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 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * @(#)radix.c 8.4 (Berkeley) 11/2/94 34 * $FreeBSD: src/sys/net/radix.c,v 1.20.2.3 2002/04/28 05:40:25 suz Exp $ 35 */ 36 37 /* 38 * Routines to build and maintain radix trees for routing lookups. 39 */ 40 #include <sys/param.h> 41 #ifdef _KERNEL 42 #include <sys/systm.h> 43 #include <sys/malloc.h> 44 #include <sys/domain.h> 45 #include <sys/globaldata.h> 46 #include <sys/thread.h> 47 #else 48 #include <stdlib.h> 49 #endif 50 #include <sys/syslog.h> 51 #include <net/radix.h> 52 53 /* 54 * The arguments to the radix functions are really counted byte arrays with 55 * the length in the first byte. struct sockaddr's fit this type structurally. 56 */ 57 #define clen(c) (*(u_char *)(c)) 58 59 static int rn_walktree_from(struct radix_node_head *h, char *a, char *m, 60 walktree_f_t *f, void *w); 61 static int rn_walktree(struct radix_node_head *, walktree_f_t *, void *); 62 63 static struct radix_node 64 *rn_insert(char *, struct radix_node_head *, boolean_t *, 65 struct radix_node [2]), 66 *rn_newpair(char *, int, struct radix_node[2]), 67 *rn_search(const char *, struct radix_node *), 68 *rn_search_m(const char *, struct radix_node *, const char *); 69 70 static struct radix_mask *rn_mkfreelist; 71 static struct radix_node_head *mask_rnheads[MAXCPU]; 72 73 static char rn_zeros[RN_MAXKEYLEN]; 74 static char rn_ones[RN_MAXKEYLEN] = RN_MAXKEYONES; 75 76 static int rn_lexobetter(char *m, char *n); 77 static struct radix_mask * 78 rn_new_radix_mask(struct radix_node *tt, struct radix_mask *nextmask); 79 static boolean_t 80 rn_satisfies_leaf(char *trial, struct radix_node *leaf, int skip); 81 82 static __inline struct radix_mask * 83 MKGet(struct radix_mask **l) 84 { 85 struct radix_mask *m; 86 87 if (*l != NULL) { 88 m = *l; 89 *l = m->rm_next; 90 } else { 91 R_Malloc(m, struct radix_mask *, sizeof *m); 92 } 93 return m; 94 } 95 96 static __inline void 97 MKFree(struct radix_mask **l, struct radix_mask *m) 98 { 99 m->rm_next = *l; 100 *l = m; 101 } 102 103 /* 104 * The data structure for the keys is a radix tree with one way 105 * branching removed. The index rn_bit at an internal node n represents a bit 106 * position to be tested. The tree is arranged so that all descendants 107 * of a node n have keys whose bits all agree up to position rn_bit - 1. 108 * (We say the index of n is rn_bit.) 109 * 110 * There is at least one descendant which has a one bit at position rn_bit, 111 * and at least one with a zero there. 112 * 113 * A route is determined by a pair of key and mask. We require that the 114 * bit-wise logical and of the key and mask to be the key. 115 * We define the index of a route to associated with the mask to be 116 * the first bit number in the mask where 0 occurs (with bit number 0 117 * representing the highest order bit). 118 * 119 * We say a mask is normal if every bit is 0, past the index of the mask. 120 * If a node n has a descendant (k, m) with index(m) == index(n) == rn_bit, 121 * and m is a normal mask, then the route applies to every descendant of n. 122 * If the index(m) < rn_bit, this implies the trailing last few bits of k 123 * before bit b are all 0, (and hence consequently true of every descendant 124 * of n), so the route applies to all descendants of the node as well. 125 * 126 * Similar logic shows that a non-normal mask m such that 127 * index(m) <= index(n) could potentially apply to many children of n. 128 * Thus, for each non-host route, we attach its mask to a list at an internal 129 * node as high in the tree as we can go. 130 * 131 * The present version of the code makes use of normal routes in short- 132 * circuiting an explict mask and compare operation when testing whether 133 * a key satisfies a normal route, and also in remembering the unique leaf 134 * that governs a subtree. 135 */ 136 137 static struct radix_node * 138 rn_search(const char *v, struct radix_node *head) 139 { 140 struct radix_node *x; 141 142 x = head; 143 while (x->rn_bit >= 0) { 144 if (x->rn_bmask & v[x->rn_offset]) 145 x = x->rn_right; 146 else 147 x = x->rn_left; 148 } 149 return (x); 150 } 151 152 static struct radix_node * 153 rn_search_m(const char *v, struct radix_node *head, const char *m) 154 { 155 struct radix_node *x; 156 157 for (x = head; x->rn_bit >= 0;) { 158 if ((x->rn_bmask & m[x->rn_offset]) && 159 (x->rn_bmask & v[x->rn_offset])) 160 x = x->rn_right; 161 else 162 x = x->rn_left; 163 } 164 return x; 165 } 166 167 boolean_t 168 rn_refines(char *m, char *n) 169 { 170 char *lim, *lim2; 171 int longer = clen(n++) - clen(m++); 172 boolean_t masks_are_equal = TRUE; 173 174 lim2 = lim = n + clen(n); 175 if (longer > 0) 176 lim -= longer; 177 while (n < lim) { 178 if (*n & ~(*m)) 179 return FALSE; 180 if (*n++ != *m++) 181 masks_are_equal = FALSE; 182 } 183 while (n < lim2) 184 if (*n++) 185 return FALSE; 186 if (masks_are_equal && (longer < 0)) 187 for (lim2 = m - longer; m < lim2; ) 188 if (*m++) 189 return TRUE; 190 return (!masks_are_equal); 191 } 192 193 struct radix_node * 194 rn_lookup(char *key, char *mask, struct radix_node_head *head) 195 { 196 struct radix_node *x; 197 char *netmask = NULL; 198 199 if (mask != NULL) { 200 x = rn_addmask(mask, TRUE, head->rnh_treetop->rn_offset, 201 head->rnh_maskhead); 202 if (x == NULL) 203 return (NULL); 204 netmask = x->rn_key; 205 } 206 x = rn_match(key, head); 207 if (x != NULL && netmask != NULL) { 208 while (x != NULL && x->rn_mask != netmask) 209 x = x->rn_dupedkey; 210 } 211 return x; 212 } 213 214 static boolean_t 215 rn_satisfies_leaf(char *trial, struct radix_node *leaf, int skip) 216 { 217 char *cp = trial, *cp2 = leaf->rn_key, *cp3 = leaf->rn_mask; 218 char *cplim; 219 int length = min(clen(cp), clen(cp2)); 220 221 if (cp3 == NULL) 222 cp3 = rn_ones; 223 else 224 length = min(length, clen(cp3)); 225 cplim = cp + length; 226 cp3 += skip; 227 cp2 += skip; 228 for (cp += skip; cp < cplim; cp++, cp2++, cp3++) 229 if ((*cp ^ *cp2) & *cp3) 230 return FALSE; 231 return TRUE; 232 } 233 234 struct radix_node * 235 rn_match(char *key, struct radix_node_head *head) 236 { 237 struct radix_node *t, *x; 238 char *cp = key, *cp2; 239 char *cplim; 240 struct radix_node *saved_t, *top = head->rnh_treetop; 241 int off = top->rn_offset, klen, matched_off; 242 int test, b, rn_bit; 243 244 t = rn_search(key, top); 245 /* 246 * See if we match exactly as a host destination 247 * or at least learn how many bits match, for normal mask finesse. 248 * 249 * It doesn't hurt us to limit how many bytes to check 250 * to the length of the mask, since if it matches we had a genuine 251 * match and the leaf we have is the most specific one anyway; 252 * if it didn't match with a shorter length it would fail 253 * with a long one. This wins big for class B&C netmasks which 254 * are probably the most common case... 255 */ 256 if (t->rn_mask != NULL) 257 klen = clen(t->rn_mask); 258 else 259 klen = clen(key); 260 cp += off; cp2 = t->rn_key + off; cplim = key + klen; 261 for (; cp < cplim; cp++, cp2++) 262 if (*cp != *cp2) 263 goto on1; 264 /* 265 * This extra grot is in case we are explicitly asked 266 * to look up the default. Ugh! 267 * 268 * Never return the root node itself, it seems to cause a 269 * lot of confusion. 270 */ 271 if (t->rn_flags & RNF_ROOT) 272 t = t->rn_dupedkey; 273 return t; 274 on1: 275 test = (*cp ^ *cp2) & 0xff; /* find first bit that differs */ 276 for (b = 7; (test >>= 1) > 0;) 277 b--; 278 matched_off = cp - key; 279 b += matched_off << 3; 280 rn_bit = -1 - b; 281 /* 282 * If there is a host route in a duped-key chain, it will be first. 283 */ 284 if ((saved_t = t)->rn_mask == NULL) 285 t = t->rn_dupedkey; 286 for (; t; t = t->rn_dupedkey) { 287 /* 288 * Even if we don't match exactly as a host, 289 * we may match if the leaf we wound up at is 290 * a route to a net. 291 */ 292 if (t->rn_flags & RNF_NORMAL) { 293 if (rn_bit <= t->rn_bit) 294 return t; 295 } else if (rn_satisfies_leaf(key, t, matched_off)) 296 return t; 297 } 298 t = saved_t; 299 /* start searching up the tree */ 300 do { 301 struct radix_mask *m; 302 303 t = t->rn_parent; 304 /* 305 * If non-contiguous masks ever become important 306 * we can restore the masking and open coding of 307 * the search and satisfaction test and put the 308 * calculation of "off" back before the "do". 309 */ 310 m = t->rn_mklist; 311 while (m != NULL) { 312 if (m->rm_flags & RNF_NORMAL) { 313 if (rn_bit <= m->rm_bit) 314 return (m->rm_leaf); 315 } else { 316 off = min(t->rn_offset, matched_off); 317 x = rn_search_m(key, t, m->rm_mask); 318 while (x != NULL && x->rn_mask != m->rm_mask) 319 x = x->rn_dupedkey; 320 if (x && rn_satisfies_leaf(key, x, off)) 321 return x; 322 } 323 m = m->rm_next; 324 } 325 } while (t != top); 326 return NULL; 327 } 328 329 #ifdef RN_DEBUG 330 int rn_nodenum; 331 struct radix_node *rn_clist; 332 int rn_saveinfo; 333 boolean_t rn_debug = TRUE; 334 #endif 335 336 static struct radix_node * 337 rn_newpair(char *key, int indexbit, struct radix_node nodes[2]) 338 { 339 struct radix_node *leaf = &nodes[0], *interior = &nodes[1]; 340 341 interior->rn_bit = indexbit; 342 interior->rn_bmask = 0x80 >> (indexbit & 0x7); 343 interior->rn_offset = indexbit >> 3; 344 interior->rn_left = leaf; 345 interior->rn_mklist = NULL; 346 347 leaf->rn_bit = -1; 348 leaf->rn_key = key; 349 leaf->rn_parent = interior; 350 leaf->rn_flags = interior->rn_flags = RNF_ACTIVE; 351 leaf->rn_mklist = NULL; 352 353 #ifdef RN_DEBUG 354 leaf->rn_info = rn_nodenum++; 355 interior->rn_info = rn_nodenum++; 356 leaf->rn_twin = interior; 357 leaf->rn_ybro = rn_clist; 358 rn_clist = leaf; 359 #endif 360 return interior; 361 } 362 363 static struct radix_node * 364 rn_insert(char *key, struct radix_node_head *head, boolean_t *dupentry, 365 struct radix_node nodes[2]) 366 { 367 struct radix_node *top = head->rnh_treetop; 368 int head_off = top->rn_offset, klen = clen(key); 369 struct radix_node *t = rn_search(key, top); 370 char *cp = key + head_off; 371 int b; 372 struct radix_node *tt; 373 374 /* 375 * Find first bit at which the key and t->rn_key differ 376 */ 377 { 378 char *cp2 = t->rn_key + head_off; 379 int cmp_res; 380 char *cplim = key + klen; 381 382 while (cp < cplim) 383 if (*cp2++ != *cp++) 384 goto on1; 385 *dupentry = TRUE; 386 return t; 387 on1: 388 *dupentry = FALSE; 389 cmp_res = (cp[-1] ^ cp2[-1]) & 0xff; 390 for (b = (cp - key) << 3; cmp_res; b--) 391 cmp_res >>= 1; 392 } 393 { 394 struct radix_node *p, *x = top; 395 396 cp = key; 397 do { 398 p = x; 399 if (cp[x->rn_offset] & x->rn_bmask) 400 x = x->rn_right; 401 else 402 x = x->rn_left; 403 } while (b > (unsigned) x->rn_bit); 404 /* x->rn_bit < b && x->rn_bit >= 0 */ 405 #ifdef RN_DEBUG 406 if (rn_debug) 407 log(LOG_DEBUG, "rn_insert: Going In:\n"), traverse(p); 408 #endif 409 t = rn_newpair(key, b, nodes); 410 tt = t->rn_left; 411 if ((cp[p->rn_offset] & p->rn_bmask) == 0) 412 p->rn_left = t; 413 else 414 p->rn_right = t; 415 x->rn_parent = t; 416 t->rn_parent = p; /* frees x, p as temp vars below */ 417 if ((cp[t->rn_offset] & t->rn_bmask) == 0) { 418 t->rn_right = x; 419 } else { 420 t->rn_right = tt; 421 t->rn_left = x; 422 } 423 #ifdef RN_DEBUG 424 if (rn_debug) 425 log(LOG_DEBUG, "rn_insert: Coming Out:\n"), traverse(p); 426 #endif 427 } 428 return (tt); 429 } 430 431 struct radix_node * 432 rn_addmask(char *netmask, boolean_t search, int skip, 433 struct radix_node_head *mask_rnh) 434 { 435 struct radix_node *x, *saved_x; 436 char *cp, *cplim; 437 int b = 0, mlen, m0, j; 438 boolean_t maskduplicated, isnormal; 439 static int last_zeroed = 0; 440 char *addmask_key; 441 442 if ((mlen = clen(netmask)) > RN_MAXKEYLEN) 443 mlen = RN_MAXKEYLEN; 444 if (skip == 0) 445 skip = 1; 446 if (mlen <= skip) 447 return (mask_rnh->rnh_nodes); 448 R_Malloc(addmask_key, char *, RN_MAXKEYLEN); 449 if (addmask_key == NULL) 450 return NULL; 451 if (skip > 1) 452 bcopy(rn_ones + 1, addmask_key + 1, skip - 1); 453 if ((m0 = mlen) > skip) 454 bcopy(netmask + skip, addmask_key + skip, mlen - skip); 455 /* 456 * Trim trailing zeroes. 457 */ 458 for (cp = addmask_key + mlen; (cp > addmask_key) && cp[-1] == 0;) 459 cp--; 460 mlen = cp - addmask_key; 461 if (mlen <= skip) { 462 if (m0 >= last_zeroed) 463 last_zeroed = mlen; 464 Free(addmask_key); 465 return (mask_rnh->rnh_nodes); 466 } 467 if (m0 < last_zeroed) 468 bzero(addmask_key + m0, last_zeroed - m0); 469 *addmask_key = last_zeroed = mlen; 470 x = rn_search(addmask_key, mask_rnh->rnh_treetop); 471 if (x->rn_key == NULL) { 472 kprintf("WARNING: radix_node->rn_key is NULL rn=%p\n", x); 473 print_backtrace(-1); 474 x = NULL; 475 } else if (bcmp(addmask_key, x->rn_key, mlen) != 0) { 476 x = NULL; 477 } 478 if (x != NULL || search) 479 goto out; 480 R_Malloc(x, struct radix_node *, RN_MAXKEYLEN + 2 * (sizeof *x)); 481 if ((saved_x = x) == NULL) 482 goto out; 483 bzero(x, RN_MAXKEYLEN + 2 * (sizeof *x)); 484 netmask = cp = (char *)(x + 2); 485 bcopy(addmask_key, cp, mlen); 486 x = rn_insert(cp, mask_rnh, &maskduplicated, x); 487 if (maskduplicated) { 488 log(LOG_ERR, "rn_addmask: mask impossibly already in tree"); 489 Free(saved_x); 490 goto out; 491 } 492 /* 493 * Calculate index of mask, and check for normalcy. 494 */ 495 isnormal = TRUE; 496 cplim = netmask + mlen; 497 for (cp = netmask + skip; cp < cplim && clen(cp) == 0xff;) 498 cp++; 499 if (cp != cplim) { 500 static const char normal_chars[] = { 501 0, 0x80, 0xc0, 0xe0, 0xf0, 0xf8, 0xfc, 0xfe, -1 502 }; 503 504 for (j = 0x80; (j & *cp) != 0; j >>= 1) 505 b++; 506 if (*cp != normal_chars[b] || cp != (cplim - 1)) 507 isnormal = FALSE; 508 } 509 b += (cp - netmask) << 3; 510 x->rn_bit = -1 - b; 511 if (isnormal) 512 x->rn_flags |= RNF_NORMAL; 513 out: 514 Free(addmask_key); 515 return (x); 516 } 517 518 /* XXX: arbitrary ordering for non-contiguous masks */ 519 static boolean_t 520 rn_lexobetter(char *mp, char *np) 521 { 522 char *lim; 523 524 if ((unsigned) *mp > (unsigned) *np) 525 return TRUE;/* not really, but need to check longer one first */ 526 if (*mp == *np) 527 for (lim = mp + clen(mp); mp < lim;) 528 if (*mp++ > *np++) 529 return TRUE; 530 return FALSE; 531 } 532 533 static struct radix_mask * 534 rn_new_radix_mask(struct radix_node *tt, struct radix_mask *nextmask) 535 { 536 struct radix_mask *m; 537 538 m = MKGet(&rn_mkfreelist); 539 if (m == NULL) { 540 log(LOG_ERR, "Mask for route not entered\n"); 541 return (NULL); 542 } 543 bzero(m, sizeof *m); 544 m->rm_bit = tt->rn_bit; 545 m->rm_flags = tt->rn_flags; 546 if (tt->rn_flags & RNF_NORMAL) 547 m->rm_leaf = tt; 548 else 549 m->rm_mask = tt->rn_mask; 550 m->rm_next = nextmask; 551 tt->rn_mklist = m; 552 return m; 553 } 554 555 struct radix_node * 556 rn_addroute(char *key, char *netmask, struct radix_node_head *head, 557 struct radix_node treenodes[2]) 558 { 559 struct radix_node *t, *x = NULL, *tt; 560 struct radix_node *saved_tt, *top = head->rnh_treetop; 561 short b = 0, b_leaf = 0; 562 boolean_t keyduplicated; 563 char *mmask; 564 struct radix_mask *m, **mp; 565 566 /* 567 * In dealing with non-contiguous masks, there may be 568 * many different routes which have the same mask. 569 * We will find it useful to have a unique pointer to 570 * the mask to speed avoiding duplicate references at 571 * nodes and possibly save time in calculating indices. 572 */ 573 if (netmask != NULL) { 574 if ((x = rn_addmask(netmask, FALSE, top->rn_offset, 575 head->rnh_maskhead)) == NULL) 576 return (NULL); 577 b_leaf = x->rn_bit; 578 b = -1 - x->rn_bit; 579 netmask = x->rn_key; 580 } 581 /* 582 * Deal with duplicated keys: attach node to previous instance 583 */ 584 saved_tt = tt = rn_insert(key, head, &keyduplicated, treenodes); 585 if (keyduplicated) { 586 for (t = tt; tt; t = tt, tt = tt->rn_dupedkey) { 587 if (tt->rn_mask == netmask) 588 return (NULL); 589 if (netmask == NULL || 590 (tt->rn_mask && 591 ((b_leaf < tt->rn_bit) /* index(netmask) > node */ 592 || rn_refines(netmask, tt->rn_mask) 593 || rn_lexobetter(netmask, tt->rn_mask)))) 594 break; 595 } 596 /* 597 * If the mask is not duplicated, we wouldn't 598 * find it among possible duplicate key entries 599 * anyway, so the above test doesn't hurt. 600 * 601 * We sort the masks for a duplicated key the same way as 602 * in a masklist -- most specific to least specific. 603 * This may require the unfortunate nuisance of relocating 604 * the head of the list. 605 */ 606 if (tt == saved_tt) { 607 struct radix_node *xx = x; 608 /* link in at head of list */ 609 (tt = treenodes)->rn_dupedkey = t; 610 tt->rn_flags = t->rn_flags; 611 tt->rn_parent = x = t->rn_parent; 612 t->rn_parent = tt; /* parent */ 613 if (x->rn_left == t) 614 x->rn_left = tt; 615 else 616 x->rn_right = tt; 617 saved_tt = tt; x = xx; 618 } else { 619 (tt = treenodes)->rn_dupedkey = t->rn_dupedkey; 620 t->rn_dupedkey = tt; 621 tt->rn_parent = t; /* parent */ 622 if (tt->rn_dupedkey != NULL) /* parent */ 623 tt->rn_dupedkey->rn_parent = tt; /* parent */ 624 } 625 #ifdef RN_DEBUG 626 t=tt+1; tt->rn_info = rn_nodenum++; t->rn_info = rn_nodenum++; 627 tt->rn_twin = t; tt->rn_ybro = rn_clist; rn_clist = tt; 628 #endif 629 tt->rn_key = key; 630 tt->rn_bit = -1; 631 tt->rn_flags = RNF_ACTIVE; 632 } 633 /* 634 * Put mask in tree. 635 */ 636 if (netmask != NULL) { 637 tt->rn_mask = netmask; 638 tt->rn_bit = x->rn_bit; 639 tt->rn_flags |= x->rn_flags & RNF_NORMAL; 640 } 641 t = saved_tt->rn_parent; 642 if (keyduplicated) 643 goto on2; 644 b_leaf = -1 - t->rn_bit; 645 if (t->rn_right == saved_tt) 646 x = t->rn_left; 647 else 648 x = t->rn_right; 649 /* Promote general routes from below */ 650 if (x->rn_bit < 0) { 651 mp = &t->rn_mklist; 652 while (x != NULL) { 653 if (x->rn_mask != NULL && 654 x->rn_bit >= b_leaf && 655 x->rn_mklist == NULL) { 656 *mp = m = rn_new_radix_mask(x, NULL); 657 if (m != NULL) 658 mp = &m->rm_next; 659 } 660 x = x->rn_dupedkey; 661 } 662 } else if (x->rn_mklist != NULL) { 663 /* 664 * Skip over masks whose index is > that of new node 665 */ 666 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_next) 667 if (m->rm_bit >= b_leaf) 668 break; 669 t->rn_mklist = m; 670 *mp = NULL; 671 } 672 on2: 673 /* Add new route to highest possible ancestor's list */ 674 if ((netmask == NULL) || (b > t->rn_bit )) 675 return tt; /* can't lift at all */ 676 b_leaf = tt->rn_bit; 677 do { 678 x = t; 679 t = t->rn_parent; 680 } while (b <= t->rn_bit && x != top); 681 /* 682 * Search through routes associated with node to 683 * insert new route according to index. 684 * Need same criteria as when sorting dupedkeys to avoid 685 * double loop on deletion. 686 */ 687 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_next) { 688 if (m->rm_bit < b_leaf) 689 continue; 690 if (m->rm_bit > b_leaf) 691 break; 692 if (m->rm_flags & RNF_NORMAL) { 693 mmask = m->rm_leaf->rn_mask; 694 if (tt->rn_flags & RNF_NORMAL) { 695 log(LOG_ERR, 696 "Non-unique normal route, mask not entered\n"); 697 return tt; 698 } 699 } else 700 mmask = m->rm_mask; 701 if (mmask == netmask) { 702 m->rm_refs++; 703 tt->rn_mklist = m; 704 return tt; 705 } 706 if (rn_refines(netmask, mmask) || rn_lexobetter(netmask, mmask)) 707 break; 708 } 709 *mp = rn_new_radix_mask(tt, *mp); 710 return tt; 711 } 712 713 struct radix_node * 714 rn_delete(char *key, char *netmask, struct radix_node_head *head) 715 { 716 struct radix_node *t, *p, *x, *tt; 717 struct radix_mask *m, *saved_m, **mp; 718 struct radix_node *dupedkey, *saved_tt, *top; 719 int b, head_off, klen; 720 721 x = head->rnh_treetop; 722 tt = rn_search(key, x); 723 head_off = x->rn_offset; 724 klen = clen(key); 725 saved_tt = tt; 726 top = x; 727 if (tt == NULL || 728 bcmp(key + head_off, tt->rn_key + head_off, klen - head_off)) 729 return (NULL); 730 /* 731 * Delete our route from mask lists. 732 */ 733 if (netmask != NULL) { 734 if ((x = rn_addmask(netmask, TRUE, head_off, 735 head->rnh_maskhead)) == NULL) 736 return (NULL); 737 netmask = x->rn_key; 738 while (tt->rn_mask != netmask) 739 if ((tt = tt->rn_dupedkey) == NULL) 740 return (NULL); 741 } 742 if (tt->rn_mask == NULL || (saved_m = m = tt->rn_mklist) == NULL) 743 goto on1; 744 if (tt->rn_flags & RNF_NORMAL) { 745 if (m->rm_leaf != tt || m->rm_refs > 0) { 746 log(LOG_ERR, "rn_delete: inconsistent annotation\n"); 747 return (NULL); /* dangling ref could cause disaster */ 748 } 749 } else { 750 if (m->rm_mask != tt->rn_mask) { 751 log(LOG_ERR, "rn_delete: inconsistent annotation\n"); 752 goto on1; 753 } 754 if (--m->rm_refs >= 0) 755 goto on1; 756 } 757 b = -1 - tt->rn_bit; 758 t = saved_tt->rn_parent; 759 if (b > t->rn_bit) 760 goto on1; /* Wasn't lifted at all */ 761 do { 762 x = t; 763 t = t->rn_parent; 764 } while (b <= t->rn_bit && x != top); 765 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_next) 766 if (m == saved_m) { 767 *mp = m->rm_next; 768 MKFree(&rn_mkfreelist, m); 769 break; 770 } 771 if (m == NULL) { 772 log(LOG_ERR, "rn_delete: couldn't find our annotation\n"); 773 if (tt->rn_flags & RNF_NORMAL) 774 return (NULL); /* Dangling ref to us */ 775 } 776 on1: 777 /* 778 * Eliminate us from tree 779 */ 780 if (tt->rn_flags & RNF_ROOT) 781 return (NULL); 782 #ifdef RN_DEBUG 783 /* Get us out of the creation list */ 784 for (t = rn_clist; t && t->rn_ybro != tt; t = t->rn_ybro) {} 785 if (t) t->rn_ybro = tt->rn_ybro; 786 #endif 787 t = tt->rn_parent; 788 dupedkey = saved_tt->rn_dupedkey; 789 if (dupedkey != NULL) { 790 /* 791 * at this point, tt is the deletion target and saved_tt 792 * is the head of the dupekey chain 793 */ 794 if (tt == saved_tt) { 795 /* remove from head of chain */ 796 x = dupedkey; x->rn_parent = t; 797 if (t->rn_left == tt) 798 t->rn_left = x; 799 else 800 t->rn_right = x; 801 } else { 802 /* find node in front of tt on the chain */ 803 for (x = p = saved_tt; p && p->rn_dupedkey != tt;) 804 p = p->rn_dupedkey; 805 if (p) { 806 p->rn_dupedkey = tt->rn_dupedkey; 807 if (tt->rn_dupedkey) /* parent */ 808 tt->rn_dupedkey->rn_parent = p; 809 /* parent */ 810 } else log(LOG_ERR, "rn_delete: couldn't find us\n"); 811 } 812 t = tt + 1; 813 if (t->rn_flags & RNF_ACTIVE) { 814 #ifndef RN_DEBUG 815 *++x = *t; 816 p = t->rn_parent; 817 #else 818 b = t->rn_info; 819 *++x = *t; 820 t->rn_info = b; 821 p = t->rn_parent; 822 #endif 823 if (p->rn_left == t) 824 p->rn_left = x; 825 else 826 p->rn_right = x; 827 x->rn_left->rn_parent = x; 828 x->rn_right->rn_parent = x; 829 } 830 goto out; 831 } 832 if (t->rn_left == tt) 833 x = t->rn_right; 834 else 835 x = t->rn_left; 836 p = t->rn_parent; 837 if (p->rn_right == t) 838 p->rn_right = x; 839 else 840 p->rn_left = x; 841 x->rn_parent = p; 842 /* 843 * Demote routes attached to us. 844 */ 845 if (t->rn_mklist != NULL) { 846 if (x->rn_bit >= 0) { 847 for (mp = &x->rn_mklist; (m = *mp);) 848 mp = &m->rm_next; 849 *mp = t->rn_mklist; 850 } else { 851 /* 852 * If there are any (key, mask) pairs in a sibling 853 * duped-key chain, some subset will appear sorted 854 * in the same order attached to our mklist. 855 */ 856 for (m = t->rn_mklist; m && x; x = x->rn_dupedkey) 857 if (m == x->rn_mklist) { 858 struct radix_mask *mm = m->rm_next; 859 860 x->rn_mklist = NULL; 861 if (--(m->rm_refs) < 0) 862 MKFree(&rn_mkfreelist, m); 863 m = mm; 864 } 865 if (m) 866 log(LOG_ERR, 867 "rn_delete: Orphaned Mask %p at %p\n", 868 (void *)m, (void *)x); 869 } 870 } 871 /* 872 * We may be holding an active internal node in the tree. 873 */ 874 x = tt + 1; 875 if (t != x) { 876 #ifndef RN_DEBUG 877 *t = *x; 878 #else 879 b = t->rn_info; 880 *t = *x; 881 t->rn_info = b; 882 #endif 883 t->rn_left->rn_parent = t; 884 t->rn_right->rn_parent = t; 885 p = x->rn_parent; 886 if (p->rn_left == x) 887 p->rn_left = t; 888 else 889 p->rn_right = t; 890 } 891 out: 892 tt->rn_flags &= ~RNF_ACTIVE; 893 tt[1].rn_flags &= ~RNF_ACTIVE; 894 return (tt); 895 } 896 897 /* 898 * This is the same as rn_walktree() except for the parameters and the 899 * exit. 900 */ 901 static int 902 rn_walktree_from(struct radix_node_head *h, char *xa, char *xm, 903 walktree_f_t *f, void *w) 904 { 905 struct radix_node *base, *next; 906 struct radix_node *rn, *last = NULL /* shut up gcc */; 907 boolean_t stopping = FALSE; 908 int lastb, error; 909 910 /* 911 * rn_search_m is sort-of-open-coded here. 912 */ 913 /* kprintf("about to search\n"); */ 914 for (rn = h->rnh_treetop; rn->rn_bit >= 0; ) { 915 last = rn; 916 /* kprintf("rn_bit %d, rn_bmask %x, xm[rn_offset] %x\n", 917 rn->rn_bit, rn->rn_bmask, xm[rn->rn_offset]); */ 918 if (!(rn->rn_bmask & xm[rn->rn_offset])) { 919 break; 920 } 921 if (rn->rn_bmask & xa[rn->rn_offset]) { 922 rn = rn->rn_right; 923 } else { 924 rn = rn->rn_left; 925 } 926 } 927 /* kprintf("done searching\n"); */ 928 929 /* 930 * Two cases: either we stepped off the end of our mask, 931 * in which case last == rn, or we reached a leaf, in which 932 * case we want to start from the last node we looked at. 933 * Either way, last is the node we want to start from. 934 */ 935 rn = last; 936 lastb = rn->rn_bit; 937 938 /* kprintf("rn %p, lastb %d\n", rn, lastb);*/ 939 940 /* 941 * This gets complicated because we may delete the node 942 * while applying the function f to it, so we need to calculate 943 * the successor node in advance. 944 */ 945 while (rn->rn_bit >= 0) 946 rn = rn->rn_left; 947 948 while (!stopping) { 949 /* kprintf("node %p (%d)\n", rn, rn->rn_bit); */ 950 base = rn; 951 /* If at right child go back up, otherwise, go right */ 952 while (rn->rn_parent->rn_right == rn && 953 !(rn->rn_flags & RNF_ROOT)) { 954 rn = rn->rn_parent; 955 956 /* if went up beyond last, stop */ 957 if (rn->rn_bit < lastb) { 958 stopping = TRUE; 959 /* kprintf("up too far\n"); */ 960 } 961 } 962 963 /* Find the next *leaf* since next node might vanish, too */ 964 for (rn = rn->rn_parent->rn_right; rn->rn_bit >= 0;) 965 rn = rn->rn_left; 966 next = rn; 967 /* Process leaves */ 968 while ((rn = base) != NULL) { 969 base = rn->rn_dupedkey; 970 /* kprintf("leaf %p\n", rn); */ 971 if (!(rn->rn_flags & RNF_ROOT) && (error = (*f)(rn, w))) 972 return (error); 973 } 974 rn = next; 975 976 if (rn->rn_flags & RNF_ROOT) { 977 /* kprintf("root, stopping"); */ 978 stopping = TRUE; 979 } 980 981 } 982 return 0; 983 } 984 985 static int 986 rn_walktree(struct radix_node_head *h, walktree_f_t *f, void *w) 987 { 988 struct radix_node *base, *next; 989 struct radix_node *rn = h->rnh_treetop; 990 int error; 991 992 /* 993 * This gets complicated because we may delete the node 994 * while applying the function f to it, so we need to calculate 995 * the successor node in advance. 996 */ 997 /* First time through node, go left */ 998 while (rn->rn_bit >= 0) 999 rn = rn->rn_left; 1000 for (;;) { 1001 base = rn; 1002 /* If at right child go back up, otherwise, go right */ 1003 while (rn->rn_parent->rn_right == rn && 1004 !(rn->rn_flags & RNF_ROOT)) 1005 rn = rn->rn_parent; 1006 /* Find the next *leaf* since next node might vanish, too */ 1007 for (rn = rn->rn_parent->rn_right; rn->rn_bit >= 0;) 1008 rn = rn->rn_left; 1009 next = rn; 1010 /* Process leaves */ 1011 while ((rn = base)) { 1012 base = rn->rn_dupedkey; 1013 if (!(rn->rn_flags & RNF_ROOT) && (error = (*f)(rn, w))) 1014 return (error); 1015 } 1016 rn = next; 1017 if (rn->rn_flags & RNF_ROOT) 1018 return (0); 1019 } 1020 /* NOTREACHED */ 1021 } 1022 1023 int 1024 rn_inithead(void **head, struct radix_node_head *maskhead, int off) 1025 { 1026 struct radix_node_head *rnh; 1027 struct radix_node *root, *left, *right; 1028 1029 if (*head != NULL) /* already initialized */ 1030 return (1); 1031 1032 R_Malloc(rnh, struct radix_node_head *, sizeof *rnh); 1033 if (rnh == NULL) 1034 return (0); 1035 bzero(rnh, sizeof *rnh); 1036 *head = rnh; 1037 1038 root = rn_newpair(rn_zeros, off, rnh->rnh_nodes); 1039 right = &rnh->rnh_nodes[2]; 1040 root->rn_parent = root; 1041 root->rn_flags = RNF_ROOT | RNF_ACTIVE; 1042 root->rn_right = right; 1043 1044 left = root->rn_left; 1045 left->rn_bit = -1 - off; 1046 left->rn_flags = RNF_ROOT | RNF_ACTIVE; 1047 1048 *right = *left; 1049 right->rn_key = rn_ones; 1050 1051 rnh->rnh_treetop = root; 1052 rnh->rnh_maskhead = maskhead; 1053 1054 rnh->rnh_addaddr = rn_addroute; 1055 rnh->rnh_deladdr = rn_delete; 1056 rnh->rnh_matchaddr = rn_match; 1057 rnh->rnh_lookup = rn_lookup; 1058 rnh->rnh_walktree = rn_walktree; 1059 rnh->rnh_walktree_from = rn_walktree_from; 1060 1061 return (1); 1062 } 1063 1064 void 1065 rn_init(void) 1066 { 1067 int cpu; 1068 #ifdef _KERNEL 1069 struct domain *dom; 1070 1071 SLIST_FOREACH(dom, &domains, dom_next) { 1072 if (dom->dom_maxrtkey > RN_MAXKEYLEN) { 1073 panic("domain %s maxkey too big %d/%d", 1074 dom->dom_name, dom->dom_maxrtkey, RN_MAXKEYLEN); 1075 } 1076 } 1077 #endif 1078 for (cpu = 0; cpu < ncpus; ++cpu) { 1079 if (rn_inithead((void **)&mask_rnheads[cpu], NULL, 0) == 0) 1080 panic("rn_init 2"); 1081 } 1082 } 1083 1084 struct radix_node_head * 1085 rn_cpumaskhead(int cpu) 1086 { 1087 KKASSERT(mask_rnheads[cpu] != NULL); 1088 return mask_rnheads[cpu]; 1089 } 1090