sbin/routed/table.c

   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 acknowledgment:
  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  * $FreeBSD: src/sbin/routed/table.c,v 1.9.2.2 2000/08/14 17:00:04 sheldonh Exp $
  34  * $DragonFly: src/sbin/routed/table.c,v 1.6 2005/03/16 21:21:34 cpressey Exp $
  35  */
  36
  37 #include "defs.h"
  38
  39 #if !defined(sgi) && !defined(__NetBSD__)
  40 static char sccsid[] __attribute__((unused)) = "@(#)tables.c    8.1 (Berkeley) 6/5/93";
  41 #elif defined(__NetBSD__)
  42 __RCSID("$NetBSD$");
  43 #endif
  44
  45 static struct rt_spare *rts_better(struct rt_entry *);
  46 static struct rt_spare rts_empty = {0,0,0,HOPCNT_INFINITY,0,0,0};
  47 static void  set_need_flash(void);
  48 #ifdef _HAVE_SIN_LEN
  49 static void masktrim(struct sockaddr_in *ap);
  50 #else
  51 static void masktrim(struct sockaddr_in_new *ap);
  52 #endif
  53
  54
  55 struct radix_node_head *rhead;          /* root of the radix tree */
  56
  57 int     need_flash = 1;                 /* flash update needed
  58                                          * start =1 to suppress the 1st
  59                                          */
  60
  61 struct timeval age_timer;               /* next check of old routes */
  62 struct timeval need_kern = {            /* need to update kernel table */
  63         EPOCH+MIN_WAITTIME-1, 0
  64 };
  65
  66 int     stopint;
  67
  68 int     total_routes;
  69
  70 /* zap any old routes through this gateway */
  71 naddr   age_bad_gate;
  72
  73
  74 /* It is desirable to "aggregate" routes, to combine differing routes of
  75  * the same metric and next hop into a common route with a smaller netmask
  76  * or to suppress redundant routes, routes that add no information to
  77  * routes with smaller netmasks.
  78  *
  79  * A route is redundant if and only if any and all routes with smaller
  80  * but matching netmasks and nets are the same.  Since routes are
  81  * kept sorted in the radix tree, redundant routes always come second.
  82  *
  83  * There are two kinds of aggregations.  First, two routes of the same bit
  84  * mask and differing only in the least significant bit of the network
  85  * number can be combined into a single route with a coarser mask.
  86  *
  87  * Second, a route can be suppressed in favor of another route with a more
  88  * coarse mask provided no incompatible routes with intermediate masks
  89  * are present.  The second kind of aggregation involves suppressing routes.
  90  * A route must not be suppressed if an incompatible route exists with
  91  * an intermediate mask, since the suppressed route would be covered
  92  * by the intermediate.
  93  *
  94  * This code relies on the radix tree walk encountering routes
  95  * sorted first by address, with the smallest address first.
  96  */
  97
  98 struct ag_info ag_slots[NUM_AG_SLOTS], *ag_avail, *ag_corsest, *ag_finest;
  99
 100 /* #define DEBUG_AG */
 101 #ifdef DEBUG_AG
 102 #define CHECK_AG() {int acnt = 0; struct ag_info *cag;          \
 103         for (cag = ag_avail; cag != 0; cag = cag->ag_fine)      \
 104                 acnt++;                                         \
 105         for (cag = ag_corsest; cag != 0; cag = cag->ag_fine)    \
 106                 acnt++;                                         \
 107         if (acnt != NUM_AG_SLOTS) {                             \
 108                 fflush(stderr);                                 \
 109                 abort();                                        \
 110         }                                                       \
 111 }
 112 #else
 113 #define CHECK_AG()
 114 #endif
 115
 116
 117 /* Output the contents of an aggregation table slot.
 118  *      This function must always be immediately followed with the deletion
 119  *      of the target slot.
 120  */
 121 static void
 122 ag_out(struct ag_info *ag,
 123          void (*out)(struct ag_info *))
 124 {
 125         struct ag_info *ag_cors;
 126         naddr bit;
 127
 128
 129         /* Forget it if this route should not be output for split-horizon. */
 130         if (ag->ag_state & AGS_SPLIT_HZ)
 131                 return;
 132
 133         /* If we output both the even and odd twins, then the immediate parent,
 134          * if it is present, is redundant, unless the parent manages to
 135          * aggregate into something coarser.
 136          * On successive calls, this code detects the even and odd twins,
 137          * and marks the parent.
 138          *
 139          * Note that the order in which the radix tree code emits routes
 140          * ensures that the twins are seen before the parent is emitted.
 141          */
 142         ag_cors = ag->ag_cors;
 143         if (ag_cors != 0
 144             && ag_cors->ag_mask == ag->ag_mask<<1
 145             && ag_cors->ag_dst_h == (ag->ag_dst_h & ag_cors->ag_mask)) {
 146                 ag_cors->ag_state |= ((ag_cors->ag_dst_h == ag->ag_dst_h)
 147                                       ? AGS_REDUN0
 148                                       : AGS_REDUN1);
 149         }
 150
 151         /* Skip it if this route is itself redundant.
 152          *
 153          * It is ok to change the contents of the slot here, since it is
 154          * always deleted next.
 155          */
 156         if (ag->ag_state & AGS_REDUN0) {
 157                 if (ag->ag_state & AGS_REDUN1)
 158                         return;         /* quit if fully redundant */
 159                 /* make it finer if it is half-redundant */
 160                 bit = (-ag->ag_mask) >> 1;
 161                 ag->ag_dst_h |= bit;
 162                 ag->ag_mask |= bit;
 163
 164         } else if (ag->ag_state & AGS_REDUN1) {
 165                 /* make it finer if it is half-redundant */
 166                 bit = (-ag->ag_mask) >> 1;
 167                 ag->ag_mask |= bit;
 168         }
 169         out(ag);
 170 }
 171
 172
 173 static void
 174 ag_del(struct ag_info *ag)
 175 {
 176         CHECK_AG();
 177
 178         if (ag->ag_cors == 0)
 179                 ag_corsest = ag->ag_fine;
 180         else
 181                 ag->ag_cors->ag_fine = ag->ag_fine;
 182
 183         if (ag->ag_fine == 0)
 184                 ag_finest = ag->ag_cors;
 185         else
 186                 ag->ag_fine->ag_cors = ag->ag_cors;
 187
 188         ag->ag_fine = ag_avail;
 189         ag_avail = ag;
 190
 191         CHECK_AG();
 192 }
 193
 194
 195 /* Flush routes waiting for aggregation.
 196  *      This must not suppress a route unless it is known that among all
 197  *      routes with coarser masks that match it, the one with the longest
 198  *      mask is appropriate.  This is ensured by scanning the routes
 199  *      in lexical order, and with the most restrictive mask first
 200  *      among routes to the same destination.
 201  */
 202 void
 203 ag_flush(naddr lim_dst_h,               /* flush routes to here */
 204          naddr lim_mask,                /* matching this mask */
 205          void (*out)(struct ag_info *))
 206 {
 207         struct ag_info *ag, *ag_cors;
 208         naddr dst_h;
 209
 210
 211         for (ag = ag_finest;
 212              ag != 0 && ag->ag_mask >= lim_mask;
 213              ag = ag_cors) {
 214                 ag_cors = ag->ag_cors;
 215
 216                 /* work on only the specified routes */
 217                 dst_h = ag->ag_dst_h;
 218                 if ((dst_h & lim_mask) != lim_dst_h)
 219                         continue;
 220
 221                 if (!(ag->ag_state & AGS_SUPPRESS))
 222                         ag_out(ag, out);
 223
 224                 else for ( ; ; ag_cors = ag_cors->ag_cors) {
 225                         /* Look for a route that can suppress the
 226                          * current route */
 227                         if (ag_cors == 0) {
 228                                 /* failed, so output it and look for
 229                                  * another route to work on
 230                                  */
 231                                 ag_out(ag, out);
 232                                 break;
 233                         }
 234
 235                         if ((dst_h & ag_cors->ag_mask) == ag_cors->ag_dst_h) {
 236                                 /* We found a route with a coarser mask that
 237                                  * aggregates the current target.
 238                                  *
 239                                  * If it has a different next hop, it
 240                                  * cannot replace the target, so output
 241                                  * the target.
 242                                  */
 243                                 if (ag->ag_gate != ag_cors->ag_gate
 244                                     && !(ag->ag_state & AGS_FINE_GATE)
 245                                     && !(ag_cors->ag_state & AGS_CORS_GATE)) {
 246                                         ag_out(ag, out);
 247                                         break;
 248                                 }
 249
 250                                 /* If the coarse route has a good enough
 251                                  * metric, it suppresses the target.
 252                                  * If the suppressed target was redundant,
 253                                  * then mark the suppressor redundant.
 254                                  */
 255                                 if (ag_cors->ag_pref <= ag->ag_pref) {
 256                                     if (ag_cors->ag_seqno > ag->ag_seqno)
 257                                         ag_cors->ag_seqno = ag->ag_seqno;
 258                                     if (AG_IS_REDUN(ag->ag_state)
 259                                         && ag_cors->ag_mask==ag->ag_mask<<1) {
 260                                         if (ag_cors->ag_dst_h == dst_h)
 261                                             ag_cors->ag_state |= AGS_REDUN0;
 262                                         else
 263                                             ag_cors->ag_state |= AGS_REDUN1;
 264                                     }
 265                                     if (ag->ag_tag != ag_cors->ag_tag)
 266                                             ag_cors->ag_tag = 0;
 267                                     if (ag->ag_nhop != ag_cors->ag_nhop)
 268                                             ag_cors->ag_nhop = 0;
 269                                     break;
 270                                 }
 271                         }
 272                 }
 273
 274                 /* That route has either been output or suppressed */
 275                 ag_cors = ag->ag_cors;
 276                 ag_del(ag);
 277         }
 278
 279         CHECK_AG();
 280 }
 281
 282
 283 /* Try to aggregate a route with previous routes.
 284  */
 285 void
 286 ag_check(naddr  dst,
 287          naddr  mask,
 288          naddr  gate,
 289          naddr  nhop,
 290          char   metric,
 291          char   pref,
 292          u_int  seqnum,
 293          u_short tag,
 294          u_short state,
 295          void (*out)(struct ag_info *)) /* output using this */
 296 {
 297         struct ag_info *ag, *nag, *ag_cors;
 298         naddr xaddr;
 299         int x;
 300
 301         dst = ntohl(dst);
 302
 303         /* Punt non-contiguous subnet masks.
 304          *
 305          * (X & -X) contains a single bit if and only if X is a power of 2.
 306          * (X + (X & -X)) == 0 if and only if X is a power of 2.
 307          */
 308         if ((mask & -mask) + mask != 0) {
 309                 struct ag_info nc_ag;
 310
 311                 nc_ag.ag_dst_h = dst;
 312                 nc_ag.ag_mask = mask;
 313                 nc_ag.ag_gate = gate;
 314                 nc_ag.ag_nhop = nhop;
 315                 nc_ag.ag_metric = metric;
 316                 nc_ag.ag_pref = pref;
 317                 nc_ag.ag_tag = tag;
 318                 nc_ag.ag_state = state;
 319                 nc_ag.ag_seqno = seqnum;
 320                 out(&nc_ag);
 321                 return;
 322         }
 323
 324         /* Search for the right slot in the aggregation table.
 325          */
 326         ag_cors = 0;
 327         ag = ag_corsest;
 328         while (ag != 0) {
 329                 if (ag->ag_mask >= mask)
 330                         break;
 331
 332                 /* Suppress old routes (i.e. combine with compatible routes
 333                  * with coarser masks) as we look for the right slot in the
 334                  * aggregation table for the new route.
 335                  * A route to an address less than the current destination
 336                  * will not be affected by the current route or any route
 337                  * seen hereafter.  That means it is safe to suppress it.
 338                  * This check keeps poor routes (e.g. with large hop counts)
 339                  * from preventing suppression of finer routes.
 340                  */
 341                 if (ag_cors != 0
 342                     && ag->ag_dst_h < dst
 343                     && (ag->ag_state & AGS_SUPPRESS)
 344                     && ag_cors->ag_pref <= ag->ag_pref
 345                     && (ag->ag_dst_h & ag_cors->ag_mask) == ag_cors->ag_dst_h
 346                     && (ag_cors->ag_gate == ag->ag_gate
 347                         || (ag->ag_state & AGS_FINE_GATE)
 348                         || (ag_cors->ag_state & AGS_CORS_GATE))) {
 349                         if (ag_cors->ag_seqno > ag->ag_seqno)
 350                                 ag_cors->ag_seqno = ag->ag_seqno;
 351                         /*  If the suppressed target was redundant,
 352                          * then mark the suppressor redundant.
 353                          */
 354                         if (AG_IS_REDUN(ag->ag_state)
 355                             && ag_cors->ag_mask == ag->ag_mask<<1) {
 356                                 if (ag_cors->ag_dst_h == dst)
 357                                         ag_cors->ag_state |= AGS_REDUN0;
 358                                 else
 359                                         ag_cors->ag_state |= AGS_REDUN1;
 360                         }
 361                         if (ag->ag_tag != ag_cors->ag_tag)
 362                                 ag_cors->ag_tag = 0;
 363                         if (ag->ag_nhop != ag_cors->ag_nhop)
 364                                 ag_cors->ag_nhop = 0;
 365                         ag_del(ag);
 366                         CHECK_AG();
 367                 } else {
 368                         ag_cors = ag;
 369                 }
 370                 ag = ag_cors->ag_fine;
 371         }
 372
 373         /* If we find the even/odd twin of the new route, and if the
 374          * masks and so forth are equal, we can aggregate them.
 375          * We can probably promote one of the pair.
 376          *
 377          * Since the routes are encountered in lexical order,
 378          * the new route must be odd.  However, the second or later
 379          * times around this loop, it could be the even twin promoted
 380          * from the even/odd pair of twins of the finer route.
 381          */
 382         while (ag != 0
 383                && ag->ag_mask == mask
 384                && ((ag->ag_dst_h ^ dst) & (mask<<1)) == 0) {
 385
 386                 /* Here we know the target route and the route in the current
 387                  * slot have the same netmasks and differ by at most the
 388                  * last bit.  They are either for the same destination, or
 389                  * for an even/odd pair of destinations.
 390                  */
 391                 if (ag->ag_dst_h == dst) {
 392                         /* We have two routes to the same destination.
 393                          * Routes are encountered in lexical order, so a
 394                          * route is never promoted until the parent route is
 395                          * already present.  So we know that the new route is
 396                          * a promoted (or aggregated) pair and the route
 397                          * already in the slot is the explicit route.
 398                          *
 399                          * Prefer the best route if their metrics differ,
 400                          * or the aggregated one if not, following a sort
 401                          * of longest-match rule.
 402                          */
 403                         if (pref <= ag->ag_pref) {
 404                                 ag->ag_gate = gate;
 405                                 ag->ag_nhop = nhop;
 406                                 ag->ag_tag = tag;
 407                                 ag->ag_metric = metric;
 408                                 ag->ag_pref = pref;
 409                                 x = ag->ag_state;
 410                                 ag->ag_state = state;
 411                                 state = x;
 412                         }
 413
 414                         /* The sequence number controls flash updating,
 415                          * and should be the smaller of the two.
 416                          */
 417                         if (ag->ag_seqno > seqnum)
 418                                 ag->ag_seqno = seqnum;
 419
 420                         /* Some bits are set if they are set on either route,
 421                          * except when the route is for an interface.
 422                          */
 423                         if (!(ag->ag_state & AGS_IF))
 424                                 ag->ag_state |= (state & (AGS_AGGREGATE_EITHER
 425                                                         | AGS_REDUN0
 426                                                         | AGS_REDUN1));
 427                         return;
 428                 }
 429
 430                 /* If one of the routes can be promoted and the other can
 431                  * be suppressed, it may be possible to combine them or
 432                  * worthwhile to promote one.
 433                  *
 434                  * Any route that can be promoted is always
 435                  * marked to be eligible to be suppressed.
 436                  */
 437                 if (!((state & AGS_AGGREGATE)
 438                       && (ag->ag_state & AGS_SUPPRESS))
 439                     && !((ag->ag_state & AGS_AGGREGATE)
 440                          && (state & AGS_SUPPRESS)))
 441                         break;
 442
 443                 /* A pair of even/odd twin routes can be combined
 444                  * if either is redundant, or if they are via the
 445                  * same gateway and have the same metric.
 446                  */
 447                 if (AG_IS_REDUN(ag->ag_state)
 448                     || AG_IS_REDUN(state)
 449                     || (ag->ag_gate == gate
 450                         && ag->ag_pref == pref
 451                         && (state & ag->ag_state & AGS_AGGREGATE) != 0)) {
 452
 453                         /* We have both the even and odd pairs.
 454                          * Since the routes are encountered in order,
 455                          * the route in the slot must be the even twin.
 456                          *
 457                          * Combine and promote (aggregate) the pair of routes.
 458                          */
 459                         if (seqnum > ag->ag_seqno)
 460                                 seqnum = ag->ag_seqno;
 461                         if (!AG_IS_REDUN(state))
 462                                 state &= ~AGS_REDUN1;
 463                         if (AG_IS_REDUN(ag->ag_state))
 464                                 state |= AGS_REDUN0;
 465                         else
 466                                 state &= ~AGS_REDUN0;
 467                         state |= (ag->ag_state & AGS_AGGREGATE_EITHER);
 468                         if (ag->ag_tag != tag)
 469                                 tag = 0;
 470                         if (ag->ag_nhop != nhop)
 471                                 nhop = 0;
 472
 473                         /* Get rid of the even twin that was already
 474                          * in the slot.
 475                          */
 476                         ag_del(ag);
 477
 478                 } else if (ag->ag_pref >= pref
 479                            && (ag->ag_state & AGS_AGGREGATE)) {
 480                         /* If we cannot combine the pair, maybe the route
 481                          * with the worse metric can be promoted.
 482                          *
 483                          * Promote the old, even twin, by giving its slot
 484                          * in the table to the new, odd twin.
 485                          */
 486                         ag->ag_dst_h = dst;
 487
 488                         xaddr = ag->ag_gate;
 489                         ag->ag_gate = gate;
 490                         gate = xaddr;
 491
 492                         xaddr = ag->ag_nhop;
 493                         ag->ag_nhop = nhop;
 494                         nhop = xaddr;
 495
 496                         x = ag->ag_tag;
 497                         ag->ag_tag = tag;
 498                         tag = x;
 499
 500                         /* The promoted route is even-redundant only if the
 501                          * even twin was fully redundant.  It is not
 502                          * odd-redundant because the odd-twin will still be
 503                          * in the table.
 504                          */
 505                         x = ag->ag_state;
 506                         if (!AG_IS_REDUN(x))
 507                                 x &= ~AGS_REDUN0;
 508                         x &= ~AGS_REDUN1;
 509                         ag->ag_state = state;
 510                         state = x;
 511
 512                         x = ag->ag_metric;
 513                         ag->ag_metric = metric;
 514                         metric = x;
 515
 516                         x = ag->ag_pref;
 517                         ag->ag_pref = pref;
 518                         pref = x;
 519
 520                         /* take the newest sequence number */
 521                         if (seqnum >= ag->ag_seqno)
 522                                 seqnum = ag->ag_seqno;
 523                         else
 524                                 ag->ag_seqno = seqnum;
 525
 526                 } else {
 527                         if (!(state & AGS_AGGREGATE))
 528                                 break;  /* cannot promote either twin */
 529
 530                         /* Promote the new, odd twin by shaving its
 531                          * mask and address.
 532                          * The promoted route is odd-redundant only if the
 533                          * odd twin was fully redundant.  It is not
 534                          * even-redundant because the even twin is still in
 535                          * the table.
 536                          */
 537                         if (!AG_IS_REDUN(state))
 538                                 state &= ~AGS_REDUN1;
 539                         state &= ~AGS_REDUN0;
 540                         if (seqnum > ag->ag_seqno)
 541                                 seqnum = ag->ag_seqno;
 542                         else
 543                                 ag->ag_seqno = seqnum;
 544                 }
 545
 546                 mask <<= 1;
 547                 dst &= mask;
 548
 549                 if (ag_cors == 0) {
 550                         ag = ag_corsest;
 551                         break;
 552                 }
 553                 ag = ag_cors;
 554                 ag_cors = ag->ag_cors;
 555         }
 556
 557         /* When we can no longer promote and combine routes,
 558          * flush the old route in the target slot.  Also flush
 559          * any finer routes that we know will never be aggregated by
 560          * the new route.
 561          *
 562          * In case we moved toward coarser masks,
 563          * get back where we belong
 564          */
 565         if (ag != 0
 566             && ag->ag_mask < mask) {
 567                 ag_cors = ag;
 568                 ag = ag->ag_fine;
 569         }
 570
 571         /* Empty the target slot
 572          */
 573         if (ag != 0 && ag->ag_mask == mask) {
 574                 ag_flush(ag->ag_dst_h, ag->ag_mask, out);
 575                 ag = (ag_cors == 0) ? ag_corsest : ag_cors->ag_fine;
 576         }
 577
 578 #ifdef DEBUG_AG
 579         fflush(stderr);
 580         if (ag == 0 && ag_cors != ag_finest)
 581                 abort();
 582         if (ag_cors == 0 && ag != ag_corsest)
 583                 abort();
 584         if (ag != 0 && ag->ag_cors != ag_cors)
 585                 abort();
 586         if (ag_cors != 0 && ag_cors->ag_fine != ag)
 587                 abort();
 588         CHECK_AG();
 589 #endif
 590
 591         /* Save the new route on the end of the table.
 592          */
 593         nag = ag_avail;
 594         ag_avail = nag->ag_fine;
 595
 596         nag->ag_dst_h = dst;
 597         nag->ag_mask = mask;
 598         nag->ag_gate = gate;
 599         nag->ag_nhop = nhop;
 600         nag->ag_metric = metric;
 601         nag->ag_pref = pref;
 602         nag->ag_tag = tag;
 603         nag->ag_state = state;
 604         nag->ag_seqno = seqnum;
 605
 606         nag->ag_fine = ag;
 607         if (ag != 0)
 608                 ag->ag_cors = nag;
 609         else
 610                 ag_finest = nag;
 611         nag->ag_cors = ag_cors;
 612         if (ag_cors == 0)
 613                 ag_corsest = nag;
 614         else
 615                 ag_cors->ag_fine = nag;
 616         CHECK_AG();
 617 }
 618
 619
 620 #define NAME0_LEN 14
 621 static const char *
 622 rtm_type_name(u_char type)
 623 {
 624         static const char *rtm_types[] = {
 625                 "RTM_ADD",
 626                 "RTM_DELETE",
 627                 "RTM_CHANGE",
 628                 "RTM_GET",
 629                 "RTM_LOSING",
 630                 "RTM_REDIRECT",
 631                 "RTM_MISS",
 632                 "RTM_LOCK",
 633                 "RTM_OLDADD",
 634                 "RTM_OLDDEL",
 635                 "RTM_RESOLVE",
 636                 "RTM_NEWADDR",
 637                 "RTM_DELADDR",
 638                 "RTM_IFINFO",
 639                 "RTM_NEWMADDR",
 640                 "RTM_DELMADDR"
 641         };
 642 #define NEW_RTM_PAT "RTM type %#x"
 643         static char name0[sizeof(NEW_RTM_PAT)+2];
 644
 645
 646         if (type > sizeof(rtm_types)/sizeof(rtm_types[0])
 647             || type == 0) {
 648                 snprintf(name0, sizeof(name0), NEW_RTM_PAT, type);
 649                 return name0;
 650         } else {
 651                 return rtm_types[type-1];
 652         }
 653 #undef NEW_RTM_PAT
 654 }
 655
 656
 657 /* Trim a mask in a sockaddr
 658  *      Produce a length of 0 for an address of 0.
 659  *      Otherwise produce the index of the first zero byte.
 660  */
 661 void
 662 #ifdef _HAVE_SIN_LEN
 663 masktrim(struct sockaddr_in *ap)
 664 #else
 665 masktrim(struct sockaddr_in_new *ap)
 666 #endif
 667 {
 668         char *cp;
 669
 670         if (ap->sin_addr.s_addr == 0) {
 671                 ap->sin_len = 0;
 672                 return;
 673         }
 674         cp = (char *)(&ap->sin_addr.s_addr+1);
 675         while (*--cp == 0)
 676                 continue;
 677         ap->sin_len = cp - (char*)ap + 1;
 678 }
 679
 680
 681 /* Tell the kernel to add, delete or change a route
 682  */
 683 static void
 684 rtioctl(int action,                     /* RTM_DELETE, etc */
 685         naddr dst,
 686         naddr gate,
 687         naddr mask,
 688         int metric,
 689         int flags)
 690 {
 691         struct {
 692                 struct rt_msghdr w_rtm;
 693                 struct sockaddr_in w_dst;
 694                 struct sockaddr_in w_gate;
 695 #ifdef _HAVE_SA_LEN
 696                 struct sockaddr_in w_mask;
 697 #else
 698                 struct sockaddr_in_new w_mask;
 699 #endif
 700         } w;
 701         long cc;
 702 #   define PAT " %-10s %s metric=%d flags=%#x"
 703 #   define ARGS rtm_type_name(action), rtname(dst,mask,gate), metric, flags
 704
 705 again:
 706         memset(&w, 0, sizeof(w));
 707         w.w_rtm.rtm_msglen = sizeof(w);
 708         w.w_rtm.rtm_version = RTM_VERSION;
 709         w.w_rtm.rtm_type = action;
 710         w.w_rtm.rtm_flags = flags;
 711         w.w_rtm.rtm_seq = ++rt_sock_seqno;
 712         w.w_rtm.rtm_addrs = RTA_DST|RTA_GATEWAY;
 713         if (metric != 0 || action == RTM_CHANGE) {
 714                 w.w_rtm.rtm_rmx.rmx_hopcount = metric;
 715                 w.w_rtm.rtm_inits |= RTV_HOPCOUNT;
 716         }
 717         w.w_dst.sin_family = AF_INET;
 718         w.w_dst.sin_addr.s_addr = dst;
 719         w.w_gate.sin_family = AF_INET;
 720         w.w_gate.sin_addr.s_addr = gate;
 721 #ifdef _HAVE_SA_LEN
 722         w.w_dst.sin_len = sizeof(w.w_dst);
 723         w.w_gate.sin_len = sizeof(w.w_gate);
 724 #endif
 725         if (mask == HOST_MASK) {
 726                 w.w_rtm.rtm_flags |= RTF_HOST;
 727                 w.w_rtm.rtm_msglen -= sizeof(w.w_mask);
 728         } else {
 729                 w.w_rtm.rtm_addrs |= RTA_NETMASK;
 730                 w.w_mask.sin_addr.s_addr = htonl(mask);
 731 #ifdef _HAVE_SA_LEN
 732                 masktrim(&w.w_mask);
 733                 if (w.w_mask.sin_len == 0)
 734                         w.w_mask.sin_len = sizeof(long);
 735                 w.w_rtm.rtm_msglen -= (sizeof(w.w_mask) - w.w_mask.sin_len);
 736 #endif
 737         }
 738
 739 #ifndef NO_INSTALL
 740         cc = write(rt_sock, &w, w.w_rtm.rtm_msglen);
 741         if (cc < 0) {
 742                 if (errno == ESRCH
 743                     && (action == RTM_CHANGE || action == RTM_DELETE)) {
 744                         trace_act("route disappeared before" PAT, ARGS);
 745                         if (action == RTM_CHANGE) {
 746                                 action = RTM_ADD;
 747                                 goto again;
 748                         }
 749                         return;
 750                 }
 751                 msglog("write(rt_sock)" PAT ": %s", ARGS, strerror(errno));
 752                 return;
 753         } else if (cc != w.w_rtm.rtm_msglen) {
 754                 msglog("write(rt_sock) wrote %ld instead of %d for" PAT,
 755                        cc, w.w_rtm.rtm_msglen, ARGS);
 756                 return;
 757         }
 758 #endif
 759         if (TRACEKERNEL)
 760                 trace_misc("write kernel" PAT, ARGS);
 761 #undef PAT
 762 #undef ARGS
 763 }
 764
 765
 766 #define KHASH_SIZE 71                   /* should be prime */
 767 #define KHASH(a,m) khash_bins[((a) ^ (m)) % KHASH_SIZE]
 768 static struct khash {
 769         struct khash *k_next;
 770         naddr   k_dst;
 771         naddr   k_mask;
 772         naddr   k_gate;
 773         short   k_metric;
 774         u_short k_state;
 775 #define     KS_NEW      0x001
 776 #define     KS_DELETE   0x002           /* need to delete the route */
 777 #define     KS_ADD      0x004           /* add to the kernel */
 778 #define     KS_CHANGE   0x008           /* tell kernel to change the route */
 779 #define     KS_DEL_ADD  0x010           /* delete & add to change the kernel */
 780 #define     KS_STATIC   0x020           /* Static flag in kernel */
 781 #define     KS_GATEWAY  0x040           /* G flag in kernel */
 782 #define     KS_DYNAMIC  0x080           /* result of redirect */
 783 #define     KS_DELETED  0x100           /* already deleted from kernel */
 784 #define     KS_CHECK    0x200
 785         time_t  k_keep;
 786 #define     K_KEEP_LIM  30
 787         time_t  k_redirect_time;        /* when redirected route 1st seen */
 788 } *khash_bins[KHASH_SIZE];
 789
 790
 791 static struct khash*
 792 kern_find(naddr dst, naddr mask, struct khash ***ppk)
 793 {
 794         struct khash *k, **pk;
 795
 796         for (pk = &KHASH(dst,mask); (k = *pk) != 0; pk = &k->k_next) {
 797                 if (k->k_dst == dst && k->k_mask == mask)
 798                         break;
 799         }
 800         if (ppk != 0)
 801                 *ppk = pk;
 802         return k;
 803 }
 804
 805
 806 static struct khash*
 807 kern_add(naddr dst, naddr mask)
 808 {
 809         struct khash *k, **pk;
 810
 811         k = kern_find(dst, mask, &pk);
 812         if (k != 0)
 813                 return k;
 814
 815         k = (struct khash *)rtmalloc(sizeof(*k), "kern_add");
 816
 817         memset(k, 0, sizeof(*k));
 818         k->k_dst = dst;
 819         k->k_mask = mask;
 820         k->k_state = KS_NEW;
 821         k->k_keep = now.tv_sec;
 822         *pk = k;
 823
 824         return k;
 825 }
 826
 827
 828 /* If a kernel route has a non-zero metric, check that it is still in the
 829  *      daemon table, and not deleted by interfaces coming and going.
 830  */
 831 static void
 832 kern_check_static(struct khash *k,
 833                   struct interface *ifp)
 834 {
 835         struct rt_entry *rt;
 836         struct rt_spare new;
 837
 838         if (k->k_metric == 0)
 839                 return;
 840
 841         memset(&new, 0, sizeof(new));
 842         new.rts_ifp = ifp;
 843         new.rts_gate = k->k_gate;
 844         new.rts_router = (ifp != 0) ? ifp->int_addr : loopaddr;
 845         new.rts_metric = k->k_metric;
 846         new.rts_time = now.tv_sec;
 847
 848         rt = rtget(k->k_dst, k->k_mask);
 849         if (rt != 0) {
 850                 if (!(rt->rt_state & RS_STATIC))
 851                         rtchange(rt, rt->rt_state | RS_STATIC, &new, 0);
 852         } else {
 853                 rtadd(k->k_dst, k->k_mask, RS_STATIC, &new);
 854         }
 855 }
 856
 857
 858 /* operate on a kernel entry
 859  */
 860 static void
 861 kern_ioctl(struct khash *k,
 862            int action,                  /* RTM_DELETE, etc */
 863            int flags)
 864
 865 {
 866         switch (action) {
 867         case RTM_DELETE:
 868                 k->k_state &= ~KS_DYNAMIC;
 869                 if (k->k_state & KS_DELETED)
 870                         return;
 871                 k->k_state |= KS_DELETED;
 872                 break;
 873         case RTM_ADD:
 874                 k->k_state &= ~KS_DELETED;
 875                 break;
 876         case RTM_CHANGE:
 877                 if (k->k_state & KS_DELETED) {
 878                         action = RTM_ADD;
 879                         k->k_state &= ~KS_DELETED;
 880                 }
 881                 break;
 882         }
 883
 884         rtioctl(action, k->k_dst, k->k_gate, k->k_mask, k->k_metric, flags);
 885 }
 886
 887
 888 /* add a route the kernel told us
 889  */
 890 static void
 891 rtm_add(struct rt_msghdr *rtm,
 892         struct rt_addrinfo *info,
 893         time_t keep)
 894 {
 895         struct khash *k;
 896         struct interface *ifp;
 897         naddr mask;
 898
 899
 900         if (rtm->rtm_flags & RTF_HOST) {
 901                 mask = HOST_MASK;
 902         } else if (INFO_MASK(info) != 0) {
 903                 mask = ntohl(S_ADDR(INFO_MASK(info)));
 904         } else {
 905                 msglog("ignore %s without mask", rtm_type_name(rtm->rtm_type));
 906                 return;
 907         }
 908
 909         k = kern_add(S_ADDR(INFO_DST(info)), mask);
 910         if (k->k_state & KS_NEW)
 911                 k->k_keep = now.tv_sec+keep;
 912         if (INFO_GATE(info) == 0) {
 913                 trace_act("note %s without gateway",
 914                           rtm_type_name(rtm->rtm_type));
 915                 k->k_metric = HOPCNT_INFINITY;
 916         } else if (INFO_GATE(info)->sa_family != AF_INET) {
 917                 trace_act("note %s with gateway AF=%d",
 918                           rtm_type_name(rtm->rtm_type),
 919                           INFO_GATE(info)->sa_family);
 920                 k->k_metric = HOPCNT_INFINITY;
 921         } else {
 922                 k->k_gate = S_ADDR(INFO_GATE(info));
 923                 k->k_metric = rtm->rtm_rmx.rmx_hopcount;
 924                 if (k->k_metric < 0)
 925                         k->k_metric = 0;
 926                 else if (k->k_metric > HOPCNT_INFINITY-1)
 927                         k->k_metric = HOPCNT_INFINITY-1;
 928         }
 929         k->k_state &= ~(KS_DELETE | KS_ADD | KS_CHANGE | KS_DEL_ADD
 930                         | KS_DELETED | KS_GATEWAY | KS_STATIC
 931                         | KS_NEW | KS_CHECK);
 932         if (rtm->rtm_flags & RTF_GATEWAY)
 933                 k->k_state |= KS_GATEWAY;
 934         if (rtm->rtm_flags & RTF_STATIC)
 935                 k->k_state |= KS_STATIC;
 936
 937         if (0 != (rtm->rtm_flags & (RTF_DYNAMIC | RTF_MODIFIED))) {
 938                 if (INFO_AUTHOR(info) != 0
 939                     && INFO_AUTHOR(info)->sa_family == AF_INET)
 940                         ifp = iflookup(S_ADDR(INFO_AUTHOR(info)));
 941                 else
 942                         ifp = 0;
 943                 if (supplier
 944                     && (ifp == 0 || !(ifp->int_state & IS_REDIRECT_OK))) {
 945                         /* Routers are not supposed to listen to redirects,
 946                          * so delete it if it came via an unknown interface
 947                          * or the interface does not have special permission.
 948                          */
 949                         k->k_state &= ~KS_DYNAMIC;
 950                         k->k_state |= KS_DELETE;
 951                         LIM_SEC(need_kern, 0);
 952                         trace_act("mark for deletion redirected %s --> %s"
 953                                   " via %s",
 954                                   addrname(k->k_dst, k->k_mask, 0),
 955                                   naddr_ntoa(k->k_gate),
 956                                   ifp ? ifp->int_name : "unknown interface");
 957                 } else {
 958                         k->k_state |= KS_DYNAMIC;
 959                         k->k_redirect_time = now.tv_sec;
 960                         trace_act("accept redirected %s --> %s via %s",
 961                                   addrname(k->k_dst, k->k_mask, 0),
 962                                   naddr_ntoa(k->k_gate),
 963                                   ifp ? ifp->int_name : "unknown interface");
 964                 }
 965                 return;
 966         }
 967
 968         /* If it is not a static route, quit until the next comparison
 969          * between the kernel and daemon tables, when it will be deleted.
 970          */
 971         if (!(k->k_state & KS_STATIC)) {
 972                 k->k_state |= KS_DELETE;
 973                 LIM_SEC(need_kern, k->k_keep);
 974                 return;
 975         }
 976
 977         /* Put static routes with real metrics into the daemon table so
 978          * they can be advertised.
 979          *
 980          * Find the interface toward the gateway.
 981          */
 982         ifp = iflookup(k->k_gate);
 983         if (ifp == 0)
 984                 msglog("static route %s --> %s impossibly lacks ifp",
 985                        addrname(S_ADDR(INFO_DST(info)), mask, 0),
 986                        naddr_ntoa(k->k_gate));
 987
 988         kern_check_static(k, ifp);
 989 }
 990
 991
 992 /* deal with packet loss
 993  */
 994 static void
 995 rtm_lose(struct rt_msghdr *rtm,
 996          struct rt_addrinfo *info)
 997 {
 998         if (INFO_GATE(info) == 0
 999             || INFO_GATE(info)->sa_family != AF_INET) {
1000                 trace_act("ignore %s without gateway",
1001                           rtm_type_name(rtm->rtm_type));
1002                 return;
1003         }
1004
1005         if (rdisc_ok)
1006                 rdisc_age(S_ADDR(INFO_GATE(info)));
1007         age(S_ADDR(INFO_GATE(info)));
1008 }
1009
1010
1011 /* Make the gateway slot of an info structure point to something
1012  * useful.  If it is not already useful, but it specifies an interface,
1013  * then fill in the sockaddr_in provided and point it there.
1014  */
1015 static int
1016 get_info_gate(struct sockaddr **sap,
1017               struct sockaddr_in *in)
1018 {
1019         struct sockaddr_dl *sdl = (struct sockaddr_dl *)*sap;
1020         struct interface *ifp;
1021
1022         if (sdl == 0)
1023                 return 0;
1024         if ((sdl)->sdl_family == AF_INET)
1025                 return 1;
1026         if ((sdl)->sdl_family != AF_LINK)
1027                 return 0;
1028
1029         ifp = ifwithindex(sdl->sdl_index, 1);
1030         if (ifp == 0)
1031                 return 0;
1032
1033         in->sin_addr.s_addr = ifp->int_addr;
1034 #ifdef _HAVE_SA_LEN
1035         in->sin_len = sizeof(*in);
1036 #endif
1037         in->sin_family = AF_INET;
1038         *sap = (struct sockaddr *)in;
1039
1040         return 1;
1041 }
1042
1043
1044 /* Clean the kernel table by copying it to the daemon image.
1045  * Eventually the daemon will delete any extra routes.
1046  */
1047 void
1048 flush_kern(void)
1049 {
1050         static char *sysctl_buf;
1051         static size_t sysctl_buf_size = 0;
1052         size_t needed;
1053         int mib[6];
1054         char *next, *lim;
1055         struct rt_msghdr *rtm;
1056         struct sockaddr_in gate_sin;
1057         struct rt_addrinfo info;
1058         int i;
1059         struct khash *k;
1060
1061
1062         for (i = 0; i < KHASH_SIZE; i++) {
1063                 for (k = khash_bins[i]; k != 0; k = k->k_next) {
1064                         k->k_state |= KS_CHECK;
1065                 }
1066         }
1067
1068         mib[0] = CTL_NET;
1069         mib[1] = PF_ROUTE;
1070         mib[2] = 0;             /* protocol */
1071         mib[3] = 0;             /* wildcard address family */
1072         mib[4] = NET_RT_DUMP;
1073         mib[5] = 0;             /* no flags */
1074         for (;;) {
1075                 if ((needed = sysctl_buf_size) != 0) {
1076                         if (sysctl(mib, 6, sysctl_buf,&needed, 0, 0) >= 0)
1077                                 break;
1078                         if (errno != ENOMEM && errno != EFAULT)
1079                                 BADERR(1,"flush_kern: sysctl(RT_DUMP)");
1080                         free(sysctl_buf);
1081                         needed = 0;
1082                 }
1083                 if (sysctl(mib, 6, 0, &needed, 0, 0) < 0)
1084                         BADERR(1,"flush_kern: sysctl(RT_DUMP) estimate");
1085                 /* Kludge around the habit of some systems, such as
1086                  * BSD/OS 3.1, to not admit how many routes are in the
1087                  * kernel, or at least to be quite wrong.
1088                  */
1089                 needed += 50*(sizeof(*rtm)+5*sizeof(struct sockaddr));
1090                 sysctl_buf = rtmalloc(sysctl_buf_size = needed,
1091                                       "flush_kern sysctl(RT_DUMP)");
1092         }
1093
1094         lim = sysctl_buf + needed;
1095         for (next = sysctl_buf; next < lim; next += rtm->rtm_msglen) {
1096                 rtm = (struct rt_msghdr *)next;
1097                 if (rtm->rtm_msglen == 0) {
1098                         msglog("zero length kernel route at "
1099                                " %#lx in buffer %#lx before %#lx",
1100                                (u_long)rtm, (u_long)sysctl_buf, (u_long)lim);
1101                         break;
1102                 }
1103
1104                 rt_xaddrs(&info,
1105                           (struct sockaddr *)(rtm+1),
1106                           (struct sockaddr *)(next + rtm->rtm_msglen),
1107                           rtm->rtm_addrs);
1108
1109                 if (INFO_DST(&info) == 0
1110                     || INFO_DST(&info)->sa_family != AF_INET)
1111                         continue;
1112
1113                 /* ignore ARP table entries on systems with a merged route
1114                  * and ARP table.
1115                  */
1116                 if (rtm->rtm_flags & RTF_LLINFO)
1117                         continue;
1118
1119                 /* ignore multicast addresses
1120                  */
1121                 if (IN_MULTICAST(ntohl(S_ADDR(INFO_DST(&info)))))
1122                         continue;
1123
1124                 if (!get_info_gate(&INFO_GATE(&info), &gate_sin))
1125                         continue;
1126
1127                 /* Note static routes and interface routes, and also
1128                  * preload the image of the kernel table so that
1129                  * we can later clean it, as well as avoid making
1130                  * unneeded changes.  Keep the old kernel routes for a
1131                  * few seconds to allow a RIP or router-discovery
1132                  * response to be heard.
1133                  */
1134                 rtm_add(rtm,&info,MIN_WAITTIME);
1135         }
1136
1137         for (i = 0; i < KHASH_SIZE; i++) {
1138                 for (k = khash_bins[i]; k != 0; k = k->k_next) {
1139                         if (k->k_state & KS_CHECK) {
1140                                 msglog("%s --> %s disappeared from kernel",
1141                                        addrname(k->k_dst, k->k_mask, 0),
1142                                        naddr_ntoa(k->k_gate));
1143                                 del_static(k->k_dst, k->k_mask, k->k_gate, 1);
1144                         }
1145                 }
1146         }
1147 }
1148
1149
1150 /* Listen to announcements from the kernel
1151  */
1152 void
1153 read_rt(void)
1154 {
1155         long cc;
1156         struct interface *ifp;
1157         struct sockaddr_in gate_sin;
1158         naddr mask, gate;
1159         union {
1160                 struct {
1161                         struct rt_msghdr rtm;
1162                         struct sockaddr addrs[RTAX_MAX];
1163                 } r;
1164                 struct if_msghdr ifm;
1165         } m;
1166         char str[100], *strp;
1167         struct rt_addrinfo info;
1168
1169
1170         for (;;) {
1171                 cc = read(rt_sock, &m, sizeof(m));
1172                 if (cc <= 0) {
1173                         if (cc < 0 && errno != EWOULDBLOCK)
1174                                 LOGERR("read(rt_sock)");
1175                         return;
1176                 }
1177
1178                 if (m.r.rtm.rtm_version != RTM_VERSION) {
1179                         msglog("bogus routing message version %d",
1180                                m.r.rtm.rtm_version);
1181                         continue;
1182                 }
1183
1184                 /* Ignore our own results.
1185                  */
1186                 if (m.r.rtm.rtm_type <= RTM_CHANGE
1187                     && m.r.rtm.rtm_pid == mypid) {
1188                         static int complained = 0;
1189                         if (!complained) {
1190                                 msglog("receiving our own change messages");
1191                                 complained = 1;
1192                         }
1193                         continue;
1194                 }
1195
1196                 if (m.r.rtm.rtm_type == RTM_IFINFO
1197                     || m.r.rtm.rtm_type == RTM_NEWADDR
1198                     || m.r.rtm.rtm_type == RTM_DELADDR) {
1199                         ifp = ifwithindex(m.ifm.ifm_index,
1200                                           m.r.rtm.rtm_type != RTM_DELADDR);
1201                         if (ifp == 0)
1202                                 trace_act("note %s with flags %#x"
1203                                           " for unknown interface index #%d",
1204                                           rtm_type_name(m.r.rtm.rtm_type),
1205                                           m.ifm.ifm_flags,
1206                                           m.ifm.ifm_index);
1207                         else
1208                                 trace_act("note %s with flags %#x for %s",
1209                                           rtm_type_name(m.r.rtm.rtm_type),
1210                                           m.ifm.ifm_flags,
1211                                           ifp->int_name);
1212
1213                         /* After being informed of a change to an interface,
1214                          * check them all now if the check would otherwise
1215                          * be a long time from now, if the interface is
1216                          * not known, or if the interface has been turned
1217                          * off or on.
1218                          */
1219                         if (ifinit_timer.tv_sec-now.tv_sec>=CHECK_BAD_INTERVAL
1220                             || ifp == 0
1221                             || ((ifp->int_if_flags ^ m.ifm.ifm_flags)
1222                                 & IFF_UP) != 0)
1223                                 ifinit_timer.tv_sec = now.tv_sec;
1224                         continue;
1225                 }
1226
1227                 strcpy(str, rtm_type_name(m.r.rtm.rtm_type));
1228                 strp = &str[strlen(str)];
1229                 if (m.r.rtm.rtm_type <= RTM_CHANGE)
1230                         strp += sprintf(strp," from pid %d",m.r.rtm.rtm_pid);
1231
1232                 rt_xaddrs(&info, m.r.addrs, &m.r.addrs[RTAX_MAX],
1233                           m.r.rtm.rtm_addrs);
1234
1235                 if (INFO_DST(&info) == 0) {
1236                         trace_act("ignore %s without dst", str);
1237                         continue;
1238                 }
1239
1240                 if (INFO_DST(&info)->sa_family != AF_INET) {
1241                         trace_act("ignore %s for AF %d", str,
1242                                   INFO_DST(&info)->sa_family);
1243                         continue;
1244                 }
1245
1246                 mask = ((INFO_MASK(&info) != 0)
1247                         ? ntohl(S_ADDR(INFO_MASK(&info)))
1248                         : (m.r.rtm.rtm_flags & RTF_HOST)
1249                         ? HOST_MASK
1250                         : std_mask(S_ADDR(INFO_DST(&info))));
1251
1252                 strp += sprintf(strp, ": %s",
1253                                 addrname(S_ADDR(INFO_DST(&info)), mask, 0));
1254
1255                 if (IN_MULTICAST(ntohl(S_ADDR(INFO_DST(&info))))) {
1256                         trace_act("ignore multicast %s", str);
1257                         continue;
1258                 }
1259
1260                 if (m.r.rtm.rtm_flags & RTF_LLINFO) {
1261                         trace_act("ignore ARP %s", str);
1262                         continue;
1263                 }
1264
1265                 if (get_info_gate(&INFO_GATE(&info), &gate_sin)) {
1266                         gate = S_ADDR(INFO_GATE(&info));
1267                         strp += sprintf(strp, " --> %s", naddr_ntoa(gate));
1268                 } else {
1269                         gate = 0;
1270                 }
1271
1272                 if (INFO_AUTHOR(&info) != 0)
1273                         strp += sprintf(strp, " by authority of %s",
1274                                         saddr_ntoa(INFO_AUTHOR(&info)));
1275
1276                 switch (m.r.rtm.rtm_type) {
1277                 case RTM_ADD:
1278                 case RTM_CHANGE:
1279                 case RTM_REDIRECT:
1280                         if (m.r.rtm.rtm_errno != 0) {
1281                                 trace_act("ignore %s with \"%s\" error",
1282                                           str, strerror(m.r.rtm.rtm_errno));
1283                         } else {
1284                                 trace_act("%s", str);
1285                                 rtm_add(&m.r.rtm,&info,0);
1286                         }
1287                         break;
1288
1289                 case RTM_DELETE:
1290                         if (m.r.rtm.rtm_errno != 0
1291                             && m.r.rtm.rtm_errno != ESRCH) {
1292                                 trace_act("ignore %s with \"%s\" error",
1293                                           str, strerror(m.r.rtm.rtm_errno));
1294                         } else {
1295                                 trace_act("%s", str);
1296                                 del_static(S_ADDR(INFO_DST(&info)), mask,
1297                                            gate, 1);
1298                         }
1299                         break;
1300
1301                 case RTM_LOSING:
1302                         trace_act("%s", str);
1303                         rtm_lose(&m.r.rtm,&info);
1304                         break;
1305
1306                 default:
1307                         trace_act("ignore %s", str);
1308                         break;
1309                 }
1310         }
1311 }
1312
1313
1314 /* after aggregating, note routes that belong in the kernel
1315  */
1316 static void
1317 kern_out(struct ag_info *ag)
1318 {
1319         struct khash *k;
1320
1321
1322         /* Do not install bad routes if they are not already present.
1323          * This includes routes that had RS_NET_SYN for interfaces that
1324          * recently died.
1325          */
1326         if (ag->ag_metric == HOPCNT_INFINITY) {
1327                 k = kern_find(htonl(ag->ag_dst_h), ag->ag_mask, 0);
1328                 if (k == 0)
1329                         return;
1330         } else {
1331                 k = kern_add(htonl(ag->ag_dst_h), ag->ag_mask);
1332         }
1333
1334         if (k->k_state & KS_NEW) {
1335                 /* will need to add new entry to the kernel table */
1336                 k->k_state = KS_ADD;
1337                 if (ag->ag_state & AGS_GATEWAY)
1338                         k->k_state |= KS_GATEWAY;
1339                 k->k_gate = ag->ag_gate;
1340                 k->k_metric = ag->ag_metric;
1341                 return;
1342         }
1343
1344         if (k->k_state & KS_STATIC)
1345                 return;
1346
1347         /* modify existing kernel entry if necessary */
1348         if (k->k_gate != ag->ag_gate
1349             || k->k_metric != ag->ag_metric) {
1350                 /* Must delete bad interface routes etc. to change them. */
1351                 if (k->k_metric == HOPCNT_INFINITY)
1352                         k->k_state |= KS_DEL_ADD;
1353                 k->k_gate = ag->ag_gate;
1354                 k->k_metric = ag->ag_metric;
1355                 k->k_state |= KS_CHANGE;
1356         }
1357
1358         /* If the daemon thinks the route should exist, forget
1359          * about any redirections.
1360          * If the daemon thinks the route should exist, eventually
1361          * override manual intervention by the operator.
1362          */
1363         if ((k->k_state & (KS_DYNAMIC | KS_DELETED)) != 0) {
1364                 k->k_state &= ~KS_DYNAMIC;
1365                 k->k_state |= (KS_ADD | KS_DEL_ADD);
1366         }
1367
1368         if ((k->k_state & KS_GATEWAY)
1369             && !(ag->ag_state & AGS_GATEWAY)) {
1370                 k->k_state &= ~KS_GATEWAY;
1371                 k->k_state |= (KS_ADD | KS_DEL_ADD);
1372         } else if (!(k->k_state & KS_GATEWAY)
1373                    && (ag->ag_state & AGS_GATEWAY)) {
1374                 k->k_state |= KS_GATEWAY;
1375                 k->k_state |= (KS_ADD | KS_DEL_ADD);
1376         }
1377
1378         /* Deleting-and-adding is necessary to change aspects of a route.
1379          * Just delete instead of deleting and then adding a bad route.
1380          * Otherwise, we want to keep the route in the kernel.
1381          */
1382         if (k->k_metric == HOPCNT_INFINITY
1383             && (k->k_state & KS_DEL_ADD))
1384                 k->k_state |= KS_DELETE;
1385         else
1386                 k->k_state &= ~KS_DELETE;
1387 #undef RT
1388 }
1389
1390
1391 /* ARGSUSED */
1392 static int
1393 walk_kern(struct radix_node *rn,
1394           struct walkarg *argp UNUSED)
1395 {
1396 #define RT ((struct rt_entry *)rn)
1397         char metric, pref;
1398         u_int ags = 0;
1399
1400
1401         /* Do not install synthetic routes */
1402         if (RT->rt_state & RS_NET_SYN)
1403                 return 0;
1404
1405         if (!(RT->rt_state & RS_IF)) {
1406                 /* This is an ordinary route, not for an interface.
1407                  */
1408
1409                 /* aggregate, ordinary good routes without regard to
1410                  * their metric
1411                  */
1412                 pref = 1;
1413                 ags |= (AGS_GATEWAY | AGS_SUPPRESS | AGS_AGGREGATE);
1414
1415                 /* Do not install host routes directly to hosts, to avoid
1416                  * interfering with ARP entries in the kernel table.
1417                  */
1418                 if (RT_ISHOST(RT)
1419                     && ntohl(RT->rt_dst) == RT->rt_gate)
1420                         return 0;
1421
1422         } else {
1423                 /* This is an interface route.
1424                  * Do not install routes for "external" remote interfaces.
1425                  */
1426                 if (RT->rt_ifp != 0 && (RT->rt_ifp->int_state & IS_EXTERNAL))
1427                         return 0;
1428
1429                 /* Interfaces should override received routes.
1430                  */
1431                 pref = 0;
1432                 ags |= (AGS_IF | AGS_CORS_GATE);
1433
1434                 /* If it is not an interface, or an alias for an interface,
1435                  * it must be a "gateway."
1436                  *
1437                  * If it is a "remote" interface, it is also a "gateway" to
1438                  * the kernel if is not a alias.
1439                  */
1440                 if (RT->rt_ifp == 0
1441                     || (RT->rt_ifp->int_state & IS_REMOTE))
1442                         ags |= (AGS_GATEWAY | AGS_SUPPRESS | AGS_AGGREGATE);
1443         }
1444
1445         /* If RIP is off and IRDP is on, let the route to the discovered
1446          * route suppress any RIP routes.  Eventually the RIP routes
1447          * will time-out and be deleted.  This reaches the steady-state
1448          * quicker.
1449          */
1450         if ((RT->rt_state & RS_RDISC) && rip_sock < 0)
1451                 ags |= AGS_CORS_GATE;
1452
1453         metric = RT->rt_metric;
1454         if (metric == HOPCNT_INFINITY) {
1455                 /* if the route is dead, so try hard to aggregate. */
1456                 pref = HOPCNT_INFINITY;
1457                 ags |= (AGS_FINE_GATE | AGS_SUPPRESS);
1458                 ags &= ~(AGS_IF | AGS_CORS_GATE);
1459         }
1460
1461         ag_check(RT->rt_dst, RT->rt_mask, RT->rt_gate, 0,
1462                  metric,pref, 0, 0, ags, kern_out);
1463         return 0;
1464 #undef RT
1465 }
1466
1467
1468 /* Update the kernel table to match the daemon table.
1469  */
1470 static void
1471 fix_kern(void)
1472 {
1473         int i;
1474         struct khash *k, **pk;
1475
1476
1477         need_kern = age_timer;
1478
1479         /* Walk daemon table, updating the copy of the kernel table.
1480          */
1481         rn_walktree(rhead, walk_kern, 0);
1482         ag_flush(0,0,kern_out);
1483
1484         for (i = 0; i < KHASH_SIZE; i++) {
1485                 for (pk = &khash_bins[i]; (k = *pk) != 0; ) {
1486                         /* Do not touch static routes */
1487                         if (k->k_state & KS_STATIC) {
1488                                 kern_check_static(k,0);
1489                                 pk = &k->k_next;
1490                                 continue;
1491                         }
1492
1493                         /* check hold on routes deleted by the operator */
1494                         if (k->k_keep > now.tv_sec) {
1495                                 /* ensure we check when the hold is over */
1496                                 LIM_SEC(need_kern, k->k_keep);
1497                                 /* mark for the next cycle */
1498                                 k->k_state |= KS_DELETE;
1499                                 pk = &k->k_next;
1500                                 continue;
1501                         }
1502
1503                         if ((k->k_state & KS_DELETE)
1504                             && !(k->k_state & KS_DYNAMIC)) {
1505                                 kern_ioctl(k, RTM_DELETE, 0);
1506                                 *pk = k->k_next;
1507                                 free(k);
1508                                 continue;
1509                         }
1510
1511                         if (k->k_state & KS_DEL_ADD)
1512                                 kern_ioctl(k, RTM_DELETE, 0);
1513
1514                         if (k->k_state & KS_ADD) {
1515                                 kern_ioctl(k, RTM_ADD,
1516                                            ((0 != (k->k_state & (KS_GATEWAY
1517                                                         | KS_DYNAMIC)))
1518                                             ? RTF_GATEWAY : 0));
1519                         } else if (k->k_state & KS_CHANGE) {
1520                                 kern_ioctl(k,  RTM_CHANGE,
1521                                            ((0 != (k->k_state & (KS_GATEWAY
1522                                                         | KS_DYNAMIC)))
1523                                             ? RTF_GATEWAY : 0));
1524                         }
1525                         k->k_state &= ~(KS_ADD|KS_CHANGE|KS_DEL_ADD);
1526
1527                         /* Mark this route to be deleted in the next cycle.
1528                          * This deletes routes that disappear from the
1529                          * daemon table, since the normal aging code
1530                          * will clear the bit for routes that have not
1531                          * disappeared from the daemon table.
1532                          */
1533                         k->k_state |= KS_DELETE;
1534                         pk = &k->k_next;
1535                 }
1536         }
1537 }
1538
1539
1540 /* Delete a static route in the image of the kernel table.
1541  */
1542 void
1543 del_static(naddr dst,
1544            naddr mask,
1545            naddr gate,
1546            int gone)
1547 {
1548         struct khash *k;
1549         struct rt_entry *rt;
1550
1551         /* Just mark it in the table to be deleted next time the kernel
1552          * table is updated.
1553          * If it has already been deleted, mark it as such, and set its
1554          * keep-timer so that it will not be deleted again for a while.
1555          * This lets the operator delete a route added by the daemon
1556          * and add a replacement.
1557          */
1558         k = kern_find(dst, mask, 0);
1559         if (k != 0 && (gate == 0 || k->k_gate == gate)) {
1560                 k->k_state &= ~(KS_STATIC | KS_DYNAMIC | KS_CHECK);
1561                 k->k_state |= KS_DELETE;
1562                 if (gone) {
1563                         k->k_state |= KS_DELETED;
1564                         k->k_keep = now.tv_sec + K_KEEP_LIM;
1565                 }
1566         }
1567
1568         rt = rtget(dst, mask);
1569         if (rt != 0 && (rt->rt_state & RS_STATIC))
1570                 rtbad(rt);
1571 }
1572
1573
1574 /* Delete all routes generated from ICMP Redirects that use a given gateway,
1575  * as well as old redirected routes.
1576  */
1577 void
1578 del_redirects(naddr bad_gate,
1579               time_t old)
1580 {
1581         int i;
1582         struct khash *k;
1583
1584
1585         for (i = 0; i < KHASH_SIZE; i++) {
1586                 for (k = khash_bins[i]; k != 0; k = k->k_next) {
1587                         if (!(k->k_state & KS_DYNAMIC)
1588                             || (k->k_state & KS_STATIC))
1589                                 continue;
1590
1591                         if (k->k_gate != bad_gate
1592                             && k->k_redirect_time > old
1593                             && !supplier)
1594                                 continue;
1595
1596                         k->k_state |= KS_DELETE;
1597                         k->k_state &= ~KS_DYNAMIC;
1598                         need_kern.tv_sec = now.tv_sec;
1599                         trace_act("mark redirected %s --> %s for deletion",
1600                                   addrname(k->k_dst, k->k_mask, 0),
1601                                   naddr_ntoa(k->k_gate));
1602                 }
1603         }
1604 }
1605
1606
1607 /* Start the daemon tables.
1608  */
1609 extern int max_keylen;
1610
1611 void
1612 rtinit(void)
1613 {
1614         int i;
1615         struct ag_info *ag;
1616
1617         /* Initialize the radix trees */
1618         max_keylen = sizeof(struct sockaddr_in);
1619         rn_init();
1620         rn_inithead(&rhead, 32);
1621
1622         /* mark all of the slots in the table free */
1623         ag_avail = ag_slots;
1624         for (ag = ag_slots, i = 1; i < NUM_AG_SLOTS; i++) {
1625                 ag->ag_fine = ag+1;
1626                 ag++;
1627         }
1628 }
1629
1630
1631 #ifdef _HAVE_SIN_LEN
1632 static struct sockaddr_in dst_sock = {sizeof(dst_sock), AF_INET, 0, {0}, {0}};
1633 static struct sockaddr_in mask_sock = {sizeof(mask_sock), AF_INET, 0, {0}, {0}};
1634 #else
1635 static struct sockaddr_in_new dst_sock = {_SIN_ADDR_SIZE, AF_INET};
1636 static struct sockaddr_in_new mask_sock = {_SIN_ADDR_SIZE, AF_INET};
1637 #endif
1638
1639
1640 static void
1641 set_need_flash(void)
1642 {
1643         if (!need_flash) {
1644                 need_flash = 1;
1645                 /* Do not send the flash update immediately.  Wait a little
1646                  * while to hear from other routers.
1647                  */
1648                 no_flash.tv_sec = now.tv_sec + MIN_WAITTIME;
1649         }
1650 }
1651
1652
1653 /* Get a particular routing table entry
1654  */
1655 struct rt_entry *
1656 rtget(naddr dst, naddr mask)
1657 {
1658         struct rt_entry *rt;
1659
1660         dst_sock.sin_addr.s_addr = dst;
1661         mask_sock.sin_addr.s_addr = htonl(mask);
1662         masktrim(&mask_sock);
1663         rt = (struct rt_entry *)rhead->rnh_lookup(&dst_sock,&mask_sock,rhead);
1664         if (!rt
1665             || rt->rt_dst != dst
1666             || rt->rt_mask != mask)
1667                 return 0;
1668
1669         return rt;
1670 }
1671
1672
1673 /* Find a route to dst as the kernel would.
1674  */
1675 struct rt_entry *
1676 rtfind(naddr dst)
1677 {
1678         dst_sock.sin_addr.s_addr = dst;
1679         return (struct rt_entry *)rhead->rnh_matchaddr(&dst_sock, rhead);
1680 }
1681
1682
1683 /* add a route to the table
1684  */
1685 void
1686 rtadd(naddr     dst,
1687       naddr     mask,
1688       u_int     state,                  /* rt_state for the entry */
1689       struct    rt_spare *new)
1690 {
1691         struct rt_entry *rt;
1692         naddr smask;
1693         int i;
1694         struct rt_spare *rts;
1695
1696         rt = (struct rt_entry *)rtmalloc(sizeof (*rt), "rtadd");
1697         memset(rt, 0, sizeof(*rt));
1698         for (rts = rt->rt_spares, i = NUM_SPARES; i != 0; i--, rts++)
1699                 rts->rts_metric = HOPCNT_INFINITY;
1700
1701         rt->rt_nodes->rn_key = (caddr_t)&rt->rt_dst_sock;
1702         rt->rt_dst = dst;
1703         rt->rt_dst_sock.sin_family = AF_INET;
1704 #ifdef _HAVE_SIN_LEN
1705         rt->rt_dst_sock.sin_len = dst_sock.sin_len;
1706 #endif
1707         if (mask != HOST_MASK) {
1708                 smask = std_mask(dst);
1709                 if ((smask & ~mask) == 0 && mask > smask)
1710                         state |= RS_SUBNET;
1711         }
1712         mask_sock.sin_addr.s_addr = htonl(mask);
1713         masktrim(&mask_sock);
1714         rt->rt_mask = mask;
1715         rt->rt_state = state;
1716         rt->rt_spares[0] = *new;
1717         rt->rt_time = now.tv_sec;
1718         rt->rt_poison_metric = HOPCNT_INFINITY;
1719         rt->rt_seqno = update_seqno;
1720
1721         if (++total_routes == MAX_ROUTES)
1722                 msglog("have maximum (%d) routes", total_routes);
1723         if (TRACEACTIONS)
1724                 trace_add_del("Add", rt);
1725
1726         need_kern.tv_sec = now.tv_sec;
1727         set_need_flash();
1728
1729         if (0 == rhead->rnh_addaddr(&rt->rt_dst_sock, &mask_sock,
1730                                     rhead, rt->rt_nodes)) {
1731                 msglog("rnh_addaddr() failed for %s mask=%#lx",
1732                        naddr_ntoa(dst), (u_long)mask);
1733                 free(rt);
1734         }
1735 }
1736
1737
1738 /* notice a changed route
1739  */
1740 void
1741 rtchange(struct rt_entry *rt,
1742          u_int  state,                  /* new state bits */
1743          struct rt_spare *new,
1744          char   *label)
1745 {
1746         if (rt->rt_metric != new->rts_metric) {
1747                 /* Fix the kernel immediately if it seems the route
1748                  * has gone bad, since there may be a working route that
1749                  * aggregates this route.
1750                  */
1751                 if (new->rts_metric == HOPCNT_INFINITY) {
1752                         need_kern.tv_sec = now.tv_sec;
1753                         if (new->rts_time >= now.tv_sec - EXPIRE_TIME)
1754                                 new->rts_time = now.tv_sec - EXPIRE_TIME;
1755                 }
1756                 rt->rt_seqno = update_seqno;
1757                 set_need_flash();
1758         }
1759
1760         if (rt->rt_gate != new->rts_gate) {
1761                 need_kern.tv_sec = now.tv_sec;
1762                 rt->rt_seqno = update_seqno;
1763                 set_need_flash();
1764         }
1765
1766         state |= (rt->rt_state & RS_SUBNET);
1767
1768         /* Keep various things from deciding ageless routes are stale.
1769          */
1770         if (!AGE_RT(state, new->rts_ifp))
1771                 new->rts_time = now.tv_sec;
1772
1773         if (TRACEACTIONS)
1774                 trace_change(rt, state, new,
1775                              label ? label : "Chg   ");
1776
1777         rt->rt_state = state;
1778         rt->rt_spares[0] = *new;
1779 }
1780
1781
1782 /* check for a better route among the spares
1783  */
1784 static struct rt_spare *
1785 rts_better(struct rt_entry *rt)
1786 {
1787         struct rt_spare *rts, *rts1;
1788         int i;
1789
1790         /* find the best alternative among the spares */
1791         rts = rt->rt_spares+1;
1792         for (i = NUM_SPARES, rts1 = rts+1; i > 2; i--, rts1++) {
1793                 if (BETTER_LINK(rt,rts1,rts))
1794                         rts = rts1;
1795         }
1796
1797         return rts;
1798 }
1799
1800
1801 /* switch to a backup route
1802  */
1803 void
1804 rtswitch(struct rt_entry *rt,
1805          struct rt_spare *rts)
1806 {
1807         struct rt_spare swap;
1808         char label[10];
1809
1810
1811         /* Do not change permanent routes */
1812         if (0 != (rt->rt_state & (RS_MHOME | RS_STATIC | RS_RDISC
1813                                   | RS_NET_SYN | RS_IF)))
1814                 return;
1815
1816         /* find the best alternative among the spares */
1817         if (rts == 0)
1818                 rts = rts_better(rt);
1819
1820         /* Do not bother if it is not worthwhile.
1821          */
1822         if (!BETTER_LINK(rt, rts, rt->rt_spares))
1823                 return;
1824
1825         swap = rt->rt_spares[0];
1826         sprintf(label, "Use #%d", (int)(rts - rt->rt_spares));
1827         rtchange(rt, rt->rt_state & ~(RS_NET_SYN | RS_RDISC), rts, label);
1828         if (swap.rts_metric == HOPCNT_INFINITY) {
1829                 *rts = rts_empty;
1830         } else {
1831                 *rts = swap;
1832         }
1833 }
1834
1835
1836 void
1837 rtdelete(struct rt_entry *rt)
1838 {
1839         struct khash *k;
1840
1841
1842         if (TRACEACTIONS)
1843                 trace_add_del("Del", rt);
1844
1845         k = kern_find(rt->rt_dst, rt->rt_mask, 0);
1846         if (k != 0) {
1847                 k->k_state |= KS_DELETE;
1848                 need_kern.tv_sec = now.tv_sec;
1849         }
1850
1851         dst_sock.sin_addr.s_addr = rt->rt_dst;
1852         mask_sock.sin_addr.s_addr = htonl(rt->rt_mask);
1853         masktrim(&mask_sock);
1854         if (rt != (struct rt_entry *)rhead->rnh_deladdr(&dst_sock, &mask_sock,
1855                                                         rhead)) {
1856                 msglog("rnh_deladdr() failed");
1857         } else {
1858                 free(rt);
1859                 total_routes--;
1860         }
1861 }
1862
1863
1864 void
1865 rts_delete(struct rt_entry *rt,
1866            struct rt_spare *rts)
1867 {
1868         trace_upslot(rt, rts, &rts_empty);
1869         *rts = rts_empty;
1870 }
1871
1872
1873 /* Get rid of a bad route, and try to switch to a replacement.
1874  */
1875 void
1876 rtbad(struct rt_entry *rt)
1877 {
1878         struct rt_spare new;
1879
1880         /* Poison the route */
1881         new = rt->rt_spares[0];
1882         new.rts_metric = HOPCNT_INFINITY;
1883         rtchange(rt, rt->rt_state & ~(RS_IF | RS_LOCAL | RS_STATIC), &new, 0);
1884         rtswitch(rt, 0);
1885 }
1886
1887
1888 /* Junk a RS_NET_SYN or RS_LOCAL route,
1889  *      unless it is needed by another interface.
1890  */
1891 void
1892 rtbad_sub(struct rt_entry *rt)
1893 {
1894         struct interface *ifp, *ifp1;
1895         struct intnet *intnetp;
1896         u_int state;
1897
1898
1899         ifp1 = 0;
1900         state = 0;
1901
1902         if (rt->rt_state & RS_LOCAL) {
1903                 /* Is this the route through loopback for the interface?
1904                  * If so, see if it is used by any other interfaces, such
1905                  * as a point-to-point interface with the same local address.
1906                  */
1907                 for (ifp = ifnet; ifp != 0; ifp = ifp->int_next) {
1908                         /* Retain it if another interface needs it.
1909                          */
1910                         if (ifp->int_addr == rt->rt_ifp->int_addr) {
1911                                 state |= RS_LOCAL;
1912                                 ifp1 = ifp;
1913                                 break;
1914                         }
1915                 }
1916
1917         }
1918
1919         if (!(state & RS_LOCAL)) {
1920                 /* Retain RIPv1 logical network route if there is another
1921                  * interface that justifies it.
1922                  */
1923                 if (rt->rt_state & RS_NET_SYN) {
1924                         for (ifp = ifnet; ifp != 0; ifp = ifp->int_next) {
1925                                 if ((ifp->int_state & IS_NEED_NET_SYN)
1926                                     && rt->rt_mask == ifp->int_std_mask
1927                                     && rt->rt_dst == ifp->int_std_addr) {
1928                                         state |= RS_NET_SYN;
1929                                         ifp1 = ifp;
1930                                         break;
1931                                 }
1932                         }
1933                 }
1934
1935                 /* or if there is an authority route that needs it. */
1936                 for (intnetp = intnets;
1937                      intnetp != 0;
1938                      intnetp = intnetp->intnet_next) {
1939                         if (intnetp->intnet_addr == rt->rt_dst
1940                             && intnetp->intnet_mask == rt->rt_mask) {
1941                                 state |= (RS_NET_SYN | RS_NET_INT);
1942                                 break;
1943                         }
1944                 }
1945         }
1946
1947         if (ifp1 != 0 || (state & RS_NET_SYN)) {
1948                 struct rt_spare new = rt->rt_spares[0];
1949                 new.rts_ifp = ifp1;
1950                 rtchange(rt, ((rt->rt_state & ~(RS_NET_SYN|RS_LOCAL)) | state),
1951                          &new, 0);
1952         } else {
1953                 rtbad(rt);
1954         }
1955 }
1956
1957
1958 /* Called while walking the table looking for sick interfaces
1959  * or after a time change.
1960  */
1961 /* ARGSUSED */
1962 int
1963 walk_bad(struct radix_node *rn,
1964          struct walkarg *argp UNUSED)
1965 {
1966 #define RT ((struct rt_entry *)rn)
1967         struct rt_spare *rts;
1968         int i;
1969
1970
1971         /* fix any spare routes through the interface
1972          */
1973         rts = RT->rt_spares;
1974         for (i = NUM_SPARES; i != 1; i--) {
1975                 rts++;
1976                 if (rts->rts_metric < HOPCNT_INFINITY
1977                     && (rts->rts_ifp == 0
1978                         || (rts->rts_ifp->int_state & IS_BROKE)))
1979                         rts_delete(RT, rts);
1980         }
1981
1982         /* Deal with the main route
1983          */
1984         /* finished if it has been handled before or if its interface is ok
1985          */
1986         if (RT->rt_ifp == 0 || !(RT->rt_ifp->int_state & IS_BROKE))
1987                 return 0;
1988
1989         /* Bad routes for other than interfaces are easy.
1990          */
1991         if (0 == (RT->rt_state & (RS_IF | RS_NET_SYN | RS_LOCAL))) {
1992                 rtbad(RT);
1993                 return 0;
1994         }
1995
1996         rtbad_sub(RT);
1997         return 0;
1998 #undef RT
1999 }
2000
2001
2002 /* Check the age of an individual route.
2003  */
2004 /* ARGSUSED */
2005 static int
2006 walk_age(struct radix_node *rn,
2007            struct walkarg *argp UNUSED)
2008 {
2009 #define RT ((struct rt_entry *)rn)
2010         struct interface *ifp;
2011         struct rt_spare *rts;
2012         int i;
2013
2014
2015         /* age all of the spare routes, including the primary route
2016          * currently in use
2017          */
2018         rts = RT->rt_spares;
2019         for (i = NUM_SPARES; i != 0; i--, rts++) {
2020
2021                 ifp = rts->rts_ifp;
2022                 if (i == NUM_SPARES) {
2023                         if (!AGE_RT(RT->rt_state, ifp)) {
2024                                 /* Keep various things from deciding ageless
2025                                  * routes are stale
2026                                  */
2027                                 rts->rts_time = now.tv_sec;
2028                                 continue;
2029                         }
2030
2031                         /* forget RIP routes after RIP has been turned off.
2032                          */
2033                         if (rip_sock < 0) {
2034                                 rtdelete(RT);
2035                                 return 0;
2036                         }
2037                 }
2038
2039                 /* age failing routes
2040                  */
2041                 if (age_bad_gate == rts->rts_gate
2042                     && rts->rts_time >= now_stale) {
2043                         rts->rts_time -= SUPPLY_INTERVAL;
2044                 }
2045
2046                 /* trash the spare routes when they go bad */
2047                 if (rts->rts_metric < HOPCNT_INFINITY
2048                     && now_garbage > rts->rts_time
2049                     && i != NUM_SPARES)
2050                         rts_delete(RT, rts);
2051         }
2052
2053
2054         /* finished if the active route is still fresh */
2055         if (now_stale <= RT->rt_time)
2056                 return 0;
2057
2058         /* try to switch to an alternative */
2059         rtswitch(RT, 0);
2060
2061         /* Delete a dead route after it has been publically mourned. */
2062         if (now_garbage > RT->rt_time) {
2063                 rtdelete(RT);
2064                 return 0;
2065         }
2066
2067         /* Start poisoning a bad route before deleting it. */
2068         if (now.tv_sec - RT->rt_time > EXPIRE_TIME) {
2069                 struct rt_spare new = RT->rt_spares[0];
2070                 new.rts_metric = HOPCNT_INFINITY;
2071                 rtchange(RT, RT->rt_state, &new, 0);
2072         }
2073         return 0;
2074 }
2075
2076
2077 /* Watch for dead routes and interfaces.
2078  */
2079 void
2080 age(naddr bad_gate)
2081 {
2082         struct interface *ifp;
2083         int need_query = 0;
2084
2085         /* If not listening to RIP, there is no need to age the routes in
2086          * the table.
2087          */
2088         age_timer.tv_sec = (now.tv_sec
2089                             + ((rip_sock < 0) ? NEVER : SUPPLY_INTERVAL));
2090
2091         /* Check for dead IS_REMOTE interfaces by timing their
2092          * transmissions.
2093          */
2094         for (ifp = ifnet; ifp; ifp = ifp->int_next) {
2095                 if (!(ifp->int_state & IS_REMOTE))
2096                         continue;
2097
2098                 /* ignore unreachable remote interfaces */
2099                 if (!check_remote(ifp))
2100                         continue;
2101
2102                 /* Restore remote interface that has become reachable
2103                  */
2104                 if (ifp->int_state & IS_BROKE)
2105                         if_ok(ifp, "remote ");
2106
2107                 if (ifp->int_act_time != NEVER
2108                     && now.tv_sec - ifp->int_act_time > EXPIRE_TIME) {
2109                         msglog("remote interface %s to %s timed out after"
2110                                " %ld:%ld",
2111                                ifp->int_name,
2112                                naddr_ntoa(ifp->int_dstaddr),
2113                                (now.tv_sec - ifp->int_act_time)/60,
2114                                (now.tv_sec - ifp->int_act_time)%60);
2115                         if_sick(ifp);
2116                 }
2117
2118                 /* If we have not heard from the other router
2119                  * recently, ask it.
2120                  */
2121                 if (now.tv_sec >= ifp->int_query_time) {
2122                         ifp->int_query_time = NEVER;
2123                         need_query = 1;
2124                 }
2125         }
2126
2127         /* Age routes. */
2128         age_bad_gate = bad_gate;
2129         rn_walktree(rhead, walk_age, 0);
2130
2131         /* delete old redirected routes to keep the kernel table small
2132          * and prevent blackholes
2133          */
2134         del_redirects(bad_gate, now.tv_sec-STALE_TIME);
2135
2136         /* Update the kernel routing table. */
2137         fix_kern();
2138
2139         /* poke reticent remote gateways */
2140         if (need_query)
2141                 rip_query();
2142 }