1 /* $OpenBSD: radix.c,v 1.21 2006/06/18 11:47:45 pascoe Exp $ */
2 /* $NetBSD: radix.c,v 1.20 2003/08/07 16:32:56 agc Exp $ */
3
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. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 *
32 * @(#)radix.c 8.6 (Berkeley) 10/17/95
33 */
34
35 /*
36 * Routines to build and maintain radix trees for routing lookups.
37 */
38
39 #ifndef _NET_RADIX_H_
40 #include <sys/param.h>
41 #ifdef _KERNEL
42 #include <sys/systm.h>
43 #include <sys/malloc.h>
44 #define M_DONTWAIT M_NOWAIT
45 #include <sys/domain.h>
46 #else
47 #include <stdlib.h>
48 #endif
49 #include <sys/syslog.h>
50 #include <net/radix.h>
51 #endif
52
53 #ifndef SMALL_KERNEL
54 #include <net/radix_mpath.h>
55 #endif
56
57 int max_keylen;
58 struct radix_mask *rn_mkfreelist;
59 struct radix_node_head *mask_rnhead;
60 static char *addmask_key;
61 static char normal_chars[] = {0, 0x80, 0xc0, 0xe0, 0xf0, 0xf8, 0xfc, 0xfe, -1};
62 static char *rn_zeros, *rn_ones;
63
64 #define rn_masktop (mask_rnhead->rnh_treetop)
65 #undef Bcmp
66 #define Bcmp(a, b, l) (l == 0 ? 0 : bcmp((caddr_t)(a), (caddr_t)(b), (u_long)l))
67
68 static int rn_satisfies_leaf(char *, struct radix_node *, int);
69 static int rn_lexobetter(void *, void *);
70 static struct radix_mask *rn_new_radix_mask(struct radix_node *,
71 struct radix_mask *);
72
73 /*
74 * The data structure for the keys is a radix tree with one way
75 * branching removed. The index rn_b at an internal node n represents a bit
76 * position to be tested. The tree is arranged so that all descendants
77 * of a node n have keys whose bits all agree up to position rn_b - 1.
78 * (We say the index of n is rn_b.)
79 *
80 * There is at least one descendant which has a one bit at position rn_b,
81 * and at least one with a zero there.
82 *
83 * A route is determined by a pair of key and mask. We require that the
84 * bit-wise logical and of the key and mask to be the key.
85 * We define the index of a route to associated with the mask to be
86 * the first bit number in the mask where 0 occurs (with bit number 0
87 * representing the highest order bit).
88 *
89 * We say a mask is normal if every bit is 0, past the index of the mask.
90 * If a node n has a descendant (k, m) with index(m) == index(n) == rn_b,
91 * and m is a normal mask, then the route applies to every descendant of n.
92 * If the index(m) < rn_b, this implies the trailing last few bits of k
93 * before bit b are all 0, (and hence consequently true of every descendant
94 * of n), so the route applies to all descendants of the node as well.
95 *
96 * Similar logic shows that a non-normal mask m such that
97 * index(m) <= index(n) could potentially apply to many children of n.
98 * Thus, for each non-host route, we attach its mask to a list at an internal
99 * node as high in the tree as we can go.
100 *
101 * The present version of the code makes use of normal routes in short-
102 * circuiting an explicit mask and compare operation when testing whether
103 * a key satisfies a normal route, and also in remembering the unique leaf
104 * that governs a subtree.
105 */
106
107 struct radix_node *
108 rn_search(void *v_arg, struct radix_node *head)
109 {
110 struct radix_node *x;
111 caddr_t v;
112
113 for (x = head, v = v_arg; x->rn_b >= 0;) {
114 if (x->rn_bmask & v[x->rn_off])
115 x = x->rn_r;
116 else
117 x = x->rn_l;
118 }
119 return (x);
120 }
121
122 struct radix_node *
123 rn_search_m(void *v_arg, struct radix_node *head, void *m_arg)
124 {
125 struct radix_node *x;
126 caddr_t v = v_arg, m = m_arg;
127
128 for (x = head; x->rn_b >= 0;) {
129 if ((x->rn_bmask & m[x->rn_off]) &&
130 (x->rn_bmask & v[x->rn_off]))
131 x = x->rn_r;
132 else
133 x = x->rn_l;
134 }
135 return x;
136 }
137
138 int
139 rn_refines(void *m_arg, void *n_arg)
140 {
141 caddr_t m = m_arg, n = n_arg;
142 caddr_t lim, lim2 = lim = n + *(u_char *)n;
143 int longer = (*(u_char *)n++) - (int)(*(u_char *)m++);
144 int masks_are_equal = 1;
145
146 if (longer > 0)
147 lim -= longer;
148 while (n < lim) {
149 if (*n & ~(*m))
150 return 0;
151 if (*n++ != *m++)
152 masks_are_equal = 0;
153 }
154 while (n < lim2)
155 if (*n++)
156 return 0;
157 if (masks_are_equal && (longer < 0))
158 for (lim2 = m - longer; m < lim2; )
159 if (*m++)
160 return 1;
161 return (!masks_are_equal);
162 }
163
164 struct radix_node *
165 rn_lookup(void *v_arg, void *m_arg, struct radix_node_head *head)
166 {
167 struct radix_node *x;
168 caddr_t netmask = 0;
169
170 if (m_arg) {
171 if ((x = rn_addmask(m_arg, 1, head->rnh_treetop->rn_off)) == 0)
172 return (0);
173 netmask = x->rn_key;
174 }
175 x = rn_match(v_arg, head);
176 if (x && netmask) {
177 while (x && x->rn_mask != netmask)
178 x = x->rn_dupedkey;
179 }
180 return x;
181 }
182
183 static int
184 rn_satisfies_leaf(char *trial, struct radix_node *leaf, int skip)
185 {
186 char *cp = trial, *cp2 = leaf->rn_key, *cp3 = leaf->rn_mask;
187 char *cplim;
188 int length = min(*(u_char *)cp, *(u_char *)cp2);
189
190 if (cp3 == 0)
191 cp3 = rn_ones;
192 else
193 length = min(length, *(u_char *)cp3);
194 cplim = cp + length; cp3 += skip; cp2 += skip;
195 for (cp += skip; cp < cplim; cp++, cp2++, cp3++)
196 if ((*cp ^ *cp2) & *cp3)
197 return 0;
198 return 1;
199 }
200
201 struct radix_node *
202 rn_match(void *v_arg, struct radix_node_head *head)
203 {
204 caddr_t v = v_arg;
205 struct radix_node *t = head->rnh_treetop, *x;
206 caddr_t cp = v, cp2;
207 caddr_t cplim;
208 struct radix_node *saved_t, *top = t;
209 int off = t->rn_off, vlen = *(u_char *)cp, matched_off;
210 int test, b, rn_b;
211
212 /*
213 * Open code rn_search(v, top) to avoid overhead of extra
214 * subroutine call.
215 */
216 for (; t->rn_b >= 0; ) {
217 if (t->rn_bmask & cp[t->rn_off])
218 t = t->rn_r;
219 else
220 t = t->rn_l;
221 }
222 /*
223 * See if we match exactly as a host destination
224 * or at least learn how many bits match, for normal mask finesse.
225 *
226 * It doesn't hurt us to limit how many bytes to check
227 * to the length of the mask, since if it matches we had a genuine
228 * match and the leaf we have is the most specific one anyway;
229 * if it didn't match with a shorter length it would fail
230 * with a long one. This wins big for class B&C netmasks which
231 * are probably the most common case...
232 */
233 if (t->rn_mask)
234 vlen = *(u_char *)t->rn_mask;
235 cp += off; cp2 = t->rn_key + off; cplim = v + vlen;
236 for (; cp < cplim; cp++, cp2++)
237 if (*cp != *cp2)
238 goto on1;
239 /*
240 * This extra grot is in case we are explicitly asked
241 * to look up the default. Ugh!
242 */
243 if ((t->rn_flags & RNF_ROOT) && t->rn_dupedkey)
244 t = t->rn_dupedkey;
245 return t;
246 on1:
247 test = (*cp ^ *cp2) & 0xff; /* find first bit that differs */
248 for (b = 7; (test >>= 1) > 0;)
249 b--;
250 matched_off = cp - v;
251 b += matched_off << 3;
252 rn_b = -1 - b;
253 /*
254 * If there is a host route in a duped-key chain, it will be first.
255 */
256 if ((saved_t = t)->rn_mask == 0)
257 t = t->rn_dupedkey;
258 for (; t; t = t->rn_dupedkey)
259 /*
260 * Even if we don't match exactly as a host,
261 * we may match if the leaf we wound up at is
262 * a route to a net.
263 */
264 if (t->rn_flags & RNF_NORMAL) {
265 if (rn_b <= t->rn_b)
266 return t;
267 } else if (rn_satisfies_leaf(v, t, matched_off))
268 return t;
269 t = saved_t;
270 /* start searching up the tree */
271 do {
272 struct radix_mask *m;
273 t = t->rn_p;
274 m = t->rn_mklist;
275 if (m) {
276 /*
277 * If non-contiguous masks ever become important
278 * we can restore the masking and open coding of
279 * the search and satisfaction test and put the
280 * calculation of "off" back before the "do".
281 */
282 do {
283 if (m->rm_flags & RNF_NORMAL) {
284 if (rn_b <= m->rm_b)
285 return (m->rm_leaf);
286 } else {
287 off = min(t->rn_off, matched_off);
288 x = rn_search_m(v, t, m->rm_mask);
289 while (x && x->rn_mask != m->rm_mask)
290 x = x->rn_dupedkey;
291 if (x && rn_satisfies_leaf(v, x, off))
292 return x;
293 }
294 m = m->rm_mklist;
295 } while (m);
296 }
297 } while (t != top);
298 return 0;
299 }
300
301 #ifdef RN_DEBUG
302 int rn_nodenum;
303 struct radix_node *rn_clist;
304 int rn_saveinfo;
305 int rn_debug = 1;
306 #endif
307
308 struct radix_node *
309 rn_newpair(void *v, int b, struct radix_node nodes[2])
310 {
311 struct radix_node *tt = nodes, *t = tt + 1;
312 t->rn_b = b;
313 t->rn_bmask = 0x80 >> (b & 7);
314 t->rn_l = tt;
315 t->rn_off = b >> 3;
316 tt->rn_b = -1;
317 tt->rn_key = (caddr_t)v;
318 tt->rn_p = t;
319 tt->rn_flags = t->rn_flags = RNF_ACTIVE;
320 #ifdef RN_DEBUG
321 tt->rn_info = rn_nodenum++;
322 t->rn_info = rn_nodenum++;
323 tt->rn_twin = t;
324 tt->rn_ybro = rn_clist;
325 rn_clist = tt;
326 #endif
327 return t;
328 }
329
330 struct radix_node *
331 rn_insert(void *v_arg, struct radix_node_head *head,
332 int *dupentry, struct radix_node nodes[2])
333 {
334 caddr_t v = v_arg;
335 struct radix_node *top = head->rnh_treetop;
336 int head_off = top->rn_off, vlen = (int)*((u_char *)v);
337 struct radix_node *t = rn_search(v_arg, top);
338 caddr_t cp = v + head_off;
339 int b;
340 struct radix_node *tt;
341 /*
342 * Find first bit at which v and t->rn_key differ
343 */
344 {
345 caddr_t cp2 = t->rn_key + head_off;
346 int cmp_res;
347 caddr_t cplim = v + vlen;
348
349 while (cp < cplim)
350 if (*cp2++ != *cp++)
351 goto on1;
352 *dupentry = 1;
353 return t;
354 on1:
355 *dupentry = 0;
356 cmp_res = (cp[-1] ^ cp2[-1]) & 0xff;
357 for (b = (cp - v) << 3; cmp_res; b--)
358 cmp_res >>= 1;
359 }
360 {
361 struct radix_node *p, *x = top;
362 cp = v;
363 do {
364 p = x;
365 if (cp[x->rn_off] & x->rn_bmask)
366 x = x->rn_r;
367 else
368 x = x->rn_l;
369 } while (b > (unsigned) x->rn_b); /* x->rn_b < b && x->rn_b >= 0 */
370 #ifdef RN_DEBUG
371 if (rn_debug)
372 log(LOG_DEBUG, "rn_insert: Going In:\n"), traverse(p);
373 #endif
374 t = rn_newpair(v_arg, b, nodes);
375 tt = t->rn_l;
376 if ((cp[p->rn_off] & p->rn_bmask) == 0)
377 p->rn_l = t;
378 else
379 p->rn_r = t;
380 x->rn_p = t;
381 t->rn_p = p; /* frees x, p as temp vars below */
382 if ((cp[t->rn_off] & t->rn_bmask) == 0) {
383 t->rn_r = x;
384 } else {
385 t->rn_r = tt;
386 t->rn_l = x;
387 }
388 #ifdef RN_DEBUG
389 if (rn_debug)
390 log(LOG_DEBUG, "rn_insert: Coming Out:\n"), traverse(p);
391 #endif
392 }
393 return (tt);
394 }
395
396 struct radix_node *
397 rn_addmask(void *n_arg, int search, int skip)
398 {
399 caddr_t netmask = (caddr_t)n_arg;
400 struct radix_node *x;
401 caddr_t cp, cplim;
402 int b = 0, mlen, j;
403 int maskduplicated, m0, isnormal;
404 struct radix_node *saved_x;
405 static int last_zeroed = 0;
406
407 if ((mlen = *(u_char *)netmask) > max_keylen)
408 mlen = max_keylen;
409 if (skip == 0)
410 skip = 1;
411 if (mlen <= skip)
412 return (mask_rnhead->rnh_nodes);
413 if (skip > 1)
414 Bcopy(rn_ones + 1, addmask_key + 1, skip - 1);
415 if ((m0 = mlen) > skip)
416 Bcopy(netmask + skip, addmask_key + skip, mlen - skip);
417 /*
418 * Trim trailing zeroes.
419 */
420 for (cp = addmask_key + mlen; (cp > addmask_key) && cp[-1] == 0;)
421 cp--;
422 mlen = cp - addmask_key;
423 if (mlen <= skip) {
424 if (m0 >= last_zeroed)
425 last_zeroed = mlen;
426 return (mask_rnhead->rnh_nodes);
427 }
428 if (m0 < last_zeroed)
429 Bzero(addmask_key + m0, last_zeroed - m0);
430 *addmask_key = last_zeroed = mlen;
431 x = rn_search(addmask_key, rn_masktop);
432 if (Bcmp(addmask_key, x->rn_key, mlen) != 0)
433 x = 0;
434 if (x || search)
435 return (x);
436 R_Malloc(x, struct radix_node *, max_keylen + 2 * sizeof (*x));
437 if ((saved_x = x) == 0)
438 return (0);
439 Bzero(x, max_keylen + 2 * sizeof (*x));
440 netmask = cp = (caddr_t)(x + 2);
441 Bcopy(addmask_key, cp, mlen);
442 x = rn_insert(cp, mask_rnhead, &maskduplicated, x);
443 if (maskduplicated) {
444 log(LOG_ERR, "rn_addmask: mask impossibly already in tree\n");
445 Free(saved_x);
446 return (x);
447 }
448 /*
449 * Calculate index of mask, and check for normalcy.
450 */
451 cplim = netmask + mlen;
452 isnormal = 1;
453 for (cp = netmask + skip; (cp < cplim) && *(u_char *)cp == 0xff;)
454 cp++;
455 if (cp != cplim) {
456 for (j = 0x80; (j & *cp) != 0; j >>= 1)
457 b++;
458 if (*cp != normal_chars[b] || cp != (cplim - 1))
459 isnormal = 0;
460 }
461 b += (cp - netmask) << 3;
462 x->rn_b = -1 - b;
463 if (isnormal)
464 x->rn_flags |= RNF_NORMAL;
465 return (x);
466 }
467
468 static int /* XXX: arbitrary ordering for non-contiguous masks */
469 rn_lexobetter(void *m_arg, void *n_arg)
470 {
471 u_char *mp = m_arg, *np = n_arg, *lim;
472
473 if (*mp > *np)
474 return 1; /* not really, but need to check longer one first */
475 if (*mp == *np)
476 for (lim = mp + *mp; mp < lim;)
477 if (*mp++ > *np++)
478 return 1;
479 return 0;
480 }
481
482 static struct radix_mask *
483 rn_new_radix_mask(struct radix_node *tt, struct radix_mask *next)
484 {
485 struct radix_mask *m;
486
487 MKGet(m);
488 if (m == 0) {
489 log(LOG_ERR, "Mask for route not entered\n");
490 return (0);
491 }
492 Bzero(m, sizeof *m);
493 m->rm_b = tt->rn_b;
494 m->rm_flags = tt->rn_flags;
495 if (tt->rn_flags & RNF_NORMAL)
496 m->rm_leaf = tt;
497 else
498 m->rm_mask = tt->rn_mask;
499 m->rm_mklist = next;
500 tt->rn_mklist = m;
501 return m;
502 }
503
504 struct radix_node *
505 rn_addroute(void *v_arg, void *n_arg, struct radix_node_head *head,
506 struct radix_node treenodes[2])
507 {
508 caddr_t v = (caddr_t)v_arg, netmask = (caddr_t)n_arg;
509 struct radix_node *t, *x = NULL, *tt;
510 struct radix_node *saved_tt, *top = head->rnh_treetop;
511 short b = 0, b_leaf = 0;
512 int keyduplicated;
513 caddr_t mmask;
514 struct radix_mask *m, **mp;
515
516 /*
517 * In dealing with non-contiguous masks, there may be
518 * many different routes which have the same mask.
519 * We will find it useful to have a unique pointer to
520 * the mask to speed avoiding duplicate references at
521 * nodes and possibly save time in calculating indices.
522 */
523 if (netmask) {
524 if ((x = rn_addmask(netmask, 0, top->rn_off)) == 0)
525 return (0);
526 b_leaf = x->rn_b;
527 b = -1 - x->rn_b;
528 netmask = x->rn_key;
529 }
530 /*
531 * Deal with duplicated keys: attach node to previous instance
532 */
533 saved_tt = tt = rn_insert(v, head, &keyduplicated, treenodes);
534 if (keyduplicated) {
535 for (t = tt; tt; t = tt, tt = tt->rn_dupedkey) {
536 #ifndef SMALL_KERNEL
537 /* permit multipath, if enabled for the family */
538 if (rn_mpath_capable(head) && netmask == tt->rn_mask) {
539 /*
540 * Try to insert the new node in the middle
541 * of the list of any preexisting multipaths,
542 * to reduce the number of path disruptions
543 * that occur as a result of an insertion,
544 * per RFC2992.
545 */
546 int mid = rn_mpath_count(tt) / 2;
547 do {
548 t = tt;
549 tt = tt->rn_dupedkey;
550 } while (tt && t->rn_mask == tt->rn_mask
551 && --mid > 0);
552 break;
553 }
554 #endif
555 if (tt->rn_mask == netmask)
556 return (0);
557 if (netmask == 0 ||
558 (tt->rn_mask &&
559 ((b_leaf < tt->rn_b) || /* index(netmask) > node */
560 rn_refines(netmask, tt->rn_mask) ||
561 rn_lexobetter(netmask, tt->rn_mask))))
562 break;
563 }
564 /*
565 * If the mask is not duplicated, we wouldn't
566 * find it among possible duplicate key entries
567 * anyway, so the above test doesn't hurt.
568 *
569 * We sort the masks for a duplicated key the same way as
570 * in a masklist -- most specific to least specific.
571 * This may require the unfortunate nuisance of relocating
572 * the head of the list.
573 *
574 * We also reverse, or doubly link the list through the
575 * parent pointer.
576 */
577 if (tt == saved_tt) {
578 struct radix_node *xx = x;
579 /* link in at head of list */
580 (tt = treenodes)->rn_dupedkey = t;
581 tt->rn_flags = t->rn_flags;
582 tt->rn_p = x = t->rn_p;
583 t->rn_p = tt;
584 if (x->rn_l == t)
585 x->rn_l = tt;
586 else
587 x->rn_r = tt;
588 saved_tt = tt;
589 x = xx;
590 } else {
591 (tt = treenodes)->rn_dupedkey = t->rn_dupedkey;
592 t->rn_dupedkey = tt;
593 tt->rn_p = t;
594 if (tt->rn_dupedkey)
595 tt->rn_dupedkey->rn_p = tt;
596 }
597 #ifdef RN_DEBUG
598 t=tt+1;
599 tt->rn_info = rn_nodenum++;
600 t->rn_info = rn_nodenum++;
601 tt->rn_twin = t;
602 tt->rn_ybro = rn_clist;
603 rn_clist = tt;
604 #endif
605 tt->rn_key = (caddr_t) v;
606 tt->rn_b = -1;
607 tt->rn_flags = RNF_ACTIVE;
608 }
609 /*
610 * Put mask in tree.
611 */
612 if (netmask) {
613 tt->rn_mask = netmask;
614 tt->rn_b = x->rn_b;
615 tt->rn_flags |= x->rn_flags & RNF_NORMAL;
616 }
617 t = saved_tt->rn_p;
618 if (keyduplicated)
619 goto on2;
620 b_leaf = -1 - t->rn_b;
621 if (t->rn_r == saved_tt)
622 x = t->rn_l;
623 else
624 x = t->rn_r;
625 /* Promote general routes from below */
626 if (x->rn_b < 0) {
627 for (mp = &t->rn_mklist; x; x = x->rn_dupedkey)
628 if (x->rn_mask && (x->rn_b >= b_leaf) && x->rn_mklist == 0) {
629 *mp = m = rn_new_radix_mask(x, 0);
630 if (m)
631 mp = &m->rm_mklist;
632 }
633 } else if (x->rn_mklist) {
634 /*
635 * Skip over masks whose index is > that of new node
636 */
637 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist)
638 if (m->rm_b >= b_leaf)
639 break;
640 t->rn_mklist = m;
641 *mp = 0;
642 }
643 on2:
644 /* Add new route to highest possible ancestor's list */
645 if ((netmask == 0) || (b > t->rn_b ))
646 return tt; /* can't lift at all */
647 b_leaf = tt->rn_b;
648 do {
649 x = t;
650 t = t->rn_p;
651 } while (b <= t->rn_b && x != top);
652 /*
653 * Search through routes associated with node to
654 * insert new route according to index.
655 * Need same criteria as when sorting dupedkeys to avoid
656 * double loop on deletion.
657 */
658 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist) {
659 if (m->rm_b < b_leaf)
660 continue;
661 if (m->rm_b > b_leaf)
662 break;
663 if (m->rm_flags & RNF_NORMAL) {
664 mmask = m->rm_leaf->rn_mask;
665 if (tt->rn_flags & RNF_NORMAL) {
666 log(LOG_ERR, "Non-unique normal route,"
667 " mask not entered\n");
668 return tt;
669 }
670 } else
671 mmask = m->rm_mask;
672 if (mmask == netmask) {
673 m->rm_refs++;
674 tt->rn_mklist = m;
675 return tt;
676 }
677 if (rn_refines(netmask, mmask) || rn_lexobetter(netmask, mmask))
678 break;
679 }
680 *mp = rn_new_radix_mask(tt, *mp);
681 return tt;
682 }
683
684 struct radix_node *
685 rn_delete(void *v_arg, void *netmask_arg, struct radix_node_head *head,
686 struct radix_node *rn)
687 {
688 struct radix_node *t, *p, *x, *tt;
689 struct radix_mask *m, *saved_m, **mp;
690 struct radix_node *dupedkey, *saved_tt, *top;
691 caddr_t v, netmask;
692 int b, head_off, vlen;
693 #ifndef SMALL_KERNEL
694 int mpath_enable = 0;
695 #endif
696
697 v = v_arg;
698 netmask = netmask_arg;
699 x = head->rnh_treetop;
700 #ifndef SMALL_KERNEL
701 if (rn) {
702 tt = rn;
703 /*
704 * Is this route(rn) a rn->dupedkey chain?
705 */
706 if (rn_mpath_next(tt->rn_p))
707 mpath_enable = 1;
708 else
709 tt = rn_search(v, x);
710 } else
711 tt = rn_search(v, x);
712 #else
713 tt = rn_search(v, x);
714 #endif
715 head_off = x->rn_off;
716 vlen = *(u_char *)v;
717 saved_tt = tt;
718 top = x;
719 if (tt == 0 ||
720 Bcmp(v + head_off, tt->rn_key + head_off, vlen - head_off))
721 return (0);
722 /*
723 * Delete our route from mask lists.
724 */
725 if (netmask) {
726 if ((x = rn_addmask(netmask, 1, head_off)) == 0)
727 return (0);
728 netmask = x->rn_key;
729 while (tt->rn_mask != netmask)
730 if ((tt = tt->rn_dupedkey) == 0)
731 return (0);
732 }
733 if (tt->rn_mask == 0 || (saved_m = m = tt->rn_mklist) == 0)
734 goto on1;
735 if (tt->rn_flags & RNF_NORMAL) {
736 if (m->rm_leaf != tt || m->rm_refs > 0) {
737 log(LOG_ERR, "rn_delete: inconsistent annotation\n");
738 return 0; /* dangling ref could cause disaster */
739 }
740 } else {
741 if (m->rm_mask != tt->rn_mask) {
742 log(LOG_ERR, "rn_delete: inconsistent annotation\n");
743 goto on1;
744 }
745 if (--m->rm_refs >= 0)
746 goto on1;
747 }
748 b = -1 - tt->rn_b;
749 t = saved_tt->rn_p;
750 if (b > t->rn_b)
751 goto on1; /* Wasn't lifted at all */
752 do {
753 x = t;
754 t = t->rn_p;
755 } while (b <= t->rn_b && x != top);
756 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist)
757 if (m == saved_m) {
758 *mp = m->rm_mklist;
759 MKFree(m);
760 break;
761 }
762 if (m == 0) {
763 log(LOG_ERR, "rn_delete: couldn't find our annotation\n");
764 if (tt->rn_flags & RNF_NORMAL)
765 return (0); /* Dangling ref to us */
766 }
767 on1:
768 /*
769 * Eliminate us from tree
770 */
771 if (tt->rn_flags & RNF_ROOT)
772 return (0);
773 #ifdef RN_DEBUG
774 /* Get us out of the creation list */
775 for (t = rn_clist; t && t->rn_ybro != tt; t = t->rn_ybro)
776 ;
777 if (t) t->rn_ybro = tt->rn_ybro;
778 #endif
779 t = tt->rn_p;
780 dupedkey = saved_tt->rn_dupedkey;
781 if (dupedkey) {
782 /*
783 * Here, tt is the deletion target, and
784 * saved_tt is the head of the dupedkey chain.
785 */
786 if (tt == saved_tt) {
787 x = dupedkey;
788 x->rn_p = t;
789 if (t->rn_l == tt)
790 t->rn_l = x;
791 else
792 t->rn_r = x;
793 } else {
794 /* find node in front of tt on the chain */
795 for (x = p = saved_tt; p && p->rn_dupedkey != tt;)
796 p = p->rn_dupedkey;
797 if (p) {
798 p->rn_dupedkey = tt->rn_dupedkey;
799 if (tt->rn_dupedkey)
800 tt->rn_dupedkey->rn_p = p;
801 } else log(LOG_ERR, "rn_delete: couldn't find us\n");
802 }
803 t = tt + 1;
804 if (t->rn_flags & RNF_ACTIVE) {
805 #ifndef RN_DEBUG
806 *++x = *t;
807 p = t->rn_p;
808 #else
809 b = t->rn_info;
810 *++x = *t;
811 t->rn_info = b;
812 p = t->rn_p;
813 #endif
814 if (p->rn_l == t)
815 p->rn_l = x;
816 else
817 p->rn_r = x;
818 x->rn_l->rn_p = x;
819 x->rn_r->rn_p = x;
820 }
821 goto out;
822 }
823 #ifndef SMALL_KERNEL
824 if (mpath_enable) {
825 /*
826 * my parent dupedkey is NULL
827 * end of mpath route.
828 */
829 t->rn_dupedkey = NULL;
830 goto out;
831 }
832 #endif
833 if (t->rn_l == tt)
834 x = t->rn_r;
835 else
836 x = t->rn_l;
837 p = t->rn_p;
838 if (p->rn_r == t)
839 p->rn_r = x;
840 else
841 p->rn_l = x;
842 x->rn_p = p;
843 /*
844 * Demote routes attached to us.
845 */
846 if (t->rn_mklist) {
847 if (x->rn_b >= 0) {
848 for (mp = &x->rn_mklist; (m = *mp);)
849 mp = &m->rm_mklist;
850 *mp = t->rn_mklist;
851 } else {
852 /* If there are any key,mask pairs in a sibling
853 duped-key chain, some subset will appear sorted
854 in the same order attached to our mklist */
855 for (m = t->rn_mklist; m && x; x = x->rn_dupedkey)
856 if (m == x->rn_mklist) {
857 struct radix_mask *mm = m->rm_mklist;
858 x->rn_mklist = 0;
859 if (--(m->rm_refs) < 0)
860 MKFree(m);
861 m = mm;
862 }
863 if (m)
864 log(LOG_ERR, "%s %p at %p\n",
865 "rn_delete: Orphaned Mask", m, x);
866 }
867 }
868 /*
869 * We may be holding an active internal node in the tree.
870 */
871 x = tt + 1;
872 if (t != x) {
873 #ifndef RN_DEBUG
874 *t = *x;
875 #else
876 b = t->rn_info;
877 *t = *x;
878 t->rn_info = b;
879 #endif
880 t->rn_l->rn_p = t;
881 t->rn_r->rn_p = t;
882 p = x->rn_p;
883 if (p->rn_l == x)
884 p->rn_l = t;
885 else
886 p->rn_r = t;
887 }
888 out:
889 tt->rn_flags &= ~RNF_ACTIVE;
890 tt[1].rn_flags &= ~RNF_ACTIVE;
891 return (tt);
892 }
893
894 int
895 rn_walktree(struct radix_node_head *h, int (*f)(struct radix_node *, void *),
896 void *w)
897 {
898 int error;
899 struct radix_node *base, *next;
900 struct radix_node *rn = h->rnh_treetop;
901 /*
902 * This gets complicated because we may delete the node
903 * while applying the function f to it, so we need to calculate
904 * the successor node in advance.
905 */
906 /* First time through node, go left */
907 while (rn->rn_b >= 0)
908 rn = rn->rn_l;
909 for (;;) {
910 base = rn;
911 /* If at right child go back up, otherwise, go right */
912 while (rn->rn_p->rn_r == rn && (rn->rn_flags & RNF_ROOT) == 0)
913 rn = rn->rn_p;
914 /* Find the next *leaf* since next node might vanish, too */
915 for (rn = rn->rn_p->rn_r; rn->rn_b >= 0;)
916 rn = rn->rn_l;
917 next = rn;
918 /* Process leaves */
919 while ((rn = base) != NULL) {
920 base = rn->rn_dupedkey;
921 if (!(rn->rn_flags & RNF_ROOT) && (error = (*f)(rn, w)))
922 return (error);
923 }
924 rn = next;
925 if (rn->rn_flags & RNF_ROOT)
926 return (0);
927 }
928 /* NOTREACHED */
929 }
930
931 int
932 rn_inithead(void **head, int off)
933 {
934 struct radix_node_head *rnh;
935
936 if (*head)
937 return (1);
938 R_Malloc(rnh, struct radix_node_head *, sizeof (*rnh));
939 if (rnh == 0)
940 return (0);
941 *head = rnh;
942 return rn_inithead0(rnh, off);
943 }
944
945 int
946 rn_inithead0(struct radix_node_head *rnh, int off)
947 {
948 struct radix_node *t, *tt, *ttt;
949
950 Bzero(rnh, sizeof (*rnh));
951 t = rn_newpair(rn_zeros, off, rnh->rnh_nodes);
952 ttt = rnh->rnh_nodes + 2;
953 t->rn_r = ttt;
954 t->rn_p = t;
955 tt = t->rn_l;
956 tt->rn_flags = t->rn_flags = RNF_ROOT | RNF_ACTIVE;
957 tt->rn_b = -1 - off;
958 *ttt = *tt;
959 ttt->rn_key = rn_ones;
960 rnh->rnh_addaddr = rn_addroute;
961 rnh->rnh_deladdr = rn_delete;
962 rnh->rnh_matchaddr = rn_match;
963 rnh->rnh_lookup = rn_lookup;
964 rnh->rnh_walktree = rn_walktree;
965 rnh->rnh_treetop = t;
966 return (1);
967 }
968
969 void
970 rn_init()
971 {
972 char *cp, *cplim;
973 #ifdef _KERNEL
974 struct domain *dom;
975
976 for (dom = domains; dom; dom = dom->dom_next)
977 if (dom->dom_maxrtkey > max_keylen)
978 max_keylen = dom->dom_maxrtkey;
979 #endif
980 if (max_keylen == 0) {
981 log(LOG_ERR,
982 "rn_init: radix functions require max_keylen be set\n");
983 return;
984 }
985 R_Malloc(rn_zeros, char *, 3 * max_keylen);
986 if (rn_zeros == NULL)
987 panic("rn_init");
988 Bzero(rn_zeros, 3 * max_keylen);
989 rn_ones = cp = rn_zeros + max_keylen;
990 addmask_key = cplim = rn_ones + max_keylen;
991 while (cp < cplim)
992 *cp++ = -1;
993 if (rn_inithead((void *)&mask_rnhead, 0) == 0)
994 panic("rn_init 2");
995 }