radix.c revision 257330
1/*- 2 * Copyright (c) 1988, 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)radix.c 8.5 (Berkeley) 5/19/95 30 * $FreeBSD: stable/10/sys/net/radix.c 257330 2013-10-29 12:53:23Z melifaro $ 31 */ 32 33/* 34 * Routines to build and maintain radix trees for routing lookups. 35 */ 36#include <sys/param.h> 37#ifdef _KERNEL 38#include <sys/lock.h> 39#include <sys/mutex.h> 40#include <sys/rwlock.h> 41#include <sys/systm.h> 42#include <sys/malloc.h> 43#include <sys/syslog.h> 44#include <net/radix.h> 45#include "opt_mpath.h" 46#ifdef RADIX_MPATH 47#include <net/radix_mpath.h> 48#endif 49#else /* !_KERNEL */ 50#include <stdio.h> 51#include <strings.h> 52#include <stdlib.h> 53#define log(x, arg...) fprintf(stderr, ## arg) 54#define panic(x) fprintf(stderr, "PANIC: %s", x), exit(1) 55#define min(a, b) ((a) < (b) ? (a) : (b) ) 56#include <net/radix.h> 57#endif /* !_KERNEL */ 58 59static int rn_walktree_from(struct radix_node_head *h, void *a, void *m, 60 walktree_f_t *f, void *w); 61static int rn_walktree(struct radix_node_head *, walktree_f_t *, void *); 62static struct radix_node 63 *rn_insert(void *, struct radix_node_head *, int *, 64 struct radix_node [2]), 65 *rn_newpair(void *, int, struct radix_node[2]), 66 *rn_search(void *, struct radix_node *), 67 *rn_search_m(void *, struct radix_node *, void *); 68 69static void rn_detachhead_internal(void **head); 70static int rn_inithead_internal(void **head, int off); 71 72#define RADIX_MAX_KEY_LEN 32 73 74static char rn_zeros[RADIX_MAX_KEY_LEN]; 75static char rn_ones[RADIX_MAX_KEY_LEN] = { 76 -1, -1, -1, -1, -1, -1, -1, -1, 77 -1, -1, -1, -1, -1, -1, -1, -1, 78 -1, -1, -1, -1, -1, -1, -1, -1, 79 -1, -1, -1, -1, -1, -1, -1, -1, 80}; 81 82/* 83 * XXX: Compat stuff for old rn_addmask() users 84 */ 85static struct radix_node_head *mask_rnhead_compat; 86#ifdef _KERNEL 87static struct mtx mask_mtx; 88#endif 89 90 91static int rn_lexobetter(void *m_arg, void *n_arg); 92static struct radix_mask * 93 rn_new_radix_mask(struct radix_node *tt, 94 struct radix_mask *next); 95static int rn_satisfies_leaf(char *trial, struct radix_node *leaf, 96 int skip); 97 98/* 99 * The data structure for the keys is a radix tree with one way 100 * branching removed. The index rn_bit at an internal node n represents a bit 101 * position to be tested. The tree is arranged so that all descendants 102 * of a node n have keys whose bits all agree up to position rn_bit - 1. 103 * (We say the index of n is rn_bit.) 104 * 105 * There is at least one descendant which has a one bit at position rn_bit, 106 * and at least one with a zero there. 107 * 108 * A route is determined by a pair of key and mask. We require that the 109 * bit-wise logical and of the key and mask to be the key. 110 * We define the index of a route to associated with the mask to be 111 * the first bit number in the mask where 0 occurs (with bit number 0 112 * representing the highest order bit). 113 * 114 * We say a mask is normal if every bit is 0, past the index of the mask. 115 * If a node n has a descendant (k, m) with index(m) == index(n) == rn_bit, 116 * and m is a normal mask, then the route applies to every descendant of n. 117 * If the index(m) < rn_bit, this implies the trailing last few bits of k 118 * before bit b are all 0, (and hence consequently true of every descendant 119 * of n), so the route applies to all descendants of the node as well. 120 * 121 * Similar logic shows that a non-normal mask m such that 122 * index(m) <= index(n) could potentially apply to many children of n. 123 * Thus, for each non-host route, we attach its mask to a list at an internal 124 * node as high in the tree as we can go. 125 * 126 * The present version of the code makes use of normal routes in short- 127 * circuiting an explict mask and compare operation when testing whether 128 * a key satisfies a normal route, and also in remembering the unique leaf 129 * that governs a subtree. 130 */ 131 132/* 133 * Most of the functions in this code assume that the key/mask arguments 134 * are sockaddr-like structures, where the first byte is an u_char 135 * indicating the size of the entire structure. 136 * 137 * To make the assumption more explicit, we use the LEN() macro to access 138 * this field. It is safe to pass an expression with side effects 139 * to LEN() as the argument is evaluated only once. 140 * We cast the result to int as this is the dominant usage. 141 */ 142#define LEN(x) ( (int) (*(const u_char *)(x)) ) 143 144/* 145 * XXX THIS NEEDS TO BE FIXED 146 * In the code, pointers to keys and masks are passed as either 147 * 'void *' (because callers use to pass pointers of various kinds), or 148 * 'caddr_t' (which is fine for pointer arithmetics, but not very 149 * clean when you dereference it to access data). Furthermore, caddr_t 150 * is really 'char *', while the natural type to operate on keys and 151 * masks would be 'u_char'. This mismatch require a lot of casts and 152 * intermediate variables to adapt types that clutter the code. 153 */ 154 155/* 156 * Search a node in the tree matching the key. 157 */ 158static struct radix_node * 159rn_search(v_arg, head) 160 void *v_arg; 161 struct radix_node *head; 162{ 163 register struct radix_node *x; 164 register caddr_t v; 165 166 for (x = head, v = v_arg; x->rn_bit >= 0;) { 167 if (x->rn_bmask & v[x->rn_offset]) 168 x = x->rn_right; 169 else 170 x = x->rn_left; 171 } 172 return (x); 173} 174 175/* 176 * Same as above, but with an additional mask. 177 * XXX note this function is used only once. 178 */ 179static struct radix_node * 180rn_search_m(v_arg, head, m_arg) 181 struct radix_node *head; 182 void *v_arg, *m_arg; 183{ 184 register struct radix_node *x; 185 register caddr_t v = v_arg, m = m_arg; 186 187 for (x = head; x->rn_bit >= 0;) { 188 if ((x->rn_bmask & m[x->rn_offset]) && 189 (x->rn_bmask & v[x->rn_offset])) 190 x = x->rn_right; 191 else 192 x = x->rn_left; 193 } 194 return x; 195} 196 197int 198rn_refines(m_arg, n_arg) 199 void *m_arg, *n_arg; 200{ 201 register caddr_t m = m_arg, n = n_arg; 202 register caddr_t lim, lim2 = lim = n + LEN(n); 203 int longer = LEN(n++) - LEN(m++); 204 int masks_are_equal = 1; 205 206 if (longer > 0) 207 lim -= longer; 208 while (n < lim) { 209 if (*n & ~(*m)) 210 return 0; 211 if (*n++ != *m++) 212 masks_are_equal = 0; 213 } 214 while (n < lim2) 215 if (*n++) 216 return 0; 217 if (masks_are_equal && (longer < 0)) 218 for (lim2 = m - longer; m < lim2; ) 219 if (*m++) 220 return 1; 221 return (!masks_are_equal); 222} 223 224struct radix_node * 225rn_lookup(v_arg, m_arg, head) 226 void *v_arg, *m_arg; 227 struct radix_node_head *head; 228{ 229 register struct radix_node *x; 230 caddr_t netmask = 0; 231 232 if (m_arg) { 233 x = rn_addmask_r(m_arg, head->rnh_masks, 1, 234 head->rnh_treetop->rn_offset); 235 if (x == 0) 236 return (0); 237 netmask = x->rn_key; 238 } 239 x = rn_match(v_arg, head); 240 if (x && netmask) { 241 while (x && x->rn_mask != netmask) 242 x = x->rn_dupedkey; 243 } 244 return x; 245} 246 247static int 248rn_satisfies_leaf(trial, leaf, skip) 249 char *trial; 250 register struct radix_node *leaf; 251 int skip; 252{ 253 register char *cp = trial, *cp2 = leaf->rn_key, *cp3 = leaf->rn_mask; 254 char *cplim; 255 int length = min(LEN(cp), LEN(cp2)); 256 257 if (cp3 == NULL) 258 cp3 = rn_ones; 259 else 260 length = min(length, LEN(cp3)); 261 cplim = cp + length; cp3 += skip; cp2 += skip; 262 for (cp += skip; cp < cplim; cp++, cp2++, cp3++) 263 if ((*cp ^ *cp2) & *cp3) 264 return 0; 265 return 1; 266} 267 268struct radix_node * 269rn_match(v_arg, head) 270 void *v_arg; 271 struct radix_node_head *head; 272{ 273 caddr_t v = v_arg; 274 register struct radix_node *t = head->rnh_treetop, *x; 275 register caddr_t cp = v, cp2; 276 caddr_t cplim; 277 struct radix_node *saved_t, *top = t; 278 int off = t->rn_offset, vlen = LEN(cp), matched_off; 279 register int test, b, rn_bit; 280 281 /* 282 * Open code rn_search(v, top) to avoid overhead of extra 283 * subroutine call. 284 */ 285 for (; t->rn_bit >= 0; ) { 286 if (t->rn_bmask & cp[t->rn_offset]) 287 t = t->rn_right; 288 else 289 t = t->rn_left; 290 } 291 /* 292 * See if we match exactly as a host destination 293 * or at least learn how many bits match, for normal mask finesse. 294 * 295 * It doesn't hurt us to limit how many bytes to check 296 * to the length of the mask, since if it matches we had a genuine 297 * match and the leaf we have is the most specific one anyway; 298 * if it didn't match with a shorter length it would fail 299 * with a long one. This wins big for class B&C netmasks which 300 * are probably the most common case... 301 */ 302 if (t->rn_mask) 303 vlen = *(u_char *)t->rn_mask; 304 cp += off; cp2 = t->rn_key + off; cplim = v + vlen; 305 for (; cp < cplim; cp++, cp2++) 306 if (*cp != *cp2) 307 goto on1; 308 /* 309 * This extra grot is in case we are explicitly asked 310 * to look up the default. Ugh! 311 * 312 * Never return the root node itself, it seems to cause a 313 * lot of confusion. 314 */ 315 if (t->rn_flags & RNF_ROOT) 316 t = t->rn_dupedkey; 317 return t; 318on1: 319 test = (*cp ^ *cp2) & 0xff; /* find first bit that differs */ 320 for (b = 7; (test >>= 1) > 0;) 321 b--; 322 matched_off = cp - v; 323 b += matched_off << 3; 324 rn_bit = -1 - b; 325 /* 326 * If there is a host route in a duped-key chain, it will be first. 327 */ 328 if ((saved_t = t)->rn_mask == 0) 329 t = t->rn_dupedkey; 330 for (; t; t = t->rn_dupedkey) 331 /* 332 * Even if we don't match exactly as a host, 333 * we may match if the leaf we wound up at is 334 * a route to a net. 335 */ 336 if (t->rn_flags & RNF_NORMAL) { 337 if (rn_bit <= t->rn_bit) 338 return t; 339 } else if (rn_satisfies_leaf(v, t, matched_off)) 340 return t; 341 t = saved_t; 342 /* start searching up the tree */ 343 do { 344 register struct radix_mask *m; 345 t = t->rn_parent; 346 m = t->rn_mklist; 347 /* 348 * If non-contiguous masks ever become important 349 * we can restore the masking and open coding of 350 * the search and satisfaction test and put the 351 * calculation of "off" back before the "do". 352 */ 353 while (m) { 354 if (m->rm_flags & RNF_NORMAL) { 355 if (rn_bit <= m->rm_bit) 356 return (m->rm_leaf); 357 } else { 358 off = min(t->rn_offset, matched_off); 359 x = rn_search_m(v, t, m->rm_mask); 360 while (x && x->rn_mask != m->rm_mask) 361 x = x->rn_dupedkey; 362 if (x && rn_satisfies_leaf(v, x, off)) 363 return x; 364 } 365 m = m->rm_mklist; 366 } 367 } while (t != top); 368 return 0; 369} 370 371#ifdef RN_DEBUG 372int rn_nodenum; 373struct radix_node *rn_clist; 374int rn_saveinfo; 375int rn_debug = 1; 376#endif 377 378/* 379 * Whenever we add a new leaf to the tree, we also add a parent node, 380 * so we allocate them as an array of two elements: the first one must be 381 * the leaf (see RNTORT() in route.c), the second one is the parent. 382 * This routine initializes the relevant fields of the nodes, so that 383 * the leaf is the left child of the parent node, and both nodes have 384 * (almost) all all fields filled as appropriate. 385 * (XXX some fields are left unset, see the '#if 0' section). 386 * The function returns a pointer to the parent node. 387 */ 388 389static struct radix_node * 390rn_newpair(v, b, nodes) 391 void *v; 392 int b; 393 struct radix_node nodes[2]; 394{ 395 register struct radix_node *tt = nodes, *t = tt + 1; 396 t->rn_bit = b; 397 t->rn_bmask = 0x80 >> (b & 7); 398 t->rn_left = tt; 399 t->rn_offset = b >> 3; 400 401#if 0 /* XXX perhaps we should fill these fields as well. */ 402 t->rn_parent = t->rn_right = NULL; 403 404 tt->rn_mask = NULL; 405 tt->rn_dupedkey = NULL; 406 tt->rn_bmask = 0; 407#endif 408 tt->rn_bit = -1; 409 tt->rn_key = (caddr_t)v; 410 tt->rn_parent = t; 411 tt->rn_flags = t->rn_flags = RNF_ACTIVE; 412 tt->rn_mklist = t->rn_mklist = 0; 413#ifdef RN_DEBUG 414 tt->rn_info = rn_nodenum++; t->rn_info = rn_nodenum++; 415 tt->rn_twin = t; 416 tt->rn_ybro = rn_clist; 417 rn_clist = tt; 418#endif 419 return t; 420} 421 422static struct radix_node * 423rn_insert(v_arg, head, dupentry, nodes) 424 void *v_arg; 425 struct radix_node_head *head; 426 int *dupentry; 427 struct radix_node nodes[2]; 428{ 429 caddr_t v = v_arg; 430 struct radix_node *top = head->rnh_treetop; 431 int head_off = top->rn_offset, vlen = LEN(v); 432 register struct radix_node *t = rn_search(v_arg, top); 433 register caddr_t cp = v + head_off; 434 register int b; 435 struct radix_node *tt; 436 /* 437 * Find first bit at which v and t->rn_key differ 438 */ 439 { 440 register caddr_t cp2 = t->rn_key + head_off; 441 register int cmp_res; 442 caddr_t cplim = v + vlen; 443 444 while (cp < cplim) 445 if (*cp2++ != *cp++) 446 goto on1; 447 *dupentry = 1; 448 return t; 449on1: 450 *dupentry = 0; 451 cmp_res = (cp[-1] ^ cp2[-1]) & 0xff; 452 for (b = (cp - v) << 3; cmp_res; b--) 453 cmp_res >>= 1; 454 } 455 { 456 register struct radix_node *p, *x = top; 457 cp = v; 458 do { 459 p = x; 460 if (cp[x->rn_offset] & x->rn_bmask) 461 x = x->rn_right; 462 else 463 x = x->rn_left; 464 } while (b > (unsigned) x->rn_bit); 465 /* x->rn_bit < b && x->rn_bit >= 0 */ 466#ifdef RN_DEBUG 467 if (rn_debug) 468 log(LOG_DEBUG, "rn_insert: Going In:\n"), traverse(p); 469#endif 470 t = rn_newpair(v_arg, b, nodes); 471 tt = t->rn_left; 472 if ((cp[p->rn_offset] & p->rn_bmask) == 0) 473 p->rn_left = t; 474 else 475 p->rn_right = t; 476 x->rn_parent = t; 477 t->rn_parent = p; /* frees x, p as temp vars below */ 478 if ((cp[t->rn_offset] & t->rn_bmask) == 0) { 479 t->rn_right = x; 480 } else { 481 t->rn_right = tt; 482 t->rn_left = x; 483 } 484#ifdef RN_DEBUG 485 if (rn_debug) 486 log(LOG_DEBUG, "rn_insert: Coming Out:\n"), traverse(p); 487#endif 488 } 489 return (tt); 490} 491 492struct radix_node * 493rn_addmask_r(void *arg, struct radix_node_head *maskhead, int search, int skip) 494{ 495 caddr_t netmask = (caddr_t)arg; 496 register struct radix_node *x; 497 register caddr_t cp, cplim; 498 register int b = 0, mlen, j; 499 int maskduplicated, isnormal; 500 struct radix_node *saved_x; 501 char addmask_key[RADIX_MAX_KEY_LEN]; 502 503 if ((mlen = LEN(netmask)) > RADIX_MAX_KEY_LEN) 504 mlen = RADIX_MAX_KEY_LEN; 505 if (skip == 0) 506 skip = 1; 507 if (mlen <= skip) 508 return (maskhead->rnh_nodes); 509 510 bzero(addmask_key, RADIX_MAX_KEY_LEN); 511 if (skip > 1) 512 bcopy(rn_ones + 1, addmask_key + 1, skip - 1); 513 bcopy(netmask + skip, addmask_key + skip, mlen - skip); 514 /* 515 * Trim trailing zeroes. 516 */ 517 for (cp = addmask_key + mlen; (cp > addmask_key) && cp[-1] == 0;) 518 cp--; 519 mlen = cp - addmask_key; 520 if (mlen <= skip) 521 return (maskhead->rnh_nodes); 522 *addmask_key = mlen; 523 x = rn_search(addmask_key, maskhead->rnh_treetop); 524 if (bcmp(addmask_key, x->rn_key, mlen) != 0) 525 x = 0; 526 if (x || search) 527 return (x); 528 R_Zalloc(x, struct radix_node *, RADIX_MAX_KEY_LEN + 2 * sizeof (*x)); 529 if ((saved_x = x) == 0) 530 return (0); 531 netmask = cp = (caddr_t)(x + 2); 532 bcopy(addmask_key, cp, mlen); 533 x = rn_insert(cp, maskhead, &maskduplicated, x); 534 if (maskduplicated) { 535 log(LOG_ERR, "rn_addmask: mask impossibly already in tree"); 536 Free(saved_x); 537 return (x); 538 } 539 /* 540 * Calculate index of mask, and check for normalcy. 541 * First find the first byte with a 0 bit, then if there are 542 * more bits left (remember we already trimmed the trailing 0's), 543 * the pattern must be one of those in normal_chars[], or we have 544 * a non-contiguous mask. 545 */ 546 cplim = netmask + mlen; 547 isnormal = 1; 548 for (cp = netmask + skip; (cp < cplim) && *(u_char *)cp == 0xff;) 549 cp++; 550 if (cp != cplim) { 551 static char normal_chars[] = { 552 0, 0x80, 0xc0, 0xe0, 0xf0, 0xf8, 0xfc, 0xfe, 0xff}; 553 554 for (j = 0x80; (j & *cp) != 0; j >>= 1) 555 b++; 556 if (*cp != normal_chars[b] || cp != (cplim - 1)) 557 isnormal = 0; 558 } 559 b += (cp - netmask) << 3; 560 x->rn_bit = -1 - b; 561 if (isnormal) 562 x->rn_flags |= RNF_NORMAL; 563 return (x); 564} 565 566struct radix_node * 567rn_addmask(void *n_arg, int search, int skip) 568{ 569 struct radix_node *tt; 570 571#ifdef _KERNEL 572 mtx_lock(&mask_mtx); 573#endif 574 tt = rn_addmask_r(&mask_rnhead_compat, n_arg, search, skip); 575 576#ifdef _KERNEL 577 mtx_unlock(&mask_mtx); 578#endif 579 580 return (tt); 581} 582 583static int /* XXX: arbitrary ordering for non-contiguous masks */ 584rn_lexobetter(m_arg, n_arg) 585 void *m_arg, *n_arg; 586{ 587 register u_char *mp = m_arg, *np = n_arg, *lim; 588 589 if (LEN(mp) > LEN(np)) 590 return 1; /* not really, but need to check longer one first */ 591 if (LEN(mp) == LEN(np)) 592 for (lim = mp + LEN(mp); mp < lim;) 593 if (*mp++ > *np++) 594 return 1; 595 return 0; 596} 597 598static struct radix_mask * 599rn_new_radix_mask(tt, next) 600 register struct radix_node *tt; 601 register struct radix_mask *next; 602{ 603 register struct radix_mask *m; 604 605 R_Malloc(m, struct radix_mask *, sizeof (struct radix_mask)); 606 if (m == 0) { 607 log(LOG_ERR, "Failed to allocate route mask\n"); 608 return (0); 609 } 610 bzero(m, sizeof(*m)); 611 m->rm_bit = tt->rn_bit; 612 m->rm_flags = tt->rn_flags; 613 if (tt->rn_flags & RNF_NORMAL) 614 m->rm_leaf = tt; 615 else 616 m->rm_mask = tt->rn_mask; 617 m->rm_mklist = next; 618 tt->rn_mklist = m; 619 return m; 620} 621 622struct radix_node * 623rn_addroute(v_arg, n_arg, head, treenodes) 624 void *v_arg, *n_arg; 625 struct radix_node_head *head; 626 struct radix_node treenodes[2]; 627{ 628 caddr_t v = (caddr_t)v_arg, netmask = (caddr_t)n_arg; 629 register struct radix_node *t, *x = 0, *tt; 630 struct radix_node *saved_tt, *top = head->rnh_treetop; 631 short b = 0, b_leaf = 0; 632 int keyduplicated; 633 caddr_t mmask; 634 struct radix_mask *m, **mp; 635 636 /* 637 * In dealing with non-contiguous masks, there may be 638 * many different routes which have the same mask. 639 * We will find it useful to have a unique pointer to 640 * the mask to speed avoiding duplicate references at 641 * nodes and possibly save time in calculating indices. 642 */ 643 if (netmask) { 644 x = rn_addmask_r(netmask, head->rnh_masks, 0, top->rn_offset); 645 if (x == NULL) 646 return (0); 647 b_leaf = x->rn_bit; 648 b = -1 - x->rn_bit; 649 netmask = x->rn_key; 650 } 651 /* 652 * Deal with duplicated keys: attach node to previous instance 653 */ 654 saved_tt = tt = rn_insert(v, head, &keyduplicated, treenodes); 655 if (keyduplicated) { 656 for (t = tt; tt; t = tt, tt = tt->rn_dupedkey) { 657#ifdef RADIX_MPATH 658 /* permit multipath, if enabled for the family */ 659 if (rn_mpath_capable(head) && netmask == tt->rn_mask) { 660 /* 661 * go down to the end of multipaths, so that 662 * new entry goes into the end of rn_dupedkey 663 * chain. 664 */ 665 do { 666 t = tt; 667 tt = tt->rn_dupedkey; 668 } while (tt && t->rn_mask == tt->rn_mask); 669 break; 670 } 671#endif 672 if (tt->rn_mask == netmask) 673 return (0); 674 if (netmask == 0 || 675 (tt->rn_mask && 676 ((b_leaf < tt->rn_bit) /* index(netmask) > node */ 677 || rn_refines(netmask, tt->rn_mask) 678 || rn_lexobetter(netmask, tt->rn_mask)))) 679 break; 680 } 681 /* 682 * If the mask is not duplicated, we wouldn't 683 * find it among possible duplicate key entries 684 * anyway, so the above test doesn't hurt. 685 * 686 * We sort the masks for a duplicated key the same way as 687 * in a masklist -- most specific to least specific. 688 * This may require the unfortunate nuisance of relocating 689 * the head of the list. 690 * 691 * We also reverse, or doubly link the list through the 692 * parent pointer. 693 */ 694 if (tt == saved_tt) { 695 struct radix_node *xx = x; 696 /* link in at head of list */ 697 (tt = treenodes)->rn_dupedkey = t; 698 tt->rn_flags = t->rn_flags; 699 tt->rn_parent = x = t->rn_parent; 700 t->rn_parent = tt; /* parent */ 701 if (x->rn_left == t) 702 x->rn_left = tt; 703 else 704 x->rn_right = tt; 705 saved_tt = tt; x = xx; 706 } else { 707 (tt = treenodes)->rn_dupedkey = t->rn_dupedkey; 708 t->rn_dupedkey = tt; 709 tt->rn_parent = t; /* parent */ 710 if (tt->rn_dupedkey) /* parent */ 711 tt->rn_dupedkey->rn_parent = tt; /* parent */ 712 } 713#ifdef RN_DEBUG 714 t=tt+1; tt->rn_info = rn_nodenum++; t->rn_info = rn_nodenum++; 715 tt->rn_twin = t; tt->rn_ybro = rn_clist; rn_clist = tt; 716#endif 717 tt->rn_key = (caddr_t) v; 718 tt->rn_bit = -1; 719 tt->rn_flags = RNF_ACTIVE; 720 } 721 /* 722 * Put mask in tree. 723 */ 724 if (netmask) { 725 tt->rn_mask = netmask; 726 tt->rn_bit = x->rn_bit; 727 tt->rn_flags |= x->rn_flags & RNF_NORMAL; 728 } 729 t = saved_tt->rn_parent; 730 if (keyduplicated) 731 goto on2; 732 b_leaf = -1 - t->rn_bit; 733 if (t->rn_right == saved_tt) 734 x = t->rn_left; 735 else 736 x = t->rn_right; 737 /* Promote general routes from below */ 738 if (x->rn_bit < 0) { 739 for (mp = &t->rn_mklist; x; x = x->rn_dupedkey) 740 if (x->rn_mask && (x->rn_bit >= b_leaf) && x->rn_mklist == 0) { 741 *mp = m = rn_new_radix_mask(x, 0); 742 if (m) 743 mp = &m->rm_mklist; 744 } 745 } else if (x->rn_mklist) { 746 /* 747 * Skip over masks whose index is > that of new node 748 */ 749 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist) 750 if (m->rm_bit >= b_leaf) 751 break; 752 t->rn_mklist = m; *mp = 0; 753 } 754on2: 755 /* Add new route to highest possible ancestor's list */ 756 if ((netmask == 0) || (b > t->rn_bit )) 757 return tt; /* can't lift at all */ 758 b_leaf = tt->rn_bit; 759 do { 760 x = t; 761 t = t->rn_parent; 762 } while (b <= t->rn_bit && x != top); 763 /* 764 * Search through routes associated with node to 765 * insert new route according to index. 766 * Need same criteria as when sorting dupedkeys to avoid 767 * double loop on deletion. 768 */ 769 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist) { 770 if (m->rm_bit < b_leaf) 771 continue; 772 if (m->rm_bit > b_leaf) 773 break; 774 if (m->rm_flags & RNF_NORMAL) { 775 mmask = m->rm_leaf->rn_mask; 776 if (tt->rn_flags & RNF_NORMAL) { 777#if !defined(RADIX_MPATH) 778 log(LOG_ERR, 779 "Non-unique normal route, mask not entered\n"); 780#endif 781 return tt; 782 } 783 } else 784 mmask = m->rm_mask; 785 if (mmask == netmask) { 786 m->rm_refs++; 787 tt->rn_mklist = m; 788 return tt; 789 } 790 if (rn_refines(netmask, mmask) 791 || rn_lexobetter(netmask, mmask)) 792 break; 793 } 794 *mp = rn_new_radix_mask(tt, *mp); 795 return tt; 796} 797 798struct radix_node * 799rn_delete(v_arg, netmask_arg, head) 800 void *v_arg, *netmask_arg; 801 struct radix_node_head *head; 802{ 803 register struct radix_node *t, *p, *x, *tt; 804 struct radix_mask *m, *saved_m, **mp; 805 struct radix_node *dupedkey, *saved_tt, *top; 806 caddr_t v, netmask; 807 int b, head_off, vlen; 808 809 v = v_arg; 810 netmask = netmask_arg; 811 x = head->rnh_treetop; 812 tt = rn_search(v, x); 813 head_off = x->rn_offset; 814 vlen = LEN(v); 815 saved_tt = tt; 816 top = x; 817 if (tt == 0 || 818 bcmp(v + head_off, tt->rn_key + head_off, vlen - head_off)) 819 return (0); 820 /* 821 * Delete our route from mask lists. 822 */ 823 if (netmask) { 824 x = rn_addmask_r(netmask, head->rnh_masks, 1, head_off); 825 if (x == NULL) 826 return (0); 827 netmask = x->rn_key; 828 while (tt->rn_mask != netmask) 829 if ((tt = tt->rn_dupedkey) == 0) 830 return (0); 831 } 832 if (tt->rn_mask == 0 || (saved_m = m = tt->rn_mklist) == 0) 833 goto on1; 834 if (tt->rn_flags & RNF_NORMAL) { 835 if (m->rm_leaf != tt || m->rm_refs > 0) { 836 log(LOG_ERR, "rn_delete: inconsistent annotation\n"); 837 return 0; /* dangling ref could cause disaster */ 838 } 839 } else { 840 if (m->rm_mask != tt->rn_mask) { 841 log(LOG_ERR, "rn_delete: inconsistent annotation\n"); 842 goto on1; 843 } 844 if (--m->rm_refs >= 0) 845 goto on1; 846 } 847 b = -1 - tt->rn_bit; 848 t = saved_tt->rn_parent; 849 if (b > t->rn_bit) 850 goto on1; /* Wasn't lifted at all */ 851 do { 852 x = t; 853 t = t->rn_parent; 854 } while (b <= t->rn_bit && x != top); 855 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist) 856 if (m == saved_m) { 857 *mp = m->rm_mklist; 858 Free(m); 859 break; 860 } 861 if (m == 0) { 862 log(LOG_ERR, "rn_delete: couldn't find our annotation\n"); 863 if (tt->rn_flags & RNF_NORMAL) 864 return (0); /* Dangling ref to us */ 865 } 866on1: 867 /* 868 * Eliminate us from tree 869 */ 870 if (tt->rn_flags & RNF_ROOT) 871 return (0); 872#ifdef RN_DEBUG 873 /* Get us out of the creation list */ 874 for (t = rn_clist; t && t->rn_ybro != tt; t = t->rn_ybro) {} 875 if (t) t->rn_ybro = tt->rn_ybro; 876#endif 877 t = tt->rn_parent; 878 dupedkey = saved_tt->rn_dupedkey; 879 if (dupedkey) { 880 /* 881 * Here, tt is the deletion target and 882 * saved_tt is the head of the dupekey chain. 883 */ 884 if (tt == saved_tt) { 885 /* remove from head of chain */ 886 x = dupedkey; x->rn_parent = t; 887 if (t->rn_left == tt) 888 t->rn_left = x; 889 else 890 t->rn_right = x; 891 } else { 892 /* find node in front of tt on the chain */ 893 for (x = p = saved_tt; p && p->rn_dupedkey != tt;) 894 p = p->rn_dupedkey; 895 if (p) { 896 p->rn_dupedkey = tt->rn_dupedkey; 897 if (tt->rn_dupedkey) /* parent */ 898 tt->rn_dupedkey->rn_parent = p; 899 /* parent */ 900 } else log(LOG_ERR, "rn_delete: couldn't find us\n"); 901 } 902 t = tt + 1; 903 if (t->rn_flags & RNF_ACTIVE) { 904#ifndef RN_DEBUG 905 *++x = *t; 906 p = t->rn_parent; 907#else 908 b = t->rn_info; 909 *++x = *t; 910 t->rn_info = b; 911 p = t->rn_parent; 912#endif 913 if (p->rn_left == t) 914 p->rn_left = x; 915 else 916 p->rn_right = x; 917 x->rn_left->rn_parent = x; 918 x->rn_right->rn_parent = x; 919 } 920 goto out; 921 } 922 if (t->rn_left == tt) 923 x = t->rn_right; 924 else 925 x = t->rn_left; 926 p = t->rn_parent; 927 if (p->rn_right == t) 928 p->rn_right = x; 929 else 930 p->rn_left = x; 931 x->rn_parent = p; 932 /* 933 * Demote routes attached to us. 934 */ 935 if (t->rn_mklist) { 936 if (x->rn_bit >= 0) { 937 for (mp = &x->rn_mklist; (m = *mp);) 938 mp = &m->rm_mklist; 939 *mp = t->rn_mklist; 940 } else { 941 /* If there are any key,mask pairs in a sibling 942 duped-key chain, some subset will appear sorted 943 in the same order attached to our mklist */ 944 for (m = t->rn_mklist; m && x; x = x->rn_dupedkey) 945 if (m == x->rn_mklist) { 946 struct radix_mask *mm = m->rm_mklist; 947 x->rn_mklist = 0; 948 if (--(m->rm_refs) < 0) 949 Free(m); 950 m = mm; 951 } 952 if (m) 953 log(LOG_ERR, 954 "rn_delete: Orphaned Mask %p at %p\n", 955 m, x); 956 } 957 } 958 /* 959 * We may be holding an active internal node in the tree. 960 */ 961 x = tt + 1; 962 if (t != x) { 963#ifndef RN_DEBUG 964 *t = *x; 965#else 966 b = t->rn_info; 967 *t = *x; 968 t->rn_info = b; 969#endif 970 t->rn_left->rn_parent = t; 971 t->rn_right->rn_parent = t; 972 p = x->rn_parent; 973 if (p->rn_left == x) 974 p->rn_left = t; 975 else 976 p->rn_right = t; 977 } 978out: 979 tt->rn_flags &= ~RNF_ACTIVE; 980 tt[1].rn_flags &= ~RNF_ACTIVE; 981 return (tt); 982} 983 984/* 985 * This is the same as rn_walktree() except for the parameters and the 986 * exit. 987 */ 988static int 989rn_walktree_from(h, a, m, f, w) 990 struct radix_node_head *h; 991 void *a, *m; 992 walktree_f_t *f; 993 void *w; 994{ 995 int error; 996 struct radix_node *base, *next; 997 u_char *xa = (u_char *)a; 998 u_char *xm = (u_char *)m; 999 register struct radix_node *rn, *last = 0 /* shut up gcc */; 1000 int stopping = 0; 1001 int lastb; 1002 1003 /* 1004 * rn_search_m is sort-of-open-coded here. We cannot use the 1005 * function because we need to keep track of the last node seen. 1006 */ 1007 /* printf("about to search\n"); */ 1008 for (rn = h->rnh_treetop; rn->rn_bit >= 0; ) { 1009 last = rn; 1010 /* printf("rn_bit %d, rn_bmask %x, xm[rn_offset] %x\n", 1011 rn->rn_bit, rn->rn_bmask, xm[rn->rn_offset]); */ 1012 if (!(rn->rn_bmask & xm[rn->rn_offset])) { 1013 break; 1014 } 1015 if (rn->rn_bmask & xa[rn->rn_offset]) { 1016 rn = rn->rn_right; 1017 } else { 1018 rn = rn->rn_left; 1019 } 1020 } 1021 /* printf("done searching\n"); */ 1022 1023 /* 1024 * Two cases: either we stepped off the end of our mask, 1025 * in which case last == rn, or we reached a leaf, in which 1026 * case we want to start from the last node we looked at. 1027 * Either way, last is the node we want to start from. 1028 */ 1029 rn = last; 1030 lastb = rn->rn_bit; 1031 1032 /* printf("rn %p, lastb %d\n", rn, lastb);*/ 1033 1034 /* 1035 * This gets complicated because we may delete the node 1036 * while applying the function f to it, so we need to calculate 1037 * the successor node in advance. 1038 */ 1039 while (rn->rn_bit >= 0) 1040 rn = rn->rn_left; 1041 1042 while (!stopping) { 1043 /* printf("node %p (%d)\n", rn, rn->rn_bit); */ 1044 base = rn; 1045 /* If at right child go back up, otherwise, go right */ 1046 while (rn->rn_parent->rn_right == rn 1047 && !(rn->rn_flags & RNF_ROOT)) { 1048 rn = rn->rn_parent; 1049 1050 /* if went up beyond last, stop */ 1051 if (rn->rn_bit <= lastb) { 1052 stopping = 1; 1053 /* printf("up too far\n"); */ 1054 /* 1055 * XXX we should jump to the 'Process leaves' 1056 * part, because the values of 'rn' and 'next' 1057 * we compute will not be used. Not a big deal 1058 * because this loop will terminate, but it is 1059 * inefficient and hard to understand! 1060 */ 1061 } 1062 } 1063 1064 /* 1065 * At the top of the tree, no need to traverse the right 1066 * half, prevent the traversal of the entire tree in the 1067 * case of default route. 1068 */ 1069 if (rn->rn_parent->rn_flags & RNF_ROOT) 1070 stopping = 1; 1071 1072 /* Find the next *leaf* since next node might vanish, too */ 1073 for (rn = rn->rn_parent->rn_right; rn->rn_bit >= 0;) 1074 rn = rn->rn_left; 1075 next = rn; 1076 /* Process leaves */ 1077 while ((rn = base) != 0) { 1078 base = rn->rn_dupedkey; 1079 /* printf("leaf %p\n", rn); */ 1080 if (!(rn->rn_flags & RNF_ROOT) 1081 && (error = (*f)(rn, w))) 1082 return (error); 1083 } 1084 rn = next; 1085 1086 if (rn->rn_flags & RNF_ROOT) { 1087 /* printf("root, stopping"); */ 1088 stopping = 1; 1089 } 1090 1091 } 1092 return 0; 1093} 1094 1095static int 1096rn_walktree(h, f, w) 1097 struct radix_node_head *h; 1098 walktree_f_t *f; 1099 void *w; 1100{ 1101 int error; 1102 struct radix_node *base, *next; 1103 register struct radix_node *rn = h->rnh_treetop; 1104 /* 1105 * This gets complicated because we may delete the node 1106 * while applying the function f to it, so we need to calculate 1107 * the successor node in advance. 1108 */ 1109 1110 /* First time through node, go left */ 1111 while (rn->rn_bit >= 0) 1112 rn = rn->rn_left; 1113 for (;;) { 1114 base = rn; 1115 /* If at right child go back up, otherwise, go right */ 1116 while (rn->rn_parent->rn_right == rn 1117 && (rn->rn_flags & RNF_ROOT) == 0) 1118 rn = rn->rn_parent; 1119 /* Find the next *leaf* since next node might vanish, too */ 1120 for (rn = rn->rn_parent->rn_right; rn->rn_bit >= 0;) 1121 rn = rn->rn_left; 1122 next = rn; 1123 /* Process leaves */ 1124 while ((rn = base)) { 1125 base = rn->rn_dupedkey; 1126 if (!(rn->rn_flags & RNF_ROOT) 1127 && (error = (*f)(rn, w))) 1128 return (error); 1129 } 1130 rn = next; 1131 if (rn->rn_flags & RNF_ROOT) 1132 return (0); 1133 } 1134 /* NOTREACHED */ 1135} 1136 1137/* 1138 * Allocate and initialize an empty tree. This has 3 nodes, which are 1139 * part of the radix_node_head (in the order <left,root,right>) and are 1140 * marked RNF_ROOT so they cannot be freed. 1141 * The leaves have all-zero and all-one keys, with significant 1142 * bits starting at 'off'. 1143 * Return 1 on success, 0 on error. 1144 */ 1145static int 1146rn_inithead_internal(void **head, int off) 1147{ 1148 register struct radix_node_head *rnh; 1149 register struct radix_node *t, *tt, *ttt; 1150 if (*head) 1151 return (1); 1152 R_Zalloc(rnh, struct radix_node_head *, sizeof (*rnh)); 1153 if (rnh == 0) 1154 return (0); 1155#ifdef _KERNEL 1156 RADIX_NODE_HEAD_LOCK_INIT(rnh); 1157#endif 1158 *head = rnh; 1159 t = rn_newpair(rn_zeros, off, rnh->rnh_nodes); 1160 ttt = rnh->rnh_nodes + 2; 1161 t->rn_right = ttt; 1162 t->rn_parent = t; 1163 tt = t->rn_left; /* ... which in turn is rnh->rnh_nodes */ 1164 tt->rn_flags = t->rn_flags = RNF_ROOT | RNF_ACTIVE; 1165 tt->rn_bit = -1 - off; 1166 *ttt = *tt; 1167 ttt->rn_key = rn_ones; 1168 rnh->rnh_addaddr = rn_addroute; 1169 rnh->rnh_deladdr = rn_delete; 1170 rnh->rnh_matchaddr = rn_match; 1171 rnh->rnh_lookup = rn_lookup; 1172 rnh->rnh_walktree = rn_walktree; 1173 rnh->rnh_walktree_from = rn_walktree_from; 1174 rnh->rnh_treetop = t; 1175 return (1); 1176} 1177 1178static void 1179rn_detachhead_internal(void **head) 1180{ 1181 struct radix_node_head *rnh; 1182 1183 KASSERT((head != NULL && *head != NULL), 1184 ("%s: head already freed", __func__)); 1185 rnh = *head; 1186 1187 /* Free <left,root,right> nodes. */ 1188 Free(rnh); 1189 1190 *head = NULL; 1191} 1192 1193int 1194rn_inithead(void **head, int off) 1195{ 1196 struct radix_node_head *rnh; 1197 1198 if (*head != NULL) 1199 return (1); 1200 1201 if (rn_inithead_internal(head, off) == 0) 1202 return (0); 1203 1204 rnh = (struct radix_node_head *)(*head); 1205 1206 if (rn_inithead_internal((void **)&rnh->rnh_masks, 0) == 0) { 1207 rn_detachhead_internal(head); 1208 return (0); 1209 } 1210 1211 return (1); 1212} 1213 1214int 1215rn_detachhead(void **head) 1216{ 1217 struct radix_node_head *rnh; 1218 1219 KASSERT((head != NULL && *head != NULL), 1220 ("%s: head already freed", __func__)); 1221 1222 rnh = *head; 1223 1224 rn_detachhead_internal((void **)&rnh->rnh_masks); 1225 rn_detachhead_internal(head); 1226 return (1); 1227} 1228 1229void 1230rn_init(int maxk) 1231{ 1232 if ((maxk <= 0) || (maxk > RADIX_MAX_KEY_LEN)) { 1233 log(LOG_ERR, 1234 "rn_init: max_keylen must be within 1..%d\n", 1235 RADIX_MAX_KEY_LEN); 1236 return; 1237 } 1238 1239 /* 1240 * XXX: Compat for old rn_addmask() users 1241 */ 1242 if (rn_inithead((void **)(void *)&mask_rnhead_compat, 0) == 0) 1243 panic("rn_init 2"); 1244#ifdef _KERNEL 1245 mtx_init(&mask_mtx, "radix_mask", NULL, MTX_DEF); 1246#endif 1247} 1248