bsd/netinet/in_rmx.c

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