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 37 #include <sys/param.h> 38 #ifdef _KERNEL 39 #include <sys/systm.h> 40 #include <sys/malloc.h> 41 #include <sys/domain.h> 42 #include <sys/globaldata.h> 43 #include <sys/thread.h> 44 #else 45 #include <stdlib.h> 46 #endif 47 #include <sys/syslog.h> 48 49 #include <net/radix.h> 50 #include <net/netmsg2.h> 51 #include <net/netisr2.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 * Cast the result to int as this is the dominant usage. 57 */ 58 #define clen(c) (int)(*(const u_char *)(c)) 59 60 61 static struct radix_mask *rn_mkfreelist[MAXCPU]; 62 static struct radix_node_head *mask_rnheads[MAXCPU]; 63 64 static char rn_zeros[RN_MAXKEYLEN]; 65 static char rn_ones[RN_MAXKEYLEN] = RN_MAXKEYONES; 66 67 #ifdef RN_DEBUG 68 static int rn_nodenum; 69 static struct radix_node *rn_clist; 70 static int rn_saveinfo; 71 static bool rn_debug = true; 72 #endif 73 74 75 static __inline struct radix_mask * 76 MKGet(struct radix_mask **l) 77 { 78 struct radix_mask *m; 79 80 if (*l != NULL) { 81 m = *l; 82 *l = m->rm_next; 83 } else { 84 R_Malloc(m, struct radix_mask *, sizeof *m); 85 } 86 return m; 87 } 88 89 static __inline void 90 MKFree(struct radix_mask **l, struct radix_mask *m) 91 { 92 m->rm_next = *l; 93 *l = m; 94 } 95 96 /* 97 * The data structure for the keys is a radix tree with one way 98 * branching removed. The index rn_bit at an internal node n represents a bit 99 * position to be tested. The tree is arranged so that all descendants 100 * of a node n have keys whose bits all agree up to position rn_bit - 1. 101 * (We say the index of n is rn_bit.) 102 * 103 * There is at least one descendant which has a one bit at position rn_bit, 104 * and at least one with a zero there. 105 * 106 * A route is determined by a pair of key and mask. We require that the 107 * bit-wise logical and of the key and mask to be the key. 108 * We define the index of a route to associated with the mask to be 109 * the first bit number in the mask where 0 occurs (with bit number 0 110 * representing the highest order bit). 111 * 112 * We say a mask is normal if every bit is 0, past the index of the mask. 113 * If a node n has a descendant (k, m) with index(m) == index(n) == rn_bit, 114 * and m is a normal mask, then the route applies to every descendant of n. 115 * If the index(m) < rn_bit, this implies the trailing last few bits of k 116 * before bit b are all 0, (and hence consequently true of every descendant 117 * of n), so the route applies to all descendants of the node as well. 118 * 119 * Similar logic shows that a non-normal mask m such that 120 * index(m) <= index(n) could potentially apply to many children of n. 121 * Thus, for each non-host route, we attach its mask to a list at an internal 122 * node as high in the tree as we can go. 123 * 124 * The present version of the code makes use of normal routes in short- 125 * circuiting an explict mask and compare operation when testing whether 126 * a key satisfies a normal route, and also in remembering the unique leaf 127 * that governs a subtree. 128 */ 129 130 static struct radix_node * 131 rn_search(const char *v, struct radix_node *head) 132 { 133 struct radix_node *x; 134 135 x = head; 136 while (x->rn_bit >= 0) { 137 if (x->rn_bmask & v[x->rn_offset]) 138 x = x->rn_right; 139 else 140 x = x->rn_left; 141 } 142 return (x); 143 } 144 145 static struct radix_node * 146 rn_search_m(const char *v, struct radix_node *head, const char *m) 147 { 148 struct radix_node *x; 149 150 x = head; 151 while (x->rn_bit >= 0) { 152 if ((x->rn_bmask & m[x->rn_offset]) && 153 (x->rn_bmask & v[x->rn_offset])) 154 x = x->rn_right; 155 else 156 x = x->rn_left; 157 } 158 return x; 159 } 160 161 /* 162 * Compare the two netmasks and return true if netmask <m> is strictly more 163 * specific than netmask <n>. 164 * 165 * NOTE: Non-contiguous netmask is supported. 166 */ 167 bool 168 rn_refines(const char *m, const char *n) 169 { 170 const char *lim, *lim2; 171 int longer; 172 bool masks_are_equal = true; 173 174 lim2 = lim = n + clen(n); 175 longer = clen(n++) - clen(m++); 176 if (longer > 0) 177 lim -= longer; 178 while (n < lim) { 179 if (*n & ~(*m)) 180 return false; 181 if (*n++ != *m++) 182 masks_are_equal = false; 183 } 184 while (n < lim2) 185 if (*n++) 186 return false; 187 if (masks_are_equal && (longer < 0)) 188 for (lim2 = m - longer; m < lim2; ) 189 if (*m++) 190 return true; 191 return (!masks_are_equal); 192 } 193 194 struct radix_node * 195 rn_lookup(const char *key, const char *mask, struct radix_node_head *head) 196 { 197 struct radix_node *x; 198 const char *netmask = NULL; 199 200 if (mask != NULL) { 201 x = rn_addmask(mask, true, head->rnh_treetop->rn_offset, 202 head->rnh_maskhead); 203 if (x == NULL) 204 return (NULL); 205 netmask = x->rn_key; 206 } 207 208 x = rn_match(key, head); 209 if (x != NULL && netmask != NULL) { 210 while (x != NULL && x->rn_mask != netmask) 211 x = x->rn_dupedkey; 212 } 213 214 return x; 215 } 216 217 static bool 218 rn_satisfies_leaf(const char *trial, struct radix_node *leaf, int skip) 219 { 220 const char *cp = trial, *cp2 = leaf->rn_key, *cp3 = leaf->rn_mask; 221 const char *cplim; 222 int length; 223 224 length = min(clen(cp), clen(cp2)); 225 if (cp3 == NULL) 226 cp3 = rn_ones; 227 else 228 length = min(length, clen(cp3)); 229 230 cplim = cp + length; 231 cp2 += skip; 232 cp3 += skip; 233 for (cp += skip; cp < cplim; cp++, cp2++, cp3++) { 234 if ((*cp ^ *cp2) & *cp3) 235 return false; 236 } 237 238 return true; 239 } 240 241 struct radix_node * 242 rn_match(const char *key, struct radix_node_head *head) 243 { 244 struct radix_node *t, *x; 245 const char *cp = key, *cp2; 246 const char *cplim; 247 struct radix_node *saved_t, *top = head->rnh_treetop; 248 int off = top->rn_offset, klen, matched_off; 249 int test, b, rn_bit; 250 251 t = rn_search(key, top); 252 /* 253 * See if we match exactly as a host destination 254 * or at least learn how many bits match, for normal mask finesse. 255 * 256 * It doesn't hurt us to limit how many bytes to check 257 * to the length of the mask, since if it matches we had a genuine 258 * match and the leaf we have is the most specific one anyway; 259 * if it didn't match with a shorter length it would fail 260 * with a long one. This wins big for class B&C netmasks which 261 * are probably the most common case... 262 */ 263 if (t->rn_mask != NULL) 264 klen = clen(t->rn_mask); 265 else 266 klen = clen(key); 267 cp += off; cp2 = t->rn_key + off; cplim = key + klen; 268 for (; cp < cplim; cp++, cp2++) 269 if (*cp != *cp2) 270 goto on1; 271 272 /* 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; 282 283 on1: 284 test = (*cp ^ *cp2) & 0xff; /* find first bit that differs */ 285 for (b = 7; (test >>= 1) > 0;) 286 b--; 287 matched_off = cp - key; 288 b += matched_off << 3; 289 rn_bit = -1 - b; 290 291 /* 292 * If there is a host route in a duped-key chain, it will be first. 293 */ 294 saved_t = t; 295 if (t->rn_mask == NULL) 296 t = t->rn_dupedkey; 297 for (; t != NULL; t = t->rn_dupedkey) { 298 /* 299 * Even if we don't match exactly as a host, 300 * we may match if the leaf we wound up at is 301 * a route to a net. 302 */ 303 if (t->rn_flags & RNF_NORMAL) { 304 if (rn_bit <= t->rn_bit) 305 return t; 306 } else if (rn_satisfies_leaf(key, t, matched_off)) 307 return t; 308 } 309 t = saved_t; 310 311 /* start searching up the tree */ 312 do { 313 struct radix_mask *m; 314 315 t = t->rn_parent; 316 /* 317 * If non-contiguous masks ever become important 318 * we can restore the masking and open coding of 319 * the search and satisfaction test and put the 320 * calculation of "off" back before the "do". 321 */ 322 m = t->rn_mklist; 323 while (m != NULL) { 324 if (m->rm_flags & RNF_NORMAL) { 325 if (rn_bit <= m->rm_bit) 326 return (m->rm_leaf); 327 } else { 328 off = min(t->rn_offset, matched_off); 329 x = rn_search_m(key, t, m->rm_mask); 330 while (x != NULL && x->rn_mask != m->rm_mask) 331 x = x->rn_dupedkey; 332 if (x != NULL && rn_satisfies_leaf(key, x, off)) 333 return x; 334 } 335 m = m->rm_next; 336 } 337 } while (t != top); 338 339 return NULL; 340 } 341 342 static struct radix_node * 343 rn_newpair(const char *key, int indexbit, struct radix_node nodes[2]) 344 { 345 struct radix_node *leaf = &nodes[0], *interior = &nodes[1]; 346 347 interior->rn_bit = indexbit; 348 interior->rn_bmask = 0x80 >> (indexbit & 0x7); 349 interior->rn_offset = indexbit >> 3; 350 interior->rn_left = leaf; 351 interior->rn_flags = RNF_ACTIVE; 352 interior->rn_mklist = NULL; 353 354 leaf->rn_bit = -1; 355 leaf->rn_key = key; 356 leaf->rn_parent = interior; 357 leaf->rn_flags = interior->rn_flags; 358 leaf->rn_mklist = NULL; 359 360 #ifdef RN_DEBUG 361 leaf->rn_info = rn_nodenum++; 362 interior->rn_info = rn_nodenum++; 363 leaf->rn_twin = interior; 364 leaf->rn_ybro = rn_clist; 365 rn_clist = leaf; 366 #endif 367 368 return interior; 369 } 370 371 static struct radix_node * 372 rn_insert(const char *key, struct radix_node_head *head, bool *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 const 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 const char *cp2 = t->rn_key + head_off; 387 int cmp_res; 388 const char *cplim = key + klen; 389 390 while (cp < cplim) { 391 if (*cp2++ != *cp++) 392 goto on1; 393 } 394 395 *dupentry = true; 396 return t; 397 398 on1: 399 *dupentry = false; 400 cmp_res = (cp[-1] ^ cp2[-1]) & 0xff; 401 for (b = (cp - key) << 3; cmp_res; b--) 402 cmp_res >>= 1; 403 } 404 { 405 struct radix_node *p, *x = top; 406 407 cp = key; 408 do { 409 p = x; 410 if (cp[x->rn_offset] & x->rn_bmask) 411 x = x->rn_right; 412 else 413 x = x->rn_left; 414 } while (b > (unsigned) x->rn_bit); 415 /* x->rn_bit < b && x->rn_bit >= 0 */ 416 #ifdef RN_DEBUG 417 if (rn_debug) { 418 log(LOG_DEBUG, "%s: Going In:\n", __func__); 419 traverse(p); 420 } 421 #endif 422 t = rn_newpair(key, b, nodes); 423 tt = t->rn_left; 424 if ((cp[p->rn_offset] & p->rn_bmask) == 0) 425 p->rn_left = t; 426 else 427 p->rn_right = t; 428 x->rn_parent = t; 429 t->rn_parent = p; /* frees x, p as temp vars below */ 430 if ((cp[t->rn_offset] & t->rn_bmask) == 0) { 431 t->rn_right = x; 432 } else { 433 t->rn_right = tt; 434 t->rn_left = x; 435 } 436 #ifdef RN_DEBUG 437 if (rn_debug) { 438 log(LOG_DEBUG, "%s: Coming Out:\n", __func__); 439 traverse(p); 440 } 441 #endif 442 } 443 return (tt); 444 } 445 446 struct radix_node * 447 rn_addmask(const char *netmask, bool search, int skip, 448 struct radix_node_head *mask_rnh) 449 { 450 struct radix_node *x, *saved_x; 451 const char *cp, *cplim; 452 char *p; 453 int b = 0, mlen, j; 454 bool maskduplicated, isnormal; 455 char addmask_key[RN_MAXKEYLEN]; 456 457 if ((mlen = clen(netmask)) > RN_MAXKEYLEN) 458 mlen = RN_MAXKEYLEN; 459 if (skip == 0) 460 skip = 1; 461 if (mlen <= skip) 462 return (mask_rnh->rnh_nodes); /* all-zero key */ 463 464 bzero(addmask_key, sizeof(addmask_key)); 465 if (skip > 1) 466 bcopy(rn_ones + 1, addmask_key + 1, skip - 1); 467 bcopy(netmask + skip, addmask_key + skip, mlen - skip); 468 /* Trim trailing zeroes. */ 469 for (cp = addmask_key + mlen; (cp > addmask_key) && cp[-1] == 0;) 470 cp--; 471 mlen = cp - addmask_key; 472 if (mlen <= skip) 473 return (mask_rnh->rnh_nodes); /* all-zero key */ 474 475 *addmask_key = mlen; 476 x = rn_search(addmask_key, mask_rnh->rnh_treetop); 477 if (x->rn_key == NULL) { 478 kprintf("WARNING: radix_node->rn_key is NULL rn=%p\n", x); 479 print_backtrace(-1); 480 x = NULL; 481 } else if (bcmp(addmask_key, x->rn_key, mlen) != 0) { 482 x = NULL; 483 } 484 if (x != NULL || search) 485 return (x); 486 487 R_Malloc(x, struct radix_node *, RN_MAXKEYLEN + 2 * (sizeof *x)); 488 if ((saved_x = x) == NULL) 489 return (NULL); 490 491 bzero(x, RN_MAXKEYLEN + 2 * (sizeof *x)); 492 netmask = p = (char *)(x + 2); 493 bcopy(addmask_key, p, mlen); 494 x = rn_insert(netmask, mask_rnh, &maskduplicated, x); 495 if (maskduplicated) { 496 log(LOG_ERR, "%s: mask impossibly already in tree", __func__); 497 R_Free(saved_x); 498 return (x); 499 } 500 501 /* 502 * Calculate index of mask, and check for normalcy. 503 */ 504 isnormal = true; 505 cplim = netmask + mlen; 506 for (cp = netmask + skip; cp < cplim && clen(cp) == 0xff;) 507 cp++; 508 if (cp != cplim) { 509 static const char normal_chars[] = { 510 0, 0x80, 0xc0, 0xe0, 0xf0, 0xf8, 0xfc, 0xfe, -1 511 }; 512 513 for (j = 0x80; (j & *cp) != 0; j >>= 1) 514 b++; 515 if (*cp != normal_chars[b] || cp != (cplim - 1)) 516 isnormal = false; 517 } 518 b += (cp - netmask) << 3; 519 x->rn_bit = -1 - b; 520 if (isnormal) 521 x->rn_flags |= RNF_NORMAL; 522 return (x); 523 } 524 525 /* XXX: arbitrary ordering for non-contiguous masks */ 526 static bool 527 rn_lexobetter(const char *mp, const char *np) 528 { 529 const 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 } 550 551 bzero(m, sizeof *m); 552 m->rm_bit = tt->rn_bit; 553 m->rm_flags = tt->rn_flags; 554 if (tt->rn_flags & RNF_NORMAL) 555 m->rm_leaf = tt; 556 else 557 m->rm_mask = tt->rn_mask; 558 m->rm_next = nextmask; 559 tt->rn_mklist = m; 560 561 return m; 562 } 563 564 struct radix_node * 565 rn_addroute(const char *key, const char *netmask, 566 struct radix_node_head *head, struct radix_node treenodes[2]) 567 { 568 struct radix_node *t, *x = NULL, *tt; 569 struct radix_node *saved_tt, *top = head->rnh_treetop; 570 short b = 0, b_leaf = 0; 571 bool keyduplicated; 572 const char *mmask; 573 struct radix_mask *m, **mp; 574 575 /* 576 * In dealing with non-contiguous masks, there may be 577 * many different routes which have the same mask. 578 * We will find it useful to have a unique pointer to 579 * the mask to speed avoiding duplicate references at 580 * nodes and possibly save time in calculating indices. 581 */ 582 if (netmask != NULL) { 583 if ((x = rn_addmask(netmask, false, top->rn_offset, 584 head->rnh_maskhead)) == NULL) 585 return (NULL); 586 b_leaf = x->rn_bit; 587 b = -1 - x->rn_bit; 588 netmask = x->rn_key; 589 } 590 /* 591 * Deal with duplicated keys: attach node to previous instance 592 */ 593 saved_tt = tt = rn_insert(key, head, &keyduplicated, treenodes); 594 if (keyduplicated) { 595 for (t = tt; tt; t = tt, tt = tt->rn_dupedkey) { 596 if (tt->rn_mask == netmask) 597 return (NULL); 598 if (netmask == NULL || 599 (tt->rn_mask && 600 ((b_leaf < tt->rn_bit) /* index(netmask) > node */ 601 || rn_refines(netmask, tt->rn_mask) 602 || rn_lexobetter(netmask, tt->rn_mask)))) 603 break; 604 } 605 /* 606 * If the mask is not duplicated, we wouldn't 607 * find it among possible duplicate key entries 608 * anyway, so the above test doesn't hurt. 609 * 610 * We sort the masks for a duplicated key the same way as 611 * in a masklist -- most specific to least specific. 612 * This may require the unfortunate nuisance of relocating 613 * the head of the list. 614 */ 615 if (tt == saved_tt) { 616 struct radix_node *xx = x; 617 /* link in at head of list */ 618 (tt = treenodes)->rn_dupedkey = t; 619 tt->rn_flags = t->rn_flags; 620 tt->rn_parent = x = t->rn_parent; 621 t->rn_parent = tt; /* parent */ 622 if (x->rn_left == t) 623 x->rn_left = tt; 624 else 625 x->rn_right = tt; 626 saved_tt = tt; x = xx; 627 } else { 628 (tt = treenodes)->rn_dupedkey = t->rn_dupedkey; 629 t->rn_dupedkey = tt; 630 tt->rn_parent = t; /* parent */ 631 if (tt->rn_dupedkey != NULL) /* parent */ 632 tt->rn_dupedkey->rn_parent = tt; /* parent */ 633 } 634 #ifdef RN_DEBUG 635 t=tt+1; tt->rn_info = rn_nodenum++; t->rn_info = rn_nodenum++; 636 tt->rn_twin = t; tt->rn_ybro = rn_clist; rn_clist = tt; 637 #endif 638 tt->rn_key = key; 639 tt->rn_bit = -1; 640 tt->rn_flags = RNF_ACTIVE; 641 } 642 /* 643 * Put mask in tree. 644 */ 645 if (netmask != NULL) { 646 tt->rn_mask = netmask; 647 tt->rn_bit = x->rn_bit; 648 tt->rn_flags |= x->rn_flags & RNF_NORMAL; 649 } 650 t = saved_tt->rn_parent; 651 if (keyduplicated) 652 goto on2; 653 b_leaf = -1 - t->rn_bit; 654 if (t->rn_right == saved_tt) 655 x = t->rn_left; 656 else 657 x = t->rn_right; 658 /* Promote general routes from below */ 659 if (x->rn_bit < 0) { 660 mp = &t->rn_mklist; 661 while (x != NULL) { 662 if (x->rn_mask != NULL && 663 x->rn_bit >= b_leaf && 664 x->rn_mklist == NULL) { 665 *mp = m = rn_new_radix_mask(x, NULL); 666 if (m != NULL) 667 mp = &m->rm_next; 668 } 669 x = x->rn_dupedkey; 670 } 671 } else if (x->rn_mklist != NULL) { 672 /* 673 * Skip over masks whose index is > that of new node 674 */ 675 for (mp = &x->rn_mklist; (m = *mp) != NULL; mp = &m->rm_next) { 676 if (m->rm_bit >= b_leaf) 677 break; 678 } 679 t->rn_mklist = m; 680 *mp = NULL; 681 } 682 on2: 683 /* Add new route to highest possible ancestor's list */ 684 if ((netmask == NULL) || (b > t->rn_bit )) 685 return tt; /* can't lift at all */ 686 b_leaf = tt->rn_bit; 687 do { 688 x = t; 689 t = t->rn_parent; 690 } while (b <= t->rn_bit && x != top); 691 /* 692 * Search through routes associated with node to 693 * insert new route according to index. 694 * Need same criteria as when sorting dupedkeys to avoid 695 * double loop on deletion. 696 */ 697 for (mp = &x->rn_mklist; (m = *mp) != NULL; mp = &m->rm_next) { 698 if (m->rm_bit < b_leaf) 699 continue; 700 if (m->rm_bit > b_leaf) 701 break; 702 if (m->rm_flags & RNF_NORMAL) { 703 mmask = m->rm_leaf->rn_mask; 704 if (tt->rn_flags & RNF_NORMAL) { 705 log(LOG_ERR, 706 "Non-unique normal route, mask not entered\n"); 707 return tt; 708 } 709 } else 710 mmask = m->rm_mask; 711 if (mmask == netmask) { 712 m->rm_refs++; 713 tt->rn_mklist = m; 714 return tt; 715 } 716 if (rn_refines(netmask, mmask) || rn_lexobetter(netmask, mmask)) 717 break; 718 } 719 *mp = rn_new_radix_mask(tt, *mp); 720 return tt; 721 } 722 723 struct radix_node * 724 rn_delete(const char *key, const char *netmask, struct radix_node_head *head) 725 { 726 struct radix_node *t, *p, *x, *tt; 727 struct radix_mask *m, *saved_m, **mp; 728 struct radix_node *dupedkey, *saved_tt, *top; 729 int b, head_off, klen; 730 int cpu = mycpuid; 731 732 x = head->rnh_treetop; 733 tt = rn_search(key, x); 734 head_off = x->rn_offset; 735 klen = clen(key); 736 saved_tt = tt; 737 top = x; 738 if (tt == NULL || 739 bcmp(key + head_off, tt->rn_key + head_off, klen - head_off)) 740 return (NULL); 741 742 /* 743 * Delete our route from mask lists. 744 */ 745 if (netmask != NULL) { 746 if ((x = rn_addmask(netmask, true, head_off, 747 head->rnh_maskhead)) == NULL) 748 return (NULL); 749 netmask = x->rn_key; 750 while (tt->rn_mask != netmask) 751 if ((tt = tt->rn_dupedkey) == NULL) 752 return (NULL); 753 } 754 if (tt->rn_mask == NULL || (saved_m = m = tt->rn_mklist) == NULL) 755 goto on1; 756 if (tt->rn_flags & RNF_NORMAL) { 757 if (m->rm_leaf != tt || m->rm_refs > 0) { 758 log(LOG_ERR, "rn_delete: inconsistent annotation\n"); 759 return (NULL); /* dangling ref could cause disaster */ 760 } 761 } else { 762 if (m->rm_mask != tt->rn_mask) { 763 log(LOG_ERR, "rn_delete: inconsistent annotation\n"); 764 goto on1; 765 } 766 if (--m->rm_refs >= 0) 767 goto on1; 768 } 769 b = -1 - tt->rn_bit; 770 t = saved_tt->rn_parent; 771 if (b > t->rn_bit) 772 goto on1; /* Wasn't lifted at all */ 773 774 do { 775 x = t; 776 t = t->rn_parent; 777 } while (b <= t->rn_bit && x != top); 778 for (mp = &x->rn_mklist; (m = *mp) != NULL; mp = &m->rm_next) 779 if (m == saved_m) { 780 *mp = m->rm_next; 781 MKFree(&rn_mkfreelist[cpu], m); 782 break; 783 } 784 if (m == NULL) { 785 log(LOG_ERR, "rn_delete: couldn't find our annotation\n"); 786 if (tt->rn_flags & RNF_NORMAL) 787 return (NULL); /* Dangling ref to us */ 788 } 789 790 on1: 791 /* 792 * Eliminate us from tree 793 */ 794 if (tt->rn_flags & RNF_ROOT) 795 return (NULL); 796 797 #ifdef RN_DEBUG 798 /* Get us out of the creation list */ 799 for (t = rn_clist; t != NULL && t->rn_ybro != tt; t = t->rn_ybro) 800 ; 801 if (t != NULL) 802 t->rn_ybro = tt->rn_ybro; 803 #endif 804 805 t = tt->rn_parent; 806 dupedkey = saved_tt->rn_dupedkey; 807 if (dupedkey != NULL) { 808 /* 809 * at this point, tt is the deletion target and saved_tt 810 * is the head of the dupekey chain 811 */ 812 if (tt == saved_tt) { 813 /* remove from head of chain */ 814 x = dupedkey; 815 x->rn_parent = t; 816 if (t->rn_left == tt) 817 t->rn_left = x; 818 else 819 t->rn_right = x; 820 } else { 821 /* find node in front of tt on the chain */ 822 for (x = p = saved_tt; p != NULL && p->rn_dupedkey != tt;) 823 p = p->rn_dupedkey; 824 if (p) { 825 p->rn_dupedkey = tt->rn_dupedkey; 826 if (tt->rn_dupedkey) /* parent */ 827 tt->rn_dupedkey->rn_parent = p; 828 /* parent */ 829 } else { 830 log(LOG_ERR, "rn_delete: couldn't find us\n"); 831 } 832 } 833 t = tt + 1; 834 if (t->rn_flags & RNF_ACTIVE) { 835 #ifndef RN_DEBUG 836 *++x = *t; 837 p = t->rn_parent; 838 #else 839 b = t->rn_info; 840 *++x = *t; 841 t->rn_info = b; 842 p = t->rn_parent; 843 #endif 844 if (p->rn_left == t) 845 p->rn_left = x; 846 else 847 p->rn_right = x; 848 x->rn_left->rn_parent = x; 849 x->rn_right->rn_parent = x; 850 } 851 goto out; 852 } 853 if (t->rn_left == tt) 854 x = t->rn_right; 855 else 856 x = t->rn_left; 857 p = t->rn_parent; 858 if (p->rn_right == t) 859 p->rn_right = x; 860 else 861 p->rn_left = x; 862 x->rn_parent = p; 863 /* 864 * Demote routes attached to us. 865 */ 866 if (t->rn_mklist != NULL) { 867 if (x->rn_bit >= 0) { 868 for (mp = &x->rn_mklist; (m = *mp) != NULL;) 869 mp = &m->rm_next; 870 *mp = t->rn_mklist; 871 } else { 872 /* 873 * If there are any (key, mask) pairs in a sibling 874 * duped-key chain, some subset will appear sorted 875 * in the same order attached to our mklist. 876 */ 877 for (m = t->rn_mklist; m && x; x = x->rn_dupedkey) 878 if (m == x->rn_mklist) { 879 struct radix_mask *mm = m->rm_next; 880 881 x->rn_mklist = NULL; 882 if (--(m->rm_refs) < 0) 883 MKFree(&rn_mkfreelist[cpu], m); 884 m = mm; 885 } 886 if (m) { 887 log(LOG_ERR, 888 "rn_delete: Orphaned Mask %p at %p\n", 889 (void *)m, (void *)x); 890 } 891 } 892 } 893 /* 894 * We may be holding an active internal node in the tree. 895 */ 896 x = tt + 1; 897 if (t != x) { 898 #ifndef RN_DEBUG 899 *t = *x; 900 #else 901 b = t->rn_info; 902 *t = *x; 903 t->rn_info = b; 904 #endif 905 t->rn_left->rn_parent = t; 906 t->rn_right->rn_parent = t; 907 p = x->rn_parent; 908 if (p->rn_left == x) 909 p->rn_left = t; 910 else 911 p->rn_right = t; 912 } 913 914 out: 915 tt->rn_flags &= ~RNF_ACTIVE; 916 tt[1].rn_flags &= ~RNF_ACTIVE; 917 return (tt); 918 } 919 920 /* 921 * This is the same as rn_walktree() except for the parameters and the 922 * exit. 923 */ 924 static int 925 rn_walktree_from(struct radix_node_head *h, const char *xa, const char *xm, 926 walktree_f_t *f, void *w) 927 { 928 struct radix_node *base, *next; 929 struct radix_node *rn, *last = NULL /* shut up gcc */; 930 bool stopping = false; 931 int lastb, error; 932 933 /* 934 * rn_search_m is sort-of-open-coded here. 935 */ 936 /* kprintf("about to search\n"); */ 937 for (rn = h->rnh_treetop; rn->rn_bit >= 0; ) { 938 last = rn; 939 /* kprintf("rn_bit %d, rn_bmask %x, xm[rn_offset] %x\n", 940 rn->rn_bit, rn->rn_bmask, xm[rn->rn_offset]); */ 941 if (!(rn->rn_bmask & xm[rn->rn_offset])) { 942 break; 943 } 944 if (rn->rn_bmask & xa[rn->rn_offset]) { 945 rn = rn->rn_right; 946 } else { 947 rn = rn->rn_left; 948 } 949 } 950 /* kprintf("done searching\n"); */ 951 952 /* 953 * Two cases: either we stepped off the end of our mask, 954 * in which case last == rn, or we reached a leaf, in which 955 * case we want to start from the last node we looked at. 956 * Either way, last is the node we want to start from. 957 */ 958 rn = last; 959 lastb = rn->rn_bit; 960 961 /* kprintf("rn %p, lastb %d\n", rn, lastb);*/ 962 963 /* 964 * This gets complicated because we may delete the node 965 * while applying the function f to it, so we need to calculate 966 * the successor node in advance. 967 */ 968 while (rn->rn_bit >= 0) 969 rn = rn->rn_left; 970 971 while (!stopping) { 972 /* kprintf("node %p (%d)\n", rn, rn->rn_bit); */ 973 base = rn; 974 /* If at right child go back up, otherwise, go right */ 975 while (rn->rn_parent->rn_right == rn && 976 !(rn->rn_flags & RNF_ROOT)) { 977 rn = rn->rn_parent; 978 979 /* if went up beyond last, stop */ 980 if (rn->rn_bit < lastb) { 981 stopping = true; 982 /* kprintf("up too far\n"); */ 983 } 984 } 985 986 /* Find the next *leaf* since next node might vanish, too */ 987 for (rn = rn->rn_parent->rn_right; rn->rn_bit >= 0;) 988 rn = rn->rn_left; 989 next = rn; 990 /* Process leaves */ 991 while ((rn = base) != NULL) { 992 base = rn->rn_dupedkey; 993 /* kprintf("leaf %p\n", rn); */ 994 if (!(rn->rn_flags & RNF_ROOT) && (error = (*f)(rn, w))) 995 return (error); 996 } 997 rn = next; 998 999 if (rn->rn_flags & RNF_ROOT) { 1000 /* kprintf("root, stopping"); */ 1001 stopping = true; 1002 } 1003 } 1004 1005 return 0; 1006 } 1007 1008 static int 1009 rn_walktree_at(struct radix_node_head *h, const char *a, const char *m, 1010 walktree_f_t *f, void *w) 1011 { 1012 struct radix_node *base, *next; 1013 struct radix_node *rn = h->rnh_treetop; 1014 int error; 1015 1016 /* 1017 * This gets complicated because we may delete the node 1018 * while applying the function f to it, so we need to calculate 1019 * the successor node in advance. 1020 */ 1021 if (a == NULL) { 1022 /* First time through node, go left */ 1023 while (rn->rn_bit >= 0) 1024 rn = rn->rn_left; 1025 } else { 1026 if (m != NULL) 1027 rn = rn_search_m(a, rn, m); 1028 else 1029 rn = rn_search(a, rn); 1030 } 1031 for (;;) { 1032 base = rn; 1033 /* If at right child go back up, otherwise, go right */ 1034 while (rn->rn_parent->rn_right == rn && 1035 !(rn->rn_flags & RNF_ROOT)) 1036 rn = rn->rn_parent; 1037 /* Find the next *leaf* since next node might vanish, too */ 1038 for (rn = rn->rn_parent->rn_right; rn->rn_bit >= 0;) 1039 rn = rn->rn_left; 1040 next = rn; 1041 /* Process leaves */ 1042 while ((rn = base)) { 1043 base = rn->rn_dupedkey; 1044 if (!(rn->rn_flags & RNF_ROOT) && (error = (*f)(rn, w))) 1045 return (error); 1046 } 1047 rn = next; 1048 if (rn->rn_flags & RNF_ROOT) 1049 return (0); 1050 } 1051 /* NOTREACHED */ 1052 } 1053 1054 static int 1055 rn_walktree(struct radix_node_head *h, walktree_f_t *f, void *w) 1056 { 1057 return rn_walktree_at(h, NULL, NULL, f, w); 1058 } 1059 1060 int 1061 rn_inithead(void **head, struct radix_node_head *maskhead, int off) 1062 { 1063 struct radix_node_head *rnh; 1064 struct radix_node *root, *left, *right; 1065 1066 if (*head != NULL) /* already initialized */ 1067 return (1); 1068 1069 R_Malloc(rnh, struct radix_node_head *, sizeof *rnh); 1070 if (rnh == NULL) 1071 return (0); 1072 1073 bzero(rnh, sizeof *rnh); 1074 *head = rnh; 1075 1076 root = rn_newpair(rn_zeros, off, rnh->rnh_nodes); 1077 right = &rnh->rnh_nodes[2]; 1078 root->rn_parent = root; 1079 root->rn_flags = RNF_ROOT | RNF_ACTIVE; 1080 root->rn_right = right; 1081 1082 left = root->rn_left; 1083 left->rn_bit = -1 - off; 1084 left->rn_flags = root->rn_flags; 1085 1086 *right = *left; 1087 right->rn_key = rn_ones; 1088 1089 rnh->rnh_treetop = root; 1090 rnh->rnh_maskhead = maskhead; 1091 1092 rnh->rnh_addaddr = rn_addroute; 1093 rnh->rnh_deladdr = rn_delete; 1094 rnh->rnh_matchaddr = rn_match; 1095 rnh->rnh_lookup = rn_lookup; 1096 rnh->rnh_walktree = rn_walktree; 1097 rnh->rnh_walktree_from = rn_walktree_from; 1098 rnh->rnh_walktree_at = rn_walktree_at; 1099 1100 return (1); 1101 } 1102 1103 static void 1104 rn_init_handler(netmsg_t msg) 1105 { 1106 int cpu = mycpuid; 1107 1108 ASSERT_NETISR_NCPUS(cpu); 1109 if (rn_inithead((void **)&mask_rnheads[cpu], NULL, 0) == 0) 1110 panic("rn_init 2"); 1111 1112 netisr_forwardmsg(&msg->base, cpu + 1); 1113 } 1114 1115 void 1116 rn_init(void) 1117 { 1118 struct netmsg_base msg; 1119 #ifdef _KERNEL 1120 struct domain *dom; 1121 1122 SLIST_FOREACH(dom, &domains, dom_next) { 1123 if (dom->dom_maxrtkey > RN_MAXKEYLEN) { 1124 panic("domain %s maxkey too big %d/%d", 1125 dom->dom_name, dom->dom_maxrtkey, RN_MAXKEYLEN); 1126 } 1127 } 1128 #endif 1129 1130 netmsg_init(&msg, NULL, &curthread->td_msgport, 0, rn_init_handler); 1131 netisr_domsg_global(&msg); 1132 } 1133 1134 struct radix_node_head * 1135 rn_cpumaskhead(int cpu) 1136 { 1137 ASSERT_NETISR_NCPUS(cpu); 1138 KKASSERT(mask_rnheads[cpu] != NULL); 1139 return mask_rnheads[cpu]; 1140 } 1141