bsd/netinet/in_rmx.c

   1 /*
   2  * Copyright (c) 2000 Apple Computer, Inc. All rights reserved.
   3  *
   4  * @APPLE_LICENSE_HEADER_START@
   5  *
   6  * The contents of this file constitute Original Code as defined in and
   7  * are subject to the Apple Public Source License Version 1.1 (the
   8  * "License").  You may not use this file except in compliance with the
   9  * License.  Please obtain a copy of the License at
  10  * http://www.apple.com/publicsource and read it before using this file.
  11  *
  12  * This Original Code and all software distributed under the License are
  13  * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  14  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  15  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  16  * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT.  Please see the
  17  * License for the specific language governing rights and limitations
  18  * under the License.
  19  *
  20  * @APPLE_LICENSE_HEADER_END@
  21  */
  22 /*
  23  * Copyright 1994, 1995 Massachusetts Institute of Technology
  24  *
  25  * Permission to use, copy, modify, and distribute this software and
  26  * its documentation for any purpose and without fee is hereby
  27  * granted, provided that both the above copyright notice and this
  28  * permission notice appear in all copies, that both the above
  29  * copyright notice and this permission notice appear in all
  30  * supporting documentation, and that the name of M.I.T. not be used
  31  * in advertising or publicity pertaining to distribution of the
  32  * software without specific, written prior permission.  M.I.T. makes
  33  * no representations about the suitability of this software for any
  34  * purpose.  It is provided "as is" without express or implied
  35  * warranty.
  36  *
  37  * THIS SOFTWARE IS PROVIDED BY M.I.T. ``AS IS''.  M.I.T. DISCLAIMS
  38  * ALL EXPRESS OR IMPLIED WARRANTIES WITH REGARD TO THIS SOFTWARE,
  39  * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  40  * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. IN NO EVENT
  41  * SHALL M.I.T. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  42  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  43  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF
  44  * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
  45  * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  46  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  47  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  48  * SUCH DAMAGE.
  49  *
  50  * $FreeBSD: src/sys/netinet/in_rmx.c,v 1.37.2.1 2001/05/14 08:23:49 ru Exp $
  51  */
  52
  53 /*
  54  * This code does two things necessary for the enhanced TCP metrics to
  55  * function in a useful manner:
  56  *  1) It marks all non-host routes as `cloning', thus ensuring that
  57  *     every actual reference to such a route actually gets turned
  58  *     into a reference to a host route to the specific destination
  59  *     requested.
  60  *  2) When such routes lose all their references, it arranges for them
  61  *     to be deleted in some random collection of circumstances, so that
  62  *     a large quantity of stale routing data is not kept in kernel memory
  63  *     indefinitely.  See in_rtqtimo() below for the exact mechanism.
  64  */
  65
  66 #include <sys/param.h>
  67 #include <sys/systm.h>
  68 #include <sys/kernel.h>
  69 #include <sys/sysctl.h>
  70 #include <sys/socket.h>
  71 #include <sys/mbuf.h>
  72 #include <sys/syslog.h>
  73
  74 #include <net/if.h>
  75 #include <net/route.h>
  76 #include <netinet/in.h>
  77 #include <netinet/in_var.h>
  78
  79 extern int      in_inithead __P((void **head, int off));
  80
  81 #ifdef __APPLE__
  82 static void in_rtqtimo(void *rock);
  83 #endif
  84
  85 #define RTPRF_OURS              RTF_PROTO3      /* set on routes we manage */
  86
  87 /*
  88  * Do what we need to do when inserting a route.
  89  */
  90 static struct radix_node *
  91 in_addroute(void *v_arg, void *n_arg, struct radix_node_head *head,
  92             struct radix_node *treenodes)
  93 {
  94         struct rtentry *rt = (struct rtentry *)treenodes;
  95         struct sockaddr_in *sin = (struct sockaddr_in *)rt_key(rt);
  96         struct radix_node *ret;
  97
  98         /*
  99          * For IP, all unicast non-host routes are automatically cloning.
 100          */
 101         if(IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
 102                 rt->rt_flags |= RTF_MULTICAST;
 103
 104         if(!(rt->rt_flags & (RTF_HOST | RTF_CLONING | RTF_MULTICAST))) {
 105                 rt->rt_flags |= RTF_PRCLONING;
 106         }
 107
 108         /*
 109          * A little bit of help for both IP output and input:
 110          *   For host routes, we make sure that RTF_BROADCAST
 111          *   is set for anything that looks like a broadcast address.
 112          *   This way, we can avoid an expensive call to in_broadcast()
 113          *   in ip_output() most of the time (because the route passed
 114          *   to ip_output() is almost always a host route).
 115          *
 116          *   We also do the same for local addresses, with the thought
 117          *   that this might one day be used to speed up ip_input().
 118          *
 119          * We also mark routes to multicast addresses as such, because
 120          * it's easy to do and might be useful (but this is much more
 121          * dubious since it's so easy to inspect the address).  (This
 122          * is done above.)
 123          */
 124         if (rt->rt_flags & RTF_HOST) {
 125                 if (in_broadcast(sin->sin_addr, rt->rt_ifp)) {
 126                         rt->rt_flags |= RTF_BROADCAST;
 127                 } else {
 128 #define satosin(sa) ((struct sockaddr_in *)sa)
 129                         if (satosin(rt->rt_ifa->ifa_addr)->sin_addr.s_addr
 130                             == sin->sin_addr.s_addr)
 131                                 rt->rt_flags |= RTF_LOCAL;
 132 #undef satosin
 133                 }
 134         }
 135
 136         if (!rt->rt_rmx.rmx_mtu && !(rt->rt_rmx.rmx_locks & RTV_MTU)
 137             && rt->rt_ifp)
 138                 rt->rt_rmx.rmx_mtu = rt->rt_ifp->if_mtu;
 139
 140         ret = rn_addroute(v_arg, n_arg, head, treenodes);
 141         if (ret == NULL && rt->rt_flags & RTF_HOST) {
 142                 struct rtentry *rt2;
 143                 /*
 144                  * We are trying to add a host route, but can't.
 145                  * Find out if it is because of an
 146                  * ARP entry and delete it if so.
 147                  */
 148                 rt2 = rtalloc1((struct sockaddr *)sin, 0,
 149                                 RTF_CLONING | RTF_PRCLONING);
 150                 if (rt2) {
 151                         if (rt2->rt_flags & RTF_LLINFO &&
 152                                 rt2->rt_flags & RTF_HOST &&
 153                                 rt2->rt_gateway &&
 154                                 rt2->rt_gateway->sa_family == AF_LINK) {
 155                                 rtrequest(RTM_DELETE,
 156                                           (struct sockaddr *)rt_key(rt2),
 157                                           rt2->rt_gateway,
 158                                           rt_mask(rt2), rt2->rt_flags, 0);
 159                                 ret = rn_addroute(v_arg, n_arg, head,
 160                                         treenodes);
 161                         }
 162                         rtfree(rt2);
 163                 }
 164         }
 165         return ret;
 166 }
 167
 168 /*
 169  * This code is the inverse of in_clsroute: on first reference, if we
 170  * were managing the route, stop doing so and set the expiration timer
 171  * back off again.
 172  */
 173 static struct radix_node *
 174 in_matroute(void *v_arg, struct radix_node_head *head)
 175 {
 176         struct radix_node *rn = rn_match(v_arg, head);
 177         struct rtentry *rt = (struct rtentry *)rn;
 178
 179         if(rt && rt->rt_refcnt == 0) { /* this is first reference */
 180                 if(rt->rt_flags & RTPRF_OURS) {
 181                         rt->rt_flags &= ~RTPRF_OURS;
 182                         rt->rt_rmx.rmx_expire = 0;
 183                 }
 184         }
 185         return rn;
 186 }
 187
 188 static int rtq_reallyold = 60*60;
 189         /* one hour is ``really old'' */
 190 SYSCTL_INT(_net_inet_ip, IPCTL_RTEXPIRE, rtexpire, CTLFLAG_RW,
 191     &rtq_reallyold , 0,
 192     "Default expiration time on dynamically learned routes");
 193
 194 static int rtq_minreallyold = 10;
 195         /* never automatically crank down to less */
 196 SYSCTL_INT(_net_inet_ip, IPCTL_RTMINEXPIRE, rtminexpire, CTLFLAG_RW,
 197     &rtq_minreallyold , 0,
 198     "Minimum time to attempt to hold onto dynamically learned routes");
 199
 200 static int rtq_toomany = 128;
 201         /* 128 cached routes is ``too many'' */
 202 SYSCTL_INT(_net_inet_ip, IPCTL_RTMAXCACHE, rtmaxcache, CTLFLAG_RW,
 203     &rtq_toomany , 0, "Upper limit on dynamically learned routes");
 204
 205 #ifdef __APPLE__
 206 /* XXX LD11JUL02 Special case for AOL 5.1.2 connectivity issue to AirPort BS (Radar 2969954)
 207  * AOL is adding a circular route ("10.0.1.1/32 10.0.1.1") when establishing its ppp tunnel
 208  * to the AP BaseStation by removing the default gateway and replacing it with their tunnel entry point.
 209  * There is no apparent reason to add this route as there is a valid 10.0.1.1/24 route to the BS.
 210  * That circular route was ignored on previous version of MacOS X because of a routing bug
 211  * corrected with the merge to FreeBSD4.4 (a route generated from an RTF_CLONING route had the RTF_WASCLONED
 212  * flag set but did not have a reference to the parent route) and that entry was left in the RT. This workaround is
 213  * made in order to provide binary compatibility with AOL.
 214  * If we catch a process adding a circular route with a /32 from the routing socket, we error it out instead of
 215  * confusing the routing table with a wrong route to the previous default gateway
 216  * If for some reason a circular route is needed, turn this sysctl (net.inet.ip.check_route_selfref) to zero.
 217  */
 218 int check_routeselfref = 1;
 219 SYSCTL_INT(_net_inet_ip, OID_AUTO, check_route_selfref, CTLFLAG_RW,
 220     &check_routeselfref , 0, "");
 221 #endif
 222
 223 __private_extern__ int use_routegenid = 1;
 224 SYSCTL_INT(_net_inet_ip, OID_AUTO, use_route_genid, CTLFLAG_RW,
 225     &use_routegenid , 0, "");
 226
 227 /*
 228  * On last reference drop, mark the route as belong to us so that it can be
 229  * timed out.
 230  */
 231 static void
 232 in_clsroute(struct radix_node *rn, struct radix_node_head *head)
 233 {
 234         struct rtentry *rt = (struct rtentry *)rn;
 235
 236         if(!(rt->rt_flags & RTF_UP))
 237                 return;         /* prophylactic measures */
 238
 239         if((rt->rt_flags & (RTF_LLINFO | RTF_HOST)) != RTF_HOST)
 240                 return;
 241
 242         if((rt->rt_flags & (RTF_WASCLONED | RTPRF_OURS))
 243            != RTF_WASCLONED)
 244                 return;
 245
 246         /*
 247          * As requested by David Greenman:
 248          * If rtq_reallyold is 0, just delete the route without
 249          * waiting for a timeout cycle to kill it.
 250          */
 251         if(rtq_reallyold != 0) {
 252                 rt->rt_flags |= RTPRF_OURS;
 253                 rt->rt_rmx.rmx_expire = time_second + rtq_reallyold;
 254         } else {
 255                 rtrequest(RTM_DELETE,
 256                           (struct sockaddr *)rt_key(rt),
 257                           rt->rt_gateway, rt_mask(rt),
 258                           rt->rt_flags, 0);
 259         }
 260 }
 261
 262 struct rtqk_arg {
 263         struct radix_node_head *rnh;
 264         int draining;
 265         int killed;
 266         int found;
 267         int updating;
 268         time_t nextstop;
 269 };
 270
 271 /*
 272  * Get rid of old routes.  When draining, this deletes everything, even when
 273  * the timeout is not expired yet.  When updating, this makes sure that
 274  * nothing has a timeout longer than the current value of rtq_reallyold.
 275  */
 276 static int
 277 in_rtqkill(struct radix_node *rn, void *rock)
 278 {
 279         struct rtqk_arg *ap = rock;
 280         struct rtentry *rt = (struct rtentry *)rn;
 281         int err;
 282
 283         if(rt->rt_flags & RTPRF_OURS) {
 284                 ap->found++;
 285
 286                 if(ap->draining || rt->rt_rmx.rmx_expire <= time_second) {
 287                         if(rt->rt_refcnt > 0)
 288                                 panic("rtqkill route really not free");
 289
 290                         err = rtrequest(RTM_DELETE,
 291                                         (struct sockaddr *)rt_key(rt),
 292                                         rt->rt_gateway, rt_mask(rt),
 293                                         rt->rt_flags, 0);
 294                         if(err) {
 295                                 log(LOG_WARNING, "in_rtqkill: error %d\n", err);
 296                         } else {
 297                                 ap->killed++;
 298                         }
 299                 } else {
 300                         if(ap->updating
 301                            && (rt->rt_rmx.rmx_expire - time_second
 302                                > rtq_reallyold)) {
 303                                 rt->rt_rmx.rmx_expire = time_second
 304                                         + rtq_reallyold;
 305                         }
 306                         ap->nextstop = lmin(ap->nextstop,
 307                                             rt->rt_rmx.rmx_expire);
 308                 }
 309         }
 310
 311         return 0;
 312 }
 313
 314 static void
 315 in_rtqtimo_funnel(void *rock)
 316 {
 317         boolean_t       funnel_state;
 318
 319         funnel_state = thread_funnel_set(network_flock, TRUE);
 320         in_rtqtimo(rock);
 321         (void) thread_funnel_set(network_flock, FALSE);
 322
 323 }
 324 #define RTQ_TIMEOUT     60*10   /* run no less than once every ten minutes */
 325 static int rtq_timeout = RTQ_TIMEOUT;
 326
 327 static void
 328 in_rtqtimo(void *rock)
 329 {
 330         struct radix_node_head *rnh = rock;
 331         struct rtqk_arg arg;
 332         struct timeval atv;
 333         static time_t last_adjusted_timeout = 0;
 334         int s;
 335
 336         arg.found = arg.killed = 0;
 337         arg.rnh = rnh;
 338         arg.nextstop = time_second + rtq_timeout;
 339         arg.draining = arg.updating = 0;
 340         s = splnet();
 341         rnh->rnh_walktree(rnh, in_rtqkill, &arg);
 342         splx(s);
 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_timeout seconds, to keep from cranking down too
 350          * hard.
 351          */
 352         if((arg.found - arg.killed > rtq_toomany)
 353            && (time_second - last_adjusted_timeout >= rtq_timeout)
 354            && rtq_reallyold > rtq_minreallyold) {
 355                 rtq_reallyold = 2*rtq_reallyold / 3;
 356                 if(rtq_reallyold < rtq_minreallyold) {
 357                         rtq_reallyold = rtq_minreallyold;
 358                 }
 359
 360                 last_adjusted_timeout = time_second;
 361 #if DIAGNOSTIC
 362                 log(LOG_DEBUG, "in_rtqtimo: adjusted rtq_reallyold to %d\n",
 363                     rtq_reallyold);
 364 #endif
 365                 arg.found = arg.killed = 0;
 366                 arg.updating = 1;
 367                 s = splnet();
 368                 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
 369                 splx(s);
 370         }
 371
 372         atv.tv_usec = 0;
 373         atv.tv_sec = arg.nextstop - time_second;
 374         timeout(in_rtqtimo_funnel, rock, tvtohz(&atv));
 375 }
 376
 377 void
 378 in_rtqdrain(void)
 379 {
 380         struct radix_node_head *rnh = rt_tables[AF_INET];
 381         struct rtqk_arg arg;
 382         int s;
 383         arg.found = arg.killed = 0;
 384         arg.rnh = rnh;
 385         arg.nextstop = 0;
 386         arg.draining = 1;
 387         arg.updating = 0;
 388         s = splnet();
 389         rnh->rnh_walktree(rnh, in_rtqkill, &arg);
 390         splx(s);
 391 }
 392
 393 /*
 394  * Initialize our routing tree.
 395  */
 396 int
 397 in_inithead(void **head, int off)
 398 {
 399         struct radix_node_head *rnh;
 400
 401 #ifdef __APPLE__
 402         if (*head)
 403                 return 1;
 404 #endif
 405
 406         if(!rn_inithead(head, off))
 407                 return 0;
 408
 409         if(head != (void **)&rt_tables[AF_INET]) /* BOGUS! */
 410                 return 1;       /* only do this for the real routing table */
 411
 412         rnh = *head;
 413         rnh->rnh_addaddr = in_addroute;
 414         rnh->rnh_matchaddr = in_matroute;
 415         rnh->rnh_close = in_clsroute;
 416         in_rtqtimo(rnh);        /* kick off timeout first time */
 417         return 1;
 418 }
 419
 420 \f
 421 /*
 422  * This zaps old routes when the interface goes down or interface
 423  * address is deleted.  In the latter case, it deletes static routes
 424  * that point to this address.  If we don't do this, we may end up
 425  * using the old address in the future.  The ones we always want to
 426  * get rid of are things like ARP entries, since the user might down
 427  * the interface, walk over to a completely different network, and
 428  * plug back in.
 429  */
 430 struct in_ifadown_arg {
 431         struct radix_node_head *rnh;
 432         struct ifaddr *ifa;
 433         int del;
 434 };
 435
 436 static int
 437 in_ifadownkill(struct radix_node *rn, void *xap)
 438 {
 439         struct in_ifadown_arg *ap = xap;
 440         struct rtentry *rt = (struct rtentry *)rn;
 441         int err;
 442
 443         if (rt->rt_ifa == ap->ifa &&
 444             (ap->del || !(rt->rt_flags & RTF_STATIC))) {
 445                 /*
 446                  * We need to disable the automatic prune that happens
 447                  * in this case in rtrequest() because it will blow
 448                  * away the pointers that rn_walktree() needs in order
 449                  * continue our descent.  We will end up deleting all
 450                  * the routes that rtrequest() would have in any case,
 451                  * so that behavior is not needed there.
 452                  */
 453                 rt->rt_flags &= ~(RTF_CLONING | RTF_PRCLONING);
 454                 err = rtrequest(RTM_DELETE, (struct sockaddr *)rt_key(rt),
 455                                 rt->rt_gateway, rt_mask(rt), rt->rt_flags, 0);
 456                 if (err) {
 457                         log(LOG_WARNING, "in_ifadownkill: error %d\n", err);
 458                 }
 459         }
 460         return 0;
 461 }
 462
 463 int
 464 in_ifadown(struct ifaddr *ifa, int delete)
 465 {
 466         struct in_ifadown_arg arg;
 467         struct radix_node_head *rnh;
 468
 469         if (ifa->ifa_addr->sa_family != AF_INET)
 470                 return 1;
 471
 472         arg.rnh = rnh = rt_tables[AF_INET];
 473         arg.ifa = ifa;
 474         arg.del = delete;
 475         rnh->rnh_walktree(rnh, in_ifadownkill, &arg);
 476         ifa->ifa_flags &= ~IFA_ROUTE;
 477         return 0;
 478 }