1 /* 2 * Copyright (c) 1983, 1988, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 */ 33 34 #if !defined(lint) && !defined(sgi) && !defined(__NetBSD__) 35 static char sccsid[] = "@(#)tables.c 8.1 (Berkeley) 6/5/93"; 36 #elif defined(__NetBSD__) 37 static char rcsid[] = "$NetBSD$"; 38 #endif 39 40 #include "defs.h" 41 42 static struct rt_spare *rts_better(struct rt_entry *); 43 44 struct radix_node_head *rhead; /* root of the radix tree */ 45 46 int need_flash = 1; /* flash update needed 47 * start =1 to suppress the 1st 48 */ 49 50 struct timeval age_timer; /* next check of old routes */ 51 struct timeval need_kern = { /* need to update kernel table */ 52 EPOCH+MIN_WAITTIME-1 53 }; 54 55 int stopint; 56 57 int total_routes; 58 59 /* zap any old routes through this gateway */ 60 naddr age_bad_gate; 61 62 63 /* It is desirable to "aggregate" routes, to combine differing routes of 64 * the same metric and next hop into a common route with a smaller netmask 65 * or to suppress redundant routes, routes that add no information to 66 * routes with smaller netmasks. 67 * 68 * A route is redundant if and only if any and all routes with smaller 69 * but matching netmasks and nets are the same. Since routes are 70 * kept sorted in the radix tree, redundant routes always come second. 71 * 72 * There are two kinds of aggregations. First, two routes of the same bit 73 * mask and differing only in the least significant bit of the network 74 * number can be combined into a single route with a coarser mask. 75 * 76 * Second, a route can be suppressed in favor of another route with a more 77 * coarse mask provided no incompatible routes with intermediate masks 78 * are present. The second kind of aggregation involves suppressing routes. 79 * A route must not be suppressed if an incompatible route exists with 80 * an intermediate mask, since the suppressed route would be covered 81 * by the intermediate. 82 * 83 * This code relies on the radix tree walk encountering routes 84 * sorted first by address, with the smallest address first. 85 */ 86 87 struct ag_info ag_slots[NUM_AG_SLOTS], *ag_avail, *ag_corsest, *ag_finest; 88 89 /* #define DEBUG_AG */ 90 #ifdef DEBUG_AG 91 #define CHECK_AG() {int acnt = 0; struct ag_info *cag; \ 92 for (cag = ag_avail; cag != 0; cag = cag->ag_fine) \ 93 acnt++; \ 94 for (cag = ag_corsest; cag != 0; cag = cag->ag_fine) \ 95 acnt++; \ 96 if (acnt != NUM_AG_SLOTS) { \ 97 (void)fflush(stderr); \ 98 abort(); \ 99 } \ 100 } 101 #else 102 #define CHECK_AG() 103 #endif 104 105 106 /* Output the contents of an aggregation table slot. 107 * This function must always be immediately followed with the deletion 108 * of the target slot. 109 */ 110 static void 111 ag_out(struct ag_info *ag, 112 void (*out)(struct ag_info *)) 113 { 114 struct ag_info *ag_cors; 115 naddr bit; 116 117 118 /* If we output both the even and odd twins, then the immediate parent, 119 * if it is present, is redundant, unless the parent manages to 120 * aggregate into something coarser. 121 * On successive calls, this code detects the even and odd twins, 122 * and marks the parent. 123 * 124 * Note that the order in which the radix tree code emits routes 125 * ensures that the twins are seen before the parent is emitted. 126 */ 127 ag_cors = ag->ag_cors; 128 if (ag_cors != 0 129 && ag_cors->ag_mask == ag->ag_mask<<1 130 && ag_cors->ag_dst_h == (ag->ag_dst_h & ag_cors->ag_mask)) { 131 ag_cors->ag_state |= ((ag_cors->ag_dst_h == ag->ag_dst_h) 132 ? AGS_REDUN0 133 : AGS_REDUN1); 134 } 135 136 /* Skip it if this route is itself redundant. 137 * 138 * It is ok to change the contents of the slot here, since it is 139 * always deleted next. 140 */ 141 if (ag->ag_state & AGS_REDUN0) { 142 if (ag->ag_state & AGS_REDUN1) 143 return; 144 bit = (-ag->ag_mask) >> 1; 145 ag->ag_dst_h |= bit; 146 ag->ag_mask |= bit; 147 148 } else if (ag->ag_state & AGS_REDUN1) { 149 bit = (-ag->ag_mask) >> 1; 150 ag->ag_mask |= bit; 151 } 152 out(ag); 153 } 154 155 156 static void 157 ag_del(struct ag_info *ag) 158 { 159 CHECK_AG(); 160 161 if (ag->ag_cors == 0) 162 ag_corsest = ag->ag_fine; 163 else 164 ag->ag_cors->ag_fine = ag->ag_fine; 165 166 if (ag->ag_fine == 0) 167 ag_finest = ag->ag_cors; 168 else 169 ag->ag_fine->ag_cors = ag->ag_cors; 170 171 ag->ag_fine = ag_avail; 172 ag_avail = ag; 173 174 CHECK_AG(); 175 } 176 177 178 /* Flush routes waiting for aggretation. 179 * This must not suppress a route unless it is known that among all 180 * routes with coarser masks that match it, the one with the longest 181 * mask is appropriate. This is ensured by scanning the routes 182 * in lexical order, and with the most restritive mask first 183 * among routes to the same destination. 184 */ 185 void 186 ag_flush(naddr lim_dst_h, /* flush routes to here */ 187 naddr lim_mask, /* matching this mask */ 188 void (*out)(struct ag_info *)) 189 { 190 struct ag_info *ag, *ag_cors; 191 naddr dst_h; 192 193 194 for (ag = ag_finest; 195 ag != 0 && ag->ag_mask >= lim_mask; 196 ag = ag_cors) { 197 ag_cors = ag->ag_cors; 198 199 /* work on only the specified routes */ 200 dst_h = ag->ag_dst_h; 201 if ((dst_h & lim_mask) != lim_dst_h) 202 continue; 203 204 if (!(ag->ag_state & AGS_SUPPRESS)) 205 ag_out(ag, out); 206 207 else for ( ; ; ag_cors = ag_cors->ag_cors) { 208 /* Look for a route that can suppress the 209 * current route */ 210 if (ag_cors == 0) { 211 /* failed, so output it and look for 212 * another route to work on 213 */ 214 ag_out(ag, out); 215 break; 216 } 217 218 if ((dst_h & ag_cors->ag_mask) == ag_cors->ag_dst_h) { 219 /* We found a route with a coarser mask that 220 * aggregates the current target. 221 * 222 * If it has a different next hop, it 223 * cannot replace the target, so output 224 * the target. 225 */ 226 if (ag->ag_gate != ag_cors->ag_gate 227 && !(ag->ag_state & AGS_FINE_GATE) 228 && !(ag_cors->ag_state & AGS_CORS_GATE)) { 229 ag_out(ag, out); 230 break; 231 } 232 233 /* If the coarse route has a good enough 234 * metric, it suppresses the target. 235 */ 236 if (ag_cors->ag_pref <= ag->ag_pref) { 237 if (ag_cors->ag_seqno > ag->ag_seqno) 238 ag_cors->ag_seqno = ag->ag_seqno; 239 if (AG_IS_REDUN(ag->ag_state) 240 && ag_cors->ag_mask==ag->ag_mask<<1) { 241 if (ag_cors->ag_dst_h == dst_h) 242 ag_cors->ag_state |= AGS_REDUN0; 243 else 244 ag_cors->ag_state |= AGS_REDUN1; 245 } 246 if (ag->ag_tag != ag_cors->ag_tag) 247 ag_cors->ag_tag = 0; 248 if (ag->ag_nhop != ag_cors->ag_nhop) 249 ag_cors->ag_nhop = 0; 250 break; 251 } 252 } 253 } 254 255 /* That route has either been output or suppressed */ 256 ag_cors = ag->ag_cors; 257 ag_del(ag); 258 } 259 260 CHECK_AG(); 261 } 262 263 264 /* Try to aggregate a route with previous routes. 265 */ 266 void 267 ag_check(naddr dst, 268 naddr mask, 269 naddr gate, 270 naddr nhop, 271 char metric, 272 char pref, 273 u_int seqno, 274 u_short tag, 275 u_short state, 276 void (*out)(struct ag_info *)) /* output using this */ 277 { 278 struct ag_info *ag, *nag, *ag_cors; 279 naddr xaddr; 280 int x; 281 282 NTOHL(dst); 283 284 /* Punt non-contiguous subnet masks. 285 * 286 * (X & -X) contains a single bit if and only if X is a power of 2. 287 * (X + (X & -X)) == 0 if and only if X is a power of 2. 288 */ 289 if ((mask & -mask) + mask != 0) { 290 struct ag_info nc_ag; 291 292 nc_ag.ag_dst_h = dst; 293 nc_ag.ag_mask = mask; 294 nc_ag.ag_gate = gate; 295 nc_ag.ag_nhop = nhop; 296 nc_ag.ag_metric = metric; 297 nc_ag.ag_pref = pref; 298 nc_ag.ag_tag = tag; 299 nc_ag.ag_state = state; 300 nc_ag.ag_seqno = seqno; 301 out(&nc_ag); 302 return; 303 } 304 305 /* Search for the right slot in the aggregation table. 306 */ 307 ag_cors = 0; 308 ag = ag_corsest; 309 while (ag != 0) { 310 if (ag->ag_mask >= mask) 311 break; 312 313 /* Suppress old routes (i.e. combine with compatible routes 314 * with coarser masks) as we look for the right slot in the 315 * aggregation table for the new route. 316 * A route to an address less than the current destination 317 * will not be affected by the current route or any route 318 * seen hereafter. That means it is safe to suppress it. 319 * This check keeps poor routes (eg. with large hop counts) 320 * from preventing suppresion of finer routes. 321 */ 322 if (ag_cors != 0 323 && ag->ag_dst_h < dst 324 && (ag->ag_state & AGS_SUPPRESS) 325 && ag_cors->ag_pref <= ag->ag_pref 326 && (ag->ag_dst_h & ag_cors->ag_mask) == ag_cors->ag_dst_h 327 && (ag_cors->ag_gate == ag->ag_gate 328 || (ag->ag_state & AGS_FINE_GATE) 329 || (ag_cors->ag_state & AGS_CORS_GATE))) { 330 if (ag_cors->ag_seqno > ag->ag_seqno) 331 ag_cors->ag_seqno = ag->ag_seqno; 332 if (AG_IS_REDUN(ag->ag_state) 333 && ag_cors->ag_mask==ag->ag_mask<<1) { 334 if (ag_cors->ag_dst_h == dst) 335 ag_cors->ag_state |= AGS_REDUN0; 336 else 337 ag_cors->ag_state |= AGS_REDUN1; 338 } 339 if (ag->ag_tag != ag_cors->ag_tag) 340 ag_cors->ag_tag = 0; 341 if (ag->ag_nhop != ag_cors->ag_nhop) 342 ag_cors->ag_nhop = 0; 343 ag_del(ag); 344 CHECK_AG(); 345 } else { 346 ag_cors = ag; 347 } 348 ag = ag_cors->ag_fine; 349 } 350 351 /* If we find the even/odd twin of the new route, and if the 352 * masks and so forth are equal, we can aggregate them. 353 * We can probably promote one of the pair. 354 * 355 * Since the routes are encountered in lexical order, 356 * the new route must be odd. However, the second or later 357 * times around this loop, it could be the even twin promoted 358 * from the even/odd pair of twins of the finer route. 359 */ 360 while (ag != 0 361 && ag->ag_mask == mask 362 && ((ag->ag_dst_h ^ dst) & (mask<<1)) == 0) { 363 364 /* Here we know the target route and the route in the current 365 * slot have the same netmasks and differ by at most the 366 * last bit. They are either for the same destination, or 367 * for an even/odd pair of destinations. 368 */ 369 if (ag->ag_dst_h == dst) { 370 /* We have two routes to the same destination. 371 * Routes are encountered in lexical order, so a 372 * route is never promoted until the parent route is 373 * already present. So we know that the new route is 374 * a promoted pair and the route already in the slot 375 * is the explicit route. 376 * 377 * Prefer the best route if their metrics differ, 378 * or the promoted one if not, following a sort 379 * of longest-match rule. 380 */ 381 if (pref <= ag->ag_pref) { 382 ag->ag_gate = gate; 383 ag->ag_nhop = nhop; 384 ag->ag_tag = tag; 385 ag->ag_metric = metric; 386 ag->ag_pref = pref; 387 x = ag->ag_state; 388 ag->ag_state = state; 389 state = x; 390 } 391 392 /* The sequence number controls flash updating, 393 * and should be the smaller of the two. 394 */ 395 if (ag->ag_seqno > seqno) 396 ag->ag_seqno = seqno; 397 398 /* some bits are set if they are set on either route */ 399 ag->ag_state |= (state & (AGS_PROMOTE_EITHER 400 | AGS_REDUN0 | AGS_REDUN1)); 401 return; 402 } 403 404 /* If one of the routes can be promoted and the other can 405 * be suppressed, it may be possible to combine them or 406 * worthwhile to promote one. 407 * 408 * Note that any route that can be promoted is always 409 * marked to be eligible to be suppressed. 410 */ 411 if (!((state & AGS_PROMOTE) 412 && (ag->ag_state & AGS_SUPPRESS)) 413 && !((ag->ag_state & AGS_PROMOTE) 414 && (state & AGS_SUPPRESS))) 415 break; 416 417 /* A pair of even/odd twin routes can be combined 418 * if either is redundant, or if they are via the 419 * same gateway and have the same metric. 420 */ 421 if (AG_IS_REDUN(ag->ag_state) 422 || AG_IS_REDUN(state) 423 || (ag->ag_gate == gate 424 && ag->ag_pref == pref 425 && (state & ag->ag_state & AGS_PROMOTE) != 0)) { 426 427 /* We have both the even and odd pairs. 428 * Since the routes are encountered in order, 429 * the route in the slot must be the even twin. 430 * 431 * Combine and promote the pair of routes. 432 */ 433 if (seqno > ag->ag_seqno) 434 seqno = ag->ag_seqno; 435 if (!AG_IS_REDUN(state)) 436 state &= ~AGS_REDUN1; 437 if (AG_IS_REDUN(ag->ag_state)) 438 state |= AGS_REDUN0; 439 else 440 state &= ~AGS_REDUN0; 441 state |= (ag->ag_state & AGS_PROMOTE_EITHER); 442 if (ag->ag_tag != tag) 443 tag = 0; 444 if (ag->ag_nhop != nhop) 445 nhop = 0; 446 447 /* Get rid of the even twin that was already 448 * in the slot. 449 */ 450 ag_del(ag); 451 452 } else if (ag->ag_pref >= pref 453 && (ag->ag_state & AGS_PROMOTE)) { 454 /* If we cannot combine the pair, maybe the route 455 * with the worse metric can be promoted. 456 * 457 * Promote the old, even twin, by giving its slot 458 * in the table to the new, odd twin. 459 */ 460 ag->ag_dst_h = dst; 461 462 xaddr = ag->ag_gate; 463 ag->ag_gate = gate; 464 gate = xaddr; 465 466 xaddr = ag->ag_nhop; 467 ag->ag_nhop = nhop; 468 nhop = xaddr; 469 470 x = ag->ag_tag; 471 ag->ag_tag = tag; 472 tag = x; 473 474 x = ag->ag_state; 475 ag->ag_state = state; 476 state = x; 477 if (!AG_IS_REDUN(state)) 478 state &= ~AGS_REDUN0; 479 480 x = ag->ag_metric; 481 ag->ag_metric = metric; 482 metric = x; 483 484 x = ag->ag_pref; 485 ag->ag_pref = pref; 486 pref = x; 487 488 if (seqno >= ag->ag_seqno) 489 seqno = ag->ag_seqno; 490 else 491 ag->ag_seqno = seqno; 492 493 } else { 494 if (!(state & AGS_PROMOTE)) 495 break; /* cannot promote either twin */ 496 497 /* promote the new, odd twin by shaving its 498 * mask and address. 499 */ 500 if (seqno > ag->ag_seqno) 501 seqno = ag->ag_seqno; 502 else 503 ag->ag_seqno = seqno; 504 if (!AG_IS_REDUN(state)) 505 state &= ~AGS_REDUN1; 506 } 507 508 mask <<= 1; 509 dst &= mask; 510 511 if (ag_cors == 0) { 512 ag = ag_corsest; 513 break; 514 } 515 ag = ag_cors; 516 ag_cors = ag->ag_cors; 517 } 518 519 /* When we can no longer promote and combine routes, 520 * flush the old route in the target slot. Also flush 521 * any finer routes that we know will never be aggregated by 522 * the new route. 523 * 524 * In case we moved toward coarser masks, 525 * get back where we belong 526 */ 527 if (ag != 0 528 && ag->ag_mask < mask) { 529 ag_cors = ag; 530 ag = ag->ag_fine; 531 } 532 533 /* Empty the target slot 534 */ 535 if (ag != 0 && ag->ag_mask == mask) { 536 ag_flush(ag->ag_dst_h, ag->ag_mask, out); 537 ag = (ag_cors == 0) ? ag_corsest : ag_cors->ag_fine; 538 } 539 540 #ifdef DEBUG_AG 541 (void)fflush(stderr); 542 if (ag == 0 && ag_cors != ag_finest) 543 abort(); 544 if (ag_cors == 0 && ag != ag_corsest) 545 abort(); 546 if (ag != 0 && ag->ag_cors != ag_cors) 547 abort(); 548 if (ag_cors != 0 && ag_cors->ag_fine != ag) 549 abort(); 550 CHECK_AG(); 551 #endif 552 553 /* Save the new route on the end of the table. 554 */ 555 nag = ag_avail; 556 ag_avail = nag->ag_fine; 557 558 nag->ag_dst_h = dst; 559 nag->ag_mask = mask; 560 nag->ag_gate = gate; 561 nag->ag_nhop = nhop; 562 nag->ag_metric = metric; 563 nag->ag_pref = pref; 564 nag->ag_tag = tag; 565 nag->ag_state = state; 566 nag->ag_seqno = seqno; 567 568 nag->ag_fine = ag; 569 if (ag != 0) 570 ag->ag_cors = nag; 571 else 572 ag_finest = nag; 573 nag->ag_cors = ag_cors; 574 if (ag_cors == 0) 575 ag_corsest = nag; 576 else 577 ag_cors->ag_fine = nag; 578 CHECK_AG(); 579 } 580 581 582 static char * 583 rtm_type_name(u_char type) 584 { 585 static char *rtm_types[] = { 586 "RTM_ADD", 587 "RTM_DELETE", 588 "RTM_CHANGE", 589 "RTM_GET", 590 "RTM_LOSING", 591 "RTM_REDIRECT", 592 "RTM_MISS", 593 "RTM_LOCK", 594 "RTM_OLDADD", 595 "RTM_OLDDEL", 596 "RTM_RESOLVE", 597 "RTM_NEWADDR", 598 "RTM_DELADDR", 599 "RTM_IFINFO" 600 }; 601 static char name0[10]; 602 603 604 if (type > sizeof(rtm_types)/sizeof(rtm_types[0]) 605 || type == 0) { 606 sprintf(name0, "RTM type %#x", type); 607 return name0; 608 } else { 609 return rtm_types[type-1]; 610 } 611 } 612 613 614 /* Trim a mask in a sockaddr 615 * Produce the index of the first zero byte. 616 * i.e. Produce a index of 4 for an mask of 0. (default route) 617 */ 618 void 619 #ifdef _HAVE_SIN_LEN 620 masktrim(struct sockaddr_in *ap) 621 #else 622 masktrim(struct sockaddr_in_new *ap) 623 #endif 624 { 625 register char *cp; 626 627 ap->sin_port = 0xffff; /* buffer zone for default route */ 628 cp = (char *)(&ap->sin_addr.s_addr+1); 629 while (*--cp == 0) 630 continue; 631 /*ap->sin_port = 0x0;*/ /* may not be needed (who cares?)*/ 632 ap->sin_len = cp - (char*)ap + 1; 633 } 634 635 636 /* Tell the kernel to add, delete or change a route 637 */ 638 static void 639 rtioctl(int action, /* RTM_DELETE, etc */ 640 naddr dst, 641 naddr gate, 642 naddr mask, 643 int metric, 644 int flags) 645 { 646 struct { 647 struct rt_msghdr w_rtm; 648 struct sockaddr_in w_dst; 649 struct sockaddr_in w_gate; 650 #ifdef _HAVE_SA_LEN 651 struct sockaddr_in w_mask; 652 #else 653 struct sockaddr_in_new w_mask; 654 #endif 655 } w; 656 long cc; 657 658 again: 659 bzero(&w, sizeof(w)); 660 w.w_rtm.rtm_msglen = sizeof(w); 661 w.w_rtm.rtm_version = RTM_VERSION; 662 w.w_rtm.rtm_type = action; 663 w.w_rtm.rtm_flags = flags; 664 w.w_rtm.rtm_seq = ++rt_sock_seqno; 665 w.w_rtm.rtm_addrs = RTA_DST|RTA_GATEWAY; 666 if (metric != 0) { 667 w.w_rtm.rtm_rmx.rmx_hopcount = metric; 668 w.w_rtm.rtm_inits |= RTV_HOPCOUNT; 669 } 670 w.w_dst.sin_family = AF_INET; 671 w.w_dst.sin_addr.s_addr = dst; 672 w.w_gate.sin_family = AF_INET; 673 w.w_gate.sin_addr.s_addr = gate; 674 #ifdef _HAVE_SA_LEN 675 w.w_dst.sin_len = sizeof(w.w_dst); 676 w.w_gate.sin_len = sizeof(w.w_gate); 677 #endif 678 if (mask == HOST_MASK) { 679 w.w_rtm.rtm_flags |= RTF_HOST; 680 w.w_rtm.rtm_msglen -= sizeof(w.w_mask); 681 } else { 682 w.w_rtm.rtm_addrs |= RTA_NETMASK; 683 w.w_mask.sin_addr.s_addr = htonl(mask); 684 #ifdef _HAVE_SA_LEN 685 masktrim(&w.w_mask); 686 w.w_rtm.rtm_msglen -= (sizeof(w.w_mask) - w.w_mask.sin_len); 687 #endif 688 } 689 690 if (TRACEKERNEL) 691 trace_kernel("write kernel %s %s->%s metric=%d flags=%#x\n", 692 rtm_type_name(action), 693 addrname(dst, mask, 0), naddr_ntoa(gate), 694 metric, flags); 695 696 #ifndef NO_INSTALL 697 cc = write(rt_sock, &w, w.w_rtm.rtm_msglen); 698 if (cc == w.w_rtm.rtm_msglen) 699 return; 700 if (cc < 0) { 701 if (errno == ESRCH 702 && (action == RTM_CHANGE || action == RTM_DELETE)) { 703 trace_act("route to %s disappeared before %s", 704 addrname(dst, mask, 0), 705 rtm_type_name(action)); 706 if (action == RTM_CHANGE) { 707 action = RTM_ADD; 708 goto again; 709 } 710 return; 711 } 712 msglog("write(rt_sock) %s %s --> %s: %s", 713 rtm_type_name(action), 714 addrname(dst, mask, 0), naddr_ntoa(gate), 715 strerror(errno)); 716 } else { 717 msglog("write(rt_sock) wrote %d instead of %d", 718 cc, w.w_rtm.rtm_msglen); 719 } 720 #endif 721 } 722 723 724 #define KHASH_SIZE 71 /* should be prime */ 725 #define KHASH(a,m) khash_bins[((a) ^ (m)) % KHASH_SIZE] 726 static struct khash { 727 struct khash *k_next; 728 naddr k_dst; 729 naddr k_mask; 730 naddr k_gate; 731 short k_metric; 732 u_short k_state; 733 #define KS_NEW 0x001 734 #define KS_DELETE 0x002 735 #define KS_ADD 0x004 /* add to the kernel */ 736 #define KS_CHANGE 0x008 /* tell kernel to change the route */ 737 #define KS_DEL_ADD 0x010 /* delete & add to change the kernel */ 738 #define KS_STATIC 0x020 /* Static flag in kernel */ 739 #define KS_GATEWAY 0x040 /* G flag in kernel */ 740 #define KS_DYNAMIC 0x080 /* result of redirect */ 741 #define KS_DELETED 0x100 /* already deleted */ 742 time_t k_keep; 743 #define K_KEEP_LIM 30 744 time_t k_redirect_time; /* when redirected route 1st seen */ 745 } *khash_bins[KHASH_SIZE]; 746 747 748 static struct khash* 749 kern_find(naddr dst, naddr mask, struct khash ***ppk) 750 { 751 struct khash *k, **pk; 752 753 for (pk = &KHASH(dst,mask); (k = *pk) != 0; pk = &k->k_next) { 754 if (k->k_dst == dst && k->k_mask == mask) 755 break; 756 } 757 if (ppk != 0) 758 *ppk = pk; 759 return k; 760 } 761 762 763 static struct khash* 764 kern_add(naddr dst, naddr mask) 765 { 766 struct khash *k, **pk; 767 768 k = kern_find(dst, mask, &pk); 769 if (k != 0) 770 return k; 771 772 k = (struct khash *)malloc(sizeof(*k)); 773 774 bzero(k, sizeof(*k)); 775 k->k_dst = dst; 776 k->k_mask = mask; 777 k->k_state = KS_NEW; 778 k->k_keep = now.tv_sec; 779 *pk = k; 780 781 return k; 782 } 783 784 785 /* If a kernel route has a non-zero metric, check that it is still in the 786 * daemon table, and not deleted by interfaces coming and going. 787 */ 788 static void 789 kern_check_static(struct khash *k, 790 struct interface *ifp) 791 { 792 struct rt_entry *rt; 793 naddr int_addr; 794 795 if (k->k_metric == 0) 796 return; 797 798 int_addr = (ifp != 0) ? ifp->int_addr : loopaddr; 799 800 rt = rtget(k->k_dst, k->k_mask); 801 if (rt != 0) { 802 if (!(rt->rt_state & RS_STATIC)) 803 rtchange(rt, rt->rt_state | RS_STATIC, 804 k->k_gate, int_addr, 805 k->k_metric, 0, ifp, now.tv_sec, 0); 806 } else { 807 rtadd(k->k_dst, k->k_mask, k->k_gate, int_addr, 808 k->k_metric, 0, RS_STATIC, ifp); 809 } 810 } 811 812 813 /* add a route the kernel told us 814 */ 815 static void 816 rtm_add(struct rt_msghdr *rtm, 817 struct rt_addrinfo *info, 818 time_t keep) 819 { 820 struct khash *k; 821 struct interface *ifp; 822 naddr mask; 823 824 825 if (rtm->rtm_flags & RTF_HOST) { 826 mask = HOST_MASK; 827 } else if (INFO_MASK(info) != 0) { 828 mask = ntohl(S_ADDR(INFO_MASK(info))); 829 } else { 830 msglog("ignore %s without mask", rtm_type_name(rtm->rtm_type)); 831 return; 832 } 833 834 if (INFO_GATE(info) == 0 835 || INFO_GATE(info)->sa_family != AF_INET) { 836 msglog("ignore %s without gateway", 837 rtm_type_name(rtm->rtm_type)); 838 return; 839 } 840 841 k = kern_add(S_ADDR(INFO_DST(info)), mask); 842 if (k->k_state & KS_NEW) 843 k->k_keep = now.tv_sec+keep; 844 k->k_gate = S_ADDR(INFO_GATE(info)); 845 k->k_metric = rtm->rtm_rmx.rmx_hopcount; 846 if (k->k_metric < 0) 847 k->k_metric = 0; 848 else if (k->k_metric > HOPCNT_INFINITY) 849 k->k_metric = HOPCNT_INFINITY; 850 k->k_state &= ~(KS_DELETED | KS_GATEWAY | KS_STATIC | KS_NEW); 851 if (rtm->rtm_flags & RTF_GATEWAY) 852 k->k_state |= KS_GATEWAY; 853 if (rtm->rtm_flags & RTF_STATIC) 854 k->k_state |= KS_STATIC; 855 856 if (0 != (rtm->rtm_flags & (RTF_DYNAMIC | RTF_MODIFIED))) { 857 if (INFO_AUTHOR(info) != 0 858 && INFO_AUTHOR(info)->sa_family == AF_INET) 859 ifp = iflookup(S_ADDR(INFO_AUTHOR(info))); 860 else 861 ifp = 0; 862 if (supplier 863 && (ifp == 0 || !(ifp->int_state & IS_REDIRECT_OK))) { 864 /* Routers are not supposed to listen to redirects, 865 * so delete it if it came via an unknown interface 866 * or the interface does not have special permission. 867 */ 868 k->k_state &= ~KS_DYNAMIC; 869 k->k_state |= KS_DELETE; 870 LIM_SEC(need_kern, 0); 871 trace_act("mark for deletion redirected %s --> %s" 872 " via %s", 873 addrname(k->k_dst, k->k_mask, 0), 874 naddr_ntoa(k->k_gate), 875 ifp ? ifp->int_name : "unknown interface"); 876 } else { 877 k->k_state |= KS_DYNAMIC; 878 k->k_redirect_time = now.tv_sec; 879 trace_act("accept redirected %s --> %s via %s", 880 addrname(k->k_dst, k->k_mask, 0), 881 naddr_ntoa(k->k_gate), 882 ifp ? ifp->int_name : "unknown interface"); 883 } 884 return; 885 } 886 887 /* If it is not a static route, quit until the next comparison 888 * between the kernel and daemon tables, when it will be deleted. 889 */ 890 if (!(k->k_state & KS_STATIC)) { 891 k->k_state |= KS_DELETE; 892 LIM_SEC(need_kern, k->k_keep); 893 return; 894 } 895 896 /* Put static routes with real metrics into the daemon table so 897 * they can be advertised. 898 * 899 * Find the interface toward the gateway. 900 */ 901 ifp = iflookup(k->k_gate); 902 if (ifp == 0) 903 msglog("static route %s --> %s impossibly lacks ifp", 904 addrname(S_ADDR(INFO_DST(info)), mask, 0), 905 naddr_ntoa(k->k_gate)); 906 907 kern_check_static(k, ifp); 908 } 909 910 911 /* deal with packet loss 912 */ 913 static void 914 rtm_lose(struct rt_msghdr *rtm, 915 struct rt_addrinfo *info) 916 { 917 if (INFO_GATE(info) == 0 918 || INFO_GATE(info)->sa_family != AF_INET) { 919 trace_act("ignore %s without gateway", 920 rtm_type_name(rtm->rtm_type)); 921 return; 922 } 923 924 if (!supplier) 925 rdisc_age(S_ADDR(INFO_GATE(info))); 926 927 age(S_ADDR(INFO_GATE(info))); 928 } 929 930 931 /* Clean the kernel table by copying it to the daemon image. 932 * Eventually the daemon will delete any extra routes. 933 */ 934 void 935 flush_kern(void) 936 { 937 size_t needed; 938 int mib[6]; 939 char *buf, *next, *lim; 940 struct rt_msghdr *rtm; 941 struct interface *ifp; 942 static struct sockaddr_in gate_sa; 943 struct rt_addrinfo info; 944 945 946 mib[0] = CTL_NET; 947 mib[1] = PF_ROUTE; 948 mib[2] = 0; /* protocol */ 949 mib[3] = 0; /* wildcard address family */ 950 mib[4] = NET_RT_DUMP; 951 mib[5] = 0; /* no flags */ 952 if (sysctl(mib, 6, 0, &needed, 0, 0) < 0) { 953 DBGERR(1,"RT_DUMP-sysctl-estimate"); 954 return; 955 } 956 buf = malloc(needed); 957 if (sysctl(mib, 6, buf, &needed, 0, 0) < 0) 958 BADERR(1,"RT_DUMP"); 959 lim = buf + needed; 960 for (next = buf; next < lim; next += rtm->rtm_msglen) { 961 rtm = (struct rt_msghdr *)next; 962 963 rt_xaddrs(&info, 964 (struct sockaddr *)(rtm+1), 965 (struct sockaddr *)(next + rtm->rtm_msglen), 966 rtm->rtm_addrs); 967 968 if (INFO_DST(&info) == 0 969 || INFO_DST(&info)->sa_family != AF_INET) 970 continue; 971 972 /* ignore ARP table entries on systems with a merged route 973 * and ARP table. 974 */ 975 if (rtm->rtm_flags & RTF_LLINFO) 976 continue; 977 978 if (INFO_GATE(&info) == 0) 979 continue; 980 if (INFO_GATE(&info)->sa_family != AF_INET) { 981 if (INFO_GATE(&info)->sa_family != AF_LINK) 982 continue; 983 ifp = ifwithindex(((struct sockaddr_dl *) 984 INFO_GATE(&info))->sdl_index); 985 if (ifp == 0) 986 continue; 987 if ((ifp->int_if_flags & IFF_POINTOPOINT) 988 || S_ADDR(INFO_DST(&info)) == ifp->int_addr) 989 gate_sa.sin_addr.s_addr = ifp->int_addr; 990 else 991 gate_sa.sin_addr.s_addr = htonl(ifp->int_net); 992 #ifdef _HAVE_SA_LEN 993 gate_sa.sin_len = sizeof(gate_sa); 994 #endif 995 gate_sa.sin_family = AF_INET; 996 INFO_GATE(&info) = (struct sockaddr *)&gate_sa; 997 } 998 999 /* ignore multicast addresses 1000 */ 1001 if (IN_MULTICAST(ntohl(S_ADDR(INFO_DST(&info))))) 1002 continue; 1003 1004 /* Note static routes and interface routes, and also 1005 * preload the image of the kernel table so that 1006 * we can later clean it, as well as avoid making 1007 * unneeded changes. Keep the old kernel routes for a 1008 * few seconds to allow a RIP or router-discovery 1009 * response to be heard. 1010 */ 1011 rtm_add(rtm,&info,MIN_WAITTIME); 1012 } 1013 free(buf); 1014 } 1015 1016 1017 /* Listen to announcements from the kernel 1018 */ 1019 void 1020 read_rt(void) 1021 { 1022 long cc; 1023 struct interface *ifp; 1024 naddr mask; 1025 union { 1026 struct { 1027 struct rt_msghdr rtm; 1028 struct sockaddr addrs[RTAX_MAX]; 1029 } r; 1030 struct if_msghdr ifm; 1031 } m; 1032 char str[100], *strp; 1033 struct rt_addrinfo info; 1034 1035 1036 for (;;) { 1037 cc = read(rt_sock, &m, sizeof(m)); 1038 if (cc <= 0) { 1039 if (cc < 0 && errno != EWOULDBLOCK) 1040 LOGERR("read(rt_sock)"); 1041 return; 1042 } 1043 1044 if (m.r.rtm.rtm_version != RTM_VERSION) { 1045 msglog("bogus routing message version %d", 1046 m.r.rtm.rtm_version); 1047 continue; 1048 } 1049 1050 /* Ignore our own results. 1051 */ 1052 if (m.r.rtm.rtm_type <= RTM_CHANGE 1053 && m.r.rtm.rtm_pid == mypid) { 1054 static int complained = 0; 1055 if (!complained) { 1056 msglog("receiving our own change messages"); 1057 complained = 1; 1058 } 1059 continue; 1060 } 1061 1062 if (m.r.rtm.rtm_type == RTM_IFINFO 1063 || m.r.rtm.rtm_type == RTM_NEWADDR 1064 || m.r.rtm.rtm_type == RTM_DELADDR) { 1065 ifp = ifwithindex(m.ifm.ifm_index); 1066 if (ifp == 0) 1067 trace_act("note %s with flags %#x" 1068 " for index #%d", 1069 rtm_type_name(m.r.rtm.rtm_type), 1070 m.ifm.ifm_flags, 1071 m.ifm.ifm_index); 1072 else 1073 trace_act("note %s with flags %#x for %s", 1074 rtm_type_name(m.r.rtm.rtm_type), 1075 m.ifm.ifm_flags, 1076 ifp->int_name); 1077 1078 /* After being informed of a change to an interface, 1079 * check them all now if the check would otherwise 1080 * be a long time from now, if the interface is 1081 * not known, or if the interface has been turned 1082 * off or on. 1083 */ 1084 if (ifinit_timer.tv_sec-now.tv_sec>=CHECK_BAD_INTERVAL 1085 || ifp == 0 1086 || ((ifp->int_if_flags ^ m.ifm.ifm_flags) 1087 & IFF_UP_RUNNING) != 0) 1088 ifinit_timer.tv_sec = now.tv_sec; 1089 continue; 1090 } 1091 1092 strcpy(str, rtm_type_name(m.r.rtm.rtm_type)); 1093 strp = &str[strlen(str)]; 1094 if (m.r.rtm.rtm_type <= RTM_CHANGE) 1095 strp += sprintf(strp," from pid %d",m.r.rtm.rtm_pid); 1096 1097 rt_xaddrs(&info, m.r.addrs, &m.r.addrs[RTAX_MAX], 1098 m.r.rtm.rtm_addrs); 1099 1100 if (INFO_DST(&info) == 0) { 1101 trace_act("ignore %s without dst", str); 1102 continue; 1103 } 1104 1105 if (INFO_DST(&info)->sa_family != AF_INET) { 1106 trace_act("ignore %s for AF %d", str, 1107 INFO_DST(&info)->sa_family); 1108 continue; 1109 } 1110 1111 mask = ((INFO_MASK(&info) != 0) 1112 ? ntohl(S_ADDR(INFO_MASK(&info))) 1113 : (m.r.rtm.rtm_flags & RTF_HOST) 1114 ? HOST_MASK 1115 : std_mask(S_ADDR(INFO_DST(&info)))); 1116 1117 strp += sprintf(strp, ": %s", 1118 addrname(S_ADDR(INFO_DST(&info)), mask, 0)); 1119 1120 if (IN_MULTICAST(ntohl(S_ADDR(INFO_DST(&info))))) { 1121 trace_act("ignore multicast %s", str); 1122 continue; 1123 } 1124 1125 if (INFO_GATE(&info) != 0 1126 && INFO_GATE(&info)->sa_family == AF_INET) 1127 strp += sprintf(strp, " --> %s", 1128 saddr_ntoa(INFO_GATE(&info))); 1129 1130 if (INFO_AUTHOR(&info) != 0) 1131 strp += sprintf(strp, " by authority of %s", 1132 saddr_ntoa(INFO_AUTHOR(&info))); 1133 1134 switch (m.r.rtm.rtm_type) { 1135 case RTM_ADD: 1136 case RTM_CHANGE: 1137 case RTM_REDIRECT: 1138 if (m.r.rtm.rtm_errno != 0) { 1139 trace_act("ignore %s with \"%s\" error", 1140 str, strerror(m.r.rtm.rtm_errno)); 1141 } else { 1142 trace_act("%s", str); 1143 rtm_add(&m.r.rtm,&info,0); 1144 } 1145 break; 1146 1147 case RTM_DELETE: 1148 if (m.r.rtm.rtm_errno != 0) { 1149 trace_act("ignore %s with \"%s\" error", 1150 str, strerror(m.r.rtm.rtm_errno)); 1151 } else { 1152 trace_act("%s", str); 1153 del_static(S_ADDR(INFO_DST(&info)), mask, 1); 1154 } 1155 break; 1156 1157 case RTM_LOSING: 1158 trace_act("%s", str); 1159 rtm_lose(&m.r.rtm,&info); 1160 break; 1161 1162 default: 1163 trace_act("ignore %s", str); 1164 break; 1165 } 1166 } 1167 } 1168 1169 1170 /* after aggregating, note routes that belong in the kernel 1171 */ 1172 static void 1173 kern_out(struct ag_info *ag) 1174 { 1175 struct khash *k; 1176 1177 1178 /* Do not install bad routes if they are not already present. 1179 * This includes routes that had RS_NET_SYN for interfaces that 1180 * recently died. 1181 */ 1182 if (ag->ag_metric == HOPCNT_INFINITY) { 1183 k = kern_find(htonl(ag->ag_dst_h), ag->ag_mask, 0); 1184 if (k == 0) 1185 return; 1186 } else { 1187 k = kern_add(htonl(ag->ag_dst_h), ag->ag_mask); 1188 } 1189 1190 if (k->k_state & KS_NEW) { 1191 /* will need to add new entry to the kernel table */ 1192 k->k_state = KS_ADD; 1193 if (ag->ag_state & AGS_GATEWAY) 1194 k->k_state |= KS_GATEWAY; 1195 k->k_gate = ag->ag_gate; 1196 k->k_metric = ag->ag_metric; 1197 return; 1198 } 1199 1200 if (k->k_state & KS_STATIC) 1201 return; 1202 1203 /* modify existing kernel entry if necessary */ 1204 if (k->k_gate != ag->ag_gate 1205 || k->k_metric != ag->ag_metric) { 1206 k->k_gate = ag->ag_gate; 1207 k->k_metric = ag->ag_metric; 1208 k->k_state |= KS_CHANGE; 1209 } 1210 1211 if (k->k_state & KS_DYNAMIC) { 1212 k->k_state &= ~KS_DYNAMIC; 1213 k->k_state |= (KS_ADD | KS_DEL_ADD); 1214 } 1215 1216 if ((k->k_state & KS_GATEWAY) 1217 && !(ag->ag_state & AGS_GATEWAY)) { 1218 k->k_state &= ~KS_GATEWAY; 1219 k->k_state |= (KS_ADD | KS_DEL_ADD); 1220 } else if (!(k->k_state & KS_GATEWAY) 1221 && (ag->ag_state & AGS_GATEWAY)) { 1222 k->k_state |= KS_GATEWAY; 1223 k->k_state |= (KS_ADD | KS_DEL_ADD); 1224 } 1225 1226 /* Deleting-and-adding is necessary to change aspects of a route. 1227 * Just delete instead of deleting and then adding a bad route. 1228 * Otherwise, we want to keep the route in the kernel. 1229 */ 1230 if (k->k_metric == HOPCNT_INFINITY 1231 && (k->k_state & KS_DEL_ADD)) 1232 k->k_state |= KS_DELETE; 1233 else 1234 k->k_state &= ~KS_DELETE; 1235 #undef RT 1236 } 1237 1238 1239 /* ARGSUSED */ 1240 static int 1241 walk_kern(struct radix_node *rn, 1242 struct walkarg *w) 1243 { 1244 #define RT ((struct rt_entry *)rn) 1245 char metric, pref; 1246 u_int ags = 0; 1247 1248 1249 /* Do not install synthetic routes */ 1250 if (RT->rt_state & RS_NET_SYN) 1251 return 0; 1252 1253 if (!(RT->rt_state & RS_IF)) { 1254 ags |= (AGS_GATEWAY | AGS_SUPPRESS | AGS_PROMOTE); 1255 1256 } else { 1257 /* Do not install routes for "external" remote interfaces. 1258 */ 1259 if (RT->rt_ifp != 0 && (RT->rt_ifp->int_state & IS_EXTERNAL)) 1260 return 0; 1261 1262 ags |= AGS_IF; 1263 1264 /* If it is not an interface, or an alias for an interface, 1265 * it must be a "gateway." 1266 * 1267 * If it is a "remote" interface, it is also a "gateway" to 1268 * the kernel if is not a alias. 1269 */ 1270 if (RT->rt_ifp == 0 1271 || (RT->rt_ifp->int_state & IS_REMOTE)) 1272 ags |= (AGS_GATEWAY | AGS_SUPPRESS | AGS_PROMOTE); 1273 } 1274 1275 if (RT->rt_state & RS_RDISC) 1276 ags |= AGS_CORS_GATE; 1277 1278 /* aggregate good routes without regard to their metric */ 1279 pref = 1; 1280 metric = RT->rt_metric; 1281 if (metric == HOPCNT_INFINITY) { 1282 /* if the route is dead, so try hard to aggregate. */ 1283 pref = HOPCNT_INFINITY; 1284 ags |= (AGS_FINE_GATE | AGS_SUPPRESS); 1285 } 1286 1287 ag_check(RT->rt_dst, RT->rt_mask, RT->rt_gate, 0, 1288 metric,pref, 0, 0, ags, kern_out); 1289 return 0; 1290 #undef RT 1291 } 1292 1293 1294 /* Update the kernel table to match the daemon table. 1295 */ 1296 static void 1297 fix_kern(void) 1298 { 1299 int i, flags; 1300 struct khash *k, **pk; 1301 1302 1303 need_kern = age_timer; 1304 1305 /* Walk daemon table, updating the copy of the kernel table. 1306 */ 1307 (void)rn_walktree(rhead, walk_kern, 0); 1308 ag_flush(0,0,kern_out); 1309 1310 for (i = 0; i < KHASH_SIZE; i++) { 1311 for (pk = &khash_bins[i]; (k = *pk) != 0; ) { 1312 /* Do not touch static routes */ 1313 if (k->k_state & KS_STATIC) { 1314 kern_check_static(k,0); 1315 pk = &k->k_next; 1316 continue; 1317 } 1318 1319 /* check hold on routes deleted by the operator */ 1320 if (k->k_keep > now.tv_sec) { 1321 LIM_SEC(need_kern, k->k_keep); 1322 k->k_state |= KS_DELETE; 1323 pk = &k->k_next; 1324 continue; 1325 } 1326 1327 if ((k->k_state & (KS_DELETE | KS_DYNAMIC)) 1328 == KS_DELETE) { 1329 if (!(k->k_state & KS_DELETED)) 1330 rtioctl(RTM_DELETE, 1331 k->k_dst, k->k_gate, k->k_mask, 1332 0, 0); 1333 *pk = k->k_next; 1334 free(k); 1335 continue; 1336 } 1337 1338 if (0 != (k->k_state&(KS_ADD|KS_CHANGE|KS_DEL_ADD))) { 1339 if (k->k_state & KS_DEL_ADD) { 1340 rtioctl(RTM_DELETE, 1341 k->k_dst,k->k_gate,k->k_mask, 1342 0, 0); 1343 k->k_state &= ~KS_DYNAMIC; 1344 } 1345 1346 flags = 0; 1347 if (0 != (k->k_state&(KS_GATEWAY|KS_DYNAMIC))) 1348 flags |= RTF_GATEWAY; 1349 1350 if (k->k_state & KS_ADD) { 1351 rtioctl(RTM_ADD, 1352 k->k_dst, k->k_gate, k->k_mask, 1353 k->k_metric, flags); 1354 } else if (k->k_state & KS_CHANGE) { 1355 rtioctl(RTM_CHANGE, 1356 k->k_dst,k->k_gate,k->k_mask, 1357 k->k_metric, flags); 1358 } 1359 k->k_state &= ~(KS_ADD|KS_CHANGE|KS_DEL_ADD); 1360 } 1361 1362 /* Mark this route to be deleted in the next cycle. 1363 * This deletes routes that disappear from the 1364 * daemon table, since the normal aging code 1365 * will clear the bit for routes that have not 1366 * disappeared from the daemon table. 1367 */ 1368 k->k_state |= KS_DELETE; 1369 pk = &k->k_next; 1370 } 1371 } 1372 } 1373 1374 1375 /* Delete a static route in the image of the kernel table. 1376 */ 1377 void 1378 del_static(naddr dst, 1379 naddr mask, 1380 int gone) 1381 { 1382 struct khash *k; 1383 struct rt_entry *rt; 1384 1385 /* Just mark it in the table to be deleted next time the kernel 1386 * table is updated. 1387 * If it has already been deleted, mark it as such, and set its 1388 * keep-timer so that it will not be deleted again for a while. 1389 * This lets the operator delete a route added by the daemon 1390 * and add a replacement. 1391 */ 1392 k = kern_find(dst, mask, 0); 1393 if (k != 0) { 1394 k->k_state &= ~(KS_STATIC | KS_DYNAMIC); 1395 k->k_state |= KS_DELETE; 1396 if (gone) { 1397 k->k_state |= KS_DELETED; 1398 k->k_keep = now.tv_sec + K_KEEP_LIM; 1399 } 1400 } 1401 1402 rt = rtget(dst, mask); 1403 if (rt != 0 && (rt->rt_state & RS_STATIC)) 1404 rtbad(rt); 1405 } 1406 1407 1408 /* Delete all routes generated from ICMP Redirects that use a given gateway, 1409 * as well as old redirected routes. 1410 */ 1411 void 1412 del_redirects(naddr bad_gate, 1413 time_t old) 1414 { 1415 int i; 1416 struct khash *k; 1417 1418 1419 for (i = 0; i < KHASH_SIZE; i++) { 1420 for (k = khash_bins[i]; k != 0; k = k->k_next) { 1421 if (!(k->k_state & KS_DYNAMIC) 1422 || (k->k_state & KS_STATIC)) 1423 continue; 1424 1425 if (k->k_gate != bad_gate 1426 && k->k_redirect_time > old 1427 && !supplier) 1428 continue; 1429 1430 k->k_state |= KS_DELETE; 1431 k->k_state &= ~KS_DYNAMIC; 1432 need_kern.tv_sec = now.tv_sec; 1433 trace_act("mark redirected %s --> %s for deletion", 1434 addrname(k->k_dst, k->k_mask, 0), 1435 naddr_ntoa(k->k_gate)); 1436 } 1437 } 1438 } 1439 1440 1441 /* Start the daemon tables. 1442 */ 1443 void 1444 rtinit(void) 1445 { 1446 extern int max_keylen; 1447 int i; 1448 struct ag_info *ag; 1449 1450 /* Initialize the radix trees */ 1451 max_keylen = sizeof(struct sockaddr_in); 1452 rn_init(); 1453 rn_inithead((void**)&rhead, 32); 1454 1455 /* mark all of the slots in the table free */ 1456 ag_avail = ag_slots; 1457 for (ag = ag_slots, i = 1; i < NUM_AG_SLOTS; i++) { 1458 ag->ag_fine = ag+1; 1459 ag++; 1460 } 1461 } 1462 1463 1464 #ifdef _HAVE_SIN_LEN 1465 static struct sockaddr_in dst_sock = {sizeof(dst_sock), AF_INET}; 1466 static struct sockaddr_in mask_sock = {sizeof(mask_sock), AF_INET}; 1467 #else 1468 static struct sockaddr_in_new dst_sock = {_SIN_ADDR_SIZE, AF_INET}; 1469 static struct sockaddr_in_new mask_sock = {_SIN_ADDR_SIZE, AF_INET}; 1470 #endif 1471 1472 1473 void 1474 set_need_flash(void) 1475 { 1476 if (!need_flash) { 1477 need_flash = 1; 1478 /* Do not send the flash update immediately. Wait a little 1479 * while to hear from other routers. 1480 */ 1481 no_flash.tv_sec = now.tv_sec + MIN_WAITTIME; 1482 } 1483 } 1484 1485 1486 /* Get a particular routing table entry 1487 */ 1488 struct rt_entry * 1489 rtget(naddr dst, naddr mask) 1490 { 1491 struct rt_entry *rt; 1492 1493 dst_sock.sin_addr.s_addr = dst; 1494 mask_sock.sin_addr.s_addr = mask; 1495 masktrim(&mask_sock); 1496 rt = (struct rt_entry *)rhead->rnh_lookup(&dst_sock,&mask_sock,rhead); 1497 if (!rt 1498 || rt->rt_dst != dst 1499 || rt->rt_mask != mask) 1500 return 0; 1501 1502 return rt; 1503 } 1504 1505 1506 /* Find a route to dst as the kernel would. 1507 */ 1508 struct rt_entry * 1509 rtfind(naddr dst) 1510 { 1511 dst_sock.sin_addr.s_addr = dst; 1512 return (struct rt_entry *)rhead->rnh_matchaddr(&dst_sock, rhead); 1513 } 1514 1515 1516 /* add a route to the table 1517 */ 1518 void 1519 rtadd(naddr dst, 1520 naddr mask, 1521 naddr gate, /* forward packets here */ 1522 naddr router, /* on the authority of this router */ 1523 int metric, 1524 u_short tag, 1525 u_int state, /* rs_state for the entry */ 1526 struct interface *ifp) 1527 { 1528 struct rt_entry *rt; 1529 naddr smask; 1530 int i; 1531 struct rt_spare *rts; 1532 1533 rt = (struct rt_entry *)rtmalloc(sizeof (*rt), "rtadd"); 1534 bzero(rt, sizeof(*rt)); 1535 for (rts = rt->rt_spares, i = NUM_SPARES; i != 0; i--, rts++) 1536 rts->rts_metric = HOPCNT_INFINITY; 1537 1538 rt->rt_nodes->rn_key = (caddr_t)&rt->rt_dst_sock; 1539 rt->rt_dst = dst; 1540 rt->rt_dst_sock.sin_family = AF_INET; 1541 #ifdef _HAVE_SIN_LEN 1542 rt->rt_dst_sock.sin_len = dst_sock.sin_len; 1543 #endif 1544 if (mask != HOST_MASK) { 1545 smask = std_mask(dst); 1546 if ((smask & ~mask) == 0 && mask > smask) 1547 state |= RS_SUBNET; 1548 } 1549 mask_sock.sin_addr.s_addr = mask; 1550 masktrim(&mask_sock); 1551 rt->rt_mask = mask; 1552 rt->rt_state = state; 1553 rt->rt_gate = gate; 1554 rt->rt_router = router; 1555 rt->rt_time = now.tv_sec; 1556 rt->rt_metric = metric; 1557 rt->rt_poison_metric = HOPCNT_INFINITY; 1558 rt->rt_tag = tag; 1559 rt->rt_ifp = ifp; 1560 rt->rt_seqno = update_seqno; 1561 1562 if (++total_routes == MAX_ROUTES) 1563 msglog("have maximum (%d) routes", total_routes); 1564 if (TRACEACTIONS) 1565 trace_add_del("Add", rt); 1566 1567 need_kern.tv_sec = now.tv_sec; 1568 set_need_flash(); 1569 1570 if (0 == rhead->rnh_addaddr(&rt->rt_dst_sock, &mask_sock, 1571 rhead, rt->rt_nodes)) { 1572 /* 1573 * This will happen if RIP1 and RIP2 routeds talk to one another and 1574 * there are variable subnets. This is only good for filling up your 1575 * syslog. -jkh 1576 */ 1577 #if 0 1578 msglog("rnh_addaddr() failed for %s mask=%#x", 1579 naddr_ntoa(dst), mask); 1580 #endif 1581 } 1582 } 1583 1584 1585 /* notice a changed route 1586 */ 1587 void 1588 rtchange(struct rt_entry *rt, 1589 u_int state, /* new state bits */ 1590 naddr gate, /* now forward packets here */ 1591 naddr router, /* on the authority of this router */ 1592 int metric, /* new metric */ 1593 u_short tag, 1594 struct interface *ifp, 1595 time_t new_time, 1596 char *label) 1597 { 1598 if (rt->rt_metric != metric) { 1599 /* Fix the kernel immediately if it seems the route 1600 * has gone bad, since there may be a working route that 1601 * aggregates this route. 1602 */ 1603 if (metric == HOPCNT_INFINITY) { 1604 need_kern.tv_sec = now.tv_sec; 1605 if (new_time >= now.tv_sec - EXPIRE_TIME) 1606 new_time = now.tv_sec - EXPIRE_TIME; 1607 } 1608 rt->rt_seqno = update_seqno; 1609 set_need_flash(); 1610 } 1611 1612 if (rt->rt_gate != gate) { 1613 need_kern.tv_sec = now.tv_sec; 1614 rt->rt_seqno = update_seqno; 1615 set_need_flash(); 1616 } 1617 1618 state |= (rt->rt_state & RS_SUBNET); 1619 1620 /* Keep various things from deciding ageless routes are stale. 1621 */ 1622 if (!AGE_RT(state, ifp)) 1623 new_time = now.tv_sec; 1624 1625 if (TRACEACTIONS) 1626 trace_change(rt, state, gate, router, metric, tag, ifp, 1627 new_time, 1628 label ? label : "Chg "); 1629 1630 rt->rt_state = state; 1631 rt->rt_gate = gate; 1632 rt->rt_router = router; 1633 rt->rt_metric = metric; 1634 rt->rt_tag = tag; 1635 rt->rt_ifp = ifp; 1636 rt->rt_time = new_time; 1637 } 1638 1639 1640 /* check for a better route among the spares 1641 */ 1642 static struct rt_spare * 1643 rts_better(struct rt_entry *rt) 1644 { 1645 struct rt_spare *rts, *rts1; 1646 int i; 1647 1648 /* find the best alternative among the spares */ 1649 rts = rt->rt_spares+1; 1650 for (i = NUM_SPARES, rts1 = rts+1; i > 2; i--, rts1++) { 1651 if (BETTER_LINK(rt,rts1,rts)) 1652 rts = rts1; 1653 } 1654 1655 return rts; 1656 } 1657 1658 1659 /* switch to a backup route 1660 */ 1661 void 1662 rtswitch(struct rt_entry *rt, 1663 struct rt_spare *rts) 1664 { 1665 struct rt_spare swap; 1666 char label[10]; 1667 1668 1669 /* Do not change permanent routes */ 1670 if (0 != (rt->rt_state & (RS_MHOME | RS_STATIC | RS_RDISC 1671 | RS_NET_SYN | RS_IF))) 1672 return; 1673 1674 /* find the best alternative among the spares */ 1675 if (rts == 0) 1676 rts = rts_better(rt); 1677 1678 /* Do not bother if it is not worthwhile. 1679 */ 1680 if (!BETTER_LINK(rt, rts, rt->rt_spares)) 1681 return; 1682 1683 swap = rt->rt_spares[0]; 1684 (void)sprintf(label, "Use #%d", rts - rt->rt_spares); 1685 rtchange(rt, rt->rt_state & ~(RS_NET_SYN | RS_RDISC), 1686 rts->rts_gate, rts->rts_router, rts->rts_metric, 1687 rts->rts_tag, rts->rts_ifp, rts->rts_time, label); 1688 *rts = swap; 1689 } 1690 1691 1692 void 1693 rtdelete(struct rt_entry *rt) 1694 { 1695 struct khash *k; 1696 1697 1698 if (TRACEACTIONS) 1699 trace_add_del("Del", rt); 1700 1701 k = kern_find(rt->rt_dst, rt->rt_mask, 0); 1702 if (k != 0) { 1703 k->k_state |= KS_DELETE; 1704 need_kern.tv_sec = now.tv_sec; 1705 } 1706 1707 dst_sock.sin_addr.s_addr = rt->rt_dst; 1708 mask_sock.sin_addr.s_addr = rt->rt_mask; 1709 masktrim(&mask_sock); 1710 if (rt != (struct rt_entry *)rhead->rnh_deladdr(&dst_sock, &mask_sock, 1711 rhead)) { 1712 msglog("rnh_deladdr() failed"); 1713 } else { 1714 free(rt); 1715 total_routes--; 1716 } 1717 } 1718 1719 1720 /* Get rid of a bad route, and try to switch to a replacement. 1721 */ 1722 void 1723 rtbad(struct rt_entry *rt) 1724 { 1725 /* Poison the route */ 1726 rtchange(rt, rt->rt_state & ~(RS_IF | RS_LOCAL | RS_STATIC), 1727 rt->rt_gate, rt->rt_router, HOPCNT_INFINITY, rt->rt_tag, 1728 0, rt->rt_time, 0); 1729 1730 rtswitch(rt, 0); 1731 } 1732 1733 1734 /* Junk a RS_NET_SYN or RS_LOCAL route, 1735 * unless it is needed by another interface. 1736 */ 1737 void 1738 rtbad_sub(struct rt_entry *rt) 1739 { 1740 struct interface *ifp, *ifp1; 1741 struct intnet *intnetp; 1742 u_int state; 1743 1744 1745 ifp1 = 0; 1746 state = 0; 1747 1748 if (rt->rt_state & RS_LOCAL) { 1749 /* Is this the route through loopback for the interface? 1750 * If so, see if it is used by any other interfaces, such 1751 * as a point-to-point interface with the same local address. 1752 */ 1753 for (ifp = ifnet; ifp != 0; ifp = ifp->int_next) { 1754 /* Retain it if another interface needs it. 1755 */ 1756 if (ifp->int_addr == rt->rt_ifp->int_addr) { 1757 state |= RS_LOCAL; 1758 ifp1 = ifp; 1759 break; 1760 } 1761 } 1762 1763 } 1764 1765 if (!(state & RS_LOCAL)) { 1766 /* Retain RIPv1 logical network route if there is another 1767 * interface that justifies it. 1768 */ 1769 if (rt->rt_state & RS_NET_SYN) { 1770 for (ifp = ifnet; ifp != 0; ifp = ifp->int_next) { 1771 if ((ifp->int_state & IS_NEED_NET_SYN) 1772 && rt->rt_mask == ifp->int_std_mask 1773 && rt->rt_dst == ifp->int_std_addr) { 1774 state |= RS_NET_SYN; 1775 ifp1 = ifp; 1776 break; 1777 } 1778 } 1779 } 1780 1781 /* or if there is an authority route that needs it. */ 1782 for (intnetp = intnets; 1783 intnetp != 0; 1784 intnetp = intnetp->intnet_next) { 1785 if (intnetp->intnet_addr == rt->rt_dst 1786 && intnetp->intnet_mask == rt->rt_mask) { 1787 state |= (RS_NET_SYN | RS_NET_INT); 1788 break; 1789 } 1790 } 1791 } 1792 1793 if (ifp1 != 0 || (state & RS_NET_SYN)) { 1794 rtchange(rt, ((rt->rt_state & ~(RS_NET_SYN | RS_LOCAL)) 1795 | state), 1796 rt->rt_gate, rt->rt_router, rt->rt_metric, 1797 rt->rt_tag, ifp1, rt->rt_time, 0); 1798 } else { 1799 rtbad(rt); 1800 } 1801 } 1802 1803 1804 /* Called while walking the table looking for sick interfaces 1805 * or after a time change. 1806 */ 1807 /* ARGSUSED */ 1808 int 1809 walk_bad(struct radix_node *rn, 1810 struct walkarg *w) 1811 { 1812 #define RT ((struct rt_entry *)rn) 1813 struct rt_spare *rts; 1814 int i; 1815 time_t new_time; 1816 1817 1818 /* fix any spare routes through the interface 1819 */ 1820 rts = RT->rt_spares; 1821 for (i = NUM_SPARES; i != 1; i--) { 1822 rts++; 1823 1824 if (rts->rts_ifp != 0 1825 && (rts->rts_ifp->int_state & IS_BROKE)) { 1826 /* mark the spare route to be deleted immediately */ 1827 new_time = rts->rts_time; 1828 if (new_time >= now_garbage) 1829 new_time = now_garbage-1; 1830 trace_upslot(RT, rts, rts->rts_gate, 1831 rts->rts_router, 0, 1832 HOPCNT_INFINITY, rts->rts_tag, 1833 new_time); 1834 rts->rts_ifp = 0; 1835 rts->rts_metric = HOPCNT_INFINITY; 1836 rts->rts_time = new_time; 1837 } 1838 } 1839 1840 /* Deal with the main route 1841 */ 1842 /* finished if it has been handled before or if its interface is ok 1843 */ 1844 if (RT->rt_ifp == 0 || !(RT->rt_ifp->int_state & IS_BROKE)) 1845 return 0; 1846 1847 /* Bad routes for other than interfaces are easy. 1848 */ 1849 if (0 == (RT->rt_state & (RS_IF | RS_NET_SYN | RS_LOCAL))) { 1850 rtbad(RT); 1851 return 0; 1852 } 1853 1854 rtbad_sub(RT); 1855 return 0; 1856 #undef RT 1857 } 1858 1859 1860 /* Check the age of an individual route. 1861 */ 1862 /* ARGSUSED */ 1863 static int 1864 walk_age(struct radix_node *rn, 1865 struct walkarg *w) 1866 { 1867 #define RT ((struct rt_entry *)rn) 1868 struct interface *ifp; 1869 struct rt_spare *rts; 1870 int i; 1871 1872 1873 /* age all of the spare routes, including the primary route 1874 * currently in use 1875 */ 1876 rts = RT->rt_spares; 1877 for (i = NUM_SPARES; i != 0; i--, rts++) { 1878 1879 ifp = rts->rts_ifp; 1880 if (i == NUM_SPARES) { 1881 if (!AGE_RT(RT->rt_state, ifp)) { 1882 /* Keep various things from deciding ageless 1883 * routes are stale 1884 */ 1885 rts->rts_time = now.tv_sec; 1886 continue; 1887 } 1888 1889 /* forget RIP routes after RIP has been turned off. 1890 */ 1891 if (rip_sock < 0) { 1892 rtdelete(RT); 1893 return 0; 1894 } 1895 } 1896 1897 /* age failing routes 1898 */ 1899 if (age_bad_gate == rts->rts_gate 1900 && rts->rts_time >= now_stale) { 1901 rts->rts_time -= SUPPLY_INTERVAL; 1902 } 1903 1904 /* trash the spare routes when they go bad */ 1905 if (rts->rts_metric < HOPCNT_INFINITY 1906 && now_garbage > rts->rts_time) { 1907 trace_upslot(RT, rts, rts->rts_gate, 1908 rts->rts_router, rts->rts_ifp, 1909 HOPCNT_INFINITY, rts->rts_tag, 1910 rts->rts_time); 1911 rts->rts_metric = HOPCNT_INFINITY; 1912 } 1913 } 1914 1915 1916 /* finished if the active route is still fresh */ 1917 if (now_stale <= RT->rt_time) 1918 return 0; 1919 1920 /* try to switch to an alternative */ 1921 rtswitch(RT, 0); 1922 1923 /* Delete a dead route after it has been publically mourned. */ 1924 if (now_garbage > RT->rt_time) { 1925 rtdelete(RT); 1926 return 0; 1927 } 1928 1929 /* Start poisoning a bad route before deleting it. */ 1930 if (now.tv_sec - RT->rt_time > EXPIRE_TIME) 1931 rtchange(RT, RT->rt_state, RT->rt_gate, RT->rt_router, 1932 HOPCNT_INFINITY, RT->rt_tag, RT->rt_ifp, 1933 RT->rt_time, 0); 1934 return 0; 1935 } 1936 1937 1938 /* Watch for dead routes and interfaces. 1939 */ 1940 void 1941 age(naddr bad_gate) 1942 { 1943 struct interface *ifp; 1944 int need_query = 0; 1945 1946 /* If not listening to RIP, there is no need to age the routes in 1947 * the table. 1948 */ 1949 age_timer.tv_sec = (now.tv_sec 1950 + ((rip_sock < 0) ? NEVER : SUPPLY_INTERVAL)); 1951 1952 /* Check for dead IS_REMOTE interfaces by timing their 1953 * transmissions. 1954 */ 1955 for (ifp = ifnet; ifp; ifp = ifp->int_next) { 1956 if (!(ifp->int_state & IS_REMOTE)) 1957 continue; 1958 1959 /* ignore unreachable remote interfaces */ 1960 if (!check_remote(ifp)) 1961 continue; 1962 /* Restore remote interface that has become reachable 1963 */ 1964 if (ifp->int_state & IS_BROKE) 1965 if_ok(ifp, "remote "); 1966 1967 if (ifp->int_act_time != NEVER 1968 && now.tv_sec - ifp->int_act_time > EXPIRE_TIME) { 1969 msglog("remote interface %s to %s timed out after" 1970 " %d:%d", 1971 ifp->int_name, 1972 naddr_ntoa(ifp->int_dstaddr), 1973 (now.tv_sec - ifp->int_act_time)/60, 1974 (now.tv_sec - ifp->int_act_time)%60); 1975 if_sick(ifp); 1976 } 1977 1978 /* If we have not heard from the other router 1979 * recently, ask it. 1980 */ 1981 if (now.tv_sec >= ifp->int_query_time) { 1982 ifp->int_query_time = NEVER; 1983 need_query = 1; 1984 } 1985 } 1986 1987 /* Age routes. */ 1988 age_bad_gate = bad_gate; 1989 (void)rn_walktree(rhead, walk_age, 0); 1990 1991 /* Update the kernel routing table. */ 1992 fix_kern(); 1993 1994 /* poke reticent remote gateways */ 1995 if (need_query) 1996 rip_query(); 1997 } 1998