Home | History | Annotate | Line # | Download | only in kern
kern_sysctl.c revision 1.163
      1 /*	$NetBSD: kern_sysctl.c,v 1.163 2004/03/24 16:55:49 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 1.163 2004/03/24 16:55:49 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 static int sysctl_cvt_in(struct lwp *, int *, const void *, size_t,
    103 			 struct sysctlnode *);
    104 static int sysctl_cvt_out(struct lwp *, int, const struct sysctlnode *,
    105 			  void *, size_t, size_t *);
    106 
    107 static int sysctl_log_add(struct sysctllog **, struct sysctlnode *);
    108 static int sysctl_log_realloc(struct sysctllog *);
    109 
    110 struct sysctllog {
    111 	struct sysctlnode *log_root;
    112 	int *log_num;
    113 	int log_size, log_left;
    114 };
    115 
    116 /*
    117  * the "root" of the new sysctl tree
    118  */
    119 static struct sysctlnode sysctl_root = {
    120 	.sysctl_flags = SYSCTL_VERSION|
    121 	    CTLFLAG_ROOT|CTLFLAG_READWRITE|
    122 	    CTLTYPE_NODE,
    123 	.sysctl_num = 0,
    124 	.sysctl_size = sizeof(struct sysctlnode),
    125 	.sysctl_name = "(root)",
    126 };
    127 
    128 /*
    129  * link set of functions that add nodes at boot time (see also
    130  * sysctl_buildtree())
    131  */
    132 __link_set_decl(sysctl_funcs, sysctl_setup_func);
    133 
    134 /*
    135  * The `sysctl_lock' is intended to serialize access to the sysctl
    136  * tree.  Given that it is now (a) dynamic, and (b) most consumers of
    137  * sysctl are going to be copying data out, the old `sysctl_memlock'
    138  * has been `upgraded' to simply guard the whole tree.
    139  *
    140  * The two new data here are to keep track of the locked chunk of
    141  * memory, if there is one, so that it can be released more easily
    142  * from anywhere.
    143  */
    144 struct lock sysctl_treelock;
    145 caddr_t sysctl_memaddr;
    146 size_t sysctl_memsize;
    147 
    148 /*
    149  * Attributes stored in the kernel.
    150  */
    151 char hostname[MAXHOSTNAMELEN];
    152 int hostnamelen;
    153 
    154 char domainname[MAXHOSTNAMELEN];
    155 int domainnamelen;
    156 
    157 long hostid;
    158 
    159 #ifdef INSECURE
    160 int securelevel = -1;
    161 #else
    162 int securelevel = 0;
    163 #endif
    164 
    165 #ifndef DEFCORENAME
    166 #define	DEFCORENAME	"%n.core"
    167 #endif
    168 char defcorename[MAXPATHLEN] = DEFCORENAME;
    169 
    170 /*
    171  * ********************************************************************
    172  * Section 0: Some simple glue
    173  * ********************************************************************
    174  * By wrapping copyin(), copyout(), and copyinstr() like this, we can
    175  * stop caring about who's calling us and simplify some code a bunch.
    176  * ********************************************************************
    177  */
    178 static inline int
    179 sysctl_copyin(const struct lwp *l, const void *uaddr, void *kaddr, size_t len)
    180 {
    181 
    182 	if (l != NULL)
    183 		return (copyin(uaddr, kaddr, len));
    184 	else
    185 		return (kcopy(uaddr, kaddr, len));
    186 }
    187 
    188 static inline int
    189 sysctl_copyout(const struct lwp *l, const void *kaddr, void *uaddr, size_t len)
    190 {
    191 
    192 	if (l != NULL)
    193 		return (copyout(kaddr, uaddr, len));
    194 	else
    195 		return (kcopy(kaddr, uaddr, len));
    196 }
    197 
    198 static inline int
    199 sysctl_copyinstr(const struct lwp *l, const void *uaddr, void *kaddr,
    200 		 size_t len, size_t *done)
    201 {
    202 
    203 	if (l != NULL)
    204 		return (copyinstr(uaddr, kaddr, len, done));
    205 	else
    206 		return (copystr(uaddr, kaddr, len, done));
    207 }
    208 
    209 /*
    210  * ********************************************************************
    211  * Initialize sysctl subsystem.
    212  * ********************************************************************
    213  */
    214 void
    215 sysctl_init(void)
    216 {
    217 	sysctl_setup_func **sysctl_setup, f;
    218 
    219 	lockinit(&sysctl_treelock, PRIBIO|PCATCH, "sysctl", 0, 0);
    220 
    221 	/*
    222 	 * dynamic mib numbers start here
    223 	 */
    224 	sysctl_root.sysctl_num = CREATE_BASE;
    225 
    226         __link_set_foreach(sysctl_setup, sysctl_funcs) {
    227 		/*
    228 		 * XXX - why do i have to coerce the pointers like this?
    229 		 */
    230 		f = (void*)*sysctl_setup;
    231 		(*f)(NULL);
    232 	}
    233 
    234 	/*
    235 	 * setting this means no more permanent nodes can be added,
    236 	 * trees that claim to be readonly at the root now are, and if
    237 	 * the main tree is readonly, *everything* is.
    238 	 */
    239 	sysctl_root.sysctl_flags |= CTLFLAG_PERMANENT;
    240 
    241 }
    242 
    243 /*
    244  * ********************************************************************
    245  * The main native sysctl system call itself.
    246  * ********************************************************************
    247  */
    248 int
    249 sys___sysctl(struct lwp *l, void *v, register_t *retval)
    250 {
    251 	struct sys___sysctl_args /* {
    252 		syscallarg(int *) name;
    253 		syscallarg(u_int) namelen;
    254 		syscallarg(void *) old;
    255 		syscallarg(size_t *) oldlenp;
    256 		syscallarg(void *) new;
    257 		syscallarg(size_t) newlen;
    258 	} */ *uap = v;
    259 	int error, nerror, name[CTL_MAXNAME];
    260 	size_t oldlen, savelen, *oldlenp;
    261 
    262 	/*
    263 	 * get oldlen
    264 	 */
    265 	oldlen = 0;
    266 	oldlenp = SCARG(uap, oldlenp);
    267 	if (oldlenp != NULL) {
    268 		error = copyin(oldlenp, &oldlen, sizeof(oldlen));
    269 		if (error)
    270 			return (error);
    271 	}
    272 	savelen = oldlen;
    273 
    274 	/*
    275 	 * top-level sysctl names may or may not be non-terminal, but
    276 	 * we don't care
    277 	 */
    278 	if (SCARG(uap, namelen) > CTL_MAXNAME || SCARG(uap, namelen) < 1)
    279 		return (EINVAL);
    280 	error = copyin(SCARG(uap, name), &name,
    281 		       SCARG(uap, namelen) * sizeof(int));
    282 	if (error)
    283 		return (error);
    284 
    285 	/*
    286 	 * wire old so that copyout() is less likely to fail?
    287 	 */
    288 	error = sysctl_lock(l, SCARG(uap, old), savelen);
    289 	if (error)
    290 		return (error);
    291 
    292 	/*
    293 	 * do sysctl work (NULL means main built-in default tree)
    294 	 */
    295 	error = sysctl_dispatch(&name[0], SCARG(uap, namelen),
    296 				SCARG(uap, old), &oldlen,
    297 				SCARG(uap, new), SCARG(uap, newlen),
    298 				&name[0], l, NULL);
    299 
    300 	/*
    301 	 * release the sysctl lock
    302 	 */
    303 	sysctl_unlock(l);
    304 
    305 	/*
    306 	 * set caller's oldlen to new value even in the face of an
    307 	 * error (if this gets an error and they didn't have one, they
    308 	 * get this one)
    309 	 */
    310 	if (oldlenp) {
    311 		nerror = copyout(&oldlen, oldlenp, sizeof(oldlen));
    312 		if (error == 0)
    313 			error = nerror;
    314 	}
    315 
    316 	/*
    317 	 * if the only problem is that we weren't given enough space,
    318 	 * that's an ENOMEM error
    319 	 */
    320 	if (error == 0 && SCARG(uap, old) != NULL && savelen < oldlen)
    321 		error = ENOMEM;
    322 
    323 	return (error);
    324 }
    325 
    326 /*
    327  * ********************************************************************
    328  * Section 1: How the tree is used
    329  * ********************************************************************
    330  * Implementations of sysctl for emulations should typically need only
    331  * these three functions in this order: lock the tree, dispatch
    332  * request into it, unlock the tree.
    333  * ********************************************************************
    334  */
    335 int
    336 sysctl_lock(struct lwp *l, void *oldp, size_t savelen)
    337 {
    338 	int error = 0;
    339 
    340 	error = lockmgr(&sysctl_treelock, LK_EXCLUSIVE, NULL);
    341 	if (error)
    342 		return (error);
    343 
    344 	if (l != NULL && oldp != NULL && savelen) {
    345 		error = uvm_vslock(l->l_proc, oldp, savelen, VM_PROT_WRITE);
    346 		if (error) {
    347 			(void) lockmgr(&sysctl_treelock, LK_RELEASE, NULL);
    348 			return (error);
    349 		}
    350 		sysctl_memaddr = oldp;
    351 		sysctl_memsize = savelen;
    352 	}
    353 
    354 	return (0);
    355 }
    356 
    357 /*
    358  * ********************************************************************
    359  * the main sysctl dispatch routine.  scans the given tree and picks a
    360  * function to call based on what it finds.
    361  * ********************************************************************
    362  */
    363 int
    364 sysctl_dispatch(SYSCTLFN_RWARGS)
    365 {
    366 	int error;
    367 	sysctlfn fn;
    368 	int ni;
    369 
    370 	if (rnode && SYSCTL_VERS(rnode->sysctl_flags) != SYSCTL_VERSION) {
    371 		printf("sysctl_dispatch: rnode %p wrong version\n", rnode);
    372 		return (EINVAL);
    373 	}
    374 
    375 	fn = NULL;
    376 	error = sysctl_locate(l, name, namelen, &rnode, &ni);
    377 
    378 	/*
    379 	 * the node we ended up at has a function, so call it.  it can
    380 	 * hand off to query or create if it wants to.
    381 	 */
    382 	if (rnode->sysctl_func != NULL)
    383 		fn = rnode->sysctl_func;
    384 
    385 	/*
    386 	 * we found the node they were looking for, so do a lookup.
    387 	 */
    388 	else if (error == 0)
    389 		fn = (sysctlfn)sysctl_lookup; /* XXX may write to rnode */
    390 
    391 	/*
    392 	 * prospective parent node found, but the terminal node was
    393 	 * not.  generic operations associate with the parent.
    394 	 */
    395 	else if (error == ENOENT && (ni + 1) == namelen && name[ni] < 0) {
    396 		switch (name[ni]) {
    397 		case CTL_QUERY:
    398 			fn = sysctl_query;
    399 			break;
    400 		case CTL_CREATE:
    401 #if NKSYMS > 0
    402 		case CTL_CREATESYM:
    403 #endif /* NKSYMS > 0 */
    404 			fn = (sysctlfn)sysctl_create; /* we own the rnode */
    405 			break;
    406 		case CTL_DESTROY:
    407 			fn = (sysctlfn)sysctl_destroy; /* we own the rnode */
    408 			break;
    409 		case CTL_MMAP:
    410 			fn = (sysctlfn)sysctl_mmap; /* we own the rnode */
    411 			break;
    412 		default:
    413 			error = EOPNOTSUPP;
    414 			break;
    415 		}
    416 	}
    417 
    418 	/*
    419 	 * after all of that, maybe we found someone who knows how to
    420 	 * get us what we want?
    421 	 */
    422 	if (fn != NULL)
    423 		error = (*fn)(name + ni, namelen - ni, oldp, oldlenp,
    424 			      newp, newlen, name, l, rnode);
    425 
    426 	else if (error == 0)
    427 		error = EOPNOTSUPP;
    428 
    429 	return (error);
    430 }
    431 
    432 /*
    433  * ********************************************************************
    434  * Releases the tree lock.  Note that if uvm_vslock() was called when
    435  * the lock was taken, we release that memory now.  By keeping track
    436  * of where and how much by ourselves, the lock can be released much
    437  * more easily from anywhere.
    438  * ********************************************************************
    439  */
    440 void
    441 sysctl_unlock(struct lwp *l)
    442 {
    443 
    444 	if (l != NULL && sysctl_memsize != 0) {
    445 		uvm_vsunlock(l->l_proc, sysctl_memaddr, sysctl_memsize);
    446 		sysctl_memsize = 0;
    447 	}
    448 
    449 	(void) lockmgr(&sysctl_treelock, LK_RELEASE, NULL);
    450 }
    451 
    452 /*
    453  * ********************************************************************
    454  * Section 2: The main tree interfaces
    455  * ********************************************************************
    456  * This is how sysctl_dispatch() does its work, and you can too, by
    457  * calling these routines from helpers (though typically only
    458  * sysctl_lookup() will be used).  The tree MUST BE LOCKED when these
    459  * are called.
    460  * ********************************************************************
    461  */
    462 
    463 /*
    464  * sysctl_locate -- Finds the node matching the given mib under the
    465  * given tree (via rv).  If no tree is given, we fall back to the
    466  * native tree.  The current process (via l) is used for access
    467  * control on the tree (some nodes may be traversable only by root) and
    468  * on return, nip will show how many numbers in the mib were consumed.
    469  */
    470 int
    471 sysctl_locate(struct lwp *l, const int *name, u_int namelen,
    472 	      struct sysctlnode **rnode, int *nip)
    473 {
    474 	struct sysctlnode *node, *pnode;
    475 	int tn, si, ni, error, alias;
    476 
    477 	/*
    478 	 * basic checks and setup
    479 	 */
    480 	if (*rnode == NULL)
    481 		*rnode = &sysctl_root;
    482 	if (nip)
    483 		*nip = 0;
    484 	if (namelen < 0)
    485 		return (EINVAL);
    486 	if (namelen == 0)
    487 		return (0);
    488 
    489 	/*
    490 	 * search starts from "root"
    491 	 */
    492 	pnode = *rnode;
    493 	if (SYSCTL_VERS(pnode->sysctl_flags) != SYSCTL_VERSION) {
    494 		printf("sysctl_locate: pnode %p wrong version\n", pnode);
    495 		return (EINVAL);
    496 	}
    497 	node = pnode->sysctl_child;
    498 	error = 0;
    499 
    500 	/*
    501 	 * scan for node to which new node should be attached
    502 	 */
    503 	for (ni = 0; ni < namelen; ni++) {
    504 		/*
    505 		 * walked off bottom of tree
    506 		 */
    507 		if (node == NULL) {
    508 			if (SYSCTL_TYPE(pnode->sysctl_flags) == CTLTYPE_NODE)
    509 				error = ENOENT;
    510 			else
    511 				error = ENOTDIR;
    512 			break;
    513 		}
    514 		/*
    515 		 * can anyone traverse this node or only root?
    516 		 */
    517 		if (l != NULL && (pnode->sysctl_flags & CTLFLAG_PRIVATE) &&
    518 		    (error = suser(l->l_proc->p_ucred, &l->l_proc->p_acflag))
    519 		    != 0)
    520 			return (error);
    521 		/*
    522 		 * find a child node with the right number
    523 		 */
    524 		tn = name[ni];
    525 		alias = 0;
    526 
    527 		si = 0;
    528 		/*
    529 		 * Note: ANYNUMBER only matches positive integers.
    530 		 * Since ANYNUMBER is only permitted on single-node
    531 		 * sub-trees (eg proc), check before the loop and skip
    532 		 * it if we can.
    533 		 */
    534 		if ((node[si].sysctl_flags & CTLFLAG_ANYNUMBER) && (tn >= 0))
    535 			goto foundit;
    536 		for (; si < pnode->sysctl_clen; si++) {
    537 			if (node[si].sysctl_num == tn) {
    538 				if (node[si].sysctl_flags & CTLFLAG_ALIAS) {
    539 					if (alias++ == 4)
    540 						break;
    541 					else {
    542 						tn = node[si].sysctl_alias;
    543 						si = -1;
    544 					}
    545 				}
    546 				else
    547 					goto foundit;
    548 			}
    549 		}
    550 		/*
    551 		 * if we ran off the end, it obviously doesn't exist
    552 		 */
    553 		error = ENOENT;
    554 		break;
    555 
    556 		/*
    557 		 * so far so good, move on down the line
    558 		 */
    559 	  foundit:
    560 		pnode = &node[si];
    561 		if (SYSCTL_TYPE(pnode->sysctl_flags) == CTLTYPE_NODE)
    562 			node = node[si].sysctl_child;
    563 		else
    564 			node = NULL;
    565 	}
    566 
    567 	*rnode = pnode;
    568 	if (nip)
    569 		*nip = ni;
    570 
    571 	return (error);
    572 }
    573 
    574 /*
    575  * sysctl_query -- The auto-discovery engine.  Copies out the
    576  * descriptions on nodes under the given node and handles overlay
    577  * trees.
    578  */
    579 int
    580 sysctl_query(SYSCTLFN_ARGS)
    581 {
    582 	int error, ni, elim, v;
    583 	size_t out, left, t;
    584 	struct sysctlnode *enode, *onode, qnode;
    585 
    586 	if (SYSCTL_VERS(rnode->sysctl_flags) != SYSCTL_VERSION) {
    587 		printf("sysctl_query: rnode %p wrong version\n", rnode);
    588 		return (EINVAL);
    589 	}
    590 
    591 	if (SYSCTL_TYPE(rnode->sysctl_flags) != CTLTYPE_NODE)
    592 		return (ENOTDIR);
    593 	if (namelen != 1 || name[0] != CTL_QUERY)
    594 		return (EINVAL);
    595 
    596 	error = 0;
    597 	out = 0;
    598 	left = *oldlenp;
    599 	elim = 0;
    600 	enode = NULL;
    601 
    602 	/*
    603 	 * translate the given request to a current node
    604 	 */
    605 	error = sysctl_cvt_in(l, &v, newp, newlen, &qnode);
    606 	if (error)
    607 		return (error);
    608 
    609 	/*
    610 	 * if the request specifies a version, check it
    611 	 */
    612 	if (qnode.sysctl_ver != 0) {
    613 		enode = (struct sysctlnode *)rnode; /* discard const */
    614 		if (qnode.sysctl_ver != enode->sysctl_ver &&
    615 		    qnode.sysctl_ver != sysctl_rootof(enode)->sysctl_ver)
    616 			return (EINVAL);
    617 	}
    618 
    619 	/*
    620 	 * process has overlay tree
    621 	 */
    622 	if (l && l->l_proc->p_emul->e_sysctlovly) {
    623 		enode = l->l_proc->p_emul->e_sysctlovly;
    624 		elim = (name - oname);
    625 		error = sysctl_locate(l, oname, elim, &enode, NULL);
    626 		if (error == 0) {
    627 			/* ah, found parent in overlay */
    628 			elim = enode->sysctl_clen;
    629 			enode = enode->sysctl_child;
    630 		}
    631 		else {
    632 			error = 0;
    633 			elim = 0;
    634 			enode = NULL;
    635 		}
    636 	}
    637 
    638 	for (ni = 0; ni < rnode->sysctl_clen; ni++) {
    639 		onode = &rnode->sysctl_child[ni];
    640 		if (enode && enode->sysctl_num == onode->sysctl_num) {
    641 			if (SYSCTL_TYPE(enode->sysctl_flags) != CTLTYPE_NODE)
    642 				onode = enode;
    643 			if (--elim > 0)
    644 				enode++;
    645 			else
    646 				enode = NULL;
    647 		}
    648 		error = sysctl_cvt_out(l, v, onode, oldp, left, &t);
    649 		if (error)
    650 			return (error);
    651 		if (oldp != NULL)
    652 			oldp = (char*)oldp + t;
    653 		out += t;
    654 		left -= MIN(left, t);
    655 	}
    656 
    657 	/*
    658 	 * overlay trees *MUST* be entirely consumed
    659 	 */
    660 	KASSERT(enode == NULL);
    661 
    662 	*oldlenp = out;
    663 
    664 	return (error);
    665 }
    666 
    667 #ifdef SYSCTL_DEBUG_CREATE
    668 #undef sysctl_create
    669 #endif /* SYSCTL_DEBUG_CREATE */
    670 
    671 /*
    672  * sysctl_create -- Adds a node (the description of which is taken
    673  * from newp) to the tree, returning a copy of it in the space pointed
    674  * to by oldp.  In the event that the requested slot is already taken
    675  * (either by name or by number), the offending node is returned
    676  * instead.  Yes, this is complex, but we want to make sure everything
    677  * is proper.
    678  */
    679 int
    680 sysctl_create(SYSCTLFN_RWARGS)
    681 {
    682 	struct sysctlnode nnode, *node, *pnode;
    683 	int error, ni, at, nm, type, sz, flags, rw, anum, v;
    684 	void *own;
    685 
    686 	error = 0;
    687 	own = NULL;
    688 	anum = -1;
    689 
    690 	if (SYSCTL_VERS(rnode->sysctl_flags) != SYSCTL_VERSION) {
    691 		printf("sysctl_create: rnode %p wrong version\n", rnode);
    692 		return (EINVAL);
    693 	}
    694 
    695 	if (namelen != 1 || (name[namelen - 1] != CTL_CREATE
    696 #if NKSYMS > 0
    697 			     && name[namelen - 1] != CTL_CREATESYM
    698 #endif /* NKSYMS > 0 */
    699 			     ))
    700 		return (EINVAL);
    701 
    702 	/*
    703 	 * processes can only add nodes at securelevel 0, must be
    704 	 * root, and can't add nodes to a parent that's not writeable
    705 	 */
    706 	if (l != NULL) {
    707 #ifndef SYSCTL_DISALLOW_CREATE
    708 		if (securelevel > 0)
    709 			return (EPERM);
    710 		error = suser(l->l_proc->p_ucred, &l->l_proc->p_acflag);
    711 		if (error)
    712 			return (error);
    713 		if (!(rnode->sysctl_flags & CTLFLAG_READWRITE))
    714 #endif /* SYSCTL_DISALLOW_CREATE */
    715 			return (EPERM);
    716 	}
    717 
    718 	/*
    719 	 * nothing can add a node if:
    720 	 * we've finished initial set up and
    721 	 * the tree itself is not writeable or
    722 	 * the entire sysctl system is not writeable
    723 	 */
    724 	if ((sysctl_root.sysctl_flags & CTLFLAG_PERMANENT) &&
    725 	    (!(sysctl_rootof(rnode)->sysctl_flags & CTLFLAG_READWRITE) ||
    726 	     !(sysctl_root.sysctl_flags & CTLFLAG_READWRITE)))
    727 		return (EPERM);
    728 
    729 	/*
    730 	 * it must be a "node", not a "int" or something
    731 	 */
    732 	if (SYSCTL_TYPE(rnode->sysctl_flags) != CTLTYPE_NODE)
    733 		return (ENOTDIR);
    734 	pnode = rnode;
    735 
    736 	if (newp == NULL)
    737 		return (EINVAL);
    738 	error = sysctl_cvt_in(l, &v, newp, newlen, &nnode);
    739 	if (error)
    740 		return (error);
    741 
    742 	/*
    743 	 * nodes passed in don't *have* parents
    744 	 */
    745 	if (nnode.sysctl_parent != NULL)
    746 		return (EINVAL);
    747 
    748 	/*
    749 	 * if we are indeed adding it, it should be a "good" name and
    750 	 * number
    751 	 */
    752 	nm = nnode.sysctl_num;
    753 #if NKSYMS > 0
    754 	if (nm == CTL_CREATESYM)
    755 		nm = CTL_CREATE;
    756 #endif /* NKSYMS > 0 */
    757 	if (nm < 0 && nm != CTL_CREATE)
    758 		return (EINVAL);
    759 	sz = 0;
    760 
    761 	/*
    762 	 * the name can't start with a digit
    763 	 */
    764 	if (nnode.sysctl_name[sz] >= '0' &&
    765 	    nnode.sysctl_name[sz] <= '9')
    766 		return (EINVAL);
    767 
    768 	/*
    769 	 * the name must be only alphanumerics or - or _, longer than
    770 	 * 0 bytes and less that SYSCTL_NAMELEN
    771 	 */
    772 	while (sz < SYSCTL_NAMELEN && nnode.sysctl_name[sz] != '\0') {
    773 		if ((nnode.sysctl_name[sz] >= '0' &&
    774 		     nnode.sysctl_name[sz] <= '9') ||
    775 		    (nnode.sysctl_name[sz] >= 'A' &&
    776 		     nnode.sysctl_name[sz] <= 'Z') ||
    777 		    (nnode.sysctl_name[sz] >= 'a' &&
    778 		     nnode.sysctl_name[sz] <= 'z') ||
    779 		    nnode.sysctl_name[sz] == '-' ||
    780 		    nnode.sysctl_name[sz] == '_')
    781 			sz++;
    782 		else
    783 			return (EINVAL);
    784 	}
    785 	if (sz == 0 || sz == SYSCTL_NAMELEN)
    786 		return (EINVAL);
    787 
    788 	/*
    789 	 * various checks revolve around size vs type, etc
    790 	 */
    791 	type = SYSCTL_TYPE(nnode.sysctl_flags);
    792 	flags = SYSCTL_FLAGS(nnode.sysctl_flags);
    793 	rw = (flags & CTLFLAG_READWRITE) ? B_WRITE : B_READ;
    794 	sz = nnode.sysctl_size;
    795 
    796 	/*
    797 	 * find out if there's a collision, and if so, let the caller
    798 	 * know what they collided with
    799 	 */
    800 	node = pnode->sysctl_child;
    801 	if (((flags & CTLFLAG_ANYNUMBER) && node) ||
    802 	    (node && node->sysctl_flags & CTLFLAG_ANYNUMBER))
    803 		return (EINVAL);
    804 	for (ni = at = 0; ni < pnode->sysctl_clen; ni++) {
    805 		if (nm == node[ni].sysctl_num ||
    806 		    strcmp(nnode.sysctl_name, node[ni].sysctl_name) == 0) {
    807 			/*
    808 			 * ignore error here, since we
    809 			 * are already fixed on EEXIST
    810 			 */
    811 			(void)sysctl_cvt_out(l, v, &node[ni], oldp,
    812 					     *oldlenp, oldlenp);
    813 			return (EEXIST);
    814 		}
    815 		if (nm > node[ni].sysctl_num)
    816 			at++;
    817 	}
    818 
    819 	/*
    820 	 * use sysctl_ver to add to the tree iff it hasn't changed
    821 	 */
    822 	if (nnode.sysctl_ver != 0) {
    823 		/*
    824 		 * a specified value must match either the parent
    825 		 * node's version or the root node's version
    826 		 */
    827 		if (nnode.sysctl_ver != sysctl_rootof(rnode)->sysctl_ver &&
    828 		    nnode.sysctl_ver != rnode->sysctl_ver) {
    829 			return (EINVAL);
    830 		}
    831 	}
    832 
    833 	/*
    834 	 * only the kernel can assign functions to entries
    835 	 */
    836 	if (l != NULL && nnode.sysctl_func != NULL)
    837 		return (EPERM);
    838 
    839 	/*
    840 	 * only the kernel can create permanent entries, and only then
    841 	 * before the kernel is finished setting itself up
    842 	 */
    843 	if (l != NULL && (flags & ~SYSCTL_USERFLAGS))
    844 		return (EPERM);
    845 	if ((flags & CTLFLAG_PERMANENT) &
    846 	    (sysctl_root.sysctl_flags & CTLFLAG_PERMANENT))
    847 		return (EPERM);
    848 	if ((flags & (CTLFLAG_OWNDATA | CTLFLAG_IMMEDIATE)) ==
    849 	    (CTLFLAG_OWNDATA | CTLFLAG_IMMEDIATE))
    850 		return (EINVAL);
    851 	if ((flags & CTLFLAG_IMMEDIATE) &&
    852 	    type != CTLTYPE_INT && type != CTLTYPE_QUAD)
    853 		return (EINVAL);
    854 
    855 	/*
    856 	 * check size, or set it if unset and we can figure it out.
    857 	 * kernel created nodes are allowed to have a function instead
    858 	 * of a size (or a data pointer).
    859 	 */
    860 	switch (type) {
    861 	case CTLTYPE_NODE:
    862 		/*
    863 		 * only *i* can assert the size of a node
    864 		 */
    865 		if (flags & CTLFLAG_ALIAS) {
    866 			anum = nnode.sysctl_alias;
    867 			if (anum < 0)
    868 				return (EINVAL);
    869 			nnode.sysctl_alias = 0;
    870 		}
    871 		if (sz != 0 || nnode.sysctl_data != NULL)
    872 			return (EINVAL);
    873 		if (nnode.sysctl_csize != 0 ||
    874 		    nnode.sysctl_clen != 0 ||
    875 		    nnode.sysctl_child != 0)
    876 			return (EINVAL);
    877 		if (flags & CTLFLAG_OWNDATA)
    878 			return (EINVAL);
    879 		sz = sizeof(struct sysctlnode);
    880 		break;
    881 	case CTLTYPE_INT:
    882 		/*
    883 		 * since an int is an int, if the size is not given or
    884 		 * is wrong, we can "int-uit" it.
    885 		 */
    886 		if (sz != 0 && sz != sizeof(int))
    887 			return (EINVAL);
    888 		sz = sizeof(int);
    889 		break;
    890 	case CTLTYPE_STRING:
    891 		/*
    892 		 * strings are a little more tricky
    893 		 */
    894 		if (sz == 0) {
    895 			if (l == NULL) {
    896 				if (nnode.sysctl_func == NULL) {
    897 					if (nnode.sysctl_data == NULL)
    898 						return (EINVAL);
    899 					else
    900 						sz = strlen(nnode.sysctl_data) +
    901 						    1;
    902 				}
    903 			}
    904 			else if (nnode.sysctl_data == NULL &&
    905 				 flags & CTLFLAG_OWNDATA) {
    906 				return (EINVAL);
    907 			}
    908 			else {
    909 				char v[PAGE_SIZE], *e;
    910 				size_t s;
    911 
    912 				/*
    913 				 * we want a rough idea of what the
    914 				 * size is now
    915 				 */
    916 				e = nnode.sysctl_data;
    917 				do {
    918 					error = copyinstr(e, &v[0], sizeof(v),
    919 							  &s);
    920 					if (error) {
    921 						if (error != ENAMETOOLONG)
    922 							return (error);
    923 						e += PAGE_SIZE;
    924 						if ((e - 32 * PAGE_SIZE) >
    925 						    (char*)nnode.sysctl_data)
    926 							return (ERANGE);
    927 					}
    928 				} while (error != 0);
    929 				sz = s + (e - (char*)nnode.sysctl_data);
    930 			}
    931 		}
    932 		break;
    933 	case CTLTYPE_QUAD:
    934 		if (sz != 0 && sz != sizeof(u_quad_t))
    935 			return (EINVAL);
    936 		sz = sizeof(u_quad_t);
    937 		break;
    938 	case CTLTYPE_STRUCT:
    939 		if (sz == 0) {
    940 			if (l != NULL || nnode.sysctl_func == NULL)
    941 				return (EINVAL);
    942 			if (flags & CTLFLAG_OWNDATA)
    943 				return (EINVAL);
    944 		}
    945 		break;
    946 	default:
    947 		return (EINVAL);
    948 	}
    949 
    950 	/*
    951 	 * at this point, if sz is zero, we *must* have a
    952 	 * function to go with it and we can't own it.
    953 	 */
    954 
    955 	/*
    956 	 *  l  ptr own
    957 	 *  0   0   0  -> EINVAL (if no func)
    958 	 *  0   0   1  -> own
    959 	 *  0   1   0  -> kptr
    960 	 *  0   1   1  -> kptr
    961 	 *  1   0   0  -> EINVAL
    962 	 *  1   0   1  -> own
    963 	 *  1   1   0  -> kptr, no own (fault on lookup)
    964 	 *  1   1   1  -> uptr, own
    965 	 */
    966 	if (type != CTLTYPE_NODE) {
    967 		if (sz != 0) {
    968 			if (flags & CTLFLAG_OWNDATA) {
    969 				own = malloc(sz, M_SYSCTLDATA,
    970 					     M_WAITOK|M_CANFAIL);
    971 				if (nnode.sysctl_data == NULL)
    972 					memset(own, 0, sz);
    973 				else {
    974 					error = sysctl_copyin(l,
    975 					    nnode.sysctl_data, own, sz);
    976 					if (error != 0) {
    977 						FREE(own, M_SYSCTLDATA);
    978 						return (error);
    979 					}
    980 				}
    981 			}
    982 			else if ((nnode.sysctl_data != NULL) &&
    983 				 !(flags & CTLFLAG_IMMEDIATE)) {
    984 #if NKSYMS > 0
    985 				if (name[namelen - 1] == CTL_CREATESYM) {
    986 					char symname[128]; /* XXX enough? */
    987 					u_long symaddr;
    988 					size_t symlen;
    989 
    990 					error = sysctl_copyinstr(l,
    991 					    nnode.sysctl_data, symname,
    992 					    sizeof(symname), &symlen);
    993 					if (error)
    994 						return (error);
    995 					error = ksyms_getval_from_kernel(NULL,
    996 					    symname, &symaddr, KSYMS_EXTERN);
    997 					if (error)
    998 						return (error); /* EINVAL? */
    999 					nnode.sysctl_data = (void*)symaddr;
   1000 				}
   1001 #endif /* NKSYMS > 0 */
   1002 				/*
   1003 				 * Ideally, we'd like to verify here
   1004 				 * that this address is acceptable,
   1005 				 * but...
   1006 				 *
   1007 				 * - it might be valid now, only to
   1008 				 *   become invalid later
   1009 				 *
   1010 				 * - it might be invalid only for the
   1011 				 *   moment and valid later
   1012 				 *
   1013 				 * - or something else.
   1014 				 *
   1015 				 * Since we can't get a good answer,
   1016 				 * we'll just accept the address as
   1017 				 * given, and fault on individual
   1018 				 * lookups.
   1019 				 */
   1020 			}
   1021 		}
   1022 		else if (nnode.sysctl_func == NULL)
   1023 			return (EINVAL);
   1024 	}
   1025 
   1026 	/*
   1027 	 * a process can't assign a function to a node, and the kernel
   1028 	 * can't create a node that has no function or data.
   1029 	 * (XXX somewhat redundant check)
   1030 	 */
   1031 	if (l != NULL || nnode.sysctl_func == NULL) {
   1032 		if (type != CTLTYPE_NODE &&
   1033 		    nnode.sysctl_data == NULL &&
   1034 		    !(flags & CTLFLAG_IMMEDIATE) &&
   1035 		    own == NULL)
   1036 			return (EINVAL);
   1037 	}
   1038 
   1039 #ifdef SYSCTL_DISALLOW_KWRITE
   1040 	/*
   1041 	 * a process can't create a writable node unless it refers to
   1042 	 * new data.
   1043 	 */
   1044 	if (l != NULL && own == NULL && type != CTLTYPE_NODE &&
   1045 	    (flags & CTLFLAG_READWRITE) != CTLFLAG_READONLY &&
   1046 	    !(flags & CTLFLAG_IMMEDIATE))
   1047 		return (EPERM);
   1048 #endif /* SYSCTL_DISALLOW_KWRITE */
   1049 
   1050 	/*
   1051 	 * make sure there's somewhere to put the new stuff.
   1052 	 */
   1053 	if (pnode->sysctl_child == NULL) {
   1054 		if (flags & CTLFLAG_ANYNUMBER)
   1055 			error = sysctl_alloc(pnode, 1);
   1056 		else
   1057 			error = sysctl_alloc(pnode, 0);
   1058 		if (error)
   1059 			return (error);
   1060 	}
   1061 	node = pnode->sysctl_child;
   1062 
   1063 	/*
   1064 	 * no collisions, so pick a good dynamic number if we need to.
   1065 	 */
   1066 	if (nm == CTL_CREATE) {
   1067 		nm = ++sysctl_root.sysctl_num;
   1068 		for (ni = 0; ni < pnode->sysctl_clen; ni++) {
   1069 			if (nm == node[ni].sysctl_num) {
   1070 				nm++;
   1071 				ni = -1;
   1072 			}
   1073 			else if (nm > node[ni].sysctl_num)
   1074 				at = ni + 1;
   1075 		}
   1076 	}
   1077 
   1078 	/*
   1079 	 * oops...ran out of space
   1080 	 */
   1081 	if (pnode->sysctl_clen == pnode->sysctl_csize) {
   1082 		error = sysctl_realloc(pnode);
   1083 		if (error)
   1084 			return (error);
   1085 		node = pnode->sysctl_child;
   1086 	}
   1087 
   1088 	/*
   1089 	 * insert new node data
   1090 	 */
   1091 	if (at < pnode->sysctl_clen) {
   1092 		int t;
   1093 
   1094 		/*
   1095 		 * move the nodes that should come after the new one
   1096 		 */
   1097 		memmove(&node[at + 1], &node[at],
   1098 			(pnode->sysctl_clen - at) * sizeof(struct sysctlnode));
   1099 		memset(&node[at], 0, sizeof(struct sysctlnode));
   1100 		node[at].sysctl_parent = pnode;
   1101 		/*
   1102 		 * and...reparent any children of any moved nodes
   1103 		 */
   1104 		for (ni = at; ni <= pnode->sysctl_clen; ni++)
   1105 			if (SYSCTL_TYPE(node[ni].sysctl_flags) == CTLTYPE_NODE)
   1106 				for (t = 0; t < node[ni].sysctl_clen; t++)
   1107 					node[ni].sysctl_child[t].sysctl_parent =
   1108 						&node[ni];
   1109 	}
   1110 	node = &node[at];
   1111 	pnode->sysctl_clen++;
   1112 
   1113 	strlcpy(node->sysctl_name, nnode.sysctl_name,
   1114 		sizeof(node->sysctl_name));
   1115 	node->sysctl_num = nm;
   1116 	node->sysctl_size = sz;
   1117 	node->sysctl_flags = SYSCTL_VERSION|type|flags; /* XXX other trees */
   1118 	node->sysctl_csize = 0;
   1119 	node->sysctl_clen = 0;
   1120 	if (own) {
   1121 		node->sysctl_data = own;
   1122 		node->sysctl_flags |= CTLFLAG_OWNDATA;
   1123 	}
   1124 	else if (flags & CTLFLAG_ALIAS) {
   1125 		node->sysctl_alias = anum;
   1126 	}
   1127 	else if (flags & CTLFLAG_IMMEDIATE) {
   1128 		switch (type) {
   1129 		case CTLTYPE_INT:
   1130 			node->sysctl_idata = nnode.sysctl_idata;
   1131 			break;
   1132 		case CTLTYPE_QUAD:
   1133 			node->sysctl_qdata = nnode.sysctl_qdata;
   1134 			break;
   1135 		}
   1136 	}
   1137 	else {
   1138 		node->sysctl_data = nnode.sysctl_data;
   1139 		node->sysctl_flags &= ~CTLFLAG_OWNDATA;
   1140 	}
   1141         node->sysctl_func = nnode.sysctl_func;
   1142         node->sysctl_child = NULL;
   1143 	/* node->sysctl_parent should already be done */
   1144 
   1145 	/*
   1146 	 * update "version" on path to "root"
   1147 	 */
   1148 	for (; rnode->sysctl_parent != NULL; rnode = rnode->sysctl_parent)
   1149 		;
   1150 	pnode = node;
   1151 	for (nm = rnode->sysctl_ver + 1; pnode != NULL;
   1152 	     pnode = pnode->sysctl_parent)
   1153 		pnode->sysctl_ver = nm;
   1154 
   1155 	error = sysctl_cvt_out(l, v, node, oldp, *oldlenp, oldlenp);
   1156 
   1157 	return (error);
   1158 }
   1159 
   1160 /*
   1161  * ********************************************************************
   1162  * A wrapper around sysctl_create() that prints the thing we're trying
   1163  * to add.
   1164  * ********************************************************************
   1165  */
   1166 #ifdef SYSCTL_DEBUG_CREATE
   1167 int _sysctl_create(SYSCTLFN_RWPROTO);
   1168 int
   1169 _sysctl_create(SYSCTLFN_RWARGS)
   1170 {
   1171 	const struct sysctlnode *node;
   1172 	int k, rc, ni, nl = namelen + (name - oname);
   1173 
   1174 	node = newp;
   1175 
   1176 	printf("namelen %d (", nl);
   1177 	for (ni = 0; ni < nl - 1; ni++)
   1178 		printf(" %d", oname[ni]);
   1179 	printf(" %d )\t[%s]\tflags %08x (%08x %d %zu)\n",
   1180 	       k = node->sysctl_num,
   1181 	       node->sysctl_name,
   1182 	       node->sysctl_flags,
   1183 	       SYSCTL_FLAGS(node->sysctl_flags),
   1184 	       SYSCTL_TYPE(node->sysctl_flags),
   1185 	       node->sysctl_size);
   1186 
   1187 	node = rnode;
   1188 	rc = sysctl_create(SYSCTLFN_CALL(rnode));
   1189 
   1190 	printf("sysctl_create(");
   1191 	for (ni = 0; ni < nl - 1; ni++)
   1192 		printf(" %d", oname[ni]);
   1193 	printf(" %d ) returned %d\n", k, rc);
   1194 
   1195 	return (rc);
   1196 }
   1197 #define sysctl_create _sysctl_create
   1198 #endif /* SYSCTL_DEBUG_CREATE */
   1199 
   1200 /*
   1201  * sysctl_destroy -- Removes a node (as described by newp) from the
   1202  * given tree, returning (if successful) a copy of the dead node in
   1203  * oldp.  Since we're removing stuff, there's not much to check.
   1204  */
   1205 int
   1206 sysctl_destroy(SYSCTLFN_RWARGS)
   1207 {
   1208 	struct sysctlnode *node, *pnode, onode, nnode;
   1209 	int ni, error, v;
   1210 
   1211 	if (SYSCTL_VERS(rnode->sysctl_flags) != SYSCTL_VERSION) {
   1212 		printf("sysctl_destroy: rnode %p wrong version\n", rnode);
   1213 		return (EINVAL);
   1214 	}
   1215 
   1216 	error = 0;
   1217 
   1218 	if (namelen != 1 || name[namelen - 1] != CTL_DESTROY)
   1219 		return (EINVAL);
   1220 
   1221 	/*
   1222 	 * processes can only destroy nodes at securelevel 0, must be
   1223 	 * root, and can't remove nodes from a parent that's not
   1224 	 * writeable
   1225 	 */
   1226 	if (l != NULL) {
   1227 #ifndef SYSCTL_DISALLOW_CREATE
   1228 		if (securelevel > 0)
   1229 			return (EPERM);
   1230 		error = suser(l->l_proc->p_ucred, &l->l_proc->p_acflag);
   1231 		if (error)
   1232 			return (error);
   1233 		if (!(rnode->sysctl_flags & CTLFLAG_READWRITE))
   1234 #endif /* SYSCTL_DISALLOW_CREATE */
   1235 			return (EPERM);
   1236 	}
   1237 
   1238 	/*
   1239 	 * nothing can remove a node if:
   1240 	 * the node is permanent (checked later) or
   1241 	 * the tree itself is not writeable or
   1242 	 * the entire sysctl system is not writeable
   1243 	 */
   1244 	if (!(sysctl_rootof(rnode)->sysctl_flags & CTLFLAG_READWRITE) ||
   1245 	    !(sysctl_root.sysctl_flags & CTLFLAG_READWRITE))
   1246 		return (EPERM);
   1247 
   1248 	if (newp == NULL)
   1249 		return (EINVAL);
   1250 	error = sysctl_cvt_in(l, &v, newp, newlen, &nnode);
   1251 	if (error)
   1252 		return (error);
   1253 	memset(&onode, 0, sizeof(struct sysctlnode));
   1254 
   1255 	node = rnode->sysctl_child;
   1256 	for (ni = 0; ni < rnode->sysctl_clen; ni++) {
   1257 		if (nnode.sysctl_num == node[ni].sysctl_num) {
   1258 			/*
   1259 			 * if name specified, must match
   1260 			 */
   1261 			if (nnode.sysctl_name[0] != '\0' &&
   1262 			    strcmp(nnode.sysctl_name, node[ni].sysctl_name))
   1263 				continue;
   1264 			/*
   1265 			 * if version specified, must match
   1266 			 */
   1267 			if (nnode.sysctl_ver != 0 &&
   1268 			    nnode.sysctl_ver != node[ni].sysctl_ver)
   1269 				continue;
   1270 			/*
   1271 			 * this must be the one
   1272 			 */
   1273 			break;
   1274 		}
   1275 	}
   1276 	if (ni == rnode->sysctl_clen)
   1277 		return (ENOENT);
   1278 	node = &node[ni];
   1279 	pnode = node->sysctl_parent;
   1280 
   1281 	/*
   1282 	 * if the kernel says permanent, it is, so there.  nyah.
   1283 	 */
   1284 	if (SYSCTL_FLAGS(node->sysctl_flags) & CTLFLAG_PERMANENT)
   1285 		return (EPERM);
   1286 
   1287 	/*
   1288 	 * can't delete non-empty nodes
   1289 	 */
   1290 	if (SYSCTL_TYPE(node->sysctl_flags) == CTLTYPE_NODE &&
   1291 	    node->sysctl_clen != 0)
   1292 		return (ENOTEMPTY);
   1293 
   1294 	/*
   1295 	 * if the node "owns" data, release it now
   1296 	 */
   1297 	if (node->sysctl_flags & CTLFLAG_OWNDATA) {
   1298 		if (node->sysctl_data != NULL)
   1299 			FREE(node->sysctl_data, M_SYSCTLDATA);
   1300 		node->sysctl_data = NULL;
   1301 	}
   1302 
   1303 	/*
   1304 	 * if the node to be removed is not the last one on the list,
   1305 	 * move the remaining nodes up, and reparent any grandchildren
   1306 	 */
   1307 	onode = *node;
   1308 	if (ni < pnode->sysctl_clen - 1) {
   1309 		int t;
   1310 
   1311 		memmove(&pnode->sysctl_child[ni], &pnode->sysctl_child[ni + 1],
   1312 			(pnode->sysctl_clen - ni - 1) *
   1313 			sizeof(struct sysctlnode));
   1314 		for (; ni < pnode->sysctl_clen - 1; ni++)
   1315 			if (SYSCTL_TYPE(pnode->sysctl_child[ni].sysctl_flags) ==
   1316 			    CTLTYPE_NODE)
   1317 				for (t = 0;
   1318 				     t < pnode->sysctl_child[ni].sysctl_clen;
   1319 				     t++)
   1320 					pnode->sysctl_child[ni].sysctl_child[t].
   1321 						sysctl_parent =
   1322 						&pnode->sysctl_child[ni];
   1323 		ni = pnode->sysctl_clen - 1;
   1324 		node = &pnode->sysctl_child[ni];
   1325 	}
   1326 
   1327 	/*
   1328 	 * reset the space we just vacated
   1329 	 */
   1330 	memset(node, 0, sizeof(struct sysctlnode));
   1331 	node->sysctl_parent = pnode;
   1332 	pnode->sysctl_clen--;
   1333 
   1334 	/*
   1335 	 * if this parent just lost its last child, nuke the creche
   1336 	 */
   1337 	if (pnode->sysctl_clen == 0) {
   1338 		FREE(pnode->sysctl_child, M_SYSCTLNODE);
   1339 		pnode->sysctl_csize = 0;
   1340 		pnode->sysctl_child = NULL;
   1341 	}
   1342 
   1343 	/*
   1344 	 * update "version" on path to "root"
   1345 	 */
   1346         for (; rnode->sysctl_parent != NULL; rnode = rnode->sysctl_parent)
   1347                 ;
   1348 	for (ni = rnode->sysctl_ver + 1; pnode != NULL;
   1349 	     pnode = pnode->sysctl_parent)
   1350 		pnode->sysctl_ver = ni;
   1351 
   1352 	error = sysctl_cvt_out(l, v, &onode, oldp, *oldlenp, oldlenp);
   1353 
   1354 	return (error);
   1355 }
   1356 
   1357 /*
   1358  * sysctl_lookup -- Handles copyin/copyout of new and old values.
   1359  * Partial reads are globally allowed.  Only root can write to things
   1360  * unless the node says otherwise.
   1361  */
   1362 int
   1363 sysctl_lookup(SYSCTLFN_RWARGS)
   1364 {
   1365 	int error, rw;
   1366 	size_t sz, len;
   1367 	void *d;
   1368 
   1369 	if (SYSCTL_VERS(rnode->sysctl_flags) != SYSCTL_VERSION) {
   1370 		printf("sysctl_lookup: rnode %p wrong version\n", rnode);
   1371 		return (EINVAL);
   1372 	}
   1373 
   1374 	error = 0;
   1375 
   1376 	/*
   1377 	 * you can't "look up" a node.  you can "query" it, but you
   1378 	 * can't "look it up".
   1379 	 */
   1380 	if (SYSCTL_TYPE(rnode->sysctl_flags) == CTLTYPE_NODE || namelen != 0)
   1381 		return (EINVAL);
   1382 
   1383 	/*
   1384 	 * some nodes are private, so only root can look into them.
   1385 	 */
   1386 	if (l != NULL && (rnode->sysctl_flags & CTLFLAG_PRIVATE) &&
   1387 	    (error = suser(l->l_proc->p_ucred, &l->l_proc->p_acflag)) != 0)
   1388 		return (error);
   1389 
   1390 	/*
   1391 	 * if a node wants to be writable according to different rules
   1392 	 * other than "only root can write to stuff unless a flag is
   1393 	 * set", then it needs its own function which should have been
   1394 	 * called and not us.
   1395 	 */
   1396 	if (l != NULL && newp != NULL &&
   1397 	    !(rnode->sysctl_flags & CTLFLAG_ANYWRITE) &&
   1398 	    (error = suser(l->l_proc->p_ucred, &l->l_proc->p_acflag)) != 0)
   1399 		return (error);
   1400 
   1401 	/*
   1402 	 * is this node supposedly writable?
   1403 	 */
   1404 	rw = 0;
   1405 	switch (rnode->sysctl_flags & CTLFLAG_READWRITE) {
   1406 	    case CTLFLAG_READONLY1:
   1407 		rw = (securelevel < 1) ? 1 : 0;
   1408 		break;
   1409 	    case CTLFLAG_READONLY2:
   1410 		rw = (securelevel < 2) ? 1 : 0;
   1411 		break;
   1412 	    case CTLFLAG_READWRITE:
   1413 		rw = 1;
   1414 		break;
   1415 	}
   1416 
   1417 	/*
   1418 	 * it appears not to be writable at this time, so if someone
   1419 	 * tried to write to it, we must tell them to go away
   1420 	 */
   1421 	if (!rw && newp != NULL)
   1422 		return (EPERM);
   1423 
   1424 	/*
   1425 	 * step one, copy out the stuff we have presently
   1426 	 */
   1427 	if (rnode->sysctl_flags & CTLFLAG_IMMEDIATE) {
   1428 		switch (SYSCTL_TYPE(rnode->sysctl_flags)) {
   1429 		case CTLTYPE_INT:
   1430 			d = &rnode->sysctl_idata;
   1431 			break;
   1432 		case CTLTYPE_QUAD:
   1433 			d = &rnode->sysctl_qdata;
   1434 			break;
   1435 		default:
   1436 			return (EINVAL);
   1437 		}
   1438 	}
   1439 	else
   1440 		d = rnode->sysctl_data;
   1441 	if (SYSCTL_TYPE(rnode->sysctl_flags) == CTLTYPE_STRING)
   1442 		sz = strlen(d) + 1; /* XXX@@@ possible fault here */
   1443 	else
   1444 		sz = rnode->sysctl_size;
   1445 	if (oldp != NULL)
   1446 		error = sysctl_copyout(l, d, oldp, MIN(sz, *oldlenp));
   1447 	if (error)
   1448 		return (error);
   1449 	*oldlenp = sz;
   1450 
   1451 	/*
   1452 	 * are we done?
   1453 	 */
   1454 	if (newp == NULL || newlen == 0)
   1455 		return (0);
   1456 
   1457 	/*
   1458 	 * hmm...not done.  must now "copy in" new value.  re-adjust
   1459 	 * sz to maximum value (strings are "weird").
   1460 	 */
   1461 	sz = rnode->sysctl_size;
   1462 	switch (SYSCTL_TYPE(rnode->sysctl_flags)) {
   1463 	case CTLTYPE_INT:
   1464 	case CTLTYPE_QUAD:
   1465 	case CTLTYPE_STRUCT:
   1466 		/*
   1467 		 * these data must be *exactly* the same size coming
   1468 		 * in.
   1469 		 */
   1470 		if (newlen != sz)
   1471 			return (EINVAL);
   1472 		error = sysctl_copyin(l, newp, d, sz);
   1473 		break;
   1474 	case CTLTYPE_STRING: {
   1475 		/*
   1476 		 * strings, on the other hand, can be shorter, and we
   1477 		 * let userland be sloppy about the trailing nul.
   1478 		 */
   1479 		char *newbuf;
   1480 
   1481 		/*
   1482 		 * too much new string?
   1483 		 */
   1484 		if (newlen > sz)
   1485 			return (EINVAL);
   1486 
   1487 		/*
   1488 		 * temporary copy of new inbound string
   1489 		 */
   1490 		len = MIN(sz, newlen);
   1491 		newbuf = malloc(len, M_SYSCTLDATA, M_WAITOK|M_CANFAIL);
   1492 		if (newbuf == NULL)
   1493 			return (ENOMEM);
   1494 		error = sysctl_copyin(l, newp, newbuf, len);
   1495 		if (error) {
   1496 			FREE(newbuf, M_SYSCTLDATA);
   1497 			return (error);
   1498 		}
   1499 
   1500 		/*
   1501 		 * did they null terminate it, or do we have space
   1502 		 * left to do it ourselves?
   1503 		 */
   1504 		if (newbuf[len - 1] != '\0' && len == sz) {
   1505 			FREE(newbuf, M_SYSCTLDATA);
   1506 			return (EINVAL);
   1507 		}
   1508 
   1509 		/*
   1510 		 * looks good, so pop it into place and zero the rest.
   1511 		 */
   1512 		if (len > 0)
   1513 			memcpy(rnode->sysctl_data, newbuf, len);
   1514 		if (sz != len)
   1515 			memset((char*)rnode->sysctl_data + len, 0, sz - len);
   1516 		FREE(newbuf, M_SYSCTLDATA);
   1517 		break;
   1518 	}
   1519 	default:
   1520 		return (EINVAL);
   1521 	}
   1522 
   1523 	return (error);
   1524 }
   1525 
   1526 /*
   1527  * sysctl_mmap -- Dispatches sysctl mmap requests to those nodes that
   1528  * purport to handle it.  This interface isn't fully fleshed out yet,
   1529  * unfortunately.
   1530  */
   1531 static int
   1532 sysctl_mmap(SYSCTLFN_RWARGS)
   1533 {
   1534 	struct sysctlnode nnode, *node;
   1535 	int error;
   1536 
   1537 	if (SYSCTL_VERS(rnode->sysctl_flags) != SYSCTL_VERSION) {
   1538 		printf("sysctl_mmap: rnode %p wrong version\n", rnode);
   1539 		return (EINVAL);
   1540 	}
   1541 
   1542 	/*
   1543 	 * let's just pretend that didn't happen, m'kay?
   1544 	 */
   1545 	if (l == NULL)
   1546 		return (EPERM);
   1547 
   1548 	/*
   1549 	 * is this a sysctlnode description of an mmap request?
   1550 	 */
   1551 	if (newp == NULL || newlen != sizeof(struct sysctlnode))
   1552 		return (EINVAL);
   1553 	error = sysctl_copyin(l, newp, &nnode, sizeof(nnode));
   1554 	if (error)
   1555 		return (error);
   1556 
   1557 	/*
   1558 	 * does the node they asked for exist?
   1559 	 */
   1560 	if (namelen != 1)
   1561 		return (EOPNOTSUPP);
   1562 	node = rnode;
   1563         error = sysctl_locate(l, &nnode.sysctl_num, 1, &node, NULL);
   1564 	if (error)
   1565 		return (error);
   1566 
   1567 	/*
   1568 	 * does this node that we have found purport to handle mmap?
   1569 	 */
   1570 	if (node->sysctl_func == NULL ||
   1571 	    !(node->sysctl_flags & CTLFLAG_MMAP))
   1572 		return (EOPNOTSUPP);
   1573 
   1574 	/*
   1575 	 * well...okay, they asked for it.
   1576 	 */
   1577 	return ((*node->sysctl_func)(SYSCTLFN_CALL(node)));
   1578 }
   1579 
   1580 /*
   1581  * ********************************************************************
   1582  * Section 3: Create and destroy from inside the kernel
   1583  * ********************************************************************
   1584  * sysctl_createv() and sysctl_destroyv() are simpler-to-use
   1585  * interfaces for the kernel to fling new entries into the mib and rip
   1586  * them out later.  In the case of sysctl_createv(), the returned copy
   1587  * of the node (see sysctl_create()) will be translated back into a
   1588  * pointer to the actual node.
   1589  *
   1590  * Note that sysctl_createv() will return 0 if the create request
   1591  * matches an existing node (ala mkdir -p), and that sysctl_destroyv()
   1592  * will return 0 if the node to be destroyed already does not exist
   1593  * (aka rm -f) or if it is a parent of other nodes.
   1594  *
   1595  * This allows two (or more) different subsystems to assert sub-tree
   1596  * existence before populating their own nodes, and to remove their
   1597  * own nodes without orphaning the others when they are done.
   1598  * ********************************************************************
   1599  */
   1600 int
   1601 sysctl_createv(struct sysctllog **log, int cflags,
   1602 	       struct sysctlnode **rnode, struct sysctlnode **cnode,
   1603 	       int flags, int type, const char *namep, const char *descr,
   1604 	       sysctlfn func, u_quad_t qv, void *newp, size_t newlen,
   1605 	       ...)
   1606 {
   1607 	va_list ap;
   1608 	int error, ni, namelen, name[CTL_MAXNAME];
   1609 	struct sysctlnode *pnode, nnode, onode;
   1610 	size_t sz;
   1611 
   1612 	if (cnode != NULL)
   1613 		*cnode = NULL;
   1614 	if (cflags != 0)
   1615 		return (EINVAL);
   1616 
   1617 	/*
   1618 	 * where are we putting this?
   1619 	 */
   1620 	if (rnode != NULL && *rnode == NULL) {
   1621 		printf("sysctl_createv: rnode NULL\n");
   1622 		return (EINVAL);
   1623 	}
   1624 
   1625 	/*
   1626 	 * what is it?
   1627 	 */
   1628 	flags = SYSCTL_VERSION|SYSCTL_TYPE(type)|SYSCTL_FLAGS(flags);
   1629 	if (log != NULL)
   1630 		flags &= ~CTLFLAG_PERMANENT;
   1631 
   1632 	/*
   1633 	 * where do we put it?
   1634 	 */
   1635 	va_start(ap, newlen);
   1636 	namelen = 0;
   1637 	ni = -1;
   1638 	do {
   1639 		if (++ni == CTL_MAXNAME)
   1640 			return (ENAMETOOLONG);
   1641 		name[ni] = va_arg(ap, int);
   1642 		/*
   1643 		 * sorry, this is not supported from here
   1644 		 */
   1645 		if (name[ni] == CTL_CREATESYM)
   1646 			return (EINVAL);
   1647 	} while (name[ni] != CTL_EOL && name[ni] != CTL_CREATE);
   1648 	namelen = ni + (name[ni] == CTL_CREATE ? 1 : 0);
   1649 	va_end(ap);
   1650 
   1651 	/*
   1652 	 * what's it called
   1653 	 */
   1654 	if (strlcpy(nnode.sysctl_name, namep, sizeof(nnode.sysctl_name)) >
   1655 	    sizeof(nnode.sysctl_name))
   1656 		return (ENAMETOOLONG);
   1657 
   1658 	/*
   1659 	 * cons up the description of the new node
   1660 	 */
   1661 	nnode.sysctl_num = name[namelen - 1];
   1662 	name[namelen - 1] = CTL_CREATE;
   1663 	nnode.sysctl_size = newlen;
   1664 	nnode.sysctl_flags = flags;
   1665 	if (type == CTLTYPE_NODE) {
   1666 		nnode.sysctl_csize = 0;
   1667 		nnode.sysctl_clen = 0;
   1668 		nnode.sysctl_child = NULL;
   1669 		if (flags & CTLFLAG_ALIAS)
   1670 			nnode.sysctl_alias = qv;
   1671 	}
   1672 	else if (flags & CTLFLAG_IMMEDIATE) {
   1673 		switch (type) {
   1674 		case CTLTYPE_INT:
   1675 			nnode.sysctl_idata = qv;
   1676 			break;
   1677 		case CTLTYPE_QUAD:
   1678 			nnode.sysctl_qdata = qv;
   1679 			break;
   1680 		default:
   1681 			return (EINVAL);
   1682 		}
   1683 	}
   1684 	else {
   1685 		nnode.sysctl_data = newp;
   1686 	}
   1687 	nnode.sysctl_func = func;
   1688 	nnode.sysctl_parent = NULL;
   1689 	nnode.sysctl_ver = 0;
   1690 
   1691 	/*
   1692 	 * initialize lock state -- we need locks if the main tree has
   1693 	 * been marked as complete, but since we could be called from
   1694 	 * either there, or from a device driver (say, at device
   1695 	 * insertion), or from an lkm (at lkm load time, say), we
   1696 	 * don't really want to "wait"...
   1697 	 */
   1698 	error = sysctl_lock(NULL, NULL, 0);
   1699 	if (error)
   1700 		return (error);
   1701 
   1702 	/*
   1703 	 * locate the prospective parent of the new node, and if we
   1704 	 * find it, add the new node.
   1705 	 */
   1706 	sz = sizeof(onode);
   1707 	pnode = rnode ? *rnode : NULL;
   1708 	error = sysctl_locate(NULL, &name[0], namelen - 1, &pnode, &ni);
   1709 	if (error) {
   1710 		sysctl_unlock(NULL);
   1711 		return (error);
   1712 	}
   1713 	error = sysctl_create(&name[ni], namelen - ni, &onode, &sz,
   1714 			      &nnode, sizeof(nnode), &name[0], NULL,
   1715 			      pnode);
   1716 
   1717 	/*
   1718 	 * unfortunately the node we wanted to create is already
   1719 	 * there.  if the node that's already there is a reasonable
   1720 	 * facsimile of the node we wanted to create, just pretend
   1721 	 * (for the caller's benefit) that we managed to create the
   1722 	 * node they wanted.
   1723 	 */
   1724 	if (error == EEXIST) {
   1725 		/* name is the same as requested... */
   1726 		if (strcmp(nnode.sysctl_name, onode.sysctl_name) == 0 &&
   1727 		    /* they want the same function... */
   1728 		    nnode.sysctl_func == onode.sysctl_func &&
   1729 		    /* number is the same as requested, or... */
   1730 		    (nnode.sysctl_num == onode.sysctl_num ||
   1731 		     /* they didn't pick a number... */
   1732 		     nnode.sysctl_num == CTL_CREATE)) {
   1733 			/*
   1734 			 * collision here from trying to create
   1735 			 * something that already existed; let's give
   1736 			 * our customers a hand and tell them they got
   1737 			 * what they wanted.
   1738 			 */
   1739 #ifdef SYSCTL_DEBUG_CREATE
   1740 			printf("cleared\n");
   1741 #endif /* SYSCTL_DEBUG_CREATE */
   1742 			error = 0;
   1743 		}
   1744 	}
   1745 
   1746 	if (error == 0 &&
   1747 	    (cnode != NULL || log != NULL)) {
   1748 		/*
   1749 		 * sysctl_create() gave us back a copy of the node,
   1750 		 * but we need to know where it actually is...
   1751 		 */
   1752 		pnode = rnode ? *rnode : NULL;
   1753 		error = sysctl_locate(NULL, &name[0], namelen - 1, &pnode, &ni);
   1754 
   1755 		/*
   1756 		 * manual scan of last layer so that aliased nodes
   1757 		 * aren't followed.
   1758 		 */
   1759 		if (error == 0) {
   1760 			for (ni = 0; ni < pnode->sysctl_clen; ni++)
   1761 				if (pnode->sysctl_child[ni].sysctl_num ==
   1762 				    onode.sysctl_num)
   1763 					break;
   1764 			if (ni < pnode->sysctl_clen)
   1765 				pnode = &pnode->sysctl_child[ni];
   1766 			else
   1767 				error = ENOENT;
   1768 		}
   1769 
   1770 		/*
   1771 		 * not expecting an error here, but...
   1772 		 */
   1773 		if (error == 0) {
   1774 			if (log != NULL)
   1775 				sysctl_log_add(log, pnode);
   1776 			if (cnode != NULL)
   1777 				*cnode = pnode;
   1778 		}
   1779 		else {
   1780 			printf("sysctl_create succeeded but node not found?!\n");
   1781 			/*
   1782 			 *  confusing, but the create said it
   1783 			 * succeeded, so...
   1784 			 */
   1785 			error = 0;
   1786 		}
   1787 	}
   1788 
   1789 	/*
   1790 	 * now it should be safe to release the lock state.  note that
   1791 	 * the pointer to the newly created node being passed back may
   1792 	 * not be "good" for very long.
   1793 	 */
   1794 	sysctl_unlock(NULL);
   1795 
   1796 	if (error != 0) {
   1797 		printf("sysctl_createv: sysctl_create(%s) returned %d\n",
   1798 		       nnode.sysctl_name, error);
   1799 #if 0
   1800 		if (error != ENOENT)
   1801 			sysctl_dump(&onode);
   1802 #endif
   1803 	}
   1804 
   1805 	return (error);
   1806 }
   1807 
   1808 int
   1809 sysctl_destroyv(struct sysctlnode *rnode, ...)
   1810 {
   1811 	va_list ap;
   1812 	int error, name[CTL_MAXNAME], namelen, ni;
   1813 	struct sysctlnode *pnode, *node, dnode;
   1814 	size_t sz;
   1815 
   1816 	va_start(ap, rnode);
   1817 	namelen = 0;
   1818 	ni = 0;
   1819 	do {
   1820 		if (ni == CTL_MAXNAME)
   1821 			return (ENAMETOOLONG);
   1822 		name[ni] = va_arg(ap, int);
   1823 	} while (name[ni++] != CTL_EOL);
   1824 	namelen = ni - 1;
   1825 	va_end(ap);
   1826 
   1827 	/*
   1828 	 * i can't imagine why we'd be destroying a node when the tree
   1829 	 * wasn't complete, but who knows?
   1830 	 */
   1831 	error = sysctl_lock(NULL, NULL, 0);
   1832 	if (error)
   1833 		return (error);
   1834 
   1835 	/*
   1836 	 * where is it?
   1837 	 */
   1838 	node = rnode;
   1839 	error = sysctl_locate(NULL, &name[0], namelen - 1, &node, &ni);
   1840 	if (error) {
   1841 		/* they want it gone and it's not there, so... */
   1842 		sysctl_unlock(NULL);
   1843 		return (error == ENOENT ? 0 : error);
   1844 	}
   1845 
   1846 	/*
   1847 	 * set up the deletion
   1848 	 */
   1849 	pnode = node;
   1850 	node = &dnode;
   1851 	memset(&dnode, 0, sizeof(dnode));
   1852 	dnode.sysctl_num = name[namelen - 1];
   1853 
   1854 	/*
   1855 	 * we found it, now let's nuke it
   1856 	 */
   1857 	name[namelen - 1] = CTL_DESTROY;
   1858 	sz = 0;
   1859 	error = sysctl_destroy(&name[namelen - 1], 1, NULL, &sz,
   1860 			       node, sizeof(*node), &name[0], NULL,
   1861 			       pnode);
   1862 	if (error == ENOTEMPTY)
   1863 		/*
   1864 		 * think of trying to delete "foo" when "foo.bar"
   1865 		 * (which someone else put there) is still in
   1866 		 * existence
   1867 		 */
   1868 		error = 0;
   1869 
   1870         sysctl_unlock(NULL);
   1871 
   1872 	return (error);
   1873 }
   1874 
   1875 #if 0
   1876 /*
   1877  * ********************************************************************
   1878  * the dump routine.  i haven't yet decided how (if at all) i'll call
   1879  * this from userland when it's in the kernel.
   1880  * ********************************************************************
   1881  */
   1882 static const char *
   1883 sf(int f)
   1884 {
   1885 	static char s[256];
   1886 	char *c;
   1887 
   1888 	s[0] = '\0';
   1889 	c = "";
   1890 
   1891 #define print_flag(_f, _s, _c, _q, _x) \
   1892 	if (((_f) & (__CONCAT(CTLFLAG_,_x))) == (__CONCAT(CTLFLAG_,_q))) { \
   1893 		strlcat((_s), (_c), sizeof(_s)); \
   1894 		strlcat((_s), __STRING(_q), sizeof(_s)); \
   1895 		(_c) = ","; \
   1896 		(_f) &= ~__CONCAT(CTLFLAG_,_x); \
   1897 	}
   1898 
   1899 	print_flag(f, s, c, READONLY,  READWRITE);
   1900 	print_flag(f, s, c, READONLY1, READWRITE);
   1901 	print_flag(f, s, c, READONLY2, READWRITE);
   1902 	print_flag(f, s, c, READWRITE, READWRITE);
   1903 	print_flag(f, s, c, ANYWRITE,  ANYWRITE);
   1904 	print_flag(f, s, c, PRIVATE,   PRIVATE);
   1905 	print_flag(f, s, c, PERMANENT, PERMANENT);
   1906 	print_flag(f, s, c, OWNDATA,   OWNDATA);
   1907 	print_flag(f, s, c, IMMEDIATE, IMMEDIATE);
   1908 	print_flag(f, s, c, HEX,       HEX);
   1909 	print_flag(f, s, c, ROOT,      ROOT);
   1910 	print_flag(f, s, c, ANYNUMBER, ANYNUMBER);
   1911 	print_flag(f, s, c, HIDDEN,    HIDDEN);
   1912 	print_flag(f, s, c, ALIAS,     ALIAS);
   1913 #undef print_flag
   1914 
   1915 	if (f) {
   1916 		char foo[9];
   1917 		snprintf(foo, sizeof(foo), "%x", f);
   1918 		strlcat(s, c, sizeof(s));
   1919 		strlcat(s, foo, sizeof(s));
   1920 	}
   1921 
   1922 	return (s);
   1923 }
   1924 
   1925 static const char *
   1926 st(int t)
   1927 {
   1928 
   1929 	switch (t) {
   1930 	case CTLTYPE_NODE:
   1931 		return "NODE";
   1932 	case CTLTYPE_INT:
   1933 		return "INT";
   1934 	case CTLTYPE_STRING:
   1935 		return "STRING";
   1936 	case CTLTYPE_QUAD:
   1937 		return "QUAD";
   1938 	case CTLTYPE_STRUCT:
   1939 		return "STRUCT";
   1940 	}
   1941 
   1942 	return "???";
   1943 }
   1944 
   1945 void
   1946 sysctl_dump(const struct sysctlnode *d)
   1947 {
   1948 	static char nmib[64], smib[256];
   1949 	static int indent;
   1950 	struct sysctlnode *n;
   1951 	char *np, *sp, tmp[20];
   1952 	int i;
   1953 
   1954 	if (d == NULL)
   1955 		return;
   1956 
   1957 	np = &nmib[strlen(nmib)];
   1958 	sp = &smib[strlen(smib)];
   1959 
   1960 	if (!(d->sysctl_flags & CTLFLAG_ROOT)) {
   1961 		snprintf(tmp, sizeof(tmp), "%d", d->sysctl_num);
   1962 		strcat(nmib, ".");
   1963 		strcat(smib, ".");
   1964 		strcat(nmib, tmp);
   1965 		strcat(smib, d->sysctl_name);
   1966 		printf("%s -> %s (%d)\n", &nmib[1], &smib[1],
   1967 		       SYSCTL_TYPE(d->sysctl_flags));
   1968 	}
   1969 
   1970 	if (1) {
   1971 		printf("%*s%p:\tsysctl_name  [%s]\n", indent, "",
   1972 		       d, d->sysctl_name);
   1973 		printf("%*s\t\tsysctl_num    %d\n",   indent, "",
   1974 		       d->sysctl_num);
   1975 		printf("%*s\t\tsysctl_flags  %x (flags=%x<%s> type=%d<%s> "
   1976 		       "size=%zu)\n",
   1977 		       indent, "", d->sysctl_flags,
   1978 		       SYSCTL_FLAGS(d->sysctl_flags),
   1979 		       sf(SYSCTL_FLAGS(d->sysctl_flags)),
   1980 		       SYSCTL_TYPE(d->sysctl_flags),
   1981 		       st(SYSCTL_TYPE(d->sysctl_flags)),
   1982 		       d->sysctl_size);
   1983 		if (SYSCTL_TYPE(d->sysctl_flags) == CTLTYPE_NODE) {
   1984 			printf("%*s\t\tsysctl_csize  %d\n",   indent, "",
   1985 			       d->sysctl_csize);
   1986 			printf("%*s\t\tsysctl_clen   %d\n",   indent, "",
   1987 			       d->sysctl_clen);
   1988 			printf("%*s\t\tsysctl_child  %p\n",   indent, "",
   1989 			       d->sysctl_child);
   1990 		}
   1991 		else
   1992 			printf("%*s\t\tsysctl_data   %p\n",   indent, "",
   1993 			       d->sysctl_data);
   1994 		printf("%*s\t\tsysctl_func   %p\n",   indent, "",
   1995 		       d->sysctl_func);
   1996 		printf("%*s\t\tsysctl_parent %p\n",   indent, "",
   1997 		       d->sysctl_parent);
   1998 		printf("%*s\t\tsysctl_ver    %d\n",   indent, "",
   1999 		       d->sysctl_ver);
   2000 	}
   2001 
   2002 	if (SYSCTL_TYPE(d->sysctl_flags) == CTLTYPE_NODE) {
   2003 		indent += 8;
   2004 		n = d->sysctl_child;
   2005 		for (i = 0; i < d->sysctl_clen; i++) {
   2006 			sysctl_dump(&n[i]);
   2007 		}
   2008 		indent -= 8;
   2009 	}
   2010 
   2011 	np[0] = '\0';
   2012 	sp[0] = '\0';
   2013 }
   2014 #endif /* 0 */
   2015 
   2016 /*
   2017  * ********************************************************************
   2018  * Deletes an entire n-ary tree.  Not recommended unless you know why
   2019  * you're doing it.  Personally, I don't know why you'd even think
   2020  * about it.
   2021  * ********************************************************************
   2022  */
   2023 void
   2024 sysctl_free(struct sysctlnode *rnode)
   2025 {
   2026 	struct sysctlnode *node, *pnode;
   2027 
   2028 	if (SYSCTL_VERS(rnode->sysctl_flags) != SYSCTL_VERSION) {
   2029 		printf("sysctl_free: rnode %p wrong version\n", rnode);
   2030 		return;
   2031 	}
   2032 
   2033 	if (rnode == NULL)
   2034 		rnode = &sysctl_root;
   2035 	pnode = rnode;
   2036 
   2037 	node = pnode->sysctl_child;
   2038 	do {
   2039 		while (node != NULL && pnode->sysctl_csize > 0) {
   2040 			while (node <
   2041 			       &pnode->sysctl_child[pnode->sysctl_clen] &&
   2042 			       (SYSCTL_TYPE(node->sysctl_flags) !=
   2043 				CTLTYPE_NODE ||
   2044 				node->sysctl_csize == 0)) {
   2045 				if (SYSCTL_FLAGS(node->sysctl_flags) &
   2046 				    CTLFLAG_OWNDATA) {
   2047 					if (node->sysctl_data != NULL) {
   2048 						FREE(node->sysctl_data,
   2049 						     M_SYSCTLDATA);
   2050 						node->sysctl_data = NULL;
   2051 					}
   2052 				}
   2053 				node++;
   2054 			}
   2055 			if (node < &pnode->sysctl_child[pnode->sysctl_clen]) {
   2056 				pnode = node;
   2057 				node = node->sysctl_child;
   2058 			}
   2059 			else
   2060 				break;
   2061 		}
   2062 		if (pnode->sysctl_child != NULL)
   2063 			FREE(pnode->sysctl_child, M_SYSCTLNODE);
   2064 		pnode->sysctl_clen = 0;
   2065 		pnode->sysctl_csize = 0;
   2066 		pnode->sysctl_child = NULL;
   2067 		node = pnode;
   2068 		pnode = node->sysctl_parent;
   2069 	} while (pnode != NULL && node != rnode);
   2070 }
   2071 
   2072 int
   2073 sysctl_log_add(struct sysctllog **logp, struct sysctlnode *node)
   2074 {
   2075 	int name[CTL_MAXNAME], namelen, i;
   2076 	struct sysctlnode *pnode;
   2077 	struct sysctllog *log;
   2078 
   2079 	if (node->sysctl_flags & CTLFLAG_PERMANENT)
   2080 		return (0);
   2081 
   2082 	if (logp == NULL)
   2083 		return (0);
   2084 
   2085 	if (*logp == NULL) {
   2086 		MALLOC(log, struct sysctllog *, sizeof(struct sysctllog),
   2087 		       M_SYSCTLDATA, M_WAITOK|M_CANFAIL);
   2088 		if (log == NULL) {
   2089 			/* XXX print error message? */
   2090 			return (-1);
   2091 		}
   2092 		MALLOC(log->log_num, int *, 16 * sizeof(int),
   2093 		       M_SYSCTLDATA, M_WAITOK|M_CANFAIL);
   2094 		if (log->log_num == NULL) {
   2095 			/* XXX print error message? */
   2096 			free(log, M_SYSCTLDATA);
   2097 			return (-1);
   2098 		}
   2099 		memset(log->log_num, 0, 16 * sizeof(int));
   2100 		log->log_root = NULL;
   2101 		log->log_size = 16;
   2102 		log->log_left = 16;
   2103 		*logp = log;
   2104 	}
   2105 	else
   2106 		log = *logp;
   2107 
   2108 	/*
   2109 	 * check that the root is proper.  it's okay to record the
   2110 	 * address of the root of a tree.  it's the only thing that's
   2111 	 * guaranteed not to shift around as nodes come and go.
   2112 	 */
   2113 	if (log->log_root == NULL)
   2114 		log->log_root = sysctl_rootof(node);
   2115 	else if (log->log_root != sysctl_rootof(node)) {
   2116 		printf("sysctl: log %p root mismatch (%p)\n",
   2117 		       log->log_root, sysctl_rootof(node));
   2118 		return (-1);
   2119 	}
   2120 
   2121 	/*
   2122 	 * we will copy out name in reverse order
   2123 	 */
   2124 	for (pnode = node, namelen = 0;
   2125 	     pnode != NULL && !(pnode->sysctl_flags & CTLFLAG_ROOT);
   2126 	     pnode = pnode->sysctl_parent)
   2127 		name[namelen++] = pnode->sysctl_num;
   2128 
   2129 	/*
   2130 	 * do we have space?
   2131 	 */
   2132 	if (log->log_left < (namelen + 3))
   2133 		sysctl_log_realloc(log);
   2134 	if (log->log_left < (namelen + 3))
   2135 		return (-1);
   2136 
   2137 	/*
   2138 	 * stuff name in, then namelen, then node type, and finally,
   2139 	 * the version for non-node nodes.
   2140 	 */
   2141 	for (i = 0; i < namelen; i++)
   2142 		log->log_num[--log->log_left] = name[i];
   2143 	log->log_num[--log->log_left] = namelen;
   2144 	log->log_num[--log->log_left] = SYSCTL_TYPE(node->sysctl_flags);
   2145 	if (log->log_num[log->log_left] != CTLTYPE_NODE)
   2146 		log->log_num[--log->log_left] = node->sysctl_ver;
   2147 	else
   2148 		log->log_num[--log->log_left] = 0;
   2149 
   2150 	return (0);
   2151 }
   2152 
   2153 void
   2154 sysctl_teardown(struct sysctllog **logp)
   2155 {
   2156 	struct sysctlnode node, *rnode;
   2157 	struct sysctllog *log;
   2158 	uint namelen;
   2159 	int *name, t, v, error, ni;
   2160 	size_t sz;
   2161 
   2162 	if (logp == NULL || *logp == NULL)
   2163 		return;
   2164 	log = *logp;
   2165 
   2166 	error = sysctl_lock(NULL, NULL, 0);
   2167 	if (error)
   2168 		return;
   2169 
   2170 	memset(&node, 0, sizeof(node));
   2171 
   2172 	while (log->log_left < log->log_size) {
   2173 		KASSERT((log->log_left + 3 < log->log_size) &&
   2174 			(log->log_left + log->log_num[log->log_left + 2] <=
   2175 			 log->log_size));
   2176 		v = log->log_num[log->log_left++];
   2177 		t = log->log_num[log->log_left++];
   2178 		namelen = log->log_num[log->log_left++];
   2179 		name = &log->log_num[log->log_left];
   2180 
   2181 		node.sysctl_num = name[namelen - 1];
   2182 		node.sysctl_flags = t;
   2183 		node.sysctl_ver = v;
   2184 
   2185 		rnode = log->log_root;
   2186 		error = sysctl_locate(NULL, &name[0], namelen, &rnode, &ni);
   2187 		if (error == 0) {
   2188 			name[namelen - 1] = CTL_DESTROY;
   2189 			rnode = rnode->sysctl_parent;
   2190 			sz = 0;
   2191 			(void)sysctl_destroy(&name[namelen - 1], 1, NULL,
   2192 					     &sz, &node, sizeof(node),
   2193 					     &name[0], NULL, rnode);
   2194 		}
   2195 
   2196 		log->log_left += namelen;
   2197 	}
   2198 
   2199 	KASSERT(log->log_size == log->log_left);
   2200 	free(log->log_num, M_SYSCTLDATA);
   2201 	free(log, M_SYSCTLDATA);
   2202 	*logp = NULL;
   2203 
   2204 	sysctl_unlock(NULL);
   2205 }
   2206 
   2207 /*
   2208  * ********************************************************************
   2209  * old_sysctl -- A routine to bridge old-style internal calls to the
   2210  * new infrastructure.
   2211  * ********************************************************************
   2212  */
   2213 int
   2214 old_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp,
   2215 	   void *newp, size_t newlen, struct lwp *l)
   2216 {
   2217 	int error;
   2218 	size_t savelen = *oldlenp;
   2219 
   2220 	error = sysctl_lock(l, oldp, savelen);
   2221 	if (error)
   2222 		return (error);
   2223 	error = sysctl_dispatch(name, namelen, oldp, oldlenp,
   2224 				newp, newlen, name, l, NULL);
   2225 	sysctl_unlock(l);
   2226 	if (error == 0 && oldp != NULL && savelen < *oldlenp)
   2227 		error = ENOMEM;
   2228 
   2229 	return (error);
   2230 }
   2231 
   2232 /*
   2233  * ********************************************************************
   2234  * Section 4: Generic helper routines
   2235  * ********************************************************************
   2236  * "helper" routines that can do more finely grained access control,
   2237  * construct structures from disparate information, create the
   2238  * appearance of more nodes and sub-trees, etc.  for example, if
   2239  * CTL_PROC wanted a helper function, it could respond to a CTL_QUERY
   2240  * with a dynamically created list of nodes that represented the
   2241  * currently running processes at that instant.
   2242  * ********************************************************************
   2243  */
   2244 
   2245 /*
   2246  * first, a few generic helpers that provide:
   2247  *
   2248  * sysctl_needfunc()		a readonly interface that emits a warning
   2249  * sysctl_notavail()		returns EOPNOTSUPP (generic error)
   2250  * sysctl_null()		an empty return buffer with no error
   2251  */
   2252 int
   2253 sysctl_needfunc(SYSCTLFN_ARGS)
   2254 {
   2255 	int error;
   2256 
   2257 	printf("!!SYSCTL_NEEDFUNC!!\n");
   2258 
   2259 	if (newp != NULL || namelen != 0)
   2260 		return (EOPNOTSUPP);
   2261 
   2262 	error = 0;
   2263 	if (oldp != NULL)
   2264 		error = sysctl_copyout(l, rnode->sysctl_data, oldp,
   2265 				       MIN(rnode->sysctl_size, *oldlenp));
   2266 	*oldlenp = rnode->sysctl_size;
   2267 
   2268 	return (error);
   2269 }
   2270 
   2271 int
   2272 sysctl_notavail(SYSCTLFN_ARGS)
   2273 {
   2274 
   2275 	if (namelen == 1 && name[0] == CTL_QUERY)
   2276 		return (sysctl_query(SYSCTLFN_CALL(rnode)));
   2277 
   2278 	return (EOPNOTSUPP);
   2279 }
   2280 
   2281 int
   2282 sysctl_null(SYSCTLFN_ARGS)
   2283 {
   2284 
   2285 	*oldlenp = 0;
   2286 
   2287 	return (0);
   2288 }
   2289 
   2290 /*
   2291  * ********************************************************************
   2292  * Section 5: The machinery that makes it all go
   2293  * ********************************************************************
   2294  * Memory "manglement" routines.  Not much to this, eh?
   2295  * ********************************************************************
   2296  */
   2297 static int
   2298 sysctl_alloc(struct sysctlnode *p, int x)
   2299 {
   2300 	int i;
   2301 	struct sysctlnode *n;
   2302 
   2303 	assert(p->sysctl_child == NULL);
   2304 
   2305 	if (x == 1)
   2306 		MALLOC(n, struct sysctlnode *,
   2307 		       sizeof(struct sysctlnode),
   2308 		       M_SYSCTLNODE, M_WAITOK|M_CANFAIL);
   2309 	else
   2310 		MALLOC(n, struct sysctlnode *,
   2311 		       SYSCTL_DEFSIZE * sizeof(struct sysctlnode),
   2312 		       M_SYSCTLNODE, M_WAITOK|M_CANFAIL);
   2313 	if (n == NULL)
   2314 		return (ENOMEM);
   2315 
   2316 	if (x == 1) {
   2317 		memset(n, 0, sizeof(struct sysctlnode));
   2318 		p->sysctl_csize = 1;
   2319 	}
   2320 	else {
   2321 		memset(n, 0, SYSCTL_DEFSIZE * sizeof(struct sysctlnode));
   2322 		p->sysctl_csize = SYSCTL_DEFSIZE;
   2323 	}
   2324 	p->sysctl_clen = 0;
   2325 
   2326 	for (i = 0; i < p->sysctl_csize; i++)
   2327 		n[i].sysctl_parent = p;
   2328 
   2329 	p->sysctl_child = n;
   2330 	return (0);
   2331 }
   2332 
   2333 static int
   2334 sysctl_realloc(struct sysctlnode *p)
   2335 {
   2336 	int i, j;
   2337 	struct sysctlnode *n;
   2338 
   2339 	assert(p->sysctl_csize == p->sysctl_clen);
   2340 
   2341 	/*
   2342 	 * how many do we have...how many should we make?
   2343 	 */
   2344 	i = p->sysctl_clen;
   2345 	n = malloc(2 * i * sizeof(struct sysctlnode), M_SYSCTLNODE,
   2346 		   M_WAITOK|M_CANFAIL);
   2347 	if (n == NULL)
   2348 		return (ENOMEM);
   2349 
   2350 	/*
   2351 	 * move old children over...initialize new children
   2352 	 */
   2353 	memcpy(n, p->sysctl_child, i * sizeof(struct sysctlnode));
   2354 	memset(&n[i], 0, i * sizeof(struct sysctlnode));
   2355 	p->sysctl_csize = 2 * i;
   2356 
   2357 	/*
   2358 	 * reattach moved (and new) children to parent; if a moved
   2359 	 * child node has children, reattach the parent pointers of
   2360 	 * grandchildren
   2361 	 */
   2362         for (i = 0; i < p->sysctl_csize; i++) {
   2363                 n[i].sysctl_parent = p;
   2364 		if (n[i].sysctl_child != NULL) {
   2365 			for (j = 0; j < n[i].sysctl_csize; j++)
   2366 				n[i].sysctl_child[j].sysctl_parent = &n[i];
   2367 		}
   2368 	}
   2369 
   2370 	/*
   2371 	 * get out with the old and in with the new
   2372 	 */
   2373 	FREE(p->sysctl_child, M_SYSCTLNODE);
   2374 	p->sysctl_child = n;
   2375 
   2376 	return (0);
   2377 }
   2378 
   2379 static int
   2380 sysctl_log_realloc(struct sysctllog *log)
   2381 {
   2382 	int *n, s, d;
   2383 
   2384 	s = log->log_size * 2;
   2385 	d = log->log_size;
   2386 
   2387 	n = malloc(s * sizeof(int), M_SYSCTLDATA, M_WAITOK|M_CANFAIL);
   2388 	if (n == NULL)
   2389 		return (-1);
   2390 
   2391 	memset(n, 0, s * sizeof(int));
   2392 	memcpy(&n[d], log->log_num, d * sizeof(int));
   2393 	free(log->log_num, M_SYSCTLDATA);
   2394 	log->log_num = n;
   2395 	if (d)
   2396 		log->log_left += d;
   2397 	else
   2398 		log->log_left = s;
   2399 	log->log_size = s;
   2400 
   2401 	return (0);
   2402 }
   2403 
   2404 /*
   2405  * ********************************************************************
   2406  * Section 6: Conversion between API versions wrt the sysctlnode
   2407  * ********************************************************************
   2408  */
   2409 static int
   2410 sysctl_cvt_in(struct lwp *l, int *vp, const void *i, size_t sz,
   2411 	      struct sysctlnode *node)
   2412 {
   2413 	int error, flags;
   2414 
   2415 	if (i == NULL) {
   2416 		memset(node, 0, sizeof(*node));
   2417 		*vp = SYSCTL_VERS_0;
   2418 		return (0);
   2419 	}
   2420 
   2421 	if (sz < sizeof(flags))
   2422 		return (EINVAL);
   2423 
   2424 	error = sysctl_copyin(l, i, &flags, sizeof(flags));
   2425 	if (error)
   2426 		return (error);
   2427 
   2428 	if (sz == sizeof(*node) &&
   2429 	    SYSCTL_VERS(flags) == SYSCTL_VERSION) {
   2430 		error = sysctl_copyin(l, i, node, sizeof(*node));
   2431 		if (error)
   2432 			return (error);
   2433 		*vp = SYSCTL_VERSION;
   2434 		return (0);
   2435 	}
   2436 
   2437 	return (EINVAL);
   2438 }
   2439 
   2440 static int
   2441 sysctl_cvt_out(struct lwp *l, int v, const struct sysctlnode *i,
   2442 	       void *ovp, size_t left, size_t *szp)
   2443 {
   2444 	size_t sz = sizeof(*i);
   2445 	const void *src = i;
   2446 	int error;
   2447 
   2448 	switch (v) {
   2449 	case SYSCTL_VERSION:
   2450 		/* nothing more to do here */
   2451 		break;
   2452 	}
   2453 
   2454 	if (ovp != NULL && left >= sz) {
   2455 		error = sysctl_copyout(l, src, ovp, sz);
   2456 		if (error)
   2457 			return (error);
   2458 	}
   2459 
   2460 	if (szp != NULL)
   2461 		*szp = sz;
   2462 
   2463 	return (0);
   2464 }
   2465