radix.c revision 200439
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 200439 2009-12-12 15:49:28Z luigi $ 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/domain.h> 44#include <sys/syslog.h> 45#include <net/radix.h> 46#include "opt_mpath.h" 47#ifdef RADIX_MPATH 48#include <net/radix_mpath.h> 49#endif 50#else /* !_KERNEL */ 51#include <stdio.h> 52#include <strings.h> 53#include <stdlib.h> 54#define log(x, arg...) fprintf(stderr, ## arg) 55#define panic(x) fprintf(stderr, "PANIC: %s", x), exit(1) 56#define min(a, b) ((a) < (b) ? (a) : (b) ) 57#include <net/radix.h> 58#endif /* !_KERNEL */ 59 60static int rn_walktree_from(struct radix_node_head *h, void *a, void *m, 61 walktree_f_t *f, void *w); 62static int rn_walktree(struct radix_node_head *, walktree_f_t *, void *); 63static struct radix_node 64 *rn_insert(void *, struct radix_node_head *, int *, 65 struct radix_node [2]), 66 *rn_newpair(void *, int, struct radix_node[2]), 67 *rn_search(void *, struct radix_node *), 68 *rn_search_m(void *, struct radix_node *, void *); 69 70static int max_keylen; 71static struct radix_mask *rn_mkfreelist; 72static struct radix_node_head *mask_rnhead; 73/* 74 * Work area -- the following point to 3 buffers of size max_keylen, 75 * allocated in this order in a block of memory malloc'ed by rn_init. 76 * rn_zeros, rn_ones are set in rn_init and used in readonly afterwards. 77 * addmask_key is used in rn_addmask in rw mode and not thread-safe. 78 */ 79static char *rn_zeros, *rn_ones, *addmask_key; 80 81#define MKGet(m) { \ 82 if (rn_mkfreelist) { \ 83 m = rn_mkfreelist; \ 84 rn_mkfreelist = (m)->rm_mklist; \ 85 } else \ 86 R_Malloc(m, struct radix_mask *, sizeof (struct radix_mask)); } 87 88#define MKFree(m) { (m)->rm_mklist = rn_mkfreelist; rn_mkfreelist = (m);} 89 90#define rn_masktop (mask_rnhead->rnh_treetop) 91 92static int rn_lexobetter(void *m_arg, void *n_arg); 93static struct radix_mask * 94 rn_new_radix_mask(struct radix_node *tt, 95 struct radix_mask *next); 96static int rn_satisfies_leaf(char *trial, struct radix_node *leaf, 97 int skip); 98 99/* 100 * The data structure for the keys is a radix tree with one way 101 * branching removed. The index rn_bit at an internal node n represents a bit 102 * position to be tested. The tree is arranged so that all descendants 103 * of a node n have keys whose bits all agree up to position rn_bit - 1. 104 * (We say the index of n is rn_bit.) 105 * 106 * There is at least one descendant which has a one bit at position rn_bit, 107 * and at least one with a zero there. 108 * 109 * A route is determined by a pair of key and mask. We require that the 110 * bit-wise logical and of the key and mask to be the key. 111 * We define the index of a route to associated with the mask to be 112 * the first bit number in the mask where 0 occurs (with bit number 0 113 * representing the highest order bit). 114 * 115 * We say a mask is normal if every bit is 0, past the index of the mask. 116 * If a node n has a descendant (k, m) with index(m) == index(n) == rn_bit, 117 * and m is a normal mask, then the route applies to every descendant of n. 118 * If the index(m) < rn_bit, this implies the trailing last few bits of k 119 * before bit b are all 0, (and hence consequently true of every descendant 120 * of n), so the route applies to all descendants of the node as well. 121 * 122 * Similar logic shows that a non-normal mask m such that 123 * index(m) <= index(n) could potentially apply to many children of n. 124 * Thus, for each non-host route, we attach its mask to a list at an internal 125 * node as high in the tree as we can go. 126 * 127 * The present version of the code makes use of normal routes in short- 128 * circuiting an explict mask and compare operation when testing whether 129 * a key satisfies a normal route, and also in remembering the unique leaf 130 * that governs a subtree. 131 */ 132 133/* 134 * Most of the functions in this code assume that the key/mask arguments 135 * are sockaddr-like structures, where the first byte is an u_char 136 * indicating the size of the entire structure. 137 * 138 * To make the assumption more explicit, we use the LEN() macro to access 139 * this field. It is safe to pass an expression with side effects 140 * to LEN() as the argument is evaluated only once. 141 * We cast the result to int as this is the dominant usage. 142 */ 143#define LEN(x) ( (int) (*(const u_char *)(x)) ) 144 145/* 146 * XXX THIS NEEDS TO BE FIXED 147 * In the code, pointers to keys and masks are passed as either 148 * 'void *' (because callers use to pass pointers of various kinds), or 149 * 'caddr_t' (which is fine for pointer arithmetics, but not very 150 * clean when you dereference it to access data). Furthermore, caddr_t 151 * is really 'char *', while the natural type to operate on keys and 152 * masks would be 'u_char'. This mismatch require a lot of casts and 153 * intermediate variables to adapt types that clutter the code. 154 */ 155 156/* 157 * Search a node in the tree matching the key. 158 */ 159static struct radix_node * 160rn_search(v_arg, head) 161 void *v_arg; 162 struct radix_node *head; 163{ 164 register struct radix_node *x; 165 register caddr_t v; 166 167 for (x = head, v = v_arg; x->rn_bit >= 0;) { 168 if (x->rn_bmask & v[x->rn_offset]) 169 x = x->rn_right; 170 else 171 x = x->rn_left; 172 } 173 return (x); 174} 175 176/* 177 * Same as above, but with an additional mask. 178 * XXX note this function is used only once. 179 */ 180static struct radix_node * 181rn_search_m(v_arg, head, m_arg) 182 struct radix_node *head; 183 void *v_arg, *m_arg; 184{ 185 register struct radix_node *x; 186 register caddr_t v = v_arg, m = m_arg; 187 188 for (x = head; x->rn_bit >= 0;) { 189 if ((x->rn_bmask & m[x->rn_offset]) && 190 (x->rn_bmask & v[x->rn_offset])) 191 x = x->rn_right; 192 else 193 x = x->rn_left; 194 } 195 return x; 196} 197 198int 199rn_refines(m_arg, n_arg) 200 void *m_arg, *n_arg; 201{ 202 register caddr_t m = m_arg, n = n_arg; 203 register caddr_t lim, lim2 = lim = n + LEN(n); 204 int longer = LEN(n++) - LEN(m++); 205 int masks_are_equal = 1; 206 207 if (longer > 0) 208 lim -= longer; 209 while (n < lim) { 210 if (*n & ~(*m)) 211 return 0; 212 if (*n++ != *m++) 213 masks_are_equal = 0; 214 } 215 while (n < lim2) 216 if (*n++) 217 return 0; 218 if (masks_are_equal && (longer < 0)) 219 for (lim2 = m - longer; m < lim2; ) 220 if (*m++) 221 return 1; 222 return (!masks_are_equal); 223} 224 225struct radix_node * 226rn_lookup(v_arg, m_arg, head) 227 void *v_arg, *m_arg; 228 struct radix_node_head *head; 229{ 230 register struct radix_node *x; 231 caddr_t netmask = 0; 232 233 if (m_arg) { 234 x = rn_addmask(m_arg, 1, 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(n_arg, search, skip) 494 int search, skip; 495 void *n_arg; 496{ 497 caddr_t netmask = (caddr_t)n_arg; 498 register struct radix_node *x; 499 register caddr_t cp, cplim; 500 register int b = 0, mlen, j; 501 int maskduplicated, m0, isnormal; 502 struct radix_node *saved_x; 503 static int last_zeroed = 0; 504 505 if ((mlen = LEN(netmask)) > max_keylen) 506 mlen = max_keylen; 507 if (skip == 0) 508 skip = 1; 509 if (mlen <= skip) 510 return (mask_rnhead->rnh_nodes); 511 if (skip > 1) 512 bcopy(rn_ones + 1, addmask_key + 1, skip - 1); 513 if ((m0 = mlen) > skip) 514 bcopy(netmask + skip, addmask_key + skip, mlen - skip); 515 /* 516 * Trim trailing zeroes. 517 */ 518 for (cp = addmask_key + mlen; (cp > addmask_key) && cp[-1] == 0;) 519 cp--; 520 mlen = cp - addmask_key; 521 if (mlen <= skip) { 522 if (m0 >= last_zeroed) 523 last_zeroed = mlen; 524 return (mask_rnhead->rnh_nodes); 525 } 526 if (m0 < last_zeroed) 527 bzero(addmask_key + m0, last_zeroed - m0); 528 *addmask_key = last_zeroed = mlen; 529 x = rn_search(addmask_key, rn_masktop); 530 if (bcmp(addmask_key, x->rn_key, mlen) != 0) 531 x = 0; 532 if (x || search) 533 return (x); 534 R_Zalloc(x, struct radix_node *, max_keylen + 2 * sizeof (*x)); 535 if ((saved_x = x) == 0) 536 return (0); 537 netmask = cp = (caddr_t)(x + 2); 538 bcopy(addmask_key, cp, mlen); 539 x = rn_insert(cp, mask_rnhead, &maskduplicated, x); 540 if (maskduplicated) { 541 log(LOG_ERR, "rn_addmask: mask impossibly already in tree"); 542 Free(saved_x); 543 return (x); 544 } 545 /* 546 * Calculate index of mask, and check for normalcy. 547 * First find the first byte with a 0 bit, then if there are 548 * more bits left (remember we already trimmed the trailing 0's), 549 * the pattern must be one of those in normal_chars[], or we have 550 * a non-contiguous mask. 551 */ 552 cplim = netmask + mlen; 553 isnormal = 1; 554 for (cp = netmask + skip; (cp < cplim) && *(u_char *)cp == 0xff;) 555 cp++; 556 if (cp != cplim) { 557 static char normal_chars[] = { 558 0, 0x80, 0xc0, 0xe0, 0xf0, 0xf8, 0xfc, 0xfe, 0xff}; 559 560 for (j = 0x80; (j & *cp) != 0; j >>= 1) 561 b++; 562 if (*cp != normal_chars[b] || cp != (cplim - 1)) 563 isnormal = 0; 564 } 565 b += (cp - netmask) << 3; 566 x->rn_bit = -1 - b; 567 if (isnormal) 568 x->rn_flags |= RNF_NORMAL; 569 return (x); 570} 571 572static int /* XXX: arbitrary ordering for non-contiguous masks */ 573rn_lexobetter(m_arg, n_arg) 574 void *m_arg, *n_arg; 575{ 576 register u_char *mp = m_arg, *np = n_arg, *lim; 577 578 if (LEN(mp) > LEN(np)) 579 return 1; /* not really, but need to check longer one first */ 580 if (LEN(mp) == LEN(np)) 581 for (lim = mp + LEN(mp); mp < lim;) 582 if (*mp++ > *np++) 583 return 1; 584 return 0; 585} 586 587static struct radix_mask * 588rn_new_radix_mask(tt, next) 589 register struct radix_node *tt; 590 register struct radix_mask *next; 591{ 592 register struct radix_mask *m; 593 594 MKGet(m); 595 if (m == 0) { 596 log(LOG_ERR, "Mask for route not entered\n"); 597 return (0); 598 } 599 bzero(m, sizeof *m); 600 m->rm_bit = tt->rn_bit; 601 m->rm_flags = tt->rn_flags; 602 if (tt->rn_flags & RNF_NORMAL) 603 m->rm_leaf = tt; 604 else 605 m->rm_mask = tt->rn_mask; 606 m->rm_mklist = next; 607 tt->rn_mklist = m; 608 return m; 609} 610 611struct radix_node * 612rn_addroute(v_arg, n_arg, head, treenodes) 613 void *v_arg, *n_arg; 614 struct radix_node_head *head; 615 struct radix_node treenodes[2]; 616{ 617 caddr_t v = (caddr_t)v_arg, netmask = (caddr_t)n_arg; 618 register struct radix_node *t, *x = 0, *tt; 619 struct radix_node *saved_tt, *top = head->rnh_treetop; 620 short b = 0, b_leaf = 0; 621 int keyduplicated; 622 caddr_t mmask; 623 struct radix_mask *m, **mp; 624 625 /* 626 * In dealing with non-contiguous masks, there may be 627 * many different routes which have the same mask. 628 * We will find it useful to have a unique pointer to 629 * the mask to speed avoiding duplicate references at 630 * nodes and possibly save time in calculating indices. 631 */ 632 if (netmask) { 633 if ((x = rn_addmask(netmask, 0, top->rn_offset)) == 0) 634 return (0); 635 b_leaf = x->rn_bit; 636 b = -1 - x->rn_bit; 637 netmask = x->rn_key; 638 } 639 /* 640 * Deal with duplicated keys: attach node to previous instance 641 */ 642 saved_tt = tt = rn_insert(v, head, &keyduplicated, treenodes); 643 if (keyduplicated) { 644 for (t = tt; tt; t = tt, tt = tt->rn_dupedkey) { 645#ifdef RADIX_MPATH 646 /* permit multipath, if enabled for the family */ 647 if (rn_mpath_capable(head) && netmask == tt->rn_mask) { 648 /* 649 * go down to the end of multipaths, so that 650 * new entry goes into the end of rn_dupedkey 651 * chain. 652 */ 653 do { 654 t = tt; 655 tt = tt->rn_dupedkey; 656 } while (tt && t->rn_mask == tt->rn_mask); 657 break; 658 } 659#endif 660 if (tt->rn_mask == netmask) 661 return (0); 662 if (netmask == 0 || 663 (tt->rn_mask && 664 ((b_leaf < tt->rn_bit) /* index(netmask) > node */ 665 || rn_refines(netmask, tt->rn_mask) 666 || rn_lexobetter(netmask, tt->rn_mask)))) 667 break; 668 } 669 /* 670 * If the mask is not duplicated, we wouldn't 671 * find it among possible duplicate key entries 672 * anyway, so the above test doesn't hurt. 673 * 674 * We sort the masks for a duplicated key the same way as 675 * in a masklist -- most specific to least specific. 676 * This may require the unfortunate nuisance of relocating 677 * the head of the list. 678 * 679 * We also reverse, or doubly link the list through the 680 * parent pointer. 681 */ 682 if (tt == saved_tt) { 683 struct radix_node *xx = x; 684 /* link in at head of list */ 685 (tt = treenodes)->rn_dupedkey = t; 686 tt->rn_flags = t->rn_flags; 687 tt->rn_parent = x = t->rn_parent; 688 t->rn_parent = tt; /* parent */ 689 if (x->rn_left == t) 690 x->rn_left = tt; 691 else 692 x->rn_right = tt; 693 saved_tt = tt; x = xx; 694 } else { 695 (tt = treenodes)->rn_dupedkey = t->rn_dupedkey; 696 t->rn_dupedkey = tt; 697 tt->rn_parent = t; /* parent */ 698 if (tt->rn_dupedkey) /* parent */ 699 tt->rn_dupedkey->rn_parent = tt; /* parent */ 700 } 701#ifdef RN_DEBUG 702 t=tt+1; tt->rn_info = rn_nodenum++; t->rn_info = rn_nodenum++; 703 tt->rn_twin = t; tt->rn_ybro = rn_clist; rn_clist = tt; 704#endif 705 tt->rn_key = (caddr_t) v; 706 tt->rn_bit = -1; 707 tt->rn_flags = RNF_ACTIVE; 708 } 709 /* 710 * Put mask in tree. 711 */ 712 if (netmask) { 713 tt->rn_mask = netmask; 714 tt->rn_bit = x->rn_bit; 715 tt->rn_flags |= x->rn_flags & RNF_NORMAL; 716 } 717 t = saved_tt->rn_parent; 718 if (keyduplicated) 719 goto on2; 720 b_leaf = -1 - t->rn_bit; 721 if (t->rn_right == saved_tt) 722 x = t->rn_left; 723 else 724 x = t->rn_right; 725 /* Promote general routes from below */ 726 if (x->rn_bit < 0) { 727 for (mp = &t->rn_mklist; x; x = x->rn_dupedkey) 728 if (x->rn_mask && (x->rn_bit >= b_leaf) && x->rn_mklist == 0) { 729 *mp = m = rn_new_radix_mask(x, 0); 730 if (m) 731 mp = &m->rm_mklist; 732 } 733 } else if (x->rn_mklist) { 734 /* 735 * Skip over masks whose index is > that of new node 736 */ 737 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist) 738 if (m->rm_bit >= b_leaf) 739 break; 740 t->rn_mklist = m; *mp = 0; 741 } 742on2: 743 /* Add new route to highest possible ancestor's list */ 744 if ((netmask == 0) || (b > t->rn_bit )) 745 return tt; /* can't lift at all */ 746 b_leaf = tt->rn_bit; 747 do { 748 x = t; 749 t = t->rn_parent; 750 } while (b <= t->rn_bit && x != top); 751 /* 752 * Search through routes associated with node to 753 * insert new route according to index. 754 * Need same criteria as when sorting dupedkeys to avoid 755 * double loop on deletion. 756 */ 757 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist) { 758 if (m->rm_bit < b_leaf) 759 continue; 760 if (m->rm_bit > b_leaf) 761 break; 762 if (m->rm_flags & RNF_NORMAL) { 763 mmask = m->rm_leaf->rn_mask; 764 if (tt->rn_flags & RNF_NORMAL) { 765 log(LOG_ERR, 766 "Non-unique normal route, mask not entered\n"); 767 return tt; 768 } 769 } else 770 mmask = m->rm_mask; 771 if (mmask == netmask) { 772 m->rm_refs++; 773 tt->rn_mklist = m; 774 return tt; 775 } 776 if (rn_refines(netmask, mmask) 777 || rn_lexobetter(netmask, mmask)) 778 break; 779 } 780 *mp = rn_new_radix_mask(tt, *mp); 781 return tt; 782} 783 784struct radix_node * 785rn_delete(v_arg, netmask_arg, head) 786 void *v_arg, *netmask_arg; 787 struct radix_node_head *head; 788{ 789 register struct radix_node *t, *p, *x, *tt; 790 struct radix_mask *m, *saved_m, **mp; 791 struct radix_node *dupedkey, *saved_tt, *top; 792 caddr_t v, netmask; 793 int b, head_off, vlen; 794 795 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 /* 989 * rn_search_m is sort-of-open-coded here. We cannot use the 990 * function because we need to keep track of the last node seen. 991 */ 992 /* printf("about to search\n"); */ 993 for (rn = h->rnh_treetop; rn->rn_bit >= 0; ) { 994 last = rn; 995 /* printf("rn_bit %d, rn_bmask %x, xm[rn_offset] %x\n", 996 rn->rn_bit, rn->rn_bmask, xm[rn->rn_offset]); */ 997 if (!(rn->rn_bmask & xm[rn->rn_offset])) { 998 break; 999 } 1000 if (rn->rn_bmask & xa[rn->rn_offset]) { 1001 rn = rn->rn_right; 1002 } else { 1003 rn = rn->rn_left; 1004 } 1005 } 1006 /* printf("done searching\n"); */ 1007 1008 /* 1009 * Two cases: either we stepped off the end of our mask, 1010 * in which case last == rn, or we reached a leaf, in which 1011 * case we want to start from the last node we looked at. 1012 * Either way, last is the node we want to start from. 1013 */ 1014 rn = last; 1015 lastb = rn->rn_bit; 1016 1017 /* printf("rn %p, lastb %d\n", rn, lastb);*/ 1018 1019 /* 1020 * This gets complicated because we may delete the node 1021 * while applying the function f to it, so we need to calculate 1022 * the successor node in advance. 1023 */ 1024 while (rn->rn_bit >= 0) 1025 rn = rn->rn_left; 1026 1027 while (!stopping) { 1028 /* printf("node %p (%d)\n", rn, rn->rn_bit); */ 1029 base = rn; 1030 /* If at right child go back up, otherwise, go right */ 1031 while (rn->rn_parent->rn_right == rn 1032 && !(rn->rn_flags & RNF_ROOT)) { 1033 rn = rn->rn_parent; 1034 1035 /* if went up beyond last, stop */ 1036 if (rn->rn_bit <= lastb) { 1037 stopping = 1; 1038 /* printf("up too far\n"); */ 1039 /* 1040 * XXX we should jump to the 'Process leaves' 1041 * part, because the values of 'rn' and 'next' 1042 * we compute will not be used. Not a big deal 1043 * because this loop will terminate, but it is 1044 * inefficient and hard to understand! 1045 */ 1046 } 1047 } 1048 1049 /* 1050 * At the top of the tree, no need to traverse the right 1051 * half, prevent the traversal of the entire tree in the 1052 * case of default route. 1053 */ 1054 if (rn->rn_parent->rn_flags & RNF_ROOT) 1055 stopping = 1; 1056 1057 /* Find the next *leaf* since next node might vanish, too */ 1058 for (rn = rn->rn_parent->rn_right; rn->rn_bit >= 0;) 1059 rn = rn->rn_left; 1060 next = rn; 1061 /* Process leaves */ 1062 while ((rn = base) != 0) { 1063 base = rn->rn_dupedkey; 1064 /* printf("leaf %p\n", rn); */ 1065 if (!(rn->rn_flags & RNF_ROOT) 1066 && (error = (*f)(rn, w))) 1067 return (error); 1068 } 1069 rn = next; 1070 1071 if (rn->rn_flags & RNF_ROOT) { 1072 /* printf("root, stopping"); */ 1073 stopping = 1; 1074 } 1075 1076 } 1077 return 0; 1078} 1079 1080static int 1081rn_walktree(h, f, w) 1082 struct radix_node_head *h; 1083 walktree_f_t *f; 1084 void *w; 1085{ 1086 int error; 1087 struct radix_node *base, *next; 1088 register struct radix_node *rn = h->rnh_treetop; 1089 /* 1090 * This gets complicated because we may delete the node 1091 * while applying the function f to it, so we need to calculate 1092 * the successor node in advance. 1093 */ 1094 1095 /* First time through node, go left */ 1096 while (rn->rn_bit >= 0) 1097 rn = rn->rn_left; 1098 for (;;) { 1099 base = rn; 1100 /* If at right child go back up, otherwise, go right */ 1101 while (rn->rn_parent->rn_right == rn 1102 && (rn->rn_flags & RNF_ROOT) == 0) 1103 rn = rn->rn_parent; 1104 /* Find the next *leaf* since next node might vanish, too */ 1105 for (rn = rn->rn_parent->rn_right; rn->rn_bit >= 0;) 1106 rn = rn->rn_left; 1107 next = rn; 1108 /* Process leaves */ 1109 while ((rn = base)) { 1110 base = rn->rn_dupedkey; 1111 if (!(rn->rn_flags & RNF_ROOT) 1112 && (error = (*f)(rn, w))) 1113 return (error); 1114 } 1115 rn = next; 1116 if (rn->rn_flags & RNF_ROOT) 1117 return (0); 1118 } 1119 /* NOTREACHED */ 1120} 1121 1122/* 1123 * Allocate and initialize an empty tree. This has 3 nodes, which are 1124 * part of the radix_node_head (in the order <left,root,right>) and are 1125 * marked RNF_ROOT so they cannot be freed. 1126 * The leaves have all-zero and all-one keys, with significant 1127 * bits starting at 'off'. 1128 * Return 1 on success, 0 on error. 1129 */ 1130int 1131rn_inithead(head, off) 1132 void **head; 1133 int off; 1134{ 1135 register struct radix_node_head *rnh; 1136 register struct radix_node *t, *tt, *ttt; 1137 if (*head) 1138 return (1); 1139 R_Zalloc(rnh, struct radix_node_head *, sizeof (*rnh)); 1140 if (rnh == 0) 1141 return (0); 1142#ifdef _KERNEL 1143 RADIX_NODE_HEAD_LOCK_INIT(rnh); 1144#endif 1145 *head = rnh; 1146 t = rn_newpair(rn_zeros, off, rnh->rnh_nodes); 1147 ttt = rnh->rnh_nodes + 2; 1148 t->rn_right = ttt; 1149 t->rn_parent = t; 1150 tt = t->rn_left; /* ... which in turn is rnh->rnh_nodes */ 1151 tt->rn_flags = t->rn_flags = RNF_ROOT | RNF_ACTIVE; 1152 tt->rn_bit = -1 - off; 1153 *ttt = *tt; 1154 ttt->rn_key = rn_ones; 1155 rnh->rnh_addaddr = rn_addroute; 1156 rnh->rnh_deladdr = rn_delete; 1157 rnh->rnh_matchaddr = rn_match; 1158 rnh->rnh_lookup = rn_lookup; 1159 rnh->rnh_walktree = rn_walktree; 1160 rnh->rnh_walktree_from = rn_walktree_from; 1161 rnh->rnh_treetop = t; 1162 return (1); 1163} 1164 1165void 1166rn_init() 1167{ 1168 char *cp, *cplim; 1169#ifdef _KERNEL 1170 struct domain *dom; 1171 1172 for (dom = domains; dom; dom = dom->dom_next) 1173 if (dom->dom_maxrtkey > max_keylen) 1174 max_keylen = dom->dom_maxrtkey; 1175#endif 1176 if (max_keylen == 0) { 1177 log(LOG_ERR, 1178 "rn_init: radix functions require max_keylen be set\n"); 1179 return; 1180 } 1181 R_Malloc(rn_zeros, char *, 3 * max_keylen); 1182 if (rn_zeros == NULL) 1183 panic("rn_init"); 1184 bzero(rn_zeros, 3 * max_keylen); 1185 rn_ones = cp = rn_zeros + max_keylen; 1186 addmask_key = cplim = rn_ones + max_keylen; 1187 while (cp < cplim) 1188 *cp++ = -1; 1189 if (rn_inithead((void **)(void *)&mask_rnhead, 0) == 0) 1190 panic("rn_init 2"); 1191} 1192 1193#ifndef _KERNEL 1194/* 1195 * A simple function to make the code usable from userland. 1196 * A proper fix (maybe later) would be to change rn_init() so that it 1197 * takes maxkeylen as an argument, and move the scan of 1198 * domains into net/route.c::route_init(). 1199 */ 1200void rn_init2(int maxk); 1201void 1202rn_init2(int maxk) 1203{ 1204 max_keylen = maxk; 1205 rn_init(); 1206} 1207#endif /* !_KERNEL */ 1208