1 /* $NetBSD: radix.c,v 1.43 2009/05/27 17:46:50 pooka Exp $ */ 2 3 /* 4 * Copyright (c) 1988, 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * @(#)radix.c 8.6 (Berkeley) 10/17/95 32 */ 33 34 /* 35 * Routines to build and maintain radix trees for routing lookups. 36 */ 37 38 #include <sys/cdefs.h> 39 __KERNEL_RCSID(0, "$NetBSD: radix.c,v 1.43 2009/05/27 17:46:50 pooka Exp $"); 40 41 #ifndef _NET_RADIX_H_ 42 #include <sys/param.h> 43 #include <sys/queue.h> 44 #include <sys/kmem.h> 45 #ifdef _KERNEL 46 #include "opt_inet.h" 47 48 #include <sys/systm.h> 49 #include <sys/malloc.h> 50 #define M_DONTWAIT M_NOWAIT 51 #include <sys/domain.h> 52 #else 53 #include <stdlib.h> 54 #endif 55 #include <machine/stdarg.h> 56 #include <sys/syslog.h> 57 #include <net/radix.h> 58 #endif 59 60 typedef void (*rn_printer_t)(void *, const char *fmt, ...); 61 62 int max_keylen; 63 struct radix_mask *rn_mkfreelist; 64 struct radix_node_head *mask_rnhead; 65 static char *addmask_key; 66 static const char normal_chars[] = 67 {0, 0x80, 0xc0, 0xe0, 0xf0, 0xf8, 0xfc, 0xfe, -1}; 68 static char *rn_zeros, *rn_ones; 69 70 #define rn_masktop (mask_rnhead->rnh_treetop) 71 72 static int rn_satisfies_leaf(const char *, struct radix_node *, int); 73 static int rn_lexobetter(const void *, const void *); 74 static struct radix_mask *rn_new_radix_mask(struct radix_node *, 75 struct radix_mask *); 76 static struct radix_node *rn_walknext(struct radix_node *, rn_printer_t, 77 void *); 78 static struct radix_node *rn_walkfirst(struct radix_node *, rn_printer_t, 79 void *); 80 static void rn_nodeprint(struct radix_node *, rn_printer_t, void *, 81 const char *); 82 83 #define SUBTREE_OPEN "[ " 84 #define SUBTREE_CLOSE " ]" 85 86 #ifdef RN_DEBUG 87 static void rn_treeprint(struct radix_node_head *, rn_printer_t, void *); 88 #endif /* RN_DEBUG */ 89 90 /* 91 * The data structure for the keys is a radix tree with one way 92 * branching removed. The index rn_b at an internal node n represents a bit 93 * position to be tested. The tree is arranged so that all descendants 94 * of a node n have keys whose bits all agree up to position rn_b - 1. 95 * (We say the index of n is rn_b.) 96 * 97 * There is at least one descendant which has a one bit at position rn_b, 98 * and at least one with a zero there. 99 * 100 * A route is determined by a pair of key and mask. We require that the 101 * bit-wise logical and of the key and mask to be the key. 102 * We define the index of a route to associated with the mask to be 103 * the first bit number in the mask where 0 occurs (with bit number 0 104 * representing the highest order bit). 105 * 106 * We say a mask is normal if every bit is 0, past the index of the mask. 107 * If a node n has a descendant (k, m) with index(m) == index(n) == rn_b, 108 * and m is a normal mask, then the route applies to every descendant of n. 109 * If the index(m) < rn_b, this implies the trailing last few bits of k 110 * before bit b are all 0, (and hence consequently true of every descendant 111 * of n), so the route applies to all descendants of the node as well. 112 * 113 * Similar logic shows that a non-normal mask m such that 114 * index(m) <= index(n) could potentially apply to many children of n. 115 * Thus, for each non-host route, we attach its mask to a list at an internal 116 * node as high in the tree as we can go. 117 * 118 * The present version of the code makes use of normal routes in short- 119 * circuiting an explicit mask and compare operation when testing whether 120 * a key satisfies a normal route, and also in remembering the unique leaf 121 * that governs a subtree. 122 */ 123 124 struct radix_node * 125 rn_search( 126 const void *v_arg, 127 struct radix_node *head) 128 { 129 const u_char * const v = v_arg; 130 struct radix_node *x; 131 132 for (x = head; x->rn_b >= 0;) { 133 if (x->rn_bmask & v[x->rn_off]) 134 x = x->rn_r; 135 else 136 x = x->rn_l; 137 } 138 return x; 139 } 140 141 struct radix_node * 142 rn_search_m( 143 const void *v_arg, 144 struct radix_node *head, 145 const void *m_arg) 146 { 147 struct radix_node *x; 148 const u_char * const v = v_arg; 149 const u_char * const m = m_arg; 150 151 for (x = head; x->rn_b >= 0;) { 152 if ((x->rn_bmask & m[x->rn_off]) && 153 (x->rn_bmask & v[x->rn_off])) 154 x = x->rn_r; 155 else 156 x = x->rn_l; 157 } 158 return x; 159 } 160 161 int 162 rn_refines( 163 const void *m_arg, 164 const void *n_arg) 165 { 166 const char *m = m_arg; 167 const char *n = n_arg; 168 const char *lim = n + *(const u_char *)n; 169 const char *lim2 = lim; 170 int longer = (*(const u_char *)n++) - (int)(*(const u_char *)m++); 171 int masks_are_equal = 1; 172 173 if (longer > 0) 174 lim -= longer; 175 while (n < lim) { 176 if (*n & ~(*m)) 177 return 0; 178 if (*n++ != *m++) 179 masks_are_equal = 0; 180 } 181 while (n < lim2) 182 if (*n++) 183 return 0; 184 if (masks_are_equal && (longer < 0)) 185 for (lim2 = m - longer; m < lim2; ) 186 if (*m++) 187 return 1; 188 return !masks_are_equal; 189 } 190 191 struct radix_node * 192 rn_lookup( 193 const void *v_arg, 194 const void *m_arg, 195 struct radix_node_head *head) 196 { 197 struct radix_node *x; 198 const char *netmask = NULL; 199 200 if (m_arg) { 201 if ((x = rn_addmask(m_arg, 1, head->rnh_treetop->rn_off)) == 0) 202 return NULL; 203 netmask = x->rn_key; 204 } 205 x = rn_match(v_arg, head); 206 if (x != NULL && netmask != NULL) { 207 while (x != NULL && x->rn_mask != netmask) 208 x = x->rn_dupedkey; 209 } 210 return x; 211 } 212 213 static int 214 rn_satisfies_leaf( 215 const char *trial, 216 struct radix_node *leaf, 217 int skip) 218 { 219 const char *cp = trial; 220 const char *cp2 = leaf->rn_key; 221 const char *cp3 = leaf->rn_mask; 222 const char *cplim; 223 int length = min(*(const u_char *)cp, *(const u_char *)cp2); 224 225 if (cp3 == 0) 226 cp3 = rn_ones; 227 else 228 length = min(length, *(const u_char *)cp3); 229 cplim = cp + length; cp3 += skip; cp2 += skip; 230 for (cp += skip; cp < cplim; cp++, cp2++, cp3++) 231 if ((*cp ^ *cp2) & *cp3) 232 return 0; 233 return 1; 234 } 235 236 struct radix_node * 237 rn_match( 238 const void *v_arg, 239 struct radix_node_head *head) 240 { 241 const char * const v = v_arg; 242 struct radix_node *t = head->rnh_treetop; 243 struct radix_node *top = t; 244 struct radix_node *x; 245 struct radix_node *saved_t; 246 const char *cp = v; 247 const char *cp2; 248 const char *cplim; 249 int off = t->rn_off; 250 int vlen = *(const u_char *)cp; 251 int matched_off; 252 int test, b, rn_b; 253 254 /* 255 * Open code rn_search(v, top) to avoid overhead of extra 256 * subroutine call. 257 */ 258 for (; t->rn_b >= 0; ) { 259 if (t->rn_bmask & cp[t->rn_off]) 260 t = t->rn_r; 261 else 262 t = t->rn_l; 263 } 264 /* 265 * See if we match exactly as a host destination 266 * or at least learn how many bits match, for normal mask finesse. 267 * 268 * It doesn't hurt us to limit how many bytes to check 269 * to the length of the mask, since if it matches we had a genuine 270 * match and the leaf we have is the most specific one anyway; 271 * if it didn't match with a shorter length it would fail 272 * with a long one. This wins big for class B&C netmasks which 273 * are probably the most common case... 274 */ 275 if (t->rn_mask) 276 vlen = *(const u_char *)t->rn_mask; 277 cp += off; cp2 = t->rn_key + off; cplim = v + vlen; 278 for (; cp < cplim; cp++, cp2++) 279 if (*cp != *cp2) 280 goto on1; 281 /* 282 * This extra grot is in case we are explicitly asked 283 * to look up the default. Ugh! 284 */ 285 if ((t->rn_flags & RNF_ROOT) && t->rn_dupedkey) 286 t = t->rn_dupedkey; 287 return t; 288 on1: 289 test = (*cp ^ *cp2) & 0xff; /* find first bit that differs */ 290 for (b = 7; (test >>= 1) > 0;) 291 b--; 292 matched_off = cp - v; 293 b += matched_off << 3; 294 rn_b = -1 - b; 295 /* 296 * If there is a host route in a duped-key chain, it will be first. 297 */ 298 if ((saved_t = t)->rn_mask == 0) 299 t = t->rn_dupedkey; 300 for (; t; t = t->rn_dupedkey) 301 /* 302 * Even if we don't match exactly as a host, 303 * we may match if the leaf we wound up at is 304 * a route to a net. 305 */ 306 if (t->rn_flags & RNF_NORMAL) { 307 if (rn_b <= t->rn_b) 308 return t; 309 } else if (rn_satisfies_leaf(v, t, matched_off)) 310 return t; 311 t = saved_t; 312 /* start searching up the tree */ 313 do { 314 struct radix_mask *m; 315 t = t->rn_p; 316 m = t->rn_mklist; 317 if (m) { 318 /* 319 * If non-contiguous masks ever become important 320 * we can restore the masking and open coding of 321 * the search and satisfaction test and put the 322 * calculation of "off" back before the "do". 323 */ 324 do { 325 if (m->rm_flags & RNF_NORMAL) { 326 if (rn_b <= m->rm_b) 327 return m->rm_leaf; 328 } else { 329 off = min(t->rn_off, matched_off); 330 x = rn_search_m(v, t, m->rm_mask); 331 while (x && x->rn_mask != m->rm_mask) 332 x = x->rn_dupedkey; 333 if (x && rn_satisfies_leaf(v, x, off)) 334 return x; 335 } 336 m = m->rm_mklist; 337 } while (m); 338 } 339 } while (t != top); 340 return NULL; 341 } 342 343 static void 344 rn_nodeprint(struct radix_node *rn, rn_printer_t printer, void *arg, 345 const char *delim) 346 { 347 (*printer)(arg, "%s(%s%p: p<%p> l<%p> r<%p>)", 348 delim, ((void *)rn == arg) ? "*" : "", rn, rn->rn_p, 349 rn->rn_l, rn->rn_r); 350 } 351 352 #ifdef RN_DEBUG 353 int rn_debug = 1; 354 355 static void 356 rn_dbg_print(void *arg, const char *fmt, ...) 357 { 358 va_list ap; 359 360 va_start(ap, fmt); 361 vlog(LOG_DEBUG, fmt, ap); 362 va_end(ap); 363 } 364 365 static void 366 rn_treeprint(struct radix_node_head *h, rn_printer_t printer, void *arg) 367 { 368 struct radix_node *dup, *rn; 369 const char *delim; 370 371 if (printer == NULL) 372 return; 373 374 rn = rn_walkfirst(h->rnh_treetop, printer, arg); 375 for (;;) { 376 /* Process leaves */ 377 delim = ""; 378 for (dup = rn; dup != NULL; dup = dup->rn_dupedkey) { 379 if ((dup->rn_flags & RNF_ROOT) != 0) 380 continue; 381 rn_nodeprint(dup, printer, arg, delim); 382 delim = ", "; 383 } 384 rn = rn_walknext(rn, printer, arg); 385 if (rn->rn_flags & RNF_ROOT) 386 return; 387 } 388 /* NOTREACHED */ 389 } 390 391 #define traverse(__head, __rn) rn_treeprint((__head), rn_dbg_print, (__rn)) 392 #endif /* RN_DEBUG */ 393 394 struct radix_node * 395 rn_newpair( 396 const void *v, 397 int b, 398 struct radix_node nodes[2]) 399 { 400 struct radix_node *tt = nodes; 401 struct radix_node *t = tt + 1; 402 t->rn_b = b; t->rn_bmask = 0x80 >> (b & 7); 403 t->rn_l = tt; t->rn_off = b >> 3; 404 tt->rn_b = -1; tt->rn_key = v; tt->rn_p = t; 405 tt->rn_flags = t->rn_flags = RNF_ACTIVE; 406 return t; 407 } 408 409 struct radix_node * 410 rn_insert( 411 const void *v_arg, 412 struct radix_node_head *head, 413 int *dupentry, 414 struct radix_node nodes[2]) 415 { 416 struct radix_node *top = head->rnh_treetop; 417 struct radix_node *t = rn_search(v_arg, top); 418 struct radix_node *tt; 419 const char *v = v_arg; 420 int head_off = top->rn_off; 421 int vlen = *((const u_char *)v); 422 const char *cp = v + head_off; 423 int b; 424 /* 425 * Find first bit at which v and t->rn_key differ 426 */ 427 { 428 const char *cp2 = t->rn_key + head_off; 429 const char *cplim = v + vlen; 430 int cmp_res; 431 432 while (cp < cplim) 433 if (*cp2++ != *cp++) 434 goto on1; 435 *dupentry = 1; 436 return t; 437 on1: 438 *dupentry = 0; 439 cmp_res = (cp[-1] ^ cp2[-1]) & 0xff; 440 for (b = (cp - v) << 3; cmp_res; b--) 441 cmp_res >>= 1; 442 } 443 { 444 struct radix_node *p, *x = top; 445 cp = v; 446 do { 447 p = x; 448 if (cp[x->rn_off] & x->rn_bmask) 449 x = x->rn_r; 450 else x = x->rn_l; 451 } while (b > (unsigned) x->rn_b); /* x->rn_b < b && x->rn_b >= 0 */ 452 #ifdef RN_DEBUG 453 if (rn_debug) 454 log(LOG_DEBUG, "%s: Going In:\n", __func__), traverse(head, p); 455 #endif 456 t = rn_newpair(v_arg, b, nodes); tt = t->rn_l; 457 if ((cp[p->rn_off] & p->rn_bmask) == 0) 458 p->rn_l = t; 459 else 460 p->rn_r = t; 461 x->rn_p = t; t->rn_p = p; /* frees x, p as temp vars below */ 462 if ((cp[t->rn_off] & t->rn_bmask) == 0) { 463 t->rn_r = x; 464 } else { 465 t->rn_r = tt; t->rn_l = x; 466 } 467 #ifdef RN_DEBUG 468 if (rn_debug) { 469 log(LOG_DEBUG, "%s: Coming Out:\n", __func__), 470 traverse(head, p); 471 } 472 #endif /* RN_DEBUG */ 473 } 474 return tt; 475 } 476 477 struct radix_node * 478 rn_addmask( 479 const void *n_arg, 480 int search, 481 int skip) 482 { 483 const char *netmask = n_arg; 484 const char *cp; 485 const char *cplim; 486 struct radix_node *x; 487 struct radix_node *saved_x; 488 int b = 0, mlen, j; 489 int maskduplicated, m0, isnormal; 490 static int last_zeroed = 0; 491 492 if ((mlen = *(const u_char *)netmask) > max_keylen) 493 mlen = max_keylen; 494 if (skip == 0) 495 skip = 1; 496 if (mlen <= skip) 497 return mask_rnhead->rnh_nodes; 498 if (skip > 1) 499 memmove(addmask_key + 1, rn_ones + 1, skip - 1); 500 if ((m0 = mlen) > skip) 501 memmove(addmask_key + skip, netmask + skip, mlen - skip); 502 /* 503 * Trim trailing zeroes. 504 */ 505 for (cp = addmask_key + mlen; (cp > addmask_key) && cp[-1] == 0;) 506 cp--; 507 mlen = cp - addmask_key; 508 if (mlen <= skip) { 509 if (m0 >= last_zeroed) 510 last_zeroed = mlen; 511 return mask_rnhead->rnh_nodes; 512 } 513 if (m0 < last_zeroed) 514 memset(addmask_key + m0, 0, last_zeroed - m0); 515 *addmask_key = last_zeroed = mlen; 516 x = rn_search(addmask_key, rn_masktop); 517 if (memcmp(addmask_key, x->rn_key, mlen) != 0) 518 x = 0; 519 if (x || search) 520 return x; 521 R_Malloc(x, struct radix_node *, max_keylen + 2 * sizeof (*x)); 522 if ((saved_x = x) == NULL) 523 return NULL; 524 memset(x, 0, max_keylen + 2 * sizeof (*x)); 525 cp = netmask = (void *)(x + 2); 526 memmove(x + 2, addmask_key, mlen); 527 x = rn_insert(cp, mask_rnhead, &maskduplicated, x); 528 if (maskduplicated) { 529 log(LOG_ERR, "rn_addmask: mask impossibly already in tree\n"); 530 Free(saved_x); 531 return x; 532 } 533 /* 534 * Calculate index of mask, and check for normalcy. 535 */ 536 cplim = netmask + mlen; isnormal = 1; 537 for (cp = netmask + skip; (cp < cplim) && *(const u_char *)cp == 0xff;) 538 cp++; 539 if (cp != cplim) { 540 for (j = 0x80; (j & *cp) != 0; j >>= 1) 541 b++; 542 if (*cp != normal_chars[b] || cp != (cplim - 1)) 543 isnormal = 0; 544 } 545 b += (cp - netmask) << 3; 546 x->rn_b = -1 - b; 547 if (isnormal) 548 x->rn_flags |= RNF_NORMAL; 549 return x; 550 } 551 552 static int /* XXX: arbitrary ordering for non-contiguous masks */ 553 rn_lexobetter( 554 const void *m_arg, 555 const void *n_arg) 556 { 557 const u_char *mp = m_arg; 558 const u_char *np = n_arg; 559 const u_char *lim; 560 561 if (*mp > *np) 562 return 1; /* not really, but need to check longer one first */ 563 if (*mp == *np) 564 for (lim = mp + *mp; mp < lim;) 565 if (*mp++ > *np++) 566 return 1; 567 return 0; 568 } 569 570 static struct radix_mask * 571 rn_new_radix_mask( 572 struct radix_node *tt, 573 struct radix_mask *next) 574 { 575 struct radix_mask *m; 576 577 MKGet(m); 578 if (m == NULL) { 579 log(LOG_ERR, "Mask for route not entered\n"); 580 return NULL; 581 } 582 memset(m, 0, sizeof(*m)); 583 m->rm_b = tt->rn_b; 584 m->rm_flags = tt->rn_flags; 585 if (tt->rn_flags & RNF_NORMAL) 586 m->rm_leaf = tt; 587 else 588 m->rm_mask = tt->rn_mask; 589 m->rm_mklist = next; 590 tt->rn_mklist = m; 591 return m; 592 } 593 594 struct radix_node * 595 rn_addroute( 596 const void *v_arg, 597 const void *n_arg, 598 struct radix_node_head *head, 599 struct radix_node treenodes[2]) 600 { 601 const char *v = v_arg, *netmask = 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 const char *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 != NULL) { 617 if ((x = rn_addmask(netmask, 0, top->rn_off)) == NULL) 618 return NULL; 619 b_leaf = x->rn_b; 620 b = -1 - x->rn_b; 621 netmask = x->rn_key; 622 } 623 /* 624 * Deal with duplicated keys: attach node to previous instance 625 */ 626 saved_tt = tt = rn_insert(v, head, &keyduplicated, treenodes); 627 if (keyduplicated) { 628 for (t = tt; tt != NULL; t = tt, tt = tt->rn_dupedkey) { 629 if (tt->rn_mask == netmask) 630 return NULL; 631 if (netmask == NULL || 632 (tt->rn_mask != NULL && 633 (b_leaf < tt->rn_b || /* index(netmask) > node */ 634 rn_refines(netmask, tt->rn_mask) || 635 rn_lexobetter(netmask, tt->rn_mask)))) 636 break; 637 } 638 /* 639 * If the mask is not duplicated, we wouldn't 640 * find it among possible duplicate key entries 641 * anyway, so the above test doesn't hurt. 642 * 643 * We sort the masks for a duplicated key the same way as 644 * in a masklist -- most specific to least specific. 645 * This may require the unfortunate nuisance of relocating 646 * the head of the list. 647 * 648 * We also reverse, or doubly link the list through the 649 * parent pointer. 650 */ 651 if (tt == saved_tt) { 652 struct radix_node *xx = x; 653 /* link in at head of list */ 654 (tt = treenodes)->rn_dupedkey = t; 655 tt->rn_flags = t->rn_flags; 656 tt->rn_p = x = t->rn_p; 657 t->rn_p = tt; 658 if (x->rn_l == t) 659 x->rn_l = tt; 660 else 661 x->rn_r = tt; 662 saved_tt = tt; 663 x = xx; 664 } else { 665 (tt = treenodes)->rn_dupedkey = t->rn_dupedkey; 666 t->rn_dupedkey = tt; 667 tt->rn_p = t; 668 if (tt->rn_dupedkey) 669 tt->rn_dupedkey->rn_p = tt; 670 } 671 tt->rn_key = v; 672 tt->rn_b = -1; 673 tt->rn_flags = RNF_ACTIVE; 674 } 675 /* 676 * Put mask in tree. 677 */ 678 if (netmask != NULL) { 679 tt->rn_mask = netmask; 680 tt->rn_b = x->rn_b; 681 tt->rn_flags |= x->rn_flags & RNF_NORMAL; 682 } 683 t = saved_tt->rn_p; 684 if (keyduplicated) 685 goto on2; 686 b_leaf = -1 - t->rn_b; 687 if (t->rn_r == saved_tt) 688 x = t->rn_l; 689 else 690 x = t->rn_r; 691 /* Promote general routes from below */ 692 if (x->rn_b < 0) { 693 for (mp = &t->rn_mklist; x != NULL; x = x->rn_dupedkey) { 694 if (x->rn_mask != NULL && x->rn_b >= b_leaf && 695 x->rn_mklist == NULL) { 696 *mp = m = rn_new_radix_mask(x, NULL); 697 if (m != NULL) 698 mp = &m->rm_mklist; 699 } 700 } 701 } else if (x->rn_mklist != NULL) { 702 /* 703 * Skip over masks whose index is > that of new node 704 */ 705 for (mp = &x->rn_mklist; (m = *mp) != NULL; mp = &m->rm_mklist) 706 if (m->rm_b >= b_leaf) 707 break; 708 t->rn_mklist = m; 709 *mp = NULL; 710 } 711 on2: 712 /* Add new route to highest possible ancestor's list */ 713 if (netmask == NULL || b > t->rn_b) 714 return tt; /* can't lift at all */ 715 b_leaf = tt->rn_b; 716 do { 717 x = t; 718 t = t->rn_p; 719 } while (b <= t->rn_b && x != top); 720 /* 721 * Search through routes associated with node to 722 * insert new route according to index. 723 * Need same criteria as when sorting dupedkeys to avoid 724 * double loop on deletion. 725 */ 726 for (mp = &x->rn_mklist; (m = *mp) != NULL; mp = &m->rm_mklist) { 727 if (m->rm_b < b_leaf) 728 continue; 729 if (m->rm_b > b_leaf) 730 break; 731 if (m->rm_flags & RNF_NORMAL) { 732 mmask = m->rm_leaf->rn_mask; 733 if (tt->rn_flags & RNF_NORMAL) { 734 log(LOG_ERR, "Non-unique normal route," 735 " mask not entered\n"); 736 return tt; 737 } 738 } else 739 mmask = m->rm_mask; 740 if (mmask == netmask) { 741 m->rm_refs++; 742 tt->rn_mklist = m; 743 return tt; 744 } 745 if (rn_refines(netmask, mmask) || rn_lexobetter(netmask, mmask)) 746 break; 747 } 748 *mp = rn_new_radix_mask(tt, *mp); 749 return tt; 750 } 751 752 struct radix_node * 753 rn_delete1( 754 const void *v_arg, 755 const void *netmask_arg, 756 struct radix_node_head *head, 757 struct radix_node *rn) 758 { 759 struct radix_node *t, *p, *x, *tt; 760 struct radix_mask *m, *saved_m, **mp; 761 struct radix_node *dupedkey, *saved_tt, *top; 762 const char *v, *netmask; 763 int b, head_off, vlen; 764 765 v = v_arg; 766 netmask = netmask_arg; 767 x = head->rnh_treetop; 768 tt = rn_search(v, x); 769 head_off = x->rn_off; 770 vlen = *(const u_char *)v; 771 saved_tt = tt; 772 top = x; 773 if (tt == NULL || 774 memcmp(v + head_off, tt->rn_key + head_off, vlen - head_off) != 0) 775 return NULL; 776 /* 777 * Delete our route from mask lists. 778 */ 779 if (netmask != NULL) { 780 if ((x = rn_addmask(netmask, 1, head_off)) == NULL) 781 return NULL; 782 netmask = x->rn_key; 783 while (tt->rn_mask != netmask) 784 if ((tt = tt->rn_dupedkey) == NULL) 785 return NULL; 786 } 787 if (tt->rn_mask == NULL || (saved_m = m = tt->rn_mklist) == NULL) 788 goto on1; 789 if (tt->rn_flags & RNF_NORMAL) { 790 if (m->rm_leaf != tt || m->rm_refs > 0) { 791 log(LOG_ERR, "rn_delete: inconsistent annotation\n"); 792 return NULL; /* dangling ref could cause disaster */ 793 } 794 } else { 795 if (m->rm_mask != tt->rn_mask) { 796 log(LOG_ERR, "rn_delete: inconsistent annotation\n"); 797 goto on1; 798 } 799 if (--m->rm_refs >= 0) 800 goto on1; 801 } 802 b = -1 - tt->rn_b; 803 t = saved_tt->rn_p; 804 if (b > t->rn_b) 805 goto on1; /* Wasn't lifted at all */ 806 do { 807 x = t; 808 t = t->rn_p; 809 } while (b <= t->rn_b && x != top); 810 for (mp = &x->rn_mklist; (m = *mp) != NULL; mp = &m->rm_mklist) { 811 if (m == saved_m) { 812 *mp = m->rm_mklist; 813 MKFree(m); 814 break; 815 } 816 } 817 if (m == NULL) { 818 log(LOG_ERR, "rn_delete: couldn't find our annotation\n"); 819 if (tt->rn_flags & RNF_NORMAL) 820 return NULL; /* Dangling ref to us */ 821 } 822 on1: 823 /* 824 * Eliminate us from tree 825 */ 826 if (tt->rn_flags & RNF_ROOT) 827 return NULL; 828 #ifdef RN_DEBUG 829 if (rn_debug) 830 log(LOG_DEBUG, "%s: Going In:\n", __func__), traverse(head, tt); 831 #endif 832 t = tt->rn_p; 833 dupedkey = saved_tt->rn_dupedkey; 834 if (dupedkey != NULL) { 835 /* 836 * Here, tt is the deletion target, and 837 * saved_tt is the head of the dupedkey chain. 838 */ 839 if (tt == saved_tt) { 840 x = dupedkey; 841 x->rn_p = t; 842 if (t->rn_l == tt) 843 t->rn_l = x; 844 else 845 t->rn_r = x; 846 } else { 847 /* find node in front of tt on the chain */ 848 for (x = p = saved_tt; 849 p != NULL && p->rn_dupedkey != tt;) 850 p = p->rn_dupedkey; 851 if (p != NULL) { 852 p->rn_dupedkey = tt->rn_dupedkey; 853 if (tt->rn_dupedkey != NULL) 854 tt->rn_dupedkey->rn_p = p; 855 } else 856 log(LOG_ERR, "rn_delete: couldn't find us\n"); 857 } 858 t = tt + 1; 859 if (t->rn_flags & RNF_ACTIVE) { 860 *++x = *t; 861 p = t->rn_p; 862 if (p->rn_l == t) 863 p->rn_l = x; 864 else 865 p->rn_r = x; 866 x->rn_l->rn_p = x; 867 x->rn_r->rn_p = x; 868 } 869 goto out; 870 } 871 if (t->rn_l == tt) 872 x = t->rn_r; 873 else 874 x = t->rn_l; 875 p = t->rn_p; 876 if (p->rn_r == t) 877 p->rn_r = x; 878 else 879 p->rn_l = x; 880 x->rn_p = p; 881 /* 882 * Demote routes attached to us. 883 */ 884 if (t->rn_mklist == NULL) 885 ; 886 else if (x->rn_b >= 0) { 887 for (mp = &x->rn_mklist; (m = *mp) != NULL; mp = &m->rm_mklist) 888 ; 889 *mp = t->rn_mklist; 890 } else { 891 /* If there are any key,mask pairs in a sibling 892 duped-key chain, some subset will appear sorted 893 in the same order attached to our mklist */ 894 for (m = t->rn_mklist; 895 m != NULL && x != NULL; 896 x = x->rn_dupedkey) { 897 if (m == x->rn_mklist) { 898 struct radix_mask *mm = m->rm_mklist; 899 x->rn_mklist = NULL; 900 if (--(m->rm_refs) < 0) 901 MKFree(m); 902 m = mm; 903 } 904 } 905 if (m != NULL) { 906 log(LOG_ERR, "rn_delete: Orphaned Mask %p at %p\n", 907 m, x); 908 } 909 } 910 /* 911 * We may be holding an active internal node in the tree. 912 */ 913 x = tt + 1; 914 if (t != x) { 915 *t = *x; 916 t->rn_l->rn_p = t; 917 t->rn_r->rn_p = t; 918 p = x->rn_p; 919 if (p->rn_l == x) 920 p->rn_l = t; 921 else 922 p->rn_r = t; 923 } 924 out: 925 #ifdef RN_DEBUG 926 if (rn_debug) { 927 log(LOG_DEBUG, "%s: Coming Out:\n", __func__), 928 traverse(head, tt); 929 } 930 #endif /* RN_DEBUG */ 931 tt->rn_flags &= ~RNF_ACTIVE; 932 tt[1].rn_flags &= ~RNF_ACTIVE; 933 return tt; 934 } 935 936 struct radix_node * 937 rn_delete( 938 const void *v_arg, 939 const void *netmask_arg, 940 struct radix_node_head *head) 941 { 942 return rn_delete1(v_arg, netmask_arg, head, NULL); 943 } 944 945 static struct radix_node * 946 rn_walknext(struct radix_node *rn, rn_printer_t printer, void *arg) 947 { 948 /* If at right child go back up, otherwise, go right */ 949 while (rn->rn_p->rn_r == rn && (rn->rn_flags & RNF_ROOT) == 0) { 950 if (printer != NULL) 951 (*printer)(arg, SUBTREE_CLOSE); 952 rn = rn->rn_p; 953 } 954 if (printer) 955 rn_nodeprint(rn->rn_p, printer, arg, ""); 956 /* Find the next *leaf* since next node might vanish, too */ 957 for (rn = rn->rn_p->rn_r; rn->rn_b >= 0;) { 958 if (printer != NULL) 959 (*printer)(arg, SUBTREE_OPEN); 960 rn = rn->rn_l; 961 } 962 return rn; 963 } 964 965 static struct radix_node * 966 rn_walkfirst(struct radix_node *rn, rn_printer_t printer, void *arg) 967 { 968 /* First time through node, go left */ 969 while (rn->rn_b >= 0) { 970 if (printer != NULL) 971 (*printer)(arg, SUBTREE_OPEN); 972 rn = rn->rn_l; 973 } 974 return rn; 975 } 976 977 int 978 rn_walktree( 979 struct radix_node_head *h, 980 int (*f)(struct radix_node *, void *), 981 void *w) 982 { 983 int error; 984 struct radix_node *base, *next, *rn; 985 /* 986 * This gets complicated because we may delete the node 987 * while applying the function f to it, so we need to calculate 988 * the successor node in advance. 989 */ 990 rn = rn_walkfirst(h->rnh_treetop, NULL, NULL); 991 for (;;) { 992 base = rn; 993 next = rn_walknext(rn, NULL, NULL); 994 /* Process leaves */ 995 while ((rn = base) != NULL) { 996 base = rn->rn_dupedkey; 997 if (!(rn->rn_flags & RNF_ROOT) && (error = (*f)(rn, w))) 998 return error; 999 } 1000 rn = next; 1001 if (rn->rn_flags & RNF_ROOT) 1002 return 0; 1003 } 1004 /* NOTREACHED */ 1005 } 1006 1007 struct delayinit { 1008 void **head; 1009 int off; 1010 SLIST_ENTRY(delayinit) entries; 1011 }; 1012 static SLIST_HEAD(, delayinit) delayinits = SLIST_HEAD_INITIALIZER(delayheads); 1013 static int radix_initialized; 1014 1015 /* 1016 * Initialize a radix tree once radix is initialized. Only for bootstrap. 1017 * Assume that no concurrency protection is necessary at this stage. 1018 */ 1019 void 1020 rn_delayedinit(void **head, int off) 1021 { 1022 struct delayinit *di; 1023 1024 KASSERT(radix_initialized == 0); 1025 1026 di = kmem_alloc(sizeof(*di), KM_SLEEP); 1027 di->head = head; 1028 di->off = off; 1029 SLIST_INSERT_HEAD(&delayinits, di, entries); 1030 } 1031 1032 int 1033 rn_inithead(void **head, int off) 1034 { 1035 struct radix_node_head *rnh; 1036 1037 if (*head != NULL) 1038 return 1; 1039 R_Malloc(rnh, struct radix_node_head *, sizeof (*rnh)); 1040 if (rnh == NULL) 1041 return 0; 1042 *head = rnh; 1043 return rn_inithead0(rnh, off); 1044 } 1045 1046 int 1047 rn_inithead0(struct radix_node_head *rnh, int off) 1048 { 1049 struct radix_node *t; 1050 struct radix_node *tt; 1051 struct radix_node *ttt; 1052 1053 memset(rnh, 0, sizeof(*rnh)); 1054 t = rn_newpair(rn_zeros, off, rnh->rnh_nodes); 1055 ttt = rnh->rnh_nodes + 2; 1056 t->rn_r = ttt; 1057 t->rn_p = t; 1058 tt = t->rn_l; 1059 tt->rn_flags = t->rn_flags = RNF_ROOT | RNF_ACTIVE; 1060 tt->rn_b = -1 - off; 1061 *ttt = *tt; 1062 ttt->rn_key = rn_ones; 1063 rnh->rnh_addaddr = rn_addroute; 1064 rnh->rnh_deladdr = rn_delete; 1065 rnh->rnh_matchaddr = rn_match; 1066 rnh->rnh_lookup = rn_lookup; 1067 rnh->rnh_treetop = t; 1068 return 1; 1069 } 1070 1071 void 1072 rn_init(void) 1073 { 1074 char *cp, *cplim; 1075 struct delayinit *di; 1076 #ifdef _KERNEL 1077 struct domain *dp; 1078 1079 if (radix_initialized) 1080 panic("radix already initialized"); 1081 radix_initialized = 1; 1082 1083 DOMAIN_FOREACH(dp) { 1084 if (dp->dom_maxrtkey > max_keylen) 1085 max_keylen = dp->dom_maxrtkey; 1086 } 1087 #endif 1088 if (max_keylen == 0) { 1089 log(LOG_ERR, 1090 "rn_init: radix functions require max_keylen be set\n"); 1091 return; 1092 } 1093 1094 R_Malloc(rn_zeros, char *, 3 * max_keylen); 1095 if (rn_zeros == NULL) 1096 panic("rn_init"); 1097 memset(rn_zeros, 0, 3 * max_keylen); 1098 rn_ones = cp = rn_zeros + max_keylen; 1099 addmask_key = cplim = rn_ones + max_keylen; 1100 while (cp < cplim) 1101 *cp++ = -1; 1102 if (rn_inithead((void *)&mask_rnhead, 0) == 0) 1103 panic("rn_init 2"); 1104 1105 while ((di = SLIST_FIRST(&delayinits)) != NULL) { 1106 if (!rn_inithead(di->head, di->off)) 1107 panic("delayed rn_inithead failed"); 1108 SLIST_REMOVE_HEAD(&delayinits, entries); 1109 kmem_free(di, sizeof(*di)); 1110 } 1111 } 1112