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 */ 31 32 /* 33 * This code does two things necessary for the enhanced TCP metrics to 34 * function in a useful manner: 35 * 1) It marks all non-host routes as `cloning', thus ensuring that 36 * every actual reference to such a route actually gets turned 37 * into a reference to a host route to the specific destination 38 * requested. 39 * 2) When such routes lose all their references, it arranges for them 40 * to be deleted in some random collection of circumstances, so that 41 * a large quantity of stale routing data is not kept in kernel memory 42 * indefinitely. See in_rtqtimo() below for the exact mechanism. 43 */ 44 45 #include <sys/param.h> 46 #include <sys/systm.h> 47 #include <sys/kernel.h> 48 #include <sys/sysctl.h> 49 #include <sys/socket.h> 50 #include <sys/mbuf.h> 51 #include <sys/syslog.h> 52 53 #include <net/if.h> 54 #include <net/route.h> 55 #include <netinet/in.h> 56 #include <netinet/in_var.h> 57 #include <netinet/ip_var.h> 58 59 extern int in_inithead __P((void **head, int off)); 60 61 #define RTPRF_OURS RTF_PROTO3 /* set on routes we manage */ 62 63 /* 64 * Do what we need to do when inserting a route. 65 */ 66 static struct radix_node * 67 in_addroute(void *v_arg, void *n_arg, struct radix_node_head *head, 68 struct radix_node *treenodes) 69 { 70 struct rtentry *rt = (struct rtentry *)treenodes; 71 struct sockaddr_in *sin = (struct sockaddr_in *)rt_key(rt); 72 struct radix_node *ret; 73 74 /* 75 * For IP, all unicast non-host routes are automatically cloning. 76 */ 77 if(IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) 78 rt->rt_flags |= RTF_MULTICAST; 79 80 if(!(rt->rt_flags & (RTF_HOST | RTF_CLONING | RTF_MULTICAST))) { 81 rt->rt_flags |= RTF_PRCLONING; 82 } 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 && !(rt->rt_rmx.rmx_locks & RTV_MTU) 111 && rt->rt_ifp) 112 rt->rt_rmx.rmx_mtu = rt->rt_ifp->if_mtu; 113 114 ret = rn_addroute(v_arg, n_arg, head, treenodes); 115 if (ret == NULL && rt->rt_flags & RTF_HOST) { 116 struct rtentry *rt2; 117 /* 118 * We are trying to add a host route, but can't. 119 * Find out if it is because of an 120 * ARP entry and delete it if so. 121 */ 122 rt2 = rtalloc1((struct sockaddr *)sin, 0, 123 RTF_CLONING | RTF_PRCLONING); 124 if (rt2) { 125 if (rt2->rt_flags & RTF_LLINFO && 126 rt2->rt_flags & RTF_HOST && 127 rt2->rt_gateway && 128 rt2->rt_gateway->sa_family == AF_LINK) { 129 rtrequest(RTM_DELETE, 130 (struct sockaddr *)rt_key(rt2), 131 rt2->rt_gateway, 132 rt_mask(rt2), rt2->rt_flags, 0); 133 ret = rn_addroute(v_arg, n_arg, head, 134 treenodes); 135 } 136 RTFREE(rt2); 137 } 138 } 139 140 /* 141 * If the new route created successfully, and we are forwarding, 142 * and there is a cached route, free it. Otherwise, we may end 143 * up using the wrong route. 144 */ 145 if (ret != NULL && ipforwarding && ipforward_rt.ro_rt) { 146 RTFREE(ipforward_rt.ro_rt); 147 ipforward_rt.ro_rt = 0; 148 } 149 150 return ret; 151 } 152 153 /* 154 * This code is the inverse of in_clsroute: on first reference, if we 155 * were managing the route, stop doing so and set the expiration timer 156 * back off again. 157 */ 158 static struct radix_node * 159 in_matroute(void *v_arg, struct radix_node_head *head) 160 { 161 struct radix_node *rn = rn_match(v_arg, head); 162 struct rtentry *rt = (struct rtentry *)rn; 163 164 if(rt && rt->rt_refcnt == 0) { /* this is first reference */ 165 if(rt->rt_flags & RTPRF_OURS) { 166 rt->rt_flags &= ~RTPRF_OURS; 167 rt->rt_rmx.rmx_expire = 0; 168 } 169 } 170 return rn; 171 } 172 173 static int rtq_reallyold = 60*60; 174 /* one hour is ``really old'' */ 175 SYSCTL_INT(_net_inet_ip, IPCTL_RTEXPIRE, rtexpire, CTLFLAG_RW, 176 &rtq_reallyold , 0, 177 "Default expiration time on dynamically learned routes"); 178 179 static int rtq_minreallyold = 10; 180 /* never automatically crank down to less */ 181 SYSCTL_INT(_net_inet_ip, IPCTL_RTMINEXPIRE, rtminexpire, CTLFLAG_RW, 182 &rtq_minreallyold , 0, 183 "Minimum time to attempt to hold onto dynamically learned routes"); 184 185 static int rtq_toomany = 128; 186 /* 128 cached routes is ``too many'' */ 187 SYSCTL_INT(_net_inet_ip, IPCTL_RTMAXCACHE, rtmaxcache, CTLFLAG_RW, 188 &rtq_toomany , 0, "Upper limit on dynamically learned routes"); 189 190 /* 191 * On last reference drop, mark the route as belong to us so that it can be 192 * timed out. 193 */ 194 static void 195 in_clsroute(struct radix_node *rn, struct radix_node_head *head) 196 { 197 struct rtentry *rt = (struct rtentry *)rn; 198 199 if(!(rt->rt_flags & RTF_UP)) 200 return; /* prophylactic measures */ 201 202 if((rt->rt_flags & (RTF_LLINFO | RTF_HOST)) != RTF_HOST) 203 return; 204 205 if((rt->rt_flags & (RTF_WASCLONED | RTPRF_OURS)) 206 != RTF_WASCLONED) 207 return; 208 209 /* 210 * As requested by David Greenman: 211 * If rtq_reallyold is 0, just delete the route without 212 * waiting for a timeout cycle to kill it. 213 */ 214 if(rtq_reallyold != 0) { 215 rt->rt_flags |= RTPRF_OURS; 216 rt->rt_rmx.rmx_expire = time_second + rtq_reallyold; 217 } else { 218 rtrequest(RTM_DELETE, 219 (struct sockaddr *)rt_key(rt), 220 rt->rt_gateway, rt_mask(rt), 221 rt->rt_flags, 0); 222 } 223 } 224 225 struct rtqk_arg { 226 struct radix_node_head *rnh; 227 int draining; 228 int killed; 229 int found; 230 int updating; 231 time_t nextstop; 232 }; 233 234 /* 235 * Get rid of old routes. When draining, this deletes everything, even when 236 * the timeout is not expired yet. When updating, this makes sure that 237 * nothing has a timeout longer than the current value of rtq_reallyold. 238 */ 239 static int 240 in_rtqkill(struct radix_node *rn, void *rock) 241 { 242 struct rtqk_arg *ap = rock; 243 struct rtentry *rt = (struct rtentry *)rn; 244 int err; 245 246 if(rt->rt_flags & RTPRF_OURS) { 247 ap->found++; 248 249 if(ap->draining || rt->rt_rmx.rmx_expire <= time_second) { 250 if(rt->rt_refcnt > 0) 251 panic("rtqkill route really not free"); 252 253 err = rtrequest(RTM_DELETE, 254 (struct sockaddr *)rt_key(rt), 255 rt->rt_gateway, rt_mask(rt), 256 rt->rt_flags, 0); 257 if(err) { 258 log(LOG_WARNING, "in_rtqkill: error %d\n", err); 259 } else { 260 ap->killed++; 261 } 262 } else { 263 if(ap->updating 264 && (rt->rt_rmx.rmx_expire - time_second 265 > rtq_reallyold)) { 266 rt->rt_rmx.rmx_expire = time_second 267 + rtq_reallyold; 268 } 269 ap->nextstop = lmin(ap->nextstop, 270 rt->rt_rmx.rmx_expire); 271 } 272 } 273 274 return 0; 275 } 276 277 #define RTQ_TIMEOUT 60*10 /* run no less than once every ten minutes */ 278 static int rtq_timeout = RTQ_TIMEOUT; 279 280 static void 281 in_rtqtimo(void *rock) 282 { 283 struct radix_node_head *rnh = rock; 284 struct rtqk_arg arg; 285 struct timeval atv; 286 static time_t last_adjusted_timeout = 0; 287 int s; 288 289 arg.found = arg.killed = 0; 290 arg.rnh = rnh; 291 arg.nextstop = time_second + rtq_timeout; 292 arg.draining = arg.updating = 0; 293 s = splnet(); 294 rnh->rnh_walktree(rnh, in_rtqkill, &arg); 295 splx(s); 296 297 /* 298 * Attempt to be somewhat dynamic about this: 299 * If there are ``too many'' routes sitting around taking up space, 300 * then crank down the timeout, and see if we can't make some more 301 * go away. However, we make sure that we will never adjust more 302 * than once in rtq_timeout seconds, to keep from cranking down too 303 * hard. 304 */ 305 if((arg.found - arg.killed > rtq_toomany) 306 && (time_second - last_adjusted_timeout >= rtq_timeout) 307 && rtq_reallyold > rtq_minreallyold) { 308 rtq_reallyold = 2*rtq_reallyold / 3; 309 if(rtq_reallyold < rtq_minreallyold) { 310 rtq_reallyold = rtq_minreallyold; 311 } 312 313 last_adjusted_timeout = time_second; 314 #ifdef DIAGNOSTIC 315 log(LOG_DEBUG, "in_rtqtimo: adjusted rtq_reallyold to %d\n", 316 rtq_reallyold); 317 #endif 318 arg.found = arg.killed = 0; 319 arg.updating = 1; 320 s = splnet(); 321 rnh->rnh_walktree(rnh, in_rtqkill, &arg); 322 splx(s); 323 } 324 325 atv.tv_usec = 0; 326 atv.tv_sec = arg.nextstop - time_second; 327 timeout(in_rtqtimo, rock, tvtohz(&atv)); 328 } 329 330 void 331 in_rtqdrain(void) 332 { 333 struct radix_node_head *rnh = rt_tables[AF_INET]; 334 struct rtqk_arg arg; 335 int s; 336 arg.found = arg.killed = 0; 337 arg.rnh = rnh; 338 arg.nextstop = 0; 339 arg.draining = 1; 340 arg.updating = 0; 341 s = splnet(); 342 rnh->rnh_walktree(rnh, in_rtqkill, &arg); 343 splx(s); 344 } 345 346 /* 347 * Initialize our routing tree. 348 */ 349 int 350 in_inithead(void **head, int off) 351 { 352 struct radix_node_head *rnh; 353 354 if(!rn_inithead(head, off)) 355 return 0; 356 357 if(head != (void **)&rt_tables[AF_INET]) /* BOGUS! */ 358 return 1; /* only do this for the real routing table */ 359 360 rnh = *head; 361 rnh->rnh_addaddr = in_addroute; 362 rnh->rnh_matchaddr = in_matroute; 363 rnh->rnh_close = in_clsroute; 364 in_rtqtimo(rnh); /* kick off timeout first time */ 365 return 1; 366 } 367 368 369 /* 370 * This zaps old routes when the interface goes down or interface 371 * address is deleted. In the latter case, it deletes static routes 372 * that point to this address. If we don't do this, we may end up 373 * using the old address in the future. The ones we always want to 374 * get rid of are things like ARP entries, since the user might down 375 * the interface, walk over to a completely different network, and 376 * plug back in. 377 */ 378 struct in_ifadown_arg { 379 struct radix_node_head *rnh; 380 struct ifaddr *ifa; 381 int del; 382 }; 383 384 static int 385 in_ifadownkill(struct radix_node *rn, void *xap) 386 { 387 struct in_ifadown_arg *ap = xap; 388 struct rtentry *rt = (struct rtentry *)rn; 389 int err; 390 391 if (rt->rt_ifa == ap->ifa && 392 (ap->del || !(rt->rt_flags & RTF_STATIC))) { 393 /* 394 * We need to disable the automatic prune that happens 395 * in this case in rtrequest() because it will blow 396 * away the pointers that rn_walktree() needs in order 397 * continue our descent. We will end up deleting all 398 * the routes that rtrequest() would have in any case, 399 * so that behavior is not needed there. 400 */ 401 rt->rt_flags &= ~(RTF_CLONING | RTF_PRCLONING); 402 err = rtrequest(RTM_DELETE, (struct sockaddr *)rt_key(rt), 403 rt->rt_gateway, rt_mask(rt), rt->rt_flags, 0); 404 if (err) { 405 log(LOG_WARNING, "in_ifadownkill: error %d\n", err); 406 } 407 } 408 return 0; 409 } 410 411 int 412 in_ifadown(struct ifaddr *ifa, int delete) 413 { 414 struct in_ifadown_arg arg; 415 struct radix_node_head *rnh; 416 417 if (ifa->ifa_addr->sa_family != AF_INET) 418 return 1; 419 420 arg.rnh = rnh = rt_tables[AF_INET]; 421 arg.ifa = ifa; 422 arg.del = delete; 423 rnh->rnh_walktree(rnh, in_ifadownkill, &arg); 424 ifa->ifa_flags &= ~IFA_ROUTE; 425 return 0; 426 } 427