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 #include <kern/lock.h>
  86
  87 #include <net/if.h>
  88 #include <net/route.h>
  89 #include <netinet/in.h>
  90 #include <netinet/ip_var.h>
  91 #include <netinet/in_var.h>
  92
  93 #include <netinet/ip6.h>
  94 #include <netinet6/ip6_var.h>
  95
  96 #include <netinet/icmp6.h>
  97
  98 #include <netinet/tcp.h>
  99 #include <netinet/tcp_seq.h>
 100 #include <netinet/tcp_timer.h>
 101 #include <netinet/tcp_var.h>
 102
 103 extern int      in6_inithead(void **head, int off);
 104 static void     in6_rtqtimo(void *rock);
 105 static void in6_mtutimo(void *rock);
 106 extern lck_mtx_t *rt_mtx;
 107
 108 #define RTPRF_OURS              RTF_PROTO3      /* set on routes we manage */
 109
 110 /*
 111  * Do what we need to do when inserting a route.
 112  */
 113 static struct radix_node *
 114 in6_addroute(void *v_arg, void *n_arg, struct radix_node_head *head,
 115             struct radix_node *treenodes)
 116 {
 117         struct rtentry *rt = (struct rtentry *)treenodes;
 118         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)rt_key(rt);
 119         struct radix_node *ret;
 120
 121         /*
 122          * For IPv6, all unicast non-host routes are automatically cloning.
 123          */
 124         if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
 125                 rt->rt_flags |= RTF_MULTICAST;
 126
 127         if (!(rt->rt_flags & (RTF_HOST | RTF_CLONING | RTF_MULTICAST))) {
 128                 rt->rt_flags |= RTF_PRCLONING;
 129         }
 130
 131         /*
 132          * A little bit of help for both IPv6 output and input:
 133          *   For local addresses, we make sure that RTF_LOCAL is set,
 134          *   with the thought that this might one day be used to speed up
 135          *   ip_input().
 136          *
 137          * We also mark routes to multicast addresses as such, because
 138          * it's easy to do and might be useful (but this is much more
 139          * dubious since it's so easy to inspect the address).  (This
 140          * is done above.)
 141          *
 142          * XXX
 143          * should elaborate the code.
 144          */
 145         if (rt->rt_flags & RTF_HOST) {
 146                 if (IN6_ARE_ADDR_EQUAL(&satosin6(rt->rt_ifa->ifa_addr)
 147                                         ->sin6_addr,
 148                                        &sin6->sin6_addr)) {
 149                         rt->rt_flags |= RTF_LOCAL;
 150                 }
 151         }
 152
 153         if (!rt->rt_rmx.rmx_mtu && !(rt->rt_rmx.rmx_locks & RTV_MTU)
 154             && rt->rt_ifp)
 155                 rt->rt_rmx.rmx_mtu = rt->rt_ifp->if_mtu;
 156
 157         ret = rn_addroute(v_arg, n_arg, head, treenodes);
 158         if (ret == NULL && rt->rt_flags & RTF_HOST) {
 159                 struct rtentry *rt2;
 160                 /*
 161                  * We are trying to add a host route, but can't.
 162                  * Find out if it is because of an
 163                  * ARP entry and delete it if so.
 164                  */
 165                 rt2 = rtalloc1_locked((struct sockaddr *)sin6, 0,
 166                                 RTF_CLONING | RTF_PRCLONING);
 167                 if (rt2) {
 168                         if (rt2->rt_flags & RTF_LLINFO &&
 169                                 rt2->rt_flags & RTF_HOST &&
 170                                 rt2->rt_gateway &&
 171                                 rt2->rt_gateway->sa_family == AF_LINK) {
 172                                 rtrequest_locked(RTM_DELETE,
 173                                           (struct sockaddr *)rt_key(rt2),
 174                                           rt2->rt_gateway,
 175                                           rt_mask(rt2), rt2->rt_flags, 0);
 176                                 ret = rn_addroute(v_arg, n_arg, head,
 177                                         treenodes);
 178                         }
 179                         rtfree_locked(rt2);
 180                 }
 181         } else if (ret == NULL && rt->rt_flags & RTF_CLONING) {
 182                 struct rtentry *rt2;
 183                 /*
 184                  * We are trying to add a net route, but can't.
 185                  * The following case should be allowed, so we'll make a
 186                  * special check for this:
 187                  *      Two IPv6 addresses with the same prefix is assigned
 188                  *      to a single interrface.
 189                  *      # ifconfig if0 inet6 3ffe:0501::1 prefix 64 alias (*1)
 190                  *      # ifconfig if0 inet6 3ffe:0501::2 prefix 64 alias (*2)
 191                  *      In this case, (*1) and (*2) want to add the same
 192                  *      net route entry, 3ffe:0501:: -> if0.
 193                  *      This case should not raise an error.
 194                  */
 195                 rt2 = rtalloc1_locked((struct sockaddr *)sin6, 0,
 196                                 RTF_CLONING | RTF_PRCLONING);
 197                 if (rt2) {
 198                         if ((rt2->rt_flags & (RTF_CLONING|RTF_HOST|RTF_GATEWAY))
 199                                         == RTF_CLONING
 200                          && rt2->rt_gateway
 201                          && rt2->rt_gateway->sa_family == AF_LINK
 202                          && rt2->rt_ifp == rt->rt_ifp) {
 203                                 ret = rt2->rt_nodes;
 204                         }
 205                         rtfree_locked(rt2);
 206                 }
 207         }
 208         return ret;
 209 }
 210
 211 /*
 212  * This code is the inverse of in6_clsroute: on first reference, if we
 213  * were managing the route, stop doing so and set the expiration timer
 214  * back off again.
 215  */
 216 static struct radix_node *
 217 in6_matroute(void *v_arg, struct radix_node_head *head)
 218 {
 219         struct radix_node *rn = rn_match(v_arg, head);
 220         struct rtentry *rt = (struct rtentry *)rn;
 221
 222         if (rt && rt->rt_refcnt == 0) { /* this is first reference */
 223                 if (rt->rt_flags & RTPRF_OURS) {
 224                         rt->rt_flags &= ~RTPRF_OURS;
 225                         rt->rt_rmx.rmx_expire = 0;
 226                 }
 227         }
 228         return rn;
 229 }
 230
 231 SYSCTL_DECL(_net_inet6_ip6);
 232
 233 static int rtq_reallyold = 60*60;
 234         /* one hour is ``really old'' */
 235 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_RTEXPIRE, rtexpire,
 236         CTLFLAG_RW, &rtq_reallyold , 0, "");
 237
 238 static int rtq_minreallyold = 10;
 239         /* never automatically crank down to less */
 240 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_RTMINEXPIRE, rtminexpire,
 241         CTLFLAG_RW, &rtq_minreallyold , 0, "");
 242
 243 static int rtq_toomany = 128;
 244         /* 128 cached routes is ``too many'' */
 245 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_RTMAXCACHE, rtmaxcache,
 246         CTLFLAG_RW, &rtq_toomany , 0, "");
 247
 248
 249 /*
 250  * On last reference drop, mark the route as belong to us so that it can be
 251  * timed out.
 252  */
 253 static void
 254 in6_clsroute(struct radix_node *rn, struct radix_node_head *head)
 255 {
 256         struct rtentry *rt = (struct rtentry *)rn;
 257         struct timeval timenow;
 258
 259
 260         if (!(rt->rt_flags & RTF_UP))
 261                 return;         /* prophylactic measures */
 262
 263         if ((rt->rt_flags & (RTF_LLINFO | RTF_HOST)) != RTF_HOST)
 264                 return;
 265
 266         if ((rt->rt_flags & (RTF_WASCLONED | RTPRF_OURS))
 267            != RTF_WASCLONED)
 268                 return;
 269
 270         /*
 271          * As requested by David Greenman:
 272          * If rtq_reallyold is 0, just delete the route without
 273          * waiting for a timeout cycle to kill it.
 274          */
 275         getmicrotime(&timenow);
 276         if (rtq_reallyold != 0) {
 277                 rt->rt_flags |= RTPRF_OURS;
 278                 rt->rt_rmx.rmx_expire = timenow.tv_sec + rtq_reallyold;
 279         } else {
 280                 rtrequest_locked(RTM_DELETE,
 281                           (struct sockaddr *)rt_key(rt),
 282                           rt->rt_gateway, rt_mask(rt),
 283                           rt->rt_flags, 0);
 284         }
 285 }
 286
 287 struct rtqk_arg {
 288         struct radix_node_head *rnh;
 289         int mode;
 290         int updating;
 291         int draining;
 292         int killed;
 293         int found;
 294         time_t nextstop;
 295 };
 296
 297 /*
 298  * Get rid of old routes.  When draining, this deletes everything, even when
 299  * the timeout is not expired yet.  When updating, this makes sure that
 300  * nothing has a timeout longer than the current value of rtq_reallyold.
 301  */
 302 static int
 303 in6_rtqkill(struct radix_node *rn, void *rock)
 304 {
 305         struct rtqk_arg *ap = rock;
 306         struct rtentry *rt = (struct rtentry *)rn;
 307         int err;
 308         struct timeval timenow;
 309
 310         getmicrotime(&timenow);
 311
 312         lck_mtx_assert(rt_mtx, LCK_MTX_ASSERT_OWNED);
 313         if (rt->rt_flags & RTPRF_OURS) {
 314                 ap->found++;
 315
 316                 if (ap->draining || rt->rt_rmx.rmx_expire <= timenow.tv_sec) {
 317                         if (rt->rt_refcnt > 0)
 318                                 panic("rtqkill route really not free");
 319
 320                         err = rtrequest_locked(RTM_DELETE,
 321                                         (struct sockaddr *)rt_key(rt),
 322                                         rt->rt_gateway, rt_mask(rt),
 323                                         rt->rt_flags, 0);
 324                         if (err) {
 325                                 log(LOG_WARNING, "in6_rtqkill: error %d", err);
 326                         } else {
 327                                 ap->killed++;
 328                         }
 329                 } else {
 330                         if (ap->updating
 331                            && (rt->rt_rmx.rmx_expire - timenow.tv_sec
 332                                > rtq_reallyold)) {
 333                                 rt->rt_rmx.rmx_expire = timenow.tv_sec
 334                                         + rtq_reallyold;
 335                         }
 336                         ap->nextstop = lmin(ap->nextstop,
 337                                             rt->rt_rmx.rmx_expire);
 338                 }
 339         }
 340
 341         return 0;
 342 }
 343
 344 #define RTQ_TIMEOUT     60*10   /* run no less than once every ten minutes */
 345 static int rtq_timeout = RTQ_TIMEOUT;
 346
 347 static void
 348 in6_rtqtimo(void *rock)
 349 {
 350         struct radix_node_head *rnh = rock;
 351         struct rtqk_arg arg;
 352         struct timeval atv;
 353         static time_t last_adjusted_timeout = 0;
 354         struct timeval timenow;
 355
 356         getmicrotime(&timenow);
 357
 358         arg.found = arg.killed = 0;
 359         arg.rnh = rnh;
 360         arg.nextstop = timenow.tv_sec + rtq_timeout;
 361         arg.draining = arg.updating = 0;
 362         lck_mtx_lock(rt_mtx);
 363         rnh->rnh_walktree(rnh, in6_rtqkill, &arg);
 364
 365         /*
 366          * Attempt to be somewhat dynamic about this:
 367          * If there are ``too many'' routes sitting around taking up space,
 368          * then crank down the timeout, and see if we can't make some more
 369          * go away.  However, we make sure that we will never adjust more
 370          * than once in rtq_timeout seconds, to keep from cranking down too
 371          * hard.
 372          */
 373         if ((arg.found - arg.killed > rtq_toomany)
 374            && (timenow.tv_sec - last_adjusted_timeout >= rtq_timeout)
 375            && rtq_reallyold > rtq_minreallyold) {
 376                 rtq_reallyold = 2*rtq_reallyold / 3;
 377                 if (rtq_reallyold < rtq_minreallyold) {
 378                         rtq_reallyold = rtq_minreallyold;
 379                 }
 380
 381                 last_adjusted_timeout = timenow.tv_sec;
 382 #if DIAGNOSTIC
 383                 log(LOG_DEBUG, "in6_rtqtimo: adjusted rtq_reallyold to %d",
 384                     rtq_reallyold);
 385 #endif
 386                 arg.found = arg.killed = 0;
 387                 arg.updating = 1;
 388                 rnh->rnh_walktree(rnh, in6_rtqkill, &arg);
 389         }
 390
 391         atv.tv_usec = 0;
 392         atv.tv_sec = arg.nextstop - timenow.tv_sec;
 393         lck_mtx_unlock(rt_mtx);
 394         timeout(in6_rtqtimo, rock, tvtohz(&atv));
 395 }
 396
 397 /*
 398  * Age old PMTUs.
 399  */
 400 struct mtuex_arg {
 401         struct radix_node_head *rnh;
 402         time_t nextstop;
 403 };
 404
 405 static int
 406 in6_mtuexpire(struct radix_node *rn, void *rock)
 407 {
 408         struct rtentry *rt = (struct rtentry *)rn;
 409         struct mtuex_arg *ap = rock;
 410         struct timeval timenow;
 411
 412         getmicrotime(&timenow);
 413
 414         /* sanity */
 415         if (!rt)
 416                 panic("rt == NULL in in6_mtuexpire");
 417
 418         if (rt->rt_rmx.rmx_expire && !(rt->rt_flags & RTF_PROBEMTU)) {
 419                 if (rt->rt_rmx.rmx_expire <= timenow.tv_sec) {
 420                         rt->rt_flags |= RTF_PROBEMTU;
 421                 } else {
 422                         ap->nextstop = lmin(ap->nextstop,
 423                                         rt->rt_rmx.rmx_expire);
 424                 }
 425         }
 426
 427         return 0;
 428 }
 429
 430 #define MTUTIMO_DEFAULT (60*1)
 431
 432 static void
 433 in6_mtutimo(void *rock)
 434 {
 435         struct radix_node_head *rnh = rock;
 436         struct mtuex_arg arg;
 437         struct timeval atv;
 438         struct timeval timenow;
 439
 440         getmicrotime(&timenow);
 441
 442         arg.rnh = rnh;
 443         arg.nextstop = timenow.tv_sec + MTUTIMO_DEFAULT;
 444         lck_mtx_lock(rt_mtx);
 445         rnh->rnh_walktree(rnh, in6_mtuexpire, &arg);
 446
 447         atv.tv_usec = 0;
 448         atv.tv_sec = arg.nextstop;
 449         if (atv.tv_sec < timenow.tv_sec) {
 450 #if DIAGNOSTIC
 451                 log(LOG_DEBUG, "IPv6: invalid mtu expiration time on routing table\n");
 452 #endif
 453                 arg.nextstop = timenow.tv_sec + 30;     /*last resort*/
 454         }
 455         atv.tv_sec -= timenow.tv_sec;
 456         lck_mtx_unlock(rt_mtx);
 457         timeout(in6_mtutimo, rock, tvtohz(&atv));
 458 }
 459
 460 #if 0
 461 void
 462 in6_rtqdrain()
 463 {
 464         struct radix_node_head *rnh = rt_tables[AF_INET6];
 465         struct rtqk_arg arg;
 466         int s;
 467         arg.found = arg.killed = 0;
 468         arg.rnh = rnh;
 469         arg.nextstop = 0;
 470         arg.draining = 1;
 471         arg.updating = 0;
 472         s = splnet();
 473         rnh->rnh_walktree(rnh, in6_rtqkill, &arg);
 474         splx(s);
 475 }
 476 #endif
 477
 478 /*
 479  * Initialize our routing tree.
 480  */
 481 int
 482 in6_inithead(void **head, int off)
 483 {
 484         struct radix_node_head *rnh;
 485
 486         if (!rn_inithead(head, off))
 487                 return 0;
 488
 489         if (head != (void **)&rt_tables[AF_INET6]) /* BOGUS! */
 490                 return 1;       /* only do this for the real routing table */
 491
 492         rnh = *head;
 493         rnh->rnh_addaddr = in6_addroute;
 494         rnh->rnh_matchaddr = in6_matroute;
 495         rnh->rnh_close = in6_clsroute;
 496         in6_rtqtimo(rnh);       /* kick off timeout first time */
 497         in6_mtutimo(rnh);       /* kick off timeout first time */
 498         return 1;
 499 }