bsd/netinet/in_rmx.c

   1 /*
   2  * Copyright (c) 2000 Apple Computer, Inc. All rights reserved.
   3  *
   4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
   5  *
   6  * This file contains Original Code and/or Modifications of Original Code
   7  * as defined in and that are subject to the Apple Public Source License
   8  * Version 2.0 (the 'License'). You may not use this file except in
   9  * compliance with the License. The rights granted to you under the License
  10  * may not be used to create, or enable the creation or redistribution of,
  11  * unlawful or unlicensed copies of an Apple operating system, or to
  12  * circumvent, violate, or enable the circumvention or violation of, any
  13  * terms of an Apple operating system software license agreement.
  14  *
  15  * Please obtain a copy of the License at
  16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
  17  *
  18  * The Original Code and all software distributed under the License are
  19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  23  * Please see the License for the specific language governing rights and
  24  * limitations under the License.
  25  *
  26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  27  */
  28 /*
  29  * Copyright 1994, 1995 Massachusetts Institute of Technology
  30  *
  31  * Permission to use, copy, modify, and distribute this software and
  32  * its documentation for any purpose and without fee is hereby
  33  * granted, provided that both the above copyright notice and this
  34  * permission notice appear in all copies, that both the above
  35  * copyright notice and this permission notice appear in all
  36  * supporting documentation, and that the name of M.I.T. not be used
  37  * in advertising or publicity pertaining to distribution of the
  38  * software without specific, written prior permission.  M.I.T. makes
  39  * no representations about the suitability of this software for any
  40  * purpose.  It is provided "as is" without express or implied
  41  * warranty.
  42  *
  43  * THIS SOFTWARE IS PROVIDED BY M.I.T. ``AS IS''.  M.I.T. DISCLAIMS
  44  * ALL EXPRESS OR IMPLIED WARRANTIES WITH REGARD TO THIS SOFTWARE,
  45  * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  46  * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. IN NO EVENT
  47  * SHALL M.I.T. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  48  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  49  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF
  50  * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
  51  * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  52  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  53  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  54  * SUCH DAMAGE.
  55  *
  56  * $FreeBSD: src/sys/netinet/in_rmx.c,v 1.37.2.1 2001/05/14 08:23:49 ru Exp $
  57  */
  58
  59 /*
  60  * This code does two things necessary for the enhanced TCP metrics to
  61  * function in a useful manner:
  62  *  1) It marks all non-host routes as `cloning', thus ensuring that
  63  *     every actual reference to such a route actually gets turned
  64  *     into a reference to a host route to the specific destination
  65  *     requested.
  66  *  2) When such routes lose all their references, it arranges for them
  67  *     to be deleted in some random collection of circumstances, so that
  68  *     a large quantity of stale routing data is not kept in kernel memory
  69  *     indefinitely.  See in_rtqtimo() below for the exact mechanism.
  70  */
  71
  72 #include <sys/param.h>
  73 #include <sys/systm.h>
  74 #include <sys/kernel.h>
  75 #include <sys/sysctl.h>
  76 #include <sys/socket.h>
  77 #include <sys/mbuf.h>
  78 #include <sys/syslog.h>
  79 #include <kern/lock.h>
  80
  81 #include <net/if.h>
  82 #include <net/route.h>
  83 #include <netinet/in.h>
  84 #include <netinet/in_var.h>
  85
  86 extern int      in_inithead(void **head, int off);
  87
  88 #ifdef __APPLE__
  89 static void in_rtqtimo(void *rock);
  90 #endif
  91
  92 #define RTPRF_OURS              RTF_PROTO3      /* set on routes we manage */
  93 extern lck_mtx_t *rt_mtx;
  94
  95 /*
  96  * Do what we need to do when inserting a route.
  97  */
  98 static struct radix_node *
  99 in_addroute(void *v_arg, void *n_arg, struct radix_node_head *head,
 100             struct radix_node *treenodes)
 101 {
 102         struct rtentry *rt = (struct rtentry *)treenodes;
 103         struct sockaddr_in *sin = (struct sockaddr_in *)rt_key(rt);
 104         struct radix_node *ret;
 105
 106         /*
 107          * For IP, all unicast non-host routes are automatically cloning.
 108          */
 109         if(IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
 110                 rt->rt_flags |= RTF_MULTICAST;
 111
 112         if(!(rt->rt_flags & (RTF_HOST | RTF_CLONING | RTF_MULTICAST))) {
 113                 rt->rt_flags |= RTF_PRCLONING;
 114         }
 115
 116         /*
 117          * A little bit of help for both IP output and input:
 118          *   For host routes, we make sure that RTF_BROADCAST
 119          *   is set for anything that looks like a broadcast address.
 120          *   This way, we can avoid an expensive call to in_broadcast()
 121          *   in ip_output() most of the time (because the route passed
 122          *   to ip_output() is almost always a host route).
 123          *
 124          *   We also do the same for local addresses, with the thought
 125          *   that this might one day be used to speed up ip_input().
 126          *
 127          * We also mark routes to multicast addresses as such, because
 128          * it's easy to do and might be useful (but this is much more
 129          * dubious since it's so easy to inspect the address).  (This
 130          * is done above.)
 131          */
 132         if (rt->rt_flags & RTF_HOST) {
 133                 if (in_broadcast(sin->sin_addr, rt->rt_ifp)) {
 134                         rt->rt_flags |= RTF_BROADCAST;
 135                 } else {
 136 #define satosin(sa) ((struct sockaddr_in *)sa)
 137                         if (satosin(rt->rt_ifa->ifa_addr)->sin_addr.s_addr
 138                             == sin->sin_addr.s_addr)
 139                                 rt->rt_flags |= RTF_LOCAL;
 140 #undef satosin
 141                 }
 142         }
 143
 144         if (!rt->rt_rmx.rmx_mtu && !(rt->rt_rmx.rmx_locks & RTV_MTU)
 145             && rt->rt_ifp)
 146                 rt->rt_rmx.rmx_mtu = rt->rt_ifp->if_mtu;
 147
 148         ret = rn_addroute(v_arg, n_arg, head, treenodes);
 149         if (ret == NULL && rt->rt_flags & RTF_HOST) {
 150                 struct rtentry *rt2;
 151                 /*
 152                  * We are trying to add a host route, but can't.
 153                  * Find out if it is because of an
 154                  * ARP entry and delete it if so.
 155                  */
 156                 rt2 = rtalloc1_locked((struct sockaddr *)sin, 0,
 157                                 RTF_CLONING | RTF_PRCLONING);
 158                 if (rt2) {
 159                         if (rt2->rt_flags & RTF_LLINFO &&
 160                                 rt2->rt_flags & RTF_HOST &&
 161                                 rt2->rt_gateway &&
 162                                 rt2->rt_gateway->sa_family == AF_LINK) {
 163                                 rtrequest_locked(RTM_DELETE,
 164                                           (struct sockaddr *)rt_key(rt2),
 165                                           rt2->rt_gateway,
 166                                           rt_mask(rt2), rt2->rt_flags, 0);
 167                                 ret = rn_addroute(v_arg, n_arg, head,
 168                                         treenodes);
 169                         }
 170                         rtfree_locked(rt2);
 171                 }
 172         }
 173         return ret;
 174 }
 175
 176 /*
 177  * This code is the inverse of in_clsroute: on first reference, if we
 178  * were managing the route, stop doing so and set the expiration timer
 179  * back off again.
 180  */
 181 static struct radix_node *
 182 in_matroute(void *v_arg, struct radix_node_head *head)
 183 {
 184         struct radix_node *rn = rn_match(v_arg, head);
 185         struct rtentry *rt = (struct rtentry *)rn;
 186
 187         if(rt && rt->rt_refcnt == 0) { /* this is first reference */
 188                 if(rt->rt_flags & RTPRF_OURS) {
 189                         rt->rt_flags &= ~RTPRF_OURS;
 190                         rt->rt_rmx.rmx_expire = 0;
 191                 }
 192         }
 193         return rn;
 194 }
 195
 196 static int rtq_reallyold = 60*60;
 197         /* one hour is ``really old'' */
 198 SYSCTL_INT(_net_inet_ip, IPCTL_RTEXPIRE, rtexpire, CTLFLAG_RW,
 199     &rtq_reallyold , 0,
 200     "Default expiration time on dynamically learned routes");
 201
 202 static int rtq_minreallyold = 10;
 203         /* never automatically crank down to less */
 204 SYSCTL_INT(_net_inet_ip, IPCTL_RTMINEXPIRE, rtminexpire, CTLFLAG_RW,
 205     &rtq_minreallyold , 0,
 206     "Minimum time to attempt to hold onto dynamically learned routes");
 207
 208 static int rtq_toomany = 128;
 209         /* 128 cached routes is ``too many'' */
 210 SYSCTL_INT(_net_inet_ip, IPCTL_RTMAXCACHE, rtmaxcache, CTLFLAG_RW,
 211     &rtq_toomany , 0, "Upper limit on dynamically learned routes");
 212
 213 #ifdef __APPLE__
 214 /* XXX LD11JUL02 Special case for AOL 5.1.2 connectivity issue to AirPort BS (Radar 2969954)
 215  * AOL is adding a circular route ("10.0.1.1/32 10.0.1.1") when establishing its ppp tunnel
 216  * to the AP BaseStation by removing the default gateway and replacing it with their tunnel entry point.
 217  * There is no apparent reason to add this route as there is a valid 10.0.1.1/24 route to the BS.
 218  * That circular route was ignored on previous version of MacOS X because of a routing bug
 219  * corrected with the merge to FreeBSD4.4 (a route generated from an RTF_CLONING route had the RTF_WASCLONED
 220  * flag set but did not have a reference to the parent route) and that entry was left in the RT. This workaround is
 221  * made in order to provide binary compatibility with AOL.
 222  * If we catch a process adding a circular route with a /32 from the routing socket, we error it out instead of
 223  * confusing the routing table with a wrong route to the previous default gateway
 224  * If for some reason a circular route is needed, turn this sysctl (net.inet.ip.check_route_selfref) to zero.
 225  */
 226 int check_routeselfref = 1;
 227 SYSCTL_INT(_net_inet_ip, OID_AUTO, check_route_selfref, CTLFLAG_RW,
 228     &check_routeselfref , 0, "");
 229 #endif
 230
 231 __private_extern__ int use_routegenid = 1;
 232 SYSCTL_INT(_net_inet_ip, OID_AUTO, use_route_genid, CTLFLAG_RW,
 233     &use_routegenid , 0, "");
 234
 235 /*
 236  * On last reference drop, mark the route as belong to us so that it can be
 237  * timed out.
 238  */
 239 static void
 240 in_clsroute(struct radix_node *rn, struct radix_node_head *head)
 241 {
 242         struct rtentry *rt = (struct rtentry *)rn;
 243         struct timeval timenow;
 244
 245         if(!(rt->rt_flags & RTF_UP))
 246                 return;         /* prophylactic measures */
 247
 248         if((rt->rt_flags & (RTF_LLINFO | RTF_HOST)) != RTF_HOST)
 249                 return;
 250
 251         if((rt->rt_flags & (RTF_WASCLONED | RTPRF_OURS))
 252            != RTF_WASCLONED)
 253                 return;
 254
 255         /*
 256          * As requested by David Greenman:
 257          * If rtq_reallyold is 0, just delete the route without
 258          * waiting for a timeout cycle to kill it.
 259          */
 260         if(rtq_reallyold != 0) {
 261                 getmicrotime(&timenow);
 262                 rt->rt_flags |= RTPRF_OURS;
 263                 rt->rt_rmx.rmx_expire = timenow.tv_sec + rtq_reallyold;
 264         } else {
 265                 rtrequest_locked(RTM_DELETE,
 266                           (struct sockaddr *)rt_key(rt),
 267                           rt->rt_gateway, rt_mask(rt),
 268                           rt->rt_flags, 0);
 269         }
 270 }
 271
 272 struct rtqk_arg {
 273         struct radix_node_head *rnh;
 274         int draining;
 275         int killed;
 276         int found;
 277         int updating;
 278         time_t nextstop;
 279 };
 280
 281 /*
 282  * Get rid of old routes.  When draining, this deletes everything, even when
 283  * the timeout is not expired yet.  When updating, this makes sure that
 284  * nothing has a timeout longer than the current value of rtq_reallyold.
 285  */
 286 static int
 287 in_rtqkill(struct radix_node *rn, void *rock)
 288 {
 289         struct rtqk_arg *ap = rock;
 290         struct rtentry *rt = (struct rtentry *)rn;
 291         int err;
 292         struct timeval timenow;
 293
 294         getmicrotime(&timenow);
 295         lck_mtx_assert(rt_mtx, LCK_MTX_ASSERT_OWNED);
 296         if(rt->rt_flags & RTPRF_OURS) {
 297                 ap->found++;
 298
 299                 if(ap->draining || rt->rt_rmx.rmx_expire <= timenow.tv_sec) {
 300                         if(rt->rt_refcnt > 0)
 301                                 panic("rtqkill route really not free");
 302
 303                         err = rtrequest_locked(RTM_DELETE,
 304                                         (struct sockaddr *)rt_key(rt),
 305                                         rt->rt_gateway, rt_mask(rt),
 306                                         rt->rt_flags, 0);
 307                         if(err) {
 308                                 log(LOG_WARNING, "in_rtqkill: error %d\n", err);
 309                         } else {
 310                                 ap->killed++;
 311                         }
 312                 } else {
 313                         if(ap->updating
 314                            && (rt->rt_rmx.rmx_expire - timenow.tv_sec
 315                                > rtq_reallyold)) {
 316                                 rt->rt_rmx.rmx_expire = timenow.tv_sec
 317                                         + rtq_reallyold;
 318                         }
 319                         ap->nextstop = lmin(ap->nextstop,
 320                                             rt->rt_rmx.rmx_expire);
 321                 }
 322         }
 323
 324         return 0;
 325 }
 326
 327 static void
 328 in_rtqtimo_funnel(void *rock)
 329 {
 330         in_rtqtimo(rock);
 331
 332 }
 333 #define RTQ_TIMEOUT     60*10   /* run no less than once every ten minutes */
 334 static int rtq_timeout = RTQ_TIMEOUT;
 335
 336 static void
 337 in_rtqtimo(void *rock)
 338 {
 339         struct radix_node_head *rnh = rock;
 340         struct rtqk_arg arg;
 341         struct timeval atv;
 342         static time_t last_adjusted_timeout = 0;
 343         struct timeval timenow;
 344
 345         getmicrotime(&timenow);
 346         arg.found = arg.killed = 0;
 347         arg.rnh = rnh;
 348         arg.nextstop = timenow.tv_sec + rtq_timeout;
 349         arg.draining = arg.updating = 0;
 350         lck_mtx_lock(rt_mtx);
 351         rnh->rnh_walktree(rnh, in_rtqkill, &arg);
 352
 353         /*
 354          * Attempt to be somewhat dynamic about this:
 355          * If there are ``too many'' routes sitting around taking up space,
 356          * then crank down the timeout, and see if we can't make some more
 357          * go away.  However, we make sure that we will never adjust more
 358          * than once in rtq_timeout seconds, to keep from cranking down too
 359          * hard.
 360          */
 361         if((arg.found - arg.killed > rtq_toomany)
 362            && (timenow.tv_sec - last_adjusted_timeout >= rtq_timeout)
 363            && rtq_reallyold > rtq_minreallyold) {
 364                 rtq_reallyold = 2*rtq_reallyold / 3;
 365                 if(rtq_reallyold < rtq_minreallyold) {
 366                         rtq_reallyold = rtq_minreallyold;
 367                 }
 368
 369                 last_adjusted_timeout = timenow.tv_sec;
 370 #if DIAGNOSTIC
 371                 log(LOG_DEBUG, "in_rtqtimo: adjusted rtq_reallyold to %d\n",
 372                     rtq_reallyold);
 373 #endif
 374                 arg.found = arg.killed = 0;
 375                 arg.updating = 1;
 376                 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
 377         }
 378
 379         atv.tv_usec = 0;
 380         atv.tv_sec = arg.nextstop - timenow.tv_sec;
 381         lck_mtx_unlock(rt_mtx);
 382         timeout(in_rtqtimo_funnel, rock, tvtohz(&atv));
 383 }
 384
 385 void
 386 in_rtqdrain(void)
 387 {
 388         struct radix_node_head *rnh = rt_tables[AF_INET];
 389         struct rtqk_arg arg;
 390         arg.found = arg.killed = 0;
 391         arg.rnh = rnh;
 392         arg.nextstop = 0;
 393         arg.draining = 1;
 394         arg.updating = 0;
 395         lck_mtx_lock(rt_mtx);
 396         rnh->rnh_walktree(rnh, in_rtqkill, &arg);
 397         lck_mtx_unlock(rt_mtx);
 398 }
 399
 400 /*
 401  * Initialize our routing tree.
 402  */
 403 int
 404 in_inithead(void **head, int off)
 405 {
 406         struct radix_node_head *rnh;
 407
 408 #ifdef __APPLE__
 409         if (*head)
 410                 return 1;
 411 #endif
 412
 413         if(!rn_inithead(head, off))
 414                 return 0;
 415
 416         if(head != (void **)&rt_tables[AF_INET]) /* BOGUS! */
 417                 return 1;       /* only do this for the real routing table */
 418
 419         rnh = *head;
 420         rnh->rnh_addaddr = in_addroute;
 421         rnh->rnh_matchaddr = in_matroute;
 422         rnh->rnh_close = in_clsroute;
 423         in_rtqtimo(rnh);        /* kick off timeout first time */
 424         return 1;
 425 }
 426
 427 \f
 428 /*
 429  * This zaps old routes when the interface goes down or interface
 430  * address is deleted.  In the latter case, it deletes static routes
 431  * that point to this address.  If we don't do this, we may end up
 432  * using the old address in the future.  The ones we always want to
 433  * get rid of are things like ARP entries, since the user might down
 434  * the interface, walk over to a completely different network, and
 435  * plug back in.
 436  */
 437 struct in_ifadown_arg {
 438         struct radix_node_head *rnh;
 439         struct ifaddr *ifa;
 440         int del;
 441 };
 442
 443 static int
 444 in_ifadownkill(struct radix_node *rn, void *xap)
 445 {
 446         struct in_ifadown_arg *ap = xap;
 447         struct rtentry *rt = (struct rtentry *)rn;
 448         int err;
 449
 450         if (rt->rt_ifa == ap->ifa &&
 451             (ap->del || !(rt->rt_flags & RTF_STATIC))) {
 452                 /*
 453                  * We need to disable the automatic prune that happens
 454                  * in this case in rtrequest() because it will blow
 455                  * away the pointers that rn_walktree() needs in order
 456                  * continue our descent.  We will end up deleting all
 457                  * the routes that rtrequest() would have in any case,
 458                  * so that behavior is not needed there.
 459                  */
 460                 rt->rt_flags &= ~(RTF_CLONING | RTF_PRCLONING);
 461                 err = rtrequest_locked(RTM_DELETE, (struct sockaddr *)rt_key(rt),
 462                                 rt->rt_gateway, rt_mask(rt), rt->rt_flags, 0);
 463                 if (err) {
 464                         log(LOG_WARNING, "in_ifadownkill: error %d\n", err);
 465                 }
 466         }
 467         return 0;
 468 }
 469
 470 int
 471 in_ifadown(struct ifaddr *ifa, int delete)
 472 {
 473         struct in_ifadown_arg arg;
 474         struct radix_node_head *rnh;
 475
 476         lck_mtx_assert(rt_mtx, LCK_MTX_ASSERT_OWNED);
 477
 478         if (ifa->ifa_addr->sa_family != AF_INET)
 479                 return 1;
 480
 481         arg.rnh = rnh = rt_tables[AF_INET];
 482         arg.ifa = ifa;
 483         arg.del = delete;
 484         rnh->rnh_walktree(rnh, in_ifadownkill, &arg);
 485         ifa->ifa_flags &= ~IFA_ROUTE;
 486         return 0;
 487 }