kern_sysctl.c revision 316128
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 * 4. 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: stable/11/sys/kern/kern_sysctl.c 316128 2017-03-29 08:00:13Z ngie $");
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
73static MALLOC_DEFINE(M_SYSCTL, "sysctl", "sysctl internal magic");
74static MALLOC_DEFINE(M_SYSCTLOID, "sysctloid", "sysctl dynamic oids");
75static 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 */
90static struct rmlock sysctllock;
91static 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
106static int sysctl_root(SYSCTL_HANDLER_ARGS);
107
108/* Root list */
109struct sysctl_oid_list sysctl__children = SLIST_HEAD_INITIALIZER(&sysctl__children);
110
111static int	sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del,
112		    int recurse);
113static int	sysctl_old_kernel(struct sysctl_req *, const void *, size_t);
114static int	sysctl_new_kernel(struct sysctl_req *, void *, size_t);
115
116static struct sysctl_oid *
117sysctl_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 */
135void
136sysctl_wlock(void)
137{
138
139	SYSCTL_WLOCK();
140}
141
142void
143sysctl_wunlock(void)
144{
145
146	SYSCTL_WUNLOCK();
147}
148
149static int
150sysctl_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
185static void
186sysctl_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
326void
327sysctl_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	 */
380retry:
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
427void
428sysctl_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. */
458int
459sysctl_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 */
476int
477sysctl_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 */
528struct sysctl_ctx_entry *
529sysctl_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 */
543struct sysctl_ctx_entry *
544sysctl_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 */
563int
564sysctl_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 */
589int
590sysctl_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
600int
601sysctl_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
621static int
622sysctl_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			free(__DECONST(char *, oidp->oid_name), M_SYSCTLOID);
683			free(oidp, M_SYSCTLOID);
684		}
685	}
686	return (0);
687}
688/*
689 * Create new sysctls at run time.
690 * clist may point to a valid context initialized with sysctl_ctx_init().
691 */
692struct sysctl_oid *
693sysctl_add_oid(struct sysctl_ctx_list *clist, struct sysctl_oid_list *parent,
694	int number, const char *name, int kind, void *arg1, intmax_t arg2,
695	int (*handler)(SYSCTL_HANDLER_ARGS), const char *fmt, const char *descr)
696{
697	struct sysctl_oid *oidp;
698
699	/* You have to hook up somewhere.. */
700	if (parent == NULL)
701		return(NULL);
702	/* Check if the node already exists, otherwise create it */
703	SYSCTL_WLOCK();
704	oidp = sysctl_find_oidname(name, parent);
705	if (oidp != NULL) {
706		if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
707			oidp->oid_refcnt++;
708			/* Update the context */
709			if (clist != NULL)
710				sysctl_ctx_entry_add(clist, oidp);
711			SYSCTL_WUNLOCK();
712			return (oidp);
713		} else {
714			SYSCTL_WUNLOCK();
715			printf("can't re-use a leaf (%s)!\n", name);
716			return (NULL);
717		}
718	}
719	oidp = malloc(sizeof(struct sysctl_oid), M_SYSCTLOID, M_WAITOK|M_ZERO);
720	oidp->oid_parent = parent;
721	SLIST_INIT(&oidp->oid_children);
722	oidp->oid_number = number;
723	oidp->oid_refcnt = 1;
724	oidp->oid_name = strdup(name, M_SYSCTLOID);
725	oidp->oid_handler = handler;
726	oidp->oid_kind = CTLFLAG_DYN | kind;
727	oidp->oid_arg1 = arg1;
728	oidp->oid_arg2 = arg2;
729	oidp->oid_fmt = fmt;
730	if (descr != NULL)
731		oidp->oid_descr = strdup(descr, M_SYSCTLOID);
732	/* Update the context, if used */
733	if (clist != NULL)
734		sysctl_ctx_entry_add(clist, oidp);
735	/* Register this oid */
736	sysctl_register_oid(oidp);
737	SYSCTL_WUNLOCK();
738	return (oidp);
739}
740
741/*
742 * Rename an existing oid.
743 */
744void
745sysctl_rename_oid(struct sysctl_oid *oidp, const char *name)
746{
747	char *newname;
748	char *oldname;
749
750	newname = strdup(name, M_SYSCTLOID);
751	SYSCTL_WLOCK();
752	oldname = __DECONST(char *, oidp->oid_name);
753	oidp->oid_name = newname;
754	SYSCTL_WUNLOCK();
755	free(oldname, M_SYSCTLOID);
756}
757
758/*
759 * Reparent an existing oid.
760 */
761int
762sysctl_move_oid(struct sysctl_oid *oid, struct sysctl_oid_list *parent)
763{
764	struct sysctl_oid *oidp;
765
766	SYSCTL_WLOCK();
767	if (oid->oid_parent == parent) {
768		SYSCTL_WUNLOCK();
769		return (0);
770	}
771	oidp = sysctl_find_oidname(oid->oid_name, parent);
772	if (oidp != NULL) {
773		SYSCTL_WUNLOCK();
774		return (EEXIST);
775	}
776	sysctl_unregister_oid(oid);
777	oid->oid_parent = parent;
778	oid->oid_number = OID_AUTO;
779	sysctl_register_oid(oid);
780	SYSCTL_WUNLOCK();
781	return (0);
782}
783
784/*
785 * Register the kernel's oids on startup.
786 */
787SET_DECLARE(sysctl_set, struct sysctl_oid);
788
789static void
790sysctl_register_all(void *arg)
791{
792	struct sysctl_oid **oidp;
793
794	sx_init(&sysctlmemlock, "sysctl mem");
795	SYSCTL_INIT();
796	SYSCTL_WLOCK();
797	SET_FOREACH(oidp, sysctl_set)
798		sysctl_register_oid(*oidp);
799	SYSCTL_WUNLOCK();
800}
801SYSINIT(sysctl, SI_SUB_KMEM, SI_ORDER_FIRST, sysctl_register_all, 0);
802
803/*
804 * "Staff-functions"
805 *
806 * These functions implement a presently undocumented interface
807 * used by the sysctl program to walk the tree, and get the type
808 * so it can print the value.
809 * This interface is under work and consideration, and should probably
810 * be killed with a big axe by the first person who can find the time.
811 * (be aware though, that the proper interface isn't as obvious as it
812 * may seem, there are various conflicting requirements.
813 *
814 * {0,0}	printf the entire MIB-tree.
815 * {0,1,...}	return the name of the "..." OID.
816 * {0,2,...}	return the next OID.
817 * {0,3}	return the OID of the name in "new"
818 * {0,4,...}	return the kind & format info for the "..." OID.
819 * {0,5,...}	return the description the "..." OID.
820 */
821
822#ifdef SYSCTL_DEBUG
823static void
824sysctl_sysctl_debug_dump_node(struct sysctl_oid_list *l, int i)
825{
826	int k;
827	struct sysctl_oid *oidp;
828
829	SYSCTL_ASSERT_LOCKED();
830	SLIST_FOREACH(oidp, l, oid_link) {
831
832		for (k=0; k<i; k++)
833			printf(" ");
834
835		printf("%d %s ", oidp->oid_number, oidp->oid_name);
836
837		printf("%c%c",
838			oidp->oid_kind & CTLFLAG_RD ? 'R':' ',
839			oidp->oid_kind & CTLFLAG_WR ? 'W':' ');
840
841		if (oidp->oid_handler)
842			printf(" *Handler");
843
844		switch (oidp->oid_kind & CTLTYPE) {
845			case CTLTYPE_NODE:
846				printf(" Node\n");
847				if (!oidp->oid_handler) {
848					sysctl_sysctl_debug_dump_node(
849					    SYSCTL_CHILDREN(oidp), i + 2);
850				}
851				break;
852			case CTLTYPE_INT:    printf(" Int\n"); break;
853			case CTLTYPE_UINT:   printf(" u_int\n"); break;
854			case CTLTYPE_LONG:   printf(" Long\n"); break;
855			case CTLTYPE_ULONG:  printf(" u_long\n"); break;
856			case CTLTYPE_STRING: printf(" String\n"); break;
857			case CTLTYPE_S8:     printf(" int8_t\n"); break;
858			case CTLTYPE_S16:    printf(" int16_t\n"); break;
859			case CTLTYPE_S32:    printf(" int32_t\n"); break;
860			case CTLTYPE_S64:    printf(" int64_t\n"); break;
861			case CTLTYPE_U8:     printf(" uint8_t\n"); break;
862			case CTLTYPE_U16:    printf(" uint16_t\n"); break;
863			case CTLTYPE_U32:    printf(" uint32_t\n"); break;
864			case CTLTYPE_U64:    printf(" uint64_t\n"); break;
865			case CTLTYPE_OPAQUE: printf(" Opaque/struct\n"); break;
866			default:	     printf("\n");
867		}
868
869	}
870}
871
872static int
873sysctl_sysctl_debug(SYSCTL_HANDLER_ARGS)
874{
875	struct rm_priotracker tracker;
876	int error;
877
878	error = priv_check(req->td, PRIV_SYSCTL_DEBUG);
879	if (error)
880		return (error);
881	SYSCTL_RLOCK(&tracker);
882	sysctl_sysctl_debug_dump_node(&sysctl__children, 0);
883	SYSCTL_RUNLOCK(&tracker);
884	return (ENOENT);
885}
886
887SYSCTL_PROC(_sysctl, 0, debug, CTLTYPE_STRING|CTLFLAG_RD|CTLFLAG_MPSAFE,
888	0, 0, sysctl_sysctl_debug, "-", "");
889#endif
890
891static int
892sysctl_sysctl_name(SYSCTL_HANDLER_ARGS)
893{
894	int *name = (int *) arg1;
895	u_int namelen = arg2;
896	int error = 0;
897	struct sysctl_oid *oid;
898	struct sysctl_oid_list *lsp = &sysctl__children, *lsp2;
899	struct rm_priotracker tracker;
900	char buf[10];
901
902	SYSCTL_RLOCK(&tracker);
903	while (namelen) {
904		if (!lsp) {
905			snprintf(buf,sizeof(buf),"%d",*name);
906			if (req->oldidx)
907				error = SYSCTL_OUT(req, ".", 1);
908			if (!error)
909				error = SYSCTL_OUT(req, buf, strlen(buf));
910			if (error)
911				goto out;
912			namelen--;
913			name++;
914			continue;
915		}
916		lsp2 = NULL;
917		SLIST_FOREACH(oid, lsp, oid_link) {
918			if (oid->oid_number != *name)
919				continue;
920
921			if (req->oldidx)
922				error = SYSCTL_OUT(req, ".", 1);
923			if (!error)
924				error = SYSCTL_OUT(req, oid->oid_name,
925					strlen(oid->oid_name));
926			if (error)
927				goto out;
928
929			namelen--;
930			name++;
931
932			if ((oid->oid_kind & CTLTYPE) != CTLTYPE_NODE)
933				break;
934
935			if (oid->oid_handler)
936				break;
937
938			lsp2 = SYSCTL_CHILDREN(oid);
939			break;
940		}
941		lsp = lsp2;
942	}
943	error = SYSCTL_OUT(req, "", 1);
944 out:
945	SYSCTL_RUNLOCK(&tracker);
946	return (error);
947}
948
949/*
950 * XXXRW/JA: Shouldn't return name data for nodes that we don't permit in
951 * capability mode.
952 */
953static SYSCTL_NODE(_sysctl, 1, name, CTLFLAG_RD | CTLFLAG_MPSAFE | CTLFLAG_CAPRD,
954    sysctl_sysctl_name, "");
955
956static int
957sysctl_sysctl_next_ls(struct sysctl_oid_list *lsp, int *name, u_int namelen,
958	int *next, int *len, int level, struct sysctl_oid **oidpp)
959{
960	struct sysctl_oid *oidp;
961
962	SYSCTL_ASSERT_LOCKED();
963	*len = level;
964	SLIST_FOREACH(oidp, lsp, oid_link) {
965		*next = oidp->oid_number;
966		*oidpp = oidp;
967
968		if (oidp->oid_kind & CTLFLAG_SKIP)
969			continue;
970
971		if (!namelen) {
972			if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
973				return (0);
974			if (oidp->oid_handler)
975				/* We really should call the handler here...*/
976				return (0);
977			lsp = SYSCTL_CHILDREN(oidp);
978			if (!sysctl_sysctl_next_ls(lsp, 0, 0, next+1,
979				len, level+1, oidpp))
980				return (0);
981			goto emptynode;
982		}
983
984		if (oidp->oid_number < *name)
985			continue;
986
987		if (oidp->oid_number > *name) {
988			if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
989				return (0);
990			if (oidp->oid_handler)
991				return (0);
992			lsp = SYSCTL_CHILDREN(oidp);
993			if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1,
994				next+1, len, level+1, oidpp))
995				return (0);
996			goto next;
997		}
998		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
999			continue;
1000
1001		if (oidp->oid_handler)
1002			continue;
1003
1004		lsp = SYSCTL_CHILDREN(oidp);
1005		if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1, next+1,
1006			len, level+1, oidpp))
1007			return (0);
1008	next:
1009		namelen = 1;
1010	emptynode:
1011		*len = level;
1012	}
1013	return (1);
1014}
1015
1016static int
1017sysctl_sysctl_next(SYSCTL_HANDLER_ARGS)
1018{
1019	int *name = (int *) arg1;
1020	u_int namelen = arg2;
1021	int i, j, error;
1022	struct sysctl_oid *oid;
1023	struct sysctl_oid_list *lsp = &sysctl__children;
1024	struct rm_priotracker tracker;
1025	int newoid[CTL_MAXNAME];
1026
1027	SYSCTL_RLOCK(&tracker);
1028	i = sysctl_sysctl_next_ls(lsp, name, namelen, newoid, &j, 1, &oid);
1029	SYSCTL_RUNLOCK(&tracker);
1030	if (i)
1031		return (ENOENT);
1032	error = SYSCTL_OUT(req, newoid, j * sizeof (int));
1033	return (error);
1034}
1035
1036/*
1037 * XXXRW/JA: Shouldn't return next data for nodes that we don't permit in
1038 * capability mode.
1039 */
1040static SYSCTL_NODE(_sysctl, 2, next, CTLFLAG_RD | CTLFLAG_MPSAFE | CTLFLAG_CAPRD,
1041    sysctl_sysctl_next, "");
1042
1043static int
1044name2oid(char *name, int *oid, int *len, struct sysctl_oid **oidpp)
1045{
1046	struct sysctl_oid *oidp;
1047	struct sysctl_oid_list *lsp = &sysctl__children;
1048	char *p;
1049
1050	SYSCTL_ASSERT_LOCKED();
1051
1052	for (*len = 0; *len < CTL_MAXNAME;) {
1053		p = strsep(&name, ".");
1054
1055		oidp = SLIST_FIRST(lsp);
1056		for (;; oidp = SLIST_NEXT(oidp, oid_link)) {
1057			if (oidp == NULL)
1058				return (ENOENT);
1059			if (strcmp(p, oidp->oid_name) == 0)
1060				break;
1061		}
1062		*oid++ = oidp->oid_number;
1063		(*len)++;
1064
1065		if (name == NULL || *name == '\0') {
1066			if (oidpp)
1067				*oidpp = oidp;
1068			return (0);
1069		}
1070
1071		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1072			break;
1073
1074		if (oidp->oid_handler)
1075			break;
1076
1077		lsp = SYSCTL_CHILDREN(oidp);
1078	}
1079	return (ENOENT);
1080}
1081
1082static int
1083sysctl_sysctl_name2oid(SYSCTL_HANDLER_ARGS)
1084{
1085	char *p;
1086	int error, oid[CTL_MAXNAME], len = 0;
1087	struct sysctl_oid *op = NULL;
1088	struct rm_priotracker tracker;
1089
1090	if (!req->newlen)
1091		return (ENOENT);
1092	if (req->newlen >= MAXPATHLEN)	/* XXX arbitrary, undocumented */
1093		return (ENAMETOOLONG);
1094
1095	p = malloc(req->newlen+1, M_SYSCTL, M_WAITOK);
1096
1097	error = SYSCTL_IN(req, p, req->newlen);
1098	if (error) {
1099		free(p, M_SYSCTL);
1100		return (error);
1101	}
1102
1103	p [req->newlen] = '\0';
1104
1105	SYSCTL_RLOCK(&tracker);
1106	error = name2oid(p, oid, &len, &op);
1107	SYSCTL_RUNLOCK(&tracker);
1108
1109	free(p, M_SYSCTL);
1110
1111	if (error)
1112		return (error);
1113
1114	error = SYSCTL_OUT(req, oid, len * sizeof *oid);
1115	return (error);
1116}
1117
1118/*
1119 * XXXRW/JA: Shouldn't return name2oid data for nodes that we don't permit in
1120 * capability mode.
1121 */
1122SYSCTL_PROC(_sysctl, 3, name2oid,
1123    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_MPSAFE
1124    | CTLFLAG_CAPRW, 0, 0, sysctl_sysctl_name2oid, "I", "");
1125
1126static int
1127sysctl_sysctl_oidfmt(SYSCTL_HANDLER_ARGS)
1128{
1129	struct sysctl_oid *oid;
1130	struct rm_priotracker tracker;
1131	int error;
1132
1133	SYSCTL_RLOCK(&tracker);
1134	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1135	if (error)
1136		goto out;
1137
1138	if (oid->oid_fmt == NULL) {
1139		error = ENOENT;
1140		goto out;
1141	}
1142	error = SYSCTL_OUT(req, &oid->oid_kind, sizeof(oid->oid_kind));
1143	if (error)
1144		goto out;
1145	error = SYSCTL_OUT(req, oid->oid_fmt, strlen(oid->oid_fmt) + 1);
1146 out:
1147	SYSCTL_RUNLOCK(&tracker);
1148	return (error);
1149}
1150
1151
1152static SYSCTL_NODE(_sysctl, 4, oidfmt, CTLFLAG_RD|CTLFLAG_MPSAFE|CTLFLAG_CAPRD,
1153    sysctl_sysctl_oidfmt, "");
1154
1155static int
1156sysctl_sysctl_oiddescr(SYSCTL_HANDLER_ARGS)
1157{
1158	struct sysctl_oid *oid;
1159	struct rm_priotracker tracker;
1160	int error;
1161
1162	SYSCTL_RLOCK(&tracker);
1163	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1164	if (error)
1165		goto out;
1166
1167	if (oid->oid_descr == NULL) {
1168		error = ENOENT;
1169		goto out;
1170	}
1171	error = SYSCTL_OUT(req, oid->oid_descr, strlen(oid->oid_descr) + 1);
1172 out:
1173	SYSCTL_RUNLOCK(&tracker);
1174	return (error);
1175}
1176
1177static SYSCTL_NODE(_sysctl, 5, oiddescr, CTLFLAG_RD|CTLFLAG_MPSAFE|CTLFLAG_CAPRD,
1178    sysctl_sysctl_oiddescr, "");
1179
1180/*
1181 * Default "handler" functions.
1182 */
1183
1184/*
1185 * Handle a bool.
1186 * Two cases:
1187 *     a variable:  point arg1 at it.
1188 *     a constant:  pass it in arg2.
1189 */
1190
1191int
1192sysctl_handle_bool(SYSCTL_HANDLER_ARGS)
1193{
1194	uint8_t temp;
1195	int error;
1196
1197	/*
1198	 * Attempt to get a coherent snapshot by making a copy of the data.
1199	 */
1200	if (arg1)
1201		temp = *(bool *)arg1 ? 1 : 0;
1202	else
1203		temp = arg2 ? 1 : 0;
1204
1205	error = SYSCTL_OUT(req, &temp, sizeof(temp));
1206	if (error || !req->newptr)
1207		return (error);
1208
1209	if (!arg1)
1210		error = EPERM;
1211	else {
1212		error = SYSCTL_IN(req, &temp, sizeof(temp));
1213		if (!error)
1214			*(bool *)arg1 = temp ? 1 : 0;
1215	}
1216	return (error);
1217}
1218
1219/*
1220 * Handle an int8_t, signed or unsigned.
1221 * Two cases:
1222 *     a variable:  point arg1 at it.
1223 *     a constant:  pass it in arg2.
1224 */
1225
1226int
1227sysctl_handle_8(SYSCTL_HANDLER_ARGS)
1228{
1229	int8_t tmpout;
1230	int error = 0;
1231
1232	/*
1233	 * Attempt to get a coherent snapshot by making a copy of the data.
1234	 */
1235	if (arg1)
1236		tmpout = *(int8_t *)arg1;
1237	else
1238		tmpout = arg2;
1239	error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout));
1240
1241	if (error || !req->newptr)
1242		return (error);
1243
1244	if (!arg1)
1245		error = EPERM;
1246	else
1247		error = SYSCTL_IN(req, arg1, sizeof(tmpout));
1248	return (error);
1249}
1250
1251/*
1252 * Handle an int16_t, signed or unsigned.
1253 * Two cases:
1254 *     a variable:  point arg1 at it.
1255 *     a constant:  pass it in arg2.
1256 */
1257
1258int
1259sysctl_handle_16(SYSCTL_HANDLER_ARGS)
1260{
1261	int16_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 = *(int16_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 int32_t, signed or unsigned.
1285 * Two cases:
1286 *     a variable:  point arg1 at it.
1287 *     a constant:  pass it in arg2.
1288 */
1289
1290int
1291sysctl_handle_32(SYSCTL_HANDLER_ARGS)
1292{
1293	int32_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 = *(int32_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 int, signed or unsigned.
1317 * Two cases:
1318 *     a variable:  point arg1 at it.
1319 *     a constant:  pass it in arg2.
1320 */
1321
1322int
1323sysctl_handle_int(SYSCTL_HANDLER_ARGS)
1324{
1325	int tmpout, error = 0;
1326
1327	/*
1328	 * Attempt to get a coherent snapshot by making a copy of the data.
1329	 */
1330	if (arg1)
1331		tmpout = *(int *)arg1;
1332	else
1333		tmpout = arg2;
1334	error = SYSCTL_OUT(req, &tmpout, sizeof(int));
1335
1336	if (error || !req->newptr)
1337		return (error);
1338
1339	if (!arg1)
1340		error = EPERM;
1341	else
1342		error = SYSCTL_IN(req, arg1, sizeof(int));
1343	return (error);
1344}
1345
1346/*
1347 * Based on on sysctl_handle_int() convert milliseconds into ticks.
1348 * Note: this is used by TCP.
1349 */
1350
1351int
1352sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
1353{
1354	int error, s, tt;
1355
1356	tt = *(int *)arg1;
1357	s = (int)((int64_t)tt * 1000 / hz);
1358
1359	error = sysctl_handle_int(oidp, &s, 0, req);
1360	if (error || !req->newptr)
1361		return (error);
1362
1363	tt = (int)((int64_t)s * hz / 1000);
1364	if (tt < 1)
1365		return (EINVAL);
1366
1367	*(int *)arg1 = tt;
1368	return (0);
1369}
1370
1371
1372/*
1373 * Handle a long, signed or unsigned.
1374 * Two cases:
1375 *     a variable:  point arg1 at it.
1376 *     a constant:  pass it in arg2.
1377 */
1378
1379int
1380sysctl_handle_long(SYSCTL_HANDLER_ARGS)
1381{
1382	int error = 0;
1383	long tmplong;
1384#ifdef SCTL_MASK32
1385	int tmpint;
1386#endif
1387
1388	/*
1389	 * Attempt to get a coherent snapshot by making a copy of the data.
1390	 */
1391	if (arg1)
1392		tmplong = *(long *)arg1;
1393	else
1394		tmplong = arg2;
1395#ifdef SCTL_MASK32
1396	if (req->flags & SCTL_MASK32) {
1397		tmpint = tmplong;
1398		error = SYSCTL_OUT(req, &tmpint, sizeof(int));
1399	} else
1400#endif
1401		error = SYSCTL_OUT(req, &tmplong, sizeof(long));
1402
1403	if (error || !req->newptr)
1404		return (error);
1405
1406	if (!arg1)
1407		error = EPERM;
1408#ifdef SCTL_MASK32
1409	else if (req->flags & SCTL_MASK32) {
1410		error = SYSCTL_IN(req, &tmpint, sizeof(int));
1411		*(long *)arg1 = (long)tmpint;
1412	}
1413#endif
1414	else
1415		error = SYSCTL_IN(req, arg1, sizeof(long));
1416	return (error);
1417}
1418
1419/*
1420 * Handle a 64 bit int, signed or unsigned.
1421 * Two cases:
1422 *     a variable:  point arg1 at it.
1423 *     a constant:  pass it in arg2.
1424 */
1425int
1426sysctl_handle_64(SYSCTL_HANDLER_ARGS)
1427{
1428	int error = 0;
1429	uint64_t tmpout;
1430
1431	/*
1432	 * Attempt to get a coherent snapshot by making a copy of the data.
1433	 */
1434	if (arg1)
1435		tmpout = *(uint64_t *)arg1;
1436	else
1437		tmpout = arg2;
1438	error = SYSCTL_OUT(req, &tmpout, sizeof(uint64_t));
1439
1440	if (error || !req->newptr)
1441		return (error);
1442
1443	if (!arg1)
1444		error = EPERM;
1445	else
1446		error = SYSCTL_IN(req, arg1, sizeof(uint64_t));
1447	return (error);
1448}
1449
1450/*
1451 * Handle our generic '\0' terminated 'C' string.
1452 * Two cases:
1453 * 	a variable string:  point arg1 at it, arg2 is max length.
1454 * 	a constant string:  point arg1 at it, arg2 is zero.
1455 */
1456
1457int
1458sysctl_handle_string(SYSCTL_HANDLER_ARGS)
1459{
1460	size_t outlen;
1461	int error = 0, ro_string = 0;
1462
1463	/*
1464	 * A zero-length buffer indicates a fixed size read-only
1465	 * string:
1466	 */
1467	if (arg2 == 0) {
1468		arg2 = strlen((char *)arg1) + 1;
1469		ro_string = 1;
1470	}
1471
1472	if (req->oldptr != NULL) {
1473		char *tmparg;
1474
1475		if (ro_string) {
1476			tmparg = arg1;
1477		} else {
1478			/* try to make a coherent snapshot of the string */
1479			tmparg = malloc(arg2, M_SYSCTLTMP, M_WAITOK);
1480			memcpy(tmparg, arg1, arg2);
1481		}
1482
1483		outlen = strnlen(tmparg, arg2 - 1) + 1;
1484		error = SYSCTL_OUT(req, tmparg, outlen);
1485
1486		if (!ro_string)
1487			free(tmparg, M_SYSCTLTMP);
1488	} else {
1489		outlen = strnlen((char *)arg1, arg2 - 1) + 1;
1490		error = SYSCTL_OUT(req, NULL, outlen);
1491	}
1492	if (error || !req->newptr)
1493		return (error);
1494
1495	if ((req->newlen - req->newidx) >= arg2) {
1496		error = EINVAL;
1497	} else {
1498		arg2 = (req->newlen - req->newidx);
1499		error = SYSCTL_IN(req, arg1, arg2);
1500		((char *)arg1)[arg2] = '\0';
1501	}
1502	return (error);
1503}
1504
1505/*
1506 * Handle any kind of opaque data.
1507 * arg1 points to it, arg2 is the size.
1508 */
1509
1510int
1511sysctl_handle_opaque(SYSCTL_HANDLER_ARGS)
1512{
1513	int error, tries;
1514	u_int generation;
1515	struct sysctl_req req2;
1516
1517	/*
1518	 * Attempt to get a coherent snapshot, by using the thread
1519	 * pre-emption counter updated from within mi_switch() to
1520	 * determine if we were pre-empted during a bcopy() or
1521	 * copyout(). Make 3 attempts at doing this before giving up.
1522	 * If we encounter an error, stop immediately.
1523	 */
1524	tries = 0;
1525	req2 = *req;
1526retry:
1527	generation = curthread->td_generation;
1528	error = SYSCTL_OUT(req, arg1, arg2);
1529	if (error)
1530		return (error);
1531	tries++;
1532	if (generation != curthread->td_generation && tries < 3) {
1533		*req = req2;
1534		goto retry;
1535	}
1536
1537	error = SYSCTL_IN(req, arg1, arg2);
1538
1539	return (error);
1540}
1541
1542/*
1543 * Transfer functions to/from kernel space.
1544 * XXX: rather untested at this point
1545 */
1546static int
1547sysctl_old_kernel(struct sysctl_req *req, const void *p, size_t l)
1548{
1549	size_t i = 0;
1550
1551	if (req->oldptr) {
1552		i = l;
1553		if (req->oldlen <= req->oldidx)
1554			i = 0;
1555		else
1556			if (i > req->oldlen - req->oldidx)
1557				i = req->oldlen - req->oldidx;
1558		if (i > 0)
1559			bcopy(p, (char *)req->oldptr + req->oldidx, i);
1560	}
1561	req->oldidx += l;
1562	if (req->oldptr && i != l)
1563		return (ENOMEM);
1564	return (0);
1565}
1566
1567static int
1568sysctl_new_kernel(struct sysctl_req *req, void *p, size_t l)
1569{
1570	if (!req->newptr)
1571		return (0);
1572	if (req->newlen - req->newidx < l)
1573		return (EINVAL);
1574	bcopy((char *)req->newptr + req->newidx, p, l);
1575	req->newidx += l;
1576	return (0);
1577}
1578
1579int
1580kernel_sysctl(struct thread *td, int *name, u_int namelen, void *old,
1581    size_t *oldlenp, void *new, size_t newlen, size_t *retval, int flags)
1582{
1583	int error = 0;
1584	struct sysctl_req req;
1585
1586	bzero(&req, sizeof req);
1587
1588	req.td = td;
1589	req.flags = flags;
1590
1591	if (oldlenp) {
1592		req.oldlen = *oldlenp;
1593	}
1594	req.validlen = req.oldlen;
1595
1596	if (old) {
1597		req.oldptr= old;
1598	}
1599
1600	if (new != NULL) {
1601		req.newlen = newlen;
1602		req.newptr = new;
1603	}
1604
1605	req.oldfunc = sysctl_old_kernel;
1606	req.newfunc = sysctl_new_kernel;
1607	req.lock = REQ_UNWIRED;
1608
1609	error = sysctl_root(0, name, namelen, &req);
1610
1611	if (req.lock == REQ_WIRED && req.validlen > 0)
1612		vsunlock(req.oldptr, req.validlen);
1613
1614	if (error && error != ENOMEM)
1615		return (error);
1616
1617	if (retval) {
1618		if (req.oldptr && req.oldidx > req.validlen)
1619			*retval = req.validlen;
1620		else
1621			*retval = req.oldidx;
1622	}
1623	return (error);
1624}
1625
1626int
1627kernel_sysctlbyname(struct thread *td, char *name, void *old, size_t *oldlenp,
1628    void *new, size_t newlen, size_t *retval, int flags)
1629{
1630        int oid[CTL_MAXNAME];
1631        size_t oidlen, plen;
1632	int error;
1633
1634	oid[0] = 0;		/* sysctl internal magic */
1635	oid[1] = 3;		/* name2oid */
1636	oidlen = sizeof(oid);
1637
1638	error = kernel_sysctl(td, oid, 2, oid, &oidlen,
1639	    (void *)name, strlen(name), &plen, flags);
1640	if (error)
1641		return (error);
1642
1643	error = kernel_sysctl(td, oid, plen / sizeof(int), old, oldlenp,
1644	    new, newlen, retval, flags);
1645	return (error);
1646}
1647
1648/*
1649 * Transfer function to/from user space.
1650 */
1651static int
1652sysctl_old_user(struct sysctl_req *req, const void *p, size_t l)
1653{
1654	size_t i, len, origidx;
1655	int error;
1656
1657	origidx = req->oldidx;
1658	req->oldidx += l;
1659	if (req->oldptr == NULL)
1660		return (0);
1661	/*
1662	 * If we have not wired the user supplied buffer and we are currently
1663	 * holding locks, drop a witness warning, as it's possible that
1664	 * write operations to the user page can sleep.
1665	 */
1666	if (req->lock != REQ_WIRED)
1667		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
1668		    "sysctl_old_user()");
1669	i = l;
1670	len = req->validlen;
1671	if (len <= origidx)
1672		i = 0;
1673	else {
1674		if (i > len - origidx)
1675			i = len - origidx;
1676		if (req->lock == REQ_WIRED) {
1677			error = copyout_nofault(p, (char *)req->oldptr +
1678			    origidx, i);
1679		} else
1680			error = copyout(p, (char *)req->oldptr + origidx, i);
1681		if (error != 0)
1682			return (error);
1683	}
1684	if (i < l)
1685		return (ENOMEM);
1686	return (0);
1687}
1688
1689static int
1690sysctl_new_user(struct sysctl_req *req, void *p, size_t l)
1691{
1692	int error;
1693
1694	if (!req->newptr)
1695		return (0);
1696	if (req->newlen - req->newidx < l)
1697		return (EINVAL);
1698	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
1699	    "sysctl_new_user()");
1700	error = copyin((char *)req->newptr + req->newidx, p, l);
1701	req->newidx += l;
1702	return (error);
1703}
1704
1705/*
1706 * Wire the user space destination buffer.  If set to a value greater than
1707 * zero, the len parameter limits the maximum amount of wired memory.
1708 */
1709int
1710sysctl_wire_old_buffer(struct sysctl_req *req, size_t len)
1711{
1712	int ret;
1713	size_t wiredlen;
1714
1715	wiredlen = (len > 0 && len < req->oldlen) ? len : req->oldlen;
1716	ret = 0;
1717	if (req->lock != REQ_WIRED && req->oldptr &&
1718	    req->oldfunc == sysctl_old_user) {
1719		if (wiredlen != 0) {
1720			ret = vslock(req->oldptr, wiredlen);
1721			if (ret != 0) {
1722				if (ret != ENOMEM)
1723					return (ret);
1724				wiredlen = 0;
1725			}
1726		}
1727		req->lock = REQ_WIRED;
1728		req->validlen = wiredlen;
1729	}
1730	return (0);
1731}
1732
1733int
1734sysctl_find_oid(int *name, u_int namelen, struct sysctl_oid **noid,
1735    int *nindx, struct sysctl_req *req)
1736{
1737	struct sysctl_oid_list *lsp;
1738	struct sysctl_oid *oid;
1739	int indx;
1740
1741	SYSCTL_ASSERT_LOCKED();
1742	lsp = &sysctl__children;
1743	indx = 0;
1744	while (indx < CTL_MAXNAME) {
1745		SLIST_FOREACH(oid, lsp, oid_link) {
1746			if (oid->oid_number == name[indx])
1747				break;
1748		}
1749		if (oid == NULL)
1750			return (ENOENT);
1751
1752		indx++;
1753		if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1754			if (oid->oid_handler != NULL || indx == namelen) {
1755				*noid = oid;
1756				if (nindx != NULL)
1757					*nindx = indx;
1758				KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
1759				    ("%s found DYING node %p", __func__, oid));
1760				return (0);
1761			}
1762			lsp = SYSCTL_CHILDREN(oid);
1763		} else if (indx == namelen) {
1764			*noid = oid;
1765			if (nindx != NULL)
1766				*nindx = indx;
1767			KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
1768			    ("%s found DYING node %p", __func__, oid));
1769			return (0);
1770		} else {
1771			return (ENOTDIR);
1772		}
1773	}
1774	return (ENOENT);
1775}
1776
1777/*
1778 * Traverse our tree, and find the right node, execute whatever it points
1779 * to, and return the resulting error code.
1780 */
1781
1782static int
1783sysctl_root(SYSCTL_HANDLER_ARGS)
1784{
1785	struct sysctl_oid *oid;
1786	struct rm_priotracker tracker;
1787	int error, indx, lvl;
1788
1789	SYSCTL_RLOCK(&tracker);
1790
1791	error = sysctl_find_oid(arg1, arg2, &oid, &indx, req);
1792	if (error)
1793		goto out;
1794
1795	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1796		/*
1797		 * You can't call a sysctl when it's a node, but has
1798		 * no handler.  Inform the user that it's a node.
1799		 * The indx may or may not be the same as namelen.
1800		 */
1801		if (oid->oid_handler == NULL) {
1802			error = EISDIR;
1803			goto out;
1804		}
1805	}
1806
1807	/* Is this sysctl writable? */
1808	if (req->newptr && !(oid->oid_kind & CTLFLAG_WR)) {
1809		error = EPERM;
1810		goto out;
1811	}
1812
1813	KASSERT(req->td != NULL, ("sysctl_root(): req->td == NULL"));
1814
1815#ifdef CAPABILITY_MODE
1816	/*
1817	 * If the process is in capability mode, then don't permit reading or
1818	 * writing unless specifically granted for the node.
1819	 */
1820	if (IN_CAPABILITY_MODE(req->td)) {
1821		if ((req->oldptr && !(oid->oid_kind & CTLFLAG_CAPRD)) ||
1822		    (req->newptr && !(oid->oid_kind & CTLFLAG_CAPWR))) {
1823			error = EPERM;
1824			goto out;
1825		}
1826	}
1827#endif
1828
1829	/* Is this sysctl sensitive to securelevels? */
1830	if (req->newptr && (oid->oid_kind & CTLFLAG_SECURE)) {
1831		lvl = (oid->oid_kind & CTLMASK_SECURE) >> CTLSHIFT_SECURE;
1832		error = securelevel_gt(req->td->td_ucred, lvl);
1833		if (error)
1834			goto out;
1835	}
1836
1837	/* Is this sysctl writable by only privileged users? */
1838	if (req->newptr && !(oid->oid_kind & CTLFLAG_ANYBODY)) {
1839		int priv;
1840
1841		if (oid->oid_kind & CTLFLAG_PRISON)
1842			priv = PRIV_SYSCTL_WRITEJAIL;
1843#ifdef VIMAGE
1844		else if ((oid->oid_kind & CTLFLAG_VNET) &&
1845		     prison_owns_vnet(req->td->td_ucred))
1846			priv = PRIV_SYSCTL_WRITEJAIL;
1847#endif
1848		else
1849			priv = PRIV_SYSCTL_WRITE;
1850		error = priv_check(req->td, priv);
1851		if (error)
1852			goto out;
1853	}
1854
1855	if (!oid->oid_handler) {
1856		error = EINVAL;
1857		goto out;
1858	}
1859
1860	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1861		arg1 = (int *)arg1 + indx;
1862		arg2 -= indx;
1863	} else {
1864		arg1 = oid->oid_arg1;
1865		arg2 = oid->oid_arg2;
1866	}
1867#ifdef MAC
1868	error = mac_system_check_sysctl(req->td->td_ucred, oid, arg1, arg2,
1869	    req);
1870	if (error != 0)
1871		goto out;
1872#endif
1873#ifdef VIMAGE
1874	if ((oid->oid_kind & CTLFLAG_VNET) && arg1 != NULL)
1875		arg1 = (void *)(curvnet->vnet_data_base + (uintptr_t)arg1);
1876#endif
1877	error = sysctl_root_handler_locked(oid, arg1, arg2, req, &tracker);
1878
1879out:
1880	SYSCTL_RUNLOCK(&tracker);
1881	return (error);
1882}
1883
1884#ifndef _SYS_SYSPROTO_H_
1885struct sysctl_args {
1886	int	*name;
1887	u_int	namelen;
1888	void	*old;
1889	size_t	*oldlenp;
1890	void	*new;
1891	size_t	newlen;
1892};
1893#endif
1894int
1895sys___sysctl(struct thread *td, struct sysctl_args *uap)
1896{
1897	int error, i, name[CTL_MAXNAME];
1898	size_t j;
1899
1900	if (uap->namelen > CTL_MAXNAME || uap->namelen < 2)
1901		return (EINVAL);
1902
1903 	error = copyin(uap->name, &name, uap->namelen * sizeof(int));
1904 	if (error)
1905		return (error);
1906
1907	error = userland_sysctl(td, name, uap->namelen,
1908		uap->old, uap->oldlenp, 0,
1909		uap->new, uap->newlen, &j, 0);
1910	if (error && error != ENOMEM)
1911		return (error);
1912	if (uap->oldlenp) {
1913		i = copyout(&j, uap->oldlenp, sizeof(j));
1914		if (i)
1915			return (i);
1916	}
1917	return (error);
1918}
1919
1920/*
1921 * This is used from various compatibility syscalls too.  That's why name
1922 * must be in kernel space.
1923 */
1924int
1925userland_sysctl(struct thread *td, int *name, u_int namelen, void *old,
1926    size_t *oldlenp, int inkernel, void *new, size_t newlen, size_t *retval,
1927    int flags)
1928{
1929	int error = 0, memlocked;
1930	struct sysctl_req req;
1931
1932	bzero(&req, sizeof req);
1933
1934	req.td = td;
1935	req.flags = flags;
1936
1937	if (oldlenp) {
1938		if (inkernel) {
1939			req.oldlen = *oldlenp;
1940		} else {
1941			error = copyin(oldlenp, &req.oldlen, sizeof(*oldlenp));
1942			if (error)
1943				return (error);
1944		}
1945	}
1946	req.validlen = req.oldlen;
1947
1948	if (old) {
1949		if (!useracc(old, req.oldlen, VM_PROT_WRITE))
1950			return (EFAULT);
1951		req.oldptr= old;
1952	}
1953
1954	if (new != NULL) {
1955		if (!useracc(new, newlen, VM_PROT_READ))
1956			return (EFAULT);
1957		req.newlen = newlen;
1958		req.newptr = new;
1959	}
1960
1961	req.oldfunc = sysctl_old_user;
1962	req.newfunc = sysctl_new_user;
1963	req.lock = REQ_UNWIRED;
1964
1965#ifdef KTRACE
1966	if (KTRPOINT(curthread, KTR_SYSCTL))
1967		ktrsysctl(name, namelen);
1968#endif
1969
1970	if (req.oldptr && req.oldlen > PAGE_SIZE) {
1971		memlocked = 1;
1972		sx_xlock(&sysctlmemlock);
1973	} else
1974		memlocked = 0;
1975	CURVNET_SET(TD_TO_VNET(td));
1976
1977	for (;;) {
1978		req.oldidx = 0;
1979		req.newidx = 0;
1980		error = sysctl_root(0, name, namelen, &req);
1981		if (error != EAGAIN)
1982			break;
1983		kern_yield(PRI_USER);
1984	}
1985
1986	CURVNET_RESTORE();
1987
1988	if (req.lock == REQ_WIRED && req.validlen > 0)
1989		vsunlock(req.oldptr, req.validlen);
1990	if (memlocked)
1991		sx_xunlock(&sysctlmemlock);
1992
1993	if (error && error != ENOMEM)
1994		return (error);
1995
1996	if (retval) {
1997		if (req.oldptr && req.oldidx > req.validlen)
1998			*retval = req.validlen;
1999		else
2000			*retval = req.oldidx;
2001	}
2002	return (error);
2003}
2004
2005/*
2006 * Drain into a sysctl struct.  The user buffer should be wired if a page
2007 * fault would cause issue.
2008 */
2009static int
2010sbuf_sysctl_drain(void *arg, const char *data, int len)
2011{
2012	struct sysctl_req *req = arg;
2013	int error;
2014
2015	error = SYSCTL_OUT(req, data, len);
2016	KASSERT(error >= 0, ("Got unexpected negative value %d", error));
2017	return (error == 0 ? len : -error);
2018}
2019
2020struct sbuf *
2021sbuf_new_for_sysctl(struct sbuf *s, char *buf, int length,
2022    struct sysctl_req *req)
2023{
2024
2025	/* Supply a default buffer size if none given. */
2026	if (buf == NULL && length == 0)
2027		length = 64;
2028	s = sbuf_new(s, buf, length, SBUF_FIXEDLEN | SBUF_INCLUDENUL);
2029	sbuf_set_drain(s, sbuf_sysctl_drain, req);
2030	return (s);
2031}
2032