xref: /freebsd/sys/kern/kern_sysctl.c (revision b249ce48ea5560afdcff57e72a9880b7d3132434)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Mike Karels at Berkeley Software Design, Inc.
9  *
10  * Quite extensively rewritten by Poul-Henning Kamp of the FreeBSD
11  * project, to make these variables more userfriendly.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. Neither the name of the University nor the names of its contributors
22  *    may be used to endorse or promote products derived from this software
23  *    without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  *
37  *	@(#)kern_sysctl.c	8.4 (Berkeley) 4/14/94
38  */
39 
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
42 
43 #include "opt_capsicum.h"
44 #include "opt_ddb.h"
45 #include "opt_ktrace.h"
46 
47 #include <sys/param.h>
48 #include <sys/fail.h>
49 #include <sys/systm.h>
50 #include <sys/capsicum.h>
51 #include <sys/kernel.h>
52 #include <sys/limits.h>
53 #include <sys/sysctl.h>
54 #include <sys/malloc.h>
55 #include <sys/priv.h>
56 #include <sys/proc.h>
57 #include <sys/jail.h>
58 #include <sys/kdb.h>
59 #include <sys/lock.h>
60 #include <sys/mutex.h>
61 #include <sys/rmlock.h>
62 #include <sys/sbuf.h>
63 #include <sys/sx.h>
64 #include <sys/sysproto.h>
65 #include <sys/uio.h>
66 #ifdef KTRACE
67 #include <sys/ktrace.h>
68 #endif
69 
70 #ifdef DDB
71 #include <ddb/ddb.h>
72 #include <ddb/db_lex.h>
73 #endif
74 
75 #include <net/vnet.h>
76 
77 #include <security/mac/mac_framework.h>
78 
79 #include <vm/vm.h>
80 #include <vm/vm_extern.h>
81 
82 static MALLOC_DEFINE(M_SYSCTL, "sysctl", "sysctl internal magic");
83 static MALLOC_DEFINE(M_SYSCTLOID, "sysctloid", "sysctl dynamic oids");
84 static MALLOC_DEFINE(M_SYSCTLTMP, "sysctltmp", "sysctl temp output buffer");
85 
86 /*
87  * The sysctllock protects the MIB tree.  It also protects sysctl
88  * contexts used with dynamic sysctls.  The sysctl_register_oid() and
89  * sysctl_unregister_oid() routines require the sysctllock to already
90  * be held, so the sysctl_wlock() and sysctl_wunlock() routines are
91  * provided for the few places in the kernel which need to use that
92  * API rather than using the dynamic API.  Use of the dynamic API is
93  * strongly encouraged for most code.
94  *
95  * The sysctlmemlock is used to limit the amount of user memory wired for
96  * sysctl requests.  This is implemented by serializing any userland
97  * sysctl requests larger than a single page via an exclusive lock.
98  */
99 static struct rmlock sysctllock;
100 static struct sx __exclusive_cache_line sysctlmemlock;
101 
102 #define	SYSCTL_WLOCK()		rm_wlock(&sysctllock)
103 #define	SYSCTL_WUNLOCK()	rm_wunlock(&sysctllock)
104 #define	SYSCTL_RLOCK(tracker)	rm_rlock(&sysctllock, (tracker))
105 #define	SYSCTL_RUNLOCK(tracker)	rm_runlock(&sysctllock, (tracker))
106 #define	SYSCTL_WLOCKED()	rm_wowned(&sysctllock)
107 #define	SYSCTL_ASSERT_LOCKED()	rm_assert(&sysctllock, RA_LOCKED)
108 #define	SYSCTL_ASSERT_WLOCKED()	rm_assert(&sysctllock, RA_WLOCKED)
109 #define	SYSCTL_ASSERT_RLOCKED()	rm_assert(&sysctllock, RA_RLOCKED)
110 #define	SYSCTL_INIT()		rm_init_flags(&sysctllock, "sysctl lock", \
111 				    RM_SLEEPABLE)
112 #define	SYSCTL_SLEEP(ch, wmesg, timo)					\
113 				rm_sleep(ch, &sysctllock, 0, wmesg, timo)
114 
115 static int sysctl_root(SYSCTL_HANDLER_ARGS);
116 
117 /* Root list */
118 struct sysctl_oid_list sysctl__children = SLIST_HEAD_INITIALIZER(&sysctl__children);
119 
120 static int	sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del,
121 		    int recurse);
122 static int	sysctl_old_kernel(struct sysctl_req *, const void *, size_t);
123 static int	sysctl_new_kernel(struct sysctl_req *, void *, size_t);
124 
125 static struct sysctl_oid *
126 sysctl_find_oidname(const char *name, struct sysctl_oid_list *list)
127 {
128 	struct sysctl_oid *oidp;
129 
130 	SYSCTL_ASSERT_LOCKED();
131 	SLIST_FOREACH(oidp, list, oid_link) {
132 		if (strcmp(oidp->oid_name, name) == 0) {
133 			return (oidp);
134 		}
135 	}
136 	return (NULL);
137 }
138 
139 /*
140  * Initialization of the MIB tree.
141  *
142  * Order by number in each list.
143  */
144 void
145 sysctl_wlock(void)
146 {
147 
148 	SYSCTL_WLOCK();
149 }
150 
151 void
152 sysctl_wunlock(void)
153 {
154 
155 	SYSCTL_WUNLOCK();
156 }
157 
158 static int
159 sysctl_root_handler_locked(struct sysctl_oid *oid, void *arg1, intmax_t arg2,
160     struct sysctl_req *req, struct rm_priotracker *tracker)
161 {
162 	int error;
163 
164 	if (oid->oid_kind & CTLFLAG_DYN)
165 		atomic_add_int(&oid->oid_running, 1);
166 
167 	if (tracker != NULL)
168 		SYSCTL_RUNLOCK(tracker);
169 	else
170 		SYSCTL_WUNLOCK();
171 
172 	if (!(oid->oid_kind & CTLFLAG_MPSAFE))
173 		mtx_lock(&Giant);
174 	error = oid->oid_handler(oid, arg1, arg2, req);
175 	if (!(oid->oid_kind & CTLFLAG_MPSAFE))
176 		mtx_unlock(&Giant);
177 
178 	KFAIL_POINT_ERROR(_debug_fail_point, sysctl_running, error);
179 
180 	if (tracker != NULL)
181 		SYSCTL_RLOCK(tracker);
182 	else
183 		SYSCTL_WLOCK();
184 
185 	if (oid->oid_kind & CTLFLAG_DYN) {
186 		if (atomic_fetchadd_int(&oid->oid_running, -1) == 1 &&
187 		    (oid->oid_kind & CTLFLAG_DYING) != 0)
188 			wakeup(&oid->oid_running);
189 	}
190 
191 	return (error);
192 }
193 
194 static void
195 sysctl_load_tunable_by_oid_locked(struct sysctl_oid *oidp)
196 {
197 	struct sysctl_req req;
198 	struct sysctl_oid *curr;
199 	char *penv = NULL;
200 	char path[96];
201 	ssize_t rem = sizeof(path);
202 	ssize_t len;
203 	uint8_t data[512] __aligned(sizeof(uint64_t));
204 	int size;
205 	int error;
206 
207 	path[--rem] = 0;
208 
209 	for (curr = oidp; curr != NULL; curr = SYSCTL_PARENT(curr)) {
210 		len = strlen(curr->oid_name);
211 		rem -= len;
212 		if (curr != oidp)
213 			rem -= 1;
214 		if (rem < 0) {
215 			printf("OID path exceeds %d bytes\n", (int)sizeof(path));
216 			return;
217 		}
218 		memcpy(path + rem, curr->oid_name, len);
219 		if (curr != oidp)
220 			path[rem + len] = '.';
221 	}
222 
223 	memset(&req, 0, sizeof(req));
224 
225 	req.td = curthread;
226 	req.oldfunc = sysctl_old_kernel;
227 	req.newfunc = sysctl_new_kernel;
228 	req.lock = REQ_UNWIRED;
229 
230 	switch (oidp->oid_kind & CTLTYPE) {
231 	case CTLTYPE_INT:
232 		if (getenv_array(path + rem, data, sizeof(data), &size,
233 		    sizeof(int), GETENV_SIGNED) == 0)
234 			return;
235 		req.newlen = size;
236 		req.newptr = data;
237 		break;
238 	case CTLTYPE_UINT:
239 		if (getenv_array(path + rem, data, sizeof(data), &size,
240 		    sizeof(int), GETENV_UNSIGNED) == 0)
241 			return;
242 		req.newlen = size;
243 		req.newptr = data;
244 		break;
245 	case CTLTYPE_LONG:
246 		if (getenv_array(path + rem, data, sizeof(data), &size,
247 		    sizeof(long), GETENV_SIGNED) == 0)
248 			return;
249 		req.newlen = size;
250 		req.newptr = data;
251 		break;
252 	case CTLTYPE_ULONG:
253 		if (getenv_array(path + rem, data, sizeof(data), &size,
254 		    sizeof(long), GETENV_UNSIGNED) == 0)
255 			return;
256 		req.newlen = size;
257 		req.newptr = data;
258 		break;
259 	case CTLTYPE_S8:
260 		if (getenv_array(path + rem, data, sizeof(data), &size,
261 		    sizeof(int8_t), GETENV_SIGNED) == 0)
262 			return;
263 		req.newlen = size;
264 		req.newptr = data;
265 		break;
266 	case CTLTYPE_S16:
267 		if (getenv_array(path + rem, data, sizeof(data), &size,
268 		    sizeof(int16_t), GETENV_SIGNED) == 0)
269 			return;
270 		req.newlen = size;
271 		req.newptr = data;
272 		break;
273 	case CTLTYPE_S32:
274 		if (getenv_array(path + rem, data, sizeof(data), &size,
275 		    sizeof(int32_t), GETENV_SIGNED) == 0)
276 			return;
277 		req.newlen = size;
278 		req.newptr = data;
279 		break;
280 	case CTLTYPE_S64:
281 		if (getenv_array(path + rem, data, sizeof(data), &size,
282 		    sizeof(int64_t), GETENV_SIGNED) == 0)
283 			return;
284 		req.newlen = size;
285 		req.newptr = data;
286 		break;
287 	case CTLTYPE_U8:
288 		if (getenv_array(path + rem, data, sizeof(data), &size,
289 		    sizeof(uint8_t), GETENV_UNSIGNED) == 0)
290 			return;
291 		req.newlen = size;
292 		req.newptr = data;
293 		break;
294 	case CTLTYPE_U16:
295 		if (getenv_array(path + rem, data, sizeof(data), &size,
296 		    sizeof(uint16_t), GETENV_UNSIGNED) == 0)
297 			return;
298 		req.newlen = size;
299 		req.newptr = data;
300 		break;
301 	case CTLTYPE_U32:
302 		if (getenv_array(path + rem, data, sizeof(data), &size,
303 		    sizeof(uint32_t), GETENV_UNSIGNED) == 0)
304 			return;
305 		req.newlen = size;
306 		req.newptr = data;
307 		break;
308 	case CTLTYPE_U64:
309 		if (getenv_array(path + rem, data, sizeof(data), &size,
310 		    sizeof(uint64_t), GETENV_UNSIGNED) == 0)
311 			return;
312 		req.newlen = size;
313 		req.newptr = data;
314 		break;
315 	case CTLTYPE_STRING:
316 		penv = kern_getenv(path + rem);
317 		if (penv == NULL)
318 			return;
319 		req.newlen = strlen(penv);
320 		req.newptr = penv;
321 		break;
322 	default:
323 		return;
324 	}
325 	error = sysctl_root_handler_locked(oidp, oidp->oid_arg1,
326 	    oidp->oid_arg2, &req, NULL);
327 	if (error != 0)
328 		printf("Setting sysctl %s failed: %d\n", path + rem, error);
329 	if (penv != NULL)
330 		freeenv(penv);
331 }
332 
333 /*
334  * Locate the path to a given oid.  Returns the length of the resulting path,
335  * or -1 if the oid was not found.  nodes must have room for CTL_MAXNAME
336  * elements and be NULL initialized.
337  */
338 static int
339 sysctl_search_oid(struct sysctl_oid **nodes, struct sysctl_oid *needle)
340 {
341 	int indx;
342 
343 	SYSCTL_ASSERT_LOCKED();
344 	indx = 0;
345 	while (indx < CTL_MAXNAME && indx >= 0) {
346 		if (nodes[indx] == NULL && indx == 0)
347 			nodes[indx] = SLIST_FIRST(&sysctl__children);
348 		else if (nodes[indx] == NULL)
349 			nodes[indx] = SLIST_FIRST(&nodes[indx - 1]->oid_children);
350 		else
351 			nodes[indx] = SLIST_NEXT(nodes[indx], oid_link);
352 
353 		if (nodes[indx] == needle)
354 			return (indx + 1);
355 
356 		if (nodes[indx] == NULL) {
357 			indx--;
358 			continue;
359 		}
360 
361 		if ((nodes[indx]->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
362 			indx++;
363 			continue;
364 		}
365 	}
366 	return (-1);
367 }
368 
369 static void
370 sysctl_warn_reuse(const char *func, struct sysctl_oid *leaf)
371 {
372 	struct sysctl_oid *nodes[CTL_MAXNAME];
373 	char buf[128];
374 	struct sbuf sb;
375 	int rc, i;
376 
377 	(void)sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN | SBUF_INCLUDENUL);
378 	sbuf_set_drain(&sb, sbuf_printf_drain, NULL);
379 
380 	sbuf_printf(&sb, "%s: can't re-use a leaf (", __func__);
381 
382 	memset(nodes, 0, sizeof(nodes));
383 	rc = sysctl_search_oid(nodes, leaf);
384 	if (rc > 0) {
385 		for (i = 0; i < rc; i++)
386 			sbuf_printf(&sb, "%s%.*s", nodes[i]->oid_name,
387 			    i != (rc - 1), ".");
388 	} else {
389 		sbuf_printf(&sb, "%s", leaf->oid_name);
390 	}
391 	sbuf_printf(&sb, ")!\n");
392 
393 	(void)sbuf_finish(&sb);
394 }
395 
396 #ifdef SYSCTL_DEBUG
397 static int
398 sysctl_reuse_test(SYSCTL_HANDLER_ARGS)
399 {
400 	struct rm_priotracker tracker;
401 
402 	SYSCTL_RLOCK(&tracker);
403 	sysctl_warn_reuse(__func__, oidp);
404 	SYSCTL_RUNLOCK(&tracker);
405 	return (0);
406 }
407 SYSCTL_PROC(_sysctl, 0, reuse_test, CTLTYPE_STRING|CTLFLAG_RD|CTLFLAG_MPSAFE,
408 	0, 0, sysctl_reuse_test, "-", "");
409 #endif
410 
411 void
412 sysctl_register_oid(struct sysctl_oid *oidp)
413 {
414 	struct sysctl_oid_list *parent = oidp->oid_parent;
415 	struct sysctl_oid *p;
416 	struct sysctl_oid *q;
417 	int oid_number;
418 	int timeout = 2;
419 
420 	/*
421 	 * First check if another oid with the same name already
422 	 * exists in the parent's list.
423 	 */
424 	SYSCTL_ASSERT_WLOCKED();
425 	p = sysctl_find_oidname(oidp->oid_name, parent);
426 	if (p != NULL) {
427 		if ((p->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
428 			p->oid_refcnt++;
429 			return;
430 		} else {
431 			sysctl_warn_reuse(__func__, p);
432 			return;
433 		}
434 	}
435 	/* get current OID number */
436 	oid_number = oidp->oid_number;
437 
438 #if (OID_AUTO >= 0)
439 #error "OID_AUTO is expected to be a negative value"
440 #endif
441 	/*
442 	 * Any negative OID number qualifies as OID_AUTO. Valid OID
443 	 * numbers should always be positive.
444 	 *
445 	 * NOTE: DO NOT change the starting value here, change it in
446 	 * <sys/sysctl.h>, and make sure it is at least 256 to
447 	 * accommodate e.g. net.inet.raw as a static sysctl node.
448 	 */
449 	if (oid_number < 0) {
450 		static int newoid;
451 
452 		/*
453 		 * By decrementing the next OID number we spend less
454 		 * time inserting the OIDs into a sorted list.
455 		 */
456 		if (--newoid < CTL_AUTO_START)
457 			newoid = 0x7fffffff;
458 
459 		oid_number = newoid;
460 	}
461 
462 	/*
463 	 * Insert the OID into the parent's list sorted by OID number.
464 	 */
465 retry:
466 	q = NULL;
467 	SLIST_FOREACH(p, parent, oid_link) {
468 		/* check if the current OID number is in use */
469 		if (oid_number == p->oid_number) {
470 			/* get the next valid OID number */
471 			if (oid_number < CTL_AUTO_START ||
472 			    oid_number == 0x7fffffff) {
473 				/* wraparound - restart */
474 				oid_number = CTL_AUTO_START;
475 				/* don't loop forever */
476 				if (!timeout--)
477 					panic("sysctl: Out of OID numbers\n");
478 				goto retry;
479 			} else {
480 				oid_number++;
481 			}
482 		} else if (oid_number < p->oid_number)
483 			break;
484 		q = p;
485 	}
486 	/* check for non-auto OID number collision */
487 	if (oidp->oid_number >= 0 && oidp->oid_number < CTL_AUTO_START &&
488 	    oid_number >= CTL_AUTO_START) {
489 		printf("sysctl: OID number(%d) is already in use for '%s'\n",
490 		    oidp->oid_number, oidp->oid_name);
491 	}
492 	/* update the OID number, if any */
493 	oidp->oid_number = oid_number;
494 	if (q != NULL)
495 		SLIST_INSERT_AFTER(q, oidp, oid_link);
496 	else
497 		SLIST_INSERT_HEAD(parent, oidp, oid_link);
498 
499 	if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE &&
500 #ifdef VIMAGE
501 	    (oidp->oid_kind & CTLFLAG_VNET) == 0 &&
502 #endif
503 	    (oidp->oid_kind & CTLFLAG_TUN) != 0 &&
504 	    (oidp->oid_kind & CTLFLAG_NOFETCH) == 0) {
505 		/* only fetch value once */
506 		oidp->oid_kind |= CTLFLAG_NOFETCH;
507 		/* try to fetch value from kernel environment */
508 		sysctl_load_tunable_by_oid_locked(oidp);
509 	}
510 }
511 
512 void
513 sysctl_register_disabled_oid(struct sysctl_oid *oidp)
514 {
515 
516 	/*
517 	 * Mark the leaf as dormant if it's not to be immediately enabled.
518 	 * We do not disable nodes as they can be shared between modules
519 	 * and it is always safe to access a node.
520 	 */
521 	KASSERT((oidp->oid_kind & CTLFLAG_DORMANT) == 0,
522 	    ("internal flag is set in oid_kind"));
523 	if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
524 		oidp->oid_kind |= CTLFLAG_DORMANT;
525 	sysctl_register_oid(oidp);
526 }
527 
528 void
529 sysctl_enable_oid(struct sysctl_oid *oidp)
530 {
531 
532 	SYSCTL_ASSERT_WLOCKED();
533 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
534 		KASSERT((oidp->oid_kind & CTLFLAG_DORMANT) == 0,
535 		    ("sysctl node is marked as dormant"));
536 		return;
537 	}
538 	KASSERT((oidp->oid_kind & CTLFLAG_DORMANT) != 0,
539 	    ("enabling already enabled sysctl oid"));
540 	oidp->oid_kind &= ~CTLFLAG_DORMANT;
541 }
542 
543 void
544 sysctl_unregister_oid(struct sysctl_oid *oidp)
545 {
546 	struct sysctl_oid *p;
547 	int error;
548 
549 	SYSCTL_ASSERT_WLOCKED();
550 	if (oidp->oid_number == OID_AUTO) {
551 		error = EINVAL;
552 	} else {
553 		error = ENOENT;
554 		SLIST_FOREACH(p, oidp->oid_parent, oid_link) {
555 			if (p == oidp) {
556 				SLIST_REMOVE(oidp->oid_parent, oidp,
557 				    sysctl_oid, oid_link);
558 				error = 0;
559 				break;
560 			}
561 		}
562 	}
563 
564 	/*
565 	 * This can happen when a module fails to register and is
566 	 * being unloaded afterwards.  It should not be a panic()
567 	 * for normal use.
568 	 */
569 	if (error) {
570 		printf("%s: failed(%d) to unregister sysctl(%s)\n",
571 		    __func__, error, oidp->oid_name);
572 	}
573 }
574 
575 /* Initialize a new context to keep track of dynamically added sysctls. */
576 int
577 sysctl_ctx_init(struct sysctl_ctx_list *c)
578 {
579 
580 	if (c == NULL) {
581 		return (EINVAL);
582 	}
583 
584 	/*
585 	 * No locking here, the caller is responsible for not adding
586 	 * new nodes to a context until after this function has
587 	 * returned.
588 	 */
589 	TAILQ_INIT(c);
590 	return (0);
591 }
592 
593 /* Free the context, and destroy all dynamic oids registered in this context */
594 int
595 sysctl_ctx_free(struct sysctl_ctx_list *clist)
596 {
597 	struct sysctl_ctx_entry *e, *e1;
598 	int error;
599 
600 	error = 0;
601 	/*
602 	 * First perform a "dry run" to check if it's ok to remove oids.
603 	 * XXX FIXME
604 	 * XXX This algorithm is a hack. But I don't know any
605 	 * XXX better solution for now...
606 	 */
607 	SYSCTL_WLOCK();
608 	TAILQ_FOREACH(e, clist, link) {
609 		error = sysctl_remove_oid_locked(e->entry, 0, 0);
610 		if (error)
611 			break;
612 	}
613 	/*
614 	 * Restore deregistered entries, either from the end,
615 	 * or from the place where error occurred.
616 	 * e contains the entry that was not unregistered
617 	 */
618 	if (error)
619 		e1 = TAILQ_PREV(e, sysctl_ctx_list, link);
620 	else
621 		e1 = TAILQ_LAST(clist, sysctl_ctx_list);
622 	while (e1 != NULL) {
623 		sysctl_register_oid(e1->entry);
624 		e1 = TAILQ_PREV(e1, sysctl_ctx_list, link);
625 	}
626 	if (error) {
627 		SYSCTL_WUNLOCK();
628 		return(EBUSY);
629 	}
630 	/* Now really delete the entries */
631 	e = TAILQ_FIRST(clist);
632 	while (e != NULL) {
633 		e1 = TAILQ_NEXT(e, link);
634 		error = sysctl_remove_oid_locked(e->entry, 1, 0);
635 		if (error)
636 			panic("sysctl_remove_oid: corrupt tree, entry: %s",
637 			    e->entry->oid_name);
638 		free(e, M_SYSCTLOID);
639 		e = e1;
640 	}
641 	SYSCTL_WUNLOCK();
642 	return (error);
643 }
644 
645 /* Add an entry to the context */
646 struct sysctl_ctx_entry *
647 sysctl_ctx_entry_add(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
648 {
649 	struct sysctl_ctx_entry *e;
650 
651 	SYSCTL_ASSERT_WLOCKED();
652 	if (clist == NULL || oidp == NULL)
653 		return(NULL);
654 	e = malloc(sizeof(struct sysctl_ctx_entry), M_SYSCTLOID, M_WAITOK);
655 	e->entry = oidp;
656 	TAILQ_INSERT_HEAD(clist, e, link);
657 	return (e);
658 }
659 
660 /* Find an entry in the context */
661 struct sysctl_ctx_entry *
662 sysctl_ctx_entry_find(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
663 {
664 	struct sysctl_ctx_entry *e;
665 
666 	SYSCTL_ASSERT_WLOCKED();
667 	if (clist == NULL || oidp == NULL)
668 		return(NULL);
669 	TAILQ_FOREACH(e, clist, link) {
670 		if(e->entry == oidp)
671 			return(e);
672 	}
673 	return (e);
674 }
675 
676 /*
677  * Delete an entry from the context.
678  * NOTE: this function doesn't free oidp! You have to remove it
679  * with sysctl_remove_oid().
680  */
681 int
682 sysctl_ctx_entry_del(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
683 {
684 	struct sysctl_ctx_entry *e;
685 
686 	if (clist == NULL || oidp == NULL)
687 		return (EINVAL);
688 	SYSCTL_WLOCK();
689 	e = sysctl_ctx_entry_find(clist, oidp);
690 	if (e != NULL) {
691 		TAILQ_REMOVE(clist, e, link);
692 		SYSCTL_WUNLOCK();
693 		free(e, M_SYSCTLOID);
694 		return (0);
695 	} else {
696 		SYSCTL_WUNLOCK();
697 		return (ENOENT);
698 	}
699 }
700 
701 /*
702  * Remove dynamically created sysctl trees.
703  * oidp - top of the tree to be removed
704  * del - if 0 - just deregister, otherwise free up entries as well
705  * recurse - if != 0 traverse the subtree to be deleted
706  */
707 int
708 sysctl_remove_oid(struct sysctl_oid *oidp, int del, int recurse)
709 {
710 	int error;
711 
712 	SYSCTL_WLOCK();
713 	error = sysctl_remove_oid_locked(oidp, del, recurse);
714 	SYSCTL_WUNLOCK();
715 	return (error);
716 }
717 
718 int
719 sysctl_remove_name(struct sysctl_oid *parent, const char *name,
720     int del, int recurse)
721 {
722 	struct sysctl_oid *p, *tmp;
723 	int error;
724 
725 	error = ENOENT;
726 	SYSCTL_WLOCK();
727 	SLIST_FOREACH_SAFE(p, SYSCTL_CHILDREN(parent), oid_link, tmp) {
728 		if (strcmp(p->oid_name, name) == 0) {
729 			error = sysctl_remove_oid_locked(p, del, recurse);
730 			break;
731 		}
732 	}
733 	SYSCTL_WUNLOCK();
734 
735 	return (error);
736 }
737 
738 
739 static int
740 sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del, int recurse)
741 {
742 	struct sysctl_oid *p, *tmp;
743 	int error;
744 
745 	SYSCTL_ASSERT_WLOCKED();
746 	if (oidp == NULL)
747 		return(EINVAL);
748 	if ((oidp->oid_kind & CTLFLAG_DYN) == 0) {
749 		printf("Warning: can't remove non-dynamic nodes (%s)!\n",
750 		    oidp->oid_name);
751 		return (EINVAL);
752 	}
753 	/*
754 	 * WARNING: normal method to do this should be through
755 	 * sysctl_ctx_free(). Use recursing as the last resort
756 	 * method to purge your sysctl tree of leftovers...
757 	 * However, if some other code still references these nodes,
758 	 * it will panic.
759 	 */
760 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
761 		if (oidp->oid_refcnt == 1) {
762 			SLIST_FOREACH_SAFE(p,
763 			    SYSCTL_CHILDREN(oidp), oid_link, tmp) {
764 				if (!recurse) {
765 					printf("Warning: failed attempt to "
766 					    "remove oid %s with child %s\n",
767 					    oidp->oid_name, p->oid_name);
768 					return (ENOTEMPTY);
769 				}
770 				error = sysctl_remove_oid_locked(p, del,
771 				    recurse);
772 				if (error)
773 					return (error);
774 			}
775 		}
776 	}
777 	if (oidp->oid_refcnt > 1 ) {
778 		oidp->oid_refcnt--;
779 	} else {
780 		if (oidp->oid_refcnt == 0) {
781 			printf("Warning: bad oid_refcnt=%u (%s)!\n",
782 				oidp->oid_refcnt, oidp->oid_name);
783 			return (EINVAL);
784 		}
785 		sysctl_unregister_oid(oidp);
786 		if (del) {
787 			/*
788 			 * Wait for all threads running the handler to drain.
789 			 * This preserves the previous behavior when the
790 			 * sysctl lock was held across a handler invocation,
791 			 * and is necessary for module unload correctness.
792 			 */
793 			while (oidp->oid_running > 0) {
794 				oidp->oid_kind |= CTLFLAG_DYING;
795 				SYSCTL_SLEEP(&oidp->oid_running, "oidrm", 0);
796 			}
797 			if (oidp->oid_descr)
798 				free(__DECONST(char *, oidp->oid_descr),
799 				    M_SYSCTLOID);
800 			if (oidp->oid_label)
801 				free(__DECONST(char *, oidp->oid_label),
802 				    M_SYSCTLOID);
803 			free(__DECONST(char *, oidp->oid_name), M_SYSCTLOID);
804 			free(oidp, M_SYSCTLOID);
805 		}
806 	}
807 	return (0);
808 }
809 /*
810  * Create new sysctls at run time.
811  * clist may point to a valid context initialized with sysctl_ctx_init().
812  */
813 struct sysctl_oid *
814 sysctl_add_oid(struct sysctl_ctx_list *clist, struct sysctl_oid_list *parent,
815 	int number, const char *name, int kind, void *arg1, intmax_t arg2,
816 	int (*handler)(SYSCTL_HANDLER_ARGS), const char *fmt, const char *descr,
817 	const char *label)
818 {
819 	struct sysctl_oid *oidp;
820 
821 	/* You have to hook up somewhere.. */
822 	if (parent == NULL)
823 		return(NULL);
824 	/* Check if the node already exists, otherwise create it */
825 	SYSCTL_WLOCK();
826 	oidp = sysctl_find_oidname(name, parent);
827 	if (oidp != NULL) {
828 		if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
829 			oidp->oid_refcnt++;
830 			/* Update the context */
831 			if (clist != NULL)
832 				sysctl_ctx_entry_add(clist, oidp);
833 			SYSCTL_WUNLOCK();
834 			return (oidp);
835 		} else {
836 			sysctl_warn_reuse(__func__, oidp);
837 			SYSCTL_WUNLOCK();
838 			return (NULL);
839 		}
840 	}
841 	oidp = malloc(sizeof(struct sysctl_oid), M_SYSCTLOID, M_WAITOK|M_ZERO);
842 	oidp->oid_parent = parent;
843 	SLIST_INIT(&oidp->oid_children);
844 	oidp->oid_number = number;
845 	oidp->oid_refcnt = 1;
846 	oidp->oid_name = strdup(name, M_SYSCTLOID);
847 	oidp->oid_handler = handler;
848 	oidp->oid_kind = CTLFLAG_DYN | kind;
849 	oidp->oid_arg1 = arg1;
850 	oidp->oid_arg2 = arg2;
851 	oidp->oid_fmt = fmt;
852 	if (descr != NULL)
853 		oidp->oid_descr = strdup(descr, M_SYSCTLOID);
854 	if (label != NULL)
855 		oidp->oid_label = strdup(label, M_SYSCTLOID);
856 	/* Update the context, if used */
857 	if (clist != NULL)
858 		sysctl_ctx_entry_add(clist, oidp);
859 	/* Register this oid */
860 	sysctl_register_oid(oidp);
861 	SYSCTL_WUNLOCK();
862 	return (oidp);
863 }
864 
865 /*
866  * Rename an existing oid.
867  */
868 void
869 sysctl_rename_oid(struct sysctl_oid *oidp, const char *name)
870 {
871 	char *newname;
872 	char *oldname;
873 
874 	newname = strdup(name, M_SYSCTLOID);
875 	SYSCTL_WLOCK();
876 	oldname = __DECONST(char *, oidp->oid_name);
877 	oidp->oid_name = newname;
878 	SYSCTL_WUNLOCK();
879 	free(oldname, M_SYSCTLOID);
880 }
881 
882 /*
883  * Reparent an existing oid.
884  */
885 int
886 sysctl_move_oid(struct sysctl_oid *oid, struct sysctl_oid_list *parent)
887 {
888 	struct sysctl_oid *oidp;
889 
890 	SYSCTL_WLOCK();
891 	if (oid->oid_parent == parent) {
892 		SYSCTL_WUNLOCK();
893 		return (0);
894 	}
895 	oidp = sysctl_find_oidname(oid->oid_name, parent);
896 	if (oidp != NULL) {
897 		SYSCTL_WUNLOCK();
898 		return (EEXIST);
899 	}
900 	sysctl_unregister_oid(oid);
901 	oid->oid_parent = parent;
902 	oid->oid_number = OID_AUTO;
903 	sysctl_register_oid(oid);
904 	SYSCTL_WUNLOCK();
905 	return (0);
906 }
907 
908 /*
909  * Register the kernel's oids on startup.
910  */
911 SET_DECLARE(sysctl_set, struct sysctl_oid);
912 
913 static void
914 sysctl_register_all(void *arg)
915 {
916 	struct sysctl_oid **oidp;
917 
918 	sx_init(&sysctlmemlock, "sysctl mem");
919 	SYSCTL_INIT();
920 	SYSCTL_WLOCK();
921 	SET_FOREACH(oidp, sysctl_set)
922 		sysctl_register_oid(*oidp);
923 	SYSCTL_WUNLOCK();
924 }
925 SYSINIT(sysctl, SI_SUB_KMEM, SI_ORDER_FIRST, sysctl_register_all, NULL);
926 
927 /*
928  * "Staff-functions"
929  *
930  * These functions implement a presently undocumented interface
931  * used by the sysctl program to walk the tree, and get the type
932  * so it can print the value.
933  * This interface is under work and consideration, and should probably
934  * be killed with a big axe by the first person who can find the time.
935  * (be aware though, that the proper interface isn't as obvious as it
936  * may seem, there are various conflicting requirements.
937  *
938  * {CTL_SYSCTL, CTL_SYSCTL_DEBUG}		printf the entire MIB-tree.
939  * {CTL_SYSCTL, CTL_SYSCTL_NAME, ...}		return the name of the "..."
940  *						OID.
941  * {CTL_SYSCTL, CTL_SYSCTL_NEXT, ...}		return the next OID.
942  * {CTL_SYSCTL, CTL_SYSCTL_NAME2OID}		return the OID of the name in
943  *						"new"
944  * {CTL_SYSCTL, CTL_SYSCTL_OIDFMT, ...}		return the kind & format info
945  *						for the "..." OID.
946  * {CTL_SYSCTL, CTL_SYSCTL_OIDDESCR, ...}	return the description of the
947  *						"..." OID.
948  * {CTL_SYSCTL, CTL_SYSCTL_OIDLABEL, ...}	return the aggregation label of
949  *						the "..." OID.
950  */
951 
952 #ifdef SYSCTL_DEBUG
953 static void
954 sysctl_sysctl_debug_dump_node(struct sysctl_oid_list *l, int i)
955 {
956 	int k;
957 	struct sysctl_oid *oidp;
958 
959 	SYSCTL_ASSERT_LOCKED();
960 	SLIST_FOREACH(oidp, l, oid_link) {
961 
962 		for (k=0; k<i; k++)
963 			printf(" ");
964 
965 		printf("%d %s ", oidp->oid_number, oidp->oid_name);
966 
967 		printf("%c%c",
968 			oidp->oid_kind & CTLFLAG_RD ? 'R':' ',
969 			oidp->oid_kind & CTLFLAG_WR ? 'W':' ');
970 
971 		if (oidp->oid_handler)
972 			printf(" *Handler");
973 
974 		switch (oidp->oid_kind & CTLTYPE) {
975 			case CTLTYPE_NODE:
976 				printf(" Node\n");
977 				if (!oidp->oid_handler) {
978 					sysctl_sysctl_debug_dump_node(
979 					    SYSCTL_CHILDREN(oidp), i + 2);
980 				}
981 				break;
982 			case CTLTYPE_INT:    printf(" Int\n"); break;
983 			case CTLTYPE_UINT:   printf(" u_int\n"); break;
984 			case CTLTYPE_LONG:   printf(" Long\n"); break;
985 			case CTLTYPE_ULONG:  printf(" u_long\n"); break;
986 			case CTLTYPE_STRING: printf(" String\n"); break;
987 			case CTLTYPE_S8:     printf(" int8_t\n"); break;
988 			case CTLTYPE_S16:    printf(" int16_t\n"); break;
989 			case CTLTYPE_S32:    printf(" int32_t\n"); break;
990 			case CTLTYPE_S64:    printf(" int64_t\n"); break;
991 			case CTLTYPE_U8:     printf(" uint8_t\n"); break;
992 			case CTLTYPE_U16:    printf(" uint16_t\n"); break;
993 			case CTLTYPE_U32:    printf(" uint32_t\n"); break;
994 			case CTLTYPE_U64:    printf(" uint64_t\n"); break;
995 			case CTLTYPE_OPAQUE: printf(" Opaque/struct\n"); break;
996 			default:	     printf("\n");
997 		}
998 
999 	}
1000 }
1001 
1002 static int
1003 sysctl_sysctl_debug(SYSCTL_HANDLER_ARGS)
1004 {
1005 	struct rm_priotracker tracker;
1006 	int error;
1007 
1008 	error = priv_check(req->td, PRIV_SYSCTL_DEBUG);
1009 	if (error)
1010 		return (error);
1011 	SYSCTL_RLOCK(&tracker);
1012 	sysctl_sysctl_debug_dump_node(&sysctl__children, 0);
1013 	SYSCTL_RUNLOCK(&tracker);
1014 	return (ENOENT);
1015 }
1016 
1017 SYSCTL_PROC(_sysctl, CTL_SYSCTL_DEBUG, debug, CTLTYPE_STRING | CTLFLAG_RD |
1018     CTLFLAG_MPSAFE, 0, 0, sysctl_sysctl_debug, "-", "");
1019 #endif
1020 
1021 static int
1022 sysctl_sysctl_name(SYSCTL_HANDLER_ARGS)
1023 {
1024 	int *name = (int *) arg1;
1025 	u_int namelen = arg2;
1026 	int error;
1027 	struct sysctl_oid *oid;
1028 	struct sysctl_oid_list *lsp = &sysctl__children, *lsp2;
1029 	struct rm_priotracker tracker;
1030 	char buf[10];
1031 
1032 	error = sysctl_wire_old_buffer(req, 0);
1033 	if (error)
1034 		return (error);
1035 
1036 	SYSCTL_RLOCK(&tracker);
1037 	while (namelen) {
1038 		if (!lsp) {
1039 			snprintf(buf,sizeof(buf),"%d",*name);
1040 			if (req->oldidx)
1041 				error = SYSCTL_OUT(req, ".", 1);
1042 			if (!error)
1043 				error = SYSCTL_OUT(req, buf, strlen(buf));
1044 			if (error)
1045 				goto out;
1046 			namelen--;
1047 			name++;
1048 			continue;
1049 		}
1050 		lsp2 = NULL;
1051 		SLIST_FOREACH(oid, lsp, oid_link) {
1052 			if (oid->oid_number != *name)
1053 				continue;
1054 
1055 			if (req->oldidx)
1056 				error = SYSCTL_OUT(req, ".", 1);
1057 			if (!error)
1058 				error = SYSCTL_OUT(req, oid->oid_name,
1059 					strlen(oid->oid_name));
1060 			if (error)
1061 				goto out;
1062 
1063 			namelen--;
1064 			name++;
1065 
1066 			if ((oid->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1067 				break;
1068 
1069 			if (oid->oid_handler)
1070 				break;
1071 
1072 			lsp2 = SYSCTL_CHILDREN(oid);
1073 			break;
1074 		}
1075 		lsp = lsp2;
1076 	}
1077 	error = SYSCTL_OUT(req, "", 1);
1078  out:
1079 	SYSCTL_RUNLOCK(&tracker);
1080 	return (error);
1081 }
1082 
1083 /*
1084  * XXXRW/JA: Shouldn't return name data for nodes that we don't permit in
1085  * capability mode.
1086  */
1087 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_NAME, name, CTLFLAG_RD |
1088     CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_name, "");
1089 
1090 static int
1091 sysctl_sysctl_next_ls(struct sysctl_oid_list *lsp, int *name, u_int namelen,
1092 	int *next, int *len, int level, struct sysctl_oid **oidpp)
1093 {
1094 	struct sysctl_oid *oidp;
1095 
1096 	SYSCTL_ASSERT_LOCKED();
1097 	*len = level;
1098 	SLIST_FOREACH(oidp, lsp, oid_link) {
1099 		*next = oidp->oid_number;
1100 		*oidpp = oidp;
1101 
1102 		if ((oidp->oid_kind & (CTLFLAG_SKIP | CTLFLAG_DORMANT)) != 0)
1103 			continue;
1104 
1105 		if (!namelen) {
1106 			if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1107 				return (0);
1108 			if (oidp->oid_handler)
1109 				/* We really should call the handler here...*/
1110 				return (0);
1111 			lsp = SYSCTL_CHILDREN(oidp);
1112 			if (!sysctl_sysctl_next_ls(lsp, 0, 0, next+1,
1113 				len, level+1, oidpp))
1114 				return (0);
1115 			goto emptynode;
1116 		}
1117 
1118 		if (oidp->oid_number < *name)
1119 			continue;
1120 
1121 		if (oidp->oid_number > *name) {
1122 			if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1123 				return (0);
1124 			if (oidp->oid_handler)
1125 				return (0);
1126 			lsp = SYSCTL_CHILDREN(oidp);
1127 			if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1,
1128 				next+1, len, level+1, oidpp))
1129 				return (0);
1130 			goto next;
1131 		}
1132 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1133 			continue;
1134 
1135 		if (oidp->oid_handler)
1136 			continue;
1137 
1138 		lsp = SYSCTL_CHILDREN(oidp);
1139 		if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1, next+1,
1140 			len, level+1, oidpp))
1141 			return (0);
1142 	next:
1143 		namelen = 1;
1144 	emptynode:
1145 		*len = level;
1146 	}
1147 	return (1);
1148 }
1149 
1150 static int
1151 sysctl_sysctl_next(SYSCTL_HANDLER_ARGS)
1152 {
1153 	int *name = (int *) arg1;
1154 	u_int namelen = arg2;
1155 	int i, j, error;
1156 	struct sysctl_oid *oid;
1157 	struct sysctl_oid_list *lsp = &sysctl__children;
1158 	struct rm_priotracker tracker;
1159 	int newoid[CTL_MAXNAME];
1160 
1161 	SYSCTL_RLOCK(&tracker);
1162 	i = sysctl_sysctl_next_ls(lsp, name, namelen, newoid, &j, 1, &oid);
1163 	SYSCTL_RUNLOCK(&tracker);
1164 	if (i)
1165 		return (ENOENT);
1166 	error = SYSCTL_OUT(req, newoid, j * sizeof (int));
1167 	return (error);
1168 }
1169 
1170 /*
1171  * XXXRW/JA: Shouldn't return next data for nodes that we don't permit in
1172  * capability mode.
1173  */
1174 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_NEXT, next, CTLFLAG_RD |
1175     CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_next, "");
1176 
1177 static int
1178 name2oid(char *name, int *oid, int *len, struct sysctl_oid **oidpp)
1179 {
1180 	struct sysctl_oid *oidp;
1181 	struct sysctl_oid_list *lsp = &sysctl__children;
1182 	char *p;
1183 
1184 	SYSCTL_ASSERT_LOCKED();
1185 
1186 	for (*len = 0; *len < CTL_MAXNAME;) {
1187 		p = strsep(&name, ".");
1188 
1189 		oidp = SLIST_FIRST(lsp);
1190 		for (;; oidp = SLIST_NEXT(oidp, oid_link)) {
1191 			if (oidp == NULL)
1192 				return (ENOENT);
1193 			if (strcmp(p, oidp->oid_name) == 0)
1194 				break;
1195 		}
1196 		*oid++ = oidp->oid_number;
1197 		(*len)++;
1198 
1199 		if (name == NULL || *name == '\0') {
1200 			if (oidpp)
1201 				*oidpp = oidp;
1202 			return (0);
1203 		}
1204 
1205 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1206 			break;
1207 
1208 		if (oidp->oid_handler)
1209 			break;
1210 
1211 		lsp = SYSCTL_CHILDREN(oidp);
1212 	}
1213 	return (ENOENT);
1214 }
1215 
1216 static int
1217 sysctl_sysctl_name2oid(SYSCTL_HANDLER_ARGS)
1218 {
1219 	char *p;
1220 	int error, oid[CTL_MAXNAME], len = 0;
1221 	struct sysctl_oid *op = NULL;
1222 	struct rm_priotracker tracker;
1223 	char buf[32];
1224 
1225 	if (!req->newlen)
1226 		return (ENOENT);
1227 	if (req->newlen >= MAXPATHLEN)	/* XXX arbitrary, undocumented */
1228 		return (ENAMETOOLONG);
1229 
1230 	p = buf;
1231 	if (req->newlen >= sizeof(buf))
1232 		p = malloc(req->newlen+1, M_SYSCTL, M_WAITOK);
1233 
1234 	error = SYSCTL_IN(req, p, req->newlen);
1235 	if (error) {
1236 		if (p != buf)
1237 			free(p, M_SYSCTL);
1238 		return (error);
1239 	}
1240 
1241 	p [req->newlen] = '\0';
1242 
1243 	SYSCTL_RLOCK(&tracker);
1244 	error = name2oid(p, oid, &len, &op);
1245 	SYSCTL_RUNLOCK(&tracker);
1246 
1247 	if (p != buf)
1248 		free(p, M_SYSCTL);
1249 
1250 	if (error)
1251 		return (error);
1252 
1253 	error = SYSCTL_OUT(req, oid, len * sizeof *oid);
1254 	return (error);
1255 }
1256 
1257 /*
1258  * XXXRW/JA: Shouldn't return name2oid data for nodes that we don't permit in
1259  * capability mode.
1260  */
1261 SYSCTL_PROC(_sysctl, CTL_SYSCTL_NAME2OID, name2oid, CTLTYPE_INT | CTLFLAG_RW |
1262     CTLFLAG_ANYBODY | CTLFLAG_MPSAFE | CTLFLAG_CAPRW, 0, 0,
1263     sysctl_sysctl_name2oid, "I", "");
1264 
1265 static int
1266 sysctl_sysctl_oidfmt(SYSCTL_HANDLER_ARGS)
1267 {
1268 	struct sysctl_oid *oid;
1269 	struct rm_priotracker tracker;
1270 	int error;
1271 
1272 	error = sysctl_wire_old_buffer(req, 0);
1273 	if (error)
1274 		return (error);
1275 
1276 	SYSCTL_RLOCK(&tracker);
1277 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1278 	if (error)
1279 		goto out;
1280 
1281 	if (oid->oid_fmt == NULL) {
1282 		error = ENOENT;
1283 		goto out;
1284 	}
1285 	error = SYSCTL_OUT(req, &oid->oid_kind, sizeof(oid->oid_kind));
1286 	if (error)
1287 		goto out;
1288 	error = SYSCTL_OUT(req, oid->oid_fmt, strlen(oid->oid_fmt) + 1);
1289  out:
1290 	SYSCTL_RUNLOCK(&tracker);
1291 	return (error);
1292 }
1293 
1294 
1295 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_OIDFMT, oidfmt, CTLFLAG_RD |
1296     CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_oidfmt, "");
1297 
1298 static int
1299 sysctl_sysctl_oiddescr(SYSCTL_HANDLER_ARGS)
1300 {
1301 	struct sysctl_oid *oid;
1302 	struct rm_priotracker tracker;
1303 	int error;
1304 
1305 	error = sysctl_wire_old_buffer(req, 0);
1306 	if (error)
1307 		return (error);
1308 
1309 	SYSCTL_RLOCK(&tracker);
1310 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1311 	if (error)
1312 		goto out;
1313 
1314 	if (oid->oid_descr == NULL) {
1315 		error = ENOENT;
1316 		goto out;
1317 	}
1318 	error = SYSCTL_OUT(req, oid->oid_descr, strlen(oid->oid_descr) + 1);
1319  out:
1320 	SYSCTL_RUNLOCK(&tracker);
1321 	return (error);
1322 }
1323 
1324 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_OIDDESCR, oiddescr, CTLFLAG_RD |
1325     CTLFLAG_MPSAFE|CTLFLAG_CAPRD, sysctl_sysctl_oiddescr, "");
1326 
1327 static int
1328 sysctl_sysctl_oidlabel(SYSCTL_HANDLER_ARGS)
1329 {
1330 	struct sysctl_oid *oid;
1331 	struct rm_priotracker tracker;
1332 	int error;
1333 
1334 	error = sysctl_wire_old_buffer(req, 0);
1335 	if (error)
1336 		return (error);
1337 
1338 	SYSCTL_RLOCK(&tracker);
1339 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1340 	if (error)
1341 		goto out;
1342 
1343 	if (oid->oid_label == NULL) {
1344 		error = ENOENT;
1345 		goto out;
1346 	}
1347 	error = SYSCTL_OUT(req, oid->oid_label, strlen(oid->oid_label) + 1);
1348  out:
1349 	SYSCTL_RUNLOCK(&tracker);
1350 	return (error);
1351 }
1352 
1353 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_OIDLABEL, oidlabel, CTLFLAG_RD |
1354     CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_oidlabel, "");
1355 
1356 /*
1357  * Default "handler" functions.
1358  */
1359 
1360 /*
1361  * Handle a bool.
1362  * Two cases:
1363  *     a variable:  point arg1 at it.
1364  *     a constant:  pass it in arg2.
1365  */
1366 
1367 int
1368 sysctl_handle_bool(SYSCTL_HANDLER_ARGS)
1369 {
1370 	uint8_t temp;
1371 	int error;
1372 
1373 	/*
1374 	 * Attempt to get a coherent snapshot by making a copy of the data.
1375 	 */
1376 	if (arg1)
1377 		temp = *(bool *)arg1 ? 1 : 0;
1378 	else
1379 		temp = arg2 ? 1 : 0;
1380 
1381 	error = SYSCTL_OUT(req, &temp, sizeof(temp));
1382 	if (error || !req->newptr)
1383 		return (error);
1384 
1385 	if (!arg1)
1386 		error = EPERM;
1387 	else {
1388 		error = SYSCTL_IN(req, &temp, sizeof(temp));
1389 		if (!error)
1390 			*(bool *)arg1 = temp ? 1 : 0;
1391 	}
1392 	return (error);
1393 }
1394 
1395 /*
1396  * Handle an int8_t, signed or unsigned.
1397  * Two cases:
1398  *     a variable:  point arg1 at it.
1399  *     a constant:  pass it in arg2.
1400  */
1401 
1402 int
1403 sysctl_handle_8(SYSCTL_HANDLER_ARGS)
1404 {
1405 	int8_t tmpout;
1406 	int error = 0;
1407 
1408 	/*
1409 	 * Attempt to get a coherent snapshot by making a copy of the data.
1410 	 */
1411 	if (arg1)
1412 		tmpout = *(int8_t *)arg1;
1413 	else
1414 		tmpout = arg2;
1415 	error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout));
1416 
1417 	if (error || !req->newptr)
1418 		return (error);
1419 
1420 	if (!arg1)
1421 		error = EPERM;
1422 	else
1423 		error = SYSCTL_IN(req, arg1, sizeof(tmpout));
1424 	return (error);
1425 }
1426 
1427 /*
1428  * Handle an int16_t, signed or unsigned.
1429  * Two cases:
1430  *     a variable:  point arg1 at it.
1431  *     a constant:  pass it in arg2.
1432  */
1433 
1434 int
1435 sysctl_handle_16(SYSCTL_HANDLER_ARGS)
1436 {
1437 	int16_t tmpout;
1438 	int error = 0;
1439 
1440 	/*
1441 	 * Attempt to get a coherent snapshot by making a copy of the data.
1442 	 */
1443 	if (arg1)
1444 		tmpout = *(int16_t *)arg1;
1445 	else
1446 		tmpout = arg2;
1447 	error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout));
1448 
1449 	if (error || !req->newptr)
1450 		return (error);
1451 
1452 	if (!arg1)
1453 		error = EPERM;
1454 	else
1455 		error = SYSCTL_IN(req, arg1, sizeof(tmpout));
1456 	return (error);
1457 }
1458 
1459 /*
1460  * Handle an int32_t, signed or unsigned.
1461  * Two cases:
1462  *     a variable:  point arg1 at it.
1463  *     a constant:  pass it in arg2.
1464  */
1465 
1466 int
1467 sysctl_handle_32(SYSCTL_HANDLER_ARGS)
1468 {
1469 	int32_t tmpout;
1470 	int error = 0;
1471 
1472 	/*
1473 	 * Attempt to get a coherent snapshot by making a copy of the data.
1474 	 */
1475 	if (arg1)
1476 		tmpout = *(int32_t *)arg1;
1477 	else
1478 		tmpout = arg2;
1479 	error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout));
1480 
1481 	if (error || !req->newptr)
1482 		return (error);
1483 
1484 	if (!arg1)
1485 		error = EPERM;
1486 	else
1487 		error = SYSCTL_IN(req, arg1, sizeof(tmpout));
1488 	return (error);
1489 }
1490 
1491 /*
1492  * Handle an int, signed or unsigned.
1493  * Two cases:
1494  *     a variable:  point arg1 at it.
1495  *     a constant:  pass it in arg2.
1496  */
1497 
1498 int
1499 sysctl_handle_int(SYSCTL_HANDLER_ARGS)
1500 {
1501 	int tmpout, error = 0;
1502 
1503 	/*
1504 	 * Attempt to get a coherent snapshot by making a copy of the data.
1505 	 */
1506 	if (arg1)
1507 		tmpout = *(int *)arg1;
1508 	else
1509 		tmpout = arg2;
1510 	error = SYSCTL_OUT(req, &tmpout, sizeof(int));
1511 
1512 	if (error || !req->newptr)
1513 		return (error);
1514 
1515 	if (!arg1)
1516 		error = EPERM;
1517 	else
1518 		error = SYSCTL_IN(req, arg1, sizeof(int));
1519 	return (error);
1520 }
1521 
1522 /*
1523  * Based on on sysctl_handle_int() convert milliseconds into ticks.
1524  * Note: this is used by TCP.
1525  */
1526 
1527 int
1528 sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
1529 {
1530 	int error, s, tt;
1531 
1532 	tt = *(int *)arg1;
1533 	s = (int)((int64_t)tt * 1000 / hz);
1534 
1535 	error = sysctl_handle_int(oidp, &s, 0, req);
1536 	if (error || !req->newptr)
1537 		return (error);
1538 
1539 	tt = (int)((int64_t)s * hz / 1000);
1540 	if (tt < 1)
1541 		return (EINVAL);
1542 
1543 	*(int *)arg1 = tt;
1544 	return (0);
1545 }
1546 
1547 
1548 /*
1549  * Handle a long, signed or unsigned.
1550  * Two cases:
1551  *     a variable:  point arg1 at it.
1552  *     a constant:  pass it in arg2.
1553  */
1554 
1555 int
1556 sysctl_handle_long(SYSCTL_HANDLER_ARGS)
1557 {
1558 	int error = 0;
1559 	long tmplong;
1560 #ifdef SCTL_MASK32
1561 	int tmpint;
1562 #endif
1563 
1564 	/*
1565 	 * Attempt to get a coherent snapshot by making a copy of the data.
1566 	 */
1567 	if (arg1)
1568 		tmplong = *(long *)arg1;
1569 	else
1570 		tmplong = arg2;
1571 #ifdef SCTL_MASK32
1572 	if (req->flags & SCTL_MASK32) {
1573 		tmpint = tmplong;
1574 		error = SYSCTL_OUT(req, &tmpint, sizeof(int));
1575 	} else
1576 #endif
1577 		error = SYSCTL_OUT(req, &tmplong, sizeof(long));
1578 
1579 	if (error || !req->newptr)
1580 		return (error);
1581 
1582 	if (!arg1)
1583 		error = EPERM;
1584 #ifdef SCTL_MASK32
1585 	else if (req->flags & SCTL_MASK32) {
1586 		error = SYSCTL_IN(req, &tmpint, sizeof(int));
1587 		*(long *)arg1 = (long)tmpint;
1588 	}
1589 #endif
1590 	else
1591 		error = SYSCTL_IN(req, arg1, sizeof(long));
1592 	return (error);
1593 }
1594 
1595 /*
1596  * Handle a 64 bit int, signed or unsigned.
1597  * Two cases:
1598  *     a variable:  point arg1 at it.
1599  *     a constant:  pass it in arg2.
1600  */
1601 int
1602 sysctl_handle_64(SYSCTL_HANDLER_ARGS)
1603 {
1604 	int error = 0;
1605 	uint64_t tmpout;
1606 
1607 	/*
1608 	 * Attempt to get a coherent snapshot by making a copy of the data.
1609 	 */
1610 	if (arg1)
1611 		tmpout = *(uint64_t *)arg1;
1612 	else
1613 		tmpout = arg2;
1614 	error = SYSCTL_OUT(req, &tmpout, sizeof(uint64_t));
1615 
1616 	if (error || !req->newptr)
1617 		return (error);
1618 
1619 	if (!arg1)
1620 		error = EPERM;
1621 	else
1622 		error = SYSCTL_IN(req, arg1, sizeof(uint64_t));
1623 	return (error);
1624 }
1625 
1626 /*
1627  * Handle our generic '\0' terminated 'C' string.
1628  * Two cases:
1629  * 	a variable string:  point arg1 at it, arg2 is max length.
1630  * 	a constant string:  point arg1 at it, arg2 is zero.
1631  */
1632 
1633 int
1634 sysctl_handle_string(SYSCTL_HANDLER_ARGS)
1635 {
1636 	size_t outlen;
1637 	int error = 0, ro_string = 0;
1638 
1639 	/*
1640 	 * A zero-length buffer indicates a fixed size read-only
1641 	 * string.  In ddb, don't worry about trying to make a malloced
1642 	 * snapshot.
1643 	 */
1644 	if (arg2 == 0 || kdb_active) {
1645 		arg2 = strlen((char *)arg1) + 1;
1646 		ro_string = 1;
1647 	}
1648 
1649 	if (req->oldptr != NULL) {
1650 		char *tmparg;
1651 
1652 		if (ro_string) {
1653 			tmparg = arg1;
1654 		} else {
1655 			/* try to make a coherent snapshot of the string */
1656 			tmparg = malloc(arg2, M_SYSCTLTMP, M_WAITOK);
1657 			memcpy(tmparg, arg1, arg2);
1658 		}
1659 
1660 		outlen = strnlen(tmparg, arg2 - 1) + 1;
1661 		error = SYSCTL_OUT(req, tmparg, outlen);
1662 
1663 		if (!ro_string)
1664 			free(tmparg, M_SYSCTLTMP);
1665 	} else {
1666 		outlen = strnlen((char *)arg1, arg2 - 1) + 1;
1667 		error = SYSCTL_OUT(req, NULL, outlen);
1668 	}
1669 	if (error || !req->newptr)
1670 		return (error);
1671 
1672 	if ((req->newlen - req->newidx) >= arg2) {
1673 		error = EINVAL;
1674 	} else {
1675 		arg2 = (req->newlen - req->newidx);
1676 		error = SYSCTL_IN(req, arg1, arg2);
1677 		((char *)arg1)[arg2] = '\0';
1678 	}
1679 	return (error);
1680 }
1681 
1682 /*
1683  * Handle any kind of opaque data.
1684  * arg1 points to it, arg2 is the size.
1685  */
1686 
1687 int
1688 sysctl_handle_opaque(SYSCTL_HANDLER_ARGS)
1689 {
1690 	int error, tries;
1691 	u_int generation;
1692 	struct sysctl_req req2;
1693 
1694 	/*
1695 	 * Attempt to get a coherent snapshot, by using the thread
1696 	 * pre-emption counter updated from within mi_switch() to
1697 	 * determine if we were pre-empted during a bcopy() or
1698 	 * copyout(). Make 3 attempts at doing this before giving up.
1699 	 * If we encounter an error, stop immediately.
1700 	 */
1701 	tries = 0;
1702 	req2 = *req;
1703 retry:
1704 	generation = curthread->td_generation;
1705 	error = SYSCTL_OUT(req, arg1, arg2);
1706 	if (error)
1707 		return (error);
1708 	tries++;
1709 	if (generation != curthread->td_generation && tries < 3) {
1710 		*req = req2;
1711 		goto retry;
1712 	}
1713 
1714 	error = SYSCTL_IN(req, arg1, arg2);
1715 
1716 	return (error);
1717 }
1718 
1719 /*
1720  * Based on on sysctl_handle_int() convert microseconds to a sbintime.
1721  */
1722 int
1723 sysctl_usec_to_sbintime(SYSCTL_HANDLER_ARGS)
1724 {
1725 	int error;
1726 	int64_t tt;
1727 	sbintime_t sb;
1728 
1729 	tt = *(int64_t *)arg1;
1730 	sb = sbttous(tt);
1731 
1732 	error = sysctl_handle_64(oidp, &sb, 0, req);
1733 	if (error || !req->newptr)
1734 		return (error);
1735 
1736 	tt = ustosbt(sb);
1737 	*(int64_t *)arg1 = tt;
1738 
1739 	return (0);
1740 }
1741 
1742 /*
1743  * Based on on sysctl_handle_int() convert milliseconds to a sbintime.
1744  */
1745 int
1746 sysctl_msec_to_sbintime(SYSCTL_HANDLER_ARGS)
1747 {
1748 	int error;
1749 	int64_t tt;
1750 	sbintime_t sb;
1751 
1752 	tt = *(int64_t *)arg1;
1753 	sb = sbttoms(tt);
1754 
1755 	error = sysctl_handle_64(oidp, &sb, 0, req);
1756 	if (error || !req->newptr)
1757 		return (error);
1758 
1759 	tt = mstosbt(sb);
1760 	*(int64_t *)arg1 = tt;
1761 
1762 	return (0);
1763 }
1764 
1765 /*
1766  * Convert seconds to a struct timeval.  Intended for use with
1767  * intervals and thus does not permit negative seconds.
1768  */
1769 int
1770 sysctl_sec_to_timeval(SYSCTL_HANDLER_ARGS)
1771 {
1772 	struct timeval *tv;
1773 	int error, secs;
1774 
1775 	tv = arg1;
1776 	secs = tv->tv_sec;
1777 
1778 	error = sysctl_handle_int(oidp, &secs, 0, req);
1779 	if (error || req->newptr == NULL)
1780 		return (error);
1781 
1782 	if (secs < 0)
1783 		return (EINVAL);
1784 	tv->tv_sec = secs;
1785 
1786 	return (0);
1787 }
1788 
1789 /*
1790  * Transfer functions to/from kernel space.
1791  * XXX: rather untested at this point
1792  */
1793 static int
1794 sysctl_old_kernel(struct sysctl_req *req, const void *p, size_t l)
1795 {
1796 	size_t i = 0;
1797 
1798 	if (req->oldptr) {
1799 		i = l;
1800 		if (req->oldlen <= req->oldidx)
1801 			i = 0;
1802 		else
1803 			if (i > req->oldlen - req->oldidx)
1804 				i = req->oldlen - req->oldidx;
1805 		if (i > 0)
1806 			bcopy(p, (char *)req->oldptr + req->oldidx, i);
1807 	}
1808 	req->oldidx += l;
1809 	if (req->oldptr && i != l)
1810 		return (ENOMEM);
1811 	return (0);
1812 }
1813 
1814 static int
1815 sysctl_new_kernel(struct sysctl_req *req, void *p, size_t l)
1816 {
1817 	if (!req->newptr)
1818 		return (0);
1819 	if (req->newlen - req->newidx < l)
1820 		return (EINVAL);
1821 	bcopy((const char *)req->newptr + req->newidx, p, l);
1822 	req->newidx += l;
1823 	return (0);
1824 }
1825 
1826 int
1827 kernel_sysctl(struct thread *td, int *name, u_int namelen, void *old,
1828     size_t *oldlenp, void *new, size_t newlen, size_t *retval, int flags)
1829 {
1830 	int error = 0;
1831 	struct sysctl_req req;
1832 
1833 	bzero(&req, sizeof req);
1834 
1835 	req.td = td;
1836 	req.flags = flags;
1837 
1838 	if (oldlenp) {
1839 		req.oldlen = *oldlenp;
1840 	}
1841 	req.validlen = req.oldlen;
1842 
1843 	if (old) {
1844 		req.oldptr= old;
1845 	}
1846 
1847 	if (new != NULL) {
1848 		req.newlen = newlen;
1849 		req.newptr = new;
1850 	}
1851 
1852 	req.oldfunc = sysctl_old_kernel;
1853 	req.newfunc = sysctl_new_kernel;
1854 	req.lock = REQ_UNWIRED;
1855 
1856 	error = sysctl_root(0, name, namelen, &req);
1857 
1858 	if (req.lock == REQ_WIRED && req.validlen > 0)
1859 		vsunlock(req.oldptr, req.validlen);
1860 
1861 	if (error && error != ENOMEM)
1862 		return (error);
1863 
1864 	if (retval) {
1865 		if (req.oldptr && req.oldidx > req.validlen)
1866 			*retval = req.validlen;
1867 		else
1868 			*retval = req.oldidx;
1869 	}
1870 	return (error);
1871 }
1872 
1873 int
1874 kernel_sysctlbyname(struct thread *td, char *name, void *old, size_t *oldlenp,
1875     void *new, size_t newlen, size_t *retval, int flags)
1876 {
1877         int oid[CTL_MAXNAME];
1878         size_t oidlen, plen;
1879 	int error;
1880 
1881 	oid[0] = CTL_SYSCTL;
1882 	oid[1] = CTL_SYSCTL_NAME2OID;
1883 	oidlen = sizeof(oid);
1884 
1885 	error = kernel_sysctl(td, oid, 2, oid, &oidlen,
1886 	    (void *)name, strlen(name), &plen, flags);
1887 	if (error)
1888 		return (error);
1889 
1890 	error = kernel_sysctl(td, oid, plen / sizeof(int), old, oldlenp,
1891 	    new, newlen, retval, flags);
1892 	return (error);
1893 }
1894 
1895 /*
1896  * Transfer function to/from user space.
1897  */
1898 static int
1899 sysctl_old_user(struct sysctl_req *req, const void *p, size_t l)
1900 {
1901 	size_t i, len, origidx;
1902 	int error;
1903 
1904 	origidx = req->oldidx;
1905 	req->oldidx += l;
1906 	if (req->oldptr == NULL)
1907 		return (0);
1908 	/*
1909 	 * If we have not wired the user supplied buffer and we are currently
1910 	 * holding locks, drop a witness warning, as it's possible that
1911 	 * write operations to the user page can sleep.
1912 	 */
1913 	if (req->lock != REQ_WIRED)
1914 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
1915 		    "sysctl_old_user()");
1916 	i = l;
1917 	len = req->validlen;
1918 	if (len <= origidx)
1919 		i = 0;
1920 	else {
1921 		if (i > len - origidx)
1922 			i = len - origidx;
1923 		if (req->lock == REQ_WIRED) {
1924 			error = copyout_nofault(p, (char *)req->oldptr +
1925 			    origidx, i);
1926 		} else
1927 			error = copyout(p, (char *)req->oldptr + origidx, i);
1928 		if (error != 0)
1929 			return (error);
1930 	}
1931 	if (i < l)
1932 		return (ENOMEM);
1933 	return (0);
1934 }
1935 
1936 static int
1937 sysctl_new_user(struct sysctl_req *req, void *p, size_t l)
1938 {
1939 	int error;
1940 
1941 	if (!req->newptr)
1942 		return (0);
1943 	if (req->newlen - req->newidx < l)
1944 		return (EINVAL);
1945 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
1946 	    "sysctl_new_user()");
1947 	error = copyin((const char *)req->newptr + req->newidx, p, l);
1948 	req->newidx += l;
1949 	return (error);
1950 }
1951 
1952 /*
1953  * Wire the user space destination buffer.  If set to a value greater than
1954  * zero, the len parameter limits the maximum amount of wired memory.
1955  */
1956 int
1957 sysctl_wire_old_buffer(struct sysctl_req *req, size_t len)
1958 {
1959 	int ret;
1960 	size_t wiredlen;
1961 
1962 	wiredlen = (len > 0 && len < req->oldlen) ? len : req->oldlen;
1963 	ret = 0;
1964 	if (req->lock != REQ_WIRED && req->oldptr &&
1965 	    req->oldfunc == sysctl_old_user) {
1966 		if (wiredlen != 0) {
1967 			ret = vslock(req->oldptr, wiredlen);
1968 			if (ret != 0) {
1969 				if (ret != ENOMEM)
1970 					return (ret);
1971 				wiredlen = 0;
1972 			}
1973 		}
1974 		req->lock = REQ_WIRED;
1975 		req->validlen = wiredlen;
1976 	}
1977 	return (0);
1978 }
1979 
1980 int
1981 sysctl_find_oid(int *name, u_int namelen, struct sysctl_oid **noid,
1982     int *nindx, struct sysctl_req *req)
1983 {
1984 	struct sysctl_oid_list *lsp;
1985 	struct sysctl_oid *oid;
1986 	int indx;
1987 
1988 	SYSCTL_ASSERT_LOCKED();
1989 	lsp = &sysctl__children;
1990 	indx = 0;
1991 	while (indx < CTL_MAXNAME) {
1992 		SLIST_FOREACH(oid, lsp, oid_link) {
1993 			if (oid->oid_number == name[indx])
1994 				break;
1995 		}
1996 		if (oid == NULL)
1997 			return (ENOENT);
1998 
1999 		indx++;
2000 		if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
2001 			if (oid->oid_handler != NULL || indx == namelen) {
2002 				*noid = oid;
2003 				if (nindx != NULL)
2004 					*nindx = indx;
2005 				KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
2006 				    ("%s found DYING node %p", __func__, oid));
2007 				return (0);
2008 			}
2009 			lsp = SYSCTL_CHILDREN(oid);
2010 		} else if (indx == namelen) {
2011 			if ((oid->oid_kind & CTLFLAG_DORMANT) != 0)
2012 				return (ENOENT);
2013 			*noid = oid;
2014 			if (nindx != NULL)
2015 				*nindx = indx;
2016 			KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
2017 			    ("%s found DYING node %p", __func__, oid));
2018 			return (0);
2019 		} else {
2020 			return (ENOTDIR);
2021 		}
2022 	}
2023 	return (ENOENT);
2024 }
2025 
2026 /*
2027  * Traverse our tree, and find the right node, execute whatever it points
2028  * to, and return the resulting error code.
2029  */
2030 
2031 static int
2032 sysctl_root(SYSCTL_HANDLER_ARGS)
2033 {
2034 	struct sysctl_oid *oid;
2035 	struct rm_priotracker tracker;
2036 	int error, indx, lvl;
2037 
2038 	SYSCTL_RLOCK(&tracker);
2039 
2040 	error = sysctl_find_oid(arg1, arg2, &oid, &indx, req);
2041 	if (error)
2042 		goto out;
2043 
2044 	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
2045 		/*
2046 		 * You can't call a sysctl when it's a node, but has
2047 		 * no handler.  Inform the user that it's a node.
2048 		 * The indx may or may not be the same as namelen.
2049 		 */
2050 		if (oid->oid_handler == NULL) {
2051 			error = EISDIR;
2052 			goto out;
2053 		}
2054 	}
2055 
2056 	/* Is this sysctl writable? */
2057 	if (req->newptr && !(oid->oid_kind & CTLFLAG_WR)) {
2058 		error = EPERM;
2059 		goto out;
2060 	}
2061 
2062 	KASSERT(req->td != NULL, ("sysctl_root(): req->td == NULL"));
2063 
2064 #ifdef CAPABILITY_MODE
2065 	/*
2066 	 * If the process is in capability mode, then don't permit reading or
2067 	 * writing unless specifically granted for the node.
2068 	 */
2069 	if (IN_CAPABILITY_MODE(req->td)) {
2070 		if ((req->oldptr && !(oid->oid_kind & CTLFLAG_CAPRD)) ||
2071 		    (req->newptr && !(oid->oid_kind & CTLFLAG_CAPWR))) {
2072 			error = EPERM;
2073 			goto out;
2074 		}
2075 	}
2076 #endif
2077 
2078 	/* Is this sysctl sensitive to securelevels? */
2079 	if (req->newptr && (oid->oid_kind & CTLFLAG_SECURE)) {
2080 		lvl = (oid->oid_kind & CTLMASK_SECURE) >> CTLSHIFT_SECURE;
2081 		error = securelevel_gt(req->td->td_ucred, lvl);
2082 		if (error)
2083 			goto out;
2084 	}
2085 
2086 	/* Is this sysctl writable by only privileged users? */
2087 	if (req->newptr && !(oid->oid_kind & CTLFLAG_ANYBODY)) {
2088 		int priv;
2089 
2090 		if (oid->oid_kind & CTLFLAG_PRISON)
2091 			priv = PRIV_SYSCTL_WRITEJAIL;
2092 #ifdef VIMAGE
2093 		else if ((oid->oid_kind & CTLFLAG_VNET) &&
2094 		     prison_owns_vnet(req->td->td_ucred))
2095 			priv = PRIV_SYSCTL_WRITEJAIL;
2096 #endif
2097 		else
2098 			priv = PRIV_SYSCTL_WRITE;
2099 		error = priv_check(req->td, priv);
2100 		if (error)
2101 			goto out;
2102 	}
2103 
2104 	if (!oid->oid_handler) {
2105 		error = EINVAL;
2106 		goto out;
2107 	}
2108 
2109 	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
2110 		arg1 = (int *)arg1 + indx;
2111 		arg2 -= indx;
2112 	} else {
2113 		arg1 = oid->oid_arg1;
2114 		arg2 = oid->oid_arg2;
2115 	}
2116 #ifdef MAC
2117 	error = mac_system_check_sysctl(req->td->td_ucred, oid, arg1, arg2,
2118 	    req);
2119 	if (error != 0)
2120 		goto out;
2121 #endif
2122 #ifdef VIMAGE
2123 	if ((oid->oid_kind & CTLFLAG_VNET) && arg1 != NULL)
2124 		arg1 = (void *)(curvnet->vnet_data_base + (uintptr_t)arg1);
2125 #endif
2126 	error = sysctl_root_handler_locked(oid, arg1, arg2, req, &tracker);
2127 
2128 out:
2129 	SYSCTL_RUNLOCK(&tracker);
2130 	return (error);
2131 }
2132 
2133 #ifndef _SYS_SYSPROTO_H_
2134 struct sysctl_args {
2135 	int	*name;
2136 	u_int	namelen;
2137 	void	*old;
2138 	size_t	*oldlenp;
2139 	void	*new;
2140 	size_t	newlen;
2141 };
2142 #endif
2143 int
2144 sys___sysctl(struct thread *td, struct sysctl_args *uap)
2145 {
2146 	int error, i, name[CTL_MAXNAME];
2147 	size_t j;
2148 
2149 	if (uap->namelen > CTL_MAXNAME || uap->namelen < 2)
2150 		return (EINVAL);
2151 
2152  	error = copyin(uap->name, &name, uap->namelen * sizeof(int));
2153  	if (error)
2154 		return (error);
2155 
2156 	error = userland_sysctl(td, name, uap->namelen,
2157 		uap->old, uap->oldlenp, 0,
2158 		uap->new, uap->newlen, &j, 0);
2159 	if (error && error != ENOMEM)
2160 		return (error);
2161 	if (uap->oldlenp) {
2162 		i = copyout(&j, uap->oldlenp, sizeof(j));
2163 		if (i)
2164 			return (i);
2165 	}
2166 	return (error);
2167 }
2168 
2169 int
2170 kern___sysctlbyname(struct thread *td, const char *oname, size_t namelen,
2171     void *old, size_t *oldlenp, void *new, size_t newlen, size_t *retval,
2172     int flags, bool inkernel)
2173 {
2174 	int oid[CTL_MAXNAME];
2175 	char namebuf[16];
2176 	char *name;
2177 	size_t oidlen;
2178 	int error;
2179 
2180 	if (namelen > MAXPATHLEN || namelen == 0)
2181 		return (EINVAL);
2182 	name = namebuf;
2183 	if (namelen > sizeof(namebuf))
2184 		name = malloc(namelen, M_SYSCTL, M_WAITOK);
2185 	error = copyin(oname, name, namelen);
2186 	if (error != 0)
2187 		goto out;
2188 
2189 	oid[0] = CTL_SYSCTL;
2190 	oid[1] = CTL_SYSCTL_NAME2OID;
2191 	oidlen = sizeof(oid);
2192 	error = kernel_sysctl(td, oid, 2, oid, &oidlen, (void *)name, namelen,
2193 	    retval, flags);
2194 	if (error != 0)
2195 		goto out;
2196 	error = userland_sysctl(td, oid, *retval / sizeof(int), old, oldlenp,
2197 	    inkernel, new, newlen, retval, flags);
2198 
2199 out:
2200 	if (namelen > sizeof(namebuf))
2201 		free(name, M_SYSCTL);
2202 	return (error);
2203 }
2204 
2205 #ifndef	_SYS_SYSPROTO_H_
2206 struct __sysctlbyname_args {
2207 	const char	*name;
2208 	size_t	namelen;
2209 	void	*old;
2210 	size_t	*oldlenp;
2211 	void	*new;
2212 	size_t	newlen;
2213 };
2214 #endif
2215 int
2216 sys___sysctlbyname(struct thread *td, struct __sysctlbyname_args *uap)
2217 {
2218 	size_t rv;
2219 	int error;
2220 
2221 	error = kern___sysctlbyname(td, uap->name, uap->namelen, uap->old,
2222 	    uap->oldlenp, uap->new, uap->newlen, &rv, 0, 0);
2223 	if (error != 0)
2224 		return (error);
2225 	if (uap->oldlenp != NULL)
2226 		error = copyout(&rv, uap->oldlenp, sizeof(rv));
2227 
2228 	return (error);
2229 }
2230 
2231 /*
2232  * This is used from various compatibility syscalls too.  That's why name
2233  * must be in kernel space.
2234  */
2235 int
2236 userland_sysctl(struct thread *td, int *name, u_int namelen, void *old,
2237     size_t *oldlenp, int inkernel, const void *new, size_t newlen,
2238     size_t *retval, int flags)
2239 {
2240 	int error = 0, memlocked;
2241 	struct sysctl_req req;
2242 
2243 	bzero(&req, sizeof req);
2244 
2245 	req.td = td;
2246 	req.flags = flags;
2247 
2248 	if (oldlenp) {
2249 		if (inkernel) {
2250 			req.oldlen = *oldlenp;
2251 		} else {
2252 			error = copyin(oldlenp, &req.oldlen, sizeof(*oldlenp));
2253 			if (error)
2254 				return (error);
2255 		}
2256 	}
2257 	req.validlen = req.oldlen;
2258 	req.oldptr = old;
2259 
2260 	if (new != NULL) {
2261 		req.newlen = newlen;
2262 		req.newptr = new;
2263 	}
2264 
2265 	req.oldfunc = sysctl_old_user;
2266 	req.newfunc = sysctl_new_user;
2267 	req.lock = REQ_UNWIRED;
2268 
2269 #ifdef KTRACE
2270 	if (KTRPOINT(curthread, KTR_SYSCTL))
2271 		ktrsysctl(name, namelen);
2272 #endif
2273 	memlocked = 0;
2274 	if (req.oldptr && req.oldlen > 4 * PAGE_SIZE) {
2275 		memlocked = 1;
2276 		sx_xlock(&sysctlmemlock);
2277 	}
2278 	CURVNET_SET(TD_TO_VNET(td));
2279 
2280 	for (;;) {
2281 		req.oldidx = 0;
2282 		req.newidx = 0;
2283 		error = sysctl_root(0, name, namelen, &req);
2284 		if (error != EAGAIN)
2285 			break;
2286 		kern_yield(PRI_USER);
2287 	}
2288 
2289 	CURVNET_RESTORE();
2290 
2291 	if (req.lock == REQ_WIRED && req.validlen > 0)
2292 		vsunlock(req.oldptr, req.validlen);
2293 	if (memlocked)
2294 		sx_xunlock(&sysctlmemlock);
2295 
2296 	if (error && error != ENOMEM)
2297 		return (error);
2298 
2299 	if (retval) {
2300 		if (req.oldptr && req.oldidx > req.validlen)
2301 			*retval = req.validlen;
2302 		else
2303 			*retval = req.oldidx;
2304 	}
2305 	return (error);
2306 }
2307 
2308 /*
2309  * Drain into a sysctl struct.  The user buffer should be wired if a page
2310  * fault would cause issue.
2311  */
2312 static int
2313 sbuf_sysctl_drain(void *arg, const char *data, int len)
2314 {
2315 	struct sysctl_req *req = arg;
2316 	int error;
2317 
2318 	error = SYSCTL_OUT(req, data, len);
2319 	KASSERT(error >= 0, ("Got unexpected negative value %d", error));
2320 	return (error == 0 ? len : -error);
2321 }
2322 
2323 struct sbuf *
2324 sbuf_new_for_sysctl(struct sbuf *s, char *buf, int length,
2325     struct sysctl_req *req)
2326 {
2327 
2328 	/* Supply a default buffer size if none given. */
2329 	if (buf == NULL && length == 0)
2330 		length = 64;
2331 	s = sbuf_new(s, buf, length, SBUF_FIXEDLEN | SBUF_INCLUDENUL);
2332 	sbuf_set_drain(s, sbuf_sysctl_drain, req);
2333 	return (s);
2334 }
2335 
2336 #ifdef DDB
2337 
2338 /* The current OID the debugger is working with */
2339 static struct sysctl_oid *g_ddb_oid;
2340 
2341 /* The current flags specified by the user */
2342 static int g_ddb_sysctl_flags;
2343 
2344 /* Check to see if the last sysctl printed */
2345 static int g_ddb_sysctl_printed;
2346 
2347 static const int ctl_sign[CTLTYPE+1] = {
2348 	[CTLTYPE_INT] = 1,
2349 	[CTLTYPE_LONG] = 1,
2350 	[CTLTYPE_S8] = 1,
2351 	[CTLTYPE_S16] = 1,
2352 	[CTLTYPE_S32] = 1,
2353 	[CTLTYPE_S64] = 1,
2354 };
2355 
2356 static const int ctl_size[CTLTYPE+1] = {
2357 	[CTLTYPE_INT] = sizeof(int),
2358 	[CTLTYPE_UINT] = sizeof(u_int),
2359 	[CTLTYPE_LONG] = sizeof(long),
2360 	[CTLTYPE_ULONG] = sizeof(u_long),
2361 	[CTLTYPE_S8] = sizeof(int8_t),
2362 	[CTLTYPE_S16] = sizeof(int16_t),
2363 	[CTLTYPE_S32] = sizeof(int32_t),
2364 	[CTLTYPE_S64] = sizeof(int64_t),
2365 	[CTLTYPE_U8] = sizeof(uint8_t),
2366 	[CTLTYPE_U16] = sizeof(uint16_t),
2367 	[CTLTYPE_U32] = sizeof(uint32_t),
2368 	[CTLTYPE_U64] = sizeof(uint64_t),
2369 };
2370 
2371 #define DB_SYSCTL_NAME_ONLY	0x001	/* Compare with -N */
2372 #define DB_SYSCTL_VALUE_ONLY	0x002	/* Compare with -n */
2373 #define DB_SYSCTL_OPAQUE	0x004	/* Compare with -o */
2374 #define DB_SYSCTL_HEX		0x008	/* Compare with -x */
2375 
2376 #define DB_SYSCTL_SAFE_ONLY	0x100	/* Only simple types */
2377 
2378 static const char db_sysctl_modifs[] = {
2379 	'N', 'n', 'o', 'x',
2380 };
2381 
2382 static const int db_sysctl_modif_values[] = {
2383 	DB_SYSCTL_NAME_ONLY, DB_SYSCTL_VALUE_ONLY,
2384 	DB_SYSCTL_OPAQUE, DB_SYSCTL_HEX,
2385 };
2386 
2387 /* Handlers considered safe to print while recursing */
2388 static int (* const db_safe_handlers[])(SYSCTL_HANDLER_ARGS) = {
2389 	sysctl_handle_bool,
2390 	sysctl_handle_8,
2391 	sysctl_handle_16,
2392 	sysctl_handle_32,
2393 	sysctl_handle_64,
2394 	sysctl_handle_int,
2395 	sysctl_handle_long,
2396 	sysctl_handle_string,
2397 	sysctl_handle_opaque,
2398 };
2399 
2400 /*
2401  * Use in place of sysctl_old_kernel to print sysctl values.
2402  *
2403  * Compare to the output handling in show_var from sbin/sysctl/sysctl.c
2404  */
2405 static int
2406 sysctl_old_ddb(struct sysctl_req *req, const void *ptr, size_t len)
2407 {
2408 	const u_char *val, *p;
2409 	const char *sep1;
2410 	size_t intlen, slen;
2411 	uintmax_t umv;
2412 	intmax_t mv;
2413 	int sign, ctltype, hexlen, xflag, error;
2414 
2415 	/* Suppress false-positive GCC uninitialized variable warnings */
2416 	mv = 0;
2417 	umv = 0;
2418 
2419 	slen = len;
2420 	val = p = ptr;
2421 
2422 	if (ptr == NULL) {
2423 		error = 0;
2424 		goto out;
2425 	}
2426 
2427 	/* We are going to print */
2428 	g_ddb_sysctl_printed = 1;
2429 
2430 	xflag = g_ddb_sysctl_flags & DB_SYSCTL_HEX;
2431 
2432 	ctltype = (g_ddb_oid->oid_kind & CTLTYPE);
2433 	sign = ctl_sign[ctltype];
2434 	intlen = ctl_size[ctltype];
2435 
2436 	switch (ctltype) {
2437 	case CTLTYPE_NODE:
2438 	case CTLTYPE_STRING:
2439 		db_printf("%.*s", (int) len, (const char *) p);
2440 		error = 0;
2441 		goto out;
2442 
2443 	case CTLTYPE_INT:
2444 	case CTLTYPE_UINT:
2445 	case CTLTYPE_LONG:
2446 	case CTLTYPE_ULONG:
2447 	case CTLTYPE_S8:
2448 	case CTLTYPE_S16:
2449 	case CTLTYPE_S32:
2450 	case CTLTYPE_S64:
2451 	case CTLTYPE_U8:
2452 	case CTLTYPE_U16:
2453 	case CTLTYPE_U32:
2454 	case CTLTYPE_U64:
2455 		hexlen = 2 + (intlen * CHAR_BIT + 3) / 4;
2456 		sep1 = "";
2457 		while (len >= intlen) {
2458 			switch (ctltype) {
2459 			case CTLTYPE_INT:
2460 			case CTLTYPE_UINT:
2461 				umv = *(const u_int *)p;
2462 				mv = *(const int *)p;
2463 				break;
2464 			case CTLTYPE_LONG:
2465 			case CTLTYPE_ULONG:
2466 				umv = *(const u_long *)p;
2467 				mv = *(const long *)p;
2468 				break;
2469 			case CTLTYPE_S8:
2470 			case CTLTYPE_U8:
2471 				umv = *(const uint8_t *)p;
2472 				mv = *(const int8_t *)p;
2473 				break;
2474 			case CTLTYPE_S16:
2475 			case CTLTYPE_U16:
2476 				umv = *(const uint16_t *)p;
2477 				mv = *(const int16_t *)p;
2478 				break;
2479 			case CTLTYPE_S32:
2480 			case CTLTYPE_U32:
2481 				umv = *(const uint32_t *)p;
2482 				mv = *(const int32_t *)p;
2483 				break;
2484 			case CTLTYPE_S64:
2485 			case CTLTYPE_U64:
2486 				umv = *(const uint64_t *)p;
2487 				mv = *(const int64_t *)p;
2488 				break;
2489 			}
2490 
2491 			db_printf("%s", sep1);
2492 			if (xflag)
2493 				db_printf("%#0*jx", hexlen, umv);
2494 			else if (!sign)
2495 				db_printf("%ju", umv);
2496 			else if (g_ddb_oid->oid_fmt[1] == 'K') {
2497 				/* Kelvins are currently unsupported. */
2498 				error = EOPNOTSUPP;
2499 				goto out;
2500 			} else
2501 				db_printf("%jd", mv);
2502 
2503 			sep1 = " ";
2504 			len -= intlen;
2505 			p += intlen;
2506 		}
2507 		error = 0;
2508 		goto out;
2509 
2510 	case CTLTYPE_OPAQUE:
2511 		/* TODO: Support struct functions. */
2512 
2513 		/* FALLTHROUGH */
2514 	default:
2515 		db_printf("Format:%s Length:%zu Dump:0x",
2516 		    g_ddb_oid->oid_fmt, len);
2517 		while (len-- && (xflag || p < val + 16))
2518 			db_printf("%02x", *p++);
2519 		if (!xflag && len > 16)
2520 			db_printf("...");
2521 		error = 0;
2522 		goto out;
2523 	}
2524 
2525 out:
2526 	req->oldidx += slen;
2527 	return (error);
2528 }
2529 
2530 /*
2531  * Avoid setting new sysctl values from the debugger
2532  */
2533 static int
2534 sysctl_new_ddb(struct sysctl_req *req, void *p, size_t l)
2535 {
2536 
2537 	if (!req->newptr)
2538 		return (0);
2539 
2540 	/* Changing sysctls from the debugger is currently unsupported */
2541 	return (EPERM);
2542 }
2543 
2544 /*
2545  * Run a sysctl handler with the DDB oldfunc and newfunc attached.
2546  * Instead of copying any output to a buffer we'll dump it right to
2547  * the console.
2548  */
2549 static int
2550 db_sysctl(struct sysctl_oid *oidp, int *name, u_int namelen,
2551     void *old, size_t *oldlenp, size_t *retval, int flags)
2552 {
2553 	struct sysctl_req req;
2554 	int error;
2555 
2556 	/* Setup the request */
2557 	bzero(&req, sizeof req);
2558 	req.td = kdb_thread;
2559 	req.oldfunc = sysctl_old_ddb;
2560 	req.newfunc = sysctl_new_ddb;
2561 	req.lock = REQ_UNWIRED;
2562 	if (oldlenp) {
2563 		req.oldlen = *oldlenp;
2564 	}
2565 	req.validlen = req.oldlen;
2566 	if (old) {
2567 		req.oldptr = old;
2568 	}
2569 
2570 	/* Setup our globals for sysctl_old_ddb */
2571 	g_ddb_oid = oidp;
2572 	g_ddb_sysctl_flags = flags;
2573 	g_ddb_sysctl_printed = 0;
2574 
2575 	error = sysctl_root(0, name, namelen, &req);
2576 
2577 	/* Reset globals */
2578 	g_ddb_oid = NULL;
2579 	g_ddb_sysctl_flags = 0;
2580 
2581 	if (retval) {
2582 		if (req.oldptr && req.oldidx > req.validlen)
2583 			*retval = req.validlen;
2584 		else
2585 			*retval = req.oldidx;
2586 	}
2587 	return (error);
2588 }
2589 
2590 /*
2591  * Show a sysctl's name
2592  */
2593 static void
2594 db_show_oid_name(int *oid, size_t nlen)
2595 {
2596 	struct sysctl_oid *oidp;
2597 	int qoid[CTL_MAXNAME+2];
2598 	int error;
2599 
2600 	qoid[0] = 0;
2601 	memcpy(qoid + 2, oid, nlen * sizeof(int));
2602 	qoid[1] = 1;
2603 
2604 	error = sysctl_find_oid(qoid, nlen + 2, &oidp, NULL, NULL);
2605 	if (error)
2606 		db_error("sysctl name oid");
2607 
2608 	error = db_sysctl(oidp, qoid, nlen + 2, NULL, NULL, NULL, 0);
2609 	if (error)
2610 		db_error("sysctl name");
2611 }
2612 
2613 /*
2614  * Check to see if an OID is safe to print from ddb.
2615  */
2616 static bool
2617 db_oid_safe(const struct sysctl_oid *oidp)
2618 {
2619 	for (unsigned int i = 0; i < nitems(db_safe_handlers); ++i) {
2620 		if (oidp->oid_handler == db_safe_handlers[i])
2621 			return (true);
2622 	}
2623 
2624 	return (false);
2625 }
2626 
2627 /*
2628  * Show a sysctl at a specific OID
2629  * Compare to the input handling in show_var from sbin/sysctl/sysctl.c
2630  */
2631 static int
2632 db_show_oid(struct sysctl_oid *oidp, int *oid, size_t nlen, int flags)
2633 {
2634 	int error, xflag, oflag, Nflag, nflag;
2635 	size_t len;
2636 
2637 	xflag = flags & DB_SYSCTL_HEX;
2638 	oflag = flags & DB_SYSCTL_OPAQUE;
2639 	nflag = flags & DB_SYSCTL_VALUE_ONLY;
2640 	Nflag = flags & DB_SYSCTL_NAME_ONLY;
2641 
2642 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_OPAQUE &&
2643 	    (!xflag && !oflag))
2644 		return (0);
2645 
2646 	if (Nflag) {
2647 		db_show_oid_name(oid, nlen);
2648 		error = 0;
2649 		goto out;
2650 	}
2651 
2652 	if (!nflag) {
2653 		db_show_oid_name(oid, nlen);
2654 		db_printf(": ");
2655 	}
2656 
2657 	if ((flags & DB_SYSCTL_SAFE_ONLY) && !db_oid_safe(oidp)) {
2658 		db_printf("Skipping, unsafe to print while recursing.");
2659 		error = 0;
2660 		goto out;
2661 	}
2662 
2663 	/* Try once, and ask about the size */
2664 	len = 0;
2665 	error = db_sysctl(oidp, oid, nlen,
2666 	    NULL, NULL, &len, flags);
2667 	if (error)
2668 		goto out;
2669 
2670 	if (!g_ddb_sysctl_printed)
2671 		/* Lie about the size */
2672 		error = db_sysctl(oidp, oid, nlen,
2673 		    (void *) 1, &len, NULL, flags);
2674 
2675 out:
2676 	db_printf("\n");
2677 	return (error);
2678 }
2679 
2680 /*
2681  * Show all sysctls under a specific OID
2682  * Compare to sysctl_all from sbin/sysctl/sysctl.c
2683  */
2684 static int
2685 db_show_sysctl_all(int *oid, size_t len, int flags)
2686 {
2687 	struct sysctl_oid *oidp;
2688 	int name1[CTL_MAXNAME + 2], name2[CTL_MAXNAME + 2];
2689 	size_t l1, l2;
2690 
2691 	name1[0] = CTL_SYSCTL;
2692 	name1[1] = CTL_SYSCTL_NEXT;
2693 	l1 = 2;
2694 	if (len) {
2695 		memcpy(name1+2, oid, len * sizeof(int));
2696 		l1 +=len;
2697 	} else {
2698 		name1[2] = 1;
2699 		l1++;
2700 	}
2701 	for (;;) {
2702 		int i, error;
2703 
2704 		l2 = sizeof(name2);
2705 		error = kernel_sysctl(kdb_thread, name1, l1,
2706 		    name2, &l2, NULL, 0, &l2, 0);
2707 		if (error != 0) {
2708 			if (error == ENOENT)
2709 				return (0);
2710 			else
2711 				db_error("sysctl(getnext)");
2712 		}
2713 
2714 		l2 /= sizeof(int);
2715 
2716 		if (l2 < (unsigned int)len)
2717 			return (0);
2718 
2719 		for (i = 0; i < len; i++)
2720 			if (name2[i] != oid[i])
2721 				return (0);
2722 
2723 		/* Find the OID in question */
2724 		error = sysctl_find_oid(name2, l2, &oidp, NULL, NULL);
2725 		if (error)
2726 			return (error);
2727 
2728 		i = db_show_oid(oidp, name2, l2, flags | DB_SYSCTL_SAFE_ONLY);
2729 
2730 		if (db_pager_quit)
2731 			return (0);
2732 
2733 		memcpy(name1+2, name2, l2 * sizeof(int));
2734 		l1 = 2 + l2;
2735 	}
2736 }
2737 
2738 /*
2739  * Show a sysctl by its user facing string
2740  */
2741 static int
2742 db_sysctlbyname(char *name, int flags)
2743 {
2744 	struct sysctl_oid *oidp;
2745 	int oid[CTL_MAXNAME];
2746 	int error, nlen;
2747 
2748 	error = name2oid(name, oid, &nlen, &oidp);
2749 	if (error) {
2750 		return (error);
2751 	}
2752 
2753 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
2754 		db_show_sysctl_all(oid, nlen, flags);
2755 	} else {
2756 		error = db_show_oid(oidp, oid, nlen, flags);
2757 	}
2758 
2759 	return (error);
2760 }
2761 
2762 static void
2763 db_sysctl_cmd_usage(void)
2764 {
2765 	db_printf(
2766 	    " sysctl [/Nnox] <sysctl>					    \n"
2767 	    "								    \n"
2768 	    " <sysctl> The name of the sysctl to show.			    \n"
2769 	    "								    \n"
2770 	    " Show a sysctl by hooking into SYSCTL_IN and SYSCTL_OUT.	    \n"
2771 	    " This will work for most sysctls, but should not be used	    \n"
2772 	    " with sysctls that are known to malloc.			    \n"
2773 	    "								    \n"
2774 	    " While recursing any \"unsafe\" sysctls will be skipped.	    \n"
2775 	    " Call sysctl directly on the sysctl to try printing the	    \n"
2776 	    " skipped sysctl. This is unsafe and may make the ddb	    \n"
2777 	    " session unusable.						    \n"
2778 	    "								    \n"
2779 	    " Arguments:						    \n"
2780 	    "	/N	Display only the name of the sysctl.		    \n"
2781 	    "	/n	Display only the value of the sysctl.		    \n"
2782 	    "	/o	Display opaque values.				    \n"
2783 	    "	/x	Display the sysctl in hex.			    \n"
2784 	    "								    \n"
2785 	    "For example:						    \n"
2786 	    "sysctl vm.v_free_min					    \n"
2787 	    "vn.v_free_min: 12669					    \n"
2788 	    );
2789 }
2790 
2791 /*
2792  * Show a specific sysctl similar to sysctl (8).
2793  */
2794 DB_FUNC(sysctl, db_sysctl_cmd, db_cmd_table, CS_OWN, NULL)
2795 {
2796 	char name[TOK_STRING_SIZE];
2797 	int error, i, t, flags;
2798 
2799 	/* Parse the modifiers */
2800 	t = db_read_token();
2801 	if (t == tSLASH || t == tMINUS) {
2802 		t = db_read_token();
2803 		if (t != tIDENT) {
2804 			db_printf("Bad modifier\n");
2805 			error = EINVAL;
2806 			goto out;
2807 		}
2808 		db_strcpy(modif, db_tok_string);
2809 	}
2810 	else {
2811 		db_unread_token(t);
2812 		modif[0] = '\0';
2813 	}
2814 
2815 	flags = 0;
2816 	for (i = 0; i < nitems(db_sysctl_modifs); i++) {
2817 		if (strchr(modif, db_sysctl_modifs[i])) {
2818 			flags |= db_sysctl_modif_values[i];
2819 		}
2820 	}
2821 
2822 	/* Parse the sysctl names */
2823 	t = db_read_token();
2824 	if (t != tIDENT) {
2825 		db_printf("Need sysctl name\n");
2826 		error = EINVAL;
2827 		goto out;
2828 	}
2829 
2830 	/* Copy the name into a temporary buffer */
2831 	db_strcpy(name, db_tok_string);
2832 
2833 	/* Ensure there is no trailing cruft */
2834 	t = db_read_token();
2835 	if (t != tEOL) {
2836 		db_printf("Unexpected sysctl argument\n");
2837 		error = EINVAL;
2838 		goto out;
2839 	}
2840 
2841 	error = db_sysctlbyname(name, flags);
2842 	if (error == ENOENT) {
2843 		db_printf("unknown oid: '%s'\n", db_tok_string);
2844 		goto out;
2845 	} else if (error) {
2846 		db_printf("%s: error: %d\n", db_tok_string, error);
2847 		goto out;
2848 	}
2849 
2850 out:
2851 	/* Ensure we eat all of our text */
2852 	db_flush_lex();
2853 
2854 	if (error == EINVAL) {
2855 		db_sysctl_cmd_usage();
2856 	}
2857 }
2858 
2859 #endif /* DDB */
2860