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  */
  51
  52 /*
  53  * This code does two things necessary for the enhanced TCP metrics to
  54  * function in a useful manner:
  55  *  1) It marks all non-host routes as `cloning', thus ensuring that
  56  *     every actual reference to such a route actually gets turned
  57  *     into a reference to a host route to the specific destination
  58  *     requested.
  59  *  2) When such routes lose all their references, it arranges for them
  60  *     to be deleted in some random collection of circumstances, so that
  61  *     a large quantity of stale routing data is not kept in kernel memory
  62  *     indefinitely.  See in_rtqtimo() below for the exact mechanism.
  63  */
  64
  65 #include <sys/param.h>
  66 #include <sys/systm.h>
  67 #include <sys/kernel.h>
  68 #include <sys/sysctl.h>
  69 #include <sys/socket.h>
  70 #include <sys/mbuf.h>
  71 #include <sys/syslog.h>
  72
  73 #include <net/if.h>
  74 #include <net/route.h>
  75 #include <netinet/in.h>
  76 #include <netinet/in_var.h>
  77
  78 extern int      in_inithead __P((void **head, int off));
  79
  80 #define RTPRF_OURS              RTF_PROTO3      /* set on routes we manage */
  81
  82 /*
  83  * Do what we need to do when inserting a route.
  84  */
  85 static struct radix_node *
  86 in_addroute(void *v_arg, void *n_arg, struct radix_node_head *head,
  87             struct radix_node *treenodes)
  88 {
  89         struct rtentry *rt = (struct rtentry *)treenodes;
  90         struct sockaddr_in *sin = (struct sockaddr_in *)rt_key(rt);
  91         struct radix_node *ret;
  92
  93         /*
  94          * For IP, all unicast non-host routes are automatically cloning.
  95          */
  96         if(IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
  97                 rt->rt_flags |= RTF_MULTICAST;
  98
  99         if(!(rt->rt_flags & (RTF_HOST | RTF_CLONING | RTF_MULTICAST))) {
 100                 rt->rt_flags |= RTF_PRCLONING;
 101         }
 102
 103         /*
 104          * A little bit of help for both IP output and input:
 105          *   For host routes, we make sure that RTF_BROADCAST
 106          *   is set for anything that looks like a broadcast address.
 107          *   This way, we can avoid an expensive call to in_broadcast()
 108          *   in ip_output() most of the time (because the route passed
 109          *   to ip_output() is almost always a host route).
 110          *
 111          *   We also do the same for local addresses, with the thought
 112          *   that this might one day be used to speed up ip_input().
 113          *
 114          * We also mark routes to multicast addresses as such, because
 115          * it's easy to do and might be useful (but this is much more
 116          * dubious since it's so easy to inspect the address).  (This
 117          * is done above.)
 118          */
 119         if (rt->rt_flags & RTF_HOST) {
 120                 if (in_broadcast(sin->sin_addr, rt->rt_ifp)) {
 121                         rt->rt_flags |= RTF_BROADCAST;
 122                 } else {
 123 #define satosin(sa) ((struct sockaddr_in *)sa)
 124                         if (satosin(rt->rt_ifa->ifa_addr)->sin_addr.s_addr
 125                             == sin->sin_addr.s_addr)
 126                                 rt->rt_flags |= RTF_LOCAL;
 127 #undef satosin
 128                 }
 129         }
 130
 131         if (!rt->rt_rmx.rmx_mtu && !(rt->rt_rmx.rmx_locks & RTV_MTU)
 132             && rt->rt_ifp)
 133                 rt->rt_rmx.rmx_mtu = rt->rt_ifp->if_mtu;
 134
 135         ret = rn_addroute(v_arg, n_arg, head, treenodes);
 136         if (ret == NULL && rt->rt_flags & RTF_HOST) {
 137                 struct rtentry *rt2;
 138                 /*
 139                  * We are trying to add a host route, but can't.
 140                  * Find out if it is because of an
 141                  * ARP entry and delete it if so.
 142                  */
 143                 rt2 = rtalloc1((struct sockaddr *)sin, 0,
 144                                 RTF_CLONING | RTF_PRCLONING);
 145                 if (rt2) {
 146                         if (rt2->rt_flags & RTF_LLINFO &&
 147                                 rt2->rt_flags & RTF_HOST &&
 148                                 rt2->rt_gateway &&
 149                                 rt2->rt_gateway->sa_family == AF_LINK) {
 150                                 rtrequest(RTM_DELETE,
 151                                           (struct sockaddr *)rt_key(rt2),
 152                                           rt2->rt_gateway,
 153                                           rt_mask(rt2), rt2->rt_flags, 0);
 154                                 ret = rn_addroute(v_arg, n_arg, head,
 155                                         treenodes);
 156                         }
 157                         RTFREE(rt2);
 158                 }
 159         }
 160         return ret;
 161 }
 162
 163 /*
 164  * This code is the inverse of in_clsroute: on first reference, if we
 165  * were managing the route, stop doing so and set the expiration timer
 166  * back off again.
 167  */
 168 static struct radix_node *
 169 in_matroute(void *v_arg, struct radix_node_head *head)
 170 {
 171         struct radix_node *rn = rn_match(v_arg, head);
 172         struct rtentry *rt = (struct rtentry *)rn;
 173
 174         if(rt && rt->rt_refcnt == 0) { /* this is first reference */
 175                 if(rt->rt_flags & RTPRF_OURS) {
 176                         rt->rt_flags &= ~RTPRF_OURS;
 177                         rt->rt_rmx.rmx_expire = 0;
 178                 }
 179         }
 180         return rn;
 181 }
 182
 183 int rtq_reallyold = 60*60;
 184         /* one hour is ``really old'' */
 185 SYSCTL_INT(_net_inet_ip, IPCTL_RTEXPIRE, rtexpire,
 186         CTLFLAG_RW, &rtq_reallyold , 0, "");
 187
 188 int rtq_minreallyold = 10;
 189         /* never automatically crank down to less */
 190 SYSCTL_INT(_net_inet_ip, IPCTL_RTMINEXPIRE, rtminexpire,
 191         CTLFLAG_RW, &rtq_minreallyold , 0, "");
 192
 193 int rtq_toomany = 128;
 194         /* 128 cached routes is ``too many'' */
 195 SYSCTL_INT(_net_inet_ip, IPCTL_RTMAXCACHE, rtmaxcache,
 196         CTLFLAG_RW, &rtq_toomany , 0, "");
 197
 198
 199 /*
 200  * On last reference drop, mark the route as belong to us so that it can be
 201  * timed out.
 202  */
 203 static void
 204 in_clsroute(struct radix_node *rn, struct radix_node_head *head)
 205 {
 206         struct rtentry *rt = (struct rtentry *)rn;
 207
 208         if(!(rt->rt_flags & RTF_UP))
 209                 return;         /* prophylactic measures */
 210
 211         if((rt->rt_flags & (RTF_LLINFO | RTF_HOST)) != RTF_HOST)
 212                 return;
 213
 214         if((rt->rt_flags & (RTF_WASCLONED | RTPRF_OURS))
 215            != RTF_WASCLONED)
 216                 return;
 217
 218         /*
 219          * As requested by David Greenman:
 220          * If rtq_reallyold is 0, just delete the route without
 221          * waiting for a timeout cycle to kill it.
 222          */
 223         if(rtq_reallyold != 0) {
 224                 rt->rt_flags |= RTPRF_OURS;
 225                 rt->rt_rmx.rmx_expire = time_second + rtq_reallyold;
 226         } else {
 227                 rtrequest(RTM_DELETE,
 228                           (struct sockaddr *)rt_key(rt),
 229                           rt->rt_gateway, rt_mask(rt),
 230                           rt->rt_flags, 0);
 231         }
 232 }
 233
 234 struct rtqk_arg {
 235         struct radix_node_head *rnh;
 236         int draining;
 237         int killed;
 238         int found;
 239         int updating;
 240         time_t nextstop;
 241 };
 242
 243 /*
 244  * Get rid of old routes.  When draining, this deletes everything, even when
 245  * the timeout is not expired yet.  When updating, this makes sure that
 246  * nothing has a timeout longer than the current value of rtq_reallyold.
 247  */
 248 static int
 249 in_rtqkill(struct radix_node *rn, void *rock)
 250 {
 251         struct rtqk_arg *ap = rock;
 252         struct rtentry *rt = (struct rtentry *)rn;
 253         int err;
 254
 255         if(rt->rt_flags & RTPRF_OURS) {
 256                 ap->found++;
 257
 258                 if(ap->draining || rt->rt_rmx.rmx_expire <= time_second) {
 259                         if(rt->rt_refcnt > 0)
 260                                 panic("rtqkill route really not free");
 261
 262                         err = rtrequest(RTM_DELETE,
 263                                         (struct sockaddr *)rt_key(rt),
 264                                         rt->rt_gateway, rt_mask(rt),
 265                                         rt->rt_flags, 0);
 266                         if(err) {
 267                                 log(LOG_WARNING, "in_rtqkill: error %d\n", err);
 268                         } else {
 269                                 ap->killed++;
 270                         }
 271                 } else {
 272                         if(ap->updating
 273                            && (rt->rt_rmx.rmx_expire - time_second
 274                                > rtq_reallyold)) {
 275                                 rt->rt_rmx.rmx_expire = time_second
 276                                         + rtq_reallyold;
 277                         }
 278                         ap->nextstop = lmin(ap->nextstop,
 279                                             rt->rt_rmx.rmx_expire);
 280                 }
 281         }
 282
 283         return 0;
 284 }
 285
 286 static void
 287 in_rtqtimo_funnel(void *rock)
 288 {
 289         boolean_t       funnel_state;
 290
 291         funnel_state = thread_funnel_set(network_flock, TRUE);
 292         in_rtqtimo(rock);
 293         (void) thread_funnel_set(network_flock, FALSE);
 294
 295 }
 296 #define RTQ_TIMEOUT     60*10   /* run no less than once every ten minutes */
 297 static int rtq_timeout = RTQ_TIMEOUT;
 298
 299 static void
 300 in_rtqtimo(void *rock)
 301 {
 302         struct radix_node_head *rnh = rock;
 303         struct rtqk_arg arg;
 304         struct timeval atv;
 305         static time_t last_adjusted_timeout = 0;
 306         int s;
 307
 308         arg.found = arg.killed = 0;
 309         arg.rnh = rnh;
 310         arg.nextstop = time_second + rtq_timeout;
 311         arg.draining = arg.updating = 0;
 312         s = splnet();
 313         rnh->rnh_walktree(rnh, in_rtqkill, &arg);
 314         splx(s);
 315
 316         /*
 317          * Attempt to be somewhat dynamic about this:
 318          * If there are ``too many'' routes sitting around taking up space,
 319          * then crank down the timeout, and see if we can't make some more
 320          * go away.  However, we make sure that we will never adjust more
 321          * than once in rtq_timeout seconds, to keep from cranking down too
 322          * hard.
 323          */
 324         if((arg.found - arg.killed > rtq_toomany)
 325            && (time_second - last_adjusted_timeout >= rtq_timeout)
 326            && rtq_reallyold > rtq_minreallyold) {
 327                 rtq_reallyold = 2*rtq_reallyold / 3;
 328                 if(rtq_reallyold < rtq_minreallyold) {
 329                         rtq_reallyold = rtq_minreallyold;
 330                 }
 331
 332                 last_adjusted_timeout = time_second;
 333 #if DIAGNOSTIC
 334                 log(LOG_DEBUG, "in_rtqtimo: adjusted rtq_reallyold to %d\n",
 335                     rtq_reallyold);
 336 #endif
 337                 arg.found = arg.killed = 0;
 338                 arg.updating = 1;
 339                 s = splnet();
 340                 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
 341                 splx(s);
 342         }
 343
 344         atv.tv_usec = 0;
 345         atv.tv_sec = arg.nextstop - time_second;
 346         timeout(in_rtqtimo_funnel, rock, tvtohz(&atv));
 347
 348 }
 349
 350 void
 351 in_rtqdrain(void)
 352 {
 353         struct radix_node_head *rnh = rt_tables[AF_INET];
 354         struct rtqk_arg arg;
 355         int s;
 356         arg.found = arg.killed = 0;
 357         arg.rnh = rnh;
 358         arg.nextstop = 0;
 359         arg.draining = 1;
 360         arg.updating = 0;
 361         s = splnet();
 362         rnh->rnh_walktree(rnh, in_rtqkill, &arg);
 363         splx(s);
 364 }
 365
 366 /*
 367  * Initialize our routing tree.
 368  */
 369 int
 370 in_inithead(void **head, int off)
 371 {
 372         struct radix_node_head *rnh;
 373
 374         if (*head)
 375             return 1;
 376
 377         if(!rn_inithead(head, off))
 378                 return 0;
 379
 380         if(head != (void **)&rt_tables[AF_INET]) /* BOGUS! */
 381                 return 1;       /* only do this for the real routing table */
 382
 383         rnh = *head;
 384         rnh->rnh_addaddr = in_addroute;
 385         rnh->rnh_matchaddr = in_matroute;
 386         rnh->rnh_close = in_clsroute;
 387         in_rtqtimo(rnh);        /* kick off timeout first time */
 388         return 1;
 389 }
 390
 391 \f
 392 /*
 393  * This zaps old routes when the interface goes down.
 394  * Currently it doesn't delete static routes; there are
 395  * arguments one could make for both behaviors.  For the moment,
 396  * we will adopt the Principle of Least Surprise and leave them
 397  * alone (with the knowledge that this will not be enough for some
 398  * people).  The ones we really want to get rid of are things like ARP
 399  * entries, since the user might down the interface, walk over to a completely
 400  * different network, and plug back in.
 401  */
 402 struct in_ifadown_arg {
 403         struct radix_node_head *rnh;
 404         struct ifaddr *ifa;
 405 };
 406
 407 static int
 408 in_ifadownkill(struct radix_node *rn, void *xap)
 409 {
 410         struct in_ifadown_arg *ap = xap;
 411         struct rtentry *rt = (struct rtentry *)rn;
 412         int err;
 413
 414         if (rt->rt_ifa == ap->ifa && !(rt->rt_flags & RTF_STATIC)) {
 415                 /*
 416                  * We need to disable the automatic prune that happens
 417                  * in this case in rtrequest() because it will blow
 418                  * away the pointers that rn_walktree() needs in order
 419                  * continue our descent.  We will end up deleting all
 420                  * the routes that rtrequest() would have in any case,
 421                  * so that behavior is not needed there.
 422                  */
 423                 rt->rt_flags &= ~RTF_PRCLONING;
 424                 err = rtrequest(RTM_DELETE, (struct sockaddr *)rt_key(rt),
 425                                 rt->rt_gateway, rt_mask(rt), rt->rt_flags, 0);
 426                 if (err) {
 427                         log(LOG_WARNING, "in_ifadownkill: error %d\n", err);
 428                 }
 429         }
 430         return 0;
 431 }
 432
 433 int
 434 in_ifadown(struct ifaddr *ifa)
 435 {
 436         struct in_ifadown_arg arg;
 437         struct radix_node_head *rnh;
 438
 439         if (ifa->ifa_addr->sa_family != AF_INET)
 440                 return 1;
 441
 442         arg.rnh = rnh = rt_tables[AF_INET];
 443         arg.ifa = ifa;
 444         rnh->rnh_walktree(rnh, in_ifadownkill, &arg);
 445         ifa->ifa_flags &= ~IFA_ROUTE;
 446         return 0;
 447 }