1 /* $NetBSD: table.c,v 1.4 1997/02/03 22:03:08 christos Exp $ */ 2 3 /* 4 * Copyright (c) 1983, 1988, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. All advertising materials mentioning features or use of this software 16 * must display the following acknowledgement: 17 * This product includes software developed by the University of 18 * California, Berkeley and its contributors. 19 * 4. Neither the name of the University nor the names of its contributors 20 * may be used to endorse or promote products derived from this software 21 * without specific prior written permission. 22 * 23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 33 * SUCH DAMAGE. 34 */ 35 36 #if !defined(lint) && !defined(sgi) && !defined(__NetBSD__) 37 static char sccsid[] = "@(#)tables.c 8.1 (Berkeley) 6/5/93"; 38 #elif defined(__NetBSD__) 39 static char rcsid[] = "$NetBSD: table.c,v 1.4 1997/02/03 22:03:08 christos Exp $"; 40 #endif 41 42 #include "defs.h" 43 44 static struct rt_spare *rts_better(struct rt_entry *); 45 static struct rt_spare rts_empty = {0,0,0,HOPCNT_INFINITY,0,0}; 46 47 struct radix_node_head *rhead; /* root of the radix tree */ 48 49 int need_flash = 1; /* flash update needed 50 * start =1 to suppress the 1st 51 */ 52 53 struct timeval age_timer; /* next check of old routes */ 54 struct timeval need_kern = { /* need to update kernel table */ 55 EPOCH+MIN_WAITTIME-1 56 }; 57 58 int stopint; 59 60 int total_routes; 61 62 /* zap any old routes through this gateway */ 63 naddr age_bad_gate; 64 65 66 /* It is desirable to "aggregate" routes, to combine differing routes of 67 * the same metric and next hop into a common route with a smaller netmask 68 * or to suppress redundant routes, routes that add no information to 69 * routes with smaller netmasks. 70 * 71 * A route is redundant if and only if any and all routes with smaller 72 * but matching netmasks and nets are the same. Since routes are 73 * kept sorted in the radix tree, redundant routes always come second. 74 * 75 * There are two kinds of aggregations. First, two routes of the same bit 76 * mask and differing only in the least significant bit of the network 77 * number can be combined into a single route with a coarser mask. 78 * 79 * Second, a route can be suppressed in favor of another route with a more 80 * coarse mask provided no incompatible routes with intermediate masks 81 * are present. The second kind of aggregation involves suppressing routes. 82 * A route must not be suppressed if an incompatible route exists with 83 * an intermediate mask, since the suppressed route would be covered 84 * by the intermediate. 85 * 86 * This code relies on the radix tree walk encountering routes 87 * sorted first by address, with the smallest address first. 88 */ 89 90 struct ag_info ag_slots[NUM_AG_SLOTS], *ag_avail, *ag_corsest, *ag_finest; 91 92 /* #define DEBUG_AG */ 93 #ifdef DEBUG_AG 94 #define CHECK_AG() {int acnt = 0; struct ag_info *cag; \ 95 for (cag = ag_avail; cag != 0; cag = cag->ag_fine) \ 96 acnt++; \ 97 for (cag = ag_corsest; cag != 0; cag = cag->ag_fine) \ 98 acnt++; \ 99 if (acnt != NUM_AG_SLOTS) { \ 100 (void)fflush(stderr); \ 101 abort(); \ 102 } \ 103 } 104 #else 105 #define CHECK_AG() 106 #endif 107 108 109 /* Output the contents of an aggregation table slot. 110 * This function must always be immediately followed with the deletion 111 * of the target slot. 112 */ 113 static void 114 ag_out(struct ag_info *ag, 115 void (*out)(struct ag_info *)) 116 { 117 struct ag_info *ag_cors; 118 naddr bit; 119 120 121 /* If we output both the even and odd twins, then the immediate parent, 122 * if it is present, is redundant, unless the parent manages to 123 * aggregate into something coarser. 124 * On successive calls, this code detects the even and odd twins, 125 * and marks the parent. 126 * 127 * Note that the order in which the radix tree code emits routes 128 * ensures that the twins are seen before the parent is emitted. 129 */ 130 ag_cors = ag->ag_cors; 131 if (ag_cors != 0 132 && ag_cors->ag_mask == ag->ag_mask<<1 133 && ag_cors->ag_dst_h == (ag->ag_dst_h & ag_cors->ag_mask)) { 134 ag_cors->ag_state |= ((ag_cors->ag_dst_h == ag->ag_dst_h) 135 ? AGS_REDUN0 136 : AGS_REDUN1); 137 } 138 139 /* Skip it if this route is itself redundant. 140 * 141 * It is ok to change the contents of the slot here, since it is 142 * always deleted next. 143 */ 144 if (ag->ag_state & AGS_REDUN0) { 145 if (ag->ag_state & AGS_REDUN1) 146 return; 147 bit = (-ag->ag_mask) >> 1; 148 ag->ag_dst_h |= bit; 149 ag->ag_mask |= bit; 150 151 } else if (ag->ag_state & AGS_REDUN1) { 152 bit = (-ag->ag_mask) >> 1; 153 ag->ag_mask |= bit; 154 } 155 out(ag); 156 } 157 158 159 static void 160 ag_del(struct ag_info *ag) 161 { 162 CHECK_AG(); 163 164 if (ag->ag_cors == 0) 165 ag_corsest = ag->ag_fine; 166 else 167 ag->ag_cors->ag_fine = ag->ag_fine; 168 169 if (ag->ag_fine == 0) 170 ag_finest = ag->ag_cors; 171 else 172 ag->ag_fine->ag_cors = ag->ag_cors; 173 174 ag->ag_fine = ag_avail; 175 ag_avail = ag; 176 177 CHECK_AG(); 178 } 179 180 181 /* Flush routes waiting for aggretation. 182 * This must not suppress a route unless it is known that among all 183 * routes with coarser masks that match it, the one with the longest 184 * mask is appropriate. This is ensured by scanning the routes 185 * in lexical order, and with the most restritive mask first 186 * among routes to the same destination. 187 */ 188 void 189 ag_flush(naddr lim_dst_h, /* flush routes to here */ 190 naddr lim_mask, /* matching this mask */ 191 void (*out)(struct ag_info *)) 192 { 193 struct ag_info *ag, *ag_cors; 194 naddr dst_h; 195 196 197 for (ag = ag_finest; 198 ag != 0 && ag->ag_mask >= lim_mask; 199 ag = ag_cors) { 200 ag_cors = ag->ag_cors; 201 202 /* work on only the specified routes */ 203 dst_h = ag->ag_dst_h; 204 if ((dst_h & lim_mask) != lim_dst_h) 205 continue; 206 207 if (!(ag->ag_state & AGS_SUPPRESS)) 208 ag_out(ag, out); 209 210 else for ( ; ; ag_cors = ag_cors->ag_cors) { 211 /* Look for a route that can suppress the 212 * current route */ 213 if (ag_cors == 0) { 214 /* failed, so output it and look for 215 * another route to work on 216 */ 217 ag_out(ag, out); 218 break; 219 } 220 221 if ((dst_h & ag_cors->ag_mask) == ag_cors->ag_dst_h) { 222 /* We found a route with a coarser mask that 223 * aggregates the current target. 224 * 225 * If it has a different next hop, it 226 * cannot replace the target, so output 227 * the target. 228 */ 229 if (ag->ag_gate != ag_cors->ag_gate 230 && !(ag->ag_state & AGS_FINE_GATE) 231 && !(ag_cors->ag_state & AGS_CORS_GATE)) { 232 ag_out(ag, out); 233 break; 234 } 235 236 /* If the coarse route has a good enough 237 * metric, it suppresses the target. 238 */ 239 if (ag_cors->ag_pref <= ag->ag_pref) { 240 if (ag_cors->ag_seqno > ag->ag_seqno) 241 ag_cors->ag_seqno = ag->ag_seqno; 242 if (AG_IS_REDUN(ag->ag_state) 243 && ag_cors->ag_mask==ag->ag_mask<<1) { 244 if (ag_cors->ag_dst_h == dst_h) 245 ag_cors->ag_state |= AGS_REDUN0; 246 else 247 ag_cors->ag_state |= AGS_REDUN1; 248 } 249 if (ag->ag_tag != ag_cors->ag_tag) 250 ag_cors->ag_tag = 0; 251 if (ag->ag_nhop != ag_cors->ag_nhop) 252 ag_cors->ag_nhop = 0; 253 break; 254 } 255 } 256 } 257 258 /* That route has either been output or suppressed */ 259 ag_cors = ag->ag_cors; 260 ag_del(ag); 261 } 262 263 CHECK_AG(); 264 } 265 266 267 /* Try to aggregate a route with previous routes. 268 */ 269 void 270 ag_check(naddr dst, 271 naddr mask, 272 naddr gate, 273 naddr nhop, 274 char metric, 275 char pref, 276 u_int seqno, 277 u_short tag, 278 u_short state, 279 void (*out)(struct ag_info *)) /* output using this */ 280 { 281 struct ag_info *ag, *nag, *ag_cors; 282 naddr xaddr; 283 int x; 284 285 NTOHL(dst); 286 287 /* Punt non-contiguous subnet masks. 288 * 289 * (X & -X) contains a single bit if and only if X is a power of 2. 290 * (X + (X & -X)) == 0 if and only if X is a power of 2. 291 */ 292 if ((mask & -mask) + mask != 0) { 293 struct ag_info nc_ag; 294 295 nc_ag.ag_dst_h = dst; 296 nc_ag.ag_mask = mask; 297 nc_ag.ag_gate = gate; 298 nc_ag.ag_nhop = nhop; 299 nc_ag.ag_metric = metric; 300 nc_ag.ag_pref = pref; 301 nc_ag.ag_tag = tag; 302 nc_ag.ag_state = state; 303 nc_ag.ag_seqno = seqno; 304 out(&nc_ag); 305 return; 306 } 307 308 /* Search for the right slot in the aggregation table. 309 */ 310 ag_cors = 0; 311 ag = ag_corsest; 312 while (ag != 0) { 313 if (ag->ag_mask >= mask) 314 break; 315 316 /* Suppress old routes (i.e. combine with compatible routes 317 * with coarser masks) as we look for the right slot in the 318 * aggregation table for the new route. 319 * A route to an address less than the current destination 320 * will not be affected by the current route or any route 321 * seen hereafter. That means it is safe to suppress it. 322 * This check keeps poor routes (eg. with large hop counts) 323 * from preventing suppresion of finer routes. 324 */ 325 if (ag_cors != 0 326 && ag->ag_dst_h < dst 327 && (ag->ag_state & AGS_SUPPRESS) 328 && ag_cors->ag_pref <= ag->ag_pref 329 && (ag->ag_dst_h & ag_cors->ag_mask) == ag_cors->ag_dst_h 330 && (ag_cors->ag_gate == ag->ag_gate 331 || (ag->ag_state & AGS_FINE_GATE) 332 || (ag_cors->ag_state & AGS_CORS_GATE))) { 333 if (ag_cors->ag_seqno > ag->ag_seqno) 334 ag_cors->ag_seqno = ag->ag_seqno; 335 if (AG_IS_REDUN(ag->ag_state) 336 && ag_cors->ag_mask==ag->ag_mask<<1) { 337 if (ag_cors->ag_dst_h == dst) 338 ag_cors->ag_state |= AGS_REDUN0; 339 else 340 ag_cors->ag_state |= AGS_REDUN1; 341 } 342 if (ag->ag_tag != ag_cors->ag_tag) 343 ag_cors->ag_tag = 0; 344 if (ag->ag_nhop != ag_cors->ag_nhop) 345 ag_cors->ag_nhop = 0; 346 ag_del(ag); 347 CHECK_AG(); 348 } else { 349 ag_cors = ag; 350 } 351 ag = ag_cors->ag_fine; 352 } 353 354 /* If we find the even/odd twin of the new route, and if the 355 * masks and so forth are equal, we can aggregate them. 356 * We can probably promote one of the pair. 357 * 358 * Since the routes are encountered in lexical order, 359 * the new route must be odd. However, the second or later 360 * times around this loop, it could be the even twin promoted 361 * from the even/odd pair of twins of the finer route. 362 */ 363 while (ag != 0 364 && ag->ag_mask == mask 365 && ((ag->ag_dst_h ^ dst) & (mask<<1)) == 0) { 366 367 /* Here we know the target route and the route in the current 368 * slot have the same netmasks and differ by at most the 369 * last bit. They are either for the same destination, or 370 * for an even/odd pair of destinations. 371 */ 372 if (ag->ag_dst_h == dst) { 373 /* We have two routes to the same destination. 374 * Routes are encountered in lexical order, so a 375 * route is never promoted until the parent route is 376 * already present. So we know that the new route is 377 * a promoted pair and the route already in the slot 378 * is the explicit route. 379 * 380 * Prefer the best route if their metrics differ, 381 * or the promoted one if not, following a sort 382 * of longest-match rule. 383 */ 384 if (pref <= ag->ag_pref) { 385 ag->ag_gate = gate; 386 ag->ag_nhop = nhop; 387 ag->ag_tag = tag; 388 ag->ag_metric = metric; 389 ag->ag_pref = pref; 390 x = ag->ag_state; 391 ag->ag_state = state; 392 state = x; 393 } 394 395 /* The sequence number controls flash updating, 396 * and should be the smaller of the two. 397 */ 398 if (ag->ag_seqno > seqno) 399 ag->ag_seqno = seqno; 400 401 /* some bits are set if they are set on either route */ 402 ag->ag_state |= (state & (AGS_PROMOTE_EITHER 403 | AGS_REDUN0 | AGS_REDUN1)); 404 return; 405 } 406 407 /* If one of the routes can be promoted and the other can 408 * be suppressed, it may be possible to combine them or 409 * worthwhile to promote one. 410 * 411 * Note that any route that can be promoted is always 412 * marked to be eligible to be suppressed. 413 */ 414 if (!((state & AGS_PROMOTE) 415 && (ag->ag_state & AGS_SUPPRESS)) 416 && !((ag->ag_state & AGS_PROMOTE) 417 && (state & AGS_SUPPRESS))) 418 break; 419 420 /* A pair of even/odd twin routes can be combined 421 * if either is redundant, or if they are via the 422 * same gateway and have the same metric. 423 */ 424 if (AG_IS_REDUN(ag->ag_state) 425 || AG_IS_REDUN(state) 426 || (ag->ag_gate == gate 427 && ag->ag_pref == pref 428 && (state & ag->ag_state & AGS_PROMOTE) != 0)) { 429 430 /* We have both the even and odd pairs. 431 * Since the routes are encountered in order, 432 * the route in the slot must be the even twin. 433 * 434 * Combine and promote the pair of routes. 435 */ 436 if (seqno > ag->ag_seqno) 437 seqno = ag->ag_seqno; 438 if (!AG_IS_REDUN(state)) 439 state &= ~AGS_REDUN1; 440 if (AG_IS_REDUN(ag->ag_state)) 441 state |= AGS_REDUN0; 442 else 443 state &= ~AGS_REDUN0; 444 state |= (ag->ag_state & AGS_PROMOTE_EITHER); 445 if (ag->ag_tag != tag) 446 tag = 0; 447 if (ag->ag_nhop != nhop) 448 nhop = 0; 449 450 /* Get rid of the even twin that was already 451 * in the slot. 452 */ 453 ag_del(ag); 454 455 } else if (ag->ag_pref >= pref 456 && (ag->ag_state & AGS_PROMOTE)) { 457 /* If we cannot combine the pair, maybe the route 458 * with the worse metric can be promoted. 459 * 460 * Promote the old, even twin, by giving its slot 461 * in the table to the new, odd twin. 462 */ 463 ag->ag_dst_h = dst; 464 465 xaddr = ag->ag_gate; 466 ag->ag_gate = gate; 467 gate = xaddr; 468 469 xaddr = ag->ag_nhop; 470 ag->ag_nhop = nhop; 471 nhop = xaddr; 472 473 x = ag->ag_tag; 474 ag->ag_tag = tag; 475 tag = x; 476 477 x = ag->ag_state; 478 ag->ag_state = state; 479 state = x; 480 if (!AG_IS_REDUN(state)) 481 state &= ~AGS_REDUN0; 482 483 x = ag->ag_metric; 484 ag->ag_metric = metric; 485 metric = x; 486 487 x = ag->ag_pref; 488 ag->ag_pref = pref; 489 pref = x; 490 491 if (seqno >= ag->ag_seqno) 492 seqno = ag->ag_seqno; 493 else 494 ag->ag_seqno = seqno; 495 496 } else { 497 if (!(state & AGS_PROMOTE)) 498 break; /* cannot promote either twin */ 499 500 /* promote the new, odd twin by shaving its 501 * mask and address. 502 */ 503 if (seqno > ag->ag_seqno) 504 seqno = ag->ag_seqno; 505 else 506 ag->ag_seqno = seqno; 507 if (!AG_IS_REDUN(state)) 508 state &= ~AGS_REDUN1; 509 } 510 511 mask <<= 1; 512 dst &= mask; 513 514 if (ag_cors == 0) { 515 ag = ag_corsest; 516 break; 517 } 518 ag = ag_cors; 519 ag_cors = ag->ag_cors; 520 } 521 522 /* When we can no longer promote and combine routes, 523 * flush the old route in the target slot. Also flush 524 * any finer routes that we know will never be aggregated by 525 * the new route. 526 * 527 * In case we moved toward coarser masks, 528 * get back where we belong 529 */ 530 if (ag != 0 531 && ag->ag_mask < mask) { 532 ag_cors = ag; 533 ag = ag->ag_fine; 534 } 535 536 /* Empty the target slot 537 */ 538 if (ag != 0 && ag->ag_mask == mask) { 539 ag_flush(ag->ag_dst_h, ag->ag_mask, out); 540 ag = (ag_cors == 0) ? ag_corsest : ag_cors->ag_fine; 541 } 542 543 #ifdef DEBUG_AG 544 (void)fflush(stderr); 545 if (ag == 0 && ag_cors != ag_finest) 546 abort(); 547 if (ag_cors == 0 && ag != ag_corsest) 548 abort(); 549 if (ag != 0 && ag->ag_cors != ag_cors) 550 abort(); 551 if (ag_cors != 0 && ag_cors->ag_fine != ag) 552 abort(); 553 CHECK_AG(); 554 #endif 555 556 /* Save the new route on the end of the table. 557 */ 558 nag = ag_avail; 559 ag_avail = nag->ag_fine; 560 561 nag->ag_dst_h = dst; 562 nag->ag_mask = mask; 563 nag->ag_gate = gate; 564 nag->ag_nhop = nhop; 565 nag->ag_metric = metric; 566 nag->ag_pref = pref; 567 nag->ag_tag = tag; 568 nag->ag_state = state; 569 nag->ag_seqno = seqno; 570 571 nag->ag_fine = ag; 572 if (ag != 0) 573 ag->ag_cors = nag; 574 else 575 ag_finest = nag; 576 nag->ag_cors = ag_cors; 577 if (ag_cors == 0) 578 ag_corsest = nag; 579 else 580 ag_cors->ag_fine = nag; 581 CHECK_AG(); 582 } 583 584 585 static char * 586 rtm_type_name(u_char type) 587 { 588 static char *rtm_types[] = { 589 "RTM_ADD", 590 "RTM_DELETE", 591 "RTM_CHANGE", 592 "RTM_GET", 593 "RTM_LOSING", 594 "RTM_REDIRECT", 595 "RTM_MISS", 596 "RTM_LOCK", 597 "RTM_OLDADD", 598 "RTM_OLDDEL", 599 "RTM_RESOLVE", 600 "RTM_NEWADDR", 601 "RTM_DELADDR", 602 "RTM_IFINFO" 603 }; 604 static char name0[10]; 605 606 607 if (type > sizeof(rtm_types)/sizeof(rtm_types[0]) 608 || type == 0) { 609 sprintf(name0, "RTM type %#x", type); 610 return name0; 611 } else { 612 return rtm_types[type-1]; 613 } 614 } 615 616 617 /* Trim a mask in a sockaddr 618 * Produce a length of 0 for an address of 0. 619 * Otherwise produce the index of the first zero byte. 620 */ 621 void 622 #ifdef _HAVE_SIN_LEN 623 masktrim(struct sockaddr_in *ap) 624 #else 625 masktrim(struct sockaddr_in_new *ap) 626 #endif 627 { 628 register char *cp; 629 630 if (ap->sin_addr.s_addr == 0) { 631 ap->sin_len = 0; 632 return; 633 } 634 cp = (char *)(&ap->sin_addr.s_addr+1); 635 while (*--cp == 0) 636 continue; 637 ap->sin_len = cp - (char*)ap + 1; 638 } 639 640 641 /* Tell the kernel to add, delete or change a route 642 */ 643 static void 644 rtioctl(int action, /* RTM_DELETE, etc */ 645 naddr dst, 646 naddr gate, 647 naddr mask, 648 int metric, 649 int flags) 650 { 651 struct { 652 struct rt_msghdr w_rtm; 653 struct sockaddr_in w_dst; 654 struct sockaddr_in w_gate; 655 #ifdef _HAVE_SA_LEN 656 struct sockaddr_in w_mask; 657 #else 658 struct sockaddr_in_new w_mask; 659 #endif 660 } w; 661 long cc; 662 # define PAT " %-10s %s metric=%d flags=%#x" 663 # define ARGS rtm_type_name(action), rtname(dst,mask,gate), metric, flags 664 665 again: 666 bzero(&w, sizeof(w)); 667 w.w_rtm.rtm_msglen = sizeof(w); 668 w.w_rtm.rtm_version = RTM_VERSION; 669 w.w_rtm.rtm_type = action; 670 w.w_rtm.rtm_flags = flags; 671 w.w_rtm.rtm_seq = ++rt_sock_seqno; 672 w.w_rtm.rtm_addrs = RTA_DST|RTA_GATEWAY; 673 if (metric != 0) { 674 w.w_rtm.rtm_rmx.rmx_hopcount = metric; 675 w.w_rtm.rtm_inits |= RTV_HOPCOUNT; 676 } 677 w.w_dst.sin_family = AF_INET; 678 w.w_dst.sin_addr.s_addr = dst; 679 w.w_gate.sin_family = AF_INET; 680 w.w_gate.sin_addr.s_addr = gate; 681 #ifdef _HAVE_SA_LEN 682 w.w_dst.sin_len = sizeof(w.w_dst); 683 w.w_gate.sin_len = sizeof(w.w_gate); 684 #endif 685 if (mask == HOST_MASK) { 686 w.w_rtm.rtm_flags |= RTF_HOST; 687 w.w_rtm.rtm_msglen -= sizeof(w.w_mask); 688 } else { 689 w.w_rtm.rtm_addrs |= RTA_NETMASK; 690 w.w_mask.sin_addr.s_addr = htonl(mask); 691 #ifdef _HAVE_SA_LEN 692 masktrim(&w.w_mask); 693 if (w.w_mask.sin_len == 0) 694 w.w_mask.sin_len = sizeof(long); 695 w.w_rtm.rtm_msglen -= (sizeof(w.w_mask) - w.w_mask.sin_len); 696 #endif 697 } 698 699 #ifndef NO_INSTALL 700 cc = write(rt_sock, &w, w.w_rtm.rtm_msglen); 701 if (cc < 0) { 702 if (errno == ESRCH 703 && (action == RTM_CHANGE || action == RTM_DELETE)) { 704 trace_act("route disappeared before" PAT, ARGS); 705 if (action == RTM_CHANGE) { 706 action = RTM_ADD; 707 goto again; 708 } 709 return; 710 } 711 msglog("write(rt_sock)" PAT ": ", ARGS, strerror(errno)); 712 return; 713 } else if (cc != w.w_rtm.rtm_msglen) { 714 msglog("write(rt_sock) wrote %d instead of %d for" PAT, 715 cc, w.w_rtm.rtm_msglen, ARGS); 716 return; 717 } 718 #endif 719 if (TRACEKERNEL) 720 trace_kernel("write kernel" PAT, ARGS); 721 #undef PAT 722 #undef ARGS 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 *)malloc(sizeof(*k)); 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 /* operate on a kernel entry 816 */ 817 static void 818 kern_ioctl(struct khash *k, 819 int action, /* RTM_DELETE, etc */ 820 int flags) 821 822 { 823 switch (action) { 824 case RTM_DELETE: 825 k->k_state &= ~KS_DYNAMIC; 826 if (k->k_state & KS_DELETED) 827 return; 828 k->k_state |= KS_DELETED; 829 break; 830 case RTM_ADD: 831 k->k_state &= ~KS_DELETED; 832 break; 833 case RTM_CHANGE: 834 if (k->k_state & KS_DELETED) { 835 action = RTM_ADD; 836 k->k_state &= ~KS_DELETED; 837 } 838 break; 839 } 840 841 rtioctl(action, k->k_dst, k->k_gate, k->k_mask, k->k_metric, flags); 842 } 843 844 845 /* add a route the kernel told us 846 */ 847 static void 848 rtm_add(struct rt_msghdr *rtm, 849 struct rt_addrinfo *info, 850 time_t keep) 851 { 852 struct khash *k; 853 struct interface *ifp; 854 naddr mask; 855 856 857 if (rtm->rtm_flags & RTF_HOST) { 858 mask = HOST_MASK; 859 } else if (INFO_MASK(info) != 0) { 860 mask = ntohl(S_ADDR(INFO_MASK(info))); 861 } else { 862 msglog("ignore %s without mask", rtm_type_name(rtm->rtm_type)); 863 return; 864 } 865 866 if (INFO_GATE(info) == 0 867 || INFO_GATE(info)->sa_family != AF_INET) { 868 msglog("ignore %s without gateway", 869 rtm_type_name(rtm->rtm_type)); 870 return; 871 } 872 873 k = kern_add(S_ADDR(INFO_DST(info)), mask); 874 if (k->k_state & KS_NEW) 875 k->k_keep = now.tv_sec+keep; 876 k->k_gate = S_ADDR(INFO_GATE(info)); 877 k->k_metric = rtm->rtm_rmx.rmx_hopcount; 878 if (k->k_metric < 0) 879 k->k_metric = 0; 880 else if (k->k_metric > HOPCNT_INFINITY) 881 k->k_metric = HOPCNT_INFINITY; 882 k->k_state &= ~(KS_DELETED | KS_GATEWAY | KS_STATIC | KS_NEW); 883 if (rtm->rtm_flags & RTF_GATEWAY) 884 k->k_state |= KS_GATEWAY; 885 if (rtm->rtm_flags & RTF_STATIC) 886 k->k_state |= KS_STATIC; 887 888 if (0 != (rtm->rtm_flags & (RTF_DYNAMIC | RTF_MODIFIED))) { 889 if (INFO_AUTHOR(info) != 0 890 && INFO_AUTHOR(info)->sa_family == AF_INET) 891 ifp = iflookup(S_ADDR(INFO_AUTHOR(info))); 892 else 893 ifp = 0; 894 if (supplier 895 && (ifp == 0 || !(ifp->int_state & IS_REDIRECT_OK))) { 896 /* Routers are not supposed to listen to redirects, 897 * so delete it if it came via an unknown interface 898 * or the interface does not have special permission. 899 */ 900 k->k_state &= ~KS_DYNAMIC; 901 k->k_state |= KS_DELETE; 902 LIM_SEC(need_kern, 0); 903 trace_act("mark for deletion redirected %s --> %s" 904 " via %s", 905 addrname(k->k_dst, k->k_mask, 0), 906 naddr_ntoa(k->k_gate), 907 ifp ? ifp->int_name : "unknown interface"); 908 } else { 909 k->k_state |= KS_DYNAMIC; 910 k->k_redirect_time = now.tv_sec; 911 trace_act("accept redirected %s --> %s via %s", 912 addrname(k->k_dst, k->k_mask, 0), 913 naddr_ntoa(k->k_gate), 914 ifp ? ifp->int_name : "unknown interface"); 915 } 916 return; 917 } 918 919 /* If it is not a static route, quit until the next comparison 920 * between the kernel and daemon tables, when it will be deleted. 921 */ 922 if (!(k->k_state & KS_STATIC)) { 923 k->k_state |= KS_DELETE; 924 LIM_SEC(need_kern, k->k_keep); 925 return; 926 } 927 928 /* Put static routes with real metrics into the daemon table so 929 * they can be advertised. 930 * 931 * Find the interface toward the gateway. 932 */ 933 ifp = iflookup(k->k_gate); 934 if (ifp == 0) 935 msglog("static route %s --> %s impossibly lacks ifp", 936 addrname(S_ADDR(INFO_DST(info)), mask, 0), 937 naddr_ntoa(k->k_gate)); 938 939 kern_check_static(k, ifp); 940 } 941 942 943 /* deal with packet loss 944 */ 945 static void 946 rtm_lose(struct rt_msghdr *rtm, 947 struct rt_addrinfo *info) 948 { 949 if (INFO_GATE(info) == 0 950 || INFO_GATE(info)->sa_family != AF_INET) { 951 trace_act("ignore %s without gateway", 952 rtm_type_name(rtm->rtm_type)); 953 return; 954 } 955 956 if (!supplier) 957 rdisc_age(S_ADDR(INFO_GATE(info))); 958 959 age(S_ADDR(INFO_GATE(info))); 960 } 961 962 963 /* Clean the kernel table by copying it to the daemon image. 964 * Eventually the daemon will delete any extra routes. 965 */ 966 void 967 flush_kern(void) 968 { 969 size_t needed; 970 int mib[6]; 971 char *buf, *next, *lim; 972 struct rt_msghdr *rtm; 973 struct interface *ifp; 974 static struct sockaddr_in gate_sa; 975 struct rt_addrinfo info; 976 977 978 mib[0] = CTL_NET; 979 mib[1] = PF_ROUTE; 980 mib[2] = 0; /* protocol */ 981 mib[3] = 0; /* wildcard address family */ 982 mib[4] = NET_RT_DUMP; 983 mib[5] = 0; /* no flags */ 984 if (sysctl(mib, 6, 0, &needed, 0, 0) < 0) { 985 DBGERR(1,"RT_DUMP-sysctl-estimate"); 986 return; 987 } 988 buf = malloc(needed); 989 if (sysctl(mib, 6, buf, &needed, 0, 0) < 0) 990 BADERR(1,"RT_DUMP"); 991 lim = buf + needed; 992 for (next = buf; next < lim; next += rtm->rtm_msglen) { 993 rtm = (struct rt_msghdr *)next; 994 995 rt_xaddrs(&info, 996 (struct sockaddr *)(rtm+1), 997 (struct sockaddr *)(next + rtm->rtm_msglen), 998 rtm->rtm_addrs); 999 1000 if (INFO_DST(&info) == 0 1001 || INFO_DST(&info)->sa_family != AF_INET) 1002 continue; 1003 1004 /* ignore ARP table entries on systems with a merged route 1005 * and ARP table. 1006 */ 1007 if (rtm->rtm_flags & RTF_LLINFO) 1008 continue; 1009 1010 if (INFO_GATE(&info) == 0) 1011 continue; 1012 if (INFO_GATE(&info)->sa_family != AF_INET) { 1013 if (INFO_GATE(&info)->sa_family != AF_LINK) 1014 continue; 1015 ifp = ifwithindex(((struct sockaddr_dl *) 1016 INFO_GATE(&info))->sdl_index, 0); 1017 if (ifp == 0) 1018 continue; 1019 if ((ifp->int_if_flags & IFF_POINTOPOINT) 1020 || S_ADDR(INFO_DST(&info)) == ifp->int_addr) 1021 gate_sa.sin_addr.s_addr = ifp->int_addr; 1022 else 1023 gate_sa.sin_addr.s_addr = htonl(ifp->int_net); 1024 #ifdef _HAVE_SA_LEN 1025 gate_sa.sin_len = sizeof(gate_sa); 1026 #endif 1027 gate_sa.sin_family = AF_INET; 1028 INFO_GATE(&info) = (struct sockaddr *)&gate_sa; 1029 } 1030 1031 /* ignore multicast addresses 1032 */ 1033 if (IN_MULTICAST(ntohl(S_ADDR(INFO_DST(&info))))) 1034 continue; 1035 1036 /* Note static routes and interface routes, and also 1037 * preload the image of the kernel table so that 1038 * we can later clean it, as well as avoid making 1039 * unneeded changes. Keep the old kernel routes for a 1040 * few seconds to allow a RIP or router-discovery 1041 * response to be heard. 1042 */ 1043 rtm_add(rtm,&info,MIN_WAITTIME); 1044 } 1045 free(buf); 1046 } 1047 1048 1049 /* Listen to announcements from the kernel 1050 */ 1051 void 1052 read_rt(void) 1053 { 1054 long cc; 1055 struct interface *ifp; 1056 naddr mask; 1057 union { 1058 struct { 1059 struct rt_msghdr rtm; 1060 struct sockaddr addrs[RTAX_MAX]; 1061 } r; 1062 struct if_msghdr ifm; 1063 } m; 1064 char str[100], *strp; 1065 struct rt_addrinfo info; 1066 1067 1068 for (;;) { 1069 cc = read(rt_sock, &m, sizeof(m)); 1070 if (cc <= 0) { 1071 if (cc < 0 && errno != EWOULDBLOCK) 1072 LOGERR("read(rt_sock)"); 1073 return; 1074 } 1075 1076 if (m.r.rtm.rtm_version != RTM_VERSION) { 1077 msglog("bogus routing message version %d", 1078 m.r.rtm.rtm_version); 1079 continue; 1080 } 1081 1082 /* Ignore our own results. 1083 */ 1084 if (m.r.rtm.rtm_type <= RTM_CHANGE 1085 && m.r.rtm.rtm_pid == mypid) { 1086 static int complained = 0; 1087 if (!complained) { 1088 msglog("receiving our own change messages"); 1089 complained = 1; 1090 } 1091 continue; 1092 } 1093 1094 if (m.r.rtm.rtm_type == RTM_IFINFO 1095 || m.r.rtm.rtm_type == RTM_NEWADDR 1096 || m.r.rtm.rtm_type == RTM_DELADDR) { 1097 ifp = ifwithindex(m.ifm.ifm_index, 1098 m.r.rtm.rtm_type != RTM_DELADDR); 1099 if (ifp == 0) 1100 trace_act("note %s with flags %#x" 1101 " for interface index #%d", 1102 rtm_type_name(m.r.rtm.rtm_type), 1103 m.ifm.ifm_flags, 1104 m.ifm.ifm_index); 1105 else 1106 trace_act("note %s with flags %#x for %s", 1107 rtm_type_name(m.r.rtm.rtm_type), 1108 m.ifm.ifm_flags, 1109 ifp->int_name); 1110 1111 /* After being informed of a change to an interface, 1112 * check them all now if the check would otherwise 1113 * be a long time from now, if the interface is 1114 * not known, or if the interface has been turned 1115 * off or on. 1116 */ 1117 if (ifinit_timer.tv_sec-now.tv_sec>=CHECK_BAD_INTERVAL 1118 || ifp == 0 1119 || ((ifp->int_if_flags ^ m.ifm.ifm_flags) 1120 & IFF_UP_RUNNING) != 0) 1121 ifinit_timer.tv_sec = now.tv_sec; 1122 continue; 1123 } 1124 1125 strcpy(str, rtm_type_name(m.r.rtm.rtm_type)); 1126 strp = &str[strlen(str)]; 1127 if (m.r.rtm.rtm_type <= RTM_CHANGE) 1128 strp += sprintf(strp," from pid %d",m.r.rtm.rtm_pid); 1129 1130 rt_xaddrs(&info, m.r.addrs, &m.r.addrs[RTAX_MAX], 1131 m.r.rtm.rtm_addrs); 1132 1133 if (INFO_DST(&info) == 0) { 1134 trace_act("ignore %s without dst", str); 1135 continue; 1136 } 1137 1138 if (INFO_DST(&info)->sa_family != AF_INET) { 1139 trace_act("ignore %s for AF %d", str, 1140 INFO_DST(&info)->sa_family); 1141 continue; 1142 } 1143 1144 mask = ((INFO_MASK(&info) != 0) 1145 ? ntohl(S_ADDR(INFO_MASK(&info))) 1146 : (m.r.rtm.rtm_flags & RTF_HOST) 1147 ? HOST_MASK 1148 : std_mask(S_ADDR(INFO_DST(&info)))); 1149 1150 strp += sprintf(strp, ": %s", 1151 addrname(S_ADDR(INFO_DST(&info)), mask, 0)); 1152 1153 if (IN_MULTICAST(ntohl(S_ADDR(INFO_DST(&info))))) { 1154 trace_act("ignore multicast %s", str); 1155 continue; 1156 } 1157 1158 if (INFO_GATE(&info) != 0 1159 && INFO_GATE(&info)->sa_family == AF_INET) 1160 strp += sprintf(strp, " --> %s", 1161 saddr_ntoa(INFO_GATE(&info))); 1162 1163 if (INFO_AUTHOR(&info) != 0) 1164 strp += sprintf(strp, " by authority of %s", 1165 saddr_ntoa(INFO_AUTHOR(&info))); 1166 1167 switch (m.r.rtm.rtm_type) { 1168 case RTM_ADD: 1169 case RTM_CHANGE: 1170 case RTM_REDIRECT: 1171 if (m.r.rtm.rtm_errno != 0) { 1172 trace_act("ignore %s with \"%s\" error", 1173 str, strerror(m.r.rtm.rtm_errno)); 1174 } else { 1175 trace_act("%s", str); 1176 rtm_add(&m.r.rtm,&info,0); 1177 } 1178 break; 1179 1180 case RTM_DELETE: 1181 if (m.r.rtm.rtm_errno != 0) { 1182 trace_act("ignore %s with \"%s\" error", 1183 str, strerror(m.r.rtm.rtm_errno)); 1184 } else { 1185 trace_act("%s", str); 1186 del_static(S_ADDR(INFO_DST(&info)), mask, 1); 1187 } 1188 break; 1189 1190 case RTM_LOSING: 1191 trace_act("%s", str); 1192 rtm_lose(&m.r.rtm,&info); 1193 break; 1194 1195 default: 1196 trace_act("ignore %s", str); 1197 break; 1198 } 1199 } 1200 } 1201 1202 1203 /* after aggregating, note routes that belong in the kernel 1204 */ 1205 static void 1206 kern_out(struct ag_info *ag) 1207 { 1208 struct khash *k; 1209 1210 1211 /* Do not install bad routes if they are not already present. 1212 * This includes routes that had RS_NET_SYN for interfaces that 1213 * recently died. 1214 */ 1215 if (ag->ag_metric == HOPCNT_INFINITY) { 1216 k = kern_find(htonl(ag->ag_dst_h), ag->ag_mask, 0); 1217 if (k == 0) 1218 return; 1219 } else { 1220 k = kern_add(htonl(ag->ag_dst_h), ag->ag_mask); 1221 } 1222 1223 if (k->k_state & KS_NEW) { 1224 /* will need to add new entry to the kernel table */ 1225 k->k_state = KS_ADD; 1226 if (ag->ag_state & AGS_GATEWAY) 1227 k->k_state |= KS_GATEWAY; 1228 k->k_gate = ag->ag_gate; 1229 k->k_metric = ag->ag_metric; 1230 return; 1231 } 1232 1233 if (k->k_state & KS_STATIC) 1234 return; 1235 1236 /* modify existing kernel entry if necessary */ 1237 if (k->k_gate != ag->ag_gate 1238 || k->k_metric != ag->ag_metric) { 1239 k->k_gate = ag->ag_gate; 1240 k->k_metric = ag->ag_metric; 1241 k->k_state |= KS_CHANGE; 1242 } 1243 1244 if (k->k_state & KS_DYNAMIC) { 1245 k->k_state &= ~KS_DYNAMIC; 1246 k->k_state |= (KS_ADD | KS_DEL_ADD); 1247 } 1248 1249 if ((k->k_state & KS_GATEWAY) 1250 && !(ag->ag_state & AGS_GATEWAY)) { 1251 k->k_state &= ~KS_GATEWAY; 1252 k->k_state |= (KS_ADD | KS_DEL_ADD); 1253 } else if (!(k->k_state & KS_GATEWAY) 1254 && (ag->ag_state & AGS_GATEWAY)) { 1255 k->k_state |= KS_GATEWAY; 1256 k->k_state |= (KS_ADD | KS_DEL_ADD); 1257 } 1258 1259 /* Deleting-and-adding is necessary to change aspects of a route. 1260 * Just delete instead of deleting and then adding a bad route. 1261 * Otherwise, we want to keep the route in the kernel. 1262 */ 1263 if (k->k_metric == HOPCNT_INFINITY 1264 && (k->k_state & KS_DEL_ADD)) 1265 k->k_state |= KS_DELETE; 1266 else 1267 k->k_state &= ~KS_DELETE; 1268 #undef RT 1269 } 1270 1271 1272 /* ARGSUSED */ 1273 static int 1274 walk_kern(struct radix_node *rn, struct walkarg *argp) 1275 { 1276 #define RT ((struct rt_entry *)rn) 1277 char metric, pref; 1278 u_int ags = 0; 1279 1280 1281 /* Do not install synthetic routes */ 1282 if (RT->rt_state & RS_NET_SYN) 1283 return 0; 1284 1285 if (!(RT->rt_state & RS_IF)) { 1286 ags |= (AGS_GATEWAY | AGS_SUPPRESS | AGS_PROMOTE); 1287 1288 } else { 1289 /* Do not install routes for "external" remote interfaces. 1290 */ 1291 if (RT->rt_ifp != 0 && (RT->rt_ifp->int_state & IS_EXTERNAL)) 1292 return 0; 1293 1294 ags |= AGS_IF; 1295 1296 /* If it is not an interface, or an alias for an interface, 1297 * it must be a "gateway." 1298 * 1299 * If it is a "remote" interface, it is also a "gateway" to 1300 * the kernel if is not a alias. 1301 */ 1302 if (RT->rt_ifp == 0 1303 || (RT->rt_ifp->int_state & IS_REMOTE)) 1304 ags |= (AGS_GATEWAY | AGS_SUPPRESS | AGS_PROMOTE); 1305 } 1306 1307 if (RT->rt_state & RS_RDISC) 1308 ags |= AGS_CORS_GATE; 1309 1310 /* aggregate good routes without regard to their metric */ 1311 pref = 1; 1312 metric = RT->rt_metric; 1313 if (metric == HOPCNT_INFINITY) { 1314 /* if the route is dead, so try hard to aggregate. */ 1315 pref = HOPCNT_INFINITY; 1316 ags |= (AGS_FINE_GATE | AGS_SUPPRESS); 1317 } 1318 1319 ag_check(RT->rt_dst, RT->rt_mask, RT->rt_gate, 0, 1320 metric,pref, 0, 0, ags, kern_out); 1321 return 0; 1322 #undef RT 1323 } 1324 1325 1326 /* Update the kernel table to match the daemon table. 1327 */ 1328 static void 1329 fix_kern(void) 1330 { 1331 int i; 1332 struct khash *k, **pk; 1333 1334 1335 need_kern = age_timer; 1336 1337 /* Walk daemon table, updating the copy of the kernel table. 1338 */ 1339 (void)rn_walktree(rhead, walk_kern, 0); 1340 ag_flush(0,0,kern_out); 1341 1342 for (i = 0; i < KHASH_SIZE; i++) { 1343 for (pk = &khash_bins[i]; (k = *pk) != 0; ) { 1344 /* Do not touch static routes */ 1345 if (k->k_state & KS_STATIC) { 1346 kern_check_static(k,0); 1347 pk = &k->k_next; 1348 continue; 1349 } 1350 1351 /* check hold on routes deleted by the operator */ 1352 if (k->k_keep > now.tv_sec) { 1353 LIM_SEC(need_kern, k->k_keep); 1354 k->k_state |= KS_DELETE; 1355 pk = &k->k_next; 1356 continue; 1357 } 1358 1359 if ((k->k_state & KS_DELETE) 1360 && !(k->k_state & KS_DYNAMIC)) { 1361 kern_ioctl(k, RTM_DELETE, 0); 1362 *pk = k->k_next; 1363 free(k); 1364 continue; 1365 } 1366 1367 if (k->k_state & KS_DEL_ADD) 1368 kern_ioctl(k, RTM_DELETE, 0); 1369 1370 if (k->k_state & KS_ADD) { 1371 kern_ioctl(k, RTM_ADD, 1372 ((0 != (k->k_state & (KS_GATEWAY 1373 | KS_DYNAMIC))) 1374 ? RTF_GATEWAY : 0)); 1375 } else if (k->k_state & KS_CHANGE) { 1376 kern_ioctl(k, RTM_CHANGE, 1377 ((0 != (k->k_state & (KS_GATEWAY 1378 | KS_DYNAMIC))) 1379 ? RTF_GATEWAY : 0)); 1380 } 1381 k->k_state &= ~(KS_ADD|KS_CHANGE|KS_DEL_ADD); 1382 1383 /* Mark this route to be deleted in the next cycle. 1384 * This deletes routes that disappear from the 1385 * daemon table, since the normal aging code 1386 * will clear the bit for routes that have not 1387 * disappeared from the daemon table. 1388 */ 1389 k->k_state |= KS_DELETE; 1390 pk = &k->k_next; 1391 } 1392 } 1393 } 1394 1395 1396 /* Delete a static route in the image of the kernel table. 1397 */ 1398 void 1399 del_static(naddr dst, 1400 naddr mask, 1401 int gone) 1402 { 1403 struct khash *k; 1404 struct rt_entry *rt; 1405 1406 /* Just mark it in the table to be deleted next time the kernel 1407 * table is updated. 1408 * If it has already been deleted, mark it as such, and set its 1409 * keep-timer so that it will not be deleted again for a while. 1410 * This lets the operator delete a route added by the daemon 1411 * and add a replacement. 1412 */ 1413 k = kern_find(dst, mask, 0); 1414 if (k != 0) { 1415 k->k_state &= ~(KS_STATIC | KS_DYNAMIC); 1416 k->k_state |= KS_DELETE; 1417 if (gone) { 1418 k->k_state |= KS_DELETED; 1419 k->k_keep = now.tv_sec + K_KEEP_LIM; 1420 } 1421 } 1422 1423 rt = rtget(dst, mask); 1424 if (rt != 0 && (rt->rt_state & RS_STATIC)) 1425 rtbad(rt); 1426 } 1427 1428 1429 /* Delete all routes generated from ICMP Redirects that use a given gateway, 1430 * as well as old redirected routes. 1431 */ 1432 void 1433 del_redirects(naddr bad_gate, 1434 time_t old) 1435 { 1436 int i; 1437 struct khash *k; 1438 1439 1440 for (i = 0; i < KHASH_SIZE; i++) { 1441 for (k = khash_bins[i]; k != 0; k = k->k_next) { 1442 if (!(k->k_state & KS_DYNAMIC) 1443 || (k->k_state & KS_STATIC)) 1444 continue; 1445 1446 if (k->k_gate != bad_gate 1447 && k->k_redirect_time > old 1448 && !supplier) 1449 continue; 1450 1451 k->k_state |= KS_DELETE; 1452 k->k_state &= ~KS_DYNAMIC; 1453 need_kern.tv_sec = now.tv_sec; 1454 trace_act("mark redirected %s --> %s for deletion", 1455 addrname(k->k_dst, k->k_mask, 0), 1456 naddr_ntoa(k->k_gate)); 1457 } 1458 } 1459 } 1460 1461 1462 /* Start the daemon tables. 1463 */ 1464 void 1465 rtinit(void) 1466 { 1467 extern int max_keylen; 1468 int i; 1469 struct ag_info *ag; 1470 1471 /* Initialize the radix trees */ 1472 max_keylen = sizeof(struct sockaddr_in); 1473 rn_init(); 1474 rn_inithead((void**)&rhead, 32); 1475 1476 /* mark all of the slots in the table free */ 1477 ag_avail = ag_slots; 1478 for (ag = ag_slots, i = 1; i < NUM_AG_SLOTS; i++) { 1479 ag->ag_fine = ag+1; 1480 ag++; 1481 } 1482 } 1483 1484 1485 #ifdef _HAVE_SIN_LEN 1486 static struct sockaddr_in dst_sock = {sizeof(dst_sock), AF_INET}; 1487 static struct sockaddr_in mask_sock = {sizeof(mask_sock), AF_INET}; 1488 #else 1489 static struct sockaddr_in_new dst_sock = {_SIN_ADDR_SIZE, AF_INET}; 1490 static struct sockaddr_in_new mask_sock = {_SIN_ADDR_SIZE, AF_INET}; 1491 #endif 1492 1493 1494 void 1495 set_need_flash(void) 1496 { 1497 if (!need_flash) { 1498 need_flash = 1; 1499 /* Do not send the flash update immediately. Wait a little 1500 * while to hear from other routers. 1501 */ 1502 no_flash.tv_sec = now.tv_sec + MIN_WAITTIME; 1503 } 1504 } 1505 1506 1507 /* Get a particular routing table entry 1508 */ 1509 struct rt_entry * 1510 rtget(naddr dst, naddr mask) 1511 { 1512 struct rt_entry *rt; 1513 1514 dst_sock.sin_addr.s_addr = dst; 1515 mask_sock.sin_addr.s_addr = mask; 1516 masktrim(&mask_sock); 1517 rt = (struct rt_entry *)rhead->rnh_lookup(&dst_sock,&mask_sock,rhead); 1518 if (!rt 1519 || rt->rt_dst != dst 1520 || rt->rt_mask != mask) 1521 return 0; 1522 1523 return rt; 1524 } 1525 1526 1527 /* Find a route to dst as the kernel would. 1528 */ 1529 struct rt_entry * 1530 rtfind(naddr dst) 1531 { 1532 dst_sock.sin_addr.s_addr = dst; 1533 return (struct rt_entry *)rhead->rnh_matchaddr(&dst_sock, rhead); 1534 } 1535 1536 1537 /* add a route to the table 1538 */ 1539 void 1540 rtadd(naddr dst, 1541 naddr mask, 1542 naddr gate, /* forward packets here */ 1543 naddr router, /* on the authority of this router */ 1544 int metric, 1545 u_short tag, 1546 u_int state, /* rs_state for the entry */ 1547 struct interface *ifp) 1548 { 1549 struct rt_entry *rt; 1550 naddr smask; 1551 int i; 1552 struct rt_spare *rts; 1553 1554 rt = (struct rt_entry *)rtmalloc(sizeof (*rt), "rtadd"); 1555 bzero(rt, sizeof(*rt)); 1556 for (rts = rt->rt_spares, i = NUM_SPARES; i != 0; i--, rts++) 1557 rts->rts_metric = HOPCNT_INFINITY; 1558 1559 rt->rt_nodes->rn_key = (caddr_t)&rt->rt_dst_sock; 1560 rt->rt_dst = dst; 1561 rt->rt_dst_sock.sin_family = AF_INET; 1562 #ifdef _HAVE_SIN_LEN 1563 rt->rt_dst_sock.sin_len = dst_sock.sin_len; 1564 #endif 1565 if (mask != HOST_MASK) { 1566 smask = std_mask(dst); 1567 if ((smask & ~mask) == 0 && mask > smask) 1568 state |= RS_SUBNET; 1569 } 1570 mask_sock.sin_addr.s_addr = mask; 1571 masktrim(&mask_sock); 1572 rt->rt_mask = mask; 1573 rt->rt_state = state; 1574 rt->rt_gate = gate; 1575 rt->rt_router = router; 1576 rt->rt_time = now.tv_sec; 1577 rt->rt_metric = metric; 1578 rt->rt_poison_metric = HOPCNT_INFINITY; 1579 rt->rt_tag = tag; 1580 rt->rt_ifp = ifp; 1581 rt->rt_seqno = update_seqno; 1582 1583 if (++total_routes == MAX_ROUTES) 1584 msglog("have maximum (%d) routes", total_routes); 1585 if (TRACEACTIONS) 1586 trace_add_del("Add", rt); 1587 1588 need_kern.tv_sec = now.tv_sec; 1589 set_need_flash(); 1590 1591 if (0 == rhead->rnh_addaddr(&rt->rt_dst_sock, &mask_sock, 1592 rhead, rt->rt_nodes)) { 1593 msglog("rnh_addaddr() failed for %s mask=%#x", 1594 naddr_ntoa(dst), mask); 1595 } 1596 } 1597 1598 1599 /* notice a changed route 1600 */ 1601 void 1602 rtchange(struct rt_entry *rt, 1603 u_int state, /* new state bits */ 1604 naddr gate, /* now forward packets here */ 1605 naddr router, /* on the authority of this router */ 1606 int metric, /* new metric */ 1607 u_short tag, 1608 struct interface *ifp, 1609 time_t new_time, 1610 char *label) 1611 { 1612 if (rt->rt_metric != metric) { 1613 /* Fix the kernel immediately if it seems the route 1614 * has gone bad, since there may be a working route that 1615 * aggregates this route. 1616 */ 1617 if (metric == HOPCNT_INFINITY) { 1618 need_kern.tv_sec = now.tv_sec; 1619 if (new_time >= now.tv_sec - EXPIRE_TIME) 1620 new_time = now.tv_sec - EXPIRE_TIME; 1621 } 1622 rt->rt_seqno = update_seqno; 1623 set_need_flash(); 1624 } 1625 1626 if (rt->rt_gate != gate) { 1627 need_kern.tv_sec = now.tv_sec; 1628 rt->rt_seqno = update_seqno; 1629 set_need_flash(); 1630 } 1631 1632 state |= (rt->rt_state & RS_SUBNET); 1633 1634 /* Keep various things from deciding ageless routes are stale. 1635 */ 1636 if (!AGE_RT(state, ifp)) 1637 new_time = now.tv_sec; 1638 1639 if (TRACEACTIONS) 1640 trace_change(rt, state, gate, router, metric, tag, ifp, 1641 new_time, 1642 label ? label : "Chg "); 1643 1644 rt->rt_state = state; 1645 rt->rt_gate = gate; 1646 rt->rt_router = router; 1647 rt->rt_metric = metric; 1648 rt->rt_tag = tag; 1649 rt->rt_ifp = ifp; 1650 rt->rt_time = new_time; 1651 } 1652 1653 1654 /* check for a better route among the spares 1655 */ 1656 static struct rt_spare * 1657 rts_better(struct rt_entry *rt) 1658 { 1659 struct rt_spare *rts, *rts1; 1660 int i; 1661 1662 /* find the best alternative among the spares */ 1663 rts = rt->rt_spares+1; 1664 for (i = NUM_SPARES, rts1 = rts+1; i > 2; i--, rts1++) { 1665 if (BETTER_LINK(rt,rts1,rts)) 1666 rts = rts1; 1667 } 1668 1669 return rts; 1670 } 1671 1672 1673 /* switch to a backup route 1674 */ 1675 void 1676 rtswitch(struct rt_entry *rt, 1677 struct rt_spare *rts) 1678 { 1679 struct rt_spare swap; 1680 char label[10]; 1681 1682 1683 /* Do not change permanent routes */ 1684 if (0 != (rt->rt_state & (RS_MHOME | RS_STATIC | RS_RDISC 1685 | RS_NET_SYN | RS_IF))) 1686 return; 1687 1688 /* find the best alternative among the spares */ 1689 if (rts == 0) 1690 rts = rts_better(rt); 1691 1692 /* Do not bother if it is not worthwhile. 1693 */ 1694 if (!BETTER_LINK(rt, rts, rt->rt_spares)) 1695 return; 1696 1697 swap = rt->rt_spares[0]; 1698 (void)sprintf(label, "Use #%d", rts - rt->rt_spares); 1699 rtchange(rt, rt->rt_state & ~(RS_NET_SYN | RS_RDISC), 1700 rts->rts_gate, rts->rts_router, rts->rts_metric, 1701 rts->rts_tag, rts->rts_ifp, rts->rts_time, label); 1702 if (swap.rts_metric == HOPCNT_INFINITY) { 1703 *rts = rts_empty; 1704 } else { 1705 *rts = swap; 1706 } 1707 } 1708 1709 1710 void 1711 rtdelete(struct rt_entry *rt) 1712 { 1713 struct khash *k; 1714 1715 1716 if (TRACEACTIONS) 1717 trace_add_del("Del", rt); 1718 1719 k = kern_find(rt->rt_dst, rt->rt_mask, 0); 1720 if (k != 0) { 1721 k->k_state |= KS_DELETE; 1722 need_kern.tv_sec = now.tv_sec; 1723 } 1724 1725 dst_sock.sin_addr.s_addr = rt->rt_dst; 1726 mask_sock.sin_addr.s_addr = rt->rt_mask; 1727 masktrim(&mask_sock); 1728 if (rt != (struct rt_entry *)rhead->rnh_deladdr(&dst_sock, &mask_sock, 1729 rhead)) { 1730 msglog("rnh_deladdr() failed"); 1731 } else { 1732 free(rt); 1733 total_routes--; 1734 } 1735 } 1736 1737 1738 void 1739 rts_delete(struct rt_entry *rt, 1740 struct rt_spare *rts) 1741 { 1742 trace_upslot(rt, rts, 0, 0, 0, HOPCNT_INFINITY, 0, 0); 1743 *rts = rts_empty; 1744 } 1745 1746 1747 /* Get rid of a bad route, and try to switch to a replacement. 1748 */ 1749 void 1750 rtbad(struct rt_entry *rt) 1751 { 1752 /* Poison the route */ 1753 rtchange(rt, rt->rt_state & ~(RS_IF | RS_LOCAL | RS_STATIC), 1754 rt->rt_gate, rt->rt_router, HOPCNT_INFINITY, rt->rt_tag, 1755 0, rt->rt_time, 0); 1756 1757 rtswitch(rt, 0); 1758 } 1759 1760 1761 /* Junk a RS_NET_SYN or RS_LOCAL route, 1762 * unless it is needed by another interface. 1763 */ 1764 void 1765 rtbad_sub(struct rt_entry *rt) 1766 { 1767 struct interface *ifp, *ifp1; 1768 struct intnet *intnetp; 1769 u_int state; 1770 1771 1772 ifp1 = 0; 1773 state = 0; 1774 1775 if (rt->rt_state & RS_LOCAL) { 1776 /* Is this the route through loopback for the interface? 1777 * If so, see if it is used by any other interfaces, such 1778 * as a point-to-point interface with the same local address. 1779 */ 1780 for (ifp = ifnet; ifp != 0; ifp = ifp->int_next) { 1781 /* Retain it if another interface needs it. 1782 */ 1783 if (ifp->int_addr == rt->rt_ifp->int_addr) { 1784 state |= RS_LOCAL; 1785 ifp1 = ifp; 1786 break; 1787 } 1788 } 1789 1790 } 1791 1792 if (!(state & RS_LOCAL)) { 1793 /* Retain RIPv1 logical network route if there is another 1794 * interface that justifies it. 1795 */ 1796 if (rt->rt_state & RS_NET_SYN) { 1797 for (ifp = ifnet; ifp != 0; ifp = ifp->int_next) { 1798 if ((ifp->int_state & IS_NEED_NET_SYN) 1799 && rt->rt_mask == ifp->int_std_mask 1800 && rt->rt_dst == ifp->int_std_addr) { 1801 state |= RS_NET_SYN; 1802 ifp1 = ifp; 1803 break; 1804 } 1805 } 1806 } 1807 1808 /* or if there is an authority route that needs it. */ 1809 for (intnetp = intnets; 1810 intnetp != 0; 1811 intnetp = intnetp->intnet_next) { 1812 if (intnetp->intnet_addr == rt->rt_dst 1813 && intnetp->intnet_mask == rt->rt_mask) { 1814 state |= (RS_NET_SYN | RS_NET_INT); 1815 break; 1816 } 1817 } 1818 } 1819 1820 if (ifp1 != 0 || (state & RS_NET_SYN)) { 1821 rtchange(rt, ((rt->rt_state & ~(RS_NET_SYN | RS_LOCAL)) 1822 | state), 1823 rt->rt_gate, rt->rt_router, rt->rt_metric, 1824 rt->rt_tag, ifp1, rt->rt_time, 0); 1825 } else { 1826 rtbad(rt); 1827 } 1828 } 1829 1830 1831 /* Called while walking the table looking for sick interfaces 1832 * or after a time change. 1833 */ 1834 /* ARGSUSED */ 1835 int 1836 walk_bad(struct radix_node *rn, struct walkarg *argp) 1837 { 1838 #define RT ((struct rt_entry *)rn) 1839 struct rt_spare *rts; 1840 int i; 1841 1842 1843 /* fix any spare routes through the interface 1844 */ 1845 rts = RT->rt_spares; 1846 for (i = NUM_SPARES; i != 1; i--) { 1847 rts++; 1848 if (rts->rts_metric < HOPCNT_INFINITY 1849 && (rts->rts_ifp == 0 1850 || (rts->rts_ifp->int_state & IS_BROKE))) 1851 rts_delete(RT, rts); 1852 } 1853 1854 /* Deal with the main route 1855 */ 1856 /* finished if it has been handled before or if its interface is ok 1857 */ 1858 if (RT->rt_ifp == 0 || !(RT->rt_ifp->int_state & IS_BROKE)) 1859 return 0; 1860 1861 /* Bad routes for other than interfaces are easy. 1862 */ 1863 if (0 == (RT->rt_state & (RS_IF | RS_NET_SYN | RS_LOCAL))) { 1864 rtbad(RT); 1865 return 0; 1866 } 1867 1868 rtbad_sub(RT); 1869 return 0; 1870 #undef RT 1871 } 1872 1873 1874 /* Check the age of an individual route. 1875 */ 1876 /* ARGSUSED */ 1877 static int 1878 walk_age(struct radix_node *rn, struct walkarg *argp) 1879 { 1880 #define RT ((struct rt_entry *)rn) 1881 struct interface *ifp; 1882 struct rt_spare *rts; 1883 int i; 1884 1885 1886 /* age all of the spare routes, including the primary route 1887 * currently in use 1888 */ 1889 rts = RT->rt_spares; 1890 for (i = NUM_SPARES; i != 0; i--, rts++) { 1891 1892 ifp = rts->rts_ifp; 1893 if (i == NUM_SPARES) { 1894 if (!AGE_RT(RT->rt_state, ifp)) { 1895 /* Keep various things from deciding ageless 1896 * routes are stale 1897 */ 1898 rts->rts_time = now.tv_sec; 1899 continue; 1900 } 1901 1902 /* forget RIP routes after RIP has been turned off. 1903 */ 1904 if (rip_sock < 0) { 1905 rtdelete(RT); 1906 return 0; 1907 } 1908 } 1909 1910 /* age failing routes 1911 */ 1912 if (age_bad_gate == rts->rts_gate 1913 && rts->rts_time >= now_stale) { 1914 rts->rts_time -= SUPPLY_INTERVAL; 1915 } 1916 1917 /* trash the spare routes when they go bad */ 1918 if (rts->rts_metric < HOPCNT_INFINITY 1919 && now_garbage > rts->rts_time) 1920 rts_delete(RT, rts); 1921 } 1922 1923 1924 /* finished if the active route is still fresh */ 1925 if (now_stale <= RT->rt_time) 1926 return 0; 1927 1928 /* try to switch to an alternative */ 1929 rtswitch(RT, 0); 1930 1931 /* Delete a dead route after it has been publically mourned. */ 1932 if (now_garbage > RT->rt_time) { 1933 rtdelete(RT); 1934 return 0; 1935 } 1936 1937 /* Start poisoning a bad route before deleting it. */ 1938 if (now.tv_sec - RT->rt_time > EXPIRE_TIME) 1939 rtchange(RT, RT->rt_state, RT->rt_gate, RT->rt_router, 1940 HOPCNT_INFINITY, RT->rt_tag, RT->rt_ifp, 1941 RT->rt_time, 0); 1942 return 0; 1943 } 1944 1945 1946 /* Watch for dead routes and interfaces. 1947 */ 1948 void 1949 age(naddr bad_gate) 1950 { 1951 struct interface *ifp; 1952 int need_query = 0; 1953 1954 /* If not listening to RIP, there is no need to age the routes in 1955 * the table. 1956 */ 1957 age_timer.tv_sec = (now.tv_sec 1958 + ((rip_sock < 0) ? NEVER : SUPPLY_INTERVAL)); 1959 1960 /* Check for dead IS_REMOTE interfaces by timing their 1961 * transmissions. 1962 */ 1963 for (ifp = ifnet; ifp; ifp = ifp->int_next) { 1964 if (!(ifp->int_state & IS_REMOTE)) 1965 continue; 1966 1967 /* ignore unreachable remote interfaces */ 1968 if (!check_remote(ifp)) 1969 continue; 1970 /* Restore remote interface that has become reachable 1971 */ 1972 if (ifp->int_state & IS_BROKE) 1973 if_ok(ifp, "remote "); 1974 1975 if (ifp->int_act_time != NEVER 1976 && now.tv_sec - ifp->int_act_time > EXPIRE_TIME) { 1977 msglog("remote interface %s to %s timed out after" 1978 " %d:%d", 1979 ifp->int_name, 1980 naddr_ntoa(ifp->int_dstaddr), 1981 (now.tv_sec - ifp->int_act_time)/60, 1982 (now.tv_sec - ifp->int_act_time)%60); 1983 if_sick(ifp); 1984 } 1985 1986 /* If we have not heard from the other router 1987 * recently, ask it. 1988 */ 1989 if (now.tv_sec >= ifp->int_query_time) { 1990 ifp->int_query_time = NEVER; 1991 need_query = 1; 1992 } 1993 } 1994 1995 /* Age routes. */ 1996 age_bad_gate = bad_gate; 1997 (void)rn_walktree(rhead, walk_age, 0); 1998 1999 /* Update the kernel routing table. */ 2000 fix_kern(); 2001 2002 /* poke reticent remote gateways */ 2003 if (need_query) 2004 rip_query(); 2005 } 2006