in6_rmx.c revision 195727
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 195727 2009-07-16 21:13:04Z rwatson $"); 77 78#include <sys/param.h> 79#include <sys/systm.h> 80#include <sys/kernel.h> 81#include <sys/lock.h> 82#include <sys/sysctl.h> 83#include <sys/queue.h> 84#include <sys/socket.h> 85#include <sys/socketvar.h> 86#include <sys/mbuf.h> 87#include <sys/rwlock.h> 88#include <sys/syslog.h> 89#include <sys/callout.h> 90#include <sys/vimage.h> 91 92#include <net/if.h> 93#include <net/route.h> 94 95#include <netinet/in.h> 96#include <netinet/ip_var.h> 97#include <netinet/in_var.h> 98 99#include <netinet/ip6.h> 100#include <netinet6/ip6_var.h> 101 102#include <netinet/icmp6.h> 103#include <netinet6/nd6.h> 104 105#include <netinet/tcp.h> 106#include <netinet/tcp_seq.h> 107#include <netinet/tcp_timer.h> 108#include <netinet/tcp_var.h> 109 110extern int in6_inithead(void **head, int off); 111#ifdef VIMAGE 112extern int in6_detachhead(void **head, int off); 113#endif 114 115#define RTPRF_OURS RTF_PROTO3 /* set on routes we manage */ 116 117/* 118 * Do what we need to do when inserting a route. 119 */ 120static struct radix_node * 121in6_addroute(void *v_arg, void *n_arg, struct radix_node_head *head, 122 struct radix_node *treenodes) 123{ 124 struct rtentry *rt = (struct rtentry *)treenodes; 125 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)rt_key(rt); 126 struct radix_node *ret; 127 128 RADIX_NODE_HEAD_WLOCK_ASSERT(head); 129 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) 130 rt->rt_flags |= RTF_MULTICAST; 131 132 /* 133 * A little bit of help for both IPv6 output and input: 134 * For local addresses, we make sure that RTF_LOCAL is set, 135 * with the thought that this might one day be used to speed up 136 * ip_input(). 137 * 138 * We also mark routes to multicast addresses as such, because 139 * it's easy to do and might be useful (but this is much more 140 * dubious since it's so easy to inspect the address). (This 141 * is done above.) 142 * 143 * XXX 144 * should elaborate the code. 145 */ 146 if (rt->rt_flags & RTF_HOST) { 147 if (IN6_ARE_ADDR_EQUAL(&satosin6(rt->rt_ifa->ifa_addr) 148 ->sin6_addr, 149 &sin6->sin6_addr)) { 150 rt->rt_flags |= RTF_LOCAL; 151 } 152 } 153 154 if (!rt->rt_rmx.rmx_mtu && rt->rt_ifp) 155 rt->rt_rmx.rmx_mtu = IN6_LINKMTU(rt->rt_ifp); 156 157 ret = rn_addroute(v_arg, n_arg, head, treenodes); 158 if (ret == NULL) { 159 struct rtentry *rt2; 160 /* 161 * We are trying to add a net route, but can't. 162 * The following case should be allowed, so we'll make a 163 * special check for this: 164 * Two IPv6 addresses with the same prefix is assigned 165 * to a single interrface. 166 * # ifconfig if0 inet6 3ffe:0501::1 prefix 64 alias (*1) 167 * # ifconfig if0 inet6 3ffe:0501::2 prefix 64 alias (*2) 168 * In this case, (*1) and (*2) want to add the same 169 * net route entry, 3ffe:0501:: -> if0. 170 * This case should not raise an error. 171 */ 172 rt2 = rtalloc1((struct sockaddr *)sin6, 0, RTF_RNH_LOCKED); 173 if (rt2) { 174 if (((rt2->rt_flags & (RTF_HOST|RTF_GATEWAY)) == 0) 175 && rt2->rt_gateway 176 && rt2->rt_gateway->sa_family == AF_LINK 177 && rt2->rt_ifp == rt->rt_ifp) { 178 ret = rt2->rt_nodes; 179 } 180 RTFREE_LOCKED(rt2); 181 } 182 } 183 return (ret); 184} 185 186/* 187 * This code is the inverse of in6_clsroute: on first reference, if we 188 * were managing the route, stop doing so and set the expiration timer 189 * back off again. 190 */ 191static struct radix_node * 192in6_matroute(void *v_arg, struct radix_node_head *head) 193{ 194 struct radix_node *rn = rn_match(v_arg, head); 195 struct rtentry *rt = (struct rtentry *)rn; 196 197 if (rt && rt->rt_refcnt == 0) { /* this is first reference */ 198 if (rt->rt_flags & RTPRF_OURS) { 199 rt->rt_flags &= ~RTPRF_OURS; 200 rt->rt_rmx.rmx_expire = 0; 201 } 202 } 203 return rn; 204} 205 206SYSCTL_DECL(_net_inet6_ip6); 207 208static VNET_DEFINE(int, rtq_reallyold6); 209static VNET_DEFINE(int, rtq_minreallyold6); 210static VNET_DEFINE(int, rtq_toomany6); 211 212#define V_rtq_reallyold6 VNET(rtq_reallyold6) 213#define V_rtq_minreallyold6 VNET(rtq_minreallyold6) 214#define V_rtq_toomany6 VNET(rtq_toomany6) 215 216SYSCTL_VNET_INT(_net_inet6_ip6, IPV6CTL_RTEXPIRE, rtexpire, CTLFLAG_RW, 217 &VNET_NAME(rtq_reallyold6) , 0, ""); 218 219SYSCTL_VNET_INT(_net_inet6_ip6, IPV6CTL_RTMINEXPIRE, rtminexpire, CTLFLAG_RW, 220 &VNET_NAME(rtq_minreallyold6) , 0, ""); 221 222SYSCTL_VNET_INT(_net_inet6_ip6, IPV6CTL_RTMAXCACHE, rtmaxcache, CTLFLAG_RW, 223 &VNET_NAME(rtq_toomany6) , 0, ""); 224 225struct rtqk_arg { 226 struct radix_node_head *rnh; 227 int mode; 228 int updating; 229 int draining; 230 int killed; 231 int found; 232 time_t nextstop; 233}; 234 235/* 236 * Get rid of old routes. When draining, this deletes everything, even when 237 * the timeout is not expired yet. When updating, this makes sure that 238 * nothing has a timeout longer than the current value of rtq_reallyold6. 239 */ 240static int 241in6_rtqkill(struct radix_node *rn, void *rock) 242{ 243 struct rtqk_arg *ap = rock; 244 struct rtentry *rt = (struct rtentry *)rn; 245 int err; 246 247 RADIX_NODE_HEAD_WLOCK_ASSERT(ap->rnh); 248 249 if (rt->rt_flags & RTPRF_OURS) { 250 ap->found++; 251 252 if (ap->draining || rt->rt_rmx.rmx_expire <= time_uptime) { 253 if (rt->rt_refcnt > 0) 254 panic("rtqkill route really not free"); 255 256 err = rtrequest(RTM_DELETE, 257 (struct sockaddr *)rt_key(rt), 258 rt->rt_gateway, rt_mask(rt), 259 rt->rt_flags|RTF_RNH_LOCKED, 0); 260 if (err) { 261 log(LOG_WARNING, "in6_rtqkill: error %d", err); 262 } else { 263 ap->killed++; 264 } 265 } else { 266 if (ap->updating 267 && (rt->rt_rmx.rmx_expire - time_uptime 268 > V_rtq_reallyold6)) { 269 rt->rt_rmx.rmx_expire = time_uptime 270 + V_rtq_reallyold6; 271 } 272 ap->nextstop = lmin(ap->nextstop, 273 rt->rt_rmx.rmx_expire); 274 } 275 } 276 277 return 0; 278} 279 280#define RTQ_TIMEOUT 60*10 /* run no less than once every ten minutes */ 281static VNET_DEFINE(int, rtq_timeout6); 282static VNET_DEFINE(struct callout, rtq_timer6); 283 284#define V_rtq_timeout6 VNET(rtq_timeout6) 285#define V_rtq_timer6 VNET(rtq_timer6) 286 287static void 288in6_rtqtimo(void *rock) 289{ 290 CURVNET_SET_QUIET((struct vnet *) rock); 291 struct radix_node_head *rnh; 292 struct rtqk_arg arg; 293 struct timeval atv; 294 static time_t last_adjusted_timeout = 0; 295 296 rnh = rt_tables_get_rnh(0, AF_INET6); 297 if (rnh == NULL) { 298 CURVNET_RESTORE(); 299 return; 300 } 301 arg.found = arg.killed = 0; 302 arg.rnh = rnh; 303 arg.nextstop = time_uptime + V_rtq_timeout6; 304 arg.draining = arg.updating = 0; 305 RADIX_NODE_HEAD_LOCK(rnh); 306 rnh->rnh_walktree(rnh, in6_rtqkill, &arg); 307 RADIX_NODE_HEAD_UNLOCK(rnh); 308 309 /* 310 * Attempt to be somewhat dynamic about this: 311 * If there are ``too many'' routes sitting around taking up space, 312 * then crank down the timeout, and see if we can't make some more 313 * go away. However, we make sure that we will never adjust more 314 * than once in rtq_timeout6 seconds, to keep from cranking down too 315 * hard. 316 */ 317 if ((arg.found - arg.killed > V_rtq_toomany6) 318 && (time_uptime - last_adjusted_timeout >= V_rtq_timeout6) 319 && V_rtq_reallyold6 > V_rtq_minreallyold6) { 320 V_rtq_reallyold6 = 2*V_rtq_reallyold6 / 3; 321 if (V_rtq_reallyold6 < V_rtq_minreallyold6) { 322 V_rtq_reallyold6 = V_rtq_minreallyold6; 323 } 324 325 last_adjusted_timeout = time_uptime; 326#ifdef DIAGNOSTIC 327 log(LOG_DEBUG, "in6_rtqtimo: adjusted rtq_reallyold6 to %d", 328 V_rtq_reallyold6); 329#endif 330 arg.found = arg.killed = 0; 331 arg.updating = 1; 332 RADIX_NODE_HEAD_LOCK(rnh); 333 rnh->rnh_walktree(rnh, in6_rtqkill, &arg); 334 RADIX_NODE_HEAD_UNLOCK(rnh); 335 } 336 337 atv.tv_usec = 0; 338 atv.tv_sec = arg.nextstop - time_uptime; 339 callout_reset(&V_rtq_timer6, tvtohz(&atv), in6_rtqtimo, rock); 340 CURVNET_RESTORE(); 341} 342 343/* 344 * Age old PMTUs. 345 */ 346struct mtuex_arg { 347 struct radix_node_head *rnh; 348 time_t nextstop; 349}; 350 351static VNET_DEFINE(struct callout, rtq_mtutimer); 352#define V_rtq_mtutimer VNET(rtq_mtutimer) 353 354static int 355in6_mtuexpire(struct radix_node *rn, void *rock) 356{ 357 struct rtentry *rt = (struct rtentry *)rn; 358 struct mtuex_arg *ap = rock; 359 360 /* sanity */ 361 if (!rt) 362 panic("rt == NULL in in6_mtuexpire"); 363 364 if (rt->rt_rmx.rmx_expire && !(rt->rt_flags & RTF_PROBEMTU)) { 365 if (rt->rt_rmx.rmx_expire <= time_uptime) { 366 rt->rt_flags |= RTF_PROBEMTU; 367 } else { 368 ap->nextstop = lmin(ap->nextstop, 369 rt->rt_rmx.rmx_expire); 370 } 371 } 372 373 return 0; 374} 375 376#define MTUTIMO_DEFAULT (60*1) 377 378static void 379in6_mtutimo(void *rock) 380{ 381 CURVNET_SET_QUIET((struct vnet *) rock); 382 struct radix_node_head *rnh; 383 struct mtuex_arg arg; 384 struct timeval atv; 385 386 rnh = rt_tables_get_rnh(0, AF_INET6); 387 if (rnh == NULL) { 388 CURVNET_RESTORE(); 389 return; 390 } 391 arg.rnh = rnh; 392 arg.nextstop = time_uptime + MTUTIMO_DEFAULT; 393 RADIX_NODE_HEAD_LOCK(rnh); 394 rnh->rnh_walktree(rnh, in6_mtuexpire, &arg); 395 RADIX_NODE_HEAD_UNLOCK(rnh); 396 397 atv.tv_usec = 0; 398 atv.tv_sec = arg.nextstop - time_uptime; 399 if (atv.tv_sec < 0) { 400 printf("invalid mtu expiration time on routing table\n"); 401 arg.nextstop = time_uptime + 30; /* last resort */ 402 atv.tv_sec = 30; 403 } 404 callout_reset(&V_rtq_mtutimer, tvtohz(&atv), in6_mtutimo, rock); 405 CURVNET_RESTORE(); 406} 407 408/* 409 * Initialize our routing tree. 410 * XXX MRT When off == 0, we are being called from vfs_export.c 411 * so just set up their table and leave. (we know what the correct 412 * value should be so just use that).. FIX AFTER RELENG_7 is MFC'd 413 * see also comments in in_inithead() vfs_export.c and domain.h 414 */ 415int 416in6_inithead(void **head, int off) 417{ 418 struct radix_node_head *rnh; 419 420 if (!rn_inithead(head, offsetof(struct sockaddr_in6, sin6_addr) << 3)) 421 return 0; /* See above */ 422 423 if (off == 0) /* See above */ 424 return 1; /* only do the rest for the real thing */ 425 426 V_rtq_reallyold6 = 60*60; /* one hour is ``really old'' */ 427 V_rtq_minreallyold6 = 10; /* never automatically crank down to less */ 428 V_rtq_toomany6 = 128; /* 128 cached routes is ``too many'' */ 429 V_rtq_timeout6 = RTQ_TIMEOUT; 430 431 rnh = *head; 432 KASSERT(rnh == rt_tables_get_rnh(0, AF_INET6), ("rnh?")); 433 rnh->rnh_addaddr = in6_addroute; 434 rnh->rnh_matchaddr = in6_matroute; 435 callout_init(&V_rtq_timer6, CALLOUT_MPSAFE); 436 callout_init(&V_rtq_mtutimer, CALLOUT_MPSAFE); 437 in6_rtqtimo(curvnet); /* kick off timeout first time */ 438 in6_mtutimo(curvnet); /* kick off timeout first time */ 439 return 1; 440} 441 442#ifdef VIMAGE 443int 444in6_detachhead(void **head, int off) 445{ 446 447 callout_drain(&V_rtq_timer6); 448 callout_drain(&V_rtq_mtutimer); 449 return (1); 450} 451#endif 452