nataraid(4): Add devstat support.
[dragonfly.git] / sys / net / radix.c
CommitLineData
984263bc
MD
1/*
2 * Copyright (c) 1988, 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 *
33 * @(#)radix.c 8.4 (Berkeley) 11/2/94
34 * $FreeBSD: src/sys/net/radix.c,v 1.20.2.3 2002/04/28 05:40:25 suz Exp $
4b1cf444 35 * $DragonFly: src/sys/net/radix.c,v 1.14 2006/12/22 23:44:54 swildner Exp $
984263bc
MD
36 */
37
38/*
39 * Routines to build and maintain radix trees for routing lookups.
40 */
984263bc
MD
41#include <sys/param.h>
42#ifdef _KERNEL
43#include <sys/systm.h>
44#include <sys/malloc.h>
984263bc 45#include <sys/domain.h>
ac0f13c5
SZ
46#include <sys/globaldata.h>
47#include <sys/thread.h>
984263bc
MD
48#else
49#include <stdlib.h>
50#endif
51#include <sys/syslog.h>
52#include <net/radix.h>
984263bc 53
2e9572df
JH
54/*
55 * The arguments to the radix functions are really counted byte arrays with
56 * the length in the first byte. struct sockaddr's fit this type structurally.
57 */
58#define clen(c) (*(u_char *)(c))
59
f23061d4 60static int rn_walktree_from(struct radix_node_head *h, char *a, char *m,
2e9572df
JH
61 walktree_f_t *f, void *w);
62static int rn_walktree(struct radix_node_head *, walktree_f_t *, void *);
63
984263bc 64static struct radix_node
2e9572df 65 *rn_insert(char *, struct radix_node_head *, boolean_t *,
f23061d4 66 struct radix_node [2]),
2e9572df
JH
67 *rn_newpair(char *, int, struct radix_node[2]),
68 *rn_search(const char *, struct radix_node *),
69 *rn_search_m(const char *, struct radix_node *, const char *);
984263bc 70
984263bc 71static struct radix_mask *rn_mkfreelist;
ac0f13c5 72static struct radix_node_head *mask_rnheads[MAXCPU];
984263bc 73
70cef922
MD
74static char rn_zeros[RN_MAXKEYLEN];
75static char rn_ones[RN_MAXKEYLEN] = RN_MAXKEYONES;
984263bc 76
2e9572df 77static int rn_lexobetter(char *m, char *n);
984263bc 78static struct radix_mask *
2e9572df 79 rn_new_radix_mask(struct radix_node *tt, struct radix_mask *nextmask);
5fe66e68 80static boolean_t
2e9572df
JH
81 rn_satisfies_leaf(char *trial, struct radix_node *leaf, int skip);
82
83static __inline struct radix_mask *
84MKGet(struct radix_mask **l)
85{
86 struct radix_mask *m;
87
88 if (*l != NULL) {
89 m = *l;
90 *l = m->rm_next;
91 } else {
5fe66e68 92 R_Malloc(m, struct radix_mask *, sizeof *m);
2e9572df
JH
93 }
94 return m;
95}
96
97static __inline void
98MKFree(struct radix_mask **l, struct radix_mask *m)
99{
100 m->rm_next = *l;
101 *l = m;
102}
984263bc
MD
103
104/*
105 * The data structure for the keys is a radix tree with one way
106 * branching removed. The index rn_bit at an internal node n represents a bit
107 * position to be tested. The tree is arranged so that all descendants
108 * of a node n have keys whose bits all agree up to position rn_bit - 1.
109 * (We say the index of n is rn_bit.)
110 *
111 * There is at least one descendant which has a one bit at position rn_bit,
112 * and at least one with a zero there.
113 *
114 * A route is determined by a pair of key and mask. We require that the
115 * bit-wise logical and of the key and mask to be the key.
116 * We define the index of a route to associated with the mask to be
117 * the first bit number in the mask where 0 occurs (with bit number 0
118 * representing the highest order bit).
119 *
120 * We say a mask is normal if every bit is 0, past the index of the mask.
121 * If a node n has a descendant (k, m) with index(m) == index(n) == rn_bit,
122 * and m is a normal mask, then the route applies to every descendant of n.
123 * If the index(m) < rn_bit, this implies the trailing last few bits of k
124 * before bit b are all 0, (and hence consequently true of every descendant
125 * of n), so the route applies to all descendants of the node as well.
126 *
127 * Similar logic shows that a non-normal mask m such that
128 * index(m) <= index(n) could potentially apply to many children of n.
129 * Thus, for each non-host route, we attach its mask to a list at an internal
130 * node as high in the tree as we can go.
131 *
132 * The present version of the code makes use of normal routes in short-
133 * circuiting an explict mask and compare operation when testing whether
134 * a key satisfies a normal route, and also in remembering the unique leaf
135 * that governs a subtree.
136 */
137
138static struct radix_node *
2e9572df 139rn_search(const char *v, struct radix_node *head)
984263bc 140{
82ed7fc2 141 struct radix_node *x;
984263bc 142
2e9572df
JH
143 x = head;
144 while (x->rn_bit >= 0) {
984263bc
MD
145 if (x->rn_bmask & v[x->rn_offset])
146 x = x->rn_right;
147 else
148 x = x->rn_left;
149 }
150 return (x);
151}
152
153static struct radix_node *
2e9572df 154rn_search_m(const char *v, struct radix_node *head, const char *m)
984263bc 155{
82ed7fc2 156 struct radix_node *x;
984263bc
MD
157
158 for (x = head; x->rn_bit >= 0;) {
159 if ((x->rn_bmask & m[x->rn_offset]) &&
160 (x->rn_bmask & v[x->rn_offset]))
161 x = x->rn_right;
162 else
163 x = x->rn_left;
164 }
165 return x;
166}
167
2e9572df
JH
168boolean_t
169rn_refines(char *m, char *n)
984263bc 170{
2e9572df
JH
171 char *lim, *lim2;
172 int longer = clen(n++) - clen(m++);
173 boolean_t masks_are_equal = TRUE;
984263bc 174
2e9572df 175 lim2 = lim = n + clen(n);
984263bc
MD
176 if (longer > 0)
177 lim -= longer;
178 while (n < lim) {
179 if (*n & ~(*m))
f23061d4 180 return FALSE;
984263bc 181 if (*n++ != *m++)
2e9572df 182 masks_are_equal = FALSE;
984263bc
MD
183 }
184 while (n < lim2)
185 if (*n++)
f23061d4 186 return FALSE;
984263bc
MD
187 if (masks_are_equal && (longer < 0))
188 for (lim2 = m - longer; m < lim2; )
189 if (*m++)
f23061d4 190 return TRUE;
984263bc
MD
191 return (!masks_are_equal);
192}
193
194struct radix_node *
2e9572df 195rn_lookup(char *key, char *mask, struct radix_node_head *head)
984263bc 196{
82ed7fc2 197 struct radix_node *x;
2e9572df 198 char *netmask = NULL;
984263bc 199
2e9572df 200 if (mask != NULL) {
b4628cf9
SZ
201 x = rn_addmask(mask, TRUE, head->rnh_treetop->rn_offset,
202 head->rnh_maskhead);
2e9572df
JH
203 if (x == NULL)
204 return (NULL);
984263bc
MD
205 netmask = x->rn_key;
206 }
2e9572df
JH
207 x = rn_match(key, head);
208 if (x != NULL && netmask != NULL) {
209 while (x != NULL && x->rn_mask != netmask)
984263bc
MD
210 x = x->rn_dupedkey;
211 }
212 return x;
213}
214
5fe66e68 215static boolean_t
2e9572df 216rn_satisfies_leaf(char *trial, struct radix_node *leaf, int skip)
984263bc 217{
82ed7fc2 218 char *cp = trial, *cp2 = leaf->rn_key, *cp3 = leaf->rn_mask;
984263bc 219 char *cplim;
2e9572df 220 int length = min(clen(cp), clen(cp2));
984263bc 221
2e9572df 222 if (cp3 == NULL)
984263bc
MD
223 cp3 = rn_ones;
224 else
2e9572df
JH
225 length = min(length, clen(cp3));
226 cplim = cp + length;
227 cp3 += skip;
228 cp2 += skip;
984263bc
MD
229 for (cp += skip; cp < cplim; cp++, cp2++, cp3++)
230 if ((*cp ^ *cp2) & *cp3)
5fe66e68
JH
231 return FALSE;
232 return TRUE;
984263bc
MD
233}
234
235struct radix_node *
2e9572df 236rn_match(char *key, struct radix_node_head *head)
984263bc 237{
2e9572df
JH
238 struct radix_node *t, *x;
239 char *cp = key, *cp2;
240 char *cplim;
241 struct radix_node *saved_t, *top = head->rnh_treetop;
242 int off = top->rn_offset, klen, matched_off;
82ed7fc2 243 int test, b, rn_bit;
984263bc 244
2e9572df 245 t = rn_search(key, top);
984263bc
MD
246 /*
247 * See if we match exactly as a host destination
248 * or at least learn how many bits match, for normal mask finesse.
249 *
250 * It doesn't hurt us to limit how many bytes to check
251 * to the length of the mask, since if it matches we had a genuine
252 * match and the leaf we have is the most specific one anyway;
253 * if it didn't match with a shorter length it would fail
254 * with a long one. This wins big for class B&C netmasks which
255 * are probably the most common case...
256 */
2e9572df
JH
257 if (t->rn_mask != NULL)
258 klen = clen(t->rn_mask);
259 else
260 klen = clen(key);
261 cp += off; cp2 = t->rn_key + off; cplim = key + klen;
984263bc
MD
262 for (; cp < cplim; cp++, cp2++)
263 if (*cp != *cp2)
264 goto on1;
265 /*
266 * This extra grot is in case we are explicitly asked
267 * to look up the default. Ugh!
268 *
269 * Never return the root node itself, it seems to cause a
270 * lot of confusion.
271 */
272 if (t->rn_flags & RNF_ROOT)
273 t = t->rn_dupedkey;
274 return t;
275on1:
276 test = (*cp ^ *cp2) & 0xff; /* find first bit that differs */
277 for (b = 7; (test >>= 1) > 0;)
278 b--;
2e9572df 279 matched_off = cp - key;
984263bc
MD
280 b += matched_off << 3;
281 rn_bit = -1 - b;
282 /*
283 * If there is a host route in a duped-key chain, it will be first.
284 */
2e9572df 285 if ((saved_t = t)->rn_mask == NULL)
984263bc 286 t = t->rn_dupedkey;
2e9572df 287 for (; t; t = t->rn_dupedkey) {
984263bc
MD
288 /*
289 * Even if we don't match exactly as a host,
290 * we may match if the leaf we wound up at is
291 * a route to a net.
292 */
293 if (t->rn_flags & RNF_NORMAL) {
294 if (rn_bit <= t->rn_bit)
295 return t;
2e9572df 296 } else if (rn_satisfies_leaf(key, t, matched_off))
984263bc 297 return t;
2e9572df 298 }
984263bc
MD
299 t = saved_t;
300 /* start searching up the tree */
301 do {
82ed7fc2 302 struct radix_mask *m;
2e9572df 303
984263bc 304 t = t->rn_parent;
984263bc
MD
305 /*
306 * If non-contiguous masks ever become important
307 * we can restore the masking and open coding of
308 * the search and satisfaction test and put the
309 * calculation of "off" back before the "do".
310 */
2e9572df
JH
311 m = t->rn_mklist;
312 while (m != NULL) {
984263bc
MD
313 if (m->rm_flags & RNF_NORMAL) {
314 if (rn_bit <= m->rm_bit)
315 return (m->rm_leaf);
316 } else {
317 off = min(t->rn_offset, matched_off);
2e9572df
JH
318 x = rn_search_m(key, t, m->rm_mask);
319 while (x != NULL && x->rn_mask != m->rm_mask)
984263bc 320 x = x->rn_dupedkey;
2e9572df 321 if (x && rn_satisfies_leaf(key, x, off))
984263bc
MD
322 return x;
323 }
2e9572df 324 m = m->rm_next;
984263bc
MD
325 }
326 } while (t != top);
2e9572df 327 return NULL;
984263bc
MD
328}
329
330#ifdef RN_DEBUG
5fe66e68
JH
331int rn_nodenum;
332struct radix_node *rn_clist;
333int rn_saveinfo;
334boolean_t rn_debug = TRUE;
984263bc
MD
335#endif
336
337static struct radix_node *
2e9572df 338rn_newpair(char *key, int indexbit, struct radix_node nodes[2])
984263bc 339{
5fe66e68 340 struct radix_node *leaf = &nodes[0], *interior = &nodes[1];
2e9572df
JH
341
342 interior->rn_bit = indexbit;
343 interior->rn_bmask = 0x80 >> (indexbit & 0x7);
344 interior->rn_offset = indexbit >> 3;
345 interior->rn_left = leaf;
346 interior->rn_mklist = NULL;
347
348 leaf->rn_bit = -1;
349 leaf->rn_key = key;
350 leaf->rn_parent = interior;
351 leaf->rn_flags = interior->rn_flags = RNF_ACTIVE;
352 leaf->rn_mklist = NULL;
353
984263bc 354#ifdef RN_DEBUG
2e9572df
JH
355 leaf->rn_info = rn_nodenum++;
356 interior->rn_info = rn_nodenum++;
357 leaf->rn_twin = interior;
358 leaf->rn_ybro = rn_clist;
359 rn_clist = leaf;
984263bc 360#endif
2e9572df 361 return interior;
984263bc
MD
362}
363
364static struct radix_node *
f23061d4
JH
365rn_insert(char *key, struct radix_node_head *head, boolean_t *dupentry,
366 struct radix_node nodes[2])
984263bc 367{
984263bc 368 struct radix_node *top = head->rnh_treetop;
2e9572df
JH
369 int head_off = top->rn_offset, klen = clen(key);
370 struct radix_node *t = rn_search(key, top);
371 char *cp = key + head_off;
82ed7fc2 372 int b;
984263bc 373 struct radix_node *tt;
5fe66e68 374
f23061d4 375 /*
2e9572df 376 * Find first bit at which the key and t->rn_key differ
984263bc
MD
377 */
378 {
2e9572df 379 char *cp2 = t->rn_key + head_off;
82ed7fc2 380 int cmp_res;
2e9572df 381 char *cplim = key + klen;
984263bc
MD
382
383 while (cp < cplim)
384 if (*cp2++ != *cp++)
385 goto on1;
2e9572df 386 *dupentry = TRUE;
984263bc
MD
387 return t;
388on1:
2e9572df 389 *dupentry = FALSE;
984263bc 390 cmp_res = (cp[-1] ^ cp2[-1]) & 0xff;
2e9572df 391 for (b = (cp - key) << 3; cmp_res; b--)
984263bc
MD
392 cmp_res >>= 1;
393 }
394 {
82ed7fc2 395 struct radix_node *p, *x = top;
2e9572df
JH
396
397 cp = key;
984263bc
MD
398 do {
399 p = x;
400 if (cp[x->rn_offset] & x->rn_bmask)
401 x = x->rn_right;
402 else
403 x = x->rn_left;
404 } while (b > (unsigned) x->rn_bit);
405 /* x->rn_bit < b && x->rn_bit >= 0 */
406#ifdef RN_DEBUG
407 if (rn_debug)
408 log(LOG_DEBUG, "rn_insert: Going In:\n"), traverse(p);
409#endif
2e9572df 410 t = rn_newpair(key, b, nodes);
984263bc
MD
411 tt = t->rn_left;
412 if ((cp[p->rn_offset] & p->rn_bmask) == 0)
413 p->rn_left = t;
414 else
415 p->rn_right = t;
416 x->rn_parent = t;
417 t->rn_parent = p; /* frees x, p as temp vars below */
418 if ((cp[t->rn_offset] & t->rn_bmask) == 0) {
419 t->rn_right = x;
420 } else {
421 t->rn_right = tt;
422 t->rn_left = x;
423 }
424#ifdef RN_DEBUG
425 if (rn_debug)
426 log(LOG_DEBUG, "rn_insert: Coming Out:\n"), traverse(p);
427#endif
428 }
429 return (tt);
430}
431
432struct radix_node *
b4628cf9
SZ
433rn_addmask(char *netmask, boolean_t search, int skip,
434 struct radix_node_head *mask_rnh)
984263bc 435{
2e9572df
JH
436 struct radix_node *x, *saved_x;
437 char *cp, *cplim;
ef87f48d
JH
438 int b = 0, mlen, m0, j;
439 boolean_t maskduplicated, isnormal;
984263bc 440 static int last_zeroed = 0;
ecdefdda 441 char *addmask_key;
984263bc 442
70cef922
MD
443 if ((mlen = clen(netmask)) > RN_MAXKEYLEN)
444 mlen = RN_MAXKEYLEN;
984263bc
MD
445 if (skip == 0)
446 skip = 1;
447 if (mlen <= skip)
ac0f13c5 448 return (mask_rnh->rnh_nodes);
70cef922 449 R_Malloc(addmask_key, char *, RN_MAXKEYLEN);
ecdefdda
MD
450 if (addmask_key == NULL)
451 return NULL;
984263bc 452 if (skip > 1)
2e9572df 453 bcopy(rn_ones + 1, addmask_key + 1, skip - 1);
984263bc 454 if ((m0 = mlen) > skip)
2e9572df 455 bcopy(netmask + skip, addmask_key + skip, mlen - skip);
984263bc
MD
456 /*
457 * Trim trailing zeroes.
458 */
459 for (cp = addmask_key + mlen; (cp > addmask_key) && cp[-1] == 0;)
460 cp--;
461 mlen = cp - addmask_key;
462 if (mlen <= skip) {
463 if (m0 >= last_zeroed)
464 last_zeroed = mlen;
ecdefdda 465 Free(addmask_key);
ac0f13c5 466 return (mask_rnh->rnh_nodes);
984263bc
MD
467 }
468 if (m0 < last_zeroed)
2e9572df 469 bzero(addmask_key + m0, last_zeroed - m0);
984263bc 470 *addmask_key = last_zeroed = mlen;
ac0f13c5 471 x = rn_search(addmask_key, mask_rnh->rnh_treetop);
70cef922
MD
472 if (x->rn_key == NULL) {
473 kprintf("WARNING: radix_node->rn_key is NULL rn=%p\n", x);
474 print_backtrace(-1);
2e9572df 475 x = NULL;
70cef922
MD
476 } else if (bcmp(addmask_key, x->rn_key, mlen) != 0) {
477 x = NULL;
478 }
2e9572df 479 if (x != NULL || search)
ecdefdda 480 goto out;
70cef922 481 R_Malloc(x, struct radix_node *, RN_MAXKEYLEN + 2 * (sizeof *x));
2e9572df 482 if ((saved_x = x) == NULL)
ecdefdda 483 goto out;
70cef922 484 bzero(x, RN_MAXKEYLEN + 2 * (sizeof *x));
2e9572df
JH
485 netmask = cp = (char *)(x + 2);
486 bcopy(addmask_key, cp, mlen);
ac0f13c5 487 x = rn_insert(cp, mask_rnh, &maskduplicated, x);
984263bc
MD
488 if (maskduplicated) {
489 log(LOG_ERR, "rn_addmask: mask impossibly already in tree");
490 Free(saved_x);
ecdefdda 491 goto out;
984263bc
MD
492 }
493 /*
494 * Calculate index of mask, and check for normalcy.
495 */
ef87f48d
JH
496 isnormal = TRUE;
497 cplim = netmask + mlen;
2e9572df 498 for (cp = netmask + skip; cp < cplim && clen(cp) == 0xff;)
984263bc
MD
499 cp++;
500 if (cp != cplim) {
f23061d4
JH
501 static const char normal_chars[] = {
502 0, 0x80, 0xc0, 0xe0, 0xf0, 0xf8, 0xfc, 0xfe, -1
503 };
504
984263bc
MD
505 for (j = 0x80; (j & *cp) != 0; j >>= 1)
506 b++;
507 if (*cp != normal_chars[b] || cp != (cplim - 1))
ef87f48d 508 isnormal = FALSE;
984263bc
MD
509 }
510 b += (cp - netmask) << 3;
511 x->rn_bit = -1 - b;
512 if (isnormal)
513 x->rn_flags |= RNF_NORMAL;
ecdefdda
MD
514out:
515 Free(addmask_key);
984263bc
MD
516 return (x);
517}
518
3bf25ce1
SW
519/* XXX: arbitrary ordering for non-contiguous masks */
520static boolean_t
2e9572df 521rn_lexobetter(char *mp, char *np)
984263bc 522{
2e9572df 523 char *lim;
984263bc 524
2e9572df
JH
525 if ((unsigned) *mp > (unsigned) *np)
526 return TRUE;/* not really, but need to check longer one first */
984263bc 527 if (*mp == *np)
2e9572df 528 for (lim = mp + clen(mp); mp < lim;)
984263bc 529 if (*mp++ > *np++)
2e9572df
JH
530 return TRUE;
531 return FALSE;
984263bc
MD
532}
533
534static struct radix_mask *
2e9572df 535rn_new_radix_mask(struct radix_node *tt, struct radix_mask *nextmask)
984263bc 536{
82ed7fc2 537 struct radix_mask *m;
984263bc 538
2e9572df
JH
539 m = MKGet(&rn_mkfreelist);
540 if (m == NULL) {
984263bc 541 log(LOG_ERR, "Mask for route not entered\n");
2e9572df 542 return (NULL);
984263bc 543 }
2e9572df 544 bzero(m, sizeof *m);
984263bc
MD
545 m->rm_bit = tt->rn_bit;
546 m->rm_flags = tt->rn_flags;
547 if (tt->rn_flags & RNF_NORMAL)
548 m->rm_leaf = tt;
549 else
550 m->rm_mask = tt->rn_mask;
2e9572df 551 m->rm_next = nextmask;
984263bc
MD
552 tt->rn_mklist = m;
553 return m;
554}
555
556struct radix_node *
2e9572df
JH
557rn_addroute(char *key, char *netmask, struct radix_node_head *head,
558 struct radix_node treenodes[2])
984263bc 559{
2e9572df 560 struct radix_node *t, *x = NULL, *tt;
984263bc
MD
561 struct radix_node *saved_tt, *top = head->rnh_treetop;
562 short b = 0, b_leaf = 0;
2e9572df
JH
563 boolean_t keyduplicated;
564 char *mmask;
984263bc
MD
565 struct radix_mask *m, **mp;
566
567 /*
568 * In dealing with non-contiguous masks, there may be
569 * many different routes which have the same mask.
570 * We will find it useful to have a unique pointer to
571 * the mask to speed avoiding duplicate references at
572 * nodes and possibly save time in calculating indices.
573 */
2e9572df 574 if (netmask != NULL) {
b4628cf9
SZ
575 if ((x = rn_addmask(netmask, FALSE, top->rn_offset,
576 head->rnh_maskhead)) == NULL)
2e9572df 577 return (NULL);
984263bc
MD
578 b_leaf = x->rn_bit;
579 b = -1 - x->rn_bit;
580 netmask = x->rn_key;
581 }
582 /*
583 * Deal with duplicated keys: attach node to previous instance
584 */
2e9572df 585 saved_tt = tt = rn_insert(key, head, &keyduplicated, treenodes);
984263bc
MD
586 if (keyduplicated) {
587 for (t = tt; tt; t = tt, tt = tt->rn_dupedkey) {
588 if (tt->rn_mask == netmask)
2e9572df
JH
589 return (NULL);
590 if (netmask == NULL ||
984263bc
MD
591 (tt->rn_mask &&
592 ((b_leaf < tt->rn_bit) /* index(netmask) > node */
593 || rn_refines(netmask, tt->rn_mask)
594 || rn_lexobetter(netmask, tt->rn_mask))))
595 break;
596 }
597 /*
598 * If the mask is not duplicated, we wouldn't
599 * find it among possible duplicate key entries
600 * anyway, so the above test doesn't hurt.
601 *
602 * We sort the masks for a duplicated key the same way as
603 * in a masklist -- most specific to least specific.
604 * This may require the unfortunate nuisance of relocating
605 * the head of the list.
606 */
607 if (tt == saved_tt) {
608 struct radix_node *xx = x;
609 /* link in at head of list */
610 (tt = treenodes)->rn_dupedkey = t;
611 tt->rn_flags = t->rn_flags;
612 tt->rn_parent = x = t->rn_parent;
f23061d4 613 t->rn_parent = tt; /* parent */
984263bc
MD
614 if (x->rn_left == t)
615 x->rn_left = tt;
616 else
617 x->rn_right = tt;
618 saved_tt = tt; x = xx;
619 } else {
620 (tt = treenodes)->rn_dupedkey = t->rn_dupedkey;
621 t->rn_dupedkey = tt;
622 tt->rn_parent = t; /* parent */
2e9572df 623 if (tt->rn_dupedkey != NULL) /* parent */
984263bc
MD
624 tt->rn_dupedkey->rn_parent = tt; /* parent */
625 }
626#ifdef RN_DEBUG
627 t=tt+1; tt->rn_info = rn_nodenum++; t->rn_info = rn_nodenum++;
628 tt->rn_twin = t; tt->rn_ybro = rn_clist; rn_clist = tt;
629#endif
2e9572df 630 tt->rn_key = key;
984263bc
MD
631 tt->rn_bit = -1;
632 tt->rn_flags = RNF_ACTIVE;
633 }
634 /*
635 * Put mask in tree.
636 */
2e9572df 637 if (netmask != NULL) {
984263bc
MD
638 tt->rn_mask = netmask;
639 tt->rn_bit = x->rn_bit;
640 tt->rn_flags |= x->rn_flags & RNF_NORMAL;
641 }
642 t = saved_tt->rn_parent;
643 if (keyduplicated)
644 goto on2;
645 b_leaf = -1 - t->rn_bit;
646 if (t->rn_right == saved_tt)
647 x = t->rn_left;
648 else
649 x = t->rn_right;
650 /* Promote general routes from below */
651 if (x->rn_bit < 0) {
2e9572df
JH
652 mp = &t->rn_mklist;
653 while (x != NULL) {
654 if (x->rn_mask != NULL &&
655 x->rn_bit >= b_leaf &&
656 x->rn_mklist == NULL) {
657 *mp = m = rn_new_radix_mask(x, NULL);
658 if (m != NULL)
659 mp = &m->rm_next;
660 }
661 x = x->rn_dupedkey;
984263bc 662 }
2e9572df 663 } else if (x->rn_mklist != NULL) {
984263bc
MD
664 /*
665 * Skip over masks whose index is > that of new node
666 */
2e9572df 667 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_next)
984263bc
MD
668 if (m->rm_bit >= b_leaf)
669 break;
2e9572df
JH
670 t->rn_mklist = m;
671 *mp = NULL;
984263bc
MD
672 }
673on2:
674 /* Add new route to highest possible ancestor's list */
2e9572df 675 if ((netmask == NULL) || (b > t->rn_bit ))
984263bc
MD
676 return tt; /* can't lift at all */
677 b_leaf = tt->rn_bit;
678 do {
679 x = t;
680 t = t->rn_parent;
681 } while (b <= t->rn_bit && x != top);
682 /*
683 * Search through routes associated with node to
684 * insert new route according to index.
685 * Need same criteria as when sorting dupedkeys to avoid
686 * double loop on deletion.
687 */
2e9572df 688 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_next) {
984263bc
MD
689 if (m->rm_bit < b_leaf)
690 continue;
691 if (m->rm_bit > b_leaf)
692 break;
693 if (m->rm_flags & RNF_NORMAL) {
694 mmask = m->rm_leaf->rn_mask;
695 if (tt->rn_flags & RNF_NORMAL) {
696 log(LOG_ERR,
697 "Non-unique normal route, mask not entered\n");
698 return tt;
699 }
700 } else
701 mmask = m->rm_mask;
702 if (mmask == netmask) {
703 m->rm_refs++;
704 tt->rn_mklist = m;
705 return tt;
706 }
2e9572df 707 if (rn_refines(netmask, mmask) || rn_lexobetter(netmask, mmask))
984263bc
MD
708 break;
709 }
710 *mp = rn_new_radix_mask(tt, *mp);
711 return tt;
712}
713
714struct radix_node *
2e9572df 715rn_delete(char *key, char *netmask, struct radix_node_head *head)
984263bc 716{
82ed7fc2 717 struct radix_node *t, *p, *x, *tt;
984263bc
MD
718 struct radix_mask *m, *saved_m, **mp;
719 struct radix_node *dupedkey, *saved_tt, *top;
2e9572df 720 int b, head_off, klen;
984263bc 721
984263bc 722 x = head->rnh_treetop;
2e9572df 723 tt = rn_search(key, x);
984263bc 724 head_off = x->rn_offset;
2e9572df 725 klen = clen(key);
984263bc
MD
726 saved_tt = tt;
727 top = x;
2e9572df
JH
728 if (tt == NULL ||
729 bcmp(key + head_off, tt->rn_key + head_off, klen - head_off))
730 return (NULL);
984263bc
MD
731 /*
732 * Delete our route from mask lists.
733 */
2e9572df 734 if (netmask != NULL) {
b4628cf9
SZ
735 if ((x = rn_addmask(netmask, TRUE, head_off,
736 head->rnh_maskhead)) == NULL)
2e9572df 737 return (NULL);
984263bc
MD
738 netmask = x->rn_key;
739 while (tt->rn_mask != netmask)
2e9572df
JH
740 if ((tt = tt->rn_dupedkey) == NULL)
741 return (NULL);
984263bc 742 }
2e9572df 743 if (tt->rn_mask == NULL || (saved_m = m = tt->rn_mklist) == NULL)
984263bc
MD
744 goto on1;
745 if (tt->rn_flags & RNF_NORMAL) {
746 if (m->rm_leaf != tt || m->rm_refs > 0) {
747 log(LOG_ERR, "rn_delete: inconsistent annotation\n");
2e9572df 748 return (NULL); /* dangling ref could cause disaster */
984263bc
MD
749 }
750 } else {
751 if (m->rm_mask != tt->rn_mask) {
752 log(LOG_ERR, "rn_delete: inconsistent annotation\n");
753 goto on1;
754 }
755 if (--m->rm_refs >= 0)
756 goto on1;
757 }
758 b = -1 - tt->rn_bit;
759 t = saved_tt->rn_parent;
760 if (b > t->rn_bit)
761 goto on1; /* Wasn't lifted at all */
762 do {
763 x = t;
764 t = t->rn_parent;
765 } while (b <= t->rn_bit && x != top);
2e9572df 766 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_next)
984263bc 767 if (m == saved_m) {
2e9572df
JH
768 *mp = m->rm_next;
769 MKFree(&rn_mkfreelist, m);
984263bc
MD
770 break;
771 }
2e9572df 772 if (m == NULL) {
984263bc
MD
773 log(LOG_ERR, "rn_delete: couldn't find our annotation\n");
774 if (tt->rn_flags & RNF_NORMAL)
2e9572df 775 return (NULL); /* Dangling ref to us */
984263bc
MD
776 }
777on1:
778 /*
779 * Eliminate us from tree
780 */
781 if (tt->rn_flags & RNF_ROOT)
2e9572df 782 return (NULL);
984263bc
MD
783#ifdef RN_DEBUG
784 /* Get us out of the creation list */
785 for (t = rn_clist; t && t->rn_ybro != tt; t = t->rn_ybro) {}
786 if (t) t->rn_ybro = tt->rn_ybro;
787#endif
788 t = tt->rn_parent;
789 dupedkey = saved_tt->rn_dupedkey;
2e9572df 790 if (dupedkey != NULL) {
984263bc
MD
791 /*
792 * at this point, tt is the deletion target and saved_tt
793 * is the head of the dupekey chain
794 */
795 if (tt == saved_tt) {
796 /* remove from head of chain */
797 x = dupedkey; x->rn_parent = t;
798 if (t->rn_left == tt)
799 t->rn_left = x;
800 else
801 t->rn_right = x;
802 } else {
803 /* find node in front of tt on the chain */
804 for (x = p = saved_tt; p && p->rn_dupedkey != tt;)
805 p = p->rn_dupedkey;
806 if (p) {
807 p->rn_dupedkey = tt->rn_dupedkey;
808 if (tt->rn_dupedkey) /* parent */
809 tt->rn_dupedkey->rn_parent = p;
810 /* parent */
811 } else log(LOG_ERR, "rn_delete: couldn't find us\n");
812 }
813 t = tt + 1;
814 if (t->rn_flags & RNF_ACTIVE) {
815#ifndef RN_DEBUG
816 *++x = *t;
817 p = t->rn_parent;
818#else
819 b = t->rn_info;
820 *++x = *t;
821 t->rn_info = b;
822 p = t->rn_parent;
823#endif
824 if (p->rn_left == t)
825 p->rn_left = x;
826 else
827 p->rn_right = x;
828 x->rn_left->rn_parent = x;
829 x->rn_right->rn_parent = x;
830 }
831 goto out;
832 }
833 if (t->rn_left == tt)
834 x = t->rn_right;
835 else
836 x = t->rn_left;
837 p = t->rn_parent;
838 if (p->rn_right == t)
839 p->rn_right = x;
840 else
841 p->rn_left = x;
842 x->rn_parent = p;
843 /*
844 * Demote routes attached to us.
845 */
2e9572df 846 if (t->rn_mklist != NULL) {
984263bc
MD
847 if (x->rn_bit >= 0) {
848 for (mp = &x->rn_mklist; (m = *mp);)
2e9572df 849 mp = &m->rm_next;
984263bc
MD
850 *mp = t->rn_mklist;
851 } else {
f23061d4
JH
852 /*
853 * If there are any (key, mask) pairs in a sibling
854 * duped-key chain, some subset will appear sorted
855 * in the same order attached to our mklist.
856 */
984263bc
MD
857 for (m = t->rn_mklist; m && x; x = x->rn_dupedkey)
858 if (m == x->rn_mklist) {
2e9572df
JH
859 struct radix_mask *mm = m->rm_next;
860
861 x->rn_mklist = NULL;
984263bc 862 if (--(m->rm_refs) < 0)
2e9572df 863 MKFree(&rn_mkfreelist, m);
984263bc
MD
864 m = mm;
865 }
866 if (m)
867 log(LOG_ERR,
868 "rn_delete: Orphaned Mask %p at %p\n",
869 (void *)m, (void *)x);
870 }
871 }
872 /*
873 * We may be holding an active internal node in the tree.
874 */
875 x = tt + 1;
876 if (t != x) {
877#ifndef RN_DEBUG
878 *t = *x;
879#else
880 b = t->rn_info;
881 *t = *x;
882 t->rn_info = b;
883#endif
884 t->rn_left->rn_parent = t;
885 t->rn_right->rn_parent = t;
886 p = x->rn_parent;
887 if (p->rn_left == x)
888 p->rn_left = t;
889 else
890 p->rn_right = t;
891 }
892out:
893 tt->rn_flags &= ~RNF_ACTIVE;
894 tt[1].rn_flags &= ~RNF_ACTIVE;
895 return (tt);
896}
897
898/*
899 * This is the same as rn_walktree() except for the parameters and the
900 * exit.
901 */
902static int
f23061d4 903rn_walktree_from(struct radix_node_head *h, char *xa, char *xm,
2e9572df 904 walktree_f_t *f, void *w)
984263bc 905{
984263bc 906 struct radix_node *base, *next;
2e9572df 907 struct radix_node *rn, *last = NULL /* shut up gcc */;
f23061d4 908 boolean_t stopping = FALSE;
2e9572df 909 int lastb, error;
984263bc
MD
910
911 /*
912 * rn_search_m is sort-of-open-coded here.
913 */
4b1cf444 914 /* kprintf("about to search\n"); */
984263bc
MD
915 for (rn = h->rnh_treetop; rn->rn_bit >= 0; ) {
916 last = rn;
4b1cf444 917 /* kprintf("rn_bit %d, rn_bmask %x, xm[rn_offset] %x\n",
984263bc
MD
918 rn->rn_bit, rn->rn_bmask, xm[rn->rn_offset]); */
919 if (!(rn->rn_bmask & xm[rn->rn_offset])) {
920 break;
921 }
922 if (rn->rn_bmask & xa[rn->rn_offset]) {
923 rn = rn->rn_right;
924 } else {
925 rn = rn->rn_left;
926 }
927 }
4b1cf444 928 /* kprintf("done searching\n"); */
984263bc
MD
929
930 /*
931 * Two cases: either we stepped off the end of our mask,
932 * in which case last == rn, or we reached a leaf, in which
933 * case we want to start from the last node we looked at.
934 * Either way, last is the node we want to start from.
935 */
936 rn = last;
937 lastb = rn->rn_bit;
938
4b1cf444 939 /* kprintf("rn %p, lastb %d\n", rn, lastb);*/
984263bc
MD
940
941 /*
942 * This gets complicated because we may delete the node
943 * while applying the function f to it, so we need to calculate
944 * the successor node in advance.
945 */
946 while (rn->rn_bit >= 0)
947 rn = rn->rn_left;
948
949 while (!stopping) {
4b1cf444 950 /* kprintf("node %p (%d)\n", rn, rn->rn_bit); */
984263bc
MD
951 base = rn;
952 /* If at right child go back up, otherwise, go right */
2e9572df
JH
953 while (rn->rn_parent->rn_right == rn &&
954 !(rn->rn_flags & RNF_ROOT)) {
984263bc
MD
955 rn = rn->rn_parent;
956
957 /* if went up beyond last, stop */
958 if (rn->rn_bit < lastb) {
f23061d4 959 stopping = TRUE;
4b1cf444 960 /* kprintf("up too far\n"); */
984263bc
MD
961 }
962 }
963
964 /* Find the next *leaf* since next node might vanish, too */
965 for (rn = rn->rn_parent->rn_right; rn->rn_bit >= 0;)
966 rn = rn->rn_left;
967 next = rn;
968 /* Process leaves */
2e9572df 969 while ((rn = base) != NULL) {
984263bc 970 base = rn->rn_dupedkey;
4b1cf444 971 /* kprintf("leaf %p\n", rn); */
2e9572df 972 if (!(rn->rn_flags & RNF_ROOT) && (error = (*f)(rn, w)))
984263bc
MD
973 return (error);
974 }
975 rn = next;
976
977 if (rn->rn_flags & RNF_ROOT) {
4b1cf444 978 /* kprintf("root, stopping"); */
f23061d4 979 stopping = TRUE;
984263bc
MD
980 }
981
982 }
983 return 0;
984}
985
986static int
2e9572df 987rn_walktree(struct radix_node_head *h, walktree_f_t *f, void *w)
984263bc 988{
984263bc 989 struct radix_node *base, *next;
82ed7fc2 990 struct radix_node *rn = h->rnh_treetop;
2e9572df
JH
991 int error;
992
984263bc
MD
993 /*
994 * This gets complicated because we may delete the node
995 * while applying the function f to it, so we need to calculate
996 * the successor node in advance.
997 */
998 /* First time through node, go left */
999 while (rn->rn_bit >= 0)
1000 rn = rn->rn_left;
1001 for (;;) {
1002 base = rn;
1003 /* If at right child go back up, otherwise, go right */
2e9572df
JH
1004 while (rn->rn_parent->rn_right == rn &&
1005 !(rn->rn_flags & RNF_ROOT))
984263bc
MD
1006 rn = rn->rn_parent;
1007 /* Find the next *leaf* since next node might vanish, too */
1008 for (rn = rn->rn_parent->rn_right; rn->rn_bit >= 0;)
1009 rn = rn->rn_left;
1010 next = rn;
1011 /* Process leaves */
1012 while ((rn = base)) {
1013 base = rn->rn_dupedkey;
2e9572df 1014 if (!(rn->rn_flags & RNF_ROOT) && (error = (*f)(rn, w)))
984263bc
MD
1015 return (error);
1016 }
1017 rn = next;
1018 if (rn->rn_flags & RNF_ROOT)
1019 return (0);
1020 }
1021 /* NOTREACHED */
1022}
1023
1024int
b4628cf9 1025rn_inithead(void **head, struct radix_node_head *maskhead, int off)
984263bc 1026{
82ed7fc2 1027 struct radix_node_head *rnh;
f23061d4 1028 struct radix_node *root, *left, *right;
2e9572df 1029
5fe66e68 1030 if (*head != NULL) /* already initialized */
984263bc 1031 return (1);
5fe66e68
JH
1032
1033 R_Malloc(rnh, struct radix_node_head *, sizeof *rnh);
2e9572df 1034 if (rnh == NULL)
984263bc 1035 return (0);
5fe66e68 1036 bzero(rnh, sizeof *rnh);
984263bc 1037 *head = rnh;
2e9572df 1038
f23061d4 1039 root = rn_newpair(rn_zeros, off, rnh->rnh_nodes);
5fe66e68 1040 right = &rnh->rnh_nodes[2];
f23061d4
JH
1041 root->rn_parent = root;
1042 root->rn_flags = RNF_ROOT | RNF_ACTIVE;
1043 root->rn_right = right;
2e9572df 1044
f23061d4 1045 left = root->rn_left;
2e9572df 1046 left->rn_bit = -1 - off;
f23061d4 1047 left->rn_flags = RNF_ROOT | RNF_ACTIVE;
2e9572df
JH
1048
1049 *right = *left;
1050 right->rn_key = rn_ones;
1051
f23061d4 1052 rnh->rnh_treetop = root;
b4628cf9 1053 rnh->rnh_maskhead = maskhead;
0c3c561c 1054
984263bc
MD
1055 rnh->rnh_addaddr = rn_addroute;
1056 rnh->rnh_deladdr = rn_delete;
1057 rnh->rnh_matchaddr = rn_match;
1058 rnh->rnh_lookup = rn_lookup;
1059 rnh->rnh_walktree = rn_walktree;
1060 rnh->rnh_walktree_from = rn_walktree_from;
2e9572df 1061
984263bc
MD
1062 return (1);
1063}
1064
1065void
3bf25ce1 1066rn_init(void)
984263bc 1067{
ac0f13c5 1068 int cpu;
984263bc
MD
1069#ifdef _KERNEL
1070 struct domain *dom;
1071
70cef922
MD
1072 SLIST_FOREACH(dom, &domains, dom_next) {
1073 if (dom->dom_maxrtkey > RN_MAXKEYLEN) {
1074 panic("domain %s maxkey too big %d/%d\n",
1075 dom->dom_name, dom->dom_maxrtkey, RN_MAXKEYLEN);
1076 }
984263bc 1077 }
70cef922 1078#endif
ac0f13c5 1079 for (cpu = 0; cpu < ncpus; ++cpu) {
b4628cf9 1080 if (rn_inithead((void **)&mask_rnheads[cpu], NULL, 0) == 0)
ac0f13c5
SZ
1081 panic("rn_init 2");
1082 }
984263bc 1083}
b4628cf9
SZ
1084
1085struct radix_node_head *
1086rn_cpumaskhead(int cpu)
1087{
1088 KKASSERT(mask_rnheads[cpu] != NULL);
1089 return mask_rnheads[cpu];
1090}