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
31 * This code does two things necessary for the enhanced TCP metrics to
32 * function in a useful manner:
33 * 1) It marks all non-host routes as `cloning', thus ensuring that
34 * every actual reference to such a route actually gets turned
35 * into a reference to a host route to the specific destination
37 * 2) When such routes lose all their references, it arranges for them
38 * to be deleted in some random collection of circumstances, so that
39 * a large quantity of stale routing data is not kept in kernel memory
40 * indefinitely. See in_rtqtimo() below for the exact mechanism.
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD$");
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/callout.h>
56 #include <net/route.h>
57 #include <netinet/in.h>
58 #include <netinet/in_var.h>
59 #include <netinet/ip_var.h>
61 extern int in_inithead(void **head, int off);
63 #define RTPRF_OURS RTF_PROTO3 /* set on routes we manage */
66 * Do what we need to do when inserting a route.
68 static struct radix_node *
69 in_addroute(void *v_arg, void *n_arg, 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 * A little bit of help for both IP output and input:
78 * For host routes, we make sure that RTF_BROADCAST
79 * is set for anything that looks like a broadcast address.
80 * This way, we can avoid an expensive call to in_broadcast()
81 * in ip_output() most of the time (because the route passed
82 * to ip_output() is almost always a host route).
84 * We also do the same for local addresses, with the thought
85 * that this might one day be used to speed up ip_input().
87 * We also 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).
91 if (rt->rt_flags & RTF_HOST) {
92 if (in_broadcast(sin->sin_addr, rt->rt_ifp)) {
93 rt->rt_flags |= RTF_BROADCAST;
94 } else if (satosin(rt->rt_ifa->ifa_addr)->sin_addr.s_addr ==
95 sin->sin_addr.s_addr) {
96 rt->rt_flags |= RTF_LOCAL;
99 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
100 rt->rt_flags |= RTF_MULTICAST;
102 if (!rt->rt_rmx.rmx_mtu && rt->rt_ifp)
103 rt->rt_rmx.rmx_mtu = rt->rt_ifp->if_mtu;
105 ret = rn_addroute(v_arg, n_arg, head, treenodes);
106 if (ret == NULL && rt->rt_flags & RTF_HOST) {
109 * We are trying to add a host route, but can't.
110 * Find out if it is because of an
111 * ARP entry and delete it if so.
113 rt2 = in_rtalloc1((struct sockaddr *)sin, 0,
114 RTF_CLONING|RTF_RNH_LOCKED, rt->rt_fibnum);
116 if (rt2->rt_flags & RTF_LLINFO &&
117 rt2->rt_flags & RTF_HOST &&
119 rt2->rt_gateway->sa_family == AF_LINK) {
122 ret = rn_addroute(v_arg, n_arg, head,
133 * This code is the inverse of in_clsroute: on first reference, if we
134 * were managing the route, stop doing so and set the expiration timer
137 static struct radix_node *
138 in_matroute(void *v_arg, struct radix_node_head *head)
140 struct radix_node *rn = rn_match(v_arg, head);
141 struct rtentry *rt = (struct rtentry *)rn;
144 if (rt && rt->rt_refcnt == 0) { /* this is first reference */
145 if (rt->rt_flags & RTPRF_OURS) {
146 rt->rt_flags &= ~RTPRF_OURS;
147 rt->rt_rmx.rmx_expire = 0;
153 static int rtq_reallyold = 60*60; /* one hour is "really old" */
154 SYSCTL_INT(_net_inet_ip, IPCTL_RTEXPIRE, rtexpire, CTLFLAG_RW,
155 &rtq_reallyold, 0, "Default expiration time on dynamically learned routes");
157 static int rtq_minreallyold = 10; /* never automatically crank down to less */
158 SYSCTL_INT(_net_inet_ip, IPCTL_RTMINEXPIRE, rtminexpire, CTLFLAG_RW,
159 &rtq_minreallyold, 0,
160 "Minimum time to attempt to hold onto dynamically learned routes");
162 static int rtq_toomany = 128; /* 128 cached routes is "too many" */
163 SYSCTL_INT(_net_inet_ip, IPCTL_RTMAXCACHE, rtmaxcache, CTLFLAG_RW,
164 &rtq_toomany, 0, "Upper limit on dynamically learned routes");
167 * On last reference drop, mark the route as belong to us so that it can be
171 in_clsroute(struct radix_node *rn, struct radix_node_head *head)
173 struct rtentry *rt = (struct rtentry *)rn;
177 if (!(rt->rt_flags & RTF_UP))
178 return; /* prophylactic measures */
180 if ((rt->rt_flags & (RTF_LLINFO | RTF_HOST)) != RTF_HOST)
183 if (rt->rt_flags & RTPRF_OURS)
186 if (!(rt->rt_flags & (RTF_WASCLONED | RTF_DYNAMIC)))
190 * If rtq_reallyold is 0, just delete the route without
191 * waiting for a timeout cycle to kill it.
193 if (rtq_reallyold != 0) {
194 rt->rt_flags |= RTPRF_OURS;
195 rt->rt_rmx.rmx_expire = time_uptime + rtq_reallyold;
202 struct radix_node_head *rnh;
211 * Get rid of old routes. When draining, this deletes everything, even when
212 * the timeout is not expired yet. When updating, this makes sure that
213 * nothing has a timeout longer than the current value of rtq_reallyold.
216 in_rtqkill(struct radix_node *rn, void *rock)
218 struct rtqk_arg *ap = rock;
219 struct rtentry *rt = (struct rtentry *)rn;
222 RADIX_NODE_HEAD_LOCK_ASSERT(ap->rnh);
224 if (rt->rt_flags & RTPRF_OURS) {
227 if (ap->draining || rt->rt_rmx.rmx_expire <= time_uptime) {
228 if (rt->rt_refcnt > 0)
229 panic("rtqkill route really not free");
231 err = in_rtrequest(RTM_DELETE,
232 (struct sockaddr *)rt_key(rt),
233 rt->rt_gateway, rt_mask(rt),
234 rt->rt_flags | RTF_RNH_LOCKED, 0,
237 log(LOG_WARNING, "in_rtqkill: error %d\n", err);
243 (rt->rt_rmx.rmx_expire - time_uptime >
245 rt->rt_rmx.rmx_expire =
246 time_uptime + rtq_reallyold;
248 ap->nextstop = lmin(ap->nextstop,
249 rt->rt_rmx.rmx_expire);
256 #define RTQ_TIMEOUT 60*10 /* run no less than once every ten minutes */
257 static int rtq_timeout = RTQ_TIMEOUT;
258 static struct callout rtq_timer;
260 static void in_rtqtimo_one(void *rock);
263 in_rtqtimo(void *rock)
269 KASSERT((rock == (void *)rt_tables[0][AF_INET]),
270 ("in_rtqtimo: unexpected arg"));
271 for (fibnum = 0; fibnum < rt_numfibs; fibnum++) {
272 if ((newrock = rt_tables[fibnum][AF_INET]) != NULL)
273 in_rtqtimo_one(newrock);
276 atv.tv_sec = rtq_timeout;
277 callout_reset(&rtq_timer, tvtohz(&atv), in_rtqtimo, rock);
281 in_rtqtimo_one(void *rock)
283 struct radix_node_head *rnh = rock;
285 static time_t last_adjusted_timeout = 0;
287 arg.found = arg.killed = 0;
289 arg.nextstop = time_uptime + rtq_timeout;
290 arg.draining = arg.updating = 0;
291 RADIX_NODE_HEAD_LOCK(rnh);
292 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
293 RADIX_NODE_HEAD_UNLOCK(rnh);
296 * Attempt to be somewhat dynamic about this:
297 * If there are ``too many'' routes sitting around taking up space,
298 * then crank down the timeout, and see if we can't make some more
299 * go away. However, we make sure that we will never adjust more
300 * than once in rtq_timeout seconds, to keep from cranking down too
303 if ((arg.found - arg.killed > rtq_toomany) &&
304 (time_uptime - last_adjusted_timeout >= rtq_timeout) &&
305 rtq_reallyold > rtq_minreallyold) {
306 rtq_reallyold = 2 * rtq_reallyold / 3;
307 if (rtq_reallyold < rtq_minreallyold) {
308 rtq_reallyold = rtq_minreallyold;
311 last_adjusted_timeout = time_uptime;
313 log(LOG_DEBUG, "in_rtqtimo: adjusted rtq_reallyold to %d\n",
316 arg.found = arg.killed = 0;
318 RADIX_NODE_HEAD_LOCK(rnh);
319 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
320 RADIX_NODE_HEAD_UNLOCK(rnh);
328 struct radix_node_head *rnh;
332 for ( fibnum = 0; fibnum < rt_numfibs; fibnum++) {
333 rnh = rt_tables[fibnum][AF_INET];
334 arg.found = arg.killed = 0;
339 RADIX_NODE_HEAD_LOCK(rnh);
340 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
341 RADIX_NODE_HEAD_UNLOCK(rnh);
345 static int _in_rt_was_here;
347 * Initialize our routing tree.
350 in_inithead(void **head, int off)
352 struct radix_node_head *rnh;
355 * This can be called from vfs_export.c too in which case 'off'
356 * will be 0. We know the correct value so just use that and
357 * return directly if it was 0.
358 * This is a hack that replaces an even worse hack on a bad hack
359 * on a bad design. After RELENG_7 this should be fixed but that
360 * will change the ABI, so for now do it this way.
362 if (!rn_inithead(head, 32))
365 if (off == 0) /* XXX MRT see above */
366 return 1; /* only do the rest for a real routing table */
369 rnh->rnh_addaddr = in_addroute;
370 rnh->rnh_matchaddr = in_matroute;
371 rnh->rnh_close = in_clsroute;
372 if (_in_rt_was_here == 0 ) {
373 callout_init(&rtq_timer, CALLOUT_MPSAFE);
374 in_rtqtimo(rnh); /* kick off timeout first time */
381 * This zaps old routes when the interface goes down or interface
382 * address is deleted. In the latter case, it deletes static routes
383 * that point to this address. If we don't do this, we may end up
384 * using the old address in the future. The ones we always want to
385 * get rid of are things like ARP entries, since the user might down
386 * the interface, walk over to a completely different network, and
389 struct in_ifadown_arg {
390 struct radix_node_head *rnh;
396 in_ifadownkill(struct radix_node *rn, void *xap)
398 struct in_ifadown_arg *ap = xap;
399 struct rtentry *rt = (struct rtentry *)rn;
402 if (rt->rt_ifa == ap->ifa &&
403 (ap->del || !(rt->rt_flags & RTF_STATIC))) {
405 * We need to disable the automatic prune that happens
406 * in this case in rtrequest() because it will blow
407 * away the pointers that rn_walktree() needs in order
408 * continue our descent. We will end up deleting all
409 * the routes that rtrequest() would have in any case,
410 * so that behavior is not needed there.
412 rt->rt_flags &= ~RTF_CLONING;
420 in_ifadown(struct ifaddr *ifa, int delete)
422 struct in_ifadown_arg arg;
423 struct radix_node_head *rnh;
426 if (ifa->ifa_addr->sa_family != AF_INET)
429 for ( fibnum = 0; fibnum < rt_numfibs; fibnum++) {
430 arg.rnh = rnh = rt_tables[fibnum][AF_INET];
433 RADIX_NODE_HEAD_LOCK(rnh);
434 rnh->rnh_walktree(rnh, in_ifadownkill, &arg);
435 RADIX_NODE_HEAD_UNLOCK(rnh);
436 ifa->ifa_flags &= ~IFA_ROUTE; /* XXXlocking? */
442 * inet versions of rt functions. These have fib extensions and
443 * for now will just reference the _fib variants.
444 * eventually this order will be reversed,
447 in_rtalloc_ign(struct route *ro, u_long ignflags, u_int fibnum)
449 rtalloc_ign_fib(ro, ignflags, fibnum);
453 in_rtrequest( int req,
454 struct sockaddr *dst,
455 struct sockaddr *gateway,
456 struct sockaddr *netmask,
458 struct rtentry **ret_nrt,
461 return (rtrequest_fib(req, dst, gateway, netmask,
462 flags, ret_nrt, fibnum));
466 in_rtalloc1(struct sockaddr *dst, int report, u_long ignflags, u_int fibnum)
468 return (rtalloc1_fib(dst, report, ignflags, fibnum));
472 in_rtredirect(struct sockaddr *dst,
473 struct sockaddr *gateway,
474 struct sockaddr *netmask,
476 struct sockaddr *src,
479 rtredirect_fib(dst, gateway, netmask, flags, src, fibnum);
483 in_rtalloc(struct route *ro, u_int fibnum)
485 rtalloc_ign_fib(ro, 0UL, fibnum);
489 int in_rt_getifa(struct rt_addrinfo *, u_int fibnum);
490 int in_rtioctl(u_long, caddr_t, u_int);
491 int in_rtrequest1(int, struct rt_addrinfo *, struct rtentry **, u_int);