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
30 /*
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
36 * requested.
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.
41 */
42
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD$");
45
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>
51 #include <sys/mbuf.h>
52 #include <sys/syslog.h>
53 #include <sys/callout.h>
54
55 #include <net/if.h>
56 #include <net/route.h>
57 #include <netinet/in.h>
58 #include <netinet/in_var.h>
59 #include <netinet/ip_var.h>
60
61 extern int in_inithead(void **head, int off);
62
63 #define RTPRF_OURS RTF_PROTO3 /* set on routes we manage */
64
65 /*
66 * Do what we need to do when inserting a route.
67 */
68 static struct radix_node *
69 in_addroute(void *v_arg, void *n_arg, struct radix_node_head *head,
70 struct radix_node *treenodes)
71 {
72 struct rtentry *rt = (struct rtentry *)treenodes;
73 struct sockaddr_in *sin = (struct sockaddr_in *)rt_key(rt);
74 struct radix_node *ret;
75
76 /*
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).
83 *
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().
86 *
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).
90 */
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;
97 }
98 }
99 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
100 rt->rt_flags |= RTF_MULTICAST;
101
102 if (!rt->rt_rmx.rmx_mtu && rt->rt_ifp)
103 rt->rt_rmx.rmx_mtu = rt->rt_ifp->if_mtu;
104
105 ret = rn_addroute(v_arg, n_arg, head, treenodes);
106 if (ret == NULL && rt->rt_flags & RTF_HOST) {
107 struct rtentry *rt2;
108 /*
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.
112 */
113 rt2 = in_rtalloc1((struct sockaddr *)sin, 0,
114 RTF_CLONING|RTF_RNH_LOCKED, rt->rt_fibnum);
115 if (rt2) {
116 if (rt2->rt_flags & RTF_LLINFO &&
117 rt2->rt_flags & RTF_HOST &&
118 rt2->rt_gateway &&
119 rt2->rt_gateway->sa_family == AF_LINK) {
120 rtexpunge(rt2);
121 RTFREE_LOCKED(rt2);
122 ret = rn_addroute(v_arg, n_arg, head,
123 treenodes);
124 } else
125 RTFREE_LOCKED(rt2);
126 }
127 }
128
129 return ret;
130 }
131
132 /*
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
135 * back off again.
136 */
137 static struct radix_node *
138 in_matroute(void *v_arg, struct radix_node_head *head)
139 {
140 struct radix_node *rn = rn_match(v_arg, head);
141 struct rtentry *rt = (struct rtentry *)rn;
142
143 if (rt) {
144 RT_LOCK(rt);
145 if (rt->rt_flags & RTPRF_OURS) {
146 rt->rt_flags &= ~RTPRF_OURS;
147 rt->rt_rmx.rmx_expire = 0;
148 }
149 RT_UNLOCK(rt);
150 }
151 return rn;
152 }
153
154 static int rtq_reallyold = 60*60; /* one hour is "really old" */
155 SYSCTL_INT(_net_inet_ip, IPCTL_RTEXPIRE, rtexpire, CTLFLAG_RW,
156 &rtq_reallyold, 0, "Default expiration time on dynamically learned routes");
157
158 static int rtq_minreallyold = 10; /* never automatically crank down to less */
159 SYSCTL_INT(_net_inet_ip, IPCTL_RTMINEXPIRE, rtminexpire, CTLFLAG_RW,
160 &rtq_minreallyold, 0,
161 "Minimum time to attempt to hold onto dynamically learned routes");
162
163 static int rtq_toomany = 128; /* 128 cached routes is "too many" */
164 SYSCTL_INT(_net_inet_ip, IPCTL_RTMAXCACHE, rtmaxcache, CTLFLAG_RW,
165 &rtq_toomany, 0, "Upper limit on dynamically learned routes");
166
167 /*
168 * On last reference drop, mark the route as belong to us so that it can be
169 * timed out.
170 */
171 static void
172 in_clsroute(struct radix_node *rn, struct radix_node_head *head)
173 {
174 struct rtentry *rt = (struct rtentry *)rn;
175
176 RT_LOCK_ASSERT(rt);
177
178 if (!(rt->rt_flags & RTF_UP))
179 return; /* prophylactic measures */
180
181 if ((rt->rt_flags & (RTF_LLINFO | RTF_HOST)) != RTF_HOST)
182 return;
183
184 if (rt->rt_flags & RTPRF_OURS)
185 return;
186
187 if (!(rt->rt_flags & (RTF_WASCLONED | RTF_DYNAMIC)))
188 return;
189
190 /*
191 * If rtq_reallyold is 0, just delete the route without
192 * waiting for a timeout cycle to kill it.
193 */
194 if (rtq_reallyold != 0) {
195 rt->rt_flags |= RTPRF_OURS;
196 rt->rt_rmx.rmx_expire = time_uptime + rtq_reallyold;
197 } else {
198 rtexpunge(rt);
199 }
200 }
201
202 struct rtqk_arg {
203 struct radix_node_head *rnh;
204 int draining;
205 int killed;
206 int found;
207 int updating;
208 time_t nextstop;
209 };
210
211 /*
212 * Get rid of old routes. When draining, this deletes everything, even when
213 * the timeout is not expired yet. When updating, this makes sure that
214 * nothing has a timeout longer than the current value of rtq_reallyold.
215 */
216 static int
217 in_rtqkill(struct radix_node *rn, void *rock)
218 {
219 struct rtqk_arg *ap = rock;
220 struct rtentry *rt = (struct rtentry *)rn;
221 int err;
222
223 RADIX_NODE_HEAD_LOCK_ASSERT(ap->rnh);
224
225 if (rt->rt_flags & RTPRF_OURS) {
226 ap->found++;
227
228 if (ap->draining || rt->rt_rmx.rmx_expire <= time_uptime) {
229 if (rt->rt_refcnt > 0)
230 panic("rtqkill route really not free");
231
232 err = in_rtrequest(RTM_DELETE,
233 (struct sockaddr *)rt_key(rt),
234 rt->rt_gateway, rt_mask(rt),
235 rt->rt_flags | RTF_RNH_LOCKED, 0,
236 rt->rt_fibnum);
237 if (err) {
238 log(LOG_WARNING, "in_rtqkill: error %d\n", err);
239 } else {
240 ap->killed++;
241 }
242 } else {
243 if (ap->updating &&
244 (rt->rt_rmx.rmx_expire - time_uptime >
245 rtq_reallyold)) {
246 rt->rt_rmx.rmx_expire =
247 time_uptime + rtq_reallyold;
248 }
249 ap->nextstop = lmin(ap->nextstop,
250 rt->rt_rmx.rmx_expire);
251 }
252 }
253
254 return 0;
255 }
256
257 #define RTQ_TIMEOUT 60*10 /* run no less than once every ten minutes */
258 static int rtq_timeout = RTQ_TIMEOUT;
259 static struct callout rtq_timer;
260
261 static void in_rtqtimo_one(void *rock);
262
263 static void
264 in_rtqtimo(void *rock)
265 {
266 int fibnum;
267 void *newrock;
268 struct timeval atv;
269
270 KASSERT((rock == (void *)rt_tables[0][AF_INET]),
271 ("in_rtqtimo: unexpected arg"));
272 for (fibnum = 0; fibnum < rt_numfibs; fibnum++) {
273 if ((newrock = rt_tables[fibnum][AF_INET]) != NULL)
274 in_rtqtimo_one(newrock);
275 }
276 atv.tv_usec = 0;
277 atv.tv_sec = rtq_timeout;
278 callout_reset(&rtq_timer, tvtohz(&atv), in_rtqtimo, rock);
279 }
280
281 static void
282 in_rtqtimo_one(void *rock)
283 {
284 struct radix_node_head *rnh = rock;
285 struct rtqk_arg arg;
286 static time_t last_adjusted_timeout = 0;
287
288 arg.found = arg.killed = 0;
289 arg.rnh = rnh;
290 arg.nextstop = time_uptime + rtq_timeout;
291 arg.draining = arg.updating = 0;
292 RADIX_NODE_HEAD_LOCK(rnh);
293 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
294 RADIX_NODE_HEAD_UNLOCK(rnh);
295
296 /*
297 * Attempt to be somewhat dynamic about this:
298 * If there are ``too many'' routes sitting around taking up space,
299 * then crank down the timeout, and see if we can't make some more
300 * go away. However, we make sure that we will never adjust more
301 * than once in rtq_timeout seconds, to keep from cranking down too
302 * hard.
303 */
304 if ((arg.found - arg.killed > rtq_toomany) &&
305 (time_uptime - last_adjusted_timeout >= rtq_timeout) &&
306 rtq_reallyold > rtq_minreallyold) {
307 rtq_reallyold = 2 * rtq_reallyold / 3;
308 if (rtq_reallyold < rtq_minreallyold) {
309 rtq_reallyold = rtq_minreallyold;
310 }
311
312 last_adjusted_timeout = time_uptime;
313 #ifdef DIAGNOSTIC
314 log(LOG_DEBUG, "in_rtqtimo: adjusted rtq_reallyold to %d\n",
315 rtq_reallyold);
316 #endif
317 arg.found = arg.killed = 0;
318 arg.updating = 1;
319 RADIX_NODE_HEAD_LOCK(rnh);
320 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
321 RADIX_NODE_HEAD_UNLOCK(rnh);
322 }
323
324 }
325
326 void
327 in_rtqdrain(void)
328 {
329 struct radix_node_head *rnh;
330 struct rtqk_arg arg;
331 int fibnum;
332
333 for ( fibnum = 0; fibnum < rt_numfibs; fibnum++) {
334 rnh = rt_tables[fibnum][AF_INET];
335 arg.found = arg.killed = 0;
336 arg.rnh = rnh;
337 arg.nextstop = 0;
338 arg.draining = 1;
339 arg.updating = 0;
340 RADIX_NODE_HEAD_LOCK(rnh);
341 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
342 RADIX_NODE_HEAD_UNLOCK(rnh);
343 }
344 }
345
346 static int _in_rt_was_here;
347 /*
348 * Initialize our routing tree.
349 */
350 int
351 in_inithead(void **head, int off)
352 {
353 struct radix_node_head *rnh;
354
355 /* XXX MRT
356 * This can be called from vfs_export.c too in which case 'off'
357 * will be 0. We know the correct value so just use that and
358 * return directly if it was 0.
359 * This is a hack that replaces an even worse hack on a bad hack
360 * on a bad design. After RELENG_7 this should be fixed but that
361 * will change the ABI, so for now do it this way.
362 */
363 if (!rn_inithead(head, 32))
364 return 0;
365
366 if (off == 0) /* XXX MRT see above */
367 return 1; /* only do the rest for a real routing table */
368
369 rnh = *head;
370 rnh->rnh_addaddr = in_addroute;
371 rnh->rnh_matchaddr = in_matroute;
372 rnh->rnh_close = in_clsroute;
373 if (_in_rt_was_here == 0 ) {
374 callout_init(&rtq_timer, CALLOUT_MPSAFE);
375 in_rtqtimo(rnh); /* kick off timeout first time */
376 _in_rt_was_here = 1;
377 }
378 return 1;
379 }
380
381 /*
382 * This zaps old routes when the interface goes down or interface
383 * address is deleted. In the latter case, it deletes static routes
384 * that point to this address. If we don't do this, we may end up
385 * using the old address in the future. The ones we always want to
386 * get rid of are things like ARP entries, since the user might down
387 * the interface, walk over to a completely different network, and
388 * plug back in.
389 */
390 struct in_ifadown_arg {
391 struct radix_node_head *rnh;
392 struct ifaddr *ifa;
393 int del;
394 };
395
396 static int
397 in_ifadownkill(struct radix_node *rn, void *xap)
398 {
399 struct in_ifadown_arg *ap = xap;
400 struct rtentry *rt = (struct rtentry *)rn;
401
402 RT_LOCK(rt);
403 if (rt->rt_ifa == ap->ifa &&
404 (ap->del || !(rt->rt_flags & RTF_STATIC))) {
405 /*
406 * Aquire a reference so that it can later be freed
407 * as the refcount would be 0 here in case of at least
408 * ap->del.
409 */
410 RT_ADDREF(rt);
411 /*
412 * We need to disable the automatic prune that happens
413 * in this case in rtrequest() because it will blow
414 * away the pointers that rn_walktree() needs in order
415 * continue our descent. We will end up deleting all
416 * the routes that rtrequest() would have in any case,
417 * so that behavior is not needed there.
418 * Disconnect it from the tree and permit protocols
419 * to cleanup.
420 */
421 rt->rt_flags &= ~RTF_CLONING;
422 rtexpunge(rt);
423 /*
424 * At this point it is an rttrash node, and in case
425 * the above is the only reference we must free it.
426 * If we do not noone will have a pointer and the
427 * rtentry will be leaked forever.
428 * In case someone else holds a reference, we are
429 * fine as we only decrement the refcount. In that
430 * case if the other entity calls RT_REMREF, we
431 * will still be leaking but at least we tried.
432 */
433 RTFREE_LOCKED(rt);
434 return (0);
435 }
436 RT_UNLOCK(rt);
437 return 0;
438 }
439
440 int
441 in_ifadown(struct ifaddr *ifa, int delete)
442 {
443 struct in_ifadown_arg arg;
444 struct radix_node_head *rnh;
445 int fibnum;
446
447 if (ifa->ifa_addr->sa_family != AF_INET)
448 return 1;
449
450 for ( fibnum = 0; fibnum < rt_numfibs; fibnum++) {
451 arg.rnh = rnh = rt_tables[fibnum][AF_INET];
452 arg.ifa = ifa;
453 arg.del = delete;
454 RADIX_NODE_HEAD_LOCK(rnh);
455 rnh->rnh_walktree(rnh, in_ifadownkill, &arg);
456 RADIX_NODE_HEAD_UNLOCK(rnh);
457 ifa->ifa_flags &= ~IFA_ROUTE; /* XXXlocking? */
458 }
459 return 0;
460 }
461
462 /*
463 * inet versions of rt functions. These have fib extensions and
464 * for now will just reference the _fib variants.
465 * eventually this order will be reversed,
466 */
467 void
468 in_rtalloc_ign(struct route *ro, u_long ignflags, u_int fibnum)
469 {
470 rtalloc_ign_fib(ro, ignflags, fibnum);
471 }
472
473 int
474 in_rtrequest( int req,
475 struct sockaddr *dst,
476 struct sockaddr *gateway,
477 struct sockaddr *netmask,
478 int flags,
479 struct rtentry **ret_nrt,
480 u_int fibnum)
481 {
482 return (rtrequest_fib(req, dst, gateway, netmask,
483 flags, ret_nrt, fibnum));
484 }
485
486 struct rtentry *
487 in_rtalloc1(struct sockaddr *dst, int report, u_long ignflags, u_int fibnum)
488 {
489 return (rtalloc1_fib(dst, report, ignflags, fibnum));
490 }
491
492 void
493 in_rtredirect(struct sockaddr *dst,
494 struct sockaddr *gateway,
495 struct sockaddr *netmask,
496 int flags,
497 struct sockaddr *src,
498 u_int fibnum)
499 {
500 rtredirect_fib(dst, gateway, netmask, flags, src, fibnum);
501 }
502
503 void
504 in_rtalloc(struct route *ro, u_int fibnum)
505 {
506 rtalloc_ign_fib(ro, 0UL, fibnum);
507 }
508
509 #if 0 /* not used */
510 int in_rt_getifa(struct rt_addrinfo *, u_int fibnum);
511 int in_rtioctl(u_long, caddr_t, u_int);
512 int in_rtrequest1(int, struct rt_addrinfo *, struct rtentry **, u_int);
513 #endif
514
515
Cache object: 7da82760adf0a07aa1929f1e5d8c7194
|