FreeBSD/Linux Kernel Cross Reference
sys/net/radix.c
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 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 * @(#)radix.c 8.5 (Berkeley) 5/19/95
30 * $FreeBSD: releng/8.0/sys/net/radix.c 186176 2008-12-16 11:01:36Z kmacy $
31 */
32
33 /*
34 * Routines to build and maintain radix trees for routing lookups.
35 */
36 #ifndef _RADIX_H_
37 #include <sys/param.h>
38 #ifdef _KERNEL
39 #include <sys/lock.h>
40 #include <sys/mutex.h>
41 #include <sys/rwlock.h>
42 #include <sys/systm.h>
43 #include <sys/malloc.h>
44 #include <sys/domain.h>
45 #else
46 #include <stdlib.h>
47 #endif
48 #include <sys/syslog.h>
49 #include <net/radix.h>
50 #endif
51
52 #include "opt_mpath.h"
53
54 #ifdef RADIX_MPATH
55 #include <net/radix_mpath.h>
56 #endif
57
58
59 static int rn_walktree_from(struct radix_node_head *h, void *a, void *m,
60 walktree_f_t *f, void *w);
61 static int rn_walktree(struct radix_node_head *, walktree_f_t *, void *);
62 static struct radix_node
63 *rn_insert(void *, struct radix_node_head *, int *,
64 struct radix_node [2]),
65 *rn_newpair(void *, int, struct radix_node[2]),
66 *rn_search(void *, struct radix_node *),
67 *rn_search_m(void *, struct radix_node *, void *);
68
69 static int max_keylen;
70 static struct radix_mask *rn_mkfreelist;
71 static struct radix_node_head *mask_rnhead;
72 /*
73 * Work area -- the following point to 3 buffers of size max_keylen,
74 * allocated in this order in a block of memory malloc'ed by rn_init.
75 */
76 static char *rn_zeros, *rn_ones, *addmask_key;
77
78 #define MKGet(m) { \
79 if (rn_mkfreelist) { \
80 m = rn_mkfreelist; \
81 rn_mkfreelist = (m)->rm_mklist; \
82 } else \
83 R_Malloc(m, struct radix_mask *, sizeof (struct radix_mask)); }
84
85 #define MKFree(m) { (m)->rm_mklist = rn_mkfreelist; rn_mkfreelist = (m);}
86
87 #define rn_masktop (mask_rnhead->rnh_treetop)
88
89 static int rn_lexobetter(void *m_arg, void *n_arg);
90 static struct radix_mask *
91 rn_new_radix_mask(struct radix_node *tt,
92 struct radix_mask *next);
93 static int rn_satisfies_leaf(char *trial, struct radix_node *leaf,
94 int skip);
95
96 /*
97 * The data structure for the keys is a radix tree with one way
98 * branching removed. The index rn_bit at an internal node n represents a bit
99 * position to be tested. The tree is arranged so that all descendants
100 * of a node n have keys whose bits all agree up to position rn_bit - 1.
101 * (We say the index of n is rn_bit.)
102 *
103 * There is at least one descendant which has a one bit at position rn_bit,
104 * and at least one with a zero there.
105 *
106 * A route is determined by a pair of key and mask. We require that the
107 * bit-wise logical and of the key and mask to be the key.
108 * We define the index of a route to associated with the mask to be
109 * the first bit number in the mask where 0 occurs (with bit number 0
110 * representing the highest order bit).
111 *
112 * We say a mask is normal if every bit is 0, past the index of the mask.
113 * If a node n has a descendant (k, m) with index(m) == index(n) == rn_bit,
114 * and m is a normal mask, then the route applies to every descendant of n.
115 * If the index(m) < rn_bit, this implies the trailing last few bits of k
116 * before bit b are all 0, (and hence consequently true of every descendant
117 * of n), so the route applies to all descendants of the node as well.
118 *
119 * Similar logic shows that a non-normal mask m such that
120 * index(m) <= index(n) could potentially apply to many children of n.
121 * Thus, for each non-host route, we attach its mask to a list at an internal
122 * node as high in the tree as we can go.
123 *
124 * The present version of the code makes use of normal routes in short-
125 * circuiting an explict mask and compare operation when testing whether
126 * a key satisfies a normal route, and also in remembering the unique leaf
127 * that governs a subtree.
128 */
129
130 /*
131 * Most of the functions in this code assume that the key/mask arguments
132 * are sockaddr-like structures, where the first byte is an u_char
133 * indicating the size of the entire structure.
134 *
135 * To make the assumption more explicit, we use the LEN() macro to access
136 * this field. It is safe to pass an expression with side effects
137 * to LEN() as the argument is evaluated only once.
138 */
139 #define LEN(x) (*(const u_char *)(x))
140
141 /*
142 * XXX THIS NEEDS TO BE FIXED
143 * In the code, pointers to keys and masks are passed as either
144 * 'void *' (because callers use to pass pointers of various kinds), or
145 * 'caddr_t' (which is fine for pointer arithmetics, but not very
146 * clean when you dereference it to access data). Furthermore, caddr_t
147 * is really 'char *', while the natural type to operate on keys and
148 * masks would be 'u_char'. This mismatch require a lot of casts and
149 * intermediate variables to adapt types that clutter the code.
150 */
151
152 /*
153 * Search a node in the tree matching the key.
154 */
155 static struct radix_node *
156 rn_search(v_arg, head)
157 void *v_arg;
158 struct radix_node *head;
159 {
160 register struct radix_node *x;
161 register caddr_t v;
162
163 for (x = head, v = v_arg; x->rn_bit >= 0;) {
164 if (x->rn_bmask & v[x->rn_offset])
165 x = x->rn_right;
166 else
167 x = x->rn_left;
168 }
169 return (x);
170 }
171
172 /*
173 * Same as above, but with an additional mask.
174 * XXX note this function is used only once.
175 */
176 static struct radix_node *
177 rn_search_m(v_arg, head, m_arg)
178 struct radix_node *head;
179 void *v_arg, *m_arg;
180 {
181 register struct radix_node *x;
182 register caddr_t v = v_arg, m = m_arg;
183
184 for (x = head; x->rn_bit >= 0;) {
185 if ((x->rn_bmask & m[x->rn_offset]) &&
186 (x->rn_bmask & v[x->rn_offset]))
187 x = x->rn_right;
188 else
189 x = x->rn_left;
190 }
191 return x;
192 }
193
194 int
195 rn_refines(m_arg, n_arg)
196 void *m_arg, *n_arg;
197 {
198 register caddr_t m = m_arg, n = n_arg;
199 register caddr_t lim, lim2 = lim = n + LEN(n);
200 int longer = LEN(n++) - (int)LEN(m++);
201 int masks_are_equal = 1;
202
203 if (longer > 0)
204 lim -= longer;
205 while (n < lim) {
206 if (*n & ~(*m))
207 return 0;
208 if (*n++ != *m++)
209 masks_are_equal = 0;
210 }
211 while (n < lim2)
212 if (*n++)
213 return 0;
214 if (masks_are_equal && (longer < 0))
215 for (lim2 = m - longer; m < lim2; )
216 if (*m++)
217 return 1;
218 return (!masks_are_equal);
219 }
220
221 struct radix_node *
222 rn_lookup(v_arg, m_arg, head)
223 void *v_arg, *m_arg;
224 struct radix_node_head *head;
225 {
226 register struct radix_node *x;
227 caddr_t netmask = 0;
228
229 if (m_arg) {
230 x = rn_addmask(m_arg, 1, head->rnh_treetop->rn_offset);
231 if (x == 0)
232 return (0);
233 netmask = x->rn_key;
234 }
235 x = rn_match(v_arg, head);
236 if (x && netmask) {
237 while (x && x->rn_mask != netmask)
238 x = x->rn_dupedkey;
239 }
240 return x;
241 }
242
243 static int
244 rn_satisfies_leaf(trial, leaf, skip)
245 char *trial;
246 register struct radix_node *leaf;
247 int skip;
248 {
249 register char *cp = trial, *cp2 = leaf->rn_key, *cp3 = leaf->rn_mask;
250 char *cplim;
251 int length = min(LEN(cp), LEN(cp2));
252
253 if (cp3 == 0)
254 cp3 = rn_ones;
255 else
256 length = min(length, *(u_char *)cp3);
257 cplim = cp + length; cp3 += skip; cp2 += skip;
258 for (cp += skip; cp < cplim; cp++, cp2++, cp3++)
259 if ((*cp ^ *cp2) & *cp3)
260 return 0;
261 return 1;
262 }
263
264 struct radix_node *
265 rn_match(v_arg, head)
266 void *v_arg;
267 struct radix_node_head *head;
268 {
269 caddr_t v = v_arg;
270 register struct radix_node *t = head->rnh_treetop, *x;
271 register caddr_t cp = v, cp2;
272 caddr_t cplim;
273 struct radix_node *saved_t, *top = t;
274 int off = t->rn_offset, vlen = LEN(cp), matched_off;
275 register int test, b, rn_bit;
276
277 /*
278 * Open code rn_search(v, top) to avoid overhead of extra
279 * subroutine call.
280 */
281 for (; t->rn_bit >= 0; ) {
282 if (t->rn_bmask & cp[t->rn_offset])
283 t = t->rn_right;
284 else
285 t = t->rn_left;
286 }
287 /*
288 * See if we match exactly as a host destination
289 * or at least learn how many bits match, for normal mask finesse.
290 *
291 * It doesn't hurt us to limit how many bytes to check
292 * to the length of the mask, since if it matches we had a genuine
293 * match and the leaf we have is the most specific one anyway;
294 * if it didn't match with a shorter length it would fail
295 * with a long one. This wins big for class B&C netmasks which
296 * are probably the most common case...
297 */
298 if (t->rn_mask)
299 vlen = *(u_char *)t->rn_mask;
300 cp += off; cp2 = t->rn_key + off; cplim = v + vlen;
301 for (; cp < cplim; cp++, cp2++)
302 if (*cp != *cp2)
303 goto on1;
304 /*
305 * This extra grot is in case we are explicitly asked
306 * to look up the default. Ugh!
307 *
308 * Never return the root node itself, it seems to cause a
309 * lot of confusion.
310 */
311 if (t->rn_flags & RNF_ROOT)
312 t = t->rn_dupedkey;
313 return t;
314 on1:
315 test = (*cp ^ *cp2) & 0xff; /* find first bit that differs */
316 for (b = 7; (test >>= 1) > 0;)
317 b--;
318 matched_off = cp - v;
319 b += matched_off << 3;
320 rn_bit = -1 - b;
321 /*
322 * If there is a host route in a duped-key chain, it will be first.
323 */
324 if ((saved_t = t)->rn_mask == 0)
325 t = t->rn_dupedkey;
326 for (; t; t = t->rn_dupedkey)
327 /*
328 * Even if we don't match exactly as a host,
329 * we may match if the leaf we wound up at is
330 * a route to a net.
331 */
332 if (t->rn_flags & RNF_NORMAL) {
333 if (rn_bit <= t->rn_bit)
334 return t;
335 } else if (rn_satisfies_leaf(v, t, matched_off))
336 return t;
337 t = saved_t;
338 /* start searching up the tree */
339 do {
340 register struct radix_mask *m;
341 t = t->rn_parent;
342 m = t->rn_mklist;
343 /*
344 * If non-contiguous masks ever become important
345 * we can restore the masking and open coding of
346 * the search and satisfaction test and put the
347 * calculation of "off" back before the "do".
348 */
349 while (m) {
350 if (m->rm_flags & RNF_NORMAL) {
351 if (rn_bit <= m->rm_bit)
352 return (m->rm_leaf);
353 } else {
354 off = min(t->rn_offset, matched_off);
355 x = rn_search_m(v, t, m->rm_mask);
356 while (x && x->rn_mask != m->rm_mask)
357 x = x->rn_dupedkey;
358 if (x && rn_satisfies_leaf(v, x, off))
359 return x;
360 }
361 m = m->rm_mklist;
362 }
363 } while (t != top);
364 return 0;
365 }
366
367 #ifdef RN_DEBUG
368 int rn_nodenum;
369 struct radix_node *rn_clist;
370 int rn_saveinfo;
371 int rn_debug = 1;
372 #endif
373
374 /*
375 * Whenever we add a new leaf to the tree, we also add a parent node,
376 * so we allocate them as an array of two elements: the first one must be
377 * the leaf (see RNTORT() in route.c), the second one is the parent.
378 * This routine initializes the relevant fields of the nodes, so that
379 * the leaf is the left child of the parent node, and both nodes have
380 * (almost) all all fields filled as appropriate.
381 * (XXX some fields are left unset, see the '#if 0' section).
382 * The function returns a pointer to the parent node.
383 */
384
385 static struct radix_node *
386 rn_newpair(v, b, nodes)
387 void *v;
388 int b;
389 struct radix_node nodes[2];
390 {
391 register struct radix_node *tt = nodes, *t = tt + 1;
392 t->rn_bit = b;
393 t->rn_bmask = 0x80 >> (b & 7);
394 t->rn_left = tt;
395 t->rn_offset = b >> 3;
396
397 #if 0 /* XXX perhaps we should fill these fields as well. */
398 t->rn_parent = t->rn_right = NULL;
399
400 tt->rn_mask = NULL;
401 tt->rn_dupedkey = NULL;
402 tt->rn_bmask = 0;
403 #endif
404 tt->rn_bit = -1;
405 tt->rn_key = (caddr_t)v;
406 tt->rn_parent = t;
407 tt->rn_flags = t->rn_flags = RNF_ACTIVE;
408 tt->rn_mklist = t->rn_mklist = 0;
409 #ifdef RN_DEBUG
410 tt->rn_info = rn_nodenum++; t->rn_info = rn_nodenum++;
411 tt->rn_twin = t;
412 tt->rn_ybro = rn_clist;
413 rn_clist = tt;
414 #endif
415 return t;
416 }
417
418 static struct radix_node *
419 rn_insert(v_arg, head, dupentry, nodes)
420 void *v_arg;
421 struct radix_node_head *head;
422 int *dupentry;
423 struct radix_node nodes[2];
424 {
425 caddr_t v = v_arg;
426 struct radix_node *top = head->rnh_treetop;
427 int head_off = top->rn_offset, vlen = (int)LEN(v);
428 register struct radix_node *t = rn_search(v_arg, top);
429 register caddr_t cp = v + head_off;
430 register int b;
431 struct radix_node *tt;
432 /*
433 * Find first bit at which v and t->rn_key differ
434 */
435 {
436 register caddr_t cp2 = t->rn_key + head_off;
437 register int cmp_res;
438 caddr_t cplim = v + vlen;
439
440 while (cp < cplim)
441 if (*cp2++ != *cp++)
442 goto on1;
443 *dupentry = 1;
444 return t;
445 on1:
446 *dupentry = 0;
447 cmp_res = (cp[-1] ^ cp2[-1]) & 0xff;
448 for (b = (cp - v) << 3; cmp_res; b--)
449 cmp_res >>= 1;
450 }
451 {
452 register struct radix_node *p, *x = top;
453 cp = v;
454 do {
455 p = x;
456 if (cp[x->rn_offset] & x->rn_bmask)
457 x = x->rn_right;
458 else
459 x = x->rn_left;
460 } while (b > (unsigned) x->rn_bit);
461 /* x->rn_bit < b && x->rn_bit >= 0 */
462 #ifdef RN_DEBUG
463 if (rn_debug)
464 log(LOG_DEBUG, "rn_insert: Going In:\n"), traverse(p);
465 #endif
466 t = rn_newpair(v_arg, b, nodes);
467 tt = t->rn_left;
468 if ((cp[p->rn_offset] & p->rn_bmask) == 0)
469 p->rn_left = t;
470 else
471 p->rn_right = t;
472 x->rn_parent = t;
473 t->rn_parent = p; /* frees x, p as temp vars below */
474 if ((cp[t->rn_offset] & t->rn_bmask) == 0) {
475 t->rn_right = x;
476 } else {
477 t->rn_right = tt;
478 t->rn_left = x;
479 }
480 #ifdef RN_DEBUG
481 if (rn_debug)
482 log(LOG_DEBUG, "rn_insert: Coming Out:\n"), traverse(p);
483 #endif
484 }
485 return (tt);
486 }
487
488 struct radix_node *
489 rn_addmask(n_arg, search, skip)
490 int search, skip;
491 void *n_arg;
492 {
493 caddr_t netmask = (caddr_t)n_arg;
494 register struct radix_node *x;
495 register caddr_t cp, cplim;
496 register int b = 0, mlen, j;
497 int maskduplicated, m0, isnormal;
498 struct radix_node *saved_x;
499 static int last_zeroed = 0;
500
501 if ((mlen = LEN(netmask)) > max_keylen)
502 mlen = max_keylen;
503 if (skip == 0)
504 skip = 1;
505 if (mlen <= skip)
506 return (mask_rnhead->rnh_nodes);
507 if (skip > 1)
508 bcopy(rn_ones + 1, addmask_key + 1, skip - 1);
509 if ((m0 = mlen) > skip)
510 bcopy(netmask + skip, addmask_key + skip, mlen - skip);
511 /*
512 * Trim trailing zeroes.
513 */
514 for (cp = addmask_key + mlen; (cp > addmask_key) && cp[-1] == 0;)
515 cp--;
516 mlen = cp - addmask_key;
517 if (mlen <= skip) {
518 if (m0 >= last_zeroed)
519 last_zeroed = mlen;
520 return (mask_rnhead->rnh_nodes);
521 }
522 if (m0 < last_zeroed)
523 bzero(addmask_key + m0, last_zeroed - m0);
524 *addmask_key = last_zeroed = mlen;
525 x = rn_search(addmask_key, rn_masktop);
526 if (bcmp(addmask_key, x->rn_key, mlen) != 0)
527 x = 0;
528 if (x || search)
529 return (x);
530 R_Zalloc(x, struct radix_node *, max_keylen + 2 * sizeof (*x));
531 if ((saved_x = x) == 0)
532 return (0);
533 netmask = cp = (caddr_t)(x + 2);
534 bcopy(addmask_key, cp, mlen);
535 x = rn_insert(cp, mask_rnhead, &maskduplicated, x);
536 if (maskduplicated) {
537 log(LOG_ERR, "rn_addmask: mask impossibly already in tree");
538 Free(saved_x);
539 return (x);
540 }
541 /*
542 * Calculate index of mask, and check for normalcy.
543 * First find the first byte with a 0 bit, then if there are
544 * more bits left (remember we already trimmed the trailing 0's),
545 * the pattern must be one of those in normal_chars[], or we have
546 * a non-contiguous mask.
547 */
548 cplim = netmask + mlen;
549 isnormal = 1;
550 for (cp = netmask + skip; (cp < cplim) && *(u_char *)cp == 0xff;)
551 cp++;
552 if (cp != cplim) {
553 static char normal_chars[] = {
554 0, 0x80, 0xc0, 0xe0, 0xf0, 0xf8, 0xfc, 0xfe, 0xff};
555
556 for (j = 0x80; (j & *cp) != 0; j >>= 1)
557 b++;
558 if (*cp != normal_chars[b] || cp != (cplim - 1))
559 isnormal = 0;
560 }
561 b += (cp - netmask) << 3;
562 x->rn_bit = -1 - b;
563 if (isnormal)
564 x->rn_flags |= RNF_NORMAL;
565 return (x);
566 }
567
568 static int /* XXX: arbitrary ordering for non-contiguous masks */
569 rn_lexobetter(m_arg, n_arg)
570 void *m_arg, *n_arg;
571 {
572 register u_char *mp = m_arg, *np = n_arg, *lim;
573
574 if (LEN(mp) > LEN(np))
575 return 1; /* not really, but need to check longer one first */
576 if (LEN(mp) == LEN(np))
577 for (lim = mp + LEN(mp); mp < lim;)
578 if (*mp++ > *np++)
579 return 1;
580 return 0;
581 }
582
583 static struct radix_mask *
584 rn_new_radix_mask(tt, next)
585 register struct radix_node *tt;
586 register struct radix_mask *next;
587 {
588 register struct radix_mask *m;
589
590 MKGet(m);
591 if (m == 0) {
592 log(LOG_ERR, "Mask for route not entered\n");
593 return (0);
594 }
595 bzero(m, sizeof *m);
596 m->rm_bit = tt->rn_bit;
597 m->rm_flags = tt->rn_flags;
598 if (tt->rn_flags & RNF_NORMAL)
599 m->rm_leaf = tt;
600 else
601 m->rm_mask = tt->rn_mask;
602 m->rm_mklist = next;
603 tt->rn_mklist = m;
604 return m;
605 }
606
607 struct radix_node *
608 rn_addroute(v_arg, n_arg, head, treenodes)
609 void *v_arg, *n_arg;
610 struct radix_node_head *head;
611 struct radix_node treenodes[2];
612 {
613 caddr_t v = (caddr_t)v_arg, netmask = (caddr_t)n_arg;
614 register struct radix_node *t, *x = 0, *tt;
615 struct radix_node *saved_tt, *top = head->rnh_treetop;
616 short b = 0, b_leaf = 0;
617 int keyduplicated;
618 caddr_t mmask;
619 struct radix_mask *m, **mp;
620
621 /*
622 * In dealing with non-contiguous masks, there may be
623 * many different routes which have the same mask.
624 * We will find it useful to have a unique pointer to
625 * the mask to speed avoiding duplicate references at
626 * nodes and possibly save time in calculating indices.
627 */
628 if (netmask) {
629 if ((x = rn_addmask(netmask, 0, top->rn_offset)) == 0)
630 return (0);
631 b_leaf = x->rn_bit;
632 b = -1 - x->rn_bit;
633 netmask = x->rn_key;
634 }
635 /*
636 * Deal with duplicated keys: attach node to previous instance
637 */
638 saved_tt = tt = rn_insert(v, head, &keyduplicated, treenodes);
639 if (keyduplicated) {
640 for (t = tt; tt; t = tt, tt = tt->rn_dupedkey) {
641 #ifdef RADIX_MPATH
642 /* permit multipath, if enabled for the family */
643 if (rn_mpath_capable(head) && netmask == tt->rn_mask) {
644 /*
645 * go down to the end of multipaths, so that
646 * new entry goes into the end of rn_dupedkey
647 * chain.
648 */
649 do {
650 t = tt;
651 tt = tt->rn_dupedkey;
652 } while (tt && t->rn_mask == tt->rn_mask);
653 break;
654 }
655 #endif
656 if (tt->rn_mask == netmask)
657 return (0);
658 if (netmask == 0 ||
659 (tt->rn_mask &&
660 ((b_leaf < tt->rn_bit) /* index(netmask) > node */
661 || rn_refines(netmask, tt->rn_mask)
662 || rn_lexobetter(netmask, tt->rn_mask))))
663 break;
664 }
665 /*
666 * If the mask is not duplicated, we wouldn't
667 * find it among possible duplicate key entries
668 * anyway, so the above test doesn't hurt.
669 *
670 * We sort the masks for a duplicated key the same way as
671 * in a masklist -- most specific to least specific.
672 * This may require the unfortunate nuisance of relocating
673 * the head of the list.
674 *
675 * We also reverse, or doubly link the list through the
676 * parent pointer.
677 */
678 if (tt == saved_tt) {
679 struct radix_node *xx = x;
680 /* link in at head of list */
681 (tt = treenodes)->rn_dupedkey = t;
682 tt->rn_flags = t->rn_flags;
683 tt->rn_parent = x = t->rn_parent;
684 t->rn_parent = tt; /* parent */
685 if (x->rn_left == t)
686 x->rn_left = tt;
687 else
688 x->rn_right = tt;
689 saved_tt = tt; x = xx;
690 } else {
691 (tt = treenodes)->rn_dupedkey = t->rn_dupedkey;
692 t->rn_dupedkey = tt;
693 tt->rn_parent = t; /* parent */
694 if (tt->rn_dupedkey) /* parent */
695 tt->rn_dupedkey->rn_parent = tt; /* parent */
696 }
697 #ifdef RN_DEBUG
698 t=tt+1; tt->rn_info = rn_nodenum++; t->rn_info = rn_nodenum++;
699 tt->rn_twin = t; tt->rn_ybro = rn_clist; rn_clist = tt;
700 #endif
701 tt->rn_key = (caddr_t) v;
702 tt->rn_bit = -1;
703 tt->rn_flags = RNF_ACTIVE;
704 }
705 /*
706 * Put mask in tree.
707 */
708 if (netmask) {
709 tt->rn_mask = netmask;
710 tt->rn_bit = x->rn_bit;
711 tt->rn_flags |= x->rn_flags & RNF_NORMAL;
712 }
713 t = saved_tt->rn_parent;
714 if (keyduplicated)
715 goto on2;
716 b_leaf = -1 - t->rn_bit;
717 if (t->rn_right == saved_tt)
718 x = t->rn_left;
719 else
720 x = t->rn_right;
721 /* Promote general routes from below */
722 if (x->rn_bit < 0) {
723 for (mp = &t->rn_mklist; x; x = x->rn_dupedkey)
724 if (x->rn_mask && (x->rn_bit >= b_leaf) && x->rn_mklist == 0) {
725 *mp = m = rn_new_radix_mask(x, 0);
726 if (m)
727 mp = &m->rm_mklist;
728 }
729 } else if (x->rn_mklist) {
730 /*
731 * Skip over masks whose index is > that of new node
732 */
733 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist)
734 if (m->rm_bit >= b_leaf)
735 break;
736 t->rn_mklist = m; *mp = 0;
737 }
738 on2:
739 /* Add new route to highest possible ancestor's list */
740 if ((netmask == 0) || (b > t->rn_bit ))
741 return tt; /* can't lift at all */
742 b_leaf = tt->rn_bit;
743 do {
744 x = t;
745 t = t->rn_parent;
746 } while (b <= t->rn_bit && x != top);
747 /*
748 * Search through routes associated with node to
749 * insert new route according to index.
750 * Need same criteria as when sorting dupedkeys to avoid
751 * double loop on deletion.
752 */
753 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist) {
754 if (m->rm_bit < b_leaf)
755 continue;
756 if (m->rm_bit > b_leaf)
757 break;
758 if (m->rm_flags & RNF_NORMAL) {
759 mmask = m->rm_leaf->rn_mask;
760 if (tt->rn_flags & RNF_NORMAL) {
761 log(LOG_ERR,
762 "Non-unique normal route, mask not entered\n");
763 return tt;
764 }
765 } else
766 mmask = m->rm_mask;
767 if (mmask == netmask) {
768 m->rm_refs++;
769 tt->rn_mklist = m;
770 return tt;
771 }
772 if (rn_refines(netmask, mmask)
773 || rn_lexobetter(netmask, mmask))
774 break;
775 }
776 *mp = rn_new_radix_mask(tt, *mp);
777 return tt;
778 }
779
780 struct radix_node *
781 rn_delete(v_arg, netmask_arg, head)
782 void *v_arg, *netmask_arg;
783 struct radix_node_head *head;
784 {
785 register struct radix_node *t, *p, *x, *tt;
786 struct radix_mask *m, *saved_m, **mp;
787 struct radix_node *dupedkey, *saved_tt, *top;
788 caddr_t v, netmask;
789 int b, head_off, vlen;
790
791 v = v_arg;
792 netmask = netmask_arg;
793 x = head->rnh_treetop;
794 tt = rn_search(v, x);
795 head_off = x->rn_offset;
796 vlen = LEN(v);
797 saved_tt = tt;
798 top = x;
799 if (tt == 0 ||
800 bcmp(v + head_off, tt->rn_key + head_off, vlen - head_off))
801 return (0);
802 /*
803 * Delete our route from mask lists.
804 */
805 if (netmask) {
806 if ((x = rn_addmask(netmask, 1, head_off)) == 0)
807 return (0);
808 netmask = x->rn_key;
809 while (tt->rn_mask != netmask)
810 if ((tt = tt->rn_dupedkey) == 0)
811 return (0);
812 }
813 if (tt->rn_mask == 0 || (saved_m = m = tt->rn_mklist) == 0)
814 goto on1;
815 if (tt->rn_flags & RNF_NORMAL) {
816 if (m->rm_leaf != tt || m->rm_refs > 0) {
817 log(LOG_ERR, "rn_delete: inconsistent annotation\n");
818 return 0; /* dangling ref could cause disaster */
819 }
820 } else {
821 if (m->rm_mask != tt->rn_mask) {
822 log(LOG_ERR, "rn_delete: inconsistent annotation\n");
823 goto on1;
824 }
825 if (--m->rm_refs >= 0)
826 goto on1;
827 }
828 b = -1 - tt->rn_bit;
829 t = saved_tt->rn_parent;
830 if (b > t->rn_bit)
831 goto on1; /* Wasn't lifted at all */
832 do {
833 x = t;
834 t = t->rn_parent;
835 } while (b <= t->rn_bit && x != top);
836 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist)
837 if (m == saved_m) {
838 *mp = m->rm_mklist;
839 MKFree(m);
840 break;
841 }
842 if (m == 0) {
843 log(LOG_ERR, "rn_delete: couldn't find our annotation\n");
844 if (tt->rn_flags & RNF_NORMAL)
845 return (0); /* Dangling ref to us */
846 }
847 on1:
848 /*
849 * Eliminate us from tree
850 */
851 if (tt->rn_flags & RNF_ROOT)
852 return (0);
853 #ifdef RN_DEBUG
854 /* Get us out of the creation list */
855 for (t = rn_clist; t && t->rn_ybro != tt; t = t->rn_ybro) {}
856 if (t) t->rn_ybro = tt->rn_ybro;
857 #endif
858 t = tt->rn_parent;
859 dupedkey = saved_tt->rn_dupedkey;
860 if (dupedkey) {
861 /*
862 * Here, tt is the deletion target and
863 * saved_tt is the head of the dupekey chain.
864 */
865 if (tt == saved_tt) {
866 /* remove from head of chain */
867 x = dupedkey; x->rn_parent = t;
868 if (t->rn_left == tt)
869 t->rn_left = x;
870 else
871 t->rn_right = x;
872 } else {
873 /* find node in front of tt on the chain */
874 for (x = p = saved_tt; p && p->rn_dupedkey != tt;)
875 p = p->rn_dupedkey;
876 if (p) {
877 p->rn_dupedkey = tt->rn_dupedkey;
878 if (tt->rn_dupedkey) /* parent */
879 tt->rn_dupedkey->rn_parent = p;
880 /* parent */
881 } else log(LOG_ERR, "rn_delete: couldn't find us\n");
882 }
883 t = tt + 1;
884 if (t->rn_flags & RNF_ACTIVE) {
885 #ifndef RN_DEBUG
886 *++x = *t;
887 p = t->rn_parent;
888 #else
889 b = t->rn_info;
890 *++x = *t;
891 t->rn_info = b;
892 p = t->rn_parent;
893 #endif
894 if (p->rn_left == t)
895 p->rn_left = x;
896 else
897 p->rn_right = x;
898 x->rn_left->rn_parent = x;
899 x->rn_right->rn_parent = x;
900 }
901 goto out;
902 }
903 if (t->rn_left == tt)
904 x = t->rn_right;
905 else
906 x = t->rn_left;
907 p = t->rn_parent;
908 if (p->rn_right == t)
909 p->rn_right = x;
910 else
911 p->rn_left = x;
912 x->rn_parent = p;
913 /*
914 * Demote routes attached to us.
915 */
916 if (t->rn_mklist) {
917 if (x->rn_bit >= 0) {
918 for (mp = &x->rn_mklist; (m = *mp);)
919 mp = &m->rm_mklist;
920 *mp = t->rn_mklist;
921 } else {
922 /* If there are any key,mask pairs in a sibling
923 duped-key chain, some subset will appear sorted
924 in the same order attached to our mklist */
925 for (m = t->rn_mklist; m && x; x = x->rn_dupedkey)
926 if (m == x->rn_mklist) {
927 struct radix_mask *mm = m->rm_mklist;
928 x->rn_mklist = 0;
929 if (--(m->rm_refs) < 0)
930 MKFree(m);
931 m = mm;
932 }
933 if (m)
934 log(LOG_ERR,
935 "rn_delete: Orphaned Mask %p at %p\n",
936 (void *)m, (void *)x);
937 }
938 }
939 /*
940 * We may be holding an active internal node in the tree.
941 */
942 x = tt + 1;
943 if (t != x) {
944 #ifndef RN_DEBUG
945 *t = *x;
946 #else
947 b = t->rn_info;
948 *t = *x;
949 t->rn_info = b;
950 #endif
951 t->rn_left->rn_parent = t;
952 t->rn_right->rn_parent = t;
953 p = x->rn_parent;
954 if (p->rn_left == x)
955 p->rn_left = t;
956 else
957 p->rn_right = t;
958 }
959 out:
960 tt->rn_flags &= ~RNF_ACTIVE;
961 tt[1].rn_flags &= ~RNF_ACTIVE;
962 return (tt);
963 }
964
965 /*
966 * This is the same as rn_walktree() except for the parameters and the
967 * exit.
968 */
969 static int
970 rn_walktree_from(h, a, m, f, w)
971 struct radix_node_head *h;
972 void *a, *m;
973 walktree_f_t *f;
974 void *w;
975 {
976 int error;
977 struct radix_node *base, *next;
978 u_char *xa = (u_char *)a;
979 u_char *xm = (u_char *)m;
980 register struct radix_node *rn, *last = 0 /* shut up gcc */;
981 int stopping = 0;
982 int lastb;
983
984 /*
985 * rn_search_m is sort-of-open-coded here. We cannot use the
986 * function because we need to keep track of the last node seen.
987 */
988 /* printf("about to search\n"); */
989 for (rn = h->rnh_treetop; rn->rn_bit >= 0; ) {
990 last = rn;
991 /* printf("rn_bit %d, rn_bmask %x, xm[rn_offset] %x\n",
992 rn->rn_bit, rn->rn_bmask, xm[rn->rn_offset]); */
993 if (!(rn->rn_bmask & xm[rn->rn_offset])) {
994 break;
995 }
996 if (rn->rn_bmask & xa[rn->rn_offset]) {
997 rn = rn->rn_right;
998 } else {
999 rn = rn->rn_left;
1000 }
1001 }
1002 /* printf("done searching\n"); */
1003
1004 /*
1005 * Two cases: either we stepped off the end of our mask,
1006 * in which case last == rn, or we reached a leaf, in which
1007 * case we want to start from the last node we looked at.
1008 * Either way, last is the node we want to start from.
1009 */
1010 rn = last;
1011 lastb = rn->rn_bit;
1012
1013 /* printf("rn %p, lastb %d\n", rn, lastb);*/
1014
1015 /*
1016 * This gets complicated because we may delete the node
1017 * while applying the function f to it, so we need to calculate
1018 * the successor node in advance.
1019 */
1020 while (rn->rn_bit >= 0)
1021 rn = rn->rn_left;
1022
1023 while (!stopping) {
1024 /* printf("node %p (%d)\n", rn, rn->rn_bit); */
1025 base = rn;
1026 /* If at right child go back up, otherwise, go right */
1027 while (rn->rn_parent->rn_right == rn
1028 && !(rn->rn_flags & RNF_ROOT)) {
1029 rn = rn->rn_parent;
1030
1031 /* if went up beyond last, stop */
1032 if (rn->rn_bit <= lastb) {
1033 stopping = 1;
1034 /* printf("up too far\n"); */
1035 /*
1036 * XXX we should jump to the 'Process leaves'
1037 * part, because the values of 'rn' and 'next'
1038 * we compute will not be used. Not a big deal
1039 * because this loop will terminate, but it is
1040 * inefficient and hard to understand!
1041 */
1042 }
1043 }
1044
1045 /*
1046 * At the top of the tree, no need to traverse the right
1047 * half, prevent the traversal of the entire tree in the
1048 * case of default route.
1049 */
1050 if (rn->rn_parent->rn_flags & RNF_ROOT)
1051 stopping = 1;
1052
1053 /* Find the next *leaf* since next node might vanish, too */
1054 for (rn = rn->rn_parent->rn_right; rn->rn_bit >= 0;)
1055 rn = rn->rn_left;
1056 next = rn;
1057 /* Process leaves */
1058 while ((rn = base) != 0) {
1059 base = rn->rn_dupedkey;
1060 /* printf("leaf %p\n", rn); */
1061 if (!(rn->rn_flags & RNF_ROOT)
1062 && (error = (*f)(rn, w)))
1063 return (error);
1064 }
1065 rn = next;
1066
1067 if (rn->rn_flags & RNF_ROOT) {
1068 /* printf("root, stopping"); */
1069 stopping = 1;
1070 }
1071
1072 }
1073 return 0;
1074 }
1075
1076 static int
1077 rn_walktree(h, f, w)
1078 struct radix_node_head *h;
1079 walktree_f_t *f;
1080 void *w;
1081 {
1082 int error;
1083 struct radix_node *base, *next;
1084 register struct radix_node *rn = h->rnh_treetop;
1085 /*
1086 * This gets complicated because we may delete the node
1087 * while applying the function f to it, so we need to calculate
1088 * the successor node in advance.
1089 */
1090
1091 /* First time through node, go left */
1092 while (rn->rn_bit >= 0)
1093 rn = rn->rn_left;
1094 for (;;) {
1095 base = rn;
1096 /* If at right child go back up, otherwise, go right */
1097 while (rn->rn_parent->rn_right == rn
1098 && (rn->rn_flags & RNF_ROOT) == 0)
1099 rn = rn->rn_parent;
1100 /* Find the next *leaf* since next node might vanish, too */
1101 for (rn = rn->rn_parent->rn_right; rn->rn_bit >= 0;)
1102 rn = rn->rn_left;
1103 next = rn;
1104 /* Process leaves */
1105 while ((rn = base)) {
1106 base = rn->rn_dupedkey;
1107 if (!(rn->rn_flags & RNF_ROOT)
1108 && (error = (*f)(rn, w)))
1109 return (error);
1110 }
1111 rn = next;
1112 if (rn->rn_flags & RNF_ROOT)
1113 return (0);
1114 }
1115 /* NOTREACHED */
1116 }
1117
1118 /*
1119 * Allocate and initialize an empty tree. This has 3 nodes, which are
1120 * part of the radix_node_head (in the order <left,root,right>) and are
1121 * marked RNF_ROOT so they cannot be freed.
1122 * The leaves have all-zero and all-one keys, with significant
1123 * bits starting at 'off'.
1124 * Return 1 on success, 0 on error.
1125 */
1126 int
1127 rn_inithead(head, off)
1128 void **head;
1129 int off;
1130 {
1131 register struct radix_node_head *rnh;
1132 register struct radix_node *t, *tt, *ttt;
1133 if (*head)
1134 return (1);
1135 R_Zalloc(rnh, struct radix_node_head *, sizeof (*rnh));
1136 if (rnh == 0)
1137 return (0);
1138 #ifdef _KERNEL
1139 RADIX_NODE_HEAD_LOCK_INIT(rnh);
1140 #endif
1141 *head = rnh;
1142 t = rn_newpair(rn_zeros, off, rnh->rnh_nodes);
1143 ttt = rnh->rnh_nodes + 2;
1144 t->rn_right = ttt;
1145 t->rn_parent = t;
1146 tt = t->rn_left; /* ... which in turn is rnh->rnh_nodes */
1147 tt->rn_flags = t->rn_flags = RNF_ROOT | RNF_ACTIVE;
1148 tt->rn_bit = -1 - off;
1149 *ttt = *tt;
1150 ttt->rn_key = rn_ones;
1151 rnh->rnh_addaddr = rn_addroute;
1152 rnh->rnh_deladdr = rn_delete;
1153 rnh->rnh_matchaddr = rn_match;
1154 rnh->rnh_lookup = rn_lookup;
1155 rnh->rnh_walktree = rn_walktree;
1156 rnh->rnh_walktree_from = rn_walktree_from;
1157 rnh->rnh_treetop = t;
1158 return (1);
1159 }
1160
1161 void
1162 rn_init()
1163 {
1164 char *cp, *cplim;
1165 #ifdef _KERNEL
1166 struct domain *dom;
1167
1168 for (dom = domains; dom; dom = dom->dom_next)
1169 if (dom->dom_maxrtkey > max_keylen)
1170 max_keylen = dom->dom_maxrtkey;
1171 #endif
1172 if (max_keylen == 0) {
1173 log(LOG_ERR,
1174 "rn_init: radix functions require max_keylen be set\n");
1175 return;
1176 }
1177 R_Malloc(rn_zeros, char *, 3 * max_keylen);
1178 if (rn_zeros == NULL)
1179 panic("rn_init");
1180 bzero(rn_zeros, 3 * max_keylen);
1181 rn_ones = cp = rn_zeros + max_keylen;
1182 addmask_key = cplim = rn_ones + max_keylen;
1183 while (cp < cplim)
1184 *cp++ = -1;
1185 if (rn_inithead((void **)(void *)&mask_rnhead, 0) == 0)
1186 panic("rn_init 2");
1187 }
Cache object: 3a0ab3ab44c85461cb2abd52a51deb86
|