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/5.3/sys/netinet/in_rmx.c 136588 2004-10-16 08:43:07Z cvs2svn $
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 #include <sys/callout.h>
53
54 #include <net/if.h>
55 #include <net/route.h>
56 #include <netinet/in.h>
57 #include <netinet/in_var.h>
58 #include <netinet/ip_var.h>
59
60 extern int in_inithead(void **head, int off);
61
62 #define RTPRF_OURS RTF_PROTO3 /* set on routes we manage */
63
64 /*
65 * Do what we need to do when inserting a route.
66 */
67 static struct radix_node *
68 in_addroute(void *v_arg, void *n_arg, struct radix_node_head *head,
69 struct radix_node *treenodes)
70 {
71 struct rtentry *rt = (struct rtentry *)treenodes;
72 struct sockaddr_in *sin = (struct sockaddr_in *)rt_key(rt);
73 struct radix_node *ret;
74
75 /*
76 * A little bit of help for both IP output and input:
77 * For host routes, we make sure that RTF_BROADCAST
78 * is set for anything that looks like a broadcast address.
79 * This way, we can avoid an expensive call to in_broadcast()
80 * in ip_output() most of the time (because the route passed
81 * to ip_output() is almost always a host route).
82 *
83 * We also do the same for local addresses, with the thought
84 * that this might one day be used to speed up ip_input().
85 *
86 * We also mark routes to multicast addresses as such, because
87 * it's easy to do and might be useful (but this is much more
88 * dubious since it's so easy to inspect the address).
89 */
90 if (rt->rt_flags & RTF_HOST) {
91 if (in_broadcast(sin->sin_addr, rt->rt_ifp)) {
92 rt->rt_flags |= RTF_BROADCAST;
93 } else if (satosin(rt->rt_ifa->ifa_addr)->sin_addr.s_addr ==
94 sin->sin_addr.s_addr) {
95 rt->rt_flags |= RTF_LOCAL;
96 }
97 }
98 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
99 rt->rt_flags |= RTF_MULTICAST;
100
101 if (!rt->rt_rmx.rmx_mtu && rt->rt_ifp)
102 rt->rt_rmx.rmx_mtu = rt->rt_ifp->if_mtu;
103
104 ret = rn_addroute(v_arg, n_arg, head, treenodes);
105 if (ret == NULL && rt->rt_flags & RTF_HOST) {
106 struct rtentry *rt2;
107 /*
108 * We are trying to add a host route, but can't.
109 * Find out if it is because of an
110 * ARP entry and delete it if so.
111 */
112 rt2 = rtalloc1((struct sockaddr *)sin, 0, RTF_CLONING);
113 if (rt2) {
114 if (rt2->rt_flags & RTF_LLINFO &&
115 rt2->rt_flags & RTF_HOST &&
116 rt2->rt_gateway &&
117 rt2->rt_gateway->sa_family == AF_LINK) {
118 rtexpunge(rt2);
119 RTFREE_LOCKED(rt2);
120 ret = rn_addroute(v_arg, n_arg, head,
121 treenodes);
122 } else
123 RTFREE_LOCKED(rt2);
124 }
125 }
126
127 return ret;
128 }
129
130 /*
131 * This code is the inverse of in_clsroute: on first reference, if we
132 * were managing the route, stop doing so and set the expiration timer
133 * back off again.
134 */
135 static struct radix_node *
136 in_matroute(void *v_arg, struct radix_node_head *head)
137 {
138 struct radix_node *rn = rn_match(v_arg, head);
139 struct rtentry *rt = (struct rtentry *)rn;
140
141 /*XXX locking? */
142 if (rt && rt->rt_refcnt == 0) { /* this is first reference */
143 if (rt->rt_flags & RTPRF_OURS) {
144 rt->rt_flags &= ~RTPRF_OURS;
145 rt->rt_rmx.rmx_expire = 0;
146 }
147 }
148 return rn;
149 }
150
151 static int rtq_reallyold = 60*60; /* one hour is "really old" */
152 SYSCTL_INT(_net_inet_ip, IPCTL_RTEXPIRE, rtexpire, CTLFLAG_RW,
153 &rtq_reallyold, 0, "Default expiration time on dynamically learned routes");
154
155 static int rtq_minreallyold = 10; /* never automatically crank down to less */
156 SYSCTL_INT(_net_inet_ip, IPCTL_RTMINEXPIRE, rtminexpire, CTLFLAG_RW,
157 &rtq_minreallyold, 0,
158 "Minimum time to attempt to hold onto dynamically learned routes");
159
160 static int rtq_toomany = 128; /* 128 cached routes is "too many" */
161 SYSCTL_INT(_net_inet_ip, IPCTL_RTMAXCACHE, rtmaxcache, CTLFLAG_RW,
162 &rtq_toomany, 0, "Upper limit on dynamically learned routes");
163
164 /*
165 * On last reference drop, mark the route as belong to us so that it can be
166 * timed out.
167 */
168 static void
169 in_clsroute(struct radix_node *rn, struct radix_node_head *head)
170 {
171 struct rtentry *rt = (struct rtentry *)rn;
172
173 RT_LOCK_ASSERT(rt);
174
175 if (!(rt->rt_flags & RTF_UP))
176 return; /* prophylactic measures */
177
178 if ((rt->rt_flags & (RTF_LLINFO | RTF_HOST)) != RTF_HOST)
179 return;
180
181 if ((rt->rt_flags & (RTF_WASCLONED | RTPRF_OURS)) != RTF_WASCLONED)
182 return;
183
184 /*
185 * If rtq_reallyold is 0, just delete the route without
186 * waiting for a timeout cycle to kill it.
187 */
188 if (rtq_reallyold != 0) {
189 rt->rt_flags |= RTPRF_OURS;
190 rt->rt_rmx.rmx_expire = time_second + rtq_reallyold;
191 } else {
192 rtexpunge(rt);
193 }
194 }
195
196 struct rtqk_arg {
197 struct radix_node_head *rnh;
198 int draining;
199 int killed;
200 int found;
201 int updating;
202 time_t nextstop;
203 };
204
205 /*
206 * Get rid of old routes. When draining, this deletes everything, even when
207 * the timeout is not expired yet. When updating, this makes sure that
208 * nothing has a timeout longer than the current value of rtq_reallyold.
209 */
210 static int
211 in_rtqkill(struct radix_node *rn, void *rock)
212 {
213 struct rtqk_arg *ap = rock;
214 struct rtentry *rt = (struct rtentry *)rn;
215 int err;
216
217 if (rt->rt_flags & RTPRF_OURS) {
218 ap->found++;
219
220 if (ap->draining || rt->rt_rmx.rmx_expire <= time_second) {
221 if (rt->rt_refcnt > 0)
222 panic("rtqkill route really not free");
223
224 err = rtrequest(RTM_DELETE,
225 (struct sockaddr *)rt_key(rt),
226 rt->rt_gateway, rt_mask(rt),
227 rt->rt_flags, 0);
228 if (err) {
229 log(LOG_WARNING, "in_rtqkill: error %d\n", err);
230 } else {
231 ap->killed++;
232 }
233 } else {
234 if (ap->updating &&
235 (rt->rt_rmx.rmx_expire - time_second >
236 rtq_reallyold)) {
237 rt->rt_rmx.rmx_expire =
238 time_second + rtq_reallyold;
239 }
240 ap->nextstop = lmin(ap->nextstop,
241 rt->rt_rmx.rmx_expire);
242 }
243 }
244
245 return 0;
246 }
247
248 #define RTQ_TIMEOUT 60*10 /* run no less than once every ten minutes */
249 static int rtq_timeout = RTQ_TIMEOUT;
250 static struct callout rtq_timer;
251
252 static void
253 in_rtqtimo(void *rock)
254 {
255 struct radix_node_head *rnh = rock;
256 struct rtqk_arg arg;
257 struct timeval atv;
258 static time_t last_adjusted_timeout = 0;
259
260 arg.found = arg.killed = 0;
261 arg.rnh = rnh;
262 arg.nextstop = time_second + rtq_timeout;
263 arg.draining = arg.updating = 0;
264 RADIX_NODE_HEAD_LOCK(rnh);
265 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
266 RADIX_NODE_HEAD_UNLOCK(rnh);
267
268 /*
269 * Attempt to be somewhat dynamic about this:
270 * If there are ``too many'' routes sitting around taking up space,
271 * then crank down the timeout, and see if we can't make some more
272 * go away. However, we make sure that we will never adjust more
273 * than once in rtq_timeout seconds, to keep from cranking down too
274 * hard.
275 */
276 if ((arg.found - arg.killed > rtq_toomany) &&
277 (time_second - last_adjusted_timeout >= rtq_timeout) &&
278 rtq_reallyold > rtq_minreallyold) {
279 rtq_reallyold = 2 * rtq_reallyold / 3;
280 if (rtq_reallyold < rtq_minreallyold) {
281 rtq_reallyold = rtq_minreallyold;
282 }
283
284 last_adjusted_timeout = time_second;
285 #ifdef DIAGNOSTIC
286 log(LOG_DEBUG, "in_rtqtimo: adjusted rtq_reallyold to %d\n",
287 rtq_reallyold);
288 #endif
289 arg.found = arg.killed = 0;
290 arg.updating = 1;
291 RADIX_NODE_HEAD_LOCK(rnh);
292 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
293 RADIX_NODE_HEAD_UNLOCK(rnh);
294 }
295
296 atv.tv_usec = 0;
297 atv.tv_sec = arg.nextstop - time_second;
298 callout_reset(&rtq_timer, tvtohz(&atv), in_rtqtimo, rock);
299 }
300
301 void
302 in_rtqdrain(void)
303 {
304 struct radix_node_head *rnh = rt_tables[AF_INET];
305 struct rtqk_arg arg;
306
307 arg.found = arg.killed = 0;
308 arg.rnh = rnh;
309 arg.nextstop = 0;
310 arg.draining = 1;
311 arg.updating = 0;
312 RADIX_NODE_HEAD_LOCK(rnh);
313 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
314 RADIX_NODE_HEAD_UNLOCK(rnh);
315 }
316
317 /*
318 * Initialize our routing tree.
319 */
320 int
321 in_inithead(void **head, int off)
322 {
323 struct radix_node_head *rnh;
324
325 if (!rn_inithead(head, off))
326 return 0;
327
328 if (head != (void **)&rt_tables[AF_INET]) /* BOGUS! */
329 return 1; /* only do this for the real routing table */
330
331 rnh = *head;
332 rnh->rnh_addaddr = in_addroute;
333 rnh->rnh_matchaddr = in_matroute;
334 rnh->rnh_close = in_clsroute;
335 callout_init(&rtq_timer, CALLOUT_MPSAFE);
336 in_rtqtimo(rnh); /* kick off timeout first time */
337 return 1;
338 }
339
340 /*
341 * This zaps old routes when the interface goes down or interface
342 * address is deleted. In the latter case, it deletes static routes
343 * that point to this address. If we don't do this, we may end up
344 * using the old address in the future. The ones we always want to
345 * get rid of are things like ARP entries, since the user might down
346 * the interface, walk over to a completely different network, and
347 * plug back in.
348 */
349 struct in_ifadown_arg {
350 struct radix_node_head *rnh;
351 struct ifaddr *ifa;
352 int del;
353 };
354
355 static int
356 in_ifadownkill(struct radix_node *rn, void *xap)
357 {
358 struct in_ifadown_arg *ap = xap;
359 struct rtentry *rt = (struct rtentry *)rn;
360
361 RT_LOCK(rt);
362 if (rt->rt_ifa == ap->ifa &&
363 (ap->del || !(rt->rt_flags & RTF_STATIC))) {
364 /*
365 * We need to disable the automatic prune that happens
366 * in this case in rtrequest() because it will blow
367 * away the pointers that rn_walktree() needs in order
368 * continue our descent. We will end up deleting all
369 * the routes that rtrequest() would have in any case,
370 * so that behavior is not needed there.
371 */
372 rt->rt_flags &= ~RTF_CLONING;
373 rtexpunge(rt);
374 }
375 RT_UNLOCK(rt);
376 return 0;
377 }
378
379 int
380 in_ifadown(struct ifaddr *ifa, int delete)
381 {
382 struct in_ifadown_arg arg;
383 struct radix_node_head *rnh;
384
385 if (ifa->ifa_addr->sa_family != AF_INET)
386 return 1;
387
388 arg.rnh = rnh = rt_tables[AF_INET];
389 arg.ifa = ifa;
390 arg.del = delete;
391 RADIX_NODE_HEAD_LOCK(rnh);
392 rnh->rnh_walktree(rnh, in_ifadownkill, &arg);
393 RADIX_NODE_HEAD_UNLOCK(rnh);
394 ifa->ifa_flags &= ~IFA_ROUTE; /* XXXlocking? */
395 return 0;
396 }
Cache object: 2e86ffd66695b5a626a141771f21c3cc
|