1 /*
2 * Copyright (c) 2000-2017 Apple Inc. All rights reserved.
3 *
4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5 *
6 * This file contains Original Code and/or Modifications of Original Code
7 * as defined in and that are subject to the Apple Public Source License
8 * Version 2.0 (the 'License'). You may not use this file except in
9 * compliance with the License. The rights granted to you under the License
10 * may not be used to create, or enable the creation or redistribution of,
11 * unlawful or unlicensed copies of an Apple operating system, or to
12 * circumvent, violate, or enable the circumvention or violation of, any
13 * terms of an Apple operating system software license agreement.
14 *
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
17 *
18 * The Original Code and all software distributed under the License are
19 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23 * Please see the License for the specific language governing rights and
24 * limitations under the License.
25 *
26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27 */
28 /*
29 * Copyright 1994, 1995 Massachusetts Institute of Technology
30 *
31 * Permission to use, copy, modify, and distribute this software and
32 * its documentation for any purpose and without fee is hereby
33 * granted, provided that both the above copyright notice and this
34 * permission notice appear in all copies, that both the above
35 * copyright notice and this permission notice appear in all
36 * supporting documentation, and that the name of M.I.T. not be used
37 * in advertising or publicity pertaining to distribution of the
38 * software without specific, written prior permission. M.I.T. makes
39 * no representations about the suitability of this software for any
40 * purpose. It is provided "as is" without express or implied
41 * warranty.
42 *
43 * THIS SOFTWARE IS PROVIDED BY M.I.T. ``AS IS''. M.I.T. DISCLAIMS
44 * ALL EXPRESS OR IMPLIED WARRANTIES WITH REGARD TO THIS SOFTWARE,
45 * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
46 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. IN NO EVENT
47 * SHALL M.I.T. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
48 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
49 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF
50 * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
51 * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
52 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
53 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
54 * SUCH DAMAGE.
55 *
56 */
57
58 /*
59 * This code does two things necessary for the enhanced TCP metrics to
60 * function in a useful manner:
61 * 1) It marks all non-host routes as `cloning', thus ensuring that
62 * every actual reference to such a route actually gets turned
63 * into a reference to a host route to the specific destination
64 * requested.
65 * 2) When such routes lose all their references, it arranges for them
66 * to be deleted in some random collection of circumstances, so that
67 * a large quantity of stale routing data is not kept in kernel memory
68 * indefinitely. See in_rtqtimo() below for the exact mechanism.
69 */
70
71 #include <sys/param.h>
72 #include <sys/systm.h>
73 #include <sys/kernel.h>
74 #include <sys/sysctl.h>
75 #include <sys/socket.h>
76 #include <sys/mbuf.h>
77 #include <sys/protosw.h>
78 #include <sys/syslog.h>
79 #include <sys/mcache.h>
80 #include <kern/locks.h>
81
82 #include <net/if.h>
83 #include <net/route.h>
84 #include <netinet/in.h>
85 #include <netinet/in_var.h>
86 #include <netinet/in_arp.h>
87 #include <netinet/ip.h>
88 #include <netinet/ip6.h>
89 #include <netinet6/nd6.h>
90
91 extern int tvtohz(struct timeval *);
92
93 static int in_rtqtimo_run; /* in_rtqtimo is scheduled to run */
94 static void in_rtqtimo(void *);
95 static void in_sched_rtqtimo(struct timeval *);
96
97 static struct radix_node *in_addroute(void *, void *, struct radix_node_head *,
98 struct radix_node *);
99 static struct radix_node *in_deleteroute(void *, void *,
100 struct radix_node_head *);
101 static struct radix_node *in_matroute(void *, struct radix_node_head *);
102 static struct radix_node *in_matroute_args(void *, struct radix_node_head *,
103 rn_matchf_t *f, void *);
104 static void in_clsroute(struct radix_node *, struct radix_node_head *);
105 static int in_rtqkill(struct radix_node *, void *);
106
107 static int in_ifadownkill(struct radix_node *, void *);
108
109 /*
110 * Do what we need to do when inserting a route.
111 */
112 static struct radix_node *
in_addroute(void * v_arg,void * n_arg,struct radix_node_head * head,struct radix_node * treenodes)113 in_addroute(void *v_arg, void *n_arg, struct radix_node_head *head,
114 struct radix_node *treenodes)
115 {
116 struct rtentry *rt = (struct rtentry *)treenodes;
117 struct sockaddr_in *sin = (struct sockaddr_in *)(void *)rt_key(rt);
118 struct radix_node *ret;
119 char dbuf[MAX_IPv4_STR_LEN], gbuf[MAX_IPv4_STR_LEN];
120 uint32_t flags = rt->rt_flags;
121 boolean_t verbose = (rt_verbose > 1);
122
123 LCK_MTX_ASSERT(rnh_lock, LCK_MTX_ASSERT_OWNED);
124 RT_LOCK_ASSERT_HELD(rt);
125
126 if (verbose) {
127 rt_str(rt, dbuf, sizeof(dbuf), gbuf, sizeof(gbuf));
128 }
129
130 /*
131 * For IP, all unicast non-host routes are automatically cloning.
132 */
133 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
134 rt->rt_flags |= RTF_MULTICAST;
135 }
136
137 if (!(rt->rt_flags & (RTF_HOST | RTF_CLONING | RTF_MULTICAST))) {
138 rt->rt_flags |= RTF_PRCLONING;
139 }
140
141 /*
142 * A little bit of help for both IP output and input:
143 * For host routes, we make sure that RTF_BROADCAST
144 * is set for anything that looks like a broadcast address.
145 * This way, we can avoid an expensive call to in_broadcast()
146 * in ip_output() most of the time (because the route passed
147 * to ip_output() is almost always a host route).
148 *
149 * We also do the same for local addresses, with the thought
150 * that this might one day be used to speed up ip_input().
151 *
152 * We also mark routes to multicast addresses as such, because
153 * it's easy to do and might be useful (but this is much more
154 * dubious since it's so easy to inspect the address). (This
155 * is done above.)
156 */
157 if (rt->rt_flags & RTF_HOST) {
158 if (in_broadcast(sin->sin_addr, rt->rt_ifp)) {
159 rt->rt_flags |= RTF_BROADCAST;
160 } else {
161 /* Become a regular mutex */
162 RT_CONVERT_LOCK(rt);
163 IFA_LOCK_SPIN(rt->rt_ifa);
164 if (satosin(rt->rt_ifa->ifa_addr)->sin_addr.s_addr ==
165 sin->sin_addr.s_addr) {
166 rt->rt_flags |= RTF_LOCAL;
167 }
168 IFA_UNLOCK(rt->rt_ifa);
169 }
170 }
171
172 if (!rt->rt_rmx.rmx_mtu && !(rt->rt_rmx.rmx_locks & RTV_MTU) &&
173 rt->rt_ifp) {
174 rt->rt_rmx.rmx_mtu = rt->rt_ifp->if_mtu;
175 if (INTF_ADJUST_MTU_FOR_CLAT46(rt->rt_ifp)) {
176 rt->rt_rmx.rmx_mtu = IN6_LINKMTU(rt->rt_ifp);
177 /* Further adjust the size for CLAT46 expansion */
178 rt->rt_rmx.rmx_mtu -= CLAT46_HDR_EXPANSION_OVERHD;
179 }
180 }
181
182 ret = rn_addroute(v_arg, n_arg, head, treenodes);
183 if (ret == NULL && (rt->rt_flags & RTF_HOST)) {
184 struct rtentry *rt2;
185 /*
186 * We are trying to add a host route, but can't.
187 * Find out if it is because of an
188 * ARP entry and delete it if so.
189 */
190 rt2 = rtalloc1_scoped_locked(rt_key(rt), 0,
191 RTF_CLONING | RTF_PRCLONING, sin_get_ifscope(rt_key(rt)));
192 if (rt2 != NULL) {
193 char dbufc[MAX_IPv4_STR_LEN];
194
195 RT_LOCK(rt2);
196 if (verbose) {
197 rt_str(rt2, dbufc, sizeof(dbufc), NULL, 0);
198 }
199
200 if ((rt2->rt_flags & RTF_LLINFO) &&
201 (rt2->rt_flags & RTF_HOST) &&
202 rt2->rt_gateway != NULL &&
203 rt2->rt_gateway->sa_family == AF_LINK) {
204 if (verbose) {
205 os_log_debug(OS_LOG_DEFAULT, "%s: unable to insert "
206 "route to %s;%s, flags=%b, due to "
207 "existing ARP route %s->%s "
208 "flags=%b, attempting to delete\n",
209 __func__, dbuf,
210 (rt->rt_ifp != NULL) ?
211 rt->rt_ifp->if_xname : "",
212 rt->rt_flags, RTF_BITS, dbufc,
213 (rt2->rt_ifp != NULL) ?
214 rt2->rt_ifp->if_xname : "",
215 rt2->rt_flags, RTF_BITS);
216 }
217 /*
218 * Safe to drop rt_lock and use rt_key,
219 * rt_gateway, since holding rnh_lock here
220 * prevents another thread from calling
221 * rt_setgate() on this route.
222 */
223 RT_UNLOCK(rt2);
224 (void) rtrequest_locked(RTM_DELETE, rt_key(rt2),
225 rt2->rt_gateway, rt_mask(rt2),
226 rt2->rt_flags, NULL);
227 ret = rn_addroute(v_arg, n_arg, head,
228 treenodes);
229 } else {
230 RT_UNLOCK(rt2);
231 }
232 rtfree_locked(rt2);
233 }
234 }
235
236 if (!verbose) {
237 goto done;
238 }
239
240 if (ret != NULL) {
241 if (flags != rt->rt_flags) {
242 os_log_debug(OS_LOG_DEFAULT, "%s: route to %s->%s->%s inserted, "
243 "oflags=%b, flags=%b\n", __func__,
244 dbuf, gbuf, (rt->rt_ifp != NULL) ?
245 rt->rt_ifp->if_xname : "", flags, RTF_BITS,
246 rt->rt_flags, RTF_BITS);
247 } else {
248 os_log_debug(OS_LOG_DEFAULT, "%s: route to %s->%s->%s inserted, "
249 "flags=%b\n", __func__, dbuf, gbuf,
250 (rt->rt_ifp != NULL) ? rt->rt_ifp->if_xname : "",
251 rt->rt_flags, RTF_BITS);
252 }
253 } else {
254 os_log_debug(OS_LOG_DEFAULT, "%s: unable to insert route to %s->%s->%s, "
255 "flags=%b, already exists\n", __func__, dbuf, gbuf,
256 (rt->rt_ifp != NULL) ? rt->rt_ifp->if_xname : "",
257 rt->rt_flags, RTF_BITS);
258 }
259 done:
260 return ret;
261 }
262
263 static struct radix_node *
in_deleteroute(void * v_arg,void * netmask_arg,struct radix_node_head * head)264 in_deleteroute(void *v_arg, void *netmask_arg, struct radix_node_head *head)
265 {
266 struct radix_node *rn;
267
268 LCK_MTX_ASSERT(rnh_lock, LCK_MTX_ASSERT_OWNED);
269
270 rn = rn_delete(v_arg, netmask_arg, head);
271 if (rt_verbose > 1 && rn != NULL) {
272 char dbuf[MAX_IPv4_STR_LEN], gbuf[MAX_IPv4_STR_LEN];
273 struct rtentry *rt = (struct rtentry *)rn;
274
275 RT_LOCK(rt);
276 rt_str(rt, dbuf, sizeof(dbuf), gbuf, sizeof(gbuf));
277 os_log_debug(OS_LOG_DEFAULT, "%s: route to %s->%s->%s deleted, "
278 "flags=%b\n", __func__, dbuf, gbuf, (rt->rt_ifp != NULL) ?
279 rt->rt_ifp->if_xname : "", rt->rt_flags, RTF_BITS);
280 RT_UNLOCK(rt);
281 }
282 return rn;
283 }
284
285 /*
286 * Validate (unexpire) an expiring AF_INET route.
287 */
288 struct radix_node *
in_validate(struct radix_node * rn)289 in_validate(struct radix_node *rn)
290 {
291 struct rtentry *rt = (struct rtentry *)rn;
292
293 RT_LOCK_ASSERT_HELD(rt);
294
295 /* This is first reference? */
296 if (rt->rt_refcnt == 0) {
297 if (rt_verbose > 2) {
298 char dbuf[MAX_IPv4_STR_LEN], gbuf[MAX_IPv4_STR_LEN];
299
300 rt_str(rt, dbuf, sizeof(dbuf), gbuf, sizeof(gbuf));
301 os_log_debug(OS_LOG_DEFAULT, "%s: route to %s->%s->%s validated, "
302 "flags=%b\n", __func__, dbuf, gbuf,
303 (rt->rt_ifp != NULL) ? rt->rt_ifp->if_xname : "",
304 rt->rt_flags, RTF_BITS);
305 }
306
307 /*
308 * It's one of ours; unexpire it. If the timer is already
309 * scheduled, let it run later as it won't re-arm itself
310 * if there's nothing to do.
311 */
312 if (rt->rt_flags & RTPRF_OURS) {
313 rt->rt_flags &= ~RTPRF_OURS;
314 rt_setexpire(rt, 0);
315 }
316 }
317 return rn;
318 }
319
320 /*
321 * Similar to in_matroute_args except without the leaf-matching parameters.
322 */
323 static struct radix_node *
in_matroute(void * v_arg,struct radix_node_head * head)324 in_matroute(void *v_arg, struct radix_node_head *head)
325 {
326 return in_matroute_args(v_arg, head, NULL, NULL);
327 }
328
329 /*
330 * This code is the inverse of in_clsroute: on first reference, if we
331 * were managing the route, stop doing so and set the expiration timer
332 * back off again.
333 */
334 static struct radix_node *
in_matroute_args(void * v_arg,struct radix_node_head * head,rn_matchf_t * f,void * w)335 in_matroute_args(void *v_arg, struct radix_node_head *head,
336 rn_matchf_t *f, void *w)
337 {
338 struct radix_node *rn = rn_match_args(v_arg, head, f, w);
339
340 if (rn != NULL) {
341 RT_LOCK_SPIN((struct rtentry *)rn);
342 in_validate(rn);
343 RT_UNLOCK((struct rtentry *)rn);
344 }
345 return rn;
346 }
347
348 /* one hour is ``really old'' */
349 static uint32_t rtq_reallyold = 60 * 60;
350 SYSCTL_UINT(_net_inet_ip, IPCTL_RTEXPIRE, rtexpire,
351 CTLFLAG_RW | CTLFLAG_LOCKED, &rtq_reallyold, 0,
352 "Default expiration time on dynamically learned routes");
353
354 /* never automatically crank down to less */
355 static uint32_t rtq_minreallyold = 10;
356 SYSCTL_UINT(_net_inet_ip, IPCTL_RTMINEXPIRE, rtminexpire,
357 CTLFLAG_RW | CTLFLAG_LOCKED, &rtq_minreallyold, 0,
358 "Minimum time to attempt to hold onto dynamically learned routes");
359
360 /* 128 cached routes is ``too many'' */
361 static uint32_t rtq_toomany = 128;
362 SYSCTL_UINT(_net_inet_ip, IPCTL_RTMAXCACHE, rtmaxcache,
363 CTLFLAG_RW | CTLFLAG_LOCKED, &rtq_toomany, 0,
364 "Upper limit on dynamically learned routes");
365
366 /*
367 * On last reference drop, mark the route as belong to us so that it can be
368 * timed out.
369 */
370 static void
in_clsroute(struct radix_node * rn,struct radix_node_head * head)371 in_clsroute(struct radix_node *rn, struct radix_node_head *head)
372 {
373 #pragma unused(head)
374 char dbuf[MAX_IPv4_STR_LEN], gbuf[MAX_IPv4_STR_LEN];
375 struct rtentry *rt = (struct rtentry *)rn;
376 boolean_t verbose = (rt_verbose > 1);
377
378 LCK_MTX_ASSERT(rnh_lock, LCK_MTX_ASSERT_OWNED);
379 RT_LOCK_ASSERT_HELD(rt);
380
381 if (!(rt->rt_flags & RTF_UP)) {
382 return; /* prophylactic measures */
383 }
384 if ((rt->rt_flags & (RTF_LLINFO | RTF_HOST)) != RTF_HOST) {
385 return;
386 }
387
388 if (rt->rt_flags & RTPRF_OURS) {
389 return;
390 }
391
392 if (!(rt->rt_flags & (RTF_WASCLONED | RTF_DYNAMIC))) {
393 return;
394 }
395
396 if (verbose) {
397 rt_str(rt, dbuf, sizeof(dbuf), gbuf, sizeof(gbuf));
398 }
399
400 /*
401 * Delete the route immediately if RTF_DELCLONE is set or
402 * if route caching is disabled (rtq_reallyold set to 0).
403 * Otherwise, let it expire and be deleted by in_rtqkill().
404 */
405 if ((rt->rt_flags & RTF_DELCLONE) || rtq_reallyold == 0) {
406 int err;
407
408 if (verbose) {
409 os_log_debug(OS_LOG_DEFAULT, "%s: deleting route to %s->%s->%s, "
410 "flags=%b\n", __func__, dbuf, gbuf,
411 (rt->rt_ifp != NULL) ? rt->rt_ifp->if_xname : "",
412 rt->rt_flags, RTF_BITS);
413 }
414 /*
415 * Delete the route from the radix tree but since we are
416 * called when the route's reference count is 0, don't
417 * deallocate it until we return from this routine by
418 * telling rtrequest that we're interested in it.
419 * Safe to drop rt_lock and use rt_key, rt_gateway since
420 * holding rnh_lock here prevents another thread from
421 * calling rt_setgate() on this route.
422 */
423 RT_UNLOCK(rt);
424 err = rtrequest_locked(RTM_DELETE, rt_key(rt),
425 rt->rt_gateway, rt_mask(rt), rt->rt_flags, &rt);
426 if (err == 0) {
427 /* Now let the caller free it */
428 RT_LOCK(rt);
429 RT_REMREF_LOCKED(rt);
430 } else {
431 RT_LOCK(rt);
432 if (!verbose) {
433 rt_str(rt, dbuf, sizeof(dbuf),
434 gbuf, sizeof(gbuf));
435 }
436 os_log_error(OS_LOG_DEFAULT, "%s: error deleting route to "
437 "%s->%s->%s, flags=%b, err=%d\n", __func__,
438 dbuf, gbuf, (rt->rt_ifp != NULL) ?
439 rt->rt_ifp->if_xname : "", rt->rt_flags,
440 RTF_BITS, err);
441 }
442 } else {
443 uint64_t timenow;
444
445 timenow = net_uptime();
446 rt->rt_flags |= RTPRF_OURS;
447 rt_setexpire(rt, timenow + rtq_reallyold);
448
449 if (verbose) {
450 os_log_debug(OS_LOG_DEFAULT, "%s: route to %s->%s->%s invalidated, "
451 "flags=%b, expire=T+%u\n", __func__, dbuf, gbuf,
452 (rt->rt_ifp != NULL) ? rt->rt_ifp->if_xname : "",
453 rt->rt_flags, RTF_BITS, rt->rt_expire - timenow);
454 }
455
456 /* We have at least one entry; arm the timer if not already */
457 in_sched_rtqtimo(NULL);
458 }
459 }
460
461 struct rtqk_arg {
462 struct radix_node_head *rnh;
463 int updating;
464 int draining;
465 uint32_t killed;
466 uint32_t found;
467 uint64_t nextstop;
468 };
469
470 /*
471 * Get rid of old routes. When draining, this deletes everything, even when
472 * the timeout is not expired yet. When updating, this makes sure that
473 * nothing has a timeout longer than the current value of rtq_reallyold.
474 */
475 static int
in_rtqkill(struct radix_node * rn,void * rock)476 in_rtqkill(struct radix_node *rn, void *rock)
477 {
478 struct rtqk_arg *ap = rock;
479 struct rtentry *rt = (struct rtentry *)rn;
480 boolean_t verbose = (rt_verbose > 1);
481 uint64_t timenow;
482 int err;
483
484 timenow = net_uptime();
485 LCK_MTX_ASSERT(rnh_lock, LCK_MTX_ASSERT_OWNED);
486
487 RT_LOCK(rt);
488 if (rt->rt_flags & RTPRF_OURS) {
489 char dbuf[MAX_IPv4_STR_LEN], gbuf[MAX_IPv4_STR_LEN];
490
491 if (verbose) {
492 rt_str(rt, dbuf, sizeof(dbuf), gbuf, sizeof(gbuf));
493 }
494
495 ap->found++;
496 VERIFY(rt->rt_expire == 0 || rt->rt_rmx.rmx_expire != 0);
497 VERIFY(rt->rt_expire != 0 || rt->rt_rmx.rmx_expire == 0);
498 if (ap->draining || rt->rt_expire <= timenow) {
499 if (rt->rt_refcnt > 0) {
500 panic("%s: route %p marked with RTPRF_OURS "
501 "with non-zero refcnt (%u)", __func__,
502 rt, rt->rt_refcnt);
503 /* NOTREACHED */
504 }
505
506 if (verbose) {
507 os_log_debug(OS_LOG_DEFAULT, "%s: deleting route to "
508 "%s->%s->%s, flags=%b, draining=%d\n",
509 __func__, dbuf, gbuf, (rt->rt_ifp != NULL) ?
510 rt->rt_ifp->if_xname : "", rt->rt_flags,
511 RTF_BITS, ap->draining);
512 }
513 RT_ADDREF_LOCKED(rt); /* for us to free below */
514 /*
515 * Delete this route since we're done with it;
516 * the route may be freed afterwards, so we
517 * can no longer refer to 'rt' upon returning
518 * from rtrequest(). Safe to drop rt_lock and
519 * use rt_key, rt_gateway since holding rnh_lock
520 * here prevents another thread from calling
521 * rt_setgate() on this route.
522 */
523 RT_UNLOCK(rt);
524 err = rtrequest_locked(RTM_DELETE, rt_key(rt),
525 rt->rt_gateway, rt_mask(rt), rt->rt_flags, NULL);
526 if (err != 0) {
527 RT_LOCK(rt);
528 if (!verbose) {
529 rt_str(rt, dbuf, sizeof(dbuf),
530 gbuf, sizeof(gbuf));
531 }
532 os_log_error(OS_LOG_DEFAULT, "%s: error deleting route to "
533 "%s->%s->%s, flags=%b, err=%d\n", __func__,
534 dbuf, gbuf, (rt->rt_ifp != NULL) ?
535 rt->rt_ifp->if_xname : "", rt->rt_flags,
536 RTF_BITS, err);
537 RT_UNLOCK(rt);
538 } else {
539 ap->killed++;
540 }
541 rtfree_locked(rt);
542 } else {
543 uint64_t expire = (rt->rt_expire - timenow);
544
545 if (ap->updating && expire > rtq_reallyold) {
546 rt_setexpire(rt, timenow + rtq_reallyold);
547 if (verbose) {
548 os_log_debug(OS_LOG_DEFAULT, "%s: route to "
549 "%s->%s->%s, flags=%b, adjusted "
550 "expire=T+%u (was T+%u)\n",
551 __func__, dbuf, gbuf,
552 (rt->rt_ifp != NULL) ?
553 rt->rt_ifp->if_xname : "",
554 rt->rt_flags, RTF_BITS,
555 (rt->rt_expire - timenow), expire);
556 }
557 }
558 ap->nextstop = lmin(ap->nextstop, rt->rt_expire);
559 RT_UNLOCK(rt);
560 }
561 } else {
562 RT_UNLOCK(rt);
563 }
564
565 return 0;
566 }
567
568 #define RTQ_TIMEOUT 60*10 /* run no less than once every ten minutes */
569 static int rtq_timeout = RTQ_TIMEOUT;
570
571 static void
in_rtqtimo(void * targ)572 in_rtqtimo(void *targ)
573 {
574 #pragma unused(targ)
575 struct radix_node_head *rnh;
576 struct rtqk_arg arg;
577 struct timeval atv;
578 static uint64_t last_adjusted_timeout = 0;
579 boolean_t verbose = (rt_verbose > 1);
580 uint64_t timenow;
581 uint32_t ours;
582
583 lck_mtx_lock(rnh_lock);
584 rnh = rt_tables[AF_INET];
585 VERIFY(rnh != NULL);
586
587 /* Get the timestamp after we acquire the lock for better accuracy */
588 timenow = net_uptime();
589 if (verbose) {
590 os_log_debug(OS_LOG_DEFAULT, "%s: initial nextstop is T+%u seconds\n",
591 __func__, rtq_timeout);
592 }
593 bzero(&arg, sizeof(arg));
594 arg.rnh = rnh;
595 arg.nextstop = timenow + rtq_timeout;
596 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
597 if (verbose) {
598 os_log_debug(OS_LOG_DEFAULT, "%s: found %u, killed %u\n", __func__,
599 arg.found, arg.killed);
600 }
601 /*
602 * Attempt to be somewhat dynamic about this:
603 * If there are ``too many'' routes sitting around taking up space,
604 * then crank down the timeout, and see if we can't make some more
605 * go away. However, we make sure that we will never adjust more
606 * than once in rtq_timeout seconds, to keep from cranking down too
607 * hard.
608 */
609 ours = (arg.found - arg.killed);
610 if (ours > rtq_toomany &&
611 ((timenow - last_adjusted_timeout) >= (uint64_t)rtq_timeout) &&
612 rtq_reallyold > rtq_minreallyold) {
613 rtq_reallyold = 2 * rtq_reallyold / 3;
614 if (rtq_reallyold < rtq_minreallyold) {
615 rtq_reallyold = rtq_minreallyold;
616 }
617
618 last_adjusted_timeout = timenow;
619 if (verbose) {
620 os_log_debug(OS_LOG_DEFAULT, "%s: adjusted rtq_reallyold to %d "
621 "seconds\n", __func__, rtq_reallyold);
622 }
623 arg.found = arg.killed = 0;
624 arg.updating = 1;
625 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
626 }
627
628 atv.tv_usec = 0;
629 atv.tv_sec = arg.nextstop - timenow;
630 /* re-arm the timer only if there's work to do */
631 in_rtqtimo_run = 0;
632 if (ours > 0) {
633 in_sched_rtqtimo(&atv);
634 } else if (verbose) {
635 os_log_debug(OS_LOG_DEFAULT, "%s: not rescheduling timer\n", __func__);
636 }
637 lck_mtx_unlock(rnh_lock);
638 }
639
640 static void
in_sched_rtqtimo(struct timeval * atv)641 in_sched_rtqtimo(struct timeval *atv)
642 {
643 LCK_MTX_ASSERT(rnh_lock, LCK_MTX_ASSERT_OWNED);
644
645 if (!in_rtqtimo_run) {
646 struct timeval tv;
647
648 if (atv == NULL) {
649 tv.tv_usec = 0;
650 tv.tv_sec = MAX(rtq_timeout / 10, 1);
651 atv = &tv;
652 }
653 if (rt_verbose > 1) {
654 os_log_debug(OS_LOG_DEFAULT, "%s: timer scheduled in "
655 "T+%llus.%lluu\n", __func__,
656 (uint64_t)atv->tv_sec, (uint64_t)atv->tv_usec);
657 }
658 in_rtqtimo_run = 1;
659 timeout(in_rtqtimo, NULL, tvtohz(atv));
660 }
661 }
662
663 void
in_rtqdrain(void)664 in_rtqdrain(void)
665 {
666 struct radix_node_head *rnh;
667 struct rtqk_arg arg;
668
669 if (rt_verbose > 1) {
670 os_log_debug(OS_LOG_DEFAULT, "%s: draining routes\n", __func__);
671 }
672
673 lck_mtx_lock(rnh_lock);
674 rnh = rt_tables[AF_INET];
675 VERIFY(rnh != NULL);
676 bzero(&arg, sizeof(arg));
677 arg.rnh = rnh;
678 arg.draining = 1;
679 rnh->rnh_walktree(rnh, in_rtqkill, &arg);
680 lck_mtx_unlock(rnh_lock);
681 }
682
683 /*
684 * Initialize our routing tree.
685 */
686 int
in_inithead(void ** head,int off)687 in_inithead(void **head, int off)
688 {
689 struct radix_node_head *rnh;
690
691 /* If called from route_init(), make sure it is exactly once */
692 VERIFY(head != (void **)&rt_tables[AF_INET] || *head == NULL);
693
694 if (!rn_inithead(head, off)) {
695 return 0;
696 }
697
698 /*
699 * We can get here from nfs_subs.c as well, in which case this
700 * won't be for the real routing table and thus we're done;
701 * this also takes care of the case when we're called more than
702 * once from anywhere but route_init().
703 */
704 if (head != (void **)&rt_tables[AF_INET]) {
705 return 1; /* only do this for the real routing table */
706 }
707 rnh = *head;
708 rnh->rnh_addaddr = in_addroute;
709 rnh->rnh_deladdr = in_deleteroute;
710 rnh->rnh_matchaddr = in_matroute;
711 rnh->rnh_matchaddr_args = in_matroute_args;
712 rnh->rnh_close = in_clsroute;
713 return 1;
714 }
715
716 /*
717 * This zaps old routes when the interface goes down or interface
718 * address is deleted. In the latter case, it deletes static routes
719 * that point to this address. If we don't do this, we may end up
720 * using the old address in the future. The ones we always want to
721 * get rid of are things like ARP entries, since the user might down
722 * the interface, walk over to a completely different network, and
723 * plug back in.
724 */
725 struct in_ifadown_arg {
726 struct radix_node_head *rnh;
727 struct ifaddr *ifa;
728 int del;
729 };
730
731 static int
in_ifadownkill(struct radix_node * rn,void * xap)732 in_ifadownkill(struct radix_node *rn, void *xap)
733 {
734 char dbuf[MAX_IPv4_STR_LEN], gbuf[MAX_IPv4_STR_LEN];
735 struct in_ifadown_arg *ap = xap;
736 struct rtentry *rt = (struct rtentry *)rn;
737 boolean_t verbose = (rt_verbose != 0);
738 int err;
739
740 LCK_MTX_ASSERT(rnh_lock, LCK_MTX_ASSERT_OWNED);
741
742 RT_LOCK(rt);
743 if (rt->rt_ifa == ap->ifa &&
744 (ap->del || !(rt->rt_flags & RTF_STATIC))) {
745 rt_str(rt, dbuf, sizeof(dbuf), gbuf, sizeof(gbuf));
746 if (verbose) {
747 os_log_debug(OS_LOG_DEFAULT, "%s: deleting route to %s->%s->%s, "
748 "flags=%b\n", __func__, dbuf, gbuf,
749 (rt->rt_ifp != NULL) ? rt->rt_ifp->if_xname : "",
750 rt->rt_flags, RTF_BITS);
751 }
752 RT_ADDREF_LOCKED(rt); /* for us to free below */
753 /*
754 * We need to disable the automatic prune that happens
755 * in this case in rtrequest() because it will blow
756 * away the pointers that rn_walktree() needs in order
757 * continue our descent. We will end up deleting all
758 * the routes that rtrequest() would have in any case,
759 * so that behavior is not needed there. Safe to drop
760 * rt_lock and use rt_key, rt_gateway, since holding
761 * rnh_lock here prevents another thread from calling
762 * rt_setgate() on this route.
763 */
764 rt->rt_flags &= ~(RTF_CLONING | RTF_PRCLONING);
765 RT_UNLOCK(rt);
766 err = rtrequest_locked(RTM_DELETE, rt_key(rt),
767 rt->rt_gateway, rt_mask(rt), rt->rt_flags, NULL);
768 if (err != 0) {
769 RT_LOCK(rt);
770 if (!verbose) {
771 rt_str(rt, dbuf, sizeof(dbuf),
772 gbuf, sizeof(gbuf));
773 }
774 os_log_error(OS_LOG_DEFAULT, "%s: error deleting route to "
775 "%s->%s->%s, flags=%b, err=%d\n", __func__,
776 dbuf, gbuf, (rt->rt_ifp != NULL) ?
777 rt->rt_ifp->if_xname : "", rt->rt_flags,
778 RTF_BITS, err);
779 RT_UNLOCK(rt);
780 }
781 rtfree_locked(rt);
782 } else {
783 RT_UNLOCK(rt);
784 }
785 return 0;
786 }
787
788 int
in_ifadown(struct ifaddr * ifa,int delete)789 in_ifadown(struct ifaddr *ifa, int delete)
790 {
791 struct in_ifadown_arg arg;
792 struct radix_node_head *rnh;
793
794 LCK_MTX_ASSERT(rnh_lock, LCK_MTX_ASSERT_OWNED);
795
796 /*
797 * Holding rnh_lock here prevents the possibility of
798 * ifa from changing (e.g. in_ifinit), so it is safe
799 * to access its ifa_addr without locking.
800 */
801 if (ifa->ifa_addr->sa_family != AF_INET) {
802 return 1;
803 }
804
805 /* trigger route cache reevaluation */
806 routegenid_inet_update();
807
808 arg.rnh = rnh = rt_tables[AF_INET];
809 arg.ifa = ifa;
810 arg.del = delete;
811 rnh->rnh_walktree(rnh, in_ifadownkill, &arg);
812 IFA_LOCK_SPIN(ifa);
813 ifa->ifa_flags &= ~IFA_ROUTE;
814 IFA_UNLOCK(ifa);
815 return 0;
816 }
817