1 /* $FreeBSD: src/sys/netinet6/in6_rmx.c,v 1.1.2.4 2004/10/06 02:35:17 suz Exp $ */ 2 /* $DragonFly: src/sys/netinet6/in6_rmx.c,v 1.13 2005/06/03 19:56:08 eirikn 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 #include <sys/thread2.h> 87 88 #include <net/if.h> 89 #include <net/route.h> 90 #include <netinet/in.h> 91 #include <netinet/ip_var.h> 92 #include <netinet/in_var.h> 93 94 #include <netinet/ip6.h> 95 #include <netinet6/ip6_var.h> 96 97 #include <netinet/icmp6.h> 98 99 #include <netinet/tcp.h> 100 #include <netinet/tcp_seq.h> 101 #include <netinet/tcp_timer.h> 102 #include <netinet/tcp_var.h> 103 104 static struct callout in6_rtqtimo_ch; 105 static struct callout in6_mtutimo_ch; 106 107 extern int in6_inithead (void **head, int off); 108 109 #define RTPRF_OURS RTF_PROTO3 /* set on routes we manage */ 110 111 /* 112 * Do what we need to do when inserting a route. 113 */ 114 static struct radix_node * 115 in6_addroute(char *key, char *mask, struct radix_node_head *head, 116 struct radix_node *treenodes) 117 { 118 struct rtentry *rt = (struct rtentry *)treenodes; 119 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)rt_key(rt); 120 struct radix_node *ret; 121 122 /* 123 * For IPv6, all unicast non-host routes are automatically cloning. 124 */ 125 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) 126 rt->rt_flags |= RTF_MULTICAST; 127 128 if (!(rt->rt_flags & (RTF_HOST | RTF_CLONING | RTF_MULTICAST))) { 129 rt->rt_flags |= RTF_PRCLONING; 130 } 131 132 /* 133 * A little bit of help for both IPv6 output and input: 134 * For local addresses, we make sure that RTF_LOCAL is set, 135 * with the thought that this might one day be used to speed up 136 * ip_input(). 137 * 138 * We also mark routes to multicast addresses as such, because 139 * it's easy to do and might be useful (but this is much more 140 * dubious since it's so easy to inspect the address). (This 141 * is done above.) 142 * 143 * XXX 144 * should elaborate the code. 145 */ 146 if (rt->rt_flags & RTF_HOST) { 147 if (IN6_ARE_ADDR_EQUAL(&satosin6(rt->rt_ifa->ifa_addr) 148 ->sin6_addr, 149 &sin6->sin6_addr)) { 150 rt->rt_flags |= RTF_LOCAL; 151 } 152 } 153 154 if (!rt->rt_rmx.rmx_mtu && !(rt->rt_rmx.rmx_locks & RTV_MTU) && 155 rt->rt_ifp != NULL) 156 rt->rt_rmx.rmx_mtu = rt->rt_ifp->if_mtu; 157 158 ret = rn_addroute(key, mask, head, treenodes); 159 if (ret == NULL && rt->rt_flags & RTF_HOST) { 160 struct rtentry *rt2; 161 162 /* 163 * We are trying to add a host route, but can't. 164 * Find out if it is because of an 165 * ARP entry and delete it if so. 166 */ 167 rt2 = rtpurelookup((struct sockaddr *)sin6); 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, rt_key(rt2), 175 rt2->rt_gateway, rt_mask(rt2), 176 rt2->rt_flags, NULL); 177 ret = rn_addroute(key, mask, head, treenodes); 178 } 179 } 180 } else if (ret == NULL && rt->rt_flags & RTF_CLONING) { 181 struct rtentry *rt2; 182 183 /* 184 * We are trying to add a net route, but can't. 185 * The following case should be allowed, so we'll make a 186 * special check for this: 187 * Two IPv6 addresses with the same prefix is assigned 188 * to a single interrface. 189 * # ifconfig if0 inet6 3ffe:0501::1 prefix 64 alias (*1) 190 * # ifconfig if0 inet6 3ffe:0501::2 prefix 64 alias (*2) 191 * In this case, (*1) and (*2) want to add the same 192 * net route entry, 3ffe:0501:: -> if0. 193 * This case should not raise an error. 194 */ 195 rt2 = rtpurelookup((struct sockaddr *)sin6); 196 if (rt2 != NULL) { 197 if ((rt2->rt_flags & (RTF_CLONING|RTF_HOST|RTF_GATEWAY)) 198 == RTF_CLONING && 199 rt2->rt_gateway && 200 rt2->rt_gateway->sa_family == AF_LINK && 201 rt2->rt_ifp == rt->rt_ifp) { 202 ret = rt2->rt_nodes; 203 } 204 --rt2->rt_refcnt; 205 } 206 } 207 return ret; 208 } 209 210 /* 211 * This code is the inverse of in6_clsroute: on first reference, if we 212 * were managing the route, stop doing so and set the expiration timer 213 * back off again. 214 */ 215 static struct radix_node * 216 in6_matchroute(char *key, struct radix_node_head *head) 217 { 218 struct radix_node *rn = rn_match(key, head); 219 struct rtentry *rt = (struct rtentry *)rn; 220 221 if (rt != NULL && rt->rt_refcnt == 0) { /* this is first reference */ 222 if (rt->rt_flags & RTPRF_OURS) { 223 rt->rt_flags &= ~RTPRF_OURS; 224 rt->rt_rmx.rmx_expire = 0; 225 } 226 } 227 return rn; 228 } 229 230 SYSCTL_DECL(_net_inet6_ip6); 231 232 static int rtq_reallyold = 60*60; 233 /* one hour is ``really old'' */ 234 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_RTEXPIRE, rtexpire, 235 CTLFLAG_RW, &rtq_reallyold , 0, ""); 236 237 static int rtq_minreallyold = 10; 238 /* never automatically crank down to less */ 239 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_RTMINEXPIRE, rtminexpire, 240 CTLFLAG_RW, &rtq_minreallyold , 0, ""); 241 242 static int rtq_toomany = 128; 243 /* 128 cached routes is ``too many'' */ 244 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_RTMAXCACHE, rtmaxcache, 245 CTLFLAG_RW, &rtq_toomany , 0, ""); 246 247 248 /* 249 * On last reference drop, mark the route as belong to us so that it can be 250 * timed out. 251 */ 252 static void 253 in6_clsroute(struct radix_node *rn, struct radix_node_head *head) 254 { 255 struct rtentry *rt = (struct rtentry *)rn; 256 257 if (!(rt->rt_flags & RTF_UP)) 258 return; /* prophylactic measures */ 259 260 if ((rt->rt_flags & (RTF_LLINFO | RTF_HOST)) != RTF_HOST) 261 return; 262 263 if ((rt->rt_flags & (RTF_WASCLONED | RTPRF_OURS)) != RTF_WASCLONED) 264 return; 265 266 /* 267 * As requested by David Greenman: 268 * If rtq_reallyold is 0, just delete the route without 269 * waiting for a timeout cycle to kill it. 270 */ 271 if (rtq_reallyold != 0) { 272 rt->rt_flags |= RTPRF_OURS; 273 rt->rt_rmx.rmx_expire = time_second + rtq_reallyold; 274 } else { 275 /* 276 * Remove route from the radix tree, but defer deallocation 277 * until we return to rtfree(). 278 */ 279 rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway, rt_mask(rt), 280 rt->rt_flags, &rt); 281 } 282 } 283 284 struct rtqk_arg { 285 struct radix_node_head *rnh; 286 int mode; 287 int updating; 288 int draining; 289 int killed; 290 int found; 291 time_t nextstop; 292 }; 293 294 /* 295 * Get rid of old routes. When draining, this deletes everything, even when 296 * the timeout is not expired yet. When updating, this makes sure that 297 * nothing has a timeout longer than the current value of rtq_reallyold. 298 */ 299 static int 300 in6_rtqkill(struct radix_node *rn, void *rock) 301 { 302 struct rtqk_arg *ap = rock; 303 struct rtentry *rt = (struct rtentry *)rn; 304 int err; 305 306 if (rt->rt_flags & RTPRF_OURS) { 307 ap->found++; 308 309 if (ap->draining || rt->rt_rmx.rmx_expire <= time_second) { 310 if (rt->rt_refcnt > 0) 311 panic("rtqkill route really not free"); 312 313 err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway, 314 rt_mask(rt), rt->rt_flags, NULL); 315 if (err) 316 log(LOG_WARNING, "in6_rtqkill: error %d", err); 317 else 318 ap->killed++; 319 } else { 320 if (ap->updating && 321 (rt->rt_rmx.rmx_expire - time_second > 322 rtq_reallyold)) { 323 rt->rt_rmx.rmx_expire = 324 time_second + rtq_reallyold; 325 } 326 ap->nextstop = lmin(ap->nextstop, 327 rt->rt_rmx.rmx_expire); 328 } 329 } 330 331 return 0; 332 } 333 334 #define RTQ_TIMEOUT 60*10 /* run no less than once every ten minutes */ 335 static int rtq_timeout = RTQ_TIMEOUT; 336 337 static void 338 in6_rtqtimo(void *rock) 339 { 340 struct radix_node_head *rnh = rock; 341 struct rtqk_arg arg; 342 struct timeval atv; 343 static time_t last_adjusted_timeout = 0; 344 345 arg.found = arg.killed = 0; 346 arg.rnh = rnh; 347 arg.nextstop = time_second + rtq_timeout; 348 arg.draining = arg.updating = 0; 349 crit_enter(); 350 rnh->rnh_walktree(rnh, in6_rtqkill, &arg); 351 crit_exit(); 352 353 /* 354 * Attempt to be somewhat dynamic about this: 355 * If there are ``too many'' routes sitting around taking up space, 356 * then crank down the timeout, and see if we can't make some more 357 * go away. However, we make sure that we will never adjust more 358 * than once in rtq_timeout seconds, to keep from cranking down too 359 * hard. 360 */ 361 if ((arg.found - arg.killed > rtq_toomany) 362 && (time_second - last_adjusted_timeout >= rtq_timeout) 363 && rtq_reallyold > rtq_minreallyold) { 364 rtq_reallyold = 2*rtq_reallyold / 3; 365 if (rtq_reallyold < rtq_minreallyold) { 366 rtq_reallyold = rtq_minreallyold; 367 } 368 369 last_adjusted_timeout = time_second; 370 #ifdef DIAGNOSTIC 371 log(LOG_DEBUG, "in6_rtqtimo: adjusted rtq_reallyold to %d", 372 rtq_reallyold); 373 #endif 374 arg.found = arg.killed = 0; 375 arg.updating = 1; 376 crit_enter(); 377 rnh->rnh_walktree(rnh, in6_rtqkill, &arg); 378 crit_exit(); 379 } 380 381 atv.tv_usec = 0; 382 atv.tv_sec = arg.nextstop; 383 callout_reset(&in6_rtqtimo_ch, tvtohz_high(&atv), in6_rtqtimo, rock); 384 } 385 386 /* 387 * Age old PMTUs. 388 */ 389 struct mtuex_arg { 390 struct radix_node_head *rnh; 391 time_t nextstop; 392 }; 393 394 static int 395 in6_mtuexpire(struct radix_node *rn, void *rock) 396 { 397 struct rtentry *rt = (struct rtentry *)rn; 398 struct mtuex_arg *ap = rock; 399 400 /* sanity */ 401 if (!rt) 402 panic("rt == NULL in in6_mtuexpire"); 403 404 if (rt->rt_rmx.rmx_expire && !(rt->rt_flags & RTF_PROBEMTU)) { 405 if (rt->rt_rmx.rmx_expire <= time_second) { 406 rt->rt_flags |= RTF_PROBEMTU; 407 } else { 408 ap->nextstop = lmin(ap->nextstop, 409 rt->rt_rmx.rmx_expire); 410 } 411 } 412 413 return 0; 414 } 415 416 #define MTUTIMO_DEFAULT (60*1) 417 418 static void 419 in6_mtutimo(void *rock) 420 { 421 struct radix_node_head *rnh = rock; 422 struct mtuex_arg arg; 423 struct timeval atv; 424 425 arg.rnh = rnh; 426 arg.nextstop = time_second + MTUTIMO_DEFAULT; 427 crit_enter(); 428 rnh->rnh_walktree(rnh, in6_mtuexpire, &arg); 429 crit_exit(); 430 431 atv.tv_usec = 0; 432 atv.tv_sec = arg.nextstop; 433 if (atv.tv_sec < time_second) { 434 printf("invalid mtu expiration time on routing table\n"); 435 arg.nextstop = time_second + 30; /* last resort */ 436 } 437 callout_reset(&in6_mtutimo_ch, tvtohz_high(&atv), in6_mtutimo, rock); 438 } 439 440 #if 0 441 void 442 in6_rtqdrain(void) 443 { 444 struct radix_node_head *rnh = rt_tables[AF_INET6]; 445 struct rtqk_arg arg; 446 447 arg.found = arg.killed = 0; 448 arg.rnh = rnh; 449 arg.nextstop = 0; 450 arg.draining = 1; 451 arg.updating = 0; 452 crit_enter(); 453 rnh->rnh_walktree(rnh, in6_rtqkill, &arg); 454 crit_exit(); 455 } 456 #endif 457 458 /* 459 * Initialize our routing tree. 460 */ 461 int 462 in6_inithead(void **head, int off) 463 { 464 struct radix_node_head *rnh; 465 466 if (!rn_inithead(head, off)) 467 return 0; 468 469 if (head != (void **)&rt_tables[AF_INET6]) /* BOGUS! */ 470 return 1; /* only do this for the real routing table */ 471 472 rnh = *head; 473 rnh->rnh_addaddr = in6_addroute; 474 rnh->rnh_matchaddr = in6_matchroute; 475 rnh->rnh_close = in6_clsroute; 476 callout_init(&in6_mtutimo_ch); 477 callout_init(&in6_rtqtimo_ch); 478 in6_rtqtimo(rnh); /* kick off timeout first time */ 479 in6_mtutimo(rnh); /* kick off timeout first time */ 480 return 1; 481 } 482