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 1.9 2005/01/06 17:59:32 hsu 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>
55 #include <net/route.h>
56 #include <netinet/in.h>
57 #include <netinet/in_var.h>
58 #include <netinet/ip_var.h>
60 #define RTPRF_EXPIRING RTF_PROTO3 /* set on routes we manage */
62 static struct callout in_rtqtimo_ch;
65 * Do what we need to do when inserting a route.
67 static struct radix_node *
68 in_addroute(char *key, char *mask, struct radix_node_head *head,
69 struct radix_node *treenodes)
71 struct rtentry *rt = (struct rtentry *)treenodes;
72 struct sockaddr_in *sin = (struct sockaddr_in *)rt_key(rt);
73 struct radix_node *ret;
76 * For IP, all unicast non-host routes are automatically cloning.
78 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
79 rt->rt_flags |= RTF_MULTICAST;
81 if (!(rt->rt_flags & (RTF_HOST | RTF_CLONING | RTF_MULTICAST)))
82 rt->rt_flags |= RTF_PRCLONING;
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).
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().
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
100 if (rt->rt_flags & RTF_HOST) {
101 if (in_broadcast(sin->sin_addr, rt->rt_ifp)) {
102 rt->rt_flags |= RTF_BROADCAST;
104 if (satosin(rt->rt_ifa->ifa_addr)->sin_addr.s_addr
105 == sin->sin_addr.s_addr)
106 rt->rt_flags |= RTF_LOCAL;
110 if (rt->rt_rmx.rmx_mtu != 0 && !(rt->rt_rmx.rmx_locks & RTV_MTU) &&
112 rt->rt_rmx.rmx_mtu = rt->rt_ifp->if_mtu;
114 ret = rn_addroute(key, mask, head, treenodes);
115 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 = rtpurelookup((struct sockaddr *)sin);
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, rt_key(rt2),
131 rt2->rt_gateway, rt_mask(rt2),
132 rt2->rt_flags, NULL);
133 ret = rn_addroute(key, mask, head, treenodes);
139 * If the new route created successfully, and we are forwarding,
140 * and there is a cached route, free it. Otherwise, we may end
141 * up using the wrong route.
143 if (ret != NULL && ipforwarding && ipforward_rt.ro_rt != NULL) {
144 RTFREE(ipforward_rt.ro_rt);
145 ipforward_rt.ro_rt = NULL;
152 * This code is the inverse of in_closeroute: on first reference, if we
153 * were managing the route, stop doing so and set the expiration timer
156 static struct radix_node *
157 in_matchroute(char *key, struct radix_node_head *head)
159 struct radix_node *rn = rn_match(key, head);
160 struct rtentry *rt = (struct rtentry *)rn;
162 if (rt != NULL && rt->rt_refcnt == 0) { /* this is first reference */
163 if (rt->rt_flags & RTPRF_EXPIRING) {
164 rt->rt_flags &= ~RTPRF_EXPIRING;
165 rt->rt_rmx.rmx_expire = 0;
171 static int rtq_reallyold = 60*60; /* one hour is ``really old'' */
172 SYSCTL_INT(_net_inet_ip, IPCTL_RTEXPIRE, rtexpire, CTLFLAG_RW,
174 "Default expiration time on cloned routes");
176 static int rtq_minreallyold = 10; /* never automatically crank down to less */
177 SYSCTL_INT(_net_inet_ip, IPCTL_RTMINEXPIRE, rtminexpire, CTLFLAG_RW,
178 &rtq_minreallyold , 0,
179 "Minimum time to attempt to hold onto cloned routes");
181 static int rtq_toomany = 128; /* 128 cached routes is ``too many'' */
182 SYSCTL_INT(_net_inet_ip, IPCTL_RTMAXCACHE, rtmaxcache, CTLFLAG_RW,
183 &rtq_toomany , 0, "Upper limit on cloned routes");
186 * On last reference drop, mark the route as belong to us so that it can be
190 in_closeroute(struct radix_node *rn, struct radix_node_head *head)
192 struct rtentry *rt = (struct rtentry *)rn;
194 if (!(rt->rt_flags & RTF_UP))
195 return; /* prophylactic measures */
197 if ((rt->rt_flags & (RTF_LLINFO | RTF_HOST)) != RTF_HOST)
200 if ((rt->rt_flags & (RTF_WASCLONED | RTPRF_EXPIRING)) != RTF_WASCLONED)
204 * As requested by David Greenman:
205 * If rtq_reallyold is 0, just delete the route without
206 * waiting for a timeout cycle to kill it.
208 if (rtq_reallyold != 0) {
209 rt->rt_flags |= RTPRF_EXPIRING;
210 rt->rt_rmx.rmx_expire = time_second + rtq_reallyold;
213 * Remove route from the radix tree, but defer deallocation
214 * until we return to rtfree().
216 rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway, rt_mask(rt),
222 struct radix_node_head *rnh;
231 * Get rid of old routes. When draining, this deletes everything, even when
232 * the timeout is not expired yet. When updating, this makes sure that
233 * nothing has a timeout longer than the current value of rtq_reallyold.
236 in_rtqkill(struct radix_node *rn, void *rock)
238 struct rtqk_arg *ap = rock;
239 struct rtentry *rt = (struct rtentry *)rn;
242 if (rt->rt_flags & RTPRF_EXPIRING) {
244 if (ap->draining || rt->rt_rmx.rmx_expire <= time_second) {
245 if (rt->rt_refcnt > 0)
246 panic("rtqkill route really not free");
248 err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
249 rt_mask(rt), rt->rt_flags, NULL);
251 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;
282 arg.found = arg.killed = 0;
284 arg.nextstop = time_second + rtq_timeout;
285 arg.draining = arg.updating = 0;
287 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
291 * Attempt to be somewhat dynamic about this:
292 * If there are ``too many'' routes sitting around taking up space,
293 * then crank down the timeout, and see if we can't make some more
294 * go away. However, we make sure that we will never adjust more
295 * than once in rtq_timeout seconds, to keep from cranking down too
298 if ((arg.found - arg.killed > rtq_toomany) &&
299 (time_second - last_adjusted_timeout >= rtq_timeout) &&
300 rtq_reallyold > rtq_minreallyold) {
301 rtq_reallyold = 2*rtq_reallyold / 3;
302 if (rtq_reallyold < rtq_minreallyold) {
303 rtq_reallyold = rtq_minreallyold;
306 last_adjusted_timeout = time_second;
308 log(LOG_DEBUG, "in_rtqtimo: adjusted rtq_reallyold to %d\n",
311 arg.found = arg.killed = 0;
314 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
319 atv.tv_sec = arg.nextstop - time_second;
320 callout_reset(&in_rtqtimo_ch, tvtohz_high(&atv), in_rtqtimo, rock);
326 struct radix_node_head *rnh = rt_tables[AF_INET];
330 arg.found = arg.killed = 0;
336 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
341 * Initialize our routing tree.
344 in_inithead(void **head, int off)
346 struct radix_node_head *rnh;
348 if (!rn_inithead(head, off))
351 if (head != (void **)&rt_tables[AF_INET]) /* BOGUS! */
352 return 1; /* only do this for the real routing table */
355 rnh->rnh_addaddr = in_addroute;
356 rnh->rnh_matchaddr = in_matchroute;
357 rnh->rnh_close = in_closeroute;
358 callout_init(&in_rtqtimo_ch);
359 in_rtqtimo(rnh); /* kick off timeout first time */
365 * This zaps old routes when the interface goes down or interface
366 * address is deleted. In the latter case, it deletes static routes
367 * that point to this address. If we don't do this, we may end up
368 * using the old address in the future. The ones we always want to
369 * get rid of are things like ARP entries, since the user might down
370 * the interface, walk over to a completely different network, and
373 struct in_ifadown_arg {
374 struct radix_node_head *rnh;
380 in_ifadownkill(struct radix_node *rn, void *xap)
382 struct in_ifadown_arg *ap = xap;
383 struct rtentry *rt = (struct rtentry *)rn;
386 if (rt->rt_ifa == ap->ifa &&
387 (ap->del || !(rt->rt_flags & RTF_STATIC))) {
389 * We need to disable the automatic prune that happens
390 * in this case in rtrequest() because it will blow
391 * away the pointers that rn_walktree() needs in order
392 * continue our descent. We will end up deleting all
393 * the routes that rtrequest() would have in any case,
394 * so that behavior is not needed there.
396 rt->rt_flags &= ~(RTF_CLONING | RTF_PRCLONING);
397 err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
398 rt_mask(rt), rt->rt_flags, NULL);
400 log(LOG_WARNING, "in_ifadownkill: error %d\n", err);
407 in_ifadown(struct ifaddr *ifa, int delete)
409 struct in_ifadown_arg arg;
410 struct radix_node_head *rnh;
412 if (ifa->ifa_addr->sa_family != AF_INET)
415 arg.rnh = rnh = rt_tables[AF_INET];
418 rnh->rnh_walktree(rnh, in_ifadownkill, &arg);
419 ifa->ifa_flags &= ~IFA_ROUTE;