1 /* 2 * Copyright 1994, 1995 Massachusetts Institute of Technology 3 * 4 * Permission to use, copy, modify, and distribute this software and 5 * its documentation for any purpose and without fee is hereby 6 * granted, provided that both the above copyright notice and this 7 * permission notice appear in all copies, that both the above 8 * copyright notice and this permission notice appear in all 9 * supporting documentation, and that the name of M.I.T. not be used 10 * in advertising or publicity pertaining to distribution of the 11 * software without specific, written prior permission. M.I.T. makes 12 * no representations about the suitability of this software for any 13 * purpose. It is provided "as is" without express or implied 14 * warranty. 15 * 16 * THIS SOFTWARE IS PROVIDED BY M.I.T. ``AS IS''. M.I.T. DISCLAIMS 17 * ALL EXPRESS OR IMPLIED WARRANTIES WITH REGARD TO THIS SOFTWARE, 18 * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF 19 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. IN NO EVENT 20 * SHALL M.I.T. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 21 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 22 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF 23 * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND 24 * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 25 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 26 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * $FreeBSD: src/sys/netinet/in_rmx.c,v 1.37.2.3 2002/08/09 14:49:23 ru Exp $ 30 * $DragonFly: src/sys/netinet/in_rmx.c,v 1.9 2005/01/06 17:59:32 hsu Exp $ 31 */ 32 33 /* 34 * This code does two things necessary for the enhanced TCP metrics to 35 * function in a useful manner: 36 * 1) It marks all non-host routes as `cloning', thus ensuring that 37 * every actual reference to such a route actually gets turned 38 * into a reference to a host route to the specific destination 39 * requested. 40 * 2) When such routes lose all their references, it arranges for them 41 * to be deleted in some random collection of circumstances, so that 42 * a large quantity of stale routing data is not kept in kernel memory 43 * indefinitely. See in_rtqtimo() below for the exact mechanism. 44 */ 45 46 #include <sys/param.h> 47 #include <sys/systm.h> 48 #include <sys/kernel.h> 49 #include <sys/sysctl.h> 50 #include <sys/socket.h> 51 #include <sys/mbuf.h> 52 #include <sys/syslog.h> 53 54 #include <net/if.h> 55 #include <net/route.h> 56 #include <netinet/in.h> 57 #include <netinet/in_var.h> 58 #include <netinet/ip_var.h> 59 60 #define RTPRF_EXPIRING RTF_PROTO3 /* set on routes we manage */ 61 62 static struct callout in_rtqtimo_ch; 63 64 /* 65 * Do what we need to do when inserting a route. 66 */ 67 static struct radix_node * 68 in_addroute(char *key, char *mask, struct radix_node_head *head, 69 struct radix_node *treenodes) 70 { 71 struct rtentry *rt = (struct rtentry *)treenodes; 72 struct sockaddr_in *sin = (struct sockaddr_in *)rt_key(rt); 73 struct radix_node *ret; 74 75 /* 76 * For IP, all unicast non-host routes are automatically cloning. 77 */ 78 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) 79 rt->rt_flags |= RTF_MULTICAST; 80 81 if (!(rt->rt_flags & (RTF_HOST | RTF_CLONING | RTF_MULTICAST))) 82 rt->rt_flags |= RTF_PRCLONING; 83 84 /* 85 * A little bit of help for both IP output and input: 86 * For host routes, we make sure that RTF_BROADCAST 87 * is set for anything that looks like a broadcast address. 88 * This way, we can avoid an expensive call to in_broadcast() 89 * in ip_output() most of the time (because the route passed 90 * to ip_output() is almost always a host route). 91 * 92 * We also do the same for local addresses, with the thought 93 * that this might one day be used to speed up ip_input(). 94 * 95 * We also mark routes to multicast addresses as such, because 96 * it's easy to do and might be useful (but this is much more 97 * dubious since it's so easy to inspect the address). (This 98 * is done above.) 99 */ 100 if (rt->rt_flags & RTF_HOST) { 101 if (in_broadcast(sin->sin_addr, rt->rt_ifp)) { 102 rt->rt_flags |= RTF_BROADCAST; 103 } else { 104 if (satosin(rt->rt_ifa->ifa_addr)->sin_addr.s_addr 105 == sin->sin_addr.s_addr) 106 rt->rt_flags |= RTF_LOCAL; 107 } 108 } 109 110 if (rt->rt_rmx.rmx_mtu != 0 && !(rt->rt_rmx.rmx_locks & RTV_MTU) && 111 rt->rt_ifp != NULL) 112 rt->rt_rmx.rmx_mtu = rt->rt_ifp->if_mtu; 113 114 ret = rn_addroute(key, mask, head, treenodes); 115 if (ret == NULL && rt->rt_flags & RTF_HOST) { 116 struct rtentry *rt2; 117 118 /* 119 * We are trying to add a host route, but can't. 120 * Find out if it is because of an 121 * ARP entry and delete it if so. 122 */ 123 rt2 = rtpurelookup((struct sockaddr *)sin); 124 if (rt2) { 125 --rt->rt_refcnt; 126 if (rt2->rt_flags & RTF_LLINFO && 127 rt2->rt_flags & RTF_HOST && 128 rt2->rt_gateway && 129 rt2->rt_gateway->sa_family == AF_LINK) { 130 rtrequest(RTM_DELETE, rt_key(rt2), 131 rt2->rt_gateway, rt_mask(rt2), 132 rt2->rt_flags, NULL); 133 ret = rn_addroute(key, mask, head, treenodes); 134 } 135 } 136 } 137 138 /* 139 * If the new route created successfully, and we are forwarding, 140 * and there is a cached route, free it. Otherwise, we may end 141 * up using the wrong route. 142 */ 143 if (ret != NULL && ipforwarding && ipforward_rt.ro_rt != NULL) { 144 RTFREE(ipforward_rt.ro_rt); 145 ipforward_rt.ro_rt = NULL; 146 } 147 148 return ret; 149 } 150 151 /* 152 * This code is the inverse of in_closeroute: on first reference, if we 153 * were managing the route, stop doing so and set the expiration timer 154 * back off again. 155 */ 156 static struct radix_node * 157 in_matchroute(char *key, struct radix_node_head *head) 158 { 159 struct radix_node *rn = rn_match(key, head); 160 struct rtentry *rt = (struct rtentry *)rn; 161 162 if (rt != NULL && rt->rt_refcnt == 0) { /* this is first reference */ 163 if (rt->rt_flags & RTPRF_EXPIRING) { 164 rt->rt_flags &= ~RTPRF_EXPIRING; 165 rt->rt_rmx.rmx_expire = 0; 166 } 167 } 168 return rn; 169 } 170 171 static int rtq_reallyold = 60*60; /* one hour is ``really old'' */ 172 SYSCTL_INT(_net_inet_ip, IPCTL_RTEXPIRE, rtexpire, CTLFLAG_RW, 173 &rtq_reallyold , 0, 174 "Default expiration time on cloned routes"); 175 176 static int rtq_minreallyold = 10; /* never automatically crank down to less */ 177 SYSCTL_INT(_net_inet_ip, IPCTL_RTMINEXPIRE, rtminexpire, CTLFLAG_RW, 178 &rtq_minreallyold , 0, 179 "Minimum time to attempt to hold onto cloned routes"); 180 181 static int rtq_toomany = 128; /* 128 cached routes is ``too many'' */ 182 SYSCTL_INT(_net_inet_ip, IPCTL_RTMAXCACHE, rtmaxcache, CTLFLAG_RW, 183 &rtq_toomany , 0, "Upper limit on cloned routes"); 184 185 /* 186 * On last reference drop, mark the route as belong to us so that it can be 187 * timed out. 188 */ 189 static void 190 in_closeroute(struct radix_node *rn, struct radix_node_head *head) 191 { 192 struct rtentry *rt = (struct rtentry *)rn; 193 194 if (!(rt->rt_flags & RTF_UP)) 195 return; /* prophylactic measures */ 196 197 if ((rt->rt_flags & (RTF_LLINFO | RTF_HOST)) != RTF_HOST) 198 return; 199 200 if ((rt->rt_flags & (RTF_WASCLONED | RTPRF_EXPIRING)) != RTF_WASCLONED) 201 return; 202 203 /* 204 * As requested by David Greenman: 205 * If rtq_reallyold is 0, just delete the route without 206 * waiting for a timeout cycle to kill it. 207 */ 208 if (rtq_reallyold != 0) { 209 rt->rt_flags |= RTPRF_EXPIRING; 210 rt->rt_rmx.rmx_expire = time_second + rtq_reallyold; 211 } else { 212 /* 213 * Remove route from the radix tree, but defer deallocation 214 * until we return to rtfree(). 215 */ 216 rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway, rt_mask(rt), 217 rt->rt_flags, &rt); 218 } 219 } 220 221 struct rtqk_arg { 222 struct radix_node_head *rnh; 223 int draining; 224 int killed; 225 int found; 226 int updating; 227 time_t nextstop; 228 }; 229 230 /* 231 * Get rid of old routes. When draining, this deletes everything, even when 232 * the timeout is not expired yet. When updating, this makes sure that 233 * nothing has a timeout longer than the current value of rtq_reallyold. 234 */ 235 static int 236 in_rtqkill(struct radix_node *rn, void *rock) 237 { 238 struct rtqk_arg *ap = rock; 239 struct rtentry *rt = (struct rtentry *)rn; 240 int err; 241 242 if (rt->rt_flags & RTPRF_EXPIRING) { 243 ap->found++; 244 if (ap->draining || rt->rt_rmx.rmx_expire <= time_second) { 245 if (rt->rt_refcnt > 0) 246 panic("rtqkill route really not free"); 247 248 err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway, 249 rt_mask(rt), rt->rt_flags, NULL); 250 if (err) { 251 log(LOG_WARNING, "in_rtqkill: error %d\n", err); 252 } else { 253 ap->killed++; 254 } 255 } else { 256 if (ap->updating && 257 (rt->rt_rmx.rmx_expire - time_second > 258 rtq_reallyold)) { 259 rt->rt_rmx.rmx_expire = time_second + 260 rtq_reallyold; 261 } 262 ap->nextstop = lmin(ap->nextstop, 263 rt->rt_rmx.rmx_expire); 264 } 265 } 266 267 return 0; 268 } 269 270 #define RTQ_TIMEOUT 60*10 /* run no less than once every ten minutes */ 271 static int rtq_timeout = RTQ_TIMEOUT; 272 273 static void 274 in_rtqtimo(void *rock) 275 { 276 struct radix_node_head *rnh = rock; 277 struct rtqk_arg arg; 278 struct timeval atv; 279 static time_t last_adjusted_timeout = 0; 280 int s; 281 282 arg.found = arg.killed = 0; 283 arg.rnh = rnh; 284 arg.nextstop = time_second + rtq_timeout; 285 arg.draining = arg.updating = 0; 286 s = splnet(); 287 rnh->rnh_walktree(rnh, in_rtqkill, &arg); 288 splx(s); 289 290 /* 291 * Attempt to be somewhat dynamic about this: 292 * If there are ``too many'' routes sitting around taking up space, 293 * then crank down the timeout, and see if we can't make some more 294 * go away. However, we make sure that we will never adjust more 295 * than once in rtq_timeout seconds, to keep from cranking down too 296 * hard. 297 */ 298 if ((arg.found - arg.killed > rtq_toomany) && 299 (time_second - last_adjusted_timeout >= rtq_timeout) && 300 rtq_reallyold > rtq_minreallyold) { 301 rtq_reallyold = 2*rtq_reallyold / 3; 302 if (rtq_reallyold < rtq_minreallyold) { 303 rtq_reallyold = rtq_minreallyold; 304 } 305 306 last_adjusted_timeout = time_second; 307 #ifdef DIAGNOSTIC 308 log(LOG_DEBUG, "in_rtqtimo: adjusted rtq_reallyold to %d\n", 309 rtq_reallyold); 310 #endif 311 arg.found = arg.killed = 0; 312 arg.updating = 1; 313 s = splnet(); 314 rnh->rnh_walktree(rnh, in_rtqkill, &arg); 315 splx(s); 316 } 317 318 atv.tv_usec = 0; 319 atv.tv_sec = arg.nextstop - time_second; 320 callout_reset(&in_rtqtimo_ch, tvtohz_high(&atv), in_rtqtimo, rock); 321 } 322 323 void 324 in_rtqdrain(void) 325 { 326 struct radix_node_head *rnh = rt_tables[AF_INET]; 327 struct rtqk_arg arg; 328 int s; 329 330 arg.found = arg.killed = 0; 331 arg.rnh = rnh; 332 arg.nextstop = 0; 333 arg.draining = 1; 334 arg.updating = 0; 335 s = splnet(); 336 rnh->rnh_walktree(rnh, in_rtqkill, &arg); 337 splx(s); 338 } 339 340 /* 341 * Initialize our routing tree. 342 */ 343 int 344 in_inithead(void **head, int off) 345 { 346 struct radix_node_head *rnh; 347 348 if (!rn_inithead(head, off)) 349 return 0; 350 351 if (head != (void **)&rt_tables[AF_INET]) /* BOGUS! */ 352 return 1; /* only do this for the real routing table */ 353 354 rnh = *head; 355 rnh->rnh_addaddr = in_addroute; 356 rnh->rnh_matchaddr = in_matchroute; 357 rnh->rnh_close = in_closeroute; 358 callout_init(&in_rtqtimo_ch); 359 in_rtqtimo(rnh); /* kick off timeout first time */ 360 return 1; 361 } 362 363 364 /* 365 * This zaps old routes when the interface goes down or interface 366 * address is deleted. In the latter case, it deletes static routes 367 * that point to this address. If we don't do this, we may end up 368 * using the old address in the future. The ones we always want to 369 * get rid of are things like ARP entries, since the user might down 370 * the interface, walk over to a completely different network, and 371 * plug back in. 372 */ 373 struct in_ifadown_arg { 374 struct radix_node_head *rnh; 375 struct ifaddr *ifa; 376 int del; 377 }; 378 379 static int 380 in_ifadownkill(struct radix_node *rn, void *xap) 381 { 382 struct in_ifadown_arg *ap = xap; 383 struct rtentry *rt = (struct rtentry *)rn; 384 int err; 385 386 if (rt->rt_ifa == ap->ifa && 387 (ap->del || !(rt->rt_flags & RTF_STATIC))) { 388 /* 389 * We need to disable the automatic prune that happens 390 * in this case in rtrequest() because it will blow 391 * away the pointers that rn_walktree() needs in order 392 * continue our descent. We will end up deleting all 393 * the routes that rtrequest() would have in any case, 394 * so that behavior is not needed there. 395 */ 396 rt->rt_flags &= ~(RTF_CLONING | RTF_PRCLONING); 397 err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway, 398 rt_mask(rt), rt->rt_flags, NULL); 399 if (err) { 400 log(LOG_WARNING, "in_ifadownkill: error %d\n", err); 401 } 402 } 403 return 0; 404 } 405 406 int 407 in_ifadown(struct ifaddr *ifa, int delete) 408 { 409 struct in_ifadown_arg arg; 410 struct radix_node_head *rnh; 411 412 if (ifa->ifa_addr->sa_family != AF_INET) 413 return 1; 414 415 arg.rnh = rnh = rt_tables[AF_INET]; 416 arg.ifa = ifa; 417 arg.del = delete; 418 rnh->rnh_walktree(rnh, in_ifadownkill, &arg); 419 ifa->ifa_flags &= ~IFA_ROUTE; 420 return 0; 421 } 422