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: src/sys/netinet/in_rmx.c,v 1.62 2008/10/02 15:37:58 zec Exp $");
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 #include <sys/vimage.h>
55
56 #include <net/if.h>
57 #include <net/route.h>
58 #include <netinet/in.h>
59 #include <netinet/in_var.h>
60 #include <netinet/ip_var.h>
61
62 extern int in_inithead(void **head, int off);
63
64 #define RTPRF_OURS RTF_PROTO3 /* set on routes we manage */
65
66 /*
67 * Do what we need to do when inserting a route.
68 */
69 static struct radix_node *
70 in_addroute(void *v_arg, void *n_arg, struct radix_node_head *head,
71 struct radix_node *treenodes)
72 {
73 struct rtentry *rt = (struct rtentry *)treenodes;
74 struct sockaddr_in *sin = (struct sockaddr_in *)rt_key(rt);
75 struct radix_node *ret;
76
77 /*
78 * A little bit of help for both IP output and input:
79 * For host routes, we make sure that RTF_BROADCAST
80 * is set for anything that looks like a broadcast address.
81 * This way, we can avoid an expensive call to in_broadcast()
82 * in ip_output() most of the time (because the route passed
83 * to ip_output() is almost always a host route).
84 *
85 * We also do the same for local addresses, with the thought
86 * that this might one day be used to speed up ip_input().
87 *
88 * We also mark routes to multicast addresses as such, because
89 * it's easy to do and might be useful (but this is much more
90 * dubious since it's so easy to inspect the address).
91 */
92 if (rt->rt_flags & RTF_HOST) {
93 if (in_broadcast(sin->sin_addr, rt->rt_ifp)) {
94 rt->rt_flags |= RTF_BROADCAST;
95 } else if (satosin(rt->rt_ifa->ifa_addr)->sin_addr.s_addr ==
96 sin->sin_addr.s_addr) {
97 rt->rt_flags |= RTF_LOCAL;
98 }
99 }
100 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
101 rt->rt_flags |= RTF_MULTICAST;
102
103 if (!rt->rt_rmx.rmx_mtu && rt->rt_ifp)
104 rt->rt_rmx.rmx_mtu = rt->rt_ifp->if_mtu;
105
106 ret = rn_addroute(v_arg, n_arg, head, treenodes);
107 if (ret == NULL && rt->rt_flags & RTF_HOST) {
108 struct rtentry *rt2;
109 /*
110 * We are trying to add a host route, but can't.
111 * Find out if it is because of an
112 * ARP entry and delete it if so.
113 */
114 rt2 = in_rtalloc1((struct sockaddr *)sin, 0,
115 RTF_CLONING, rt->rt_fibnum);
116 if (rt2) {
117 if (rt2->rt_flags & RTF_LLINFO &&
118 rt2->rt_flags & RTF_HOST &&
119 rt2->rt_gateway &&
120 rt2->rt_gateway->sa_family == AF_LINK) {
121 rtexpunge(rt2);
122 RTFREE_LOCKED(rt2);
123 ret = rn_addroute(v_arg, n_arg, head,
124 treenodes);
125 } else
126 RTFREE_LOCKED(rt2);
127 }
128 }
129
130 return ret;
131 }
132
133 /*
134 * This code is the inverse of in_clsroute: on first reference, if we
135 * were managing the route, stop doing so and set the expiration timer
136 * back off again.
137 */
138 static struct radix_node *
139 in_matroute(void *v_arg, struct radix_node_head *head)
140 {
141 struct radix_node *rn = rn_match(v_arg, head);
142 struct rtentry *rt = (struct rtentry *)rn;
143
144 /*XXX locking? */
145 if (rt && rt->rt_refcnt == 0) { /* this is first reference */
146 if (rt->rt_flags & RTPRF_OURS) {
147 rt->rt_flags &= ~RTPRF_OURS;
148 rt->rt_rmx.rmx_expire = 0;
149 }
150 }
151 return rn;
152 }
153
154 static int rtq_reallyold = 60*60; /* one hour is "really old" */
155 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_ip, IPCTL_RTEXPIRE, rtexpire,
156 CTLFLAG_RW, rtq_reallyold, 0,
157 "Default expiration time on dynamically learned routes");
158
159 static int rtq_minreallyold = 10; /* never automatically crank down to less */
160 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_ip, IPCTL_RTMINEXPIRE,
161 rtminexpire, CTLFLAG_RW, rtq_minreallyold, 0,
162 "Minimum time to attempt to hold onto dynamically learned routes");
163
164 static int rtq_toomany = 128; /* 128 cached routes is "too many" */
165 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_ip, IPCTL_RTMAXCACHE,
166 rtmaxcache, CTLFLAG_RW, rtq_toomany, 0,
167 "Upper limit on dynamically learned routes");
168
169 /*
170 * On last reference drop, mark the route as belong to us so that it can be
171 * timed out.
172 */
173 static void
174 in_clsroute(struct radix_node *rn, struct radix_node_head *head)
175 {
176 INIT_VNET_INET(curvnet);
177 struct rtentry *rt = (struct rtentry *)rn;
178
179 RT_LOCK_ASSERT(rt);
180
181 if (!(rt->rt_flags & RTF_UP))
182 return; /* prophylactic measures */
183
184 if ((rt->rt_flags & (RTF_LLINFO | RTF_HOST)) != RTF_HOST)
185 return;
186
187 if (rt->rt_flags & RTPRF_OURS)
188 return;
189
190 if (!(rt->rt_flags & (RTF_WASCLONED | RTF_DYNAMIC)))
191 return;
192
193 /*
194 * If rtq_reallyold is 0, just delete the route without
195 * waiting for a timeout cycle to kill it.
196 */
197 if (V_rtq_reallyold != 0) {
198 rt->rt_flags |= RTPRF_OURS;
199 rt->rt_rmx.rmx_expire = time_uptime + V_rtq_reallyold;
200 } else {
201 rtexpunge(rt);
202 }
203 }
204
205 struct rtqk_arg {
206 struct radix_node_head *rnh;
207 int draining;
208 int killed;
209 int found;
210 int updating;
211 time_t nextstop;
212 };
213
214 /*
215 * Get rid of old routes. When draining, this deletes everything, even when
216 * the timeout is not expired yet. When updating, this makes sure that
217 * nothing has a timeout longer than the current value of rtq_reallyold.
218 */
219 static int
220 in_rtqkill(struct radix_node *rn, void *rock)
221 {
222 INIT_VNET_INET(curvnet);
223 struct rtqk_arg *ap = rock;
224 struct rtentry *rt = (struct rtentry *)rn;
225 int err;
226
227 if (rt->rt_flags & RTPRF_OURS) {
228 ap->found++;
229
230 if (ap->draining || rt->rt_rmx.rmx_expire <= time_uptime) {
231 if (rt->rt_refcnt > 0)
232 panic("rtqkill route really not free");
233
234 err = in_rtrequest(RTM_DELETE,
235 (struct sockaddr *)rt_key(rt),
236 rt->rt_gateway, rt_mask(rt),
237 rt->rt_flags, 0, rt->rt_fibnum);
238 if (err) {
239 log(LOG_WARNING, "in_rtqkill: error %d\n", err);
240 } else {
241 ap->killed++;
242 }
243 } else {
244 if (ap->updating &&
245 (rt->rt_rmx.rmx_expire - time_uptime >
246 V_rtq_reallyold)) {
247 rt->rt_rmx.rmx_expire =
248 time_uptime + V_rtq_reallyold;
249 }
250 ap->nextstop = lmin(ap->nextstop,
251 rt->rt_rmx.rmx_expire);
252 }
253 }
254
255 return 0;
256 }
257
258 #define RTQ_TIMEOUT 60*10 /* run no less than once every ten minutes */
259 static int rtq_timeout = RTQ_TIMEOUT;
260 static struct callout rtq_timer;
261
262 static void in_rtqtimo_one(void *rock);
263
264 static void
265 in_rtqtimo(void *rock)
266 {
267 int fibnum;
268 void *newrock;
269 struct timeval atv;
270
271 KASSERT((rock == (void *)V_rt_tables[0][AF_INET]),
272 ("in_rtqtimo: unexpected arg"));
273 for (fibnum = 0; fibnum < rt_numfibs; fibnum++) {
274 if ((newrock = V_rt_tables[fibnum][AF_INET]) != NULL)
275 in_rtqtimo_one(newrock);
276 }
277 atv.tv_usec = 0;
278 atv.tv_sec = V_rtq_timeout;
279 callout_reset(&V_rtq_timer, tvtohz(&atv), in_rtqtimo, rock);
280 }
281
282 static void
283 in_rtqtimo_one(void *rock)
284 {
285 struct radix_node_head *rnh = rock;
286 struct rtqk_arg arg;
287 static time_t last_adjusted_timeout = 0;
288
289 arg.found = arg.killed = 0;
290 arg.rnh = rnh;
291 arg.nextstop = time_uptime + V_rtq_timeout;
292 arg.draining = arg.updating = 0;
293 RADIX_NODE_HEAD_LOCK(rnh);
294 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
295 RADIX_NODE_HEAD_UNLOCK(rnh);
296
297 /*
298 * Attempt to be somewhat dynamic about this:
299 * If there are ``too many'' routes sitting around taking up space,
300 * then crank down the timeout, and see if we can't make some more
301 * go away. However, we make sure that we will never adjust more
302 * than once in rtq_timeout seconds, to keep from cranking down too
303 * hard.
304 */
305 if ((arg.found - arg.killed > V_rtq_toomany) &&
306 (time_uptime - last_adjusted_timeout >= V_rtq_timeout) &&
307 V_rtq_reallyold > V_rtq_minreallyold) {
308 V_rtq_reallyold = 2 * V_rtq_reallyold / 3;
309 if (V_rtq_reallyold < V_rtq_minreallyold) {
310 V_rtq_reallyold = V_rtq_minreallyold;
311 }
312
313 last_adjusted_timeout = time_uptime;
314 #ifdef DIAGNOSTIC
315 log(LOG_DEBUG, "in_rtqtimo: adjusted rtq_reallyold to %d\n",
316 V_rtq_reallyold);
317 #endif
318 arg.found = arg.killed = 0;
319 arg.updating = 1;
320 RADIX_NODE_HEAD_LOCK(rnh);
321 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
322 RADIX_NODE_HEAD_UNLOCK(rnh);
323 }
324
325 }
326
327 void
328 in_rtqdrain(void)
329 {
330 VNET_ITERATOR_DECL(vnet_iter);
331 struct radix_node_head *rnh;
332 struct rtqk_arg arg;
333 int fibnum;
334
335 VNET_LIST_RLOCK();
336 VNET_FOREACH(vnet_iter) {
337 CURVNET_SET(vnet_iter);
338 INIT_VNET_NET(vnet_iter);
339 for ( fibnum = 0; fibnum < rt_numfibs; fibnum++) {
340 rnh = V_rt_tables[fibnum][AF_INET];
341 arg.found = arg.killed = 0;
342 arg.rnh = rnh;
343 arg.nextstop = 0;
344 arg.draining = 1;
345 arg.updating = 0;
346 RADIX_NODE_HEAD_LOCK(rnh);
347 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
348 RADIX_NODE_HEAD_UNLOCK(rnh);
349 }
350 CURVNET_RESTORE();
351 }
352 VNET_LIST_RUNLOCK();
353 }
354
355 static int _in_rt_was_here;
356 /*
357 * Initialize our routing tree.
358 */
359 int
360 in_inithead(void **head, int off)
361 {
362 INIT_VNET_INET(curvnet);
363 struct radix_node_head *rnh;
364
365 /* XXX MRT
366 * This can be called from vfs_export.c too in which case 'off'
367 * will be 0. We know the correct value so just use that and
368 * return directly if it was 0.
369 * This is a hack that replaces an even worse hack on a bad hack
370 * on a bad design. After RELENG_7 this should be fixed but that
371 * will change the ABI, so for now do it this way.
372 */
373 if (!rn_inithead(head, 32))
374 return 0;
375
376 if (off == 0) /* XXX MRT see above */
377 return 1; /* only do the rest for a real routing table */
378
379 rnh = *head;
380 rnh->rnh_addaddr = in_addroute;
381 rnh->rnh_matchaddr = in_matroute;
382 rnh->rnh_close = in_clsroute;
383 if (_in_rt_was_here == 0 ) {
384 callout_init(&V_rtq_timer, CALLOUT_MPSAFE);
385 in_rtqtimo(rnh); /* kick off timeout first time */
386 _in_rt_was_here = 1;
387 }
388 return 1;
389 }
390
391 /*
392 * This zaps old routes when the interface goes down or interface
393 * address is deleted. In the latter case, it deletes static routes
394 * that point to this address. If we don't do this, we may end up
395 * using the old address in the future. The ones we always want to
396 * get rid of are things like ARP entries, since the user might down
397 * the interface, walk over to a completely different network, and
398 * plug back in.
399 */
400 struct in_ifadown_arg {
401 struct ifaddr *ifa;
402 int del;
403 };
404
405 static int
406 in_ifadownkill(struct radix_node *rn, void *xap)
407 {
408 struct in_ifadown_arg *ap = xap;
409 struct rtentry *rt = (struct rtentry *)rn;
410
411 RT_LOCK(rt);
412 if (rt->rt_ifa == ap->ifa &&
413 (ap->del || !(rt->rt_flags & RTF_STATIC))) {
414 /*
415 * We need to disable the automatic prune that happens
416 * in this case in rtrequest() because it will blow
417 * away the pointers that rn_walktree() needs in order
418 * continue our descent. We will end up deleting all
419 * the routes that rtrequest() would have in any case,
420 * so that behavior is not needed there.
421 */
422 rt->rt_flags &= ~RTF_CLONING;
423 rtexpunge(rt);
424 }
425 RT_UNLOCK(rt);
426 return 0;
427 }
428
429 int
430 in_ifadown(struct ifaddr *ifa, int delete)
431 {
432 INIT_VNET_NET(curvnet);
433 struct in_ifadown_arg arg;
434 struct radix_node_head *rnh;
435 int fibnum;
436
437 if (ifa->ifa_addr->sa_family != AF_INET)
438 return 1;
439
440 for ( fibnum = 0; fibnum < rt_numfibs; fibnum++) {
441 rnh = V_rt_tables[fibnum][AF_INET];
442 arg.ifa = ifa;
443 arg.del = delete;
444 RADIX_NODE_HEAD_LOCK(rnh);
445 rnh->rnh_walktree(rnh, in_ifadownkill, &arg);
446 RADIX_NODE_HEAD_UNLOCK(rnh);
447 ifa->ifa_flags &= ~IFA_ROUTE; /* XXXlocking? */
448 }
449 return 0;
450 }
451
452 /*
453 * inet versions of rt functions. These have fib extensions and
454 * for now will just reference the _fib variants.
455 * eventually this order will be reversed,
456 */
457 void
458 in_rtalloc_ign(struct route *ro, u_long ignflags, u_int fibnum)
459 {
460 rtalloc_ign_fib(ro, ignflags, fibnum);
461 }
462
463 int
464 in_rtrequest( int req,
465 struct sockaddr *dst,
466 struct sockaddr *gateway,
467 struct sockaddr *netmask,
468 int flags,
469 struct rtentry **ret_nrt,
470 u_int fibnum)
471 {
472 return (rtrequest_fib(req, dst, gateway, netmask,
473 flags, ret_nrt, fibnum));
474 }
475
476 struct rtentry *
477 in_rtalloc1(struct sockaddr *dst, int report, u_long ignflags, u_int fibnum)
478 {
479 return (rtalloc1_fib(dst, report, ignflags, fibnum));
480 }
481
482 void
483 in_rtredirect(struct sockaddr *dst,
484 struct sockaddr *gateway,
485 struct sockaddr *netmask,
486 int flags,
487 struct sockaddr *src,
488 u_int fibnum)
489 {
490 rtredirect_fib(dst, gateway, netmask, flags, src, fibnum);
491 }
492
493 void
494 in_rtalloc(struct route *ro, u_int fibnum)
495 {
496 rtalloc_ign_fib(ro, 0UL, fibnum);
497 }
498
499 #if 0
500 int in_rt_getifa(struct rt_addrinfo *, u_int fibnum);
501 int in_rtioctl(u_long, caddr_t, u_int);
502 int in_rtrequest1(int, struct rt_addrinfo *, struct rtentry **, u_int);
503 #endif
504
505
506
|