xref: /freebsd/sys/kern/kern_sysctl.c (revision 3416500aef140042c64bc149cb1ec6620483bc44)
1 /*-
2  * Copyright (c) 1982, 1986, 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Mike Karels at Berkeley Software Design, Inc.
7  *
8  * Quite extensively rewritten by Poul-Henning Kamp of the FreeBSD
9  * project, to make these variables more userfriendly.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. Neither the name of the University nor the names of its contributors
20  *    may be used to endorse or promote products derived from this software
21  *    without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  *
35  *	@(#)kern_sysctl.c	8.4 (Berkeley) 4/14/94
36  */
37 
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
40 
41 #include "opt_capsicum.h"
42 #include "opt_compat.h"
43 #include "opt_ktrace.h"
44 
45 #include <sys/param.h>
46 #include <sys/fail.h>
47 #include <sys/systm.h>
48 #include <sys/capsicum.h>
49 #include <sys/kernel.h>
50 #include <sys/sysctl.h>
51 #include <sys/malloc.h>
52 #include <sys/priv.h>
53 #include <sys/proc.h>
54 #include <sys/jail.h>
55 #include <sys/lock.h>
56 #include <sys/mutex.h>
57 #include <sys/rmlock.h>
58 #include <sys/sbuf.h>
59 #include <sys/sx.h>
60 #include <sys/sysproto.h>
61 #include <sys/uio.h>
62 #ifdef KTRACE
63 #include <sys/ktrace.h>
64 #endif
65 
66 #include <net/vnet.h>
67 
68 #include <security/mac/mac_framework.h>
69 
70 #include <vm/vm.h>
71 #include <vm/vm_extern.h>
72 
73 static MALLOC_DEFINE(M_SYSCTL, "sysctl", "sysctl internal magic");
74 static MALLOC_DEFINE(M_SYSCTLOID, "sysctloid", "sysctl dynamic oids");
75 static MALLOC_DEFINE(M_SYSCTLTMP, "sysctltmp", "sysctl temp output buffer");
76 
77 /*
78  * The sysctllock protects the MIB tree.  It also protects sysctl
79  * contexts used with dynamic sysctls.  The sysctl_register_oid() and
80  * sysctl_unregister_oid() routines require the sysctllock to already
81  * be held, so the sysctl_wlock() and sysctl_wunlock() routines are
82  * provided for the few places in the kernel which need to use that
83  * API rather than using the dynamic API.  Use of the dynamic API is
84  * strongly encouraged for most code.
85  *
86  * The sysctlmemlock is used to limit the amount of user memory wired for
87  * sysctl requests.  This is implemented by serializing any userland
88  * sysctl requests larger than a single page via an exclusive lock.
89  */
90 static struct rmlock sysctllock;
91 static struct sx sysctlmemlock;
92 
93 #define	SYSCTL_WLOCK()		rm_wlock(&sysctllock)
94 #define	SYSCTL_WUNLOCK()	rm_wunlock(&sysctllock)
95 #define	SYSCTL_RLOCK(tracker)	rm_rlock(&sysctllock, (tracker))
96 #define	SYSCTL_RUNLOCK(tracker)	rm_runlock(&sysctllock, (tracker))
97 #define	SYSCTL_WLOCKED()	rm_wowned(&sysctllock)
98 #define	SYSCTL_ASSERT_LOCKED()	rm_assert(&sysctllock, RA_LOCKED)
99 #define	SYSCTL_ASSERT_WLOCKED()	rm_assert(&sysctllock, RA_WLOCKED)
100 #define	SYSCTL_ASSERT_RLOCKED()	rm_assert(&sysctllock, RA_RLOCKED)
101 #define	SYSCTL_INIT()		rm_init_flags(&sysctllock, "sysctl lock", \
102 				    RM_SLEEPABLE)
103 #define	SYSCTL_SLEEP(ch, wmesg, timo)					\
104 				rm_sleep(ch, &sysctllock, 0, wmesg, timo)
105 
106 static int sysctl_root(SYSCTL_HANDLER_ARGS);
107 
108 /* Root list */
109 struct sysctl_oid_list sysctl__children = SLIST_HEAD_INITIALIZER(&sysctl__children);
110 
111 static int	sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del,
112 		    int recurse);
113 static int	sysctl_old_kernel(struct sysctl_req *, const void *, size_t);
114 static int	sysctl_new_kernel(struct sysctl_req *, void *, size_t);
115 
116 static struct sysctl_oid *
117 sysctl_find_oidname(const char *name, struct sysctl_oid_list *list)
118 {
119 	struct sysctl_oid *oidp;
120 
121 	SYSCTL_ASSERT_LOCKED();
122 	SLIST_FOREACH(oidp, list, oid_link) {
123 		if (strcmp(oidp->oid_name, name) == 0) {
124 			return (oidp);
125 		}
126 	}
127 	return (NULL);
128 }
129 
130 /*
131  * Initialization of the MIB tree.
132  *
133  * Order by number in each list.
134  */
135 void
136 sysctl_wlock(void)
137 {
138 
139 	SYSCTL_WLOCK();
140 }
141 
142 void
143 sysctl_wunlock(void)
144 {
145 
146 	SYSCTL_WUNLOCK();
147 }
148 
149 static int
150 sysctl_root_handler_locked(struct sysctl_oid *oid, void *arg1, intmax_t arg2,
151     struct sysctl_req *req, struct rm_priotracker *tracker)
152 {
153 	int error;
154 
155 	if (oid->oid_kind & CTLFLAG_DYN)
156 		atomic_add_int(&oid->oid_running, 1);
157 
158 	if (tracker != NULL)
159 		SYSCTL_RUNLOCK(tracker);
160 	else
161 		SYSCTL_WUNLOCK();
162 
163 	if (!(oid->oid_kind & CTLFLAG_MPSAFE))
164 		mtx_lock(&Giant);
165 	error = oid->oid_handler(oid, arg1, arg2, req);
166 	if (!(oid->oid_kind & CTLFLAG_MPSAFE))
167 		mtx_unlock(&Giant);
168 
169 	KFAIL_POINT_ERROR(_debug_fail_point, sysctl_running, error);
170 
171 	if (tracker != NULL)
172 		SYSCTL_RLOCK(tracker);
173 	else
174 		SYSCTL_WLOCK();
175 
176 	if (oid->oid_kind & CTLFLAG_DYN) {
177 		if (atomic_fetchadd_int(&oid->oid_running, -1) == 1 &&
178 		    (oid->oid_kind & CTLFLAG_DYING) != 0)
179 			wakeup(&oid->oid_running);
180 	}
181 
182 	return (error);
183 }
184 
185 static void
186 sysctl_load_tunable_by_oid_locked(struct sysctl_oid *oidp)
187 {
188 	struct sysctl_req req;
189 	struct sysctl_oid *curr;
190 	char *penv = NULL;
191 	char path[64];
192 	ssize_t rem = sizeof(path);
193 	ssize_t len;
194 	uint8_t val_8;
195 	uint16_t val_16;
196 	uint32_t val_32;
197 	int val_int;
198 	long val_long;
199 	int64_t val_64;
200 	quad_t val_quad;
201 	int error;
202 
203 	path[--rem] = 0;
204 
205 	for (curr = oidp; curr != NULL; curr = SYSCTL_PARENT(curr)) {
206 		len = strlen(curr->oid_name);
207 		rem -= len;
208 		if (curr != oidp)
209 			rem -= 1;
210 		if (rem < 0) {
211 			printf("OID path exceeds %d bytes\n", (int)sizeof(path));
212 			return;
213 		}
214 		memcpy(path + rem, curr->oid_name, len);
215 		if (curr != oidp)
216 			path[rem + len] = '.';
217 	}
218 
219 	memset(&req, 0, sizeof(req));
220 
221 	req.td = curthread;
222 	req.oldfunc = sysctl_old_kernel;
223 	req.newfunc = sysctl_new_kernel;
224 	req.lock = REQ_UNWIRED;
225 
226 	switch (oidp->oid_kind & CTLTYPE) {
227 	case CTLTYPE_INT:
228 		if (getenv_int(path + rem, &val_int) == 0)
229 			return;
230 		req.newlen = sizeof(val_int);
231 		req.newptr = &val_int;
232 		break;
233 	case CTLTYPE_UINT:
234 		if (getenv_uint(path + rem, (unsigned int *)&val_int) == 0)
235 			return;
236 		req.newlen = sizeof(val_int);
237 		req.newptr = &val_int;
238 		break;
239 	case CTLTYPE_LONG:
240 		if (getenv_long(path + rem, &val_long) == 0)
241 			return;
242 		req.newlen = sizeof(val_long);
243 		req.newptr = &val_long;
244 		break;
245 	case CTLTYPE_ULONG:
246 		if (getenv_ulong(path + rem, (unsigned long *)&val_long) == 0)
247 			return;
248 		req.newlen = sizeof(val_long);
249 		req.newptr = &val_long;
250 		break;
251 	case CTLTYPE_S8:
252 		if (getenv_int(path + rem, &val_int) == 0)
253 			return;
254 		val_8 = val_int;
255 		req.newlen = sizeof(val_8);
256 		req.newptr = &val_8;
257 		break;
258 	case CTLTYPE_S16:
259 		if (getenv_int(path + rem, &val_int) == 0)
260 			return;
261 		val_16 = val_int;
262 		req.newlen = sizeof(val_16);
263 		req.newptr = &val_16;
264 		break;
265 	case CTLTYPE_S32:
266 		if (getenv_long(path + rem, &val_long) == 0)
267 			return;
268 		val_32 = val_long;
269 		req.newlen = sizeof(val_32);
270 		req.newptr = &val_32;
271 		break;
272 	case CTLTYPE_S64:
273 		if (getenv_quad(path + rem, &val_quad) == 0)
274 			return;
275 		val_64 = val_quad;
276 		req.newlen = sizeof(val_64);
277 		req.newptr = &val_64;
278 		break;
279 	case CTLTYPE_U8:
280 		if (getenv_uint(path + rem, (unsigned int *)&val_int) == 0)
281 			return;
282 		val_8 = val_int;
283 		req.newlen = sizeof(val_8);
284 		req.newptr = &val_8;
285 		break;
286 	case CTLTYPE_U16:
287 		if (getenv_uint(path + rem, (unsigned int *)&val_int) == 0)
288 			return;
289 		val_16 = val_int;
290 		req.newlen = sizeof(val_16);
291 		req.newptr = &val_16;
292 		break;
293 	case CTLTYPE_U32:
294 		if (getenv_ulong(path + rem, (unsigned long *)&val_long) == 0)
295 			return;
296 		val_32 = val_long;
297 		req.newlen = sizeof(val_32);
298 		req.newptr = &val_32;
299 		break;
300 	case CTLTYPE_U64:
301 		/* XXX there is no getenv_uquad() */
302 		if (getenv_quad(path + rem, &val_quad) == 0)
303 			return;
304 		val_64 = val_quad;
305 		req.newlen = sizeof(val_64);
306 		req.newptr = &val_64;
307 		break;
308 	case CTLTYPE_STRING:
309 		penv = kern_getenv(path + rem);
310 		if (penv == NULL)
311 			return;
312 		req.newlen = strlen(penv);
313 		req.newptr = penv;
314 		break;
315 	default:
316 		return;
317 	}
318 	error = sysctl_root_handler_locked(oidp, oidp->oid_arg1,
319 	    oidp->oid_arg2, &req, NULL);
320 	if (error != 0)
321 		printf("Setting sysctl %s failed: %d\n", path + rem, error);
322 	if (penv != NULL)
323 		freeenv(penv);
324 }
325 
326 void
327 sysctl_register_oid(struct sysctl_oid *oidp)
328 {
329 	struct sysctl_oid_list *parent = oidp->oid_parent;
330 	struct sysctl_oid *p;
331 	struct sysctl_oid *q;
332 	int oid_number;
333 	int timeout = 2;
334 
335 	/*
336 	 * First check if another oid with the same name already
337 	 * exists in the parent's list.
338 	 */
339 	SYSCTL_ASSERT_WLOCKED();
340 	p = sysctl_find_oidname(oidp->oid_name, parent);
341 	if (p != NULL) {
342 		if ((p->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
343 			p->oid_refcnt++;
344 			return;
345 		} else {
346 			printf("can't re-use a leaf (%s)!\n", p->oid_name);
347 			return;
348 		}
349 	}
350 	/* get current OID number */
351 	oid_number = oidp->oid_number;
352 
353 #if (OID_AUTO >= 0)
354 #error "OID_AUTO is expected to be a negative value"
355 #endif
356 	/*
357 	 * Any negative OID number qualifies as OID_AUTO. Valid OID
358 	 * numbers should always be positive.
359 	 *
360 	 * NOTE: DO NOT change the starting value here, change it in
361 	 * <sys/sysctl.h>, and make sure it is at least 256 to
362 	 * accommodate e.g. net.inet.raw as a static sysctl node.
363 	 */
364 	if (oid_number < 0) {
365 		static int newoid;
366 
367 		/*
368 		 * By decrementing the next OID number we spend less
369 		 * time inserting the OIDs into a sorted list.
370 		 */
371 		if (--newoid < CTL_AUTO_START)
372 			newoid = 0x7fffffff;
373 
374 		oid_number = newoid;
375 	}
376 
377 	/*
378 	 * Insert the OID into the parent's list sorted by OID number.
379 	 */
380 retry:
381 	q = NULL;
382 	SLIST_FOREACH(p, parent, oid_link) {
383 		/* check if the current OID number is in use */
384 		if (oid_number == p->oid_number) {
385 			/* get the next valid OID number */
386 			if (oid_number < CTL_AUTO_START ||
387 			    oid_number == 0x7fffffff) {
388 				/* wraparound - restart */
389 				oid_number = CTL_AUTO_START;
390 				/* don't loop forever */
391 				if (!timeout--)
392 					panic("sysctl: Out of OID numbers\n");
393 				goto retry;
394 			} else {
395 				oid_number++;
396 			}
397 		} else if (oid_number < p->oid_number)
398 			break;
399 		q = p;
400 	}
401 	/* check for non-auto OID number collision */
402 	if (oidp->oid_number >= 0 && oidp->oid_number < CTL_AUTO_START &&
403 	    oid_number >= CTL_AUTO_START) {
404 		printf("sysctl: OID number(%d) is already in use for '%s'\n",
405 		    oidp->oid_number, oidp->oid_name);
406 	}
407 	/* update the OID number, if any */
408 	oidp->oid_number = oid_number;
409 	if (q != NULL)
410 		SLIST_INSERT_AFTER(q, oidp, oid_link);
411 	else
412 		SLIST_INSERT_HEAD(parent, oidp, oid_link);
413 
414 	if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE &&
415 #ifdef VIMAGE
416 	    (oidp->oid_kind & CTLFLAG_VNET) == 0 &&
417 #endif
418 	    (oidp->oid_kind & CTLFLAG_TUN) != 0 &&
419 	    (oidp->oid_kind & CTLFLAG_NOFETCH) == 0) {
420 		/* only fetch value once */
421 		oidp->oid_kind |= CTLFLAG_NOFETCH;
422 		/* try to fetch value from kernel environment */
423 		sysctl_load_tunable_by_oid_locked(oidp);
424 	}
425 }
426 
427 void
428 sysctl_unregister_oid(struct sysctl_oid *oidp)
429 {
430 	struct sysctl_oid *p;
431 	int error;
432 
433 	SYSCTL_ASSERT_WLOCKED();
434 	error = ENOENT;
435 	if (oidp->oid_number == OID_AUTO) {
436 		error = EINVAL;
437 	} else {
438 		SLIST_FOREACH(p, oidp->oid_parent, oid_link) {
439 			if (p == oidp) {
440 				SLIST_REMOVE(oidp->oid_parent, oidp,
441 				    sysctl_oid, oid_link);
442 				error = 0;
443 				break;
444 			}
445 		}
446 	}
447 
448 	/*
449 	 * This can happen when a module fails to register and is
450 	 * being unloaded afterwards.  It should not be a panic()
451 	 * for normal use.
452 	 */
453 	if (error)
454 		printf("%s: failed to unregister sysctl\n", __func__);
455 }
456 
457 /* Initialize a new context to keep track of dynamically added sysctls. */
458 int
459 sysctl_ctx_init(struct sysctl_ctx_list *c)
460 {
461 
462 	if (c == NULL) {
463 		return (EINVAL);
464 	}
465 
466 	/*
467 	 * No locking here, the caller is responsible for not adding
468 	 * new nodes to a context until after this function has
469 	 * returned.
470 	 */
471 	TAILQ_INIT(c);
472 	return (0);
473 }
474 
475 /* Free the context, and destroy all dynamic oids registered in this context */
476 int
477 sysctl_ctx_free(struct sysctl_ctx_list *clist)
478 {
479 	struct sysctl_ctx_entry *e, *e1;
480 	int error;
481 
482 	error = 0;
483 	/*
484 	 * First perform a "dry run" to check if it's ok to remove oids.
485 	 * XXX FIXME
486 	 * XXX This algorithm is a hack. But I don't know any
487 	 * XXX better solution for now...
488 	 */
489 	SYSCTL_WLOCK();
490 	TAILQ_FOREACH(e, clist, link) {
491 		error = sysctl_remove_oid_locked(e->entry, 0, 0);
492 		if (error)
493 			break;
494 	}
495 	/*
496 	 * Restore deregistered entries, either from the end,
497 	 * or from the place where error occurred.
498 	 * e contains the entry that was not unregistered
499 	 */
500 	if (error)
501 		e1 = TAILQ_PREV(e, sysctl_ctx_list, link);
502 	else
503 		e1 = TAILQ_LAST(clist, sysctl_ctx_list);
504 	while (e1 != NULL) {
505 		sysctl_register_oid(e1->entry);
506 		e1 = TAILQ_PREV(e1, sysctl_ctx_list, link);
507 	}
508 	if (error) {
509 		SYSCTL_WUNLOCK();
510 		return(EBUSY);
511 	}
512 	/* Now really delete the entries */
513 	e = TAILQ_FIRST(clist);
514 	while (e != NULL) {
515 		e1 = TAILQ_NEXT(e, link);
516 		error = sysctl_remove_oid_locked(e->entry, 1, 0);
517 		if (error)
518 			panic("sysctl_remove_oid: corrupt tree, entry: %s",
519 			    e->entry->oid_name);
520 		free(e, M_SYSCTLOID);
521 		e = e1;
522 	}
523 	SYSCTL_WUNLOCK();
524 	return (error);
525 }
526 
527 /* Add an entry to the context */
528 struct sysctl_ctx_entry *
529 sysctl_ctx_entry_add(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
530 {
531 	struct sysctl_ctx_entry *e;
532 
533 	SYSCTL_ASSERT_WLOCKED();
534 	if (clist == NULL || oidp == NULL)
535 		return(NULL);
536 	e = malloc(sizeof(struct sysctl_ctx_entry), M_SYSCTLOID, M_WAITOK);
537 	e->entry = oidp;
538 	TAILQ_INSERT_HEAD(clist, e, link);
539 	return (e);
540 }
541 
542 /* Find an entry in the context */
543 struct sysctl_ctx_entry *
544 sysctl_ctx_entry_find(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
545 {
546 	struct sysctl_ctx_entry *e;
547 
548 	SYSCTL_ASSERT_WLOCKED();
549 	if (clist == NULL || oidp == NULL)
550 		return(NULL);
551 	TAILQ_FOREACH(e, clist, link) {
552 		if(e->entry == oidp)
553 			return(e);
554 	}
555 	return (e);
556 }
557 
558 /*
559  * Delete an entry from the context.
560  * NOTE: this function doesn't free oidp! You have to remove it
561  * with sysctl_remove_oid().
562  */
563 int
564 sysctl_ctx_entry_del(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
565 {
566 	struct sysctl_ctx_entry *e;
567 
568 	if (clist == NULL || oidp == NULL)
569 		return (EINVAL);
570 	SYSCTL_WLOCK();
571 	e = sysctl_ctx_entry_find(clist, oidp);
572 	if (e != NULL) {
573 		TAILQ_REMOVE(clist, e, link);
574 		SYSCTL_WUNLOCK();
575 		free(e, M_SYSCTLOID);
576 		return (0);
577 	} else {
578 		SYSCTL_WUNLOCK();
579 		return (ENOENT);
580 	}
581 }
582 
583 /*
584  * Remove dynamically created sysctl trees.
585  * oidp - top of the tree to be removed
586  * del - if 0 - just deregister, otherwise free up entries as well
587  * recurse - if != 0 traverse the subtree to be deleted
588  */
589 int
590 sysctl_remove_oid(struct sysctl_oid *oidp, int del, int recurse)
591 {
592 	int error;
593 
594 	SYSCTL_WLOCK();
595 	error = sysctl_remove_oid_locked(oidp, del, recurse);
596 	SYSCTL_WUNLOCK();
597 	return (error);
598 }
599 
600 int
601 sysctl_remove_name(struct sysctl_oid *parent, const char *name,
602     int del, int recurse)
603 {
604 	struct sysctl_oid *p, *tmp;
605 	int error;
606 
607 	error = ENOENT;
608 	SYSCTL_WLOCK();
609 	SLIST_FOREACH_SAFE(p, SYSCTL_CHILDREN(parent), oid_link, tmp) {
610 		if (strcmp(p->oid_name, name) == 0) {
611 			error = sysctl_remove_oid_locked(p, del, recurse);
612 			break;
613 		}
614 	}
615 	SYSCTL_WUNLOCK();
616 
617 	return (error);
618 }
619 
620 
621 static int
622 sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del, int recurse)
623 {
624 	struct sysctl_oid *p, *tmp;
625 	int error;
626 
627 	SYSCTL_ASSERT_WLOCKED();
628 	if (oidp == NULL)
629 		return(EINVAL);
630 	if ((oidp->oid_kind & CTLFLAG_DYN) == 0) {
631 		printf("Warning: can't remove non-dynamic nodes (%s)!\n",
632 		    oidp->oid_name);
633 		return (EINVAL);
634 	}
635 	/*
636 	 * WARNING: normal method to do this should be through
637 	 * sysctl_ctx_free(). Use recursing as the last resort
638 	 * method to purge your sysctl tree of leftovers...
639 	 * However, if some other code still references these nodes,
640 	 * it will panic.
641 	 */
642 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
643 		if (oidp->oid_refcnt == 1) {
644 			SLIST_FOREACH_SAFE(p,
645 			    SYSCTL_CHILDREN(oidp), oid_link, tmp) {
646 				if (!recurse) {
647 					printf("Warning: failed attempt to "
648 					    "remove oid %s with child %s\n",
649 					    oidp->oid_name, p->oid_name);
650 					return (ENOTEMPTY);
651 				}
652 				error = sysctl_remove_oid_locked(p, del,
653 				    recurse);
654 				if (error)
655 					return (error);
656 			}
657 		}
658 	}
659 	if (oidp->oid_refcnt > 1 ) {
660 		oidp->oid_refcnt--;
661 	} else {
662 		if (oidp->oid_refcnt == 0) {
663 			printf("Warning: bad oid_refcnt=%u (%s)!\n",
664 				oidp->oid_refcnt, oidp->oid_name);
665 			return (EINVAL);
666 		}
667 		sysctl_unregister_oid(oidp);
668 		if (del) {
669 			/*
670 			 * Wait for all threads running the handler to drain.
671 			 * This preserves the previous behavior when the
672 			 * sysctl lock was held across a handler invocation,
673 			 * and is necessary for module unload correctness.
674 			 */
675 			while (oidp->oid_running > 0) {
676 				oidp->oid_kind |= CTLFLAG_DYING;
677 				SYSCTL_SLEEP(&oidp->oid_running, "oidrm", 0);
678 			}
679 			if (oidp->oid_descr)
680 				free(__DECONST(char *, oidp->oid_descr),
681 				    M_SYSCTLOID);
682 			if (oidp->oid_label)
683 				free(__DECONST(char *, oidp->oid_label),
684 				    M_SYSCTLOID);
685 			free(__DECONST(char *, oidp->oid_name), M_SYSCTLOID);
686 			free(oidp, M_SYSCTLOID);
687 		}
688 	}
689 	return (0);
690 }
691 /*
692  * Create new sysctls at run time.
693  * clist may point to a valid context initialized with sysctl_ctx_init().
694  */
695 struct sysctl_oid *
696 sysctl_add_oid(struct sysctl_ctx_list *clist, struct sysctl_oid_list *parent,
697 	int number, const char *name, int kind, void *arg1, intmax_t arg2,
698 	int (*handler)(SYSCTL_HANDLER_ARGS), const char *fmt, const char *descr,
699 	const char *label)
700 {
701 	struct sysctl_oid *oidp;
702 
703 	/* You have to hook up somewhere.. */
704 	if (parent == NULL)
705 		return(NULL);
706 	/* Check if the node already exists, otherwise create it */
707 	SYSCTL_WLOCK();
708 	oidp = sysctl_find_oidname(name, parent);
709 	if (oidp != NULL) {
710 		if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
711 			oidp->oid_refcnt++;
712 			/* Update the context */
713 			if (clist != NULL)
714 				sysctl_ctx_entry_add(clist, oidp);
715 			SYSCTL_WUNLOCK();
716 			return (oidp);
717 		} else {
718 			SYSCTL_WUNLOCK();
719 			printf("can't re-use a leaf (%s)!\n", name);
720 			return (NULL);
721 		}
722 	}
723 	oidp = malloc(sizeof(struct sysctl_oid), M_SYSCTLOID, M_WAITOK|M_ZERO);
724 	oidp->oid_parent = parent;
725 	SLIST_INIT(&oidp->oid_children);
726 	oidp->oid_number = number;
727 	oidp->oid_refcnt = 1;
728 	oidp->oid_name = strdup(name, M_SYSCTLOID);
729 	oidp->oid_handler = handler;
730 	oidp->oid_kind = CTLFLAG_DYN | kind;
731 	oidp->oid_arg1 = arg1;
732 	oidp->oid_arg2 = arg2;
733 	oidp->oid_fmt = fmt;
734 	if (descr != NULL)
735 		oidp->oid_descr = strdup(descr, M_SYSCTLOID);
736 	if (label != NULL)
737 		oidp->oid_label = strdup(label, M_SYSCTLOID);
738 	/* Update the context, if used */
739 	if (clist != NULL)
740 		sysctl_ctx_entry_add(clist, oidp);
741 	/* Register this oid */
742 	sysctl_register_oid(oidp);
743 	SYSCTL_WUNLOCK();
744 	return (oidp);
745 }
746 
747 /*
748  * Rename an existing oid.
749  */
750 void
751 sysctl_rename_oid(struct sysctl_oid *oidp, const char *name)
752 {
753 	char *newname;
754 	char *oldname;
755 
756 	newname = strdup(name, M_SYSCTLOID);
757 	SYSCTL_WLOCK();
758 	oldname = __DECONST(char *, oidp->oid_name);
759 	oidp->oid_name = newname;
760 	SYSCTL_WUNLOCK();
761 	free(oldname, M_SYSCTLOID);
762 }
763 
764 /*
765  * Reparent an existing oid.
766  */
767 int
768 sysctl_move_oid(struct sysctl_oid *oid, struct sysctl_oid_list *parent)
769 {
770 	struct sysctl_oid *oidp;
771 
772 	SYSCTL_WLOCK();
773 	if (oid->oid_parent == parent) {
774 		SYSCTL_WUNLOCK();
775 		return (0);
776 	}
777 	oidp = sysctl_find_oidname(oid->oid_name, parent);
778 	if (oidp != NULL) {
779 		SYSCTL_WUNLOCK();
780 		return (EEXIST);
781 	}
782 	sysctl_unregister_oid(oid);
783 	oid->oid_parent = parent;
784 	oid->oid_number = OID_AUTO;
785 	sysctl_register_oid(oid);
786 	SYSCTL_WUNLOCK();
787 	return (0);
788 }
789 
790 /*
791  * Register the kernel's oids on startup.
792  */
793 SET_DECLARE(sysctl_set, struct sysctl_oid);
794 
795 static void
796 sysctl_register_all(void *arg)
797 {
798 	struct sysctl_oid **oidp;
799 
800 	sx_init(&sysctlmemlock, "sysctl mem");
801 	SYSCTL_INIT();
802 	SYSCTL_WLOCK();
803 	SET_FOREACH(oidp, sysctl_set)
804 		sysctl_register_oid(*oidp);
805 	SYSCTL_WUNLOCK();
806 }
807 SYSINIT(sysctl, SI_SUB_KMEM, SI_ORDER_FIRST, sysctl_register_all, 0);
808 
809 /*
810  * "Staff-functions"
811  *
812  * These functions implement a presently undocumented interface
813  * used by the sysctl program to walk the tree, and get the type
814  * so it can print the value.
815  * This interface is under work and consideration, and should probably
816  * be killed with a big axe by the first person who can find the time.
817  * (be aware though, that the proper interface isn't as obvious as it
818  * may seem, there are various conflicting requirements.
819  *
820  * {0,0}	printf the entire MIB-tree.
821  * {0,1,...}	return the name of the "..." OID.
822  * {0,2,...}	return the next OID.
823  * {0,3}	return the OID of the name in "new"
824  * {0,4,...}	return the kind & format info for the "..." OID.
825  * {0,5,...}	return the description of the "..." OID.
826  * {0,6,...}	return the aggregation label of the "..." OID.
827  */
828 
829 #ifdef SYSCTL_DEBUG
830 static void
831 sysctl_sysctl_debug_dump_node(struct sysctl_oid_list *l, int i)
832 {
833 	int k;
834 	struct sysctl_oid *oidp;
835 
836 	SYSCTL_ASSERT_LOCKED();
837 	SLIST_FOREACH(oidp, l, oid_link) {
838 
839 		for (k=0; k<i; k++)
840 			printf(" ");
841 
842 		printf("%d %s ", oidp->oid_number, oidp->oid_name);
843 
844 		printf("%c%c",
845 			oidp->oid_kind & CTLFLAG_RD ? 'R':' ',
846 			oidp->oid_kind & CTLFLAG_WR ? 'W':' ');
847 
848 		if (oidp->oid_handler)
849 			printf(" *Handler");
850 
851 		switch (oidp->oid_kind & CTLTYPE) {
852 			case CTLTYPE_NODE:
853 				printf(" Node\n");
854 				if (!oidp->oid_handler) {
855 					sysctl_sysctl_debug_dump_node(
856 					    SYSCTL_CHILDREN(oidp), i + 2);
857 				}
858 				break;
859 			case CTLTYPE_INT:    printf(" Int\n"); break;
860 			case CTLTYPE_UINT:   printf(" u_int\n"); break;
861 			case CTLTYPE_LONG:   printf(" Long\n"); break;
862 			case CTLTYPE_ULONG:  printf(" u_long\n"); break;
863 			case CTLTYPE_STRING: printf(" String\n"); break;
864 			case CTLTYPE_S8:     printf(" int8_t\n"); break;
865 			case CTLTYPE_S16:    printf(" int16_t\n"); break;
866 			case CTLTYPE_S32:    printf(" int32_t\n"); break;
867 			case CTLTYPE_S64:    printf(" int64_t\n"); break;
868 			case CTLTYPE_U8:     printf(" uint8_t\n"); break;
869 			case CTLTYPE_U16:    printf(" uint16_t\n"); break;
870 			case CTLTYPE_U32:    printf(" uint32_t\n"); break;
871 			case CTLTYPE_U64:    printf(" uint64_t\n"); break;
872 			case CTLTYPE_OPAQUE: printf(" Opaque/struct\n"); break;
873 			default:	     printf("\n");
874 		}
875 
876 	}
877 }
878 
879 static int
880 sysctl_sysctl_debug(SYSCTL_HANDLER_ARGS)
881 {
882 	struct rm_priotracker tracker;
883 	int error;
884 
885 	error = priv_check(req->td, PRIV_SYSCTL_DEBUG);
886 	if (error)
887 		return (error);
888 	SYSCTL_RLOCK(&tracker);
889 	sysctl_sysctl_debug_dump_node(&sysctl__children, 0);
890 	SYSCTL_RUNLOCK(&tracker);
891 	return (ENOENT);
892 }
893 
894 SYSCTL_PROC(_sysctl, 0, debug, CTLTYPE_STRING|CTLFLAG_RD|CTLFLAG_MPSAFE,
895 	0, 0, sysctl_sysctl_debug, "-", "");
896 #endif
897 
898 static int
899 sysctl_sysctl_name(SYSCTL_HANDLER_ARGS)
900 {
901 	int *name = (int *) arg1;
902 	u_int namelen = arg2;
903 	int error = 0;
904 	struct sysctl_oid *oid;
905 	struct sysctl_oid_list *lsp = &sysctl__children, *lsp2;
906 	struct rm_priotracker tracker;
907 	char buf[10];
908 
909 	SYSCTL_RLOCK(&tracker);
910 	while (namelen) {
911 		if (!lsp) {
912 			snprintf(buf,sizeof(buf),"%d",*name);
913 			if (req->oldidx)
914 				error = SYSCTL_OUT(req, ".", 1);
915 			if (!error)
916 				error = SYSCTL_OUT(req, buf, strlen(buf));
917 			if (error)
918 				goto out;
919 			namelen--;
920 			name++;
921 			continue;
922 		}
923 		lsp2 = NULL;
924 		SLIST_FOREACH(oid, lsp, oid_link) {
925 			if (oid->oid_number != *name)
926 				continue;
927 
928 			if (req->oldidx)
929 				error = SYSCTL_OUT(req, ".", 1);
930 			if (!error)
931 				error = SYSCTL_OUT(req, oid->oid_name,
932 					strlen(oid->oid_name));
933 			if (error)
934 				goto out;
935 
936 			namelen--;
937 			name++;
938 
939 			if ((oid->oid_kind & CTLTYPE) != CTLTYPE_NODE)
940 				break;
941 
942 			if (oid->oid_handler)
943 				break;
944 
945 			lsp2 = SYSCTL_CHILDREN(oid);
946 			break;
947 		}
948 		lsp = lsp2;
949 	}
950 	error = SYSCTL_OUT(req, "", 1);
951  out:
952 	SYSCTL_RUNLOCK(&tracker);
953 	return (error);
954 }
955 
956 /*
957  * XXXRW/JA: Shouldn't return name data for nodes that we don't permit in
958  * capability mode.
959  */
960 static SYSCTL_NODE(_sysctl, 1, name, CTLFLAG_RD | CTLFLAG_MPSAFE | CTLFLAG_CAPRD,
961     sysctl_sysctl_name, "");
962 
963 static int
964 sysctl_sysctl_next_ls(struct sysctl_oid_list *lsp, int *name, u_int namelen,
965 	int *next, int *len, int level, struct sysctl_oid **oidpp)
966 {
967 	struct sysctl_oid *oidp;
968 
969 	SYSCTL_ASSERT_LOCKED();
970 	*len = level;
971 	SLIST_FOREACH(oidp, lsp, oid_link) {
972 		*next = oidp->oid_number;
973 		*oidpp = oidp;
974 
975 		if (oidp->oid_kind & CTLFLAG_SKIP)
976 			continue;
977 
978 		if (!namelen) {
979 			if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
980 				return (0);
981 			if (oidp->oid_handler)
982 				/* We really should call the handler here...*/
983 				return (0);
984 			lsp = SYSCTL_CHILDREN(oidp);
985 			if (!sysctl_sysctl_next_ls(lsp, 0, 0, next+1,
986 				len, level+1, oidpp))
987 				return (0);
988 			goto emptynode;
989 		}
990 
991 		if (oidp->oid_number < *name)
992 			continue;
993 
994 		if (oidp->oid_number > *name) {
995 			if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
996 				return (0);
997 			if (oidp->oid_handler)
998 				return (0);
999 			lsp = SYSCTL_CHILDREN(oidp);
1000 			if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1,
1001 				next+1, len, level+1, oidpp))
1002 				return (0);
1003 			goto next;
1004 		}
1005 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1006 			continue;
1007 
1008 		if (oidp->oid_handler)
1009 			continue;
1010 
1011 		lsp = SYSCTL_CHILDREN(oidp);
1012 		if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1, next+1,
1013 			len, level+1, oidpp))
1014 			return (0);
1015 	next:
1016 		namelen = 1;
1017 	emptynode:
1018 		*len = level;
1019 	}
1020 	return (1);
1021 }
1022 
1023 static int
1024 sysctl_sysctl_next(SYSCTL_HANDLER_ARGS)
1025 {
1026 	int *name = (int *) arg1;
1027 	u_int namelen = arg2;
1028 	int i, j, error;
1029 	struct sysctl_oid *oid;
1030 	struct sysctl_oid_list *lsp = &sysctl__children;
1031 	struct rm_priotracker tracker;
1032 	int newoid[CTL_MAXNAME];
1033 
1034 	SYSCTL_RLOCK(&tracker);
1035 	i = sysctl_sysctl_next_ls(lsp, name, namelen, newoid, &j, 1, &oid);
1036 	SYSCTL_RUNLOCK(&tracker);
1037 	if (i)
1038 		return (ENOENT);
1039 	error = SYSCTL_OUT(req, newoid, j * sizeof (int));
1040 	return (error);
1041 }
1042 
1043 /*
1044  * XXXRW/JA: Shouldn't return next data for nodes that we don't permit in
1045  * capability mode.
1046  */
1047 static SYSCTL_NODE(_sysctl, 2, next, CTLFLAG_RD | CTLFLAG_MPSAFE | CTLFLAG_CAPRD,
1048     sysctl_sysctl_next, "");
1049 
1050 static int
1051 name2oid(char *name, int *oid, int *len, struct sysctl_oid **oidpp)
1052 {
1053 	struct sysctl_oid *oidp;
1054 	struct sysctl_oid_list *lsp = &sysctl__children;
1055 	char *p;
1056 
1057 	SYSCTL_ASSERT_LOCKED();
1058 
1059 	for (*len = 0; *len < CTL_MAXNAME;) {
1060 		p = strsep(&name, ".");
1061 
1062 		oidp = SLIST_FIRST(lsp);
1063 		for (;; oidp = SLIST_NEXT(oidp, oid_link)) {
1064 			if (oidp == NULL)
1065 				return (ENOENT);
1066 			if (strcmp(p, oidp->oid_name) == 0)
1067 				break;
1068 		}
1069 		*oid++ = oidp->oid_number;
1070 		(*len)++;
1071 
1072 		if (name == NULL || *name == '\0') {
1073 			if (oidpp)
1074 				*oidpp = oidp;
1075 			return (0);
1076 		}
1077 
1078 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1079 			break;
1080 
1081 		if (oidp->oid_handler)
1082 			break;
1083 
1084 		lsp = SYSCTL_CHILDREN(oidp);
1085 	}
1086 	return (ENOENT);
1087 }
1088 
1089 static int
1090 sysctl_sysctl_name2oid(SYSCTL_HANDLER_ARGS)
1091 {
1092 	char *p;
1093 	int error, oid[CTL_MAXNAME], len = 0;
1094 	struct sysctl_oid *op = NULL;
1095 	struct rm_priotracker tracker;
1096 
1097 	if (!req->newlen)
1098 		return (ENOENT);
1099 	if (req->newlen >= MAXPATHLEN)	/* XXX arbitrary, undocumented */
1100 		return (ENAMETOOLONG);
1101 
1102 	p = malloc(req->newlen+1, M_SYSCTL, M_WAITOK);
1103 
1104 	error = SYSCTL_IN(req, p, req->newlen);
1105 	if (error) {
1106 		free(p, M_SYSCTL);
1107 		return (error);
1108 	}
1109 
1110 	p [req->newlen] = '\0';
1111 
1112 	SYSCTL_RLOCK(&tracker);
1113 	error = name2oid(p, oid, &len, &op);
1114 	SYSCTL_RUNLOCK(&tracker);
1115 
1116 	free(p, M_SYSCTL);
1117 
1118 	if (error)
1119 		return (error);
1120 
1121 	error = SYSCTL_OUT(req, oid, len * sizeof *oid);
1122 	return (error);
1123 }
1124 
1125 /*
1126  * XXXRW/JA: Shouldn't return name2oid data for nodes that we don't permit in
1127  * capability mode.
1128  */
1129 SYSCTL_PROC(_sysctl, 3, name2oid,
1130     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_MPSAFE
1131     | CTLFLAG_CAPRW, 0, 0, sysctl_sysctl_name2oid, "I", "");
1132 
1133 static int
1134 sysctl_sysctl_oidfmt(SYSCTL_HANDLER_ARGS)
1135 {
1136 	struct sysctl_oid *oid;
1137 	struct rm_priotracker tracker;
1138 	int error;
1139 
1140 	SYSCTL_RLOCK(&tracker);
1141 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1142 	if (error)
1143 		goto out;
1144 
1145 	if (oid->oid_fmt == NULL) {
1146 		error = ENOENT;
1147 		goto out;
1148 	}
1149 	error = SYSCTL_OUT(req, &oid->oid_kind, sizeof(oid->oid_kind));
1150 	if (error)
1151 		goto out;
1152 	error = SYSCTL_OUT(req, oid->oid_fmt, strlen(oid->oid_fmt) + 1);
1153  out:
1154 	SYSCTL_RUNLOCK(&tracker);
1155 	return (error);
1156 }
1157 
1158 
1159 static SYSCTL_NODE(_sysctl, 4, oidfmt, CTLFLAG_RD|CTLFLAG_MPSAFE|CTLFLAG_CAPRD,
1160     sysctl_sysctl_oidfmt, "");
1161 
1162 static int
1163 sysctl_sysctl_oiddescr(SYSCTL_HANDLER_ARGS)
1164 {
1165 	struct sysctl_oid *oid;
1166 	struct rm_priotracker tracker;
1167 	int error;
1168 
1169 	SYSCTL_RLOCK(&tracker);
1170 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1171 	if (error)
1172 		goto out;
1173 
1174 	if (oid->oid_descr == NULL) {
1175 		error = ENOENT;
1176 		goto out;
1177 	}
1178 	error = SYSCTL_OUT(req, oid->oid_descr, strlen(oid->oid_descr) + 1);
1179  out:
1180 	SYSCTL_RUNLOCK(&tracker);
1181 	return (error);
1182 }
1183 
1184 static SYSCTL_NODE(_sysctl, 5, oiddescr, CTLFLAG_RD|CTLFLAG_MPSAFE|CTLFLAG_CAPRD,
1185     sysctl_sysctl_oiddescr, "");
1186 
1187 static int
1188 sysctl_sysctl_oidlabel(SYSCTL_HANDLER_ARGS)
1189 {
1190 	struct sysctl_oid *oid;
1191 	struct rm_priotracker tracker;
1192 	int error;
1193 
1194 	SYSCTL_RLOCK(&tracker);
1195 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1196 	if (error)
1197 		goto out;
1198 
1199 	if (oid->oid_label == NULL) {
1200 		error = ENOENT;
1201 		goto out;
1202 	}
1203 	error = SYSCTL_OUT(req, oid->oid_label, strlen(oid->oid_label) + 1);
1204  out:
1205 	SYSCTL_RUNLOCK(&tracker);
1206 	return (error);
1207 }
1208 
1209 static SYSCTL_NODE(_sysctl, 6, oidlabel,
1210     CTLFLAG_RD | CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_oidlabel, "");
1211 
1212 /*
1213  * Default "handler" functions.
1214  */
1215 
1216 /*
1217  * Handle a bool.
1218  * Two cases:
1219  *     a variable:  point arg1 at it.
1220  *     a constant:  pass it in arg2.
1221  */
1222 
1223 int
1224 sysctl_handle_bool(SYSCTL_HANDLER_ARGS)
1225 {
1226 	uint8_t temp;
1227 	int error;
1228 
1229 	/*
1230 	 * Attempt to get a coherent snapshot by making a copy of the data.
1231 	 */
1232 	if (arg1)
1233 		temp = *(bool *)arg1 ? 1 : 0;
1234 	else
1235 		temp = arg2 ? 1 : 0;
1236 
1237 	error = SYSCTL_OUT(req, &temp, sizeof(temp));
1238 	if (error || !req->newptr)
1239 		return (error);
1240 
1241 	if (!arg1)
1242 		error = EPERM;
1243 	else {
1244 		error = SYSCTL_IN(req, &temp, sizeof(temp));
1245 		if (!error)
1246 			*(bool *)arg1 = temp ? 1 : 0;
1247 	}
1248 	return (error);
1249 }
1250 
1251 /*
1252  * Handle an int8_t, signed or unsigned.
1253  * Two cases:
1254  *     a variable:  point arg1 at it.
1255  *     a constant:  pass it in arg2.
1256  */
1257 
1258 int
1259 sysctl_handle_8(SYSCTL_HANDLER_ARGS)
1260 {
1261 	int8_t tmpout;
1262 	int error = 0;
1263 
1264 	/*
1265 	 * Attempt to get a coherent snapshot by making a copy of the data.
1266 	 */
1267 	if (arg1)
1268 		tmpout = *(int8_t *)arg1;
1269 	else
1270 		tmpout = arg2;
1271 	error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout));
1272 
1273 	if (error || !req->newptr)
1274 		return (error);
1275 
1276 	if (!arg1)
1277 		error = EPERM;
1278 	else
1279 		error = SYSCTL_IN(req, arg1, sizeof(tmpout));
1280 	return (error);
1281 }
1282 
1283 /*
1284  * Handle an int16_t, signed or unsigned.
1285  * Two cases:
1286  *     a variable:  point arg1 at it.
1287  *     a constant:  pass it in arg2.
1288  */
1289 
1290 int
1291 sysctl_handle_16(SYSCTL_HANDLER_ARGS)
1292 {
1293 	int16_t tmpout;
1294 	int error = 0;
1295 
1296 	/*
1297 	 * Attempt to get a coherent snapshot by making a copy of the data.
1298 	 */
1299 	if (arg1)
1300 		tmpout = *(int16_t *)arg1;
1301 	else
1302 		tmpout = arg2;
1303 	error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout));
1304 
1305 	if (error || !req->newptr)
1306 		return (error);
1307 
1308 	if (!arg1)
1309 		error = EPERM;
1310 	else
1311 		error = SYSCTL_IN(req, arg1, sizeof(tmpout));
1312 	return (error);
1313 }
1314 
1315 /*
1316  * Handle an int32_t, signed or unsigned.
1317  * Two cases:
1318  *     a variable:  point arg1 at it.
1319  *     a constant:  pass it in arg2.
1320  */
1321 
1322 int
1323 sysctl_handle_32(SYSCTL_HANDLER_ARGS)
1324 {
1325 	int32_t tmpout;
1326 	int error = 0;
1327 
1328 	/*
1329 	 * Attempt to get a coherent snapshot by making a copy of the data.
1330 	 */
1331 	if (arg1)
1332 		tmpout = *(int32_t *)arg1;
1333 	else
1334 		tmpout = arg2;
1335 	error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout));
1336 
1337 	if (error || !req->newptr)
1338 		return (error);
1339 
1340 	if (!arg1)
1341 		error = EPERM;
1342 	else
1343 		error = SYSCTL_IN(req, arg1, sizeof(tmpout));
1344 	return (error);
1345 }
1346 
1347 /*
1348  * Handle an int, signed or unsigned.
1349  * Two cases:
1350  *     a variable:  point arg1 at it.
1351  *     a constant:  pass it in arg2.
1352  */
1353 
1354 int
1355 sysctl_handle_int(SYSCTL_HANDLER_ARGS)
1356 {
1357 	int tmpout, error = 0;
1358 
1359 	/*
1360 	 * Attempt to get a coherent snapshot by making a copy of the data.
1361 	 */
1362 	if (arg1)
1363 		tmpout = *(int *)arg1;
1364 	else
1365 		tmpout = arg2;
1366 	error = SYSCTL_OUT(req, &tmpout, sizeof(int));
1367 
1368 	if (error || !req->newptr)
1369 		return (error);
1370 
1371 	if (!arg1)
1372 		error = EPERM;
1373 	else
1374 		error = SYSCTL_IN(req, arg1, sizeof(int));
1375 	return (error);
1376 }
1377 
1378 /*
1379  * Based on on sysctl_handle_int() convert milliseconds into ticks.
1380  * Note: this is used by TCP.
1381  */
1382 
1383 int
1384 sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
1385 {
1386 	int error, s, tt;
1387 
1388 	tt = *(int *)arg1;
1389 	s = (int)((int64_t)tt * 1000 / hz);
1390 
1391 	error = sysctl_handle_int(oidp, &s, 0, req);
1392 	if (error || !req->newptr)
1393 		return (error);
1394 
1395 	tt = (int)((int64_t)s * hz / 1000);
1396 	if (tt < 1)
1397 		return (EINVAL);
1398 
1399 	*(int *)arg1 = tt;
1400 	return (0);
1401 }
1402 
1403 
1404 /*
1405  * Handle a long, signed or unsigned.
1406  * Two cases:
1407  *     a variable:  point arg1 at it.
1408  *     a constant:  pass it in arg2.
1409  */
1410 
1411 int
1412 sysctl_handle_long(SYSCTL_HANDLER_ARGS)
1413 {
1414 	int error = 0;
1415 	long tmplong;
1416 #ifdef SCTL_MASK32
1417 	int tmpint;
1418 #endif
1419 
1420 	/*
1421 	 * Attempt to get a coherent snapshot by making a copy of the data.
1422 	 */
1423 	if (arg1)
1424 		tmplong = *(long *)arg1;
1425 	else
1426 		tmplong = arg2;
1427 #ifdef SCTL_MASK32
1428 	if (req->flags & SCTL_MASK32) {
1429 		tmpint = tmplong;
1430 		error = SYSCTL_OUT(req, &tmpint, sizeof(int));
1431 	} else
1432 #endif
1433 		error = SYSCTL_OUT(req, &tmplong, sizeof(long));
1434 
1435 	if (error || !req->newptr)
1436 		return (error);
1437 
1438 	if (!arg1)
1439 		error = EPERM;
1440 #ifdef SCTL_MASK32
1441 	else if (req->flags & SCTL_MASK32) {
1442 		error = SYSCTL_IN(req, &tmpint, sizeof(int));
1443 		*(long *)arg1 = (long)tmpint;
1444 	}
1445 #endif
1446 	else
1447 		error = SYSCTL_IN(req, arg1, sizeof(long));
1448 	return (error);
1449 }
1450 
1451 /*
1452  * Handle a 64 bit int, signed or unsigned.
1453  * Two cases:
1454  *     a variable:  point arg1 at it.
1455  *     a constant:  pass it in arg2.
1456  */
1457 int
1458 sysctl_handle_64(SYSCTL_HANDLER_ARGS)
1459 {
1460 	int error = 0;
1461 	uint64_t tmpout;
1462 
1463 	/*
1464 	 * Attempt to get a coherent snapshot by making a copy of the data.
1465 	 */
1466 	if (arg1)
1467 		tmpout = *(uint64_t *)arg1;
1468 	else
1469 		tmpout = arg2;
1470 	error = SYSCTL_OUT(req, &tmpout, sizeof(uint64_t));
1471 
1472 	if (error || !req->newptr)
1473 		return (error);
1474 
1475 	if (!arg1)
1476 		error = EPERM;
1477 	else
1478 		error = SYSCTL_IN(req, arg1, sizeof(uint64_t));
1479 	return (error);
1480 }
1481 
1482 /*
1483  * Handle our generic '\0' terminated 'C' string.
1484  * Two cases:
1485  * 	a variable string:  point arg1 at it, arg2 is max length.
1486  * 	a constant string:  point arg1 at it, arg2 is zero.
1487  */
1488 
1489 int
1490 sysctl_handle_string(SYSCTL_HANDLER_ARGS)
1491 {
1492 	size_t outlen;
1493 	int error = 0, ro_string = 0;
1494 
1495 	/*
1496 	 * A zero-length buffer indicates a fixed size read-only
1497 	 * string:
1498 	 */
1499 	if (arg2 == 0) {
1500 		arg2 = strlen((char *)arg1) + 1;
1501 		ro_string = 1;
1502 	}
1503 
1504 	if (req->oldptr != NULL) {
1505 		char *tmparg;
1506 
1507 		if (ro_string) {
1508 			tmparg = arg1;
1509 		} else {
1510 			/* try to make a coherent snapshot of the string */
1511 			tmparg = malloc(arg2, M_SYSCTLTMP, M_WAITOK);
1512 			memcpy(tmparg, arg1, arg2);
1513 		}
1514 
1515 		outlen = strnlen(tmparg, arg2 - 1) + 1;
1516 		error = SYSCTL_OUT(req, tmparg, outlen);
1517 
1518 		if (!ro_string)
1519 			free(tmparg, M_SYSCTLTMP);
1520 	} else {
1521 		outlen = strnlen((char *)arg1, arg2 - 1) + 1;
1522 		error = SYSCTL_OUT(req, NULL, outlen);
1523 	}
1524 	if (error || !req->newptr)
1525 		return (error);
1526 
1527 	if ((req->newlen - req->newidx) >= arg2) {
1528 		error = EINVAL;
1529 	} else {
1530 		arg2 = (req->newlen - req->newidx);
1531 		error = SYSCTL_IN(req, arg1, arg2);
1532 		((char *)arg1)[arg2] = '\0';
1533 	}
1534 	return (error);
1535 }
1536 
1537 /*
1538  * Handle any kind of opaque data.
1539  * arg1 points to it, arg2 is the size.
1540  */
1541 
1542 int
1543 sysctl_handle_opaque(SYSCTL_HANDLER_ARGS)
1544 {
1545 	int error, tries;
1546 	u_int generation;
1547 	struct sysctl_req req2;
1548 
1549 	/*
1550 	 * Attempt to get a coherent snapshot, by using the thread
1551 	 * pre-emption counter updated from within mi_switch() to
1552 	 * determine if we were pre-empted during a bcopy() or
1553 	 * copyout(). Make 3 attempts at doing this before giving up.
1554 	 * If we encounter an error, stop immediately.
1555 	 */
1556 	tries = 0;
1557 	req2 = *req;
1558 retry:
1559 	generation = curthread->td_generation;
1560 	error = SYSCTL_OUT(req, arg1, arg2);
1561 	if (error)
1562 		return (error);
1563 	tries++;
1564 	if (generation != curthread->td_generation && tries < 3) {
1565 		*req = req2;
1566 		goto retry;
1567 	}
1568 
1569 	error = SYSCTL_IN(req, arg1, arg2);
1570 
1571 	return (error);
1572 }
1573 
1574 /*
1575  * Transfer functions to/from kernel space.
1576  * XXX: rather untested at this point
1577  */
1578 static int
1579 sysctl_old_kernel(struct sysctl_req *req, const void *p, size_t l)
1580 {
1581 	size_t i = 0;
1582 
1583 	if (req->oldptr) {
1584 		i = l;
1585 		if (req->oldlen <= req->oldidx)
1586 			i = 0;
1587 		else
1588 			if (i > req->oldlen - req->oldidx)
1589 				i = req->oldlen - req->oldidx;
1590 		if (i > 0)
1591 			bcopy(p, (char *)req->oldptr + req->oldidx, i);
1592 	}
1593 	req->oldidx += l;
1594 	if (req->oldptr && i != l)
1595 		return (ENOMEM);
1596 	return (0);
1597 }
1598 
1599 static int
1600 sysctl_new_kernel(struct sysctl_req *req, void *p, size_t l)
1601 {
1602 	if (!req->newptr)
1603 		return (0);
1604 	if (req->newlen - req->newidx < l)
1605 		return (EINVAL);
1606 	bcopy((char *)req->newptr + req->newidx, p, l);
1607 	req->newidx += l;
1608 	return (0);
1609 }
1610 
1611 int
1612 kernel_sysctl(struct thread *td, int *name, u_int namelen, void *old,
1613     size_t *oldlenp, void *new, size_t newlen, size_t *retval, int flags)
1614 {
1615 	int error = 0;
1616 	struct sysctl_req req;
1617 
1618 	bzero(&req, sizeof req);
1619 
1620 	req.td = td;
1621 	req.flags = flags;
1622 
1623 	if (oldlenp) {
1624 		req.oldlen = *oldlenp;
1625 	}
1626 	req.validlen = req.oldlen;
1627 
1628 	if (old) {
1629 		req.oldptr= old;
1630 	}
1631 
1632 	if (new != NULL) {
1633 		req.newlen = newlen;
1634 		req.newptr = new;
1635 	}
1636 
1637 	req.oldfunc = sysctl_old_kernel;
1638 	req.newfunc = sysctl_new_kernel;
1639 	req.lock = REQ_UNWIRED;
1640 
1641 	error = sysctl_root(0, name, namelen, &req);
1642 
1643 	if (req.lock == REQ_WIRED && req.validlen > 0)
1644 		vsunlock(req.oldptr, req.validlen);
1645 
1646 	if (error && error != ENOMEM)
1647 		return (error);
1648 
1649 	if (retval) {
1650 		if (req.oldptr && req.oldidx > req.validlen)
1651 			*retval = req.validlen;
1652 		else
1653 			*retval = req.oldidx;
1654 	}
1655 	return (error);
1656 }
1657 
1658 int
1659 kernel_sysctlbyname(struct thread *td, char *name, void *old, size_t *oldlenp,
1660     void *new, size_t newlen, size_t *retval, int flags)
1661 {
1662         int oid[CTL_MAXNAME];
1663         size_t oidlen, plen;
1664 	int error;
1665 
1666 	oid[0] = 0;		/* sysctl internal magic */
1667 	oid[1] = 3;		/* name2oid */
1668 	oidlen = sizeof(oid);
1669 
1670 	error = kernel_sysctl(td, oid, 2, oid, &oidlen,
1671 	    (void *)name, strlen(name), &plen, flags);
1672 	if (error)
1673 		return (error);
1674 
1675 	error = kernel_sysctl(td, oid, plen / sizeof(int), old, oldlenp,
1676 	    new, newlen, retval, flags);
1677 	return (error);
1678 }
1679 
1680 /*
1681  * Transfer function to/from user space.
1682  */
1683 static int
1684 sysctl_old_user(struct sysctl_req *req, const void *p, size_t l)
1685 {
1686 	size_t i, len, origidx;
1687 	int error;
1688 
1689 	origidx = req->oldidx;
1690 	req->oldidx += l;
1691 	if (req->oldptr == NULL)
1692 		return (0);
1693 	/*
1694 	 * If we have not wired the user supplied buffer and we are currently
1695 	 * holding locks, drop a witness warning, as it's possible that
1696 	 * write operations to the user page can sleep.
1697 	 */
1698 	if (req->lock != REQ_WIRED)
1699 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
1700 		    "sysctl_old_user()");
1701 	i = l;
1702 	len = req->validlen;
1703 	if (len <= origidx)
1704 		i = 0;
1705 	else {
1706 		if (i > len - origidx)
1707 			i = len - origidx;
1708 		if (req->lock == REQ_WIRED) {
1709 			error = copyout_nofault(p, (char *)req->oldptr +
1710 			    origidx, i);
1711 		} else
1712 			error = copyout(p, (char *)req->oldptr + origidx, i);
1713 		if (error != 0)
1714 			return (error);
1715 	}
1716 	if (i < l)
1717 		return (ENOMEM);
1718 	return (0);
1719 }
1720 
1721 static int
1722 sysctl_new_user(struct sysctl_req *req, void *p, size_t l)
1723 {
1724 	int error;
1725 
1726 	if (!req->newptr)
1727 		return (0);
1728 	if (req->newlen - req->newidx < l)
1729 		return (EINVAL);
1730 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
1731 	    "sysctl_new_user()");
1732 	error = copyin((char *)req->newptr + req->newidx, p, l);
1733 	req->newidx += l;
1734 	return (error);
1735 }
1736 
1737 /*
1738  * Wire the user space destination buffer.  If set to a value greater than
1739  * zero, the len parameter limits the maximum amount of wired memory.
1740  */
1741 int
1742 sysctl_wire_old_buffer(struct sysctl_req *req, size_t len)
1743 {
1744 	int ret;
1745 	size_t wiredlen;
1746 
1747 	wiredlen = (len > 0 && len < req->oldlen) ? len : req->oldlen;
1748 	ret = 0;
1749 	if (req->lock != REQ_WIRED && req->oldptr &&
1750 	    req->oldfunc == sysctl_old_user) {
1751 		if (wiredlen != 0) {
1752 			ret = vslock(req->oldptr, wiredlen);
1753 			if (ret != 0) {
1754 				if (ret != ENOMEM)
1755 					return (ret);
1756 				wiredlen = 0;
1757 			}
1758 		}
1759 		req->lock = REQ_WIRED;
1760 		req->validlen = wiredlen;
1761 	}
1762 	return (0);
1763 }
1764 
1765 int
1766 sysctl_find_oid(int *name, u_int namelen, struct sysctl_oid **noid,
1767     int *nindx, struct sysctl_req *req)
1768 {
1769 	struct sysctl_oid_list *lsp;
1770 	struct sysctl_oid *oid;
1771 	int indx;
1772 
1773 	SYSCTL_ASSERT_LOCKED();
1774 	lsp = &sysctl__children;
1775 	indx = 0;
1776 	while (indx < CTL_MAXNAME) {
1777 		SLIST_FOREACH(oid, lsp, oid_link) {
1778 			if (oid->oid_number == name[indx])
1779 				break;
1780 		}
1781 		if (oid == NULL)
1782 			return (ENOENT);
1783 
1784 		indx++;
1785 		if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1786 			if (oid->oid_handler != NULL || indx == namelen) {
1787 				*noid = oid;
1788 				if (nindx != NULL)
1789 					*nindx = indx;
1790 				KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
1791 				    ("%s found DYING node %p", __func__, oid));
1792 				return (0);
1793 			}
1794 			lsp = SYSCTL_CHILDREN(oid);
1795 		} else if (indx == namelen) {
1796 			*noid = oid;
1797 			if (nindx != NULL)
1798 				*nindx = indx;
1799 			KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
1800 			    ("%s found DYING node %p", __func__, oid));
1801 			return (0);
1802 		} else {
1803 			return (ENOTDIR);
1804 		}
1805 	}
1806 	return (ENOENT);
1807 }
1808 
1809 /*
1810  * Traverse our tree, and find the right node, execute whatever it points
1811  * to, and return the resulting error code.
1812  */
1813 
1814 static int
1815 sysctl_root(SYSCTL_HANDLER_ARGS)
1816 {
1817 	struct sysctl_oid *oid;
1818 	struct rm_priotracker tracker;
1819 	int error, indx, lvl;
1820 
1821 	SYSCTL_RLOCK(&tracker);
1822 
1823 	error = sysctl_find_oid(arg1, arg2, &oid, &indx, req);
1824 	if (error)
1825 		goto out;
1826 
1827 	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1828 		/*
1829 		 * You can't call a sysctl when it's a node, but has
1830 		 * no handler.  Inform the user that it's a node.
1831 		 * The indx may or may not be the same as namelen.
1832 		 */
1833 		if (oid->oid_handler == NULL) {
1834 			error = EISDIR;
1835 			goto out;
1836 		}
1837 	}
1838 
1839 	/* Is this sysctl writable? */
1840 	if (req->newptr && !(oid->oid_kind & CTLFLAG_WR)) {
1841 		error = EPERM;
1842 		goto out;
1843 	}
1844 
1845 	KASSERT(req->td != NULL, ("sysctl_root(): req->td == NULL"));
1846 
1847 #ifdef CAPABILITY_MODE
1848 	/*
1849 	 * If the process is in capability mode, then don't permit reading or
1850 	 * writing unless specifically granted for the node.
1851 	 */
1852 	if (IN_CAPABILITY_MODE(req->td)) {
1853 		if ((req->oldptr && !(oid->oid_kind & CTLFLAG_CAPRD)) ||
1854 		    (req->newptr && !(oid->oid_kind & CTLFLAG_CAPWR))) {
1855 			error = EPERM;
1856 			goto out;
1857 		}
1858 	}
1859 #endif
1860 
1861 	/* Is this sysctl sensitive to securelevels? */
1862 	if (req->newptr && (oid->oid_kind & CTLFLAG_SECURE)) {
1863 		lvl = (oid->oid_kind & CTLMASK_SECURE) >> CTLSHIFT_SECURE;
1864 		error = securelevel_gt(req->td->td_ucred, lvl);
1865 		if (error)
1866 			goto out;
1867 	}
1868 
1869 	/* Is this sysctl writable by only privileged users? */
1870 	if (req->newptr && !(oid->oid_kind & CTLFLAG_ANYBODY)) {
1871 		int priv;
1872 
1873 		if (oid->oid_kind & CTLFLAG_PRISON)
1874 			priv = PRIV_SYSCTL_WRITEJAIL;
1875 #ifdef VIMAGE
1876 		else if ((oid->oid_kind & CTLFLAG_VNET) &&
1877 		     prison_owns_vnet(req->td->td_ucred))
1878 			priv = PRIV_SYSCTL_WRITEJAIL;
1879 #endif
1880 		else
1881 			priv = PRIV_SYSCTL_WRITE;
1882 		error = priv_check(req->td, priv);
1883 		if (error)
1884 			goto out;
1885 	}
1886 
1887 	if (!oid->oid_handler) {
1888 		error = EINVAL;
1889 		goto out;
1890 	}
1891 
1892 	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1893 		arg1 = (int *)arg1 + indx;
1894 		arg2 -= indx;
1895 	} else {
1896 		arg1 = oid->oid_arg1;
1897 		arg2 = oid->oid_arg2;
1898 	}
1899 #ifdef MAC
1900 	error = mac_system_check_sysctl(req->td->td_ucred, oid, arg1, arg2,
1901 	    req);
1902 	if (error != 0)
1903 		goto out;
1904 #endif
1905 #ifdef VIMAGE
1906 	if ((oid->oid_kind & CTLFLAG_VNET) && arg1 != NULL)
1907 		arg1 = (void *)(curvnet->vnet_data_base + (uintptr_t)arg1);
1908 #endif
1909 	error = sysctl_root_handler_locked(oid, arg1, arg2, req, &tracker);
1910 
1911 out:
1912 	SYSCTL_RUNLOCK(&tracker);
1913 	return (error);
1914 }
1915 
1916 #ifndef _SYS_SYSPROTO_H_
1917 struct sysctl_args {
1918 	int	*name;
1919 	u_int	namelen;
1920 	void	*old;
1921 	size_t	*oldlenp;
1922 	void	*new;
1923 	size_t	newlen;
1924 };
1925 #endif
1926 int
1927 sys___sysctl(struct thread *td, struct sysctl_args *uap)
1928 {
1929 	int error, i, name[CTL_MAXNAME];
1930 	size_t j;
1931 
1932 	if (uap->namelen > CTL_MAXNAME || uap->namelen < 2)
1933 		return (EINVAL);
1934 
1935  	error = copyin(uap->name, &name, uap->namelen * sizeof(int));
1936  	if (error)
1937 		return (error);
1938 
1939 	error = userland_sysctl(td, name, uap->namelen,
1940 		uap->old, uap->oldlenp, 0,
1941 		uap->new, uap->newlen, &j, 0);
1942 	if (error && error != ENOMEM)
1943 		return (error);
1944 	if (uap->oldlenp) {
1945 		i = copyout(&j, uap->oldlenp, sizeof(j));
1946 		if (i)
1947 			return (i);
1948 	}
1949 	return (error);
1950 }
1951 
1952 /*
1953  * This is used from various compatibility syscalls too.  That's why name
1954  * must be in kernel space.
1955  */
1956 int
1957 userland_sysctl(struct thread *td, int *name, u_int namelen, void *old,
1958     size_t *oldlenp, int inkernel, void *new, size_t newlen, size_t *retval,
1959     int flags)
1960 {
1961 	int error = 0, memlocked;
1962 	struct sysctl_req req;
1963 
1964 	bzero(&req, sizeof req);
1965 
1966 	req.td = td;
1967 	req.flags = flags;
1968 
1969 	if (oldlenp) {
1970 		if (inkernel) {
1971 			req.oldlen = *oldlenp;
1972 		} else {
1973 			error = copyin(oldlenp, &req.oldlen, sizeof(*oldlenp));
1974 			if (error)
1975 				return (error);
1976 		}
1977 	}
1978 	req.validlen = req.oldlen;
1979 
1980 	if (old) {
1981 		if (!useracc(old, req.oldlen, VM_PROT_WRITE))
1982 			return (EFAULT);
1983 		req.oldptr= old;
1984 	}
1985 
1986 	if (new != NULL) {
1987 		if (!useracc(new, newlen, VM_PROT_READ))
1988 			return (EFAULT);
1989 		req.newlen = newlen;
1990 		req.newptr = new;
1991 	}
1992 
1993 	req.oldfunc = sysctl_old_user;
1994 	req.newfunc = sysctl_new_user;
1995 	req.lock = REQ_UNWIRED;
1996 
1997 #ifdef KTRACE
1998 	if (KTRPOINT(curthread, KTR_SYSCTL))
1999 		ktrsysctl(name, namelen);
2000 #endif
2001 
2002 	if (req.oldptr && req.oldlen > PAGE_SIZE) {
2003 		memlocked = 1;
2004 		sx_xlock(&sysctlmemlock);
2005 	} else
2006 		memlocked = 0;
2007 	CURVNET_SET(TD_TO_VNET(td));
2008 
2009 	for (;;) {
2010 		req.oldidx = 0;
2011 		req.newidx = 0;
2012 		error = sysctl_root(0, name, namelen, &req);
2013 		if (error != EAGAIN)
2014 			break;
2015 		kern_yield(PRI_USER);
2016 	}
2017 
2018 	CURVNET_RESTORE();
2019 
2020 	if (req.lock == REQ_WIRED && req.validlen > 0)
2021 		vsunlock(req.oldptr, req.validlen);
2022 	if (memlocked)
2023 		sx_xunlock(&sysctlmemlock);
2024 
2025 	if (error && error != ENOMEM)
2026 		return (error);
2027 
2028 	if (retval) {
2029 		if (req.oldptr && req.oldidx > req.validlen)
2030 			*retval = req.validlen;
2031 		else
2032 			*retval = req.oldidx;
2033 	}
2034 	return (error);
2035 }
2036 
2037 /*
2038  * Drain into a sysctl struct.  The user buffer should be wired if a page
2039  * fault would cause issue.
2040  */
2041 static int
2042 sbuf_sysctl_drain(void *arg, const char *data, int len)
2043 {
2044 	struct sysctl_req *req = arg;
2045 	int error;
2046 
2047 	error = SYSCTL_OUT(req, data, len);
2048 	KASSERT(error >= 0, ("Got unexpected negative value %d", error));
2049 	return (error == 0 ? len : -error);
2050 }
2051 
2052 struct sbuf *
2053 sbuf_new_for_sysctl(struct sbuf *s, char *buf, int length,
2054     struct sysctl_req *req)
2055 {
2056 
2057 	/* Supply a default buffer size if none given. */
2058 	if (buf == NULL && length == 0)
2059 		length = 64;
2060 	s = sbuf_new(s, buf, length, SBUF_FIXEDLEN | SBUF_INCLUDENUL);
2061 	sbuf_set_drain(s, sbuf_sysctl_drain, req);
2062 	return (s);
2063 }
2064