in6_rmx.c revision 193731
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 193731 2009-06-08 17:15:40Z zec $"); 77 78#include "opt_route.h" 79 80#include <sys/param.h> 81#include <sys/systm.h> 82#include <sys/kernel.h> 83#include <sys/lock.h> 84#include <sys/sysctl.h> 85#include <sys/queue.h> 86#include <sys/socket.h> 87#include <sys/socketvar.h> 88#include <sys/mbuf.h> 89#include <sys/rwlock.h> 90#include <sys/syslog.h> 91#include <sys/callout.h> 92#include <sys/vimage.h> 93 94#include <net/if.h> 95#include <net/route.h> 96#include <net/vnet.h> 97 98#include <netinet/in.h> 99#include <netinet/ip_var.h> 100#include <netinet/in_var.h> 101 102#include <netinet/ip6.h> 103#include <netinet6/ip6_var.h> 104 105#include <netinet/icmp6.h> 106#include <netinet6/nd6.h> 107#include <netinet6/vinet6.h> 108 109#include <netinet/tcp.h> 110#include <netinet/tcp_seq.h> 111#include <netinet/tcp_timer.h> 112#include <netinet/tcp_var.h> 113 114extern int in6_inithead(void **head, int off); 115#ifdef VIMAGE 116extern int in6_detachhead(void **head, int off); 117#endif 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 */ 124static struct radix_node * 125in6_addroute(void *v_arg, void *n_arg, 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 RADIX_NODE_HEAD_WLOCK_ASSERT(head); 133 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) 134 rt->rt_flags |= RTF_MULTICAST; 135 136 /* 137 * A little bit of help for both IPv6 output and input: 138 * For local addresses, we make sure that RTF_LOCAL is set, 139 * with the thought that this might one day be used to speed up 140 * ip_input(). 141 * 142 * We also mark routes to multicast addresses as such, because 143 * it's easy to do and might be useful (but this is much more 144 * dubious since it's so easy to inspect the address). (This 145 * is done above.) 146 * 147 * XXX 148 * should elaborate the code. 149 */ 150 if (rt->rt_flags & RTF_HOST) { 151 if (IN6_ARE_ADDR_EQUAL(&satosin6(rt->rt_ifa->ifa_addr) 152 ->sin6_addr, 153 &sin6->sin6_addr)) { 154 rt->rt_flags |= RTF_LOCAL; 155 } 156 } 157 158 if (!rt->rt_rmx.rmx_mtu && rt->rt_ifp) 159 rt->rt_rmx.rmx_mtu = IN6_LINKMTU(rt->rt_ifp); 160 161 ret = rn_addroute(v_arg, n_arg, head, treenodes); 162 if (ret == NULL) { 163 struct rtentry *rt2; 164 /* 165 * We are trying to add a net route, but can't. 166 * The following case should be allowed, so we'll make a 167 * special check for this: 168 * Two IPv6 addresses with the same prefix is assigned 169 * to a single interrface. 170 * # ifconfig if0 inet6 3ffe:0501::1 prefix 64 alias (*1) 171 * # ifconfig if0 inet6 3ffe:0501::2 prefix 64 alias (*2) 172 * In this case, (*1) and (*2) want to add the same 173 * net route entry, 3ffe:0501:: -> if0. 174 * This case should not raise an error. 175 */ 176 rt2 = rtalloc1((struct sockaddr *)sin6, 0, RTF_RNH_LOCKED); 177 if (rt2) { 178 if (((rt2->rt_flags & (RTF_HOST|RTF_GATEWAY)) == 0) 179 && rt2->rt_gateway 180 && rt2->rt_gateway->sa_family == AF_LINK 181 && rt2->rt_ifp == rt->rt_ifp) { 182 ret = rt2->rt_nodes; 183 } 184 RTFREE_LOCKED(rt2); 185 } 186 } 187 return (ret); 188} 189 190/* 191 * This code is the inverse of in6_clsroute: on first reference, if we 192 * were managing the route, stop doing so and set the expiration timer 193 * back off again. 194 */ 195static struct radix_node * 196in6_matroute(void *v_arg, struct radix_node_head *head) 197{ 198 struct radix_node *rn = rn_match(v_arg, head); 199 struct rtentry *rt = (struct rtentry *)rn; 200 201 if (rt && rt->rt_refcnt == 0) { /* this is first reference */ 202 if (rt->rt_flags & RTPRF_OURS) { 203 rt->rt_flags &= ~RTPRF_OURS; 204 rt->rt_rmx.rmx_expire = 0; 205 } 206 } 207 return rn; 208} 209 210SYSCTL_DECL(_net_inet6_ip6); 211 212#ifdef VIMAGE_GLOBALS 213static int rtq_reallyold6; 214static int rtq_minreallyold6; 215static int rtq_toomany6; 216#endif 217 218SYSCTL_V_INT(V_NET, vnet_inet6, _net_inet6_ip6, IPV6CTL_RTEXPIRE, 219 rtexpire, CTLFLAG_RW, rtq_reallyold6 , 0, ""); 220 221SYSCTL_V_INT(V_NET, vnet_inet6, _net_inet6_ip6, IPV6CTL_RTMINEXPIRE, 222 rtminexpire, CTLFLAG_RW, rtq_minreallyold6 , 0, ""); 223 224SYSCTL_V_INT(V_NET, vnet_inet6, _net_inet6_ip6, IPV6CTL_RTMAXCACHE, 225 rtmaxcache, CTLFLAG_RW, rtq_toomany6 , 0, ""); 226 227 228 229struct rtqk_arg { 230 struct radix_node_head *rnh; 231 int mode; 232 int updating; 233 int draining; 234 int killed; 235 int found; 236 time_t nextstop; 237}; 238 239/* 240 * Get rid of old routes. When draining, this deletes everything, even when 241 * the timeout is not expired yet. When updating, this makes sure that 242 * nothing has a timeout longer than the current value of rtq_reallyold6. 243 */ 244static int 245in6_rtqkill(struct radix_node *rn, void *rock) 246{ 247 INIT_VNET_INET6(curvnet); 248 struct rtqk_arg *ap = rock; 249 struct rtentry *rt = (struct rtentry *)rn; 250 int err; 251 252 RADIX_NODE_HEAD_WLOCK_ASSERT(ap->rnh); 253 254 if (rt->rt_flags & RTPRF_OURS) { 255 ap->found++; 256 257 if (ap->draining || rt->rt_rmx.rmx_expire <= time_uptime) { 258 if (rt->rt_refcnt > 0) 259 panic("rtqkill route really not free"); 260 261 err = rtrequest(RTM_DELETE, 262 (struct sockaddr *)rt_key(rt), 263 rt->rt_gateway, rt_mask(rt), 264 rt->rt_flags|RTF_RNH_LOCKED, 0); 265 if (err) { 266 log(LOG_WARNING, "in6_rtqkill: error %d", err); 267 } else { 268 ap->killed++; 269 } 270 } else { 271 if (ap->updating 272 && (rt->rt_rmx.rmx_expire - time_uptime 273 > V_rtq_reallyold6)) { 274 rt->rt_rmx.rmx_expire = time_uptime 275 + V_rtq_reallyold6; 276 } 277 ap->nextstop = lmin(ap->nextstop, 278 rt->rt_rmx.rmx_expire); 279 } 280 } 281 282 return 0; 283} 284 285#define RTQ_TIMEOUT 60*10 /* run no less than once every ten minutes */ 286#ifdef VIMAGE_GLOBALS 287static int rtq_timeout6; 288static struct callout rtq_timer6; 289#endif 290 291static void 292in6_rtqtimo(void *rock) 293{ 294 CURVNET_SET_QUIET((struct vnet *) rock); 295 INIT_VNET_INET6(curvnet); 296 struct radix_node_head *rnh; 297 struct rtqk_arg arg; 298 struct timeval atv; 299 static time_t last_adjusted_timeout = 0; 300 301 rnh = rt_tables_get_rnh(0, AF_INET6); 302 if (rnh == NULL) { 303 CURVNET_RESTORE(); 304 return; 305 } 306 arg.found = arg.killed = 0; 307 arg.rnh = rnh; 308 arg.nextstop = time_uptime + V_rtq_timeout6; 309 arg.draining = arg.updating = 0; 310 RADIX_NODE_HEAD_LOCK(rnh); 311 rnh->rnh_walktree(rnh, in6_rtqkill, &arg); 312 RADIX_NODE_HEAD_UNLOCK(rnh); 313 314 /* 315 * Attempt to be somewhat dynamic about this: 316 * If there are ``too many'' routes sitting around taking up space, 317 * then crank down the timeout, and see if we can't make some more 318 * go away. However, we make sure that we will never adjust more 319 * than once in rtq_timeout6 seconds, to keep from cranking down too 320 * hard. 321 */ 322 if ((arg.found - arg.killed > V_rtq_toomany6) 323 && (time_uptime - last_adjusted_timeout >= V_rtq_timeout6) 324 && V_rtq_reallyold6 > V_rtq_minreallyold6) { 325 V_rtq_reallyold6 = 2*V_rtq_reallyold6 / 3; 326 if (V_rtq_reallyold6 < V_rtq_minreallyold6) { 327 V_rtq_reallyold6 = V_rtq_minreallyold6; 328 } 329 330 last_adjusted_timeout = time_uptime; 331#ifdef DIAGNOSTIC 332 log(LOG_DEBUG, "in6_rtqtimo: adjusted rtq_reallyold6 to %d", 333 V_rtq_reallyold6); 334#endif 335 arg.found = arg.killed = 0; 336 arg.updating = 1; 337 RADIX_NODE_HEAD_LOCK(rnh); 338 rnh->rnh_walktree(rnh, in6_rtqkill, &arg); 339 RADIX_NODE_HEAD_UNLOCK(rnh); 340 } 341 342 atv.tv_usec = 0; 343 atv.tv_sec = arg.nextstop - time_uptime; 344 callout_reset(&V_rtq_timer6, tvtohz(&atv), in6_rtqtimo, rock); 345 CURVNET_RESTORE(); 346} 347 348/* 349 * Age old PMTUs. 350 */ 351struct mtuex_arg { 352 struct radix_node_head *rnh; 353 time_t nextstop; 354}; 355#ifdef VIMAGE_GLOBALS 356static struct callout rtq_mtutimer; 357#endif 358 359static int 360in6_mtuexpire(struct radix_node *rn, void *rock) 361{ 362 struct rtentry *rt = (struct rtentry *)rn; 363 struct mtuex_arg *ap = rock; 364 365 /* sanity */ 366 if (!rt) 367 panic("rt == NULL in in6_mtuexpire"); 368 369 if (rt->rt_rmx.rmx_expire && !(rt->rt_flags & RTF_PROBEMTU)) { 370 if (rt->rt_rmx.rmx_expire <= time_uptime) { 371 rt->rt_flags |= RTF_PROBEMTU; 372 } else { 373 ap->nextstop = lmin(ap->nextstop, 374 rt->rt_rmx.rmx_expire); 375 } 376 } 377 378 return 0; 379} 380 381#define MTUTIMO_DEFAULT (60*1) 382 383static void 384in6_mtutimo(void *rock) 385{ 386 CURVNET_SET_QUIET((struct vnet *) rock); 387 INIT_VNET_INET6(curvnet); 388 struct radix_node_head *rnh; 389 struct mtuex_arg arg; 390 struct timeval atv; 391 392 rnh = rt_tables_get_rnh(0, AF_INET6); 393 if (rnh == NULL) { 394 CURVNET_RESTORE(); 395 return; 396 } 397 arg.rnh = rnh; 398 arg.nextstop = time_uptime + MTUTIMO_DEFAULT; 399 RADIX_NODE_HEAD_LOCK(rnh); 400 rnh->rnh_walktree(rnh, in6_mtuexpire, &arg); 401 RADIX_NODE_HEAD_UNLOCK(rnh); 402 403 atv.tv_usec = 0; 404 atv.tv_sec = arg.nextstop - time_uptime; 405 if (atv.tv_sec < 0) { 406 printf("invalid mtu expiration time on routing table\n"); 407 arg.nextstop = time_uptime + 30; /* last resort */ 408 atv.tv_sec = 30; 409 } 410 callout_reset(&V_rtq_mtutimer, tvtohz(&atv), in6_mtutimo, rock); 411 CURVNET_RESTORE(); 412} 413 414#if 0 415void 416in6_rtqdrain(void) 417{ 418 INIT_VNET_NET(curvnet); 419 struct radix_node_head *rnh; 420 struct rtqk_arg arg; 421 422 rnh = rt_tables_get_rnh(0, AF_INET6); 423 if (rnh == NULL) 424 panic("%s: rnh == NULL", __func__); 425 arg.found = arg.killed = 0; 426 arg.rnh = rnh; 427 arg.nextstop = 0; 428 arg.draining = 1; 429 arg.updating = 0; 430 RADIX_NODE_HEAD_LOCK(rnh); 431 rnh->rnh_walktree(rnh, in6_rtqkill, &arg); 432 RADIX_NODE_HEAD_UNLOCK(rnh); 433} 434#endif 435 436/* 437 * Initialize our routing tree. 438 * XXX MRT When off == 0, we are being called from vfs_export.c 439 * so just set up their table and leave. (we know what the correct 440 * value should be so just use that).. FIX AFTER RELENG_7 is MFC'd 441 * see also comments in in_inithead() vfs_export.c and domain.h 442 */ 443int 444in6_inithead(void **head, int off) 445{ 446 INIT_VNET_INET6(curvnet); 447 struct radix_node_head *rnh; 448 449 if (!rn_inithead(head, offsetof(struct sockaddr_in6, sin6_addr) << 3)) 450 return 0; /* See above */ 451 452 if (off == 0) /* See above */ 453 return 1; /* only do the rest for the real thing */ 454 455 V_rtq_reallyold6 = 60*60; /* one hour is ``really old'' */ 456 V_rtq_minreallyold6 = 10; /* never automatically crank down to less */ 457 V_rtq_toomany6 = 128; /* 128 cached routes is ``too many'' */ 458 V_rtq_timeout6 = RTQ_TIMEOUT; 459 460 rnh = *head; 461 KASSERT(rnh == rt_tables_get_rnh(0, AF_INET6), ("rnh?")); 462 rnh->rnh_addaddr = in6_addroute; 463 rnh->rnh_matchaddr = in6_matroute; 464 callout_init(&V_rtq_timer6, CALLOUT_MPSAFE); 465 callout_init(&V_rtq_mtutimer, CALLOUT_MPSAFE); 466 in6_rtqtimo(curvnet); /* kick off timeout first time */ 467 in6_mtutimo(curvnet); /* kick off timeout first time */ 468 return 1; 469} 470 471#ifdef VIMAGE 472int 473in6_detachhead(void **head, int off) 474{ 475 INIT_VNET_INET6(curvnet); 476 477 callout_drain(&V_rtq_timer6); 478 callout_drain(&V_rtq_mtutimer); 479 return (1); 480} 481#endif 482