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 1.14 2006/04/11 06:59:34 dillon 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 "opt_carp.h" 47 48 #include <sys/param.h> 49 #include <sys/systm.h> 50 #include <sys/kernel.h> 51 #include <sys/sysctl.h> 52 #include <sys/socket.h> 53 #include <sys/mbuf.h> 54 #include <sys/syslog.h> 55 #include <sys/globaldata.h> 56 #include <sys/thread2.h> 57 58 #include <net/if.h> 59 #include <net/route.h> 60 #include <net/if_var.h> 61 #ifdef CARP 62 #include <net/if_types.h> 63 #endif 64 #include <netinet/in.h> 65 #include <netinet/in_var.h> 66 #include <netinet/ip_var.h> 67 #include <netinet/ip_flow.h> 68 69 #define RTPRF_EXPIRING RTF_PROTO3 /* set on routes we manage */ 70 71 static struct callout in_rtqtimo_ch[MAXCPU]; 72 73 /* 74 * Do what we need to do when inserting a route. 75 */ 76 static struct radix_node * 77 in_addroute(char *key, char *mask, struct radix_node_head *head, 78 struct radix_node *treenodes) 79 { 80 struct rtentry *rt = (struct rtentry *)treenodes; 81 struct sockaddr_in *sin = (struct sockaddr_in *)rt_key(rt); 82 struct radix_node *ret; 83 struct in_ifaddr_container *iac; 84 struct in_ifaddr *ia; 85 86 /* 87 * For IP, mark routes to multicast addresses as such, because 88 * it's easy to do and might be useful (but this is much more 89 * dubious since it's so easy to inspect the address). 90 * 91 * For IP, all unicast non-host routes are automatically cloning. 92 */ 93 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) 94 rt->rt_flags |= RTF_MULTICAST; 95 96 if (!(rt->rt_flags & (RTF_HOST | RTF_CLONING | RTF_MULTICAST))) 97 rt->rt_flags |= RTF_PRCLONING; 98 99 /* 100 * For host routes, we make sure that RTF_BROADCAST 101 * is set for anything that looks like a broadcast address. 102 * This way, we can avoid an expensive call to in_broadcast() 103 * in ip_output() most of the time (because the route passed 104 * to ip_output() is almost always a host route). 105 * 106 * For local routes we set RTF_LOCAL allowing various shortcuts. 107 * 108 * A cloned network route will point to one of several possible 109 * addresses if an interface has aliases and must be repointed 110 * back to the correct address or arp_rtrequest() will not properly 111 * detect the local ip. 112 */ 113 if (rt->rt_flags & RTF_HOST) { 114 if (in_broadcast(sin->sin_addr, rt->rt_ifp)) { 115 rt->rt_flags |= RTF_BROADCAST; 116 } else if (satosin(rt->rt_ifa->ifa_addr)->sin_addr.s_addr == 117 sin->sin_addr.s_addr) { 118 rt->rt_flags |= RTF_LOCAL; 119 } else { 120 LIST_FOREACH(iac, INADDR_HASH(sin->sin_addr.s_addr), 121 ia_hash) { 122 ia = iac->ia; 123 if (sin->sin_addr.s_addr == 124 ia->ia_addr.sin_addr.s_addr) { 125 rt->rt_flags |= RTF_LOCAL; 126 IFAREF(&ia->ia_ifa); 127 IFAFREE(rt->rt_ifa); 128 rt->rt_ifa = &ia->ia_ifa; 129 rt->rt_ifp = rt->rt_ifa->ifa_ifp; 130 break; 131 } 132 } 133 } 134 } 135 136 if (rt->rt_rmx.rmx_mtu != 0 && !(rt->rt_rmx.rmx_locks & RTV_MTU) && 137 rt->rt_ifp != NULL) 138 rt->rt_rmx.rmx_mtu = rt->rt_ifp->if_mtu; 139 140 ret = rn_addroute(key, mask, head, treenodes); 141 if (ret == NULL && (rt->rt_flags & RTF_HOST)) { 142 struct rtentry *oldrt; 143 144 /* 145 * We are trying to add a host route, but can't. 146 * Find out if it is because of an ARP entry and 147 * delete it if so. 148 */ 149 oldrt = rtpurelookup((struct sockaddr *)sin); 150 if (oldrt != NULL) { 151 --oldrt->rt_refcnt; 152 if ((oldrt->rt_flags & RTF_LLINFO) && 153 (oldrt->rt_flags & RTF_HOST) && 154 oldrt->rt_gateway && 155 oldrt->rt_gateway->sa_family == AF_LINK) { 156 rtrequest(RTM_DELETE, rt_key(oldrt), 157 oldrt->rt_gateway, rt_mask(oldrt), 158 oldrt->rt_flags, NULL); 159 ret = rn_addroute(key, mask, head, treenodes); 160 } 161 } 162 } 163 164 /* 165 * If the new route has been created successfully, and it is 166 * not a multicast/broadcast or cloned route, then we will 167 * have to flush the ipflow. Otherwise, we may end up using 168 * the wrong route. 169 */ 170 if (ret != NULL && 171 (rt->rt_flags & 172 (RTF_MULTICAST | RTF_BROADCAST | RTF_WASCLONED)) == 0) { 173 ipflow_flush_oncpu(); 174 } 175 return ret; 176 } 177 178 /* 179 * This code is the inverse of in_closeroute: on first reference, if we 180 * were managing the route, stop doing so and set the expiration timer 181 * back off again. 182 */ 183 static struct radix_node * 184 in_matchroute(char *key, struct radix_node_head *head) 185 { 186 struct radix_node *rn = rn_match(key, head); 187 struct rtentry *rt = (struct rtentry *)rn; 188 189 if (rt != NULL && rt->rt_refcnt == 0) { /* this is first reference */ 190 if (rt->rt_flags & RTPRF_EXPIRING) { 191 rt->rt_flags &= ~RTPRF_EXPIRING; 192 rt->rt_rmx.rmx_expire = 0; 193 } 194 } 195 return rn; 196 } 197 198 static int rtq_reallyold = 60*60; /* one hour is ``really old'' */ 199 SYSCTL_INT(_net_inet_ip, IPCTL_RTEXPIRE, rtexpire, CTLFLAG_RW, 200 &rtq_reallyold , 0, 201 "Default expiration time on cloned routes"); 202 203 static int rtq_minreallyold = 10; /* never automatically crank down to less */ 204 SYSCTL_INT(_net_inet_ip, IPCTL_RTMINEXPIRE, rtminexpire, CTLFLAG_RW, 205 &rtq_minreallyold , 0, 206 "Minimum time to attempt to hold onto cloned routes"); 207 208 static int rtq_toomany = 128; /* 128 cached routes is ``too many'' */ 209 SYSCTL_INT(_net_inet_ip, IPCTL_RTMAXCACHE, rtmaxcache, CTLFLAG_RW, 210 &rtq_toomany , 0, "Upper limit on cloned routes"); 211 212 /* 213 * On last reference drop, mark the route as belong to us so that it can be 214 * timed out. 215 */ 216 static void 217 in_closeroute(struct radix_node *rn, struct radix_node_head *head) 218 { 219 struct rtentry *rt = (struct rtentry *)rn; 220 221 if (!(rt->rt_flags & RTF_UP)) 222 return; /* prophylactic measures */ 223 224 if ((rt->rt_flags & (RTF_LLINFO | RTF_HOST)) != RTF_HOST) 225 return; 226 227 if ((rt->rt_flags & (RTF_WASCLONED | RTPRF_EXPIRING)) != RTF_WASCLONED) 228 return; 229 230 /* 231 * As requested by David Greenman: 232 * If rtq_reallyold is 0, just delete the route without 233 * waiting for a timeout cycle to kill it. 234 */ 235 if (rtq_reallyold != 0) { 236 rt->rt_flags |= RTPRF_EXPIRING; 237 rt->rt_rmx.rmx_expire = time_second + rtq_reallyold; 238 } else { 239 /* 240 * Remove route from the radix tree, but defer deallocation 241 * until we return to rtfree(). 242 */ 243 rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway, rt_mask(rt), 244 rt->rt_flags, &rt); 245 } 246 } 247 248 struct rtqk_arg { 249 struct radix_node_head *rnh; 250 int draining; 251 int killed; 252 int found; 253 int updating; 254 time_t nextstop; 255 }; 256 257 /* 258 * Get rid of old routes. When draining, this deletes everything, even when 259 * the timeout is not expired yet. When updating, this makes sure that 260 * nothing has a timeout longer than the current value of rtq_reallyold. 261 */ 262 static int 263 in_rtqkill(struct radix_node *rn, void *rock) 264 { 265 struct rtqk_arg *ap = rock; 266 struct rtentry *rt = (struct rtentry *)rn; 267 int err; 268 269 if (rt->rt_flags & RTPRF_EXPIRING) { 270 ap->found++; 271 if (ap->draining || rt->rt_rmx.rmx_expire <= time_second) { 272 if (rt->rt_refcnt > 0) 273 panic("rtqkill route really not free"); 274 275 err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway, 276 rt_mask(rt), rt->rt_flags, NULL); 277 if (err) 278 log(LOG_WARNING, "in_rtqkill: error %d\n", err); 279 else 280 ap->killed++; 281 } else { 282 if (ap->updating && 283 (rt->rt_rmx.rmx_expire - time_second > 284 rtq_reallyold)) { 285 rt->rt_rmx.rmx_expire = time_second + 286 rtq_reallyold; 287 } 288 ap->nextstop = lmin(ap->nextstop, 289 rt->rt_rmx.rmx_expire); 290 } 291 } 292 293 return 0; 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; 306 307 arg.found = arg.killed = 0; 308 arg.rnh = rnh; 309 arg.nextstop = time_second + rtq_timeout; 310 arg.draining = arg.updating = 0; 311 crit_enter(); 312 rnh->rnh_walktree(rnh, in_rtqkill, &arg); 313 crit_exit(); 314 315 /* 316 * Attempt to be somewhat dynamic about this: 317 * If there are ``too many'' routes sitting around taking up space, 318 * then crank down the timeout, and see if we can't make some more 319 * go away. However, we make sure that we will never adjust more 320 * than once in rtq_timeout seconds, to keep from cranking down too 321 * hard. 322 */ 323 if ((arg.found - arg.killed > rtq_toomany) && 324 (time_second - last_adjusted_timeout >= rtq_timeout) && 325 rtq_reallyold > rtq_minreallyold) { 326 rtq_reallyold = 2*rtq_reallyold / 3; 327 if (rtq_reallyold < rtq_minreallyold) { 328 rtq_reallyold = rtq_minreallyold; 329 } 330 331 last_adjusted_timeout = time_second; 332 #ifdef DIAGNOSTIC 333 log(LOG_DEBUG, "in_rtqtimo: adjusted rtq_reallyold to %d\n", 334 rtq_reallyold); 335 #endif 336 arg.found = arg.killed = 0; 337 arg.updating = 1; 338 crit_enter(); 339 rnh->rnh_walktree(rnh, in_rtqkill, &arg); 340 crit_exit(); 341 } 342 343 atv.tv_usec = 0; 344 atv.tv_sec = arg.nextstop - time_second; 345 callout_reset(&in_rtqtimo_ch[mycpuid], tvtohz_high(&atv), in_rtqtimo, 346 rock); 347 } 348 349 void 350 in_rtqdrain(void) 351 { 352 struct radix_node_head *rnh = rt_tables[mycpuid][AF_INET]; 353 struct rtqk_arg arg; 354 355 arg.found = arg.killed = 0; 356 arg.rnh = rnh; 357 arg.nextstop = 0; 358 arg.draining = 1; 359 arg.updating = 0; 360 crit_enter(); 361 rnh->rnh_walktree(rnh, in_rtqkill, &arg); 362 crit_exit(); 363 } 364 365 /* 366 * Initialize our routing tree. 367 */ 368 int 369 in_inithead(void **head, int off) 370 { 371 struct radix_node_head *rnh; 372 373 if (!rn_inithead(head, rn_cpumaskhead(mycpuid), off)) 374 return 0; 375 376 if (head != (void **)&rt_tables[mycpuid][AF_INET]) /* BOGUS! */ 377 return 1; /* only do this for the real routing table */ 378 379 rnh = *head; 380 rnh->rnh_addaddr = in_addroute; 381 rnh->rnh_matchaddr = in_matchroute; 382 rnh->rnh_close = in_closeroute; 383 callout_init(&in_rtqtimo_ch[mycpuid]); 384 in_rtqtimo(rnh); /* kick off timeout first time */ 385 return 1; 386 } 387 388 /* 389 * This zaps old routes when the interface goes down or interface 390 * address is deleted. In the latter case, it deletes static routes 391 * that point to this address. If we don't do this, we may end up 392 * using the old address in the future. The ones we always want to 393 * get rid of are things like ARP entries, since the user might down 394 * the interface, walk over to a completely different network, and 395 * plug back in. 396 * 397 * in_ifadown() is typically called when an interface is being brought 398 * down. We must iterate through all per-cpu route tables and clean 399 * them up. 400 */ 401 struct in_ifadown_arg { 402 struct radix_node_head *rnh; 403 struct ifaddr *ifa; 404 int del; 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 && 415 (ap->del || !(rt->rt_flags & RTF_STATIC))) { 416 /* 417 * We need to disable the automatic prune that happens 418 * in this case in rtrequest() because it will blow 419 * away the pointers that rn_walktree() needs in order 420 * continue our descent. We will end up deleting all 421 * the routes that rtrequest() would have in any case, 422 * so that behavior is not needed there. 423 */ 424 rt->rt_flags &= ~(RTF_CLONING | RTF_PRCLONING); 425 err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway, 426 rt_mask(rt), rt->rt_flags, NULL); 427 if (err) 428 log(LOG_WARNING, "in_ifadownkill: error %d\n", err); 429 } 430 return 0; 431 } 432 433 int 434 in_ifadown_force(struct ifaddr *ifa, int delete) 435 { 436 struct in_ifadown_arg arg; 437 struct radix_node_head *rnh; 438 int origcpu; 439 int cpu; 440 441 if (ifa->ifa_addr->sa_family != AF_INET) 442 return 1; 443 444 /* 445 * XXX individual requests are not independantly chained, 446 * which means that the per-cpu route tables will not be 447 * consistent in the middle of the operation. If routes 448 * related to the interface are manipulated while we are 449 * doing this the inconsistancy could trigger a panic. 450 */ 451 origcpu = mycpuid; 452 for (cpu = 0; cpu < ncpus; cpu++) { 453 lwkt_migratecpu(cpu); 454 455 arg.rnh = rnh = rt_tables[cpu][AF_INET]; 456 arg.ifa = ifa; 457 arg.del = delete; 458 rnh->rnh_walktree(rnh, in_ifadownkill, &arg); 459 ifa->ifa_flags &= ~IFA_ROUTE; 460 } 461 lwkt_migratecpu(origcpu); 462 return 0; 463 } 464 465 int 466 in_ifadown(struct ifaddr *ifa, int delete) 467 { 468 #ifdef CARP 469 if (ifa->ifa_ifp->if_type == IFT_CARP) 470 return 0; 471 #endif 472 return in_ifadown_force(ifa, delete); 473 } 474