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 $ 35 * $DragonFly: src/sys/net/radix.c,v 1.14 2006/12/22 23:44:54 swildner Exp $ 36 */ 37 38 /* 39 * Routines to build and maintain radix trees for routing lookups. 40 */ 41 #include <sys/param.h> 42 #ifdef _KERNEL 43 #include <sys/systm.h> 44 #include <sys/malloc.h> 45 #include <sys/domain.h> 46 #include <sys/globaldata.h> 47 #include <sys/thread.h> 48 #else 49 #include <stdlib.h> 50 #endif 51 #include <sys/syslog.h> 52 #include <net/radix.h> 53 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 60 static int rn_walktree_from(struct radix_node_head *h, char *a, char *m, 61 walktree_f_t *f, void *w); 62 static int rn_walktree(struct radix_node_head *, walktree_f_t *, void *); 63 64 static struct radix_node 65 *rn_insert(char *, struct radix_node_head *, boolean_t *, 66 struct radix_node [2]), 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 *); 70 71 static struct radix_mask *rn_mkfreelist; 72 static struct radix_node_head *mask_rnheads[MAXCPU]; 73 74 static int max_keylen; 75 static char *rn_zeros, *rn_ones; 76 77 static int rn_lexobetter(char *m, char *n); 78 static struct radix_mask * 79 rn_new_radix_mask(struct radix_node *tt, struct radix_mask *nextmask); 80 static boolean_t 81 rn_satisfies_leaf(char *trial, struct radix_node *leaf, int skip); 82 83 static __inline struct radix_mask * 84 MKGet(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 { 92 R_Malloc(m, struct radix_mask *, sizeof *m); 93 } 94 return m; 95 } 96 97 static __inline void 98 MKFree(struct radix_mask **l, struct radix_mask *m) 99 { 100 m->rm_next = *l; 101 *l = m; 102 } 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 */ 137 138 static struct radix_node * 139 rn_search(const char *v, struct radix_node *head) 140 { 141 struct radix_node *x; 142 143 x = head; 144 while (x->rn_bit >= 0) { 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 } 152 153 static struct radix_node * 154 rn_search_m(const char *v, struct radix_node *head, const char *m) 155 { 156 struct radix_node *x; 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 } 167 168 boolean_t 169 rn_refines(char *m, char *n) 170 { 171 char *lim, *lim2; 172 int longer = clen(n++) - clen(m++); 173 boolean_t masks_are_equal = TRUE; 174 175 lim2 = lim = n + clen(n); 176 if (longer > 0) 177 lim -= longer; 178 while (n < lim) { 179 if (*n & ~(*m)) 180 return FALSE; 181 if (*n++ != *m++) 182 masks_are_equal = FALSE; 183 } 184 while (n < lim2) 185 if (*n++) 186 return FALSE; 187 if (masks_are_equal && (longer < 0)) 188 for (lim2 = m - longer; m < lim2; ) 189 if (*m++) 190 return TRUE; 191 return (!masks_are_equal); 192 } 193 194 struct radix_node * 195 rn_lookup(char *key, char *mask, struct radix_node_head *head) 196 { 197 struct radix_node *x; 198 char *netmask = NULL; 199 200 if (mask != NULL) { 201 x = rn_addmask(mask, TRUE, head->rnh_treetop->rn_offset, 202 head->rnh_maskhead); 203 if (x == NULL) 204 return (NULL); 205 netmask = x->rn_key; 206 } 207 x = rn_match(key, head); 208 if (x != NULL && netmask != NULL) { 209 while (x != NULL && x->rn_mask != netmask) 210 x = x->rn_dupedkey; 211 } 212 return x; 213 } 214 215 static boolean_t 216 rn_satisfies_leaf(char *trial, struct radix_node *leaf, int skip) 217 { 218 char *cp = trial, *cp2 = leaf->rn_key, *cp3 = leaf->rn_mask; 219 char *cplim; 220 int length = min(clen(cp), clen(cp2)); 221 222 if (cp3 == NULL) 223 cp3 = rn_ones; 224 else 225 length = min(length, clen(cp3)); 226 cplim = cp + length; 227 cp3 += skip; 228 cp2 += skip; 229 for (cp += skip; cp < cplim; cp++, cp2++, cp3++) 230 if ((*cp ^ *cp2) & *cp3) 231 return FALSE; 232 return TRUE; 233 } 234 235 struct radix_node * 236 rn_match(char *key, struct radix_node_head *head) 237 { 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; 243 int test, b, rn_bit; 244 245 t = rn_search(key, top); 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 */ 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; 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; 275 on1: 276 test = (*cp ^ *cp2) & 0xff; /* find first bit that differs */ 277 for (b = 7; (test >>= 1) > 0;) 278 b--; 279 matched_off = cp - key; 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 */ 285 if ((saved_t = t)->rn_mask == NULL) 286 t = t->rn_dupedkey; 287 for (; t; t = t->rn_dupedkey) { 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; 296 } else if (rn_satisfies_leaf(key, t, matched_off)) 297 return t; 298 } 299 t = saved_t; 300 /* start searching up the tree */ 301 do { 302 struct radix_mask *m; 303 304 t = t->rn_parent; 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 */ 311 m = t->rn_mklist; 312 while (m != NULL) { 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); 318 x = rn_search_m(key, t, m->rm_mask); 319 while (x != NULL && x->rn_mask != m->rm_mask) 320 x = x->rn_dupedkey; 321 if (x && rn_satisfies_leaf(key, x, off)) 322 return x; 323 } 324 m = m->rm_next; 325 } 326 } while (t != top); 327 return NULL; 328 } 329 330 #ifdef RN_DEBUG 331 int rn_nodenum; 332 struct radix_node *rn_clist; 333 int rn_saveinfo; 334 boolean_t rn_debug = TRUE; 335 #endif 336 337 static struct radix_node * 338 rn_newpair(char *key, int indexbit, struct radix_node nodes[2]) 339 { 340 struct radix_node *leaf = &nodes[0], *interior = &nodes[1]; 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; 353 354 #ifdef RN_DEBUG 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; 360 #endif 361 return interior; 362 } 363 364 static struct radix_node * 365 rn_insert(char *key, struct radix_node_head *head, boolean_t *dupentry, 366 struct radix_node nodes[2]) 367 { 368 struct radix_node *top = head->rnh_treetop; 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; 372 int b; 373 struct radix_node *tt; 374 375 /* 376 * Find first bit at which the key and t->rn_key differ 377 */ 378 { 379 char *cp2 = t->rn_key + head_off; 380 int cmp_res; 381 char *cplim = key + klen; 382 383 while (cp < cplim) 384 if (*cp2++ != *cp++) 385 goto on1; 386 *dupentry = TRUE; 387 return t; 388 on1: 389 *dupentry = FALSE; 390 cmp_res = (cp[-1] ^ cp2[-1]) & 0xff; 391 for (b = (cp - key) << 3; cmp_res; b--) 392 cmp_res >>= 1; 393 } 394 { 395 struct radix_node *p, *x = top; 396 397 cp = key; 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 410 t = rn_newpair(key, b, nodes); 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 } 431 432 struct radix_node * 433 rn_addmask(char *netmask, boolean_t search, int skip, 434 struct radix_node_head *mask_rnh) 435 { 436 struct radix_node *x, *saved_x; 437 char *cp, *cplim; 438 int b = 0, mlen, m0, j; 439 boolean_t maskduplicated, isnormal; 440 static int last_zeroed = 0; 441 char *addmask_key; 442 443 if ((mlen = clen(netmask)) > max_keylen) 444 mlen = max_keylen; 445 if (skip == 0) 446 skip = 1; 447 if (mlen <= skip) 448 return (mask_rnh->rnh_nodes); 449 R_Malloc(addmask_key, char *, max_keylen); 450 if (addmask_key == NULL) 451 return NULL; 452 if (skip > 1) 453 bcopy(rn_ones + 1, addmask_key + 1, skip - 1); 454 if ((m0 = mlen) > skip) 455 bcopy(netmask + skip, addmask_key + skip, mlen - skip); 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; 465 Free(addmask_key); 466 return (mask_rnh->rnh_nodes); 467 } 468 if (m0 < last_zeroed) 469 bzero(addmask_key + m0, last_zeroed - m0); 470 *addmask_key = last_zeroed = mlen; 471 x = rn_search(addmask_key, mask_rnh->rnh_treetop); 472 if (bcmp(addmask_key, x->rn_key, mlen) != 0) 473 x = NULL; 474 if (x != NULL || search) 475 goto out; 476 R_Malloc(x, struct radix_node *, max_keylen + 2 * (sizeof *x)); 477 if ((saved_x = x) == NULL) 478 goto out; 479 bzero(x, max_keylen + 2 * (sizeof *x)); 480 netmask = cp = (char *)(x + 2); 481 bcopy(addmask_key, cp, mlen); 482 x = rn_insert(cp, mask_rnh, &maskduplicated, x); 483 if (maskduplicated) { 484 log(LOG_ERR, "rn_addmask: mask impossibly already in tree"); 485 Free(saved_x); 486 goto out; 487 } 488 /* 489 * Calculate index of mask, and check for normalcy. 490 */ 491 isnormal = TRUE; 492 cplim = netmask + mlen; 493 for (cp = netmask + skip; cp < cplim && clen(cp) == 0xff;) 494 cp++; 495 if (cp != cplim) { 496 static const char normal_chars[] = { 497 0, 0x80, 0xc0, 0xe0, 0xf0, 0xf8, 0xfc, 0xfe, -1 498 }; 499 500 for (j = 0x80; (j & *cp) != 0; j >>= 1) 501 b++; 502 if (*cp != normal_chars[b] || cp != (cplim - 1)) 503 isnormal = FALSE; 504 } 505 b += (cp - netmask) << 3; 506 x->rn_bit = -1 - b; 507 if (isnormal) 508 x->rn_flags |= RNF_NORMAL; 509 out: 510 Free(addmask_key); 511 return (x); 512 } 513 514 /* XXX: arbitrary ordering for non-contiguous masks */ 515 static boolean_t 516 rn_lexobetter(char *mp, char *np) 517 { 518 char *lim; 519 520 if ((unsigned) *mp > (unsigned) *np) 521 return TRUE;/* not really, but need to check longer one first */ 522 if (*mp == *np) 523 for (lim = mp + clen(mp); mp < lim;) 524 if (*mp++ > *np++) 525 return TRUE; 526 return FALSE; 527 } 528 529 static struct radix_mask * 530 rn_new_radix_mask(struct radix_node *tt, struct radix_mask *nextmask) 531 { 532 struct radix_mask *m; 533 534 m = MKGet(&rn_mkfreelist); 535 if (m == NULL) { 536 log(LOG_ERR, "Mask for route not entered\n"); 537 return (NULL); 538 } 539 bzero(m, sizeof *m); 540 m->rm_bit = tt->rn_bit; 541 m->rm_flags = tt->rn_flags; 542 if (tt->rn_flags & RNF_NORMAL) 543 m->rm_leaf = tt; 544 else 545 m->rm_mask = tt->rn_mask; 546 m->rm_next = nextmask; 547 tt->rn_mklist = m; 548 return m; 549 } 550 551 struct radix_node * 552 rn_addroute(char *key, char *netmask, struct radix_node_head *head, 553 struct radix_node treenodes[2]) 554 { 555 struct radix_node *t, *x = NULL, *tt; 556 struct radix_node *saved_tt, *top = head->rnh_treetop; 557 short b = 0, b_leaf = 0; 558 boolean_t keyduplicated; 559 char *mmask; 560 struct radix_mask *m, **mp; 561 562 /* 563 * In dealing with non-contiguous masks, there may be 564 * many different routes which have the same mask. 565 * We will find it useful to have a unique pointer to 566 * the mask to speed avoiding duplicate references at 567 * nodes and possibly save time in calculating indices. 568 */ 569 if (netmask != NULL) { 570 if ((x = rn_addmask(netmask, FALSE, top->rn_offset, 571 head->rnh_maskhead)) == NULL) 572 return (NULL); 573 b_leaf = x->rn_bit; 574 b = -1 - x->rn_bit; 575 netmask = x->rn_key; 576 } 577 /* 578 * Deal with duplicated keys: attach node to previous instance 579 */ 580 saved_tt = tt = rn_insert(key, head, &keyduplicated, treenodes); 581 if (keyduplicated) { 582 for (t = tt; tt; t = tt, tt = tt->rn_dupedkey) { 583 if (tt->rn_mask == netmask) 584 return (NULL); 585 if (netmask == NULL || 586 (tt->rn_mask && 587 ((b_leaf < tt->rn_bit) /* index(netmask) > node */ 588 || rn_refines(netmask, tt->rn_mask) 589 || rn_lexobetter(netmask, tt->rn_mask)))) 590 break; 591 } 592 /* 593 * If the mask is not duplicated, we wouldn't 594 * find it among possible duplicate key entries 595 * anyway, so the above test doesn't hurt. 596 * 597 * We sort the masks for a duplicated key the same way as 598 * in a masklist -- most specific to least specific. 599 * This may require the unfortunate nuisance of relocating 600 * the head of the list. 601 */ 602 if (tt == saved_tt) { 603 struct radix_node *xx = x; 604 /* link in at head of list */ 605 (tt = treenodes)->rn_dupedkey = t; 606 tt->rn_flags = t->rn_flags; 607 tt->rn_parent = x = t->rn_parent; 608 t->rn_parent = tt; /* parent */ 609 if (x->rn_left == t) 610 x->rn_left = tt; 611 else 612 x->rn_right = tt; 613 saved_tt = tt; x = xx; 614 } else { 615 (tt = treenodes)->rn_dupedkey = t->rn_dupedkey; 616 t->rn_dupedkey = tt; 617 tt->rn_parent = t; /* parent */ 618 if (tt->rn_dupedkey != NULL) /* parent */ 619 tt->rn_dupedkey->rn_parent = tt; /* parent */ 620 } 621 #ifdef RN_DEBUG 622 t=tt+1; tt->rn_info = rn_nodenum++; t->rn_info = rn_nodenum++; 623 tt->rn_twin = t; tt->rn_ybro = rn_clist; rn_clist = tt; 624 #endif 625 tt->rn_key = key; 626 tt->rn_bit = -1; 627 tt->rn_flags = RNF_ACTIVE; 628 } 629 /* 630 * Put mask in tree. 631 */ 632 if (netmask != NULL) { 633 tt->rn_mask = netmask; 634 tt->rn_bit = x->rn_bit; 635 tt->rn_flags |= x->rn_flags & RNF_NORMAL; 636 } 637 t = saved_tt->rn_parent; 638 if (keyduplicated) 639 goto on2; 640 b_leaf = -1 - t->rn_bit; 641 if (t->rn_right == saved_tt) 642 x = t->rn_left; 643 else 644 x = t->rn_right; 645 /* Promote general routes from below */ 646 if (x->rn_bit < 0) { 647 mp = &t->rn_mklist; 648 while (x != NULL) { 649 if (x->rn_mask != NULL && 650 x->rn_bit >= b_leaf && 651 x->rn_mklist == NULL) { 652 *mp = m = rn_new_radix_mask(x, NULL); 653 if (m != NULL) 654 mp = &m->rm_next; 655 } 656 x = x->rn_dupedkey; 657 } 658 } else if (x->rn_mklist != NULL) { 659 /* 660 * Skip over masks whose index is > that of new node 661 */ 662 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_next) 663 if (m->rm_bit >= b_leaf) 664 break; 665 t->rn_mklist = m; 666 *mp = NULL; 667 } 668 on2: 669 /* Add new route to highest possible ancestor's list */ 670 if ((netmask == NULL) || (b > t->rn_bit )) 671 return tt; /* can't lift at all */ 672 b_leaf = tt->rn_bit; 673 do { 674 x = t; 675 t = t->rn_parent; 676 } while (b <= t->rn_bit && x != top); 677 /* 678 * Search through routes associated with node to 679 * insert new route according to index. 680 * Need same criteria as when sorting dupedkeys to avoid 681 * double loop on deletion. 682 */ 683 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_next) { 684 if (m->rm_bit < b_leaf) 685 continue; 686 if (m->rm_bit > b_leaf) 687 break; 688 if (m->rm_flags & RNF_NORMAL) { 689 mmask = m->rm_leaf->rn_mask; 690 if (tt->rn_flags & RNF_NORMAL) { 691 log(LOG_ERR, 692 "Non-unique normal route, mask not entered\n"); 693 return tt; 694 } 695 } else 696 mmask = m->rm_mask; 697 if (mmask == netmask) { 698 m->rm_refs++; 699 tt->rn_mklist = m; 700 return tt; 701 } 702 if (rn_refines(netmask, mmask) || rn_lexobetter(netmask, mmask)) 703 break; 704 } 705 *mp = rn_new_radix_mask(tt, *mp); 706 return tt; 707 } 708 709 struct radix_node * 710 rn_delete(char *key, char *netmask, struct radix_node_head *head) 711 { 712 struct radix_node *t, *p, *x, *tt; 713 struct radix_mask *m, *saved_m, **mp; 714 struct radix_node *dupedkey, *saved_tt, *top; 715 int b, head_off, klen; 716 717 x = head->rnh_treetop; 718 tt = rn_search(key, x); 719 head_off = x->rn_offset; 720 klen = clen(key); 721 saved_tt = tt; 722 top = x; 723 if (tt == NULL || 724 bcmp(key + head_off, tt->rn_key + head_off, klen - head_off)) 725 return (NULL); 726 /* 727 * Delete our route from mask lists. 728 */ 729 if (netmask != NULL) { 730 if ((x = rn_addmask(netmask, TRUE, head_off, 731 head->rnh_maskhead)) == NULL) 732 return (NULL); 733 netmask = x->rn_key; 734 while (tt->rn_mask != netmask) 735 if ((tt = tt->rn_dupedkey) == NULL) 736 return (NULL); 737 } 738 if (tt->rn_mask == NULL || (saved_m = m = tt->rn_mklist) == NULL) 739 goto on1; 740 if (tt->rn_flags & RNF_NORMAL) { 741 if (m->rm_leaf != tt || m->rm_refs > 0) { 742 log(LOG_ERR, "rn_delete: inconsistent annotation\n"); 743 return (NULL); /* dangling ref could cause disaster */ 744 } 745 } else { 746 if (m->rm_mask != tt->rn_mask) { 747 log(LOG_ERR, "rn_delete: inconsistent annotation\n"); 748 goto on1; 749 } 750 if (--m->rm_refs >= 0) 751 goto on1; 752 } 753 b = -1 - tt->rn_bit; 754 t = saved_tt->rn_parent; 755 if (b > t->rn_bit) 756 goto on1; /* Wasn't lifted at all */ 757 do { 758 x = t; 759 t = t->rn_parent; 760 } while (b <= t->rn_bit && x != top); 761 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_next) 762 if (m == saved_m) { 763 *mp = m->rm_next; 764 MKFree(&rn_mkfreelist, m); 765 break; 766 } 767 if (m == NULL) { 768 log(LOG_ERR, "rn_delete: couldn't find our annotation\n"); 769 if (tt->rn_flags & RNF_NORMAL) 770 return (NULL); /* Dangling ref to us */ 771 } 772 on1: 773 /* 774 * Eliminate us from tree 775 */ 776 if (tt->rn_flags & RNF_ROOT) 777 return (NULL); 778 #ifdef RN_DEBUG 779 /* Get us out of the creation list */ 780 for (t = rn_clist; t && t->rn_ybro != tt; t = t->rn_ybro) {} 781 if (t) t->rn_ybro = tt->rn_ybro; 782 #endif 783 t = tt->rn_parent; 784 dupedkey = saved_tt->rn_dupedkey; 785 if (dupedkey != NULL) { 786 /* 787 * at this point, tt is the deletion target and saved_tt 788 * is the head of the dupekey chain 789 */ 790 if (tt == saved_tt) { 791 /* remove from head of chain */ 792 x = dupedkey; x->rn_parent = t; 793 if (t->rn_left == tt) 794 t->rn_left = x; 795 else 796 t->rn_right = x; 797 } else { 798 /* find node in front of tt on the chain */ 799 for (x = p = saved_tt; p && p->rn_dupedkey != tt;) 800 p = p->rn_dupedkey; 801 if (p) { 802 p->rn_dupedkey = tt->rn_dupedkey; 803 if (tt->rn_dupedkey) /* parent */ 804 tt->rn_dupedkey->rn_parent = p; 805 /* parent */ 806 } else log(LOG_ERR, "rn_delete: couldn't find us\n"); 807 } 808 t = tt + 1; 809 if (t->rn_flags & RNF_ACTIVE) { 810 #ifndef RN_DEBUG 811 *++x = *t; 812 p = t->rn_parent; 813 #else 814 b = t->rn_info; 815 *++x = *t; 816 t->rn_info = b; 817 p = t->rn_parent; 818 #endif 819 if (p->rn_left == t) 820 p->rn_left = x; 821 else 822 p->rn_right = x; 823 x->rn_left->rn_parent = x; 824 x->rn_right->rn_parent = x; 825 } 826 goto out; 827 } 828 if (t->rn_left == tt) 829 x = t->rn_right; 830 else 831 x = t->rn_left; 832 p = t->rn_parent; 833 if (p->rn_right == t) 834 p->rn_right = x; 835 else 836 p->rn_left = x; 837 x->rn_parent = p; 838 /* 839 * Demote routes attached to us. 840 */ 841 if (t->rn_mklist != NULL) { 842 if (x->rn_bit >= 0) { 843 for (mp = &x->rn_mklist; (m = *mp);) 844 mp = &m->rm_next; 845 *mp = t->rn_mklist; 846 } else { 847 /* 848 * If there are any (key, mask) pairs in a sibling 849 * duped-key chain, some subset will appear sorted 850 * in the same order attached to our mklist. 851 */ 852 for (m = t->rn_mklist; m && x; x = x->rn_dupedkey) 853 if (m == x->rn_mklist) { 854 struct radix_mask *mm = m->rm_next; 855 856 x->rn_mklist = NULL; 857 if (--(m->rm_refs) < 0) 858 MKFree(&rn_mkfreelist, m); 859 m = mm; 860 } 861 if (m) 862 log(LOG_ERR, 863 "rn_delete: Orphaned Mask %p at %p\n", 864 (void *)m, (void *)x); 865 } 866 } 867 /* 868 * We may be holding an active internal node in the tree. 869 */ 870 x = tt + 1; 871 if (t != x) { 872 #ifndef RN_DEBUG 873 *t = *x; 874 #else 875 b = t->rn_info; 876 *t = *x; 877 t->rn_info = b; 878 #endif 879 t->rn_left->rn_parent = t; 880 t->rn_right->rn_parent = t; 881 p = x->rn_parent; 882 if (p->rn_left == x) 883 p->rn_left = t; 884 else 885 p->rn_right = t; 886 } 887 out: 888 tt->rn_flags &= ~RNF_ACTIVE; 889 tt[1].rn_flags &= ~RNF_ACTIVE; 890 return (tt); 891 } 892 893 /* 894 * This is the same as rn_walktree() except for the parameters and the 895 * exit. 896 */ 897 static int 898 rn_walktree_from(struct radix_node_head *h, char *xa, char *xm, 899 walktree_f_t *f, void *w) 900 { 901 struct radix_node *base, *next; 902 struct radix_node *rn, *last = NULL /* shut up gcc */; 903 boolean_t stopping = FALSE; 904 int lastb, error; 905 906 /* 907 * rn_search_m is sort-of-open-coded here. 908 */ 909 /* kprintf("about to search\n"); */ 910 for (rn = h->rnh_treetop; rn->rn_bit >= 0; ) { 911 last = rn; 912 /* kprintf("rn_bit %d, rn_bmask %x, xm[rn_offset] %x\n", 913 rn->rn_bit, rn->rn_bmask, xm[rn->rn_offset]); */ 914 if (!(rn->rn_bmask & xm[rn->rn_offset])) { 915 break; 916 } 917 if (rn->rn_bmask & xa[rn->rn_offset]) { 918 rn = rn->rn_right; 919 } else { 920 rn = rn->rn_left; 921 } 922 } 923 /* kprintf("done searching\n"); */ 924 925 /* 926 * Two cases: either we stepped off the end of our mask, 927 * in which case last == rn, or we reached a leaf, in which 928 * case we want to start from the last node we looked at. 929 * Either way, last is the node we want to start from. 930 */ 931 rn = last; 932 lastb = rn->rn_bit; 933 934 /* kprintf("rn %p, lastb %d\n", rn, lastb);*/ 935 936 /* 937 * This gets complicated because we may delete the node 938 * while applying the function f to it, so we need to calculate 939 * the successor node in advance. 940 */ 941 while (rn->rn_bit >= 0) 942 rn = rn->rn_left; 943 944 while (!stopping) { 945 /* kprintf("node %p (%d)\n", rn, rn->rn_bit); */ 946 base = rn; 947 /* If at right child go back up, otherwise, go right */ 948 while (rn->rn_parent->rn_right == rn && 949 !(rn->rn_flags & RNF_ROOT)) { 950 rn = rn->rn_parent; 951 952 /* if went up beyond last, stop */ 953 if (rn->rn_bit < lastb) { 954 stopping = TRUE; 955 /* kprintf("up too far\n"); */ 956 } 957 } 958 959 /* Find the next *leaf* since next node might vanish, too */ 960 for (rn = rn->rn_parent->rn_right; rn->rn_bit >= 0;) 961 rn = rn->rn_left; 962 next = rn; 963 /* Process leaves */ 964 while ((rn = base) != NULL) { 965 base = rn->rn_dupedkey; 966 /* kprintf("leaf %p\n", rn); */ 967 if (!(rn->rn_flags & RNF_ROOT) && (error = (*f)(rn, w))) 968 return (error); 969 } 970 rn = next; 971 972 if (rn->rn_flags & RNF_ROOT) { 973 /* kprintf("root, stopping"); */ 974 stopping = TRUE; 975 } 976 977 } 978 return 0; 979 } 980 981 static int 982 rn_walktree(struct radix_node_head *h, walktree_f_t *f, void *w) 983 { 984 struct radix_node *base, *next; 985 struct radix_node *rn = h->rnh_treetop; 986 int error; 987 988 /* 989 * This gets complicated because we may delete the node 990 * while applying the function f to it, so we need to calculate 991 * the successor node in advance. 992 */ 993 /* First time through node, go left */ 994 while (rn->rn_bit >= 0) 995 rn = rn->rn_left; 996 for (;;) { 997 base = rn; 998 /* If at right child go back up, otherwise, go right */ 999 while (rn->rn_parent->rn_right == rn && 1000 !(rn->rn_flags & RNF_ROOT)) 1001 rn = rn->rn_parent; 1002 /* Find the next *leaf* since next node might vanish, too */ 1003 for (rn = rn->rn_parent->rn_right; rn->rn_bit >= 0;) 1004 rn = rn->rn_left; 1005 next = rn; 1006 /* Process leaves */ 1007 while ((rn = base)) { 1008 base = rn->rn_dupedkey; 1009 if (!(rn->rn_flags & RNF_ROOT) && (error = (*f)(rn, w))) 1010 return (error); 1011 } 1012 rn = next; 1013 if (rn->rn_flags & RNF_ROOT) 1014 return (0); 1015 } 1016 /* NOTREACHED */ 1017 } 1018 1019 int 1020 rn_inithead(void **head, struct radix_node_head *maskhead, int off) 1021 { 1022 struct radix_node_head *rnh; 1023 struct radix_node *root, *left, *right; 1024 1025 if (*head != NULL) /* already initialized */ 1026 return (1); 1027 1028 R_Malloc(rnh, struct radix_node_head *, sizeof *rnh); 1029 if (rnh == NULL) 1030 return (0); 1031 bzero(rnh, sizeof *rnh); 1032 *head = rnh; 1033 1034 root = rn_newpair(rn_zeros, off, rnh->rnh_nodes); 1035 right = &rnh->rnh_nodes[2]; 1036 root->rn_parent = root; 1037 root->rn_flags = RNF_ROOT | RNF_ACTIVE; 1038 root->rn_right = right; 1039 1040 left = root->rn_left; 1041 left->rn_bit = -1 - off; 1042 left->rn_flags = RNF_ROOT | RNF_ACTIVE; 1043 1044 *right = *left; 1045 right->rn_key = rn_ones; 1046 1047 rnh->rnh_treetop = root; 1048 rnh->rnh_maskhead = maskhead; 1049 1050 rnh->rnh_addaddr = rn_addroute; 1051 rnh->rnh_deladdr = rn_delete; 1052 rnh->rnh_matchaddr = rn_match; 1053 rnh->rnh_lookup = rn_lookup; 1054 rnh->rnh_walktree = rn_walktree; 1055 rnh->rnh_walktree_from = rn_walktree_from; 1056 1057 return (1); 1058 } 1059 1060 void 1061 rn_init(void) 1062 { 1063 char *cp, *cplim; 1064 int cpu; 1065 #ifdef _KERNEL 1066 struct domain *dom; 1067 1068 SLIST_FOREACH(dom, &domains, dom_next) 1069 if (dom->dom_maxrtkey > max_keylen) 1070 max_keylen = dom->dom_maxrtkey; 1071 #endif 1072 if (max_keylen == 0) { 1073 log(LOG_ERR, 1074 "rn_init: radix functions require max_keylen be set\n"); 1075 return; 1076 } 1077 R_Malloc(rn_zeros, char *, 2 * max_keylen); 1078 if (rn_zeros == NULL) 1079 panic("rn_init"); 1080 bzero(rn_zeros, 2 * max_keylen); 1081 rn_ones = cp = rn_zeros + max_keylen; 1082 cplim = rn_ones + max_keylen; 1083 while (cp < cplim) 1084 *cp++ = -1; 1085 1086 for (cpu = 0; cpu < ncpus; ++cpu) { 1087 if (rn_inithead((void **)&mask_rnheads[cpu], NULL, 0) == 0) 1088 panic("rn_init 2"); 1089 } 1090 } 1091 1092 struct radix_node_head * 1093 rn_cpumaskhead(int cpu) 1094 { 1095 KKASSERT(mask_rnheads[cpu] != NULL); 1096 return mask_rnheads[cpu]; 1097 } 1098