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 #ifndef lint 35 #if 0 36 static char sccsid[] = "@(#)tables.c 8.1 (Berkeley) 6/5/93"; 37 #endif 38 static const char rcsid[] = 39 "$Id$"; 40 #endif /* not lint */ 41 42 #include "defs.h" 43 44 static struct rt_spare *rts_better(struct rt_entry *); 45 46 struct radix_node_head *rhead; /* root of the radix tree */ 47 48 int need_flash = 1; /* flash update needed 49 * start =1 to suppress the 1st 50 */ 51 52 struct timeval age_timer; /* next check of old routes */ 53 struct timeval need_kern = { /* need to update kernel table */ 54 EPOCH+MIN_WAITTIME-1 55 }; 56 57 int stopint; 58 59 int total_routes; 60 61 /* zap any old routes through this gateway */ 62 naddr age_bad_gate; 63 64 65 /* It is desirable to "aggregate" routes, to combine differing routes of 66 * the same metric and next hop into a common route with a smaller netmask 67 * or to suppress redundant routes, routes that add no information to 68 * routes with smaller netmasks. 69 * 70 * A route is redundant if and only if any and all routes with smaller 71 * but matching netmasks and nets are the same. Since routes are 72 * kept sorted in the radix tree, redundant routes always come second. 73 * 74 * There are two kinds of aggregations. First, two routes of the same bit 75 * mask and differing only in the least significant bit of the network 76 * number can be combined into a single route with a coarser mask. 77 * 78 * Second, a route can be suppressed in favor of another route with a more 79 * coarse mask provided no incompatible routes with intermediate masks 80 * are present. The second kind of aggregation involves suppressing routes. 81 * A route must not be suppressed if an incompatible route exists with 82 * an intermediate mask, since the suppressed route would be covered 83 * by the intermediate. 84 * 85 * This code relies on the radix tree walk encountering routes 86 * sorted first by address, with the smallest address first. 87 */ 88 89 struct ag_info ag_slots[NUM_AG_SLOTS], *ag_avail, *ag_corsest, *ag_finest; 90 91 /* #define DEBUG_AG */ 92 #ifdef DEBUG_AG 93 #define CHECK_AG() {int acnt = 0; struct ag_info *cag; \ 94 for (cag = ag_avail; cag != 0; cag = cag->ag_fine) \ 95 acnt++; \ 96 for (cag = ag_corsest; cag != 0; cag = cag->ag_fine) \ 97 acnt++; \ 98 if (acnt != NUM_AG_SLOTS) { \ 99 (void)fflush(stderr); \ 100 abort(); \ 101 } \ 102 } 103 #else 104 #define CHECK_AG() 105 #endif 106 107 108 /* Output the contents of an aggregation table slot. 109 * This function must always be immediately followed with the deletion 110 * of the target slot. 111 */ 112 static void 113 ag_out(struct ag_info *ag, 114 void (*out)(struct ag_info *)) 115 { 116 struct ag_info *ag_cors; 117 naddr bit; 118 119 120 /* If we output both the even and odd twins, then the immediate parent, 121 * if it is present, is redundant, unless the parent manages to 122 * aggregate into something coarser. 123 * On successive calls, this code detects the even and odd twins, 124 * and marks the parent. 125 * 126 * Note that the order in which the radix tree code emits routes 127 * ensures that the twins are seen before the parent is emitted. 128 */ 129 ag_cors = ag->ag_cors; 130 if (ag_cors != 0 131 && ag_cors->ag_mask == ag->ag_mask<<1 132 && ag_cors->ag_dst_h == (ag->ag_dst_h & ag_cors->ag_mask)) { 133 ag_cors->ag_state |= ((ag_cors->ag_dst_h == ag->ag_dst_h) 134 ? AGS_REDUN0 135 : AGS_REDUN1); 136 } 137 138 /* Skip it if this route is itself redundant. 139 * 140 * It is ok to change the contents of the slot here, since it is 141 * always deleted next. 142 */ 143 if (ag->ag_state & AGS_REDUN0) { 144 if (ag->ag_state & AGS_REDUN1) 145 return; 146 bit = (-ag->ag_mask) >> 1; 147 ag->ag_dst_h |= bit; 148 ag->ag_mask |= bit; 149 150 } else if (ag->ag_state & AGS_REDUN1) { 151 bit = (-ag->ag_mask) >> 1; 152 ag->ag_mask |= bit; 153 } 154 out(ag); 155 } 156 157 158 static void 159 ag_del(struct ag_info *ag) 160 { 161 CHECK_AG(); 162 163 if (ag->ag_cors == 0) 164 ag_corsest = ag->ag_fine; 165 else 166 ag->ag_cors->ag_fine = ag->ag_fine; 167 168 if (ag->ag_fine == 0) 169 ag_finest = ag->ag_cors; 170 else 171 ag->ag_fine->ag_cors = ag->ag_cors; 172 173 ag->ag_fine = ag_avail; 174 ag_avail = ag; 175 176 CHECK_AG(); 177 } 178 179 180 /* Flush routes waiting for aggregation. 181 * This must not suppress a route unless it is known that among all 182 * routes with coarser masks that match it, the one with the longest 183 * mask is appropriate. This is ensured by scanning the routes 184 * in lexical order, and with the most restrictive mask first 185 * among routes to the same destination. 186 */ 187 void 188 ag_flush(naddr lim_dst_h, /* flush routes to here */ 189 naddr lim_mask, /* matching this mask */ 190 void (*out)(struct ag_info *)) 191 { 192 struct ag_info *ag, *ag_cors; 193 naddr dst_h; 194 195 196 for (ag = ag_finest; 197 ag != 0 && ag->ag_mask >= lim_mask; 198 ag = ag_cors) { 199 ag_cors = ag->ag_cors; 200 201 /* work on only the specified routes */ 202 dst_h = ag->ag_dst_h; 203 if ((dst_h & lim_mask) != lim_dst_h) 204 continue; 205 206 if (!(ag->ag_state & AGS_SUPPRESS)) 207 ag_out(ag, out); 208 209 else for ( ; ; ag_cors = ag_cors->ag_cors) { 210 /* Look for a route that can suppress the 211 * current route */ 212 if (ag_cors == 0) { 213 /* failed, so output it and look for 214 * another route to work on 215 */ 216 ag_out(ag, out); 217 break; 218 } 219 220 if ((dst_h & ag_cors->ag_mask) == ag_cors->ag_dst_h) { 221 /* We found a route with a coarser mask that 222 * aggregates the current target. 223 * 224 * If it has a different next hop, it 225 * cannot replace the target, so output 226 * the target. 227 */ 228 if (ag->ag_gate != ag_cors->ag_gate 229 && !(ag->ag_state & AGS_FINE_GATE) 230 && !(ag_cors->ag_state & AGS_CORS_GATE)) { 231 ag_out(ag, out); 232 break; 233 } 234 235 /* If the coarse route has a good enough 236 * metric, it suppresses the target. 237 */ 238 if (ag_cors->ag_pref <= ag->ag_pref) { 239 if (ag_cors->ag_seqno > ag->ag_seqno) 240 ag_cors->ag_seqno = ag->ag_seqno; 241 if (AG_IS_REDUN(ag->ag_state) 242 && ag_cors->ag_mask==ag->ag_mask<<1) { 243 if (ag_cors->ag_dst_h == dst_h) 244 ag_cors->ag_state |= AGS_REDUN0; 245 else 246 ag_cors->ag_state |= AGS_REDUN1; 247 } 248 if (ag->ag_tag != ag_cors->ag_tag) 249 ag_cors->ag_tag = 0; 250 if (ag->ag_nhop != ag_cors->ag_nhop) 251 ag_cors->ag_nhop = 0; 252 break; 253 } 254 } 255 } 256 257 /* That route has either been output or suppressed */ 258 ag_cors = ag->ag_cors; 259 ag_del(ag); 260 } 261 262 CHECK_AG(); 263 } 264 265 266 /* Try to aggregate a route with previous routes. 267 */ 268 void 269 ag_check(naddr dst, 270 naddr mask, 271 naddr gate, 272 naddr nhop, 273 char metric, 274 char pref, 275 u_int seqno, 276 u_short tag, 277 u_short state, 278 void (*out)(struct ag_info *)) /* output using this */ 279 { 280 struct ag_info *ag, *nag, *ag_cors; 281 naddr xaddr; 282 int x; 283 284 NTOHL(dst); 285 286 /* Punt non-contiguous subnet masks. 287 * 288 * (X & -X) contains a single bit if and only if X is a power of 2. 289 * (X + (X & -X)) == 0 if and only if X is a power of 2. 290 */ 291 if ((mask & -mask) + mask != 0) { 292 struct ag_info nc_ag; 293 294 nc_ag.ag_dst_h = dst; 295 nc_ag.ag_mask = mask; 296 nc_ag.ag_gate = gate; 297 nc_ag.ag_nhop = nhop; 298 nc_ag.ag_metric = metric; 299 nc_ag.ag_pref = pref; 300 nc_ag.ag_tag = tag; 301 nc_ag.ag_state = state; 302 nc_ag.ag_seqno = seqno; 303 out(&nc_ag); 304 return; 305 } 306 307 /* Search for the right slot in the aggregation table. 308 */ 309 ag_cors = 0; 310 ag = ag_corsest; 311 while (ag != 0) { 312 if (ag->ag_mask >= mask) 313 break; 314 315 /* Suppress old routes (i.e. combine with compatible routes 316 * with coarser masks) as we look for the right slot in the 317 * aggregation table for the new route. 318 * A route to an address less than the current destination 319 * will not be affected by the current route or any route 320 * seen hereafter. That means it is safe to suppress it. 321 * This check keeps poor routes (e.g. with large hop counts) 322 * from preventing suppression of finer routes. 323 */ 324 if (ag_cors != 0 325 && ag->ag_dst_h < dst 326 && (ag->ag_state & AGS_SUPPRESS) 327 && ag_cors->ag_pref <= ag->ag_pref 328 && (ag->ag_dst_h & ag_cors->ag_mask) == ag_cors->ag_dst_h 329 && (ag_cors->ag_gate == ag->ag_gate 330 || (ag->ag_state & AGS_FINE_GATE) 331 || (ag_cors->ag_state & AGS_CORS_GATE))) { 332 if (ag_cors->ag_seqno > ag->ag_seqno) 333 ag_cors->ag_seqno = ag->ag_seqno; 334 if (AG_IS_REDUN(ag->ag_state) 335 && ag_cors->ag_mask==ag->ag_mask<<1) { 336 if (ag_cors->ag_dst_h == dst) 337 ag_cors->ag_state |= AGS_REDUN0; 338 else 339 ag_cors->ag_state |= AGS_REDUN1; 340 } 341 if (ag->ag_tag != ag_cors->ag_tag) 342 ag_cors->ag_tag = 0; 343 if (ag->ag_nhop != ag_cors->ag_nhop) 344 ag_cors->ag_nhop = 0; 345 ag_del(ag); 346 CHECK_AG(); 347 } else { 348 ag_cors = ag; 349 } 350 ag = ag_cors->ag_fine; 351 } 352 353 /* If we find the even/odd twin of the new route, and if the 354 * masks and so forth are equal, we can aggregate them. 355 * We can probably promote one of the pair. 356 * 357 * Since the routes are encountered in lexical order, 358 * the new route must be odd. However, the second or later 359 * times around this loop, it could be the even twin promoted 360 * from the even/odd pair of twins of the finer route. 361 */ 362 while (ag != 0 363 && ag->ag_mask == mask 364 && ((ag->ag_dst_h ^ dst) & (mask<<1)) == 0) { 365 366 /* Here we know the target route and the route in the current 367 * slot have the same netmasks and differ by at most the 368 * last bit. They are either for the same destination, or 369 * for an even/odd pair of destinations. 370 */ 371 if (ag->ag_dst_h == dst) { 372 /* We have two routes to the same destination. 373 * Routes are encountered in lexical order, so a 374 * route is never promoted until the parent route is 375 * already present. So we know that the new route is 376 * a promoted pair and the route already in the slot 377 * is the explicit route. 378 * 379 * Prefer the best route if their metrics differ, 380 * or the promoted one if not, following a sort 381 * of longest-match rule. 382 */ 383 if (pref <= ag->ag_pref) { 384 ag->ag_gate = gate; 385 ag->ag_nhop = nhop; 386 ag->ag_tag = tag; 387 ag->ag_metric = metric; 388 ag->ag_pref = pref; 389 x = ag->ag_state; 390 ag->ag_state = state; 391 state = x; 392 } 393 394 /* The sequence number controls flash updating, 395 * and should be the smaller of the two. 396 */ 397 if (ag->ag_seqno > seqno) 398 ag->ag_seqno = seqno; 399 400 /* some bits are set if they are set on either route */ 401 ag->ag_state |= (state & (AGS_PROMOTE_EITHER 402 | AGS_REDUN0 | AGS_REDUN1)); 403 return; 404 } 405 406 /* If one of the routes can be promoted and the other can 407 * be suppressed, it may be possible to combine them or 408 * worthwhile to promote one. 409 * 410 * Note that any route that can be promoted is always 411 * marked to be eligible to be suppressed. 412 */ 413 if (!((state & AGS_PROMOTE) 414 && (ag->ag_state & AGS_SUPPRESS)) 415 && !((ag->ag_state & AGS_PROMOTE) 416 && (state & AGS_SUPPRESS))) 417 break; 418 419 /* A pair of even/odd twin routes can be combined 420 * if either is redundant, or if they are via the 421 * same gateway and have the same metric. 422 */ 423 if (AG_IS_REDUN(ag->ag_state) 424 || AG_IS_REDUN(state) 425 || (ag->ag_gate == gate 426 && ag->ag_pref == pref 427 && (state & ag->ag_state & AGS_PROMOTE) != 0)) { 428 429 /* We have both the even and odd pairs. 430 * Since the routes are encountered in order, 431 * the route in the slot must be the even twin. 432 * 433 * Combine and promote the pair of routes. 434 */ 435 if (seqno > ag->ag_seqno) 436 seqno = ag->ag_seqno; 437 if (!AG_IS_REDUN(state)) 438 state &= ~AGS_REDUN1; 439 if (AG_IS_REDUN(ag->ag_state)) 440 state |= AGS_REDUN0; 441 else 442 state &= ~AGS_REDUN0; 443 state |= (ag->ag_state & AGS_PROMOTE_EITHER); 444 if (ag->ag_tag != tag) 445 tag = 0; 446 if (ag->ag_nhop != nhop) 447 nhop = 0; 448 449 /* Get rid of the even twin that was already 450 * in the slot. 451 */ 452 ag_del(ag); 453 454 } else if (ag->ag_pref >= pref 455 && (ag->ag_state & AGS_PROMOTE)) { 456 /* If we cannot combine the pair, maybe the route 457 * with the worse metric can be promoted. 458 * 459 * Promote the old, even twin, by giving its slot 460 * in the table to the new, odd twin. 461 */ 462 ag->ag_dst_h = dst; 463 464 xaddr = ag->ag_gate; 465 ag->ag_gate = gate; 466 gate = xaddr; 467 468 xaddr = ag->ag_nhop; 469 ag->ag_nhop = nhop; 470 nhop = xaddr; 471 472 x = ag->ag_tag; 473 ag->ag_tag = tag; 474 tag = x; 475 476 x = ag->ag_state; 477 ag->ag_state = state; 478 state = x; 479 if (!AG_IS_REDUN(state)) 480 state &= ~AGS_REDUN0; 481 482 x = ag->ag_metric; 483 ag->ag_metric = metric; 484 metric = x; 485 486 x = ag->ag_pref; 487 ag->ag_pref = pref; 488 pref = x; 489 490 if (seqno >= ag->ag_seqno) 491 seqno = ag->ag_seqno; 492 else 493 ag->ag_seqno = seqno; 494 495 } else { 496 if (!(state & AGS_PROMOTE)) 497 break; /* cannot promote either twin */ 498 499 /* promote the new, odd twin by shaving its 500 * mask and address. 501 */ 502 if (seqno > ag->ag_seqno) 503 seqno = ag->ag_seqno; 504 else 505 ag->ag_seqno = seqno; 506 if (!AG_IS_REDUN(state)) 507 state &= ~AGS_REDUN1; 508 } 509 510 mask <<= 1; 511 dst &= mask; 512 513 if (ag_cors == 0) { 514 ag = ag_corsest; 515 break; 516 } 517 ag = ag_cors; 518 ag_cors = ag->ag_cors; 519 } 520 521 /* When we can no longer promote and combine routes, 522 * flush the old route in the target slot. Also flush 523 * any finer routes that we know will never be aggregated by 524 * the new route. 525 * 526 * In case we moved toward coarser masks, 527 * get back where we belong 528 */ 529 if (ag != 0 530 && ag->ag_mask < mask) { 531 ag_cors = ag; 532 ag = ag->ag_fine; 533 } 534 535 /* Empty the target slot 536 */ 537 if (ag != 0 && ag->ag_mask == mask) { 538 ag_flush(ag->ag_dst_h, ag->ag_mask, out); 539 ag = (ag_cors == 0) ? ag_corsest : ag_cors->ag_fine; 540 } 541 542 #ifdef DEBUG_AG 543 (void)fflush(stderr); 544 if (ag == 0 && ag_cors != ag_finest) 545 abort(); 546 if (ag_cors == 0 && ag != ag_corsest) 547 abort(); 548 if (ag != 0 && ag->ag_cors != ag_cors) 549 abort(); 550 if (ag_cors != 0 && ag_cors->ag_fine != ag) 551 abort(); 552 CHECK_AG(); 553 #endif 554 555 /* Save the new route on the end of the table. 556 */ 557 nag = ag_avail; 558 ag_avail = nag->ag_fine; 559 560 nag->ag_dst_h = dst; 561 nag->ag_mask = mask; 562 nag->ag_gate = gate; 563 nag->ag_nhop = nhop; 564 nag->ag_metric = metric; 565 nag->ag_pref = pref; 566 nag->ag_tag = tag; 567 nag->ag_state = state; 568 nag->ag_seqno = seqno; 569 570 nag->ag_fine = ag; 571 if (ag != 0) 572 ag->ag_cors = nag; 573 else 574 ag_finest = nag; 575 nag->ag_cors = ag_cors; 576 if (ag_cors == 0) 577 ag_corsest = nag; 578 else 579 ag_cors->ag_fine = nag; 580 CHECK_AG(); 581 } 582 583 584 static char * 585 rtm_type_name(u_char type) 586 { 587 static char *rtm_types[] = { 588 "RTM_ADD", 589 "RTM_DELETE", 590 "RTM_CHANGE", 591 "RTM_GET", 592 "RTM_LOSING", 593 "RTM_REDIRECT", 594 "RTM_MISS", 595 "RTM_LOCK", 596 "RTM_OLDADD", 597 "RTM_OLDDEL", 598 "RTM_RESOLVE", 599 "RTM_NEWADDR", 600 "RTM_DELADDR", 601 "RTM_IFINFO" 602 }; 603 static char name0[10]; 604 605 606 if (type > sizeof(rtm_types)/sizeof(rtm_types[0]) 607 || type == 0) { 608 sprintf(name0, "RTM type %#x", type); 609 return name0; 610 } else { 611 return rtm_types[type-1]; 612 } 613 } 614 615 616 /* Trim a mask in a sockaddr 617 * Produce the index of the first zero byte. 618 * i.e. Produce a index of 4 for an mask of 0. (default route) 619 */ 620 void 621 #ifdef _HAVE_SIN_LEN 622 masktrim(struct sockaddr_in *ap) 623 #else 624 masktrim(struct sockaddr_in_new *ap) 625 #endif 626 { 627 register char *cp; 628 629 ap->sin_port = 0xffff; /* buffer zone for default route */ 630 cp = (char *)(&ap->sin_addr.s_addr+1); 631 while (*--cp == 0) 632 continue; 633 /*ap->sin_port = 0x0;*/ /* may not be needed (who cares?)*/ 634 ap->sin_len = cp - (char*)ap + 1; 635 } 636 637 638 /* Tell the kernel to add, delete or change a route 639 */ 640 static void 641 rtioctl(int action, /* RTM_DELETE, etc */ 642 naddr dst, 643 naddr gate, 644 naddr mask, 645 int metric, 646 int flags) 647 { 648 struct { 649 struct rt_msghdr w_rtm; 650 struct sockaddr_in w_dst; 651 struct sockaddr_in w_gate; 652 #ifdef _HAVE_SA_LEN 653 struct sockaddr_in w_mask; 654 #else 655 struct sockaddr_in_new w_mask; 656 #endif 657 } w; 658 long cc; 659 660 again: 661 bzero(&w, sizeof(w)); 662 w.w_rtm.rtm_msglen = sizeof(w); 663 w.w_rtm.rtm_version = RTM_VERSION; 664 w.w_rtm.rtm_type = action; 665 w.w_rtm.rtm_flags = flags; 666 w.w_rtm.rtm_seq = ++rt_sock_seqno; 667 w.w_rtm.rtm_addrs = RTA_DST|RTA_GATEWAY; 668 if (metric != 0) { 669 w.w_rtm.rtm_rmx.rmx_hopcount = metric; 670 w.w_rtm.rtm_inits |= RTV_HOPCOUNT; 671 } 672 w.w_dst.sin_family = AF_INET; 673 w.w_dst.sin_addr.s_addr = dst; 674 w.w_gate.sin_family = AF_INET; 675 w.w_gate.sin_addr.s_addr = gate; 676 #ifdef _HAVE_SA_LEN 677 w.w_dst.sin_len = sizeof(w.w_dst); 678 w.w_gate.sin_len = sizeof(w.w_gate); 679 #endif 680 if (mask == HOST_MASK) { 681 w.w_rtm.rtm_flags |= RTF_HOST; 682 w.w_rtm.rtm_msglen -= sizeof(w.w_mask); 683 } else { 684 w.w_rtm.rtm_addrs |= RTA_NETMASK; 685 w.w_mask.sin_addr.s_addr = htonl(mask); 686 #ifdef _HAVE_SA_LEN 687 masktrim(&w.w_mask); 688 w.w_rtm.rtm_msglen -= (sizeof(w.w_mask) - w.w_mask.sin_len); 689 #endif 690 } 691 692 if (TRACEKERNEL) 693 trace_kernel("write kernel %s %s->%s metric=%d flags=%#x\n", 694 rtm_type_name(action), 695 addrname(dst, mask, 0), naddr_ntoa(gate), 696 metric, flags); 697 698 #ifndef NO_INSTALL 699 cc = write(rt_sock, &w, w.w_rtm.rtm_msglen); 700 if (cc == w.w_rtm.rtm_msglen) 701 return; 702 if (cc < 0) { 703 if (errno == ESRCH 704 && (action == RTM_CHANGE || action == RTM_DELETE)) { 705 trace_act("route to %s disappeared before %s", 706 addrname(dst, mask, 0), 707 rtm_type_name(action)); 708 if (action == RTM_CHANGE) { 709 action = RTM_ADD; 710 goto again; 711 } 712 return; 713 } 714 msglog("write(rt_sock) %s %s --> %s: %s", 715 rtm_type_name(action), 716 addrname(dst, mask, 0), naddr_ntoa(gate), 717 strerror(errno)); 718 } else { 719 msglog("write(rt_sock) wrote %d instead of %d", 720 cc, w.w_rtm.rtm_msglen); 721 } 722 #endif 723 } 724 725 726 #define KHASH_SIZE 71 /* should be prime */ 727 #define KHASH(a,m) khash_bins[((a) ^ (m)) % KHASH_SIZE] 728 static struct khash { 729 struct khash *k_next; 730 naddr k_dst; 731 naddr k_mask; 732 naddr k_gate; 733 short k_metric; 734 u_short k_state; 735 #define KS_NEW 0x001 736 #define KS_DELETE 0x002 737 #define KS_ADD 0x004 /* add to the kernel */ 738 #define KS_CHANGE 0x008 /* tell kernel to change the route */ 739 #define KS_DEL_ADD 0x010 /* delete & add to change the kernel */ 740 #define KS_STATIC 0x020 /* Static flag in kernel */ 741 #define KS_GATEWAY 0x040 /* G flag in kernel */ 742 #define KS_DYNAMIC 0x080 /* result of redirect */ 743 #define KS_DELETED 0x100 /* already deleted */ 744 time_t k_keep; 745 #define K_KEEP_LIM 30 746 time_t k_redirect_time; /* when redirected route 1st seen */ 747 } *khash_bins[KHASH_SIZE]; 748 749 750 static struct khash* 751 kern_find(naddr dst, naddr mask, struct khash ***ppk) 752 { 753 struct khash *k, **pk; 754 755 for (pk = &KHASH(dst,mask); (k = *pk) != 0; pk = &k->k_next) { 756 if (k->k_dst == dst && k->k_mask == mask) 757 break; 758 } 759 if (ppk != 0) 760 *ppk = pk; 761 return k; 762 } 763 764 765 static struct khash* 766 kern_add(naddr dst, naddr mask) 767 { 768 struct khash *k, **pk; 769 770 k = kern_find(dst, mask, &pk); 771 if (k != 0) 772 return k; 773 774 k = (struct khash *)rtmalloc(sizeof(*k), "kern_add"); 775 776 bzero(k, sizeof(*k)); 777 k->k_dst = dst; 778 k->k_mask = mask; 779 k->k_state = KS_NEW; 780 k->k_keep = now.tv_sec; 781 *pk = k; 782 783 return k; 784 } 785 786 787 /* If a kernel route has a non-zero metric, check that it is still in the 788 * daemon table, and not deleted by interfaces coming and going. 789 */ 790 static void 791 kern_check_static(struct khash *k, 792 struct interface *ifp) 793 { 794 struct rt_entry *rt; 795 naddr int_addr; 796 797 if (k->k_metric == 0) 798 return; 799 800 int_addr = (ifp != 0) ? ifp->int_addr : loopaddr; 801 802 rt = rtget(k->k_dst, k->k_mask); 803 if (rt != 0) { 804 if (!(rt->rt_state & RS_STATIC)) 805 rtchange(rt, rt->rt_state | RS_STATIC, 806 k->k_gate, int_addr, 807 k->k_metric, 0, ifp, now.tv_sec, 0); 808 } else { 809 rtadd(k->k_dst, k->k_mask, k->k_gate, int_addr, 810 k->k_metric, 0, RS_STATIC, ifp); 811 } 812 } 813 814 815 /* add a route the kernel told us 816 */ 817 static void 818 rtm_add(struct rt_msghdr *rtm, 819 struct rt_addrinfo *info, 820 time_t keep) 821 { 822 struct khash *k; 823 struct interface *ifp; 824 naddr mask; 825 826 827 if (rtm->rtm_flags & RTF_HOST) { 828 mask = HOST_MASK; 829 } else if (INFO_MASK(info) != 0) { 830 mask = ntohl(S_ADDR(INFO_MASK(info))); 831 } else { 832 msglog("ignore %s without mask", rtm_type_name(rtm->rtm_type)); 833 return; 834 } 835 836 if (INFO_GATE(info) == 0 837 || INFO_GATE(info)->sa_family != AF_INET) { 838 msglog("ignore %s without gateway", 839 rtm_type_name(rtm->rtm_type)); 840 return; 841 } 842 843 k = kern_add(S_ADDR(INFO_DST(info)), mask); 844 if (k->k_state & KS_NEW) 845 k->k_keep = now.tv_sec+keep; 846 k->k_gate = S_ADDR(INFO_GATE(info)); 847 k->k_metric = rtm->rtm_rmx.rmx_hopcount; 848 if (k->k_metric < 0) 849 k->k_metric = 0; 850 else if (k->k_metric > HOPCNT_INFINITY) 851 k->k_metric = HOPCNT_INFINITY; 852 k->k_state &= ~(KS_DELETED | KS_GATEWAY | KS_STATIC | KS_NEW); 853 if (rtm->rtm_flags & RTF_GATEWAY) 854 k->k_state |= KS_GATEWAY; 855 if (rtm->rtm_flags & RTF_STATIC) 856 k->k_state |= KS_STATIC; 857 858 if (0 != (rtm->rtm_flags & (RTF_DYNAMIC | RTF_MODIFIED))) { 859 if (INFO_AUTHOR(info) != 0 860 && INFO_AUTHOR(info)->sa_family == AF_INET) 861 ifp = iflookup(S_ADDR(INFO_AUTHOR(info))); 862 else 863 ifp = 0; 864 if (supplier 865 && (ifp == 0 || !(ifp->int_state & IS_REDIRECT_OK))) { 866 /* Routers are not supposed to listen to redirects, 867 * so delete it if it came via an unknown interface 868 * or the interface does not have special permission. 869 */ 870 k->k_state &= ~KS_DYNAMIC; 871 k->k_state |= KS_DELETE; 872 LIM_SEC(need_kern, 0); 873 trace_act("mark for deletion redirected %s --> %s" 874 " via %s", 875 addrname(k->k_dst, k->k_mask, 0), 876 naddr_ntoa(k->k_gate), 877 ifp ? ifp->int_name : "unknown interface"); 878 } else { 879 k->k_state |= KS_DYNAMIC; 880 k->k_redirect_time = now.tv_sec; 881 trace_act("accept redirected %s --> %s via %s", 882 addrname(k->k_dst, k->k_mask, 0), 883 naddr_ntoa(k->k_gate), 884 ifp ? ifp->int_name : "unknown interface"); 885 } 886 return; 887 } 888 889 /* If it is not a static route, quit until the next comparison 890 * between the kernel and daemon tables, when it will be deleted. 891 */ 892 if (!(k->k_state & KS_STATIC)) { 893 k->k_state |= KS_DELETE; 894 LIM_SEC(need_kern, k->k_keep); 895 return; 896 } 897 898 /* Put static routes with real metrics into the daemon table so 899 * they can be advertised. 900 * 901 * Find the interface toward the gateway. 902 */ 903 ifp = iflookup(k->k_gate); 904 if (ifp == 0) 905 msglog("static route %s --> %s impossibly lacks ifp", 906 addrname(S_ADDR(INFO_DST(info)), mask, 0), 907 naddr_ntoa(k->k_gate)); 908 909 kern_check_static(k, ifp); 910 } 911 912 913 /* deal with packet loss 914 */ 915 static void 916 rtm_lose(struct rt_msghdr *rtm, 917 struct rt_addrinfo *info) 918 { 919 if (INFO_GATE(info) == 0 920 || INFO_GATE(info)->sa_family != AF_INET) { 921 trace_act("ignore %s without gateway", 922 rtm_type_name(rtm->rtm_type)); 923 return; 924 } 925 926 if (!supplier) 927 rdisc_age(S_ADDR(INFO_GATE(info))); 928 929 age(S_ADDR(INFO_GATE(info))); 930 } 931 932 933 /* Clean the kernel table by copying it to the daemon image. 934 * Eventually the daemon will delete any extra routes. 935 */ 936 void 937 flush_kern(void) 938 { 939 size_t needed; 940 int mib[6]; 941 char *buf, *next, *lim; 942 struct rt_msghdr *rtm; 943 struct interface *ifp; 944 static struct sockaddr_in gate_sa; 945 struct rt_addrinfo info; 946 947 948 mib[0] = CTL_NET; 949 mib[1] = PF_ROUTE; 950 mib[2] = 0; /* protocol */ 951 mib[3] = 0; /* wildcard address family */ 952 mib[4] = NET_RT_DUMP; 953 mib[5] = 0; /* no flags */ 954 if (sysctl(mib, 6, 0, &needed, 0, 0) < 0) { 955 DBGERR(1,"RT_DUMP-sysctl-estimate"); 956 return; 957 } 958 buf = (char *)rtmalloc(needed, "flush_kern"); 959 if (sysctl(mib, 6, buf, &needed, 0, 0) < 0) 960 BADERR(1,"RT_DUMP"); 961 lim = buf + needed; 962 for (next = buf; next < lim; next += rtm->rtm_msglen) { 963 rtm = (struct rt_msghdr *)next; 964 965 rt_xaddrs(&info, 966 (struct sockaddr *)(rtm+1), 967 (struct sockaddr *)(next + rtm->rtm_msglen), 968 rtm->rtm_addrs); 969 970 if (INFO_DST(&info) == 0 971 || INFO_DST(&info)->sa_family != AF_INET) 972 continue; 973 974 /* ignore ARP table entries on systems with a merged route 975 * and ARP table. 976 */ 977 if (rtm->rtm_flags & RTF_LLINFO) 978 continue; 979 980 if (INFO_GATE(&info) == 0) 981 continue; 982 if (INFO_GATE(&info)->sa_family != AF_INET) { 983 if (INFO_GATE(&info)->sa_family != AF_LINK) 984 continue; 985 ifp = ifwithindex(((struct sockaddr_dl *) 986 INFO_GATE(&info))->sdl_index); 987 if (ifp == 0) 988 continue; 989 if ((ifp->int_if_flags & IFF_POINTOPOINT) 990 || S_ADDR(INFO_DST(&info)) == ifp->int_addr) 991 gate_sa.sin_addr.s_addr = ifp->int_addr; 992 else 993 gate_sa.sin_addr.s_addr = htonl(ifp->int_net); 994 #ifdef _HAVE_SA_LEN 995 gate_sa.sin_len = sizeof(gate_sa); 996 #endif 997 gate_sa.sin_family = AF_INET; 998 INFO_GATE(&info) = (struct sockaddr *)&gate_sa; 999 } 1000 1001 /* ignore multicast addresses 1002 */ 1003 if (IN_MULTICAST(ntohl(S_ADDR(INFO_DST(&info))))) 1004 continue; 1005 1006 /* Note static routes and interface routes, and also 1007 * preload the image of the kernel table so that 1008 * we can later clean it, as well as avoid making 1009 * unneeded changes. Keep the old kernel routes for a 1010 * few seconds to allow a RIP or router-discovery 1011 * response to be heard. 1012 */ 1013 rtm_add(rtm,&info,MIN_WAITTIME); 1014 } 1015 free(buf); 1016 } 1017 1018 1019 /* Listen to announcements from the kernel 1020 */ 1021 void 1022 read_rt(void) 1023 { 1024 long cc; 1025 struct interface *ifp; 1026 naddr mask; 1027 union { 1028 struct { 1029 struct rt_msghdr rtm; 1030 struct sockaddr addrs[RTAX_MAX]; 1031 } r; 1032 struct if_msghdr ifm; 1033 } m; 1034 char str[100], *strp; 1035 struct rt_addrinfo info; 1036 1037 1038 for (;;) { 1039 cc = read(rt_sock, &m, sizeof(m)); 1040 if (cc <= 0) { 1041 if (cc < 0 && errno != EWOULDBLOCK) 1042 LOGERR("read(rt_sock)"); 1043 return; 1044 } 1045 1046 if (m.r.rtm.rtm_version != RTM_VERSION) { 1047 msglog("bogus routing message version %d", 1048 m.r.rtm.rtm_version); 1049 continue; 1050 } 1051 1052 /* Ignore our own results. 1053 */ 1054 if (m.r.rtm.rtm_type <= RTM_CHANGE 1055 && m.r.rtm.rtm_pid == mypid) { 1056 static int complained = 0; 1057 if (!complained) { 1058 msglog("receiving our own change messages"); 1059 complained = 1; 1060 } 1061 continue; 1062 } 1063 1064 if (m.r.rtm.rtm_type == RTM_IFINFO 1065 || m.r.rtm.rtm_type == RTM_NEWADDR 1066 || m.r.rtm.rtm_type == RTM_DELADDR) { 1067 ifp = ifwithindex(m.ifm.ifm_index); 1068 if (ifp == 0) 1069 trace_act("note %s with flags %#x" 1070 " for index #%d", 1071 rtm_type_name(m.r.rtm.rtm_type), 1072 m.ifm.ifm_flags, 1073 m.ifm.ifm_index); 1074 else 1075 trace_act("note %s with flags %#x for %s", 1076 rtm_type_name(m.r.rtm.rtm_type), 1077 m.ifm.ifm_flags, 1078 ifp->int_name); 1079 1080 /* After being informed of a change to an interface, 1081 * check them all now if the check would otherwise 1082 * be a long time from now, if the interface is 1083 * not known, or if the interface has been turned 1084 * off or on. 1085 */ 1086 if (ifinit_timer.tv_sec-now.tv_sec>=CHECK_BAD_INTERVAL 1087 || ifp == 0 1088 || ((ifp->int_if_flags ^ m.ifm.ifm_flags) 1089 & IFF_UP_RUNNING) != 0) 1090 ifinit_timer.tv_sec = now.tv_sec; 1091 continue; 1092 } 1093 1094 strcpy(str, rtm_type_name(m.r.rtm.rtm_type)); 1095 strp = &str[strlen(str)]; 1096 if (m.r.rtm.rtm_type <= RTM_CHANGE) 1097 strp += sprintf(strp," from pid %d",m.r.rtm.rtm_pid); 1098 1099 rt_xaddrs(&info, m.r.addrs, &m.r.addrs[RTAX_MAX], 1100 m.r.rtm.rtm_addrs); 1101 1102 if (INFO_DST(&info) == 0) { 1103 trace_act("ignore %s without dst", str); 1104 continue; 1105 } 1106 1107 if (INFO_DST(&info)->sa_family != AF_INET) { 1108 trace_act("ignore %s for AF %d", str, 1109 INFO_DST(&info)->sa_family); 1110 continue; 1111 } 1112 1113 mask = ((INFO_MASK(&info) != 0) 1114 ? ntohl(S_ADDR(INFO_MASK(&info))) 1115 : (m.r.rtm.rtm_flags & RTF_HOST) 1116 ? HOST_MASK 1117 : std_mask(S_ADDR(INFO_DST(&info)))); 1118 1119 strp += sprintf(strp, ": %s", 1120 addrname(S_ADDR(INFO_DST(&info)), mask, 0)); 1121 1122 if (IN_MULTICAST(ntohl(S_ADDR(INFO_DST(&info))))) { 1123 trace_act("ignore multicast %s", str); 1124 continue; 1125 } 1126 1127 if (INFO_GATE(&info) != 0 1128 && INFO_GATE(&info)->sa_family == AF_INET) 1129 strp += sprintf(strp, " --> %s", 1130 saddr_ntoa(INFO_GATE(&info))); 1131 1132 if (INFO_AUTHOR(&info) != 0) 1133 strp += sprintf(strp, " by authority of %s", 1134 saddr_ntoa(INFO_AUTHOR(&info))); 1135 1136 switch (m.r.rtm.rtm_type) { 1137 case RTM_ADD: 1138 case RTM_CHANGE: 1139 case RTM_REDIRECT: 1140 if (m.r.rtm.rtm_errno != 0) { 1141 trace_act("ignore %s with \"%s\" error", 1142 str, strerror(m.r.rtm.rtm_errno)); 1143 } else { 1144 trace_act("%s", str); 1145 rtm_add(&m.r.rtm,&info,0); 1146 } 1147 break; 1148 1149 case RTM_DELETE: 1150 if (m.r.rtm.rtm_errno != 0) { 1151 trace_act("ignore %s with \"%s\" error", 1152 str, strerror(m.r.rtm.rtm_errno)); 1153 } else { 1154 trace_act("%s", str); 1155 del_static(S_ADDR(INFO_DST(&info)), mask, 1); 1156 } 1157 break; 1158 1159 case RTM_LOSING: 1160 trace_act("%s", str); 1161 rtm_lose(&m.r.rtm,&info); 1162 break; 1163 1164 default: 1165 trace_act("ignore %s", str); 1166 break; 1167 } 1168 } 1169 } 1170 1171 1172 /* after aggregating, note routes that belong in the kernel 1173 */ 1174 static void 1175 kern_out(struct ag_info *ag) 1176 { 1177 struct khash *k; 1178 1179 1180 /* Do not install bad routes if they are not already present. 1181 * This includes routes that had RS_NET_SYN for interfaces that 1182 * recently died. 1183 */ 1184 if (ag->ag_metric == HOPCNT_INFINITY) { 1185 k = kern_find(htonl(ag->ag_dst_h), ag->ag_mask, 0); 1186 if (k == 0) 1187 return; 1188 } else { 1189 k = kern_add(htonl(ag->ag_dst_h), ag->ag_mask); 1190 } 1191 1192 if (k->k_state & KS_NEW) { 1193 /* will need to add new entry to the kernel table */ 1194 k->k_state = KS_ADD; 1195 if (ag->ag_state & AGS_GATEWAY) 1196 k->k_state |= KS_GATEWAY; 1197 k->k_gate = ag->ag_gate; 1198 k->k_metric = ag->ag_metric; 1199 return; 1200 } 1201 1202 if (k->k_state & KS_STATIC) 1203 return; 1204 1205 /* modify existing kernel entry if necessary */ 1206 if (k->k_gate != ag->ag_gate 1207 || k->k_metric != ag->ag_metric) { 1208 k->k_gate = ag->ag_gate; 1209 k->k_metric = ag->ag_metric; 1210 k->k_state |= KS_CHANGE; 1211 } 1212 1213 if (k->k_state & KS_DYNAMIC) { 1214 k->k_state &= ~KS_DYNAMIC; 1215 k->k_state |= (KS_ADD | KS_DEL_ADD); 1216 } 1217 1218 if ((k->k_state & KS_GATEWAY) 1219 && !(ag->ag_state & AGS_GATEWAY)) { 1220 k->k_state &= ~KS_GATEWAY; 1221 k->k_state |= (KS_ADD | KS_DEL_ADD); 1222 } else if (!(k->k_state & KS_GATEWAY) 1223 && (ag->ag_state & AGS_GATEWAY)) { 1224 k->k_state |= KS_GATEWAY; 1225 k->k_state |= (KS_ADD | KS_DEL_ADD); 1226 } 1227 1228 /* Deleting-and-adding is necessary to change aspects of a route. 1229 * Just delete instead of deleting and then adding a bad route. 1230 * Otherwise, we want to keep the route in the kernel. 1231 */ 1232 if (k->k_metric == HOPCNT_INFINITY 1233 && (k->k_state & KS_DEL_ADD)) 1234 k->k_state |= KS_DELETE; 1235 else 1236 k->k_state &= ~KS_DELETE; 1237 #undef RT 1238 } 1239 1240 1241 /* ARGSUSED */ 1242 static int 1243 walk_kern(struct radix_node *rn, 1244 struct walkarg *w) 1245 { 1246 #define RT ((struct rt_entry *)rn) 1247 char metric, pref; 1248 u_int ags = 0; 1249 1250 1251 /* Do not install synthetic routes */ 1252 if (RT->rt_state & RS_NET_SYN) 1253 return 0; 1254 1255 if (!(RT->rt_state & RS_IF)) { 1256 ags |= (AGS_GATEWAY | AGS_SUPPRESS | AGS_PROMOTE); 1257 1258 } else { 1259 /* Do not install routes for "external" remote interfaces. 1260 */ 1261 if (RT->rt_ifp != 0 && (RT->rt_ifp->int_state & IS_EXTERNAL)) 1262 return 0; 1263 1264 ags |= AGS_IF; 1265 1266 /* If it is not an interface, or an alias for an interface, 1267 * it must be a "gateway." 1268 * 1269 * If it is a "remote" interface, it is also a "gateway" to 1270 * the kernel if is not a alias. 1271 */ 1272 if (RT->rt_ifp == 0 1273 || (RT->rt_ifp->int_state & IS_REMOTE)) 1274 ags |= (AGS_GATEWAY | AGS_SUPPRESS | AGS_PROMOTE); 1275 } 1276 1277 if (RT->rt_state & RS_RDISC) 1278 ags |= AGS_CORS_GATE; 1279 1280 /* aggregate good routes without regard to their metric */ 1281 pref = 1; 1282 metric = RT->rt_metric; 1283 if (metric == HOPCNT_INFINITY) { 1284 /* if the route is dead, so try hard to aggregate. */ 1285 pref = HOPCNT_INFINITY; 1286 ags |= (AGS_FINE_GATE | AGS_SUPPRESS); 1287 } 1288 1289 ag_check(RT->rt_dst, RT->rt_mask, RT->rt_gate, 0, 1290 metric,pref, 0, 0, ags, kern_out); 1291 return 0; 1292 #undef RT 1293 } 1294 1295 1296 /* Update the kernel table to match the daemon table. 1297 */ 1298 static void 1299 fix_kern(void) 1300 { 1301 int i, flags; 1302 struct khash *k, **pk; 1303 1304 1305 need_kern = age_timer; 1306 1307 /* Walk daemon table, updating the copy of the kernel table. 1308 */ 1309 (void)rn_walktree(rhead, walk_kern, 0); 1310 ag_flush(0,0,kern_out); 1311 1312 for (i = 0; i < KHASH_SIZE; i++) { 1313 for (pk = &khash_bins[i]; (k = *pk) != 0; ) { 1314 /* Do not touch static routes */ 1315 if (k->k_state & KS_STATIC) { 1316 kern_check_static(k,0); 1317 pk = &k->k_next; 1318 continue; 1319 } 1320 1321 /* check hold on routes deleted by the operator */ 1322 if (k->k_keep > now.tv_sec) { 1323 LIM_SEC(need_kern, k->k_keep); 1324 k->k_state |= KS_DELETE; 1325 pk = &k->k_next; 1326 continue; 1327 } 1328 1329 if ((k->k_state & (KS_DELETE | KS_DYNAMIC)) 1330 == KS_DELETE) { 1331 if (!(k->k_state & KS_DELETED)) 1332 rtioctl(RTM_DELETE, 1333 k->k_dst, k->k_gate, k->k_mask, 1334 0, 0); 1335 *pk = k->k_next; 1336 free(k); 1337 continue; 1338 } 1339 1340 if (0 != (k->k_state&(KS_ADD|KS_CHANGE|KS_DEL_ADD))) { 1341 if (k->k_state & KS_DEL_ADD) { 1342 rtioctl(RTM_DELETE, 1343 k->k_dst,k->k_gate,k->k_mask, 1344 0, 0); 1345 k->k_state &= ~KS_DYNAMIC; 1346 } 1347 1348 flags = 0; 1349 if (0 != (k->k_state&(KS_GATEWAY|KS_DYNAMIC))) 1350 flags |= RTF_GATEWAY; 1351 1352 if (k->k_state & KS_ADD) { 1353 rtioctl(RTM_ADD, 1354 k->k_dst, k->k_gate, k->k_mask, 1355 k->k_metric, flags); 1356 } else if (k->k_state & KS_CHANGE) { 1357 rtioctl(RTM_CHANGE, 1358 k->k_dst,k->k_gate,k->k_mask, 1359 k->k_metric, flags); 1360 } 1361 k->k_state &= ~(KS_ADD|KS_CHANGE|KS_DEL_ADD); 1362 } 1363 1364 /* Mark this route to be deleted in the next cycle. 1365 * This deletes routes that disappear from the 1366 * daemon table, since the normal aging code 1367 * will clear the bit for routes that have not 1368 * disappeared from the daemon table. 1369 */ 1370 k->k_state |= KS_DELETE; 1371 pk = &k->k_next; 1372 } 1373 } 1374 } 1375 1376 1377 /* Delete a static route in the image of the kernel table. 1378 */ 1379 void 1380 del_static(naddr dst, 1381 naddr mask, 1382 int gone) 1383 { 1384 struct khash *k; 1385 struct rt_entry *rt; 1386 1387 /* Just mark it in the table to be deleted next time the kernel 1388 * table is updated. 1389 * If it has already been deleted, mark it as such, and set its 1390 * keep-timer so that it will not be deleted again for a while. 1391 * This lets the operator delete a route added by the daemon 1392 * and add a replacement. 1393 */ 1394 k = kern_find(dst, mask, 0); 1395 if (k != 0) { 1396 k->k_state &= ~(KS_STATIC | KS_DYNAMIC); 1397 k->k_state |= KS_DELETE; 1398 if (gone) { 1399 k->k_state |= KS_DELETED; 1400 k->k_keep = now.tv_sec + K_KEEP_LIM; 1401 } 1402 } 1403 1404 rt = rtget(dst, mask); 1405 if (rt != 0 && (rt->rt_state & RS_STATIC)) 1406 rtbad(rt); 1407 } 1408 1409 1410 /* Delete all routes generated from ICMP Redirects that use a given gateway, 1411 * as well as old redirected routes. 1412 */ 1413 void 1414 del_redirects(naddr bad_gate, 1415 time_t old) 1416 { 1417 int i; 1418 struct khash *k; 1419 1420 1421 for (i = 0; i < KHASH_SIZE; i++) { 1422 for (k = khash_bins[i]; k != 0; k = k->k_next) { 1423 if (!(k->k_state & KS_DYNAMIC) 1424 || (k->k_state & KS_STATIC)) 1425 continue; 1426 1427 if (k->k_gate != bad_gate 1428 && k->k_redirect_time > old 1429 && !supplier) 1430 continue; 1431 1432 k->k_state |= KS_DELETE; 1433 k->k_state &= ~KS_DYNAMIC; 1434 need_kern.tv_sec = now.tv_sec; 1435 trace_act("mark redirected %s --> %s for deletion", 1436 addrname(k->k_dst, k->k_mask, 0), 1437 naddr_ntoa(k->k_gate)); 1438 } 1439 } 1440 } 1441 1442 1443 /* Start the daemon tables. 1444 */ 1445 void 1446 rtinit(void) 1447 { 1448 extern int max_keylen; 1449 int i; 1450 struct ag_info *ag; 1451 1452 /* Initialize the radix trees */ 1453 max_keylen = sizeof(struct sockaddr_in); 1454 rn_init(); 1455 rn_inithead((void**)&rhead, 32); 1456 1457 /* mark all of the slots in the table free */ 1458 ag_avail = ag_slots; 1459 for (ag = ag_slots, i = 1; i < NUM_AG_SLOTS; i++) { 1460 ag->ag_fine = ag+1; 1461 ag++; 1462 } 1463 } 1464 1465 1466 #ifdef _HAVE_SIN_LEN 1467 static struct sockaddr_in dst_sock = {sizeof(dst_sock), AF_INET}; 1468 static struct sockaddr_in mask_sock = {sizeof(mask_sock), AF_INET}; 1469 #else 1470 static struct sockaddr_in_new dst_sock = {_SIN_ADDR_SIZE, AF_INET}; 1471 static struct sockaddr_in_new mask_sock = {_SIN_ADDR_SIZE, AF_INET}; 1472 #endif 1473 1474 1475 void 1476 set_need_flash(void) 1477 { 1478 if (!need_flash) { 1479 need_flash = 1; 1480 /* Do not send the flash update immediately. Wait a little 1481 * while to hear from other routers. 1482 */ 1483 no_flash.tv_sec = now.tv_sec + MIN_WAITTIME; 1484 } 1485 } 1486 1487 1488 /* Get a particular routing table entry 1489 */ 1490 struct rt_entry * 1491 rtget(naddr dst, naddr mask) 1492 { 1493 struct rt_entry *rt; 1494 1495 dst_sock.sin_addr.s_addr = dst; 1496 mask_sock.sin_addr.s_addr = mask; 1497 masktrim(&mask_sock); 1498 rt = (struct rt_entry *)rhead->rnh_lookup(&dst_sock,&mask_sock,rhead); 1499 if (!rt 1500 || rt->rt_dst != dst 1501 || rt->rt_mask != mask) 1502 return 0; 1503 1504 return rt; 1505 } 1506 1507 1508 /* Find a route to dst as the kernel would. 1509 */ 1510 struct rt_entry * 1511 rtfind(naddr dst) 1512 { 1513 dst_sock.sin_addr.s_addr = dst; 1514 return (struct rt_entry *)rhead->rnh_matchaddr(&dst_sock, rhead); 1515 } 1516 1517 1518 /* add a route to the table 1519 */ 1520 void 1521 rtadd(naddr dst, 1522 naddr mask, 1523 naddr gate, /* forward packets here */ 1524 naddr router, /* on the authority of this router */ 1525 int metric, 1526 u_short tag, 1527 u_int state, /* rs_state for the entry */ 1528 struct interface *ifp) 1529 { 1530 struct rt_entry *rt; 1531 naddr smask; 1532 int i; 1533 struct rt_spare *rts; 1534 1535 rt = (struct rt_entry *)rtmalloc(sizeof (*rt), "rtadd"); 1536 bzero(rt, sizeof(*rt)); 1537 for (rts = rt->rt_spares, i = NUM_SPARES; i != 0; i--, rts++) 1538 rts->rts_metric = HOPCNT_INFINITY; 1539 1540 rt->rt_nodes->rn_key = (caddr_t)&rt->rt_dst_sock; 1541 rt->rt_dst = dst; 1542 rt->rt_dst_sock.sin_family = AF_INET; 1543 #ifdef _HAVE_SIN_LEN 1544 rt->rt_dst_sock.sin_len = dst_sock.sin_len; 1545 #endif 1546 if (mask != HOST_MASK) { 1547 smask = std_mask(dst); 1548 if ((smask & ~mask) == 0 && mask > smask) 1549 state |= RS_SUBNET; 1550 } 1551 mask_sock.sin_addr.s_addr = mask; 1552 masktrim(&mask_sock); 1553 rt->rt_mask = mask; 1554 rt->rt_state = state; 1555 rt->rt_gate = gate; 1556 rt->rt_router = router; 1557 rt->rt_time = now.tv_sec; 1558 rt->rt_metric = metric; 1559 rt->rt_poison_metric = HOPCNT_INFINITY; 1560 rt->rt_tag = tag; 1561 rt->rt_ifp = ifp; 1562 rt->rt_seqno = update_seqno; 1563 1564 if (++total_routes == MAX_ROUTES) 1565 msglog("have maximum (%d) routes", total_routes); 1566 if (TRACEACTIONS) 1567 trace_add_del("Add", rt); 1568 1569 need_kern.tv_sec = now.tv_sec; 1570 set_need_flash(); 1571 1572 if (0 == rhead->rnh_addaddr(&rt->rt_dst_sock, &mask_sock, 1573 rhead, rt->rt_nodes)) { 1574 /* 1575 * This will happen if RIP1 and RIP2 routeds talk to one another and 1576 * there are variable subnets. This is only good for filling up your 1577 * syslog. -jkh 1578 */ 1579 #if 0 1580 msglog("rnh_addaddr() failed for %s mask=%#x", 1581 naddr_ntoa(dst), mask); 1582 #endif 1583 } 1584 } 1585 1586 1587 /* notice a changed route 1588 */ 1589 void 1590 rtchange(struct rt_entry *rt, 1591 u_int state, /* new state bits */ 1592 naddr gate, /* now forward packets here */ 1593 naddr router, /* on the authority of this router */ 1594 int metric, /* new metric */ 1595 u_short tag, 1596 struct interface *ifp, 1597 time_t new_time, 1598 char *label) 1599 { 1600 if (rt->rt_metric != metric) { 1601 /* Fix the kernel immediately if it seems the route 1602 * has gone bad, since there may be a working route that 1603 * aggregates this route. 1604 */ 1605 if (metric == HOPCNT_INFINITY) { 1606 need_kern.tv_sec = now.tv_sec; 1607 if (new_time >= now.tv_sec - EXPIRE_TIME) 1608 new_time = now.tv_sec - EXPIRE_TIME; 1609 } 1610 rt->rt_seqno = update_seqno; 1611 set_need_flash(); 1612 } 1613 1614 if (rt->rt_gate != gate) { 1615 need_kern.tv_sec = now.tv_sec; 1616 rt->rt_seqno = update_seqno; 1617 set_need_flash(); 1618 } 1619 1620 state |= (rt->rt_state & RS_SUBNET); 1621 1622 /* Keep various things from deciding ageless routes are stale. 1623 */ 1624 if (!AGE_RT(state, ifp)) 1625 new_time = now.tv_sec; 1626 1627 if (TRACEACTIONS) 1628 trace_change(rt, state, gate, router, metric, tag, ifp, 1629 new_time, 1630 label ? label : "Chg "); 1631 1632 rt->rt_state = state; 1633 rt->rt_gate = gate; 1634 rt->rt_router = router; 1635 rt->rt_metric = metric; 1636 rt->rt_tag = tag; 1637 rt->rt_ifp = ifp; 1638 rt->rt_time = new_time; 1639 } 1640 1641 1642 /* check for a better route among the spares 1643 */ 1644 static struct rt_spare * 1645 rts_better(struct rt_entry *rt) 1646 { 1647 struct rt_spare *rts, *rts1; 1648 int i; 1649 1650 /* find the best alternative among the spares */ 1651 rts = rt->rt_spares+1; 1652 for (i = NUM_SPARES, rts1 = rts+1; i > 2; i--, rts1++) { 1653 if (BETTER_LINK(rt,rts1,rts)) 1654 rts = rts1; 1655 } 1656 1657 return rts; 1658 } 1659 1660 1661 /* switch to a backup route 1662 */ 1663 void 1664 rtswitch(struct rt_entry *rt, 1665 struct rt_spare *rts) 1666 { 1667 struct rt_spare swap; 1668 char label[10]; 1669 1670 1671 /* Do not change permanent routes */ 1672 if (0 != (rt->rt_state & (RS_MHOME | RS_STATIC | RS_RDISC 1673 | RS_NET_SYN | RS_IF))) 1674 return; 1675 1676 /* find the best alternative among the spares */ 1677 if (rts == 0) 1678 rts = rts_better(rt); 1679 1680 /* Do not bother if it is not worthwhile. 1681 */ 1682 if (!BETTER_LINK(rt, rts, rt->rt_spares)) 1683 return; 1684 1685 swap = rt->rt_spares[0]; 1686 (void)sprintf(label, "Use #%d", rts - rt->rt_spares); 1687 rtchange(rt, rt->rt_state & ~(RS_NET_SYN | RS_RDISC), 1688 rts->rts_gate, rts->rts_router, rts->rts_metric, 1689 rts->rts_tag, rts->rts_ifp, rts->rts_time, label); 1690 *rts = swap; 1691 } 1692 1693 1694 void 1695 rtdelete(struct rt_entry *rt) 1696 { 1697 struct khash *k; 1698 1699 1700 if (TRACEACTIONS) 1701 trace_add_del("Del", rt); 1702 1703 k = kern_find(rt->rt_dst, rt->rt_mask, 0); 1704 if (k != 0) { 1705 k->k_state |= KS_DELETE; 1706 need_kern.tv_sec = now.tv_sec; 1707 } 1708 1709 dst_sock.sin_addr.s_addr = rt->rt_dst; 1710 mask_sock.sin_addr.s_addr = rt->rt_mask; 1711 masktrim(&mask_sock); 1712 if (rt != (struct rt_entry *)rhead->rnh_deladdr(&dst_sock, &mask_sock, 1713 rhead)) { 1714 msglog("rnh_deladdr() failed"); 1715 } else { 1716 free(rt); 1717 total_routes--; 1718 } 1719 } 1720 1721 1722 /* Get rid of a bad route, and try to switch to a replacement. 1723 */ 1724 void 1725 rtbad(struct rt_entry *rt) 1726 { 1727 /* Poison the route */ 1728 rtchange(rt, rt->rt_state & ~(RS_IF | RS_LOCAL | RS_STATIC), 1729 rt->rt_gate, rt->rt_router, HOPCNT_INFINITY, rt->rt_tag, 1730 0, rt->rt_time, 0); 1731 1732 rtswitch(rt, 0); 1733 } 1734 1735 1736 /* Junk a RS_NET_SYN or RS_LOCAL route, 1737 * unless it is needed by another interface. 1738 */ 1739 void 1740 rtbad_sub(struct rt_entry *rt) 1741 { 1742 struct interface *ifp, *ifp1; 1743 struct intnet *intnetp; 1744 u_int state; 1745 1746 1747 ifp1 = 0; 1748 state = 0; 1749 1750 if (rt->rt_state & RS_LOCAL) { 1751 /* Is this the route through loopback for the interface? 1752 * If so, see if it is used by any other interfaces, such 1753 * as a point-to-point interface with the same local address. 1754 */ 1755 for (ifp = ifnet; ifp != 0; ifp = ifp->int_next) { 1756 /* Retain it if another interface needs it. 1757 */ 1758 if (ifp->int_addr == rt->rt_ifp->int_addr) { 1759 state |= RS_LOCAL; 1760 ifp1 = ifp; 1761 break; 1762 } 1763 } 1764 1765 } 1766 1767 if (!(state & RS_LOCAL)) { 1768 /* Retain RIPv1 logical network route if there is another 1769 * interface that justifies it. 1770 */ 1771 if (rt->rt_state & RS_NET_SYN) { 1772 for (ifp = ifnet; ifp != 0; ifp = ifp->int_next) { 1773 if ((ifp->int_state & IS_NEED_NET_SYN) 1774 && rt->rt_mask == ifp->int_std_mask 1775 && rt->rt_dst == ifp->int_std_addr) { 1776 state |= RS_NET_SYN; 1777 ifp1 = ifp; 1778 break; 1779 } 1780 } 1781 } 1782 1783 /* or if there is an authority route that needs it. */ 1784 for (intnetp = intnets; 1785 intnetp != 0; 1786 intnetp = intnetp->intnet_next) { 1787 if (intnetp->intnet_addr == rt->rt_dst 1788 && intnetp->intnet_mask == rt->rt_mask) { 1789 state |= (RS_NET_SYN | RS_NET_INT); 1790 break; 1791 } 1792 } 1793 } 1794 1795 if (ifp1 != 0 || (state & RS_NET_SYN)) { 1796 rtchange(rt, ((rt->rt_state & ~(RS_NET_SYN | RS_LOCAL)) 1797 | state), 1798 rt->rt_gate, rt->rt_router, rt->rt_metric, 1799 rt->rt_tag, ifp1, rt->rt_time, 0); 1800 } else { 1801 rtbad(rt); 1802 } 1803 } 1804 1805 1806 /* Called while walking the table looking for sick interfaces 1807 * or after a time change. 1808 */ 1809 /* ARGSUSED */ 1810 int 1811 walk_bad(struct radix_node *rn, 1812 struct walkarg *w) 1813 { 1814 #define RT ((struct rt_entry *)rn) 1815 struct rt_spare *rts; 1816 int i; 1817 time_t new_time; 1818 1819 1820 /* fix any spare routes through the interface 1821 */ 1822 rts = RT->rt_spares; 1823 for (i = NUM_SPARES; i != 1; i--) { 1824 rts++; 1825 1826 if (rts->rts_ifp != 0 1827 && (rts->rts_ifp->int_state & IS_BROKE)) { 1828 /* mark the spare route to be deleted immediately */ 1829 new_time = rts->rts_time; 1830 if (new_time >= now_garbage) 1831 new_time = now_garbage-1; 1832 trace_upslot(RT, rts, rts->rts_gate, 1833 rts->rts_router, 0, 1834 HOPCNT_INFINITY, rts->rts_tag, 1835 new_time); 1836 rts->rts_ifp = 0; 1837 rts->rts_metric = HOPCNT_INFINITY; 1838 rts->rts_time = new_time; 1839 } 1840 } 1841 1842 /* Deal with the main route 1843 */ 1844 /* finished if it has been handled before or if its interface is ok 1845 */ 1846 if (RT->rt_ifp == 0 || !(RT->rt_ifp->int_state & IS_BROKE)) 1847 return 0; 1848 1849 /* Bad routes for other than interfaces are easy. 1850 */ 1851 if (0 == (RT->rt_state & (RS_IF | RS_NET_SYN | RS_LOCAL))) { 1852 rtbad(RT); 1853 return 0; 1854 } 1855 1856 rtbad_sub(RT); 1857 return 0; 1858 #undef RT 1859 } 1860 1861 1862 /* Check the age of an individual route. 1863 */ 1864 /* ARGSUSED */ 1865 static int 1866 walk_age(struct radix_node *rn, 1867 struct walkarg *w) 1868 { 1869 #define RT ((struct rt_entry *)rn) 1870 struct interface *ifp; 1871 struct rt_spare *rts; 1872 int i; 1873 1874 1875 /* age all of the spare routes, including the primary route 1876 * currently in use 1877 */ 1878 rts = RT->rt_spares; 1879 for (i = NUM_SPARES; i != 0; i--, rts++) { 1880 1881 ifp = rts->rts_ifp; 1882 if (i == NUM_SPARES) { 1883 if (!AGE_RT(RT->rt_state, ifp)) { 1884 /* Keep various things from deciding ageless 1885 * routes are stale 1886 */ 1887 rts->rts_time = now.tv_sec; 1888 continue; 1889 } 1890 1891 /* forget RIP routes after RIP has been turned off. 1892 */ 1893 if (rip_sock < 0) { 1894 rtdelete(RT); 1895 return 0; 1896 } 1897 } 1898 1899 /* age failing routes 1900 */ 1901 if (age_bad_gate == rts->rts_gate 1902 && rts->rts_time >= now_stale) { 1903 rts->rts_time -= SUPPLY_INTERVAL; 1904 } 1905 1906 /* trash the spare routes when they go bad */ 1907 if (rts->rts_metric < HOPCNT_INFINITY 1908 && now_garbage > rts->rts_time) { 1909 trace_upslot(RT, rts, rts->rts_gate, 1910 rts->rts_router, rts->rts_ifp, 1911 HOPCNT_INFINITY, rts->rts_tag, 1912 rts->rts_time); 1913 rts->rts_metric = HOPCNT_INFINITY; 1914 } 1915 } 1916 1917 1918 /* finished if the active route is still fresh */ 1919 if (now_stale <= RT->rt_time) 1920 return 0; 1921 1922 /* try to switch to an alternative */ 1923 rtswitch(RT, 0); 1924 1925 /* Delete a dead route after it has been publically mourned. */ 1926 if (now_garbage > RT->rt_time) { 1927 rtdelete(RT); 1928 return 0; 1929 } 1930 1931 /* Start poisoning a bad route before deleting it. */ 1932 if (now.tv_sec - RT->rt_time > EXPIRE_TIME) 1933 rtchange(RT, RT->rt_state, RT->rt_gate, RT->rt_router, 1934 HOPCNT_INFINITY, RT->rt_tag, RT->rt_ifp, 1935 RT->rt_time, 0); 1936 return 0; 1937 } 1938 1939 1940 /* Watch for dead routes and interfaces. 1941 */ 1942 void 1943 age(naddr bad_gate) 1944 { 1945 struct interface *ifp; 1946 int need_query = 0; 1947 1948 /* If not listening to RIP, there is no need to age the routes in 1949 * the table. 1950 */ 1951 age_timer.tv_sec = (now.tv_sec 1952 + ((rip_sock < 0) ? NEVER : SUPPLY_INTERVAL)); 1953 1954 /* Check for dead IS_REMOTE interfaces by timing their 1955 * transmissions. 1956 */ 1957 for (ifp = ifnet; ifp; ifp = ifp->int_next) { 1958 if (!(ifp->int_state & IS_REMOTE)) 1959 continue; 1960 1961 /* ignore unreachable remote interfaces */ 1962 if (!check_remote(ifp)) 1963 continue; 1964 /* Restore remote interface that has become reachable 1965 */ 1966 if (ifp->int_state & IS_BROKE) 1967 if_ok(ifp, "remote "); 1968 1969 if (ifp->int_act_time != NEVER 1970 && now.tv_sec - ifp->int_act_time > EXPIRE_TIME) { 1971 msglog("remote interface %s to %s timed out after" 1972 " %d:%d", 1973 ifp->int_name, 1974 naddr_ntoa(ifp->int_dstaddr), 1975 (now.tv_sec - ifp->int_act_time)/60, 1976 (now.tv_sec - ifp->int_act_time)%60); 1977 if_sick(ifp); 1978 } 1979 1980 /* If we have not heard from the other router 1981 * recently, ask it. 1982 */ 1983 if (now.tv_sec >= ifp->int_query_time) { 1984 ifp->int_query_time = NEVER; 1985 need_query = 1; 1986 } 1987 } 1988 1989 /* Age routes. */ 1990 age_bad_gate = bad_gate; 1991 (void)rn_walktree(rhead, walk_age, 0); 1992 1993 /* Update the kernel routing table. */ 1994 fix_kern(); 1995 1996 /* poke reticent remote gateways */ 1997 if (need_query) 1998 rip_query(); 1999 } 2000