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 #include <kern/lock.h>
  74
  75 #include <net/if.h>
  76 #include <net/route.h>
  77 #include <netinet/in.h>
  78 #include <netinet/in_var.h>
  79
  80 extern int      in_inithead(void **head, int off);
  81
  82 #ifdef __APPLE__
  83 static void in_rtqtimo(void *rock);
  84 #endif
  85
  86 #define RTPRF_OURS              RTF_PROTO3      /* set on routes we manage */
  87 extern lck_mtx_t *rt_mtx;
  88
  89 /*
  90  * Do what we need to do when inserting a route.
  91  */
  92 static struct radix_node *
  93 in_addroute(void *v_arg, void *n_arg, struct radix_node_head *head,
  94             struct radix_node *treenodes)
  95 {
  96         struct rtentry *rt = (struct rtentry *)treenodes;
  97         struct sockaddr_in *sin = (struct sockaddr_in *)rt_key(rt);
  98         struct radix_node *ret;
  99
 100         /*
 101          * For IP, all unicast non-host routes are automatically cloning.
 102          */
 103         if(IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
 104                 rt->rt_flags |= RTF_MULTICAST;
 105
 106         if(!(rt->rt_flags & (RTF_HOST | RTF_CLONING | RTF_MULTICAST))) {
 107                 rt->rt_flags |= RTF_PRCLONING;
 108         }
 109
 110         /*
 111          * A little bit of help for both IP output and input:
 112          *   For host routes, we make sure that RTF_BROADCAST
 113          *   is set for anything that looks like a broadcast address.
 114          *   This way, we can avoid an expensive call to in_broadcast()
 115          *   in ip_output() most of the time (because the route passed
 116          *   to ip_output() is almost always a host route).
 117          *
 118          *   We also do the same for local addresses, with the thought
 119          *   that this might one day be used to speed up ip_input().
 120          *
 121          * We also mark routes to multicast addresses as such, because
 122          * it's easy to do and might be useful (but this is much more
 123          * dubious since it's so easy to inspect the address).  (This
 124          * is done above.)
 125          */
 126         if (rt->rt_flags & RTF_HOST) {
 127                 if (in_broadcast(sin->sin_addr, rt->rt_ifp)) {
 128                         rt->rt_flags |= RTF_BROADCAST;
 129                 } else {
 130 #define satosin(sa) ((struct sockaddr_in *)sa)
 131                         if (satosin(rt->rt_ifa->ifa_addr)->sin_addr.s_addr
 132                             == sin->sin_addr.s_addr)
 133                                 rt->rt_flags |= RTF_LOCAL;
 134 #undef satosin
 135                 }
 136         }
 137
 138         if (!rt->rt_rmx.rmx_mtu && !(rt->rt_rmx.rmx_locks & RTV_MTU)
 139             && rt->rt_ifp)
 140                 rt->rt_rmx.rmx_mtu = rt->rt_ifp->if_mtu;
 141
 142         ret = rn_addroute(v_arg, n_arg, head, treenodes);
 143         if (ret == NULL && rt->rt_flags & RTF_HOST) {
 144                 struct rtentry *rt2;
 145                 /*
 146                  * We are trying to add a host route, but can't.
 147                  * Find out if it is because of an
 148                  * ARP entry and delete it if so.
 149                  */
 150                 rt2 = rtalloc1_locked((struct sockaddr *)sin, 0,
 151                                 RTF_CLONING | RTF_PRCLONING);
 152                 if (rt2) {
 153                         if (rt2->rt_flags & RTF_LLINFO &&
 154                                 rt2->rt_flags & RTF_HOST &&
 155                                 rt2->rt_gateway &&
 156                                 rt2->rt_gateway->sa_family == AF_LINK) {
 157                                 rtrequest_locked(RTM_DELETE,
 158                                           (struct sockaddr *)rt_key(rt2),
 159                                           rt2->rt_gateway,
 160                                           rt_mask(rt2), rt2->rt_flags, 0);
 161                                 ret = rn_addroute(v_arg, n_arg, head,
 162                                         treenodes);
 163                         }
 164                         rtfree_locked(rt2);
 165                 }
 166         }
 167         return ret;
 168 }
 169
 170 /*
 171  * This code is the inverse of in_clsroute: on first reference, if we
 172  * were managing the route, stop doing so and set the expiration timer
 173  * back off again.
 174  */
 175 static struct radix_node *
 176 in_matroute(void *v_arg, struct radix_node_head *head)
 177 {
 178         struct radix_node *rn = rn_match(v_arg, head);
 179         struct rtentry *rt = (struct rtentry *)rn;
 180
 181         if(rt && rt->rt_refcnt == 0) { /* this is first reference */
 182                 if(rt->rt_flags & RTPRF_OURS) {
 183                         rt->rt_flags &= ~RTPRF_OURS;
 184                         rt->rt_rmx.rmx_expire = 0;
 185                 }
 186         }
 187         return rn;
 188 }
 189
 190 static int rtq_reallyold = 60*60;
 191         /* one hour is ``really old'' */
 192 SYSCTL_INT(_net_inet_ip, IPCTL_RTEXPIRE, rtexpire, CTLFLAG_RW,
 193     &rtq_reallyold , 0,
 194     "Default expiration time on dynamically learned routes");
 195
 196 static int rtq_minreallyold = 10;
 197         /* never automatically crank down to less */
 198 SYSCTL_INT(_net_inet_ip, IPCTL_RTMINEXPIRE, rtminexpire, CTLFLAG_RW,
 199     &rtq_minreallyold , 0,
 200     "Minimum time to attempt to hold onto dynamically learned routes");
 201
 202 static int rtq_toomany = 128;
 203         /* 128 cached routes is ``too many'' */
 204 SYSCTL_INT(_net_inet_ip, IPCTL_RTMAXCACHE, rtmaxcache, CTLFLAG_RW,
 205     &rtq_toomany , 0, "Upper limit on dynamically learned routes");
 206
 207 #ifdef __APPLE__
 208 /* XXX LD11JUL02 Special case for AOL 5.1.2 connectivity issue to AirPort BS (Radar 2969954)
 209  * AOL is adding a circular route ("10.0.1.1/32 10.0.1.1") when establishing its ppp tunnel
 210  * to the AP BaseStation by removing the default gateway and replacing it with their tunnel entry point.
 211  * There is no apparent reason to add this route as there is a valid 10.0.1.1/24 route to the BS.
 212  * That circular route was ignored on previous version of MacOS X because of a routing bug
 213  * corrected with the merge to FreeBSD4.4 (a route generated from an RTF_CLONING route had the RTF_WASCLONED
 214  * flag set but did not have a reference to the parent route) and that entry was left in the RT. This workaround is
 215  * made in order to provide binary compatibility with AOL.
 216  * If we catch a process adding a circular route with a /32 from the routing socket, we error it out instead of
 217  * confusing the routing table with a wrong route to the previous default gateway
 218  * If for some reason a circular route is needed, turn this sysctl (net.inet.ip.check_route_selfref) to zero.
 219  */
 220 int check_routeselfref = 1;
 221 SYSCTL_INT(_net_inet_ip, OID_AUTO, check_route_selfref, CTLFLAG_RW,
 222     &check_routeselfref , 0, "");
 223 #endif
 224
 225 __private_extern__ int use_routegenid = 1;
 226 SYSCTL_INT(_net_inet_ip, OID_AUTO, use_route_genid, CTLFLAG_RW,
 227     &use_routegenid , 0, "");
 228
 229 /*
 230  * On last reference drop, mark the route as belong to us so that it can be
 231  * timed out.
 232  */
 233 static void
 234 in_clsroute(struct radix_node *rn, struct radix_node_head *head)
 235 {
 236         struct rtentry *rt = (struct rtentry *)rn;
 237         struct timeval timenow;
 238
 239         if(!(rt->rt_flags & RTF_UP))
 240                 return;         /* prophylactic measures */
 241
 242         if((rt->rt_flags & (RTF_LLINFO | RTF_HOST)) != RTF_HOST)
 243                 return;
 244
 245         if((rt->rt_flags & (RTF_WASCLONED | RTPRF_OURS))
 246            != RTF_WASCLONED)
 247                 return;
 248
 249         /*
 250          * As requested by David Greenman:
 251          * If rtq_reallyold is 0, just delete the route without
 252          * waiting for a timeout cycle to kill it.
 253          */
 254         if(rtq_reallyold != 0) {
 255                 getmicrotime(&timenow);
 256                 rt->rt_flags |= RTPRF_OURS;
 257                 rt->rt_rmx.rmx_expire = timenow.tv_sec + rtq_reallyold;
 258         } else {
 259                 rtrequest_locked(RTM_DELETE,
 260                           (struct sockaddr *)rt_key(rt),
 261                           rt->rt_gateway, rt_mask(rt),
 262                           rt->rt_flags, 0);
 263         }
 264 }
 265
 266 struct rtqk_arg {
 267         struct radix_node_head *rnh;
 268         int draining;
 269         int killed;
 270         int found;
 271         int updating;
 272         time_t nextstop;
 273 };
 274
 275 /*
 276  * Get rid of old routes.  When draining, this deletes everything, even when
 277  * the timeout is not expired yet.  When updating, this makes sure that
 278  * nothing has a timeout longer than the current value of rtq_reallyold.
 279  */
 280 static int
 281 in_rtqkill(struct radix_node *rn, void *rock)
 282 {
 283         struct rtqk_arg *ap = rock;
 284         struct rtentry *rt = (struct rtentry *)rn;
 285         int err;
 286         struct timeval timenow;
 287
 288         getmicrotime(&timenow);
 289         lck_mtx_assert(rt_mtx, LCK_MTX_ASSERT_OWNED);
 290         if(rt->rt_flags & RTPRF_OURS) {
 291                 ap->found++;
 292
 293                 if(ap->draining || rt->rt_rmx.rmx_expire <= timenow.tv_sec) {
 294                         if(rt->rt_refcnt > 0)
 295                                 panic("rtqkill route really not free");
 296
 297                         err = rtrequest_locked(RTM_DELETE,
 298                                         (struct sockaddr *)rt_key(rt),
 299                                         rt->rt_gateway, rt_mask(rt),
 300                                         rt->rt_flags, 0);
 301                         if(err) {
 302                                 log(LOG_WARNING, "in_rtqkill: error %d\n", err);
 303                         } else {
 304                                 ap->killed++;
 305                         }
 306                 } else {
 307                         if(ap->updating
 308                            && (rt->rt_rmx.rmx_expire - timenow.tv_sec
 309                                > rtq_reallyold)) {
 310                                 rt->rt_rmx.rmx_expire = timenow.tv_sec
 311                                         + rtq_reallyold;
 312                         }
 313                         ap->nextstop = lmin(ap->nextstop,
 314                                             rt->rt_rmx.rmx_expire);
 315                 }
 316         }
 317
 318         return 0;
 319 }
 320
 321 static void
 322 in_rtqtimo_funnel(void *rock)
 323 {
 324         in_rtqtimo(rock);
 325
 326 }
 327 #define RTQ_TIMEOUT     60*10   /* run no less than once every ten minutes */
 328 static int rtq_timeout = RTQ_TIMEOUT;
 329
 330 static void
 331 in_rtqtimo(void *rock)
 332 {
 333         struct radix_node_head *rnh = rock;
 334         struct rtqk_arg arg;
 335         struct timeval atv;
 336         static time_t last_adjusted_timeout = 0;
 337         struct timeval timenow;
 338
 339         getmicrotime(&timenow);
 340         arg.found = arg.killed = 0;
 341         arg.rnh = rnh;
 342         arg.nextstop = timenow.tv_sec + rtq_timeout;
 343         arg.draining = arg.updating = 0;
 344         lck_mtx_lock(rt_mtx);
 345         rnh->rnh_walktree(rnh, in_rtqkill, &arg);
 346
 347         /*
 348          * Attempt to be somewhat dynamic about this:
 349          * If there are ``too many'' routes sitting around taking up space,
 350          * then crank down the timeout, and see if we can't make some more
 351          * go away.  However, we make sure that we will never adjust more
 352          * than once in rtq_timeout seconds, to keep from cranking down too
 353          * hard.
 354          */
 355         if((arg.found - arg.killed > rtq_toomany)
 356            && (timenow.tv_sec - last_adjusted_timeout >= rtq_timeout)
 357            && rtq_reallyold > rtq_minreallyold) {
 358                 rtq_reallyold = 2*rtq_reallyold / 3;
 359                 if(rtq_reallyold < rtq_minreallyold) {
 360                         rtq_reallyold = rtq_minreallyold;
 361                 }
 362
 363                 last_adjusted_timeout = timenow.tv_sec;
 364 #if DIAGNOSTIC
 365                 log(LOG_DEBUG, "in_rtqtimo: adjusted rtq_reallyold to %d\n",
 366                     rtq_reallyold);
 367 #endif
 368                 arg.found = arg.killed = 0;
 369                 arg.updating = 1;
 370                 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
 371         }
 372
 373         atv.tv_usec = 0;
 374         atv.tv_sec = arg.nextstop - timenow.tv_sec;
 375         lck_mtx_unlock(rt_mtx);
 376         timeout(in_rtqtimo_funnel, rock, tvtohz(&atv));
 377 }
 378
 379 void
 380 in_rtqdrain(void)
 381 {
 382         struct radix_node_head *rnh = rt_tables[AF_INET];
 383         struct rtqk_arg arg;
 384         arg.found = arg.killed = 0;
 385         arg.rnh = rnh;
 386         arg.nextstop = 0;
 387         arg.draining = 1;
 388         arg.updating = 0;
 389         lck_mtx_lock(rt_mtx);
 390         rnh->rnh_walktree(rnh, in_rtqkill, &arg);
 391         lck_mtx_unlock(rt_mtx);
 392 }
 393
 394 /*
 395  * Initialize our routing tree.
 396  */
 397 int
 398 in_inithead(void **head, int off)
 399 {
 400         struct radix_node_head *rnh;
 401
 402 #ifdef __APPLE__
 403         if (*head)
 404                 return 1;
 405 #endif
 406
 407         if(!rn_inithead(head, off))
 408                 return 0;
 409
 410         if(head != (void **)&rt_tables[AF_INET]) /* BOGUS! */
 411                 return 1;       /* only do this for the real routing table */
 412
 413         rnh = *head;
 414         rnh->rnh_addaddr = in_addroute;
 415         rnh->rnh_matchaddr = in_matroute;
 416         rnh->rnh_close = in_clsroute;
 417         in_rtqtimo(rnh);        /* kick off timeout first time */
 418         return 1;
 419 }
 420
 421 \f
 422 /*
 423  * This zaps old routes when the interface goes down or interface
 424  * address is deleted.  In the latter case, it deletes static routes
 425  * that point to this address.  If we don't do this, we may end up
 426  * using the old address in the future.  The ones we always want to
 427  * get rid of are things like ARP entries, since the user might down
 428  * the interface, walk over to a completely different network, and
 429  * plug back in.
 430  */
 431 struct in_ifadown_arg {
 432         struct radix_node_head *rnh;
 433         struct ifaddr *ifa;
 434         int del;
 435 };
 436
 437 static int
 438 in_ifadownkill(struct radix_node *rn, void *xap)
 439 {
 440         struct in_ifadown_arg *ap = xap;
 441         struct rtentry *rt = (struct rtentry *)rn;
 442         int err;
 443
 444         if (rt->rt_ifa == ap->ifa &&
 445             (ap->del || !(rt->rt_flags & RTF_STATIC))) {
 446                 /*
 447                  * We need to disable the automatic prune that happens
 448                  * in this case in rtrequest() because it will blow
 449                  * away the pointers that rn_walktree() needs in order
 450                  * continue our descent.  We will end up deleting all
 451                  * the routes that rtrequest() would have in any case,
 452                  * so that behavior is not needed there.
 453                  */
 454                 rt->rt_flags &= ~(RTF_CLONING | RTF_PRCLONING);
 455                 err = rtrequest_locked(RTM_DELETE, (struct sockaddr *)rt_key(rt),
 456                                 rt->rt_gateway, rt_mask(rt), rt->rt_flags, 0);
 457                 if (err) {
 458                         log(LOG_WARNING, "in_ifadownkill: error %d\n", err);
 459                 }
 460         }
 461         return 0;
 462 }
 463
 464 int
 465 in_ifadown(struct ifaddr *ifa, int delete)
 466 {
 467         struct in_ifadown_arg arg;
 468         struct radix_node_head *rnh;
 469
 470         lck_mtx_assert(rt_mtx, LCK_MTX_ASSERT_OWNED);
 471
 472         if (ifa->ifa_addr->sa_family != AF_INET)
 473                 return 1;
 474
 475         arg.rnh = rnh = rt_tables[AF_INET];
 476         arg.ifa = ifa;
 477         arg.del = delete;
 478         rnh->rnh_walktree(rnh, in_ifadownkill, &arg);
 479         ifa->ifa_flags &= ~IFA_ROUTE;
 480         return 0;
 481 }