_leaf != tt || m->rm_refs > 0) {
789 			log(LOG_ERR, "rn_delete: inconsistent annotation\n");
790 			return NULL;  /* dangling ref could cause disaster */
791 		}
792 	} else {
793 		if (m->rm_mask != tt->rn_mask) {
794 			log(LOG_ERR, "rn_delete: inconsistent annotation\n");
795 			goto on1;
796 		}
797 		if (--m->rm_refs >= 0)
798 			goto on1;
799 	}
800 	b = -1 - tt->rn_b;
801 	t = saved_tt->rn_p;
802 	if (b > t->rn_b)
803 		goto on1; /* Wasn't lifted at all */
804 	do {
805 		x = t;
806 		t = t->rn_p;
807 	} while (b <= t->rn_b && x != top);
808 	for (mp = &x->rn_mklist; (m = *mp) != NULL; mp = &m->rm_mklist) {
809 		if (m == saved_m) {
810 			*mp = m->rm_mklist;
811 			MKFree(m);
812 			break;
813 		}
814 	}
815 	if (m == NULL) {
816 		log(LOG_ERR, "rn_delete: couldn't find our annotation\n");
817 		if (tt->rn_flags & RNF_NORMAL)
818 			return NULL; /* Dangling ref to us */
819 	}
820 on1:
821 	/*
822 	 * Eliminate us from tree
823 	 */
824 	if (tt->rn_flags & RNF_ROOT)
825 		return NULL;
826 #ifdef RN_DEBUG
827 	if (rn_debug)
828 		log(LOG_DEBUG, "%s: Going In:\n", __func__), traverse(head, tt);
829 #endif
830 	t = tt->rn_p;
831 	dupedkey = saved_tt->rn_dupedkey;
832 	if (dupedkey != NULL) {
833 		/*
834 		 * Here, tt is the deletion target, and
835 		 * saved_tt is the head of the dupedkey chain.
836 		 */
837 		if (tt == saved_tt) {
838 			x = dupedkey;
839 			x->rn_p = t;
840 			if (t->rn_l == tt)
841 				t->rn_l = x;
842 			else
843 				t->rn_r = x;
844 		} else {
845 			/* find node in front of tt on the chain */
846 			for (x = p = saved_tt;
847 			     p != NULL && p->rn_dupedkey != tt;)
848 				p = p->rn_dupedkey;
849 			if (p != NULL) {
850 				p->rn_dupedkey = tt->rn_dupedkey;
851 				if (tt->rn_dupedkey != NULL)
852 					tt->rn_dupedkey->rn_p = p;
853 			} else
854 				log(LOG_ERR, "rn_delete: couldn't find us\n");
855 		}
856 		t = tt + 1;
857 		if  (t->rn_flags & RNF_ACTIVE) {
858 			*++x = *t;
859 			p = t->rn_p;
860 			if (p->rn_l == t)
861 				p->rn_l = x;
862 			else
863 				p->rn_r = x;
864 			x->rn_l->rn_p = x;
865 			x->rn_r->rn_p = x;
866 		}
867 		goto out;
868 	}
869 	if (t->rn_l == tt)
870 		x = t->rn_r;
871 	else
872 		x = t->rn_l;
873 	p = t->rn_p;
874 	if (p->rn_r == t)
875 		p->rn_r = x;
876 	else
877 		p->rn_l = x;
878 	x->rn_p = p;
879 	/*
880 	 * Demote routes attached to us.
881 	 */
882 	if (t->rn_mklist == NULL)
883 		;
884 	else if (x->rn_b >= 0) {
885 		for (mp = &x->rn_mklist; (m = *mp) != NULL; mp = &m->rm_mklist)
886 			;
887 		*mp = t->rn_mklist;
888 	} else {
889 		/* If there are any key,mask pairs in a sibling
890 		   duped-key chain, some subset will appear sorted
891 		   in the same order attached to our mklist */
892 		for (m = t->rn_mklist;
893 		     m != NULL && x != NULL;
894 		     x = x->rn_dupedkey) {
895 			if (m == x->rn_mklist) {
896 				struct radix_mask *mm = m->rm_mklist;
897 				x->rn_mklist = NULL;
898 				if (--(m->rm_refs) < 0)
899 					MKFree(m);
900 				m = mm;
901 			}
902 		}
903 		if (m != NULL) {
904 			log(LOG_ERR, "rn_delete: Orphaned Mask %p at %p\n",
905 			    m, x);
906 		}
907 	}
908 	/*
909 	 * We may be holding an active internal node in the tree.
910 	 */
911 	x = tt + 1;
912 	if (t != x) {
913 		*t = *x;
914 		t->rn_l->rn_p = t;
915 		t->rn_r->rn_p = t;
916 		p = x->rn_p;
917 		if (p->rn_l == x)
918 			p->rn_l = t;
919 		else
920 			p->rn_r = t;
921 	}
922 out:
923 #ifdef RN_DEBUG
924 	if (rn_debug) {
925 		log(LOG_DEBUG, "%s: Coming Out:\n", __func__),
926 		    traverse(head, tt);
927 	}
928 #endif /* RN_DEBUG */
929 	tt->rn_flags &= ~RNF_ACTIVE;
930 	tt[1].rn_flags &= ~RNF_ACTIVE;
931 	return tt;
932 }
933 
934 struct radix_node *
935 rn_delete(
936 	const void *v_arg,
937 	const void *netmask_arg,
938 	struct radix_node_head *head)
939 {
940 	return rn_delete1(v_arg, netmask_arg, head, NULL);
941 }
942 
943 static struct radix_node *
944 rn_walknext(struct radix_node *rn, rn_printer_t printer, void *arg)
945 {
946 	/* If at right child go back up, otherwise, go right */
947 	while (rn->rn_p->rn_r == rn && (rn->rn_flags & RNF_ROOT) == 0) {
948 		if (printer != NULL)
949 			(*printer)(arg, SUBTREE_CLOSE);
950 		rn = rn->rn_p;
951 	}
952 	if (printer)
953 		rn_nodeprint(rn->rn_p, printer, arg, "");
954 	/* Find the next *leaf* since next node might vanish, too */
955 	for (rn = rn->rn_p->rn_r; rn->rn_b >= 0;) {
956 		if (printer != NULL)
957 			(*printer)(arg, SUBTREE_OPEN);
958 		rn = rn->rn_l;
959 	}
960 	return rn;
961 }
962 
963 static struct radix_node *
964 rn_walkfirst(struct radix_node *rn, rn_printer_t printer, void *arg)
965 {
966 	/* First time through node, go left */
967 	while (rn->rn_b >= 0) {
968 		if (printer != NULL)
969 			(*printer)(arg, SUBTREE_OPEN);
970 		rn = rn->rn_l;
971 	}
972 	return rn;
973 }
974 
975 int
976 rn_walktree(
977 	struct radix_node_head *h,
978 	int (*f)(struct radix_node *, void *),
979 	void *w)
980 {
981 	int error;
982 	struct radix_node *base, *next, *rn;
983 	/*
984 	 * This gets complicated because we may delete the node
985 	 * while applying the function f to it, so we need to calculate
986 	 * the successor node in advance.
987 	 */
988 	rn = rn_walkfirst(h->rnh_treetop, NULL, NULL);
989 	for (;;) {
990 		base = rn;
991 		next = rn_walknext(rn, NULL, NULL);
992 		/* Process leaves */
993 		while ((rn = base) != NULL) {
994 			base = rn->rn_dupedkey;
995 			if (!(rn->rn_flags & RNF_ROOT) && (error = (*f)(rn, w)))
996 				return error;
997 		}
998 		rn = next;
999 		if (rn->rn_flags & RNF_ROOT)
1000 			return 0;
1001 	}
1002 	/* NOTREACHED */
1003 }
1004 
1005 int
1006 rn_inithead(head, off)
1007 	void **head;
1008 	int off;
1009 {
1010 	struct radix_node_head *rnh;
1011 
1012 	if (*head != NULL)
1013 		return 1;
1014 	R_Malloc(rnh, struct radix_node_head *, sizeof (*rnh));
1015 	if (rnh == NULL)
1016 		return 0;
1017 	*head = rnh;
1018 	return rn_inithead0(rnh, off);
1019 }
1020 
1021 int
1022 rn_inithead0(rnh, off)
1023 	struct radix_node_head *rnh;
1024 	int off;
1025 {
1026 	struct radix_node *t;
1027 	struct radix_node *tt;
1028 	struct radix_node *ttt;
1029 
1030 	memset(rnh, 0, sizeof(*rnh));
1031 	t = rn_newpair(rn_zeros, off, rnh->rnh_nodes);
1032 	ttt = rnh->rnh_nodes + 2;
1033 	t->rn_r = ttt;
1034 	t->rn_p = t;
1035 	tt = t->rn_l;
1036 	tt->rn_flags = t->rn_flags = RNF_ROOT | RNF_ACTIVE;
1037 	tt->rn_b = -1 - off;
1038 	*ttt = *tt;
1039 	ttt->rn_key = rn_ones;
1040 	rnh->rnh_addaddr = rn_addroute;
1041 	rnh->rnh_deladdr = rn_delete;
1042 	rnh->rnh_matchaddr = rn_match;
1043 	rnh->rnh_lookup = rn_lookup;
1044 	rnh->rnh_treetop = t;
1045 	return 1;
1046 }
1047 
1048 void
1049 rn_init()
1050 {
1051 	char *cp, *cplim;
1052 #ifdef _KERNEL
1053 	static int initialized;
1054 	__link_set_decl(domains, struct domain);
1055 	struct domain *const *dpp;
1056 
1057 	if (initialized)
1058 		return;
1059 	initialized = 1;
1060 
1061 	__link_set_foreach(dpp, domains) {
1062 		if ((*dpp)->dom_maxrtkey > max_keylen)
1063 			max_keylen = (*dpp)->dom_maxrtkey;
1064 	}
1065 #endif
1066 	if (max_keylen == 0) {
1067 		log(LOG_ERR,
1068 		    "rn_init: radix functions require max_keylen be set\n");
1069 		return;
1070 	}
1071 	R_Malloc(rn_zeros, char *, 3 * max_keylen);
1072 	if (rn_zeros == NULL)
1073 		panic("rn_init");
1074 	memset(rn_zeros, 0, 3 * max_keylen);
1075 	rn_ones = cp = rn_zeros + max_keylen;
1076 	addmask_key = cplim = rn_ones + max_keylen;
1077 	while (cp < cplim)
1078 		*cp++ = -1;
1079 	if (rn_inithead((void *)&mask_rnhead, 0) == 0)
1080 		panic("rn_init 2");
1081 }
1082