1 /* $OpenBSD: radix.c,v 1.29 2011/07/22 13:05:29 henning Exp $ */ 2 /* $NetBSD: radix.c,v 1.20 2003/08/07 16:32:56 agc Exp $ */ 3 4 /* 5 * Copyright (c) 1988, 1989, 1993 6 * The Regents of the University of California. All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * @(#)radix.c 8.6 (Berkeley) 10/17/95 33 */ 34 35 /* 36 * Routines to build and maintain radix trees for routing lookups. 37 */ 38 39 #ifndef _NET_RADIX_H_ 40 #include <sys/param.h> 41 #ifdef _KERNEL 42 #include <sys/systm.h> 43 #include <sys/malloc.h> 44 #define M_DONTWAIT M_NOWAIT 45 #include <sys/domain.h> 46 #else 47 #include <stdlib.h> 48 #endif 49 #include <sys/syslog.h> 50 #include <net/radix.h> 51 #endif 52 53 #ifndef SMALL_KERNEL 54 #include <sys/socket.h> 55 #include <net/route.h> 56 #include <net/radix_mpath.h> 57 #endif 58 59 int max_keylen; 60 struct radix_mask *rn_mkfreelist; 61 struct radix_node_head *mask_rnhead; 62 static char *addmask_key; 63 static char normal_chars[] = {0, 0x80, 0xc0, 0xe0, 0xf0, 0xf8, 0xfc, 0xfe, -1}; 64 static char *rn_zeros, *rn_ones; 65 66 #define rn_masktop (mask_rnhead->rnh_treetop) 67 #undef Bcmp 68 #define Bcmp(a, b, l) (l == 0 ? 0 : bcmp((caddr_t)(a), (caddr_t)(b), (u_long)l)) 69 70 static int rn_satisfies_leaf(char *, struct radix_node *, int); 71 static int rn_lexobetter(void *, void *); 72 static struct radix_mask *rn_new_radix_mask(struct radix_node *, 73 struct radix_mask *); 74 75 /* 76 * The data structure for the keys is a radix tree with one way 77 * branching removed. The index rn_b at an internal node n represents a bit 78 * position to be tested. The tree is arranged so that all descendants 79 * of a node n have keys whose bits all agree up to position rn_b - 1. 80 * (We say the index of n is rn_b.) 81 * 82 * There is at least one descendant which has a one bit at position rn_b, 83 * and at least one with a zero there. 84 * 85 * A route is determined by a pair of key and mask. We require that the 86 * bit-wise logical and of the key and mask to be the key. 87 * We define the index of a route to associated with the mask to be 88 * the first bit number in the mask where 0 occurs (with bit number 0 89 * representing the highest order bit). 90 * 91 * We say a mask is normal if every bit is 0, past the index of the mask. 92 * If a node n has a descendant (k, m) with index(m) == index(n) == rn_b, 93 * and m is a normal mask, then the route applies to every descendant of n. 94 * If the index(m) < rn_b, this implies the trailing last few bits of k 95 * before bit b are all 0, (and hence consequently true of every descendant 96 * of n), so the route applies to all descendants of the node as well. 97 * 98 * Similar logic shows that a non-normal mask m such that 99 * index(m) <= index(n) could potentially apply to many children of n. 100 * Thus, for each non-host route, we attach its mask to a list at an internal 101 * node as high in the tree as we can go. 102 * 103 * The present version of the code makes use of normal routes in short- 104 * circuiting an explicit mask and compare operation when testing whether 105 * a key satisfies a normal route, and also in remembering the unique leaf 106 * that governs a subtree. 107 */ 108 109 struct radix_node * 110 rn_search(void *v_arg, struct radix_node *head) 111 { 112 struct radix_node *x; 113 caddr_t v; 114 115 for (x = head, v = v_arg; x->rn_b >= 0;) { 116 if (x->rn_bmask & v[x->rn_off]) 117 x = x->rn_r; 118 else 119 x = x->rn_l; 120 } 121 return (x); 122 } 123 124 struct radix_node * 125 rn_search_m(void *v_arg, struct radix_node *head, void *m_arg) 126 { 127 struct radix_node *x; 128 caddr_t v = v_arg, m = m_arg; 129 130 for (x = head; x->rn_b >= 0;) { 131 if ((x->rn_bmask & m[x->rn_off]) && 132 (x->rn_bmask & v[x->rn_off])) 133 x = x->rn_r; 134 else 135 x = x->rn_l; 136 } 137 return x; 138 } 139 140 int 141 rn_refines(void *m_arg, void *n_arg) 142 { 143 caddr_t m = m_arg, n = n_arg; 144 caddr_t lim, lim2 = lim = n + *(u_char *)n; 145 int longer = (*(u_char *)n++) - (int)(*(u_char *)m++); 146 int masks_are_equal = 1; 147 148 if (longer > 0) 149 lim -= longer; 150 while (n < lim) { 151 if (*n & ~(*m)) 152 return 0; 153 if (*n++ != *m++) 154 masks_are_equal = 0; 155 } 156 while (n < lim2) 157 if (*n++) 158 return 0; 159 if (masks_are_equal && (longer < 0)) 160 for (lim2 = m - longer; m < lim2; ) 161 if (*m++) 162 return 1; 163 return (!masks_are_equal); 164 } 165 166 struct radix_node * 167 rn_lookup(void *v_arg, void *m_arg, struct radix_node_head *head) 168 { 169 struct radix_node *x; 170 caddr_t netmask = 0; 171 172 if (m_arg) { 173 if ((x = rn_addmask(m_arg, 1, head->rnh_treetop->rn_off)) == 0) 174 return (0); 175 netmask = x->rn_key; 176 } 177 x = rn_match(v_arg, head); 178 if (x && netmask) { 179 while (x && x->rn_mask != netmask) 180 x = x->rn_dupedkey; 181 } 182 return x; 183 } 184 185 static int 186 rn_satisfies_leaf(char *trial, struct radix_node *leaf, int skip) 187 { 188 char *cp = trial, *cp2 = leaf->rn_key, *cp3 = leaf->rn_mask; 189 char *cplim; 190 int length = min(*(u_char *)cp, *(u_char *)cp2); 191 192 if (cp3 == 0) 193 cp3 = rn_ones; 194 else 195 length = min(length, *(u_char *)cp3); 196 cplim = cp + length; cp3 += skip; cp2 += skip; 197 for (cp += skip; cp < cplim; cp++, cp2++, cp3++) 198 if ((*cp ^ *cp2) & *cp3) 199 return 0; 200 return 1; 201 } 202 203 struct radix_node * 204 rn_match(void *v_arg, struct radix_node_head *head) 205 { 206 caddr_t v = v_arg; 207 struct radix_node *t = head->rnh_treetop, *x; 208 caddr_t cp = v, cp2; 209 caddr_t cplim; 210 struct radix_node *saved_t, *top = t; 211 int off = t->rn_off, vlen = *(u_char *)cp, matched_off; 212 int test, b, rn_b; 213 214 /* 215 * Open code rn_search(v, top) to avoid overhead of extra 216 * subroutine call. 217 */ 218 for (; t->rn_b >= 0; ) { 219 if (t->rn_bmask & cp[t->rn_off]) 220 t = t->rn_r; 221 else 222 t = t->rn_l; 223 } 224 /* 225 * See if we match exactly as a host destination 226 * or at least learn how many bits match, for normal mask finesse. 227 * 228 * It doesn't hurt us to limit how many bytes to check 229 * to the length of the mask, since if it matches we had a genuine 230 * match and the leaf we have is the most specific one anyway; 231 * if it didn't match with a shorter length it would fail 232 * with a long one. This wins big for class B&C netmasks which 233 * are probably the most common case... 234 */ 235 if (t->rn_mask) 236 vlen = *(u_char *)t->rn_mask; 237 cp += off; cp2 = t->rn_key + off; cplim = v + vlen; 238 for (; cp < cplim; cp++, cp2++) 239 if (*cp != *cp2) 240 goto on1; 241 /* 242 * This extra grot is in case we are explicitly asked 243 * to look up the default. Ugh! 244 */ 245 if ((t->rn_flags & RNF_ROOT) && t->rn_dupedkey) 246 t = t->rn_dupedkey; 247 return t; 248 on1: 249 test = (*cp ^ *cp2) & 0xff; /* find first bit that differs */ 250 for (b = 7; (test >>= 1) > 0;) 251 b--; 252 matched_off = cp - v; 253 b += matched_off << 3; 254 rn_b = -1 - b; 255 /* 256 * If there is a host route in a duped-key chain, it will be first. 257 */ 258 if ((saved_t = t)->rn_mask == 0) 259 t = t->rn_dupedkey; 260 for (; t; t = t->rn_dupedkey) 261 /* 262 * Even if we don't match exactly as a host, 263 * we may match if the leaf we wound up at is 264 * a route to a net. 265 */ 266 if (t->rn_flags & RNF_NORMAL) { 267 if (rn_b <= t->rn_b) 268 return t; 269 } else if (rn_satisfies_leaf(v, t, matched_off)) 270 return t; 271 t = saved_t; 272 /* start searching up the tree */ 273 do { 274 struct radix_mask *m; 275 t = t->rn_p; 276 m = t->rn_mklist; 277 if (m) { 278 /* 279 * If non-contiguous masks ever become important 280 * we can restore the masking and open coding of 281 * the search and satisfaction test and put the 282 * calculation of "off" back before the "do". 283 */ 284 do { 285 if (m->rm_flags & RNF_NORMAL) { 286 if (rn_b <= m->rm_b) 287 return (m->rm_leaf); 288 } else { 289 off = min(t->rn_off, matched_off); 290 x = rn_search_m(v, t, m->rm_mask); 291 while (x && x->rn_mask != m->rm_mask) 292 x = x->rn_dupedkey; 293 if (x && rn_satisfies_leaf(v, x, off)) 294 return x; 295 } 296 m = m->rm_mklist; 297 } while (m); 298 } 299 } while (t != top); 300 return 0; 301 } 302 303 #ifdef RN_DEBUG 304 int rn_nodenum; 305 struct radix_node *rn_clist; 306 int rn_saveinfo; 307 int rn_debug = 1; 308 #endif 309 310 struct radix_node * 311 rn_newpair(void *v, int b, struct radix_node nodes[2]) 312 { 313 struct radix_node *tt = nodes, *t = tt + 1; 314 t->rn_b = b; 315 t->rn_bmask = 0x80 >> (b & 7); 316 t->rn_l = tt; 317 t->rn_off = b >> 3; 318 tt->rn_b = -1; 319 tt->rn_key = (caddr_t)v; 320 tt->rn_p = t; 321 tt->rn_flags = t->rn_flags = RNF_ACTIVE; 322 #ifdef RN_DEBUG 323 tt->rn_info = rn_nodenum++; 324 t->rn_info = rn_nodenum++; 325 tt->rn_twin = t; 326 tt->rn_ybro = rn_clist; 327 rn_clist = tt; 328 #endif 329 return t; 330 } 331 332 struct radix_node * 333 rn_insert(void *v_arg, struct radix_node_head *head, 334 int *dupentry, struct radix_node nodes[2]) 335 { 336 caddr_t v = v_arg; 337 struct radix_node *top = head->rnh_treetop; 338 int head_off = top->rn_off, vlen = (int)*((u_char *)v); 339 struct radix_node *t = rn_search(v_arg, top); 340 caddr_t cp = v + head_off; 341 int b; 342 struct radix_node *tt; 343 /* 344 * Find first bit at which v and t->rn_key differ 345 */ 346 { 347 caddr_t cp2 = t->rn_key + head_off; 348 int cmp_res; 349 caddr_t cplim = v + vlen; 350 351 while (cp < cplim) 352 if (*cp2++ != *cp++) 353 goto on1; 354 *dupentry = 1; 355 return t; 356 on1: 357 *dupentry = 0; 358 cmp_res = (cp[-1] ^ cp2[-1]) & 0xff; 359 for (b = (cp - v) << 3; cmp_res; b--) 360 cmp_res >>= 1; 361 } 362 { 363 struct radix_node *p, *x = top; 364 cp = v; 365 do { 366 p = x; 367 if (cp[x->rn_off] & x->rn_bmask) 368 x = x->rn_r; 369 else 370 x = x->rn_l; 371 } while (b > (unsigned) x->rn_b); /* x->rn_b < b && x->rn_b >= 0 */ 372 #ifdef RN_DEBUG 373 if (rn_debug) 374 log(LOG_DEBUG, "rn_insert: Going In:\n"), traverse(p); 375 #endif 376 t = rn_newpair(v_arg, b, nodes); 377 tt = t->rn_l; 378 if ((cp[p->rn_off] & p->rn_bmask) == 0) 379 p->rn_l = t; 380 else 381 p->rn_r = t; 382 x->rn_p = t; 383 t->rn_p = p; /* frees x, p as temp vars below */ 384 if ((cp[t->rn_off] & t->rn_bmask) == 0) { 385 t->rn_r = x; 386 } else { 387 t->rn_r = tt; 388 t->rn_l = x; 389 } 390 #ifdef RN_DEBUG 391 if (rn_debug) 392 log(LOG_DEBUG, "rn_insert: Coming Out:\n"), traverse(p); 393 #endif 394 } 395 return (tt); 396 } 397 398 struct radix_node * 399 rn_addmask(void *n_arg, int search, int skip) 400 { 401 caddr_t netmask = (caddr_t)n_arg; 402 struct radix_node *x; 403 caddr_t cp, cplim; 404 int b = 0, mlen, j; 405 int maskduplicated, m0, isnormal; 406 struct radix_node *saved_x; 407 static int last_zeroed = 0; 408 409 if ((mlen = *(u_char *)netmask) > max_keylen) 410 mlen = max_keylen; 411 if (skip == 0) 412 skip = 1; 413 if (mlen <= skip) 414 return (mask_rnhead->rnh_nodes); 415 if (skip > 1) 416 Bcopy(rn_ones + 1, addmask_key + 1, skip - 1); 417 if ((m0 = mlen) > skip) 418 Bcopy(netmask + skip, addmask_key + skip, mlen - skip); 419 /* 420 * Trim trailing zeroes. 421 */ 422 for (cp = addmask_key + mlen; (cp > addmask_key) && cp[-1] == 0;) 423 cp--; 424 mlen = cp - addmask_key; 425 if (mlen <= skip) { 426 if (m0 >= last_zeroed) 427 last_zeroed = mlen; 428 return (mask_rnhead->rnh_nodes); 429 } 430 if (m0 < last_zeroed) 431 Bzero(addmask_key + m0, last_zeroed - m0); 432 *addmask_key = last_zeroed = mlen; 433 x = rn_search(addmask_key, rn_masktop); 434 if (Bcmp(addmask_key, x->rn_key, mlen) != 0) 435 x = 0; 436 if (x || search) 437 return (x); 438 R_Malloc(x, struct radix_node *, max_keylen + 2 * sizeof (*x)); 439 if ((saved_x = x) == 0) 440 return (0); 441 Bzero(x, max_keylen + 2 * sizeof (*x)); 442 netmask = cp = (caddr_t)(x + 2); 443 Bcopy(addmask_key, cp, mlen); 444 x = rn_insert(cp, mask_rnhead, &maskduplicated, x); 445 if (maskduplicated) { 446 log(LOG_ERR, "rn_addmask: mask impossibly already in tree\n"); 447 Free(saved_x); 448 return (x); 449 } 450 /* 451 * Calculate index of mask, and check for normalcy. 452 */ 453 cplim = netmask + mlen; 454 isnormal = 1; 455 for (cp = netmask + skip; (cp < cplim) && *(u_char *)cp == 0xff;) 456 cp++; 457 if (cp != cplim) { 458 for (j = 0x80; (j & *cp) != 0; j >>= 1) 459 b++; 460 if (*cp != normal_chars[b] || cp != (cplim - 1)) 461 isnormal = 0; 462 } 463 b += (cp - netmask) << 3; 464 x->rn_b = -1 - b; 465 if (isnormal) 466 x->rn_flags |= RNF_NORMAL; 467 return (x); 468 } 469 470 static int /* XXX: arbitrary ordering for non-contiguous masks */ 471 rn_lexobetter(void *m_arg, void *n_arg) 472 { 473 u_char *mp = m_arg, *np = n_arg, *lim; 474 475 /* 476 * Longer masks might not really be lexicographically better, 477 * but longer masks always have precedence since they must be checked 478 * first. The netmasks were normalized before calling this function and 479 * don't have unneeded trailing zeros. 480 */ 481 if (*mp > *np) 482 return 1; 483 if (*mp < *np) 484 return 0; 485 /* 486 * Must return the first difference between the masks 487 * to ensure deterministic sorting. 488 */ 489 lim = mp + *mp; 490 return (memcmp(mp, np, *lim) > 0); 491 } 492 493 static struct radix_mask * 494 rn_new_radix_mask(struct radix_node *tt, struct radix_mask *next) 495 { 496 struct radix_mask *m; 497 498 MKGet(m); 499 if (m == 0) { 500 log(LOG_ERR, "Mask for route not entered\n"); 501 return (0); 502 } 503 Bzero(m, sizeof *m); 504 m->rm_b = tt->rn_b; 505 m->rm_flags = tt->rn_flags; 506 if (tt->rn_flags & RNF_NORMAL) 507 m->rm_leaf = tt; 508 else 509 m->rm_mask = tt->rn_mask; 510 m->rm_mklist = next; 511 tt->rn_mklist = m; 512 return m; 513 } 514 515 struct radix_node * 516 rn_addroute(void *v_arg, void *n_arg, struct radix_node_head *head, 517 struct radix_node treenodes[2], u_int8_t prio) 518 { 519 caddr_t v = (caddr_t)v_arg, netmask = (caddr_t)n_arg; 520 struct radix_node *t, *x = NULL, *tt; 521 struct radix_node *saved_tt, *top = head->rnh_treetop; 522 short b = 0, b_leaf = 0; 523 int keyduplicated, prioinv = -1; 524 caddr_t mmask; 525 struct radix_mask *m, **mp; 526 527 /* 528 * In dealing with non-contiguous masks, there may be 529 * many different routes which have the same mask. 530 * We will find it useful to have a unique pointer to 531 * the mask to speed avoiding duplicate references at 532 * nodes and possibly save time in calculating indices. 533 */ 534 if (netmask) { 535 if ((x = rn_addmask(netmask, 0, top->rn_off)) == 0) 536 return (0); 537 b_leaf = x->rn_b; 538 b = -1 - x->rn_b; 539 netmask = x->rn_key; 540 } 541 /* 542 * Deal with duplicated keys: attach node to previous instance 543 */ 544 saved_tt = tt = rn_insert(v, head, &keyduplicated, treenodes); 545 if (keyduplicated) { 546 for (t = tt; tt; t = tt, tt = tt->rn_dupedkey) { 547 #ifndef SMALL_KERNEL 548 /* permit multipath, if enabled for the family */ 549 if (rn_mpath_capable(head) && netmask == tt->rn_mask) { 550 int mid; 551 /* 552 * Try to insert the new node in the middle 553 * of the list of any preexisting multipaths, 554 * to reduce the number of path disruptions 555 * that occur as a result of an insertion, 556 * per RFC2992. 557 * Additionally keep the list sorted by route 558 * priority. 559 */ 560 prioinv = 0; 561 tt = rn_mpath_prio(tt, prio); 562 if (((struct rtentry *)tt)->rt_priority != 563 prio) { 564 /* 565 * rn_mpath_prio returns the previous 566 * element if no element with the 567 * requested priority exists. It could 568 * be that the previous element comes 569 * with a bigger priority. 570 */ 571 if (((struct rtentry *)tt)-> 572 rt_priority > prio) 573 prioinv = 1; 574 t = tt; 575 break; 576 } 577 578 mid = rn_mpath_count(tt) / 2; 579 do { 580 t = tt; 581 tt = rn_mpath_next(tt, 0); 582 } while (tt && --mid > 0); 583 break; 584 } 585 #endif 586 if (tt->rn_mask == netmask) 587 return (0); 588 if (netmask == 0 || 589 (tt->rn_mask && 590 ((b_leaf < tt->rn_b) || /* index(netmask) > node */ 591 rn_refines(netmask, tt->rn_mask) || 592 rn_lexobetter(netmask, tt->rn_mask)))) 593 break; 594 } 595 /* 596 * If the mask is not duplicated, we wouldn't 597 * find it among possible duplicate key entries 598 * anyway, so the above test doesn't hurt. 599 * 600 * We sort the masks for a duplicated key the same way as 601 * in a masklist -- most specific to least specific. 602 * This may require the unfortunate nuisance of relocating 603 * the head of the list. 604 * 605 * We also reverse, or doubly link the list through the 606 * parent pointer. 607 */ 608 if (tt == saved_tt && prioinv) { 609 struct radix_node *xx; 610 /* link in at head of list */ 611 (tt = treenodes)->rn_dupedkey = t; 612 tt->rn_flags = t->rn_flags; 613 tt->rn_p = xx = t->rn_p; 614 t->rn_p = tt; 615 if (xx->rn_l == t) 616 xx->rn_l = tt; 617 else 618 xx->rn_r = tt; 619 saved_tt = tt; 620 } else if (prioinv == 1) { 621 (tt = treenodes)->rn_dupedkey = t; 622 if (t->rn_p == NULL) 623 panic("rn_addroute: t->rn_p is NULL"); 624 t->rn_p->rn_dupedkey = tt; 625 tt->rn_p = t->rn_p; 626 t->rn_p = tt; 627 } else { 628 (tt = treenodes)->rn_dupedkey = t->rn_dupedkey; 629 t->rn_dupedkey = tt; 630 tt->rn_p = t; 631 if (tt->rn_dupedkey) 632 tt->rn_dupedkey->rn_p = tt; 633 } 634 #ifdef RN_DEBUG 635 t=tt+1; 636 tt->rn_info = rn_nodenum++; 637 t->rn_info = rn_nodenum++; 638 tt->rn_twin = t; 639 tt->rn_ybro = rn_clist; 640 rn_clist = tt; 641 #endif 642 tt->rn_key = (caddr_t) v; 643 tt->rn_b = -1; 644 tt->rn_flags = RNF_ACTIVE; 645 } 646 /* 647 * Put mask in tree. 648 */ 649 if (netmask) { 650 tt->rn_mask = netmask; 651 tt->rn_b = x->rn_b; 652 tt->rn_flags |= x->rn_flags & RNF_NORMAL; 653 } 654 t = saved_tt->rn_p; 655 if (keyduplicated) 656 goto on2; 657 b_leaf = -1 - t->rn_b; 658 if (t->rn_r == saved_tt) 659 x = t->rn_l; 660 else 661 x = t->rn_r; 662 /* Promote general routes from below */ 663 if (x->rn_b < 0) { 664 struct radix_node *xx = NULL; 665 for (mp = &t->rn_mklist; x; xx = x, x = x->rn_dupedkey) { 666 if (xx && xx->rn_mklist && xx->rn_mask == x->rn_mask && 667 x->rn_mklist == 0) { 668 /* multipath route, bump refcount on first mklist */ 669 x->rn_mklist = xx->rn_mklist; 670 x->rn_mklist->rm_refs++; 671 } 672 if (x->rn_mask && (x->rn_b >= b_leaf) && x->rn_mklist == 0) { 673 *mp = m = rn_new_radix_mask(x, 0); 674 if (m) 675 mp = &m->rm_mklist; 676 } 677 } 678 } else if (x->rn_mklist) { 679 /* 680 * Skip over masks whose index is > that of new node 681 */ 682 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist) 683 if (m->rm_b >= b_leaf) 684 break; 685 t->rn_mklist = m; 686 *mp = 0; 687 } 688 on2: 689 /* Add new route to highest possible ancestor's list */ 690 if ((netmask == 0) || (b > t->rn_b )) 691 return tt; /* can't lift at all */ 692 b_leaf = tt->rn_b; 693 do { 694 x = t; 695 t = t->rn_p; 696 } while (b <= t->rn_b && x != top); 697 /* 698 * Search through routes associated with node to 699 * insert new route according to index. 700 * Need same criteria as when sorting dupedkeys to avoid 701 * double loop on deletion. 702 */ 703 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist) { 704 if (m->rm_b < b_leaf) 705 continue; 706 if (m->rm_b > b_leaf) 707 break; 708 if (m->rm_flags & RNF_NORMAL) { 709 mmask = m->rm_leaf->rn_mask; 710 if (keyduplicated) { 711 if (m->rm_leaf->rn_p == tt) 712 /* new route is better */ 713 m->rm_leaf = tt; 714 #ifdef DIAGNOSTIC 715 else { 716 for (t = m->rm_leaf; t; 717 t = t->rn_dupedkey) 718 if (t == tt) 719 break; 720 if (t == NULL) { 721 log(LOG_ERR, "Non-unique " 722 "normal route on dupedkey, " 723 "mask not entered\n"); 724 return tt; 725 } 726 } 727 #endif 728 m->rm_refs++; 729 tt->rn_mklist = m; 730 return tt; 731 } else if (tt->rn_flags & RNF_NORMAL) { 732 log(LOG_ERR, "Non-unique normal route," 733 " mask not entered\n"); 734 return tt; 735 } 736 } else 737 mmask = m->rm_mask; 738 if (mmask == netmask) { 739 m->rm_refs++; 740 tt->rn_mklist = m; 741 return tt; 742 } 743 if (rn_refines(netmask, mmask) || rn_lexobetter(netmask, mmask)) 744 break; 745 } 746 *mp = rn_new_radix_mask(tt, *mp); 747 return tt; 748 } 749 750 struct radix_node * 751 rn_delete(void *v_arg, void *netmask_arg, struct radix_node_head *head, 752 struct radix_node *rn) 753 { 754 struct radix_node *t, *p, *x, *tt; 755 struct radix_mask *m, *saved_m, **mp; 756 struct radix_node *dupedkey, *saved_tt, *top; 757 caddr_t v, netmask; 758 int b, head_off, vlen; 759 760 v = v_arg; 761 netmask = netmask_arg; 762 x = head->rnh_treetop; 763 tt = rn_search(v, x); 764 head_off = x->rn_off; 765 vlen = *(u_char *)v; 766 saved_tt = tt; 767 top = x; 768 if (tt == 0 || 769 Bcmp(v + head_off, tt->rn_key + head_off, vlen - head_off)) 770 return (0); 771 /* 772 * Delete our route from mask lists. 773 */ 774 if (netmask) { 775 if ((x = rn_addmask(netmask, 1, head_off)) == 0) 776 return (0); 777 netmask = x->rn_key; 778 while (tt->rn_mask != netmask) 779 if ((tt = tt->rn_dupedkey) == 0) 780 return (0); 781 } 782 #ifndef SMALL_KERNEL 783 if (rn) { 784 while (tt != rn) 785 if ((tt = tt->rn_dupedkey) == 0) 786 return (0); 787 } 788 #endif 789 if (tt->rn_mask == 0 || (saved_m = m = tt->rn_mklist) == 0) 790 goto on1; 791 if (tt->rn_flags & RNF_NORMAL) { 792 if (m->rm_leaf != tt && m->rm_refs == 0) { 793 log(LOG_ERR, "rn_delete: inconsistent normal " 794 "annotation\n"); 795 return (0); 796 } 797 if (m->rm_leaf != tt) { 798 if (--m->rm_refs >= 0) 799 goto on1; 800 } 801 /* tt is currently the head of the possible multipath chain */ 802 if (m->rm_refs > 0) { 803 if (tt->rn_dupedkey == NULL || 804 tt->rn_dupedkey->rn_mklist != m) { 805 log(LOG_ERR, "rn_delete: inconsistent " 806 "dupedkey list\n"); 807 return (0); 808 } 809 m->rm_leaf = tt->rn_dupedkey; 810 --m->rm_refs; 811 goto on1; 812 } 813 /* else tt is last and only route */ 814 } else { 815 if (m->rm_mask != tt->rn_mask) { 816 log(LOG_ERR, "rn_delete: inconsistent annotation\n"); 817 goto on1; 818 } 819 if (--m->rm_refs >= 0) 820 goto on1; 821 } 822 b = -1 - tt->rn_b; 823 t = saved_tt->rn_p; 824 if (b > t->rn_b) 825 goto on1; /* Wasn't lifted at all */ 826 do { 827 x = t; 828 t = t->rn_p; 829 } while (b <= t->rn_b && x != top); 830 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_mklist) 831 if (m == saved_m) { 832 *mp = m->rm_mklist; 833 MKFree(m); 834 break; 835 } 836 if (m == 0) { 837 log(LOG_ERR, "rn_delete: couldn't find our annotation\n"); 838 if (tt->rn_flags & RNF_NORMAL) 839 return (0); /* Dangling ref to us */ 840 } 841 on1: 842 /* 843 * Eliminate us from tree 844 */ 845 if (tt->rn_flags & RNF_ROOT) 846 return (0); 847 #ifdef RN_DEBUG 848 /* Get us out of the creation list */ 849 for (t = rn_clist; t && t->rn_ybro != tt; t = t->rn_ybro) 850 ; 851 if (t) t->rn_ybro = tt->rn_ybro; 852 #endif 853 t = tt->rn_p; 854 dupedkey = saved_tt->rn_dupedkey; 855 if (dupedkey) { 856 /* 857 * Here, tt is the deletion target, and 858 * saved_tt is the head of the dupedkey chain. 859 */ 860 if (tt == saved_tt) { 861 x = dupedkey; 862 x->rn_p = t; 863 if (t->rn_l == tt) 864 t->rn_l = x; 865 else 866 t->rn_r = x; 867 } else { 868 x = saved_tt; 869 t->rn_dupedkey = tt->rn_dupedkey; 870 if (tt->rn_dupedkey) 871 tt->rn_dupedkey->rn_p = t; 872 } 873 t = tt + 1; 874 if (t->rn_flags & RNF_ACTIVE) { 875 #ifndef RN_DEBUG 876 *++x = *t; 877 p = t->rn_p; 878 #else 879 b = t->rn_info; 880 *++x = *t; 881 t->rn_info = b; 882 p = t->rn_p; 883 #endif 884 if (p->rn_l == t) 885 p->rn_l = x; 886 else 887 p->rn_r = x; 888 x->rn_l->rn_p = x; 889 x->rn_r->rn_p = x; 890 } 891 goto out; 892 } 893 if (t->rn_l == tt) 894 x = t->rn_r; 895 else 896 x = t->rn_l; 897 p = t->rn_p; 898 if (p->rn_r == t) 899 p->rn_r = x; 900 else 901 p->rn_l = x; 902 x->rn_p = p; 903 /* 904 * Demote routes attached to us. 905 */ 906 if (t->rn_mklist) { 907 if (x->rn_b >= 0) { 908 for (mp = &x->rn_mklist; (m = *mp);) 909 mp = &m->rm_mklist; 910 *mp = t->rn_mklist; 911 } else { 912 /* If there are any key,mask pairs in a sibling 913 duped-key chain, some subset will appear sorted 914 in the same order attached to our mklist */ 915 for (m = t->rn_mklist; m && x; x = x->rn_dupedkey) 916 if (m == x->rn_mklist) { 917 struct radix_mask *mm = m->rm_mklist; 918 x->rn_mklist = 0; 919 if (--(m->rm_refs) < 0) 920 MKFree(m); 921 else if (m->rm_flags & RNF_NORMAL) 922 /* 923 * don't progress because this 924 * a multipath route. Next 925 * route will use the same m. 926 */ 927 mm = m; 928 m = mm; 929 } 930 if (m) 931 log(LOG_ERR, "%s %p at %p\n", 932 "rn_delete: Orphaned Mask", m, x); 933 } 934 } 935 /* 936 * We may be holding an active internal node in the tree. 937 */ 938 x = tt + 1; 939 if (t != x) { 940 #ifndef RN_DEBUG 941 *t = *x; 942 #else 943 b = t->rn_info; 944 *t = *x; 945 t->rn_info = b; 946 #endif 947 t->rn_l->rn_p = t; 948 t->rn_r->rn_p = t; 949 p = x->rn_p; 950 if (p->rn_l == x) 951 p->rn_l = t; 952 else 953 p->rn_r = t; 954 } 955 out: 956 tt->rn_flags &= ~RNF_ACTIVE; 957 tt[1].rn_flags &= ~RNF_ACTIVE; 958 return (tt); 959 } 960 961 int 962 rn_walktree(struct radix_node_head *h, int (*f)(struct radix_node *, void *, 963 u_int), void *w) 964 { 965 int error; 966 struct radix_node *base, *next; 967 struct radix_node *rn = h->rnh_treetop; 968 /* 969 * This gets complicated because we may delete the node 970 * while applying the function f to it, so we need to calculate 971 * the successor node in advance. 972 */ 973 /* First time through node, go left */ 974 while (rn->rn_b >= 0) 975 rn = rn->rn_l; 976 for (;;) { 977 base = rn; 978 /* If at right child go back up, otherwise, go right */ 979 while (rn->rn_p->rn_r == rn && (rn->rn_flags & RNF_ROOT) == 0) 980 rn = rn->rn_p; 981 /* Find the next *leaf* since next node might vanish, too */ 982 for (rn = rn->rn_p->rn_r; rn->rn_b >= 0;) 983 rn = rn->rn_l; 984 next = rn; 985 /* Process leaves */ 986 while ((rn = base) != NULL) { 987 base = rn->rn_dupedkey; 988 if (!(rn->rn_flags & RNF_ROOT) && 989 (error = (*f)(rn, w, h->rnh_rtableid))) 990 return (error); 991 } 992 rn = next; 993 if (rn->rn_flags & RNF_ROOT) 994 return (0); 995 } 996 /* NOTREACHED */ 997 } 998 999 int 1000 rn_inithead(void **head, int off) 1001 { 1002 struct radix_node_head *rnh; 1003 1004 if (*head) 1005 return (1); 1006 R_Malloc(rnh, struct radix_node_head *, sizeof (*rnh)); 1007 if (rnh == 0) 1008 return (0); 1009 *head = rnh; 1010 return rn_inithead0(rnh, off); 1011 } 1012 1013 int 1014 rn_inithead0(struct radix_node_head *rnh, int off) 1015 { 1016 struct radix_node *t, *tt, *ttt; 1017 1018 Bzero(rnh, sizeof (*rnh)); 1019 t = rn_newpair(rn_zeros, off, rnh->rnh_nodes); 1020 ttt = rnh->rnh_nodes + 2; 1021 t->rn_r = ttt; 1022 t->rn_p = t; 1023 tt = t->rn_l; 1024 tt->rn_flags = t->rn_flags = RNF_ROOT | RNF_ACTIVE; 1025 tt->rn_b = -1 - off; 1026 *ttt = *tt; 1027 ttt->rn_key = rn_ones; 1028 rnh->rnh_addaddr = rn_addroute; 1029 rnh->rnh_deladdr = rn_delete; 1030 rnh->rnh_matchaddr = rn_match; 1031 rnh->rnh_lookup = rn_lookup; 1032 rnh->rnh_walktree = rn_walktree; 1033 rnh->rnh_treetop = t; 1034 return (1); 1035 } 1036 1037 void 1038 rn_init() 1039 { 1040 char *cp, *cplim; 1041 #ifdef _KERNEL 1042 struct domain *dom; 1043 1044 for (dom = domains; dom; dom = dom->dom_next) 1045 if (dom->dom_maxrtkey > max_keylen) 1046 max_keylen = dom->dom_maxrtkey; 1047 #endif 1048 if (max_keylen == 0) { 1049 log(LOG_ERR, 1050 "rn_init: radix functions require max_keylen be set\n"); 1051 return; 1052 } 1053 R_Malloc(rn_zeros, char *, 3 * max_keylen); 1054 if (rn_zeros == NULL) 1055 panic("rn_init"); 1056 Bzero(rn_zeros, 3 * max_keylen); 1057 rn_ones = cp = rn_zeros + max_keylen; 1058 addmask_key = cplim = rn_ones + max_keylen; 1059 while (cp < cplim) 1060 *cp++ = -1; 1061 if (rn_inithead((void *)&mask_rnhead, 0) == 0) 1062 panic("rn_init 2"); 1063 } 1064