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
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 1.12 2005/06/02 23:52:42 dillon Exp $
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
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.
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>
52 #include <sys/syslog.h>
53 #include <sys/thread2.h>
56 #include <net/route.h>
57 #include <netinet/in.h>
58 #include <netinet/in_var.h>
59 #include <netinet/ip_var.h>
61 #define RTPRF_EXPIRING RTF_PROTO3 /* set on routes we manage */
63 static struct callout in_rtqtimo_ch;
66 * Do what we need to do when inserting a route.
68 static struct radix_node *
69 in_addroute(char *key, char *mask, struct radix_node_head *head,
70 struct radix_node *treenodes)
72 struct rtentry *rt = (struct rtentry *)treenodes;
73 struct sockaddr_in *sin = (struct sockaddr_in *)rt_key(rt);
74 struct radix_node *ret;
77 * For IP, all unicast non-host routes are automatically cloning.
79 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
80 rt->rt_flags |= RTF_MULTICAST;
82 if (!(rt->rt_flags & (RTF_HOST | RTF_CLONING | RTF_MULTICAST)))
83 rt->rt_flags |= RTF_PRCLONING;
86 * A little bit of help for both IP output and input:
87 * For host routes, we make sure that RTF_BROADCAST
88 * is set for anything that looks like a broadcast address.
89 * This way, we can avoid an expensive call to in_broadcast()
90 * in ip_output() most of the time (because the route passed
91 * to ip_output() is almost always a host route).
93 * We also do the same for local addresses, with the thought
94 * that this might one day be used to speed up ip_input().
96 * We also mark routes to multicast addresses as such, because
97 * it's easy to do and might be useful (but this is much more
98 * dubious since it's so easy to inspect the address). (This
101 if (rt->rt_flags & RTF_HOST) {
102 if (in_broadcast(sin->sin_addr, rt->rt_ifp)) {
103 rt->rt_flags |= RTF_BROADCAST;
105 if (satosin(rt->rt_ifa->ifa_addr)->sin_addr.s_addr
106 == sin->sin_addr.s_addr)
107 rt->rt_flags |= RTF_LOCAL;
111 if (rt->rt_rmx.rmx_mtu != 0 && !(rt->rt_rmx.rmx_locks & RTV_MTU) &&
113 rt->rt_rmx.rmx_mtu = rt->rt_ifp->if_mtu;
115 ret = rn_addroute(key, mask, head, treenodes);
116 if (ret == NULL && rt->rt_flags & RTF_HOST) {
117 struct rtentry *oldrt;
120 * We are trying to add a host route, but can't.
121 * Find out if it is because of an ARP entry and
124 oldrt = rtpurelookup((struct sockaddr *)sin);
127 if (oldrt->rt_flags & RTF_LLINFO &&
128 oldrt->rt_flags & RTF_HOST &&
130 oldrt->rt_gateway->sa_family == AF_LINK) {
131 rtrequest(RTM_DELETE, rt_key(oldrt),
132 oldrt->rt_gateway, rt_mask(oldrt),
133 oldrt->rt_flags, NULL);
134 ret = rn_addroute(key, mask, head, treenodes);
140 * If the new route created successfully, and we are forwarding,
141 * and there is a cached route, free it. Otherwise, we may end
142 * up using the wrong route.
144 if (ret != NULL && ipforwarding && ipforward_rt.ro_rt != NULL) {
145 RTFREE(ipforward_rt.ro_rt);
146 ipforward_rt.ro_rt = NULL;
153 * This code is the inverse of in_closeroute: on first reference, if we
154 * were managing the route, stop doing so and set the expiration timer
157 static struct radix_node *
158 in_matchroute(char *key, struct radix_node_head *head)
160 struct radix_node *rn = rn_match(key, head);
161 struct rtentry *rt = (struct rtentry *)rn;
163 if (rt != NULL && rt->rt_refcnt == 0) { /* this is first reference */
164 if (rt->rt_flags & RTPRF_EXPIRING) {
165 rt->rt_flags &= ~RTPRF_EXPIRING;
166 rt->rt_rmx.rmx_expire = 0;
172 static int rtq_reallyold = 60*60; /* one hour is ``really old'' */
173 SYSCTL_INT(_net_inet_ip, IPCTL_RTEXPIRE, rtexpire, CTLFLAG_RW,
175 "Default expiration time on cloned routes");
177 static int rtq_minreallyold = 10; /* never automatically crank down to less */
178 SYSCTL_INT(_net_inet_ip, IPCTL_RTMINEXPIRE, rtminexpire, CTLFLAG_RW,
179 &rtq_minreallyold , 0,
180 "Minimum time to attempt to hold onto cloned routes");
182 static int rtq_toomany = 128; /* 128 cached routes is ``too many'' */
183 SYSCTL_INT(_net_inet_ip, IPCTL_RTMAXCACHE, rtmaxcache, CTLFLAG_RW,
184 &rtq_toomany , 0, "Upper limit on cloned routes");
187 * On last reference drop, mark the route as belong to us so that it can be
191 in_closeroute(struct radix_node *rn, struct radix_node_head *head)
193 struct rtentry *rt = (struct rtentry *)rn;
195 if (!(rt->rt_flags & RTF_UP))
196 return; /* prophylactic measures */
198 if ((rt->rt_flags & (RTF_LLINFO | RTF_HOST)) != RTF_HOST)
201 if ((rt->rt_flags & (RTF_WASCLONED | RTPRF_EXPIRING)) != RTF_WASCLONED)
205 * As requested by David Greenman:
206 * If rtq_reallyold is 0, just delete the route without
207 * waiting for a timeout cycle to kill it.
209 if (rtq_reallyold != 0) {
210 rt->rt_flags |= RTPRF_EXPIRING;
211 rt->rt_rmx.rmx_expire = time_second + rtq_reallyold;
214 * Remove route from the radix tree, but defer deallocation
215 * until we return to rtfree().
217 rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway, rt_mask(rt),
223 struct radix_node_head *rnh;
232 * Get rid of old routes. When draining, this deletes everything, even when
233 * the timeout is not expired yet. When updating, this makes sure that
234 * nothing has a timeout longer than the current value of rtq_reallyold.
237 in_rtqkill(struct radix_node *rn, void *rock)
239 struct rtqk_arg *ap = rock;
240 struct rtentry *rt = (struct rtentry *)rn;
243 if (rt->rt_flags & RTPRF_EXPIRING) {
245 if (ap->draining || rt->rt_rmx.rmx_expire <= time_second) {
246 if (rt->rt_refcnt > 0)
247 panic("rtqkill route really not free");
249 err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
250 rt_mask(rt), rt->rt_flags, NULL);
252 log(LOG_WARNING, "in_rtqkill: error %d\n", err);
257 (rt->rt_rmx.rmx_expire - time_second >
259 rt->rt_rmx.rmx_expire = time_second +
262 ap->nextstop = lmin(ap->nextstop,
263 rt->rt_rmx.rmx_expire);
270 #define RTQ_TIMEOUT 60*10 /* run no less than once every ten minutes */
271 static int rtq_timeout = RTQ_TIMEOUT;
274 in_rtqtimo(void *rock)
276 struct radix_node_head *rnh = rock;
279 static time_t last_adjusted_timeout = 0;
281 arg.found = arg.killed = 0;
283 arg.nextstop = time_second + rtq_timeout;
284 arg.draining = arg.updating = 0;
286 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
290 * Attempt to be somewhat dynamic about this:
291 * If there are ``too many'' routes sitting around taking up space,
292 * then crank down the timeout, and see if we can't make some more
293 * go away. However, we make sure that we will never adjust more
294 * than once in rtq_timeout seconds, to keep from cranking down too
297 if ((arg.found - arg.killed > rtq_toomany) &&
298 (time_second - last_adjusted_timeout >= rtq_timeout) &&
299 rtq_reallyold > rtq_minreallyold) {
300 rtq_reallyold = 2*rtq_reallyold / 3;
301 if (rtq_reallyold < rtq_minreallyold) {
302 rtq_reallyold = rtq_minreallyold;
305 last_adjusted_timeout = time_second;
307 log(LOG_DEBUG, "in_rtqtimo: adjusted rtq_reallyold to %d\n",
310 arg.found = arg.killed = 0;
313 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
318 atv.tv_sec = arg.nextstop - time_second;
319 callout_reset(&in_rtqtimo_ch, tvtohz_high(&atv), in_rtqtimo, rock);
325 struct radix_node_head *rnh = rt_tables[AF_INET];
328 arg.found = arg.killed = 0;
334 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
339 * Initialize our routing tree.
342 in_inithead(void **head, int off)
344 struct radix_node_head *rnh;
346 if (!rn_inithead(head, off))
349 if (head != (void **)&rt_tables[AF_INET]) /* BOGUS! */
350 return 1; /* only do this for the real routing table */
353 rnh->rnh_addaddr = in_addroute;
354 rnh->rnh_matchaddr = in_matchroute;
355 rnh->rnh_close = in_closeroute;
356 callout_init(&in_rtqtimo_ch);
357 in_rtqtimo(rnh); /* kick off timeout first time */
362 * This zaps old routes when the interface goes down or interface
363 * address is deleted. In the latter case, it deletes static routes
364 * that point to this address. If we don't do this, we may end up
365 * using the old address in the future. The ones we always want to
366 * get rid of are things like ARP entries, since the user might down
367 * the interface, walk over to a completely different network, and
370 struct in_ifadown_arg {
371 struct radix_node_head *rnh;
377 in_ifadownkill(struct radix_node *rn, void *xap)
379 struct in_ifadown_arg *ap = xap;
380 struct rtentry *rt = (struct rtentry *)rn;
383 if (rt->rt_ifa == ap->ifa &&
384 (ap->del || !(rt->rt_flags & RTF_STATIC))) {
386 * We need to disable the automatic prune that happens
387 * in this case in rtrequest() because it will blow
388 * away the pointers that rn_walktree() needs in order
389 * continue our descent. We will end up deleting all
390 * the routes that rtrequest() would have in any case,
391 * so that behavior is not needed there.
393 rt->rt_flags &= ~(RTF_CLONING | RTF_PRCLONING);
394 err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
395 rt_mask(rt), rt->rt_flags, NULL);
397 log(LOG_WARNING, "in_ifadownkill: error %d\n", err);
403 in_ifadown(struct ifaddr *ifa, int delete)
405 struct in_ifadown_arg arg;
406 struct radix_node_head *rnh;
408 if (ifa->ifa_addr->sa_family != AF_INET)
411 arg.rnh = rnh = rt_tables[AF_INET];
414 rnh->rnh_walktree(rnh, in_ifadownkill, &arg);
415 ifa->ifa_flags &= ~IFA_ROUTE;