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 * $FreeBSD$
31 */
32
33 /*
34 * Routines to build and maintain radix trees for routing lookups.
35 */
36 #include <sys/param.h>
37 #ifdef _KERNEL
38 #include <sys/lock.h>
39 #include <sys/mutex.h>
40 #include <sys/rwlock.h>
41 #include <sys/systm.h>
42 #include <sys/malloc.h>
43 #include <sys/syslog.h>
44 #include <net/radix.h>
45 #include "opt_mpath.h"
46 #ifdef RADIX_MPATH
47 #include <net/radix_mpath.h>
48 #endif
49 #else /* !_KERNEL */
50 #include <stdio.h>
51 #include <strings.h>
52 #include <stdlib.h>
53 #define log(x, arg...) fprintf(stderr, ## arg)
54 #define panic(x) fprintf(stderr, "PANIC: %s", x), exit(1)
55 #define min(a, b) ((a) < (b) ? (a) : (b) )
56 #include <net/radix.h>
57 #endif /* !_KERNEL */
58
59 static struct radix_node
60 *rn_insert(void *, struct radix_head *, int *,
61 struct radix_node [2]),
62 *rn_newpair(void *, int, struct radix_node[2]),
63 *rn_search(void *, struct radix_node *),
64 *rn_search_m(void *, struct radix_node *, void *);
65 static struct radix_node *rn_addmask(void *, struct radix_mask_head *, int,int);
66
67 static void rn_detachhead_internal(struct radix_head *);
68
69 #define RADIX_MAX_KEY_LEN 32
70
71 static char rn_zeros[RADIX_MAX_KEY_LEN];
72 static char rn_ones[RADIX_MAX_KEY_LEN] = {
73 -1, -1, -1, -1, -1, -1, -1, -1,
74 -1, -1, -1, -1, -1, -1, -1, -1,
75 -1, -1, -1, -1, -1, -1, -1, -1,
76 -1, -1, -1, -1, -1, -1, -1, -1,
77 };
78
79
80 static int rn_lexobetter(void *m_arg, void *n_arg);
81 static struct radix_mask *
82 rn_new_radix_mask(struct radix_node *tt,
83 struct radix_mask *next);
84 static int rn_satisfies_leaf(char *trial, struct radix_node *leaf,
85 int skip);
86
87 /*
88 * The data structure for the keys is a radix tree with one way
89 * branching removed. The index rn_bit at an internal node n represents a bit
90 * position to be tested. The tree is arranged so that all descendants
91 * of a node n have keys whose bits all agree up to position rn_bit - 1.
92 * (We say the index of n is rn_bit.)
93 *
94 * There is at least one descendant which has a one bit at position rn_bit,
95 * and at least one with a zero there.
96 *
97 * A route is determined by a pair of key and mask. We require that the
98 * bit-wise logical and of the key and mask to be the key.
99 * We define the index of a route to associated with the mask to be
100 * the first bit number in the mask where 0 occurs (with bit number 0
101 * representing the highest order bit).
102 *
103 * We say a mask is normal if every bit is 0, past the index of the mask.
104 * If a node n has a descendant (k, m) with index(m) == index(n) == rn_bit,
105 * and m is a normal mask, then the route applies to every descendant of n.
106 * If the index(m) < rn_bit, this implies the trailing last few bits of k
107 * before bit b are all 0, (and hence consequently true of every descendant
108 * of n), so the route applies to all descendants of the node as well.
109 *
110 * Similar logic shows that a non-normal mask m such that
111 * index(m) <= index(n) could potentially apply to many children of n.
112 * Thus, for each non-host route, we attach its mask to a list at an internal
113 * node as high in the tree as we can go.
114 *
115 * The present version of the code makes use of normal routes in short-
116 * circuiting an explict mask and compare operation when testing whether
117 * a key satisfies a normal route, and also in remembering the unique leaf
118 * that governs a subtree.
119 */
120
121 /*
122 * Most of the functions in this code assume that the key/mask arguments
123 * are sockaddr-like structures, where the first byte is an u_char
124 * indicating the size of the entire structure.
125 *
126 * To make the assumption more explicit, we use the LEN() macro to access
127 * this field. It is safe to pass an expression with side effects
128 * to LEN() as the argument is evaluated only once.
129 * We cast the result to int as this is the dominant usage.
130 */
131 #define LEN(x) ( (int) (*(const u_char *)(x)) )
132
133 /*
134 * XXX THIS NEEDS TO BE FIXED
135 * In the code, pointers to keys and masks are passed as either
136 * 'void *' (because callers use to pass pointers of various kinds), or
137 * 'caddr_t' (which is fine for pointer arithmetics, but not very
138 * clean when you dereference it to access data). Furthermore, caddr_t
139 * is really 'char *', while the natural type to operate on keys and
140 * masks would be 'u_char'. This mismatch require a lot of casts and
141 * intermediate variables to adapt types that clutter the code.
142 */
143
144 /*
145 * Search a node in the tree matching the key.
146 */
147 static struct radix_node *
148 rn_search(void *v_arg, struct radix_node *head)
149 {
150 struct radix_node *x;
151 caddr_t v;
152
153 for (x = head, v = v_arg; x->rn_bit >= 0;) {
154 if (x->rn_bmask & v[x->rn_offset])
155 x = x->rn_right;
156 else
157 x = x->rn_left;
158 }
159 return (x);
160 }
161
162 /*
163 * Same as above, but with an additional mask.
164 * XXX note this function is used only once.
165 */
166 static struct radix_node *
167 rn_search_m(void *v_arg, struct radix_node *head, void *m_arg)
168 {
169 struct radix_node *x;
170 caddr_t v = v_arg, m = m_arg;
171
172 for (x = head; x->rn_bit >= 0;) {
173 if ((x->rn_bmask & m[x->rn_offset]) &&
174 (x->rn_bmask & v[x->rn_offset]))
175 x = x->rn_right;
176 else
177 x = x->rn_left;
178 }
179 return (x);
180 }
181
182 int
183 rn_refines(void *m_arg, void *n_arg)
184 {
185 caddr_t m = m_arg, n = n_arg;
186 caddr_t lim, lim2 = lim = n + LEN(n);
187 int longer = LEN(n++) - LEN(m++);
188 int masks_are_equal = 1;
189
190 if (longer > 0)
191 lim -= longer;
192 while (n < lim) {
193 if (*n & ~(*m))
194 return (0);
195 if (*n++ != *m++)
196 masks_are_equal = 0;
197 }
198 while (n < lim2)
199 if (*n++)
200 return (0);
201 if (masks_are_equal && (longer < 0))
202 for (lim2 = m - longer; m < lim2; )
203 if (*m++)
204 return (1);
205 return (!masks_are_equal);
206 }
207
208 /*
209 * Search for exact match in given @head.
210 * Assume host bits are cleared in @v_arg if @m_arg is not NULL
211 * Note that prefixes with /32 or /128 masks are treated differently
212 * from host routes.
213 */
214 struct radix_node *
215 rn_lookup(void *v_arg, void *m_arg, struct radix_head *head)
216 {
217 struct radix_node *x;
218 caddr_t netmask;
219
220 if (m_arg != NULL) {
221 /*
222 * Most common case: search exact prefix/mask
223 */
224 x = rn_addmask(m_arg, head->rnh_masks, 1,
225 head->rnh_treetop->rn_offset);
226 if (x == NULL)
227 return (NULL);
228 netmask = x->rn_key;
229
230 x = rn_match(v_arg, head);
231
232 while (x != NULL && x->rn_mask != netmask)
233 x = x->rn_dupedkey;
234
235 return (x);
236 }
237
238 /*
239 * Search for host address.
240 */
241 if ((x = rn_match(v_arg, head)) == NULL)
242 return (NULL);
243
244 /* Check if found key is the same */
245 if (LEN(x->rn_key) != LEN(v_arg) || bcmp(x->rn_key, v_arg, LEN(v_arg)))
246 return (NULL);
247
248 /* Check if this is not host route */
249 if (x->rn_mask != NULL)
250 return (NULL);
251
252 return (x);
253 }
254
255 static int
256 rn_satisfies_leaf(char *trial, struct radix_node *leaf, int skip)
257 {
258 char *cp = trial, *cp2 = leaf->rn_key, *cp3 = leaf->rn_mask;
259 char *cplim;
260 int length = min(LEN(cp), LEN(cp2));
261
262 if (cp3 == NULL)
263 cp3 = rn_ones;
264 else
265 length = min(length, LEN(cp3));
266 cplim = cp + length; cp3 += skip; cp2 += skip;
267 for (cp += skip; cp < cplim; cp++, cp2++, cp3++)
268 if ((*cp ^ *cp2) & *cp3)
269 return (0);
270 return (1);
271 }
272
273 /*
274 * Search for longest-prefix match in given @head
275 */
276 struct radix_node *
277 rn_match(void *v_arg, struct radix_head *head)
278 {
279 caddr_t v = v_arg;
280 struct radix_node *t = head->rnh_treetop, *x;
281 caddr_t cp = v, cp2;
282 caddr_t cplim;
283 struct radix_node *saved_t, *top = t;
284 int off = t->rn_offset, vlen = LEN(cp), matched_off;
285 int test, b, rn_bit;
286
287 /*
288 * Open code rn_search(v, top) to avoid overhead of extra
289 * subroutine call.
290 */
291 for (; t->rn_bit >= 0; ) {
292 if (t->rn_bmask & cp[t->rn_offset])
293 t = t->rn_right;
294 else
295 t = t->rn_left;
296 }
297 /*
298 * See if we match exactly as a host destination
299 * or at least learn how many bits match, for normal mask finesse.
300 *
301 * It doesn't hurt us to limit how many bytes to check
302 * to the length of the mask, since if it matches we had a genuine
303 * match and the leaf we have is the most specific one anyway;
304 * if it didn't match with a shorter length it would fail
305 * with a long one. This wins big for class B&C netmasks which
306 * are probably the most common case...
307 */
308 if (t->rn_mask)
309 vlen = *(u_char *)t->rn_mask;
310 cp += off; cp2 = t->rn_key + off; cplim = v + vlen;
311 for (; cp < cplim; cp++, cp2++)
312 if (*cp != *cp2)
313 goto on1;
314 /*
315 * This extra grot is in case we are explicitly asked
316 * to look up the default. Ugh!
317 *
318 * Never return the root node itself, it seems to cause a
319 * lot of confusion.
320 */
321 if (t->rn_flags & RNF_ROOT)
322 t = t->rn_dupedkey;
323 return (t);
324 on1:
325 test = (*cp ^ *cp2) & 0xff; /* find first bit that differs */
326 for (b = 7; (test >>= 1) > 0;)
327 b--;
328 matched_off = cp - v;
329 b += matched_off << 3;
330 rn_bit = -1 - b;
331 /*
332 * If there is a host route in a duped-key chain, it will be first.
333 */
334 if ((saved_t = t)->rn_mask == 0)
335 t = t->rn_dupedkey;
336 for (; t; t = t->rn_dupedkey)
337 /*
338 * Even if we don't match exactly as a host,
339 * we may match if the leaf we wound up at is
340 * a route to a net.
341 */
342 if (t->rn_flags & RNF_NORMAL) {
343 if (rn_bit <= t->rn_bit)
344 return (t);
345 } else if (rn_satisfies_leaf(v, t, matched_off))
346 return (t);
347 t = saved_t;
348 /* start searching up the tree */
349 do {
350 struct radix_mask *m;
351 t = t->rn_parent;
352 m = t->rn_mklist;
353 /*
354 * If non-contiguous masks ever become important
355 * we can restore the masking and open coding of
356 * the search and satisfaction test and put the
357 * calculation of "off" back before the "do".
358 */
359 while (m) {
360 if (m->rm_flags & RNF_NORMAL) {
361 if (rn_bit <= m->rm_bit)
362 return (m->rm_leaf);
363 } else {
364 off = min(t->rn_offset, matched_off);
365 x = rn_search_m(v, t, m->rm_mask);
366 while (x && x->rn_mask != m->rm_mask)
367 x = x->rn_dupedkey;
368 if (x && rn_satisfies_leaf(v, x, off))
369 return (x);
370 }
371 m = m->rm_mklist;
372 }
373 } while (t != top);
374 return (0);
375 }
376
377 #ifdef RN_DEBUG
378 int rn_nodenum;
379 struct radix_node *rn_clist;
380 int rn_saveinfo;
381 int rn_debug = 1;
382 #endif
383
384 /*
385 * Whenever we add a new leaf to the tree, we also add a parent node,
386 * so we allocate them as an array of two elements: the first one must be
387 * the leaf (see RNTORT() in route.c), the second one is the parent.
388 * This routine initializes the relevant fields of the nodes, so that
389 * the leaf is the left child of the parent node, and both nodes have
390 * (almost) all all fields filled as appropriate.
391 * (XXX some fields are left unset, see the '#if 0' section).
392 * The function returns a pointer to the parent node.
393 */
394
395 static struct radix_node *
396 rn_newpair(void *v, int b, struct radix_node nodes[2])
397 {
398 struct radix_node *tt = nodes, *t = tt + 1;
399 t->rn_bit = b;
400 t->rn_bmask = 0x80 >> (b & 7);
401 t->rn_left = tt;
402 t->rn_offset = b >> 3;
403
404 #if 0 /* XXX perhaps we should fill these fields as well. */
405 t->rn_parent = t->rn_right = NULL;
406
407 tt->rn_mask = NULL;
408 tt->rn_dupedkey = NULL;
409 tt->rn_bmask = 0;
410 #endif
411 tt->rn_bit = -1;
412 tt->rn_key = (caddr_t)v;
413 tt->rn_parent = t;
414 tt->rn_flags = t->rn_flags = RNF_ACTIVE;
415 tt->rn_mklist = t->rn_mklist = 0;
416 #ifdef RN_DEBUG
417 tt->rn_info = rn_nodenum++; t->rn_info = rn_nodenum++;
418 tt->rn_twin = t;
419 tt->rn_ybro = rn_clist;
420 rn_clist = tt;
421 #endif
422 return (t);
423 }
424
425 static struct radix_node *
426 rn_insert(void *v_arg, struct radix_head *head, int *dupentry,
427 struct radix_node nodes[2])
428 {
429 caddr_t v = v_arg;
430 struct radix_node *top = head->rnh_treetop;
431 int head_off = top->rn_offset, vlen = LEN(v);
432 struct radix_node *t = rn_search(v_arg, top);
433 caddr_t cp = v + head_off;
434 int b;
435 struct radix_node *p, *tt, *x;
436 /*
437 * Find first bit at which v and t->rn_key differ
438 */
439 caddr_t cp2 = t->rn_key + head_off;
440 int cmp_res;
441 caddr_t cplim = v + vlen;
442
443 while (cp < cplim)
444 if (*cp2++ != *cp++)
445 goto on1;
446 *dupentry = 1;
447 return (t);
448 on1:
449 *dupentry = 0;
450 cmp_res = (cp[-1] ^ cp2[-1]) & 0xff;
451 for (b = (cp - v) << 3; cmp_res; b--)
452 cmp_res >>= 1;
453
454 x = top;
455 cp = v;
456 do {
457 p = x;
458 if (cp[x->rn_offset] & x->rn_bmask)
459 x = x->rn_right;
460 else
461 x = x->rn_left;
462 } while (b > (unsigned) x->rn_bit);
463 /* x->rn_bit < b && x->rn_bit >= 0 */
464 #ifdef RN_DEBUG
465 if (rn_debug)
466 log(LOG_DEBUG, "rn_insert: Going In:\n"), traverse(p);
467 #endif
468 t = rn_newpair(v_arg, b, nodes);
469 tt = t->rn_left;
470 if ((cp[p->rn_offset] & p->rn_bmask) == 0)
471 p->rn_left = t;
472 else
473 p->rn_right = t;
474 x->rn_parent = t;
475 t->rn_parent = p; /* frees x, p as temp vars below */
476 if ((cp[t->rn_offset] & t->rn_bmask) == 0) {
477 t->rn_right = x;
478 } else {
479 t->rn_right = tt;
480 t->rn_left = x;
481 }
482 #ifdef RN_DEBUG
483 if (rn_debug)
484 log(LOG_DEBUG, "rn_insert: Coming Out:\n"), traverse(p);
485 #endif
486 return (tt);
487 }
488
489 struct radix_node *
490 rn_addmask(void *n_arg, struct radix_mask_head *maskhead, int search, int skip)
491 {
492 unsigned char *netmask = n_arg;
493 unsigned char *cp, *cplim;
494 struct radix_node *x;
495 int b = 0, mlen, j;
496 int maskduplicated, isnormal;
497 struct radix_node *saved_x;
498 unsigned char addmask_key[RADIX_MAX_KEY_LEN];
499
500 if ((mlen = LEN(netmask)) > RADIX_MAX_KEY_LEN)
501 mlen = RADIX_MAX_KEY_LEN;
502 if (skip == 0)
503 skip = 1;
504 if (mlen <= skip)
505 return (maskhead->mask_nodes);
506
507 bzero(addmask_key, RADIX_MAX_KEY_LEN);
508 if (skip > 1)
509 bcopy(rn_ones + 1, addmask_key + 1, skip - 1);
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 return (maskhead->mask_nodes);
519 *addmask_key = mlen;
520 x = rn_search(addmask_key, maskhead->head.rnh_treetop);
521 if (bcmp(addmask_key, x->rn_key, mlen) != 0)
522 x = NULL;
523 if (x || search)
524 return (x);
525 R_Zalloc(x, struct radix_node *, RADIX_MAX_KEY_LEN + 2 * sizeof (*x));
526 if ((saved_x = x) == NULL)
527 return (0);
528 netmask = cp = (unsigned char *)(x + 2);
529 bcopy(addmask_key, cp, mlen);
530 x = rn_insert(cp, &maskhead->head, &maskduplicated, x);
531 if (maskduplicated) {
532 log(LOG_ERR, "rn_addmask: mask impossibly already in tree");
533 R_Free(saved_x);
534 return (x);
535 }
536 /*
537 * Calculate index of mask, and check for normalcy.
538 * First find the first byte with a 0 bit, then if there are
539 * more bits left (remember we already trimmed the trailing 0's),
540 * the bits should be contiguous, otherwise we have got
541 * a non-contiguous mask.
542 */
543 #define CONTIG(_c) (((~(_c) + 1) & (_c)) == (unsigned char)(~(_c) + 1))
544 cplim = netmask + mlen;
545 isnormal = 1;
546 for (cp = netmask + skip; (cp < cplim) && *(u_char *)cp == 0xff;)
547 cp++;
548 if (cp != cplim) {
549 for (j = 0x80; (j & *cp) != 0; j >>= 1)
550 b++;
551 if (!CONTIG(*cp) || cp != (cplim - 1))
552 isnormal = 0;
553 }
554 b += (cp - netmask) << 3;
555 x->rn_bit = -1 - b;
556 if (isnormal)
557 x->rn_flags |= RNF_NORMAL;
558 return (x);
559 }
560
561 static int /* XXX: arbitrary ordering for non-contiguous masks */
562 rn_lexobetter(void *m_arg, void *n_arg)
563 {
564 u_char *mp = m_arg, *np = n_arg, *lim;
565
566 if (LEN(mp) > LEN(np))
567 return (1); /* not really, but need to check longer one first */
568 if (LEN(mp) == LEN(np))
569 for (lim = mp + LEN(mp); mp < lim;)
570 if (*mp++ > *np++)
571 return (1);
572 return (0);
573 }
574
575 static struct radix_mask *
576 rn_new_radix_mask(struct radix_node *tt, struct radix_mask *next)
577 {
578 struct radix_mask *m;
579
580 R_Malloc(m, struct radix_mask *, sizeof (struct radix_mask));
581 if (m == NULL) {
582 log(LOG_ERR, "Failed to allocate route mask\n");
583 return (0);
584 }
585 bzero(m, sizeof(*m));
586 m->rm_bit = tt->rn_bit;
587 m->rm_flags = tt->rn_flags;
588 if (tt->rn_flags & RNF_NORMAL)
589 m->rm_leaf = tt;
590 else
591 m->rm_mask = tt->rn_mask;
592 m->rm_mklist = next;
593 tt->rn_mklist = m;
594 return (m);
595 }
596
597 struct radix_node *
598 rn_addroute(void *v_arg, void *n_arg, struct radix_head *head,
599 struct radix_node treenodes[2])
600 {
601 caddr_t v = (caddr_t)v_arg, netmask = (caddr_t)n_arg;
602 struct radix_node *t, *x = NULL, *tt;
603 struct radix_node *saved_tt, *top = head->rnh_treetop;
604 short b = 0, b_leaf = 0;
605 int keyduplicated;
606 caddr_t mmask;
607 struct radix_mask *m, **mp;
608
609 /*
610 * In dealing with non-contiguous masks, there may be
611 * many different routes which have the same mask.
612 * We will find it useful to have a unique pointer to
613 * the mask to speed avoiding duplicate references at
614 * nodes and possibly save time in calculating indices.
615 */
616 if (netmask) {
617 x = rn_addmask(netmask, head->rnh_masks, 0, top->rn_offset);
618 if (x == NULL)
619 return (0);
620 b_leaf = x->rn_bit;
621 b = -1 - x->rn_bit;
622 netmask = x->rn_key;
623 }
624 /*
625 * Deal with duplicated keys: attach node to previous instance
626 */
627 saved_tt = tt = rn_insert(v, head, &keyduplicated, treenodes);
628 if (keyduplicated) {
629 for (t = tt; tt; t = tt, tt = tt->rn_dupedkey) {
630 #ifdef RADIX_MPATH
631 /* permit multipath, if enabled for the family */
632 if (rn_mpath_capable(head) && netmask == tt->rn_mask) {
633 /*
634 * go down to the end of multipaths, so that
635 * new entry goes into the end of rn_dupedkey
636 * chain.
637 */
638 do {
639 t = tt;
640 tt = tt->rn_dupedkey;
641 } while (tt && t->rn_mask == tt->rn_mask);
642 break;
643 }
644 #endif
645 if (tt->rn_mask == netmask)
646 return (0);
647 if (netmask == 0 ||
648 (tt->rn_mask &&
649 ((b_leaf < tt->rn_bit) /* index(netmask) > node */
650 || rn_refines(netmask, tt->rn_mask)
651 || rn_lexobetter(netmask, tt->rn_mask))))
652 break;
653 }
654 /*
655 * If the mask is not duplicated, we wouldn't
656 * find it among possible duplicate key entries
657 * anyway, so the above test doesn't hurt.
658 *
659 * We sort the masks for a duplicated key the same way as
660 * in a masklist -- most specific to least specific.
661 * This may require the unfortunate nuisance of relocating
662 * the head of the list.
663 *
664 * We also reverse, or doubly link the list through the
665 * parent pointer.
666 */
667 if (tt == saved_tt) {
668 struct radix_node *xx = x;
669 /* link in at head of list */
670 (tt = treenodes)->rn_dupedkey = t;
671 tt->rn_flags = t->rn_flags;
672 tt->rn_parent = x = t->rn_parent;
673 t->rn_parent = tt; /* parent */
674 if (x->rn_left == t)
675 x->rn_left = tt;
676 else
677 x->rn_right = tt;
678 saved_tt = tt; x = xx;
679 } else {
680 (tt = treenodes)->rn_dupedkey = t->rn_dupedkey;
681 t->rn_dupedkey = tt;
682 tt->rn_parent = t; /* parent */
683 if (tt->rn_dupedkey) /* parent */
684 tt->rn_dupedkey->rn_parent = tt; /* parent */
685 }
686 #ifdef RN_DEBUG
687 t=tt+1; tt->rn_info = rn_nodenum++; t->rn_info = rn_nodenum++;
688 tt->rn_twin = t; tt->rn_ybro = rn_clist; rn_clist = tt;
689 #endif
690 tt->rn_key = (caddr_t) v;
691 tt->rn_bit = -1;
692 tt->rn_flags = RNF_ACTIVE;
693 }
694 /*
695 * Put mask in tree.
696 */
697 if (netmask) {
698 tt->rn_mask = netmask;
699 tt->rn_bit = x->rn_bit;
700 tt->rn_flags |= x->rn_flags & RNF_NORMAL;
701 }
702 t = saved_tt->rn_parent;
703 if (keyduplicated)
704 goto on2;
705 b_leaf = -1 - t->rn_bit;
706 if (t->rn_right == saved_tt)
707 x = t->rn_left;
708 else
709 x = t->rn_right;
710 /* Promote general routes from below */
711 if (x->rn_bit < 0) {
712 for (mp = &t->rn_mklist; x; x = x->rn_dupedkey)
713 if (x->rn_mask && (x->rn_bit >= b_leaf) && x->rn_mklist == 0) {
714 *mp = m = rn_new_radix_mask(x, 0);
715 if (m)
716 mp = &m->rm_mklist;
717 }
718 } else if (x->rn_mklist) {
719 /*
720 * Skip over masks whose index is > that of new node
721 */
722 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist)
723 if (m->rm_bit >= b_leaf)
724 break;
725 t->rn_mklist = m; *mp = NULL;
726 }
727 on2:
728 /* Add new route to highest possible ancestor's list */
729 if ((netmask == 0) || (b > t->rn_bit ))
730 return (tt); /* can't lift at all */
731 b_leaf = tt->rn_bit;
732 do {
733 x = t;
734 t = t->rn_parent;
735 } while (b <= t->rn_bit && x != top);
736 /*
737 * Search through routes associated with node to
738 * insert new route according to index.
739 * Need same criteria as when sorting dupedkeys to avoid
740 * double loop on deletion.
741 */
742 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist) {
743 if (m->rm_bit < b_leaf)
744 continue;
745 if (m->rm_bit > b_leaf)
746 break;
747 if (m->rm_flags & RNF_NORMAL) {
748 mmask = m->rm_leaf->rn_mask;
749 if (tt->rn_flags & RNF_NORMAL) {
750 #if !defined(RADIX_MPATH)
751 log(LOG_ERR,
752 "Non-unique normal route, mask not entered\n");
753 #endif
754 return (tt);
755 }
756 } else
757 mmask = m->rm_mask;
758 if (mmask == netmask) {
759 m->rm_refs++;
760 tt->rn_mklist = m;
761 return (tt);
762 }
763 if (rn_refines(netmask, mmask)
764 || rn_lexobetter(netmask, mmask))
765 break;
766 }
767 *mp = rn_new_radix_mask(tt, *mp);
768 return (tt);
769 }
770
771 struct radix_node *
772 rn_delete(void *v_arg, void *netmask_arg, struct radix_head *head)
773 {
774 struct radix_node *t, *p, *x, *tt;
775 struct radix_mask *m, *saved_m, **mp;
776 struct radix_node *dupedkey, *saved_tt, *top;
777 caddr_t v, netmask;
778 int b, head_off, vlen;
779
780 v = v_arg;
781 netmask = netmask_arg;
782 x = head->rnh_treetop;
783 tt = rn_search(v, x);
784 head_off = x->rn_offset;
785 vlen = LEN(v);
786 saved_tt = tt;
787 top = x;
788 if (tt == NULL ||
789 bcmp(v + head_off, tt->rn_key + head_off, vlen - head_off))
790 return (0);
791 /*
792 * Delete our route from mask lists.
793 */
794 if (netmask) {
795 x = rn_addmask(netmask, head->rnh_masks, 1, head_off);
796 if (x == NULL)
797 return (0);
798 netmask = x->rn_key;
799 while (tt->rn_mask != netmask)
800 if ((tt = tt->rn_dupedkey) == NULL)
801 return (0);
802 }
803 if (tt->rn_mask == 0 || (saved_m = m = tt->rn_mklist) == NULL)
804 goto on1;
805 if (tt->rn_flags & RNF_NORMAL) {
806 if (m->rm_leaf != tt || m->rm_refs > 0) {
807 log(LOG_ERR, "rn_delete: inconsistent annotation\n");
808 return (0); /* dangling ref could cause disaster */
809 }
810 } else {
811 if (m->rm_mask != tt->rn_mask) {
812 log(LOG_ERR, "rn_delete: inconsistent annotation\n");
813 goto on1;
814 }
815 if (--m->rm_refs >= 0)
816 goto on1;
817 }
818 b = -1 - tt->rn_bit;
819 t = saved_tt->rn_parent;
820 if (b > t->rn_bit)
821 goto on1; /* Wasn't lifted at all */
822 do {
823 x = t;
824 t = t->rn_parent;
825 } while (b <= t->rn_bit && x != top);
826 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist)
827 if (m == saved_m) {
828 *mp = m->rm_mklist;
829 R_Free(m);
830 break;
831 }
832 if (m == NULL) {
833 log(LOG_ERR, "rn_delete: couldn't find our annotation\n");
834 if (tt->rn_flags & RNF_NORMAL)
835 return (0); /* Dangling ref to us */
836 }
837 on1:
838 /*
839 * Eliminate us from tree
840 */
841 if (tt->rn_flags & RNF_ROOT)
842 return (0);
843 #ifdef RN_DEBUG
844 /* Get us out of the creation list */
845 for (t = rn_clist; t && t->rn_ybro != tt; t = t->rn_ybro) {}
846 if (t) t->rn_ybro = tt->rn_ybro;
847 #endif
848 t = tt->rn_parent;
849 dupedkey = saved_tt->rn_dupedkey;
850 if (dupedkey) {
851 /*
852 * Here, tt is the deletion target and
853 * saved_tt is the head of the dupekey chain.
854 */
855 if (tt == saved_tt) {
856 /* remove from head of chain */
857 x = dupedkey; x->rn_parent = t;
858 if (t->rn_left == tt)
859 t->rn_left = x;
860 else
861 t->rn_right = x;
862 } else {
863 /* find node in front of tt on the chain */
864 for (x = p = saved_tt; p && p->rn_dupedkey != tt;)
865 p = p->rn_dupedkey;
866 if (p) {
867 p->rn_dupedkey = tt->rn_dupedkey;
868 if (tt->rn_dupedkey) /* parent */
869 tt->rn_dupedkey->rn_parent = p;
870 /* parent */
871 } else log(LOG_ERR, "rn_delete: couldn't find us\n");
872 }
873 t = tt + 1;
874 if (t->rn_flags & RNF_ACTIVE) {
875 #ifndef RN_DEBUG
876 *++x = *t;
877 p = t->rn_parent;
878 #else
879 b = t->rn_info;
880 *++x = *t;
881 t->rn_info = b;
882 p = t->rn_parent;
883 #endif
884 if (p->rn_left == t)
885 p->rn_left = x;
886 else
887 p->rn_right = x;
888 x->rn_left->rn_parent = x;
889 x->rn_right->rn_parent = x;
890 }
891 goto out;
892 }
893 if (t->rn_left == tt)
894 x = t->rn_right;
895 else
896 x = t->rn_left;
897 p = t->rn_parent;
898 if (p->rn_right == t)
899 p->rn_right = x;
900 else
901 p->rn_left = x;
902 x->rn_parent = p;
903 /*
904 * Demote routes attached to us.
905 */
906 if (t->rn_mklist) {
907 if (x->rn_bit >= 0) {
908 for (mp = &x->rn_mklist; (m = *mp);)
909 mp = &m->rm_mklist;
910 *mp = t->rn_mklist;
911 } else {
912 /* If there are any key,mask pairs in a sibling
913 duped-key chain, some subset will appear sorted
914 in the same order attached to our mklist */
915 for (m = t->rn_mklist; m && x; x = x->rn_dupedkey)
916 if (m == x->rn_mklist) {
917 struct radix_mask *mm = m->rm_mklist;
918 x->rn_mklist = 0;
919 if (--(m->rm_refs) < 0)
920 R_Free(m);
921 m = mm;
922 }
923 if (m)
924 log(LOG_ERR,
925 "rn_delete: Orphaned Mask %p at %p\n",
926 m, x);
927 }
928 }
929 /*
930 * We may be holding an active internal node in the tree.
931 */
932 x = tt + 1;
933 if (t != x) {
934 #ifndef RN_DEBUG
935 *t = *x;
936 #else
937 b = t->rn_info;
938 *t = *x;
939 t->rn_info = b;
940 #endif
941 t->rn_left->rn_parent = t;
942 t->rn_right->rn_parent = t;
943 p = x->rn_parent;
944 if (p->rn_left == x)
945 p->rn_left = t;
946 else
947 p->rn_right = t;
948 }
949 out:
950 tt->rn_flags &= ~RNF_ACTIVE;
951 tt[1].rn_flags &= ~RNF_ACTIVE;
952 return (tt);
953 }
954
955 /*
956 * This is the same as rn_walktree() except for the parameters and the
957 * exit.
958 */
959 int
960 rn_walktree_from(struct radix_head *h, void *a, void *m,
961 walktree_f_t *f, void *w)
962 {
963 int error;
964 struct radix_node *base, *next;
965 u_char *xa = (u_char *)a;
966 u_char *xm = (u_char *)m;
967 struct radix_node *rn, *last = NULL; /* shut up gcc */
968 int stopping = 0;
969 int lastb;
970
971 KASSERT(m != NULL, ("%s: mask needs to be specified", __func__));
972
973 /*
974 * rn_search_m is sort-of-open-coded here. We cannot use the
975 * function because we need to keep track of the last node seen.
976 */
977 /* printf("about to search\n"); */
978 for (rn = h->rnh_treetop; rn->rn_bit >= 0; ) {
979 last = rn;
980 /* printf("rn_bit %d, rn_bmask %x, xm[rn_offset] %x\n",
981 rn->rn_bit, rn->rn_bmask, xm[rn->rn_offset]); */
982 if (!(rn->rn_bmask & xm[rn->rn_offset])) {
983 break;
984 }
985 if (rn->rn_bmask & xa[rn->rn_offset]) {
986 rn = rn->rn_right;
987 } else {
988 rn = rn->rn_left;
989 }
990 }
991 /* printf("done searching\n"); */
992
993 /*
994 * Two cases: either we stepped off the end of our mask,
995 * in which case last == rn, or we reached a leaf, in which
996 * case we want to start from the leaf.
997 */
998 if (rn->rn_bit >= 0)
999 rn = last;
1000 lastb = last->rn_bit;
1001
1002 /* printf("rn %p, lastb %d\n", rn, lastb);*/
1003
1004 /*
1005 * This gets complicated because we may delete the node
1006 * while applying the function f to it, so we need to calculate
1007 * the successor node in advance.
1008 */
1009 while (rn->rn_bit >= 0)
1010 rn = rn->rn_left;
1011
1012 while (!stopping) {
1013 /* printf("node %p (%d)\n", rn, rn->rn_bit); */
1014 base = rn;
1015 /* If at right child go back up, otherwise, go right */
1016 while (rn->rn_parent->rn_right == rn
1017 && !(rn->rn_flags & RNF_ROOT)) {
1018 rn = rn->rn_parent;
1019
1020 /* if went up beyond last, stop */
1021 if (rn->rn_bit <= lastb) {
1022 stopping = 1;
1023 /* printf("up too far\n"); */
1024 /*
1025 * XXX we should jump to the 'Process leaves'
1026 * part, because the values of 'rn' and 'next'
1027 * we compute will not be used. Not a big deal
1028 * because this loop will terminate, but it is
1029 * inefficient and hard to understand!
1030 */
1031 }
1032 }
1033
1034 /*
1035 * At the top of the tree, no need to traverse the right
1036 * half, prevent the traversal of the entire tree in the
1037 * case of default route.
1038 */
1039 if (rn->rn_parent->rn_flags & RNF_ROOT)
1040 stopping = 1;
1041
1042 /* Find the next *leaf* since next node might vanish, too */
1043 for (rn = rn->rn_parent->rn_right; rn->rn_bit >= 0;)
1044 rn = rn->rn_left;
1045 next = rn;
1046 /* Process leaves */
1047 while ((rn = base) != NULL) {
1048 base = rn->rn_dupedkey;
1049 /* printf("leaf %p\n", rn); */
1050 if (!(rn->rn_flags & RNF_ROOT)
1051 && (error = (*f)(rn, w)))
1052 return (error);
1053 }
1054 rn = next;
1055
1056 if (rn->rn_flags & RNF_ROOT) {
1057 /* printf("root, stopping"); */
1058 stopping = 1;
1059 }
1060
1061 }
1062 return (0);
1063 }
1064
1065 int
1066 rn_walktree(struct radix_head *h, walktree_f_t *f, void *w)
1067 {
1068 int error;
1069 struct radix_node *base, *next;
1070 struct radix_node *rn = h->rnh_treetop;
1071 /*
1072 * This gets complicated because we may delete the node
1073 * while applying the function f to it, so we need to calculate
1074 * the successor node in advance.
1075 */
1076
1077 /* First time through node, go left */
1078 while (rn->rn_bit >= 0)
1079 rn = rn->rn_left;
1080 for (;;) {
1081 base = rn;
1082 /* If at right child go back up, otherwise, go right */
1083 while (rn->rn_parent->rn_right == rn
1084 && (rn->rn_flags & RNF_ROOT) == 0)
1085 rn = rn->rn_parent;
1086 /* Find the next *leaf* since next node might vanish, too */
1087 for (rn = rn->rn_parent->rn_right; rn->rn_bit >= 0;)
1088 rn = rn->rn_left;
1089 next = rn;
1090 /* Process leaves */
1091 while ((rn = base)) {
1092 base = rn->rn_dupedkey;
1093 if (!(rn->rn_flags & RNF_ROOT)
1094 && (error = (*f)(rn, w)))
1095 return (error);
1096 }
1097 rn = next;
1098 if (rn->rn_flags & RNF_ROOT)
1099 return (0);
1100 }
1101 /* NOTREACHED */
1102 }
1103
1104 /*
1105 * Initialize an empty tree. This has 3 nodes, which are passed
1106 * via base_nodes (in the order <left,root,right>) and are
1107 * marked RNF_ROOT so they cannot be freed.
1108 * The leaves have all-zero and all-one keys, with significant
1109 * bits starting at 'off'.
1110 */
1111 void
1112 rn_inithead_internal(struct radix_head *rh, struct radix_node *base_nodes, int off)
1113 {
1114 struct radix_node *t, *tt, *ttt;
1115
1116 t = rn_newpair(rn_zeros, off, base_nodes);
1117 ttt = base_nodes + 2;
1118 t->rn_right = ttt;
1119 t->rn_parent = t;
1120 tt = t->rn_left; /* ... which in turn is base_nodes */
1121 tt->rn_flags = t->rn_flags = RNF_ROOT | RNF_ACTIVE;
1122 tt->rn_bit = -1 - off;
1123 *ttt = *tt;
1124 ttt->rn_key = rn_ones;
1125
1126 rh->rnh_treetop = t;
1127 }
1128
1129 static void
1130 rn_detachhead_internal(struct radix_head *head)
1131 {
1132
1133 KASSERT((head != NULL),
1134 ("%s: head already freed", __func__));
1135
1136 /* Free <left,root,right> nodes. */
1137 R_Free(head);
1138 }
1139
1140 /* Functions used by 'struct radix_node_head' users */
1141
1142 int
1143 rn_inithead(void **head, int off)
1144 {
1145 struct radix_node_head *rnh;
1146 struct radix_mask_head *rmh;
1147
1148 rnh = *head;
1149 rmh = NULL;
1150
1151 if (*head != NULL)
1152 return (1);
1153
1154 R_Zalloc(rnh, struct radix_node_head *, sizeof (*rnh));
1155 R_Zalloc(rmh, struct radix_mask_head *, sizeof (*rmh));
1156 if (rnh == NULL || rmh == NULL) {
1157 if (rnh != NULL)
1158 R_Free(rnh);
1159 if (rmh != NULL)
1160 R_Free(rmh);
1161 return (0);
1162 }
1163
1164 /* Init trees */
1165 rn_inithead_internal(&rnh->rh, rnh->rnh_nodes, off);
1166 rn_inithead_internal(&rmh->head, rmh->mask_nodes, 0);
1167 *head = rnh;
1168 rnh->rh.rnh_masks = rmh;
1169
1170 /* Finally, set base callbacks */
1171 rnh->rnh_addaddr = rn_addroute;
1172 rnh->rnh_deladdr = rn_delete;
1173 rnh->rnh_matchaddr = rn_match;
1174 rnh->rnh_lookup = rn_lookup;
1175 rnh->rnh_walktree = rn_walktree;
1176 rnh->rnh_walktree_from = rn_walktree_from;
1177
1178 return (1);
1179 }
1180
1181 static int
1182 rn_freeentry(struct radix_node *rn, void *arg)
1183 {
1184 struct radix_head * const rnh = arg;
1185 struct radix_node *x;
1186
1187 x = (struct radix_node *)rn_delete(rn + 2, NULL, rnh);
1188 if (x != NULL)
1189 R_Free(x);
1190 return (0);
1191 }
1192
1193 int
1194 rn_detachhead(void **head)
1195 {
1196 struct radix_node_head *rnh;
1197
1198 KASSERT((head != NULL && *head != NULL),
1199 ("%s: head already freed", __func__));
1200
1201 rnh = (struct radix_node_head *)(*head);
1202
1203 rn_walktree(&rnh->rh.rnh_masks->head, rn_freeentry, rnh->rh.rnh_masks);
1204 rn_detachhead_internal(&rnh->rh.rnh_masks->head);
1205 rn_detachhead_internal(&rnh->rh);
1206
1207 *head = NULL;
1208
1209 return (1);
1210 }
1211
Cache object: e5bc75bc781bc43cecfd2e5644099fd2
|