2 * Copyright 1994, 1995 Massachusetts Institute of Technology
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
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
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
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.
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>
51 #include <sys/syslog.h>
54 #include <net/route.h>
55 #include <netinet/in.h>
56 #include <netinet/in_var.h>
58 extern int in_inithead __P((void **head, int off));
60 #define RTPRF_OURS RTF_PROTO3 /* set on routes we manage */
63 * Do what we need to do when inserting a route.
65 static struct radix_node *
66 in_addroute(void *v_arg, void *n_arg, struct radix_node_head *head,
67 struct radix_node *treenodes)
69 struct rtentry *rt = (struct rtentry *)treenodes;
70 struct sockaddr_in *sin = (struct sockaddr_in *)rt_key(rt);
71 struct radix_node *ret;
74 * For IP, all unicast non-host routes are automatically cloning.
76 if(IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
77 rt->rt_flags |= RTF_MULTICAST;
79 if(!(rt->rt_flags & (RTF_HOST | RTF_CLONING | RTF_MULTICAST))) {
80 rt->rt_flags |= RTF_PRCLONING;
84 * A little bit of help for both IP output and input:
85 * For host routes, we make sure that RTF_BROADCAST
86 * is set for anything that looks like a broadcast address.
87 * This way, we can avoid an expensive call to in_broadcast()
88 * in ip_output() most of the time (because the route passed
89 * to ip_output() is almost always a host route).
91 * We also do the same for local addresses, with the thought
92 * that this might one day be used to speed up ip_input().
94 * We also mark routes to multicast addresses as such, because
95 * it's easy to do and might be useful (but this is much more
96 * dubious since it's so easy to inspect the address). (This
99 if (rt->rt_flags & RTF_HOST) {
100 if (in_broadcast(sin->sin_addr, rt->rt_ifp)) {
101 rt->rt_flags |= RTF_BROADCAST;
103 #define satosin(sa) ((struct sockaddr_in *)sa)
104 if (satosin(rt->rt_ifa->ifa_addr)->sin_addr.s_addr
105 == sin->sin_addr.s_addr)
106 rt->rt_flags |= RTF_LOCAL;
111 if (!rt->rt_rmx.rmx_mtu && !(rt->rt_rmx.rmx_locks & RTV_MTU)
113 rt->rt_rmx.rmx_mtu = rt->rt_ifp->if_mtu;
115 ret = rn_addroute(v_arg, n_arg, head, treenodes);
116 if (ret == NULL && rt->rt_flags & RTF_HOST) {
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.
123 rt2 = rtalloc1((struct sockaddr *)sin, 0,
124 RTF_CLONING | RTF_PRCLONING);
126 if (rt2->rt_flags & RTF_LLINFO &&
127 rt2->rt_flags & RTF_HOST &&
129 rt2->rt_gateway->sa_family == AF_LINK) {
130 rtrequest(RTM_DELETE,
131 (struct sockaddr *)rt_key(rt2),
133 rt_mask(rt2), rt2->rt_flags, 0);
134 ret = rn_addroute(v_arg, n_arg, head,
144 * This code is the inverse of in_clsroute: on first reference, if we
145 * were managing the route, stop doing so and set the expiration timer
148 static struct radix_node *
149 in_matroute(void *v_arg, struct radix_node_head *head)
151 struct radix_node *rn = rn_match(v_arg, head);
152 struct rtentry *rt = (struct rtentry *)rn;
154 if(rt && rt->rt_refcnt == 0) { /* this is first reference */
155 if(rt->rt_flags & RTPRF_OURS) {
156 rt->rt_flags &= ~RTPRF_OURS;
157 rt->rt_rmx.rmx_expire = 0;
163 static int rtq_reallyold = 60*60;
164 /* one hour is ``really old'' */
165 SYSCTL_INT(_net_inet_ip, IPCTL_RTEXPIRE, rtexpire, CTLFLAG_RW,
167 "Default expiration time on dynamically learned routes");
169 static int rtq_minreallyold = 10;
170 /* never automatically crank down to less */
171 SYSCTL_INT(_net_inet_ip, IPCTL_RTMINEXPIRE, rtminexpire, CTLFLAG_RW,
172 &rtq_minreallyold , 0,
173 "Minimum time to attempt to hold onto dynamically learned routes");
175 static int rtq_toomany = 128;
176 /* 128 cached routes is ``too many'' */
177 SYSCTL_INT(_net_inet_ip, IPCTL_RTMAXCACHE, rtmaxcache, CTLFLAG_RW,
178 &rtq_toomany , 0, "Upper limit on dynamically learned routes");
181 * On last reference drop, mark the route as belong to us so that it can be
185 in_clsroute(struct radix_node *rn, struct radix_node_head *head)
187 struct rtentry *rt = (struct rtentry *)rn;
189 if(!(rt->rt_flags & RTF_UP))
190 return; /* prophylactic measures */
192 if((rt->rt_flags & (RTF_LLINFO | RTF_HOST)) != RTF_HOST)
195 if((rt->rt_flags & (RTF_WASCLONED | RTPRF_OURS))
200 * As requested by David Greenman:
201 * If rtq_reallyold is 0, just delete the route without
202 * waiting for a timeout cycle to kill it.
204 if(rtq_reallyold != 0) {
205 rt->rt_flags |= RTPRF_OURS;
206 rt->rt_rmx.rmx_expire = time_second + rtq_reallyold;
208 rtrequest(RTM_DELETE,
209 (struct sockaddr *)rt_key(rt),
210 rt->rt_gateway, rt_mask(rt),
216 struct radix_node_head *rnh;
225 * Get rid of old routes. When draining, this deletes everything, even when
226 * the timeout is not expired yet. When updating, this makes sure that
227 * nothing has a timeout longer than the current value of rtq_reallyold.
230 in_rtqkill(struct radix_node *rn, void *rock)
232 struct rtqk_arg *ap = rock;
233 struct rtentry *rt = (struct rtentry *)rn;
236 if(rt->rt_flags & RTPRF_OURS) {
239 if(ap->draining || rt->rt_rmx.rmx_expire <= time_second) {
240 if(rt->rt_refcnt > 0)
241 panic("rtqkill route really not free");
243 err = rtrequest(RTM_DELETE,
244 (struct sockaddr *)rt_key(rt),
245 rt->rt_gateway, rt_mask(rt),
248 log(LOG_WARNING, "in_rtqkill: error %d\n", err);
254 && (rt->rt_rmx.rmx_expire - time_second
256 rt->rt_rmx.rmx_expire = time_second
259 ap->nextstop = lmin(ap->nextstop,
260 rt->rt_rmx.rmx_expire);
267 #define RTQ_TIMEOUT 60*10 /* run no less than once every ten minutes */
268 static int rtq_timeout = RTQ_TIMEOUT;
271 in_rtqtimo(void *rock)
273 struct radix_node_head *rnh = rock;
276 static time_t last_adjusted_timeout = 0;
279 arg.found = arg.killed = 0;
281 arg.nextstop = time_second + rtq_timeout;
282 arg.draining = arg.updating = 0;
284 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
288 * Attempt to be somewhat dynamic about this:
289 * If there are ``too many'' routes sitting around taking up space,
290 * then crank down the timeout, and see if we can't make some more
291 * go away. However, we make sure that we will never adjust more
292 * than once in rtq_timeout seconds, to keep from cranking down too
295 if((arg.found - arg.killed > rtq_toomany)
296 && (time_second - last_adjusted_timeout >= rtq_timeout)
297 && rtq_reallyold > rtq_minreallyold) {
298 rtq_reallyold = 2*rtq_reallyold / 3;
299 if(rtq_reallyold < rtq_minreallyold) {
300 rtq_reallyold = rtq_minreallyold;
303 last_adjusted_timeout = time_second;
305 log(LOG_DEBUG, "in_rtqtimo: adjusted rtq_reallyold to %d\n",
308 arg.found = arg.killed = 0;
311 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
316 atv.tv_sec = arg.nextstop - time_second;
317 timeout(in_rtqtimo, rock, tvtohz(&atv));
323 struct radix_node_head *rnh = rt_tables[AF_INET];
326 arg.found = arg.killed = 0;
332 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
337 * Initialize our routing tree.
340 in_inithead(void **head, int off)
342 struct radix_node_head *rnh;
344 if(!rn_inithead(head, off))
347 if(head != (void **)&rt_tables[AF_INET]) /* BOGUS! */
348 return 1; /* only do this for the real routing table */
351 rnh->rnh_addaddr = in_addroute;
352 rnh->rnh_matchaddr = in_matroute;
353 rnh->rnh_close = in_clsroute;
354 in_rtqtimo(rnh); /* kick off timeout first time */
360 * This zaps old routes when the interface goes down.
361 * Currently it doesn't delete static routes; there are
362 * arguments one could make for both behaviors. For the moment,
363 * we will adopt the Principle of Least Surprise and leave them
364 * alone (with the knowledge that this will not be enough for some
365 * people). The ones we really want to get rid of are things like ARP
366 * entries, since the user might down the interface, walk over to a completely
367 * different network, and plug back in.
369 struct in_ifadown_arg {
370 struct radix_node_head *rnh;
375 in_ifadownkill(struct radix_node *rn, void *xap)
377 struct in_ifadown_arg *ap = xap;
378 struct rtentry *rt = (struct rtentry *)rn;
381 if (rt->rt_ifa == ap->ifa && !(rt->rt_flags & RTF_STATIC)) {
383 * We need to disable the automatic prune that happens
384 * in this case in rtrequest() because it will blow
385 * away the pointers that rn_walktree() needs in order
386 * continue our descent. We will end up deleting all
387 * the routes that rtrequest() would have in any case,
388 * so that behavior is not needed there.
390 rt->rt_flags &= ~RTF_PRCLONING;
391 err = rtrequest(RTM_DELETE, (struct sockaddr *)rt_key(rt),
392 rt->rt_gateway, rt_mask(rt), rt->rt_flags, 0);
394 log(LOG_WARNING, "in_ifadownkill: error %d\n", err);
401 in_ifadown(struct ifaddr *ifa)
403 struct in_ifadown_arg arg;
404 struct radix_node_head *rnh;
406 if (ifa->ifa_addr->sa_family != AF_INET)
409 arg.rnh = rnh = rt_tables[AF_INET];
411 rnh->rnh_walktree(rnh, in_ifadownkill, &arg);
412 ifa->ifa_flags &= ~IFA_ROUTE;