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 1.14 2006/01/31 19:05:42 dillon 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/globaldata.h> 87 #include <sys/thread2.h> 88 89 #include <net/if.h> 90 #include <net/route.h> 91 #include <netinet/in.h> 92 #include <netinet/ip_var.h> 93 #include <netinet/in_var.h> 94 95 #include <netinet/ip6.h> 96 #include <netinet6/ip6_var.h> 97 98 #include <netinet/icmp6.h> 99 100 #include <netinet/tcp.h> 101 #include <netinet/tcp_seq.h> 102 #include <netinet/tcp_timer.h> 103 #include <netinet/tcp_var.h> 104 105 static struct callout in6_rtqtimo_ch[MAXCPU]; 106 static struct callout in6_mtutimo_ch[MAXCPU]; 107 108 extern int in6_inithead (void **head, int off); 109 110 #define RTPRF_OURS RTF_PROTO3 /* set on routes we manage */ 111 112 /* 113 * Do what we need to do when inserting a route. 114 */ 115 static struct radix_node * 116 in6_addroute(char *key, char *mask, struct radix_node_head *head, 117 struct radix_node *treenodes) 118 { 119 struct rtentry *rt = (struct rtentry *)treenodes; 120 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)rt_key(rt); 121 struct radix_node *ret; 122 123 /* 124 * For IPv6, all unicast non-host routes are automatically cloning. 125 */ 126 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) 127 rt->rt_flags |= RTF_MULTICAST; 128 129 if (!(rt->rt_flags & (RTF_HOST | RTF_CLONING | RTF_MULTICAST))) { 130 rt->rt_flags |= RTF_PRCLONING; 131 } 132 133 /* 134 * A little bit of help for both IPv6 output and input: 135 * For local addresses, we make sure that RTF_LOCAL is set, 136 * with the thought that this might one day be used to speed up 137 * ip_input(). 138 * 139 * We also mark routes to multicast addresses as such, because 140 * it's easy to do and might be useful (but this is much more 141 * dubious since it's so easy to inspect the address). (This 142 * is done above.) 143 * 144 * XXX 145 * should elaborate the code. 146 */ 147 if (rt->rt_flags & RTF_HOST) { 148 if (IN6_ARE_ADDR_EQUAL(&satosin6(rt->rt_ifa->ifa_addr) 149 ->sin6_addr, 150 &sin6->sin6_addr)) { 151 rt->rt_flags |= RTF_LOCAL; 152 } 153 } 154 155 if (!rt->rt_rmx.rmx_mtu && !(rt->rt_rmx.rmx_locks & RTV_MTU) && 156 rt->rt_ifp != NULL) 157 rt->rt_rmx.rmx_mtu = rt->rt_ifp->if_mtu; 158 159 ret = rn_addroute(key, mask, head, treenodes); 160 if (ret == NULL && rt->rt_flags & RTF_HOST) { 161 struct rtentry *rt2; 162 163 /* 164 * We are trying to add a host route, but can't. 165 * Find out if it is because of an 166 * ARP entry and delete it if so. 167 */ 168 rt2 = rtpurelookup((struct sockaddr *)sin6); 169 if (rt2 != NULL) { 170 --rt2->rt_refcnt; 171 if (rt2->rt_flags & RTF_LLINFO && 172 rt2->rt_flags & RTF_HOST && 173 rt2->rt_gateway && 174 rt2->rt_gateway->sa_family == AF_LINK) { 175 rtrequest(RTM_DELETE, rt_key(rt2), 176 rt2->rt_gateway, rt_mask(rt2), 177 rt2->rt_flags, NULL); 178 ret = rn_addroute(key, mask, head, treenodes); 179 } 180 } 181 } else if (ret == NULL && rt->rt_flags & RTF_CLONING) { 182 struct rtentry *rt2; 183 184 /* 185 * We are trying to add a net route, but can't. 186 * The following case should be allowed, so we'll make a 187 * special check for this: 188 * Two IPv6 addresses with the same prefix is assigned 189 * to a single interrface. 190 * # ifconfig if0 inet6 3ffe:0501::1 prefix 64 alias (*1) 191 * # ifconfig if0 inet6 3ffe:0501::2 prefix 64 alias (*2) 192 * In this case, (*1) and (*2) want to add the same 193 * net route entry, 3ffe:0501:: -> if0. 194 * This case should not raise an error. 195 */ 196 rt2 = rtpurelookup((struct sockaddr *)sin6); 197 if (rt2 != NULL) { 198 if ((rt2->rt_flags & (RTF_CLONING|RTF_HOST|RTF_GATEWAY)) 199 == RTF_CLONING && 200 rt2->rt_gateway && 201 rt2->rt_gateway->sa_family == AF_LINK && 202 rt2->rt_ifp == rt->rt_ifp) { 203 ret = rt2->rt_nodes; 204 } 205 --rt2->rt_refcnt; 206 } 207 } 208 return ret; 209 } 210 211 /* 212 * This code is the inverse of in6_clsroute: on first reference, if we 213 * were managing the route, stop doing so and set the expiration timer 214 * back off again. 215 */ 216 static struct radix_node * 217 in6_matchroute(char *key, struct radix_node_head *head) 218 { 219 struct radix_node *rn = rn_match(key, head); 220 struct rtentry *rt = (struct rtentry *)rn; 221 222 if (rt != NULL && rt->rt_refcnt == 0) { /* this is first reference */ 223 if (rt->rt_flags & RTPRF_OURS) { 224 rt->rt_flags &= ~RTPRF_OURS; 225 rt->rt_rmx.rmx_expire = 0; 226 } 227 } 228 return rn; 229 } 230 231 SYSCTL_DECL(_net_inet6_ip6); 232 233 static int rtq_reallyold = 60*60; 234 /* one hour is ``really old'' */ 235 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_RTEXPIRE, rtexpire, 236 CTLFLAG_RW, &rtq_reallyold , 0, ""); 237 238 static int rtq_minreallyold = 10; 239 /* never automatically crank down to less */ 240 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_RTMINEXPIRE, rtminexpire, 241 CTLFLAG_RW, &rtq_minreallyold , 0, ""); 242 243 static int rtq_toomany = 128; 244 /* 128 cached routes is ``too many'' */ 245 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_RTMAXCACHE, rtmaxcache, 246 CTLFLAG_RW, &rtq_toomany , 0, ""); 247 248 249 /* 250 * On last reference drop, mark the route as belong to us so that it can be 251 * timed out. 252 */ 253 static void 254 in6_clsroute(struct radix_node *rn, struct radix_node_head *head) 255 { 256 struct rtentry *rt = (struct rtentry *)rn; 257 258 if (!(rt->rt_flags & RTF_UP)) 259 return; /* prophylactic measures */ 260 261 if ((rt->rt_flags & (RTF_LLINFO | RTF_HOST)) != RTF_HOST) 262 return; 263 264 if ((rt->rt_flags & (RTF_WASCLONED | RTPRF_OURS)) != RTF_WASCLONED) 265 return; 266 267 /* 268 * As requested by David Greenman: 269 * If rtq_reallyold is 0, just delete the route without 270 * waiting for a timeout cycle to kill it. 271 */ 272 if (rtq_reallyold != 0) { 273 rt->rt_flags |= RTPRF_OURS; 274 rt->rt_rmx.rmx_expire = time_second + rtq_reallyold; 275 } else { 276 /* 277 * Remove route from the radix tree, but defer deallocation 278 * until we return to rtfree(). 279 */ 280 rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway, rt_mask(rt), 281 rt->rt_flags, &rt); 282 } 283 } 284 285 struct rtqk_arg { 286 struct radix_node_head *rnh; 287 int mode; 288 int updating; 289 int draining; 290 int killed; 291 int found; 292 time_t nextstop; 293 }; 294 295 /* 296 * Get rid of old routes. When draining, this deletes everything, even when 297 * the timeout is not expired yet. When updating, this makes sure that 298 * nothing has a timeout longer than the current value of rtq_reallyold. 299 */ 300 static int 301 in6_rtqkill(struct radix_node *rn, void *rock) 302 { 303 struct rtqk_arg *ap = rock; 304 struct rtentry *rt = (struct rtentry *)rn; 305 int err; 306 307 if (rt->rt_flags & RTPRF_OURS) { 308 ap->found++; 309 310 if (ap->draining || rt->rt_rmx.rmx_expire <= time_second) { 311 if (rt->rt_refcnt > 0) 312 panic("rtqkill route really not free"); 313 314 err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway, 315 rt_mask(rt), rt->rt_flags, NULL); 316 if (err) 317 log(LOG_WARNING, "in6_rtqkill: error %d", err); 318 else 319 ap->killed++; 320 } else { 321 if (ap->updating && 322 (rt->rt_rmx.rmx_expire - time_second > 323 rtq_reallyold)) { 324 rt->rt_rmx.rmx_expire = 325 time_second + rtq_reallyold; 326 } 327 ap->nextstop = lmin(ap->nextstop, 328 rt->rt_rmx.rmx_expire); 329 } 330 } 331 332 return 0; 333 } 334 335 #define RTQ_TIMEOUT 60*10 /* run no less than once every ten minutes */ 336 static int rtq_timeout = RTQ_TIMEOUT; 337 338 static void 339 in6_rtqtimo(void *rock) 340 { 341 struct radix_node_head *rnh = rock; 342 struct rtqk_arg arg; 343 struct timeval atv; 344 static time_t last_adjusted_timeout = 0; 345 346 arg.found = arg.killed = 0; 347 arg.rnh = rnh; 348 arg.nextstop = time_second + rtq_timeout; 349 arg.draining = arg.updating = 0; 350 crit_enter(); 351 rnh->rnh_walktree(rnh, in6_rtqkill, &arg); 352 crit_exit(); 353 354 /* 355 * Attempt to be somewhat dynamic about this: 356 * If there are ``too many'' routes sitting around taking up space, 357 * then crank down the timeout, and see if we can't make some more 358 * go away. However, we make sure that we will never adjust more 359 * than once in rtq_timeout seconds, to keep from cranking down too 360 * hard. 361 */ 362 if ((arg.found - arg.killed > rtq_toomany) 363 && (time_second - last_adjusted_timeout >= rtq_timeout) 364 && rtq_reallyold > rtq_minreallyold) { 365 rtq_reallyold = 2*rtq_reallyold / 3; 366 if (rtq_reallyold < rtq_minreallyold) { 367 rtq_reallyold = rtq_minreallyold; 368 } 369 370 last_adjusted_timeout = time_second; 371 #ifdef DIAGNOSTIC 372 log(LOG_DEBUG, "in6_rtqtimo: adjusted rtq_reallyold to %d", 373 rtq_reallyold); 374 #endif 375 arg.found = arg.killed = 0; 376 arg.updating = 1; 377 crit_enter(); 378 rnh->rnh_walktree(rnh, in6_rtqkill, &arg); 379 crit_exit(); 380 } 381 382 atv.tv_usec = 0; 383 atv.tv_sec = arg.nextstop; 384 callout_reset(&in6_rtqtimo_ch[mycpuid], tvtohz_high(&atv), in6_rtqtimo, 385 rock); 386 } 387 388 /* 389 * Age old PMTUs. 390 */ 391 struct mtuex_arg { 392 struct radix_node_head *rnh; 393 time_t nextstop; 394 }; 395 396 static int 397 in6_mtuexpire(struct radix_node *rn, void *rock) 398 { 399 struct rtentry *rt = (struct rtentry *)rn; 400 struct mtuex_arg *ap = rock; 401 402 /* sanity */ 403 if (!rt) 404 panic("rt == NULL in in6_mtuexpire"); 405 406 if (rt->rt_rmx.rmx_expire && !(rt->rt_flags & RTF_PROBEMTU)) { 407 if (rt->rt_rmx.rmx_expire <= time_second) { 408 rt->rt_flags |= RTF_PROBEMTU; 409 } else { 410 ap->nextstop = lmin(ap->nextstop, 411 rt->rt_rmx.rmx_expire); 412 } 413 } 414 415 return 0; 416 } 417 418 #define MTUTIMO_DEFAULT (60*1) 419 420 static void 421 in6_mtutimo(void *rock) 422 { 423 struct radix_node_head *rnh = rock; 424 struct mtuex_arg arg; 425 struct timeval atv; 426 427 arg.rnh = rnh; 428 arg.nextstop = time_second + MTUTIMO_DEFAULT; 429 crit_enter(); 430 rnh->rnh_walktree(rnh, in6_mtuexpire, &arg); 431 crit_exit(); 432 433 atv.tv_usec = 0; 434 atv.tv_sec = arg.nextstop; 435 if (atv.tv_sec < time_second) { 436 printf("invalid mtu expiration time on routing table\n"); 437 arg.nextstop = time_second + 30; /* last resort */ 438 } 439 callout_reset(&in6_mtutimo_ch[mycpuid], tvtohz_high(&atv), in6_mtutimo, 440 rock); 441 } 442 443 #if 0 444 void 445 in6_rtqdrain(void) 446 { 447 struct radix_node_head *rnh = rt_tables[mycpuid][AF_INET6]; 448 struct rtqk_arg arg; 449 450 arg.found = arg.killed = 0; 451 arg.rnh = rnh; 452 arg.nextstop = 0; 453 arg.draining = 1; 454 arg.updating = 0; 455 crit_enter(); 456 rnh->rnh_walktree(rnh, in6_rtqkill, &arg); 457 crit_exit(); 458 } 459 #endif 460 461 /* 462 * Initialize our routing tree. 463 */ 464 int 465 in6_inithead(void **head, int off) 466 { 467 struct radix_node_head *rnh; 468 469 if (!rn_inithead(head, off)) 470 return 0; 471 472 if (head != (void **)&rt_tables[mycpuid][AF_INET6]) /* BOGUS! */ 473 return 1; /* only do this for the real routing table */ 474 475 rnh = *head; 476 rnh->rnh_addaddr = in6_addroute; 477 rnh->rnh_matchaddr = in6_matchroute; 478 rnh->rnh_close = in6_clsroute; 479 callout_init(&in6_mtutimo_ch[mycpuid]); 480 callout_init(&in6_rtqtimo_ch[mycpuid]); 481 in6_rtqtimo(rnh); /* kick off timeout first time */ 482 in6_mtutimo(rnh); /* kick off timeout first time */ 483 return 1; 484 } 485