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 * $Id: in_rmx.c,v 1.16 1995/09/18 15:51:32 wollman 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/queue.h> 49 #include <sys/socket.h> 50 #include <sys/socketvar.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_systm.h> 58 #include <netinet/in_var.h> 59 60 #include <netinet/ip.h> 61 #include <netinet/ip_var.h> 62 63 #include <netinet/tcp.h> 64 #include <netinet/tcp_seq.h> 65 #include <netinet/tcp_timer.h> 66 #include <netinet/tcp_var.h> 67 #ifndef MTUDISC 68 #include <netinet/tcpip.h> 69 #endif /* not MTUDISC */ 70 71 #define RTPRF_OURS RTF_PROTO3 /* set on routes we manage */ 72 73 /* 74 * Do what we need to do when inserting a route. 75 */ 76 static struct radix_node * 77 in_addroute(void *v_arg, void *n_arg, 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 83 /* 84 * For IP, all unicast non-host routes are automatically cloning. 85 */ 86 if(!(rt->rt_flags & (RTF_HOST | RTF_CLONING))) { 87 if(!IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) { 88 rt->rt_flags |= RTF_PRCLONING; 89 } 90 } 91 92 /* 93 * We also specify a send and receive pipe size for every 94 * route added, to help TCP a bit. TCP doesn't actually 95 * want a true pipe size, which would be prohibitive in memory 96 * costs and is hard to compute anyway; it simply uses these 97 * values to size its buffers. So, we fill them in with the 98 * same values that TCP would have used anyway, and allow the 99 * installing program or the link layer to override these values 100 * as it sees fit. This will hopefully allow TCP more 101 * opportunities to save its ssthresh value. 102 */ 103 if (!rt->rt_rmx.rmx_sendpipe && !(rt->rt_rmx.rmx_locks & RTV_SPIPE)) 104 rt->rt_rmx.rmx_sendpipe = tcp_sendspace; 105 106 if (!rt->rt_rmx.rmx_recvpipe && !(rt->rt_rmx.rmx_locks & RTV_RPIPE)) 107 rt->rt_rmx.rmx_recvpipe = tcp_recvspace; 108 109 #ifndef MTUDISC 110 /* 111 * Finally, set an MTU, again duplicating logic in TCP. 112 * The in_localaddr() business will go away when we have 113 * proper PMTU discovery. 114 */ 115 #endif /* not MTUDISC */ 116 if (!rt->rt_rmx.rmx_mtu && !(rt->rt_rmx.rmx_locks & RTV_MTU) 117 && rt->rt_ifp) 118 #ifndef MTUDISC 119 rt->rt_rmx.rmx_mtu = (in_localaddr(sin->sin_addr) 120 ? rt->rt_ifp->if_mtu 121 : tcp_mssdflt + sizeof(struct tcpiphdr)); 122 #else /* MTUDISC */ 123 rt->rt_rmx.rmx_mtu = rt->rt_ifp->if_mtu; 124 #endif /* MTUDISC */ 125 126 return rn_addroute(v_arg, n_arg, head, treenodes); 127 } 128 129 /* 130 * This code is the inverse of in_clsroute: on first reference, if we 131 * were managing the route, stop doing so and set the expiration timer 132 * back off again. 133 */ 134 static struct radix_node * 135 in_matroute(void *v_arg, struct radix_node_head *head) 136 { 137 struct radix_node *rn = rn_match(v_arg, head); 138 struct rtentry *rt = (struct rtentry *)rn; 139 140 if(rt && rt->rt_refcnt == 0) { /* this is first reference */ 141 if(rt->rt_flags & RTPRF_OURS) { 142 rt->rt_flags &= ~RTPRF_OURS; 143 rt->rt_rmx.rmx_expire = 0; 144 } 145 } 146 return rn; 147 } 148 149 /* MIB variables: net.inet.ip.{rtexpire,rtmaxcache,rtminexpire}. */ 150 int rtq_reallyold = 60*60; /* one hour is ``really old'' */ 151 int rtq_toomany = 128; /* 128 cached routes is ``too many'' */ 152 int rtq_minreallyold = 10; /* never automatically crank down to less */ 153 154 /* 155 * On last reference drop, mark the route as belong to us so that it can be 156 * timed out. 157 */ 158 static void 159 in_clsroute(struct radix_node *rn, struct radix_node_head *head) 160 { 161 struct rtentry *rt = (struct rtentry *)rn; 162 163 if(!(rt->rt_flags & RTF_UP)) 164 return; /* prophylactic measures */ 165 166 if((rt->rt_flags & (RTF_LLINFO | RTF_HOST)) != RTF_HOST) 167 return; 168 169 if((rt->rt_flags & (RTF_WASCLONED | RTPRF_OURS)) 170 != RTF_WASCLONED) 171 return; 172 173 /* 174 * As requested by David Greenman: 175 * If rtq_reallyold is 0, just delete the route without 176 * waiting for a timeout cycle to kill it. 177 */ 178 if(rtq_reallyold != 0) { 179 rt->rt_flags |= RTPRF_OURS; 180 rt->rt_rmx.rmx_expire = time.tv_sec + rtq_reallyold; 181 } else { 182 rtrequest(RTM_DELETE, 183 (struct sockaddr *)rt_key(rt), 184 rt->rt_gateway, rt_mask(rt), 185 rt->rt_flags, 0); 186 } 187 } 188 189 struct rtqk_arg { 190 struct radix_node_head *rnh; 191 int draining; 192 int killed; 193 int found; 194 int updating; 195 time_t nextstop; 196 }; 197 198 /* 199 * Get rid of old routes. When draining, this deletes everything, even when 200 * the timeout is not expired yet. When updating, this makes sure that 201 * nothing has a timeout longer than the current value of rtq_reallyold. 202 */ 203 static int 204 in_rtqkill(struct radix_node *rn, void *rock) 205 { 206 struct rtqk_arg *ap = rock; 207 struct rtentry *rt = (struct rtentry *)rn; 208 int err; 209 210 if(rt->rt_flags & RTPRF_OURS) { 211 ap->found++; 212 213 if(ap->draining || rt->rt_rmx.rmx_expire <= time.tv_sec) { 214 if(rt->rt_refcnt > 0) 215 panic("rtqkill route really not free"); 216 217 err = rtrequest(RTM_DELETE, 218 (struct sockaddr *)rt_key(rt), 219 rt->rt_gateway, rt_mask(rt), 220 rt->rt_flags, 0); 221 if(err) { 222 log(LOG_WARNING, "in_rtqkill: error %d\n", err); 223 } else { 224 ap->killed++; 225 } 226 } else { 227 if(ap->updating 228 && (rt->rt_rmx.rmx_expire - time.tv_sec 229 > rtq_reallyold)) { 230 rt->rt_rmx.rmx_expire = time.tv_sec 231 + rtq_reallyold; 232 } 233 ap->nextstop = lmin(ap->nextstop, 234 rt->rt_rmx.rmx_expire); 235 } 236 } 237 238 return 0; 239 } 240 241 #define RTQ_TIMEOUT 60*10 /* run no less than once every ten minutes */ 242 int rtq_timeout = RTQ_TIMEOUT; 243 244 static void 245 in_rtqtimo(void *rock) 246 { 247 struct radix_node_head *rnh = rock; 248 struct rtqk_arg arg; 249 struct timeval atv; 250 static time_t last_adjusted_timeout = 0; 251 int s; 252 253 arg.found = arg.killed = 0; 254 arg.rnh = rnh; 255 arg.nextstop = time.tv_sec + rtq_timeout; 256 arg.draining = arg.updating = 0; 257 s = splnet(); 258 rnh->rnh_walktree(rnh, in_rtqkill, &arg); 259 splx(s); 260 261 /* 262 * Attempt to be somewhat dynamic about this: 263 * If there are ``too many'' routes sitting around taking up space, 264 * then crank down the timeout, and see if we can't make some more 265 * go away. However, we make sure that we will never adjust more 266 * than once in rtq_timeout seconds, to keep from cranking down too 267 * hard. 268 */ 269 if((arg.found - arg.killed > rtq_toomany) 270 && (time.tv_sec - last_adjusted_timeout >= rtq_timeout) 271 && rtq_reallyold > rtq_minreallyold) { 272 rtq_reallyold = 2*rtq_reallyold / 3; 273 if(rtq_reallyold < rtq_minreallyold) { 274 rtq_reallyold = rtq_minreallyold; 275 } 276 277 last_adjusted_timeout = time.tv_sec; 278 log(LOG_DEBUG, "in_rtqtimo: adjusted rtq_reallyold to %d\n", 279 rtq_reallyold); 280 arg.found = arg.killed = 0; 281 arg.updating = 1; 282 s = splnet(); 283 rnh->rnh_walktree(rnh, in_rtqkill, &arg); 284 splx(s); 285 } 286 287 atv.tv_usec = 0; 288 atv.tv_sec = arg.nextstop; 289 timeout(in_rtqtimo, rock, hzto(&atv)); 290 } 291 292 void 293 in_rtqdrain(void) 294 { 295 struct radix_node_head *rnh = rt_tables[AF_INET]; 296 struct rtqk_arg arg; 297 int s; 298 arg.found = arg.killed = 0; 299 arg.rnh = rnh; 300 arg.nextstop = 0; 301 arg.draining = 1; 302 arg.updating = 0; 303 s = splnet(); 304 rnh->rnh_walktree(rnh, in_rtqkill, &arg); 305 splx(s); 306 } 307 308 /* 309 * Initialize our routing tree. 310 */ 311 int 312 in_inithead(void **head, int off) 313 { 314 struct radix_node_head *rnh; 315 316 if(!rn_inithead(head, off)) 317 return 0; 318 319 if(head != (void **)&rt_tables[AF_INET]) /* BOGUS! */ 320 return 1; /* only do this for the real routing table */ 321 322 rnh = *head; 323 rnh->rnh_addaddr = in_addroute; 324 rnh->rnh_matchaddr = in_matroute; 325 rnh->rnh_close = in_clsroute; 326 in_rtqtimo(rnh); /* kick off timeout first time */ 327 return 1; 328 } 329 330