Home | History | Annotate | Line # | Download | only in kern
kern_sysctl.c revision 1.155
      1 /*	$NetBSD: kern_sysctl.c,v 1.155 2003/12/28 22:36:37 atatat 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 1.155 2003/12/28 22:36:37 atatat Exp $");
     79 
     80 #include "opt_defcorename.h"
     81 #include "opt_insecure.h"
     82 #include "ksyms.h"
     83 
     84 #include <sys/param.h>
     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 <machine/stdarg.h>
     94 
     95 MALLOC_DEFINE(M_SYSCTLNODE, "sysctlnode", "sysctl node structures");
     96 MALLOC_DEFINE(M_SYSCTLDATA, "sysctldata", "misc sysctl data");
     97 
     98 static int sysctl_mmap(SYSCTLFN_RWPROTO);
     99 static int sysctl_alloc(struct sysctlnode *, int);
    100 static int sysctl_realloc(struct sysctlnode *);
    101 
    102 /*
    103  * the "root" of the new sysctl tree
    104  */
    105 static struct sysctlnode sysctl_root = {
    106 	.sysctl_flags = SYSCTL_ROOT|
    107 	    SYSCTL_READWRITE|
    108 	    CTLTYPE_NODE,
    109 	.sysctl_num = 0,
    110 	.sysctl_size = sizeof(struct sysctlnode),
    111 	.sysctl_name = "(root)",
    112 };
    113 
    114 /*
    115  * link set of functions that add nodes at boot time (see also
    116  * sysctl_buildtree())
    117  */
    118 __link_set_decl(sysctl_funcs, sysctl_setup_func);
    119 
    120 /*
    121  * The `sysctl_lock' is intended to serialize access to the sysctl
    122  * tree.  Given that it is now (a) dynamic, and (b) most consumers of
    123  * sysctl are going to be copying data out, the old `sysctl_memlock'
    124  * has been `upgraded' to simply guard the whole tree.
    125  *
    126  * The two new data here are to keep track of the locked chunk of
    127  * memory, if there is one, so that it can be released more easily
    128  * from anywhere.
    129  */
    130 struct lock sysctl_treelock;
    131 caddr_t sysctl_memaddr;
    132 size_t sysctl_memsize;
    133 
    134 /*
    135  * Attributes stored in the kernel.
    136  */
    137 char hostname[MAXHOSTNAMELEN];
    138 int hostnamelen;
    139 
    140 char domainname[MAXHOSTNAMELEN];
    141 int domainnamelen;
    142 
    143 long hostid;
    144 
    145 #ifdef INSECURE
    146 int securelevel = -1;
    147 #else
    148 int securelevel = 0;
    149 #endif
    150 
    151 #ifndef DEFCORENAME
    152 #define	DEFCORENAME	"%n.core"
    153 #endif
    154 char defcorename[MAXPATHLEN] = DEFCORENAME;
    155 
    156 /*
    157  * ********************************************************************
    158  * Section 0: Some simple glue
    159  * ********************************************************************
    160  * By wrapping copyin(), copyout(), and copyinstr() like this, we can
    161  * stop caring about who's calling us and simplify some code a bunch.
    162  * ********************************************************************
    163  */
    164 static inline int
    165 sysctl_copyin(const struct lwp *l, const void *uaddr, void *kaddr, size_t len)
    166 {
    167 
    168 	if (l != NULL)
    169 		return (copyin(uaddr, kaddr, len));
    170 
    171 	memcpy(kaddr, uaddr, len);
    172 
    173 	return (0);
    174 }
    175 
    176 static inline int
    177 sysctl_copyout(const struct lwp *l, const void *kaddr, void *uaddr, size_t len)
    178 {
    179 
    180 	if (l != NULL)
    181 		return (copyout(kaddr, uaddr, len));
    182 
    183 	memcpy(uaddr, kaddr, len);
    184 
    185 	return (0);
    186 }
    187 
    188 static inline int
    189 sysctl_copyinstr(const struct lwp *l, const void *uaddr, void *kaddr,
    190 		 size_t len, size_t *done)
    191 {
    192 
    193 	if (l != NULL)
    194 		return (copyinstr(uaddr, kaddr, len, done));
    195 	else
    196 		return (copystr(uaddr, kaddr, len, done));
    197 }
    198 
    199 /*
    200  * ********************************************************************
    201  * Initialize sysctl subsystem.
    202  * ********************************************************************
    203  */
    204 void
    205 sysctl_init(void)
    206 {
    207 	sysctl_setup_func **sysctl_setup, f;
    208 
    209 	lockinit(&sysctl_treelock, PRIBIO|PCATCH, "sysctl", 0, 0);
    210 
    211 	/*
    212 	 * dynamic mib numbers start here
    213 	 */
    214 	sysctl_root.sysctl_num = CREATE_BASE;
    215 
    216         __link_set_foreach(sysctl_setup, sysctl_funcs) {
    217 		/*
    218 		 * XXX - why do i have to coerce the pointers like this?
    219 		 */
    220 		f = (void*)*sysctl_setup;
    221 		(*f)();
    222 	}
    223 
    224 	/*
    225 	 * setting this means no more permanent nodes can be added,
    226 	 * trees that claim to be readonly at the root now are, and if
    227 	 * the main tree is readonly, *everything* is.
    228 	 */
    229 	sysctl_root.sysctl_flags |= SYSCTL_PERMANENT;
    230 
    231 }
    232 
    233 /*
    234  * ********************************************************************
    235  * The main native sysctl system call itself.
    236  * ********************************************************************
    237  */
    238 int
    239 sys___sysctl(struct lwp *l, void *v, register_t *retval)
    240 {
    241 	struct sys___sysctl_args /* {
    242 		syscallarg(int *) name;
    243 		syscallarg(u_int) namelen;
    244 		syscallarg(void *) old;
    245 		syscallarg(size_t *) oldlenp;
    246 		syscallarg(void *) new;
    247 		syscallarg(size_t) newlen;
    248 	} */ *uap = v;
    249 	int error, nerror, name[CTL_MAXNAME];
    250 	size_t oldlen, savelen, *oldlenp;
    251 
    252 	/*
    253 	 * get oldlen
    254 	 */
    255 	oldlen = 0;
    256 	oldlenp = SCARG(uap, oldlenp);
    257 	if (oldlenp != NULL) {
    258 		error = copyin(oldlenp, &oldlen, sizeof(oldlen));
    259 		if (error)
    260 			return (error);
    261 	}
    262 	savelen = oldlen;
    263 
    264 	/*
    265 	 * top-level sysctl names may or may not be non-terminal, but
    266 	 * we don't care
    267 	 */
    268 	if (SCARG(uap, namelen) > CTL_MAXNAME || SCARG(uap, namelen) < 1)
    269 		return (EINVAL);
    270 	error = copyin(SCARG(uap, name), &name,
    271 		       SCARG(uap, namelen) * sizeof(int));
    272 	if (error)
    273 		return (error);
    274 
    275 	/*
    276 	 * wire old so that copyout() is less likely to fail?
    277 	 */
    278 	error = sysctl_lock(l, SCARG(uap, old), savelen);
    279 	if (error)
    280 		return (error);
    281 
    282 	/*
    283 	 * do sysctl work (NULL means main built-in default tree)
    284 	 */
    285 	error = sysctl_dispatch(&name[0], SCARG(uap, namelen),
    286 				SCARG(uap, old), &oldlen,
    287 				SCARG(uap, new), SCARG(uap, newlen),
    288 				&name[0], l, NULL);
    289 
    290 	/*
    291 	 * release the sysctl lock
    292 	 */
    293 	sysctl_unlock(l);
    294 
    295 	/*
    296 	 * set caller's oldlen to new value even in the face of an
    297 	 * error (if this gets an error and they didn't have one, they
    298 	 * get this one)
    299 	 */
    300 	if (oldlenp) {
    301 		nerror = copyout(&oldlen, oldlenp, sizeof(oldlen));
    302 		if (error == 0)
    303 			error = nerror;
    304 	}
    305 
    306 	/*
    307 	 * if the only problem is that we weren't given enough space,
    308 	 * that's an ENOMEM error
    309 	 */
    310 	if (error == 0 && SCARG(uap, old) != NULL && savelen < oldlen)
    311 		error = ENOMEM;
    312 
    313 	return (error);
    314 }
    315 
    316 /*
    317  * ********************************************************************
    318  * Section 1: How the tree is used
    319  * ********************************************************************
    320  * Implementations of sysctl for emulations should typically need only
    321  * these three functions in this order: lock the tree, dispatch
    322  * request into it, unlock the tree.
    323  * ********************************************************************
    324  */
    325 int
    326 sysctl_lock(struct lwp *l, void *oldp, size_t savelen)
    327 {
    328 	int error = 0;
    329 
    330 	error = lockmgr(&sysctl_treelock, LK_EXCLUSIVE, NULL);
    331 	if (error)
    332 		return (error);
    333 
    334 	if (l != NULL && oldp != NULL && savelen) {
    335 		error = uvm_vslock(l->l_proc, oldp, savelen, VM_PROT_WRITE);
    336 		if (error) {
    337 			(void) lockmgr(&sysctl_treelock, LK_RELEASE, NULL);
    338 			return (error);
    339 		}
    340 		sysctl_memaddr = oldp;
    341 		sysctl_memsize = savelen;
    342 	}
    343 
    344 	return (0);
    345 }
    346 
    347 /*
    348  * ********************************************************************
    349  * the main sysctl dispatch routine.  scans the given tree and picks a
    350  * function to call based on what it finds.
    351  * ********************************************************************
    352  */
    353 int
    354 sysctl_dispatch(SYSCTLFN_RWARGS)
    355 {
    356 	int error;
    357 	sysctlfn fn;
    358 	int ni;
    359 
    360 	fn = NULL;
    361 	error = sysctl_locate(l, name, namelen, &rnode, &ni);
    362 
    363 	/*
    364 	 * the node we ended up at has a function, so call it.  it can
    365 	 * hand off to query or create if it wants to.
    366 	 */
    367 	if (rnode->sysctl_func != NULL)
    368 		fn = rnode->sysctl_func;
    369 
    370 	/*
    371 	 * we found the node they were looking for, so do a lookup.
    372 	 */
    373 	else if (error == 0)
    374 		fn = (sysctlfn)sysctl_lookup; /* XXX may write to rnode */
    375 
    376 	/*
    377 	 * prospective parent node found, but the terminal node was
    378 	 * not.  generic operations associate with the parent.
    379 	 */
    380 	else if (error == ENOENT && (ni + 1) == namelen && name[ni] < 0) {
    381 		switch (name[ni]) {
    382 		case CTL_QUERY:
    383 			fn = sysctl_query;
    384 			break;
    385 		case CTL_CREATE:
    386 #if NKSYMS > 0
    387 		case CTL_CREATESYM:
    388 #endif /* NKSYMS > 0 */
    389 			fn = (sysctlfn)sysctl_create; /* we own the rnode */
    390 			break;
    391 		case CTL_DESTROY:
    392 			fn = (sysctlfn)sysctl_destroy; /* we own the rnode */
    393 			break;
    394 		case CTL_MMAP:
    395 			fn = (sysctlfn)sysctl_mmap; /* we own the rnode */
    396 			break;
    397 		default:
    398 			error = EOPNOTSUPP;
    399 			break;
    400 		}
    401 	}
    402 
    403 	/*
    404 	 * after all of that, maybe we found someone who knows how to
    405 	 * get us what we want?
    406 	 */
    407 	if (fn != NULL)
    408 		error = (*fn)(name + ni, namelen - ni, oldp, oldlenp,
    409 			      newp, newlen, name, l, rnode);
    410 
    411 	else if (error == 0)
    412 		error = EOPNOTSUPP;
    413 
    414 	return (error);
    415 }
    416 
    417 /*
    418  * ********************************************************************
    419  * Releases the tree lock.  Note that if uvm_vslock() was called when
    420  * the lock was taken, we release that memory now.  By keeping track
    421  * of where and how much by ourselves, the lock can be released much
    422  * more easily from anywhere.
    423  * ********************************************************************
    424  */
    425 void
    426 sysctl_unlock(struct lwp *l)
    427 {
    428 
    429 	if (l != NULL && sysctl_memsize != 0) {
    430 		uvm_vsunlock(l->l_proc, sysctl_memaddr, sysctl_memsize);
    431 		sysctl_memsize = 0;
    432 	}
    433 
    434 	(void) lockmgr(&sysctl_treelock, LK_RELEASE, NULL);
    435 }
    436 
    437 /*
    438  * ********************************************************************
    439  * Section 2: The main tree interfaces
    440  * ********************************************************************
    441  * This is how sysctl_dispatch() does its work, and you can too, by
    442  * calling these routines from helpers (though typically only
    443  * sysctl_lookup() will be used).  The tree MUST BE LOCKED when these
    444  * are called.
    445  * ********************************************************************
    446  */
    447 
    448 /*
    449  * sysctl_locate -- Finds the node matching the given mib under the
    450  * given tree (via rv).  If no tree is given, we fall back to the
    451  * native tree.  The current process (via l) is used for access
    452  * control on the tree (some nodes may be traverable only by root) and
    453  * on return, nip will show how many numbers in the mib were consumed.
    454  */
    455 int
    456 sysctl_locate(struct lwp *l, const int *name, u_int namelen,
    457 	      struct sysctlnode **rv, int *nip)
    458 {
    459 	struct sysctlnode *node, *pnode;
    460 	int tn, si, ni, error, alias;
    461 
    462 	/*
    463 	 * basic checks and setup
    464 	 */
    465 	if (*rv == NULL)
    466 		*rv = &sysctl_root;
    467 	if (nip)
    468 		*nip = 0;
    469 	if (namelen < 0)
    470 		return (EINVAL);
    471 	if (namelen == 0)
    472 		return (0);
    473 
    474 	/*
    475 	 * search starts from "root"
    476 	 */
    477 	pnode = *rv;
    478 	node = pnode->sysctl_child;
    479 	error = 0;
    480 
    481 	/*
    482 	 * scan for node to which new node should be attached
    483 	 */
    484 	for (ni = 0; ni < namelen; ni++) {
    485 		/*
    486 		 * walked off bottom of tree
    487 		 */
    488 		if (node == NULL) {
    489 			if (SYSCTL_TYPE(pnode->sysctl_flags) == CTLTYPE_NODE)
    490 				error = ENOENT;
    491 			else
    492 				error = ENOTDIR;
    493 			break;
    494 		}
    495 		/*
    496 		 * can anyone traverse this node or only root?
    497 		 */
    498 		if (l != NULL && (pnode->sysctl_flags & SYSCTL_PRIVATE) &&
    499 		    (error = suser(l->l_proc->p_ucred, &l->l_proc->p_acflag))
    500 		    != 0)
    501 			return (error);
    502 		/*
    503 		 * find a child node with the right number
    504 		 */
    505 		tn = name[ni];
    506 		alias = 0;
    507 		for (si = 0; si < pnode->sysctl_clen; si++) {
    508 			if (node[si].sysctl_num == tn ||
    509 			    (tn >= 0 &&
    510 			     node[si].sysctl_flags & SYSCTL_ANYNUMBER)) {
    511 				if (node[si].sysctl_flags & SYSCTL_ALIAS) {
    512 					if (alias++ == 4)
    513 						si = pnode->sysctl_clen - 1;
    514 					else {
    515 						tn = node[si].sysctl_alias;
    516 						si = -1;
    517 					}
    518 				}
    519 				else
    520 					break;
    521 			}
    522 		}
    523 		/*
    524 		 * if we ran off the end, it obviously doesn't exist
    525 		 */
    526 		if (si == pnode->sysctl_clen) {
    527 			error = ENOENT;
    528 			break;
    529 		}
    530 		/*
    531 		 * so far so good, move on down the line
    532 		 */
    533 		pnode = &node[si];
    534 		if (SYSCTL_TYPE(pnode->sysctl_flags) == CTLTYPE_NODE)
    535 			node = node[si].sysctl_child;
    536 		else
    537 			node = NULL;
    538 	}
    539 
    540 	*rv = pnode;
    541 	if (nip)
    542 		*nip = ni;
    543 
    544 	return (error);
    545 }
    546 
    547 /*
    548  * sysctl_query -- The auto-discovery engine.  Copies out the
    549  * descriptions on nodes under the given node and handles overlay
    550  * trees.
    551  */
    552 int
    553 sysctl_query(SYSCTLFN_ARGS)
    554 {
    555 	int error, ni, elim;
    556 	size_t out, left, t;
    557 	struct sysctlnode *enode, *onode;
    558 
    559 	if (newp != NULL)
    560 		return (EPERM);
    561 	if (SYSCTL_TYPE(rnode->sysctl_flags) != CTLTYPE_NODE)
    562 		return (ENOTDIR);
    563 	if (namelen != 1 || name[0] != CTL_QUERY)
    564 		return (EINVAL);
    565 
    566 	error = 0;
    567 	out = 0;
    568 	left = *oldlenp;
    569 	elim = 0;
    570 	enode = NULL;
    571 
    572 	/*
    573 	 * process has overlay tree
    574 	 */
    575 	if (l && l->l_proc->p_emul->e_sysctlovly) {
    576 		enode = (void*)l->l_proc->p_emul->e_sysctlovly;
    577 		elim = (name - oname);
    578 		error = sysctl_locate(l, oname, elim, &enode, NULL);
    579 		if (error == 0) {
    580 			/* ah, found parent in overlay */
    581 			elim = enode->sysctl_clen;
    582 			enode = enode->sysctl_child;
    583 		}
    584 		else {
    585 			error = 0;
    586 			elim = 0;
    587 			enode = NULL;
    588 		}
    589 	}
    590 
    591 	for (ni = 0; ni < rnode->sysctl_clen; ni++) {
    592 		t = MIN(left, sizeof(struct sysctlnode));
    593 		onode = &rnode->sysctl_child[ni];
    594 		if (enode && enode->sysctl_num == onode->sysctl_num) {
    595 			if (SYSCTL_TYPE(enode->sysctl_flags) !=
    596 			    CTLTYPE_NODE)
    597 				onode = enode;
    598 			if (--elim > 0)
    599 				enode++;
    600 			else
    601 				enode = NULL;
    602 		}
    603 		if (oldp != NULL && t > 0)
    604 			error = sysctl_copyout(l, onode, (char*)oldp + out, t);
    605 		if (error)
    606 			return (error);
    607 		out += sizeof(struct sysctlnode);
    608 		left -= t;
    609 	}
    610 
    611 	/*
    612 	 * overlay trees *MUST* be entirely consumed
    613 	 */
    614 	KASSERT(enode == NULL);
    615 
    616 	*oldlenp = out;
    617 
    618 	return (error);
    619 }
    620 
    621 #ifdef SYSCTL_DEBUG_CREATE
    622 #undef sysctl_create
    623 #endif /* SYSCTL_DEBUG_CREATE */
    624 
    625 /*
    626  * sysctl_create -- Adds a node (the description of which is taken
    627  * from newp) to the tree, returning a copy of it in the space pointed
    628  * to by oldp.  In the event that the requested slot is already taken
    629  * (either by name or by number), the offending node is returned
    630  * instead.  Yes, this is complex, but we want to make sure everything
    631  * is proper.
    632  */
    633 int
    634 sysctl_create(SYSCTLFN_RWARGS)
    635 {
    636 	struct sysctlnode nnode, *node, *pnode;
    637 	int error, ni, at, nm, type, sz, flags, rw, anum;
    638 	void *own;
    639 
    640 	error = 0;
    641 	own = NULL;
    642 	anum = -1;
    643 
    644 	if (namelen != 1 || (name[namelen - 1] != CTL_CREATE
    645 #if NKSYMS > 0
    646 			     && name[namelen - 1] != CTL_CREATESYM
    647 #endif /* NKSYMS > 0 */
    648 			     ))
    649 		return (EINVAL);
    650 
    651 	/*
    652 	 * processes can only add nodes at securelevel 0, must be
    653 	 * root, and can't add nodes to a parent that's not writeable
    654 	 */
    655 	if (l != NULL) {
    656 		if (securelevel > 0)
    657 			return (EPERM);
    658 		error = suser(l->l_proc->p_ucred, &l->l_proc->p_acflag);
    659 		if (error)
    660 			return (error);
    661 #ifndef SYSCTL_DISALLOW_CREATE
    662 		if (!(rnode->sysctl_flags & SYSCTL_READWRITE))
    663 #endif /* SYSCTL_DISALLOW_CREATE */
    664 			return (EPERM);
    665 	}
    666 
    667 	/*
    668 	 * nothing can add a node if:
    669 	 * we've finished initial set up and
    670 	 * the tree itself is not writeable or
    671 	 * the entire sysctl system is not writeable
    672 	 */
    673 	if ((sysctl_root.sysctl_flags & SYSCTL_PERMANENT) &&
    674 	    (!(sysctl_rootof(rnode)->sysctl_flags & SYSCTL_READWRITE) ||
    675 	     !(sysctl_root.sysctl_flags & SYSCTL_READWRITE)))
    676 		return (EPERM);
    677 
    678 	/*
    679 	 * it must be a "node", not a "int" or something
    680 	 */
    681 	if (SYSCTL_TYPE(rnode->sysctl_flags) != CTLTYPE_NODE)
    682 		return (ENOTDIR);
    683 	pnode = rnode;
    684 
    685 	if (newp == NULL || newlen != sizeof(struct sysctlnode))
    686 		return (EINVAL);
    687 	error = sysctl_copyin(l, newp, &nnode, sizeof(struct sysctlnode));
    688 	if (error)
    689 		return (error);
    690 
    691 	/*
    692 	 * nodes passed in don't *have* parents
    693 	 */
    694 	if (nnode.sysctl_parent != NULL)
    695 		return (EINVAL);
    696 
    697 	/*
    698 	 * if we are indeed adding it, it should be a "good" name and
    699 	 * number
    700 	 */
    701 	nm = nnode.sysctl_num;
    702 #if NKSYMS > 0
    703 	if (nm == CTL_CREATESYM)
    704 		nm = CTL_CREATE;
    705 #endif /* NKSYMS > 0 */
    706 	if (nm < 0 && nm != CTL_CREATE)
    707 		return (EINVAL);
    708 	sz = 0;
    709 
    710 	/*
    711 	 * the name can't start with a digit
    712 	 */
    713 	if (nnode.sysctl_name[sz] >= '0' &&
    714 	    nnode.sysctl_name[sz] <= '9')
    715 		return (EINVAL);
    716 
    717 	/*
    718 	 * the name must be only alphanumerics or - or _, longer than
    719 	 * 0 bytes and less that SYSCTL_NAMELEN
    720 	 */
    721 	while (sz < SYSCTL_NAMELEN && nnode.sysctl_name[sz] != '\0') {
    722 		if ((nnode.sysctl_name[sz] >= '0' &&
    723 		     nnode.sysctl_name[sz] <= '9') ||
    724 		    (nnode.sysctl_name[sz] >= 'A' &&
    725 		     nnode.sysctl_name[sz] <= 'Z') ||
    726 		    (nnode.sysctl_name[sz] >= 'a' &&
    727 		     nnode.sysctl_name[sz] <= 'z') ||
    728 		    nnode.sysctl_name[sz] == '-' ||
    729 		    nnode.sysctl_name[sz] == '_')
    730 			sz++;
    731 		else
    732 			return (EINVAL);
    733 	}
    734 	if (sz == 0 || sz == SYSCTL_NAMELEN)
    735 		return (EINVAL);
    736 
    737 	/*
    738 	 * various checks revolve around size vs type, etc
    739 	 */
    740 	type = SYSCTL_TYPE(nnode.sysctl_flags);
    741 	flags = SYSCTL_FLAGS(nnode.sysctl_flags);
    742 	rw = (flags & SYSCTL_READWRITE) ? B_WRITE : B_READ;
    743 	sz = nnode.sysctl_size;
    744 
    745 	/*
    746 	 * find out if there's a collision, and if so, let the caller
    747 	 * know what they collided with
    748 	 */
    749 	node = pnode->sysctl_child;
    750 	if (((flags & SYSCTL_ANYNUMBER) && node) ||
    751 	    (node && node->sysctl_flags & SYSCTL_ANYNUMBER))
    752 		return (EINVAL);
    753 	for (ni = at = 0; ni < pnode->sysctl_clen; ni++) {
    754 		if (nm == node[ni].sysctl_num ||
    755 		    strcmp(nnode.sysctl_name, node[ni].sysctl_name) == 0) {
    756 			if (oldp != NULL) {
    757 				/*
    758 				 * ignore error here, since we
    759 				 * are already fixed on EEXIST
    760 				 */
    761 				(void)sysctl_copyout(l, &node[ni], oldp,
    762 				     MIN(*oldlenp, sizeof(struct sysctlnode)));
    763 			}
    764 			*oldlenp = sizeof(struct sysctlnode);
    765 			return (EEXIST);
    766 		}
    767 		if (nm > node[ni].sysctl_num)
    768 			at++;
    769 	}
    770 
    771 	/*
    772 	 * use sysctl_ver to add to the tree iff it hasn't changed
    773 	 */
    774 	if (nnode.sysctl_ver != 0) {
    775 		/*
    776 		 * a specified value must match either the parent
    777 		 * node's version or the root node's version
    778 		 */
    779 		if (nnode.sysctl_ver != sysctl_rootof(rnode)->sysctl_ver &&
    780 		    nnode.sysctl_ver != rnode->sysctl_ver) {
    781 			return (EINVAL);
    782 		}
    783 	}
    784 
    785 	/*
    786 	 * only the kernel can assign functions to entries
    787 	 */
    788 	if (l != NULL && nnode.sysctl_func != NULL)
    789 		return (EPERM);
    790 
    791 	/*
    792 	 * only the kernel can create permanent entries, and only then
    793 	 * before the kernel is finished setting itself up
    794 	 */
    795 	if (l != NULL && (flags & ~SYSCTL_USERFLAGS))
    796 		return (EPERM);
    797 	if ((flags & SYSCTL_PERMANENT) &
    798 	    (sysctl_root.sysctl_flags & SYSCTL_PERMANENT))
    799 		return (EPERM);
    800 	if ((flags & (SYSCTL_OWNDATA | SYSCTL_IMMEDIATE)) ==
    801 	    (SYSCTL_OWNDATA | SYSCTL_IMMEDIATE))
    802 		return (EINVAL);
    803 	if ((flags & SYSCTL_IMMEDIATE) &&
    804 	    type != CTLTYPE_INT && type != CTLTYPE_QUAD)
    805 		return (EINVAL);
    806 
    807 	/*
    808 	 * check size, or set it if unset and we can figure it out.
    809 	 * kernel created nodes are allowed to have a function instead
    810 	 * of a size (or a data pointer).
    811 	 */
    812 	switch (type) {
    813 	case CTLTYPE_NODE:
    814 		/*
    815 		 * only *i* can assert the size of a node
    816 		 */
    817 		if (flags & SYSCTL_ALIAS) {
    818 			anum = nnode.sysctl_alias;
    819 			if (anum < 0)
    820 				return (EINVAL);
    821 			nnode.sysctl_alias = 0;
    822 		}
    823 		if (sz != 0 || nnode.sysctl_data != NULL)
    824 			return (EINVAL);
    825 		if (nnode.sysctl_csize != 0 ||
    826 		    nnode.sysctl_clen != 0 ||
    827 		    nnode.sysctl_child != 0)
    828 			return (EINVAL);
    829 		if (flags & SYSCTL_OWNDATA)
    830 			return (EINVAL);
    831 		sz = sizeof(struct sysctlnode);
    832 		break;
    833 	case CTLTYPE_INT:
    834 		/*
    835 		 * since an int is an int, if the size is not given or
    836 		 * is wrong, we can "int-uit" it.
    837 		 */
    838 		if (sz != 0 && sz != sizeof(int))
    839 			return (EINVAL);
    840 		sz = sizeof(int);
    841 		break;
    842 	case CTLTYPE_STRING:
    843 		/*
    844 		 * strings are a little more tricky
    845 		 */
    846 		if (sz == 0) {
    847 			if (l == NULL) {
    848 				if (nnode.sysctl_func == NULL) {
    849 					if (nnode.sysctl_data == NULL)
    850 						return (EINVAL);
    851 					else
    852 						sz = strlen(nnode.sysctl_data) +
    853 						    1;
    854 				}
    855 			}
    856 			else if (nnode.sysctl_data == NULL &&
    857 				 flags & SYSCTL_OWNDATA) {
    858 				return (EINVAL);
    859 			}
    860 			else {
    861 				char *v, *e;
    862 				size_t s;
    863 
    864 				/*
    865 				 * arbitrary limit here...
    866 				 */
    867 				e = NULL; /* XXX: gcc on NetBSD/sparc */
    868 				for (s = PAGE_SIZE, v = nnode.sysctl_data;
    869 				     s < 32 * PAGE_SIZE;
    870 				     s += PAGE_SIZE, v += PAGE_SIZE) {
    871 /*
    872  * XXX @@@ the use of uvm_kernacc() can generate false negatives on
    873  * some ports, so this needs to be refined shortly.
    874  */
    875 					if (!uvm_kernacc(v, PAGE_SIZE, rw))
    876 						return (EFAULT);
    877 					e = memchr(v, '\0', PAGE_SIZE);
    878 					if (e != NULL)
    879 						break;
    880 				}
    881 				if (s >= 32 * PAGE_SIZE)
    882 					return (ERANGE);
    883 				sz = e - ((char*)nnode.sysctl_data) + 1;
    884 			}
    885 		}
    886 		break;
    887 	case CTLTYPE_QUAD:
    888 		if (sz != 0 && sz != sizeof(u_quad_t))
    889 			return (EINVAL);
    890 		sz = sizeof(u_quad_t);
    891 		break;
    892 	case CTLTYPE_STRUCT:
    893 		if (sz == 0) {
    894 			if (l != NULL || nnode.sysctl_func == NULL)
    895 				return (EINVAL);
    896 			if (flags & SYSCTL_OWNDATA)
    897 				return (EINVAL);
    898 		}
    899 		break;
    900 	default:
    901 		return (EINVAL);
    902 	}
    903 
    904 	/*
    905 	 * at this point, if sz is zero, we *must* have a
    906 	 * function to go with it and we can't own it.
    907 	 */
    908 
    909 	/*
    910 	 *  l  ptr own
    911 	 *  0   0   0  -> EINVAL (if no func)
    912 	 *  0   0   1  -> own
    913 	 *  0   1   0  -> kptr
    914 	 *  0   1   1  -> kptr
    915 	 *  1   0   0  -> EINVAL
    916 	 *  1   0   1  -> own
    917 	 *  1   1   0  -> kptr, no own (check via uvm_kernacc)
    918 	 *  1   1   1  -> uptr, own
    919 	 */
    920 	if (type != CTLTYPE_NODE) {
    921 		if (sz != 0) {
    922 			if (flags & SYSCTL_OWNDATA) {
    923 				own = malloc(sz, M_SYSCTLDATA,
    924 					     M_WAITOK|M_CANFAIL);
    925 				if (nnode.sysctl_data == NULL)
    926 					memset(own, 0, sz);
    927 				else {
    928 					error = sysctl_copyin(l,
    929 					    nnode.sysctl_data, own, sz);
    930 					if (error != 0) {
    931 						FREE(own, M_SYSCTLDATA);
    932 						return (error);
    933 					}
    934 				}
    935 			}
    936 			else if ((nnode.sysctl_data != NULL) &&
    937 				 !(flags & SYSCTL_IMMEDIATE)) {
    938 #if NKSYMS > 0
    939 				if (name[namelen - 1] == CTL_CREATESYM) {
    940 					char symname[128]; /* XXX enough? */
    941 					u_long symaddr;
    942 					size_t symlen;
    943 
    944 					error = sysctl_copyinstr(l,
    945 					    nnode.sysctl_data, symname,
    946 					    sizeof(symname), &symlen);
    947 					if (error)
    948 						return (error);
    949 					error = ksyms_getval_from_kernel(NULL,
    950 					    symname, &symaddr, KSYMS_EXTERN);
    951 					if (error)
    952 						return (error); /* EINVAL? */
    953 					nnode.sysctl_data = (void*)symaddr;
    954 				}
    955 #endif /* NKSYMS > 0 */
    956 				if (!uvm_kernacc(nnode.sysctl_data, sz, rw)) {
    957 #ifdef HAVE_SOLUTION_TO_UVM_KERNACC_PROBLEM
    958 					/* XXX @@@ what is fix? */
    959 					return (EFAULT);
    960 #else /* HAVE_SOLUTION_TO_UVM_KERNACC_PROBLEM */
    961 /*
    962  * XXX @@@ the use of uvm_kernacc() can generate false negatives on
    963  * some ports, so this needs to be refined shortly.  by checking here
    964  * to see if SYSCTL_PERMANENT is set in the root, we can differentiate
    965  * between nodes being created from sysctl_init() during bootstrap and
    966  * "other nodes", so we can at least allow the bootstrap to succeed by
    967  * simply "trusting" the kernel not to shoot itself in the foot right
    968  * from the start.
    969  */
    970 					if ((sysctl_root.sysctl_flags &
    971 					    SYSCTL_PERMANENT)) {
    972 						printf("fault 2 %p %lu %d\n", nnode.sysctl_data, (unsigned long)sz, rw);
    973 						return (EFAULT);
    974 					}
    975 #endif /* HAVE_SOLUTION_TO_UVM_KERNACC_PROBLEM */
    976 				}
    977 			}
    978 		}
    979 		else if (nnode.sysctl_func == NULL)
    980 			return (EINVAL);
    981 	}
    982 
    983 	/*
    984 	 * a process can't assign a function to a node, and the kernel
    985 	 * can't create a node that has no function or data.
    986 	 * (XXX somewhat redundant check)
    987 	 */
    988 	if (l != NULL || nnode.sysctl_func == NULL) {
    989 		if (type != CTLTYPE_NODE &&
    990 		    nnode.sysctl_data == NULL &&
    991 		    !(flags & SYSCTL_IMMEDIATE) &&
    992 		    own == NULL)
    993 			return (EINVAL);
    994 	}
    995 
    996 #ifdef SYSCTL_DISALLOW_KWRITE
    997 	/*
    998 	 * a process can't create a writable node unless it refers to
    999 	 * new data.
   1000 	 */
   1001 	if (l != NULL && own == NULL && type != CTLTYPE_NODE &&
   1002 	    (flags & SYSCTL_READWRITE) != SYSCTL_READONLY &&
   1003 	    !(flags & SYSCTL_IMMEDIATE))
   1004 		return (EPERM);
   1005 #endif /* SYSCTL_DISALLOW_KWRITE */
   1006 
   1007 	/*
   1008 	 * make sure there's somewhere to put the new stuff.
   1009 	 */
   1010 	if (pnode->sysctl_child == NULL) {
   1011 		if (flags & SYSCTL_ANYNUMBER)
   1012 			error = sysctl_alloc(pnode, 1);
   1013 		else
   1014 			error = sysctl_alloc(pnode, 0);
   1015 		if (error)
   1016 			return (error);
   1017 	}
   1018 	node = pnode->sysctl_child;
   1019 
   1020 	/*
   1021 	 * no collisions, so pick a good dynamic number if we need to.
   1022 	 */
   1023 	if (nm == CTL_CREATE) {
   1024 		nm = ++sysctl_root.sysctl_num;
   1025 		for (ni = 0; ni < pnode->sysctl_clen; ni++) {
   1026 			if (nm == node[ni].sysctl_num) {
   1027 				nm++;
   1028 				ni = -1;
   1029 			}
   1030 			else if (nm > node[ni].sysctl_num)
   1031 				at = ni + 1;
   1032 		}
   1033 	}
   1034 
   1035 	/*
   1036 	 * oops...ran out of space
   1037 	 */
   1038 	if (pnode->sysctl_clen == pnode->sysctl_csize) {
   1039 		error = sysctl_realloc(pnode);
   1040 		if (error)
   1041 			return (error);
   1042 		node = pnode->sysctl_child;
   1043 	}
   1044 
   1045 	/*
   1046 	 * insert new node data
   1047 	 */
   1048 	if (at < pnode->sysctl_clen) {
   1049 		int t;
   1050 
   1051 		/*
   1052 		 * move the nodes that should come after the new one
   1053 		 */
   1054 		memmove(&node[at + 1], &node[at],
   1055 			(pnode->sysctl_clen - at) * sizeof(struct sysctlnode));
   1056 		memset(&node[at], 0, sizeof(struct sysctlnode));
   1057 		node[at].sysctl_parent = pnode;
   1058 		/*
   1059 		 * and...reparent any children of any moved nodes
   1060 		 */
   1061 		for (ni = at; ni <= pnode->sysctl_clen; ni++)
   1062 			if (SYSCTL_TYPE(node[ni].sysctl_flags) == CTLTYPE_NODE)
   1063 				for (t = 0; t < node[ni].sysctl_clen; t++)
   1064 					node[ni].sysctl_child[t].sysctl_parent =
   1065 						&node[ni];
   1066 	}
   1067 	node = &node[at];
   1068 	pnode->sysctl_clen++;
   1069 
   1070 	strlcpy(node->sysctl_name, nnode.sysctl_name,
   1071 		sizeof(node->sysctl_name));
   1072 	node->sysctl_num = nm;
   1073 	node->sysctl_size = sz;
   1074 	node->sysctl_flags = type|flags;
   1075 	node->sysctl_csize = 0;
   1076 	node->sysctl_clen = 0;
   1077 	if (own) {
   1078 		node->sysctl_data = own;
   1079 		node->sysctl_flags |= SYSCTL_OWNDATA;
   1080 	}
   1081 	else if (flags & SYSCTL_ALIAS) {
   1082 		node->sysctl_alias = anum;
   1083 	}
   1084 	else if (flags & SYSCTL_IMMEDIATE) {
   1085 		switch (type) {
   1086 		case CTLTYPE_INT:
   1087 			node->sysctl_idata = nnode.sysctl_idata;
   1088 			break;
   1089 		case CTLTYPE_QUAD:
   1090 			node->sysctl_qdata = nnode.sysctl_qdata;
   1091 			break;
   1092 		}
   1093 	}
   1094 	else {
   1095 		node->sysctl_data = nnode.sysctl_data;
   1096 		node->sysctl_flags &= ~SYSCTL_OWNDATA;
   1097 	}
   1098         node->sysctl_func = nnode.sysctl_func;
   1099         node->sysctl_child = NULL;
   1100 	/* node->sysctl_parent should already be done */
   1101 
   1102 	/*
   1103 	 * update "version" on path to "root"
   1104 	 */
   1105 	for (; rnode->sysctl_parent != NULL; rnode = rnode->sysctl_parent)
   1106 		;
   1107 	pnode = node;
   1108 	for (nm = rnode->sysctl_ver + 1; pnode != NULL;
   1109 	     pnode = pnode->sysctl_parent)
   1110 		pnode->sysctl_ver = nm;
   1111 
   1112 	if (oldp != NULL)
   1113 		error = sysctl_copyout(l, node, oldp,
   1114 		    MIN(*oldlenp, sizeof(struct sysctlnode)));
   1115 	*oldlenp = sizeof(struct sysctlnode);
   1116 
   1117 	return (error);
   1118 }
   1119 
   1120 /*
   1121  * ********************************************************************
   1122  * A wrapper around sysctl_create() that prints the thing we're trying
   1123  * to add.
   1124  * ********************************************************************
   1125  */
   1126 #ifdef SYSCTL_DEBUG_CREATE
   1127 int _sysctl_create(SYSCTLFN_RWPROTO);
   1128 int
   1129 _sysctl_create(SYSCTLFN_RWARGS)
   1130 {
   1131 	const struct sysctlnode *node;
   1132 	int k, rc, ni, nl = namelen + (name - oname);
   1133 
   1134 	node = newp;
   1135 
   1136 	printf("namelen %d (", nl);
   1137 	for (ni = 0; ni < nl - 1; ni++)
   1138 		printf(" %d", oname[ni]);
   1139 	printf(" %d )\t[%s]\tflags %08x (%08x %d %zu)\n",
   1140 	       k = node->sysctl_num,
   1141 	       node->sysctl_name,
   1142 	       node->sysctl_flags,
   1143 	       SYSCTL_FLAGS(node->sysctl_flags),
   1144 	       SYSCTL_TYPE(node->sysctl_flags),
   1145 	       node->sysctl_size);
   1146 
   1147 	node = rnode;
   1148 	rc = sysctl_create(SYSCTLFN_CALL(rnode));
   1149 
   1150 	printf("sysctl_create(");
   1151 	for (ni = 0; ni < nl - 1; ni++)
   1152 		printf(" %d", oname[ni]);
   1153 	printf(" %d ) returned %d\n", k, rc);
   1154 
   1155 	return (rc);
   1156 }
   1157 #define sysctl_create _sysctl_create
   1158 #endif /* SYSCTL_DEBUG_CREATE */
   1159 
   1160 /*
   1161  * sysctl_destroy -- Removes a node (as described by newp) from the
   1162  * given tree, returning (if successful) a copy of the dead node in
   1163  * oldp.  Since we're removing stuff, there's not much to check.
   1164  */
   1165 int
   1166 sysctl_destroy(SYSCTLFN_RWARGS)
   1167 {
   1168 	struct sysctlnode *node, *pnode, onode, nnode;
   1169 	int ni, error;
   1170 
   1171 	error = 0;
   1172 
   1173 	if (namelen != 1 || name[namelen - 1] != CTL_DESTROY)
   1174 		return (EINVAL);
   1175 
   1176 	/*
   1177 	 * processes can only destroy nodes at securelevel 0, must be
   1178 	 * root, and can't remove nodes from a parent that's not
   1179 	 * writeable
   1180 	 */
   1181 	if (l != NULL) {
   1182 		if (securelevel > 0)
   1183 			return (EPERM);
   1184 		error = suser(l->l_proc->p_ucred, &l->l_proc->p_acflag);
   1185 		if (error)
   1186 			return (error);
   1187 		if (!(rnode->sysctl_flags & SYSCTL_READWRITE))
   1188 			return (EPERM);
   1189 	}
   1190 
   1191 	/*
   1192 	 * nothing can remove a node if:
   1193 	 * the node is permanent (checked later) or
   1194 	 * the tree itself is not writeable or
   1195 	 * the entire sysctl system is not writeable
   1196 	 */
   1197 	if (!(sysctl_rootof(rnode)->sysctl_flags & SYSCTL_READWRITE) ||
   1198 	    !(sysctl_root.sysctl_flags & SYSCTL_READWRITE))
   1199 		return (EPERM);
   1200 
   1201 	if (newp == NULL || newlen != sizeof(struct sysctlnode))
   1202 		return (EINVAL);
   1203 	error = sysctl_copyin(l, newp, &nnode, sizeof(struct sysctlnode));
   1204 	if (error)
   1205 		return (error);
   1206 	memset(&onode, 0, sizeof(struct sysctlnode));
   1207 
   1208 	node = rnode->sysctl_child;
   1209 	for (ni = 0; ni < rnode->sysctl_clen; ni++) {
   1210 		if (nnode.sysctl_num == node[ni].sysctl_num) {
   1211 			/*
   1212 			 * if name specified, must match
   1213 			 */
   1214 			if (nnode.sysctl_name[0] != '\0' &&
   1215 			    strcmp(nnode.sysctl_name, node[ni].sysctl_name))
   1216 				continue;
   1217 			/*
   1218 			 * if version specified, must match
   1219 			 */
   1220 			if (nnode.sysctl_ver != 0 &&
   1221 			    nnode.sysctl_ver != node[ni].sysctl_ver)
   1222 				continue;
   1223 			/*
   1224 			 * this must be the one
   1225 			 */
   1226 			break;
   1227 		}
   1228 	}
   1229 	if (ni == rnode->sysctl_clen)
   1230 		return (ENOENT);
   1231 	node = &node[ni];
   1232 	pnode = node->sysctl_parent;
   1233 
   1234 	/*
   1235 	 * if the kernel says permanent, it is, so there.  nyah.
   1236 	 */
   1237 	if (SYSCTL_FLAGS(node->sysctl_flags) & SYSCTL_PERMANENT)
   1238 		return (EPERM);
   1239 
   1240 	/*
   1241 	 * can't delete non-empty nodes
   1242 	 */
   1243 	if (SYSCTL_TYPE(node->sysctl_flags) == CTLTYPE_NODE &&
   1244 	    node->sysctl_clen != 0)
   1245 		return (ENOTEMPTY);
   1246 
   1247 	/*
   1248 	 * if the node "owns" data, release it now
   1249 	 */
   1250 	if (node->sysctl_flags & SYSCTL_OWNDATA) {
   1251 		if (node->sysctl_data != NULL)
   1252 			FREE(node->sysctl_data, M_SYSCTLDATA);
   1253 		node->sysctl_data = NULL;
   1254 	}
   1255 
   1256 	/*
   1257 	 * if the node to be removed is not the last one on the list,
   1258 	 * move the remaining nodes up, and reparent any grandchildren
   1259 	 */
   1260 	onode = *node;
   1261 	if (ni < pnode->sysctl_clen - 1) {
   1262 		int t;
   1263 
   1264 		memmove(&pnode->sysctl_child[ni], &pnode->sysctl_child[ni + 1],
   1265 			(pnode->sysctl_clen - ni - 1) *
   1266 			sizeof(struct sysctlnode));
   1267 		for (; ni < pnode->sysctl_clen - 1; ni++)
   1268 			if (SYSCTL_TYPE(pnode->sysctl_child[ni].sysctl_flags) ==
   1269 			    CTLTYPE_NODE)
   1270 				for (t = 0; t < pnode->sysctl_child[ni].sysctl_clen;
   1271 				     t++)
   1272 					pnode->sysctl_child[ni].sysctl_child[t].
   1273 						sysctl_parent =
   1274 						&pnode->sysctl_child[ni];
   1275 		ni = pnode->sysctl_clen - 1;
   1276 		node = &pnode->sysctl_child[ni];
   1277 	}
   1278 
   1279 	/*
   1280 	 * reset the space we just vacated
   1281 	 */
   1282 	memset(node, 0, sizeof(struct sysctlnode));
   1283 	node->sysctl_parent = pnode;
   1284 	pnode->sysctl_clen--;
   1285 
   1286 	/*
   1287 	 * if this parent just lost its last child, nuke the creche
   1288 	 */
   1289 	if (pnode->sysctl_clen == 0) {
   1290 		FREE(pnode->sysctl_child, M_SYSCTLNODE);
   1291 		pnode->sysctl_csize = 0;
   1292 		pnode->sysctl_child = NULL;
   1293 	}
   1294 
   1295 	/*
   1296 	 * update "version" on path to "root"
   1297 	 */
   1298         for (; rnode->sysctl_parent != NULL; rnode = rnode->sysctl_parent)
   1299                 ;
   1300 	for (ni = rnode->sysctl_ver + 1; pnode != NULL;
   1301 	     pnode = pnode->sysctl_parent)
   1302 		pnode->sysctl_ver = ni;
   1303 
   1304 	if (oldp != NULL)
   1305 		error = sysctl_copyout(l, &onode, oldp,
   1306 		    MIN(*oldlenp, sizeof(struct sysctlnode)));
   1307 	*oldlenp = sizeof(struct sysctlnode);
   1308 
   1309 	return (error);
   1310 }
   1311 
   1312 /*
   1313  * sysctl_lookup -- Handles copyin/copyout of new and old values.
   1314  * Partial reads are globally allowed.  Only root can write to things
   1315  * unless the node says otherwise.
   1316  */
   1317 int
   1318 sysctl_lookup(SYSCTLFN_RWARGS)
   1319 {
   1320 	struct proc *p = l->l_proc;
   1321 	int error, rw;
   1322 	size_t sz, len;
   1323 	void *d;
   1324 
   1325 	error = 0;
   1326 
   1327 	/*
   1328 	 * you can't "look up" a node.  you can "query" it, but you
   1329 	 * can't "look it up".
   1330 	 */
   1331 	if (SYSCTL_TYPE(rnode->sysctl_flags) == CTLTYPE_NODE || namelen != 0)
   1332 		return (EINVAL);
   1333 
   1334 	/*
   1335 	 * some nodes are private, so only root can look into them.
   1336 	 */
   1337 	if ((rnode->sysctl_flags & SYSCTL_PRIVATE) &&
   1338 	    (error = suser(p->p_ucred, &p->p_acflag)) != 0)
   1339 		return (error);
   1340 
   1341 	/*
   1342 	 * if a node wants to be writable according to different rules
   1343 	 * other than "only root can write to stuff unless a flag is
   1344 	 * set", then it needs its own function which should have been
   1345 	 * called and not us.
   1346 	 */
   1347 	if (l != NULL && newp != NULL &&
   1348 	    !(rnode->sysctl_flags & SYSCTL_ANYWRITE) &&
   1349 	    (error = suser(l->l_proc->p_ucred, &l->l_proc->p_acflag)) != 0)
   1350 		return (error);
   1351 
   1352 	/*
   1353 	 * is this node supposedly writable?
   1354 	 */
   1355 	rw = 0;
   1356 	switch (rnode->sysctl_flags & SYSCTL_READWRITE) {
   1357 	    case SYSCTL_READONLY1:
   1358 		rw = (securelevel < 1) ? 1 : 0;
   1359 		break;
   1360 	    case SYSCTL_READONLY2:
   1361 		rw = (securelevel < 2) ? 1 : 0;
   1362 		break;
   1363 	    case SYSCTL_READWRITE:
   1364 		rw = 1;
   1365 		break;
   1366 	}
   1367 
   1368 	/*
   1369 	 * it appears not to be writable at this time, so if someone
   1370 	 * tried to write to it, we must tell them to go away
   1371 	 */
   1372 	if (!rw && newp != NULL)
   1373 		return (EPERM);
   1374 
   1375 	/*
   1376 	 * step one, copy out the stuff we have presently
   1377 	 */
   1378 	if (rnode->sysctl_flags & SYSCTL_IMMEDIATE) {
   1379 		switch (SYSCTL_TYPE(rnode->sysctl_flags)) {
   1380 		case CTLTYPE_INT:
   1381 			d = &rnode->sysctl_idata;
   1382 			break;
   1383 		case CTLTYPE_QUAD:
   1384 			d = &rnode->sysctl_qdata;
   1385 			break;
   1386 		default:
   1387 			return (EINVAL);
   1388 		}
   1389 	}
   1390 	else
   1391 		d = rnode->sysctl_data;
   1392 	if (SYSCTL_TYPE(rnode->sysctl_flags) == CTLTYPE_STRING)
   1393 		sz = strlen(d) + 1;
   1394 	else
   1395 		sz = rnode->sysctl_size;
   1396 	if (oldp != NULL)
   1397 		error = sysctl_copyout(l, d, oldp, MIN(sz, *oldlenp));
   1398 	if (error)
   1399 		return (error);
   1400 	*oldlenp = sz;
   1401 
   1402 	/*
   1403 	 * are we done?
   1404 	 */
   1405 	if (newp == NULL || newlen == 0)
   1406 		return (0);
   1407 
   1408 	/*
   1409 	 * hmm...not done.  must now "copy in" new value.  re-adjust
   1410 	 * sz to maximum value (strings are "weird").
   1411 	 */
   1412 	sz = rnode->sysctl_size;
   1413 	switch (SYSCTL_TYPE(rnode->sysctl_flags)) {
   1414 	case CTLTYPE_INT:
   1415 	case CTLTYPE_QUAD:
   1416 	case CTLTYPE_STRUCT:
   1417 		/*
   1418 		 * these data must be *exactly* the same size coming
   1419 		 * in.
   1420 		 */
   1421 		if (newlen != sz)
   1422 			return (EINVAL);
   1423 		error = sysctl_copyin(l, newp, d, sz);
   1424 		break;
   1425 	case CTLTYPE_STRING: {
   1426 		/*
   1427 		 * strings, on the other hand, can be shorter, and we
   1428 		 * let userland be sloppy about the trailing nul.
   1429 		 */
   1430 		char *newbuf;
   1431 
   1432 		/*
   1433 		 * too much new string?
   1434 		 */
   1435 		if (newlen > sz)
   1436 			return (EINVAL);
   1437 
   1438 		/*
   1439 		 * temporary copy of new inbound string
   1440 		 */
   1441 		len = MIN(sz, newlen);
   1442 		newbuf = malloc(len, M_SYSCTLDATA, M_WAITOK|M_CANFAIL);
   1443 		if (newbuf == NULL)
   1444 			return (ENOMEM);
   1445 		error = sysctl_copyin(l, newp, newbuf, len);
   1446 		if (error) {
   1447 			FREE(newbuf, M_SYSCTLDATA);
   1448 			return (error);
   1449 		}
   1450 
   1451 		/*
   1452 		 * did they null terminate it, or do we have space
   1453 		 * left to do it ourselves?
   1454 		 */
   1455 		if (newbuf[len - 1] != '\0' && len == sz) {
   1456 			FREE(newbuf, M_SYSCTLDATA);
   1457 			return (EINVAL);
   1458 		}
   1459 
   1460 		/*
   1461 		 * looks good, so pop it into place and zero the rest.
   1462 		 */
   1463 		if (len > 0)
   1464 			memcpy(rnode->sysctl_data, newbuf, len);
   1465 		if (sz != len)
   1466 			memset((char*)rnode->sysctl_data + len, 0, sz - len);
   1467 		FREE(newbuf, M_SYSCTLDATA);
   1468 		break;
   1469 	}
   1470 	default:
   1471 		return (EINVAL);
   1472 	}
   1473 
   1474 	return (error);
   1475 }
   1476 
   1477 /*
   1478  * sysctl_mmap -- Dispatches sysctl mmap requests to those nodes that
   1479  * purport to handle it.  This interface isn't fully fleshed out yet,
   1480  * unfortunately.
   1481  */
   1482 static int
   1483 sysctl_mmap(SYSCTLFN_RWARGS)
   1484 {
   1485 	struct sysctlnode nnode, *node;
   1486 	int error;
   1487 
   1488 	/*
   1489 	 * let's just pretend that didn't happen, m'kay?
   1490 	 */
   1491 	if (l == NULL)
   1492 		return (EPERM);
   1493 
   1494 	/*
   1495 	 * is this a sysctlnode description of an mmap request?
   1496 	 */
   1497 	if (newp == NULL || newlen != sizeof(struct sysctlnode))
   1498 		return (EINVAL);
   1499 	error = sysctl_copyin(l, newp, &nnode, sizeof(struct sysctlnode));
   1500 	if (error)
   1501 		return (error);
   1502 
   1503 	/*
   1504 	 * does the node they asked for exist?
   1505 	 */
   1506 	if (namelen != 1)
   1507 		return (EOPNOTSUPP);
   1508 	node = rnode;
   1509         error = sysctl_locate(l, &nnode.sysctl_num, 1, &node, NULL);
   1510 	if (error)
   1511 		return (error);
   1512 
   1513 	/*
   1514 	 * does this node that we have found purport to handle mmap?
   1515 	 */
   1516 	if (node->sysctl_func == NULL ||
   1517 	    !(node->sysctl_flags & SYSCTL_MMAP))
   1518 		return (EOPNOTSUPP);
   1519 
   1520 	/*
   1521 	 * well...okay, they asked for it.
   1522 	 */
   1523 	return ((*node->sysctl_func)(SYSCTLFN_CALL(node)));
   1524 }
   1525 
   1526 /*
   1527  * ********************************************************************
   1528  * Section 3: Create and destroy from inside the kernel
   1529  * ********************************************************************
   1530  * sysctl_createv() and sysctl_destroyv() are simpler-to-use
   1531  * interfaces for the kernel to fling new entries into the mib and rip
   1532  * them out later.  In the case of sysctl_createv(), the returned copy
   1533  * of the node (see sysctl_create()) will be translated back into a
   1534  * pointer to the actual node.
   1535  *
   1536  * Note that sysctl_createv() will return 0 if the create request
   1537  * matches an existing node (ala mkdir -p), and that sysctl_destroyv()
   1538  * will return 0 if the node to be destroyed already does not exist
   1539  * (aka rm -f) or if it is a parent of other nodes.
   1540  *
   1541  * This allows two (or more) different subsystems to assert sub-tree
   1542  * existence before populating their own nodes, and to remove their
   1543  * own nodes without orphaning the others when they are done.
   1544  * ********************************************************************
   1545  */
   1546 int
   1547 sysctl_createv(int flags, int type,
   1548 	       const char *namep, struct sysctlnode **rnode,
   1549 	       sysctlfn func, u_quad_t qv, void *newp, size_t newlen,
   1550 	       ...)
   1551 {
   1552 	va_list ap;
   1553 	int error, ni, namelen, name[CTL_MAXNAME];
   1554 	struct sysctlnode *pnode, nnode, onode;
   1555 	size_t sz;
   1556 
   1557 	/*
   1558 	 * what is it?
   1559 	 */
   1560 	flags = SYSCTL_TYPE(type)|SYSCTL_FLAGS(flags);
   1561 
   1562 	/*
   1563 	 * where do we put it?
   1564 	 */
   1565 	va_start(ap, newlen);
   1566 	namelen = 0;
   1567 	ni = -1;
   1568 	do {
   1569 		if (++ni == CTL_MAXNAME)
   1570 			return (ENAMETOOLONG);
   1571 		name[ni] = va_arg(ap, int);
   1572 		/*
   1573 		 * sorry, this is not supported from here
   1574 		 */
   1575 		if (name[ni] == CTL_CREATESYM)
   1576 			return (EINVAL);
   1577 	} while (name[ni] != CTL_EOL && name[ni] != CTL_CREATE);
   1578 	namelen = ni + (name[ni] == CTL_CREATE ? 1 : 0);
   1579 	va_end(ap);
   1580 
   1581 	/*
   1582 	 * what's it called
   1583 	 */
   1584 	if (strlcpy(nnode.sysctl_name, namep, sizeof(nnode.sysctl_name)) >
   1585 	    sizeof(nnode.sysctl_name))
   1586 		return (ENAMETOOLONG);
   1587 
   1588 	/*
   1589 	 * cons up the description of the new node
   1590 	 */
   1591 	nnode.sysctl_num = name[namelen - 1];
   1592 	name[namelen - 1] = CTL_CREATE;
   1593 	nnode.sysctl_size = newlen;
   1594 	nnode.sysctl_flags = flags;
   1595 	if (type == CTLTYPE_NODE) {
   1596 		nnode.sysctl_csize = 0;
   1597 		nnode.sysctl_clen = 0;
   1598 		nnode.sysctl_child = NULL;
   1599 		if (flags & SYSCTL_ALIAS)
   1600 			nnode.sysctl_alias = qv;
   1601 	}
   1602 	else if (flags & SYSCTL_IMMEDIATE) {
   1603 		switch (type) {
   1604 		case CTLTYPE_INT:
   1605 			nnode.sysctl_idata = qv;
   1606 			break;
   1607 		case CTLTYPE_QUAD:
   1608 			nnode.sysctl_qdata = qv;
   1609 			break;
   1610 		default:
   1611 			return (EINVAL);
   1612 		}
   1613 	}
   1614 	else {
   1615 		nnode.sysctl_data = newp;
   1616 	}
   1617 	nnode.sysctl_func = func;
   1618 	nnode.sysctl_parent = NULL;
   1619 	nnode.sysctl_ver = 0;
   1620 
   1621 	/*
   1622 	 * initialize lock state -- we need locks if the main tree has
   1623 	 * been marked as complete, but since we could be called from
   1624 	 * either there, or from a device driver (say, at device
   1625 	 * insertion), or from an lkm (at lkm load time, say), we
   1626 	 * don't really want to "wait"...
   1627 	 */
   1628 	error = sysctl_lock(NULL, NULL, 0);
   1629 	if (error)
   1630 		return (error);
   1631 
   1632 	/*
   1633 	 * locate the prospective parent of the new node, and if we
   1634 	 * find it, add the new node.
   1635 	 */
   1636 	sz = sizeof(onode);
   1637 	pnode = (rnode != NULL) ? *rnode : NULL;
   1638 	error = sysctl_locate(NULL, &name[0], namelen - 1, &pnode, &ni);
   1639 	if (error == 0)
   1640 		error = sysctl_create(&name[ni], namelen - ni, &onode, &sz,
   1641 				      &nnode, sizeof(nnode), &name[0], NULL,
   1642 				      pnode);
   1643 
   1644 	/*
   1645 	 * unfortunately the node we wanted to create is already
   1646 	 * there.  if the node that's already there is a reasonable
   1647 	 * facsimile of the node we wanted to create, just pretend
   1648 	 * (for the caller's benefit) that we managed to create the
   1649 	 * node they wanted.
   1650 	 */
   1651 	if (error == EEXIST) {
   1652 		/* name is the same as requested... */
   1653 		if (strcmp(nnode.sysctl_name, onode.sysctl_name) == 0 &&
   1654 		    /* they want the same function... */
   1655 		    nnode.sysctl_func == onode.sysctl_func &&
   1656 		    /* number is the same as requested, or... */
   1657 		    (nnode.sysctl_num == onode.sysctl_num ||
   1658 		     /* they didn't pick a number... */
   1659 		     nnode.sysctl_num == CTL_CREATE)) {
   1660 			/*
   1661 			 * collision here from trying to create
   1662 			 * something that already existed; let's give
   1663 			 * our customers a hand and tell them they got
   1664 			 * what they wanted.
   1665 			 */
   1666 #ifdef SYSCTL_DEBUG_CREATE
   1667 			printf("cleared\n");
   1668 #endif /* SYSCTL_DEBUG_CREATE */
   1669 			error = 0;
   1670 		}
   1671 	}
   1672 
   1673 	/*
   1674 	 * if they want to know where the new node is, go find the
   1675 	 * address of the actual node, not the copy that
   1676 	 * sysctl_create() gave us.
   1677 	 */
   1678 	if (rnode != NULL && error == 0) {
   1679 		/*
   1680 		 * sysctl_create() gave us back a copy of the node,
   1681 		 * but we need to know where it actually is...
   1682 		 */
   1683 		name[namelen - 1] = onode.sysctl_num;
   1684 		pnode = *rnode;
   1685 		error = sysctl_locate(NULL, &name[0], namelen, &pnode, &ni);
   1686 		/*
   1687 		 * not expecting an error here, but...
   1688 		 */
   1689 		if (error == 0)
   1690 			*rnode = pnode;
   1691 	}
   1692 
   1693 	/*
   1694 	 * now it should be safe to release the lock state.
   1695 	 */
   1696 	sysctl_unlock(NULL);
   1697 
   1698 	if (error != 0) {
   1699 		printf("sysctl_createv: sysctl_create(%s) returned %d\n",
   1700 		       nnode.sysctl_name, error);
   1701 #if 0
   1702 		if (error != ENOENT)
   1703 			sysctl_dump(&onode);
   1704 #endif
   1705 	}
   1706 
   1707 	return (error);
   1708 }
   1709 
   1710 int
   1711 sysctl_destroyv(struct sysctlnode *rnode, ...)
   1712 {
   1713 	va_list ap;
   1714 	int error, name[CTL_MAXNAME], namelen, ni;
   1715 	struct sysctlnode *pnode, *node;
   1716 
   1717 	va_start(ap, rnode);
   1718 	namelen = 0;
   1719 	ni = 0;
   1720 	do {
   1721 		if (ni == CTL_MAXNAME)
   1722 			return (ENAMETOOLONG);
   1723 		name[ni] = va_arg(ap, int);
   1724 	} while (name[ni++] != CTL_EOL);
   1725 	namelen = ni - 1;
   1726 	va_end(ap);
   1727 
   1728 	/*
   1729 	 * i can't imagine why we'd be destroying a node when the tree
   1730 	 * wasn't complete, but who knows?
   1731 	 */
   1732 	error = sysctl_lock(NULL, NULL, 0);
   1733 	if (error)
   1734 		return (error);
   1735 
   1736 	/*
   1737 	 * where is it?
   1738 	 */
   1739 	node = rnode;
   1740         error = sysctl_locate(NULL, &name[0], namelen, &node, &ni);
   1741 	if (error) {
   1742 		/* they want it gone and it's not there, so... */
   1743 		sysctl_unlock(NULL);
   1744 		return (error == ENOENT ? 0 : error);
   1745 	}
   1746 
   1747 	/*
   1748 	 * we found it, now let's nuke it
   1749 	 */
   1750 	name[namelen - 1] = CTL_DESTROY;
   1751 	pnode = node->sysctl_parent;
   1752 	error = sysctl_destroy(&name[namelen - 1], 1, NULL, NULL,
   1753 			       node, sizeof(*node), &name[0], NULL,
   1754 			       pnode);
   1755 	if (error == ENOTEMPTY)
   1756 		/*
   1757 		 * think of trying to delete "foo" when "foo.bar"
   1758 		 * (which someone else put there) is still in
   1759 		 * existence
   1760 		 */
   1761 		error = 0;
   1762 
   1763         sysctl_unlock(NULL);
   1764 
   1765 	return (error);
   1766 }
   1767 
   1768 #if 0
   1769 /*
   1770  * ********************************************************************
   1771  * the dump routine.  i haven't yet decided how (if at all) i'll call
   1772  * this from userland when it's in the kernel.
   1773  * ********************************************************************
   1774  */
   1775 static const char *
   1776 sf(int f)
   1777 {
   1778 	static char s[256];
   1779 	char *c;
   1780 
   1781 	s[0] = '\0';
   1782 	c = "";
   1783 
   1784 #define print_flag(_f, _s, _c, _q, _x) \
   1785 	if (((_x) && (((_f) & (_x)) == (__CONCAT(SYSCTL_,_q)))) || \
   1786 	    (!(_x) && ((_f) & (__CONCAT(SYSCTL_,_q))))) { \
   1787 		strlcat((_s), (_c), sizeof(_s)); \
   1788 		strlcat((_s), __STRING(_q), sizeof(_s)); \
   1789 		(_c) = ","; \
   1790 		(_f) &= ~(__CONCAT(SYSCTL_,_q)|(_x)); \
   1791 	}
   1792 	print_flag(f, s, c, READONLY, SYSCTL_READWRITE);
   1793 	print_flag(f, s, c, READONLY1, SYSCTL_READWRITE);
   1794 	print_flag(f, s, c, READONLY2, SYSCTL_READWRITE);
   1795 	print_flag(f, s, c, READWRITE, SYSCTL_READWRITE);
   1796 	print_flag(f, s, c, ANYWRITE, 0);
   1797 	print_flag(f, s, c, PRIVATE, 0);
   1798 	print_flag(f, s, c, PERMANENT, 0);
   1799 	print_flag(f, s, c, OWNDATA, 0);
   1800 	print_flag(f, s, c, IMMEDIATE, 0);
   1801 	print_flag(f, s, c, HEX, 0);
   1802 	print_flag(f, s, c, ROOT, 0);
   1803 	print_flag(f, s, c, ANYNUMBER, 0);
   1804 	print_flag(f, s, c, HIDDEN, 0);
   1805 	print_flag(f, s, c, ALIAS, 0);
   1806 #undef print_flag
   1807 
   1808 	if (f) {
   1809 		char foo[9];
   1810 		snprintf(foo, sizeof(foo), "%x", f);
   1811 		strlcat(s, c, sizeof(s));
   1812 		strlcat(s, foo, sizeof(s));
   1813 	}
   1814 
   1815 	return (s);
   1816 }
   1817 
   1818 static const char *
   1819 st(int t)
   1820 {
   1821 
   1822 	switch (t) {
   1823 	case CTLTYPE_NODE:
   1824 		return "NODE";
   1825 	case CTLTYPE_INT:
   1826 		return "INT";
   1827 	case CTLTYPE_STRING:
   1828 		return "STRING";
   1829 	case CTLTYPE_QUAD:
   1830 		return "QUAD";
   1831 	case CTLTYPE_STRUCT:
   1832 		return "STRUCT";
   1833 	}
   1834 
   1835 	return "???";
   1836 }
   1837 
   1838 void
   1839 sysctl_dump(const struct sysctlnode *d)
   1840 {
   1841 	static char nmib[64], smib[256];
   1842 	static int indent;
   1843 	struct sysctlnode *n;
   1844 	char *np, *sp, tmp[20];
   1845 	int i;
   1846 
   1847 	if (d == NULL)
   1848 		return;
   1849 
   1850 	np = &nmib[strlen(nmib)];
   1851 	sp = &smib[strlen(smib)];
   1852 
   1853 	if (!(d->sysctl_flags & SYSCTL_ROOT)) {
   1854 		snprintf(tmp, sizeof(tmp), "%d", d->sysctl_num);
   1855 		strcat(nmib, ".");
   1856 		strcat(smib, ".");
   1857 		strcat(nmib, tmp);
   1858 		strcat(smib, d->sysctl_name);
   1859 		printf("%s -> %s (%d)\n", &nmib[1], &smib[1],
   1860 		       SYSCTL_TYPE(d->sysctl_flags));
   1861 	}
   1862 
   1863 	if (1) {
   1864 		printf("%*s%p:\tsysctl_name  [%s]\n", indent, "",
   1865 		       d, d->sysctl_name);
   1866 		printf("%*s\t\tsysctl_num    %d\n",   indent, "",
   1867 		       d->sysctl_num);
   1868 		printf("%*s\t\tsysctl_flags  %x (flags=%x<%s> type=%d<%s> "
   1869 		       "size=%zu)\n",
   1870 		       indent, "", d->sysctl_flags,
   1871 		       SYSCTL_FLAGS(d->sysctl_flags),
   1872 		       sf(SYSCTL_FLAGS(d->sysctl_flags)),
   1873 		       SYSCTL_TYPE(d->sysctl_flags),
   1874 		       st(SYSCTL_TYPE(d->sysctl_flags)),
   1875 		       d->sysctl_size);
   1876 		if (SYSCTL_TYPE(d->sysctl_flags) == CTLTYPE_NODE) {
   1877 			printf("%*s\t\tsysctl_csize  %d\n",   indent, "",
   1878 			       d->sysctl_csize);
   1879 			printf("%*s\t\tsysctl_clen   %d\n",   indent, "",
   1880 			       d->sysctl_clen);
   1881 			printf("%*s\t\tsysctl_child  %p\n",   indent, "",
   1882 			       d->sysctl_child);
   1883 		}
   1884 		else
   1885 			printf("%*s\t\tsysctl_data   %p\n",   indent, "",
   1886 			       d->sysctl_data);
   1887 		printf("%*s\t\tsysctl_func   %p\n",   indent, "",
   1888 		       d->sysctl_func);
   1889 		printf("%*s\t\tsysctl_parent %p\n",   indent, "",
   1890 		       d->sysctl_parent);
   1891 		printf("%*s\t\tsysctl_ver    %d\n",   indent, "",
   1892 		       d->sysctl_ver);
   1893 	}
   1894 
   1895 	if (SYSCTL_TYPE(d->sysctl_flags) == CTLTYPE_NODE) {
   1896 		indent += 8;
   1897 		n = d->sysctl_child;
   1898 		for (i = 0; i < d->sysctl_clen; i++) {
   1899 			sysctl_dump(&n[i]);
   1900 		}
   1901 		indent -= 8;
   1902 	}
   1903 
   1904 	np[0] = '\0';
   1905 	sp[0] = '\0';
   1906 }
   1907 #endif /* 0 */
   1908 
   1909 /*
   1910  * ********************************************************************
   1911  * Deletes an entire n-ary tree.  Not recommended unless you know why
   1912  * you're doing it.  Personally, I don't know why you'd even think
   1913  * about it.
   1914  * ********************************************************************
   1915  */
   1916 void
   1917 sysctl_free(struct sysctlnode *rnode)
   1918 {
   1919 	struct sysctlnode *node, *pnode;
   1920 
   1921 	if (rnode == NULL)
   1922 		rnode = &sysctl_root;
   1923 	pnode = rnode;
   1924 
   1925 	node = pnode->sysctl_child;
   1926 	do {
   1927 		while (node != NULL && pnode->sysctl_csize > 0) {
   1928 			while (node <
   1929 			       &pnode->sysctl_child[pnode->sysctl_clen] &&
   1930 			       (SYSCTL_TYPE(node->sysctl_flags) !=
   1931 				CTLTYPE_NODE ||
   1932 				node->sysctl_csize == 0)) {
   1933 				if (SYSCTL_FLAGS(node->sysctl_flags) &
   1934 				    SYSCTL_OWNDATA) {
   1935 					if (node->sysctl_data != NULL) {
   1936 						FREE(node->sysctl_data,
   1937 						     M_SYSCTLDATA);
   1938 						node->sysctl_data = NULL;
   1939 					}
   1940 				}
   1941 				node++;
   1942 			}
   1943 			if (node < &pnode->sysctl_child[pnode->sysctl_clen]) {
   1944 				pnode = node;
   1945 				node = node->sysctl_child;
   1946 			}
   1947 			else
   1948 				break;
   1949 		}
   1950 		if (pnode->sysctl_child != NULL)
   1951 			FREE(pnode->sysctl_child, M_SYSCTLNODE);
   1952 		pnode->sysctl_clen = 0;
   1953 		pnode->sysctl_csize = 0;
   1954 		pnode->sysctl_child = NULL;
   1955 		node = pnode;
   1956 		pnode = node->sysctl_parent;
   1957 	} while (pnode != NULL && pnode != rnode);
   1958 }
   1959 
   1960 /*
   1961  * ********************************************************************
   1962  * old_sysctl -- A routine to bridge old-style internal calls to the
   1963  * new infrastructure.
   1964  * ********************************************************************
   1965  */
   1966 int
   1967 old_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp,
   1968 	   void *newp, size_t newlen, struct lwp *l)
   1969 {
   1970 	int error;
   1971 	size_t savelen = *oldlenp;
   1972 
   1973 	error = sysctl_lock(l, oldp, savelen);
   1974 	if (error)
   1975 		return (error);
   1976 	error = sysctl_dispatch(name, namelen, oldp, oldlenp,
   1977 				newp, newlen, name, l, NULL);
   1978 	sysctl_unlock(l);
   1979 	if (error == 0 && oldp != NULL && savelen < *oldlenp)
   1980 		error = ENOMEM;
   1981 
   1982 	return (error);
   1983 }
   1984 
   1985 /*
   1986  * ********************************************************************
   1987  * Section 4: Generic helper routines
   1988  * ********************************************************************
   1989  * "helper" routines that can do more finely grained access control,
   1990  * construct structures from disparate information, create the
   1991  * appearance of more nodes and sub-trees, etc.  for example, if
   1992  * CTL_PROC wanted a helper function, it could respond to a CTL_QUERY
   1993  * with a dynamically created list of nodes that represented the
   1994  * currently running processes at that instant.
   1995  * ********************************************************************
   1996  */
   1997 
   1998 /*
   1999  * first, a few generic helpers that provide:
   2000  *
   2001  * sysctl_needfunc()		a readonly interface that emits a warning
   2002  * sysctl_notavail()		returns EOPNOTSUPP (generic error)
   2003  * sysctl_null()		an empty return buffer with no error
   2004  */
   2005 int
   2006 sysctl_needfunc(SYSCTLFN_ARGS)
   2007 {
   2008 	int error;
   2009 
   2010 	printf("!!SYSCTL_NEEDFUNC!!\n");
   2011 
   2012 	if (newp != NULL || namelen != 0)
   2013 		return (EOPNOTSUPP);
   2014 
   2015 	error = 0;
   2016 	if (oldp != NULL)
   2017 		error = sysctl_copyout(l, rnode->sysctl_data, oldp,
   2018 				       MIN(rnode->sysctl_size, *oldlenp));
   2019 	*oldlenp = rnode->sysctl_size;
   2020 
   2021 	return (error);
   2022 }
   2023 
   2024 int
   2025 sysctl_notavail(SYSCTLFN_ARGS)
   2026 {
   2027 
   2028 	if (namelen == 1 && name[0] == CTL_QUERY)
   2029 		return (sysctl_query(SYSCTLFN_CALL(rnode)));
   2030 
   2031 	return (EOPNOTSUPP);
   2032 }
   2033 
   2034 int
   2035 sysctl_null(SYSCTLFN_ARGS)
   2036 {
   2037 
   2038 	*oldlenp = 0;
   2039 
   2040 	return (0);
   2041 }
   2042 
   2043 /*
   2044  * ********************************************************************
   2045  * Section 5: The machinery that makes it all go
   2046  * ********************************************************************
   2047  * Memory "manglement" routines.  Not much to this, eh?
   2048  * ********************************************************************
   2049  */
   2050 static int
   2051 sysctl_alloc(struct sysctlnode *p, int x)
   2052 {
   2053 	int i;
   2054 	struct sysctlnode *n;
   2055 
   2056 	assert(p->sysctl_child == NULL);
   2057 
   2058 	if (x == 1)
   2059 		MALLOC(n, struct sysctlnode *,
   2060 		       sizeof(struct sysctlnode),
   2061 		       M_SYSCTLNODE, M_WAITOK|M_CANFAIL);
   2062 	else
   2063 		MALLOC(n, struct sysctlnode *,
   2064 		       SYSCTL_DEFSIZE * sizeof(struct sysctlnode),
   2065 		       M_SYSCTLNODE, M_WAITOK|M_CANFAIL);
   2066 	if (n == NULL)
   2067 		return (ENOMEM);
   2068 
   2069 	if (x == 1) {
   2070 		memset(n, 0, sizeof(struct sysctlnode));
   2071 		p->sysctl_csize = 1;
   2072 	}
   2073 	else {
   2074 		memset(n, 0, SYSCTL_DEFSIZE * sizeof(struct sysctlnode));
   2075 		p->sysctl_csize = SYSCTL_DEFSIZE;
   2076 	}
   2077 	p->sysctl_clen = 0;
   2078 
   2079 	for (i = 0; i < p->sysctl_csize; i++)
   2080 		n[i].sysctl_parent = p;
   2081 
   2082 	p->sysctl_child = n;
   2083 	return (0);
   2084 }
   2085 
   2086 static int
   2087 sysctl_realloc(struct sysctlnode *p)
   2088 {
   2089 	int i, j;
   2090 	struct sysctlnode *n;
   2091 
   2092 	assert(p->sysctl_csize == p->sysctl_clen);
   2093 
   2094 	/*
   2095 	 * how many do we have...how many should we make?
   2096 	 */
   2097 	i = p->sysctl_clen;
   2098 	n = malloc(2 * i * sizeof(struct sysctlnode), M_SYSCTLNODE,
   2099 		   M_WAITOK|M_CANFAIL);
   2100 	if (n == NULL)
   2101 		return (ENOMEM);
   2102 
   2103 	/*
   2104 	 * move old children over...initialize new children
   2105 	 */
   2106 	memcpy(n, p->sysctl_child, i * sizeof(struct sysctlnode));
   2107 	memset(&n[i], 0, i * sizeof(struct sysctlnode));
   2108 	p->sysctl_csize = 2 * i;
   2109 	p->sysctl_clen = i;
   2110 
   2111 	/*
   2112 	 * reattach moved (and new) children to parent; if a moved
   2113 	 * child node has children, reattach the parent pointers of
   2114 	 * grandchildren
   2115 	 */
   2116         for (i = 0; i < p->sysctl_csize; i++) {
   2117                 n[i].sysctl_parent = p;
   2118 		if (n[i].sysctl_child != NULL) {
   2119 			for (j = 0; j < n[i].sysctl_csize; j++)
   2120 				n[i].sysctl_child[j].sysctl_parent = &n[i];
   2121 		}
   2122 	}
   2123 
   2124 	/*
   2125 	 * get out with the old and in with the new
   2126 	 */
   2127 	FREE(p->sysctl_child, M_SYSCTLNODE);
   2128 	p->sysctl_child = n;
   2129 
   2130 	return (0);
   2131 }
   2132