xref: /xnu-12377.1.9/bsd/netinet/in_rmx.c (revision f6217f891ac0bb64f3d375211650a4c1ff8ca1ea)
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 #include <kern/uipc_domain.h>
82 
83 #include <net/if.h>
84 #include <net/route.h>
85 #include <netinet/in.h>
86 #include <netinet/in_var.h>
87 #include <netinet/in_arp.h>
88 #include <netinet/ip.h>
89 #include <netinet/ip6.h>
90 #include <netinet6/nd6.h>
91 
92 #include <net/sockaddr_utils.h>
93 
94 extern int tvtohz(struct timeval *);
95 
96 static int in_rtqtimo_run;              /* in_rtqtimo is scheduled to run */
97 static void in_rtqtimo(void *);
98 static void in_sched_rtqtimo(struct timeval *);
99 
100 static struct radix_node *in_addroute(void *, void *, struct radix_node_head *,
101     struct radix_node *);
102 static struct radix_node *in_deleteroute(void *, void *,
103     struct radix_node_head *);
104 static struct radix_node *in_matroute(void *, struct radix_node_head *);
105 static struct radix_node *in_matroute_args(void *, struct radix_node_head *,
106     rn_matchf_t *f, void *);
107 static void in_clsroute(struct radix_node *, struct radix_node_head *);
108 static int in_rtqkill(struct radix_node *, void *);
109 
110 static int in_ifadownkill(struct radix_node *, void *);
111 
112 /*
113  * Do what we need to do when inserting a route.
114  */
115 static struct radix_node *
in_addroute(void * v_arg,void * n_arg,struct radix_node_head * head,struct radix_node * rn)116 in_addroute(void *v_arg, void *n_arg, struct radix_node_head *head,
117     struct radix_node *rn)
118 {
119 	rtentry_ref_t rt = RT(rn);
120 	struct sockaddr_in *sin = SIN(rt_key(rt));
121 	struct radix_node *ret;
122 	char dbuf[MAX_IPv4_STR_LEN], gbuf[MAX_IPv4_STR_LEN];
123 	uint32_t flags = rt->rt_flags;
124 	boolean_t verbose = (rt_verbose > 0);
125 
126 	LCK_MTX_ASSERT(rnh_lock, LCK_MTX_ASSERT_OWNED);
127 	RT_LOCK_ASSERT_HELD(rt);
128 
129 	if (verbose) {
130 		rt_str(rt, dbuf, sizeof(dbuf), gbuf, sizeof(gbuf));
131 	}
132 
133 	/*
134 	 * For IP, all unicast non-host routes are automatically cloning.
135 	 */
136 	if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
137 		rt->rt_flags |= RTF_MULTICAST;
138 	}
139 
140 	if (!(rt->rt_flags & (RTF_HOST | RTF_CLONING | RTF_MULTICAST))) {
141 		rt->rt_flags |= RTF_PRCLONING;
142 	}
143 
144 	/*
145 	 * A little bit of help for both IP output and input:
146 	 *   For host routes, we make sure that RTF_BROADCAST
147 	 *   is set for anything that looks like a broadcast address.
148 	 *   This way, we can avoid an expensive call to in_broadcast()
149 	 *   in ip_output() most of the time (because the route passed
150 	 *   to ip_output() is almost always a host route).
151 	 *
152 	 *   We also do the same for local addresses, with the thought
153 	 *   that this might one day be used to speed up ip_input().
154 	 *
155 	 * We also mark routes to multicast addresses as such, because
156 	 * it's easy to do and might be useful (but this is much more
157 	 * dubious since it's so easy to inspect the address).  (This
158 	 * is done above.)
159 	 */
160 	if (rt->rt_flags & RTF_HOST) {
161 		if (in_broadcast(sin->sin_addr, rt->rt_ifp)) {
162 			rt->rt_flags |= RTF_BROADCAST;
163 		} else {
164 			/* Become a regular mutex */
165 			RT_CONVERT_LOCK(rt);
166 			IFA_LOCK_SPIN(rt->rt_ifa);
167 			if (satosin(rt->rt_ifa->ifa_addr)->sin_addr.s_addr ==
168 			    sin->sin_addr.s_addr) {
169 				rt->rt_flags |= RTF_LOCAL;
170 			}
171 			IFA_UNLOCK(rt->rt_ifa);
172 		}
173 	}
174 
175 	if (!rt->rt_rmx.rmx_mtu && !(rt->rt_rmx.rmx_locks & RTV_MTU) &&
176 	    rt->rt_ifp) {
177 		rt->rt_rmx.rmx_mtu = rt->rt_ifp->if_mtu;
178 		if (INTF_ADJUST_MTU_FOR_CLAT46(rt->rt_ifp)) {
179 			rt->rt_rmx.rmx_mtu = IN6_LINKMTU(rt->rt_ifp);
180 			/* Further adjust the size for CLAT46 expansion */
181 			rt->rt_rmx.rmx_mtu -= CLAT46_HDR_EXPANSION_OVERHD;
182 		}
183 	}
184 
185 	ret = rn_addroute(v_arg, n_arg, head, rt->rt_nodes);
186 	if (ret == NULL && (rt->rt_flags & RTF_HOST)) {
187 		rtentry_ref_t rt2;
188 		/*
189 		 * We are trying to add a host route, but can't.
190 		 * Find out if it is because of an
191 		 * ARP entry and delete it if so.
192 		 */
193 		rt2 = rtalloc1_scoped_locked(rt_key(rt), 0,
194 		    RTF_CLONING | RTF_PRCLONING, sin_get_ifscope(rt_key(rt)));
195 		if (rt2 != NULL) {
196 			char dbufc[MAX_IPv4_STR_LEN];
197 
198 			RT_LOCK(rt2);
199 			if (verbose) {
200 				rt_str(rt2, dbufc, sizeof(dbufc), NULL, 0);
201 			}
202 
203 			if ((rt2->rt_flags & RTF_LLINFO) &&
204 			    (rt2->rt_flags & RTF_HOST) &&
205 			    rt2->rt_gateway != NULL &&
206 			    rt2->rt_gateway->sa_family == AF_LINK) {
207 				if (verbose) {
208 					os_log_debug(OS_LOG_DEFAULT, "%s: unable to insert "
209 					    "route to %s;%s, flags=0x%x, due to "
210 					    "existing ARP route %s->%s "
211 					    "flags=0x%x, attempting to delete\n",
212 					    __func__, dbuf,
213 					    (rt->rt_ifp != NULL) ?
214 					    rt->rt_ifp->if_xname : "",
215 					    rt->rt_flags, dbufc,
216 					    (rt2->rt_ifp != NULL) ?
217 					    rt2->rt_ifp->if_xname : "",
218 					    rt2->rt_flags);
219 				}
220 				/*
221 				 * Safe to drop rt_lock and use rt_key,
222 				 * rt_gateway, since holding rnh_lock here
223 				 * prevents another thread from calling
224 				 * rt_setgate() on this route.
225 				 */
226 				RT_UNLOCK(rt2);
227 				(void) rtrequest_locked(RTM_DELETE, rt_key(rt2),
228 				    rt2->rt_gateway, rt_mask(rt2),
229 				    rt2->rt_flags, NULL);
230 				ret = rn_addroute(v_arg, n_arg, head, rt->rt_nodes);
231 			} else {
232 				RT_UNLOCK(rt2);
233 			}
234 			rtfree_locked(rt2);
235 		}
236 	}
237 
238 	if (!verbose) {
239 		goto done;
240 	}
241 
242 	if (ret != NULL) {
243 		if (flags != rt->rt_flags) {
244 			os_log_debug(OS_LOG_DEFAULT, "%s: route to %s->%s->%s inserted, "
245 			    "oflags=0x%x, flags=0x%x\n", __func__,
246 			    dbuf, gbuf, (rt->rt_ifp != NULL) ?
247 			    rt->rt_ifp->if_xname : "", flags,
248 			    rt->rt_flags);
249 		} else {
250 			os_log_debug(OS_LOG_DEFAULT, "%s: route to %s->%s->%s inserted, "
251 			    "flags=0x%x\n", __func__, dbuf, gbuf,
252 			    (rt->rt_ifp != NULL) ? rt->rt_ifp->if_xname : "",
253 			    rt->rt_flags);
254 		}
255 	} else {
256 		os_log_debug(OS_LOG_DEFAULT, "%s: unable to insert route to %s->%s->%s, "
257 		    "flags=0x%x, already exists\n", __func__, dbuf, gbuf,
258 		    (rt->rt_ifp != NULL) ? rt->rt_ifp->if_xname : "",
259 		    rt->rt_flags);
260 	}
261 done:
262 	return ret;
263 }
264 
265 static struct radix_node *
in_deleteroute(void * v_arg,void * netmask_arg,struct radix_node_head * head)266 in_deleteroute(void *v_arg, void *netmask_arg, struct radix_node_head *head)
267 {
268 	struct radix_node *rn;
269 
270 	LCK_MTX_ASSERT(rnh_lock, LCK_MTX_ASSERT_OWNED);
271 
272 	rn = rn_delete(v_arg, netmask_arg, head);
273 	if (rt_verbose > 0 && rn != NULL) {
274 		char dbuf[MAX_IPv4_STR_LEN], gbuf[MAX_IPv4_STR_LEN];
275 		rtentry_ref_t rt = RT(rn);
276 
277 		RT_LOCK(rt);
278 		rt_str(rt, dbuf, sizeof(dbuf), gbuf, sizeof(gbuf));
279 		os_log_debug(OS_LOG_DEFAULT, "%s: route to %s->%s->%s deleted, "
280 		    "flags=0x%x\n", __func__, dbuf, gbuf, (rt->rt_ifp != NULL) ?
281 		    rt->rt_ifp->if_xname : "", rt->rt_flags);
282 		RT_UNLOCK(rt);
283 	}
284 	return rn;
285 }
286 
287 /*
288  * Validate (unexpire) an expiring AF_INET route.
289  */
290 struct radix_node *
in_validate(struct radix_node * rn)291 in_validate(struct radix_node *rn)
292 {
293 	rtentry_ref_t rt = RT(rn);
294 
295 	RT_LOCK_ASSERT_HELD(rt);
296 
297 	/* This is first reference? */
298 	if (rt->rt_refcnt == 0) {
299 		if (rt_verbose > 2) {
300 			char dbuf[MAX_IPv4_STR_LEN], gbuf[MAX_IPv4_STR_LEN];
301 
302 			rt_str(rt, dbuf, sizeof(dbuf), gbuf, sizeof(gbuf));
303 			os_log_debug(OS_LOG_DEFAULT, "%s: route to %s->%s->%s validated, "
304 			    "flags=0x%x\n", __func__, dbuf, gbuf,
305 			    (rt->rt_ifp != NULL) ? rt->rt_ifp->if_xname : "",
306 			    rt->rt_flags);
307 		}
308 
309 		/*
310 		 * It's one of ours; unexpire it.  If the timer is already
311 		 * scheduled, let it run later as it won't re-arm itself
312 		 * if there's nothing to do.
313 		 */
314 		if (rt->rt_flags & RTPRF_OURS) {
315 			rt->rt_flags &= ~RTPRF_OURS;
316 			rt_setexpire(rt, 0);
317 		}
318 	}
319 	return rn;
320 }
321 
322 /*
323  * Similar to in_matroute_args except without the leaf-matching parameters.
324  */
325 static struct radix_node *
in_matroute(void * v_arg,struct radix_node_head * head)326 in_matroute(void *v_arg, struct radix_node_head *head)
327 {
328 	return in_matroute_args(v_arg, head, NULL, NULL);
329 }
330 
331 /*
332  * This code is the inverse of in_clsroute: on first reference, if we
333  * were managing the route, stop doing so and set the expiration timer
334  * back off again.
335  */
336 static struct radix_node *
in_matroute_args(void * v_arg,struct radix_node_head * head,rn_matchf_t * f,void * w)337 in_matroute_args(void *v_arg, struct radix_node_head *head,
338     rn_matchf_t *f, void *w)
339 {
340 	struct radix_node *rn = rn_match_args(v_arg, head, f, w);
341 
342 	if (rn != NULL) {
343 		rtentry_ref_t rt = RT(rn);
344 		RT_LOCK_SPIN(rt);
345 		in_validate(rn);
346 		RT_UNLOCK(rt);
347 	}
348 	return rn;
349 }
350 
351 /* one hour is ``really old'' */
352 static uint32_t rtq_reallyold = 60 * 60;
353 SYSCTL_UINT(_net_inet_ip, IPCTL_RTEXPIRE, rtexpire,
354     CTLFLAG_RW | CTLFLAG_LOCKED, &rtq_reallyold, 0,
355     "Default expiration time on dynamically learned routes");
356 
357 /* never automatically crank down to less */
358 static uint32_t rtq_minreallyold = 10;
359 SYSCTL_UINT(_net_inet_ip, IPCTL_RTMINEXPIRE, rtminexpire,
360     CTLFLAG_RW | CTLFLAG_LOCKED, &rtq_minreallyold, 0,
361     "Minimum time to attempt to hold onto dynamically learned routes");
362 
363 /* 128 cached routes is ``too many'' */
364 static uint32_t rtq_toomany = 128;
365 SYSCTL_UINT(_net_inet_ip, IPCTL_RTMAXCACHE, rtmaxcache,
366     CTLFLAG_RW | CTLFLAG_LOCKED, &rtq_toomany, 0,
367     "Upper limit on dynamically learned routes");
368 
369 /*
370  * On last reference drop, mark the route as belong to us so that it can be
371  * timed out.
372  */
373 static void
in_clsroute(struct radix_node * rn,struct radix_node_head * head)374 in_clsroute(struct radix_node *rn, struct radix_node_head *head)
375 {
376 #pragma unused(head)
377 	char dbuf[MAX_IPv4_STR_LEN], gbuf[MAX_IPv4_STR_LEN];
378 	rtentry_ref_t rt = RT(rn);
379 	boolean_t verbose = (rt_verbose > 1);
380 
381 	LCK_MTX_ASSERT(rnh_lock, LCK_MTX_ASSERT_OWNED);
382 	RT_LOCK_ASSERT_HELD(rt);
383 
384 	if (!(rt->rt_flags & RTF_UP)) {
385 		return;         /* prophylactic measures */
386 	}
387 	if ((rt->rt_flags & (RTF_LLINFO | RTF_HOST)) != RTF_HOST) {
388 		return;
389 	}
390 
391 	if (rt->rt_flags & RTPRF_OURS) {
392 		return;
393 	}
394 
395 	if (!(rt->rt_flags & (RTF_WASCLONED | RTF_DYNAMIC))) {
396 		return;
397 	}
398 
399 	if (verbose) {
400 		rt_str(rt, dbuf, sizeof(dbuf), gbuf, sizeof(gbuf));
401 	}
402 
403 	/*
404 	 * Delete the route immediately if RTF_DELCLONE is set or
405 	 * if route caching is disabled (rtq_reallyold set to 0).
406 	 * Otherwise, let it expire and be deleted by in_rtqkill().
407 	 */
408 	if ((rt->rt_flags & RTF_DELCLONE) || rtq_reallyold == 0) {
409 		int err;
410 
411 		if (verbose) {
412 			os_log_debug(OS_LOG_DEFAULT, "%s: deleting route to %s->%s->%s, "
413 			    "flags=0x%x\n", __func__, dbuf, gbuf,
414 			    (rt->rt_ifp != NULL) ? rt->rt_ifp->if_xname : "",
415 			    rt->rt_flags);
416 		}
417 		/*
418 		 * Delete the route from the radix tree but since we are
419 		 * called when the route's reference count is 0, don't
420 		 * deallocate it until we return from this routine by
421 		 * telling rtrequest that we're interested in it.
422 		 * Safe to drop rt_lock and use rt_key, rt_gateway since
423 		 * holding rnh_lock here prevents another thread from
424 		 * calling rt_setgate() on this route.
425 		 */
426 		RT_UNLOCK(rt);
427 		err = rtrequest_locked(RTM_DELETE, rt_key(rt),
428 		    rt->rt_gateway, rt_mask(rt), rt->rt_flags, &rt);
429 		if (err == 0) {
430 			/* Now let the caller free it */
431 			RT_LOCK(rt);
432 			RT_REMREF_LOCKED(rt);
433 		} else {
434 			RT_LOCK(rt);
435 			rt_str(rt, dbuf, sizeof(dbuf), gbuf, sizeof(gbuf));
436 			os_log_error(OS_LOG_DEFAULT, "%s: error deleting route to "
437 			    "%s->%s->%s, flags=0x%x, err=%d\n", __func__,
438 			    dbuf, gbuf, (rt->rt_ifp != NULL) ?
439 			    rt->rt_ifp->if_xname : "", rt->rt_flags,
440 			    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 (rt_verbose > 2) {
450 			os_log_debug(OS_LOG_DEFAULT, "%s: route to %s->%s->%s invalidated, "
451 			    "flags=0x%x, expire=T+%u\n", __func__, dbuf, gbuf,
452 			    (rt->rt_ifp != NULL) ? rt->rt_ifp->if_xname : "",
453 			    rt->rt_flags, 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 __CCT_DECLARE_CONSTRAINED_PTR_TYPE(struct rtqk_arg, rtqk_arg, __CCT_REF);
470 
471 /*
472  * Get rid of old routes.  When draining, this deletes everything, even when
473  * the timeout is not expired yet.  When updating, this makes sure that
474  * nothing has a timeout longer than the current value of rtq_reallyold.
475  */
476 static int
in_rtqkill(struct radix_node * rn,void * rock)477 in_rtqkill(struct radix_node *rn, void *rock)
478 {
479 	rtqk_arg_ref_t ap = rock;
480 	rtentry_ref_t rt = RT(rn);
481 	boolean_t verbose = (rt_verbose > 1);
482 	uint64_t timenow;
483 	int err;
484 
485 	timenow = net_uptime();
486 	LCK_MTX_ASSERT(rnh_lock, LCK_MTX_ASSERT_OWNED);
487 
488 	RT_LOCK(rt);
489 	if (rt->rt_flags & RTPRF_OURS) {
490 		char dbuf[MAX_IPv4_STR_LEN], gbuf[MAX_IPv4_STR_LEN];
491 
492 		if (verbose) {
493 			rt_str(rt, dbuf, sizeof(dbuf), gbuf, sizeof(gbuf));
494 		}
495 
496 		ap->found++;
497 		VERIFY(rt->rt_expire == 0 || rt->rt_rmx.rmx_expire != 0);
498 		VERIFY(rt->rt_expire != 0 || rt->rt_rmx.rmx_expire == 0);
499 		if (ap->draining || rt->rt_expire <= timenow) {
500 			if (rt->rt_refcnt > 0) {
501 				panic("%s: route %p marked with RTPRF_OURS "
502 				    "with non-zero refcnt (%u)", __func__,
503 				    rt, rt->rt_refcnt);
504 				/* NOTREACHED */
505 			}
506 
507 			if (verbose) {
508 				os_log_debug(OS_LOG_DEFAULT, "%s: deleting route to "
509 				    "%s->%s->%s, flags=0x%x, draining=%d\n",
510 				    __func__, dbuf, gbuf, (rt->rt_ifp != NULL) ?
511 				    rt->rt_ifp->if_xname : "", rt->rt_flags,
512 				    ap->draining);
513 			}
514 			RT_ADDREF_LOCKED(rt);   /* for us to free below */
515 			/*
516 			 * Delete this route since we're done with it;
517 			 * the route may be freed afterwards, so we
518 			 * can no longer refer to 'rt' upon returning
519 			 * from rtrequest().  Safe to drop rt_lock and
520 			 * use rt_key, rt_gateway since holding rnh_lock
521 			 * here prevents another thread from calling
522 			 * rt_setgate() on this route.
523 			 */
524 			RT_UNLOCK(rt);
525 			err = rtrequest_locked(RTM_DELETE, rt_key(rt),
526 			    rt->rt_gateway, rt_mask(rt), rt->rt_flags, NULL);
527 			if (err != 0) {
528 				RT_LOCK(rt);
529 				if (!verbose) {
530 					rt_str(rt, dbuf, sizeof(dbuf),
531 					    gbuf, sizeof(gbuf));
532 				}
533 				os_log_error(OS_LOG_DEFAULT, "%s: error deleting route to "
534 				    "%s->%s->%s, flags=0x%x, err=%d\n", __func__,
535 				    dbuf, gbuf, (rt->rt_ifp != NULL) ?
536 				    rt->rt_ifp->if_xname : "", rt->rt_flags,
537 				    err);
538 				RT_UNLOCK(rt);
539 			} else {
540 				ap->killed++;
541 			}
542 			rtfree_locked(rt);
543 		} else {
544 			uint64_t expire = (rt->rt_expire - timenow);
545 
546 			if (ap->updating && expire > rtq_reallyold) {
547 				rt_setexpire(rt, timenow + rtq_reallyold);
548 				if (verbose) {
549 					os_log_debug(OS_LOG_DEFAULT, "%s: route to "
550 					    "%s->%s->%s, flags=0x%x, adjusted "
551 					    "expire=T+%u (was T+%u)\n",
552 					    __func__, dbuf, gbuf,
553 					    (rt->rt_ifp != NULL) ?
554 					    rt->rt_ifp->if_xname : "",
555 					    rt->rt_flags,
556 					    (rt->rt_expire - timenow), expire);
557 				}
558 			}
559 			ap->nextstop = lmin(ap->nextstop, rt->rt_expire);
560 			RT_UNLOCK(rt);
561 		}
562 	} else {
563 		RT_UNLOCK(rt);
564 	}
565 
566 	return 0;
567 }
568 
569 #define RTQ_TIMEOUT     60*10   /* run no less than once every ten minutes */
570 static int rtq_timeout = RTQ_TIMEOUT;
571 
572 static void
in_rtqtimo(void * targ)573 in_rtqtimo(void *targ)
574 {
575 #pragma unused(targ)
576 	struct radix_node_head *rnh;
577 	struct rtqk_arg arg;
578 	struct timeval atv;
579 	static uint64_t last_adjusted_timeout = 0;
580 	boolean_t verbose = (rt_verbose > 1);
581 	uint64_t timenow;
582 	uint32_t ours;
583 
584 	lck_mtx_lock(rnh_lock);
585 	rnh = rt_tables[AF_INET];
586 	VERIFY(rnh != NULL);
587 
588 	/* Get the timestamp after we acquire the lock for better accuracy */
589 	timenow = net_uptime();
590 	if (verbose) {
591 		os_log_debug(OS_LOG_DEFAULT, "%s: initial nextstop is T+%u seconds\n",
592 		    __func__, rtq_timeout);
593 	}
594 	bzero(&arg, sizeof(arg));
595 	arg.rnh = rnh;
596 	arg.nextstop = timenow + rtq_timeout;
597 	rnh->rnh_walktree(rnh, in_rtqkill, &arg);
598 	if (verbose) {
599 		os_log_debug(OS_LOG_DEFAULT, "%s: found %u, killed %u\n", __func__,
600 		    arg.found, arg.killed);
601 	}
602 	/*
603 	 * Attempt to be somewhat dynamic about this:
604 	 * If there are ``too many'' routes sitting around taking up space,
605 	 * then crank down the timeout, and see if we can't make some more
606 	 * go away.  However, we make sure that we will never adjust more
607 	 * than once in rtq_timeout seconds, to keep from cranking down too
608 	 * hard.
609 	 */
610 	ours = (arg.found - arg.killed);
611 	if (ours > rtq_toomany &&
612 	    ((timenow - last_adjusted_timeout) >= (uint64_t)rtq_timeout) &&
613 	    rtq_reallyold > rtq_minreallyold) {
614 		rtq_reallyold = 2 * rtq_reallyold / 3;
615 		if (rtq_reallyold < rtq_minreallyold) {
616 			rtq_reallyold = rtq_minreallyold;
617 		}
618 
619 		last_adjusted_timeout = timenow;
620 		if (verbose) {
621 			os_log_debug(OS_LOG_DEFAULT, "%s: adjusted rtq_reallyold to %d "
622 			    "seconds\n", __func__, rtq_reallyold);
623 		}
624 		arg.found = arg.killed = 0;
625 		arg.updating = 1;
626 		rnh->rnh_walktree(rnh, in_rtqkill, &arg);
627 	}
628 
629 	atv.tv_usec = 0;
630 	atv.tv_sec = arg.nextstop - timenow;
631 	/* re-arm the timer only if there's work to do */
632 	in_rtqtimo_run = 0;
633 	if (ours > 0) {
634 		in_sched_rtqtimo(&atv);
635 	} else if (verbose) {
636 		os_log_debug(OS_LOG_DEFAULT, "%s: not rescheduling timer\n", __func__);
637 	}
638 	lck_mtx_unlock(rnh_lock);
639 }
640 
641 static void
in_sched_rtqtimo(struct timeval * atv)642 in_sched_rtqtimo(struct timeval *atv)
643 {
644 	LCK_MTX_ASSERT(rnh_lock, LCK_MTX_ASSERT_OWNED);
645 
646 	if (!in_rtqtimo_run) {
647 		struct timeval tv;
648 
649 		if (atv == NULL) {
650 			tv.tv_usec = 0;
651 			tv.tv_sec = MAX(rtq_timeout / 10, 1);
652 			atv = &tv;
653 		}
654 		if (rt_verbose > 1) {
655 			os_log_debug(OS_LOG_DEFAULT, "%s: timer scheduled in "
656 			    "T+%llus.%lluu\n", __func__,
657 			    (uint64_t)atv->tv_sec, (uint64_t)atv->tv_usec);
658 		}
659 		in_rtqtimo_run = 1;
660 		timeout(in_rtqtimo, NULL, tvtohz(atv));
661 	}
662 }
663 
664 void
in_rtqdrain(void)665 in_rtqdrain(void)
666 {
667 	struct radix_node_head *rnh;
668 	struct rtqk_arg arg;
669 
670 	if (rt_verbose > 1) {
671 		os_log_debug(OS_LOG_DEFAULT, "%s: draining routes\n", __func__);
672 	}
673 
674 	lck_mtx_lock(rnh_lock);
675 	rnh = rt_tables[AF_INET];
676 	VERIFY(rnh != NULL);
677 	bzero(&arg, sizeof(arg));
678 	arg.rnh = rnh;
679 	arg.draining = 1;
680 	rnh->rnh_walktree(rnh, in_rtqkill, &arg);
681 	lck_mtx_unlock(rnh_lock);
682 }
683 
684 /*
685  * Initialize our routing tree.
686  */
687 int
in_inithead(void ** head,int off)688 in_inithead(void **head, int off)
689 {
690 	radix_node_head_ref_t rnh;
691 
692 	void *__single *__single inet_rt_table = (void *__single *__single)&rt_tables[AF_INET];
693 
694 	/* If called from route_init(), make sure it is exactly once */
695 	VERIFY(head != inet_rt_table || *head == NULL);
696 
697 	if (!rn_inithead(head, off)) {
698 		return 0;
699 	}
700 
701 	/*
702 	 * We can get here from nfs_subs.c as well, in which case this
703 	 * won't be for the real routing table and thus we're done;
704 	 * this also takes care of the case when we're called more than
705 	 * once from anywhere but route_init().
706 	 */
707 	if (head != inet_rt_table) {
708 		return 1;     /* only do this for the real routing table */
709 	}
710 	rnh = *(void * __single * __single)head;
711 	rnh->rnh_addaddr = in_addroute;
712 	rnh->rnh_deladdr = in_deleteroute;
713 	rnh->rnh_matchaddr = in_matroute;
714 	rnh->rnh_matchaddr_args = in_matroute_args;
715 	rnh->rnh_close = in_clsroute;
716 	return 1;
717 }
718 
719 /*
720  * This zaps old routes when the interface goes down or interface
721  * address is deleted.  In the latter case, it deletes static routes
722  * that point to this address.  If we don't do this, we may end up
723  * using the old address in the future.  The ones we always want to
724  * get rid of are things like ARP entries, since the user might down
725  * the interface, walk over to a completely different network, and
726  * plug back in.
727  */
728 struct in_ifadown_arg {
729 	struct radix_node_head *rnh;
730 	struct ifaddr *ifa;
731 	int del;
732 };
733 __CCT_DECLARE_CONSTRAINED_PTR_TYPE(struct in_ifadown_arg, in_ifadown_arg, __CCT_REF);
734 
735 static int
in_ifadownkill(struct radix_node * rn,void * xap)736 in_ifadownkill(struct radix_node *rn, void *xap)
737 {
738 	char dbuf[MAX_IPv4_STR_LEN], gbuf[MAX_IPv4_STR_LEN];
739 	in_ifadown_arg_ref_t ap = xap;
740 	rtentry_ref_t rt = RT(rn);
741 	boolean_t verbose = (rt_verbose > 1);
742 	int err;
743 
744 	LCK_MTX_ASSERT(rnh_lock, LCK_MTX_ASSERT_OWNED);
745 
746 	RT_LOCK(rt);
747 	if (rt->rt_ifa == ap->ifa &&
748 	    (ap->del || !(rt->rt_flags & RTF_STATIC))) {
749 		rt_str(rt, dbuf, sizeof(dbuf), gbuf, sizeof(gbuf));
750 		if (verbose) {
751 			os_log_debug(OS_LOG_DEFAULT, "%s: deleting route to %s->%s->%s, "
752 			    "flags=0x%x\n", __func__, dbuf, gbuf,
753 			    (rt->rt_ifp != NULL) ? rt->rt_ifp->if_xname : "",
754 			    rt->rt_flags);
755 		}
756 
757 		RT_ADDREF_LOCKED(rt);   /* for us to free below */
758 		/*
759 		 * We need to disable the automatic prune that happens
760 		 * in this case in rtrequest() because it will blow
761 		 * away the pointers that rn_walktree() needs in order
762 		 * continue our descent.  We will end up deleting all
763 		 * the routes that rtrequest() would have in any case,
764 		 * so that behavior is not needed there.  Safe to drop
765 		 * rt_lock and use rt_key, rt_gateway, since holding
766 		 * rnh_lock here prevents another thread from calling
767 		 * rt_setgate() on this route.
768 		 */
769 		rt->rt_flags &= ~(RTF_CLONING | RTF_PRCLONING);
770 		RT_UNLOCK(rt);
771 		err = rtrequest_locked(RTM_DELETE, rt_key(rt),
772 		    rt->rt_gateway, rt_mask(rt), rt->rt_flags, NULL);
773 		if (err != 0) {
774 			RT_LOCK(rt);
775 			if (!verbose) {
776 				rt_str(rt, dbuf, sizeof(dbuf),
777 				    gbuf, sizeof(gbuf));
778 			}
779 			os_log_error(OS_LOG_DEFAULT, "%s: error deleting route to "
780 			    "%s->%s->%s, flags=0x%x, err=%d\n", __func__,
781 			    dbuf, gbuf, (rt->rt_ifp != NULL) ?
782 			    rt->rt_ifp->if_xname : "", rt->rt_flags,
783 			    err);
784 			RT_UNLOCK(rt);
785 		}
786 		rtfree_locked(rt);
787 	} else {
788 		RT_UNLOCK(rt);
789 	}
790 	return 0;
791 }
792 
793 int
in_ifadown(struct ifaddr * ifa,int delete)794 in_ifadown(struct ifaddr *ifa, int delete)
795 {
796 	struct in_ifadown_arg arg;
797 	struct radix_node_head *rnh;
798 
799 	LCK_MTX_ASSERT(rnh_lock, LCK_MTX_ASSERT_OWNED);
800 
801 	/*
802 	 * Holding rnh_lock here prevents the possibility of
803 	 * ifa from changing (e.g. in_ifinit), so it is safe
804 	 * to access its ifa_addr without locking.
805 	 */
806 	if (ifa->ifa_addr->sa_family != AF_INET) {
807 		return 1;
808 	}
809 
810 	/* trigger route cache reevaluation */
811 	routegenid_inet_update();
812 
813 	arg.rnh = rnh = rt_tables[AF_INET];
814 	arg.ifa = ifa;
815 	arg.del = delete;
816 	rnh->rnh_walktree(rnh, in_ifadownkill, &arg);
817 	IFA_LOCK_SPIN(ifa);
818 	ifa->ifa_flags &= ~IFA_ROUTE;
819 	IFA_UNLOCK(ifa);
820 	return 0;
821 }
822