1 | /* |
2 | * Copyright (c) 2003-2016 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 | /* |
30 | * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. |
31 | * All rights reserved. |
32 | * |
33 | * Redistribution and use in source and binary forms, with or without |
34 | * modification, are permitted provided that the following conditions |
35 | * are met: |
36 | * 1. Redistributions of source code must retain the above copyright |
37 | * notice, this list of conditions and the following disclaimer. |
38 | * 2. Redistributions in binary form must reproduce the above copyright |
39 | * notice, this list of conditions and the following disclaimer in the |
40 | * documentation and/or other materials provided with the distribution. |
41 | * 3. Neither the name of the project nor the names of its contributors |
42 | * may be used to endorse or promote products derived from this software |
43 | * without specific prior written permission. |
44 | * |
45 | * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND |
46 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
47 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
48 | * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE |
49 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
50 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS |
51 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
52 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
53 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY |
54 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
55 | * SUCH DAMAGE. |
56 | */ |
57 | |
58 | /* |
59 | * Copyright 1994, 1995 Massachusetts Institute of Technology |
60 | * |
61 | * Permission to use, copy, modify, and distribute this software and |
62 | * its documentation for any purpose and without fee is hereby |
63 | * granted, provided that both the above copyright notice and this |
64 | * permission notice appear in all copies, that both the above |
65 | * copyright notice and this permission notice appear in all |
66 | * supporting documentation, and that the name of M.I.T. not be used |
67 | * in advertising or publicity pertaining to distribution of the |
68 | * software without specific, written prior permission. M.I.T. makes |
69 | * no representations about the suitability of this software for any |
70 | * purpose. It is provided "as is" without express or implied |
71 | * warranty. |
72 | * |
73 | * THIS SOFTWARE IS PROVIDED BY M.I.T. ``AS IS''. M.I.T. DISCLAIMS |
74 | * ALL EXPRESS OR IMPLIED WARRANTIES WITH REGARD TO THIS SOFTWARE, |
75 | * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF |
76 | * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. IN NO EVENT |
77 | * SHALL M.I.T. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, |
78 | * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT |
79 | * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF |
80 | * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND |
81 | * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, |
82 | * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT |
83 | * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
84 | * SUCH DAMAGE. |
85 | * |
86 | */ |
87 | |
88 | /* |
89 | * This code does two things necessary for the enhanced TCP metrics to |
90 | * function in a useful manner: |
91 | * 1) It marks all non-host routes as `cloning', thus ensuring that |
92 | * every actual reference to such a route actually gets turned |
93 | * into a reference to a host route to the specific destination |
94 | * requested. |
95 | * 2) When such routes lose all their references, it arranges for them |
96 | * to be deleted in some random collection of circumstances, so that |
97 | * a large quantity of stale routing data is not kept in kernel memory |
98 | * indefinitely. See in6_rtqtimo() below for the exact mechanism. |
99 | */ |
100 | |
101 | #include <sys/param.h> |
102 | #include <sys/systm.h> |
103 | #include <sys/kernel.h> |
104 | #include <sys/sysctl.h> |
105 | #include <kern/queue.h> |
106 | #include <sys/socket.h> |
107 | #include <sys/socketvar.h> |
108 | #include <sys/protosw.h> |
109 | #include <sys/mbuf.h> |
110 | #include <sys/syslog.h> |
111 | #include <sys/mcache.h> |
112 | #include <kern/locks.h> |
113 | |
114 | #include <net/if.h> |
115 | #include <net/route.h> |
116 | #include <netinet/in.h> |
117 | #include <netinet/ip_var.h> |
118 | #include <netinet/in_var.h> |
119 | |
120 | #include <netinet/ip6.h> |
121 | #include <netinet6/ip6_var.h> |
122 | |
123 | #include <netinet/icmp6.h> |
124 | |
125 | #include <netinet/tcp.h> |
126 | #include <netinet/tcp_seq.h> |
127 | #include <netinet/tcp_timer.h> |
128 | #include <netinet/tcp_var.h> |
129 | |
130 | extern int tvtohz(struct timeval *); |
131 | |
132 | static int in6_rtqtimo_run; /* in6_rtqtimo is scheduled to run */ |
133 | static void in6_rtqtimo(void *); |
134 | static void in6_sched_rtqtimo(struct timeval *); |
135 | |
136 | static struct radix_node *in6_addroute(void *, void *, struct radix_node_head *, |
137 | struct radix_node *); |
138 | static struct radix_node *in6_deleteroute(void *, void *, |
139 | struct radix_node_head *); |
140 | static struct radix_node *in6_matroute(void *, struct radix_node_head *); |
141 | static struct radix_node *in6_matroute_args(void *, struct radix_node_head *, |
142 | rn_matchf_t *, void *); |
143 | static void in6_clsroute(struct radix_node *, struct radix_node_head *); |
144 | static int in6_rtqkill(struct radix_node *, void *); |
145 | |
146 | /* |
147 | * Accessed by in6_addroute(), in6_deleteroute() and in6_rtqkill(), during |
148 | * which the routing lock (rnh_lock) is held and thus protects the variable. |
149 | */ |
150 | static int in6dynroutes; |
151 | |
152 | /* |
153 | * Do what we need to do when inserting a route. |
154 | */ |
155 | static struct radix_node * |
156 | in6_addroute(void *v_arg, void *n_arg, struct radix_node_head *head, |
157 | struct radix_node *treenodes) |
158 | { |
159 | struct rtentry *rt = (struct rtentry *)treenodes; |
160 | struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)(void *)rt_key(rt); |
161 | struct radix_node *ret; |
162 | char dbuf[MAX_IPv6_STR_LEN], gbuf[MAX_IPv6_STR_LEN]; |
163 | uint32_t flags = rt->rt_flags; |
164 | boolean_t verbose = (rt_verbose > 1); |
165 | |
166 | LCK_MTX_ASSERT(rnh_lock, LCK_MTX_ASSERT_OWNED); |
167 | RT_LOCK_ASSERT_HELD(rt); |
168 | |
169 | if (verbose) |
170 | rt_str(rt, dbuf, sizeof (dbuf), gbuf, sizeof (gbuf)); |
171 | |
172 | /* |
173 | * If this is a dynamic route (which is created via Redirect) and |
174 | * we already have the maximum acceptable number of such route entries, |
175 | * reject creating a new one. We could initiate garbage collection to |
176 | * make available space right now, but the benefit would probably not |
177 | * be worth the cleaning overhead; we only have to endure a slightly |
178 | * suboptimal path even without the redirected route. |
179 | */ |
180 | if ((rt->rt_flags & RTF_DYNAMIC) && |
181 | ip6_maxdynroutes >= 0 && in6dynroutes >= ip6_maxdynroutes) |
182 | return (NULL); |
183 | |
184 | /* |
185 | * For IPv6, all unicast non-host routes are automatically cloning. |
186 | */ |
187 | if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) |
188 | rt->rt_flags |= RTF_MULTICAST; |
189 | |
190 | if (!(rt->rt_flags & (RTF_HOST | RTF_CLONING | RTF_MULTICAST))) |
191 | rt->rt_flags |= RTF_PRCLONING; |
192 | |
193 | /* |
194 | * A little bit of help for both IPv6 output and input: |
195 | * For local addresses, we make sure that RTF_LOCAL is set, |
196 | * with the thought that this might one day be used to speed up |
197 | * ip_input(). |
198 | * |
199 | * We also mark routes to multicast addresses as such, because |
200 | * it's easy to do and might be useful (but this is much more |
201 | * dubious since it's so easy to inspect the address). (This |
202 | * is done above.) |
203 | * |
204 | * XXX |
205 | * should elaborate the code. |
206 | */ |
207 | if (rt->rt_flags & RTF_HOST) { |
208 | IFA_LOCK_SPIN(rt->rt_ifa); |
209 | if (IN6_ARE_ADDR_EQUAL(&satosin6(rt->rt_ifa->ifa_addr)-> |
210 | sin6_addr, &sin6->sin6_addr)) { |
211 | rt->rt_flags |= RTF_LOCAL; |
212 | } |
213 | IFA_UNLOCK(rt->rt_ifa); |
214 | } |
215 | |
216 | if (!rt->rt_rmx.rmx_mtu && !(rt->rt_rmx.rmx_locks & RTV_MTU) && |
217 | rt->rt_ifp) |
218 | rt->rt_rmx.rmx_mtu = rt->rt_ifp->if_mtu; |
219 | |
220 | ret = rn_addroute(v_arg, n_arg, head, treenodes); |
221 | if (ret == NULL && (rt->rt_flags & RTF_HOST)) { |
222 | struct rtentry *rt2; |
223 | /* |
224 | * We are trying to add a host route, but can't. |
225 | * Find out if it is because of an |
226 | * ND6 entry and delete it if so. |
227 | */ |
228 | rt2 = rtalloc1_scoped_locked((struct sockaddr *)sin6, 0, |
229 | RTF_CLONING | RTF_PRCLONING, sin6_get_ifscope(rt_key(rt))); |
230 | if (rt2 != NULL) { |
231 | char dbufc[MAX_IPv6_STR_LEN]; |
232 | |
233 | RT_LOCK(rt2); |
234 | if (verbose) |
235 | rt_str(rt2, dbufc, sizeof (dbufc), NULL, 0); |
236 | |
237 | if ((rt2->rt_flags & RTF_LLINFO) && |
238 | (rt2->rt_flags & RTF_HOST) && |
239 | rt2->rt_gateway != NULL && |
240 | rt2->rt_gateway->sa_family == AF_LINK) { |
241 | if (verbose) { |
242 | log(LOG_DEBUG, "%s: unable to insert " |
243 | "route to %s:%s, flags=%b, due to " |
244 | "existing ND6 route %s->%s " |
245 | "flags=%b, attempting to delete\n" , |
246 | __func__, dbuf, |
247 | (rt->rt_ifp != NULL) ? |
248 | rt->rt_ifp->if_xname : "" , |
249 | rt->rt_flags, RTF_BITS, |
250 | dbufc, (rt2->rt_ifp != NULL) ? |
251 | rt2->rt_ifp->if_xname : "" , |
252 | rt2->rt_flags, RTF_BITS); |
253 | } |
254 | /* |
255 | * Safe to drop rt_lock and use rt_key, |
256 | * rt_gateway, since holding rnh_lock here |
257 | * prevents another thread from calling |
258 | * rt_setgate() on this route. |
259 | */ |
260 | RT_UNLOCK(rt2); |
261 | (void) rtrequest_locked(RTM_DELETE, rt_key(rt2), |
262 | rt2->rt_gateway, rt_mask(rt2), |
263 | rt2->rt_flags, NULL); |
264 | ret = rn_addroute(v_arg, n_arg, head, |
265 | treenodes); |
266 | } else { |
267 | RT_UNLOCK(rt2); |
268 | } |
269 | rtfree_locked(rt2); |
270 | } |
271 | } else if (ret == NULL && (rt->rt_flags & RTF_CLONING)) { |
272 | struct rtentry *rt2; |
273 | /* |
274 | * We are trying to add a net route, but can't. |
275 | * The following case should be allowed, so we'll make a |
276 | * special check for this: |
277 | * Two IPv6 addresses with the same prefix is assigned |
278 | * to a single interrface. |
279 | * # ifconfig if0 inet6 3ffe:0501::1 prefix 64 alias (*1) |
280 | * # ifconfig if0 inet6 3ffe:0501::2 prefix 64 alias (*2) |
281 | * In this case, (*1) and (*2) want to add the same |
282 | * net route entry, 3ffe:0501:: -> if0. |
283 | * This case should not raise an error. |
284 | */ |
285 | rt2 = rtalloc1_scoped_locked((struct sockaddr *)sin6, 0, |
286 | RTF_CLONING | RTF_PRCLONING, sin6_get_ifscope(rt_key(rt))); |
287 | if (rt2 != NULL) { |
288 | RT_LOCK(rt2); |
289 | if ((rt2->rt_flags & (RTF_CLONING|RTF_HOST| |
290 | RTF_GATEWAY)) == RTF_CLONING && |
291 | rt2->rt_gateway && |
292 | rt2->rt_gateway->sa_family == AF_LINK && |
293 | rt2->rt_ifp == rt->rt_ifp) { |
294 | ret = rt2->rt_nodes; |
295 | } |
296 | RT_UNLOCK(rt2); |
297 | rtfree_locked(rt2); |
298 | } |
299 | } |
300 | |
301 | if (ret != NULL && (rt->rt_flags & RTF_DYNAMIC)) |
302 | in6dynroutes++; |
303 | |
304 | if (!verbose) |
305 | goto done; |
306 | |
307 | if (ret != NULL) { |
308 | if (flags != rt->rt_flags) { |
309 | log(LOG_DEBUG, "%s: route to %s->%s->%s inserted, " |
310 | "oflags=%b, flags=%b\n" , __func__, |
311 | dbuf, gbuf, (rt->rt_ifp != NULL) ? |
312 | rt->rt_ifp->if_xname : "" , flags, RTF_BITS, |
313 | rt->rt_flags, RTF_BITS); |
314 | } else { |
315 | log(LOG_DEBUG, "%s: route to %s->%s->%s inserted, " |
316 | "flags=%b\n" , __func__, dbuf, gbuf, |
317 | (rt->rt_ifp != NULL) ? rt->rt_ifp->if_xname : "" , |
318 | rt->rt_flags, RTF_BITS); |
319 | } |
320 | } else { |
321 | log(LOG_DEBUG, "%s: unable to insert route to %s->%s->%s, " |
322 | "flags=%b, already exists\n" , __func__, dbuf, gbuf, |
323 | (rt->rt_ifp != NULL) ? rt->rt_ifp->if_xname : "" , |
324 | rt->rt_flags, RTF_BITS); |
325 | } |
326 | done: |
327 | return (ret); |
328 | } |
329 | |
330 | static struct radix_node * |
331 | in6_deleteroute(void *v_arg, void *netmask_arg, struct radix_node_head *head) |
332 | { |
333 | struct radix_node *rn; |
334 | |
335 | LCK_MTX_ASSERT(rnh_lock, LCK_MTX_ASSERT_OWNED); |
336 | |
337 | rn = rn_delete(v_arg, netmask_arg, head); |
338 | if (rn != NULL) { |
339 | struct rtentry *rt = (struct rtentry *)rn; |
340 | |
341 | RT_LOCK(rt); |
342 | if (rt->rt_flags & RTF_DYNAMIC) |
343 | in6dynroutes--; |
344 | if (rt_verbose > 1) { |
345 | char dbuf[MAX_IPv6_STR_LEN], gbuf[MAX_IPv6_STR_LEN]; |
346 | |
347 | rt_str(rt, dbuf, sizeof (dbuf), gbuf, sizeof (gbuf)); |
348 | log(LOG_DEBUG, "%s: route to %s->%s->%s deleted, " |
349 | "flags=%b\n" , __func__, dbuf, gbuf, |
350 | (rt->rt_ifp != NULL) ? rt->rt_ifp->if_xname : "" , |
351 | rt->rt_flags, RTF_BITS); |
352 | } |
353 | RT_UNLOCK(rt); |
354 | } |
355 | return (rn); |
356 | } |
357 | |
358 | /* |
359 | * Validate (unexpire) an expiring AF_INET6 route. |
360 | */ |
361 | struct radix_node * |
362 | in6_validate(struct radix_node *rn) |
363 | { |
364 | struct rtentry *rt = (struct rtentry *)rn; |
365 | |
366 | RT_LOCK_ASSERT_HELD(rt); |
367 | |
368 | /* This is first reference? */ |
369 | if (rt->rt_refcnt == 0) { |
370 | if (rt_verbose > 2) { |
371 | char dbuf[MAX_IPv6_STR_LEN], gbuf[MAX_IPv6_STR_LEN]; |
372 | |
373 | rt_str(rt, dbuf, sizeof (dbuf), gbuf, sizeof (gbuf)); |
374 | log(LOG_DEBUG, "%s: route to %s->%s->%s validated, " |
375 | "flags=%b\n" , __func__, dbuf, gbuf, |
376 | (rt->rt_ifp != NULL) ? rt->rt_ifp->if_xname : "" , |
377 | rt->rt_flags, RTF_BITS); |
378 | } |
379 | |
380 | /* |
381 | * It's one of ours; unexpire it. If the timer is already |
382 | * scheduled, let it run later as it won't re-arm itself |
383 | * if there's nothing to do. |
384 | */ |
385 | if (rt->rt_flags & RTPRF_OURS) { |
386 | rt->rt_flags &= ~RTPRF_OURS; |
387 | rt_setexpire(rt, 0); |
388 | } |
389 | } |
390 | return (rn); |
391 | } |
392 | |
393 | /* |
394 | * Similar to in6_matroute_args except without the leaf-matching parameters. |
395 | */ |
396 | static struct radix_node * |
397 | in6_matroute(void *v_arg, struct radix_node_head *head) |
398 | { |
399 | return (in6_matroute_args(v_arg, head, NULL, NULL)); |
400 | } |
401 | |
402 | /* |
403 | * This code is the inverse of in6_clsroute: on first reference, if we |
404 | * were managing the route, stop doing so and set the expiration timer |
405 | * back off again. |
406 | */ |
407 | static struct radix_node * |
408 | in6_matroute_args(void *v_arg, struct radix_node_head *head, |
409 | rn_matchf_t *f, void *w) |
410 | { |
411 | struct radix_node *rn = rn_match_args(v_arg, head, f, w); |
412 | |
413 | if (rn != NULL) { |
414 | RT_LOCK_SPIN((struct rtentry *)rn); |
415 | in6_validate(rn); |
416 | RT_UNLOCK((struct rtentry *)rn); |
417 | } |
418 | return (rn); |
419 | } |
420 | |
421 | SYSCTL_DECL(_net_inet6_ip6); |
422 | |
423 | /* one hour is ``really old'' */ |
424 | static uint32_t rtq_reallyold = 60*60; |
425 | SYSCTL_UINT(_net_inet6_ip6, IPV6CTL_RTEXPIRE, rtexpire, |
426 | CTLFLAG_RW | CTLFLAG_LOCKED, &rtq_reallyold, 0, "" ); |
427 | |
428 | /* never automatically crank down to less */ |
429 | static uint32_t rtq_minreallyold = 10; |
430 | SYSCTL_UINT(_net_inet6_ip6, IPV6CTL_RTMINEXPIRE, rtminexpire, |
431 | CTLFLAG_RW | CTLFLAG_LOCKED, &rtq_minreallyold, 0, "" ); |
432 | |
433 | /* 128 cached routes is ``too many'' */ |
434 | static uint32_t rtq_toomany = 128; |
435 | SYSCTL_UINT(_net_inet6_ip6, IPV6CTL_RTMAXCACHE, rtmaxcache, |
436 | CTLFLAG_RW | CTLFLAG_LOCKED, &rtq_toomany, 0, "" ); |
437 | |
438 | /* |
439 | * On last reference drop, mark the route as belong to us so that it can be |
440 | * timed out. |
441 | */ |
442 | static void |
443 | in6_clsroute(struct radix_node *rn, struct radix_node_head *head) |
444 | { |
445 | #pragma unused(head) |
446 | char dbuf[MAX_IPv6_STR_LEN], gbuf[MAX_IPv6_STR_LEN]; |
447 | struct rtentry *rt = (struct rtentry *)rn; |
448 | boolean_t verbose = (rt_verbose > 1); |
449 | |
450 | LCK_MTX_ASSERT(rnh_lock, LCK_MTX_ASSERT_OWNED); |
451 | RT_LOCK_ASSERT_HELD(rt); |
452 | |
453 | if (!(rt->rt_flags & RTF_UP)) |
454 | return; /* prophylactic measures */ |
455 | |
456 | if ((rt->rt_flags & (RTF_LLINFO | RTF_HOST)) != RTF_HOST) |
457 | return; |
458 | |
459 | if (rt->rt_flags & RTPRF_OURS) |
460 | return; |
461 | |
462 | if (!(rt->rt_flags & (RTF_WASCLONED | RTF_DYNAMIC))) |
463 | return; |
464 | |
465 | if (verbose) |
466 | rt_str(rt, dbuf, sizeof (dbuf), gbuf, sizeof (gbuf)); |
467 | |
468 | /* |
469 | * Delete the route immediately if RTF_DELCLONE is set or |
470 | * if route caching is disabled (rtq_reallyold set to 0). |
471 | * Otherwise, let it expire and be deleted by in6_rtqkill(). |
472 | */ |
473 | if ((rt->rt_flags & RTF_DELCLONE) || rtq_reallyold == 0) { |
474 | int err; |
475 | |
476 | if (verbose) { |
477 | log(LOG_DEBUG, "%s: deleting route to %s->%s->%s, " |
478 | "flags=%b\n" , __func__, dbuf, gbuf, |
479 | (rt->rt_ifp != NULL) ? rt->rt_ifp->if_xname : "" , |
480 | rt->rt_flags, RTF_BITS); |
481 | } |
482 | /* |
483 | * Delete the route from the radix tree but since we are |
484 | * called when the route's reference count is 0, don't |
485 | * deallocate it until we return from this routine by |
486 | * telling rtrequest that we're interested in it. |
487 | * Safe to drop rt_lock and use rt_key, rt_gateway, |
488 | * since holding rnh_lock here prevents another thread |
489 | * from calling rt_setgate() on this route. |
490 | */ |
491 | RT_UNLOCK(rt); |
492 | err = rtrequest_locked(RTM_DELETE, rt_key(rt), |
493 | rt->rt_gateway, rt_mask(rt), rt->rt_flags, &rt); |
494 | if (err == 0) { |
495 | /* Now let the caller free it */ |
496 | RT_LOCK(rt); |
497 | RT_REMREF_LOCKED(rt); |
498 | } else { |
499 | RT_LOCK(rt); |
500 | if (!verbose) |
501 | rt_str(rt, dbuf, sizeof (dbuf), |
502 | gbuf, sizeof (gbuf)); |
503 | log(LOG_ERR, "%s: error deleting route to " |
504 | "%s->%s->%s, flags=%b, err=%d\n" , __func__, |
505 | dbuf, gbuf, (rt->rt_ifp != NULL) ? |
506 | rt->rt_ifp->if_xname : "" , rt->rt_flags, |
507 | RTF_BITS, err); |
508 | } |
509 | } else { |
510 | uint64_t timenow; |
511 | |
512 | timenow = net_uptime(); |
513 | rt->rt_flags |= RTPRF_OURS; |
514 | rt_setexpire(rt, timenow + rtq_reallyold); |
515 | |
516 | if (verbose) { |
517 | log(LOG_DEBUG, "%s: route to %s->%s->%s invalidated, " |
518 | "flags=%b, expire=T+%u\n" , __func__, dbuf, gbuf, |
519 | (rt->rt_ifp != NULL) ? rt->rt_ifp->if_xname : "" , |
520 | rt->rt_flags, RTF_BITS, rt->rt_expire - timenow); |
521 | } |
522 | |
523 | /* We have at least one entry; arm the timer if not already */ |
524 | in6_sched_rtqtimo(NULL); |
525 | } |
526 | } |
527 | |
528 | struct rtqk_arg { |
529 | struct radix_node_head *rnh; |
530 | int updating; |
531 | int draining; |
532 | uint32_t killed; |
533 | uint32_t found; |
534 | uint64_t nextstop; |
535 | }; |
536 | |
537 | /* |
538 | * Get rid of old routes. When draining, this deletes everything, even when |
539 | * the timeout is not expired yet. This also applies if the route is dynamic |
540 | * and there are sufficiently large number of such routes (more than a half of |
541 | * maximum). When updating, this makes sure that nothing has a timeout longer |
542 | * than the current value of rtq_reallyold. |
543 | */ |
544 | static int |
545 | in6_rtqkill(struct radix_node *rn, void *rock) |
546 | { |
547 | struct rtqk_arg *ap = rock; |
548 | struct rtentry *rt = (struct rtentry *)rn; |
549 | boolean_t verbose = (rt_verbose > 1); |
550 | uint64_t timenow; |
551 | int err; |
552 | |
553 | timenow = net_uptime(); |
554 | LCK_MTX_ASSERT(rnh_lock, LCK_MTX_ASSERT_OWNED); |
555 | |
556 | RT_LOCK(rt); |
557 | if (rt->rt_flags & RTPRF_OURS) { |
558 | char dbuf[MAX_IPv6_STR_LEN], gbuf[MAX_IPv6_STR_LEN]; |
559 | |
560 | if (verbose) |
561 | rt_str(rt, dbuf, sizeof (dbuf), gbuf, sizeof (gbuf)); |
562 | |
563 | ap->found++; |
564 | VERIFY(rt->rt_expire == 0 || rt->rt_rmx.rmx_expire != 0); |
565 | VERIFY(rt->rt_expire != 0 || rt->rt_rmx.rmx_expire == 0); |
566 | if (ap->draining || rt->rt_expire <= timenow || |
567 | ((rt->rt_flags & RTF_DYNAMIC) && ip6_maxdynroutes >= 0 && |
568 | in6dynroutes > ip6_maxdynroutes / 2)) { |
569 | if (rt->rt_refcnt > 0) { |
570 | panic("%s: route %p marked with RTPRF_OURS " |
571 | "with non-zero refcnt (%u)" , __func__, |
572 | rt, rt->rt_refcnt); |
573 | /* NOTREACHED */ |
574 | } |
575 | |
576 | if (verbose) { |
577 | log(LOG_DEBUG, "%s: deleting route to " |
578 | "%s->%s->%s, flags=%b, draining=%d\n" , |
579 | __func__, dbuf, gbuf, (rt->rt_ifp != NULL) ? |
580 | rt->rt_ifp->if_xname : "" , rt->rt_flags, |
581 | RTF_BITS, ap->draining); |
582 | } |
583 | RT_ADDREF_LOCKED(rt); /* for us to free below */ |
584 | /* |
585 | * Delete this route since we're done with it; |
586 | * the route may be freed afterwards, so we |
587 | * can no longer refer to 'rt' upon returning |
588 | * from rtrequest(). Safe to drop rt_lock and |
589 | * use rt_key, rt_gateway, since holding rnh_lock |
590 | * here prevents another thread from calling |
591 | * rt_setgate() on this route. |
592 | */ |
593 | RT_UNLOCK(rt); |
594 | err = rtrequest_locked(RTM_DELETE, rt_key(rt), |
595 | rt->rt_gateway, rt_mask(rt), rt->rt_flags, NULL); |
596 | if (err != 0) { |
597 | RT_LOCK(rt); |
598 | if (!verbose) |
599 | rt_str(rt, dbuf, sizeof (dbuf), |
600 | gbuf, sizeof (gbuf)); |
601 | log(LOG_ERR, "%s: error deleting route to " |
602 | "%s->%s->%s, flags=%b, err=%d\n" , __func__, |
603 | dbuf, gbuf, (rt->rt_ifp != NULL) ? |
604 | rt->rt_ifp->if_xname : "" , rt->rt_flags, |
605 | RTF_BITS, err); |
606 | RT_UNLOCK(rt); |
607 | } else { |
608 | ap->killed++; |
609 | } |
610 | rtfree_locked(rt); |
611 | } else { |
612 | uint64_t expire = (rt->rt_expire - timenow); |
613 | |
614 | if (ap->updating && expire > rtq_reallyold) { |
615 | rt_setexpire(rt, timenow + rtq_reallyold); |
616 | if (verbose) { |
617 | log(LOG_DEBUG, "%s: route to " |
618 | "%s->%s->%s, flags=%b, adjusted " |
619 | "expire=T+%u (was T+%u)\n" , |
620 | __func__, dbuf, gbuf, |
621 | (rt->rt_ifp != NULL) ? |
622 | rt->rt_ifp->if_xname : "" , |
623 | rt->rt_flags, RTF_BITS, |
624 | (rt->rt_expire - timenow), expire); |
625 | } |
626 | } |
627 | ap->nextstop = lmin(ap->nextstop, rt->rt_expire); |
628 | RT_UNLOCK(rt); |
629 | } |
630 | } else { |
631 | RT_UNLOCK(rt); |
632 | } |
633 | |
634 | return (0); |
635 | } |
636 | |
637 | #define RTQ_TIMEOUT 60*10 /* run no less than once every ten minutes */ |
638 | static int rtq_timeout = RTQ_TIMEOUT; |
639 | |
640 | static void |
641 | in6_rtqtimo(void *targ) |
642 | { |
643 | #pragma unused(targ) |
644 | struct radix_node_head *rnh; |
645 | struct rtqk_arg arg; |
646 | struct timeval atv; |
647 | static uint64_t last_adjusted_timeout = 0; |
648 | boolean_t verbose = (rt_verbose > 1); |
649 | uint64_t timenow; |
650 | uint32_t ours; |
651 | |
652 | lck_mtx_lock(rnh_lock); |
653 | rnh = rt_tables[AF_INET6]; |
654 | VERIFY(rnh != NULL); |
655 | |
656 | /* Get the timestamp after we acquire the lock for better accuracy */ |
657 | timenow = net_uptime(); |
658 | if (verbose) { |
659 | log(LOG_DEBUG, "%s: initial nextstop is T+%u seconds\n" , |
660 | __func__, rtq_timeout); |
661 | } |
662 | bzero(&arg, sizeof (arg)); |
663 | arg.rnh = rnh; |
664 | arg.nextstop = timenow + rtq_timeout; |
665 | rnh->rnh_walktree(rnh, in6_rtqkill, &arg); |
666 | if (verbose) { |
667 | log(LOG_DEBUG, "%s: found %u, killed %u\n" , __func__, |
668 | arg.found, arg.killed); |
669 | } |
670 | /* |
671 | * Attempt to be somewhat dynamic about this: |
672 | * If there are ``too many'' routes sitting around taking up space, |
673 | * then crank down the timeout, and see if we can't make some more |
674 | * go away. However, we make sure that we will never adjust more |
675 | * than once in rtq_timeout seconds, to keep from cranking down too |
676 | * hard. |
677 | */ |
678 | ours = (arg.found - arg.killed); |
679 | if (ours > rtq_toomany && |
680 | ((timenow - last_adjusted_timeout) >= (uint64_t)rtq_timeout) && |
681 | rtq_reallyold > rtq_minreallyold) { |
682 | rtq_reallyold = 2 * rtq_reallyold / 3; |
683 | if (rtq_reallyold < rtq_minreallyold) |
684 | rtq_reallyold = rtq_minreallyold; |
685 | |
686 | last_adjusted_timeout = timenow; |
687 | if (verbose) { |
688 | log(LOG_DEBUG, "%s: adjusted rtq_reallyold to %d " |
689 | "seconds\n" , __func__, rtq_reallyold); |
690 | } |
691 | arg.found = arg.killed = 0; |
692 | arg.updating = 1; |
693 | rnh->rnh_walktree(rnh, in6_rtqkill, &arg); |
694 | } |
695 | |
696 | atv.tv_usec = 0; |
697 | atv.tv_sec = arg.nextstop - timenow; |
698 | /* re-arm the timer only if there's work to do */ |
699 | in6_rtqtimo_run = 0; |
700 | if (ours > 0) |
701 | in6_sched_rtqtimo(&atv); |
702 | else if (verbose) |
703 | log(LOG_DEBUG, "%s: not rescheduling timer\n" , __func__); |
704 | lck_mtx_unlock(rnh_lock); |
705 | } |
706 | |
707 | static void |
708 | in6_sched_rtqtimo(struct timeval *atv) |
709 | { |
710 | LCK_MTX_ASSERT(rnh_lock, LCK_MTX_ASSERT_OWNED); |
711 | |
712 | if (!in6_rtqtimo_run) { |
713 | struct timeval tv; |
714 | |
715 | if (atv == NULL) { |
716 | tv.tv_usec = 0; |
717 | tv.tv_sec = MAX(rtq_timeout / 10, 1); |
718 | atv = &tv; |
719 | } |
720 | if (rt_verbose > 1) { |
721 | log(LOG_DEBUG, "%s: timer scheduled in " |
722 | "T+%llus.%lluu\n" , __func__, |
723 | (uint64_t)atv->tv_sec, (uint64_t)atv->tv_usec); |
724 | } |
725 | in6_rtqtimo_run = 1; |
726 | timeout(in6_rtqtimo, NULL, tvtohz(atv)); |
727 | } |
728 | } |
729 | |
730 | void |
731 | in6_rtqdrain(void) |
732 | { |
733 | struct radix_node_head *rnh; |
734 | struct rtqk_arg arg; |
735 | |
736 | if (rt_verbose > 1) |
737 | log(LOG_DEBUG, "%s: draining routes\n" , __func__); |
738 | |
739 | lck_mtx_lock(rnh_lock); |
740 | rnh = rt_tables[AF_INET6]; |
741 | VERIFY(rnh != NULL); |
742 | bzero(&arg, sizeof (arg)); |
743 | arg.rnh = rnh; |
744 | arg.draining = 1; |
745 | rnh->rnh_walktree(rnh, in6_rtqkill, &arg); |
746 | lck_mtx_unlock(rnh_lock); |
747 | } |
748 | |
749 | /* |
750 | * Initialize our routing tree. |
751 | */ |
752 | int |
753 | in6_inithead(void **head, int off) |
754 | { |
755 | struct radix_node_head *rnh; |
756 | |
757 | /* If called from route_init(), make sure it is exactly once */ |
758 | VERIFY(head != (void **)&rt_tables[AF_INET6] || *head == NULL); |
759 | |
760 | if (!rn_inithead(head, off)) |
761 | return (0); |
762 | |
763 | /* |
764 | * We can get here from nfs_subs.c as well, in which case this |
765 | * won't be for the real routing table and thus we're done; |
766 | * this also takes care of the case when we're called more than |
767 | * once from anywhere but route_init(). |
768 | */ |
769 | if (head != (void **)&rt_tables[AF_INET6]) |
770 | return (1); /* only do this for the real routing table */ |
771 | |
772 | rnh = *head; |
773 | rnh->rnh_addaddr = in6_addroute; |
774 | rnh->rnh_deladdr = in6_deleteroute; |
775 | rnh->rnh_matchaddr = in6_matroute; |
776 | rnh->rnh_matchaddr_args = in6_matroute_args; |
777 | rnh->rnh_close = in6_clsroute; |
778 | return (1); |
779 | } |
780 | |