1/*
2 * Copyright (c) 2000-2008 Apple Inc. All rights reserved.
3 *
4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5 *
6 * This file contains Original Code and/or Modifications of Original Code
7 * as defined in and that are subject to the Apple Public Source License
8 * Version 2.0 (the 'License'). You may not use this file except in
9 * compliance with the License. The rights granted to you under the License
10 * may not be used to create, or enable the creation or redistribution of,
11 * unlawful or unlicensed copies of an Apple operating system, or to
12 * circumvent, violate, or enable the circumvention or violation of, any
13 * terms of an Apple operating system software license agreement.
14 *
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
17 *
18 * The Original Code and all software distributed under the License are
19 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23 * Please see the License for the specific language governing rights and
24 * limitations under the License.
25 *
26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27 *
28 *
29 * Copyright (c) 1982, 1986, 1989, 1993
30 *	The Regents of the University of California.  All rights reserved.
31 *
32 * This code is derived from software contributed to Berkeley by
33 * Mike Karels at Berkeley Software Design, Inc.
34 *
35 * Quite extensively rewritten by Poul-Henning Kamp of the FreeBSD
36 * project, to make these variables more userfriendly.
37 *
38 * Redistribution and use in source and binary forms, with or without
39 * modification, are permitted provided that the following conditions
40 * are met:
41 * 1. Redistributions of source code must retain the above copyright
42 *    notice, this list of conditions and the following disclaimer.
43 * 2. Redistributions in binary form must reproduce the above copyright
44 *    notice, this list of conditions and the following disclaimer in the
45 *    documentation and/or other materials provided with the distribution.
46 * 3. All advertising materials mentioning features or use of this software
47 *    must display the following acknowledgement:
48 *	This product includes software developed by the University of
49 *	California, Berkeley and its contributors.
50 * 4. Neither the name of the University nor the names of its contributors
51 *    may be used to endorse or promote products derived from this software
52 *    without specific prior written permission.
53 *
54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64 * SUCH DAMAGE.
65 *
66 *	@(#)kern_sysctl.c	8.4 (Berkeley) 4/14/94
67 */
68
69
70#include <sys/param.h>
71#include <sys/buf.h>
72#include <sys/kernel.h>
73#include <sys/sysctl.h>
74#include <sys/malloc.h>
75#include <sys/proc_internal.h>
76#include <sys/systm.h>
77
78#include <security/audit/audit.h>
79
80/*
81struct sysctl_oid_list sysctl__debug_children;
82struct sysctl_oid_list sysctl__kern_children;
83struct sysctl_oid_list sysctl__net_children;
84struct sysctl_oid_list sysctl__sysctl_children;
85*/
86
87lck_rw_t * sysctl_geometry_lock = NULL;
88
89/*
90 * Conditionally allow dtrace to see these functions for debugging purposes.
91 */
92#ifdef STATIC
93#undef STATIC
94#endif
95#if 0
96#define	STATIC
97#else
98#define STATIC static
99#endif
100
101/* forward declarations  of static functions */
102STATIC funnel_t *spl_kernel_funnel(void);
103STATIC void sysctl_sysctl_debug_dump_node(struct sysctl_oid_list *l, int i);
104STATIC int sysctl_sysctl_debug(struct sysctl_oid *oidp, void *arg1,
105	int arg2, struct sysctl_req *req);
106STATIC int sysctl_sysctl_name(struct sysctl_oid *oidp, void *arg1,
107	int arg2, struct sysctl_req *req);
108STATIC int sysctl_sysctl_next_ls (struct sysctl_oid_list *lsp,
109	int *name, u_int namelen, int *next, int *len, int level,
110	struct sysctl_oid **oidpp);
111STATIC int sysctl_old_kernel(struct sysctl_req *req, const void *p, size_t l);
112STATIC int sysctl_new_kernel(struct sysctl_req *req, void *p, size_t l);
113STATIC int name2oid (char *name, int *oid, int *len);
114STATIC int sysctl_sysctl_name2oid(struct sysctl_oid *oidp, void *arg1, int arg2, struct sysctl_req *req);
115STATIC int sysctl_sysctl_next(struct sysctl_oid *oidp, void *arg1, int arg2,
116        struct sysctl_req *req);
117STATIC int sysctl_sysctl_oidfmt(struct sysctl_oid *oidp, void *arg1, int arg2, struct sysctl_req *req);
118STATIC void splx_kernel_funnel(funnel_t *saved);
119STATIC int sysctl_old_user(struct sysctl_req *req, const void *p, size_t l);
120STATIC int sysctl_new_user(struct sysctl_req *req, void *p, size_t l);
121STATIC int sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp, void *newp, size_t newlen);
122STATIC int sysctlnametomib(const char *name, int *mibp, size_t *sizep);
123
124
125
126/*
127 * Locking and stats
128 */
129
130/*
131 * XXX this does not belong here
132 */
133STATIC funnel_t *
134spl_kernel_funnel(void)
135{
136	funnel_t *cfunnel;
137
138	cfunnel = thread_funnel_get();
139	if (cfunnel != kernel_flock) {
140		if (cfunnel != NULL)
141			thread_funnel_set(cfunnel, FALSE);
142		thread_funnel_set(kernel_flock, TRUE);
143	}
144	return(cfunnel);
145}
146
147STATIC void
148splx_kernel_funnel(funnel_t *saved)
149{
150	if (saved != kernel_flock) {
151		thread_funnel_set(kernel_flock, FALSE);
152		if (saved != NULL)
153			thread_funnel_set(saved, TRUE);
154	}
155}
156
157STATIC int sysctl_root SYSCTL_HANDLER_ARGS;
158
159struct sysctl_oid_list sysctl__children; /* root list */
160
161/*
162 * Initialization of the MIB tree.
163 *
164 * Order by number in each list.
165 */
166
167void
168sysctl_register_oid(struct sysctl_oid *new_oidp)
169{
170	struct sysctl_oid *oidp = NULL;
171	struct sysctl_oid_list *parent = new_oidp->oid_parent;
172	struct sysctl_oid *p;
173	struct sysctl_oid *q;
174	int n;
175	funnel_t *fnl = NULL;	/* compiler doesn't notice CTLFLAG_LOCKED */
176
177	/*
178	 * The OID can be old-style (needs copy), new style without an earlier
179	 * version (also needs copy), or new style with a matching version (no
180	 * copy needed).  Later versions are rejected (presumably, the OID
181	 * structure was changed for a necessary reason).
182	 */
183	if (!(new_oidp->oid_kind & CTLFLAG_OID2)) {
184		/*
185		 * XXX:	M_TEMP is perhaps not the most apropriate zone, as it
186		 * XXX:	will subject us to use-after-free by other consumers.
187		 */
188		MALLOC(oidp, struct sysctl_oid *, sizeof(*oidp), M_TEMP, M_WAITOK | M_ZERO);
189		if (oidp == NULL)
190			return;		/* reject: no memory */
191
192		/*
193		 * Copy the structure only through the oid_fmt field, which
194		 * is the last field in a non-OID2 OID structure.
195		 *
196		 * Note:	We may want to set the oid_descr to the
197		 *		oid_name (or "") at some future date.
198		 */
199		memcpy(oidp, new_oidp, offsetof(struct sysctl_oid, oid_descr));
200	} else {
201		/* It's a later version; handle the versions we know about */
202		switch (new_oidp->oid_version) {
203		case SYSCTL_OID_VERSION:
204			/* current version */
205			oidp = new_oidp;
206			break;
207		default:
208			return;			/* rejects unknown version */
209		}
210	}
211
212	/*
213	 * If it's a locked OID being registered, we can assume that the
214	 * caller is doing their own reentrancy locking before calling us.
215	 */
216	if (!(oidp->oid_kind & CTLFLAG_LOCKED))
217		fnl = spl_kernel_funnel();
218
219	if(sysctl_geometry_lock == NULL)
220	{
221		/*
222		 * Initialise the geometry lock for reading/modifying the
223		 * sysctl tree. This is done here because IOKit registers
224		 * some sysctl's before bsd_init() calls
225		 * sysctl_register_fixed().
226		 */
227
228		lck_grp_t* lck_grp  = lck_grp_alloc_init("sysctl", NULL);
229		sysctl_geometry_lock = lck_rw_alloc_init(lck_grp, NULL);
230	}
231	/* Get the write lock to modify the geometry */
232	lck_rw_lock_exclusive(sysctl_geometry_lock);
233
234	/*
235	 * If this oid has a number OID_AUTO, give it a number which
236	 * is greater than any current oid.  Make sure it is at least
237	 * OID_AUTO_START to leave space for pre-assigned oid numbers.
238	 */
239	if (oidp->oid_number == OID_AUTO) {
240		/* First, find the highest oid in the parent list >OID_AUTO_START-1 */
241		n = OID_AUTO_START;
242		SLIST_FOREACH(p, parent, oid_link) {
243			if (p->oid_number > n)
244				n = p->oid_number;
245		}
246		oidp->oid_number = n + 1;
247		/*
248		 * Reflect the number in an llocated OID into the template
249		 * of the caller for sysctl_unregister_oid() compares.
250		 */
251		if (oidp != new_oidp)
252			new_oidp->oid_number = oidp->oid_number;
253	}
254
255	/*
256	 * Insert the oid into the parent's list in order.
257	 */
258	q = NULL;
259	SLIST_FOREACH(p, parent, oid_link) {
260		if (oidp->oid_number < p->oid_number)
261			break;
262		q = p;
263	}
264	if (q)
265		SLIST_INSERT_AFTER(q, oidp, oid_link);
266	else
267		SLIST_INSERT_HEAD(parent, oidp, oid_link);
268
269	/* Release the write lock */
270	lck_rw_unlock_exclusive(sysctl_geometry_lock);
271
272	if (!(oidp->oid_kind & CTLFLAG_LOCKED))
273		splx_kernel_funnel(fnl);
274}
275
276void
277sysctl_unregister_oid(struct sysctl_oid *oidp)
278{
279	struct sysctl_oid *removed_oidp = NULL;	/* OID removed from tree */
280	struct sysctl_oid *old_oidp = NULL;	/* OID compatibility copy */
281	funnel_t *fnl = NULL;	/* compiler doesn't notice CTLFLAG_LOCKED */
282
283	if (!(oidp->oid_kind & CTLFLAG_LOCKED))
284		fnl = spl_kernel_funnel();
285
286	/* Get the write lock to modify the geometry */
287	lck_rw_lock_exclusive(sysctl_geometry_lock);
288
289	if (!(oidp->oid_kind & CTLFLAG_OID2)) {
290		/*
291		 * We're using a copy so we can get the new fields in an
292		 * old structure, so we have to iterate to compare the
293		 * partial structure; when we find a match, we remove it
294		 * normally and free the memory.
295		 */
296		SLIST_FOREACH(old_oidp, oidp->oid_parent, oid_link) {
297			if (!memcmp(&oidp->oid_number, &old_oidp->oid_number, (offsetof(struct sysctl_oid, oid_descr)-offsetof(struct sysctl_oid, oid_number)))) {
298                break;
299            }
300		}
301		if (old_oidp != NULL) {
302			SLIST_REMOVE(old_oidp->oid_parent, old_oidp, sysctl_oid, oid_link);
303			removed_oidp = old_oidp;
304		}
305	} else {
306		/* It's a later version; handle the versions we know about */
307		switch (oidp->oid_version) {
308		case SYSCTL_OID_VERSION:
309			/* We can just remove the OID directly... */
310			SLIST_REMOVE(oidp->oid_parent, oidp, sysctl_oid, oid_link);
311			removed_oidp = oidp;
312			break;
313		default:
314			 /* XXX: Can't happen; probably tree coruption.*/
315			break;			/* rejects unknown version */
316		}
317	}
318
319	/*
320	 * We've removed it from the list at this point, but we don't want
321	 * to return to the caller until all handler references have drained
322	 * out.  Doing things in this order prevent other people coming in
323	 * and starting new operations against the OID node we want removed.
324	 *
325	 * Note:	oidp could be NULL if it wasn't found.
326	 */
327	while(removed_oidp && removed_oidp->oid_refcnt) {
328		lck_rw_sleep(sysctl_geometry_lock, LCK_SLEEP_EXCLUSIVE, &removed_oidp->oid_refcnt, THREAD_UNINT);
329	}
330
331	/* Release the write lock */
332	lck_rw_unlock_exclusive(sysctl_geometry_lock);
333
334	/* If it was allocated, free it after dropping the lock */
335	if (old_oidp != NULL) {
336		FREE(old_oidp, M_TEMP);
337	}
338
339	/* And drop the funnel interlock, if needed */
340	if (!(oidp->oid_kind & CTLFLAG_LOCKED))
341		splx_kernel_funnel(fnl);
342}
343
344/*
345 * Bulk-register all the oids in a linker_set.
346 */
347void
348sysctl_register_set(const char *set)
349{
350	struct sysctl_oid **oidpp, *oidp;
351
352	LINKER_SET_FOREACH(oidpp, struct sysctl_oid **, set) {
353		oidp = *oidpp;
354		if (!(oidp->oid_kind & CTLFLAG_NOAUTO)) {
355		    sysctl_register_oid(oidp);
356		}
357	}
358}
359
360void
361sysctl_unregister_set(const char *set)
362{
363	struct sysctl_oid **oidpp, *oidp;
364
365	LINKER_SET_FOREACH(oidpp, struct sysctl_oid **, set) {
366		oidp = *oidpp;
367		if (!(oidp->oid_kind & CTLFLAG_NOAUTO)) {
368		    sysctl_unregister_oid(oidp);
369		}
370	}
371}
372
373
374/*
375 * Register the kernel's oids on startup.
376 */
377
378void
379sysctl_register_all()
380{
381	sysctl_register_set("__sysctl_set");
382}
383
384void
385sysctl_register_fixed(void)
386{
387	sysctl_register_all();
388}
389
390/*
391 * New handler interface
392 *   If the sysctl caller (user mode or kernel mode) is interested in the
393 *   value (req->oldptr != NULL), we copy the data (bigValue etc.) out,
394 *   if the caller wants to set the value (req->newptr), we copy
395 *   the data in (*pValue etc.).
396 */
397
398int
399sysctl_io_number(struct sysctl_req *req, long long bigValue, size_t valueSize, void *pValue, int *changed) {
400	int		smallValue;
401	int		error;
402
403	if (changed) *changed = 0;
404
405	/*
406	 * Handle the various combinations of caller buffer size and
407	 * data value size.  We are generous in the case where the
408	 * caller has specified a 32-bit buffer but the value is 64-bit
409	 * sized.
410	 */
411
412	/* 32 bit value expected or 32 bit buffer offered */
413	if (((valueSize == sizeof(int)) ||
414	    ((req->oldlen == sizeof(int)) && (valueSize == sizeof(long long))))
415			&& (req->oldptr)) {
416		smallValue = (int)bigValue;
417		if ((long long)smallValue != bigValue)
418			return(ERANGE);
419		error = SYSCTL_OUT(req, &smallValue, sizeof(smallValue));
420	} else {
421		/* any other case is either size-equal or a bug */
422		error = SYSCTL_OUT(req, &bigValue, valueSize);
423	}
424	/* error or nothing to set */
425	if (error || !req->newptr)
426		return(error);
427
428	/* set request for constant */
429	if (pValue == NULL)
430		return(EPERM);
431
432	/* set request needs to convert? */
433	if ((req->newlen == sizeof(int)) && (valueSize == sizeof(long long))) {
434		/* new value is 32 bits, upconvert to 64 bits */
435		error = SYSCTL_IN(req, &smallValue, sizeof(smallValue));
436		if (!error)
437			*(long long *)pValue = (long long)smallValue;
438	} else if ((req->newlen == sizeof(long long)) && (valueSize == sizeof(int))) {
439		/* new value is 64 bits, downconvert to 32 bits and range check */
440		error = SYSCTL_IN(req, &bigValue, sizeof(bigValue));
441		if (!error) {
442			smallValue = (int)bigValue;
443			if ((long long)smallValue != bigValue)
444				return(ERANGE);
445			*(int *)pValue = smallValue;
446		}
447	} else {
448		/* sizes match, just copy in */
449		error = SYSCTL_IN(req, pValue, valueSize);
450	}
451	if (!error && changed)
452		*changed = 1;
453	return(error);
454}
455
456int
457sysctl_io_string(struct sysctl_req *req, char *pValue, size_t valueSize, int trunc, int *changed)
458{
459	int error;
460
461	if (changed) *changed = 0;
462
463	if (trunc && req->oldptr && req->oldlen && (req->oldlen<strlen(pValue) + 1)) {
464		/* If trunc != 0, if you give it a too small (but larger than
465		 * 0 bytes) buffer, instead of returning ENOMEM, it truncates the
466		 * returned string to the buffer size.  This preserves the semantics
467		 * of some library routines implemented via sysctl, which truncate
468		 * their returned data, rather than simply returning an error. The
469		 * returned string is always NUL terminated. */
470		error = SYSCTL_OUT(req, pValue, req->oldlen-1);
471		if (!error) {
472			char c = 0;
473			error = SYSCTL_OUT(req, &c, 1);
474		}
475	} else {
476		/* Copy string out */
477		error = SYSCTL_OUT(req, pValue, strlen(pValue) + 1);
478	}
479
480	/* error or no new value */
481	if (error || !req->newptr)
482		return(error);
483
484	/* attempt to set read-only value */
485	if (valueSize == 0)
486		return(EPERM);
487
488	/* make sure there's room for the new string */
489	if (req->newlen >= valueSize)
490		return(EINVAL);
491
492	/* copy the string in and force NUL termination */
493	error = SYSCTL_IN(req, pValue, req->newlen);
494	pValue[req->newlen] = '\0';
495
496	if (!error && changed)
497		*changed = 1;
498	return(error);
499}
500
501int sysctl_io_opaque(struct sysctl_req *req,void *pValue, size_t valueSize, int *changed)
502{
503	int error;
504
505	if (changed) *changed = 0;
506
507	/* Copy blob out */
508	error = SYSCTL_OUT(req, pValue, valueSize);
509
510	/* error or nothing to set */
511	if (error || !req->newptr)
512		return(error);
513
514	error = SYSCTL_IN(req, pValue, valueSize);
515
516	if (!error && changed)
517		*changed = 1;
518	return(error);
519}
520
521/*
522 * "Staff-functions"
523 *
524 * These functions implement a presently undocumented interface
525 * used by the sysctl program to walk the tree, and get the type
526 * so it can print the value.
527 * This interface is under work and consideration, and should probably
528 * be killed with a big axe by the first person who can find the time.
529 * (be aware though, that the proper interface isn't as obvious as it
530 * may seem, there are various conflicting requirements.
531 *
532 * {0,0}	printf the entire MIB-tree.
533 * {0,1,...}	return the name of the "..." OID.
534 * {0,2,...}	return the next OID.
535 * {0,3}	return the OID of the name in "new"
536 * {0,4,...}	return the kind & format info for the "..." OID.
537 */
538
539/*
540 * sysctl_sysctl_debug_dump_node
541 *
542 * Description:	Dump debug information for a given sysctl_oid_list at the
543 *		given oid depth out to the kernel log, via printf
544 *
545 * Parameters:	l				sysctl_oid_list pointer
546 *		i				current node depth
547 *
548 * Returns:	(void)
549 *
550 * Implicit:	kernel log, modified
551 *
552 * Locks:	Assumes sysctl_geometry_lock is held prior to calling
553 *
554 * Notes:	This function may call itself recursively to resolve Node
555 *		values, which potentially have an inferioer sysctl_oid_list
556 *
557 *		This function is only callable indirectly via the function
558 *		sysctl_sysctl_debug()
559 *
560 * Bugs:	The node depth indentation does not work; this may be an
561 *		artifact of leading space removal by the log daemon itself
562 *		or some intermediate routine.
563 */
564STATIC void
565sysctl_sysctl_debug_dump_node(struct sysctl_oid_list *l, int i)
566{
567	int k;
568	struct sysctl_oid *oidp;
569
570	SLIST_FOREACH(oidp, l, oid_link) {
571
572		for (k=0; k<i; k++)
573			printf(" ");
574
575		printf("%d %s ", oidp->oid_number, oidp->oid_name);
576
577		printf("%c%c%c",
578			oidp->oid_kind & CTLFLAG_LOCKED ? 'L':' ',
579			oidp->oid_kind & CTLFLAG_RD ? 'R':' ',
580			oidp->oid_kind & CTLFLAG_WR ? 'W':' ');
581
582		if (oidp->oid_handler)
583			printf(" *Handler");
584
585		switch (oidp->oid_kind & CTLTYPE) {
586			case CTLTYPE_NODE:
587				printf(" Node\n");
588				if (!oidp->oid_handler) {
589					sysctl_sysctl_debug_dump_node(
590						oidp->oid_arg1, i+2);
591				}
592				break;
593			case CTLTYPE_INT:    printf(" Int\n"); break;
594			case CTLTYPE_STRING: printf(" String\n"); break;
595			case CTLTYPE_QUAD:   printf(" Quad\n"); break;
596			case CTLTYPE_OPAQUE: printf(" Opaque/struct\n"); break;
597			default:	     printf("\n");
598		}
599
600	}
601}
602
603/*
604 * sysctl_sysctl_debug
605 *
606 * Description:	This function implements the "sysctl.debug" portion of the
607 *		OID space for sysctl.
608 *
609 * OID:		0, 0
610 *
611 * Parameters:	__unused
612 *
613 * Returns:	ENOENT
614 *
615 * Implicit:	kernel log, modified
616 *
617 * Locks:	Acquires and then releases a read lock on the
618 *		sysctl_geometry_lock
619 */
620STATIC int
621sysctl_sysctl_debug(__unused struct sysctl_oid *oidp, __unused void *arg1,
622	__unused int arg2, __unused struct sysctl_req *req)
623{
624	lck_rw_lock_shared(sysctl_geometry_lock);
625	sysctl_sysctl_debug_dump_node(&sysctl__children, 0);
626	lck_rw_done(sysctl_geometry_lock);
627	return ENOENT;
628}
629
630SYSCTL_PROC(_sysctl, 0, debug, CTLTYPE_STRING|CTLFLAG_RD | CTLFLAG_LOCKED,
631	0, 0, sysctl_sysctl_debug, "-", "");
632
633/*
634 * sysctl_sysctl_name
635 *
636 * Description:	Convert an OID into a string name; this is used by the user
637 *		space sysctl() command line utility; this is done in a purely
638 *		advisory capacity (e.g. to provide node names for "sysctl -A"
639 *		output).
640 *
641 * OID:		0, 1
642 *
643 * Parameters:	oidp				__unused
644 *		arg1				A pointer to the OID name list
645 *						integer array, beginning at
646 *						adjusted option base 2
647 *		arg2				The number of elements which
648 *						remain in the name array
649 *
650 * Returns:	0				Success
651 *	SYSCTL_OUT:EPERM			Permission denied
652 *	SYSCTL_OUT:EFAULT			Bad user supplied buffer
653 *	SYSCTL_OUT:???				Return value from user function
654 *						for SYSCTL_PROC leaf node
655 *
656 * Implict:	Contents of user request buffer, modified
657 *
658 * Locks:	Acquires and then releases a read lock on the
659 *		sysctl_geometry_lock
660 *
661 * Notes:	SPI (System Programming Interface); this is subject to change
662 *		and may not be relied upon by third party applications; use
663 *		a subprocess to communicate with the "sysctl" command line
664 *		command instead, if you believe you need this functionality.
665 *		Preferrably, use sysctlbyname() instead.
666 *
667 *		Setting of the NULL termination of the output string is
668 *		delayed until after the geometry lock is dropped.  If there
669 *		are no Entries remaining in the OID name list when this
670 *		function is called, it will still write out the termination
671 *		byte.
672 *
673 *		This function differs from other sysctl functions in that
674 *		it can not take an output buffer length of 0 to determine the
675 *		space which will be required.  It is suggested that the buffer
676 *		length be PATH_MAX, and that authors of new sysctl's refrain
677 *		from exceeding this string length.
678 */
679STATIC int
680sysctl_sysctl_name(__unused struct sysctl_oid *oidp, void *arg1, int arg2,
681        struct sysctl_req *req)
682{
683	int *name = (int *) arg1;
684	u_int namelen = arg2;
685	int error = 0;
686	struct sysctl_oid *oid;
687	struct sysctl_oid_list *lsp = &sysctl__children, *lsp2;
688	char tempbuf[10];
689
690	lck_rw_lock_shared(sysctl_geometry_lock);
691	while (namelen) {
692		if (!lsp) {
693			snprintf(tempbuf,sizeof(tempbuf),"%d",*name);
694			if (req->oldidx)
695				error = SYSCTL_OUT(req, ".", 1);
696			if (!error)
697				error = SYSCTL_OUT(req, tempbuf, strlen(tempbuf));
698			if (error) {
699				lck_rw_done(sysctl_geometry_lock);
700				return (error);
701			}
702			namelen--;
703			name++;
704			continue;
705		}
706		lsp2 = 0;
707		SLIST_FOREACH(oid, lsp, oid_link) {
708			if (oid->oid_number != *name)
709				continue;
710
711			if (req->oldidx)
712				error = SYSCTL_OUT(req, ".", 1);
713			if (!error)
714				error = SYSCTL_OUT(req, oid->oid_name,
715					strlen(oid->oid_name));
716			if (error) {
717				lck_rw_done(sysctl_geometry_lock);
718				return (error);
719			}
720
721			namelen--;
722			name++;
723
724			if ((oid->oid_kind & CTLTYPE) != CTLTYPE_NODE)
725				break;
726
727			if (oid->oid_handler)
728				break;
729
730			lsp2 = (struct sysctl_oid_list *)oid->oid_arg1;
731			break;
732		}
733		lsp = lsp2;
734	}
735	lck_rw_done(sysctl_geometry_lock);
736	return (SYSCTL_OUT(req, "", 1));
737}
738
739SYSCTL_NODE(_sysctl, 1, name, CTLFLAG_RD | CTLFLAG_LOCKED, sysctl_sysctl_name, "");
740
741/*
742 * sysctl_sysctl_next_ls
743 *
744 * Description:	For a given OID name value, return the next consecutive OID
745 *		name value within the geometry tree
746 *
747 * Parameters:	lsp				The OID list to look in
748 *		name				The OID name to start from
749 *		namelen				The length of the OID name
750 *		next				Pointer to new oid storage to
751 *						fill in
752 *		len				Pointer to receive new OID
753 *						length value of storage written
754 *		level				OID tree depth (used to compute
755 *						len value)
756 *		oidpp				Pointer to OID list entry
757 *						pointer; used to walk the list
758 *						forward across recursion
759 *
760 * Returns:	0				Returning a new entry
761 *		1				End of geometry list reached
762 *
763 * Implicit:	*next				Modified to contain the new OID
764 *		*len				Modified to contain new length
765 *
766 * Locks:	Assumes sysctl_geometry_lock is held prior to calling
767 *
768 * Notes:	This function will not return OID values that have special
769 *		handlers, since we can not tell wheter these handlers consume
770 *		elements from the OID space as parameters.  For this reason,
771 *		we STRONGLY discourage these types of handlers
772 */
773STATIC int
774sysctl_sysctl_next_ls (struct sysctl_oid_list *lsp, int *name, u_int namelen,
775	int *next, int *len, int level, struct sysctl_oid **oidpp)
776{
777	struct sysctl_oid *oidp;
778
779	*len = level;
780	SLIST_FOREACH(oidp, lsp, oid_link) {
781		*next = oidp->oid_number;
782		*oidpp = oidp;
783
784		if (!namelen) {
785			if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
786				return 0;
787			if (oidp->oid_handler)
788				/* We really should call the handler here...*/
789				return 0;
790			lsp = (struct sysctl_oid_list *)oidp->oid_arg1;
791
792			if (!SLIST_FIRST(lsp))
793				/* This node had no children - skip it! */
794				continue;
795
796			if (!sysctl_sysctl_next_ls (lsp, 0, 0, next+1,
797				len, level+1, oidpp))
798				return 0;
799			goto next;
800		}
801
802		if (oidp->oid_number < *name)
803			continue;
804
805		if (oidp->oid_number > *name) {
806			if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
807				return 0;
808			if (oidp->oid_handler)
809				return 0;
810			lsp = (struct sysctl_oid_list *)oidp->oid_arg1;
811			if (!sysctl_sysctl_next_ls (lsp, name+1, namelen-1,
812				next+1, len, level+1, oidpp))
813				return (0);
814			goto next;
815		}
816		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
817			continue;
818
819		if (oidp->oid_handler)
820			continue;
821
822		lsp = (struct sysctl_oid_list *)oidp->oid_arg1;
823		if (!sysctl_sysctl_next_ls (lsp, name+1, namelen-1, next+1,
824			len, level+1, oidpp))
825			return (0);
826	next:
827		namelen = 1;
828		*len = level;
829	}
830	return 1;
831}
832
833/*
834 * sysctl_sysctl_next
835 *
836 * Description:	This is an iterator function designed to iterate the oid tree
837 *		and provide a list of OIDs for use by the user space "sysctl"
838 *		command line tool
839 *
840 * OID:		0, 2
841 *
842 * Parameters:	oidp				__unused
843 *		arg1				Pointer to start OID name
844 *		arg2				Start OID name length
845 *		req				Pointer to user request buffer
846 *
847 * Returns:	0				Success
848 *		ENOENT				Reached end of OID space
849 *	SYSCTL_OUT:EPERM			Permission denied
850 *	SYSCTL_OUT:EFAULT			Bad user supplied buffer
851 *	SYSCTL_OUT:???				Return value from user function
852 *						for SYSCTL_PROC leaf node
853 *
854 * Implict:	Contents of user request buffer, modified
855 *
856 * Locks:	Acquires and then releases a read lock on the
857 *		sysctl_geometry_lock
858 *
859 * Notes:	SPI (System Programming Interface); this is subject to change
860 *		and may not be relied upon by third party applications; use
861 *		a subprocess to communicate with the "sysctl" command line
862 *		command instead, if you believe you need this functionality.
863 *		Preferrably, use sysctlbyname() instead.
864 *
865 *		This function differs from other sysctl functions in that
866 *		it can not take an output buffer length of 0 to determine the
867 *		space which will be required.  It is suggested that the buffer
868 *		length be PATH_MAX, and that authors of new sysctl's refrain
869 *		from exceeding this string length.
870 */
871STATIC int
872sysctl_sysctl_next(__unused struct sysctl_oid *oidp, void *arg1, int arg2,
873        struct sysctl_req *req)
874{
875	int *name = (int *) arg1;
876	u_int namelen = arg2;
877	int i, j, error;
878	struct sysctl_oid *oid;
879	struct sysctl_oid_list *lsp = &sysctl__children;
880	int newoid[CTL_MAXNAME];
881
882	lck_rw_lock_shared(sysctl_geometry_lock);
883	i = sysctl_sysctl_next_ls (lsp, name, namelen, newoid, &j, 1, &oid);
884	lck_rw_done(sysctl_geometry_lock);
885	if (i)
886		return ENOENT;
887	error = SYSCTL_OUT(req, newoid, j * sizeof (int));
888	return (error);
889}
890
891SYSCTL_NODE(_sysctl, 2, next, CTLFLAG_RD | CTLFLAG_LOCKED, sysctl_sysctl_next, "");
892
893/*
894 * name2oid
895 *
896 * Description:	Support function for use by sysctl_sysctl_name2oid(); looks
897 *		up an OID name given a string name.
898 *
899 * Parameters:	name				NULL terminated string name
900 *		oid				Pointer to receive OID name
901 *		len				Pointer to receive OID length
902 *						pointer value (see "Notes")
903 *
904 * Returns:	0				Success
905 *		ENOENT				Entry not found
906 *
907 * Implicit:	*oid				Modified to contain OID value
908 *		*len				Modified to contain OID length
909 *
910 * Locks:	Assumes sysctl_geometry_lock is held prior to calling
911 */
912STATIC int
913name2oid (char *name, int *oid, int *len)
914{
915	int i;
916	struct sysctl_oid *oidp;
917	struct sysctl_oid_list *lsp = &sysctl__children;
918	char *p;
919
920	if (!*name)
921		return ENOENT;
922
923	p = name + strlen(name) - 1 ;
924	if (*p == '.')
925		*p = '\0';
926
927	*len = 0;
928
929	for (p = name; *p && *p != '.'; p++)
930		;
931	i = *p;
932	if (i == '.')
933		*p = '\0';
934
935	oidp = SLIST_FIRST(lsp);
936
937	while (oidp && *len < CTL_MAXNAME) {
938		if (strcmp(name, oidp->oid_name)) {
939			oidp = SLIST_NEXT(oidp, oid_link);
940			continue;
941		}
942		*oid++ = oidp->oid_number;
943		(*len)++;
944
945		if (!i) {
946			return (0);
947		}
948
949		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
950			break;
951
952		if (oidp->oid_handler)
953			break;
954
955		lsp = (struct sysctl_oid_list *)oidp->oid_arg1;
956		oidp = SLIST_FIRST(lsp);
957		name = p+1;
958		for (p = name; *p && *p != '.'; p++)
959				;
960		i = *p;
961		if (i == '.')
962			*p = '\0';
963	}
964	return ENOENT;
965}
966
967/*
968 * sysctl_sysctl_name2oid
969 *
970 * Description:	Translate a string name to an OID name value; this is used by
971 *		the sysctlbyname() function as well as by the "sysctl" command
972 *		line command.
973 *
974 * OID:		0, 3
975 *
976 * Parameters:	oidp				__unused
977 *		arg1				__unused
978 *		arg2				__unused
979 *		req				Request structure
980 *
981 * Returns:	ENOENT				Input length too short
982 *		ENAMETOOLONG			Input length too long
983 *		ENOMEM				Could not allocate work area
984 *	SYSCTL_IN/OUT:EPERM			Permission denied
985 *	SYSCTL_IN/OUT:EFAULT			Bad user supplied buffer
986 *	SYSCTL_IN/OUT:???			Return value from user function
987 *	name2oid:ENOENT				Not found
988 *
989 * Implicit:	*req				Contents of request, modified
990 *
991 * Locks:	Acquires and then releases a read lock on the
992 *		sysctl_geometry_lock
993 *
994 * Notes:	SPI (System Programming Interface); this is subject to change
995 *		and may not be relied upon by third party applications; use
996 *		a subprocess to communicate with the "sysctl" command line
997 *		command instead, if you believe you need this functionality.
998 *		Preferrably, use sysctlbyname() instead.
999 *
1000 *		This function differs from other sysctl functions in that
1001 *		it can not take an output buffer length of 0 to determine the
1002 *		space which will be required.  It is suggested that the buffer
1003 *		length be PATH_MAX, and that authors of new sysctl's refrain
1004 *		from exceeding this string length.
1005 */
1006STATIC int
1007sysctl_sysctl_name2oid(__unused struct sysctl_oid *oidp, __unused void *arg1,
1008	__unused int arg2, struct sysctl_req *req)
1009{
1010	char *p;
1011	int error, oid[CTL_MAXNAME];
1012	int len = 0;		/* set by name2oid() */
1013
1014	if (req->newlen < 1)
1015		return ENOENT;
1016	if (req->newlen >= MAXPATHLEN)	/* XXX arbitrary, undocumented */
1017		return (ENAMETOOLONG);
1018
1019	MALLOC(p, char *,req->newlen+1, M_TEMP, M_WAITOK);
1020	if (!p)
1021	    return ENOMEM;
1022
1023	error = SYSCTL_IN(req, p, req->newlen);
1024	if (error) {
1025		FREE(p, M_TEMP);
1026		return (error);
1027	}
1028
1029	p [req->newlen] = '\0';
1030
1031	/*
1032	 * Note:	We acquire and release the geometry lock here to
1033	 *		avoid making name2oid needlessly complex.
1034	 */
1035	lck_rw_lock_shared(sysctl_geometry_lock);
1036	error = name2oid(p, oid, &len);
1037	lck_rw_done(sysctl_geometry_lock);
1038
1039	FREE(p, M_TEMP);
1040
1041	if (error)
1042		return (error);
1043
1044	error = SYSCTL_OUT(req, oid, len * sizeof *oid);
1045	return (error);
1046}
1047
1048SYSCTL_PROC(_sysctl, 3, name2oid, CTLFLAG_RW|CTLFLAG_ANYBODY|CTLFLAG_KERN | CTLFLAG_LOCKED, 0, 0,
1049	sysctl_sysctl_name2oid, "I", "");
1050
1051/*
1052 * sysctl_sysctl_oidfmt
1053 *
1054 * Description:	For a given OID name, determine the format of the data which
1055 *		is associated with it.  This is used by the "sysctl" command
1056 *		line command.
1057 *
1058 * OID:		0, 4
1059 *
1060 * Parameters:	oidp				__unused
1061 *		arg1				The OID name to look up
1062 *		arg2				The length of the OID name
1063 *		req				Pointer to user request buffer
1064 *
1065 * Returns:	0				Success
1066 *		EISDIR				Malformed request
1067 *		ENOENT				No such OID name
1068 *	SYSCTL_OUT:EPERM			Permission denied
1069 *	SYSCTL_OUT:EFAULT			Bad user supplied buffer
1070 *	SYSCTL_OUT:???				Return value from user function
1071 *
1072 * Implict:	Contents of user request buffer, modified
1073 *
1074 * Locks:	Acquires and then releases a read lock on the
1075 *		sysctl_geometry_lock
1076 *
1077 * Notes:	SPI (System Programming Interface); this is subject to change
1078 *		and may not be relied upon by third party applications; use
1079 *		a subprocess to communicate with the "sysctl" command line
1080 *		command instead, if you believe you need this functionality.
1081 *
1082 *		This function differs from other sysctl functions in that
1083 *		it can not take an output buffer length of 0 to determine the
1084 *		space which will be required.  It is suggested that the buffer
1085 *		length be PATH_MAX, and that authors of new sysctl's refrain
1086 *		from exceeding this string length.
1087 */
1088STATIC int
1089sysctl_sysctl_oidfmt(__unused struct sysctl_oid *oidp, void *arg1, int arg2,
1090        struct sysctl_req *req)
1091{
1092	int *name = (int *) arg1;
1093	int error = ENOENT;		/* default error: not found */
1094	u_int namelen = arg2;
1095	u_int indx;
1096	struct sysctl_oid *oid;
1097	struct sysctl_oid_list *lsp = &sysctl__children;
1098
1099	lck_rw_lock_shared(sysctl_geometry_lock);
1100	oid = SLIST_FIRST(lsp);
1101
1102	indx = 0;
1103	while (oid && indx < CTL_MAXNAME) {
1104		if (oid->oid_number == name[indx]) {
1105			indx++;
1106			if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1107				if (oid->oid_handler)
1108					goto found;
1109				if (indx == namelen)
1110					goto found;
1111				lsp = (struct sysctl_oid_list *)oid->oid_arg1;
1112				oid = SLIST_FIRST(lsp);
1113			} else {
1114				if (indx != namelen) {
1115					error =  EISDIR;
1116					goto err;
1117				}
1118				goto found;
1119			}
1120		} else {
1121			oid = SLIST_NEXT(oid, oid_link);
1122		}
1123	}
1124	/* Not found */
1125	goto err;
1126
1127found:
1128	if (!oid->oid_fmt)
1129		goto err;
1130	error = SYSCTL_OUT(req,
1131		&oid->oid_kind, sizeof(oid->oid_kind));
1132	if (!error)
1133		error = SYSCTL_OUT(req, oid->oid_fmt,
1134			strlen(oid->oid_fmt)+1);
1135err:
1136	lck_rw_done(sysctl_geometry_lock);
1137	return (error);
1138}
1139
1140SYSCTL_NODE(_sysctl, 4, oidfmt, CTLFLAG_RD | CTLFLAG_LOCKED, sysctl_sysctl_oidfmt, "");
1141
1142
1143/*
1144 * Default "handler" functions.
1145 */
1146
1147/*
1148 * Handle an int, signed or unsigned.
1149 * Two cases:
1150 *     a variable:  point arg1 at it.
1151 *     a constant:  pass it in arg2.
1152 */
1153
1154int
1155sysctl_handle_int(__unused struct sysctl_oid *oidp, void *arg1, int arg2,
1156        struct sysctl_req *req)
1157{
1158	return sysctl_io_number(req, arg1? *(int*)arg1: arg2, sizeof(int), arg1, NULL);
1159}
1160
1161/*
1162 * Handle a long, signed or unsigned.  arg1 points to it.
1163 */
1164
1165int
1166sysctl_handle_long(__unused struct sysctl_oid *oidp, void *arg1,
1167	__unused int arg2, struct sysctl_req *req)
1168{
1169	if (!arg1)
1170		return (EINVAL);
1171	return sysctl_io_number(req, *(long*)arg1, sizeof(long), arg1, NULL);
1172}
1173
1174/*
1175 * Handle a quad, signed or unsigned.  arg1 points to it.
1176 */
1177
1178int
1179sysctl_handle_quad(__unused struct sysctl_oid *oidp, void *arg1,
1180	__unused int arg2, struct sysctl_req *req)
1181{
1182	if (!arg1)
1183		return (EINVAL);
1184	return sysctl_io_number(req, *(long long*)arg1, sizeof(long long), arg1, NULL);
1185}
1186
1187/*
1188 * Expose an int value as a quad.
1189 *
1190 * This interface allows us to support interfaces defined
1191 * as using quad values while the implementation is still
1192 * using ints.
1193 */
1194int
1195sysctl_handle_int2quad(__unused struct sysctl_oid *oidp, void *arg1,
1196	__unused int arg2, struct sysctl_req *req)
1197{
1198	int error = 0;
1199	long long val;
1200	int newval;
1201
1202	if (!arg1)
1203		return (EINVAL);
1204	val = (long long)*(int *)arg1;
1205	error = SYSCTL_OUT(req, &val, sizeof(long long));
1206
1207	if (error || !req->newptr)
1208		return (error);
1209
1210	error = SYSCTL_IN(req, &val, sizeof(long long));
1211	if (!error) {
1212		/*
1213		 * Value must be representable; check by
1214		 * casting and then casting back.
1215		 */
1216		newval = (int)val;
1217		if ((long long)newval != val) {
1218			error = ERANGE;
1219		} else {
1220			*(int *)arg1 = newval;
1221		}
1222	}
1223	return (error);
1224}
1225
1226/*
1227 * Handle our generic '\0' terminated 'C' string.
1228 * Two cases:
1229 * 	a variable string:  point arg1 at it, arg2 is max length.
1230 * 	a constant string:  point arg1 at it, arg2 is zero.
1231 */
1232
1233int
1234sysctl_handle_string( __unused struct sysctl_oid *oidp, void *arg1, int arg2,
1235        struct sysctl_req *req)
1236{
1237	return sysctl_io_string(req, arg1, arg2, 0, NULL);
1238}
1239
1240/*
1241 * Handle any kind of opaque data.
1242 * arg1 points to it, arg2 is the size.
1243 */
1244
1245int
1246sysctl_handle_opaque(__unused struct sysctl_oid *oidp, void *arg1, int arg2,
1247        struct sysctl_req *req)
1248{
1249	return sysctl_io_opaque(req, arg1, arg2, NULL);
1250}
1251
1252/*
1253 * Transfer functions to/from kernel space.
1254 */
1255STATIC int
1256sysctl_old_kernel(struct sysctl_req *req, const void *p, size_t l)
1257{
1258	size_t i = 0;
1259
1260	if (req->oldptr) {
1261		i = l;
1262		if (i > req->oldlen - req->oldidx)
1263			i = req->oldlen - req->oldidx;
1264		if (i > 0)
1265			bcopy((const void*)p, CAST_DOWN(char *, (req->oldptr + req->oldidx)), i);
1266	}
1267	req->oldidx += l;
1268	if (req->oldptr && i != l)
1269		return (ENOMEM);
1270	return (0);
1271}
1272
1273STATIC int
1274sysctl_new_kernel(struct sysctl_req *req, void *p, size_t l)
1275{
1276	if (!req->newptr)
1277		return 0;
1278	if (req->newlen - req->newidx < l)
1279		return (EINVAL);
1280	bcopy(CAST_DOWN(char *, (req->newptr + req->newidx)), p, l);
1281	req->newidx += l;
1282	return (0);
1283}
1284
1285int
1286kernel_sysctl(struct proc *p, int *name, u_int namelen, void *old, size_t *oldlenp, void *new, size_t newlen)
1287{
1288	int error = 0;
1289	struct sysctl_req req;
1290
1291	/*
1292	 * Construct request.
1293	 */
1294	bzero(&req, sizeof req);
1295	req.p = p;
1296	if (oldlenp)
1297		req.oldlen = *oldlenp;
1298	if (old)
1299		req.oldptr = CAST_USER_ADDR_T(old);
1300	if (newlen) {
1301		req.newlen = newlen;
1302		req.newptr = CAST_USER_ADDR_T(new);
1303	}
1304	req.oldfunc = sysctl_old_kernel;
1305	req.newfunc = sysctl_new_kernel;
1306	req.lock = 1;
1307
1308	/* make the request */
1309	error = sysctl_root(0, name, namelen, &req);
1310
1311	/* unlock memory if required */
1312	if (req.lock == 2)
1313		vsunlock(req.oldptr, (user_size_t)req.oldlen, B_WRITE);
1314
1315	if (error && error != ENOMEM)
1316		return (error);
1317
1318	if (oldlenp)
1319		*oldlenp = req.oldidx;
1320
1321	return (error);
1322}
1323
1324/*
1325 * Transfer function to/from user space.
1326 */
1327STATIC int
1328sysctl_old_user(struct sysctl_req *req, const void *p, size_t l)
1329{
1330	int error = 0;
1331	size_t i = 0;
1332
1333	if (req->oldptr) {
1334                if (req->oldlen - req->oldidx < l)
1335                    return (ENOMEM);
1336		i = l;
1337		if (i > req->oldlen - req->oldidx)
1338			i = req->oldlen - req->oldidx;
1339		if (i > 0)
1340			error = copyout((const void*)p, (req->oldptr + req->oldidx), i);
1341	}
1342	req->oldidx += l;
1343	if (error)
1344		return (error);
1345	if (req->oldptr && i < l)
1346		return (ENOMEM);
1347	return (0);
1348}
1349
1350STATIC int
1351sysctl_new_user(struct sysctl_req *req, void *p, size_t l)
1352{
1353	int error;
1354
1355	if (!req->newptr)
1356		return 0;
1357	if (req->newlen - req->newidx < l)
1358		return (EINVAL);
1359	error = copyin((req->newptr + req->newidx), p, l);
1360	req->newidx += l;
1361	return (error);
1362}
1363
1364/*
1365 * Traverse our tree, and find the right node, execute whatever it points
1366 * at, and return the resulting error code.
1367 */
1368
1369int
1370sysctl_root(__unused struct sysctl_oid *oidp, void *arg1, int arg2,
1371        struct sysctl_req *req)
1372{
1373	int *name = (int *) arg1;
1374	u_int namelen = arg2;
1375	u_int indx;
1376	int i;
1377	struct sysctl_oid *oid;
1378	struct sysctl_oid_list *lsp = &sysctl__children;
1379	int error;
1380	funnel_t *fnl = NULL;
1381	boolean_t funnel_held = FALSE;
1382
1383	/* Get the read lock on the geometry */
1384	lck_rw_lock_shared(sysctl_geometry_lock);
1385
1386	oid = SLIST_FIRST(lsp);
1387
1388	indx = 0;
1389	while (oid && indx < CTL_MAXNAME) {
1390		if (oid->oid_number == name[indx]) {
1391			indx++;
1392			if (!(oid->oid_kind & CTLFLAG_LOCKED))
1393			{
1394/*
1395printf("sysctl_root: missing CTLFLAG_LOCKED: ");
1396for(i = 0; i < (int)(indx - 1); i++)
1397printf("oid[%d] = %d ", i, name[i]);
1398printf("\n");
1399*/
1400				funnel_held = TRUE;
1401			}
1402			if (oid->oid_kind & CTLFLAG_NOLOCK)
1403				req->lock = 0;
1404			/*
1405			 * For SYSCTL_PROC() functions which are for sysctl's
1406			 * which have parameters at the end of their OID
1407			 * space, you need to OR CTLTYPE_NODE into their
1408			 * access value.
1409			 *
1410			 * NOTE: For binary backward compatibility ONLY! Do
1411			 * NOT add new sysctl's that do this!  Existing
1412			 * sysctl's which do this will eventually have
1413			 * compatibility code in user space, and this method
1414			 * will become unsupported.
1415			 */
1416			if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1417				if (oid->oid_handler)
1418					goto found;
1419				if (indx == namelen)
1420				{
1421					error = ENOENT;
1422					goto err;
1423				}
1424
1425				lsp = (struct sysctl_oid_list *)oid->oid_arg1;
1426				oid = SLIST_FIRST(lsp);
1427			} else {
1428				if (indx != namelen)
1429				{
1430					error = EISDIR;
1431					goto err;
1432				}
1433				goto found;
1434			}
1435		} else {
1436			oid = SLIST_NEXT(oid, oid_link);
1437		}
1438	}
1439	error = ENOENT;
1440	goto err;
1441found:
1442	/* If writing isn't allowed */
1443	if (req->newptr && (!(oid->oid_kind & CTLFLAG_WR) ||
1444			    ((oid->oid_kind & CTLFLAG_SECURE) && securelevel > 0))) {
1445		error = (EPERM);
1446		goto err;
1447	}
1448
1449	/*
1450	 * If we're inside the kernel, the OID must be marked as kernel-valid.
1451	 * XXX This mechanism for testing is bad.
1452	 */
1453	if ((req->oldfunc == sysctl_old_kernel) && !(oid->oid_kind & CTLFLAG_KERN))
1454	{
1455		error = (EPERM);
1456		goto err;
1457	}
1458
1459	/*
1460	 * This is where legacy enforcement of permissions occurs.  If the
1461	 * flag does not say CTLFLAG_ANYBODY, then we prohibit anyone but
1462	 * root from writing new values down.  If local enforcement happens
1463	 * at the leaf node, then it needs to be set as CTLFLAG_ANYBODY.  In
1464	 * addition, if the leaf node is set this way, then in order to do
1465	 * specific enforcement, it has to be of type SYSCTL_PROC.
1466	 */
1467	if (!(oid->oid_kind & CTLFLAG_ANYBODY) &&
1468	    req->newptr && req->p &&
1469	    (error = proc_suser(req->p)))
1470		goto err;
1471
1472	if (!oid->oid_handler) {
1473	    error = EINVAL;
1474		goto err;
1475	}
1476
1477	/*
1478	 * Reference the OID and drop the geometry lock; this prevents the
1479	 * OID from being deleted out from under the handler call, but does
1480	 * not prevent other calls into handlers or calls to manage the
1481	 * geometry elsewhere from blocking...
1482	 */
1483	OSAddAtomic(1, &oid->oid_refcnt);
1484
1485	lck_rw_done(sysctl_geometry_lock);
1486
1487	/*
1488	 * ...however, we still have to grab the funnel for those calls which
1489	 * may be into code whose reentrancy is protected by the funnel; a
1490	 * blocking operation should not prevent reentrancy, at this point.
1491	 */
1492	if (funnel_held)
1493	{
1494		fnl = spl_kernel_funnel();
1495	}
1496
1497	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1498		i = (oid->oid_handler) (oid,
1499					name + indx, namelen - indx,
1500					req);
1501	} else {
1502		i = (oid->oid_handler) (oid,
1503					oid->oid_arg1, oid->oid_arg2,
1504					req);
1505	}
1506	error = i;
1507
1508	if (funnel_held)
1509	{
1510		splx_kernel_funnel(fnl);
1511	}
1512
1513	/*
1514	 * This is tricky... we re-grab the geometry lock in order to drop
1515	 * the reference and wake on the address; since the geometry
1516	 * lock is a reader/writer lock rather than a mutex, we have to
1517	 * wake on all apparent 1->0 transitions.  This abuses the drop
1518	 * after the reference decrement in order to wake any lck_rw_sleep()
1519	 * in progress in sysctl_unregister_oid() that slept because of a
1520	 * non-zero reference count.
1521	 *
1522	 * Note:	OSAddAtomic() is defined to return the previous value;
1523	 *		we use this and the fact that the lock itself is a
1524	 *		barrier to avoid waking every time through on "hot"
1525	 *		OIDs.
1526	 */
1527	lck_rw_lock_shared(sysctl_geometry_lock);
1528	if (OSAddAtomic(-1, &oid->oid_refcnt) == 1)
1529		wakeup(&oid->oid_refcnt);
1530
1531err:
1532	lck_rw_done(sysctl_geometry_lock);
1533	return (error);
1534}
1535
1536#ifndef _SYS_SYSPROTO_H_
1537struct sysctl_args {
1538	int	*name;
1539	u_int	namelen;
1540	void	*old;
1541	size_t	*oldlenp;
1542	void	*new;
1543	size_t	newlen;
1544};
1545#endif
1546
1547int
1548/* __sysctl(struct proc *p, struct sysctl_args *uap) */
1549new_sysctl(struct proc *p, struct sysctl_args *uap)
1550{
1551	int error, i, name[CTL_MAXNAME];
1552	size_t j;
1553
1554	if (uap->namelen > CTL_MAXNAME || uap->namelen < 2)
1555		return (EINVAL);
1556
1557 	error = copyin(CAST_USER_ADDR_T(uap->name), &name, uap->namelen * sizeof(int));
1558 	if (error)
1559		return (error);
1560
1561	error = userland_sysctl(p, name, uap->namelen,
1562		                    CAST_USER_ADDR_T(uap->old), uap->oldlenp,
1563		                    CAST_USER_ADDR_T(uap->new), uap->newlen, &j);
1564	if (error && error != ENOMEM)
1565		return (error);
1566	if (uap->oldlenp) {
1567		i = copyout(&j, CAST_USER_ADDR_T(uap->oldlenp), sizeof(j));
1568		if (i)
1569			return (i);
1570	}
1571	return (error);
1572}
1573
1574/*
1575 * This is used from various compatibility syscalls too.  That's why name
1576 * must be in kernel space.
1577 */
1578int
1579userland_sysctl(struct proc *p, int *name, u_int namelen, user_addr_t oldp,
1580                size_t *oldlenp, user_addr_t newp, size_t newlen,
1581                size_t *retval)
1582{
1583	int error = 0;
1584	struct sysctl_req req, req2;
1585
1586	bzero(&req, sizeof req);
1587
1588	req.p = p;
1589
1590	if (oldlenp) {
1591		req.oldlen = *oldlenp;
1592	}
1593
1594	if (oldp) {
1595		req.oldptr = oldp;
1596	}
1597
1598	if (newlen) {
1599		req.newlen = newlen;
1600		req.newptr = newp;
1601	}
1602
1603	req.oldfunc = sysctl_old_user;
1604	req.newfunc = sysctl_new_user;
1605	req.lock = 1;
1606
1607	do {
1608	    req2 = req;
1609	    error = sysctl_root(0, name, namelen, &req2);
1610	} while (error == EAGAIN);
1611
1612	req = req2;
1613
1614	if (error && error != ENOMEM)
1615		return (error);
1616
1617	if (retval) {
1618		if (req.oldptr && req.oldidx > req.oldlen)
1619			*retval = req.oldlen;
1620		else
1621			*retval = req.oldidx;
1622	}
1623	return (error);
1624}
1625
1626/* Non-standard BSDI extension - only present on their 4.3 net-2 releases */
1627#define	KINFO_BSDI_SYSINFO	(101<<8)
1628
1629/*
1630 * Kernel versions of the userland sysctl helper functions.
1631 *
1632 * These allow sysctl to be used in the same fashion in both
1633 * userland and the kernel.
1634 *
1635 * Note that some sysctl handlers use copyin/copyout, which
1636 * may not work correctly.
1637 */
1638
1639STATIC int
1640sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp, void *newp, size_t newlen)
1641{
1642
1643	return(kernel_sysctl(current_proc(), name, namelen, oldp, oldlenp, newp, newlen));
1644}
1645
1646STATIC int
1647sysctlnametomib(const char *name, int *mibp, size_t *sizep)
1648{
1649	int oid[2];
1650	int error;
1651	char *non_const_name;
1652
1653	/*
1654	 * NOTE: This cast is safe because the service node does not modify
1655	 * the contents of the string as part of its operation.
1656	 */
1657	non_const_name = __CAST_AWAY_QUALIFIER(name, const, char *);
1658
1659	/* magic service node */
1660	oid[0] = 0;
1661	oid[1] = 3;
1662
1663	/* look up OID for name */
1664	*sizep *= sizeof(int);
1665	error = sysctl(oid, 2, mibp, sizep, non_const_name, strlen(name));
1666	*sizep /= sizeof(int);
1667
1668	return(error);
1669}
1670
1671int
1672sysctlbyname(const char *name, void *oldp, size_t *oldlenp, void *newp, size_t newlen)
1673{
1674	int oid[CTL_MAXNAME + 2];
1675	int error;
1676	size_t oidlen;
1677
1678	/* look up the OID */
1679	oidlen = CTL_MAXNAME;
1680	error = sysctlnametomib(name, oid, &oidlen);
1681
1682	/* now use the OID */
1683	if (error == 0)
1684		error = sysctl(oid, oidlen, oldp, oldlenp, newp, newlen);
1685	return(error);
1686}
1687
1688