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 1.15 2006/12/22 23:57:53 swildner 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 <net/netisr2.h> 92 #include <net/netmsg2.h> 93 #include <netinet/in.h> 94 #include <netinet/ip_var.h> 95 #include <netinet/in_var.h> 96 97 #include <netinet/ip6.h> 98 #include <netinet6/ip6_var.h> 99 100 #include <netinet/icmp6.h> 101 #include <netinet6/nd6.h> 102 103 #include <netinet/tcp.h> 104 #include <netinet/tcp_seq.h> 105 #include <netinet/tcp_timer.h> 106 #include <netinet/tcp_var.h> 107 108 struct in6_rttimo_ctx { 109 struct callout timo_ch; 110 struct netmsg_base timo_nmsg; 111 struct radix_node_head *timo_rnh; 112 } __cachealign; 113 114 static struct in6_rttimo_ctx in6_rtqtimo_ctx[MAXCPU]; 115 static struct in6_rttimo_ctx in6_mtutimo_ctx[MAXCPU]; 116 117 extern int in6_inithead (void **head, int off); 118 119 #define RTPRF_OURS RTF_PROTO3 /* set on routes we manage */ 120 121 /* 122 * Do what we need to do when inserting a route. 123 */ 124 static struct radix_node * 125 in6_addroute(char *key, char *mask, struct radix_node_head *head, 126 struct radix_node *treenodes) 127 { 128 struct rtentry *rt = (struct rtentry *)treenodes; 129 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)rt_key(rt); 130 struct radix_node *ret; 131 132 /* 133 * For IPv6, all unicast non-host routes are automatically cloning. 134 */ 135 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) 136 rt->rt_flags |= RTF_MULTICAST; 137 138 if (!(rt->rt_flags & (RTF_HOST | RTF_CLONING | RTF_MULTICAST))) { 139 rt->rt_flags |= RTF_PRCLONING; 140 } 141 142 /* 143 * A little bit of help for both IPv6 output and input: 144 * For local addresses, we make sure that RTF_LOCAL is set, 145 * with the thought that this might one day be used to speed up 146 * ip_input(). 147 * 148 * We also mark routes to multicast addresses as such, because 149 * it's easy to do and might be useful (but this is much more 150 * dubious since it's so easy to inspect the address). (This 151 * is done above.) 152 * 153 * XXX 154 * should elaborate the code. 155 */ 156 if (rt->rt_flags & RTF_HOST) { 157 if (IN6_ARE_ADDR_EQUAL(&satosin6(rt->rt_ifa->ifa_addr) 158 ->sin6_addr, 159 &sin6->sin6_addr)) { 160 rt->rt_flags |= RTF_LOCAL; 161 } 162 } 163 164 if (!rt->rt_rmx.rmx_mtu && !(rt->rt_rmx.rmx_locks & RTV_MTU) && 165 rt->rt_ifp != NULL) 166 rt->rt_rmx.rmx_mtu = IN6_LINKMTU(rt->rt_ifp); 167 168 ret = rn_addroute(key, mask, head, treenodes); 169 if (ret == NULL && rt->rt_flags & RTF_HOST) { 170 struct rtentry *rt2; 171 172 /* 173 * We are trying to add a host route, but can't. 174 * Find out if it is because of an 175 * ARP entry and delete it if so. 176 */ 177 rt2 = rtpurelookup((struct sockaddr *)sin6); 178 if (rt2 != NULL) { 179 --rt2->rt_refcnt; 180 if (rt2->rt_flags & RTF_LLINFO && 181 rt2->rt_flags & RTF_HOST && 182 rt2->rt_gateway && 183 rt2->rt_gateway->sa_family == AF_LINK) { 184 rtrequest(RTM_DELETE, rt_key(rt2), 185 rt2->rt_gateway, rt_mask(rt2), 186 rt2->rt_flags, NULL); 187 ret = rn_addroute(key, mask, head, treenodes); 188 } 189 } 190 } else if (ret == NULL && rt->rt_flags & RTF_CLONING) { 191 struct rtentry *rt2; 192 193 /* 194 * We are trying to add a net route, but can't. 195 * The following case should be allowed, so we'll make a 196 * special check for this: 197 * Two IPv6 addresses with the same prefix is assigned 198 * to a single interrface. 199 * # ifconfig if0 inet6 3ffe:0501::1 prefix 64 alias (*1) 200 * # ifconfig if0 inet6 3ffe:0501::2 prefix 64 alias (*2) 201 * In this case, (*1) and (*2) want to add the same 202 * net route entry, 3ffe:0501:: -> if0. 203 * This case should not raise an error. 204 */ 205 rt2 = rtpurelookup((struct sockaddr *)sin6); 206 if (rt2 != NULL) { 207 if ((rt2->rt_flags & (RTF_CLONING|RTF_HOST|RTF_GATEWAY)) 208 == RTF_CLONING && 209 rt2->rt_gateway && 210 rt2->rt_gateway->sa_family == AF_LINK && 211 rt2->rt_ifp == rt->rt_ifp) { 212 ret = rt2->rt_nodes; 213 } 214 --rt2->rt_refcnt; 215 } 216 } 217 return ret; 218 } 219 220 /* 221 * This code is the inverse of in6_clsroute: on first reference, if we 222 * were managing the route, stop doing so and set the expiration timer 223 * back off again. 224 */ 225 static struct radix_node * 226 in6_matchroute(char *key, struct radix_node_head *head) 227 { 228 struct radix_node *rn = rn_match(key, head); 229 struct rtentry *rt = (struct rtentry *)rn; 230 231 if (rt != NULL && rt->rt_refcnt == 0) { /* this is first reference */ 232 if (rt->rt_flags & RTPRF_OURS) { 233 rt->rt_flags &= ~RTPRF_OURS; 234 rt->rt_rmx.rmx_expire = 0; 235 } 236 } 237 return rn; 238 } 239 240 SYSCTL_DECL(_net_inet6_ip6); 241 242 static int rtq_reallyold = 60*60; 243 /* one hour is ``really old'' */ 244 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_RTEXPIRE, rtexpire, 245 CTLFLAG_RW, &rtq_reallyold , 0, "Default expiration time on cloned routes"); 246 247 static int rtq_minreallyold = 10; 248 /* never automatically crank down to less */ 249 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_RTMINEXPIRE, rtminexpire, CTLFLAG_RW, 250 &rtq_minreallyold , 0, "Minimum time to attempt to hold onto cloned routes"); 251 252 static int rtq_toomany = 128; 253 /* 128 cached routes is ``too many'' */ 254 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_RTMAXCACHE, rtmaxcache, 255 CTLFLAG_RW, &rtq_toomany , 0, "Upper limit on cloned routes"); 256 257 258 /* 259 * On last reference drop, mark the route as belong to us so that it can be 260 * timed out. 261 */ 262 static void 263 in6_clsroute(struct radix_node *rn, struct radix_node_head *head) 264 { 265 struct rtentry *rt = (struct rtentry *)rn; 266 267 if (!(rt->rt_flags & RTF_UP)) 268 return; /* prophylactic measures */ 269 270 if ((rt->rt_flags & (RTF_LLINFO | RTF_HOST)) != RTF_HOST) 271 return; 272 273 if ((rt->rt_flags & (RTF_WASCLONED | RTPRF_OURS)) != RTF_WASCLONED) 274 return; 275 276 /* 277 * As requested by David Greenman: 278 * If rtq_reallyold is 0, just delete the route without 279 * waiting for a timeout cycle to kill it. 280 */ 281 if (rtq_reallyold != 0) { 282 rt->rt_flags |= RTPRF_OURS; 283 rt->rt_rmx.rmx_expire = time_uptime + rtq_reallyold; 284 } else { 285 /* 286 * Remove route from the radix tree, but defer deallocation 287 * until we return to rtfree(). 288 */ 289 rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway, rt_mask(rt), 290 rt->rt_flags, &rt); 291 } 292 } 293 294 struct rtqk_arg { 295 struct radix_node_head *rnh; 296 int mode; 297 int updating; 298 int draining; 299 int killed; 300 int found; 301 time_t nextstop; 302 }; 303 304 /* 305 * Get rid of old routes. When draining, this deletes everything, even when 306 * the timeout is not expired yet. When updating, this makes sure that 307 * nothing has a timeout longer than the current value of rtq_reallyold. 308 */ 309 static int 310 in6_rtqkill(struct radix_node *rn, void *rock) 311 { 312 struct rtqk_arg *ap = rock; 313 struct rtentry *rt = (struct rtentry *)rn; 314 int err; 315 316 if (rt->rt_flags & RTPRF_OURS) { 317 ap->found++; 318 319 if (ap->draining || rt->rt_rmx.rmx_expire <= time_uptime) { 320 if (rt->rt_refcnt > 0) 321 panic("rtqkill route really not free"); 322 323 err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway, 324 rt_mask(rt), rt->rt_flags, NULL); 325 if (err) 326 log(LOG_WARNING, "in6_rtqkill: error %d", err); 327 else 328 ap->killed++; 329 } else { 330 if (ap->updating && 331 (rt->rt_rmx.rmx_expire - time_uptime > 332 rtq_reallyold)) { 333 rt->rt_rmx.rmx_expire = 334 time_uptime + rtq_reallyold; 335 } 336 ap->nextstop = lmin(ap->nextstop, 337 rt->rt_rmx.rmx_expire); 338 } 339 } 340 341 return 0; 342 } 343 344 #define RTQ_TIMEOUT 60*10 /* run no less than once every ten minutes */ 345 static int rtq_timeout = RTQ_TIMEOUT; 346 347 static void 348 in6_rtqtimo(void *arg __unused) 349 { 350 int cpuid = mycpuid; 351 struct lwkt_msg *lmsg = &in6_rtqtimo_ctx[cpuid].timo_nmsg.lmsg; 352 353 crit_enter(); 354 if (lmsg->ms_flags & MSGF_DONE) 355 lwkt_sendmsg_oncpu(netisr_cpuport(cpuid), lmsg); 356 crit_exit(); 357 } 358 359 static void 360 in6_rtqtimo_dispatch(netmsg_t nmsg) 361 { 362 struct in6_rttimo_ctx *ctx = &in6_rtqtimo_ctx[mycpuid]; 363 struct radix_node_head *rnh = ctx->timo_rnh; 364 struct rtqk_arg arg; 365 struct timeval atv; 366 static time_t last_adjusted_timeout = 0; 367 368 /* Reply ASAP */ 369 crit_enter(); 370 lwkt_replymsg(&nmsg->lmsg, 0); 371 crit_exit(); 372 373 arg.found = arg.killed = 0; 374 arg.rnh = rnh; 375 arg.nextstop = time_uptime + rtq_timeout; 376 arg.draining = arg.updating = 0; 377 rnh->rnh_walktree(rnh, in6_rtqkill, &arg); 378 379 /* 380 * Attempt to be somewhat dynamic about this: 381 * If there are ``too many'' routes sitting around taking up space, 382 * then crank down the timeout, and see if we can't make some more 383 * go away. However, we make sure that we will never adjust more 384 * than once in rtq_timeout seconds, to keep from cranking down too 385 * hard. 386 */ 387 if ((arg.found - arg.killed > rtq_toomany) 388 && (int)(time_uptime - last_adjusted_timeout) >= rtq_timeout 389 && rtq_reallyold > rtq_minreallyold) { 390 rtq_reallyold = 2*rtq_reallyold / 3; 391 if (rtq_reallyold < rtq_minreallyold) { 392 rtq_reallyold = rtq_minreallyold; 393 } 394 395 last_adjusted_timeout = time_uptime; 396 #ifdef DIAGNOSTIC 397 log(LOG_DEBUG, "in6_rtqtimo: adjusted rtq_reallyold to %d", 398 rtq_reallyold); 399 #endif 400 arg.found = arg.killed = 0; 401 arg.updating = 1; 402 rnh->rnh_walktree(rnh, in6_rtqkill, &arg); 403 } 404 405 atv.tv_usec = 0; 406 atv.tv_sec = arg.nextstop - time_uptime; 407 if ((int)atv.tv_sec < 1) { /* time shift safety */ 408 atv.tv_sec = 1; 409 arg.nextstop = time_uptime + atv.tv_sec; 410 } 411 if ((int)atv.tv_sec > rtq_timeout) { /* time shift safety */ 412 atv.tv_sec = rtq_timeout; 413 arg.nextstop = time_uptime + atv.tv_sec; 414 } 415 callout_reset(&ctx->timo_ch, tvtohz_high(&atv), in6_rtqtimo, NULL); 416 } 417 418 /* 419 * Age old PMTUs. 420 */ 421 struct mtuex_arg { 422 struct radix_node_head *rnh; 423 time_t nextstop; 424 }; 425 426 static int 427 in6_mtuexpire(struct radix_node *rn, void *rock) 428 { 429 struct rtentry *rt = (struct rtentry *)rn; 430 struct mtuex_arg *ap = rock; 431 432 /* sanity */ 433 if (!rt) 434 panic("rt == NULL in in6_mtuexpire"); 435 436 if (rt->rt_rmx.rmx_expire && !(rt->rt_flags & RTF_PROBEMTU)) { 437 if (rt->rt_rmx.rmx_expire <= time_uptime) { 438 rt->rt_flags |= RTF_PROBEMTU; 439 } else { 440 ap->nextstop = lmin(ap->nextstop, 441 rt->rt_rmx.rmx_expire); 442 } 443 } 444 445 return 0; 446 } 447 448 #define MTUTIMO_DEFAULT (60*1) 449 450 static void 451 in6_mtutimo(void *arg __unused) 452 { 453 int cpuid = mycpuid; 454 struct lwkt_msg *lmsg = &in6_mtutimo_ctx[cpuid].timo_nmsg.lmsg; 455 456 crit_enter(); 457 if (lmsg->ms_flags & MSGF_DONE) 458 lwkt_sendmsg_oncpu(netisr_cpuport(cpuid), lmsg); 459 crit_exit(); 460 } 461 462 static void 463 in6_mtutimo_dispatch(netmsg_t nmsg) 464 { 465 struct in6_rttimo_ctx *ctx = &in6_mtutimo_ctx[mycpuid]; 466 struct radix_node_head *rnh = ctx->timo_rnh; 467 struct mtuex_arg arg; 468 struct timeval atv; 469 470 /* Reply ASAP */ 471 crit_enter(); 472 lwkt_replymsg(&nmsg->lmsg, 0); 473 crit_exit(); 474 475 arg.rnh = rnh; 476 arg.nextstop = time_uptime + MTUTIMO_DEFAULT; 477 rnh->rnh_walktree(rnh, in6_mtuexpire, &arg); 478 479 atv.tv_usec = 0; 480 atv.tv_sec = arg.nextstop - time_uptime; 481 if ((int)atv.tv_sec < 1) { /* time shift safety */ 482 atv.tv_sec = 1; 483 arg.nextstop = time_uptime + atv.tv_sec; 484 } 485 if ((int)atv.tv_sec > rtq_timeout) { /* time shift safety */ 486 atv.tv_sec = rtq_timeout; 487 arg.nextstop = time_uptime + atv.tv_sec; 488 } 489 callout_reset(&ctx->timo_ch, tvtohz_high(&atv), in6_mtutimo, NULL); 490 } 491 492 #if 0 493 void 494 in6_rtqdrain(void) 495 { 496 struct radix_node_head *rnh = rt_tables[mycpuid][AF_INET6]; 497 struct rtqk_arg arg; 498 499 arg.found = arg.killed = 0; 500 arg.rnh = rnh; 501 arg.nextstop = 0; 502 arg.draining = 1; 503 arg.updating = 0; 504 crit_enter(); 505 rnh->rnh_walktree(rnh, in6_rtqkill, &arg); 506 crit_exit(); 507 } 508 #endif 509 510 /* 511 * Initialize our routing tree. 512 */ 513 int 514 in6_inithead(void **head, int off) 515 { 516 struct radix_node_head *rnh; 517 struct in6_rttimo_ctx *ctx; 518 int cpuid = mycpuid; 519 520 KKASSERT(head == (void **)&rt_tables[cpuid][AF_INET6]); 521 522 if (!rn_inithead(head, rn_cpumaskhead(cpuid), off)) 523 return 0; 524 525 rnh = *head; 526 rnh->rnh_addaddr = in6_addroute; 527 rnh->rnh_matchaddr = in6_matchroute; 528 rnh->rnh_close = in6_clsroute; 529 530 ctx = &in6_rtqtimo_ctx[cpuid]; 531 ctx->timo_rnh = rnh; 532 callout_init_mp(&ctx->timo_ch); 533 netmsg_init(&ctx->timo_nmsg, NULL, &netisr_adone_rport, MSGF_PRIORITY, 534 in6_rtqtimo_dispatch); 535 536 ctx = &in6_mtutimo_ctx[cpuid]; 537 ctx->timo_rnh = rnh; 538 callout_init_mp(&ctx->timo_ch); 539 netmsg_init(&ctx->timo_nmsg, NULL, &netisr_adone_rport, MSGF_PRIORITY, 540 in6_mtutimo_dispatch); 541 542 in6_rtqtimo(NULL); /* kick off timeout first time */ 543 in6_mtutimo(NULL); /* kick off timeout first time */ 544 545 return 1; 546 } 547