xref: /netbsd-src/sys/kern/kern_sysctl.c (revision d48f14661dda8638fee055ba15d35bdfb29b9fa8)
1 /*	$NetBSD: kern_sysctl.c,v 1.197 2006/06/12 01:25:05 christos Exp $	*/
2 
3 /*-
4  * Copyright (c) 2003 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Andrew Brown.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *      This product includes software developed by the NetBSD
21  *      Foundation, Inc. and its contributors.
22  * 4. Neither the name of The NetBSD Foundation nor the names of its
23  *    contributors may be used to endorse or promote products derived
24  *    from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36  * POSSIBILITY OF SUCH DAMAGE.
37  */
38 
39 /*-
40  * Copyright (c) 1982, 1986, 1989, 1993
41  *	The Regents of the University of California.  All rights reserved.
42  *
43  * This code is derived from software contributed to Berkeley by
44  * Mike Karels at Berkeley Software Design, Inc.
45  *
46  * Redistribution and use in source and binary forms, with or without
47  * modification, are permitted provided that the following conditions
48  * are met:
49  * 1. Redistributions of source code must retain the above copyright
50  *    notice, this list of conditions and the following disclaimer.
51  * 2. Redistributions in binary form must reproduce the above copyright
52  *    notice, this list of conditions and the following disclaimer in the
53  *    documentation and/or other materials provided with the distribution.
54  * 3. Neither the name of the University nor the names of its contributors
55  *    may be used to endorse or promote products derived from this software
56  *    without specific prior written permission.
57  *
58  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
59  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
60  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
61  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
62  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
63  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
64  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
65  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
66  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
67  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
68  * SUCH DAMAGE.
69  *
70  *	@(#)kern_sysctl.c	8.9 (Berkeley) 5/20/95
71  */
72 
73 /*
74  * sysctl system call.
75  */
76 
77 #include <sys/cdefs.h>
78 __KERNEL_RCSID(0, "$NetBSD: kern_sysctl.c,v 1.197 2006/06/12 01:25:05 christos Exp $");
79 
80 #include "opt_defcorename.h"
81 #include "ksyms.h"
82 
83 #include <sys/param.h>
84 #define __COMPAT_SYSCTL
85 #include <sys/sysctl.h>
86 #include <sys/systm.h>
87 #include <sys/buf.h>
88 #include <sys/ksyms.h>
89 #include <sys/malloc.h>
90 #include <sys/mount.h>
91 #include <sys/sa.h>
92 #include <sys/syscallargs.h>
93 #include <sys/kauth.h>
94 #include <machine/stdarg.h>
95 
96 #define	MAXDESCLEN	1024
97 MALLOC_DEFINE(M_SYSCTLNODE, "sysctlnode", "sysctl node structures");
98 MALLOC_DEFINE(M_SYSCTLDATA, "sysctldata", "misc sysctl data");
99 
100 static int sysctl_mmap(SYSCTLFN_PROTO);
101 static int sysctl_alloc(struct sysctlnode *, int);
102 static int sysctl_realloc(struct sysctlnode *);
103 
104 static int sysctl_cvt_in(struct lwp *, int *, const void *, size_t,
105 			 struct sysctlnode *);
106 static int sysctl_cvt_out(struct lwp *, int, const struct sysctlnode *,
107 			  void *, size_t, size_t *);
108 
109 static int sysctl_log_add(struct sysctllog **, const struct sysctlnode *);
110 static int sysctl_log_realloc(struct sysctllog *);
111 
112 struct sysctllog {
113 	const struct sysctlnode *log_root;
114 	int *log_num;
115 	int log_size, log_left;
116 };
117 
118 /*
119  * the "root" of the new sysctl tree
120  */
121 struct sysctlnode sysctl_root = {
122 	.sysctl_flags = SYSCTL_VERSION|
123 	    CTLFLAG_ROOT|CTLFLAG_READWRITE|
124 	    CTLTYPE_NODE,
125 	.sysctl_num = 0,
126 	/*
127 	 * XXX once all ports are on gcc3, we can get rid of this
128 	 * ugliness and simply make it into
129 	 *
130 	 *	.sysctl_size = sizeof(struct sysctlnode),
131 	 */
132 	sysc_init_field(_sysctl_size, sizeof(struct sysctlnode)),
133 	.sysctl_name = "(root)",
134 };
135 
136 /*
137  * link set of functions that add nodes at boot time (see also
138  * sysctl_buildtree())
139  */
140 __link_set_decl(sysctl_funcs, sysctl_setup_func);
141 
142 /*
143  * The `sysctl_lock' is intended to serialize access to the sysctl
144  * tree.  Given that it is now (a) dynamic, and (b) most consumers of
145  * sysctl are going to be copying data out, the old `sysctl_memlock'
146  * has been `upgraded' to simply guard the whole tree.
147  *
148  * The two new data here are to keep track of the locked chunk of
149  * memory, if there is one, so that it can be released more easily
150  * from anywhere.
151  */
152 struct lock sysctl_treelock;
153 caddr_t sysctl_memaddr;
154 size_t sysctl_memsize;
155 
156 /*
157  * Attributes stored in the kernel.
158  */
159 char hostname[MAXHOSTNAMELEN];
160 int hostnamelen;
161 
162 char domainname[MAXHOSTNAMELEN];
163 int domainnamelen;
164 
165 long hostid;
166 
167 #ifndef DEFCORENAME
168 #define	DEFCORENAME	"%n.core"
169 #endif
170 char defcorename[MAXPATHLEN] = DEFCORENAME;
171 
172 /*
173  * ********************************************************************
174  * Section 0: Some simple glue
175  * ********************************************************************
176  * By wrapping copyin(), copyout(), and copyinstr() like this, we can
177  * stop caring about who's calling us and simplify some code a bunch.
178  * ********************************************************************
179  */
180 static inline int
181 sysctl_copyin(const struct lwp *l, const void *uaddr, void *kaddr, size_t len)
182 {
183 
184 	if (l != NULL)
185 		return (copyin(uaddr, kaddr, len));
186 	else
187 		return (kcopy(uaddr, kaddr, len));
188 }
189 
190 static inline int
191 sysctl_copyout(const struct lwp *l, const void *kaddr, void *uaddr, size_t len)
192 {
193 
194 	if (l != NULL)
195 		return (copyout(kaddr, uaddr, len));
196 	else
197 		return (kcopy(kaddr, uaddr, len));
198 }
199 
200 static inline int
201 sysctl_copyinstr(const struct lwp *l, const void *uaddr, void *kaddr,
202 		 size_t len, size_t *done)
203 {
204 
205 	if (l != NULL)
206 		return (copyinstr(uaddr, kaddr, len, done));
207 	else
208 		return (copystr(uaddr, kaddr, len, done));
209 }
210 
211 /*
212  * ********************************************************************
213  * Initialize sysctl subsystem.
214  * ********************************************************************
215  */
216 void
217 sysctl_init(void)
218 {
219 	sysctl_setup_func * const *sysctl_setup, f;
220 
221 	lockinit(&sysctl_treelock, PRIBIO|PCATCH, "sysctl", 0, 0);
222 
223 	/*
224 	 * dynamic mib numbers start here
225 	 */
226 	sysctl_root.sysctl_num = CREATE_BASE;
227 
228         __link_set_foreach(sysctl_setup, sysctl_funcs) {
229 		/*
230 		 * XXX - why do i have to coerce the pointers like this?
231 		 */
232 		f = (void*)*sysctl_setup;
233 		(*f)(NULL);
234 	}
235 
236 	/*
237 	 * setting this means no more permanent nodes can be added,
238 	 * trees that claim to be readonly at the root now are, and if
239 	 * the main tree is readonly, *everything* is.
240 	 */
241 	sysctl_root.sysctl_flags |= CTLFLAG_PERMANENT;
242 
243 }
244 
245 /*
246  * ********************************************************************
247  * The main native sysctl system call itself.
248  * ********************************************************************
249  */
250 int
251 sys___sysctl(struct lwp *l, void *v, register_t *retval)
252 {
253 	struct sys___sysctl_args /* {
254 		syscallarg(const int *) name;
255 		syscallarg(u_int) namelen;
256 		syscallarg(void *) old;
257 		syscallarg(size_t *) oldlenp;
258 		syscallarg(const void *) new;
259 		syscallarg(size_t) newlen;
260 	} */ *uap = v;
261 	int error, nerror, name[CTL_MAXNAME];
262 	size_t oldlen, savelen, *oldlenp;
263 
264 	/*
265 	 * get oldlen
266 	 */
267 	oldlen = 0;
268 	oldlenp = SCARG(uap, oldlenp);
269 	if (oldlenp != NULL) {
270 		error = copyin(oldlenp, &oldlen, sizeof(oldlen));
271 		if (error)
272 			return (error);
273 	}
274 	savelen = oldlen;
275 
276 	/*
277 	 * top-level sysctl names may or may not be non-terminal, but
278 	 * we don't care
279 	 */
280 	if (SCARG(uap, namelen) > CTL_MAXNAME || SCARG(uap, namelen) < 1)
281 		return (EINVAL);
282 	error = copyin(SCARG(uap, name), &name,
283 		       SCARG(uap, namelen) * sizeof(int));
284 	if (error)
285 		return (error);
286 
287 	/*
288 	 * wire old so that copyout() is less likely to fail?
289 	 */
290 	error = sysctl_lock(l, SCARG(uap, old), savelen);
291 	if (error)
292 		return (error);
293 
294 	/*
295 	 * do sysctl work (NULL means main built-in default tree)
296 	 */
297 	error = sysctl_dispatch(&name[0], SCARG(uap, namelen),
298 				SCARG(uap, old), &oldlen,
299 				SCARG(uap, new), SCARG(uap, newlen),
300 				&name[0], l, NULL);
301 
302 	/*
303 	 * release the sysctl lock
304 	 */
305 	sysctl_unlock(l);
306 
307 	/*
308 	 * set caller's oldlen to new value even in the face of an
309 	 * error (if this gets an error and they didn't have one, they
310 	 * get this one)
311 	 */
312 	if (oldlenp) {
313 		nerror = copyout(&oldlen, oldlenp, sizeof(oldlen));
314 		if (error == 0)
315 			error = nerror;
316 	}
317 
318 	/*
319 	 * if the only problem is that we weren't given enough space,
320 	 * that's an ENOMEM error
321 	 */
322 	if (error == 0 && SCARG(uap, old) != NULL && savelen < oldlen)
323 		error = ENOMEM;
324 
325 	return (error);
326 }
327 
328 /*
329  * ********************************************************************
330  * Section 1: How the tree is used
331  * ********************************************************************
332  * Implementations of sysctl for emulations should typically need only
333  * these three functions in this order: lock the tree, dispatch
334  * request into it, unlock the tree.
335  * ********************************************************************
336  */
337 int
338 sysctl_lock(struct lwp *l, void *oldp, size_t savelen)
339 {
340 	int error = 0;
341 
342 	error = lockmgr(&sysctl_treelock, LK_EXCLUSIVE, NULL);
343 	if (error)
344 		return (error);
345 
346 	if (l != NULL && oldp != NULL && savelen) {
347 		/*
348 		 * be lazy - memory is locked for short time only, so
349 		 * just do a basic check against system limit
350 		 */
351 		if (uvmexp.wired + atop(savelen) > uvmexp.wiredmax) {
352 			lockmgr(&sysctl_treelock, LK_RELEASE, NULL);
353 			return (ENOMEM);
354 		}
355 		error = uvm_vslock(l->l_proc, oldp, savelen, VM_PROT_WRITE);
356 		if (error) {
357 			(void) lockmgr(&sysctl_treelock, LK_RELEASE, NULL);
358 			return (error);
359 		}
360 		sysctl_memaddr = oldp;
361 		sysctl_memsize = savelen;
362 	}
363 
364 	return (0);
365 }
366 
367 /*
368  * ********************************************************************
369  * the main sysctl dispatch routine.  scans the given tree and picks a
370  * function to call based on what it finds.
371  * ********************************************************************
372  */
373 int
374 sysctl_dispatch(SYSCTLFN_ARGS)
375 {
376 	int error;
377 	sysctlfn fn;
378 	int ni;
379 
380 	if (rnode && SYSCTL_VERS(rnode->sysctl_flags) != SYSCTL_VERSION) {
381 		printf("sysctl_dispatch: rnode %p wrong version\n", rnode);
382 		return (EINVAL);
383 	}
384 
385 	fn = NULL;
386 	error = sysctl_locate(l, name, namelen, &rnode, &ni);
387 
388 	if (rnode->sysctl_func != NULL) {
389 		/*
390 		 * the node we ended up at has a function, so call it.  it can
391 		 * hand off to query or create if it wants to.
392 		 */
393 		fn = rnode->sysctl_func;
394 	} else if (error == 0) {
395 		/*
396 		 * we found the node they were looking for, so do a lookup.
397 		 */
398 		fn = (sysctlfn)sysctl_lookup; /* XXX may write to rnode */
399 	} else if (error == ENOENT && (ni + 1) == namelen && name[ni] < 0) {
400 		/*
401 		 * prospective parent node found, but the terminal node was
402 		 * not.  generic operations associate with the parent.
403 		 */
404 		switch (name[ni]) {
405 		case CTL_QUERY:
406 			fn = sysctl_query;
407 			break;
408 		case CTL_CREATE:
409 #if NKSYMS > 0
410 		case CTL_CREATESYM:
411 #endif /* NKSYMS > 0 */
412 			fn = (sysctlfn)sysctl_create; /* we own the rnode */
413 			break;
414 		case CTL_DESTROY:
415 			fn = (sysctlfn)sysctl_destroy; /* we own the rnode */
416 			break;
417 		case CTL_MMAP:
418 			fn = (sysctlfn)sysctl_mmap; /* we own the rnode */
419 			break;
420 		case CTL_DESCRIBE:
421 			fn = sysctl_describe;
422 			break;
423 		default:
424 			error = EOPNOTSUPP;
425 			break;
426 		}
427 	}
428 
429 	/*
430 	 * after all of that, maybe we found someone who knows how to
431 	 * get us what we want?
432 	 */
433 	if (fn != NULL)
434 		error = (*fn)(name + ni, namelen - ni, oldp, oldlenp,
435 			      newp, newlen, name, l, rnode);
436 	else if (error == 0)
437 		error = EOPNOTSUPP;
438 
439 	return (error);
440 }
441 
442 /*
443  * ********************************************************************
444  * Releases the tree lock.  Note that if uvm_vslock() was called when
445  * the lock was taken, we release that memory now.  By keeping track
446  * of where and how much by ourselves, the lock can be released much
447  * more easily from anywhere.
448  * ********************************************************************
449  */
450 void
451 sysctl_unlock(struct lwp *l)
452 {
453 
454 	if (l != NULL && sysctl_memsize != 0) {
455 		uvm_vsunlock(l->l_proc, sysctl_memaddr, sysctl_memsize);
456 		sysctl_memsize = 0;
457 	}
458 
459 	(void) lockmgr(&sysctl_treelock, LK_RELEASE, NULL);
460 }
461 
462 /*
463  * ********************************************************************
464  * Section 2: The main tree interfaces
465  * ********************************************************************
466  * This is how sysctl_dispatch() does its work, and you can too, by
467  * calling these routines from helpers (though typically only
468  * sysctl_lookup() will be used).  The tree MUST BE LOCKED when these
469  * are called.
470  * ********************************************************************
471  */
472 
473 /*
474  * sysctl_locate -- Finds the node matching the given mib under the
475  * given tree (via rv).  If no tree is given, we fall back to the
476  * native tree.  The current process (via l) is used for access
477  * control on the tree (some nodes may be traversable only by root) and
478  * on return, nip will show how many numbers in the mib were consumed.
479  */
480 int
481 sysctl_locate(struct lwp *l, const int *name, u_int namelen,
482 	      const struct sysctlnode **rnode, int *nip)
483 {
484 	const struct sysctlnode *node, *pnode;
485 	int tn, si, ni, error, alias;
486 
487 	/*
488 	 * basic checks and setup
489 	 */
490 	if (*rnode == NULL)
491 		*rnode = &sysctl_root;
492 	if (nip)
493 		*nip = 0;
494 	if (namelen < 0)
495 		return (EINVAL);
496 	if (namelen == 0)
497 		return (0);
498 
499 	/*
500 	 * search starts from "root"
501 	 */
502 	pnode = *rnode;
503 	if (SYSCTL_VERS(pnode->sysctl_flags) != SYSCTL_VERSION) {
504 		printf("sysctl_locate: pnode %p wrong version\n", pnode);
505 		return (EINVAL);
506 	}
507 	node = pnode->sysctl_child;
508 	error = 0;
509 
510 	/*
511 	 * scan for node to which new node should be attached
512 	 */
513 	for (ni = 0; ni < namelen; ni++) {
514 		/*
515 		 * walked off bottom of tree
516 		 */
517 		if (node == NULL) {
518 			if (SYSCTL_TYPE(pnode->sysctl_flags) == CTLTYPE_NODE)
519 				error = ENOENT;
520 			else
521 				error = ENOTDIR;
522 			break;
523 		}
524 		/*
525 		 * can anyone traverse this node or only root?
526 		 */
527 		if (l != NULL && (pnode->sysctl_flags & CTLFLAG_PRIVATE) &&
528 		    (error = kauth_authorize_generic(l->l_proc->p_cred,
529 				KAUTH_GENERIC_ISSUSER, &l->l_proc->p_acflag))
530 		    != 0)
531 			return (error);
532 		/*
533 		 * find a child node with the right number
534 		 */
535 		tn = name[ni];
536 		alias = 0;
537 
538 		si = 0;
539 		/*
540 		 * Note: ANYNUMBER only matches positive integers.
541 		 * Since ANYNUMBER is only permitted on single-node
542 		 * sub-trees (eg proc), check before the loop and skip
543 		 * it if we can.
544 		 */
545 		if ((node[si].sysctl_flags & CTLFLAG_ANYNUMBER) && (tn >= 0))
546 			goto foundit;
547 		for (; si < pnode->sysctl_clen; si++) {
548 			if (node[si].sysctl_num == tn) {
549 				if (node[si].sysctl_flags & CTLFLAG_ALIAS) {
550 					if (alias++ == 4)
551 						break;
552 					else {
553 						tn = node[si].sysctl_alias;
554 						si = -1;
555 					}
556 				} else
557 					goto foundit;
558 			}
559 		}
560 		/*
561 		 * if we ran off the end, it obviously doesn't exist
562 		 */
563 		error = ENOENT;
564 		break;
565 
566 		/*
567 		 * so far so good, move on down the line
568 		 */
569 	  foundit:
570 		pnode = &node[si];
571 		if (SYSCTL_TYPE(pnode->sysctl_flags) == CTLTYPE_NODE)
572 			node = node[si].sysctl_child;
573 		else
574 			node = NULL;
575 	}
576 
577 	*rnode = pnode;
578 	if (nip)
579 		*nip = ni;
580 
581 	return (error);
582 }
583 
584 /*
585  * sysctl_query -- The auto-discovery engine.  Copies out the structs
586  * describing nodes under the given node and handles overlay trees.
587  */
588 int
589 sysctl_query(SYSCTLFN_ARGS)
590 {
591 	int error, ni, elim, v;
592 	size_t out, left, t;
593 	const struct sysctlnode *enode, *onode;
594 	struct sysctlnode qnode;
595 
596 	if (SYSCTL_VERS(rnode->sysctl_flags) != SYSCTL_VERSION) {
597 		printf("sysctl_query: rnode %p wrong version\n", rnode);
598 		return (EINVAL);
599 	}
600 
601 	if (SYSCTL_TYPE(rnode->sysctl_flags) != CTLTYPE_NODE)
602 		return (ENOTDIR);
603 	if (namelen != 1 || name[0] != CTL_QUERY)
604 		return (EINVAL);
605 
606 	error = 0;
607 	out = 0;
608 	left = *oldlenp;
609 	elim = 0;
610 	enode = NULL;
611 
612 	/*
613 	 * translate the given request to a current node
614 	 */
615 	error = sysctl_cvt_in(l, &v, newp, newlen, &qnode);
616 	if (error)
617 		return (error);
618 
619 	/*
620 	 * if the request specifies a version, check it
621 	 */
622 	if (qnode.sysctl_ver != 0) {
623 		enode = rnode;
624 		if (qnode.sysctl_ver != enode->sysctl_ver &&
625 		    qnode.sysctl_ver != sysctl_rootof(enode)->sysctl_ver)
626 			return (EINVAL);
627 	}
628 
629 	/*
630 	 * process has overlay tree
631 	 */
632 	if (l && l->l_proc->p_emul->e_sysctlovly) {
633 		enode = l->l_proc->p_emul->e_sysctlovly;
634 		elim = (name - oname);
635 		error = sysctl_locate(l, oname, elim, &enode, NULL);
636 		if (error == 0) {
637 			/* ah, found parent in overlay */
638 			elim = enode->sysctl_clen;
639 			enode = enode->sysctl_child;
640 		} else {
641 			error = 0;
642 			elim = 0;
643 			enode = NULL;
644 		}
645 	}
646 
647 	for (ni = 0; ni < rnode->sysctl_clen; ni++) {
648 		onode = &rnode->sysctl_child[ni];
649 		if (enode && enode->sysctl_num == onode->sysctl_num) {
650 			if (SYSCTL_TYPE(enode->sysctl_flags) != CTLTYPE_NODE)
651 				onode = enode;
652 			if (--elim > 0)
653 				enode++;
654 			else
655 				enode = NULL;
656 		}
657 		error = sysctl_cvt_out(l, v, onode, oldp, left, &t);
658 		if (error)
659 			return (error);
660 		if (oldp != NULL)
661 			oldp = (char*)oldp + t;
662 		out += t;
663 		left -= MIN(left, t);
664 	}
665 
666 	/*
667 	 * overlay trees *MUST* be entirely consumed
668 	 */
669 	KASSERT(enode == NULL);
670 
671 	*oldlenp = out;
672 
673 	return (error);
674 }
675 
676 #ifdef SYSCTL_DEBUG_CREATE
677 #undef sysctl_create
678 #endif /* SYSCTL_DEBUG_CREATE */
679 
680 /*
681  * sysctl_create -- Adds a node (the description of which is taken
682  * from newp) to the tree, returning a copy of it in the space pointed
683  * to by oldp.  In the event that the requested slot is already taken
684  * (either by name or by number), the offending node is returned
685  * instead.  Yes, this is complex, but we want to make sure everything
686  * is proper.
687  */
688 int
689 sysctl_create(SYSCTLFN_ARGS)
690 {
691 	struct sysctlnode nnode, *node, *pnode;
692 	int error, ni, at, nm, type, sz, flags, anum, v;
693 	void *own;
694 
695 	error = 0;
696 	own = NULL;
697 	anum = -1;
698 
699 	if (SYSCTL_VERS(rnode->sysctl_flags) != SYSCTL_VERSION) {
700 		printf("sysctl_create: rnode %p wrong version\n", rnode);
701 		return (EINVAL);
702 	}
703 
704 	if (namelen != 1 || (name[namelen - 1] != CTL_CREATE
705 #if NKSYMS > 0
706 			     && name[namelen - 1] != CTL_CREATESYM
707 #endif /* NKSYMS > 0 */
708 			     ))
709 		return (EINVAL);
710 
711 	/*
712 	 * processes can only add nodes at securelevel 0, must be
713 	 * root, and can't add nodes to a parent that's not writeable
714 	 */
715 	if (l != NULL) {
716 #ifndef SYSCTL_DISALLOW_CREATE
717 		if (securelevel > 0)
718 			return (EPERM);
719 		error = kauth_authorize_generic(l->l_proc->p_cred,
720 			    KAUTH_GENERIC_ISSUSER, &l->l_proc->p_acflag);
721 		if (error)
722 			return (error);
723 		if (!(rnode->sysctl_flags & CTLFLAG_READWRITE))
724 #endif /* SYSCTL_DISALLOW_CREATE */
725 			return (EPERM);
726 	}
727 
728 	/*
729 	 * nothing can add a node if:
730 	 * we've finished initial set up and
731 	 * the tree itself is not writeable or
732 	 * the entire sysctl system is not writeable
733 	 */
734 	if ((sysctl_root.sysctl_flags & CTLFLAG_PERMANENT) &&
735 	    (!(sysctl_rootof(rnode)->sysctl_flags & CTLFLAG_READWRITE) ||
736 	     !(sysctl_root.sysctl_flags & CTLFLAG_READWRITE)))
737 		return (EPERM);
738 
739 	/*
740 	 * it must be a "node", not a "int" or something
741 	 */
742 	if (SYSCTL_TYPE(rnode->sysctl_flags) != CTLTYPE_NODE)
743 		return (ENOTDIR);
744 	if (rnode->sysctl_flags & CTLFLAG_ALIAS) {
745 		printf("sysctl_create: attempt to add node to aliased "
746 		       "node %p\n", rnode);
747 		return (EINVAL);
748 	}
749 	pnode = __UNCONST(rnode); /* we are adding children to this node */
750 
751 	if (newp == NULL)
752 		return (EINVAL);
753 	error = sysctl_cvt_in(l, &v, newp, newlen, &nnode);
754 	if (error)
755 		return (error);
756 
757 	/*
758 	 * nodes passed in don't *have* parents
759 	 */
760 	if (nnode.sysctl_parent != NULL)
761 		return (EINVAL);
762 
763 	/*
764 	 * if we are indeed adding it, it should be a "good" name and
765 	 * number
766 	 */
767 	nm = nnode.sysctl_num;
768 #if NKSYMS > 0
769 	if (nm == CTL_CREATESYM)
770 		nm = CTL_CREATE;
771 #endif /* NKSYMS > 0 */
772 	if (nm < 0 && nm != CTL_CREATE)
773 		return (EINVAL);
774 	sz = 0;
775 
776 	/*
777 	 * the name can't start with a digit
778 	 */
779 	if (nnode.sysctl_name[sz] >= '0' &&
780 	    nnode.sysctl_name[sz] <= '9')
781 		return (EINVAL);
782 
783 	/*
784 	 * the name must be only alphanumerics or - or _, longer than
785 	 * 0 bytes and less that SYSCTL_NAMELEN
786 	 */
787 	while (sz < SYSCTL_NAMELEN && nnode.sysctl_name[sz] != '\0') {
788 		if ((nnode.sysctl_name[sz] >= '0' &&
789 		     nnode.sysctl_name[sz] <= '9') ||
790 		    (nnode.sysctl_name[sz] >= 'A' &&
791 		     nnode.sysctl_name[sz] <= 'Z') ||
792 		    (nnode.sysctl_name[sz] >= 'a' &&
793 		     nnode.sysctl_name[sz] <= 'z') ||
794 		    nnode.sysctl_name[sz] == '-' ||
795 		    nnode.sysctl_name[sz] == '_')
796 			sz++;
797 		else
798 			return (EINVAL);
799 	}
800 	if (sz == 0 || sz == SYSCTL_NAMELEN)
801 		return (EINVAL);
802 
803 	/*
804 	 * various checks revolve around size vs type, etc
805 	 */
806 	type = SYSCTL_TYPE(nnode.sysctl_flags);
807 	flags = SYSCTL_FLAGS(nnode.sysctl_flags);
808 	sz = nnode.sysctl_size;
809 
810 	/*
811 	 * find out if there's a collision, and if so, let the caller
812 	 * know what they collided with
813 	 */
814 	node = pnode->sysctl_child;
815 	at = 0;
816 	if (node) {
817 		if ((flags | node->sysctl_flags) & CTLFLAG_ANYNUMBER)
818 			/* No siblings for a CTLFLAG_ANYNUMBER node */
819 			return EINVAL;
820 		for (ni = 0; ni < pnode->sysctl_clen; ni++) {
821 			if (nm == node[ni].sysctl_num ||
822 			    strcmp(nnode.sysctl_name, node[ni].sysctl_name) == 0) {
823 				/*
824 				 * ignore error here, since we
825 				 * are already fixed on EEXIST
826 				 */
827 				(void)sysctl_cvt_out(l, v, &node[ni], oldp,
828 						     *oldlenp, oldlenp);
829 				return (EEXIST);
830 			}
831 			if (nm > node[ni].sysctl_num)
832 				at++;
833 		}
834 	}
835 
836 	/*
837 	 * use sysctl_ver to add to the tree iff it hasn't changed
838 	 */
839 	if (nnode.sysctl_ver != 0) {
840 		/*
841 		 * a specified value must match either the parent
842 		 * node's version or the root node's version
843 		 */
844 		if (nnode.sysctl_ver != sysctl_rootof(rnode)->sysctl_ver &&
845 		    nnode.sysctl_ver != rnode->sysctl_ver) {
846 			return (EINVAL);
847 		}
848 	}
849 
850 	/*
851 	 * only the kernel can assign functions to entries
852 	 */
853 	if (l != NULL && nnode.sysctl_func != NULL)
854 		return (EPERM);
855 
856 	/*
857 	 * only the kernel can create permanent entries, and only then
858 	 * before the kernel is finished setting itself up
859 	 */
860 	if (l != NULL && (flags & ~SYSCTL_USERFLAGS))
861 		return (EPERM);
862 	if ((flags & CTLFLAG_PERMANENT) &
863 	    (sysctl_root.sysctl_flags & CTLFLAG_PERMANENT))
864 		return (EPERM);
865 	if ((flags & (CTLFLAG_OWNDATA | CTLFLAG_IMMEDIATE)) ==
866 	    (CTLFLAG_OWNDATA | CTLFLAG_IMMEDIATE))
867 		return (EINVAL);
868 	if ((flags & CTLFLAG_IMMEDIATE) &&
869 	    type != CTLTYPE_INT && type != CTLTYPE_QUAD)
870 		return (EINVAL);
871 
872 	/*
873 	 * check size, or set it if unset and we can figure it out.
874 	 * kernel created nodes are allowed to have a function instead
875 	 * of a size (or a data pointer).
876 	 */
877 	switch (type) {
878 	case CTLTYPE_NODE:
879 		/*
880 		 * only *i* can assert the size of a node
881 		 */
882 		if (flags & CTLFLAG_ALIAS) {
883 			anum = nnode.sysctl_alias;
884 			if (anum < 0)
885 				return (EINVAL);
886 			nnode.sysctl_alias = 0;
887 		}
888 		if (sz != 0 || nnode.sysctl_data != NULL)
889 			return (EINVAL);
890 		if (nnode.sysctl_csize != 0 ||
891 		    nnode.sysctl_clen != 0 ||
892 		    nnode.sysctl_child != 0)
893 			return (EINVAL);
894 		if (flags & CTLFLAG_OWNDATA)
895 			return (EINVAL);
896 		sz = sizeof(struct sysctlnode);
897 		break;
898 	case CTLTYPE_INT:
899 		/*
900 		 * since an int is an int, if the size is not given or
901 		 * is wrong, we can "int-uit" it.
902 		 */
903 		if (sz != 0 && sz != sizeof(int))
904 			return (EINVAL);
905 		sz = sizeof(int);
906 		break;
907 	case CTLTYPE_STRING:
908 		/*
909 		 * strings are a little more tricky
910 		 */
911 		if (sz == 0) {
912 			if (l == NULL) {
913 				if (nnode.sysctl_func == NULL) {
914 					if (nnode.sysctl_data == NULL)
915 						return (EINVAL);
916 					else
917 						sz = strlen(nnode.sysctl_data) +
918 						    1;
919 				}
920 			} else if (nnode.sysctl_data == NULL &&
921 				 flags & CTLFLAG_OWNDATA) {
922 				return (EINVAL);
923 			} else {
924 				char *vp, *e;
925 				size_t s;
926 
927 				/*
928 				 * we want a rough idea of what the
929 				 * size is now
930 				 */
931 				vp = malloc(PAGE_SIZE, M_SYSCTLDATA,
932 					     M_WAITOK|M_CANFAIL);
933 				if (vp == NULL)
934 					return (ENOMEM);
935 				e = nnode.sysctl_data;
936 				do {
937 					error = copyinstr(e, vp, PAGE_SIZE, &s);
938 					if (error) {
939 						if (error != ENAMETOOLONG) {
940 							free(vp, M_SYSCTLDATA);
941 							return (error);
942 						}
943 						e += PAGE_SIZE;
944 						if ((e - 32 * PAGE_SIZE) >
945 						    (char*)nnode.sysctl_data) {
946 							free(vp, M_SYSCTLDATA);
947 							return (ERANGE);
948 						}
949 					}
950 				} while (error != 0);
951 				sz = s + (e - (char*)nnode.sysctl_data);
952 				free(vp, M_SYSCTLDATA);
953 			}
954 		}
955 		break;
956 	case CTLTYPE_QUAD:
957 		if (sz != 0 && sz != sizeof(u_quad_t))
958 			return (EINVAL);
959 		sz = sizeof(u_quad_t);
960 		break;
961 	case CTLTYPE_STRUCT:
962 		if (sz == 0) {
963 			if (l != NULL || nnode.sysctl_func == NULL)
964 				return (EINVAL);
965 			if (flags & CTLFLAG_OWNDATA)
966 				return (EINVAL);
967 		}
968 		break;
969 	default:
970 		return (EINVAL);
971 	}
972 
973 	/*
974 	 * at this point, if sz is zero, we *must* have a
975 	 * function to go with it and we can't own it.
976 	 */
977 
978 	/*
979 	 *  l  ptr own
980 	 *  0   0   0  -> EINVAL (if no func)
981 	 *  0   0   1  -> own
982 	 *  0   1   0  -> kptr
983 	 *  0   1   1  -> kptr
984 	 *  1   0   0  -> EINVAL
985 	 *  1   0   1  -> own
986 	 *  1   1   0  -> kptr, no own (fault on lookup)
987 	 *  1   1   1  -> uptr, own
988 	 */
989 	if (type != CTLTYPE_NODE) {
990 		if (sz != 0) {
991 			if (flags & CTLFLAG_OWNDATA) {
992 				own = malloc(sz, M_SYSCTLDATA,
993 					     M_WAITOK|M_CANFAIL);
994 				if (own == NULL)
995 					return ENOMEM;
996 				if (nnode.sysctl_data == NULL)
997 					memset(own, 0, sz);
998 				else {
999 					error = sysctl_copyin(l,
1000 					    nnode.sysctl_data, own, sz);
1001 					if (error != 0) {
1002 						free(own, M_SYSCTLDATA);
1003 						return (error);
1004 					}
1005 				}
1006 			} else if ((nnode.sysctl_data != NULL) &&
1007 				 !(flags & CTLFLAG_IMMEDIATE)) {
1008 #if NKSYMS > 0
1009 				if (name[namelen - 1] == CTL_CREATESYM) {
1010 					char symname[128]; /* XXX enough? */
1011 					u_long symaddr;
1012 					size_t symlen;
1013 
1014 					error = sysctl_copyinstr(l,
1015 					    nnode.sysctl_data, symname,
1016 					    sizeof(symname), &symlen);
1017 					if (error)
1018 						return (error);
1019 					error = ksyms_getval(NULL, symname,
1020 					    &symaddr, KSYMS_EXTERN);
1021 					if (error)
1022 						return (error); /* EINVAL? */
1023 					nnode.sysctl_data = (void*)symaddr;
1024 				}
1025 #endif /* NKSYMS > 0 */
1026 				/*
1027 				 * Ideally, we'd like to verify here
1028 				 * that this address is acceptable,
1029 				 * but...
1030 				 *
1031 				 * - it might be valid now, only to
1032 				 *   become invalid later
1033 				 *
1034 				 * - it might be invalid only for the
1035 				 *   moment and valid later
1036 				 *
1037 				 * - or something else.
1038 				 *
1039 				 * Since we can't get a good answer,
1040 				 * we'll just accept the address as
1041 				 * given, and fault on individual
1042 				 * lookups.
1043 				 */
1044 			}
1045 		} else if (nnode.sysctl_func == NULL)
1046 			return (EINVAL);
1047 	}
1048 
1049 	/*
1050 	 * a process can't assign a function to a node, and the kernel
1051 	 * can't create a node that has no function or data.
1052 	 * (XXX somewhat redundant check)
1053 	 */
1054 	if (l != NULL || nnode.sysctl_func == NULL) {
1055 		if (type != CTLTYPE_NODE &&
1056 		    nnode.sysctl_data == NULL &&
1057 		    !(flags & CTLFLAG_IMMEDIATE) &&
1058 		    own == NULL)
1059 			return (EINVAL);
1060 	}
1061 
1062 #ifdef SYSCTL_DISALLOW_KWRITE
1063 	/*
1064 	 * a process can't create a writable node unless it refers to
1065 	 * new data.
1066 	 */
1067 	if (l != NULL && own == NULL && type != CTLTYPE_NODE &&
1068 	    (flags & CTLFLAG_READWRITE) != CTLFLAG_READONLY &&
1069 	    !(flags & CTLFLAG_IMMEDIATE))
1070 		return (EPERM);
1071 #endif /* SYSCTL_DISALLOW_KWRITE */
1072 
1073 	/*
1074 	 * make sure there's somewhere to put the new stuff.
1075 	 */
1076 	if (pnode->sysctl_child == NULL) {
1077 		if (flags & CTLFLAG_ANYNUMBER)
1078 			error = sysctl_alloc(pnode, 1);
1079 		else
1080 			error = sysctl_alloc(pnode, 0);
1081 		if (error) {
1082 			if (own != NULL)
1083 				free(own, M_SYSCTLDATA);
1084 			return (error);
1085 		}
1086 	}
1087 	node = pnode->sysctl_child;
1088 
1089 	/*
1090 	 * no collisions, so pick a good dynamic number if we need to.
1091 	 */
1092 	if (nm == CTL_CREATE) {
1093 		nm = ++sysctl_root.sysctl_num;
1094 		for (ni = 0; ni < pnode->sysctl_clen; ni++) {
1095 			if (nm == node[ni].sysctl_num) {
1096 				nm++;
1097 				ni = -1;
1098 			} else if (nm > node[ni].sysctl_num)
1099 				at = ni + 1;
1100 		}
1101 	}
1102 
1103 	/*
1104 	 * oops...ran out of space
1105 	 */
1106 	if (pnode->sysctl_clen == pnode->sysctl_csize) {
1107 		error = sysctl_realloc(pnode);
1108 		if (error) {
1109 			if (own != NULL)
1110 				free(own, M_SYSCTLDATA);
1111 			return (error);
1112 		}
1113 		node = pnode->sysctl_child;
1114 	}
1115 
1116 	/*
1117 	 * insert new node data
1118 	 */
1119 	if (at < pnode->sysctl_clen) {
1120 		int t;
1121 
1122 		/*
1123 		 * move the nodes that should come after the new one
1124 		 */
1125 		memmove(&node[at + 1], &node[at],
1126 			(pnode->sysctl_clen - at) * sizeof(struct sysctlnode));
1127 		memset(&node[at], 0, sizeof(struct sysctlnode));
1128 		node[at].sysctl_parent = pnode;
1129 		/*
1130 		 * and...reparent any children of any moved nodes
1131 		 */
1132 		for (ni = at; ni <= pnode->sysctl_clen; ni++)
1133 			if (SYSCTL_TYPE(node[ni].sysctl_flags) == CTLTYPE_NODE)
1134 				for (t = 0; t < node[ni].sysctl_clen; t++)
1135 					node[ni].sysctl_child[t].sysctl_parent =
1136 						&node[ni];
1137 	}
1138 	node = &node[at];
1139 	pnode->sysctl_clen++;
1140 
1141 	strlcpy(node->sysctl_name, nnode.sysctl_name,
1142 		sizeof(node->sysctl_name));
1143 	node->sysctl_num = nm;
1144 	node->sysctl_size = sz;
1145 	node->sysctl_flags = SYSCTL_VERSION|type|flags; /* XXX other trees */
1146 	node->sysctl_csize = 0;
1147 	node->sysctl_clen = 0;
1148 	if (own) {
1149 		node->sysctl_data = own;
1150 		node->sysctl_flags |= CTLFLAG_OWNDATA;
1151 	} else if (flags & CTLFLAG_ALIAS) {
1152 		node->sysctl_alias = anum;
1153 	} else if (flags & CTLFLAG_IMMEDIATE) {
1154 		switch (type) {
1155 		case CTLTYPE_INT:
1156 			node->sysctl_idata = nnode.sysctl_idata;
1157 			break;
1158 		case CTLTYPE_QUAD:
1159 			node->sysctl_qdata = nnode.sysctl_qdata;
1160 			break;
1161 		}
1162 	} else {
1163 		node->sysctl_data = nnode.sysctl_data;
1164 		node->sysctl_flags &= ~CTLFLAG_OWNDATA;
1165 	}
1166         node->sysctl_func = nnode.sysctl_func;
1167         node->sysctl_child = NULL;
1168 	/* node->sysctl_parent should already be done */
1169 
1170 	/*
1171 	 * update "version" on path to "root"
1172 	 */
1173 	for (; rnode->sysctl_parent != NULL; rnode = rnode->sysctl_parent)
1174 		;
1175 	pnode = node;
1176 	for (nm = rnode->sysctl_ver + 1; pnode != NULL;
1177 	     pnode = pnode->sysctl_parent)
1178 		pnode->sysctl_ver = nm;
1179 
1180 	/* If this fails, the node is already added - the user won't know! */
1181 	error = sysctl_cvt_out(l, v, node, oldp, *oldlenp, oldlenp);
1182 
1183 	return (error);
1184 }
1185 
1186 /*
1187  * ********************************************************************
1188  * A wrapper around sysctl_create() that prints the thing we're trying
1189  * to add.
1190  * ********************************************************************
1191  */
1192 #ifdef SYSCTL_DEBUG_CREATE
1193 int _sysctl_create(SYSCTLFN_PROTO);
1194 int
1195 _sysctl_create(SYSCTLFN_ARGS)
1196 {
1197 	const struct sysctlnode *node;
1198 	int k, rc, ni, nl = namelen + (name - oname);
1199 
1200 	node = newp;
1201 
1202 	printf("namelen %d (", nl);
1203 	for (ni = 0; ni < nl - 1; ni++)
1204 		printf(" %d", oname[ni]);
1205 	printf(" %d )\t[%s]\tflags %08x (%08x %d %zu)\n",
1206 	       k = node->sysctl_num,
1207 	       node->sysctl_name,
1208 	       node->sysctl_flags,
1209 	       SYSCTL_FLAGS(node->sysctl_flags),
1210 	       SYSCTL_TYPE(node->sysctl_flags),
1211 	       node->sysctl_size);
1212 
1213 	node = rnode;
1214 	rc = sysctl_create(SYSCTLFN_CALL(rnode));
1215 
1216 	printf("sysctl_create(");
1217 	for (ni = 0; ni < nl - 1; ni++)
1218 		printf(" %d", oname[ni]);
1219 	printf(" %d ) returned %d\n", k, rc);
1220 
1221 	return (rc);
1222 }
1223 #define sysctl_create _sysctl_create
1224 #endif /* SYSCTL_DEBUG_CREATE */
1225 
1226 /*
1227  * sysctl_destroy -- Removes a node (as described by newp) from the
1228  * given tree, returning (if successful) a copy of the dead node in
1229  * oldp.  Since we're removing stuff, there's not much to check.
1230  */
1231 int
1232 sysctl_destroy(SYSCTLFN_ARGS)
1233 {
1234 	struct sysctlnode *node, *pnode, onode, nnode;
1235 	int ni, error, v;
1236 
1237 	if (SYSCTL_VERS(rnode->sysctl_flags) != SYSCTL_VERSION) {
1238 		printf("sysctl_destroy: rnode %p wrong version\n", rnode);
1239 		return (EINVAL);
1240 	}
1241 
1242 	error = 0;
1243 
1244 	if (namelen != 1 || name[namelen - 1] != CTL_DESTROY)
1245 		return (EINVAL);
1246 
1247 	/*
1248 	 * processes can only destroy nodes at securelevel 0, must be
1249 	 * root, and can't remove nodes from a parent that's not
1250 	 * writeable
1251 	 */
1252 	if (l != NULL) {
1253 #ifndef SYSCTL_DISALLOW_CREATE
1254 		if (securelevel > 0)
1255 			return (EPERM);
1256 		error = kauth_authorize_generic(l->l_proc->p_cred,
1257 			    KAUTH_GENERIC_ISSUSER, &l->l_proc->p_acflag);
1258 		if (error)
1259 			return (error);
1260 		if (!(rnode->sysctl_flags & CTLFLAG_READWRITE))
1261 #endif /* SYSCTL_DISALLOW_CREATE */
1262 			return (EPERM);
1263 	}
1264 
1265 	/*
1266 	 * nothing can remove a node if:
1267 	 * the node is permanent (checked later) or
1268 	 * the tree itself is not writeable or
1269 	 * the entire sysctl system is not writeable
1270 	 *
1271 	 * note that we ignore whether setup is complete or not,
1272 	 * because these rules always apply.
1273 	 */
1274 	if (!(sysctl_rootof(rnode)->sysctl_flags & CTLFLAG_READWRITE) ||
1275 	    !(sysctl_root.sysctl_flags & CTLFLAG_READWRITE))
1276 		return (EPERM);
1277 
1278 	if (newp == NULL)
1279 		return (EINVAL);
1280 	error = sysctl_cvt_in(l, &v, newp, newlen, &nnode);
1281 	if (error)
1282 		return (error);
1283 	memset(&onode, 0, sizeof(struct sysctlnode));
1284 
1285 	node = rnode->sysctl_child;
1286 	for (ni = 0; ni < rnode->sysctl_clen; ni++) {
1287 		if (nnode.sysctl_num == node[ni].sysctl_num) {
1288 			/*
1289 			 * if name specified, must match
1290 			 */
1291 			if (nnode.sysctl_name[0] != '\0' &&
1292 			    strcmp(nnode.sysctl_name, node[ni].sysctl_name))
1293 				continue;
1294 			/*
1295 			 * if version specified, must match
1296 			 */
1297 			if (nnode.sysctl_ver != 0 &&
1298 			    nnode.sysctl_ver != node[ni].sysctl_ver)
1299 				continue;
1300 			/*
1301 			 * this must be the one
1302 			 */
1303 			break;
1304 		}
1305 	}
1306 	if (ni == rnode->sysctl_clen)
1307 		return (ENOENT);
1308 	node = &node[ni];
1309 	pnode = node->sysctl_parent;
1310 
1311 	/*
1312 	 * if the kernel says permanent, it is, so there.  nyah.
1313 	 */
1314 	if (SYSCTL_FLAGS(node->sysctl_flags) & CTLFLAG_PERMANENT)
1315 		return (EPERM);
1316 
1317 	/*
1318 	 * can't delete non-empty nodes
1319 	 */
1320 	if (SYSCTL_TYPE(node->sysctl_flags) == CTLTYPE_NODE &&
1321 	    node->sysctl_clen != 0)
1322 		return (ENOTEMPTY);
1323 
1324 	/*
1325 	 * if the node "owns" data, release it now
1326 	 */
1327 	if (node->sysctl_flags & CTLFLAG_OWNDATA) {
1328 		if (node->sysctl_data != NULL)
1329 			free(node->sysctl_data, M_SYSCTLDATA);
1330 		node->sysctl_data = NULL;
1331 	}
1332 	if (node->sysctl_flags & CTLFLAG_OWNDESC) {
1333 		if (node->sysctl_desc != NULL)
1334 			/*XXXUNCONST*/
1335 			free(__UNCONST(node->sysctl_desc), M_SYSCTLDATA);
1336 		node->sysctl_desc = NULL;
1337 	}
1338 
1339 	/*
1340 	 * if the node to be removed is not the last one on the list,
1341 	 * move the remaining nodes up, and reparent any grandchildren
1342 	 */
1343 	onode = *node;
1344 	if (ni < pnode->sysctl_clen - 1) {
1345 		int t;
1346 
1347 		memmove(&pnode->sysctl_child[ni], &pnode->sysctl_child[ni + 1],
1348 			(pnode->sysctl_clen - ni - 1) *
1349 			sizeof(struct sysctlnode));
1350 		for (; ni < pnode->sysctl_clen - 1; ni++)
1351 			if (SYSCTL_TYPE(pnode->sysctl_child[ni].sysctl_flags) ==
1352 			    CTLTYPE_NODE)
1353 				for (t = 0;
1354 				     t < pnode->sysctl_child[ni].sysctl_clen;
1355 				     t++)
1356 					pnode->sysctl_child[ni].sysctl_child[t].
1357 						sysctl_parent =
1358 						&pnode->sysctl_child[ni];
1359 		ni = pnode->sysctl_clen - 1;
1360 		node = &pnode->sysctl_child[ni];
1361 	}
1362 
1363 	/*
1364 	 * reset the space we just vacated
1365 	 */
1366 	memset(node, 0, sizeof(struct sysctlnode));
1367 	node->sysctl_parent = pnode;
1368 	pnode->sysctl_clen--;
1369 
1370 	/*
1371 	 * if this parent just lost its last child, nuke the creche
1372 	 */
1373 	if (pnode->sysctl_clen == 0) {
1374 		free(pnode->sysctl_child, M_SYSCTLNODE);
1375 		pnode->sysctl_csize = 0;
1376 		pnode->sysctl_child = NULL;
1377 	}
1378 
1379 	/*
1380 	 * update "version" on path to "root"
1381 	 */
1382         for (; rnode->sysctl_parent != NULL; rnode = rnode->sysctl_parent)
1383                 ;
1384 	for (ni = rnode->sysctl_ver + 1; pnode != NULL;
1385 	     pnode = pnode->sysctl_parent)
1386 		pnode->sysctl_ver = ni;
1387 
1388 	error = sysctl_cvt_out(l, v, &onode, oldp, *oldlenp, oldlenp);
1389 
1390 	return (error);
1391 }
1392 
1393 /*
1394  * sysctl_lookup -- Handles copyin/copyout of new and old values.
1395  * Partial reads are globally allowed.  Only root can write to things
1396  * unless the node says otherwise.
1397  */
1398 int
1399 sysctl_lookup(SYSCTLFN_ARGS)
1400 {
1401 	int error, rw;
1402 	size_t sz, len;
1403 	void *d;
1404 
1405 	if (SYSCTL_VERS(rnode->sysctl_flags) != SYSCTL_VERSION) {
1406 		printf("sysctl_lookup: rnode %p wrong version\n", rnode);
1407 		return (EINVAL);
1408 	}
1409 
1410 	error = 0;
1411 
1412 	/*
1413 	 * you can't "look up" a node.  you can "query" it, but you
1414 	 * can't "look it up".
1415 	 */
1416 	if (SYSCTL_TYPE(rnode->sysctl_flags) == CTLTYPE_NODE || namelen != 0)
1417 		return (EINVAL);
1418 
1419 	/*
1420 	 * some nodes are private, so only root can look into them.
1421 	 */
1422 	if (l != NULL && (rnode->sysctl_flags & CTLFLAG_PRIVATE) &&
1423 	    (error = kauth_authorize_generic(l->l_proc->p_cred,
1424 			KAUTH_GENERIC_ISSUSER, &l->l_proc->p_acflag)) != 0)
1425 		return (error);
1426 
1427 	/*
1428 	 * if a node wants to be writable according to different rules
1429 	 * other than "only root can write to stuff unless a flag is
1430 	 * set", then it needs its own function which should have been
1431 	 * called and not us.
1432 	 */
1433 	if (l != NULL && newp != NULL &&
1434 	    !(rnode->sysctl_flags & CTLFLAG_ANYWRITE) &&
1435 	    (error = kauth_authorize_generic(l->l_proc->p_cred,
1436 			KAUTH_GENERIC_ISSUSER, &l->l_proc->p_acflag)) != 0)
1437 		return (error);
1438 
1439 	/*
1440 	 * is this node supposedly writable?
1441 	 */
1442 	rw = 0;
1443 	switch (rnode->sysctl_flags & CTLFLAG_READWRITE) {
1444 	    case CTLFLAG_READONLY1:
1445 		rw = (securelevel < 1) ? 1 : 0;
1446 		break;
1447 	    case CTLFLAG_READONLY2:
1448 		rw = (securelevel < 2) ? 1 : 0;
1449 		break;
1450 	    case CTLFLAG_READWRITE:
1451 		rw = 1;
1452 		break;
1453 	}
1454 
1455 	/*
1456 	 * it appears not to be writable at this time, so if someone
1457 	 * tried to write to it, we must tell them to go away
1458 	 */
1459 	if (!rw && newp != NULL)
1460 		return (EPERM);
1461 
1462 	/*
1463 	 * step one, copy out the stuff we have presently
1464 	 */
1465 	if (rnode->sysctl_flags & CTLFLAG_IMMEDIATE) {
1466 		/*
1467 		 * note that we discard const here because we are
1468 		 * modifying the contents of the node (which is okay
1469 		 * because it's ours)
1470 		 */
1471 		switch (SYSCTL_TYPE(rnode->sysctl_flags)) {
1472 		case CTLTYPE_INT:
1473 			d = __UNCONST(&rnode->sysctl_idata);
1474 			break;
1475 		case CTLTYPE_QUAD:
1476 			d = __UNCONST(&rnode->sysctl_qdata);
1477 			break;
1478 		default:
1479 			return (EINVAL);
1480 		}
1481 	} else
1482 		d = rnode->sysctl_data;
1483 	if (SYSCTL_TYPE(rnode->sysctl_flags) == CTLTYPE_STRING)
1484 		sz = strlen(d) + 1; /* XXX@@@ possible fault here */
1485 	else
1486 		sz = rnode->sysctl_size;
1487 	if (oldp != NULL)
1488 		error = sysctl_copyout(l, d, oldp, MIN(sz, *oldlenp));
1489 	if (error)
1490 		return (error);
1491 	*oldlenp = sz;
1492 
1493 	/*
1494 	 * are we done?
1495 	 */
1496 	if (newp == NULL || newlen == 0)
1497 		return (0);
1498 
1499 	/*
1500 	 * hmm...not done.  must now "copy in" new value.  re-adjust
1501 	 * sz to maximum value (strings are "weird").
1502 	 */
1503 	sz = rnode->sysctl_size;
1504 	switch (SYSCTL_TYPE(rnode->sysctl_flags)) {
1505 	case CTLTYPE_INT:
1506 	case CTLTYPE_QUAD:
1507 	case CTLTYPE_STRUCT:
1508 		/*
1509 		 * these data must be *exactly* the same size coming
1510 		 * in.
1511 		 */
1512 		if (newlen != sz)
1513 			return (EINVAL);
1514 		error = sysctl_copyin(l, newp, d, sz);
1515 		break;
1516 	case CTLTYPE_STRING: {
1517 		/*
1518 		 * strings, on the other hand, can be shorter, and we
1519 		 * let userland be sloppy about the trailing nul.
1520 		 */
1521 		char *newbuf;
1522 
1523 		/*
1524 		 * too much new string?
1525 		 */
1526 		if (newlen > sz)
1527 			return (EINVAL);
1528 
1529 		/*
1530 		 * temporary copy of new inbound string
1531 		 */
1532 		len = MIN(sz, newlen);
1533 		newbuf = malloc(len, M_SYSCTLDATA, M_WAITOK|M_CANFAIL);
1534 		if (newbuf == NULL)
1535 			return (ENOMEM);
1536 		error = sysctl_copyin(l, newp, newbuf, len);
1537 		if (error) {
1538 			free(newbuf, M_SYSCTLDATA);
1539 			return (error);
1540 		}
1541 
1542 		/*
1543 		 * did they null terminate it, or do we have space
1544 		 * left to do it ourselves?
1545 		 */
1546 		if (newbuf[len - 1] != '\0' && len == sz) {
1547 			free(newbuf, M_SYSCTLDATA);
1548 			return (EINVAL);
1549 		}
1550 
1551 		/*
1552 		 * looks good, so pop it into place and zero the rest.
1553 		 */
1554 		if (len > 0)
1555 			memcpy(d, newbuf, len);
1556 		if (sz != len)
1557 			memset((char*)d + len, 0, sz - len);
1558 		free(newbuf, M_SYSCTLDATA);
1559 		break;
1560 	}
1561 	default:
1562 		return (EINVAL);
1563 	}
1564 
1565 	return (error);
1566 }
1567 
1568 /*
1569  * sysctl_mmap -- Dispatches sysctl mmap requests to those nodes that
1570  * purport to handle it.  This interface isn't fully fleshed out yet,
1571  * unfortunately.
1572  */
1573 static int
1574 sysctl_mmap(SYSCTLFN_ARGS)
1575 {
1576 	const struct sysctlnode *node;
1577 	struct sysctlnode nnode;
1578 	int error;
1579 
1580 	if (SYSCTL_VERS(rnode->sysctl_flags) != SYSCTL_VERSION) {
1581 		printf("sysctl_mmap: rnode %p wrong version\n", rnode);
1582 		return (EINVAL);
1583 	}
1584 
1585 	/*
1586 	 * let's just pretend that didn't happen, m'kay?
1587 	 */
1588 	if (l == NULL)
1589 		return (EPERM);
1590 
1591 	/*
1592 	 * is this a sysctlnode description of an mmap request?
1593 	 */
1594 	if (newp == NULL || newlen != sizeof(struct sysctlnode))
1595 		return (EINVAL);
1596 	error = sysctl_copyin(l, newp, &nnode, sizeof(nnode));
1597 	if (error)
1598 		return (error);
1599 
1600 	/*
1601 	 * does the node they asked for exist?
1602 	 */
1603 	if (namelen != 1)
1604 		return (EOPNOTSUPP);
1605 	node = rnode;
1606         error = sysctl_locate(l, &nnode.sysctl_num, 1, &node, NULL);
1607 	if (error)
1608 		return (error);
1609 
1610 	/*
1611 	 * does this node that we have found purport to handle mmap?
1612 	 */
1613 	if (node->sysctl_func == NULL ||
1614 	    !(node->sysctl_flags & CTLFLAG_MMAP))
1615 		return (EOPNOTSUPP);
1616 
1617 	/*
1618 	 * well...okay, they asked for it.
1619 	 */
1620 	return ((*node->sysctl_func)(SYSCTLFN_CALL(node)));
1621 }
1622 
1623 int
1624 sysctl_describe(SYSCTLFN_ARGS)
1625 {
1626 	struct sysctldesc *d;
1627 	void *bf;
1628 	size_t sz, left, tot;
1629 	int i, error, v = -1;
1630 	struct sysctlnode *node;
1631 	struct sysctlnode dnode;
1632 
1633 	if (SYSCTL_VERS(rnode->sysctl_flags) != SYSCTL_VERSION) {
1634 		printf("sysctl_query: rnode %p wrong version\n", rnode);
1635 		return (EINVAL);
1636 	}
1637 
1638 	if (SYSCTL_TYPE(rnode->sysctl_flags) != CTLTYPE_NODE)
1639 		return (ENOTDIR);
1640 	if (namelen != 1 || name[0] != CTL_DESCRIBE)
1641 		return (EINVAL);
1642 
1643 	/*
1644 	 * get ready...
1645 	 */
1646 	error = 0;
1647 	d = bf = malloc(MAXDESCLEN, M_TEMP, M_WAITOK|M_CANFAIL);
1648 	if (bf == NULL)
1649 		return ENOMEM;
1650 	tot = 0;
1651 	node = rnode->sysctl_child;
1652 	left = *oldlenp;
1653 
1654 	/*
1655 	 * no request -> all descriptions at this level
1656 	 * request with desc unset -> just this node
1657 	 * request with desc set -> set descr for this node
1658 	 */
1659 	if (newp != NULL) {
1660 		error = sysctl_cvt_in(l, &v, newp, newlen, &dnode);
1661 		if (error)
1662 			goto out;
1663 		if (dnode.sysctl_desc != NULL) {
1664 			/*
1665 			 * processes cannot set descriptions above
1666 			 * securelevel 0.  and must be root.  blah
1667 			 * blah blah.  a couple more checks are made
1668 			 * once we find the node we want.
1669 			 */
1670 			if (l != NULL) {
1671 #ifndef SYSCTL_DISALLOW_CREATE
1672 				if (securelevel > 0) {
1673 					error = EPERM;
1674 					goto out;
1675 				}
1676 				error = kauth_authorize_generic(l->l_proc->p_cred,
1677 					    KAUTH_GENERIC_ISSUSER,
1678 					      &l->l_proc->p_acflag);
1679 				if (error)
1680 					goto out;
1681 #else /* SYSCTL_DISALLOW_CREATE */
1682 				error = EPERM;
1683 				goto out;
1684 #endif /* SYSCTL_DISALLOW_CREATE */
1685 			}
1686 
1687 			/*
1688 			 * find node and try to set the description on it
1689 			 */
1690 			for (i = 0; i < rnode->sysctl_clen; i++)
1691 				if (node[i].sysctl_num == dnode.sysctl_num)
1692 					break;
1693 			if (i == rnode->sysctl_clen) {
1694 				error = ENOENT;
1695 				goto out;
1696 			}
1697 			node = &node[i];
1698 
1699 			/*
1700 			 * did the caller specify a node version?
1701 			 */
1702 			if (dnode.sysctl_ver != 0 &&
1703 			    dnode.sysctl_ver != node->sysctl_ver) {
1704 				error = EINVAL;
1705 				goto out;
1706 			}
1707 
1708 			/*
1709 			 * okay...some rules:
1710 			 * (1) if setup is done and the tree is
1711 			 *     read-only or the whole system is
1712 			 *     read-only
1713 			 * (2) no one can set a description on a
1714 			 *     permanent node (it must be set when
1715 			 *     using createv)
1716 			 * (3) processes cannot *change* a description
1717 			 * (4) processes *can*, however, set a
1718 			 *     description on a read-only node so that
1719 			 *     one can be created and then described
1720 			 *     in two steps
1721 			 * anything else come to mind?
1722 			 */
1723 			if ((sysctl_root.sysctl_flags & CTLFLAG_PERMANENT) &&
1724 			    (!(sysctl_rootof(node)->sysctl_flags &
1725 			       CTLFLAG_READWRITE) ||
1726 			     !(sysctl_root.sysctl_flags & CTLFLAG_READWRITE))) {
1727 				error = EPERM;
1728 				goto out;
1729 			}
1730 			if (node->sysctl_flags & CTLFLAG_PERMANENT) {
1731 				error = EPERM;
1732 				goto out;
1733 			}
1734 			if (l != NULL && node->sysctl_desc != NULL) {
1735 				error = EPERM;
1736 				goto out;
1737 			}
1738 
1739 			/*
1740 			 * right, let's go ahead.  the first step is
1741 			 * making the description into something the
1742 			 * node can "own", if need be.
1743 			 */
1744 			if (l != NULL ||
1745 			    dnode.sysctl_flags & CTLFLAG_OWNDESC) {
1746 				char *nd, *k;
1747 
1748 				k = malloc(MAXDESCLEN, M_TEMP,
1749 				    M_WAITOK|M_CANFAIL);
1750 				if (k == NULL) {
1751 					error = ENOMEM;
1752 					goto out;
1753 				}
1754 				error = sysctl_copyinstr(l, dnode.sysctl_desc,
1755 							 k, MAXDESCLEN, &sz);
1756 				if (error) {
1757 					free(k, M_TEMP);
1758 					goto out;
1759 				}
1760 				nd = malloc(sz, M_SYSCTLDATA,
1761 					    M_WAITOK|M_CANFAIL);
1762 				if (nd == NULL) {
1763 					free(k, M_TEMP);
1764 					error = ENOMEM;
1765 					goto out;
1766 				}
1767 				memcpy(nd, k, sz);
1768 				dnode.sysctl_flags |= CTLFLAG_OWNDESC;
1769 				dnode.sysctl_desc = nd;
1770 				free(k, M_TEMP);
1771 			}
1772 
1773 			/*
1774 			 * now "release" the old description and
1775 			 * attach the new one.  ta-da.
1776 			 */
1777 			if ((node->sysctl_flags & CTLFLAG_OWNDESC) &&
1778 			    node->sysctl_desc != NULL)
1779 				/*XXXUNCONST*/
1780 				free(__UNCONST(node->sysctl_desc), M_SYSCTLDATA);
1781 			node->sysctl_desc = dnode.sysctl_desc;
1782 			node->sysctl_flags |=
1783 				(dnode.sysctl_flags & CTLFLAG_OWNDESC);
1784 
1785 			/*
1786 			 * now we "fall out" and into the loop which
1787 			 * will copy the new description back out for
1788 			 * those interested parties
1789 			 */
1790 		}
1791 	}
1792 
1793 	/*
1794 	 * scan for one description or just retrieve all descriptions
1795 	 */
1796 	for (i = 0; i < rnode->sysctl_clen; i++) {
1797 		/*
1798 		 * did they ask for the description of only one node?
1799 		 */
1800 		if (v != -1 && node[i].sysctl_num != dnode.sysctl_num)
1801 			continue;
1802 
1803 		/*
1804 		 * don't describe "private" nodes to non-suser users
1805 		 */
1806 		if ((node[i].sysctl_flags & CTLFLAG_PRIVATE) && (l != NULL) &&
1807 		    !(kauth_authorize_generic(l->l_proc->p_cred,
1808 			KAUTH_GENERIC_ISSUSER, &l->l_proc->p_acflag)))
1809 			continue;
1810 
1811 		/*
1812 		 * is this description "valid"?
1813 		 */
1814 		memset(bf, 0, MAXDESCLEN);
1815 		if (node[i].sysctl_desc == NULL)
1816 			sz = 1;
1817 		else if (copystr(node[i].sysctl_desc, &d->descr_str[0],
1818 				 MAXDESCLEN - sizeof(*d), &sz) != 0) {
1819 			/*
1820 			 * erase possible partial description
1821 			 */
1822 			memset(bf, 0, MAXDESCLEN);
1823 			sz = 1;
1824 		}
1825 
1826 		/*
1827 		 * we've got it, stuff it into the caller's buffer
1828 		 */
1829 		d->descr_num = node[i].sysctl_num;
1830 		d->descr_ver = node[i].sysctl_ver;
1831 		d->descr_len = sz; /* includes trailing nul */
1832 		sz = (caddr_t)NEXT_DESCR(d) - (caddr_t)d;
1833 		if (oldp != NULL && left >= sz) {
1834 			error = sysctl_copyout(l, d, oldp, sz);
1835 			if (error)
1836 				goto out;
1837 			left -= sz;
1838 			oldp = (void *)__sysc_desc_adv(oldp, d->descr_len);
1839 		}
1840 		tot += sz;
1841 
1842 		/*
1843 		 * if we get this far with v not "unset", they asked
1844 		 * for a specific node and we found it
1845 		 */
1846 		if (v != -1)
1847 			break;
1848 	}
1849 
1850 	/*
1851 	 * did we find it after all?
1852 	 */
1853 	if (v != -1 && tot == 0)
1854 		error = ENOENT;
1855 	else
1856 		*oldlenp = tot;
1857 
1858 out:
1859 	free(bf, M_TEMP);
1860 	return (error);
1861 }
1862 
1863 /*
1864  * ********************************************************************
1865  * Section 3: Create and destroy from inside the kernel
1866  * ********************************************************************
1867  * sysctl_createv() and sysctl_destroyv() are simpler-to-use
1868  * interfaces for the kernel to fling new entries into the mib and rip
1869  * them out later.  In the case of sysctl_createv(), the returned copy
1870  * of the node (see sysctl_create()) will be translated back into a
1871  * pointer to the actual node.
1872  *
1873  * Note that sysctl_createv() will return 0 if the create request
1874  * matches an existing node (ala mkdir -p), and that sysctl_destroyv()
1875  * will return 0 if the node to be destroyed already does not exist
1876  * (aka rm -f) or if it is a parent of other nodes.
1877  *
1878  * This allows two (or more) different subsystems to assert sub-tree
1879  * existence before populating their own nodes, and to remove their
1880  * own nodes without orphaning the others when they are done.
1881  * ********************************************************************
1882  */
1883 int
1884 sysctl_createv(struct sysctllog **log, int cflags,
1885 	       const struct sysctlnode **rnode, const struct sysctlnode **cnode,
1886 	       int flags, int type, const char *namep, const char *descr,
1887 	       sysctlfn func, u_quad_t qv, void *newp, size_t newlen,
1888 	       ...)
1889 {
1890 	va_list ap;
1891 	int error, ni, namelen, name[CTL_MAXNAME];
1892 	const struct sysctlnode *root, *pnode;
1893 	struct sysctlnode nnode, onode, *dnode;
1894 	size_t sz;
1895 
1896 	/*
1897 	 * where are we putting this?
1898 	 */
1899 	if (rnode != NULL && *rnode == NULL) {
1900 		printf("sysctl_createv: rnode NULL\n");
1901 		return (EINVAL);
1902 	}
1903 	root = rnode ? *rnode : NULL;
1904 	if (cnode != NULL)
1905 		*cnode = NULL;
1906 	if (cflags != 0)
1907 		return (EINVAL);
1908 
1909 	/*
1910 	 * what is it?
1911 	 */
1912 	flags = SYSCTL_VERSION|SYSCTL_TYPE(type)|SYSCTL_FLAGS(flags);
1913 	if (log != NULL)
1914 		flags &= ~CTLFLAG_PERMANENT;
1915 
1916 	/*
1917 	 * where do we put it?
1918 	 */
1919 	va_start(ap, newlen);
1920 	namelen = 0;
1921 	ni = -1;
1922 	do {
1923 		if (++ni == CTL_MAXNAME)
1924 			return (ENAMETOOLONG);
1925 		name[ni] = va_arg(ap, int);
1926 		/*
1927 		 * sorry, this is not supported from here
1928 		 */
1929 		if (name[ni] == CTL_CREATESYM)
1930 			return (EINVAL);
1931 	} while (name[ni] != CTL_EOL && name[ni] != CTL_CREATE);
1932 	namelen = ni + (name[ni] == CTL_CREATE ? 1 : 0);
1933 	va_end(ap);
1934 
1935 	/*
1936 	 * what's it called
1937 	 */
1938 	if (strlcpy(nnode.sysctl_name, namep, sizeof(nnode.sysctl_name)) >=
1939 	    sizeof(nnode.sysctl_name))
1940 		return (ENAMETOOLONG);
1941 
1942 	/*
1943 	 * cons up the description of the new node
1944 	 */
1945 	nnode.sysctl_num = name[namelen - 1];
1946 	name[namelen - 1] = CTL_CREATE;
1947 	nnode.sysctl_size = newlen;
1948 	nnode.sysctl_flags = flags;
1949 	if (type == CTLTYPE_NODE) {
1950 		nnode.sysctl_csize = 0;
1951 		nnode.sysctl_clen = 0;
1952 		nnode.sysctl_child = NULL;
1953 		if (flags & CTLFLAG_ALIAS)
1954 			nnode.sysctl_alias = qv;
1955 	} else if (flags & CTLFLAG_IMMEDIATE) {
1956 		switch (type) {
1957 		case CTLTYPE_INT:
1958 			nnode.sysctl_idata = qv;
1959 			break;
1960 		case CTLTYPE_QUAD:
1961 			nnode.sysctl_qdata = qv;
1962 			break;
1963 		default:
1964 			return (EINVAL);
1965 		}
1966 	} else {
1967 		nnode.sysctl_data = newp;
1968 	}
1969 	nnode.sysctl_func = func;
1970 	nnode.sysctl_parent = NULL;
1971 	nnode.sysctl_ver = 0;
1972 
1973 	/*
1974 	 * initialize lock state -- we need locks if the main tree has
1975 	 * been marked as complete, but since we could be called from
1976 	 * either there, or from a device driver (say, at device
1977 	 * insertion), or from an lkm (at lkm load time, say), we
1978 	 * don't really want to "wait"...
1979 	 */
1980 	error = sysctl_lock(NULL, NULL, 0);
1981 	if (error)
1982 		return (error);
1983 
1984 	/*
1985 	 * locate the prospective parent of the new node, and if we
1986 	 * find it, add the new node.
1987 	 */
1988 	sz = sizeof(onode);
1989 	pnode = root;
1990 	error = sysctl_locate(NULL, &name[0], namelen - 1, &pnode, &ni);
1991 	if (error) {
1992 		printf("sysctl_createv: sysctl_locate(%s) returned %d\n",
1993 		       nnode.sysctl_name, error);
1994 		sysctl_unlock(NULL);
1995 		return (error);
1996 	}
1997 	error = sysctl_create(&name[ni], namelen - ni, &onode, &sz,
1998 			      &nnode, sizeof(nnode), &name[0], NULL,
1999 			      pnode);
2000 
2001 	/*
2002 	 * unfortunately the node we wanted to create is already
2003 	 * there.  if the node that's already there is a reasonable
2004 	 * facsimile of the node we wanted to create, just pretend
2005 	 * (for the caller's benefit) that we managed to create the
2006 	 * node they wanted.
2007 	 */
2008 	if (error == EEXIST) {
2009 		/* name is the same as requested... */
2010 		if (strcmp(nnode.sysctl_name, onode.sysctl_name) == 0 &&
2011 		    /* they want the same function... */
2012 		    nnode.sysctl_func == onode.sysctl_func &&
2013 		    /* number is the same as requested, or... */
2014 		    (nnode.sysctl_num == onode.sysctl_num ||
2015 		     /* they didn't pick a number... */
2016 		     nnode.sysctl_num == CTL_CREATE)) {
2017 			/*
2018 			 * collision here from trying to create
2019 			 * something that already existed; let's give
2020 			 * our customers a hand and tell them they got
2021 			 * what they wanted.
2022 			 */
2023 #ifdef SYSCTL_DEBUG_CREATE
2024 			printf("cleared\n");
2025 #endif /* SYSCTL_DEBUG_CREATE */
2026 			error = 0;
2027 		}
2028 	}
2029 
2030 	if (error == 0 &&
2031 	    (cnode != NULL || log != NULL || descr != NULL)) {
2032 		/*
2033 		 * sysctl_create() gave us back a copy of the node,
2034 		 * but we need to know where it actually is...
2035 		 */
2036 		pnode = root;
2037 		error = sysctl_locate(NULL, &name[0], namelen - 1, &pnode, &ni);
2038 
2039 		/*
2040 		 * manual scan of last layer so that aliased nodes
2041 		 * aren't followed.
2042 		 */
2043 		if (error == 0) {
2044 			for (ni = 0; ni < pnode->sysctl_clen; ni++)
2045 				if (pnode->sysctl_child[ni].sysctl_num ==
2046 				    onode.sysctl_num)
2047 					break;
2048 			if (ni < pnode->sysctl_clen)
2049 				pnode = &pnode->sysctl_child[ni];
2050 			else
2051 				error = ENOENT;
2052 		}
2053 
2054 		/*
2055 		 * not expecting an error here, but...
2056 		 */
2057 		if (error == 0) {
2058 			if (log != NULL)
2059 				sysctl_log_add(log, pnode);
2060 			if (cnode != NULL)
2061 				*cnode = pnode;
2062 			if (descr != NULL) {
2063 				/*
2064 				 * allow first caller to *set* a
2065 				 * description actually to set it
2066 				 *
2067 				 * discard const here so we can attach
2068 				 * the description
2069 				 */
2070 				dnode = __UNCONST(pnode);
2071 				if (pnode->sysctl_desc != NULL)
2072 					/* skip it...we've got one */;
2073 				else if (flags & CTLFLAG_OWNDESC) {
2074 					size_t l = strlen(descr) + 1;
2075 					char *d = malloc(l, M_SYSCTLDATA,
2076 							 M_WAITOK|M_CANFAIL);
2077 					if (d != NULL) {
2078 						memcpy(d, descr, l);
2079 						dnode->sysctl_desc = d;
2080 						dnode->sysctl_flags |=
2081 						    CTLFLAG_OWNDESC;
2082 					}
2083 				} else
2084 					dnode->sysctl_desc = descr;
2085 			}
2086 		} else {
2087 			printf("sysctl_create succeeded but node not found?!\n");
2088 			/*
2089 			 *  confusing, but the create said it
2090 			 * succeeded, so...
2091 			 */
2092 			error = 0;
2093 		}
2094 	}
2095 
2096 	/*
2097 	 * now it should be safe to release the lock state.  note that
2098 	 * the pointer to the newly created node being passed back may
2099 	 * not be "good" for very long.
2100 	 */
2101 	sysctl_unlock(NULL);
2102 
2103 	if (error != 0) {
2104 		printf("sysctl_createv: sysctl_create(%s) returned %d\n",
2105 		       nnode.sysctl_name, error);
2106 #if 0
2107 		if (error != ENOENT)
2108 			sysctl_dump(&onode);
2109 #endif
2110 	}
2111 
2112 	return (error);
2113 }
2114 
2115 int
2116 sysctl_destroyv(struct sysctlnode *rnode, ...)
2117 {
2118 	va_list ap;
2119 	int error, name[CTL_MAXNAME], namelen, ni;
2120 	const struct sysctlnode *pnode, *node;
2121 	struct sysctlnode dnode, *onode;
2122 	size_t sz;
2123 
2124 	va_start(ap, rnode);
2125 	namelen = 0;
2126 	ni = 0;
2127 	do {
2128 		if (ni == CTL_MAXNAME)
2129 			return (ENAMETOOLONG);
2130 		name[ni] = va_arg(ap, int);
2131 	} while (name[ni++] != CTL_EOL);
2132 	namelen = ni - 1;
2133 	va_end(ap);
2134 
2135 	/*
2136 	 * i can't imagine why we'd be destroying a node when the tree
2137 	 * wasn't complete, but who knows?
2138 	 */
2139 	error = sysctl_lock(NULL, NULL, 0);
2140 	if (error)
2141 		return (error);
2142 
2143 	/*
2144 	 * where is it?
2145 	 */
2146 	node = rnode;
2147 	error = sysctl_locate(NULL, &name[0], namelen - 1, &node, &ni);
2148 	if (error) {
2149 		/* they want it gone and it's not there, so... */
2150 		sysctl_unlock(NULL);
2151 		return (error == ENOENT ? 0 : error);
2152 	}
2153 
2154 	/*
2155 	 * set up the deletion
2156 	 */
2157 	pnode = node;
2158 	node = &dnode;
2159 	memset(&dnode, 0, sizeof(dnode));
2160 	dnode.sysctl_flags = SYSCTL_VERSION;
2161 	dnode.sysctl_num = name[namelen - 1];
2162 
2163 	/*
2164 	 * we found it, now let's nuke it
2165 	 */
2166 	name[namelen - 1] = CTL_DESTROY;
2167 	sz = 0;
2168 	error = sysctl_destroy(&name[namelen - 1], 1, NULL, &sz,
2169 			       node, sizeof(*node), &name[0], NULL,
2170 			       pnode);
2171 	if (error == ENOTEMPTY) {
2172 		/*
2173 		 * think of trying to delete "foo" when "foo.bar"
2174 		 * (which someone else put there) is still in
2175 		 * existence
2176 		 */
2177 		error = 0;
2178 
2179 		/*
2180 		 * dunno who put the description there, but if this
2181 		 * node can ever be removed, we need to make sure the
2182 		 * string doesn't go out of context.  that means we
2183 		 * need to find the node that's still there (don't use
2184 		 * sysctl_locate() because that follows aliasing).
2185 		 */
2186 		node = pnode->sysctl_child;
2187 		for (ni = 0; ni < pnode->sysctl_clen; ni++)
2188 			if (node[ni].sysctl_num == dnode.sysctl_num)
2189 				break;
2190 		node = (ni < pnode->sysctl_clen) ? &node[ni] : NULL;
2191 
2192 		/*
2193 		 * if we found it, and this node has a description,
2194 		 * and this node can be released, and it doesn't
2195 		 * already own its own description...sigh.  :)
2196 		 */
2197 		if (node != NULL && node->sysctl_desc != NULL &&
2198 		    !(node->sysctl_flags & CTLFLAG_PERMANENT) &&
2199 		    !(node->sysctl_flags & CTLFLAG_OWNDESC)) {
2200 			char *d;
2201 
2202 			sz = strlen(node->sysctl_desc) + 1;
2203 			d = malloc(sz, M_SYSCTLDATA, M_WAITOK|M_CANFAIL);
2204 			if (d != NULL) {
2205 				/*
2206 				 * discard const so that we can
2207 				 * re-attach the description
2208 				 */
2209 				memcpy(d, node->sysctl_desc, sz);
2210 				onode = __UNCONST(node);
2211 				onode->sysctl_desc = d;
2212 				onode->sysctl_flags |= CTLFLAG_OWNDESC;
2213 			} else {
2214 				/*
2215 				 * XXX drop the description?  be
2216 				 * afraid?  don't care?
2217 				 */
2218 			}
2219 		}
2220 	}
2221 
2222         sysctl_unlock(NULL);
2223 
2224 	return (error);
2225 }
2226 
2227 #if 0
2228 /*
2229  * ********************************************************************
2230  * the dump routine.  i haven't yet decided how (if at all) i'll call
2231  * this from userland when it's in the kernel.
2232  * ********************************************************************
2233  */
2234 static const char *
2235 sf(int f)
2236 {
2237 	static char s[256];
2238 	char *c;
2239 
2240 	s[0] = '\0';
2241 	c = "";
2242 
2243 #define print_flag(_f, _s, _c, _q, _x) \
2244 	if (((_f) & (__CONCAT(CTLFLAG_,_x))) == (__CONCAT(CTLFLAG_,_q))) { \
2245 		strlcat((_s), (_c), sizeof(_s)); \
2246 		strlcat((_s), __STRING(_q), sizeof(_s)); \
2247 		(_c) = ","; \
2248 		(_f) &= ~__CONCAT(CTLFLAG_,_x); \
2249 	}
2250 
2251 	print_flag(f, s, c, READONLY,  READWRITE);
2252 	print_flag(f, s, c, READONLY1, READWRITE);
2253 	print_flag(f, s, c, READONLY2, READWRITE);
2254 	print_flag(f, s, c, READWRITE, READWRITE);
2255 	print_flag(f, s, c, ANYWRITE,  ANYWRITE);
2256 	print_flag(f, s, c, PRIVATE,   PRIVATE);
2257 	print_flag(f, s, c, PERMANENT, PERMANENT);
2258 	print_flag(f, s, c, OWNDATA,   OWNDATA);
2259 	print_flag(f, s, c, IMMEDIATE, IMMEDIATE);
2260 	print_flag(f, s, c, HEX,       HEX);
2261 	print_flag(f, s, c, ROOT,      ROOT);
2262 	print_flag(f, s, c, ANYNUMBER, ANYNUMBER);
2263 	print_flag(f, s, c, HIDDEN,    HIDDEN);
2264 	print_flag(f, s, c, ALIAS,     ALIAS);
2265 #undef print_flag
2266 
2267 	if (f) {
2268 		char foo[9];
2269 		snprintf(foo, sizeof(foo), "%x", f);
2270 		strlcat(s, c, sizeof(s));
2271 		strlcat(s, foo, sizeof(s));
2272 	}
2273 
2274 	return (s);
2275 }
2276 
2277 static const char *
2278 st(int t)
2279 {
2280 
2281 	switch (t) {
2282 	case CTLTYPE_NODE:
2283 		return "NODE";
2284 	case CTLTYPE_INT:
2285 		return "INT";
2286 	case CTLTYPE_STRING:
2287 		return "STRING";
2288 	case CTLTYPE_QUAD:
2289 		return "QUAD";
2290 	case CTLTYPE_STRUCT:
2291 		return "STRUCT";
2292 	}
2293 
2294 	return "???";
2295 }
2296 
2297 void
2298 sysctl_dump(const struct sysctlnode *d)
2299 {
2300 	static char nmib[64], smib[256];
2301 	static int indent;
2302 	struct sysctlnode *n;
2303 	char *np, *sp, tmp[20];
2304 	int i;
2305 
2306 	if (d == NULL)
2307 		return;
2308 
2309 	np = &nmib[strlen(nmib)];
2310 	sp = &smib[strlen(smib)];
2311 
2312 	if (!(d->sysctl_flags & CTLFLAG_ROOT)) {
2313 		snprintf(tmp, sizeof(tmp), "%d", d->sysctl_num);
2314 		strcat(nmib, ".");
2315 		strcat(smib, ".");
2316 		strcat(nmib, tmp);
2317 		strcat(smib, d->sysctl_name);
2318 		printf("%s -> %s (%d)\n", &nmib[1], &smib[1],
2319 		       SYSCTL_TYPE(d->sysctl_flags));
2320 	}
2321 
2322 	if (1) {
2323 		printf("%*s%p:\tsysctl_name  [%s]\n", indent, "",
2324 		       d, d->sysctl_name);
2325 		printf("%*s\t\tsysctl_num    %d\n",   indent, "",
2326 		       d->sysctl_num);
2327 		printf("%*s\t\tsysctl_flags  %x (flags=%x<%s> type=%d<%s> "
2328 		       "size=%zu)\n",
2329 		       indent, "", d->sysctl_flags,
2330 		       SYSCTL_FLAGS(d->sysctl_flags),
2331 		       sf(SYSCTL_FLAGS(d->sysctl_flags)),
2332 		       SYSCTL_TYPE(d->sysctl_flags),
2333 		       st(SYSCTL_TYPE(d->sysctl_flags)),
2334 		       d->sysctl_size);
2335 		if (SYSCTL_TYPE(d->sysctl_flags) == CTLTYPE_NODE) {
2336 			printf("%*s\t\tsysctl_csize  %d\n",   indent, "",
2337 			       d->sysctl_csize);
2338 			printf("%*s\t\tsysctl_clen   %d\n",   indent, "",
2339 			       d->sysctl_clen);
2340 			printf("%*s\t\tsysctl_child  %p\n",   indent, "",
2341 			       d->sysctl_child);
2342 		} else
2343 			printf("%*s\t\tsysctl_data   %p\n",   indent, "",
2344 			       d->sysctl_data);
2345 		printf("%*s\t\tsysctl_func   %p\n",   indent, "",
2346 		       d->sysctl_func);
2347 		printf("%*s\t\tsysctl_parent %p\n",   indent, "",
2348 		       d->sysctl_parent);
2349 		printf("%*s\t\tsysctl_ver    %d\n",   indent, "",
2350 		       d->sysctl_ver);
2351 	}
2352 
2353 	if (SYSCTL_TYPE(d->sysctl_flags) == CTLTYPE_NODE) {
2354 		indent += 8;
2355 		n = d->sysctl_child;
2356 		for (i = 0; i < d->sysctl_clen; i++) {
2357 			sysctl_dump(&n[i]);
2358 		}
2359 		indent -= 8;
2360 	}
2361 
2362 	np[0] = '\0';
2363 	sp[0] = '\0';
2364 }
2365 #endif /* 0 */
2366 
2367 /*
2368  * ********************************************************************
2369  * Deletes an entire n-ary tree.  Not recommended unless you know why
2370  * you're doing it.  Personally, I don't know why you'd even think
2371  * about it.
2372  * ********************************************************************
2373  */
2374 void
2375 sysctl_free(struct sysctlnode *rnode)
2376 {
2377 	struct sysctlnode *node, *pnode;
2378 
2379 	if (rnode == NULL)
2380 		rnode = &sysctl_root;
2381 
2382 	if (SYSCTL_VERS(rnode->sysctl_flags) != SYSCTL_VERSION) {
2383 		printf("sysctl_free: rnode %p wrong version\n", rnode);
2384 		return;
2385 	}
2386 
2387 	pnode = rnode;
2388 
2389 	node = pnode->sysctl_child;
2390 	do {
2391 		while (node != NULL && pnode->sysctl_csize > 0) {
2392 			while (node <
2393 			       &pnode->sysctl_child[pnode->sysctl_clen] &&
2394 			       (SYSCTL_TYPE(node->sysctl_flags) !=
2395 				CTLTYPE_NODE ||
2396 				node->sysctl_csize == 0)) {
2397 				if (SYSCTL_FLAGS(node->sysctl_flags) &
2398 				    CTLFLAG_OWNDATA) {
2399 					if (node->sysctl_data != NULL) {
2400 						free(node->sysctl_data,
2401 						     M_SYSCTLDATA);
2402 						node->sysctl_data = NULL;
2403 					}
2404 				}
2405 				if (SYSCTL_FLAGS(node->sysctl_flags) &
2406 				    CTLFLAG_OWNDESC) {
2407 					if (node->sysctl_desc != NULL) {
2408 						/*XXXUNCONST*/
2409 						free(__UNCONST(node->sysctl_desc),
2410 						     M_SYSCTLDATA);
2411 						node->sysctl_desc = NULL;
2412 					}
2413 				}
2414 				node++;
2415 			}
2416 			if (node < &pnode->sysctl_child[pnode->sysctl_clen]) {
2417 				pnode = node;
2418 				node = node->sysctl_child;
2419 			} else
2420 				break;
2421 		}
2422 		if (pnode->sysctl_child != NULL)
2423 			free(pnode->sysctl_child, M_SYSCTLNODE);
2424 		pnode->sysctl_clen = 0;
2425 		pnode->sysctl_csize = 0;
2426 		pnode->sysctl_child = NULL;
2427 		node = pnode;
2428 		pnode = node->sysctl_parent;
2429 	} while (pnode != NULL && node != rnode);
2430 }
2431 
2432 int
2433 sysctl_log_add(struct sysctllog **logp, const struct sysctlnode *node)
2434 {
2435 	int name[CTL_MAXNAME], namelen, i;
2436 	const struct sysctlnode *pnode;
2437 	struct sysctllog *log;
2438 
2439 	if (node->sysctl_flags & CTLFLAG_PERMANENT)
2440 		return (0);
2441 
2442 	if (logp == NULL)
2443 		return (0);
2444 
2445 	if (*logp == NULL) {
2446 		log = malloc(sizeof(struct sysctllog),
2447 		       M_SYSCTLDATA, M_WAITOK|M_CANFAIL);
2448 		if (log == NULL) {
2449 			/* XXX print error message? */
2450 			return (-1);
2451 		}
2452 		log->log_num = malloc(16 * sizeof(int),
2453 		       M_SYSCTLDATA, M_WAITOK|M_CANFAIL);
2454 		if (log->log_num == NULL) {
2455 			/* XXX print error message? */
2456 			free(log, M_SYSCTLDATA);
2457 			return (-1);
2458 		}
2459 		memset(log->log_num, 0, 16 * sizeof(int));
2460 		log->log_root = NULL;
2461 		log->log_size = 16;
2462 		log->log_left = 16;
2463 		*logp = log;
2464 	} else
2465 		log = *logp;
2466 
2467 	/*
2468 	 * check that the root is proper.  it's okay to record the
2469 	 * address of the root of a tree.  it's the only thing that's
2470 	 * guaranteed not to shift around as nodes come and go.
2471 	 */
2472 	if (log->log_root == NULL)
2473 		log->log_root = sysctl_rootof(node);
2474 	else if (log->log_root != sysctl_rootof(node)) {
2475 		printf("sysctl: log %p root mismatch (%p)\n",
2476 		       log->log_root, sysctl_rootof(node));
2477 		return (-1);
2478 	}
2479 
2480 	/*
2481 	 * we will copy out name in reverse order
2482 	 */
2483 	for (pnode = node, namelen = 0;
2484 	     pnode != NULL && !(pnode->sysctl_flags & CTLFLAG_ROOT);
2485 	     pnode = pnode->sysctl_parent)
2486 		name[namelen++] = pnode->sysctl_num;
2487 
2488 	/*
2489 	 * do we have space?
2490 	 */
2491 	if (log->log_left < (namelen + 3))
2492 		sysctl_log_realloc(log);
2493 	if (log->log_left < (namelen + 3))
2494 		return (-1);
2495 
2496 	/*
2497 	 * stuff name in, then namelen, then node type, and finally,
2498 	 * the version for non-node nodes.
2499 	 */
2500 	for (i = 0; i < namelen; i++)
2501 		log->log_num[--log->log_left] = name[i];
2502 	log->log_num[--log->log_left] = namelen;
2503 	log->log_num[--log->log_left] = SYSCTL_TYPE(node->sysctl_flags);
2504 	if (log->log_num[log->log_left] != CTLTYPE_NODE)
2505 		log->log_num[--log->log_left] = node->sysctl_ver;
2506 	else
2507 		log->log_num[--log->log_left] = 0;
2508 
2509 	return (0);
2510 }
2511 
2512 void
2513 sysctl_teardown(struct sysctllog **logp)
2514 {
2515 	const struct sysctlnode *rnode;
2516 	struct sysctlnode node;
2517 	struct sysctllog *log;
2518 	uint namelen;
2519 	int *name, t, v, error, ni;
2520 	size_t sz;
2521 
2522 	if (logp == NULL || *logp == NULL)
2523 		return;
2524 	log = *logp;
2525 
2526 	error = sysctl_lock(NULL, NULL, 0);
2527 	if (error)
2528 		return;
2529 
2530 	memset(&node, 0, sizeof(node));
2531 
2532 	while (log->log_left < log->log_size) {
2533 		KASSERT((log->log_left + 3 < log->log_size) &&
2534 			(log->log_left + log->log_num[log->log_left + 2] <=
2535 			 log->log_size));
2536 		v = log->log_num[log->log_left++];
2537 		t = log->log_num[log->log_left++];
2538 		namelen = log->log_num[log->log_left++];
2539 		name = &log->log_num[log->log_left];
2540 
2541 		node.sysctl_num = name[namelen - 1];
2542 		node.sysctl_flags = SYSCTL_VERSION|t;
2543 		node.sysctl_ver = v;
2544 
2545 		rnode = log->log_root;
2546 		error = sysctl_locate(NULL, &name[0], namelen, &rnode, &ni);
2547 		if (error == 0) {
2548 			name[namelen - 1] = CTL_DESTROY;
2549 			rnode = rnode->sysctl_parent;
2550 			sz = 0;
2551 			(void)sysctl_destroy(&name[namelen - 1], 1, NULL,
2552 					     &sz, &node, sizeof(node),
2553 					     &name[0], NULL, rnode);
2554 		}
2555 
2556 		log->log_left += namelen;
2557 	}
2558 
2559 	KASSERT(log->log_size == log->log_left);
2560 	free(log->log_num, M_SYSCTLDATA);
2561 	free(log, M_SYSCTLDATA);
2562 	*logp = NULL;
2563 
2564 	sysctl_unlock(NULL);
2565 }
2566 
2567 /*
2568  * ********************************************************************
2569  * old_sysctl -- A routine to bridge old-style internal calls to the
2570  * new infrastructure.
2571  * ********************************************************************
2572  */
2573 int
2574 old_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp,
2575 	   void *newp, size_t newlen, struct lwp *l)
2576 {
2577 	int error;
2578 	size_t oldlen = 0;
2579 	size_t savelen;
2580 
2581 	if (oldlenp) {
2582 		oldlen = *oldlenp;
2583 	}
2584 	savelen = oldlen;
2585 
2586 	error = sysctl_lock(l, oldp, savelen);
2587 	if (error)
2588 		return (error);
2589 	error = sysctl_dispatch(name, namelen, oldp, &oldlen,
2590 				newp, newlen, name, l, NULL);
2591 	sysctl_unlock(l);
2592 	if (error == 0 && oldp != NULL && savelen < oldlen)
2593 		error = ENOMEM;
2594 
2595 	if (oldlenp) {
2596 		*oldlenp = oldlen;
2597 	}
2598 
2599 	return (error);
2600 }
2601 
2602 /*
2603  * ********************************************************************
2604  * Section 4: Generic helper routines
2605  * ********************************************************************
2606  * "helper" routines that can do more finely grained access control,
2607  * construct structures from disparate information, create the
2608  * appearance of more nodes and sub-trees, etc.  for example, if
2609  * CTL_PROC wanted a helper function, it could respond to a CTL_QUERY
2610  * with a dynamically created list of nodes that represented the
2611  * currently running processes at that instant.
2612  * ********************************************************************
2613  */
2614 
2615 /*
2616  * first, a few generic helpers that provide:
2617  *
2618  * sysctl_needfunc()		a readonly interface that emits a warning
2619  * sysctl_notavail()		returns EOPNOTSUPP (generic error)
2620  * sysctl_null()		an empty return buffer with no error
2621  */
2622 int
2623 sysctl_needfunc(SYSCTLFN_ARGS)
2624 {
2625 	int error;
2626 
2627 	printf("!!SYSCTL_NEEDFUNC!!\n");
2628 
2629 	if (newp != NULL || namelen != 0)
2630 		return (EOPNOTSUPP);
2631 
2632 	error = 0;
2633 	if (oldp != NULL)
2634 		error = sysctl_copyout(l, rnode->sysctl_data, oldp,
2635 				       MIN(rnode->sysctl_size, *oldlenp));
2636 	*oldlenp = rnode->sysctl_size;
2637 
2638 	return (error);
2639 }
2640 
2641 int
2642 sysctl_notavail(SYSCTLFN_ARGS)
2643 {
2644 
2645 	if (namelen == 1 && name[0] == CTL_QUERY)
2646 		return (sysctl_query(SYSCTLFN_CALL(rnode)));
2647 
2648 	return (EOPNOTSUPP);
2649 }
2650 
2651 int
2652 sysctl_null(SYSCTLFN_ARGS)
2653 {
2654 
2655 	*oldlenp = 0;
2656 
2657 	return (0);
2658 }
2659 
2660 /*
2661  * ********************************************************************
2662  * Section 5: The machinery that makes it all go
2663  * ********************************************************************
2664  * Memory "manglement" routines.  Not much to this, eh?
2665  * ********************************************************************
2666  */
2667 static int
2668 sysctl_alloc(struct sysctlnode *p, int x)
2669 {
2670 	int i;
2671 	struct sysctlnode *n;
2672 
2673 	assert(p->sysctl_child == NULL);
2674 
2675 	if (x == 1)
2676 		n = malloc(sizeof(struct sysctlnode),
2677 		       M_SYSCTLNODE, M_WAITOK|M_CANFAIL);
2678 	else
2679 		n = malloc(SYSCTL_DEFSIZE * sizeof(struct sysctlnode),
2680 		       M_SYSCTLNODE, M_WAITOK|M_CANFAIL);
2681 	if (n == NULL)
2682 		return (ENOMEM);
2683 
2684 	if (x == 1) {
2685 		memset(n, 0, sizeof(struct sysctlnode));
2686 		p->sysctl_csize = 1;
2687 	} else {
2688 		memset(n, 0, SYSCTL_DEFSIZE * sizeof(struct sysctlnode));
2689 		p->sysctl_csize = SYSCTL_DEFSIZE;
2690 	}
2691 	p->sysctl_clen = 0;
2692 
2693 	for (i = 0; i < p->sysctl_csize; i++)
2694 		n[i].sysctl_parent = p;
2695 
2696 	p->sysctl_child = n;
2697 	return (0);
2698 }
2699 
2700 static int
2701 sysctl_realloc(struct sysctlnode *p)
2702 {
2703 	int i, j;
2704 	struct sysctlnode *n;
2705 
2706 	assert(p->sysctl_csize == p->sysctl_clen);
2707 
2708 	/*
2709 	 * how many do we have...how many should we make?
2710 	 */
2711 	i = p->sysctl_clen;
2712 	n = malloc(2 * i * sizeof(struct sysctlnode), M_SYSCTLNODE,
2713 		   M_WAITOK|M_CANFAIL);
2714 	if (n == NULL)
2715 		return (ENOMEM);
2716 
2717 	/*
2718 	 * move old children over...initialize new children
2719 	 */
2720 	memcpy(n, p->sysctl_child, i * sizeof(struct sysctlnode));
2721 	memset(&n[i], 0, i * sizeof(struct sysctlnode));
2722 	p->sysctl_csize = 2 * i;
2723 
2724 	/*
2725 	 * reattach moved (and new) children to parent; if a moved
2726 	 * child node has children, reattach the parent pointers of
2727 	 * grandchildren
2728 	 */
2729         for (i = 0; i < p->sysctl_csize; i++) {
2730                 n[i].sysctl_parent = p;
2731 		if (n[i].sysctl_child != NULL) {
2732 			for (j = 0; j < n[i].sysctl_csize; j++)
2733 				n[i].sysctl_child[j].sysctl_parent = &n[i];
2734 		}
2735 	}
2736 
2737 	/*
2738 	 * get out with the old and in with the new
2739 	 */
2740 	free(p->sysctl_child, M_SYSCTLNODE);
2741 	p->sysctl_child = n;
2742 
2743 	return (0);
2744 }
2745 
2746 static int
2747 sysctl_log_realloc(struct sysctllog *log)
2748 {
2749 	int *n, s, d;
2750 
2751 	s = log->log_size * 2;
2752 	d = log->log_size;
2753 
2754 	n = malloc(s * sizeof(int), M_SYSCTLDATA, M_WAITOK|M_CANFAIL);
2755 	if (n == NULL)
2756 		return (-1);
2757 
2758 	memset(n, 0, s * sizeof(int));
2759 	memcpy(&n[d], log->log_num, d * sizeof(int));
2760 	free(log->log_num, M_SYSCTLDATA);
2761 	log->log_num = n;
2762 	if (d)
2763 		log->log_left += d;
2764 	else
2765 		log->log_left = s;
2766 	log->log_size = s;
2767 
2768 	return (0);
2769 }
2770 
2771 /*
2772  * ********************************************************************
2773  * Section 6: Conversion between API versions wrt the sysctlnode
2774  * ********************************************************************
2775  */
2776 static int
2777 sysctl_cvt_in(struct lwp *l, int *vp, const void *i, size_t sz,
2778 	      struct sysctlnode *node)
2779 {
2780 	int error, flags;
2781 
2782 	if (i == NULL || sz < sizeof(flags))
2783 		return (EINVAL);
2784 
2785 	error = sysctl_copyin(l, i, &flags, sizeof(flags));
2786 	if (error)
2787 		return (error);
2788 
2789 #if (SYSCTL_VERSION != SYSCTL_VERS_1)
2790 #error sysctl_cvt_in: no support for SYSCTL_VERSION
2791 #endif /*  (SYSCTL_VERSION != SYSCTL_VERS_1) */
2792 
2793 	if (sz == sizeof(*node) &&
2794 	    SYSCTL_VERS(flags) == SYSCTL_VERSION) {
2795 		error = sysctl_copyin(l, i, node, sizeof(*node));
2796 		if (error)
2797 			return (error);
2798 		*vp = SYSCTL_VERSION;
2799 		return (0);
2800 	}
2801 
2802 	return (EINVAL);
2803 }
2804 
2805 static int
2806 sysctl_cvt_out(struct lwp *l, int v, const struct sysctlnode *i,
2807 	       void *ovp, size_t left, size_t *szp)
2808 {
2809 	size_t sz = sizeof(*i);
2810 	const void *src = i;
2811 	int error;
2812 
2813 	switch (v) {
2814 	case SYSCTL_VERS_0:
2815 		return (EINVAL);
2816 
2817 #if (SYSCTL_VERSION != SYSCTL_VERS_1)
2818 #error sysctl_cvt_out: no support for SYSCTL_VERSION
2819 #endif /*  (SYSCTL_VERSION != SYSCTL_VERS_1) */
2820 
2821 	case SYSCTL_VERSION:
2822 		/* nothing more to do here */
2823 		break;
2824 	}
2825 
2826 	if (ovp != NULL && left >= sz) {
2827 		error = sysctl_copyout(l, src, ovp, sz);
2828 		if (error)
2829 			return (error);
2830 	}
2831 
2832 	if (szp != NULL)
2833 		*szp = sz;
2834 
2835 	return (0);
2836 }
2837