radix.c revision 186166
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: head/sys/net/radix.c 186166 2008-12-16 04:40:43Z kmacy $ 31 */ 32 33/* 34 * Routines to build and maintain radix trees for routing lookups. 35 */ 36#ifndef _RADIX_H_ 37#include <sys/param.h> 38#ifdef _KERNEL 39#include <sys/lock.h> 40#include <sys/mutex.h> 41#include <sys/rwlock.h> 42#include <sys/systm.h> 43#include <sys/malloc.h> 44#include <sys/domain.h> 45#else 46#include <stdlib.h> 47#endif 48#include <sys/syslog.h> 49#include <net/radix.h> 50#endif 51 52#include "opt_mpath.h" 53 54#ifdef RADIX_MPATH 55#include <net/radix_mpath.h> 56#endif 57 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 int max_keylen; 70static struct radix_mask *rn_mkfreelist; 71static struct radix_node_head *mask_rnhead; 72/* 73 * Work area -- the following point to 3 buffers of size max_keylen, 74 * allocated in this order in a block of memory malloc'ed by rn_init. 75 */ 76static char *rn_zeros, *rn_ones, *addmask_key; 77 78#define MKGet(m) { \ 79 if (rn_mkfreelist) { \ 80 m = rn_mkfreelist; \ 81 rn_mkfreelist = (m)->rm_mklist; \ 82 } else \ 83 R_Malloc(m, struct radix_mask *, sizeof (struct radix_mask)); } 84 85#define MKFree(m) { (m)->rm_mklist = rn_mkfreelist; rn_mkfreelist = (m);} 86 87#define rn_masktop (mask_rnhead->rnh_treetop) 88 89static int rn_lexobetter(void *m_arg, void *n_arg); 90static struct radix_mask * 91 rn_new_radix_mask(struct radix_node *tt, 92 struct radix_mask *next); 93static int rn_satisfies_leaf(char *trial, struct radix_node *leaf, 94 int skip); 95 96/* 97 * The data structure for the keys is a radix tree with one way 98 * branching removed. The index rn_bit at an internal node n represents a bit 99 * position to be tested. The tree is arranged so that all descendants 100 * of a node n have keys whose bits all agree up to position rn_bit - 1. 101 * (We say the index of n is rn_bit.) 102 * 103 * There is at least one descendant which has a one bit at position rn_bit, 104 * and at least one with a zero there. 105 * 106 * A route is determined by a pair of key and mask. We require that the 107 * bit-wise logical and of the key and mask to be the key. 108 * We define the index of a route to associated with the mask to be 109 * the first bit number in the mask where 0 occurs (with bit number 0 110 * representing the highest order bit). 111 * 112 * We say a mask is normal if every bit is 0, past the index of the mask. 113 * If a node n has a descendant (k, m) with index(m) == index(n) == rn_bit, 114 * and m is a normal mask, then the route applies to every descendant of n. 115 * If the index(m) < rn_bit, this implies the trailing last few bits of k 116 * before bit b are all 0, (and hence consequently true of every descendant 117 * of n), so the route applies to all descendants of the node as well. 118 * 119 * Similar logic shows that a non-normal mask m such that 120 * index(m) <= index(n) could potentially apply to many children of n. 121 * Thus, for each non-host route, we attach its mask to a list at an internal 122 * node as high in the tree as we can go. 123 * 124 * The present version of the code makes use of normal routes in short- 125 * circuiting an explict mask and compare operation when testing whether 126 * a key satisfies a normal route, and also in remembering the unique leaf 127 * that governs a subtree. 128 */ 129 130/* 131 * Most of the functions in this code assume that the key/mask arguments 132 * are sockaddr-like structures, where the first byte is an u_char 133 * indicating the size of the entire structure. 134 * 135 * To make the assumption more explicit, we use the LEN() macro to access 136 * this field. It is safe to pass an expression with side effects 137 * to LEN() as the argument is evaluated only once. 138 */ 139#define LEN(x) (*(const u_char *)(x)) 140 141/* 142 * XXX THIS NEEDS TO BE FIXED 143 * In the code, pointers to keys and masks are passed as either 144 * 'void *' (because callers use to pass pointers of various kinds), or 145 * 'caddr_t' (which is fine for pointer arithmetics, but not very 146 * clean when you dereference it to access data). Furthermore, caddr_t 147 * is really 'char *', while the natural type to operate on keys and 148 * masks would be 'u_char'. This mismatch require a lot of casts and 149 * intermediate variables to adapt types that clutter the code. 150 */ 151 152/* 153 * Search a node in the tree matching the key. 154 */ 155static struct radix_node * 156rn_search(v_arg, head) 157 void *v_arg; 158 struct radix_node *head; 159{ 160 register struct radix_node *x; 161 register caddr_t v; 162 163 for (x = head, v = v_arg; x->rn_bit >= 0;) { 164 if (x->rn_bmask & v[x->rn_offset]) 165 x = x->rn_right; 166 else 167 x = x->rn_left; 168 } 169 return (x); 170} 171 172/* 173 * Same as above, but with an additional mask. 174 * XXX note this function is used only once. 175 */ 176static struct radix_node * 177rn_search_m(v_arg, head, m_arg) 178 struct radix_node *head; 179 void *v_arg, *m_arg; 180{ 181 register struct radix_node *x; 182 register caddr_t v = v_arg, m = m_arg; 183 184 for (x = head; x->rn_bit >= 0;) { 185 if ((x->rn_bmask & m[x->rn_offset]) && 186 (x->rn_bmask & v[x->rn_offset])) 187 x = x->rn_right; 188 else 189 x = x->rn_left; 190 } 191 return x; 192} 193 194int 195rn_refines(m_arg, n_arg) 196 void *m_arg, *n_arg; 197{ 198 register caddr_t m = m_arg, n = n_arg; 199 register caddr_t lim, lim2 = lim = n + LEN(n); 200 int longer = LEN(n++) - (int)LEN(m++); 201 int masks_are_equal = 1; 202 203 if (longer > 0) 204 lim -= longer; 205 while (n < lim) { 206 if (*n & ~(*m)) 207 return 0; 208 if (*n++ != *m++) 209 masks_are_equal = 0; 210 } 211 while (n < lim2) 212 if (*n++) 213 return 0; 214 if (masks_are_equal && (longer < 0)) 215 for (lim2 = m - longer; m < lim2; ) 216 if (*m++) 217 return 1; 218 return (!masks_are_equal); 219} 220 221struct radix_node * 222rn_lookup(v_arg, m_arg, head) 223 void *v_arg, *m_arg; 224 struct radix_node_head *head; 225{ 226 register struct radix_node *x; 227 caddr_t netmask = 0; 228 229 if (m_arg) { 230 x = rn_addmask(m_arg, 1, head->rnh_treetop->rn_offset); 231 if (x == 0) 232 return (0); 233 netmask = x->rn_key; 234 } 235 x = rn_match(v_arg, head); 236 if (x && netmask) { 237 while (x && x->rn_mask != netmask) 238 x = x->rn_dupedkey; 239 } 240 return x; 241} 242 243static int 244rn_satisfies_leaf(trial, leaf, skip) 245 char *trial; 246 register struct radix_node *leaf; 247 int skip; 248{ 249 register char *cp = trial, *cp2 = leaf->rn_key, *cp3 = leaf->rn_mask; 250 char *cplim; 251 int length = min(LEN(cp), LEN(cp2)); 252 253 if (cp3 == 0) 254 cp3 = rn_ones; 255 else 256 length = min(length, *(u_char *)cp3); 257 cplim = cp + length; cp3 += skip; cp2 += skip; 258 for (cp += skip; cp < cplim; cp++, cp2++, cp3++) 259 if ((*cp ^ *cp2) & *cp3) 260 return 0; 261 return 1; 262} 263 264struct radix_node * 265rn_match(v_arg, head) 266 void *v_arg; 267 struct radix_node_head *head; 268{ 269 caddr_t v = v_arg; 270 register struct radix_node *t = head->rnh_treetop, *x; 271 register caddr_t cp = v, cp2; 272 caddr_t cplim; 273 struct radix_node *saved_t, *top = t; 274 int off = t->rn_offset, vlen = LEN(cp), matched_off; 275 register int test, b, rn_bit; 276 277 RADIX_NODE_HEAD_LOCK_ASSERT(head); 278 /* 279 * Open code rn_search(v, top) to avoid overhead of extra 280 * subroutine call. 281 */ 282 for (; t->rn_bit >= 0; ) { 283 if (t->rn_bmask & cp[t->rn_offset]) 284 t = t->rn_right; 285 else 286 t = t->rn_left; 287 } 288 /* 289 * See if we match exactly as a host destination 290 * or at least learn how many bits match, for normal mask finesse. 291 * 292 * It doesn't hurt us to limit how many bytes to check 293 * to the length of the mask, since if it matches we had a genuine 294 * match and the leaf we have is the most specific one anyway; 295 * if it didn't match with a shorter length it would fail 296 * with a long one. This wins big for class B&C netmasks which 297 * are probably the most common case... 298 */ 299 if (t->rn_mask) 300 vlen = *(u_char *)t->rn_mask; 301 cp += off; cp2 = t->rn_key + off; cplim = v + vlen; 302 for (; cp < cplim; cp++, cp2++) 303 if (*cp != *cp2) 304 goto on1; 305 /* 306 * This extra grot is in case we are explicitly asked 307 * to look up the default. Ugh! 308 * 309 * Never return the root node itself, it seems to cause a 310 * lot of confusion. 311 */ 312 if (t->rn_flags & RNF_ROOT) 313 t = t->rn_dupedkey; 314 return t; 315on1: 316 test = (*cp ^ *cp2) & 0xff; /* find first bit that differs */ 317 for (b = 7; (test >>= 1) > 0;) 318 b--; 319 matched_off = cp - v; 320 b += matched_off << 3; 321 rn_bit = -1 - b; 322 /* 323 * If there is a host route in a duped-key chain, it will be first. 324 */ 325 if ((saved_t = t)->rn_mask == 0) 326 t = t->rn_dupedkey; 327 for (; t; t = t->rn_dupedkey) 328 /* 329 * Even if we don't match exactly as a host, 330 * we may match if the leaf we wound up at is 331 * a route to a net. 332 */ 333 if (t->rn_flags & RNF_NORMAL) { 334 if (rn_bit <= t->rn_bit) 335 return t; 336 } else if (rn_satisfies_leaf(v, t, matched_off)) 337 return t; 338 t = saved_t; 339 /* start searching up the tree */ 340 do { 341 register struct radix_mask *m; 342 t = t->rn_parent; 343 m = t->rn_mklist; 344 /* 345 * If non-contiguous masks ever become important 346 * we can restore the masking and open coding of 347 * the search and satisfaction test and put the 348 * calculation of "off" back before the "do". 349 */ 350 while (m) { 351 if (m->rm_flags & RNF_NORMAL) { 352 if (rn_bit <= m->rm_bit) 353 return (m->rm_leaf); 354 } else { 355 off = min(t->rn_offset, matched_off); 356 x = rn_search_m(v, t, m->rm_mask); 357 while (x && x->rn_mask != m->rm_mask) 358 x = x->rn_dupedkey; 359 if (x && rn_satisfies_leaf(v, x, off)) 360 return x; 361 } 362 m = m->rm_mklist; 363 } 364 } while (t != top); 365 return 0; 366} 367 368#ifdef RN_DEBUG 369int rn_nodenum; 370struct radix_node *rn_clist; 371int rn_saveinfo; 372int rn_debug = 1; 373#endif 374 375/* 376 * Whenever we add a new leaf to the tree, we also add a parent node, 377 * so we allocate them as an array of two elements: the first one must be 378 * the leaf (see RNTORT() in route.c), the second one is the parent. 379 * This routine initializes the relevant fields of the nodes, so that 380 * the leaf is the left child of the parent node, and both nodes have 381 * (almost) all all fields filled as appropriate. 382 * (XXX some fields are left unset, see the '#if 0' section). 383 * The function returns a pointer to the parent node. 384 */ 385 386static struct radix_node * 387rn_newpair(v, b, nodes) 388 void *v; 389 int b; 390 struct radix_node nodes[2]; 391{ 392 register struct radix_node *tt = nodes, *t = tt + 1; 393 t->rn_bit = b; 394 t->rn_bmask = 0x80 >> (b & 7); 395 t->rn_left = tt; 396 t->rn_offset = b >> 3; 397 398#if 0 /* XXX perhaps we should fill these fields as well. */ 399 t->rn_parent = t->rn_right = NULL; 400 401 tt->rn_mask = NULL; 402 tt->rn_dupedkey = NULL; 403 tt->rn_bmask = 0; 404#endif 405 tt->rn_bit = -1; 406 tt->rn_key = (caddr_t)v; 407 tt->rn_parent = t; 408 tt->rn_flags = t->rn_flags = RNF_ACTIVE; 409 tt->rn_mklist = t->rn_mklist = 0; 410#ifdef RN_DEBUG 411 tt->rn_info = rn_nodenum++; t->rn_info = rn_nodenum++; 412 tt->rn_twin = t; 413 tt->rn_ybro = rn_clist; 414 rn_clist = tt; 415#endif 416 return t; 417} 418 419static struct radix_node * 420rn_insert(v_arg, head, dupentry, nodes) 421 void *v_arg; 422 struct radix_node_head *head; 423 int *dupentry; 424 struct radix_node nodes[2]; 425{ 426 caddr_t v = v_arg; 427 struct radix_node *top = head->rnh_treetop; 428 int head_off = top->rn_offset, vlen = (int)LEN(v); 429 register struct radix_node *t = rn_search(v_arg, top); 430 register caddr_t cp = v + head_off; 431 register int b; 432 struct radix_node *tt; 433 /* 434 * Find first bit at which v and t->rn_key differ 435 */ 436 { 437 register caddr_t cp2 = t->rn_key + head_off; 438 register int cmp_res; 439 caddr_t cplim = v + vlen; 440 441 while (cp < cplim) 442 if (*cp2++ != *cp++) 443 goto on1; 444 *dupentry = 1; 445 return t; 446on1: 447 *dupentry = 0; 448 cmp_res = (cp[-1] ^ cp2[-1]) & 0xff; 449 for (b = (cp - v) << 3; cmp_res; b--) 450 cmp_res >>= 1; 451 } 452 { 453 register struct radix_node *p, *x = top; 454 cp = v; 455 do { 456 p = x; 457 if (cp[x->rn_offset] & x->rn_bmask) 458 x = x->rn_right; 459 else 460 x = x->rn_left; 461 } while (b > (unsigned) x->rn_bit); 462 /* x->rn_bit < b && x->rn_bit >= 0 */ 463#ifdef RN_DEBUG 464 if (rn_debug) 465 log(LOG_DEBUG, "rn_insert: Going In:\n"), traverse(p); 466#endif 467 t = rn_newpair(v_arg, b, nodes); 468 tt = t->rn_left; 469 if ((cp[p->rn_offset] & p->rn_bmask) == 0) 470 p->rn_left = t; 471 else 472 p->rn_right = t; 473 x->rn_parent = t; 474 t->rn_parent = p; /* frees x, p as temp vars below */ 475 if ((cp[t->rn_offset] & t->rn_bmask) == 0) { 476 t->rn_right = x; 477 } else { 478 t->rn_right = tt; 479 t->rn_left = x; 480 } 481#ifdef RN_DEBUG 482 if (rn_debug) 483 log(LOG_DEBUG, "rn_insert: Coming Out:\n"), traverse(p); 484#endif 485 } 486 return (tt); 487} 488 489struct radix_node * 490rn_addmask(n_arg, search, skip) 491 int search, skip; 492 void *n_arg; 493{ 494 caddr_t netmask = (caddr_t)n_arg; 495 register struct radix_node *x; 496 register caddr_t cp, cplim; 497 register int b = 0, mlen, j; 498 int maskduplicated, m0, isnormal; 499 struct radix_node *saved_x; 500 static int last_zeroed = 0; 501 502 if ((mlen = LEN(netmask)) > max_keylen) 503 mlen = max_keylen; 504 if (skip == 0) 505 skip = 1; 506 if (mlen <= skip) 507 return (mask_rnhead->rnh_nodes); 508 if (skip > 1) 509 bcopy(rn_ones + 1, addmask_key + 1, skip - 1); 510 if ((m0 = mlen) > skip) 511 bcopy(netmask + skip, addmask_key + skip, mlen - skip); 512 /* 513 * Trim trailing zeroes. 514 */ 515 for (cp = addmask_key + mlen; (cp > addmask_key) && cp[-1] == 0;) 516 cp--; 517 mlen = cp - addmask_key; 518 if (mlen <= skip) { 519 if (m0 >= last_zeroed) 520 last_zeroed = mlen; 521 return (mask_rnhead->rnh_nodes); 522 } 523 if (m0 < last_zeroed) 524 bzero(addmask_key + m0, last_zeroed - m0); 525 *addmask_key = last_zeroed = mlen; 526 x = rn_search(addmask_key, rn_masktop); 527 if (bcmp(addmask_key, x->rn_key, mlen) != 0) 528 x = 0; 529 if (x || search) 530 return (x); 531 R_Zalloc(x, struct radix_node *, max_keylen + 2 * sizeof (*x)); 532 if ((saved_x = x) == 0) 533 return (0); 534 netmask = cp = (caddr_t)(x + 2); 535 bcopy(addmask_key, cp, mlen); 536 x = rn_insert(cp, mask_rnhead, &maskduplicated, x); 537 if (maskduplicated) { 538 log(LOG_ERR, "rn_addmask: mask impossibly already in tree"); 539 Free(saved_x); 540 return (x); 541 } 542 /* 543 * Calculate index of mask, and check for normalcy. 544 * First find the first byte with a 0 bit, then if there are 545 * more bits left (remember we already trimmed the trailing 0's), 546 * the pattern must be one of those in normal_chars[], or we have 547 * a non-contiguous mask. 548 */ 549 cplim = netmask + mlen; 550 isnormal = 1; 551 for (cp = netmask + skip; (cp < cplim) && *(u_char *)cp == 0xff;) 552 cp++; 553 if (cp != cplim) { 554 static char normal_chars[] = { 555 0, 0x80, 0xc0, 0xe0, 0xf0, 0xf8, 0xfc, 0xfe, 0xff}; 556 557 for (j = 0x80; (j & *cp) != 0; j >>= 1) 558 b++; 559 if (*cp != normal_chars[b] || cp != (cplim - 1)) 560 isnormal = 0; 561 } 562 b += (cp - netmask) << 3; 563 x->rn_bit = -1 - b; 564 if (isnormal) 565 x->rn_flags |= RNF_NORMAL; 566 return (x); 567} 568 569static int /* XXX: arbitrary ordering for non-contiguous masks */ 570rn_lexobetter(m_arg, n_arg) 571 void *m_arg, *n_arg; 572{ 573 register u_char *mp = m_arg, *np = n_arg, *lim; 574 575 if (LEN(mp) > LEN(np)) 576 return 1; /* not really, but need to check longer one first */ 577 if (LEN(mp) == LEN(np)) 578 for (lim = mp + LEN(mp); mp < lim;) 579 if (*mp++ > *np++) 580 return 1; 581 return 0; 582} 583 584static struct radix_mask * 585rn_new_radix_mask(tt, next) 586 register struct radix_node *tt; 587 register struct radix_mask *next; 588{ 589 register struct radix_mask *m; 590 591 MKGet(m); 592 if (m == 0) { 593 log(LOG_ERR, "Mask for route not entered\n"); 594 return (0); 595 } 596 bzero(m, sizeof *m); 597 m->rm_bit = tt->rn_bit; 598 m->rm_flags = tt->rn_flags; 599 if (tt->rn_flags & RNF_NORMAL) 600 m->rm_leaf = tt; 601 else 602 m->rm_mask = tt->rn_mask; 603 m->rm_mklist = next; 604 tt->rn_mklist = m; 605 return m; 606} 607 608struct radix_node * 609rn_addroute(v_arg, n_arg, head, treenodes) 610 void *v_arg, *n_arg; 611 struct radix_node_head *head; 612 struct radix_node treenodes[2]; 613{ 614 caddr_t v = (caddr_t)v_arg, netmask = (caddr_t)n_arg; 615 register struct radix_node *t, *x = 0, *tt; 616 struct radix_node *saved_tt, *top = head->rnh_treetop; 617 short b = 0, b_leaf = 0; 618 int keyduplicated; 619 caddr_t mmask; 620 struct radix_mask *m, **mp; 621 622 623 RADIX_NODE_HEAD_WLOCK_ASSERT(head); 624 /* 625 * In dealing with non-contiguous masks, there may be 626 * many different routes which have the same mask. 627 * We will find it useful to have a unique pointer to 628 * the mask to speed avoiding duplicate references at 629 * nodes and possibly save time in calculating indices. 630 */ 631 if (netmask) { 632 if ((x = rn_addmask(netmask, 0, top->rn_offset)) == 0) 633 return (0); 634 b_leaf = x->rn_bit; 635 b = -1 - x->rn_bit; 636 netmask = x->rn_key; 637 } 638 /* 639 * Deal with duplicated keys: attach node to previous instance 640 */ 641 saved_tt = tt = rn_insert(v, head, &keyduplicated, treenodes); 642 if (keyduplicated) { 643 for (t = tt; tt; t = tt, tt = tt->rn_dupedkey) { 644#ifdef RADIX_MPATH 645 /* permit multipath, if enabled for the family */ 646 if (rn_mpath_capable(head) && netmask == tt->rn_mask) { 647 /* 648 * go down to the end of multipaths, so that 649 * new entry goes into the end of rn_dupedkey 650 * chain. 651 */ 652 do { 653 t = tt; 654 tt = tt->rn_dupedkey; 655 } while (tt && t->rn_mask == tt->rn_mask); 656 break; 657 } 658#endif 659 if (tt->rn_mask == netmask) 660 return (0); 661 if (netmask == 0 || 662 (tt->rn_mask && 663 ((b_leaf < tt->rn_bit) /* index(netmask) > node */ 664 || rn_refines(netmask, tt->rn_mask) 665 || rn_lexobetter(netmask, tt->rn_mask)))) 666 break; 667 } 668 /* 669 * If the mask is not duplicated, we wouldn't 670 * find it among possible duplicate key entries 671 * anyway, so the above test doesn't hurt. 672 * 673 * We sort the masks for a duplicated key the same way as 674 * in a masklist -- most specific to least specific. 675 * This may require the unfortunate nuisance of relocating 676 * the head of the list. 677 * 678 * We also reverse, or doubly link the list through the 679 * parent pointer. 680 */ 681 if (tt == saved_tt) { 682 struct radix_node *xx = x; 683 /* link in at head of list */ 684 (tt = treenodes)->rn_dupedkey = t; 685 tt->rn_flags = t->rn_flags; 686 tt->rn_parent = x = t->rn_parent; 687 t->rn_parent = tt; /* parent */ 688 if (x->rn_left == t) 689 x->rn_left = tt; 690 else 691 x->rn_right = tt; 692 saved_tt = tt; x = xx; 693 } else { 694 (tt = treenodes)->rn_dupedkey = t->rn_dupedkey; 695 t->rn_dupedkey = tt; 696 tt->rn_parent = t; /* parent */ 697 if (tt->rn_dupedkey) /* parent */ 698 tt->rn_dupedkey->rn_parent = tt; /* parent */ 699 } 700#ifdef RN_DEBUG 701 t=tt+1; tt->rn_info = rn_nodenum++; t->rn_info = rn_nodenum++; 702 tt->rn_twin = t; tt->rn_ybro = rn_clist; rn_clist = tt; 703#endif 704 tt->rn_key = (caddr_t) v; 705 tt->rn_bit = -1; 706 tt->rn_flags = RNF_ACTIVE; 707 } 708 /* 709 * Put mask in tree. 710 */ 711 if (netmask) { 712 tt->rn_mask = netmask; 713 tt->rn_bit = x->rn_bit; 714 tt->rn_flags |= x->rn_flags & RNF_NORMAL; 715 } 716 t = saved_tt->rn_parent; 717 if (keyduplicated) 718 goto on2; 719 b_leaf = -1 - t->rn_bit; 720 if (t->rn_right == saved_tt) 721 x = t->rn_left; 722 else 723 x = t->rn_right; 724 /* Promote general routes from below */ 725 if (x->rn_bit < 0) { 726 for (mp = &t->rn_mklist; x; x = x->rn_dupedkey) 727 if (x->rn_mask && (x->rn_bit >= b_leaf) && x->rn_mklist == 0) { 728 *mp = m = rn_new_radix_mask(x, 0); 729 if (m) 730 mp = &m->rm_mklist; 731 } 732 } else if (x->rn_mklist) { 733 /* 734 * Skip over masks whose index is > that of new node 735 */ 736 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist) 737 if (m->rm_bit >= b_leaf) 738 break; 739 t->rn_mklist = m; *mp = 0; 740 } 741on2: 742 /* Add new route to highest possible ancestor's list */ 743 if ((netmask == 0) || (b > t->rn_bit )) 744 return tt; /* can't lift at all */ 745 b_leaf = tt->rn_bit; 746 do { 747 x = t; 748 t = t->rn_parent; 749 } while (b <= t->rn_bit && x != top); 750 /* 751 * Search through routes associated with node to 752 * insert new route according to index. 753 * Need same criteria as when sorting dupedkeys to avoid 754 * double loop on deletion. 755 */ 756 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist) { 757 if (m->rm_bit < b_leaf) 758 continue; 759 if (m->rm_bit > b_leaf) 760 break; 761 if (m->rm_flags & RNF_NORMAL) { 762 mmask = m->rm_leaf->rn_mask; 763 if (tt->rn_flags & RNF_NORMAL) { 764 log(LOG_ERR, 765 "Non-unique normal route, mask not entered\n"); 766 return tt; 767 } 768 } else 769 mmask = m->rm_mask; 770 if (mmask == netmask) { 771 m->rm_refs++; 772 tt->rn_mklist = m; 773 return tt; 774 } 775 if (rn_refines(netmask, mmask) 776 || rn_lexobetter(netmask, mmask)) 777 break; 778 } 779 *mp = rn_new_radix_mask(tt, *mp); 780 return tt; 781} 782 783struct radix_node * 784rn_delete(v_arg, netmask_arg, head) 785 void *v_arg, *netmask_arg; 786 struct radix_node_head *head; 787{ 788 register struct radix_node *t, *p, *x, *tt; 789 struct radix_mask *m, *saved_m, **mp; 790 struct radix_node *dupedkey, *saved_tt, *top; 791 caddr_t v, netmask; 792 int b, head_off, vlen; 793 794 RADIX_NODE_HEAD_WLOCK_ASSERT(head); 795 v = v_arg; 796 netmask = netmask_arg; 797 x = head->rnh_treetop; 798 tt = rn_search(v, x); 799 head_off = x->rn_offset; 800 vlen = LEN(v); 801 saved_tt = tt; 802 top = x; 803 if (tt == 0 || 804 bcmp(v + head_off, tt->rn_key + head_off, vlen - head_off)) 805 return (0); 806 /* 807 * Delete our route from mask lists. 808 */ 809 if (netmask) { 810 if ((x = rn_addmask(netmask, 1, head_off)) == 0) 811 return (0); 812 netmask = x->rn_key; 813 while (tt->rn_mask != netmask) 814 if ((tt = tt->rn_dupedkey) == 0) 815 return (0); 816 } 817 if (tt->rn_mask == 0 || (saved_m = m = tt->rn_mklist) == 0) 818 goto on1; 819 if (tt->rn_flags & RNF_NORMAL) { 820 if (m->rm_leaf != tt || m->rm_refs > 0) { 821 log(LOG_ERR, "rn_delete: inconsistent annotation\n"); 822 return 0; /* dangling ref could cause disaster */ 823 } 824 } else { 825 if (m->rm_mask != tt->rn_mask) { 826 log(LOG_ERR, "rn_delete: inconsistent annotation\n"); 827 goto on1; 828 } 829 if (--m->rm_refs >= 0) 830 goto on1; 831 } 832 b = -1 - tt->rn_bit; 833 t = saved_tt->rn_parent; 834 if (b > t->rn_bit) 835 goto on1; /* Wasn't lifted at all */ 836 do { 837 x = t; 838 t = t->rn_parent; 839 } while (b <= t->rn_bit && x != top); 840 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist) 841 if (m == saved_m) { 842 *mp = m->rm_mklist; 843 MKFree(m); 844 break; 845 } 846 if (m == 0) { 847 log(LOG_ERR, "rn_delete: couldn't find our annotation\n"); 848 if (tt->rn_flags & RNF_NORMAL) 849 return (0); /* Dangling ref to us */ 850 } 851on1: 852 /* 853 * Eliminate us from tree 854 */ 855 if (tt->rn_flags & RNF_ROOT) 856 return (0); 857#ifdef RN_DEBUG 858 /* Get us out of the creation list */ 859 for (t = rn_clist; t && t->rn_ybro != tt; t = t->rn_ybro) {} 860 if (t) t->rn_ybro = tt->rn_ybro; 861#endif 862 t = tt->rn_parent; 863 dupedkey = saved_tt->rn_dupedkey; 864 if (dupedkey) { 865 /* 866 * Here, tt is the deletion target and 867 * saved_tt is the head of the dupekey chain. 868 */ 869 if (tt == saved_tt) { 870 /* remove from head of chain */ 871 x = dupedkey; x->rn_parent = t; 872 if (t->rn_left == tt) 873 t->rn_left = x; 874 else 875 t->rn_right = x; 876 } else { 877 /* find node in front of tt on the chain */ 878 for (x = p = saved_tt; p && p->rn_dupedkey != tt;) 879 p = p->rn_dupedkey; 880 if (p) { 881 p->rn_dupedkey = tt->rn_dupedkey; 882 if (tt->rn_dupedkey) /* parent */ 883 tt->rn_dupedkey->rn_parent = p; 884 /* parent */ 885 } else log(LOG_ERR, "rn_delete: couldn't find us\n"); 886 } 887 t = tt + 1; 888 if (t->rn_flags & RNF_ACTIVE) { 889#ifndef RN_DEBUG 890 *++x = *t; 891 p = t->rn_parent; 892#else 893 b = t->rn_info; 894 *++x = *t; 895 t->rn_info = b; 896 p = t->rn_parent; 897#endif 898 if (p->rn_left == t) 899 p->rn_left = x; 900 else 901 p->rn_right = x; 902 x->rn_left->rn_parent = x; 903 x->rn_right->rn_parent = x; 904 } 905 goto out; 906 } 907 if (t->rn_left == tt) 908 x = t->rn_right; 909 else 910 x = t->rn_left; 911 p = t->rn_parent; 912 if (p->rn_right == t) 913 p->rn_right = x; 914 else 915 p->rn_left = x; 916 x->rn_parent = p; 917 /* 918 * Demote routes attached to us. 919 */ 920 if (t->rn_mklist) { 921 if (x->rn_bit >= 0) { 922 for (mp = &x->rn_mklist; (m = *mp);) 923 mp = &m->rm_mklist; 924 *mp = t->rn_mklist; 925 } else { 926 /* If there are any key,mask pairs in a sibling 927 duped-key chain, some subset will appear sorted 928 in the same order attached to our mklist */ 929 for (m = t->rn_mklist; m && x; x = x->rn_dupedkey) 930 if (m == x->rn_mklist) { 931 struct radix_mask *mm = m->rm_mklist; 932 x->rn_mklist = 0; 933 if (--(m->rm_refs) < 0) 934 MKFree(m); 935 m = mm; 936 } 937 if (m) 938 log(LOG_ERR, 939 "rn_delete: Orphaned Mask %p at %p\n", 940 (void *)m, (void *)x); 941 } 942 } 943 /* 944 * We may be holding an active internal node in the tree. 945 */ 946 x = tt + 1; 947 if (t != x) { 948#ifndef RN_DEBUG 949 *t = *x; 950#else 951 b = t->rn_info; 952 *t = *x; 953 t->rn_info = b; 954#endif 955 t->rn_left->rn_parent = t; 956 t->rn_right->rn_parent = t; 957 p = x->rn_parent; 958 if (p->rn_left == x) 959 p->rn_left = t; 960 else 961 p->rn_right = t; 962 } 963out: 964 tt->rn_flags &= ~RNF_ACTIVE; 965 tt[1].rn_flags &= ~RNF_ACTIVE; 966 return (tt); 967} 968 969/* 970 * This is the same as rn_walktree() except for the parameters and the 971 * exit. 972 */ 973static int 974rn_walktree_from(h, a, m, f, w) 975 struct radix_node_head *h; 976 void *a, *m; 977 walktree_f_t *f; 978 void *w; 979{ 980 int error; 981 struct radix_node *base, *next; 982 u_char *xa = (u_char *)a; 983 u_char *xm = (u_char *)m; 984 register struct radix_node *rn, *last = 0 /* shut up gcc */; 985 int stopping = 0; 986 int lastb; 987 988 RADIX_NODE_HEAD_LOCK_ASSERT(h); 989 /* 990 * rn_search_m is sort-of-open-coded here. We cannot use the 991 * function because we need to keep track of the last node seen. 992 */ 993 /* printf("about to search\n"); */ 994 for (rn = h->rnh_treetop; rn->rn_bit >= 0; ) { 995 last = rn; 996 /* printf("rn_bit %d, rn_bmask %x, xm[rn_offset] %x\n", 997 rn->rn_bit, rn->rn_bmask, xm[rn->rn_offset]); */ 998 if (!(rn->rn_bmask & xm[rn->rn_offset])) { 999 break; 1000 } 1001 if (rn->rn_bmask & xa[rn->rn_offset]) { 1002 rn = rn->rn_right; 1003 } else { 1004 rn = rn->rn_left; 1005 } 1006 } 1007 /* printf("done searching\n"); */ 1008 1009 /* 1010 * Two cases: either we stepped off the end of our mask, 1011 * in which case last == rn, or we reached a leaf, in which 1012 * case we want to start from the last node we looked at. 1013 * Either way, last is the node we want to start from. 1014 */ 1015 rn = last; 1016 lastb = rn->rn_bit; 1017 1018 /* printf("rn %p, lastb %d\n", rn, lastb);*/ 1019 1020 /* 1021 * This gets complicated because we may delete the node 1022 * while applying the function f to it, so we need to calculate 1023 * the successor node in advance. 1024 */ 1025 while (rn->rn_bit >= 0) 1026 rn = rn->rn_left; 1027 1028 while (!stopping) { 1029 /* printf("node %p (%d)\n", rn, rn->rn_bit); */ 1030 base = rn; 1031 /* If at right child go back up, otherwise, go right */ 1032 while (rn->rn_parent->rn_right == rn 1033 && !(rn->rn_flags & RNF_ROOT)) { 1034 rn = rn->rn_parent; 1035 1036 /* if went up beyond last, stop */ 1037 if (rn->rn_bit <= lastb) { 1038 stopping = 1; 1039 /* printf("up too far\n"); */ 1040 /* 1041 * XXX we should jump to the 'Process leaves' 1042 * part, because the values of 'rn' and 'next' 1043 * we compute will not be used. Not a big deal 1044 * because this loop will terminate, but it is 1045 * inefficient and hard to understand! 1046 */ 1047 } 1048 } 1049 1050 /* 1051 * At the top of the tree, no need to traverse the right 1052 * half, prevent the traversal of the entire tree in the 1053 * case of default route. 1054 */ 1055 if (rn->rn_parent->rn_flags & RNF_ROOT) 1056 stopping = 1; 1057 1058 /* Find the next *leaf* since next node might vanish, too */ 1059 for (rn = rn->rn_parent->rn_right; rn->rn_bit >= 0;) 1060 rn = rn->rn_left; 1061 next = rn; 1062 /* Process leaves */ 1063 while ((rn = base) != 0) { 1064 base = rn->rn_dupedkey; 1065 /* printf("leaf %p\n", rn); */ 1066 if (!(rn->rn_flags & RNF_ROOT) 1067 && (error = (*f)(rn, w))) 1068 return (error); 1069 } 1070 rn = next; 1071 1072 if (rn->rn_flags & RNF_ROOT) { 1073 /* printf("root, stopping"); */ 1074 stopping = 1; 1075 } 1076 1077 } 1078 return 0; 1079} 1080 1081static int 1082rn_walktree(h, f, w) 1083 struct radix_node_head *h; 1084 walktree_f_t *f; 1085 void *w; 1086{ 1087 int error; 1088 struct radix_node *base, *next; 1089 register struct radix_node *rn = h->rnh_treetop; 1090 /* 1091 * This gets complicated because we may delete the node 1092 * while applying the function f to it, so we need to calculate 1093 * the successor node in advance. 1094 */ 1095 1096 RADIX_NODE_HEAD_LOCK_ASSERT(h); 1097 /* First time through node, go left */ 1098 while (rn->rn_bit >= 0) 1099 rn = rn->rn_left; 1100 for (;;) { 1101 base = rn; 1102 /* If at right child go back up, otherwise, go right */ 1103 while (rn->rn_parent->rn_right == rn 1104 && (rn->rn_flags & RNF_ROOT) == 0) 1105 rn = rn->rn_parent; 1106 /* Find the next *leaf* since next node might vanish, too */ 1107 for (rn = rn->rn_parent->rn_right; rn->rn_bit >= 0;) 1108 rn = rn->rn_left; 1109 next = rn; 1110 /* Process leaves */ 1111 while ((rn = base)) { 1112 base = rn->rn_dupedkey; 1113 if (!(rn->rn_flags & RNF_ROOT) 1114 && (error = (*f)(rn, w))) 1115 return (error); 1116 } 1117 rn = next; 1118 if (rn->rn_flags & RNF_ROOT) 1119 return (0); 1120 } 1121 /* NOTREACHED */ 1122} 1123 1124/* 1125 * Allocate and initialize an empty tree. This has 3 nodes, which are 1126 * part of the radix_node_head (in the order <left,root,right>) and are 1127 * marked RNF_ROOT so they cannot be freed. 1128 * The leaves have all-zero and all-one keys, with significant 1129 * bits starting at 'off'. 1130 * Return 1 on success, 0 on error. 1131 */ 1132int 1133rn_inithead(head, off) 1134 void **head; 1135 int off; 1136{ 1137 register struct radix_node_head *rnh; 1138 register struct radix_node *t, *tt, *ttt; 1139 if (*head) 1140 return (1); 1141 R_Zalloc(rnh, struct radix_node_head *, sizeof (*rnh)); 1142 if (rnh == 0) 1143 return (0); 1144#ifdef _KERNEL 1145 RADIX_NODE_HEAD_LOCK_INIT(rnh); 1146#endif 1147 *head = rnh; 1148 t = rn_newpair(rn_zeros, off, rnh->rnh_nodes); 1149 ttt = rnh->rnh_nodes + 2; 1150 t->rn_right = ttt; 1151 t->rn_parent = t; 1152 tt = t->rn_left; /* ... which in turn is rnh->rnh_nodes */ 1153 tt->rn_flags = t->rn_flags = RNF_ROOT | RNF_ACTIVE; 1154 tt->rn_bit = -1 - off; 1155 *ttt = *tt; 1156 ttt->rn_key = rn_ones; 1157 rnh->rnh_addaddr = rn_addroute; 1158 rnh->rnh_deladdr = rn_delete; 1159 rnh->rnh_matchaddr = rn_match; 1160 rnh->rnh_lookup = rn_lookup; 1161 rnh->rnh_walktree = rn_walktree; 1162 rnh->rnh_walktree_from = rn_walktree_from; 1163 rnh->rnh_treetop = t; 1164 return (1); 1165} 1166 1167void 1168rn_init() 1169{ 1170 char *cp, *cplim; 1171#ifdef _KERNEL 1172 struct domain *dom; 1173 1174 for (dom = domains; dom; dom = dom->dom_next) 1175 if (dom->dom_maxrtkey > max_keylen) 1176 max_keylen = dom->dom_maxrtkey; 1177#endif 1178 if (max_keylen == 0) { 1179 log(LOG_ERR, 1180 "rn_init: radix functions require max_keylen be set\n"); 1181 return; 1182 } 1183 R_Malloc(rn_zeros, char *, 3 * max_keylen); 1184 if (rn_zeros == NULL) 1185 panic("rn_init"); 1186 bzero(rn_zeros, 3 * max_keylen); 1187 rn_ones = cp = rn_zeros + max_keylen; 1188 addmask_key = cplim = rn_ones + max_keylen; 1189 while (cp < cplim) 1190 *cp++ = -1; 1191 if (rn_inithead((void **)(void *)&mask_rnhead, 0) == 0) 1192 panic("rn_init 2"); 1193} 1194