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 48 #include <net/radix.h> 49 #include <net/netmsg2.h> 50 #include <net/netisr2.h> 51 52 /* 53 * The arguments to the radix functions are really counted byte arrays with 54 * the length in the first byte. struct sockaddr's fit this type structurally. 55 */ 56 #define clen(c) (*(u_char *)(c)) 57 58 static int rn_walktree_from(struct radix_node_head *h, char *a, char *m, 59 walktree_f_t *f, void *w); 60 static int rn_walktree(struct radix_node_head *, walktree_f_t *, void *); 61 static int rn_walktree_at(struct radix_node_head *h, const char *a, 62 const char *m, walktree_f_t *f, void *w); 63 64 static struct radix_node 65 *rn_insert(char *, struct radix_node_head *, boolean_t *, 66 struct radix_node [2]), 67 *rn_newpair(char *, int, struct radix_node[2]), 68 *rn_search(const char *, struct radix_node *), 69 *rn_search_m(const char *, struct radix_node *, const char *); 70 71 static struct radix_mask *rn_mkfreelist[MAXCPU]; 72 static struct radix_node_head *mask_rnheads[MAXCPU]; 73 74 static char rn_zeros[RN_MAXKEYLEN]; 75 static char rn_ones[RN_MAXKEYLEN] = RN_MAXKEYONES; 76 77 static boolean_t rn_lexobetter(char *m, char *n); 78 static struct radix_mask * 79 rn_new_radix_mask(struct radix_node *tt, struct radix_mask *nextmask); 80 static boolean_t 81 rn_satisfies_leaf(char *trial, struct radix_node *leaf, int skip); 82 83 static __inline struct radix_mask * 84 MKGet(struct radix_mask **l) 85 { 86 struct radix_mask *m; 87 88 if (*l != NULL) { 89 m = *l; 90 *l = m->rm_next; 91 } else { 92 R_Malloc(m, struct radix_mask *, sizeof *m); 93 } 94 return m; 95 } 96 97 static __inline void 98 MKFree(struct radix_mask **l, struct radix_mask *m) 99 { 100 m->rm_next = *l; 101 *l = m; 102 } 103 104 /* 105 * The data structure for the keys is a radix tree with one way 106 * branching removed. The index rn_bit at an internal node n represents a bit 107 * position to be tested. The tree is arranged so that all descendants 108 * of a node n have keys whose bits all agree up to position rn_bit - 1. 109 * (We say the index of n is rn_bit.) 110 * 111 * There is at least one descendant which has a one bit at position rn_bit, 112 * and at least one with a zero there. 113 * 114 * A route is determined by a pair of key and mask. We require that the 115 * bit-wise logical and of the key and mask to be the key. 116 * We define the index of a route to associated with the mask to be 117 * the first bit number in the mask where 0 occurs (with bit number 0 118 * representing the highest order bit). 119 * 120 * We say a mask is normal if every bit is 0, past the index of the mask. 121 * If a node n has a descendant (k, m) with index(m) == index(n) == rn_bit, 122 * and m is a normal mask, then the route applies to every descendant of n. 123 * If the index(m) < rn_bit, this implies the trailing last few bits of k 124 * before bit b are all 0, (and hence consequently true of every descendant 125 * of n), so the route applies to all descendants of the node as well. 126 * 127 * Similar logic shows that a non-normal mask m such that 128 * index(m) <= index(n) could potentially apply to many children of n. 129 * Thus, for each non-host route, we attach its mask to a list at an internal 130 * node as high in the tree as we can go. 131 * 132 * The present version of the code makes use of normal routes in short- 133 * circuiting an explict mask and compare operation when testing whether 134 * a key satisfies a normal route, and also in remembering the unique leaf 135 * that governs a subtree. 136 */ 137 138 static struct radix_node * 139 rn_search(const char *v, struct radix_node *head) 140 { 141 struct radix_node *x; 142 143 x = head; 144 while (x->rn_bit >= 0) { 145 if (x->rn_bmask & v[x->rn_offset]) 146 x = x->rn_right; 147 else 148 x = x->rn_left; 149 } 150 return (x); 151 } 152 153 static struct radix_node * 154 rn_search_m(const char *v, struct radix_node *head, const char *m) 155 { 156 struct radix_node *x; 157 158 for (x = head; x->rn_bit >= 0;) { 159 if ((x->rn_bmask & m[x->rn_offset]) && 160 (x->rn_bmask & v[x->rn_offset])) 161 x = x->rn_right; 162 else 163 x = x->rn_left; 164 } 165 return x; 166 } 167 168 /* 169 * Compare the two netmasks and return true if netmask <m> is strictly more 170 * specific than netmask <n>. 171 * 172 * NOTE: Non-contiguous netmask is supported. 173 */ 174 boolean_t 175 rn_refines(char *m, char *n) 176 { 177 char *lim, *lim2; 178 int longer; 179 boolean_t masks_are_equal = TRUE; 180 181 lim2 = lim = n + clen(n); 182 longer = clen(n++) - clen(m++); 183 if (longer > 0) 184 lim -= longer; 185 while (n < lim) { 186 if (*n & ~(*m)) 187 return FALSE; 188 if (*n++ != *m++) 189 masks_are_equal = FALSE; 190 } 191 while (n < lim2) 192 if (*n++) 193 return FALSE; 194 if (masks_are_equal && (longer < 0)) 195 for (lim2 = m - longer; m < lim2; ) 196 if (*m++) 197 return TRUE; 198 return (!masks_are_equal); 199 } 200 201 struct radix_node * 202 rn_lookup(char *key, char *mask, struct radix_node_head *head) 203 { 204 struct radix_node *x; 205 char *netmask = NULL; 206 207 if (mask != NULL) { 208 x = rn_addmask(mask, TRUE, head->rnh_treetop->rn_offset, 209 head->rnh_maskhead); 210 if (x == NULL) 211 return (NULL); 212 netmask = x->rn_key; 213 } 214 x = rn_match(key, head); 215 if (x != NULL && netmask != NULL) { 216 while (x != NULL && x->rn_mask != netmask) 217 x = x->rn_dupedkey; 218 } 219 return x; 220 } 221 222 static boolean_t 223 rn_satisfies_leaf(char *trial, struct radix_node *leaf, int skip) 224 { 225 char *cp = trial, *cp2 = leaf->rn_key, *cp3 = leaf->rn_mask; 226 char *cplim; 227 int length = min(clen(cp), clen(cp2)); 228 229 if (cp3 == NULL) 230 cp3 = rn_ones; 231 else 232 length = min(length, clen(cp3)); 233 cplim = cp + length; 234 cp3 += skip; 235 cp2 += skip; 236 for (cp += skip; cp < cplim; cp++, cp2++, cp3++) 237 if ((*cp ^ *cp2) & *cp3) 238 return FALSE; 239 return TRUE; 240 } 241 242 struct radix_node * 243 rn_match(char *key, struct radix_node_head *head) 244 { 245 struct radix_node *t, *x; 246 char *cp = key, *cp2; 247 char *cplim; 248 struct radix_node *saved_t, *top = head->rnh_treetop; 249 int off = top->rn_offset, klen, matched_off; 250 int test, b, rn_bit; 251 252 t = rn_search(key, top); 253 /* 254 * See if we match exactly as a host destination 255 * or at least learn how many bits match, for normal mask finesse. 256 * 257 * It doesn't hurt us to limit how many bytes to check 258 * to the length of the mask, since if it matches we had a genuine 259 * match and the leaf we have is the most specific one anyway; 260 * if it didn't match with a shorter length it would fail 261 * with a long one. This wins big for class B&C netmasks which 262 * are probably the most common case... 263 */ 264 if (t->rn_mask != NULL) 265 klen = clen(t->rn_mask); 266 else 267 klen = clen(key); 268 cp += off; cp2 = t->rn_key + off; cplim = key + klen; 269 for (; cp < cplim; cp++, cp2++) 270 if (*cp != *cp2) 271 goto on1; 272 /* 273 * This extra grot is in case we are explicitly asked 274 * to look up the default. Ugh! 275 * 276 * Never return the root node itself, it seems to cause a 277 * lot of confusion. 278 */ 279 if (t->rn_flags & RNF_ROOT) 280 t = t->rn_dupedkey; 281 return t; 282 on1: 283 test = (*cp ^ *cp2) & 0xff; /* find first bit that differs */ 284 for (b = 7; (test >>= 1) > 0;) 285 b--; 286 matched_off = cp - key; 287 b += matched_off << 3; 288 rn_bit = -1 - b; 289 /* 290 * If there is a host route in a duped-key chain, it will be first. 291 */ 292 if ((saved_t = t)->rn_mask == NULL) 293 t = t->rn_dupedkey; 294 for (; t; t = t->rn_dupedkey) { 295 /* 296 * Even if we don't match exactly as a host, 297 * we may match if the leaf we wound up at is 298 * a route to a net. 299 */ 300 if (t->rn_flags & RNF_NORMAL) { 301 if (rn_bit <= t->rn_bit) 302 return t; 303 } else if (rn_satisfies_leaf(key, t, matched_off)) 304 return t; 305 } 306 t = saved_t; 307 /* start searching up the tree */ 308 do { 309 struct radix_mask *m; 310 311 t = t->rn_parent; 312 /* 313 * If non-contiguous masks ever become important 314 * we can restore the masking and open coding of 315 * the search and satisfaction test and put the 316 * calculation of "off" back before the "do". 317 */ 318 m = t->rn_mklist; 319 while (m != NULL) { 320 if (m->rm_flags & RNF_NORMAL) { 321 if (rn_bit <= m->rm_bit) 322 return (m->rm_leaf); 323 } else { 324 off = min(t->rn_offset, matched_off); 325 x = rn_search_m(key, t, m->rm_mask); 326 while (x != NULL && x->rn_mask != m->rm_mask) 327 x = x->rn_dupedkey; 328 if (x && rn_satisfies_leaf(key, x, off)) 329 return x; 330 } 331 m = m->rm_next; 332 } 333 } while (t != top); 334 return NULL; 335 } 336 337 #ifdef RN_DEBUG 338 int rn_nodenum; 339 struct radix_node *rn_clist; 340 int rn_saveinfo; 341 boolean_t rn_debug = TRUE; 342 #endif 343 344 static struct radix_node * 345 rn_newpair(char *key, int indexbit, struct radix_node nodes[2]) 346 { 347 struct radix_node *leaf = &nodes[0], *interior = &nodes[1]; 348 349 interior->rn_bit = indexbit; 350 interior->rn_bmask = 0x80 >> (indexbit & 0x7); 351 interior->rn_offset = indexbit >> 3; 352 interior->rn_left = leaf; 353 interior->rn_mklist = NULL; 354 355 leaf->rn_bit = -1; 356 leaf->rn_key = key; 357 leaf->rn_parent = interior; 358 leaf->rn_flags = interior->rn_flags = RNF_ACTIVE; 359 leaf->rn_mklist = NULL; 360 361 #ifdef RN_DEBUG 362 leaf->rn_info = rn_nodenum++; 363 interior->rn_info = rn_nodenum++; 364 leaf->rn_twin = interior; 365 leaf->rn_ybro = rn_clist; 366 rn_clist = leaf; 367 #endif 368 return interior; 369 } 370 371 static struct radix_node * 372 rn_insert(char *key, struct radix_node_head *head, boolean_t *dupentry, 373 struct radix_node nodes[2]) 374 { 375 struct radix_node *top = head->rnh_treetop; 376 int head_off = top->rn_offset, klen = clen(key); 377 struct radix_node *t = rn_search(key, top); 378 char *cp = key + head_off; 379 int b; 380 struct radix_node *tt; 381 382 /* 383 * Find first bit at which the key and t->rn_key differ 384 */ 385 { 386 char *cp2 = t->rn_key + head_off; 387 int cmp_res; 388 char *cplim = key + klen; 389 390 while (cp < cplim) 391 if (*cp2++ != *cp++) 392 goto on1; 393 *dupentry = TRUE; 394 return t; 395 on1: 396 *dupentry = FALSE; 397 cmp_res = (cp[-1] ^ cp2[-1]) & 0xff; 398 for (b = (cp - key) << 3; cmp_res; b--) 399 cmp_res >>= 1; 400 } 401 { 402 struct radix_node *p, *x = top; 403 404 cp = key; 405 do { 406 p = x; 407 if (cp[x->rn_offset] & x->rn_bmask) 408 x = x->rn_right; 409 else 410 x = x->rn_left; 411 } while (b > (unsigned) x->rn_bit); 412 /* x->rn_bit < b && x->rn_bit >= 0 */ 413 #ifdef RN_DEBUG 414 if (rn_debug) 415 log(LOG_DEBUG, "rn_insert: Going In:\n"), traverse(p); 416 #endif 417 t = rn_newpair(key, b, nodes); 418 tt = t->rn_left; 419 if ((cp[p->rn_offset] & p->rn_bmask) == 0) 420 p->rn_left = t; 421 else 422 p->rn_right = t; 423 x->rn_parent = t; 424 t->rn_parent = p; /* frees x, p as temp vars below */ 425 if ((cp[t->rn_offset] & t->rn_bmask) == 0) { 426 t->rn_right = x; 427 } else { 428 t->rn_right = tt; 429 t->rn_left = x; 430 } 431 #ifdef RN_DEBUG 432 if (rn_debug) 433 log(LOG_DEBUG, "rn_insert: Coming Out:\n"), traverse(p); 434 #endif 435 } 436 return (tt); 437 } 438 439 struct radix_node * 440 rn_addmask(char *netmask, boolean_t search, int skip, 441 struct radix_node_head *mask_rnh) 442 { 443 struct radix_node *x, *saved_x; 444 char *cp, *cplim; 445 int b = 0, mlen, m0, j; 446 boolean_t maskduplicated, isnormal; 447 char *addmask_key; 448 449 if ((mlen = clen(netmask)) > RN_MAXKEYLEN) 450 mlen = RN_MAXKEYLEN; 451 if (skip == 0) 452 skip = 1; 453 if (mlen <= skip) 454 return (mask_rnh->rnh_nodes); 455 R_Malloc(addmask_key, char *, RN_MAXKEYLEN); 456 if (addmask_key == NULL) 457 return NULL; 458 if (skip > 1) 459 bcopy(rn_ones + 1, addmask_key + 1, skip - 1); 460 if ((m0 = mlen) > skip) 461 bcopy(netmask + skip, addmask_key + skip, mlen - skip); 462 /* 463 * Trim trailing zeroes. 464 */ 465 for (cp = addmask_key + mlen; (cp > addmask_key) && cp[-1] == 0;) 466 cp--; 467 mlen = cp - addmask_key; 468 if (mlen <= skip) { 469 if (m0 >= mask_rnh->rnh_last_zeroed) 470 mask_rnh->rnh_last_zeroed = mlen; 471 Free(addmask_key); 472 return (mask_rnh->rnh_nodes); 473 } 474 if (m0 < mask_rnh->rnh_last_zeroed) 475 bzero(addmask_key + m0, mask_rnh->rnh_last_zeroed - m0); 476 *addmask_key = mask_rnh->rnh_last_zeroed = mlen; 477 x = rn_search(addmask_key, mask_rnh->rnh_treetop); 478 if (x->rn_key == NULL) { 479 kprintf("WARNING: radix_node->rn_key is NULL rn=%p\n", x); 480 print_backtrace(-1); 481 x = NULL; 482 } else if (bcmp(addmask_key, x->rn_key, mlen) != 0) { 483 x = NULL; 484 } 485 if (x != NULL || search) 486 goto out; 487 R_Malloc(x, struct radix_node *, RN_MAXKEYLEN + 2 * (sizeof *x)); 488 if ((saved_x = x) == NULL) 489 goto out; 490 bzero(x, RN_MAXKEYLEN + 2 * (sizeof *x)); 491 netmask = cp = (char *)(x + 2); 492 bcopy(addmask_key, cp, mlen); 493 x = rn_insert(cp, mask_rnh, &maskduplicated, x); 494 if (maskduplicated) { 495 log(LOG_ERR, "rn_addmask: mask impossibly already in tree"); 496 Free(saved_x); 497 goto out; 498 } 499 /* 500 * Calculate index of mask, and check for normalcy. 501 */ 502 isnormal = TRUE; 503 cplim = netmask + mlen; 504 for (cp = netmask + skip; cp < cplim && clen(cp) == 0xff;) 505 cp++; 506 if (cp != cplim) { 507 static const char normal_chars[] = { 508 0, 0x80, 0xc0, 0xe0, 0xf0, 0xf8, 0xfc, 0xfe, -1 509 }; 510 511 for (j = 0x80; (j & *cp) != 0; j >>= 1) 512 b++; 513 if (*cp != normal_chars[b] || cp != (cplim - 1)) 514 isnormal = FALSE; 515 } 516 b += (cp - netmask) << 3; 517 x->rn_bit = -1 - b; 518 if (isnormal) 519 x->rn_flags |= RNF_NORMAL; 520 out: 521 Free(addmask_key); 522 return (x); 523 } 524 525 /* XXX: arbitrary ordering for non-contiguous masks */ 526 static boolean_t 527 rn_lexobetter(char *mp, char *np) 528 { 529 char *lim; 530 531 if ((unsigned) *mp > (unsigned) *np) 532 return TRUE;/* not really, but need to check longer one first */ 533 if (*mp == *np) 534 for (lim = mp + clen(mp); mp < lim;) 535 if (*mp++ > *np++) 536 return TRUE; 537 return FALSE; 538 } 539 540 static struct radix_mask * 541 rn_new_radix_mask(struct radix_node *tt, struct radix_mask *nextmask) 542 { 543 struct radix_mask *m; 544 545 m = MKGet(&rn_mkfreelist[mycpuid]); 546 if (m == NULL) { 547 log(LOG_ERR, "Mask for route not entered\n"); 548 return (NULL); 549 } 550 bzero(m, sizeof *m); 551 m->rm_bit = tt->rn_bit; 552 m->rm_flags = tt->rn_flags; 553 if (tt->rn_flags & RNF_NORMAL) 554 m->rm_leaf = tt; 555 else 556 m->rm_mask = tt->rn_mask; 557 m->rm_next = nextmask; 558 tt->rn_mklist = m; 559 return m; 560 } 561 562 struct radix_node * 563 rn_addroute(char *key, char *netmask, struct radix_node_head *head, 564 struct radix_node treenodes[2]) 565 { 566 struct radix_node *t, *x = NULL, *tt; 567 struct radix_node *saved_tt, *top = head->rnh_treetop; 568 short b = 0, b_leaf = 0; 569 boolean_t keyduplicated; 570 char *mmask; 571 struct radix_mask *m, **mp; 572 573 /* 574 * In dealing with non-contiguous masks, there may be 575 * many different routes which have the same mask. 576 * We will find it useful to have a unique pointer to 577 * the mask to speed avoiding duplicate references at 578 * nodes and possibly save time in calculating indices. 579 */ 580 if (netmask != NULL) { 581 if ((x = rn_addmask(netmask, FALSE, top->rn_offset, 582 head->rnh_maskhead)) == NULL) 583 return (NULL); 584 b_leaf = x->rn_bit; 585 b = -1 - x->rn_bit; 586 netmask = x->rn_key; 587 } 588 /* 589 * Deal with duplicated keys: attach node to previous instance 590 */ 591 saved_tt = tt = rn_insert(key, head, &keyduplicated, treenodes); 592 if (keyduplicated) { 593 for (t = tt; tt; t = tt, tt = tt->rn_dupedkey) { 594 if (tt->rn_mask == netmask) 595 return (NULL); 596 if (netmask == NULL || 597 (tt->rn_mask && 598 ((b_leaf < tt->rn_bit) /* index(netmask) > node */ 599 || rn_refines(netmask, tt->rn_mask) 600 || rn_lexobetter(netmask, tt->rn_mask)))) 601 break; 602 } 603 /* 604 * If the mask is not duplicated, we wouldn't 605 * find it among possible duplicate key entries 606 * anyway, so the above test doesn't hurt. 607 * 608 * We sort the masks for a duplicated key the same way as 609 * in a masklist -- most specific to least specific. 610 * This may require the unfortunate nuisance of relocating 611 * the head of the list. 612 */ 613 if (tt == saved_tt) { 614 struct radix_node *xx = x; 615 /* link in at head of list */ 616 (tt = treenodes)->rn_dupedkey = t; 617 tt->rn_flags = t->rn_flags; 618 tt->rn_parent = x = t->rn_parent; 619 t->rn_parent = tt; /* parent */ 620 if (x->rn_left == t) 621 x->rn_left = tt; 622 else 623 x->rn_right = tt; 624 saved_tt = tt; x = xx; 625 } else { 626 (tt = treenodes)->rn_dupedkey = t->rn_dupedkey; 627 t->rn_dupedkey = tt; 628 tt->rn_parent = t; /* parent */ 629 if (tt->rn_dupedkey != NULL) /* parent */ 630 tt->rn_dupedkey->rn_parent = tt; /* parent */ 631 } 632 #ifdef RN_DEBUG 633 t=tt+1; tt->rn_info = rn_nodenum++; t->rn_info = rn_nodenum++; 634 tt->rn_twin = t; tt->rn_ybro = rn_clist; rn_clist = tt; 635 #endif 636 tt->rn_key = key; 637 tt->rn_bit = -1; 638 tt->rn_flags = RNF_ACTIVE; 639 } 640 /* 641 * Put mask in tree. 642 */ 643 if (netmask != NULL) { 644 tt->rn_mask = netmask; 645 tt->rn_bit = x->rn_bit; 646 tt->rn_flags |= x->rn_flags & RNF_NORMAL; 647 } 648 t = saved_tt->rn_parent; 649 if (keyduplicated) 650 goto on2; 651 b_leaf = -1 - t->rn_bit; 652 if (t->rn_right == saved_tt) 653 x = t->rn_left; 654 else 655 x = t->rn_right; 656 /* Promote general routes from below */ 657 if (x->rn_bit < 0) { 658 mp = &t->rn_mklist; 659 while (x != NULL) { 660 if (x->rn_mask != NULL && 661 x->rn_bit >= b_leaf && 662 x->rn_mklist == NULL) { 663 *mp = m = rn_new_radix_mask(x, NULL); 664 if (m != NULL) 665 mp = &m->rm_next; 666 } 667 x = x->rn_dupedkey; 668 } 669 } else if (x->rn_mklist != NULL) { 670 /* 671 * Skip over masks whose index is > that of new node 672 */ 673 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_next) 674 if (m->rm_bit >= b_leaf) 675 break; 676 t->rn_mklist = m; 677 *mp = NULL; 678 } 679 on2: 680 /* Add new route to highest possible ancestor's list */ 681 if ((netmask == NULL) || (b > t->rn_bit )) 682 return tt; /* can't lift at all */ 683 b_leaf = tt->rn_bit; 684 do { 685 x = t; 686 t = t->rn_parent; 687 } while (b <= t->rn_bit && x != top); 688 /* 689 * Search through routes associated with node to 690 * insert new route according to index. 691 * Need same criteria as when sorting dupedkeys to avoid 692 * double loop on deletion. 693 */ 694 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_next) { 695 if (m->rm_bit < b_leaf) 696 continue; 697 if (m->rm_bit > b_leaf) 698 break; 699 if (m->rm_flags & RNF_NORMAL) { 700 mmask = m->rm_leaf->rn_mask; 701 if (tt->rn_flags & RNF_NORMAL) { 702 log(LOG_ERR, 703 "Non-unique normal route, mask not entered\n"); 704 return tt; 705 } 706 } else 707 mmask = m->rm_mask; 708 if (mmask == netmask) { 709 m->rm_refs++; 710 tt->rn_mklist = m; 711 return tt; 712 } 713 if (rn_refines(netmask, mmask) || rn_lexobetter(netmask, mmask)) 714 break; 715 } 716 *mp = rn_new_radix_mask(tt, *mp); 717 return tt; 718 } 719 720 struct radix_node * 721 rn_delete(char *key, char *netmask, struct radix_node_head *head) 722 { 723 struct radix_node *t, *p, *x, *tt; 724 struct radix_mask *m, *saved_m, **mp; 725 struct radix_node *dupedkey, *saved_tt, *top; 726 int b, head_off, klen; 727 int cpu = mycpuid; 728 729 x = head->rnh_treetop; 730 tt = rn_search(key, x); 731 head_off = x->rn_offset; 732 klen = clen(key); 733 saved_tt = tt; 734 top = x; 735 if (tt == NULL || 736 bcmp(key + head_off, tt->rn_key + head_off, klen - head_off)) 737 return (NULL); 738 /* 739 * Delete our route from mask lists. 740 */ 741 if (netmask != NULL) { 742 if ((x = rn_addmask(netmask, TRUE, head_off, 743 head->rnh_maskhead)) == NULL) 744 return (NULL); 745 netmask = x->rn_key; 746 while (tt->rn_mask != netmask) 747 if ((tt = tt->rn_dupedkey) == NULL) 748 return (NULL); 749 } 750 if (tt->rn_mask == NULL || (saved_m = m = tt->rn_mklist) == NULL) 751 goto on1; 752 if (tt->rn_flags & RNF_NORMAL) { 753 if (m->rm_leaf != tt || m->rm_refs > 0) { 754 log(LOG_ERR, "rn_delete: inconsistent annotation\n"); 755 return (NULL); /* dangling ref could cause disaster */ 756 } 757 } else { 758 if (m->rm_mask != tt->rn_mask) { 759 log(LOG_ERR, "rn_delete: inconsistent annotation\n"); 760 goto on1; 761 } 762 if (--m->rm_refs >= 0) 763 goto on1; 764 } 765 b = -1 - tt->rn_bit; 766 t = saved_tt->rn_parent; 767 if (b > t->rn_bit) 768 goto on1; /* Wasn't lifted at all */ 769 do { 770 x = t; 771 t = t->rn_parent; 772 } while (b <= t->rn_bit && x != top); 773 for (mp = &x->rn_mklist; (m = *mp); mp = &m->rm_next) 774 if (m == saved_m) { 775 *mp = m->rm_next; 776 MKFree(&rn_mkfreelist[cpu], m); 777 break; 778 } 779 if (m == NULL) { 780 log(LOG_ERR, "rn_delete: couldn't find our annotation\n"); 781 if (tt->rn_flags & RNF_NORMAL) 782 return (NULL); /* Dangling ref to us */ 783 } 784 on1: 785 /* 786 * Eliminate us from tree 787 */ 788 if (tt->rn_flags & RNF_ROOT) 789 return (NULL); 790 #ifdef RN_DEBUG 791 /* Get us out of the creation list */ 792 for (t = rn_clist; t && t->rn_ybro != tt; t = t->rn_ybro) {} 793 if (t) t->rn_ybro = tt->rn_ybro; 794 #endif 795 t = tt->rn_parent; 796 dupedkey = saved_tt->rn_dupedkey; 797 if (dupedkey != NULL) { 798 /* 799 * at this point, tt is the deletion target and saved_tt 800 * is the head of the dupekey chain 801 */ 802 if (tt == saved_tt) { 803 /* remove from head of chain */ 804 x = dupedkey; x->rn_parent = t; 805 if (t->rn_left == tt) 806 t->rn_left = x; 807 else 808 t->rn_right = x; 809 } else { 810 /* find node in front of tt on the chain */ 811 for (x = p = saved_tt; p && p->rn_dupedkey != tt;) 812 p = p->rn_dupedkey; 813 if (p) { 814 p->rn_dupedkey = tt->rn_dupedkey; 815 if (tt->rn_dupedkey) /* parent */ 816 tt->rn_dupedkey->rn_parent = p; 817 /* parent */ 818 } else log(LOG_ERR, "rn_delete: couldn't find us\n"); 819 } 820 t = tt + 1; 821 if (t->rn_flags & RNF_ACTIVE) { 822 #ifndef RN_DEBUG 823 *++x = *t; 824 p = t->rn_parent; 825 #else 826 b = t->rn_info; 827 *++x = *t; 828 t->rn_info = b; 829 p = t->rn_parent; 830 #endif 831 if (p->rn_left == t) 832 p->rn_left = x; 833 else 834 p->rn_right = x; 835 x->rn_left->rn_parent = x; 836 x->rn_right->rn_parent = x; 837 } 838 goto out; 839 } 840 if (t->rn_left == tt) 841 x = t->rn_right; 842 else 843 x = t->rn_left; 844 p = t->rn_parent; 845 if (p->rn_right == t) 846 p->rn_right = x; 847 else 848 p->rn_left = x; 849 x->rn_parent = p; 850 /* 851 * Demote routes attached to us. 852 */ 853 if (t->rn_mklist != NULL) { 854 if (x->rn_bit >= 0) { 855 for (mp = &x->rn_mklist; (m = *mp);) 856 mp = &m->rm_next; 857 *mp = t->rn_mklist; 858 } else { 859 /* 860 * If there are any (key, mask) pairs in a sibling 861 * duped-key chain, some subset will appear sorted 862 * in the same order attached to our mklist. 863 */ 864 for (m = t->rn_mklist; m && x; x = x->rn_dupedkey) 865 if (m == x->rn_mklist) { 866 struct radix_mask *mm = m->rm_next; 867 868 x->rn_mklist = NULL; 869 if (--(m->rm_refs) < 0) 870 MKFree(&rn_mkfreelist[cpu], m); 871 m = mm; 872 } 873 if (m) 874 log(LOG_ERR, 875 "rn_delete: Orphaned Mask %p at %p\n", 876 (void *)m, (void *)x); 877 } 878 } 879 /* 880 * We may be holding an active internal node in the tree. 881 */ 882 x = tt + 1; 883 if (t != x) { 884 #ifndef RN_DEBUG 885 *t = *x; 886 #else 887 b = t->rn_info; 888 *t = *x; 889 t->rn_info = b; 890 #endif 891 t->rn_left->rn_parent = t; 892 t->rn_right->rn_parent = t; 893 p = x->rn_parent; 894 if (p->rn_left == x) 895 p->rn_left = t; 896 else 897 p->rn_right = t; 898 } 899 out: 900 tt->rn_flags &= ~RNF_ACTIVE; 901 tt[1].rn_flags &= ~RNF_ACTIVE; 902 return (tt); 903 } 904 905 /* 906 * This is the same as rn_walktree() except for the parameters and the 907 * exit. 908 */ 909 static int 910 rn_walktree_from(struct radix_node_head *h, char *xa, char *xm, 911 walktree_f_t *f, void *w) 912 { 913 struct radix_node *base, *next; 914 struct radix_node *rn, *last = NULL /* shut up gcc */; 915 boolean_t stopping = FALSE; 916 int lastb, error; 917 918 /* 919 * rn_search_m is sort-of-open-coded here. 920 */ 921 /* kprintf("about to search\n"); */ 922 for (rn = h->rnh_treetop; rn->rn_bit >= 0; ) { 923 last = rn; 924 /* kprintf("rn_bit %d, rn_bmask %x, xm[rn_offset] %x\n", 925 rn->rn_bit, rn->rn_bmask, xm[rn->rn_offset]); */ 926 if (!(rn->rn_bmask & xm[rn->rn_offset])) { 927 break; 928 } 929 if (rn->rn_bmask & xa[rn->rn_offset]) { 930 rn = rn->rn_right; 931 } else { 932 rn = rn->rn_left; 933 } 934 } 935 /* kprintf("done searching\n"); */ 936 937 /* 938 * Two cases: either we stepped off the end of our mask, 939 * in which case last == rn, or we reached a leaf, in which 940 * case we want to start from the last node we looked at. 941 * Either way, last is the node we want to start from. 942 */ 943 rn = last; 944 lastb = rn->rn_bit; 945 946 /* kprintf("rn %p, lastb %d\n", rn, lastb);*/ 947 948 /* 949 * This gets complicated because we may delete the node 950 * while applying the function f to it, so we need to calculate 951 * the successor node in advance. 952 */ 953 while (rn->rn_bit >= 0) 954 rn = rn->rn_left; 955 956 while (!stopping) { 957 /* kprintf("node %p (%d)\n", rn, rn->rn_bit); */ 958 base = rn; 959 /* If at right child go back up, otherwise, go right */ 960 while (rn->rn_parent->rn_right == rn && 961 !(rn->rn_flags & RNF_ROOT)) { 962 rn = rn->rn_parent; 963 964 /* if went up beyond last, stop */ 965 if (rn->rn_bit < lastb) { 966 stopping = TRUE; 967 /* kprintf("up too far\n"); */ 968 } 969 } 970 971 /* Find the next *leaf* since next node might vanish, too */ 972 for (rn = rn->rn_parent->rn_right; rn->rn_bit >= 0;) 973 rn = rn->rn_left; 974 next = rn; 975 /* Process leaves */ 976 while ((rn = base) != NULL) { 977 base = rn->rn_dupedkey; 978 /* kprintf("leaf %p\n", rn); */ 979 if (!(rn->rn_flags & RNF_ROOT) && (error = (*f)(rn, w))) 980 return (error); 981 } 982 rn = next; 983 984 if (rn->rn_flags & RNF_ROOT) { 985 /* kprintf("root, stopping"); */ 986 stopping = TRUE; 987 } 988 989 } 990 return 0; 991 } 992 993 static int 994 rn_walktree_at(struct radix_node_head *h, const char *a, const char *m, 995 walktree_f_t *f, void *w) 996 { 997 struct radix_node *base, *next; 998 struct radix_node *rn = h->rnh_treetop; 999 int error; 1000 1001 /* 1002 * This gets complicated because we may delete the node 1003 * while applying the function f to it, so we need to calculate 1004 * the successor node in advance. 1005 */ 1006 if (a == NULL) { 1007 /* First time through node, go left */ 1008 while (rn->rn_bit >= 0) 1009 rn = rn->rn_left; 1010 } else { 1011 if (m != NULL) 1012 rn = rn_search_m(a, rn, m); 1013 else 1014 rn = rn_search(a, rn); 1015 } 1016 for (;;) { 1017 base = rn; 1018 /* If at right child go back up, otherwise, go right */ 1019 while (rn->rn_parent->rn_right == rn && 1020 !(rn->rn_flags & RNF_ROOT)) 1021 rn = rn->rn_parent; 1022 /* Find the next *leaf* since next node might vanish, too */ 1023 for (rn = rn->rn_parent->rn_right; rn->rn_bit >= 0;) 1024 rn = rn->rn_left; 1025 next = rn; 1026 /* Process leaves */ 1027 while ((rn = base)) { 1028 base = rn->rn_dupedkey; 1029 if (!(rn->rn_flags & RNF_ROOT) && (error = (*f)(rn, w))) 1030 return (error); 1031 } 1032 rn = next; 1033 if (rn->rn_flags & RNF_ROOT) 1034 return (0); 1035 } 1036 /* NOTREACHED */ 1037 } 1038 1039 static int 1040 rn_walktree(struct radix_node_head *h, walktree_f_t *f, void *w) 1041 { 1042 return rn_walktree_at(h, NULL, NULL, f, w); 1043 } 1044 1045 int 1046 rn_inithead(void **head, struct radix_node_head *maskhead, int off) 1047 { 1048 struct radix_node_head *rnh; 1049 struct radix_node *root, *left, *right; 1050 1051 if (*head != NULL) /* already initialized */ 1052 return (1); 1053 1054 R_Malloc(rnh, struct radix_node_head *, sizeof *rnh); 1055 if (rnh == NULL) 1056 return (0); 1057 bzero(rnh, sizeof *rnh); 1058 *head = rnh; 1059 1060 root = rn_newpair(rn_zeros, off, rnh->rnh_nodes); 1061 right = &rnh->rnh_nodes[2]; 1062 root->rn_parent = root; 1063 root->rn_flags = RNF_ROOT | RNF_ACTIVE; 1064 root->rn_right = right; 1065 1066 left = root->rn_left; 1067 left->rn_bit = -1 - off; 1068 left->rn_flags = RNF_ROOT | RNF_ACTIVE; 1069 1070 *right = *left; 1071 right->rn_key = rn_ones; 1072 1073 rnh->rnh_treetop = root; 1074 rnh->rnh_maskhead = maskhead; 1075 1076 rnh->rnh_addaddr = rn_addroute; 1077 rnh->rnh_deladdr = rn_delete; 1078 rnh->rnh_matchaddr = rn_match; 1079 rnh->rnh_lookup = rn_lookup; 1080 rnh->rnh_walktree = rn_walktree; 1081 rnh->rnh_walktree_from = rn_walktree_from; 1082 rnh->rnh_walktree_at = rn_walktree_at; 1083 1084 return (1); 1085 } 1086 1087 static void 1088 rn_init_handler(netmsg_t msg) 1089 { 1090 int cpu = mycpuid; 1091 1092 ASSERT_NETISR_NCPUS(cpu); 1093 if (rn_inithead((void **)&mask_rnheads[cpu], NULL, 0) == 0) 1094 panic("rn_init 2"); 1095 1096 netisr_forwardmsg(&msg->base, cpu + 1); 1097 } 1098 1099 void 1100 rn_init(void) 1101 { 1102 struct netmsg_base msg; 1103 #ifdef _KERNEL 1104 struct domain *dom; 1105 1106 SLIST_FOREACH(dom, &domains, dom_next) { 1107 if (dom->dom_maxrtkey > RN_MAXKEYLEN) { 1108 panic("domain %s maxkey too big %d/%d", 1109 dom->dom_name, dom->dom_maxrtkey, RN_MAXKEYLEN); 1110 } 1111 } 1112 #endif 1113 netmsg_init(&msg, NULL, &curthread->td_msgport, 0, rn_init_handler); 1114 netisr_domsg_global(&msg); 1115 } 1116 1117 struct radix_node_head * 1118 rn_cpumaskhead(int cpu) 1119 { 1120 1121 ASSERT_NETISR_NCPUS(cpu); 1122 KKASSERT(mask_rnheads[cpu] != NULL); 1123 return mask_rnheads[cpu]; 1124 } 1125