FreeBSD/Linux Kernel Cross Reference
sys/netinet/in_rmx.c
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 * $FreeBSD: releng/5.1/sys/netinet/in_rmx.c 110656 2003-02-10 22:01:34Z hsu $
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/sysctl.h>
49 #include <sys/socket.h>
50 #include <sys/mbuf.h>
51 #include <sys/syslog.h>
52
53 #include <net/if.h>
54 #include <net/route.h>
55 #include <netinet/in.h>
56 #include <netinet/in_var.h>
57 #include <netinet/ip_var.h>
58
59 extern int in_inithead(void **head, int off);
60
61 #define RTPRF_OURS RTF_PROTO3 /* set on routes we manage */
62
63 /*
64 * Do what we need to do when inserting a route.
65 */
66 static struct radix_node *
67 in_addroute(void *v_arg, void *n_arg, struct radix_node_head *head,
68 struct radix_node *treenodes)
69 {
70 struct rtentry *rt = (struct rtentry *)treenodes;
71 struct sockaddr_in *sin = (struct sockaddr_in *)rt_key(rt);
72 struct radix_node *ret;
73
74 /*
75 * For IP, all unicast non-host routes are automatically cloning.
76 */
77 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
78 rt->rt_flags |= RTF_MULTICAST;
79
80 if (!(rt->rt_flags & (RTF_HOST | RTF_CLONING | RTF_MULTICAST)))
81 rt->rt_flags |= RTF_PRCLONING;
82
83 /*
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).
90 *
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().
93 *
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
97 * is done above.)
98 */
99 if (rt->rt_flags & RTF_HOST) {
100 if (in_broadcast(sin->sin_addr, rt->rt_ifp)) {
101 rt->rt_flags |= RTF_BROADCAST;
102 } else if (satosin(rt->rt_ifa->ifa_addr)->sin_addr.s_addr ==
103 sin->sin_addr.s_addr) {
104 rt->rt_flags |= RTF_LOCAL;
105 }
106 }
107
108 if (!rt->rt_rmx.rmx_mtu && !(rt->rt_rmx.rmx_locks & RTV_MTU) &&
109 rt->rt_ifp)
110 rt->rt_rmx.rmx_mtu = rt->rt_ifp->if_mtu;
111
112 ret = rn_addroute(v_arg, n_arg, head, treenodes);
113 if (ret == NULL && rt->rt_flags & RTF_HOST) {
114 struct rtentry *rt2;
115 /*
116 * We are trying to add a host route, but can't.
117 * Find out if it is because of an
118 * ARP entry and delete it if so.
119 */
120 rt2 = rtalloc1((struct sockaddr *)sin, 0,
121 RTF_CLONING | RTF_PRCLONING);
122 if (rt2) {
123 if (rt2->rt_flags & RTF_LLINFO &&
124 rt2->rt_flags & RTF_HOST &&
125 rt2->rt_gateway &&
126 rt2->rt_gateway->sa_family == AF_LINK) {
127 rtrequest(RTM_DELETE,
128 (struct sockaddr *)rt_key(rt2),
129 rt2->rt_gateway, rt_mask(rt2),
130 rt2->rt_flags, 0);
131 ret = rn_addroute(v_arg, n_arg, head,
132 treenodes);
133 }
134 RTFREE(rt2);
135 }
136 }
137
138 /*
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.
142 */
143 if (ret != NULL && ipforwarding && ipforward_rt.ro_rt) {
144 RTFREE(ipforward_rt.ro_rt);
145 ipforward_rt.ro_rt = 0;
146 }
147
148 return ret;
149 }
150
151 /*
152 * This code is the inverse of in_clsroute: on first reference, if we
153 * were managing the route, stop doing so and set the expiration timer
154 * back off again.
155 */
156 static struct radix_node *
157 in_matroute(void *v_arg, struct radix_node_head *head)
158 {
159 struct radix_node *rn = rn_match(v_arg, head);
160 struct rtentry *rt = (struct rtentry *)rn;
161
162 if (rt && rt->rt_refcnt == 0) { /* this is first reference */
163 if (rt->rt_flags & RTPRF_OURS) {
164 rt->rt_flags &= ~RTPRF_OURS;
165 rt->rt_rmx.rmx_expire = 0;
166 }
167 }
168 return rn;
169 }
170
171 static int rtq_reallyold = 60*60; /* one hour is "really old" */
172 SYSCTL_INT(_net_inet_ip, IPCTL_RTEXPIRE, rtexpire, CTLFLAG_RW,
173 &rtq_reallyold, 0, "Default expiration time on dynamically learned routes");
174
175 static int rtq_minreallyold = 10; /* never automatically crank down to less */
176 SYSCTL_INT(_net_inet_ip, IPCTL_RTMINEXPIRE, rtminexpire, CTLFLAG_RW,
177 &rtq_minreallyold, 0,
178 "Minimum time to attempt to hold onto dynamically learned routes");
179
180 static int rtq_toomany = 128; /* 128 cached routes is "too many" */
181 SYSCTL_INT(_net_inet_ip, IPCTL_RTMAXCACHE, rtmaxcache, CTLFLAG_RW,
182 &rtq_toomany, 0, "Upper limit on dynamically learned routes");
183
184 /*
185 * On last reference drop, mark the route as belong to us so that it can be
186 * timed out.
187 */
188 static void
189 in_clsroute(struct radix_node *rn, struct radix_node_head *head)
190 {
191 struct rtentry *rt = (struct rtentry *)rn;
192
193 if (!(rt->rt_flags & RTF_UP))
194 return; /* prophylactic measures */
195
196 if ((rt->rt_flags & (RTF_LLINFO | RTF_HOST)) != RTF_HOST)
197 return;
198
199 if ((rt->rt_flags & (RTF_WASCLONED | RTPRF_OURS)) != RTF_WASCLONED)
200 return;
201
202 /*
203 * If rtq_reallyold is 0, just delete the route without
204 * waiting for a timeout cycle to kill it.
205 */
206 if (rtq_reallyold != 0) {
207 rt->rt_flags |= RTPRF_OURS;
208 rt->rt_rmx.rmx_expire = time_second + rtq_reallyold;
209 } else {
210 rtrequest(RTM_DELETE,
211 (struct sockaddr *)rt_key(rt),
212 rt->rt_gateway, rt_mask(rt),
213 rt->rt_flags, 0);
214 }
215 }
216
217 struct rtqk_arg {
218 struct radix_node_head *rnh;
219 int draining;
220 int killed;
221 int found;
222 int updating;
223 time_t nextstop;
224 };
225
226 /*
227 * Get rid of old routes. When draining, this deletes everything, even when
228 * the timeout is not expired yet. When updating, this makes sure that
229 * nothing has a timeout longer than the current value of rtq_reallyold.
230 */
231 static int
232 in_rtqkill(struct radix_node *rn, void *rock)
233 {
234 struct rtqk_arg *ap = rock;
235 struct rtentry *rt = (struct rtentry *)rn;
236 int err;
237
238 if (rt->rt_flags & RTPRF_OURS) {
239 ap->found++;
240
241 if (ap->draining || rt->rt_rmx.rmx_expire <= time_second) {
242 if (rt->rt_refcnt > 0)
243 panic("rtqkill route really not free");
244
245 err = rtrequest(RTM_DELETE,
246 (struct sockaddr *)rt_key(rt),
247 rt->rt_gateway, rt_mask(rt),
248 rt->rt_flags, 0);
249 if (err) {
250 log(LOG_WARNING, "in_rtqkill: error %d\n", err);
251 } else {
252 ap->killed++;
253 }
254 } else {
255 if (ap->updating &&
256 (rt->rt_rmx.rmx_expire - time_second >
257 rtq_reallyold)) {
258 rt->rt_rmx.rmx_expire =
259 time_second + rtq_reallyold;
260 }
261 ap->nextstop = lmin(ap->nextstop,
262 rt->rt_rmx.rmx_expire);
263 }
264 }
265
266 return 0;
267 }
268
269 #define RTQ_TIMEOUT 60*10 /* run no less than once every ten minutes */
270 static int rtq_timeout = RTQ_TIMEOUT;
271
272 static void
273 in_rtqtimo(void *rock)
274 {
275 struct radix_node_head *rnh = rock;
276 struct rtqk_arg arg;
277 struct timeval atv;
278 static time_t last_adjusted_timeout = 0;
279 int s;
280
281 arg.found = arg.killed = 0;
282 arg.rnh = rnh;
283 arg.nextstop = time_second + rtq_timeout;
284 arg.draining = arg.updating = 0;
285 s = splnet();
286 RADIX_NODE_HEAD_LOCK(rnh);
287 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
288 RADIX_NODE_HEAD_UNLOCK(rnh);
289 splx(s);
290
291 /*
292 * Attempt to be somewhat dynamic about this:
293 * If there are ``too many'' routes sitting around taking up space,
294 * then crank down the timeout, and see if we can't make some more
295 * go away. However, we make sure that we will never adjust more
296 * than once in rtq_timeout seconds, to keep from cranking down too
297 * hard.
298 */
299 if ((arg.found - arg.killed > rtq_toomany) &&
300 (time_second - last_adjusted_timeout >= rtq_timeout) &&
301 rtq_reallyold > rtq_minreallyold) {
302 rtq_reallyold = 2 * rtq_reallyold / 3;
303 if (rtq_reallyold < rtq_minreallyold) {
304 rtq_reallyold = rtq_minreallyold;
305 }
306
307 last_adjusted_timeout = time_second;
308 #ifdef DIAGNOSTIC
309 log(LOG_DEBUG, "in_rtqtimo: adjusted rtq_reallyold to %d\n",
310 rtq_reallyold);
311 #endif
312 arg.found = arg.killed = 0;
313 arg.updating = 1;
314 s = splnet();
315 RADIX_NODE_HEAD_LOCK(rnh);
316 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
317 RADIX_NODE_HEAD_UNLOCK(rnh);
318 splx(s);
319 }
320
321 atv.tv_usec = 0;
322 atv.tv_sec = arg.nextstop - time_second;
323 timeout(in_rtqtimo, rock, tvtohz(&atv));
324 }
325
326 void
327 in_rtqdrain(void)
328 {
329 struct radix_node_head *rnh = rt_tables[AF_INET];
330 struct rtqk_arg arg;
331 int s;
332 arg.found = arg.killed = 0;
333 arg.rnh = rnh;
334 arg.nextstop = 0;
335 arg.draining = 1;
336 arg.updating = 0;
337 s = splnet();
338 RADIX_NODE_HEAD_LOCK(rnh);
339 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
340 RADIX_NODE_HEAD_UNLOCK(rnh);
341 splx(s);
342 }
343
344 /*
345 * Initialize our routing tree.
346 */
347 int
348 in_inithead(void **head, int off)
349 {
350 struct radix_node_head *rnh;
351
352 if (!rn_inithead(head, off))
353 return 0;
354
355 if (head != (void **)&rt_tables[AF_INET]) /* BOGUS! */
356 return 1; /* only do this for the real routing table */
357
358 rnh = *head;
359 rnh->rnh_addaddr = in_addroute;
360 rnh->rnh_matchaddr = in_matroute;
361 rnh->rnh_close = in_clsroute;
362 in_rtqtimo(rnh); /* kick off timeout first time */
363 return 1;
364 }
365
366 /*
367 * This zaps old routes when the interface goes down or interface
368 * address is deleted. In the latter case, it deletes static routes
369 * that point to this address. If we don't do this, we may end up
370 * using the old address in the future. The ones we always want to
371 * get rid of are things like ARP entries, since the user might down
372 * the interface, walk over to a completely different network, and
373 * plug back in.
374 */
375 struct in_ifadown_arg {
376 struct radix_node_head *rnh;
377 struct ifaddr *ifa;
378 int del;
379 };
380
381 static int
382 in_ifadownkill(struct radix_node *rn, void *xap)
383 {
384 struct in_ifadown_arg *ap = xap;
385 struct rtentry *rt = (struct rtentry *)rn;
386 int err;
387
388 if (rt->rt_ifa == ap->ifa &&
389 (ap->del || !(rt->rt_flags & RTF_STATIC))) {
390 /*
391 * We need to disable the automatic prune that happens
392 * in this case in rtrequest() because it will blow
393 * away the pointers that rn_walktree() needs in order
394 * continue our descent. We will end up deleting all
395 * the routes that rtrequest() would have in any case,
396 * so that behavior is not needed there.
397 */
398 rt->rt_flags &= ~(RTF_CLONING | RTF_PRCLONING);
399 err = rtrequest(RTM_DELETE, (struct sockaddr *)rt_key(rt),
400 rt->rt_gateway, rt_mask(rt), rt->rt_flags, 0);
401 if (err) {
402 log(LOG_WARNING, "in_ifadownkill: error %d\n", err);
403 }
404 }
405 return 0;
406 }
407
408 int
409 in_ifadown(struct ifaddr *ifa, int delete)
410 {
411 struct in_ifadown_arg arg;
412 struct radix_node_head *rnh;
413
414 if (ifa->ifa_addr->sa_family != AF_INET)
415 return 1;
416
417 arg.rnh = rnh = rt_tables[AF_INET];
418 arg.ifa = ifa;
419 arg.del = delete;
420 RADIX_NODE_HEAD_LOCK(rnh);
421 rnh->rnh_walktree(rnh, in_ifadownkill, &arg);
422 RADIX_NODE_HEAD_UNLOCK(rnh);
423 ifa->ifa_flags &= ~IFA_ROUTE;
424 return 0;
425 }
Cache object: d65d0f1da871165f4f568061b91da015
|