bsd/netinet6/in6_rmx.c

   1 /*      $FreeBSD: src/sys/netinet6/in6_rmx.c,v 1.1.2.2 2001/07/03 11:01:52 ume Exp $    */
   2 /*      $KAME: in6_rmx.c,v 1.10 2001/05/24 05:44:58 itojun 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 <kern/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 static void     in6_rtqtimo __P((void *rock));
 104 static void in6_mtutimo __P((void *rock));
 105
 106 #define RTPRF_OURS              RTF_PROTO3      /* set on routes we manage */
 107
 108 /*
 109  * Do what we need to do when inserting a route.
 110  */
 111 static struct radix_node *
 112 in6_addroute(void *v_arg, void *n_arg, struct radix_node_head *head,
 113             struct radix_node *treenodes)
 114 {
 115         struct rtentry *rt = (struct rtentry *)treenodes;
 116         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)rt_key(rt);
 117         struct radix_node *ret;
 118
 119         /*
 120          * For IPv6, all unicast non-host routes are automatically cloning.
 121          */
 122         if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
 123                 rt->rt_flags |= RTF_MULTICAST;
 124
 125         if (!(rt->rt_flags & (RTF_HOST | RTF_CLONING | RTF_MULTICAST))) {
 126                 rt->rt_flags |= RTF_PRCLONING;
 127         }
 128
 129         /*
 130          * A little bit of help for both IPv6 output and input:
 131          *   For local addresses, we make sure that RTF_LOCAL is set,
 132          *   with the thought that this might one day be used to speed up
 133          *   ip_input().
 134          *
 135          * We also mark routes to multicast addresses as such, because
 136          * it's easy to do and might be useful (but this is much more
 137          * dubious since it's so easy to inspect the address).  (This
 138          * is done above.)
 139          *
 140          * XXX
 141          * should elaborate the code.
 142          */
 143         if (rt->rt_flags & RTF_HOST) {
 144                 if (IN6_ARE_ADDR_EQUAL(&satosin6(rt->rt_ifa->ifa_addr)
 145                                         ->sin6_addr,
 146                                        &sin6->sin6_addr)) {
 147                         rt->rt_flags |= RTF_LOCAL;
 148                 }
 149         }
 150
 151         if (!rt->rt_rmx.rmx_mtu && !(rt->rt_rmx.rmx_locks & RTV_MTU)
 152             && rt->rt_ifp)
 153                 rt->rt_rmx.rmx_mtu = rt->rt_ifp->if_mtu;
 154
 155         ret = rn_addroute(v_arg, n_arg, head, treenodes);
 156         if (ret == NULL && rt->rt_flags & RTF_HOST) {
 157                 struct rtentry *rt2;
 158                 /*
 159                  * We are trying to add a host route, but can't.
 160                  * Find out if it is because of an
 161                  * ARP entry and delete it if so.
 162                  */
 163                 rt2 = rtalloc1((struct sockaddr *)sin6, 0,
 164                                 RTF_CLONING | RTF_PRCLONING);
 165                 if (rt2) {
 166                         if (rt2->rt_flags & RTF_LLINFO &&
 167                                 rt2->rt_flags & RTF_HOST &&
 168                                 rt2->rt_gateway &&
 169                                 rt2->rt_gateway->sa_family == AF_LINK) {
 170                                 rtrequest(RTM_DELETE,
 171                                           (struct sockaddr *)rt_key(rt2),
 172                                           rt2->rt_gateway,
 173                                           rt_mask(rt2), rt2->rt_flags, 0);
 174                                 ret = rn_addroute(v_arg, n_arg, head,
 175                                         treenodes);
 176                         }
 177                         rtfree(rt2);
 178                 }
 179         } else if (ret == NULL && rt->rt_flags & RTF_CLONING) {
 180                 struct rtentry *rt2;
 181                 /*
 182                  * We are trying to add a net route, but can't.
 183                  * The following case should be allowed, so we'll make a
 184                  * special check for this:
 185                  *      Two IPv6 addresses with the same prefix is assigned
 186                  *      to a single interrface.
 187                  *      # ifconfig if0 inet6 3ffe:0501::1 prefix 64 alias (*1)
 188                  *      # ifconfig if0 inet6 3ffe:0501::2 prefix 64 alias (*2)
 189                  *      In this case, (*1) and (*2) want to add the same
 190                  *      net route entry, 3ffe:0501:: -> if0.
 191                  *      This case should not raise an error.
 192                  */
 193                 rt2 = rtalloc1((struct sockaddr *)sin6, 0,
 194                                 RTF_CLONING | RTF_PRCLONING);
 195                 if (rt2) {
 196                         if ((rt2->rt_flags & (RTF_CLONING|RTF_HOST|RTF_GATEWAY))
 197                                         == RTF_CLONING
 198                          && rt2->rt_gateway
 199                          && rt2->rt_gateway->sa_family == AF_LINK
 200                          && rt2->rt_ifp == rt->rt_ifp) {
 201                                 ret = rt2->rt_nodes;
 202                         }
 203                         rtfree(rt2);
 204                 }
 205         }
 206         return ret;
 207 }
 208
 209 /*
 210  * This code is the inverse of in6_clsroute: on first reference, if we
 211  * were managing the route, stop doing so and set the expiration timer
 212  * back off again.
 213  */
 214 static struct radix_node *
 215 in6_matroute(void *v_arg, struct radix_node_head *head)
 216 {
 217         struct radix_node *rn = rn_match(v_arg, head);
 218         struct rtentry *rt = (struct rtentry *)rn;
 219
 220         if (rt && rt->rt_refcnt == 0) { /* this is first reference */
 221                 if (rt->rt_flags & RTPRF_OURS) {
 222                         rt->rt_flags &= ~RTPRF_OURS;
 223                         rt->rt_rmx.rmx_expire = 0;
 224                 }
 225         }
 226         return rn;
 227 }
 228
 229 SYSCTL_DECL(_net_inet6_ip6);
 230
 231 static int rtq_reallyold = 60*60;
 232         /* one hour is ``really old'' */
 233 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_RTEXPIRE, rtexpire,
 234         CTLFLAG_RW, &rtq_reallyold , 0, "");
 235
 236 static int rtq_minreallyold = 10;
 237         /* never automatically crank down to less */
 238 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_RTMINEXPIRE, rtminexpire,
 239         CTLFLAG_RW, &rtq_minreallyold , 0, "");
 240
 241 static int rtq_toomany = 128;
 242         /* 128 cached routes is ``too many'' */
 243 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_RTMAXCACHE, rtmaxcache,
 244         CTLFLAG_RW, &rtq_toomany , 0, "");
 245
 246
 247 /*
 248  * On last reference drop, mark the route as belong to us so that it can be
 249  * timed out.
 250  */
 251 static void
 252 in6_clsroute(struct radix_node *rn, struct radix_node_head *head)
 253 {
 254         struct rtentry *rt = (struct rtentry *)rn;
 255
 256         if (!(rt->rt_flags & RTF_UP))
 257                 return;         /* prophylactic measures */
 258
 259         if ((rt->rt_flags & (RTF_LLINFO | RTF_HOST)) != RTF_HOST)
 260                 return;
 261
 262         if ((rt->rt_flags & (RTF_WASCLONED | RTPRF_OURS))
 263            != RTF_WASCLONED)
 264                 return;
 265
 266         /*
 267          * As requested by David Greenman:
 268          * If rtq_reallyold is 0, just delete the route without
 269          * waiting for a timeout cycle to kill it.
 270          */
 271         if (rtq_reallyold != 0) {
 272                 rt->rt_flags |= RTPRF_OURS;
 273                 rt->rt_rmx.rmx_expire = time_second + rtq_reallyold;
 274         } else {
 275                 rtrequest(RTM_DELETE,
 276                           (struct sockaddr *)rt_key(rt),
 277                           rt->rt_gateway, rt_mask(rt),
 278                           rt->rt_flags, 0);
 279         }
 280 }
 281
 282 struct rtqk_arg {
 283         struct radix_node_head *rnh;
 284         int mode;
 285         int updating;
 286         int draining;
 287         int killed;
 288         int found;
 289         time_t nextstop;
 290 };
 291
 292 /*
 293  * Get rid of old routes.  When draining, this deletes everything, even when
 294  * the timeout is not expired yet.  When updating, this makes sure that
 295  * nothing has a timeout longer than the current value of rtq_reallyold.
 296  */
 297 static int
 298 in6_rtqkill(struct radix_node *rn, void *rock)
 299 {
 300         struct rtqk_arg *ap = rock;
 301         struct rtentry *rt = (struct rtentry *)rn;
 302         int err;
 303
 304         if (rt->rt_flags & RTPRF_OURS) {
 305                 ap->found++;
 306
 307                 if (ap->draining || rt->rt_rmx.rmx_expire <= time_second) {
 308                         if (rt->rt_refcnt > 0)
 309                                 panic("rtqkill route really not free");
 310
 311                         err = rtrequest(RTM_DELETE,
 312                                         (struct sockaddr *)rt_key(rt),
 313                                         rt->rt_gateway, rt_mask(rt),
 314                                         rt->rt_flags, 0);
 315                         if (err) {
 316                                 log(LOG_WARNING, "in6_rtqkill: error %d", err);
 317                         } else {
 318                                 ap->killed++;
 319                         }
 320                 } else {
 321                         if (ap->updating
 322                            && (rt->rt_rmx.rmx_expire - time_second
 323                                > rtq_reallyold)) {
 324                                 rt->rt_rmx.rmx_expire = time_second
 325                                         + rtq_reallyold;
 326                         }
 327                         ap->nextstop = lmin(ap->nextstop,
 328                                             rt->rt_rmx.rmx_expire);
 329                 }
 330         }
 331
 332         return 0;
 333 }
 334
 335 #define RTQ_TIMEOUT     60*10   /* run no less than once every ten minutes */
 336 static int rtq_timeout = RTQ_TIMEOUT;
 337
 338 static void
 339 in6_rtqtimo_funneled(void *rock)
 340 {
 341 #ifdef __APPLE__
 342         boolean_t   funnel_state;
 343         funnel_state = thread_funnel_set(network_flock, TRUE);
 344         in6_rtqtimo(rock);
 345 #endif
 346 #ifdef __APPLE__
 347         (void) thread_funnel_set(network_flock, FALSE);
 348 #endif
 349 }
 350
 351 static void
 352 in6_rtqtimo(void *rock)
 353 {
 354         struct radix_node_head *rnh = rock;
 355         struct rtqk_arg arg;
 356         struct timeval atv;
 357         static time_t last_adjusted_timeout = 0;
 358         int s;
 359
 360         arg.found = arg.killed = 0;
 361         arg.rnh = rnh;
 362         arg.nextstop = time_second + rtq_timeout;
 363         arg.draining = arg.updating = 0;
 364         s = splnet();
 365         rnh->rnh_walktree(rnh, in6_rtqkill, &arg);
 366         splx(s);
 367
 368         /*
 369          * Attempt to be somewhat dynamic about this:
 370          * If there are ``too many'' routes sitting around taking up space,
 371          * then crank down the timeout, and see if we can't make some more
 372          * go away.  However, we make sure that we will never adjust more
 373          * than once in rtq_timeout seconds, to keep from cranking down too
 374          * hard.
 375          */
 376         if ((arg.found - arg.killed > rtq_toomany)
 377            && (time_second - last_adjusted_timeout >= rtq_timeout)
 378            && rtq_reallyold > rtq_minreallyold) {
 379                 rtq_reallyold = 2*rtq_reallyold / 3;
 380                 if (rtq_reallyold < rtq_minreallyold) {
 381                         rtq_reallyold = rtq_minreallyold;
 382                 }
 383
 384                 last_adjusted_timeout = time_second;
 385 #if DIAGNOSTIC
 386                 log(LOG_DEBUG, "in6_rtqtimo: adjusted rtq_reallyold to %d",
 387                     rtq_reallyold);
 388 #endif
 389                 arg.found = arg.killed = 0;
 390                 arg.updating = 1;
 391                 s = splnet();
 392                 rnh->rnh_walktree(rnh, in6_rtqkill, &arg);
 393                 splx(s);
 394         }
 395
 396         atv.tv_usec = 0;
 397         atv.tv_sec = arg.nextstop - time_second;
 398         timeout(in6_rtqtimo_funneled, rock, tvtohz(&atv));
 399 }
 400
 401 /*
 402  * Age old PMTUs.
 403  */
 404 struct mtuex_arg {
 405         struct radix_node_head *rnh;
 406         time_t nextstop;
 407 };
 408
 409 static int
 410 in6_mtuexpire(struct radix_node *rn, void *rock)
 411 {
 412         struct rtentry *rt = (struct rtentry *)rn;
 413         struct mtuex_arg *ap = rock;
 414
 415         /* sanity */
 416         if (!rt)
 417                 panic("rt == NULL in in6_mtuexpire");
 418
 419         if (rt->rt_rmx.rmx_expire && !(rt->rt_flags & RTF_PROBEMTU)) {
 420                 if (rt->rt_rmx.rmx_expire <= time_second) {
 421                         rt->rt_flags |= RTF_PROBEMTU;
 422                 } else {
 423                         ap->nextstop = lmin(ap->nextstop,
 424                                         rt->rt_rmx.rmx_expire);
 425                 }
 426         }
 427
 428         return 0;
 429 }
 430
 431 #define MTUTIMO_DEFAULT (60*1)
 432
 433 static void
 434 in6_mtutimo_funneled(void *rock)
 435 {
 436 #ifdef __APPLE__
 437         boolean_t   funnel_state;
 438         funnel_state = thread_funnel_set(network_flock, TRUE);
 439         in6_mtutimo(rock);
 440 #endif
 441 #ifdef __APPLE__
 442         (void) thread_funnel_set(network_flock, FALSE);
 443 #endif
 444 }
 445
 446 static void
 447 in6_mtutimo(void *rock)
 448 {
 449         struct radix_node_head *rnh = rock;
 450         struct mtuex_arg arg;
 451         struct timeval atv;
 452         int s;
 453
 454         arg.rnh = rnh;
 455         arg.nextstop = time_second + MTUTIMO_DEFAULT;
 456         s = splnet();
 457         rnh->rnh_walktree(rnh, in6_mtuexpire, &arg);
 458         splx(s);
 459
 460         atv.tv_usec = 0;
 461         atv.tv_sec = arg.nextstop;
 462         if (atv.tv_sec < time_second) {
 463 #if DIAGNOSTIC
 464                 log(LOG_DEBUG, "IPv6: invalid mtu expiration time on routing table\n");
 465 #endif
 466                 arg.nextstop = time_second + 30;        /*last resort*/
 467         }
 468         atv.tv_sec -= time_second;
 469         timeout(in6_mtutimo_funneled, rock, tvtohz(&atv));
 470 }
 471
 472 #if 0
 473 void
 474 in6_rtqdrain()
 475 {
 476         struct radix_node_head *rnh = rt_tables[AF_INET6];
 477         struct rtqk_arg arg;
 478         int s;
 479         arg.found = arg.killed = 0;
 480         arg.rnh = rnh;
 481         arg.nextstop = 0;
 482         arg.draining = 1;
 483         arg.updating = 0;
 484         s = splnet();
 485         rnh->rnh_walktree(rnh, in6_rtqkill, &arg);
 486         splx(s);
 487 }
 488 #endif
 489
 490 /*
 491  * Initialize our routing tree.
 492  */
 493 int
 494 in6_inithead(void **head, int off)
 495 {
 496         struct radix_node_head *rnh;
 497
 498         if (!rn_inithead(head, off))
 499                 return 0;
 500
 501         if (head != (void **)&rt_tables[AF_INET6]) /* BOGUS! */
 502                 return 1;       /* only do this for the real routing table */
 503
 504         rnh = *head;
 505         rnh->rnh_addaddr = in6_addroute;
 506         rnh->rnh_matchaddr = in6_matroute;
 507         rnh->rnh_close = in6_clsroute;
 508         in6_rtqtimo(rnh);       /* kick off timeout first time */
 509         in6_mtutimo(rnh);       /* kick off timeout first time */
 510         return 1;
 511 }