1 /* $FreeBSD: src/sys/netinet6/in6_rmx.c,v 1.1.2.3 2002/04/28 05:40:27 suz Exp $ */ 2 /* $DragonFly: src/sys/netinet6/in6_rmx.c,v 1.8 2004/12/21 02:54:47 hsu Exp $ */ 3 /* $KAME: in6_rmx.c,v 1.11 2001/07/26 06:53:16 jinmei Exp $ */ 4 5 /* 6 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 7 * All rights reserved. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 3. Neither the name of the project 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 PROJECT 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 PROJECT OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 */ 33 34 /* 35 * Copyright 1994, 1995 Massachusetts Institute of Technology 36 * 37 * Permission to use, copy, modify, and distribute this software and 38 * its documentation for any purpose and without fee is hereby 39 * granted, provided that both the above copyright notice and this 40 * permission notice appear in all copies, that both the above 41 * copyright notice and this permission notice appear in all 42 * supporting documentation, and that the name of M.I.T. not be used 43 * in advertising or publicity pertaining to distribution of the 44 * software without specific, written prior permission. M.I.T. makes 45 * no representations about the suitability of this software for any 46 * purpose. It is provided "as is" without express or implied 47 * warranty. 48 * 49 * THIS SOFTWARE IS PROVIDED BY M.I.T. ``AS IS''. M.I.T. DISCLAIMS 50 * ALL EXPRESS OR IMPLIED WARRANTIES WITH REGARD TO THIS SOFTWARE, 51 * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF 52 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. IN NO EVENT 53 * SHALL M.I.T. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 54 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 55 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF 56 * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND 57 * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 58 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 59 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 60 * SUCH DAMAGE. 61 * 62 */ 63 64 /* 65 * This code does two things necessary for the enhanced TCP metrics to 66 * function in a useful manner: 67 * 1) It marks all non-host routes as `cloning', thus ensuring that 68 * every actual reference to such a route actually gets turned 69 * into a reference to a host route to the specific destination 70 * requested. 71 * 2) When such routes lose all their references, it arranges for them 72 * to be deleted in some random collection of circumstances, so that 73 * a large quantity of stale routing data is not kept in kernel memory 74 * indefinitely. See in6_rtqtimo() below for the exact mechanism. 75 */ 76 77 #include <sys/param.h> 78 #include <sys/systm.h> 79 #include <sys/kernel.h> 80 #include <sys/sysctl.h> 81 #include <sys/queue.h> 82 #include <sys/socket.h> 83 #include <sys/socketvar.h> 84 #include <sys/mbuf.h> 85 #include <sys/syslog.h> 86 87 #include <net/if.h> 88 #include <net/route.h> 89 #include <netinet/in.h> 90 #include <netinet/ip_var.h> 91 #include <netinet/in_var.h> 92 93 #include <netinet/ip6.h> 94 #include <netinet6/ip6_var.h> 95 96 #include <netinet/icmp6.h> 97 98 #include <netinet/tcp.h> 99 #include <netinet/tcp_seq.h> 100 #include <netinet/tcp_timer.h> 101 #include <netinet/tcp_var.h> 102 103 static struct callout in6_rtqtimo_ch; 104 static struct callout in6_mtutimo_ch; 105 106 extern int in6_inithead (void **head, int off); 107 108 #define RTPRF_OURS RTF_PROTO3 /* set on routes we manage */ 109 110 /* 111 * Do what we need to do when inserting a route. 112 */ 113 static struct radix_node * 114 in6_addroute(char *key, char *mask, struct radix_node_head *head, 115 struct radix_node *treenodes) 116 { 117 struct rtentry *rt = (struct rtentry *)treenodes; 118 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)rt_key(rt); 119 struct radix_node *ret; 120 121 /* 122 * For IPv6, all unicast non-host routes are automatically cloning. 123 */ 124 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) 125 rt->rt_flags |= RTF_MULTICAST; 126 127 if (!(rt->rt_flags & (RTF_HOST | RTF_CLONING | RTF_MULTICAST))) { 128 rt->rt_flags |= RTF_PRCLONING; 129 } 130 131 /* 132 * A little bit of help for both IPv6 output and input: 133 * For local addresses, we make sure that RTF_LOCAL is set, 134 * with the thought that this might one day be used to speed up 135 * ip_input(). 136 * 137 * We also mark routes to multicast addresses as such, because 138 * it's easy to do and might be useful (but this is much more 139 * dubious since it's so easy to inspect the address). (This 140 * is done above.) 141 * 142 * XXX 143 * should elaborate the code. 144 */ 145 if (rt->rt_flags & RTF_HOST) { 146 if (IN6_ARE_ADDR_EQUAL(&satosin6(rt->rt_ifa->ifa_addr) 147 ->sin6_addr, 148 &sin6->sin6_addr)) { 149 rt->rt_flags |= RTF_LOCAL; 150 } 151 } 152 153 if (!rt->rt_rmx.rmx_mtu && !(rt->rt_rmx.rmx_locks & RTV_MTU) && 154 rt->rt_ifp != NULL) 155 rt->rt_rmx.rmx_mtu = rt->rt_ifp->if_mtu; 156 157 ret = rn_addroute(key, mask, head, treenodes); 158 if (ret == NULL && rt->rt_flags & RTF_HOST) { 159 struct rtentry *rt2; 160 161 /* 162 * We are trying to add a host route, but can't. 163 * Find out if it is because of an 164 * ARP entry and delete it if so. 165 */ 166 rt2 = rtlookup((struct sockaddr *)sin6, 0, 167 RTF_CLONING | RTF_PRCLONING); 168 if (rt2 != NULL) { 169 --rt2->rt_refcnt; 170 if (rt2->rt_flags & RTF_LLINFO && 171 rt2->rt_flags & RTF_HOST && 172 rt2->rt_gateway && 173 rt2->rt_gateway->sa_family == AF_LINK) { 174 rtrequest(RTM_DELETE, 175 (struct sockaddr *)rt_key(rt2), 176 rt2->rt_gateway, 177 rt_mask(rt2), rt2->rt_flags, 0); 178 ret = rn_addroute(key, mask, head, 179 treenodes); 180 } 181 } 182 } else if (ret == NULL && rt->rt_flags & RTF_CLONING) { 183 struct rtentry *rt2; 184 185 /* 186 * We are trying to add a net route, but can't. 187 * The following case should be allowed, so we'll make a 188 * special check for this: 189 * Two IPv6 addresses with the same prefix is assigned 190 * to a single interrface. 191 * # ifconfig if0 inet6 3ffe:0501::1 prefix 64 alias (*1) 192 * # ifconfig if0 inet6 3ffe:0501::2 prefix 64 alias (*2) 193 * In this case, (*1) and (*2) want to add the same 194 * net route entry, 3ffe:0501:: -> if0. 195 * This case should not raise an error. 196 */ 197 rt2 = rtlookup((struct sockaddr *)sin6, 0, 198 RTF_CLONING | RTF_PRCLONING); 199 if (rt2 != NULL) { 200 if ((rt2->rt_flags & (RTF_CLONING|RTF_HOST|RTF_GATEWAY)) 201 == RTF_CLONING && 202 rt2->rt_gateway && 203 rt2->rt_gateway->sa_family == AF_LINK && 204 rt2->rt_ifp == rt->rt_ifp) { 205 ret = rt2->rt_nodes; 206 } 207 --rt2->rt_refcnt; 208 } 209 } 210 return ret; 211 } 212 213 /* 214 * This code is the inverse of in6_clsroute: on first reference, if we 215 * were managing the route, stop doing so and set the expiration timer 216 * back off again. 217 */ 218 static struct radix_node * 219 in6_matchroute(char *key, struct radix_node_head *head) 220 { 221 struct radix_node *rn = rn_match(key, head); 222 struct rtentry *rt = (struct rtentry *)rn; 223 224 if (rt != NULL && rt->rt_refcnt == 0) { /* this is first reference */ 225 if (rt->rt_flags & RTPRF_OURS) { 226 rt->rt_flags &= ~RTPRF_OURS; 227 rt->rt_rmx.rmx_expire = 0; 228 } 229 } 230 return rn; 231 } 232 233 SYSCTL_DECL(_net_inet6_ip6); 234 235 static int rtq_reallyold = 60*60; 236 /* one hour is ``really old'' */ 237 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_RTEXPIRE, rtexpire, 238 CTLFLAG_RW, &rtq_reallyold , 0, ""); 239 240 static int rtq_minreallyold = 10; 241 /* never automatically crank down to less */ 242 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_RTMINEXPIRE, rtminexpire, 243 CTLFLAG_RW, &rtq_minreallyold , 0, ""); 244 245 static int rtq_toomany = 128; 246 /* 128 cached routes is ``too many'' */ 247 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_RTMAXCACHE, rtmaxcache, 248 CTLFLAG_RW, &rtq_toomany , 0, ""); 249 250 251 /* 252 * On last reference drop, mark the route as belong to us so that it can be 253 * timed out. 254 */ 255 static void 256 in6_clsroute(struct radix_node *rn, struct radix_node_head *head) 257 { 258 struct rtentry *rt = (struct rtentry *)rn; 259 260 if (!(rt->rt_flags & RTF_UP)) 261 return; /* prophylactic measures */ 262 263 if ((rt->rt_flags & (RTF_LLINFO | RTF_HOST)) != RTF_HOST) 264 return; 265 266 if ((rt->rt_flags & (RTF_WASCLONED | RTPRF_OURS)) 267 != RTF_WASCLONED) 268 return; 269 270 /* 271 * As requested by David Greenman: 272 * If rtq_reallyold is 0, just delete the route without 273 * waiting for a timeout cycle to kill it. 274 */ 275 if (rtq_reallyold != 0) { 276 rt->rt_flags |= RTPRF_OURS; 277 rt->rt_rmx.rmx_expire = time_second + rtq_reallyold; 278 } else { 279 rtrequest(RTM_DELETE, 280 (struct sockaddr *)rt_key(rt), 281 rt->rt_gateway, rt_mask(rt), 282 rt->rt_flags, 0); 283 } 284 } 285 286 struct rtqk_arg { 287 struct radix_node_head *rnh; 288 int mode; 289 int updating; 290 int draining; 291 int killed; 292 int found; 293 time_t nextstop; 294 }; 295 296 /* 297 * Get rid of old routes. When draining, this deletes everything, even when 298 * the timeout is not expired yet. When updating, this makes sure that 299 * nothing has a timeout longer than the current value of rtq_reallyold. 300 */ 301 static int 302 in6_rtqkill(struct radix_node *rn, void *rock) 303 { 304 struct rtqk_arg *ap = rock; 305 struct rtentry *rt = (struct rtentry *)rn; 306 int err; 307 308 if (rt->rt_flags & RTPRF_OURS) { 309 ap->found++; 310 311 if (ap->draining || rt->rt_rmx.rmx_expire <= time_second) { 312 if (rt->rt_refcnt > 0) 313 panic("rtqkill route really not free"); 314 315 err = rtrequest(RTM_DELETE, 316 (struct sockaddr *)rt_key(rt), 317 rt->rt_gateway, rt_mask(rt), 318 rt->rt_flags, 0); 319 if (err) { 320 log(LOG_WARNING, "in6_rtqkill: error %d", err); 321 } else { 322 ap->killed++; 323 } 324 } else { 325 if (ap->updating 326 && (rt->rt_rmx.rmx_expire - time_second 327 > rtq_reallyold)) { 328 rt->rt_rmx.rmx_expire = time_second 329 + rtq_reallyold; 330 } 331 ap->nextstop = lmin(ap->nextstop, 332 rt->rt_rmx.rmx_expire); 333 } 334 } 335 336 return 0; 337 } 338 339 #define RTQ_TIMEOUT 60*10 /* run no less than once every ten minutes */ 340 static int rtq_timeout = RTQ_TIMEOUT; 341 342 static void 343 in6_rtqtimo(void *rock) 344 { 345 struct radix_node_head *rnh = rock; 346 struct rtqk_arg arg; 347 struct timeval atv; 348 static time_t last_adjusted_timeout = 0; 349 int s; 350 351 arg.found = arg.killed = 0; 352 arg.rnh = rnh; 353 arg.nextstop = time_second + rtq_timeout; 354 arg.draining = arg.updating = 0; 355 s = splnet(); 356 rnh->rnh_walktree(rnh, in6_rtqkill, &arg); 357 splx(s); 358 359 /* 360 * Attempt to be somewhat dynamic about this: 361 * If there are ``too many'' routes sitting around taking up space, 362 * then crank down the timeout, and see if we can't make some more 363 * go away. However, we make sure that we will never adjust more 364 * than once in rtq_timeout seconds, to keep from cranking down too 365 * hard. 366 */ 367 if ((arg.found - arg.killed > rtq_toomany) 368 && (time_second - last_adjusted_timeout >= rtq_timeout) 369 && rtq_reallyold > rtq_minreallyold) { 370 rtq_reallyold = 2*rtq_reallyold / 3; 371 if (rtq_reallyold < rtq_minreallyold) { 372 rtq_reallyold = rtq_minreallyold; 373 } 374 375 last_adjusted_timeout = time_second; 376 #ifdef DIAGNOSTIC 377 log(LOG_DEBUG, "in6_rtqtimo: adjusted rtq_reallyold to %d", 378 rtq_reallyold); 379 #endif 380 arg.found = arg.killed = 0; 381 arg.updating = 1; 382 s = splnet(); 383 rnh->rnh_walktree(rnh, in6_rtqkill, &arg); 384 splx(s); 385 } 386 387 atv.tv_usec = 0; 388 atv.tv_sec = arg.nextstop; 389 callout_reset(&in6_rtqtimo_ch, tvtohz_high(&atv), in6_rtqtimo, rock); 390 } 391 392 /* 393 * Age old PMTUs. 394 */ 395 struct mtuex_arg { 396 struct radix_node_head *rnh; 397 time_t nextstop; 398 }; 399 400 static int 401 in6_mtuexpire(struct radix_node *rn, void *rock) 402 { 403 struct rtentry *rt = (struct rtentry *)rn; 404 struct mtuex_arg *ap = rock; 405 406 /* sanity */ 407 if (!rt) 408 panic("rt == NULL in in6_mtuexpire"); 409 410 if (rt->rt_rmx.rmx_expire && !(rt->rt_flags & RTF_PROBEMTU)) { 411 if (rt->rt_rmx.rmx_expire <= time_second) { 412 rt->rt_flags |= RTF_PROBEMTU; 413 } else { 414 ap->nextstop = lmin(ap->nextstop, 415 rt->rt_rmx.rmx_expire); 416 } 417 } 418 419 return 0; 420 } 421 422 #define MTUTIMO_DEFAULT (60*1) 423 424 static void 425 in6_mtutimo(void *rock) 426 { 427 struct radix_node_head *rnh = rock; 428 struct mtuex_arg arg; 429 struct timeval atv; 430 int s; 431 432 arg.rnh = rnh; 433 arg.nextstop = time_second + MTUTIMO_DEFAULT; 434 s = splnet(); 435 rnh->rnh_walktree(rnh, in6_mtuexpire, &arg); 436 splx(s); 437 438 atv.tv_usec = 0; 439 atv.tv_sec = arg.nextstop; 440 if (atv.tv_sec < time_second) { 441 printf("invalid mtu expiration time on routing table\n"); 442 arg.nextstop = time_second + 30; /* last resort */ 443 } 444 callout_reset(&in6_mtutimo_ch, tvtohz_high(&atv), in6_mtutimo, rock); 445 } 446 447 #if 0 448 void 449 in6_rtqdrain(void) 450 { 451 struct radix_node_head *rnh = rt_tables[AF_INET6]; 452 struct rtqk_arg arg; 453 int s; 454 arg.found = arg.killed = 0; 455 arg.rnh = rnh; 456 arg.nextstop = 0; 457 arg.draining = 1; 458 arg.updating = 0; 459 s = splnet(); 460 rnh->rnh_walktree(rnh, in6_rtqkill, &arg); 461 splx(s); 462 } 463 #endif 464 465 /* 466 * Initialize our routing tree. 467 */ 468 int 469 in6_inithead(void **head, int off) 470 { 471 struct radix_node_head *rnh; 472 473 if (!rn_inithead(head, off)) 474 return 0; 475 476 if (head != (void **)&rt_tables[AF_INET6]) /* BOGUS! */ 477 return 1; /* only do this for the real routing table */ 478 479 rnh = *head; 480 rnh->rnh_addaddr = in6_addroute; 481 rnh->rnh_matchaddr = in6_matchroute; 482 rnh->rnh_close = in6_clsroute; 483 callout_init(&in6_mtutimo_ch); 484 callout_init(&in6_rtqtimo_ch); 485 in6_rtqtimo(rnh); /* kick off timeout first time */ 486 in6_mtutimo(rnh); /* kick off timeout first time */ 487 return 1; 488 } 489