1 /* $FreeBSD$ */ 2 /* $KAME: in6_rmx.c,v 1.7 2000/04/06 08:30:43 sumikawa Exp $ */ 3 4 /* 5 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 6 * All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. Neither the name of the project nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 */ 32 33 /* 34 * Copyright 1994, 1995 Massachusetts Institute of Technology 35 * 36 * Permission to use, copy, modify, and distribute this software and 37 * its documentation for any purpose and without fee is hereby 38 * granted, provided that both the above copyright notice and this 39 * permission notice appear in all copies, that both the above 40 * copyright notice and this permission notice appear in all 41 * supporting documentation, and that the name of M.I.T. not be used 42 * in advertising or publicity pertaining to distribution of the 43 * software without specific, written prior permission. M.I.T. makes 44 * no representations about the suitability of this software for any 45 * purpose. It is provided "as is" without express or implied 46 * warranty. 47 * 48 * THIS SOFTWARE IS PROVIDED BY M.I.T. ``AS IS''. M.I.T. DISCLAIMS 49 * ALL EXPRESS OR IMPLIED WARRANTIES WITH REGARD TO THIS SOFTWARE, 50 * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF 51 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. IN NO EVENT 52 * SHALL M.I.T. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 53 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 54 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF 55 * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND 56 * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 57 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 58 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 59 * SUCH DAMAGE. 60 * 61 */ 62 63 /* 64 * This code does two things necessary for the enhanced TCP metrics to 65 * function in a useful manner: 66 * 1) It marks all non-host routes as `cloning', thus ensuring that 67 * every actual reference to such a route actually gets turned 68 * into a reference to a host route to the specific destination 69 * requested. 70 * 2) When such routes lose all their references, it arranges for them 71 * to be deleted in some random collection of circumstances, so that 72 * a large quantity of stale routing data is not kept in kernel memory 73 * indefinitely. See in6_rtqtimo() below for the exact mechanism. 74 */ 75 76 #include <sys/param.h> 77 #include <sys/systm.h> 78 #include <sys/kernel.h> 79 #include <sys/sysctl.h> 80 #include <sys/queue.h> 81 #include <sys/socket.h> 82 #include <sys/socketvar.h> 83 #include <sys/mbuf.h> 84 #include <sys/syslog.h> 85 86 #include <net/if.h> 87 #include <net/route.h> 88 #include <netinet/in.h> 89 #include <netinet/ip_var.h> 90 #include <netinet/in_var.h> 91 92 #include <netinet/ip6.h> 93 #include <netinet6/ip6_var.h> 94 95 #include <netinet/icmp6.h> 96 97 #include <netinet/tcp.h> 98 #include <netinet/tcp_seq.h> 99 #include <netinet/tcp_timer.h> 100 #include <netinet/tcp_var.h> 101 102 extern int in6_inithead __P((void **head, int off)); 103 104 #define RTPRF_OURS RTF_PROTO3 /* set on routes we manage */ 105 106 /* 107 * Do what we need to do when inserting a route. 108 */ 109 static struct radix_node * 110 in6_addroute(void *v_arg, void *n_arg, struct radix_node_head *head, 111 struct radix_node *treenodes) 112 { 113 struct rtentry *rt = (struct rtentry *)treenodes; 114 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)rt_key(rt); 115 struct radix_node *ret; 116 117 /* 118 * For IPv6, all unicast non-host routes are automatically cloning. 119 */ 120 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) 121 rt->rt_flags |= RTF_MULTICAST; 122 123 if (!(rt->rt_flags & (RTF_HOST | RTF_CLONING | RTF_MULTICAST))) { 124 rt->rt_flags |= RTF_PRCLONING; 125 } 126 127 /* 128 * A little bit of help for both IPv6 output and input: 129 * For local addresses, we make sure that RTF_LOCAL is set, 130 * with the thought that this might one day be used to speed up 131 * ip_input(). 132 * 133 * We also mark routes to multicast addresses as such, because 134 * it's easy to do and might be useful (but this is much more 135 * dubious since it's so easy to inspect the address). (This 136 * is done above.) 137 * 138 * XXX 139 * should elaborate the code. 140 */ 141 if (rt->rt_flags & RTF_HOST) { 142 if (IN6_ARE_ADDR_EQUAL(&satosin6(rt->rt_ifa->ifa_addr) 143 ->sin6_addr, 144 &sin6->sin6_addr)) { 145 rt->rt_flags |= RTF_LOCAL; 146 } 147 } 148 149 /* 150 * We also specify a send and receive pipe size for every 151 * route added, to help TCP a bit. TCP doesn't actually 152 * want a true pipe size, which would be prohibitive in memory 153 * costs and is hard to compute anyway; it simply uses these 154 * values to size its buffers. So, we fill them in with the 155 * same values that TCP would have used anyway, and allow the 156 * installing program or the link layer to override these values 157 * as it sees fit. This will hopefully allow TCP more 158 * opportunities to save its ssthresh value. 159 */ 160 if (!rt->rt_rmx.rmx_sendpipe && !(rt->rt_rmx.rmx_locks & RTV_SPIPE)) 161 rt->rt_rmx.rmx_sendpipe = tcp_sendspace; 162 163 if (!rt->rt_rmx.rmx_recvpipe && !(rt->rt_rmx.rmx_locks & RTV_RPIPE)) 164 rt->rt_rmx.rmx_recvpipe = tcp_recvspace; 165 166 if (!rt->rt_rmx.rmx_mtu && !(rt->rt_rmx.rmx_locks & RTV_MTU) 167 && rt->rt_ifp) 168 rt->rt_rmx.rmx_mtu = rt->rt_ifp->if_mtu; 169 170 ret = rn_addroute(v_arg, n_arg, head, treenodes); 171 if (ret == NULL && rt->rt_flags & RTF_HOST) { 172 struct rtentry *rt2; 173 /* 174 * We are trying to add a host route, but can't. 175 * Find out if it is because of an 176 * ARP entry and delete it if so. 177 */ 178 rt2 = rtalloc1((struct sockaddr *)sin6, 0, 179 RTF_CLONING | RTF_PRCLONING); 180 if (rt2) { 181 if (rt2->rt_flags & RTF_LLINFO && 182 rt2->rt_flags & RTF_HOST && 183 rt2->rt_gateway && 184 rt2->rt_gateway->sa_family == AF_LINK) { 185 rtrequest(RTM_DELETE, 186 (struct sockaddr *)rt_key(rt2), 187 rt2->rt_gateway, 188 rt_mask(rt2), rt2->rt_flags, 0); 189 ret = rn_addroute(v_arg, n_arg, head, 190 treenodes); 191 } 192 RTFREE(rt2); 193 } 194 } else if (ret == NULL && rt->rt_flags & RTF_CLONING) { 195 struct rtentry *rt2; 196 /* 197 * We are trying to add a net route, but can't. 198 * The following case should be allowed, so we'll make a 199 * special check for this: 200 * Two IPv6 addresses with the same prefix is assigned 201 * to a single interrface. 202 * # ifconfig if0 inet6 3ffe:0501::1 prefix 64 alias (*1) 203 * # ifconfig if0 inet6 3ffe:0501::2 prefix 64 alias (*2) 204 * In this case, (*1) and (*2) want to add the same 205 * net route entry, 3ffe:0501:: -> if0. 206 * This case should not raise an error. 207 */ 208 rt2 = rtalloc1((struct sockaddr *)sin6, 0, 209 RTF_CLONING | RTF_PRCLONING); 210 if (rt2) { 211 if ((rt2->rt_flags & (RTF_CLONING|RTF_HOST|RTF_GATEWAY)) 212 == RTF_CLONING 213 && rt2->rt_gateway 214 && rt2->rt_gateway->sa_family == AF_LINK 215 && rt2->rt_ifp == rt->rt_ifp) { 216 ret = rt2->rt_nodes; 217 } 218 RTFREE(rt2); 219 } 220 } 221 return ret; 222 } 223 224 /* 225 * This code is the inverse of in6_clsroute: on first reference, if we 226 * were managing the route, stop doing so and set the expiration timer 227 * back off again. 228 */ 229 static struct radix_node * 230 in6_matroute(void *v_arg, struct radix_node_head *head) 231 { 232 struct radix_node *rn = rn_match(v_arg, head); 233 struct rtentry *rt = (struct rtentry *)rn; 234 235 if (rt && rt->rt_refcnt == 0) { /* this is first reference */ 236 if (rt->rt_flags & RTPRF_OURS) { 237 rt->rt_flags &= ~RTPRF_OURS; 238 rt->rt_rmx.rmx_expire = 0; 239 } 240 } 241 return rn; 242 } 243 244 SYSCTL_DECL(_net_inet6_ip6); 245 246 static int rtq_reallyold = 60*60; 247 /* one hour is ``really old'' */ 248 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_RTEXPIRE, rtexpire, 249 CTLFLAG_RW, &rtq_reallyold , 0, ""); 250 251 static int rtq_minreallyold = 10; 252 /* never automatically crank down to less */ 253 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_RTMINEXPIRE, rtminexpire, 254 CTLFLAG_RW, &rtq_minreallyold , 0, ""); 255 256 static int rtq_toomany = 128; 257 /* 128 cached routes is ``too many'' */ 258 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_RTMAXCACHE, rtmaxcache, 259 CTLFLAG_RW, &rtq_toomany , 0, ""); 260 261 262 /* 263 * On last reference drop, mark the route as belong to us so that it can be 264 * timed out. 265 */ 266 static void 267 in6_clsroute(struct radix_node *rn, struct radix_node_head *head) 268 { 269 struct rtentry *rt = (struct rtentry *)rn; 270 271 if (!(rt->rt_flags & RTF_UP)) 272 return; /* prophylactic measures */ 273 274 if ((rt->rt_flags & (RTF_LLINFO | RTF_HOST)) != RTF_HOST) 275 return; 276 277 if ((rt->rt_flags & (RTF_WASCLONED | RTPRF_OURS)) 278 != RTF_WASCLONED) 279 return; 280 281 /* 282 * As requested by David Greenman: 283 * If rtq_reallyold is 0, just delete the route without 284 * waiting for a timeout cycle to kill it. 285 */ 286 if (rtq_reallyold != 0) { 287 rt->rt_flags |= RTPRF_OURS; 288 rt->rt_rmx.rmx_expire = time_second + rtq_reallyold; 289 } else { 290 rtrequest(RTM_DELETE, 291 (struct sockaddr *)rt_key(rt), 292 rt->rt_gateway, rt_mask(rt), 293 rt->rt_flags, 0); 294 } 295 } 296 297 struct rtqk_arg { 298 struct radix_node_head *rnh; 299 int mode; 300 int updating; 301 int draining; 302 int killed; 303 int found; 304 time_t nextstop; 305 }; 306 307 /* 308 * Get rid of old routes. When draining, this deletes everything, even when 309 * the timeout is not expired yet. When updating, this makes sure that 310 * nothing has a timeout longer than the current value of rtq_reallyold. 311 */ 312 static int 313 in6_rtqkill(struct radix_node *rn, void *rock) 314 { 315 struct rtqk_arg *ap = rock; 316 struct rtentry *rt = (struct rtentry *)rn; 317 int err; 318 319 if (rt->rt_flags & RTPRF_OURS) { 320 ap->found++; 321 322 if (ap->draining || rt->rt_rmx.rmx_expire <= time_second) { 323 if (rt->rt_refcnt > 0) 324 panic("rtqkill route really not free"); 325 326 err = rtrequest(RTM_DELETE, 327 (struct sockaddr *)rt_key(rt), 328 rt->rt_gateway, rt_mask(rt), 329 rt->rt_flags, 0); 330 if (err) { 331 log(LOG_WARNING, "in6_rtqkill: error %d", err); 332 } else { 333 ap->killed++; 334 } 335 } else { 336 if (ap->updating 337 && (rt->rt_rmx.rmx_expire - time_second 338 > rtq_reallyold)) { 339 rt->rt_rmx.rmx_expire = time_second 340 + rtq_reallyold; 341 } 342 ap->nextstop = lmin(ap->nextstop, 343 rt->rt_rmx.rmx_expire); 344 } 345 } 346 347 return 0; 348 } 349 350 #define RTQ_TIMEOUT 60*10 /* run no less than once every ten minutes */ 351 static int rtq_timeout = RTQ_TIMEOUT; 352 353 static void 354 in6_rtqtimo(void *rock) 355 { 356 struct radix_node_head *rnh = rock; 357 struct rtqk_arg arg; 358 struct timeval atv; 359 static time_t last_adjusted_timeout = 0; 360 int s; 361 362 arg.found = arg.killed = 0; 363 arg.rnh = rnh; 364 arg.nextstop = time_second + rtq_timeout; 365 arg.draining = arg.updating = 0; 366 s = splnet(); 367 rnh->rnh_walktree(rnh, in6_rtqkill, &arg); 368 splx(s); 369 370 /* 371 * Attempt to be somewhat dynamic about this: 372 * If there are ``too many'' routes sitting around taking up space, 373 * then crank down the timeout, and see if we can't make some more 374 * go away. However, we make sure that we will never adjust more 375 * than once in rtq_timeout seconds, to keep from cranking down too 376 * hard. 377 */ 378 if ((arg.found - arg.killed > rtq_toomany) 379 && (time_second - last_adjusted_timeout >= rtq_timeout) 380 && rtq_reallyold > rtq_minreallyold) { 381 rtq_reallyold = 2*rtq_reallyold / 3; 382 if (rtq_reallyold < rtq_minreallyold) { 383 rtq_reallyold = rtq_minreallyold; 384 } 385 386 last_adjusted_timeout = time_second; 387 #ifdef DIAGNOSTIC 388 log(LOG_DEBUG, "in6_rtqtimo: adjusted rtq_reallyold to %d", 389 rtq_reallyold); 390 #endif 391 arg.found = arg.killed = 0; 392 arg.updating = 1; 393 s = splnet(); 394 rnh->rnh_walktree(rnh, in6_rtqkill, &arg); 395 splx(s); 396 } 397 398 atv.tv_usec = 0; 399 atv.tv_sec = arg.nextstop; 400 timeout(in6_rtqtimo, rock, tvtohz(&atv)); 401 } 402 403 /* 404 * Age old PMTUs. 405 */ 406 struct mtuex_arg { 407 struct radix_node_head *rnh; 408 time_t nextstop; 409 }; 410 411 static int 412 in6_mtuexpire(struct radix_node *rn, void *rock) 413 { 414 struct rtentry *rt = (struct rtentry *)rn; 415 struct mtuex_arg *ap = rock; 416 417 /* sanity */ 418 if (!rt) 419 panic("rt == NULL in in6_mtuexpire"); 420 421 if (rt->rt_rmx.rmx_expire && !(rt->rt_flags & RTF_PROBEMTU)) { 422 if (rt->rt_rmx.rmx_expire <= time_second) { 423 rt->rt_flags |= RTF_PROBEMTU; 424 } else { 425 ap->nextstop = lmin(ap->nextstop, 426 rt->rt_rmx.rmx_expire); 427 } 428 } 429 430 return 0; 431 } 432 433 #define MTUTIMO_DEFAULT (60*1) 434 435 static void 436 in6_mtutimo(void *rock) 437 { 438 struct radix_node_head *rnh = rock; 439 struct mtuex_arg arg; 440 struct timeval atv; 441 int s; 442 443 arg.rnh = rnh; 444 arg.nextstop = time_second + MTUTIMO_DEFAULT; 445 s = splnet(); 446 rnh->rnh_walktree(rnh, in6_mtuexpire, &arg); 447 splx(s); 448 449 atv.tv_usec = 0; 450 atv.tv_sec = arg.nextstop; 451 if (atv.tv_sec < time_second) { 452 printf("invalid mtu expiration time on routing table\n"); 453 arg.nextstop = time_second + 30; /*last resort*/ 454 } 455 timeout(in6_mtutimo, rock, tvtohz(&atv)); 456 } 457 458 #if 0 459 void 460 in6_rtqdrain() 461 { 462 struct radix_node_head *rnh = rt_tables[AF_INET6]; 463 struct rtqk_arg arg; 464 int s; 465 arg.found = arg.killed = 0; 466 arg.rnh = rnh; 467 arg.nextstop = 0; 468 arg.draining = 1; 469 arg.updating = 0; 470 s = splnet(); 471 rnh->rnh_walktree(rnh, in6_rtqkill, &arg); 472 splx(s); 473 } 474 #endif 475 476 /* 477 * Initialize our routing tree. 478 */ 479 int 480 in6_inithead(void **head, int off) 481 { 482 struct radix_node_head *rnh; 483 484 if (!rn_inithead(head, off)) 485 return 0; 486 487 if (head != (void **)&rt_tables[AF_INET6]) /* BOGUS! */ 488 return 1; /* only do this for the real routing table */ 489 490 rnh = *head; 491 rnh->rnh_addaddr = in6_addroute; 492 rnh->rnh_matchaddr = in6_matroute; 493 rnh->rnh_close = in6_clsroute; 494 in6_rtqtimo(rnh); /* kick off timeout first time */ 495 in6_mtutimo(rnh); /* kick off timeout first time */ 496 return 1; 497 } 498