in6_rmx.c revision 181803
1/*- 2 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. Neither the name of the project nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * $KAME: in6_rmx.c,v 1.11 2001/07/26 06:53:16 jinmei Exp $ 30 */ 31 32/*- 33 * Copyright 1994, 1995 Massachusetts Institute of Technology 34 * 35 * Permission to use, copy, modify, and distribute this software and 36 * its documentation for any purpose and without fee is hereby 37 * granted, provided that both the above copyright notice and this 38 * permission notice appear in all copies, that both the above 39 * copyright notice and this permission notice appear in all 40 * supporting documentation, and that the name of M.I.T. not be used 41 * in advertising or publicity pertaining to distribution of the 42 * software without specific, written prior permission. M.I.T. makes 43 * no representations about the suitability of this software for any 44 * purpose. It is provided "as is" without express or implied 45 * warranty. 46 * 47 * THIS SOFTWARE IS PROVIDED BY M.I.T. ``AS IS''. M.I.T. DISCLAIMS 48 * ALL EXPRESS OR IMPLIED WARRANTIES WITH REGARD TO THIS SOFTWARE, 49 * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF 50 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. IN NO EVENT 51 * SHALL M.I.T. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 52 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 53 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF 54 * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND 55 * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 56 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 57 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 58 * SUCH DAMAGE. 59 * 60 */ 61 62/* 63 * This code does two things necessary for the enhanced TCP metrics to 64 * function in a useful manner: 65 * 1) It marks all non-host routes as `cloning', thus ensuring that 66 * every actual reference to such a route actually gets turned 67 * into a reference to a host route to the specific destination 68 * requested. 69 * 2) When such routes lose all their references, it arranges for them 70 * to be deleted in some random collection of circumstances, so that 71 * a large quantity of stale routing data is not kept in kernel memory 72 * indefinitely. See in6_rtqtimo() below for the exact mechanism. 73 */ 74 75#include <sys/cdefs.h> 76__FBSDID("$FreeBSD: head/sys/netinet6/in6_rmx.c 181803 2008-08-17 23:27:27Z bz $"); 77 78#include <sys/param.h> 79#include <sys/systm.h> 80#include <sys/kernel.h> 81#include <sys/sysctl.h> 82#include <sys/queue.h> 83#include <sys/socket.h> 84#include <sys/socketvar.h> 85#include <sys/mbuf.h> 86#include <sys/syslog.h> 87#include <sys/callout.h> 88#include <sys/vimage.h> 89 90#include <net/if.h> 91#include <net/route.h> 92#include <netinet/in.h> 93#include <netinet/ip_var.h> 94#include <netinet/in_var.h> 95 96#include <netinet/ip6.h> 97#include <netinet6/ip6_var.h> 98 99#include <netinet/icmp6.h> 100#include <netinet6/nd6.h> 101 102#include <netinet/tcp.h> 103#include <netinet/tcp_seq.h> 104#include <netinet/tcp_timer.h> 105#include <netinet/tcp_var.h> 106 107extern 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 */ 114static struct radix_node * 115in6_addroute(void *v_arg, void *n_arg, 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 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) 123 rt->rt_flags |= RTF_MULTICAST; 124 125 /* 126 * A little bit of help for both IPv6 output and input: 127 * For local addresses, we make sure that RTF_LOCAL is set, 128 * with the thought that this might one day be used to speed up 129 * ip_input(). 130 * 131 * We also mark routes to multicast addresses as such, because 132 * it's easy to do and might be useful (but this is much more 133 * dubious since it's so easy to inspect the address). (This 134 * is done above.) 135 * 136 * XXX 137 * should elaborate the code. 138 */ 139 if (rt->rt_flags & RTF_HOST) { 140 if (IN6_ARE_ADDR_EQUAL(&satosin6(rt->rt_ifa->ifa_addr) 141 ->sin6_addr, 142 &sin6->sin6_addr)) { 143 rt->rt_flags |= RTF_LOCAL; 144 } 145 } 146 147 if (!rt->rt_rmx.rmx_mtu && rt->rt_ifp) 148 rt->rt_rmx.rmx_mtu = IN6_LINKMTU(rt->rt_ifp); 149 150 ret = rn_addroute(v_arg, n_arg, head, treenodes); 151 if (ret == NULL && rt->rt_flags & RTF_HOST) { 152 struct rtentry *rt2; 153 /* 154 * We are trying to add a host route, but can't. 155 * Find out if it is because of an 156 * ARP entry and delete it if so. 157 */ 158 rt2 = rtalloc1((struct sockaddr *)sin6, 0, RTF_CLONING); 159 if (rt2) { 160 if (rt2->rt_flags & RTF_LLINFO && 161 rt2->rt_flags & RTF_HOST && 162 rt2->rt_gateway && 163 rt2->rt_gateway->sa_family == AF_LINK) { 164 rtexpunge(rt2); 165 RTFREE_LOCKED(rt2); 166 ret = rn_addroute(v_arg, n_arg, head, 167 treenodes); 168 } else 169 RTFREE_LOCKED(rt2); 170 } 171 } else if (ret == NULL && rt->rt_flags & RTF_CLONING) { 172 struct rtentry *rt2; 173 /* 174 * We are trying to add a net route, but can't. 175 * The following case should be allowed, so we'll make a 176 * special check for this: 177 * Two IPv6 addresses with the same prefix is assigned 178 * to a single interrface. 179 * # ifconfig if0 inet6 3ffe:0501::1 prefix 64 alias (*1) 180 * # ifconfig if0 inet6 3ffe:0501::2 prefix 64 alias (*2) 181 * In this case, (*1) and (*2) want to add the same 182 * net route entry, 3ffe:0501:: -> if0. 183 * This case should not raise an error. 184 */ 185 rt2 = rtalloc1((struct sockaddr *)sin6, 0, RTF_CLONING); 186 if (rt2) { 187 if ((rt2->rt_flags & (RTF_CLONING|RTF_HOST|RTF_GATEWAY)) 188 == RTF_CLONING 189 && rt2->rt_gateway 190 && rt2->rt_gateway->sa_family == AF_LINK 191 && rt2->rt_ifp == rt->rt_ifp) { 192 ret = rt2->rt_nodes; 193 } 194 RTFREE_LOCKED(rt2); 195 } 196 } 197 return ret; 198} 199 200/* 201 * This code is the inverse of in6_clsroute: on first reference, if we 202 * were managing the route, stop doing so and set the expiration timer 203 * back off again. 204 */ 205static struct radix_node * 206in6_matroute(void *v_arg, struct radix_node_head *head) 207{ 208 struct radix_node *rn = rn_match(v_arg, head); 209 struct rtentry *rt = (struct rtentry *)rn; 210 211 if (rt && rt->rt_refcnt == 0) { /* this is first reference */ 212 if (rt->rt_flags & RTPRF_OURS) { 213 rt->rt_flags &= ~RTPRF_OURS; 214 rt->rt_rmx.rmx_expire = 0; 215 } 216 } 217 return rn; 218} 219 220SYSCTL_DECL(_net_inet6_ip6); 221 222static int rtq_reallyold = 60*60; 223 /* one hour is ``really old'' */ 224SYSCTL_INT(_net_inet6_ip6, IPV6CTL_RTEXPIRE, rtexpire, 225 CTLFLAG_RW, &rtq_reallyold , 0, ""); 226 227static int rtq_minreallyold = 10; 228 /* never automatically crank down to less */ 229SYSCTL_INT(_net_inet6_ip6, IPV6CTL_RTMINEXPIRE, rtminexpire, 230 CTLFLAG_RW, &rtq_minreallyold , 0, ""); 231 232static int rtq_toomany = 128; 233 /* 128 cached routes is ``too many'' */ 234SYSCTL_INT(_net_inet6_ip6, IPV6CTL_RTMAXCACHE, rtmaxcache, 235 CTLFLAG_RW, &rtq_toomany , 0, ""); 236 237 238/* 239 * On last reference drop, mark the route as belong to us so that it can be 240 * timed out. 241 */ 242static void 243in6_clsroute(struct radix_node *rn, struct radix_node_head *head) 244{ 245 struct rtentry *rt = (struct rtentry *)rn; 246 247 RT_LOCK_ASSERT(rt); 248 249 if (!(rt->rt_flags & RTF_UP)) 250 return; /* prophylactic measures */ 251 252 if ((rt->rt_flags & (RTF_LLINFO | RTF_HOST)) != RTF_HOST) 253 return; 254 255 if ((rt->rt_flags & (RTF_WASCLONED | RTPRF_OURS)) != RTF_WASCLONED) 256 return; 257 258 /* 259 * As requested by David Greenman: 260 * If rtq_reallyold is 0, just delete the route without 261 * waiting for a timeout cycle to kill it. 262 */ 263 if (V_rtq_reallyold != 0) { 264 rt->rt_flags |= RTPRF_OURS; 265 rt->rt_rmx.rmx_expire = time_uptime + V_rtq_reallyold; 266 } else { 267 rtexpunge(rt); 268 } 269} 270 271struct rtqk_arg { 272 struct radix_node_head *rnh; 273 int mode; 274 int updating; 275 int draining; 276 int killed; 277 int found; 278 time_t nextstop; 279}; 280 281/* 282 * Get rid of old routes. When draining, this deletes everything, even when 283 * the timeout is not expired yet. When updating, this makes sure that 284 * nothing has a timeout longer than the current value of rtq_reallyold. 285 */ 286static int 287in6_rtqkill(struct radix_node *rn, void *rock) 288{ 289 struct rtqk_arg *ap = rock; 290 struct rtentry *rt = (struct rtentry *)rn; 291 int err; 292 293 if (rt->rt_flags & RTPRF_OURS) { 294 ap->found++; 295 296 if (ap->draining || rt->rt_rmx.rmx_expire <= time_uptime) { 297 if (rt->rt_refcnt > 0) 298 panic("rtqkill route really not free"); 299 300 err = rtrequest(RTM_DELETE, 301 (struct sockaddr *)rt_key(rt), 302 rt->rt_gateway, rt_mask(rt), 303 rt->rt_flags, 0); 304 if (err) { 305 log(LOG_WARNING, "in6_rtqkill: error %d", err); 306 } else { 307 ap->killed++; 308 } 309 } else { 310 if (ap->updating 311 && (rt->rt_rmx.rmx_expire - time_uptime 312 > V_rtq_reallyold)) { 313 rt->rt_rmx.rmx_expire = time_uptime 314 + V_rtq_reallyold; 315 } 316 ap->nextstop = lmin(ap->nextstop, 317 rt->rt_rmx.rmx_expire); 318 } 319 } 320 321 return 0; 322} 323 324#define RTQ_TIMEOUT 60*10 /* run no less than once every ten minutes */ 325static int rtq_timeout6 = RTQ_TIMEOUT; 326static struct callout rtq_timer6; 327 328static void 329in6_rtqtimo(void *rock) 330{ 331 struct radix_node_head *rnh = rock; 332 struct rtqk_arg arg; 333 struct timeval atv; 334 static time_t last_adjusted_timeout = 0; 335 336 arg.found = arg.killed = 0; 337 arg.rnh = rnh; 338 arg.nextstop = time_uptime + V_rtq_timeout6; 339 arg.draining = arg.updating = 0; 340 RADIX_NODE_HEAD_LOCK(rnh); 341 rnh->rnh_walktree(rnh, in6_rtqkill, &arg); 342 RADIX_NODE_HEAD_UNLOCK(rnh); 343 344 /* 345 * Attempt to be somewhat dynamic about this: 346 * If there are ``too many'' routes sitting around taking up space, 347 * then crank down the timeout, and see if we can't make some more 348 * go away. However, we make sure that we will never adjust more 349 * than once in rtq_timeout6 seconds, to keep from cranking down too 350 * hard. 351 */ 352 if ((arg.found - arg.killed > V_rtq_toomany) 353 && (time_uptime - last_adjusted_timeout >= V_rtq_timeout6) 354 && V_rtq_reallyold > V_rtq_minreallyold) { 355 V_rtq_reallyold = 2*V_rtq_reallyold / 3; 356 if (V_rtq_reallyold < V_rtq_minreallyold) { 357 V_rtq_reallyold = V_rtq_minreallyold; 358 } 359 360 last_adjusted_timeout = time_uptime; 361#ifdef DIAGNOSTIC 362 log(LOG_DEBUG, "in6_rtqtimo: adjusted rtq_reallyold to %d", 363 V_rtq_reallyold); 364#endif 365 arg.found = arg.killed = 0; 366 arg.updating = 1; 367 RADIX_NODE_HEAD_LOCK(rnh); 368 rnh->rnh_walktree(rnh, in6_rtqkill, &arg); 369 RADIX_NODE_HEAD_UNLOCK(rnh); 370 } 371 372 atv.tv_usec = 0; 373 atv.tv_sec = arg.nextstop - time_uptime; 374 callout_reset(&V_rtq_timer6, tvtohz(&atv), in6_rtqtimo, rock); 375} 376 377/* 378 * Age old PMTUs. 379 */ 380struct mtuex_arg { 381 struct radix_node_head *rnh; 382 time_t nextstop; 383}; 384static struct callout rtq_mtutimer; 385 386static int 387in6_mtuexpire(struct radix_node *rn, void *rock) 388{ 389 struct rtentry *rt = (struct rtentry *)rn; 390 struct mtuex_arg *ap = rock; 391 392 /* sanity */ 393 if (!rt) 394 panic("rt == NULL in in6_mtuexpire"); 395 396 if (rt->rt_rmx.rmx_expire && !(rt->rt_flags & RTF_PROBEMTU)) { 397 if (rt->rt_rmx.rmx_expire <= time_uptime) { 398 rt->rt_flags |= RTF_PROBEMTU; 399 } else { 400 ap->nextstop = lmin(ap->nextstop, 401 rt->rt_rmx.rmx_expire); 402 } 403 } 404 405 return 0; 406} 407 408#define MTUTIMO_DEFAULT (60*1) 409 410static void 411in6_mtutimo(void *rock) 412{ 413 struct radix_node_head *rnh = rock; 414 struct mtuex_arg arg; 415 struct timeval atv; 416 417 arg.rnh = rnh; 418 arg.nextstop = time_uptime + MTUTIMO_DEFAULT; 419 RADIX_NODE_HEAD_LOCK(rnh); 420 rnh->rnh_walktree(rnh, in6_mtuexpire, &arg); 421 RADIX_NODE_HEAD_UNLOCK(rnh); 422 423 atv.tv_usec = 0; 424 atv.tv_sec = arg.nextstop - time_uptime; 425 if (atv.tv_sec < 0) { 426 printf("invalid mtu expiration time on routing table\n"); 427 arg.nextstop = time_uptime + 30; /* last resort */ 428 atv.tv_sec = 30; 429 } 430 callout_reset(&V_rtq_mtutimer, tvtohz(&atv), in6_mtutimo, rock); 431} 432 433#if 0 434void 435in6_rtqdrain(void) 436{ 437 struct radix_node_head *rnh = V_rt_tables[AF_INET6]; 438 struct rtqk_arg arg; 439 440 arg.found = arg.killed = 0; 441 arg.rnh = rnh; 442 arg.nextstop = 0; 443 arg.draining = 1; 444 arg.updating = 0; 445 RADIX_NODE_HEAD_LOCK(rnh); 446 rnh->rnh_walktree(rnh, in6_rtqkill, &arg); 447 RADIX_NODE_HEAD_UNLOCK(rnh); 448} 449#endif 450 451/* 452 * Initialize our routing tree. 453 * XXX MRT When off == 0, we are being called from vfs_export.c 454 * so just set up their table and leave. (we know what the correct 455 * value should be so just use that).. FIX AFTER RELENG_7 is MFC'd 456 * see also comments in in_inithead() vfs_export.c and domain.h 457 */ 458int 459in6_inithead(void **head, int off) 460{ 461 struct radix_node_head *rnh; 462 463 if (!rn_inithead(head, offsetof(struct sockaddr_in6, sin6_addr) << 3)) 464 return 0; /* See above */ 465 466 if (off == 0) /* See above */ 467 return 1; /* only do the rest for the real thing */ 468 469 rnh = *head; 470 rnh->rnh_addaddr = in6_addroute; 471 rnh->rnh_matchaddr = in6_matroute; 472 rnh->rnh_close = in6_clsroute; 473 callout_init(&V_rtq_timer6, CALLOUT_MPSAFE); 474 in6_rtqtimo(rnh); /* kick off timeout first time */ 475 callout_init(&V_rtq_mtutimer, CALLOUT_MPSAFE); 476 in6_mtutimo(rnh); /* kick off timeout first time */ 477 return 1; 478} 479