xref: /titanic_41/usr/src/cmd/cmd-inet/usr.sbin/in.routed/radix.c (revision 7c478bd95313f5f23a4c958a745db2134aa03244)
1 /*
2  * Copyright 2001-2003 Sun Microsystems, Inc.  All rights reserved.
3  * Use is subject to license terms.
4  *
5  * Copyright (c) 1988, 1989, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgment:
18  *	This product includes software developed by the University of
19  *	California, Berkeley and its contributors.
20  * 4. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)radix.c	8.4 (Berkeley) 11/2/94
37  *
38  * $FreeBSD: src/sbin/routed/radix.c,v 1.6 2000/08/11 08:24:38 sheldonh Exp $
39  */
40 
41 #pragma ident	"%Z%%M%	%I%	%E% SMI"
42 
43 /*
44  * Routines to build and maintain radix trees for routing lookups.
45  */
46 
47 #include "defs.h"
48 
49 static const size_t max_keylen = sizeof (struct sockaddr_in);
50 static struct radix_mask *rn_mkfreelist;
51 static struct radix_node_head *mask_rnhead;
52 static uint8_t *rn_zeros, *rn_ones, *addmask_key;
53 
54 #define	rn_masktop (mask_rnhead->rnh_treetop)
55 
56 static boolean_t rn_satisfies_leaf(uint8_t *, struct radix_node *, int);
57 
58 static boolean_t rn_refines(void *, void *);
59 
60 static struct radix_node
61 	 *rn_addmask(void *, uint_t, uint_t),
62 	 *rn_addroute(void *, void *, struct radix_node_head *,
63 	    struct radix_node [2]),
64 	 *rn_delete(void *, void *, struct radix_node_head *),
65 	 *rn_insert(void *, struct radix_node_head *, boolean_t *,
66 	    struct radix_node [2]),
67 	 *rn_match(void *, struct radix_node_head *),
68 	 *rn_newpair(void *, uint_t, struct radix_node[2]),
69 	 *rn_search(void *, struct radix_node *),
70 	 *rn_search_m(void *, struct radix_node *, void *);
71 
72 static struct radix_node *rn_lookup(void *, void *, struct radix_node_head *);
73 
74 #ifdef DEBUG
75 #define	DBGMSG(x)	msglog x
76 #else
77 #define	DBGMSG(x)	(void) 0
78 #endif
79 
80 /*
81  * The data structure for the keys is a radix tree with one way
82  * branching removed.  The index rn_b at an internal node n represents a bit
83  * position to be tested.  The tree is arranged so that all descendants
84  * of a node n have keys whose bits all agree up to position rn_b - 1.
85  * (We say the index of n is rn_b.)
86  *
87  * There is at least one descendant which has a one bit at position rn_b,
88  * and at least one with a zero there.
89  *
90  * A route is determined by a pair of key and mask.  We require that the
91  * bit-wise logical and of the key and mask to be the key.
92  * We define the index of a route to associated with the mask to be
93  * the first bit number in the mask where 0 occurs (with bit number 0
94  * representing the highest order bit).
95  *
96  * We say a mask is normal if every bit is 0, past the index of the mask.
97  * If a node n has a descendant (k, m) with index(m) == index(n) == rn_b,
98  * and m is a normal mask, then the route applies to every descendant of n.
99  * If the index(m) < rn_b, this implies the trailing last few bits of k
100  * before bit b are all 0, (and hence consequently true of every descendant
101  * of n), so the route applies to all descendants of the node as well.
102  *
103  * Similar logic shows that a non-normal mask m such that
104  * index(m) <= index(n) could potentially apply to many children of n.
105  * Thus, for each non-host route, we attach its mask to a list at an internal
106  * node as high in the tree as we can go.
107  *
108  * The present version of the code makes use of normal routes in short-
109  * circuiting an explict mask and compare operation when testing whether
110  * a key satisfies a normal route, and also in remembering the unique leaf
111  * that governs a subtree.
112  */
113 
114 static struct radix_node *
rn_search(void * v_arg,struct radix_node * head)115 rn_search(void *v_arg, struct radix_node *head)
116 {
117 	struct radix_node *x;
118 	uint8_t *v;
119 
120 	for (x = head, v = v_arg; x->rn_b >= 0; ) {
121 		if (x->rn_bmask & v[x->rn_off])
122 			x = x->rn_r;
123 		else
124 			x = x->rn_l;
125 	}
126 	return (x);
127 }
128 
129 static struct radix_node *
rn_search_m(void * v_arg,struct radix_node * head,void * m_arg)130 rn_search_m(void *v_arg, struct radix_node *head, void *m_arg)
131 {
132 	struct radix_node *x;
133 	uint8_t *v = v_arg, *m = m_arg;
134 
135 	for (x = head; x->rn_b >= 0; ) {
136 		if (x->rn_bmask & m[x->rn_off] & v[x->rn_off])
137 			x = x->rn_r;
138 		else
139 			x = x->rn_l;
140 	}
141 	return (x);
142 }
143 
144 /*
145  * Returns true if there are no bits set in n_arg that are zero in
146  * m_arg and the masks aren't equal.  In other words, it returns true
147  * when m_arg is a finer-granularity netmask -- it represents a subset
148  * of the destinations implied by n_arg.
149  */
150 static boolean_t
rn_refines(void * m_arg,void * n_arg)151 rn_refines(void* m_arg, void *n_arg)
152 {
153 	uint8_t *m = m_arg, *n = n_arg;
154 	uint8_t *lim;
155 	boolean_t masks_are_equal = _B_TRUE;
156 
157 	lim = n + sizeof (struct sockaddr);
158 
159 	while (n < lim) {
160 		if (*n & ~(*m))
161 			return (_B_FALSE);
162 		if (*n++ != *m++)
163 			masks_are_equal = _B_FALSE;
164 	}
165 	return (!masks_are_equal);
166 }
167 
168 static struct radix_node *
rn_lookup(void * v_arg,void * m_arg,struct radix_node_head * head)169 rn_lookup(void *v_arg, void *m_arg, struct radix_node_head *head)
170 {
171 	struct radix_node *x;
172 	uint8_t *netmask = NULL;
173 
174 	if (m_arg) {
175 		if ((x = rn_addmask(m_arg, 1, head->rnh_treetop->rn_off)) ==
176 		    NULL) {
177 			DBGMSG(("rn_lookup: failed to add mask"));
178 			return (NULL);
179 		}
180 		netmask = x->rn_key;
181 	}
182 	x = rn_match(v_arg, head);
183 	if (x && netmask) {
184 		while (x && x->rn_mask != netmask)
185 			x = x->rn_dupedkey;
186 	}
187 	return (x);
188 }
189 
190 /*
191  * Returns true if address 'trial' has no bits differing from the
192  * leaf's key when compared under the leaf's mask.  In other words,
193  * returns true when 'trial' matches leaf.
194  */
195 static boolean_t
rn_satisfies_leaf(uint8_t * trial,struct radix_node * leaf,int skip)196 rn_satisfies_leaf(uint8_t *trial,
197     struct radix_node *leaf,
198     int skip)
199 {
200 	uint8_t *cp = trial, *cp2 = leaf->rn_key, *cp3 = leaf->rn_mask;
201 	uint8_t *cplim;
202 	size_t length;
203 
204 	length = sizeof (struct sockaddr);
205 
206 	if (cp3 == NULL)
207 		cp3 = rn_ones;
208 	cplim = cp + length;
209 	cp3 += skip;
210 	cp2 += skip;
211 	for (cp += skip; cp < cplim; cp++, cp2++, cp3++)
212 		if ((*cp ^ *cp2) & *cp3)
213 			return (_B_FALSE);
214 	return (_B_TRUE);
215 }
216 
217 static struct radix_node *
rn_match(void * v_arg,struct radix_node_head * head)218 rn_match(void *v_arg, struct radix_node_head *head)
219 {
220 	uint8_t *v = v_arg;
221 	struct radix_node *t = head->rnh_treetop, *x;
222 	uint8_t *cp = v, *cp2;
223 	uint8_t *cplim;
224 	struct radix_node *saved_t, *top = t;
225 	uint_t off = t->rn_off, vlen, matched_off;
226 	int test, b, rn_b;
227 
228 	vlen = sizeof (struct sockaddr);
229 
230 	/*
231 	 * Open code rn_search(v, top) to avoid overhead of extra
232 	 * subroutine call.
233 	 */
234 	for (; t->rn_b >= 0; ) {
235 		if (t->rn_bmask & cp[t->rn_off])
236 			t = t->rn_r;
237 		else
238 			t = t->rn_l;
239 	}
240 
241 	cp += off;
242 	cp2 = t->rn_key + off;
243 	cplim = v + vlen;
244 	for (; cp < cplim; cp++, cp2++)
245 		if (*cp != *cp2)
246 			goto found_difference_with_key;
247 	/*
248 	 * This extra grot is in case we are explicitly asked
249 	 * to look up the default.  Ugh!
250 	 * Or 255.255.255.255
251 	 *
252 	 * In this case, we have a complete match of the key.  Unless
253 	 * the node is one of the roots, we are finished.
254 	 * If it is the zeros root, then take what we have, prefering
255 	 * any real data.
256 	 * If it is the ones root, then pretend the target key was followed
257 	 * by a byte of zeros.
258 	 */
259 	if (!(t->rn_flags & RNF_ROOT))
260 		return (t);		/* not a root */
261 	if (t->rn_dupedkey) {
262 		t = t->rn_dupedkey;
263 		return (t);		/* have some real data */
264 	}
265 	if (*(cp-1) == 0)
266 		return (t);		/* not the ones root */
267 	b = 0;				/* fake a zero after 255.255.255.255 */
268 	goto calculated_differing_bit;
269 found_difference_with_key:
270 	test = (*cp ^ *cp2) & 0xff; /* find first bit that differs */
271 	for (b = 7; (test >>= 1) > 0; )
272 		b--;
273 calculated_differing_bit:
274 	matched_off = cp - v;
275 	b += matched_off << 3;
276 	rn_b = -1 - b;
277 	/*
278 	 * If there is a host route in a duped-key chain, it will be first.
279 	 */
280 	if ((saved_t = t)->rn_mask == NULL)
281 		t = t->rn_dupedkey;
282 	for (; t; t = t->rn_dupedkey) {
283 		/*
284 		 * Even if we don't match exactly as a host,
285 		 * we may match if the leaf we wound up at is
286 		 * a route to a net.
287 		 */
288 		if (t->rn_flags & RNF_NORMAL) {
289 			if (rn_b <= t->rn_b)
290 				return (t);
291 		} else if (rn_satisfies_leaf(v, t, matched_off)) {
292 			return (t);
293 		}
294 	}
295 	t = saved_t;
296 	/* start searching up the tree */
297 	do {
298 		struct radix_mask *m;
299 		t = t->rn_p;
300 		if ((m = t->rn_mklist) != NULL) {
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 			do {
308 				if (m->rm_flags & RNF_NORMAL) {
309 					if (rn_b <= m->rm_b)
310 						return (m->rm_leaf);
311 				} else {
312 					off = MIN(t->rn_off, matched_off);
313 					x = rn_search_m(v, t, m->rm_mask);
314 					while (x != NULL &&
315 					    x->rn_mask != m->rm_mask)
316 						x = x->rn_dupedkey;
317 					if (x != NULL &&
318 					    rn_satisfies_leaf(v, x, off))
319 						return (x);
320 				}
321 			} while ((m = m->rm_mklist) != NULL);
322 		}
323 	} while (t != top);
324 	return (NULL);
325 }
326 
327 #ifdef RN_DEBUG
328 int	rn_nodenum;
329 struct	radix_node *rn_clist;
330 int	rn_saveinfo;
331 boolean_t	rn_debug =  1;
332 #endif
333 
334 static struct radix_node *
rn_newpair(void * v,uint_t b,struct radix_node nodes[2])335 rn_newpair(void *v, uint_t b, struct radix_node nodes[2])
336 {
337 	struct radix_node *tt = nodes, *t = tt + 1;
338 
339 	t->rn_b = b;
340 	t->rn_bmask = 0x80 >> (b & 7);
341 	t->rn_l = tt;
342 	t->rn_off = b >> 3;
343 	tt->rn_b = -1;
344 	tt->rn_key = v;
345 	tt->rn_p = t;
346 	tt->rn_flags = t->rn_flags = RNF_ACTIVE;
347 #ifdef RN_DEBUG
348 	tt->rn_info = rn_nodenum++;
349 	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 *
rn_insert(void * v_arg,struct radix_node_head * head,boolean_t * dupentry,struct radix_node nodes[2])358 rn_insert(void* v_arg, struct radix_node_head *head, boolean_t *dupentry,
359     struct radix_node nodes[2])
360 {
361 	uint8_t *v = v_arg;
362 	struct radix_node *top = head->rnh_treetop;
363 	uint_t head_off = top->rn_off, vlen;
364 	struct radix_node *t = rn_search(v_arg, top);
365 	uint8_t *cp = v + head_off, b;
366 	struct radix_node *tt;
367 
368 	vlen = sizeof (struct sockaddr);
369 
370 	/*
371 	 * Find first bit at which v and t->rn_key differ
372 	 */
373 	{
374 		uint8_t *cp2 = t->rn_key + head_off;
375 		uint8_t cmp_res;
376 		uint8_t *cplim = v + vlen;
377 
378 		while (cp < cplim)
379 			if (*cp2++ != *cp++)
380 				goto found_differing_byte;
381 		/* handle adding 255.255.255.255 */
382 		if (!(t->rn_flags & RNF_ROOT) || *(cp2-1) == 0) {
383 			*dupentry = _B_TRUE;
384 			return (t);
385 		}
386 found_differing_byte:
387 		*dupentry = _B_FALSE;
388 		cmp_res = cp[-1] ^ cp2[-1];
389 		for (b = (cp - v) << 3; cmp_res != 0; b--)
390 			cmp_res >>= 1;
391 	}
392 	{
393 		struct radix_node *p, *x = top;
394 		cp = v;
395 		do {
396 			p = x;
397 			if (cp[x->rn_off] & x->rn_bmask)
398 				x = x->rn_r;
399 			else
400 				x = x->rn_l;
401 		} while (b > (unsigned)x->rn_b);
402 #ifdef RN_DEBUG
403 		if (rn_debug) {
404 			msglog("rn_insert: Going In:");
405 			traverse(p);
406 		}
407 #endif
408 		t = rn_newpair(v_arg, b, nodes);
409 		tt = t->rn_l;
410 		if (!(cp[p->rn_off] & p->rn_bmask))
411 			p->rn_l = t;
412 		else
413 			p->rn_r = t;
414 		x->rn_p = t; /* frees x, p as temp vars below */
415 		t->rn_p = p;
416 		if (!(cp[t->rn_off] & t->rn_bmask)) {
417 			t->rn_r = x;
418 		} else {
419 			t->rn_r = tt;
420 			t->rn_l = x;
421 		}
422 #ifdef RN_DEBUG
423 		if (rn_debug) {
424 			msglog("rn_insert: Coming Out:");
425 			traverse(p);
426 		}
427 #endif
428 	}
429 	return (tt);
430 }
431 
432 static struct radix_node *
rn_addmask(void * n_arg,uint_t search,uint_t skip)433 rn_addmask(void *n_arg, uint_t search, uint_t skip)
434 {
435 	uint8_t *netmask = n_arg;
436 	struct radix_node *x;
437 	uint8_t *cp, *cplim;
438 	int b = 0, mlen, j, m0;
439 	boolean_t maskduplicated;
440 	struct radix_node *saved_x;
441 	static int last_zeroed = 0;
442 
443 	mlen = sizeof (struct sockaddr);
444 	if (skip == 0)
445 		skip = 1;
446 	if (mlen <= skip)
447 		return (mask_rnhead->rnh_nodes);
448 	if (skip > 1)
449 		(void) memmove(addmask_key + 1, rn_ones + 1, skip - 1);
450 	if ((m0 = mlen) > skip)
451 		(void) memmove(addmask_key + skip, netmask + skip, mlen - skip);
452 	/*
453 	 * Trim trailing zeroes.
454 	 */
455 	for (cp = addmask_key + mlen; (cp > addmask_key) && cp[-1] == 0; )
456 		cp--;
457 	mlen = cp - addmask_key;
458 	if (mlen <= skip) {
459 		if (m0 >= last_zeroed)
460 			last_zeroed = mlen;
461 		return (mask_rnhead->rnh_nodes);
462 	}
463 	if (m0 < last_zeroed)
464 		(void) memset(addmask_key + m0, 0, last_zeroed - m0);
465 	*addmask_key = last_zeroed = mlen;
466 	x = rn_search(addmask_key, rn_masktop);
467 	if (memcmp(addmask_key, x->rn_key, mlen) != 0)
468 		x = NULL;
469 	if (x != NULL || search != 0)
470 		return (x);
471 	x = rtmalloc(max_keylen + 2*sizeof (*x), "rn_addmask");
472 	saved_x = x;
473 	(void) memset(x, 0, max_keylen + 2 * sizeof (*x));
474 	netmask = cp = (uint8_t *)(x + 2);
475 	(void) memmove(cp, addmask_key, mlen);
476 	x = rn_insert(cp, mask_rnhead, &maskduplicated, x);
477 	if (maskduplicated) {
478 #ifdef DEBUG
479 		logbad(1, "rn_addmask: mask impossibly already in tree");
480 #else
481 		msglog("rn_addmask: mask impossibly already in tree");
482 #endif
483 		free(saved_x);
484 		return (x);
485 	}
486 	/*
487 	 * Calculate index of mask, and check for normalcy.
488 	 */
489 	cplim = netmask + mlen;
490 	x->rn_flags |= RNF_NORMAL;
491 	for (cp = netmask + skip; (cp < cplim) && *cp == 0xff; )
492 		cp++;
493 	if (cp != cplim) {
494 		for (j = 0x80; (j & *cp) != 0; j >>= 1)
495 			b++;
496 		if (*cp != (0xFF & ~(0xFF >> b)) || cp != (cplim - 1))
497 			x->rn_flags &= ~RNF_NORMAL;
498 	}
499 	b += (cp - netmask) << 3;
500 	x->rn_b = -1 - b;
501 	return (x);
502 }
503 
504 static boolean_t /* Note: arbitrary ordering for non-contiguous masks */
rn_lexobetter(void * m_arg,void * n_arg)505 rn_lexobetter(void *m_arg, void *n_arg)
506 {
507 	uint8_t *mp = m_arg, *np = n_arg, *lim;
508 
509 	lim = mp + sizeof (struct sockaddr);
510 	while (mp < lim)
511 		if (*mp++ > *np++)
512 			return (_B_TRUE);
513 	return (_B_FALSE);
514 }
515 
516 static struct radix_mask *
rn_new_radix_mask(struct radix_node * tt,struct radix_mask * next)517 rn_new_radix_mask(struct radix_node *tt,
518     struct radix_mask *next)
519 {
520 	struct radix_mask *m;
521 
522 	MKGet(m);
523 	if (m == NULL) {
524 #ifdef DEBUG
525 		logbad(1, "Mask for route not entered");
526 #else
527 		msglog("Mask for route not entered");
528 #endif
529 		return (NULL);
530 	}
531 	(void) memset(m, 0, sizeof (*m));
532 	m->rm_b = tt->rn_b;
533 	m->rm_flags = tt->rn_flags;
534 	if (tt->rn_flags & RNF_NORMAL)
535 		m->rm_leaf = tt;
536 	else
537 		m->rm_mask = tt->rn_mask;
538 	m->rm_mklist = next;
539 	tt->rn_mklist = m;
540 	return (m);
541 }
542 
543 static struct radix_node *
rn_addroute(void * v_arg,void * n_arg,struct radix_node_head * head,struct radix_node treenodes[2])544 rn_addroute(void *v_arg, void *n_arg, struct radix_node_head *head,
545     struct radix_node treenodes[2])
546 {
547 	uint8_t *v = v_arg, *netmask = n_arg;
548 	struct radix_node *t, *x = 0, *tt;
549 	struct radix_node *saved_tt, *top = head->rnh_treetop;
550 	short b = 0, b_leaf = 0;
551 	boolean_t keyduplicated;
552 	uint8_t *mmask;
553 	struct radix_mask *m, **mp;
554 
555 	/*
556 	 * In dealing with non-contiguous masks, there may be
557 	 * many different routes which have the same mask.
558 	 * We will find it useful to have a unique pointer to
559 	 * the mask to speed avoiding duplicate references at
560 	 * nodes and possibly save time in calculating indices.
561 	 */
562 	if (netmask)  {
563 		if ((x = rn_addmask(netmask, 0, top->rn_off)) == NULL) {
564 			DBGMSG(("rn_addroute: addmask failed"));
565 			return (NULL);
566 		}
567 		b_leaf = x->rn_b;
568 		b = -1 - x->rn_b;
569 		netmask = x->rn_key;
570 	}
571 	/*
572 	 * Deal with duplicated keys: attach node to previous instance
573 	 */
574 	saved_tt = tt = rn_insert(v, head, &keyduplicated, treenodes);
575 	if (keyduplicated) {
576 		for (t = tt; tt; t = tt, tt = tt->rn_dupedkey) {
577 			if (tt->rn_mask == netmask) {
578 				DBGMSG(("rn_addroute: duplicated route and "
579 				    "mask"));
580 				return (NULL);
581 			}
582 			if (netmask == NULL ||
583 			    (tt->rn_mask &&
584 				((b_leaf < tt->rn_b) ||
585 				    rn_refines(netmask, tt->rn_mask) ||
586 				    rn_lexobetter(netmask, tt->rn_mask))))
587 				break;
588 		}
589 		/*
590 		 * If the mask is not duplicated, we wouldn't
591 		 * find it among possible duplicate key entries
592 		 * anyway, so the above test doesn't hurt.
593 		 *
594 		 * We sort the masks for a duplicated key the same way as
595 		 * in a masklist -- most specific to least specific.
596 		 * This may require the unfortunate nuisance of relocating
597 		 * the head of the list.
598 		 */
599 		if (tt == saved_tt) {
600 			struct	radix_node *xx = x;
601 			/* link in at head of list */
602 			(tt = treenodes)->rn_dupedkey = t;
603 			tt->rn_flags = t->rn_flags;
604 			tt->rn_p = x = t->rn_p;
605 			if (x->rn_l == t)
606 				x->rn_l = tt;
607 			else
608 				x->rn_r = tt;
609 			saved_tt = tt;
610 			x = xx;
611 		} else {
612 			(tt = treenodes)->rn_dupedkey = t->rn_dupedkey;
613 			t->rn_dupedkey = tt;
614 		}
615 #ifdef RN_DEBUG
616 		t = tt + 1;
617 		tt->rn_info = rn_nodenum++;
618 		t->rn_info = rn_nodenum++;
619 		tt->rn_twin = t;
620 		tt->rn_ybro = rn_clist;
621 		rn_clist = tt;
622 #endif
623 		tt->rn_key = v;
624 		tt->rn_b = -1;
625 		tt->rn_flags = RNF_ACTIVE;
626 	}
627 	/*
628 	 * Put mask in tree.
629 	 */
630 	if (netmask) {
631 		tt->rn_mask = netmask;
632 		tt->rn_b = x->rn_b;
633 		tt->rn_flags |= x->rn_flags & RNF_NORMAL;
634 	}
635 	t = saved_tt->rn_p;
636 	if (keyduplicated)
637 		goto key_already_in_tree;
638 	b_leaf = -1 - t->rn_b;
639 	if (t->rn_r == saved_tt)
640 		x = t->rn_l;
641 	else
642 		x = t->rn_r;
643 	/* Promote general routes from below */
644 	if (x->rn_b < 0) {
645 		for (mp = &t->rn_mklist; x; x = x->rn_dupedkey)
646 			if (x->rn_mask != NULL && (x->rn_b >= b_leaf) &&
647 			    x->rn_mklist == NULL) {
648 				if ((*mp = m = rn_new_radix_mask(x, 0)) != NULL)
649 					mp = &m->rm_mklist;
650 			}
651 	} else if (x->rn_mklist) {
652 		/*
653 		 * Skip over masks whose index is > that of new node
654 		 */
655 		for (mp = &x->rn_mklist; (m = *mp) != NULL; mp = &m->rm_mklist)
656 			if (m->rm_b >= b_leaf)
657 				break;
658 		t->rn_mklist = m;
659 		*mp = 0;
660 	}
661 key_already_in_tree:
662 	/* Add new route to highest possible ancestor's list */
663 	if ((netmask == NULL) || (b > t->rn_b)) {
664 		return (tt); /* can't lift at all */
665 	}
666 	b_leaf = tt->rn_b;
667 	do {
668 		x = t;
669 		t = t->rn_p;
670 	} while (b <= t->rn_b && x != top);
671 	/*
672 	 * Search through routes associated with node to
673 	 * insert new route according to index.
674 	 * Need same criteria as when sorting dupedkeys to avoid
675 	 * double loop on deletion.
676 	 */
677 	for (mp = &x->rn_mklist; (m = *mp) != NULL; mp = &m->rm_mklist) {
678 		if (m->rm_b < b_leaf)
679 			continue;
680 		if (m->rm_b > b_leaf)
681 			break;
682 		if (m->rm_flags & RNF_NORMAL) {
683 			mmask = m->rm_leaf->rn_mask;
684 			if (tt->rn_flags & RNF_NORMAL) {
685 #ifdef DEBUG
686 				logbad(1, "Non-unique normal route, mask "
687 				    "not entered");
688 #else
689 				msglog("Non-unique normal route, mask "
690 				    "not entered");
691 #endif
692 				return (tt);
693 			}
694 		} else
695 			mmask = m->rm_mask;
696 		if (mmask == netmask) {
697 			m->rm_refs++;
698 			tt->rn_mklist = m;
699 			return (tt);
700 		}
701 		if (rn_refines(netmask, mmask) || rn_lexobetter(netmask, mmask))
702 			break;
703 	}
704 	*mp = rn_new_radix_mask(tt, *mp);
705 	return (tt);
706 }
707 
708 static struct radix_node *
rn_delete(void * v_arg,void * netmask_arg,struct radix_node_head * head)709 rn_delete(void *v_arg, void *netmask_arg, struct radix_node_head *head)
710 {
711 	struct radix_node *t, *p, *x, *tt;
712 	struct radix_mask *m, *saved_m, **mp;
713 	struct radix_node *dupedkey, *saved_tt, *top;
714 	uint8_t *v, *netmask;
715 	int b;
716 	uint_t head_off, vlen;
717 
718 	v = v_arg;
719 	netmask = netmask_arg;
720 	x = head->rnh_treetop;
721 	tt = rn_search(v, x);
722 	head_off = x->rn_off;
723 	vlen = sizeof (struct sockaddr);
724 	saved_tt = tt;
725 	top = x;
726 	if (tt == NULL ||
727 	    memcmp(v + head_off, tt->rn_key + head_off, vlen - head_off) != 0) {
728 		DBGMSG(("rn_delete: unable to locate route to delete"));
729 		return (NULL);
730 	}
731 	/*
732 	 * Delete our route from mask lists.
733 	 */
734 	if (netmask) {
735 		if ((x = rn_addmask(netmask, 1, head_off)) == NULL) {
736 			DBGMSG(("rn_delete: cannot add mask"));
737 			return (NULL);
738 		}
739 		netmask = x->rn_key;
740 		while (tt->rn_mask != netmask)
741 			if ((tt = tt->rn_dupedkey) == NULL) {
742 				DBGMSG(("rn_delete: cannot locate mask"));
743 				return (NULL);
744 			}
745 	}
746 	if (tt->rn_mask == NULL || (saved_m = m = tt->rn_mklist) == NULL)
747 		goto annotation_removed;
748 	if (tt->rn_flags & RNF_NORMAL) {
749 		if (m->rm_leaf != tt || m->rm_refs > 0) {
750 #ifdef DEBUG
751 			logbad(1, "rn_delete: inconsistent annotation");
752 #else
753 			msglog("rn_delete: inconsistent annotation");
754 #endif
755 			return (NULL);  /* dangling ref could cause disaster */
756 		}
757 	} else {
758 		if (m->rm_mask != tt->rn_mask) {
759 #ifdef DEBUG
760 			logbad(1, "rn_delete: inconsistent annotation");
761 #else
762 			msglog("rn_delete: inconsistent annotation");
763 #endif
764 			goto annotation_removed;
765 		}
766 		if (--m->rm_refs >= 0)
767 			goto annotation_removed;
768 	}
769 	b = -1 - tt->rn_b;
770 	t = saved_tt->rn_p;
771 	if (b > t->rn_b)
772 		goto annotation_removed; /* Wasn't lifted at all */
773 	do {
774 		x = t;
775 		t = t->rn_p;
776 	} while (b <= t->rn_b && x != top);
777 	for (mp = &x->rn_mklist; (m = *mp) != NULL; mp = &m->rm_mklist)
778 		if (m == saved_m) {
779 			*mp = m->rm_mklist;
780 			MKFree(m);
781 			break;
782 		}
783 	if (m == NULL) {
784 #ifdef DEBUG
785 		logbad(1, "rn_delete: couldn't find our annotation");
786 #else
787 		msglog("rn_delete: couldn't find our annotation");
788 #endif
789 		if (tt->rn_flags & RNF_NORMAL)
790 			return (NULL); /* Dangling ref to us */
791 	}
792 annotation_removed:
793 	/*
794 	 * Eliminate us from tree
795 	 */
796 	if (tt->rn_flags & RNF_ROOT) {
797 		DBGMSG(("rn_delete: cannot delete root"));
798 		return (NULL);
799 	}
800 #ifdef RN_DEBUG
801 	/* Get us out of the creation list */
802 	for (t = rn_clist; t && t->rn_ybro != tt; t = t->rn_ybro) {}
803 	if (t != NULL)
804 		t->rn_ybro = tt->rn_ybro;
805 #endif
806 	t = tt->rn_p;
807 	if ((dupedkey = saved_tt->rn_dupedkey) != NULL) {
808 		if (tt == saved_tt) {
809 			x = dupedkey;
810 			x->rn_p = t;
811 			if (t->rn_l == tt)
812 				t->rn_l = x;
813 			else
814 				t->rn_r = x;
815 		} else {
816 			for (x = p = saved_tt; p && p->rn_dupedkey != tt; )
817 				p = p->rn_dupedkey;
818 			if (p != NULL) {
819 				p->rn_dupedkey = tt->rn_dupedkey;
820 			} else {
821 #ifdef DEBUG
822 				logbad(1, "rn_delete: couldn't find us");
823 #else
824 				msglog("rn_delete: couldn't find us");
825 #endif
826 			}
827 		}
828 		t = tt + 1;
829 		if (t->rn_flags & RNF_ACTIVE) {
830 #ifndef RN_DEBUG
831 			*++x = *t;
832 			p = t->rn_p;
833 #else
834 			b = t->rn_info;
835 			*++x = *t;
836 			t->rn_info = b;
837 			p = t->rn_p;
838 #endif
839 			if (p->rn_l == t)
840 				p->rn_l = x;
841 			else
842 				p->rn_r = x;
843 			x->rn_l->rn_p = x;
844 			x->rn_r->rn_p = x;
845 		}
846 		goto out;
847 	}
848 	if (t->rn_l == tt)
849 		x = t->rn_r;
850 	else
851 		x = t->rn_l;
852 	p = t->rn_p;
853 	if (p->rn_r == t)
854 		p->rn_r = x;
855 	else
856 		p->rn_l = x;
857 	x->rn_p = p;
858 	/*
859 	 * Demote routes attached to us.
860 	 */
861 	if (t->rn_mklist) {
862 		if (x->rn_b >= 0) {
863 			for (mp = &x->rn_mklist; (m = *mp) != NULL; )
864 				mp = &m->rm_mklist;
865 			*mp = t->rn_mklist;
866 		} else {
867 			/*
868 			 * If there are any key,mask pairs in a sibling
869 			 * duped-key chain, some subset will appear sorted
870 			 * in the same order attached to our mklist
871 			 */
872 			for (m = t->rn_mklist; m && x; x = x->rn_dupedkey)
873 				if (m == x->rn_mklist) {
874 					struct radix_mask *mm = m->rm_mklist;
875 					x->rn_mklist = 0;
876 					if (--(m->rm_refs) < 0)
877 						MKFree(m);
878 					m = mm;
879 				}
880 			if (m != NULL) {
881 #ifdef DEBUG
882 				logbad(1, "rn_delete: Orphaned Mask %p at %p\n",
883 				    m, x);
884 #else
885 				msglog("rn_delete: Orphaned Mask %p at %p\n", m,
886 				    x);
887 #endif
888 			}
889 		}
890 	}
891 	/*
892 	 * We may be holding an active internal node in the tree.
893 	 */
894 	x = tt + 1;
895 	if (t != x) {
896 #ifndef RN_DEBUG
897 		*t = *x;
898 #else
899 		b = t->rn_info;
900 		*t = *x;
901 		t->rn_info = b;
902 #endif
903 		t->rn_l->rn_p = t;
904 		t->rn_r->rn_p = t;
905 		p = x->rn_p;
906 		if (p->rn_l == x)
907 			p->rn_l = t;
908 		else
909 			p->rn_r = t;
910 	}
911 out:
912 	tt->rn_flags &= ~RNF_ACTIVE;
913 	tt[1].rn_flags &= ~RNF_ACTIVE;
914 	return (tt);
915 }
916 
917 int
rn_walktree(struct radix_node_head * h,int (* f)(struct radix_node *,void *),void * w)918 rn_walktree(struct radix_node_head *h,
919     int (*f)(struct radix_node *, void *),
920     void *w)
921 {
922 	int error;
923 	struct radix_node *base, *next;
924 	struct radix_node *rn = h->rnh_treetop;
925 	/*
926 	 * This gets complicated because we may delete the node
927 	 * while applying the function f to it, so we need to calculate
928 	 * the successor node in advance.
929 	 */
930 	/* First time through node, go left */
931 	while (rn->rn_b >= 0)
932 		rn = rn->rn_l;
933 	do {
934 		base = rn;
935 		/* If at right child go back up, otherwise, go right */
936 		while (rn->rn_p->rn_r == rn && !(rn->rn_flags & RNF_ROOT))
937 			rn = rn->rn_p;
938 		/* Find the next *leaf* since next node might vanish, too */
939 		for (rn = rn->rn_p->rn_r; rn->rn_b >= 0; )
940 			rn = rn->rn_l;
941 		next = rn;
942 		/* Process leaves */
943 		while ((rn = base) != NULL) {
944 			base = rn->rn_dupedkey;
945 			if (!(rn->rn_flags & RNF_ROOT) && (error = (*f)(rn, w)))
946 				return (error);
947 		}
948 		rn = next;
949 	} while (!(rn->rn_flags & RNF_ROOT));
950 	return (0);
951 }
952 
953 int
rn_inithead(void ** head,uint_t off)954 rn_inithead(void **head, uint_t off)
955 {
956 	struct radix_node_head *rnh;
957 	struct radix_node *t, *tt, *ttt;
958 	if (*head)
959 		return (1);
960 	rnh = rtmalloc(sizeof (*rnh), "rn_inithead");
961 	(void) memset(rnh, 0, sizeof (*rnh));
962 	*head = rnh;
963 	t = rn_newpair(rn_zeros, off, rnh->rnh_nodes);
964 	ttt = rnh->rnh_nodes + 2;
965 	t->rn_r = ttt;
966 	t->rn_p = t;
967 	tt = t->rn_l;
968 	tt->rn_flags = t->rn_flags = RNF_ROOT | RNF_ACTIVE;
969 	tt->rn_b = -1 - off;
970 	*ttt = *tt;
971 	ttt->rn_key = rn_ones;
972 	rnh->rnh_addaddr = rn_addroute;
973 	rnh->rnh_deladdr = rn_delete;
974 	rnh->rnh_matchaddr = rn_match;
975 	rnh->rnh_lookup = rn_lookup;
976 	rnh->rnh_walktree = rn_walktree;
977 	rnh->rnh_treetop = t;
978 	return (1);
979 }
980 
981 void
rn_init(void)982 rn_init(void)
983 {
984 	uint8_t *cp, *cplim;
985 
986 	if (max_keylen == 0) {
987 		logbad(1, "radix functions require max_keylen be set");
988 		return;
989 	}
990 	rn_zeros = rtmalloc(3 * max_keylen, "rn_init");
991 	(void) memset(rn_zeros, 0, 3 * max_keylen);
992 	rn_ones = cp = rn_zeros + max_keylen;
993 	addmask_key = cplim = rn_ones + max_keylen;
994 	while (cp < cplim)
995 		*cp++ = 0xFF;
996 	if (rn_inithead((void **)&mask_rnhead, 0) == 0) {
997 		logbad(0, "rn_init: could not initialize radix tree");
998 	}
999 }
1000