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. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)tables.c 8.1 (Berkeley) 6/5/93 30 * $FreeBSD: src/sbin/routed/table.c,v 1.9.2.2 2000/08/14 17:00:04 sheldonh Exp $ 31 */ 32 33 #include "defs.h" 34 35 static struct rt_spare *rts_better(struct rt_entry *); 36 static struct rt_spare rts_empty = {0,0,0,HOPCNT_INFINITY,0,0,0}; 37 static void set_need_flash(void); 38 static void masktrim(struct sockaddr_in *ap); 39 40 41 struct radix_node_head *rhead; /* root of the radix tree */ 42 43 int need_flash = 1; /* flash update needed 44 * start =1 to suppress the 1st 45 */ 46 47 struct timeval age_timer; /* next check of old routes */ 48 struct timeval need_kern = { /* need to update kernel table */ 49 EPOCH+MIN_WAITTIME-1, 0 50 }; 51 52 int stopint; 53 54 int total_routes; 55 56 /* zap any old routes through this gateway */ 57 naddr age_bad_gate; 58 59 60 /* It is desirable to "aggregate" routes, to combine differing routes of 61 * the same metric and next hop into a common route with a smaller netmask 62 * or to suppress redundant routes, routes that add no information to 63 * routes with smaller netmasks. 64 * 65 * A route is redundant if and only if any and all routes with smaller 66 * but matching netmasks and nets are the same. Since routes are 67 * kept sorted in the radix tree, redundant routes always come second. 68 * 69 * There are two kinds of aggregations. First, two routes of the same bit 70 * mask and differing only in the least significant bit of the network 71 * number can be combined into a single route with a coarser mask. 72 * 73 * Second, a route can be suppressed in favor of another route with a more 74 * coarse mask provided no incompatible routes with intermediate masks 75 * are present. The second kind of aggregation involves suppressing routes. 76 * A route must not be suppressed if an incompatible route exists with 77 * an intermediate mask, since the suppressed route would be covered 78 * by the intermediate. 79 * 80 * This code relies on the radix tree walk encountering routes 81 * sorted first by address, with the smallest address first. 82 */ 83 84 struct ag_info ag_slots[NUM_AG_SLOTS], *ag_avail, *ag_corsest, *ag_finest; 85 86 /* #define DEBUG_AG */ 87 #ifdef DEBUG_AG 88 #define CHECK_AG() {int acnt = 0; struct ag_info *cag; \ 89 for (cag = ag_avail; cag != NULL; cag = cag->ag_fine) \ 90 acnt++; \ 91 for (cag = ag_corsest; cag != NULL; cag = cag->ag_fine) \ 92 acnt++; \ 93 if (acnt != NUM_AG_SLOTS) { \ 94 fflush(stderr); \ 95 abort(); \ 96 } \ 97 } 98 #else 99 #define CHECK_AG() 100 #endif 101 102 103 /* Output the contents of an aggregation table slot. 104 * This function must always be immediately followed with the deletion 105 * of the target slot. 106 */ 107 static void 108 ag_out(struct ag_info *ag, 109 void (*out)(struct ag_info *)) 110 { 111 struct ag_info *ag_cors; 112 naddr bit; 113 114 115 /* Forget it if this route should not be output for split-horizon. */ 116 if (ag->ag_state & AGS_SPLIT_HZ) 117 return; 118 119 /* If we output both the even and odd twins, then the immediate parent, 120 * if it is present, is redundant, unless the parent manages to 121 * aggregate into something coarser. 122 * On successive calls, this code detects the even and odd twins, 123 * and marks the parent. 124 * 125 * Note that the order in which the radix tree code emits routes 126 * ensures that the twins are seen before the parent is emitted. 127 */ 128 ag_cors = ag->ag_cors; 129 if (ag_cors != NULL 130 && ag_cors->ag_mask == ag->ag_mask<<1 131 && ag_cors->ag_dst_h == (ag->ag_dst_h & ag_cors->ag_mask)) { 132 ag_cors->ag_state |= ((ag_cors->ag_dst_h == ag->ag_dst_h) 133 ? AGS_REDUN0 134 : AGS_REDUN1); 135 } 136 137 /* Skip it if this route is itself redundant. 138 * 139 * It is ok to change the contents of the slot here, since it is 140 * always deleted next. 141 */ 142 if (ag->ag_state & AGS_REDUN0) { 143 if (ag->ag_state & AGS_REDUN1) 144 return; /* quit if fully redundant */ 145 /* make it finer if it is half-redundant */ 146 bit = (-ag->ag_mask) >> 1; 147 ag->ag_dst_h |= bit; 148 ag->ag_mask |= bit; 149 150 } else if (ag->ag_state & AGS_REDUN1) { 151 /* make it finer if it is half-redundant */ 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 aggregation. 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 restrictive 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 != NULL && 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 == NULL) { 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 * If the suppressed target was redundant, 239 * then mark the suppressor redundant. 240 */ 241 if (ag_cors->ag_pref <= ag->ag_pref) { 242 if (ag_cors->ag_seqno > ag->ag_seqno) 243 ag_cors->ag_seqno = ag->ag_seqno; 244 if (AG_IS_REDUN(ag->ag_state) 245 && ag_cors->ag_mask==ag->ag_mask<<1) { 246 if (ag_cors->ag_dst_h == dst_h) 247 ag_cors->ag_state |= AGS_REDUN0; 248 else 249 ag_cors->ag_state |= AGS_REDUN1; 250 } 251 if (ag->ag_tag != ag_cors->ag_tag) 252 ag_cors->ag_tag = 0; 253 if (ag->ag_nhop != ag_cors->ag_nhop) 254 ag_cors->ag_nhop = 0; 255 break; 256 } 257 } 258 } 259 260 /* That route has either been output or suppressed */ 261 ag_cors = ag->ag_cors; 262 ag_del(ag); 263 } 264 265 CHECK_AG(); 266 } 267 268 269 /* Try to aggregate a route with previous routes. 270 */ 271 void 272 ag_check(naddr dst, 273 naddr mask, 274 naddr gate, 275 naddr nhop, 276 char metric, 277 char pref, 278 u_int seqnum, 279 u_short tag, 280 u_short state, 281 void (*out)(struct ag_info *)) /* output using this */ 282 { 283 struct ag_info *ag, *nag, *ag_cors; 284 naddr xaddr; 285 int x; 286 287 dst = ntohl(dst); 288 289 /* Punt non-contiguous subnet masks. 290 * 291 * (X & -X) contains a single bit if and only if X is a power of 2. 292 * (X + (X & -X)) == 0 if and only if X is a power of 2. 293 */ 294 if ((mask & -mask) + mask != 0) { 295 struct ag_info nc_ag; 296 297 nc_ag.ag_dst_h = dst; 298 nc_ag.ag_mask = mask; 299 nc_ag.ag_gate = gate; 300 nc_ag.ag_nhop = nhop; 301 nc_ag.ag_metric = metric; 302 nc_ag.ag_pref = pref; 303 nc_ag.ag_tag = tag; 304 nc_ag.ag_state = state; 305 nc_ag.ag_seqno = seqnum; 306 out(&nc_ag); 307 return; 308 } 309 310 /* Search for the right slot in the aggregation table. 311 */ 312 ag_cors = NULL; 313 ag = ag_corsest; 314 while (ag != NULL) { 315 if (ag->ag_mask >= mask) 316 break; 317 318 /* Suppress old routes (i.e. combine with compatible routes 319 * with coarser masks) as we look for the right slot in the 320 * aggregation table for the new route. 321 * A route to an address less than the current destination 322 * will not be affected by the current route or any route 323 * seen hereafter. That means it is safe to suppress it. 324 * This check keeps poor routes (e.g. with large hop counts) 325 * from preventing suppression of finer routes. 326 */ 327 if (ag_cors != NULL 328 && ag->ag_dst_h < dst 329 && (ag->ag_state & AGS_SUPPRESS) 330 && ag_cors->ag_pref <= ag->ag_pref 331 && (ag->ag_dst_h & ag_cors->ag_mask) == ag_cors->ag_dst_h 332 && (ag_cors->ag_gate == ag->ag_gate 333 || (ag->ag_state & AGS_FINE_GATE) 334 || (ag_cors->ag_state & AGS_CORS_GATE))) { 335 if (ag_cors->ag_seqno > ag->ag_seqno) 336 ag_cors->ag_seqno = ag->ag_seqno; 337 /* If the suppressed target was redundant, 338 * then mark the suppressor redundant. 339 */ 340 if (AG_IS_REDUN(ag->ag_state) 341 && ag_cors->ag_mask == ag->ag_mask<<1) { 342 if (ag_cors->ag_dst_h == dst) 343 ag_cors->ag_state |= AGS_REDUN0; 344 else 345 ag_cors->ag_state |= AGS_REDUN1; 346 } 347 if (ag->ag_tag != ag_cors->ag_tag) 348 ag_cors->ag_tag = 0; 349 if (ag->ag_nhop != ag_cors->ag_nhop) 350 ag_cors->ag_nhop = 0; 351 ag_del(ag); 352 CHECK_AG(); 353 } else { 354 ag_cors = ag; 355 } 356 ag = ag_cors->ag_fine; 357 } 358 359 /* If we find the even/odd twin of the new route, and if the 360 * masks and so forth are equal, we can aggregate them. 361 * We can probably promote one of the pair. 362 * 363 * Since the routes are encountered in lexical order, 364 * the new route must be odd. However, the second or later 365 * times around this loop, it could be the even twin promoted 366 * from the even/odd pair of twins of the finer route. 367 */ 368 while (ag != NULL 369 && ag->ag_mask == mask 370 && ((ag->ag_dst_h ^ dst) & (mask<<1)) == 0) { 371 372 /* Here we know the target route and the route in the current 373 * slot have the same netmasks and differ by at most the 374 * last bit. They are either for the same destination, or 375 * for an even/odd pair of destinations. 376 */ 377 if (ag->ag_dst_h == dst) { 378 /* We have two routes to the same destination. 379 * Routes are encountered in lexical order, so a 380 * route is never promoted until the parent route is 381 * already present. So we know that the new route is 382 * a promoted (or aggregated) pair and the route 383 * already in the slot is the explicit route. 384 * 385 * Prefer the best route if their metrics differ, 386 * or the aggregated one if not, following a sort 387 * of longest-match rule. 388 */ 389 if (pref <= ag->ag_pref) { 390 ag->ag_gate = gate; 391 ag->ag_nhop = nhop; 392 ag->ag_tag = tag; 393 ag->ag_metric = metric; 394 ag->ag_pref = pref; 395 x = ag->ag_state; 396 ag->ag_state = state; 397 state = x; 398 } 399 400 /* The sequence number controls flash updating, 401 * and should be the smaller of the two. 402 */ 403 if (ag->ag_seqno > seqnum) 404 ag->ag_seqno = seqnum; 405 406 /* Some bits are set if they are set on either route, 407 * except when the route is for an interface. 408 */ 409 if (!(ag->ag_state & AGS_IF)) 410 ag->ag_state |= (state & (AGS_AGGREGATE_EITHER 411 | AGS_REDUN0 412 | AGS_REDUN1)); 413 return; 414 } 415 416 /* If one of the routes can be promoted and the other can 417 * be suppressed, it may be possible to combine them or 418 * worthwhile to promote one. 419 * 420 * Any route that can be promoted is always 421 * marked to be eligible to be suppressed. 422 */ 423 if (!((state & AGS_AGGREGATE) 424 && (ag->ag_state & AGS_SUPPRESS)) 425 && !((ag->ag_state & AGS_AGGREGATE) 426 && (state & AGS_SUPPRESS))) 427 break; 428 429 /* A pair of even/odd twin routes can be combined 430 * if either is redundant, or if they are via the 431 * same gateway and have the same metric. 432 */ 433 if (AG_IS_REDUN(ag->ag_state) 434 || AG_IS_REDUN(state) 435 || (ag->ag_gate == gate 436 && ag->ag_pref == pref 437 && (state & ag->ag_state & AGS_AGGREGATE) != 0)) { 438 439 /* We have both the even and odd pairs. 440 * Since the routes are encountered in order, 441 * the route in the slot must be the even twin. 442 * 443 * Combine and promote (aggregate) the pair of routes. 444 */ 445 if (seqnum > ag->ag_seqno) 446 seqnum = ag->ag_seqno; 447 if (!AG_IS_REDUN(state)) 448 state &= ~AGS_REDUN1; 449 if (AG_IS_REDUN(ag->ag_state)) 450 state |= AGS_REDUN0; 451 else 452 state &= ~AGS_REDUN0; 453 state |= (ag->ag_state & AGS_AGGREGATE_EITHER); 454 if (ag->ag_tag != tag) 455 tag = 0; 456 if (ag->ag_nhop != nhop) 457 nhop = 0; 458 459 /* Get rid of the even twin that was already 460 * in the slot. 461 */ 462 ag_del(ag); 463 464 } else if (ag->ag_pref >= pref 465 && (ag->ag_state & AGS_AGGREGATE)) { 466 /* If we cannot combine the pair, maybe the route 467 * with the worse metric can be promoted. 468 * 469 * Promote the old, even twin, by giving its slot 470 * in the table to the new, odd twin. 471 */ 472 ag->ag_dst_h = dst; 473 474 xaddr = ag->ag_gate; 475 ag->ag_gate = gate; 476 gate = xaddr; 477 478 xaddr = ag->ag_nhop; 479 ag->ag_nhop = nhop; 480 nhop = xaddr; 481 482 x = ag->ag_tag; 483 ag->ag_tag = tag; 484 tag = x; 485 486 /* The promoted route is even-redundant only if the 487 * even twin was fully redundant. It is not 488 * odd-redundant because the odd-twin will still be 489 * in the table. 490 */ 491 x = ag->ag_state; 492 if (!AG_IS_REDUN(x)) 493 x &= ~AGS_REDUN0; 494 x &= ~AGS_REDUN1; 495 ag->ag_state = state; 496 state = x; 497 498 x = ag->ag_metric; 499 ag->ag_metric = metric; 500 metric = x; 501 502 x = ag->ag_pref; 503 ag->ag_pref = pref; 504 pref = x; 505 506 /* take the newest sequence number */ 507 if (seqnum >= ag->ag_seqno) 508 seqnum = ag->ag_seqno; 509 else 510 ag->ag_seqno = seqnum; 511 512 } else { 513 if (!(state & AGS_AGGREGATE)) 514 break; /* cannot promote either twin */ 515 516 /* Promote the new, odd twin by shaving its 517 * mask and address. 518 * The promoted route is odd-redundant only if the 519 * odd twin was fully redundant. It is not 520 * even-redundant because the even twin is still in 521 * the table. 522 */ 523 if (!AG_IS_REDUN(state)) 524 state &= ~AGS_REDUN1; 525 state &= ~AGS_REDUN0; 526 if (seqnum > ag->ag_seqno) 527 seqnum = ag->ag_seqno; 528 else 529 ag->ag_seqno = seqnum; 530 } 531 532 mask <<= 1; 533 dst &= mask; 534 535 if (ag_cors == NULL) { 536 ag = ag_corsest; 537 break; 538 } 539 ag = ag_cors; 540 ag_cors = ag->ag_cors; 541 } 542 543 /* When we can no longer promote and combine routes, 544 * flush the old route in the target slot. Also flush 545 * any finer routes that we know will never be aggregated by 546 * the new route. 547 * 548 * In case we moved toward coarser masks, 549 * get back where we belong 550 */ 551 if (ag != NULL 552 && ag->ag_mask < mask) { 553 ag_cors = ag; 554 ag = ag->ag_fine; 555 } 556 557 /* Empty the target slot 558 */ 559 if (ag != NULL && ag->ag_mask == mask) { 560 ag_flush(ag->ag_dst_h, ag->ag_mask, out); 561 ag = (ag_cors == NULL) ? ag_corsest : ag_cors->ag_fine; 562 } 563 564 #ifdef DEBUG_AG 565 fflush(stderr); 566 if (ag == NULL && ag_cors != ag_finest) 567 abort(); 568 if (ag_cors == NULL && ag != ag_corsest) 569 abort(); 570 if (ag != NULL && ag->ag_cors != ag_cors) 571 abort(); 572 if (ag_cors != NULL && ag_cors->ag_fine != ag) 573 abort(); 574 CHECK_AG(); 575 #endif 576 577 /* Save the new route on the end of the table. 578 */ 579 nag = ag_avail; 580 ag_avail = nag->ag_fine; 581 582 nag->ag_dst_h = dst; 583 nag->ag_mask = mask; 584 nag->ag_gate = gate; 585 nag->ag_nhop = nhop; 586 nag->ag_metric = metric; 587 nag->ag_pref = pref; 588 nag->ag_tag = tag; 589 nag->ag_state = state; 590 nag->ag_seqno = seqnum; 591 592 nag->ag_fine = ag; 593 if (ag != NULL) 594 ag->ag_cors = nag; 595 else 596 ag_finest = nag; 597 nag->ag_cors = ag_cors; 598 if (ag_cors == NULL) 599 ag_corsest = nag; 600 else 601 ag_cors->ag_fine = nag; 602 CHECK_AG(); 603 } 604 605 606 #define NAME0_LEN 14 607 static const char * 608 rtm_type_name(u_char type) 609 { 610 static const char *rtm_types[] = { 611 "RTM_ADD", 612 "RTM_DELETE", 613 "RTM_CHANGE", 614 "RTM_GET", 615 "RTM_LOSING", 616 "RTM_REDIRECT", 617 "RTM_MISS", 618 "RTM_LOCK", 619 "unused 0x9", 620 "unused 0xa", 621 "RTM_RESOLVE", 622 "RTM_NEWADDR", 623 "RTM_DELADDR", 624 "RTM_IFINFO", 625 "RTM_NEWMADDR", 626 "RTM_DELMADDR" 627 }; 628 #define NEW_RTM_PAT "RTM type %#x" 629 static char name0[sizeof(NEW_RTM_PAT)+2]; 630 631 632 if (type > sizeof(rtm_types)/sizeof(rtm_types[0]) 633 || type == 0) { 634 snprintf(name0, sizeof(name0), NEW_RTM_PAT, type); 635 return name0; 636 } else { 637 return rtm_types[type-1]; 638 } 639 #undef NEW_RTM_PAT 640 } 641 642 643 /* Trim a mask in a sockaddr 644 * Produce a length of 0 for an address of 0. 645 * Otherwise produce the index of the first zero byte. 646 */ 647 static void 648 masktrim(struct sockaddr_in *ap) 649 { 650 char *cp; 651 652 if (ap->sin_addr.s_addr == 0) { 653 ap->sin_len = 0; 654 return; 655 } 656 cp = (char *)(&ap->sin_addr.s_addr+1); 657 while (*--cp == 0) 658 continue; 659 ap->sin_len = cp - (char*)ap + 1; 660 } 661 662 663 /* Tell the kernel to add, delete or change a route 664 */ 665 static void 666 rtioctl(int action, /* RTM_DELETE, etc */ 667 naddr dst, 668 naddr gate, 669 naddr mask, 670 int metric, 671 int flags) 672 { 673 struct { 674 struct rt_msghdr w_rtm; 675 struct sockaddr_in w_dst; 676 struct sockaddr_in w_gate; 677 struct sockaddr_in w_mask; 678 } w; 679 long cc; 680 # define PAT " %-10s %s metric=%d flags=%#x" 681 # define ARGS rtm_type_name(action), rtname(dst,mask,gate), metric, flags 682 683 again: 684 memset(&w, 0, sizeof(w)); 685 w.w_rtm.rtm_msglen = sizeof(w); 686 w.w_rtm.rtm_version = RTM_VERSION; 687 w.w_rtm.rtm_type = action; 688 w.w_rtm.rtm_flags = flags; 689 w.w_rtm.rtm_seq = ++rt_sock_seqno; 690 w.w_rtm.rtm_addrs = RTA_DST|RTA_GATEWAY; 691 if (metric != 0 || action == RTM_CHANGE) { 692 w.w_rtm.rtm_rmx.rmx_hopcount = metric; 693 w.w_rtm.rtm_inits |= RTV_HOPCOUNT; 694 } 695 w.w_dst.sin_family = AF_INET; 696 w.w_dst.sin_addr.s_addr = dst; 697 w.w_gate.sin_family = AF_INET; 698 w.w_gate.sin_addr.s_addr = gate; 699 w.w_dst.sin_len = sizeof(w.w_dst); 700 w.w_gate.sin_len = sizeof(w.w_gate); 701 if (mask == HOST_MASK) { 702 w.w_rtm.rtm_flags |= RTF_HOST; 703 w.w_rtm.rtm_msglen -= sizeof(w.w_mask); 704 } else { 705 w.w_rtm.rtm_addrs |= RTA_NETMASK; 706 w.w_mask.sin_addr.s_addr = htonl(mask); 707 masktrim(&w.w_mask); 708 if (w.w_mask.sin_len == 0) 709 w.w_mask.sin_len = sizeof(long); 710 w.w_rtm.rtm_msglen -= (sizeof(w.w_mask) - w.w_mask.sin_len); 711 } 712 713 #ifndef NO_INSTALL 714 cc = write(rt_sock, &w, w.w_rtm.rtm_msglen); 715 if (cc < 0) { 716 if (errno == ESRCH 717 && (action == RTM_CHANGE || action == RTM_DELETE)) { 718 trace_act("route disappeared before" PAT, ARGS); 719 if (action == RTM_CHANGE) { 720 action = RTM_ADD; 721 goto again; 722 } 723 return; 724 } 725 msglog("write(rt_sock)" PAT ": %s", ARGS, strerror(errno)); 726 return; 727 } else if (cc != w.w_rtm.rtm_msglen) { 728 msglog("write(rt_sock) wrote %ld instead of %d for" PAT, 729 cc, w.w_rtm.rtm_msglen, ARGS); 730 return; 731 } 732 #endif 733 if (TRACEKERNEL) 734 trace_misc("write kernel" PAT, ARGS); 735 #undef PAT 736 #undef ARGS 737 } 738 739 740 #define KHASH_SIZE 71 /* should be prime */ 741 #define KHASH(a,m) khash_bins[((a) ^ (m)) % KHASH_SIZE] 742 static struct khash { 743 struct khash *k_next; 744 naddr k_dst; 745 naddr k_mask; 746 naddr k_gate; 747 short k_metric; 748 u_short k_state; 749 #define KS_NEW 0x001 750 #define KS_DELETE 0x002 /* need to delete the route */ 751 #define KS_ADD 0x004 /* add to the kernel */ 752 #define KS_CHANGE 0x008 /* tell kernel to change the route */ 753 #define KS_DEL_ADD 0x010 /* delete & add to change the kernel */ 754 #define KS_STATIC 0x020 /* Static flag in kernel */ 755 #define KS_GATEWAY 0x040 /* G flag in kernel */ 756 #define KS_DYNAMIC 0x080 /* result of redirect */ 757 #define KS_DELETED 0x100 /* already deleted from kernel */ 758 #define KS_CHECK 0x200 759 time_t k_keep; 760 #define K_KEEP_LIM 30 761 time_t k_redirect_time; /* when redirected route 1st seen */ 762 } *khash_bins[KHASH_SIZE]; 763 764 765 static struct khash* 766 kern_find(naddr dst, naddr mask, struct khash ***ppk) 767 { 768 struct khash *k, **pk; 769 770 for (pk = &KHASH(dst,mask); (k = *pk) != NULL; pk = &k->k_next) { 771 if (k->k_dst == dst && k->k_mask == mask) 772 break; 773 } 774 if (ppk != NULL) 775 *ppk = pk; 776 return k; 777 } 778 779 780 static struct khash* 781 kern_add(naddr dst, naddr mask) 782 { 783 struct khash *k, **pk; 784 785 k = kern_find(dst, mask, &pk); 786 if (k != NULL) 787 return k; 788 789 k = (struct khash *)rtmalloc(sizeof(*k), "kern_add"); 790 791 memset(k, 0, sizeof(*k)); 792 k->k_dst = dst; 793 k->k_mask = mask; 794 k->k_state = KS_NEW; 795 k->k_keep = now.tv_sec; 796 *pk = k; 797 798 return k; 799 } 800 801 802 /* If a kernel route has a non-zero metric, check that it is still in the 803 * daemon table, and not deleted by interfaces coming and going. 804 */ 805 static void 806 kern_check_static(struct khash *k, 807 struct interface *ifp) 808 { 809 struct rt_entry *rt; 810 struct rt_spare new; 811 812 if (k->k_metric == 0) 813 return; 814 815 memset(&new, 0, sizeof(new)); 816 new.rts_ifp = ifp; 817 new.rts_gate = k->k_gate; 818 new.rts_router = (ifp != NULL) ? ifp->int_addr : loopaddr; 819 new.rts_metric = k->k_metric; 820 new.rts_time = now.tv_sec; 821 822 rt = rtget(k->k_dst, k->k_mask); 823 if (rt != NULL) { 824 if (!(rt->rt_state & RS_STATIC)) 825 rtchange(rt, rt->rt_state | RS_STATIC, &new, 0); 826 } else { 827 rtadd(k->k_dst, k->k_mask, RS_STATIC, &new); 828 } 829 } 830 831 832 /* operate on a kernel entry 833 */ 834 static void 835 kern_ioctl(struct khash *k, 836 int action, /* RTM_DELETE, etc */ 837 int flags) 838 839 { 840 switch (action) { 841 case RTM_DELETE: 842 k->k_state &= ~KS_DYNAMIC; 843 if (k->k_state & KS_DELETED) 844 return; 845 k->k_state |= KS_DELETED; 846 break; 847 case RTM_ADD: 848 k->k_state &= ~KS_DELETED; 849 break; 850 case RTM_CHANGE: 851 if (k->k_state & KS_DELETED) { 852 action = RTM_ADD; 853 k->k_state &= ~KS_DELETED; 854 } 855 break; 856 } 857 858 rtioctl(action, k->k_dst, k->k_gate, k->k_mask, k->k_metric, flags); 859 } 860 861 862 /* add a route the kernel told us 863 */ 864 static void 865 rtm_add(struct rt_msghdr *rtm, 866 struct rt_addrinfo *info, 867 time_t keep) 868 { 869 struct khash *k; 870 struct interface *ifp; 871 naddr mask; 872 873 874 if (rtm->rtm_flags & RTF_HOST) { 875 mask = HOST_MASK; 876 } else if (INFO_MASK(info) != 0) { 877 mask = ntohl(S_ADDR(INFO_MASK(info))); 878 } else { 879 msglog("ignore %s without mask", rtm_type_name(rtm->rtm_type)); 880 return; 881 } 882 883 k = kern_add(S_ADDR(INFO_DST(info)), mask); 884 if (k->k_state & KS_NEW) 885 k->k_keep = now.tv_sec+keep; 886 if (INFO_GATE(info) == 0) { 887 trace_act("note %s without gateway", 888 rtm_type_name(rtm->rtm_type)); 889 k->k_metric = HOPCNT_INFINITY; 890 } else if (INFO_GATE(info)->sa_family != AF_INET) { 891 trace_act("note %s with gateway AF=%d", 892 rtm_type_name(rtm->rtm_type), 893 INFO_GATE(info)->sa_family); 894 k->k_metric = HOPCNT_INFINITY; 895 } else { 896 k->k_gate = S_ADDR(INFO_GATE(info)); 897 k->k_metric = rtm->rtm_rmx.rmx_hopcount; 898 if (k->k_metric < 0) 899 k->k_metric = 0; 900 else if (k->k_metric > HOPCNT_INFINITY-1) 901 k->k_metric = HOPCNT_INFINITY-1; 902 } 903 k->k_state &= ~(KS_DELETE | KS_ADD | KS_CHANGE | KS_DEL_ADD 904 | KS_DELETED | KS_GATEWAY | KS_STATIC 905 | KS_NEW | KS_CHECK); 906 if (rtm->rtm_flags & RTF_GATEWAY) 907 k->k_state |= KS_GATEWAY; 908 if (rtm->rtm_flags & RTF_STATIC) 909 k->k_state |= KS_STATIC; 910 911 if (0 != (rtm->rtm_flags & (RTF_DYNAMIC | RTF_MODIFIED))) { 912 if (INFO_AUTHOR(info) != 0 913 && INFO_AUTHOR(info)->sa_family == AF_INET) 914 ifp = iflookup(S_ADDR(INFO_AUTHOR(info))); 915 else 916 ifp = NULL; 917 if (supplier 918 && (ifp == NULL || !(ifp->int_state & IS_REDIRECT_OK))) { 919 /* Routers are not supposed to listen to redirects, 920 * so delete it if it came via an unknown interface 921 * or the interface does not have special permission. 922 */ 923 k->k_state &= ~KS_DYNAMIC; 924 k->k_state |= KS_DELETE; 925 LIM_SEC(need_kern, 0); 926 trace_act("mark for deletion redirected %s --> %s" 927 " via %s", 928 addrname(k->k_dst, k->k_mask, 0), 929 naddr_ntoa(k->k_gate), 930 ifp ? ifp->int_name : "unknown interface"); 931 } else { 932 k->k_state |= KS_DYNAMIC; 933 k->k_redirect_time = now.tv_sec; 934 trace_act("accept redirected %s --> %s via %s", 935 addrname(k->k_dst, k->k_mask, 0), 936 naddr_ntoa(k->k_gate), 937 ifp ? ifp->int_name : "unknown interface"); 938 } 939 return; 940 } 941 942 /* If it is not a static route, quit until the next comparison 943 * between the kernel and daemon tables, when it will be deleted. 944 */ 945 if (!(k->k_state & KS_STATIC)) { 946 k->k_state |= KS_DELETE; 947 LIM_SEC(need_kern, k->k_keep); 948 return; 949 } 950 951 /* Put static routes with real metrics into the daemon table so 952 * they can be advertised. 953 * 954 * Find the interface toward the gateway. 955 */ 956 ifp = iflookup(k->k_gate); 957 if (ifp == NULL) 958 msglog("static route %s --> %s impossibly lacks ifp", 959 addrname(S_ADDR(INFO_DST(info)), mask, 0), 960 naddr_ntoa(k->k_gate)); 961 962 kern_check_static(k, ifp); 963 } 964 965 966 /* deal with packet loss 967 */ 968 static void 969 rtm_lose(struct rt_msghdr *rtm, 970 struct rt_addrinfo *info) 971 { 972 if (INFO_GATE(info) == 0 973 || INFO_GATE(info)->sa_family != AF_INET) { 974 trace_act("ignore %s without gateway", 975 rtm_type_name(rtm->rtm_type)); 976 return; 977 } 978 979 if (rdisc_ok) 980 rdisc_age(S_ADDR(INFO_GATE(info))); 981 age(S_ADDR(INFO_GATE(info))); 982 } 983 984 985 /* Make the gateway slot of an info structure point to something 986 * useful. If it is not already useful, but it specifies an interface, 987 * then fill in the sockaddr_in provided and point it there. 988 */ 989 static int 990 get_info_gate(struct sockaddr **sap, 991 struct sockaddr_in *in) 992 { 993 struct sockaddr_dl *sdl = (struct sockaddr_dl *)*sap; 994 struct interface *ifp; 995 996 if (sdl == NULL) 997 return 0; 998 if ((sdl)->sdl_family == AF_INET) 999 return 1; 1000 if ((sdl)->sdl_family != AF_LINK) 1001 return 0; 1002 1003 ifp = ifwithindex(sdl->sdl_index, 1); 1004 if (ifp == NULL) 1005 return 0; 1006 1007 in->sin_addr.s_addr = ifp->int_addr; 1008 in->sin_len = sizeof(*in); 1009 in->sin_family = AF_INET; 1010 *sap = (struct sockaddr *)in; 1011 1012 return 1; 1013 } 1014 1015 1016 /* Clean the kernel table by copying it to the daemon image. 1017 * Eventually the daemon will delete any extra routes. 1018 */ 1019 void 1020 flush_kern(void) 1021 { 1022 static char *sysctl_buf; 1023 static size_t sysctl_buf_size = 0; 1024 size_t needed; 1025 int mib[6]; 1026 char *next, *lim; 1027 struct rt_msghdr *rtm; 1028 struct sockaddr_in gate_sin; 1029 struct rt_addrinfo info; 1030 int i; 1031 struct khash *k; 1032 1033 1034 for (i = 0; i < KHASH_SIZE; i++) { 1035 for (k = khash_bins[i]; k != NULL; k = k->k_next) { 1036 k->k_state |= KS_CHECK; 1037 } 1038 } 1039 1040 mib[0] = CTL_NET; 1041 mib[1] = PF_ROUTE; 1042 mib[2] = 0; /* protocol */ 1043 mib[3] = 0; /* wildcard address family */ 1044 mib[4] = NET_RT_DUMP; 1045 mib[5] = 0; /* no flags */ 1046 for (;;) { 1047 if ((needed = sysctl_buf_size) != 0) { 1048 if (sysctl(mib, 6, sysctl_buf,&needed, 0, 0) >= 0) 1049 break; 1050 if (errno != ENOMEM && errno != EFAULT) 1051 BADERR(1,"flush_kern: sysctl(RT_DUMP)"); 1052 free(sysctl_buf); 1053 needed = 0; 1054 } 1055 if (sysctl(mib, 6, 0, &needed, 0, 0) < 0) 1056 BADERR(1,"flush_kern: sysctl(RT_DUMP) estimate"); 1057 /* Kludge around the habit of some systems, such as 1058 * BSD/OS 3.1, to not admit how many routes are in the 1059 * kernel, or at least to be quite wrong. 1060 */ 1061 needed += 50*(sizeof(*rtm)+5*sizeof(struct sockaddr)); 1062 sysctl_buf = rtmalloc(sysctl_buf_size = needed, 1063 "flush_kern sysctl(RT_DUMP)"); 1064 } 1065 1066 lim = sysctl_buf + needed; 1067 for (next = sysctl_buf; next < lim; next += rtm->rtm_msglen) { 1068 rtm = (struct rt_msghdr *)next; 1069 if (rtm->rtm_msglen == 0) { 1070 msglog("zero length kernel route at " 1071 " %#lx in buffer %#lx before %#lx", 1072 (u_long)rtm, (u_long)sysctl_buf, (u_long)lim); 1073 break; 1074 } 1075 1076 rt_xaddrs(&info, 1077 (struct sockaddr *)(rtm+1), 1078 (struct sockaddr *)(next + rtm->rtm_msglen), 1079 rtm->rtm_addrs); 1080 1081 if (INFO_DST(&info) == 0 1082 || INFO_DST(&info)->sa_family != AF_INET) 1083 continue; 1084 1085 /* ignore ARP table entries on systems with a merged route 1086 * and ARP table. 1087 */ 1088 if (rtm->rtm_flags & RTF_LLINFO) 1089 continue; 1090 1091 /* ignore multicast addresses 1092 */ 1093 if (IN_MULTICAST(ntohl(S_ADDR(INFO_DST(&info))))) 1094 continue; 1095 1096 if (!get_info_gate(&INFO_GATE(&info), &gate_sin)) 1097 continue; 1098 1099 /* Note static routes and interface routes, and also 1100 * preload the image of the kernel table so that 1101 * we can later clean it, as well as avoid making 1102 * unneeded changes. Keep the old kernel routes for a 1103 * few seconds to allow a RIP or router-discovery 1104 * response to be heard. 1105 */ 1106 rtm_add(rtm,&info,MIN_WAITTIME); 1107 } 1108 1109 for (i = 0; i < KHASH_SIZE; i++) { 1110 for (k = khash_bins[i]; k != NULL; k = k->k_next) { 1111 if (k->k_state & KS_CHECK) { 1112 msglog("%s --> %s disappeared from kernel", 1113 addrname(k->k_dst, k->k_mask, 0), 1114 naddr_ntoa(k->k_gate)); 1115 del_static(k->k_dst, k->k_mask, k->k_gate, 1); 1116 } 1117 } 1118 } 1119 } 1120 1121 1122 /* Listen to announcements from the kernel 1123 */ 1124 void 1125 read_rt(void) 1126 { 1127 long cc; 1128 struct interface *ifp; 1129 struct sockaddr_in gate_sin; 1130 naddr mask, gate; 1131 union { 1132 struct { 1133 struct rt_msghdr rtm; 1134 struct sockaddr addrs[RTAX_MAX]; 1135 } r; 1136 struct if_msghdr ifm; 1137 } m; 1138 char str[100], *strp; 1139 struct rt_addrinfo info; 1140 1141 1142 for (;;) { 1143 cc = read(rt_sock, &m, sizeof(m)); 1144 if (cc <= 0) { 1145 if (cc < 0 && errno != EWOULDBLOCK) 1146 LOGERR("read(rt_sock)"); 1147 return; 1148 } 1149 1150 if (m.r.rtm.rtm_version != RTM_VERSION) { 1151 msglog("bogus routing message version %d", 1152 m.r.rtm.rtm_version); 1153 continue; 1154 } 1155 1156 /* Ignore our own results. 1157 */ 1158 if (m.r.rtm.rtm_type <= RTM_CHANGE 1159 && m.r.rtm.rtm_pid == mypid) { 1160 static int complained = 0; 1161 if (!complained) { 1162 msglog("receiving our own change messages"); 1163 complained = 1; 1164 } 1165 continue; 1166 } 1167 1168 if (m.r.rtm.rtm_type == RTM_IFINFO 1169 || m.r.rtm.rtm_type == RTM_NEWADDR 1170 || m.r.rtm.rtm_type == RTM_DELADDR) { 1171 ifp = ifwithindex(m.ifm.ifm_index, 1172 m.r.rtm.rtm_type != RTM_DELADDR); 1173 if (ifp == NULL) 1174 trace_act("note %s with flags %#x" 1175 " for unknown interface index #%d", 1176 rtm_type_name(m.r.rtm.rtm_type), 1177 m.ifm.ifm_flags, 1178 m.ifm.ifm_index); 1179 else 1180 trace_act("note %s with flags %#x for %s", 1181 rtm_type_name(m.r.rtm.rtm_type), 1182 m.ifm.ifm_flags, 1183 ifp->int_name); 1184 1185 /* After being informed of a change to an interface, 1186 * check them all now if the check would otherwise 1187 * be a long time from now, if the interface is 1188 * not known, or if the interface has been turned 1189 * off or on. 1190 */ 1191 if (ifinit_timer.tv_sec-now.tv_sec>=CHECK_BAD_INTERVAL 1192 || ifp == NULL 1193 || ((ifp->int_if_flags ^ m.ifm.ifm_flags) 1194 & IFF_UP) != 0) 1195 ifinit_timer.tv_sec = now.tv_sec; 1196 continue; 1197 } 1198 1199 strcpy(str, rtm_type_name(m.r.rtm.rtm_type)); 1200 strp = &str[strlen(str)]; 1201 if (m.r.rtm.rtm_type <= RTM_CHANGE) 1202 strp += sprintf(strp," from pid %d",m.r.rtm.rtm_pid); 1203 1204 rt_xaddrs(&info, m.r.addrs, &m.r.addrs[RTAX_MAX], 1205 m.r.rtm.rtm_addrs); 1206 1207 if (INFO_DST(&info) == 0) { 1208 trace_act("ignore %s without dst", str); 1209 continue; 1210 } 1211 1212 if (INFO_DST(&info)->sa_family != AF_INET) { 1213 trace_act("ignore %s for AF %d", str, 1214 INFO_DST(&info)->sa_family); 1215 continue; 1216 } 1217 1218 mask = ((INFO_MASK(&info) != 0) 1219 ? ntohl(S_ADDR(INFO_MASK(&info))) 1220 : (m.r.rtm.rtm_flags & RTF_HOST) 1221 ? HOST_MASK 1222 : std_mask(S_ADDR(INFO_DST(&info)))); 1223 1224 strp += sprintf(strp, ": %s", 1225 addrname(S_ADDR(INFO_DST(&info)), mask, 0)); 1226 1227 if (IN_MULTICAST(ntohl(S_ADDR(INFO_DST(&info))))) { 1228 trace_act("ignore multicast %s", str); 1229 continue; 1230 } 1231 1232 if (m.r.rtm.rtm_flags & RTF_LLINFO) { 1233 trace_act("ignore ARP %s", str); 1234 continue; 1235 } 1236 1237 if (get_info_gate(&INFO_GATE(&info), &gate_sin)) { 1238 gate = S_ADDR(INFO_GATE(&info)); 1239 strp += sprintf(strp, " --> %s", naddr_ntoa(gate)); 1240 } else { 1241 gate = 0; 1242 } 1243 1244 if (INFO_AUTHOR(&info) != 0) 1245 strp += sprintf(strp, " by authority of %s", 1246 saddr_ntoa(INFO_AUTHOR(&info))); 1247 1248 switch (m.r.rtm.rtm_type) { 1249 case RTM_ADD: 1250 case RTM_CHANGE: 1251 case RTM_REDIRECT: 1252 if (m.r.rtm.rtm_errno != 0) { 1253 trace_act("ignore %s with \"%s\" error", 1254 str, strerror(m.r.rtm.rtm_errno)); 1255 } else { 1256 trace_act("%s", str); 1257 rtm_add(&m.r.rtm,&info,0); 1258 } 1259 break; 1260 1261 case RTM_DELETE: 1262 if (m.r.rtm.rtm_errno != 0 1263 && m.r.rtm.rtm_errno != ESRCH) { 1264 trace_act("ignore %s with \"%s\" error", 1265 str, strerror(m.r.rtm.rtm_errno)); 1266 } else { 1267 trace_act("%s", str); 1268 del_static(S_ADDR(INFO_DST(&info)), mask, 1269 gate, 1); 1270 } 1271 break; 1272 1273 case RTM_LOSING: 1274 trace_act("%s", str); 1275 rtm_lose(&m.r.rtm,&info); 1276 break; 1277 1278 default: 1279 trace_act("ignore %s", str); 1280 break; 1281 } 1282 } 1283 } 1284 1285 1286 /* after aggregating, note routes that belong in the kernel 1287 */ 1288 static void 1289 kern_out(struct ag_info *ag) 1290 { 1291 struct khash *k; 1292 1293 1294 /* Do not install bad routes if they are not already present. 1295 * This includes routes that had RS_NET_SYN for interfaces that 1296 * recently died. 1297 */ 1298 if (ag->ag_metric == HOPCNT_INFINITY) { 1299 k = kern_find(htonl(ag->ag_dst_h), ag->ag_mask, 0); 1300 if (k == NULL) 1301 return; 1302 } else { 1303 k = kern_add(htonl(ag->ag_dst_h), ag->ag_mask); 1304 } 1305 1306 if (k->k_state & KS_NEW) { 1307 /* will need to add new entry to the kernel table */ 1308 k->k_state = KS_ADD; 1309 if (ag->ag_state & AGS_GATEWAY) 1310 k->k_state |= KS_GATEWAY; 1311 k->k_gate = ag->ag_gate; 1312 k->k_metric = ag->ag_metric; 1313 return; 1314 } 1315 1316 if (k->k_state & KS_STATIC) 1317 return; 1318 1319 /* modify existing kernel entry if necessary */ 1320 if (k->k_gate != ag->ag_gate 1321 || k->k_metric != ag->ag_metric) { 1322 /* Must delete bad interface routes etc. to change them. */ 1323 if (k->k_metric == HOPCNT_INFINITY) 1324 k->k_state |= KS_DEL_ADD; 1325 k->k_gate = ag->ag_gate; 1326 k->k_metric = ag->ag_metric; 1327 k->k_state |= KS_CHANGE; 1328 } 1329 1330 /* If the daemon thinks the route should exist, forget 1331 * about any redirections. 1332 * If the daemon thinks the route should exist, eventually 1333 * override manual intervention by the operator. 1334 */ 1335 if ((k->k_state & (KS_DYNAMIC | KS_DELETED)) != 0) { 1336 k->k_state &= ~KS_DYNAMIC; 1337 k->k_state |= (KS_ADD | KS_DEL_ADD); 1338 } 1339 1340 if ((k->k_state & KS_GATEWAY) 1341 && !(ag->ag_state & AGS_GATEWAY)) { 1342 k->k_state &= ~KS_GATEWAY; 1343 k->k_state |= (KS_ADD | KS_DEL_ADD); 1344 } else if (!(k->k_state & KS_GATEWAY) 1345 && (ag->ag_state & AGS_GATEWAY)) { 1346 k->k_state |= KS_GATEWAY; 1347 k->k_state |= (KS_ADD | KS_DEL_ADD); 1348 } 1349 1350 /* Deleting-and-adding is necessary to change aspects of a route. 1351 * Just delete instead of deleting and then adding a bad route. 1352 * Otherwise, we want to keep the route in the kernel. 1353 */ 1354 if (k->k_metric == HOPCNT_INFINITY 1355 && (k->k_state & KS_DEL_ADD)) 1356 k->k_state |= KS_DELETE; 1357 else 1358 k->k_state &= ~KS_DELETE; 1359 #undef RT 1360 } 1361 1362 1363 static int 1364 walk_kern(struct radix_node *rn, void *argp __unused) 1365 { 1366 #define RT ((struct rt_entry *)rn) 1367 char metric, pref; 1368 u_int ags = 0; 1369 1370 1371 /* Do not install synthetic routes */ 1372 if (RT->rt_state & RS_NET_SYN) 1373 return 0; 1374 1375 if (!(RT->rt_state & RS_IF)) { 1376 /* This is an ordinary route, not for an interface. 1377 */ 1378 1379 /* aggregate, ordinary good routes without regard to 1380 * their metric 1381 */ 1382 pref = 1; 1383 ags |= (AGS_GATEWAY | AGS_SUPPRESS | AGS_AGGREGATE); 1384 1385 /* Do not install host routes directly to hosts, to avoid 1386 * interfering with ARP entries in the kernel table. 1387 */ 1388 if (RT_ISHOST(RT) 1389 && ntohl(RT->rt_dst) == RT->rt_gate) 1390 return 0; 1391 1392 } else { 1393 /* This is an interface route. 1394 * Do not install routes for "external" remote interfaces. 1395 */ 1396 if (RT->rt_ifp != 0 && (RT->rt_ifp->int_state & IS_EXTERNAL)) 1397 return 0; 1398 1399 /* Interfaces should override received routes. 1400 */ 1401 pref = 0; 1402 ags |= (AGS_IF | AGS_CORS_GATE); 1403 1404 /* If it is not an interface, or an alias for an interface, 1405 * it must be a "gateway." 1406 * 1407 * If it is a "remote" interface, it is also a "gateway" to 1408 * the kernel if is not a alias. 1409 */ 1410 if (RT->rt_ifp == 0 1411 || (RT->rt_ifp->int_state & IS_REMOTE)) 1412 ags |= (AGS_GATEWAY | AGS_SUPPRESS | AGS_AGGREGATE); 1413 } 1414 1415 /* If RIP is off and IRDP is on, let the route to the discovered 1416 * route suppress any RIP routes. Eventually the RIP routes 1417 * will time-out and be deleted. This reaches the steady-state 1418 * quicker. 1419 */ 1420 if ((RT->rt_state & RS_RDISC) && rip_sock < 0) 1421 ags |= AGS_CORS_GATE; 1422 1423 metric = RT->rt_metric; 1424 if (metric == HOPCNT_INFINITY) { 1425 /* if the route is dead, so try hard to aggregate. */ 1426 pref = HOPCNT_INFINITY; 1427 ags |= (AGS_FINE_GATE | AGS_SUPPRESS); 1428 ags &= ~(AGS_IF | AGS_CORS_GATE); 1429 } 1430 1431 ag_check(RT->rt_dst, RT->rt_mask, RT->rt_gate, 0, 1432 metric,pref, 0, 0, ags, kern_out); 1433 return 0; 1434 #undef RT 1435 } 1436 1437 1438 /* Update the kernel table to match the daemon table. 1439 */ 1440 static void 1441 fix_kern(void) 1442 { 1443 int i; 1444 struct khash *k, **pk; 1445 1446 1447 need_kern = age_timer; 1448 1449 /* Walk daemon table, updating the copy of the kernel table. 1450 */ 1451 rhead->rnh_walktree(rhead, walk_kern, NULL); 1452 ag_flush(0, 0, kern_out); 1453 1454 for (i = 0; i < KHASH_SIZE; i++) { 1455 for (pk = &khash_bins[i]; (k = *pk) != NULL; ) { 1456 /* Do not touch static routes */ 1457 if (k->k_state & KS_STATIC) { 1458 kern_check_static(k,0); 1459 pk = &k->k_next; 1460 continue; 1461 } 1462 1463 /* check hold on routes deleted by the operator */ 1464 if (k->k_keep > now.tv_sec) { 1465 /* ensure we check when the hold is over */ 1466 LIM_SEC(need_kern, k->k_keep); 1467 /* mark for the next cycle */ 1468 k->k_state |= KS_DELETE; 1469 pk = &k->k_next; 1470 continue; 1471 } 1472 1473 if ((k->k_state & KS_DELETE) 1474 && !(k->k_state & KS_DYNAMIC)) { 1475 kern_ioctl(k, RTM_DELETE, 0); 1476 *pk = k->k_next; 1477 free(k); 1478 continue; 1479 } 1480 1481 if (k->k_state & KS_DEL_ADD) 1482 kern_ioctl(k, RTM_DELETE, 0); 1483 1484 if (k->k_state & KS_ADD) { 1485 kern_ioctl(k, RTM_ADD, 1486 ((0 != (k->k_state & (KS_GATEWAY 1487 | KS_DYNAMIC))) 1488 ? RTF_GATEWAY : 0)); 1489 } else if (k->k_state & KS_CHANGE) { 1490 kern_ioctl(k, RTM_CHANGE, 1491 ((0 != (k->k_state & (KS_GATEWAY 1492 | KS_DYNAMIC))) 1493 ? RTF_GATEWAY : 0)); 1494 } 1495 k->k_state &= ~(KS_ADD|KS_CHANGE|KS_DEL_ADD); 1496 1497 /* Mark this route to be deleted in the next cycle. 1498 * This deletes routes that disappear from the 1499 * daemon table, since the normal aging code 1500 * will clear the bit for routes that have not 1501 * disappeared from the daemon table. 1502 */ 1503 k->k_state |= KS_DELETE; 1504 pk = &k->k_next; 1505 } 1506 } 1507 } 1508 1509 1510 /* Delete a static route in the image of the kernel table. 1511 */ 1512 void 1513 del_static(naddr dst, 1514 naddr mask, 1515 naddr gate, 1516 int gone) 1517 { 1518 struct khash *k; 1519 struct rt_entry *rt; 1520 1521 /* Just mark it in the table to be deleted next time the kernel 1522 * table is updated. 1523 * If it has already been deleted, mark it as such, and set its 1524 * keep-timer so that it will not be deleted again for a while. 1525 * This lets the operator delete a route added by the daemon 1526 * and add a replacement. 1527 */ 1528 k = kern_find(dst, mask, 0); 1529 if (k != NULL && (gate == 0 || k->k_gate == gate)) { 1530 k->k_state &= ~(KS_STATIC | KS_DYNAMIC | KS_CHECK); 1531 k->k_state |= KS_DELETE; 1532 if (gone) { 1533 k->k_state |= KS_DELETED; 1534 k->k_keep = now.tv_sec + K_KEEP_LIM; 1535 } 1536 } 1537 1538 rt = rtget(dst, mask); 1539 if (rt != NULL && (rt->rt_state & RS_STATIC)) 1540 rtbad(rt); 1541 } 1542 1543 1544 /* Delete all routes generated from ICMP Redirects that use a given gateway, 1545 * as well as old redirected routes. 1546 */ 1547 void 1548 del_redirects(naddr bad_gate, 1549 time_t old) 1550 { 1551 int i; 1552 struct khash *k; 1553 1554 1555 for (i = 0; i < KHASH_SIZE; i++) { 1556 for (k = khash_bins[i]; k != NULL; k = k->k_next) { 1557 if (!(k->k_state & KS_DYNAMIC) 1558 || (k->k_state & KS_STATIC)) 1559 continue; 1560 1561 if (k->k_gate != bad_gate 1562 && k->k_redirect_time > old 1563 && !supplier) 1564 continue; 1565 1566 k->k_state |= KS_DELETE; 1567 k->k_state &= ~KS_DYNAMIC; 1568 need_kern.tv_sec = now.tv_sec; 1569 trace_act("mark redirected %s --> %s for deletion", 1570 addrname(k->k_dst, k->k_mask, 0), 1571 naddr_ntoa(k->k_gate)); 1572 } 1573 } 1574 } 1575 1576 1577 /* Start the daemon tables. 1578 */ 1579 1580 void 1581 rtinit(void) 1582 { 1583 int i; 1584 struct ag_info *ag; 1585 1586 /* Initialize the radix trees */ 1587 rn_init(); 1588 rn_inithead(&rhead, rn_cpumaskhead(0), 1589 offsetof(struct sockaddr_in, sin_addr)); 1590 1591 /* mark all of the slots in the table free */ 1592 ag_avail = ag_slots; 1593 for (ag = ag_slots, i = 1; i < NUM_AG_SLOTS; i++) { 1594 ag->ag_fine = ag+1; 1595 ag++; 1596 } 1597 } 1598 1599 1600 static struct sockaddr_in dst_sock = {sizeof(dst_sock), AF_INET, 0, {0}, {0}}; 1601 static struct sockaddr_in mask_sock = {sizeof(mask_sock), AF_INET, 0, {0}, {0}}; 1602 1603 1604 static void 1605 set_need_flash(void) 1606 { 1607 if (!need_flash) { 1608 need_flash = 1; 1609 /* Do not send the flash update immediately. Wait a little 1610 * while to hear from other routers. 1611 */ 1612 no_flash.tv_sec = now.tv_sec + MIN_WAITTIME; 1613 } 1614 } 1615 1616 1617 /* Get a particular routing table entry 1618 */ 1619 struct rt_entry * 1620 rtget(naddr dst, naddr mask) 1621 { 1622 struct rt_entry *rt; 1623 1624 dst_sock.sin_addr.s_addr = dst; 1625 mask_sock.sin_addr.s_addr = htonl(mask); 1626 masktrim(&mask_sock); 1627 rt = (struct rt_entry *)rhead->rnh_lookup(&dst_sock, &mask_sock, rhead); 1628 if (!rt || rt->rt_dst != dst || rt->rt_mask != mask) 1629 return 0; 1630 1631 return rt; 1632 } 1633 1634 1635 /* Find a route to dst as the kernel would. 1636 */ 1637 struct rt_entry * 1638 rtfind(naddr dst) 1639 { 1640 dst_sock.sin_addr.s_addr = dst; 1641 return (struct rt_entry *)rhead->rnh_matchaddr(&dst_sock, rhead); 1642 } 1643 1644 1645 /* add a route to the table 1646 */ 1647 void 1648 rtadd(naddr dst, 1649 naddr mask, 1650 u_int state, /* rt_state for the entry */ 1651 struct rt_spare *new) 1652 { 1653 struct rt_entry *rt; 1654 naddr smask; 1655 int i; 1656 struct rt_spare *rts; 1657 1658 rt = (struct rt_entry *)rtmalloc(sizeof (*rt), "rtadd"); 1659 memset(rt, 0, sizeof(*rt)); 1660 for (rts = rt->rt_spares, i = NUM_SPARES; i != 0; i--, rts++) 1661 rts->rts_metric = HOPCNT_INFINITY; 1662 1663 rt->rt_nodes->rn_key = (caddr_t)&rt->rt_dst_sock; 1664 rt->rt_dst = dst; 1665 rt->rt_dst_sock.sin_family = AF_INET; 1666 rt->rt_dst_sock.sin_len = dst_sock.sin_len; 1667 if (mask != HOST_MASK) { 1668 smask = std_mask(dst); 1669 if ((smask & ~mask) == 0 && mask > smask) 1670 state |= RS_SUBNET; 1671 } 1672 mask_sock.sin_addr.s_addr = htonl(mask); 1673 masktrim(&mask_sock); 1674 rt->rt_mask = mask; 1675 rt->rt_state = state; 1676 rt->rt_spares[0] = *new; 1677 rt->rt_time = now.tv_sec; 1678 rt->rt_poison_metric = HOPCNT_INFINITY; 1679 rt->rt_seqno = update_seqno; 1680 1681 if (++total_routes == MAX_ROUTES) 1682 msglog("have maximum (%d) routes", total_routes); 1683 if (TRACEACTIONS) 1684 trace_add_del("Add", rt); 1685 1686 need_kern.tv_sec = now.tv_sec; 1687 set_need_flash(); 1688 1689 if (rhead->rnh_addaddr(&rt->rt_dst_sock, &mask_sock, rhead, 1690 rt->rt_nodes) == NULL) { 1691 msglog("rnh_addaddr() failed for %s mask=%#lx", 1692 naddr_ntoa(dst), (u_long)mask); 1693 free(rt); 1694 } 1695 } 1696 1697 1698 /* notice a changed route 1699 */ 1700 void 1701 rtchange(struct rt_entry *rt, 1702 u_int state, /* new state bits */ 1703 struct rt_spare *new, 1704 char *label) 1705 { 1706 if (rt->rt_metric != new->rts_metric) { 1707 /* Fix the kernel immediately if it seems the route 1708 * has gone bad, since there may be a working route that 1709 * aggregates this route. 1710 */ 1711 if (new->rts_metric == HOPCNT_INFINITY) { 1712 need_kern.tv_sec = now.tv_sec; 1713 if (new->rts_time >= now.tv_sec - EXPIRE_TIME) 1714 new->rts_time = now.tv_sec - EXPIRE_TIME; 1715 } 1716 rt->rt_seqno = update_seqno; 1717 set_need_flash(); 1718 } 1719 1720 if (rt->rt_gate != new->rts_gate) { 1721 need_kern.tv_sec = now.tv_sec; 1722 rt->rt_seqno = update_seqno; 1723 set_need_flash(); 1724 } 1725 1726 state |= (rt->rt_state & RS_SUBNET); 1727 1728 /* Keep various things from deciding ageless routes are stale. 1729 */ 1730 if (!AGE_RT(state, new->rts_ifp)) 1731 new->rts_time = now.tv_sec; 1732 1733 if (TRACEACTIONS) 1734 trace_change(rt, state, new, 1735 label ? label : "Chg "); 1736 1737 rt->rt_state = state; 1738 rt->rt_spares[0] = *new; 1739 } 1740 1741 1742 /* check for a better route among the spares 1743 */ 1744 static struct rt_spare * 1745 rts_better(struct rt_entry *rt) 1746 { 1747 struct rt_spare *rts, *rts1; 1748 int i; 1749 1750 /* find the best alternative among the spares */ 1751 rts = rt->rt_spares+1; 1752 for (i = NUM_SPARES, rts1 = rts+1; i > 2; i--, rts1++) { 1753 if (BETTER_LINK(rt,rts1,rts)) 1754 rts = rts1; 1755 } 1756 1757 return rts; 1758 } 1759 1760 1761 /* switch to a backup route 1762 */ 1763 void 1764 rtswitch(struct rt_entry *rt, 1765 struct rt_spare *rts) 1766 { 1767 struct rt_spare swap; 1768 char label[10]; 1769 1770 1771 /* Do not change permanent routes */ 1772 if (0 != (rt->rt_state & (RS_MHOME | RS_STATIC | RS_RDISC 1773 | RS_NET_SYN | RS_IF))) 1774 return; 1775 1776 /* find the best alternative among the spares */ 1777 if (rts == NULL) 1778 rts = rts_better(rt); 1779 1780 /* Do not bother if it is not worthwhile. 1781 */ 1782 if (!BETTER_LINK(rt, rts, rt->rt_spares)) 1783 return; 1784 1785 swap = rt->rt_spares[0]; 1786 sprintf(label, "Use #%d", (int)(rts - rt->rt_spares)); 1787 rtchange(rt, rt->rt_state & ~(RS_NET_SYN | RS_RDISC), rts, label); 1788 if (swap.rts_metric == HOPCNT_INFINITY) { 1789 *rts = rts_empty; 1790 } else { 1791 *rts = swap; 1792 } 1793 } 1794 1795 1796 void 1797 rtdelete(struct rt_entry *rt) 1798 { 1799 struct khash *k; 1800 1801 1802 if (TRACEACTIONS) 1803 trace_add_del("Del", rt); 1804 1805 k = kern_find(rt->rt_dst, rt->rt_mask, 0); 1806 if (k != NULL) { 1807 k->k_state |= KS_DELETE; 1808 need_kern.tv_sec = now.tv_sec; 1809 } 1810 1811 dst_sock.sin_addr.s_addr = rt->rt_dst; 1812 mask_sock.sin_addr.s_addr = htonl(rt->rt_mask); 1813 masktrim(&mask_sock); 1814 if (rt != (struct rt_entry *)rhead->rnh_deladdr( 1815 &dst_sock, &mask_sock, rhead)) { 1816 msglog("rnh_deladdr() failed"); 1817 } else { 1818 free(rt); 1819 total_routes--; 1820 } 1821 } 1822 1823 1824 void 1825 rts_delete(struct rt_entry *rt, 1826 struct rt_spare *rts) 1827 { 1828 trace_upslot(rt, rts, &rts_empty); 1829 *rts = rts_empty; 1830 } 1831 1832 1833 /* Get rid of a bad route, and try to switch to a replacement. 1834 */ 1835 void 1836 rtbad(struct rt_entry *rt) 1837 { 1838 struct rt_spare new; 1839 1840 /* Poison the route */ 1841 new = rt->rt_spares[0]; 1842 new.rts_metric = HOPCNT_INFINITY; 1843 rtchange(rt, rt->rt_state & ~(RS_IF | RS_LOCAL | RS_STATIC), &new, 0); 1844 rtswitch(rt, 0); 1845 } 1846 1847 1848 /* Junk a RS_NET_SYN or RS_LOCAL route, 1849 * unless it is needed by another interface. 1850 */ 1851 void 1852 rtbad_sub(struct rt_entry *rt) 1853 { 1854 struct interface *ifp, *ifp1; 1855 struct intnet *intnetp; 1856 u_int state; 1857 1858 1859 ifp1 = NULL; 1860 state = 0; 1861 1862 if (rt->rt_state & RS_LOCAL) { 1863 /* Is this the route through loopback for the interface? 1864 * If so, see if it is used by any other interfaces, such 1865 * as a point-to-point interface with the same local address. 1866 */ 1867 for (ifp = ifnet; ifp != NULL; ifp = ifp->int_next) { 1868 /* Retain it if another interface needs it. 1869 */ 1870 if (ifp->int_addr == rt->rt_ifp->int_addr) { 1871 state |= RS_LOCAL; 1872 ifp1 = ifp; 1873 break; 1874 } 1875 } 1876 } 1877 1878 if (!(state & RS_LOCAL)) { 1879 /* Retain RIPv1 logical network route if there is another 1880 * interface that justifies it. 1881 */ 1882 if (rt->rt_state & RS_NET_SYN) { 1883 for (ifp = ifnet; ifp != NULL; ifp = ifp->int_next) { 1884 if ((ifp->int_state & IS_NEED_NET_SYN) 1885 && rt->rt_mask == ifp->int_std_mask 1886 && rt->rt_dst == ifp->int_std_addr) { 1887 state |= RS_NET_SYN; 1888 ifp1 = ifp; 1889 break; 1890 } 1891 } 1892 } 1893 1894 /* or if there is an authority route that needs it. */ 1895 for (intnetp = intnets; 1896 intnetp != NULL; 1897 intnetp = intnetp->intnet_next) { 1898 if (intnetp->intnet_addr == rt->rt_dst 1899 && intnetp->intnet_mask == rt->rt_mask) { 1900 state |= (RS_NET_SYN | RS_NET_INT); 1901 break; 1902 } 1903 } 1904 } 1905 1906 if (ifp1 != NULL || (state & RS_NET_SYN)) { 1907 struct rt_spare new = rt->rt_spares[0]; 1908 new.rts_ifp = ifp1; 1909 rtchange(rt, ((rt->rt_state & ~(RS_NET_SYN|RS_LOCAL)) | state), 1910 &new, 0); 1911 } else { 1912 rtbad(rt); 1913 } 1914 } 1915 1916 1917 /* Called while walking the table looking for sick interfaces 1918 * or after a time change. 1919 */ 1920 int 1921 walk_bad(struct radix_node *rn, void *argp __unused) 1922 { 1923 #define RT ((struct rt_entry *)rn) 1924 struct rt_spare *rts; 1925 int i; 1926 1927 1928 /* fix any spare routes through the interface 1929 */ 1930 rts = RT->rt_spares; 1931 for (i = NUM_SPARES; i != 1; i--) { 1932 rts++; 1933 if (rts->rts_metric < HOPCNT_INFINITY 1934 && (rts->rts_ifp == 0 1935 || (rts->rts_ifp->int_state & IS_BROKE))) 1936 rts_delete(RT, rts); 1937 } 1938 1939 /* Deal with the main route 1940 */ 1941 /* finished if it has been handled before or if its interface is ok 1942 */ 1943 if (RT->rt_ifp == 0 || !(RT->rt_ifp->int_state & IS_BROKE)) 1944 return 0; 1945 1946 /* Bad routes for other than interfaces are easy. 1947 */ 1948 if (0 == (RT->rt_state & (RS_IF | RS_NET_SYN | RS_LOCAL))) { 1949 rtbad(RT); 1950 return 0; 1951 } 1952 1953 rtbad_sub(RT); 1954 return 0; 1955 #undef RT 1956 } 1957 1958 1959 /* Check the age of an individual route. 1960 */ 1961 static int 1962 walk_age(struct radix_node *rn, void *argp __unused) 1963 { 1964 #define RT ((struct rt_entry *)rn) 1965 struct interface *ifp; 1966 struct rt_spare *rts; 1967 int i; 1968 1969 1970 /* age all of the spare routes, including the primary route 1971 * currently in use 1972 */ 1973 rts = RT->rt_spares; 1974 for (i = NUM_SPARES; i != 0; i--, rts++) { 1975 1976 ifp = rts->rts_ifp; 1977 if (i == NUM_SPARES) { 1978 if (!AGE_RT(RT->rt_state, ifp)) { 1979 /* Keep various things from deciding ageless 1980 * routes are stale 1981 */ 1982 rts->rts_time = now.tv_sec; 1983 continue; 1984 } 1985 1986 /* forget RIP routes after RIP has been turned off. 1987 */ 1988 if (rip_sock < 0) { 1989 rtdelete(RT); 1990 return 0; 1991 } 1992 } 1993 1994 /* age failing routes 1995 */ 1996 if (age_bad_gate == rts->rts_gate 1997 && rts->rts_time >= now_stale) { 1998 rts->rts_time -= SUPPLY_INTERVAL; 1999 } 2000 2001 /* trash the spare routes when they go bad */ 2002 if (rts->rts_metric < HOPCNT_INFINITY 2003 && now_garbage > rts->rts_time 2004 && i != NUM_SPARES) 2005 rts_delete(RT, rts); 2006 } 2007 2008 2009 /* finished if the active route is still fresh */ 2010 if (now_stale <= RT->rt_time) 2011 return 0; 2012 2013 /* try to switch to an alternative */ 2014 rtswitch(RT, 0); 2015 2016 /* Delete a dead route after it has been publically mourned. */ 2017 if (now_garbage > RT->rt_time) { 2018 rtdelete(RT); 2019 return 0; 2020 } 2021 2022 /* Start poisoning a bad route before deleting it. */ 2023 if (now.tv_sec - RT->rt_time > EXPIRE_TIME) { 2024 struct rt_spare new = RT->rt_spares[0]; 2025 new.rts_metric = HOPCNT_INFINITY; 2026 rtchange(RT, RT->rt_state, &new, 0); 2027 } 2028 return 0; 2029 } 2030 2031 2032 /* Watch for dead routes and interfaces. 2033 */ 2034 void 2035 age(naddr bad_gate) 2036 { 2037 struct interface *ifp; 2038 int need_query = 0; 2039 2040 /* If not listening to RIP, there is no need to age the routes in 2041 * the table. 2042 */ 2043 age_timer.tv_sec = (now.tv_sec 2044 + ((rip_sock < 0) ? NEVER : SUPPLY_INTERVAL)); 2045 2046 /* Check for dead IS_REMOTE interfaces by timing their 2047 * transmissions. 2048 */ 2049 for (ifp = ifnet; ifp; ifp = ifp->int_next) { 2050 if (!(ifp->int_state & IS_REMOTE)) 2051 continue; 2052 2053 /* ignore unreachable remote interfaces */ 2054 if (!check_remote(ifp)) 2055 continue; 2056 2057 /* Restore remote interface that has become reachable 2058 */ 2059 if (ifp->int_state & IS_BROKE) 2060 if_ok(ifp, "remote "); 2061 2062 if (ifp->int_act_time != NEVER 2063 && now.tv_sec - ifp->int_act_time > EXPIRE_TIME) { 2064 msglog("remote interface %s to %s timed out after" 2065 " %ld:%ld", 2066 ifp->int_name, 2067 naddr_ntoa(ifp->int_dstaddr), 2068 (now.tv_sec - ifp->int_act_time)/60, 2069 (now.tv_sec - ifp->int_act_time)%60); 2070 if_sick(ifp); 2071 } 2072 2073 /* If we have not heard from the other router 2074 * recently, ask it. 2075 */ 2076 if (now.tv_sec >= ifp->int_query_time) { 2077 ifp->int_query_time = NEVER; 2078 need_query = 1; 2079 } 2080 } 2081 2082 /* Age routes. */ 2083 age_bad_gate = bad_gate; 2084 rhead->rnh_walktree(rhead, walk_age, NULL); 2085 2086 /* delete old redirected routes to keep the kernel table small 2087 * and prevent blackholes 2088 */ 2089 del_redirects(bad_gate, now.tv_sec-STALE_TIME); 2090 2091 /* Update the kernel routing table. */ 2092 fix_kern(); 2093 2094 /* poke reticent remote gateways */ 2095 if (need_query) 2096 rip_query(); 2097 } 2098