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