xref: /dragonfly/sys/net/radix.c (revision 17b61719)
1 /*
2  * Copyright (c) 1988, 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by the University of
16  *	California, Berkeley and its contributors.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *	@(#)radix.c	8.4 (Berkeley) 11/2/94
34  * $FreeBSD: src/sys/net/radix.c,v 1.20.2.3 2002/04/28 05:40:25 suz Exp $
35  * $DragonFly: src/sys/net/radix.c,v 1.5 2003/08/26 20:49:47 rob Exp $
36  */
37 
38 /*
39  * Routines to build and maintain radix trees for routing lookups.
40  */
41 #ifndef _RADIX_H_
42 #include <sys/param.h>
43 #ifdef	_KERNEL
44 #include <sys/systm.h>
45 #include <sys/malloc.h>
46 #include <sys/domain.h>
47 #else
48 #include <stdlib.h>
49 #endif
50 #include <sys/syslog.h>
51 #include <net/radix.h>
52 #endif
53 
54 static int	rn_walktree_from (struct radix_node_head *h, void *a,
55 				      void *m, walktree_f_t *f, void *w);
56 static int rn_walktree (struct radix_node_head *, walktree_f_t *, void *);
57 static struct radix_node
58 	 *rn_insert (void *, struct radix_node_head *, int *,
59 			struct radix_node [2]),
60 	 *rn_newpair (void *, int, struct radix_node[2]),
61 	 *rn_search (void *, struct radix_node *),
62 	 *rn_search_m (void *, struct radix_node *, void *);
63 
64 static int	max_keylen;
65 static struct radix_mask *rn_mkfreelist;
66 static struct radix_node_head *mask_rnhead;
67 static char *addmask_key;
68 static char normal_chars[] = {0, 0x80, 0xc0, 0xe0, 0xf0, 0xf8, 0xfc, 0xfe, -1};
69 static char *rn_zeros, *rn_ones;
70 
71 #define rn_masktop (mask_rnhead->rnh_treetop)
72 #undef Bcmp
73 #define Bcmp(a, b, l) \
74 	(l == 0 ? 0 : bcmp((caddr_t)(a), (caddr_t)(b), (u_long)l))
75 
76 static int	rn_lexobetter (void *m_arg, void *n_arg);
77 static struct radix_mask *
78 		rn_new_radix_mask (struct radix_node *tt,
79 				       struct radix_mask *next);
80 static int	rn_satsifies_leaf (char *trial, struct radix_node *leaf,
81 				       int skip);
82 
83 /*
84  * The data structure for the keys is a radix tree with one way
85  * branching removed.  The index rn_bit at an internal node n represents a bit
86  * position to be tested.  The tree is arranged so that all descendants
87  * of a node n have keys whose bits all agree up to position rn_bit - 1.
88  * (We say the index of n is rn_bit.)
89  *
90  * There is at least one descendant which has a one bit at position rn_bit,
91  * and at least one with a zero there.
92  *
93  * A route is determined by a pair of key and mask.  We require that the
94  * bit-wise logical and of the key and mask to be the key.
95  * We define the index of a route to associated with the mask to be
96  * the first bit number in the mask where 0 occurs (with bit number 0
97  * representing the highest order bit).
98  *
99  * We say a mask is normal if every bit is 0, past the index of the mask.
100  * If a node n has a descendant (k, m) with index(m) == index(n) == rn_bit,
101  * and m is a normal mask, then the route applies to every descendant of n.
102  * If the index(m) < rn_bit, this implies the trailing last few bits of k
103  * before bit b are all 0, (and hence consequently true of every descendant
104  * of n), so the route applies to all descendants of the node as well.
105  *
106  * Similar logic shows that a non-normal mask m such that
107  * index(m) <= index(n) could potentially apply to many children of n.
108  * Thus, for each non-host route, we attach its mask to a list at an internal
109  * node as high in the tree as we can go.
110  *
111  * The present version of the code makes use of normal routes in short-
112  * circuiting an explict mask and compare operation when testing whether
113  * a key satisfies a normal route, and also in remembering the unique leaf
114  * that governs a subtree.
115  */
116 
117 static struct radix_node *
118 rn_search(v_arg, head)
119 	void *v_arg;
120 	struct radix_node *head;
121 {
122 	struct radix_node *x;
123 	caddr_t v;
124 
125 	for (x = head, v = v_arg; x->rn_bit >= 0;) {
126 		if (x->rn_bmask & v[x->rn_offset])
127 			x = x->rn_right;
128 		else
129 			x = x->rn_left;
130 	}
131 	return (x);
132 }
133 
134 static struct radix_node *
135 rn_search_m(v_arg, head, m_arg)
136 	struct radix_node *head;
137 	void *v_arg, *m_arg;
138 {
139 	struct radix_node *x;
140 	caddr_t v = v_arg, m = m_arg;
141 
142 	for (x = head; x->rn_bit >= 0;) {
143 		if ((x->rn_bmask & m[x->rn_offset]) &&
144 		    (x->rn_bmask & v[x->rn_offset]))
145 			x = x->rn_right;
146 		else
147 			x = x->rn_left;
148 	}
149 	return x;
150 }
151 
152 int
153 rn_refines(m_arg, n_arg)
154 	void *m_arg, *n_arg;
155 {
156 	caddr_t m = m_arg, n = n_arg;
157 	caddr_t lim, lim2 = lim = n + *(u_char *)n;
158 	int longer = (*(u_char *)n++) - (int)(*(u_char *)m++);
159 	int masks_are_equal = 1;
160 
161 	if (longer > 0)
162 		lim -= longer;
163 	while (n < lim) {
164 		if (*n & ~(*m))
165 			return 0;
166 		if (*n++ != *m++)
167 			masks_are_equal = 0;
168 	}
169 	while (n < lim2)
170 		if (*n++)
171 			return 0;
172 	if (masks_are_equal && (longer < 0))
173 		for (lim2 = m - longer; m < lim2; )
174 			if (*m++)
175 				return 1;
176 	return (!masks_are_equal);
177 }
178 
179 struct radix_node *
180 rn_lookup(v_arg, m_arg, head)
181 	void *v_arg, *m_arg;
182 	struct radix_node_head *head;
183 {
184 	struct radix_node *x;
185 	caddr_t netmask = 0;
186 
187 	if (m_arg) {
188 		x = rn_addmask(m_arg, 1, head->rnh_treetop->rn_offset);
189 		if (x == 0)
190 			return (0);
191 		netmask = x->rn_key;
192 	}
193 	x = rn_match(v_arg, head);
194 	if (x && netmask) {
195 		while (x && x->rn_mask != netmask)
196 			x = x->rn_dupedkey;
197 	}
198 	return x;
199 }
200 
201 static int
202 rn_satsifies_leaf(trial, leaf, skip)
203 	char *trial;
204 	struct radix_node *leaf;
205 	int skip;
206 {
207 	char *cp = trial, *cp2 = leaf->rn_key, *cp3 = leaf->rn_mask;
208 	char *cplim;
209 	int length = min(*(u_char *)cp, *(u_char *)cp2);
210 
211 	if (cp3 == 0)
212 		cp3 = rn_ones;
213 	else
214 		length = min(length, *(u_char *)cp3);
215 	cplim = cp + length; cp3 += skip; cp2 += skip;
216 	for (cp += skip; cp < cplim; cp++, cp2++, cp3++)
217 		if ((*cp ^ *cp2) & *cp3)
218 			return 0;
219 	return 1;
220 }
221 
222 struct radix_node *
223 rn_match(v_arg, head)
224 	void *v_arg;
225 	struct radix_node_head *head;
226 {
227 	caddr_t v = v_arg;
228 	struct radix_node *t = head->rnh_treetop, *x;
229 	caddr_t cp = v, cp2;
230 	caddr_t cplim;
231 	struct radix_node *saved_t, *top = t;
232 	int off = t->rn_offset, vlen = *(u_char *)cp, matched_off;
233 	int test, b, rn_bit;
234 
235 	/*
236 	 * Open code rn_search(v, top) to avoid overhead of extra
237 	 * subroutine call.
238 	 */
239 	for (; t->rn_bit >= 0; ) {
240 		if (t->rn_bmask & cp[t->rn_offset])
241 			t = t->rn_right;
242 		else
243 			t = t->rn_left;
244 	}
245 	/*
246 	 * See if we match exactly as a host destination
247 	 * or at least learn how many bits match, for normal mask finesse.
248 	 *
249 	 * It doesn't hurt us to limit how many bytes to check
250 	 * to the length of the mask, since if it matches we had a genuine
251 	 * match and the leaf we have is the most specific one anyway;
252 	 * if it didn't match with a shorter length it would fail
253 	 * with a long one.  This wins big for class B&C netmasks which
254 	 * are probably the most common case...
255 	 */
256 	if (t->rn_mask)
257 		vlen = *(u_char *)t->rn_mask;
258 	cp += off; cp2 = t->rn_key + off; cplim = v + vlen;
259 	for (; cp < cplim; cp++, cp2++)
260 		if (*cp != *cp2)
261 			goto on1;
262 	/*
263 	 * This extra grot is in case we are explicitly asked
264 	 * to look up the default.  Ugh!
265 	 *
266 	 * Never return the root node itself, it seems to cause a
267 	 * lot of confusion.
268 	 */
269 	if (t->rn_flags & RNF_ROOT)
270 		t = t->rn_dupedkey;
271 	return t;
272 on1:
273 	test = (*cp ^ *cp2) & 0xff; /* find first bit that differs */
274 	for (b = 7; (test >>= 1) > 0;)
275 		b--;
276 	matched_off = cp - v;
277 	b += matched_off << 3;
278 	rn_bit = -1 - b;
279 	/*
280 	 * If there is a host route in a duped-key chain, it will be first.
281 	 */
282 	if ((saved_t = t)->rn_mask == 0)
283 		t = t->rn_dupedkey;
284 	for (; t; t = t->rn_dupedkey)
285 		/*
286 		 * Even if we don't match exactly as a host,
287 		 * we may match if the leaf we wound up at is
288 		 * a route to a net.
289 		 */
290 		if (t->rn_flags & RNF_NORMAL) {
291 			if (rn_bit <= t->rn_bit)
292 				return t;
293 		} else if (rn_satsifies_leaf(v, t, matched_off))
294 				return t;
295 	t = saved_t;
296 	/* start searching up the tree */
297 	do {
298 		struct radix_mask *m;
299 		t = t->rn_parent;
300 		m = t->rn_mklist;
301 		/*
302 		 * If non-contiguous masks ever become important
303 		 * we can restore the masking and open coding of
304 		 * the search and satisfaction test and put the
305 		 * calculation of "off" back before the "do".
306 		 */
307 		while (m) {
308 			if (m->rm_flags & RNF_NORMAL) {
309 				if (rn_bit <= m->rm_bit)
310 					return (m->rm_leaf);
311 			} else {
312 				off = min(t->rn_offset, matched_off);
313 				x = rn_search_m(v, t, m->rm_mask);
314 				while (x && x->rn_mask != m->rm_mask)
315 					x = x->rn_dupedkey;
316 				if (x && rn_satsifies_leaf(v, x, off))
317 					return x;
318 			}
319 			m = m->rm_mklist;
320 		}
321 	} while (t != top);
322 	return 0;
323 }
324 
325 #ifdef RN_DEBUG
326 int	rn_nodenum;
327 struct	radix_node *rn_clist;
328 int	rn_saveinfo;
329 int	rn_debug =  1;
330 #endif
331 
332 static struct radix_node *
333 rn_newpair(v, b, nodes)
334 	void *v;
335 	int b;
336 	struct radix_node nodes[2];
337 {
338 	struct radix_node *tt = nodes, *t = tt + 1;
339 	t->rn_bit = b;
340 	t->rn_bmask = 0x80 >> (b & 7);
341 	t->rn_left = tt;
342 	t->rn_offset = b >> 3;
343 	tt->rn_bit = -1;
344 	tt->rn_key = (caddr_t)v;
345 	tt->rn_parent = t;
346 	tt->rn_flags = t->rn_flags = RNF_ACTIVE;
347 	tt->rn_mklist = t->rn_mklist = 0;
348 #ifdef RN_DEBUG
349 	tt->rn_info = rn_nodenum++; t->rn_info = rn_nodenum++;
350 	tt->rn_twin = t;
351 	tt->rn_ybro = rn_clist;
352 	rn_clist = tt;
353 #endif
354 	return t;
355 }
356 
357 static struct radix_node *
358 rn_insert(v_arg, head, dupentry, nodes)
359 	void *v_arg;
360 	struct radix_node_head *head;
361 	int *dupentry;
362 	struct radix_node nodes[2];
363 {
364 	caddr_t v = v_arg;
365 	struct radix_node *top = head->rnh_treetop;
366 	int head_off = top->rn_offset, vlen = (int)*((u_char *)v);
367 	struct radix_node *t = rn_search(v_arg, top);
368 	caddr_t cp = v + head_off;
369 	int b;
370 	struct radix_node *tt;
371     	/*
372 	 * Find first bit at which v and t->rn_key differ
373 	 */
374     {
375 	caddr_t cp2 = t->rn_key + head_off;
376 	int cmp_res;
377 	caddr_t cplim = v + vlen;
378 
379 	while (cp < cplim)
380 		if (*cp2++ != *cp++)
381 			goto on1;
382 	*dupentry = 1;
383 	return t;
384 on1:
385 	*dupentry = 0;
386 	cmp_res = (cp[-1] ^ cp2[-1]) & 0xff;
387 	for (b = (cp - v) << 3; cmp_res; b--)
388 		cmp_res >>= 1;
389     }
390     {
391 	struct radix_node *p, *x = top;
392 	cp = v;
393 	do {
394 		p = x;
395 		if (cp[x->rn_offset] & x->rn_bmask)
396 			x = x->rn_right;
397 		else
398 			x = x->rn_left;
399 	} while (b > (unsigned) x->rn_bit);
400 				/* x->rn_bit < b && x->rn_bit >= 0 */
401 #ifdef RN_DEBUG
402 	if (rn_debug)
403 		log(LOG_DEBUG, "rn_insert: Going In:\n"), traverse(p);
404 #endif
405 	t = rn_newpair(v_arg, b, nodes);
406 	tt = t->rn_left;
407 	if ((cp[p->rn_offset] & p->rn_bmask) == 0)
408 		p->rn_left = t;
409 	else
410 		p->rn_right = t;
411 	x->rn_parent = t;
412 	t->rn_parent = p; /* frees x, p as temp vars below */
413 	if ((cp[t->rn_offset] & t->rn_bmask) == 0) {
414 		t->rn_right = x;
415 	} else {
416 		t->rn_right = tt;
417 		t->rn_left = x;
418 	}
419 #ifdef RN_DEBUG
420 	if (rn_debug)
421 		log(LOG_DEBUG, "rn_insert: Coming Out:\n"), traverse(p);
422 #endif
423     }
424 	return (tt);
425 }
426 
427 struct radix_node *
428 rn_addmask(n_arg, search, skip)
429 	int search, skip;
430 	void *n_arg;
431 {
432 	caddr_t netmask = (caddr_t)n_arg;
433 	struct radix_node *x;
434 	caddr_t cp, cplim;
435 	int b = 0, mlen, j;
436 	int maskduplicated, m0, isnormal;
437 	struct radix_node *saved_x;
438 	static int last_zeroed = 0;
439 
440 	if ((mlen = *(u_char *)netmask) > max_keylen)
441 		mlen = max_keylen;
442 	if (skip == 0)
443 		skip = 1;
444 	if (mlen <= skip)
445 		return (mask_rnhead->rnh_nodes);
446 	if (skip > 1)
447 		Bcopy(rn_ones + 1, addmask_key + 1, skip - 1);
448 	if ((m0 = mlen) > skip)
449 		Bcopy(netmask + skip, addmask_key + skip, mlen - skip);
450 	/*
451 	 * Trim trailing zeroes.
452 	 */
453 	for (cp = addmask_key + mlen; (cp > addmask_key) && cp[-1] == 0;)
454 		cp--;
455 	mlen = cp - addmask_key;
456 	if (mlen <= skip) {
457 		if (m0 >= last_zeroed)
458 			last_zeroed = mlen;
459 		return (mask_rnhead->rnh_nodes);
460 	}
461 	if (m0 < last_zeroed)
462 		Bzero(addmask_key + m0, last_zeroed - m0);
463 	*addmask_key = last_zeroed = mlen;
464 	x = rn_search(addmask_key, rn_masktop);
465 	if (Bcmp(addmask_key, x->rn_key, mlen) != 0)
466 		x = 0;
467 	if (x || search)
468 		return (x);
469 	R_Malloc(x, struct radix_node *, max_keylen + 2 * sizeof (*x));
470 	if ((saved_x = x) == 0)
471 		return (0);
472 	Bzero(x, max_keylen + 2 * sizeof (*x));
473 	netmask = cp = (caddr_t)(x + 2);
474 	Bcopy(addmask_key, cp, mlen);
475 	x = rn_insert(cp, mask_rnhead, &maskduplicated, x);
476 	if (maskduplicated) {
477 		log(LOG_ERR, "rn_addmask: mask impossibly already in tree");
478 		Free(saved_x);
479 		return (x);
480 	}
481 	/*
482 	 * Calculate index of mask, and check for normalcy.
483 	 */
484 	cplim = netmask + mlen; isnormal = 1;
485 	for (cp = netmask + skip; (cp < cplim) && *(u_char *)cp == 0xff;)
486 		cp++;
487 	if (cp != cplim) {
488 		for (j = 0x80; (j & *cp) != 0; j >>= 1)
489 			b++;
490 		if (*cp != normal_chars[b] || cp != (cplim - 1))
491 			isnormal = 0;
492 	}
493 	b += (cp - netmask) << 3;
494 	x->rn_bit = -1 - b;
495 	if (isnormal)
496 		x->rn_flags |= RNF_NORMAL;
497 	return (x);
498 }
499 
500 static int	/* XXX: arbitrary ordering for non-contiguous masks */
501 rn_lexobetter(m_arg, n_arg)
502 	void *m_arg, *n_arg;
503 {
504 	u_char *mp = m_arg, *np = n_arg, *lim;
505 
506 	if (*mp > *np)
507 		return 1;  /* not really, but need to check longer one first */
508 	if (*mp == *np)
509 		for (lim = mp + *mp; mp < lim;)
510 			if (*mp++ > *np++)
511 				return 1;
512 	return 0;
513 }
514 
515 static struct radix_mask *
516 rn_new_radix_mask(tt, next)
517 	struct radix_node *tt;
518 	struct radix_mask *next;
519 {
520 	struct radix_mask *m;
521 
522 	MKGet(m);
523 	if (m == 0) {
524 		log(LOG_ERR, "Mask for route not entered\n");
525 		return (0);
526 	}
527 	Bzero(m, sizeof *m);
528 	m->rm_bit = tt->rn_bit;
529 	m->rm_flags = tt->rn_flags;
530 	if (tt->rn_flags & RNF_NORMAL)
531 		m->rm_leaf = tt;
532 	else
533 		m->rm_mask = tt->rn_mask;
534 	m->rm_mklist = next;
535 	tt->rn_mklist = m;
536 	return m;
537 }
538 
539 struct radix_node *
540 rn_addroute(v_arg, n_arg, head, treenodes)
541 	void *v_arg, *n_arg;
542 	struct radix_node_head *head;
543 	struct radix_node treenodes[2];
544 {
545 	caddr_t v = (caddr_t)v_arg, netmask = (caddr_t)n_arg;
546 	struct radix_node *t, *x = 0, *tt;
547 	struct radix_node *saved_tt, *top = head->rnh_treetop;
548 	short b = 0, b_leaf = 0;
549 	int keyduplicated;
550 	caddr_t mmask;
551 	struct radix_mask *m, **mp;
552 
553 	/*
554 	 * In dealing with non-contiguous masks, there may be
555 	 * many different routes which have the same mask.
556 	 * We will find it useful to have a unique pointer to
557 	 * the mask to speed avoiding duplicate references at
558 	 * nodes and possibly save time in calculating indices.
559 	 */
560 	if (netmask)  {
561 		if ((x = rn_addmask(netmask, 0, top->rn_offset)) == 0)
562 			return (0);
563 		b_leaf = x->rn_bit;
564 		b = -1 - x->rn_bit;
565 		netmask = x->rn_key;
566 	}
567 	/*
568 	 * Deal with duplicated keys: attach node to previous instance
569 	 */
570 	saved_tt = tt = rn_insert(v, head, &keyduplicated, treenodes);
571 	if (keyduplicated) {
572 		for (t = tt; tt; t = tt, tt = tt->rn_dupedkey) {
573 			if (tt->rn_mask == netmask)
574 				return (0);
575 			if (netmask == 0 ||
576 			    (tt->rn_mask &&
577 			     ((b_leaf < tt->rn_bit) /* index(netmask) > node */
578 			      || rn_refines(netmask, tt->rn_mask)
579 			      || rn_lexobetter(netmask, tt->rn_mask))))
580 				break;
581 		}
582 		/*
583 		 * If the mask is not duplicated, we wouldn't
584 		 * find it among possible duplicate key entries
585 		 * anyway, so the above test doesn't hurt.
586 		 *
587 		 * We sort the masks for a duplicated key the same way as
588 		 * in a masklist -- most specific to least specific.
589 		 * This may require the unfortunate nuisance of relocating
590 		 * the head of the list.
591 		 */
592 		if (tt == saved_tt) {
593 			struct	radix_node *xx = x;
594 			/* link in at head of list */
595 			(tt = treenodes)->rn_dupedkey = t;
596 			tt->rn_flags = t->rn_flags;
597 			tt->rn_parent = x = t->rn_parent;
598 			t->rn_parent = tt;	 		/* parent */
599 			if (x->rn_left == t)
600 				x->rn_left = tt;
601 			else
602 				x->rn_right = tt;
603 			saved_tt = tt; x = xx;
604 		} else {
605 			(tt = treenodes)->rn_dupedkey = t->rn_dupedkey;
606 			t->rn_dupedkey = tt;
607 			tt->rn_parent = t;			/* parent */
608 			if (tt->rn_dupedkey)			/* parent */
609 				tt->rn_dupedkey->rn_parent = tt; /* parent */
610 		}
611 #ifdef RN_DEBUG
612 		t=tt+1; tt->rn_info = rn_nodenum++; t->rn_info = rn_nodenum++;
613 		tt->rn_twin = t; tt->rn_ybro = rn_clist; rn_clist = tt;
614 #endif
615 		tt->rn_key = (caddr_t) v;
616 		tt->rn_bit = -1;
617 		tt->rn_flags = RNF_ACTIVE;
618 	}
619 	/*
620 	 * Put mask in tree.
621 	 */
622 	if (netmask) {
623 		tt->rn_mask = netmask;
624 		tt->rn_bit = x->rn_bit;
625 		tt->rn_flags |= x->rn_flags & RNF_NORMAL;
626 	}
627 	t = saved_tt->rn_parent;
628 	if (keyduplicated)
629 		goto on2;
630 	b_leaf = -1 - t->rn_bit;
631 	if (t->rn_right == saved_tt)
632 		x = t->rn_left;
633 	else
634 		x = t->rn_right;
635 	/* Promote general routes from below */
636 	if (x->rn_bit < 0) {
637 	    for (mp = &t->rn_mklist; x; x = x->rn_dupedkey)
638 		if (x->rn_mask && (x->rn_bit >= b_leaf) && x->rn_mklist == 0) {
639 			*mp = m = rn_new_radix_mask(x, 0);
640 			if (m)
641 				mp = &m->rm_mklist;
642 		}
643 	} else if (x->rn_mklist) {
644 		/*
645 		 * Skip over masks whose index is > that of new node
646 		 */
647 		for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist)
648 			if (m->rm_bit >= b_leaf)
649 				break;
650 		t->rn_mklist = m; *mp = 0;
651 	}
652 on2:
653 	/* Add new route to highest possible ancestor's list */
654 	if ((netmask == 0) || (b > t->rn_bit ))
655 		return tt; /* can't lift at all */
656 	b_leaf = tt->rn_bit;
657 	do {
658 		x = t;
659 		t = t->rn_parent;
660 	} while (b <= t->rn_bit && x != top);
661 	/*
662 	 * Search through routes associated with node to
663 	 * insert new route according to index.
664 	 * Need same criteria as when sorting dupedkeys to avoid
665 	 * double loop on deletion.
666 	 */
667 	for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist) {
668 		if (m->rm_bit < b_leaf)
669 			continue;
670 		if (m->rm_bit > b_leaf)
671 			break;
672 		if (m->rm_flags & RNF_NORMAL) {
673 			mmask = m->rm_leaf->rn_mask;
674 			if (tt->rn_flags & RNF_NORMAL) {
675 			    log(LOG_ERR,
676 			        "Non-unique normal route, mask not entered\n");
677 				return tt;
678 			}
679 		} else
680 			mmask = m->rm_mask;
681 		if (mmask == netmask) {
682 			m->rm_refs++;
683 			tt->rn_mklist = m;
684 			return tt;
685 		}
686 		if (rn_refines(netmask, mmask)
687 		    || rn_lexobetter(netmask, mmask))
688 			break;
689 	}
690 	*mp = rn_new_radix_mask(tt, *mp);
691 	return tt;
692 }
693 
694 struct radix_node *
695 rn_delete(v_arg, netmask_arg, head)
696 	void *v_arg, *netmask_arg;
697 	struct radix_node_head *head;
698 {
699 	struct radix_node *t, *p, *x, *tt;
700 	struct radix_mask *m, *saved_m, **mp;
701 	struct radix_node *dupedkey, *saved_tt, *top;
702 	caddr_t v, netmask;
703 	int b, head_off, vlen;
704 
705 	v = v_arg;
706 	netmask = netmask_arg;
707 	x = head->rnh_treetop;
708 	tt = rn_search(v, x);
709 	head_off = x->rn_offset;
710 	vlen =  *(u_char *)v;
711 	saved_tt = tt;
712 	top = x;
713 	if (tt == 0 ||
714 	    Bcmp(v + head_off, tt->rn_key + head_off, vlen - head_off))
715 		return (0);
716 	/*
717 	 * Delete our route from mask lists.
718 	 */
719 	if (netmask) {
720 		if ((x = rn_addmask(netmask, 1, head_off)) == 0)
721 			return (0);
722 		netmask = x->rn_key;
723 		while (tt->rn_mask != netmask)
724 			if ((tt = tt->rn_dupedkey) == 0)
725 				return (0);
726 	}
727 	if (tt->rn_mask == 0 || (saved_m = m = tt->rn_mklist) == 0)
728 		goto on1;
729 	if (tt->rn_flags & RNF_NORMAL) {
730 		if (m->rm_leaf != tt || m->rm_refs > 0) {
731 			log(LOG_ERR, "rn_delete: inconsistent annotation\n");
732 			return 0;  /* dangling ref could cause disaster */
733 		}
734 	} else {
735 		if (m->rm_mask != tt->rn_mask) {
736 			log(LOG_ERR, "rn_delete: inconsistent annotation\n");
737 			goto on1;
738 		}
739 		if (--m->rm_refs >= 0)
740 			goto on1;
741 	}
742 	b = -1 - tt->rn_bit;
743 	t = saved_tt->rn_parent;
744 	if (b > t->rn_bit)
745 		goto on1; /* Wasn't lifted at all */
746 	do {
747 		x = t;
748 		t = t->rn_parent;
749 	} while (b <= t->rn_bit && x != top);
750 	for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist)
751 		if (m == saved_m) {
752 			*mp = m->rm_mklist;
753 			MKFree(m);
754 			break;
755 		}
756 	if (m == 0) {
757 		log(LOG_ERR, "rn_delete: couldn't find our annotation\n");
758 		if (tt->rn_flags & RNF_NORMAL)
759 			return (0); /* Dangling ref to us */
760 	}
761 on1:
762 	/*
763 	 * Eliminate us from tree
764 	 */
765 	if (tt->rn_flags & RNF_ROOT)
766 		return (0);
767 #ifdef RN_DEBUG
768 	/* Get us out of the creation list */
769 	for (t = rn_clist; t && t->rn_ybro != tt; t = t->rn_ybro) {}
770 	if (t) t->rn_ybro = tt->rn_ybro;
771 #endif
772 	t = tt->rn_parent;
773 	dupedkey = saved_tt->rn_dupedkey;
774 	if (dupedkey) {
775 		/*
776 		 * at this point, tt is the deletion target and saved_tt
777 		 * is the head of the dupekey chain
778 		 */
779 		if (tt == saved_tt) {
780 			/* remove from head of chain */
781 			x = dupedkey; x->rn_parent = t;
782 			if (t->rn_left == tt)
783 				t->rn_left = x;
784 			else
785 				t->rn_right = x;
786 		} else {
787 			/* find node in front of tt on the chain */
788 			for (x = p = saved_tt; p && p->rn_dupedkey != tt;)
789 				p = p->rn_dupedkey;
790 			if (p) {
791 				p->rn_dupedkey = tt->rn_dupedkey;
792 				if (tt->rn_dupedkey)		/* parent */
793 					tt->rn_dupedkey->rn_parent = p;
794 								/* parent */
795 			} else log(LOG_ERR, "rn_delete: couldn't find us\n");
796 		}
797 		t = tt + 1;
798 		if  (t->rn_flags & RNF_ACTIVE) {
799 #ifndef RN_DEBUG
800 			*++x = *t;
801 			p = t->rn_parent;
802 #else
803 			b = t->rn_info;
804 			*++x = *t;
805 			t->rn_info = b;
806 			p = t->rn_parent;
807 #endif
808 			if (p->rn_left == t)
809 				p->rn_left = x;
810 			else
811 				p->rn_right = x;
812 			x->rn_left->rn_parent = x;
813 			x->rn_right->rn_parent = x;
814 		}
815 		goto out;
816 	}
817 	if (t->rn_left == tt)
818 		x = t->rn_right;
819 	else
820 		x = t->rn_left;
821 	p = t->rn_parent;
822 	if (p->rn_right == t)
823 		p->rn_right = x;
824 	else
825 		p->rn_left = x;
826 	x->rn_parent = p;
827 	/*
828 	 * Demote routes attached to us.
829 	 */
830 	if (t->rn_mklist) {
831 		if (x->rn_bit >= 0) {
832 			for (mp = &x->rn_mklist; (m = *mp);)
833 				mp = &m->rm_mklist;
834 			*mp = t->rn_mklist;
835 		} else {
836 			/* If there are any key,mask pairs in a sibling
837 			   duped-key chain, some subset will appear sorted
838 			   in the same order attached to our mklist */
839 			for (m = t->rn_mklist; m && x; x = x->rn_dupedkey)
840 				if (m == x->rn_mklist) {
841 					struct radix_mask *mm = m->rm_mklist;
842 					x->rn_mklist = 0;
843 					if (--(m->rm_refs) < 0)
844 						MKFree(m);
845 					m = mm;
846 				}
847 			if (m)
848 				log(LOG_ERR,
849 				    "rn_delete: Orphaned Mask %p at %p\n",
850 				    (void *)m, (void *)x);
851 		}
852 	}
853 	/*
854 	 * We may be holding an active internal node in the tree.
855 	 */
856 	x = tt + 1;
857 	if (t != x) {
858 #ifndef RN_DEBUG
859 		*t = *x;
860 #else
861 		b = t->rn_info;
862 		*t = *x;
863 		t->rn_info = b;
864 #endif
865 		t->rn_left->rn_parent = t;
866 		t->rn_right->rn_parent = t;
867 		p = x->rn_parent;
868 		if (p->rn_left == x)
869 			p->rn_left = t;
870 		else
871 			p->rn_right = t;
872 	}
873 out:
874 	tt->rn_flags &= ~RNF_ACTIVE;
875 	tt[1].rn_flags &= ~RNF_ACTIVE;
876 	return (tt);
877 }
878 
879 /*
880  * This is the same as rn_walktree() except for the parameters and the
881  * exit.
882  */
883 static int
884 rn_walktree_from(h, a, m, f, w)
885 	struct radix_node_head *h;
886 	void *a, *m;
887 	walktree_f_t *f;
888 	void *w;
889 {
890 	int error;
891 	struct radix_node *base, *next;
892 	u_char *xa = (u_char *)a;
893 	u_char *xm = (u_char *)m;
894 	struct radix_node *rn, *last = 0 /* shut up gcc */;
895 	int stopping = 0;
896 	int lastb;
897 
898 	/*
899 	 * rn_search_m is sort-of-open-coded here.
900 	 */
901 	/* printf("about to search\n"); */
902 	for (rn = h->rnh_treetop; rn->rn_bit >= 0; ) {
903 		last = rn;
904 		/* printf("rn_bit %d, rn_bmask %x, xm[rn_offset] %x\n",
905 		       rn->rn_bit, rn->rn_bmask, xm[rn->rn_offset]); */
906 		if (!(rn->rn_bmask & xm[rn->rn_offset])) {
907 			break;
908 		}
909 		if (rn->rn_bmask & xa[rn->rn_offset]) {
910 			rn = rn->rn_right;
911 		} else {
912 			rn = rn->rn_left;
913 		}
914 	}
915 	/* printf("done searching\n"); */
916 
917 	/*
918 	 * Two cases: either we stepped off the end of our mask,
919 	 * in which case last == rn, or we reached a leaf, in which
920 	 * case we want to start from the last node we looked at.
921 	 * Either way, last is the node we want to start from.
922 	 */
923 	rn = last;
924 	lastb = rn->rn_bit;
925 
926 	/* printf("rn %p, lastb %d\n", rn, lastb);*/
927 
928 	/*
929 	 * This gets complicated because we may delete the node
930 	 * while applying the function f to it, so we need to calculate
931 	 * the successor node in advance.
932 	 */
933 	while (rn->rn_bit >= 0)
934 		rn = rn->rn_left;
935 
936 	while (!stopping) {
937 		/* printf("node %p (%d)\n", rn, rn->rn_bit); */
938 		base = rn;
939 		/* If at right child go back up, otherwise, go right */
940 		while (rn->rn_parent->rn_right == rn
941 		       && !(rn->rn_flags & RNF_ROOT)) {
942 			rn = rn->rn_parent;
943 
944 			/* if went up beyond last, stop */
945 			if (rn->rn_bit < lastb) {
946 				stopping = 1;
947 				/* printf("up too far\n"); */
948 			}
949 		}
950 
951 		/* Find the next *leaf* since next node might vanish, too */
952 		for (rn = rn->rn_parent->rn_right; rn->rn_bit >= 0;)
953 			rn = rn->rn_left;
954 		next = rn;
955 		/* Process leaves */
956 		while ((rn = base) != 0) {
957 			base = rn->rn_dupedkey;
958 			/* printf("leaf %p\n", rn); */
959 			if (!(rn->rn_flags & RNF_ROOT)
960 			    && (error = (*f)(rn, w)))
961 				return (error);
962 		}
963 		rn = next;
964 
965 		if (rn->rn_flags & RNF_ROOT) {
966 			/* printf("root, stopping"); */
967 			stopping = 1;
968 		}
969 
970 	}
971 	return 0;
972 }
973 
974 static int
975 rn_walktree(h, f, w)
976 	struct radix_node_head *h;
977 	walktree_f_t *f;
978 	void *w;
979 {
980 	int error;
981 	struct radix_node *base, *next;
982 	struct radix_node *rn = h->rnh_treetop;
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 	/* First time through node, go left */
989 	while (rn->rn_bit >= 0)
990 		rn = rn->rn_left;
991 	for (;;) {
992 		base = rn;
993 		/* If at right child go back up, otherwise, go right */
994 		while (rn->rn_parent->rn_right == rn
995 		       && (rn->rn_flags & RNF_ROOT) == 0)
996 			rn = rn->rn_parent;
997 		/* Find the next *leaf* since next node might vanish, too */
998 		for (rn = rn->rn_parent->rn_right; rn->rn_bit >= 0;)
999 			rn = rn->rn_left;
1000 		next = rn;
1001 		/* Process leaves */
1002 		while ((rn = base)) {
1003 			base = rn->rn_dupedkey;
1004 			if (!(rn->rn_flags & RNF_ROOT)
1005 			    && (error = (*f)(rn, w)))
1006 				return (error);
1007 		}
1008 		rn = next;
1009 		if (rn->rn_flags & RNF_ROOT)
1010 			return (0);
1011 	}
1012 	/* NOTREACHED */
1013 }
1014 
1015 int
1016 rn_inithead(head, off)
1017 	void **head;
1018 	int off;
1019 {
1020 	struct radix_node_head *rnh;
1021 	struct radix_node *t, *tt, *ttt;
1022 	if (*head)
1023 		return (1);
1024 	R_Malloc(rnh, struct radix_node_head *, sizeof (*rnh));
1025 	if (rnh == 0)
1026 		return (0);
1027 	Bzero(rnh, sizeof (*rnh));
1028 	*head = rnh;
1029 	t = rn_newpair(rn_zeros, off, rnh->rnh_nodes);
1030 	ttt = rnh->rnh_nodes + 2;
1031 	t->rn_right = ttt;
1032 	t->rn_parent = t;
1033 	tt = t->rn_left;
1034 	tt->rn_flags = t->rn_flags = RNF_ROOT | RNF_ACTIVE;
1035 	tt->rn_bit = -1 - off;
1036 	*ttt = *tt;
1037 	ttt->rn_key = rn_ones;
1038 	rnh->rnh_addaddr = rn_addroute;
1039 	rnh->rnh_deladdr = rn_delete;
1040 	rnh->rnh_matchaddr = rn_match;
1041 	rnh->rnh_lookup = rn_lookup;
1042 	rnh->rnh_walktree = rn_walktree;
1043 	rnh->rnh_walktree_from = rn_walktree_from;
1044 	rnh->rnh_treetop = t;
1045 	return (1);
1046 }
1047 
1048 void
1049 rn_init()
1050 {
1051 	char *cp, *cplim;
1052 #ifdef _KERNEL
1053 	struct domain *dom;
1054 
1055 	for (dom = domains; dom; dom = dom->dom_next)
1056 		if (dom->dom_maxrtkey > max_keylen)
1057 			max_keylen = dom->dom_maxrtkey;
1058 #endif
1059 	if (max_keylen == 0) {
1060 		log(LOG_ERR,
1061 		    "rn_init: radix functions require max_keylen be set\n");
1062 		return;
1063 	}
1064 	R_Malloc(rn_zeros, char *, 3 * max_keylen);
1065 	if (rn_zeros == NULL)
1066 		panic("rn_init");
1067 	Bzero(rn_zeros, 3 * max_keylen);
1068 	rn_ones = cp = rn_zeros + max_keylen;
1069 	addmask_key = cplim = rn_ones + max_keylen;
1070 	while (cp < cplim)
1071 		*cp++ = -1;
1072 	if (rn_inithead((void **)&mask_rnhead, 0) == 0)
1073 		panic("rn_init 2");
1074 }
1075