xref: /dragonfly/sys/kern/kern_sysctl.c (revision 8f2ce533)
1 /*-
2  * Copyright (c) 1982, 1986, 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Mike Karels at Berkeley Software Design, Inc.
7  *
8  * Quite extensively rewritten by Poul-Henning Kamp of the FreeBSD
9  * project, to make these variables more userfriendly.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. Neither the name of the University nor the names of its contributors
20  *    may be used to endorse or promote products derived from this software
21  *    without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  *
35  *	@(#)kern_sysctl.c	8.4 (Berkeley) 4/14/94
36  * $FreeBSD: src/sys/kern/kern_sysctl.c,v 1.92.2.9 2003/05/01 22:48:09 trhodes Exp $
37  */
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/kernel.h>
42 #include <sys/buf.h>
43 #include <sys/sysctl.h>
44 #include <sys/malloc.h>
45 #include <sys/proc.h>
46 #include <sys/priv.h>
47 #include <sys/sysmsg.h>
48 #include <sys/lock.h>
49 #include <sys/sbuf.h>
50 
51 #include <vm/vm.h>
52 #include <vm/vm_extern.h>
53 
54 #include "opt_sysctl.h"
55 
56 static MALLOC_DEFINE(M_SYSCTL, "sysctl", "sysctl internal magic");
57 static MALLOC_DEFINE(M_SYSCTLOID, "sysctloid", "sysctl dynamic oids");
58 
59 int sysctl_debugx = 0;
60 SYSCTL_INT(_debug, OID_AUTO, sysctl, CTLFLAG_RW, &sysctl_debugx, 0, "");
61 
62 /*
63  * The sysctllock protects the MIB tree.  It also protects sysctl
64  * contexts used with dynamic sysctls.  The sysctl_register_oid() and
65  * sysctl_unregister_oid() routines require the sysctllock to already
66  * be held, so the sysctl_lock() and sysctl_unlock() routines are
67  * provided for the few places in the kernel which need to use that
68  * API rather than using the dynamic API.  Use of the dynamic API is
69  * strongly encouraged for most code.
70  */
71 
72 static int	sysctl_root(SYSCTL_HANDLER_ARGS);
73 static void	sysctl_register_oid_int(struct sysctl_oid *oipd);
74 static void	sysctl_unregister_oid_int(struct sysctl_oid *oipd);
75 
76 struct sysctl_oid_list sysctl__children; /* root list */
77 
78 static int	sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del,
79 		    int recurse);
80 
81 static struct sysctl_oid *
82 sysctl_find_oidname(const char *name, struct sysctl_oid_list *list, int lock)
83 {
84 	struct sysctl_oid *oidp;
85 
86 	SLIST_FOREACH(oidp, list, oid_link) {
87 		if (strcmp(oidp->oid_name, name) == 0) {
88 			break;
89 		}
90 	}
91 	return (oidp);
92 }
93 
94 /*
95  * Initialization of the MIB tree.
96  *
97  * Order by number in each list.
98  */
99 
100 void
101 sysctl_register_oid(struct sysctl_oid *oidp)
102 {
103 	SYSCTL_XLOCK();
104 	sysctl_register_oid_int(oidp);
105 	SYSCTL_XUNLOCK();
106 }
107 
108 static void
109 sysctl_register_oid_int(struct sysctl_oid *oidp)
110 {
111 	struct sysctl_oid_list *parent = oidp->oid_parent;
112 	struct sysctl_oid *p;
113 	struct sysctl_oid *q;
114 
115 	/*
116 	 * Finish initialization from sysctl_set or add.
117 	 */
118 	lockinit(&oidp->oid_lock, "oidlk", 0, LK_CANRECURSE);
119 
120 	/*
121 	 * First check if another oid with the same name already
122 	 * exists in the parent's list.
123 	 */
124 	p = sysctl_find_oidname(oidp->oid_name, parent, 0);
125 	if (p != NULL) {
126 		if ((p->oid_kind & CTLTYPE) == CTLTYPE_NODE)
127 			p->oid_refcnt++;
128 		else
129 			kprintf("can't re-use a leaf (%s)!\n", p->oid_name);
130 		return;
131 	}
132 
133 	/*
134 	 * If this oid has a number OID_AUTO, give it a number which
135 	 * is greater than any current oid.  Make sure it is at least
136 	 * 256 to leave space for pre-assigned oid numbers.
137 	 */
138 	if (oidp->oid_number == OID_AUTO) {
139 		int newoid = 0x100;	/* minimum AUTO oid */
140 
141 		/*
142 		 * Adjust based on highest oid in parent list
143 		 */
144 		SLIST_FOREACH(p, parent, oid_link) {
145 			if (newoid <= p->oid_number)
146 				newoid = p->oid_number + 1;
147 		}
148 		oidp->oid_number = newoid;
149 	}
150 
151 	/*
152 	 * Insert the oid into the parent's list in order.
153 	 */
154 	q = NULL;
155 	SLIST_FOREACH(p, parent, oid_link) {
156 		if (oidp->oid_number < p->oid_number)
157 			break;
158 		q = p;
159 	}
160 	if (q)
161 		SLIST_INSERT_AFTER(q, oidp, oid_link);
162 	else
163 		SLIST_INSERT_HEAD(parent, oidp, oid_link);
164 }
165 
166 void
167 sysctl_unregister_oid(struct sysctl_oid *oidp)
168 {
169 	SYSCTL_XLOCK();
170 	sysctl_unregister_oid_int(oidp);
171 	SYSCTL_XUNLOCK();
172 }
173 
174 static void
175 sysctl_unregister_oid_int(struct sysctl_oid *oidp)
176 {
177 	struct sysctl_oid *p;
178 
179 	if (oidp->oid_number == OID_AUTO)
180 		panic("Trying to unregister OID_AUTO entry: %p", oidp);
181 
182 	SLIST_FOREACH(p, oidp->oid_parent, oid_link) {
183 		if (p != oidp)
184 			continue;
185 		SLIST_REMOVE(oidp->oid_parent, oidp, sysctl_oid, oid_link);
186 		return;
187 	}
188 
189 	/*
190 	 * This can happen when a module fails to register and is
191 	 * being unloaded afterwards.  It should not be a panic()
192 	 * for normal use.
193 	 */
194 	kprintf("%s: failed to unregister sysctl\n", __func__);
195 }
196 
197 /* Initialize a new context to keep track of dynamically added sysctls. */
198 int
199 sysctl_ctx_init(struct sysctl_ctx_list *c)
200 {
201 	if (c == NULL)
202 		return(EINVAL);
203 	TAILQ_INIT(c);
204 	return(0);
205 }
206 
207 /* Free the context, and destroy all dynamic oids registered in this context */
208 int
209 sysctl_ctx_free(struct sysctl_ctx_list *clist)
210 {
211 	struct sysctl_ctx_entry *e, *e1;
212 	int error;
213 
214 	error = 0;
215 	/*
216 	 * First perform a "dry run" to check if it's ok to remove oids.
217 	 * XXX FIXME
218 	 * XXX This algorithm is a hack. But I don't know any
219 	 * XXX better solution for now...
220 	 */
221 	SYSCTL_XLOCK();
222 	TAILQ_FOREACH(e, clist, link) {
223 		error = sysctl_remove_oid_locked(e->entry, 0, 0);
224 		if (error)
225 			break;
226 	}
227 	/*
228 	 * Restore deregistered entries, either from the end,
229 	 * or from the place where error occured.
230 	 * e contains the entry that was not unregistered
231 	 */
232 	if (error)
233 		e1 = TAILQ_PREV(e, sysctl_ctx_list, link);
234 	else
235 		e1 = TAILQ_LAST(clist, sysctl_ctx_list);
236 	while (e1 != NULL) {
237 		sysctl_register_oid(e1->entry);
238 		e1 = TAILQ_PREV(e1, sysctl_ctx_list, link);
239 	}
240 	if (error) {
241 		SYSCTL_XUNLOCK();
242 		return(EBUSY);
243 	}
244 	/* Now really delete the entries */
245 	e = TAILQ_FIRST(clist);
246 	while (e != NULL) {
247 		e1 = TAILQ_NEXT(e, link);
248 		error = sysctl_remove_oid_locked(e->entry, 1, 0);
249 		if (error)
250 			panic("sysctl_remove_oid: corrupt tree, entry: %s",
251 			    e->entry->oid_name);
252 		kfree(e, M_SYSCTLOID);
253 		e = e1;
254 	}
255 	SYSCTL_XUNLOCK();
256 	return (error);
257 }
258 
259 /* Add an entry to the context */
260 struct sysctl_ctx_entry *
261 sysctl_ctx_entry_add(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
262 {
263 	struct sysctl_ctx_entry *e;
264 
265 	SYSCTL_ASSERT_LOCKED();
266 	if (clist == NULL || oidp == NULL)
267 		return(NULL);
268 	e = kmalloc(sizeof(struct sysctl_ctx_entry), M_SYSCTLOID, M_WAITOK);
269 	e->entry = oidp;
270 	TAILQ_INSERT_HEAD(clist, e, link);
271 	return (e);
272 }
273 
274 /* Find an entry in the context */
275 struct sysctl_ctx_entry *
276 sysctl_ctx_entry_find(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
277 {
278 	struct sysctl_ctx_entry *e;
279 
280 	SYSCTL_ASSERT_LOCKED();
281 	if (clist == NULL || oidp == NULL)
282 		return(NULL);
283 	TAILQ_FOREACH(e, clist, link) {
284 		if(e->entry == oidp)
285 			return(e);
286 	}
287 	return (e);
288 }
289 
290 /*
291  * Delete an entry from the context.
292  * NOTE: this function doesn't free oidp! You have to remove it
293  * with sysctl_remove_oid().
294  */
295 int
296 sysctl_ctx_entry_del(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
297 {
298 	struct sysctl_ctx_entry *e;
299 
300 	if (clist == NULL || oidp == NULL)
301 		return (EINVAL);
302 	SYSCTL_XLOCK();
303 	e = sysctl_ctx_entry_find(clist, oidp);
304 	if (e != NULL) {
305 		TAILQ_REMOVE(clist, e, link);
306 		SYSCTL_XUNLOCK();
307 		kfree(e, M_SYSCTLOID);
308 		return (0);
309 	} else {
310 		SYSCTL_XUNLOCK();
311 		return (ENOENT);
312 	}
313 }
314 
315 /*
316  * Remove dynamically created sysctl trees.
317  * oidp - top of the tree to be removed
318  * del - if 0 - just deregister, otherwise free up entries as well
319  * recurse - if != 0 traverse the subtree to be deleted
320  */
321 int
322 sysctl_remove_oid(struct sysctl_oid *oidp, int del, int recurse)
323 {
324 	int error;
325 
326 	SYSCTL_XLOCK();
327 	error = sysctl_remove_oid_locked(oidp, del, recurse);
328 	SYSCTL_XUNLOCK();
329 	return (error);
330 }
331 
332 static int
333 sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del, int recurse)
334 {
335 	struct sysctl_oid *p, *tmp;
336 	int error;
337 
338 	SYSCTL_ASSERT_LOCKED();
339 	if (oidp == NULL)
340 		return(EINVAL);
341 	if ((oidp->oid_kind & CTLFLAG_DYN) == 0) {
342 		kprintf("can't remove non-dynamic nodes!\n");
343 		return (EINVAL);
344 	}
345 	/*
346 	 * WARNING: normal method to do this should be through
347 	 * sysctl_ctx_free(). Use recursing as the last resort
348 	 * method to purge your sysctl tree of leftovers...
349 	 * However, if some other code still references these nodes,
350 	 * it will panic.
351 	 */
352 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
353 		if (oidp->oid_refcnt == 1) {
354 			SLIST_FOREACH_MUTABLE(p,
355 			    SYSCTL_CHILDREN(oidp), oid_link, tmp) {
356 				if (!recurse) {
357 					kprintf("Warning: failed attempt to "
358 					    "remove oid %s with child %s\n",
359 					    oidp->oid_name, p->oid_name);
360 					return (ENOTEMPTY);
361 				}
362 				error = sysctl_remove_oid_locked(p, del,
363 								 recurse);
364 				if (error)
365 					return (error);
366 			}
367 			if (del)
368 				kfree(SYSCTL_CHILDREN(oidp), M_SYSCTLOID);
369 		}
370 	}
371 	if (oidp->oid_refcnt > 1 ) {
372 		oidp->oid_refcnt--;
373 	} else {
374 		if (oidp->oid_refcnt == 0) {
375 			kprintf("Warning: bad oid_refcnt=%u (%s)!\n",
376 				oidp->oid_refcnt, oidp->oid_name);
377 			return (EINVAL);
378 		}
379 		sysctl_unregister_oid(oidp);
380 		if (del) {
381 			/*
382 			 * Wait for all threads running the handler to drain.
383 			 * This preserves the previous behavior when the
384 			 * sysctl lock was held across a handler invocation,
385 			 * and is necessary for module unload correctness.
386 			 */
387 			while (oidp->oid_running > 0) {
388 				oidp->oid_kind |= CTLFLAG_DYING;
389 				tsleep_interlock(&oidp->oid_running, 0);
390 				SYSCTL_XUNLOCK();
391 				tsleep(&oidp->oid_running, PINTERLOCKED,
392 				       "oidrm", 0);
393 				SYSCTL_XLOCK();
394 			}
395 			if (oidp->oid_descr)
396 				kfree(__DECONST(char *, oidp->oid_descr),
397 				    M_SYSCTLOID);
398 			kfree(__DECONST(char *, oidp->oid_name), M_SYSCTLOID);
399 			lockuninit(&oidp->oid_lock);
400 			kfree(oidp, M_SYSCTLOID);
401 		}
402 	}
403 	return (0);
404 }
405 
406 int
407 sysctl_remove_name(struct sysctl_oid *parent, const char *name,
408     int del, int recurse)
409 {
410 	struct sysctl_oid *p, *tmp;
411 	int error;
412 
413 	error = ENOENT;
414 	SYSCTL_XLOCK();
415 	SLIST_FOREACH_MUTABLE(p, SYSCTL_CHILDREN(parent), oid_link, tmp) {
416 		if (strcmp(p->oid_name, name) == 0) {
417 			error = sysctl_remove_oid_locked(p, del, recurse);
418 			break;
419 		}
420 	}
421 	SYSCTL_XUNLOCK();
422 
423 	return (error);
424 }
425 
426 /*
427  * Create new sysctls at run time.
428  * clist may point to a valid context initialized with sysctl_ctx_init().
429  */
430 struct sysctl_oid *
431 sysctl_add_oid(struct sysctl_ctx_list *clist, struct sysctl_oid_list *parent,
432 	int number, const char *name, int kind, void *arg1, int arg2,
433 	int (*handler)(SYSCTL_HANDLER_ARGS), const char *fmt, const char *descr)
434 {
435 	struct sysctl_oid *oidp;
436 	ssize_t len;
437 	char *newname;
438 
439 	/* You have to hook up somewhere.. */
440 	if (parent == NULL)
441 		return(NULL);
442 	SYSCTL_XLOCK();
443 	/* Check if the node already exists, otherwise create it */
444 	oidp = sysctl_find_oidname(name, parent, 0);
445 	if (oidp != NULL) {
446 		if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
447 			oidp->oid_refcnt++;
448 			/* Update the context */
449 			if (clist != NULL)
450 				sysctl_ctx_entry_add(clist, oidp);
451 			SYSCTL_XUNLOCK();
452 			return (oidp);
453 		} else {
454 			kprintf("can't re-use a leaf (%s)!\n", name);
455 			SYSCTL_XUNLOCK();
456 			return (NULL);
457 		}
458 	}
459 	oidp = kmalloc(sizeof(struct sysctl_oid), M_SYSCTLOID,
460 		       M_WAITOK | M_ZERO);
461 	oidp->oid_parent = parent;
462 	SLIST_NEXT(oidp, oid_link) = NULL;
463 	oidp->oid_number = number;
464 	oidp->oid_refcnt = 1;
465 	len = strlen(name);
466 	newname = kmalloc(len + 1, M_SYSCTLOID, M_WAITOK);
467 	bcopy(name, newname, len + 1);
468 	newname[len] = '\0';
469 	oidp->oid_name = newname;
470 	oidp->oid_handler = handler;
471 	oidp->oid_kind = CTLFLAG_DYN | kind;
472 	if ((kind & CTLTYPE) == CTLTYPE_NODE) {
473 		struct sysctl_oid_list *children;
474 
475 		/* Allocate space for children */
476 		children = kmalloc(sizeof(*children), M_SYSCTLOID, M_WAITOK);
477 		SYSCTL_SET_CHILDREN(oidp, children);
478 		SLIST_INIT(children);
479 	} else {
480 		oidp->oid_arg1 = arg1;
481 		oidp->oid_arg2 = arg2;
482 	}
483 	oidp->oid_fmt = fmt;
484 	if (descr) {
485 		int len = strlen(descr) + 1;
486 		oidp->oid_descr = kmalloc(len, M_SYSCTLOID, M_WAITOK);
487 		strcpy((char *)(uintptr_t)(const void *)oidp->oid_descr, descr);
488 	};
489 	/* Update the context, if used */
490 	if (clist != NULL)
491 		sysctl_ctx_entry_add(clist, oidp);
492 	/* Register this oid */
493 	sysctl_register_oid_int(oidp);
494 	SYSCTL_XUNLOCK();
495 	return (oidp);
496 }
497 
498 /*
499  * Rename an existing oid.
500  */
501 void
502 sysctl_rename_oid(struct sysctl_oid *oidp, const char *name)
503 {
504 	char *newname;
505 	char *oldname;
506 
507 	newname = kstrdup(name, M_SYSCTLOID);
508 	SYSCTL_XLOCK();
509 	oldname = __DECONST(char *, oidp->oid_name);
510 	oidp->oid_name = newname;
511 	SYSCTL_XUNLOCK();
512 	kfree(oldname, M_SYSCTLOID);
513 }
514 
515 /*
516  * Register the kernel's oids on startup.
517  */
518 SET_DECLARE(sysctl_set, struct sysctl_oid);
519 
520 static void
521 sysctl_register_all(void *arg)
522 {
523 	struct sysctl_oid **oidp;
524 
525 	SYSCTL_XLOCK();
526 	SET_FOREACH(oidp, sysctl_set)
527 		sysctl_register_oid(*oidp);
528 	SYSCTL_XUNLOCK();
529 }
530 SYSINIT(sysctl, SI_BOOT1_POST, SI_ORDER_ANY, sysctl_register_all, 0);
531 
532 #ifdef SYSCTL_DEBUG
533 /*
534  * "Staff-functions"
535  *
536  * These functions implement a presently undocumented interface
537  * used by the sysctl program to walk the tree, and get the type
538  * so it can print the value.
539  * This interface is under work and consideration, and should probably
540  * be killed with a big axe by the first person who can find the time.
541  * (be aware though, that the proper interface isn't as obvious as it
542  * may seem, there are various conflicting requirements.
543  *
544  * {CTL_SYSCTL, CTL_SYSCTL_DEBUG}		kprintf the entire MIB-tree.
545  * {CTL_SYSCTL, CTL_SYSCTL_NAME, ...}		return the name of the "..."
546  *						OID.
547  * {CTL_SYSCTL, CTL_SYSCTL_NEXT, ...}		return the next OID.
548  * {CTL_SYSCTL, CTL_SYSCTL_NAME2OID}		return the OID of the name in
549  *						"new"
550  * {CTL_SYSCTL, CTL_SYSCTL_OIDFMT, ...}		return the kind & format info
551  *						for the "..." OID.
552  * {CTL_SYSCTL, CTL_SYSCTL_OIDDESCR, ...}	return the description of the
553  *						"..." OID.
554  */
555 
556 static void
557 sysctl_sysctl_debug_dump_node(struct sysctl_oid_list *l, int i)
558 {
559 	int k;
560 	struct sysctl_oid *oidp;
561 
562 	SLIST_FOREACH(oidp, l, oid_link) {
563 
564 		for (k=0; k<i; k++)
565 			kprintf(" ");
566 
567 		kprintf("%d %s ", oidp->oid_number, oidp->oid_name);
568 
569 		kprintf("%c%c",
570 			oidp->oid_kind & CTLFLAG_RD ? 'R':' ',
571 			oidp->oid_kind & CTLFLAG_WR ? 'W':' ');
572 
573 		if (oidp->oid_handler)
574 			kprintf(" *Handler");
575 
576 		switch (oidp->oid_kind & CTLTYPE) {
577 		case CTLTYPE_NODE:
578 			kprintf(" Node\n");
579 			if (!oidp->oid_handler) {
580 				sysctl_sysctl_debug_dump_node(
581 					oidp->oid_arg1, i+2);
582 			}
583 			break;
584 		case CTLTYPE_INT:
585 			kprintf(" Int\n");
586 			break;
587 		case CTLTYPE_UINT:
588 			kprintf(" u_int\n");
589 			break;
590 		case CTLTYPE_LONG:
591 			kprintf(" Long\n");
592 			break;
593 		case CTLTYPE_ULONG:
594 			kprintf(" u_long\n");
595 			break;
596 		case CTLTYPE_STRING:
597 			kprintf(" String\n");
598 			break;
599 		case CTLTYPE_S8:
600 			kprintf(" int8_t\n");
601 			break;
602 		case CTLTYPE_S16:
603 			kprintf(" int16_t\n");
604 			break;
605 		case CTLTYPE_S32:
606 			kprintf(" int32_t\n");
607 			break;
608 		case CTLTYPE_S64:
609 			kprintf(" int64_t\n");
610 			break;
611 		case CTLTYPE_U8:
612 			kprintf(" uint8_t\n");
613 			break;
614 		case CTLTYPE_U16:
615 			kprintf(" uint16_t\n");
616 			break;
617 		case CTLTYPE_U32:
618 			kprintf(" uint32_t\n");
619 			break;
620 		case CTLTYPE_U64:
621 			kprintf(" uint64_t\n");
622 			break;
623 		case CTLTYPE_BIT32(0):
624 			kprintf(" Int\n");
625 			break;
626 		case CTLTYPE_BIT64(0):
627 			kprintf(" Int\n");
628 			break;
629 		case CTLTYPE_OPAQUE:
630 			kprintf(" Opaque/struct\n");
631 			break;
632 		default:
633 			kprintf("\n");
634 			break;
635 		}
636 
637 	}
638 }
639 
640 static int
641 sysctl_sysctl_debug(SYSCTL_HANDLER_ARGS)
642 {
643 	int error;
644 
645 	error = priv_check(req->td, PRIV_SYSCTL_DEBUG);
646 	if (error)
647 		return (error);
648 	sysctl_sysctl_debug_dump_node(&sysctl__children, 0);
649 
650 	return (ENOENT);
651 }
652 
653 SYSCTL_PROC(_sysctl, CTL_SYSCTL_DEBUG, debug, CTLTYPE_STRING | CTLFLAG_RD,
654 	    0, 0, sysctl_sysctl_debug, "-", "");
655 #endif /* SYSCTL_DEBUG */
656 
657 static int
658 sysctl_sysctl_name(SYSCTL_HANDLER_ARGS)
659 {
660 	int *name = (int *) arg1;
661 	u_int namelen = arg2;
662 	int error = 0;
663 	struct sysctl_oid *oid;
664 	struct sysctl_oid_list *lsp = &sysctl__children, *lsp2;
665 	char buf[10];
666 
667 	while (namelen) {
668 		if (!lsp) {
669 			ksnprintf(buf, sizeof(buf), "%d",  *name);
670 			if (req->oldidx)
671 				error = SYSCTL_OUT(req, ".", 1);
672 			if (!error)
673 				error = SYSCTL_OUT(req, buf, strlen(buf));
674 			if (error)
675 				goto out;
676 			namelen--;
677 			name++;
678 			continue;
679 		}
680 		lsp2 = NULL;
681 		SLIST_FOREACH(oid, lsp, oid_link) {
682 			if (oid->oid_number != *name)
683 				continue;
684 
685 			if (req->oldidx)
686 				error = SYSCTL_OUT(req, ".", 1);
687 			if (!error)
688 				error = SYSCTL_OUT(req, oid->oid_name,
689 					strlen(oid->oid_name));
690 			if (error)
691 				goto out;
692 
693 			namelen--;
694 			name++;
695 
696 			if ((oid->oid_kind & CTLTYPE) != CTLTYPE_NODE)
697 				break;
698 
699 			if (oid->oid_handler)
700 				break;
701 
702 			lsp2 = SYSCTL_CHILDREN(oid);
703 			break;
704 		}
705 		lsp = lsp2;
706 	}
707 	error = SYSCTL_OUT(req, "", 1);
708  out:
709 	return (error);
710 }
711 
712 SYSCTL_NODE(_sysctl, CTL_SYSCTL_NAME, name, CTLFLAG_RD | CTLFLAG_NOLOCK,
713 	    sysctl_sysctl_name, "");
714 
715 static int
716 sysctl_sysctl_next_ls(struct sysctl_oid_list *lsp, int *name, u_int namelen,
717 	int *next, int *len, int level, struct sysctl_oid **oidpp)
718 {
719 	struct sysctl_oid *oidp;
720 
721 	*len = level;
722 	SLIST_FOREACH(oidp, lsp, oid_link) {
723 		*next = oidp->oid_number;
724 		*oidpp = oidp;
725 
726 		if (oidp->oid_kind & CTLFLAG_SKIP)
727 			continue;
728 
729 		if (!namelen) {
730 			if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
731 				return (0);
732 			if (oidp->oid_handler)
733 				/* We really should call the handler here...*/
734 				return (0);
735 			lsp = SYSCTL_CHILDREN(oidp);
736 			if (!sysctl_sysctl_next_ls(lsp, 0, 0, next+1,
737 				len, level+1, oidpp))
738 				return (0);
739 			goto emptynode;
740 		}
741 
742 		if (oidp->oid_number < *name)
743 			continue;
744 
745 		if (oidp->oid_number > *name) {
746 			if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
747 				return (0);
748 			if (oidp->oid_handler)
749 				return (0);
750 			lsp = SYSCTL_CHILDREN(oidp);
751 			if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1,
752 				next+1, len, level+1, oidpp))
753 				return (0);
754 			goto next;
755 		}
756 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
757 			continue;
758 
759 		if (oidp->oid_handler)
760 			continue;
761 
762 		lsp = SYSCTL_CHILDREN(oidp);
763 		if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1, next+1,
764 			len, level+1, oidpp))
765 			return (0);
766 	next:
767 		namelen = 1;
768 	emptynode:
769 		*len = level;
770 	}
771 	return (1);
772 }
773 
774 static int
775 sysctl_sysctl_next(SYSCTL_HANDLER_ARGS)
776 {
777 	int *name = (int *) arg1;
778 	u_int namelen = arg2;
779 	int i, j, error;
780 	struct sysctl_oid *oid;
781 	struct sysctl_oid_list *lsp = &sysctl__children;
782 	int newoid[CTL_MAXNAME];
783 
784 	i = sysctl_sysctl_next_ls(lsp, name, namelen, newoid, &j, 1, &oid);
785 	if (i)
786 		return ENOENT;
787 	error = SYSCTL_OUT(req, newoid, j * sizeof (int));
788 
789 	return (error);
790 }
791 
792 SYSCTL_NODE(_sysctl, CTL_SYSCTL_NEXT, next, CTLFLAG_RD | CTLFLAG_NOLOCK,
793 	    sysctl_sysctl_next, "");
794 
795 static int
796 name2oid(char *name, int *oid, int *len, struct sysctl_oid **oidpp)
797 {
798 	struct sysctl_oid *oidp;
799 	struct sysctl_oid_list *lsp = &sysctl__children;
800 	char *p;
801 
802 	SYSCTL_ASSERT_LOCKED();
803 
804 	for (*len = 0; *len < CTL_MAXNAME;) {
805 		p = strsep(&name, ".");
806 
807 		oidp = SLIST_FIRST(lsp);
808 		for (;; oidp = SLIST_NEXT(oidp, oid_link)) {
809 			if (oidp == NULL)
810 				return (ENOENT);
811 			if (strcmp(p, oidp->oid_name) == 0)
812 				break;
813 		}
814 		*oid++ = oidp->oid_number;
815 		(*len)++;
816 
817 		if (name == NULL || *name == '\0') {
818 			if (oidpp)
819 				*oidpp = oidp;
820 			return (0);
821 		}
822 
823 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
824 			break;
825 
826 		if (oidp->oid_handler)
827 			break;
828 
829 		lsp = SYSCTL_CHILDREN(oidp);
830 	}
831 	return (ENOENT);
832 }
833 
834 static int
835 sysctl_sysctl_name2oid(SYSCTL_HANDLER_ARGS)
836 {
837 	char *p;
838 	int error, oid[CTL_MAXNAME], len;
839 	struct sysctl_oid *op = NULL;
840 
841 	if (!req->newlen)
842 		return ENOENT;
843 	if (req->newlen >= MAXPATHLEN)	/* XXX arbitrary, undocumented */
844 		return (ENAMETOOLONG);
845 
846 	p = kmalloc(req->newlen+1, M_SYSCTL, M_WAITOK);
847 
848 	error = SYSCTL_IN(req, p, req->newlen);
849 	if (error) {
850 		kfree(p, M_SYSCTL);
851 		return (error);
852 	}
853 
854 	p [req->newlen] = '\0';
855 
856 	error = name2oid(p, oid, &len, &op);
857 
858 	kfree(p, M_SYSCTL);
859 
860 	if (error)
861 		return (error);
862 
863 	error = SYSCTL_OUT(req, oid, len * sizeof *oid);
864 	return (error);
865 }
866 
867 SYSCTL_PROC(_sysctl, CTL_SYSCTL_NAME2OID, name2oid,
868 	    CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_NOLOCK,
869 	    0, 0, sysctl_sysctl_name2oid, "I", "");
870 
871 static int
872 sysctl_sysctl_oidfmt(SYSCTL_HANDLER_ARGS)
873 {
874 	struct sysctl_oid *oid;
875 	int error;
876 
877 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
878 	if (error)
879 		return (error);
880 
881 	if (!oid->oid_fmt)
882 		return (ENOENT);
883 	error = SYSCTL_OUT(req, &oid->oid_kind, sizeof(oid->oid_kind));
884 	if (error)
885 		return (error);
886 	error = SYSCTL_OUT(req, oid->oid_fmt, strlen(oid->oid_fmt) + 1);
887 	return (error);
888 }
889 
890 
891 SYSCTL_NODE(_sysctl, CTL_SYSCTL_OIDFMT, oidfmt, CTLFLAG_RD | CTLFLAG_NOLOCK,
892 	    sysctl_sysctl_oidfmt, "");
893 
894 static int
895 sysctl_sysctl_oiddescr(SYSCTL_HANDLER_ARGS)
896 {
897 	struct sysctl_oid *oid;
898 	int error;
899 
900 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
901 	if (error)
902 		return (error);
903 
904 	if (!oid->oid_descr)
905 		return (ENOENT);
906 	error = SYSCTL_OUT(req, oid->oid_descr, strlen(oid->oid_descr) + 1);
907 	return (error);
908 }
909 
910 SYSCTL_NODE(_sysctl, CTL_SYSCTL_OIDDESCR, oiddescr,
911 	    CTLFLAG_RD | CTLFLAG_NOLOCK,
912 	    sysctl_sysctl_oiddescr, "");
913 
914 /*
915  * Default "handler" functions.
916  */
917 
918 /*
919  * Handle an 8-bit number, signed or unsigned.  arg1 points to it.
920  */
921 
922 int
923 sysctl_handle_8(SYSCTL_HANDLER_ARGS)
924 {
925 	int error = 0;
926 
927 	if (!arg1)
928 		return (EINVAL);
929 	error = SYSCTL_OUT(req, arg1, sizeof(int8_t));
930 
931 	if (error || !req->newptr)
932 		return (error);
933 
934 	error = SYSCTL_IN(req, arg1, sizeof(int8_t));
935 	return (error);
936 }
937 
938 /*
939  * Handle a 16-bit number, signed or unsigned.  arg1 points to it.
940  */
941 
942 int
943 sysctl_handle_16(SYSCTL_HANDLER_ARGS)
944 {
945 	int error = 0;
946 
947 	if (!arg1)
948 		return (EINVAL);
949 	error = SYSCTL_OUT(req, arg1, sizeof(int16_t));
950 
951 	if (error || !req->newptr)
952 		return (error);
953 
954 	error = SYSCTL_IN(req, arg1, sizeof(int16_t));
955 	return (error);
956 }
957 
958 /*
959  * Handle a 32-bit number, signed or unsigned.  arg1 points to it.
960  */
961 
962 int
963 sysctl_handle_32(SYSCTL_HANDLER_ARGS)
964 {
965 	int error = 0;
966 
967 	if (!arg1)
968 		return (EINVAL);
969 	error = SYSCTL_OUT(req, arg1, sizeof(int32_t));
970 
971 	if (error || !req->newptr)
972 		return (error);
973 
974 	error = SYSCTL_IN(req, arg1, sizeof(int32_t));
975 	return (error);
976 }
977 
978 /*
979  * Handle a 64-bit number, signed or unsigned.  arg1 points to it.
980  */
981 
982 int
983 sysctl_handle_64(SYSCTL_HANDLER_ARGS)
984 {
985 	int error = 0;
986 
987 	if (!arg1)
988 		return (EINVAL);
989 	error = SYSCTL_OUT(req, arg1, sizeof(int64_t));
990 
991 	if (error || !req->newptr)
992 		return (error);
993 
994 	error = SYSCTL_IN(req, arg1, sizeof(int64_t));
995 	return (error);
996 }
997 
998 /*
999  * Handle an int, signed or unsigned.
1000  * Two cases:
1001  *     a variable:  point arg1 at it.
1002  *     a constant:  pass it in arg2.
1003  */
1004 
1005 int
1006 sysctl_handle_int(SYSCTL_HANDLER_ARGS)
1007 {
1008 	int error = 0;
1009 
1010 	if (arg1)
1011 		error = SYSCTL_OUT(req, arg1, sizeof(int));
1012 	else
1013 		error = SYSCTL_OUT(req, &arg2, sizeof(int));
1014 
1015 	if (error || !req->newptr)
1016 		return (error);
1017 
1018 	if (!arg1)
1019 		error = EPERM;
1020 	else
1021 		error = SYSCTL_IN(req, arg1, sizeof(int));
1022 	return (error);
1023 }
1024 
1025 /*
1026  * Handle a long, signed or unsigned.  arg1 points to it.
1027  */
1028 
1029 int
1030 sysctl_handle_long(SYSCTL_HANDLER_ARGS)
1031 {
1032 	int error = 0;
1033 
1034 	if (!arg1)
1035 		return (EINVAL);
1036 	if (req->oldlen == sizeof(int) &&
1037 	    *(long *)arg1 >= INT_MIN &&
1038 	    *(long *)arg1 <= INT_MAX) {
1039 		/*
1040 		 * Backwards compatibility for read-only fields promoted
1041 		 * from int to long.  Allow userland to request the field
1042 		 * as an integer if the value is in-range.
1043 		 */
1044 		int val = (int)*(long *)arg1;
1045 		error = SYSCTL_OUT(req, &val, sizeof(int));
1046 	} else {
1047 		/*
1048 		 * Normal operation fo a long
1049 		 */
1050 		error = SYSCTL_OUT(req, arg1, sizeof(long));
1051 	}
1052 
1053 	if (error || !req->newptr)
1054 		return (error);
1055 
1056 	error = SYSCTL_IN(req, arg1, sizeof(long));
1057 
1058 	return (error);
1059 }
1060 
1061 /*
1062  * Handle a quad, signed or unsigned.  arg1 points to it.
1063  */
1064 
1065 int
1066 sysctl_handle_quad(SYSCTL_HANDLER_ARGS)
1067 {
1068 	int error = 0;
1069 
1070 	if (!arg1)
1071 		return (EINVAL);
1072 	error = SYSCTL_OUT(req, arg1, sizeof(quad_t));
1073 
1074 	if (error || !req->newptr)
1075 		return (error);
1076 
1077 	error = SYSCTL_IN(req, arg1, sizeof(quad_t));
1078 	return (error);
1079 }
1080 
1081 /*
1082  * Handle an bit in a 32-bit field, pass and return an 'int'
1083  * Two cases:
1084  *     a variable:  point arg1 at it.
1085  *     a constant:  pass it in arg2.
1086  */
1087 
1088 int
1089 sysctl_handle_bit32(SYSCTL_HANDLER_ARGS)
1090 {
1091 	int error = 0;
1092 	uint32_t mask;
1093 	int v;
1094 	int bit;
1095 
1096 	bit = (oidp->oid_kind & CTLMASK_BITFLD) >> CTLSHIFT_BITFLD;
1097 	mask = arg1 ? *(uint32_t *)arg1 : (uint32_t)arg2;
1098 	v = (mask & (1U << bit)) ? 1 : 0;
1099 	error = SYSCTL_OUT(req, &v, sizeof(int));
1100 
1101 	if (error || !req->newptr)
1102 		return (error);
1103 
1104 	if (!arg1) {
1105 		error = EPERM;
1106 	} else {
1107 		error = SYSCTL_IN(req, &v, sizeof(int));
1108 		if (error == 0) {
1109 			if (v)
1110 				atomic_set_int((uint32_t *)arg1, 1U << bit);
1111 			else
1112 				atomic_clear_int((uint32_t *)arg1, 1U << bit);
1113 		}
1114 	}
1115 	return (error);
1116 }
1117 
1118 /*
1119  * Handle an bit in a 64-bit field, pass and return an 'int'
1120  * Two cases:
1121  *     a variable:  point arg1 at it.
1122  *     a constant:  pass it in arg2.  (NOTE: arg2 is only 32bits)
1123  */
1124 
1125 int
1126 sysctl_handle_bit64(SYSCTL_HANDLER_ARGS)
1127 {
1128 	int error = 0;
1129 	uint64_t mask;
1130 	int v;
1131 	int bit;
1132 
1133 	bit = (oidp->oid_kind & CTLMASK_BITFLD) >> CTLSHIFT_BITFLD;
1134 	mask = arg1 ? *(uint64_t *)arg1 : (uint64_t)(uint32_t)arg2;
1135 	v = (mask & (1LU << bit)) ? 1 : 0;
1136 	error = SYSCTL_OUT(req, &v, sizeof(int));
1137 
1138 	if (error || !req->newptr)
1139 		return (error);
1140 
1141 	if (!arg1) {
1142 		error = EPERM;
1143 	} else {
1144 		error = SYSCTL_IN(req, &v, sizeof(int));
1145 		if (error == 0) {
1146 			if (v)
1147 				atomic_set_long((uint64_t *)arg1, 1LU << bit);
1148 			else
1149 				atomic_clear_long((uint64_t *)arg1, 1LU << bit);
1150 		}
1151 	}
1152 	return (error);
1153 }
1154 
1155 /*
1156  * Handle our generic '\0' terminated 'C' string.
1157  * Two cases:
1158  *	a variable string:  point arg1 at it, arg2 is max length.
1159  *	a constant string:  point arg1 at it, arg2 is zero.
1160  */
1161 
1162 int
1163 sysctl_handle_string(SYSCTL_HANDLER_ARGS)
1164 {
1165 	int error=0;
1166 
1167 	error = SYSCTL_OUT(req, arg1, strlen((char *)arg1)+1);
1168 
1169 	if (error || !req->newptr)
1170 		return (error);
1171 
1172 	if ((req->newlen - req->newidx) >= arg2) {
1173 		error = EINVAL;
1174 	} else {
1175 		arg2 = (req->newlen - req->newidx);
1176 		error = SYSCTL_IN(req, arg1, arg2);
1177 		((char *)arg1)[arg2] = '\0';
1178 	}
1179 
1180 	return (error);
1181 }
1182 
1183 /*
1184  * Handle any kind of opaque data.
1185  * arg1 points to it, arg2 is the size.
1186  */
1187 
1188 int
1189 sysctl_handle_opaque(SYSCTL_HANDLER_ARGS)
1190 {
1191 	int error;
1192 
1193 	error = SYSCTL_OUT(req, arg1, arg2);
1194 
1195 	if (error || !req->newptr)
1196 		return (error);
1197 
1198 	error = SYSCTL_IN(req, arg1, arg2);
1199 
1200 	return (error);
1201 }
1202 
1203 /*
1204  * Transfer functions to/from kernel space.
1205  * XXX: rather untested at this point
1206  */
1207 static int
1208 sysctl_old_kernel(struct sysctl_req *req, const void *p, size_t l)
1209 {
1210 	size_t i = 0;
1211 
1212 	if (req->oldptr) {
1213 		i = l;
1214 		if (i > req->oldlen - req->oldidx)
1215 			i = req->oldlen - req->oldidx;
1216 		if (i > 0)
1217 			bcopy(p, (char *)req->oldptr + req->oldidx, i);
1218 	}
1219 	req->oldidx += l;
1220 	if (req->oldptr && i != l)
1221 		return (ENOMEM);
1222 	return (0);
1223 }
1224 
1225 static int
1226 sysctl_new_kernel(struct sysctl_req *req, void *p, size_t l)
1227 {
1228 
1229 	if (!req->newptr)
1230 		return 0;
1231 	if (req->newlen - req->newidx < l)
1232 		return (EINVAL);
1233 	bcopy((char *)req->newptr + req->newidx, p, l);
1234 	req->newidx += l;
1235 	return (0);
1236 }
1237 
1238 int
1239 kernel_sysctl(int *name, u_int namelen,
1240 	      void *old, size_t *oldlenp,
1241 	      void *new, size_t newlen, size_t *retval)
1242 {
1243 	int error = 0;
1244 	struct sysctl_req req;
1245 
1246 	bzero(&req, sizeof req);
1247 
1248 	req.td = curthread;
1249 
1250 	if (oldlenp) {
1251 		req.oldlen = *oldlenp;
1252 	}
1253 	req.validlen = req.oldlen;
1254 
1255 	if (old) {
1256 		req.oldptr= old;
1257 	}
1258 
1259 	if (new != NULL) {
1260 		req.newlen = newlen;
1261 		req.newptr = new;
1262 	}
1263 
1264 	req.oldfunc = sysctl_old_kernel;
1265 	req.newfunc = sysctl_new_kernel;
1266 #if 0
1267 	req.lock = REQ_UNWIRED;
1268 #endif
1269 
1270 	SYSCTL_SLOCK();
1271 	error = sysctl_root(0, name, namelen, &req);
1272 	SYSCTL_SUNLOCK();
1273 
1274 #if 0
1275 	if (req.lock == REQ_WIRED && req.validlen > 0)
1276 		vsunlock(req.oldptr, req.validlen);
1277 #endif
1278 
1279 	if (error && error != ENOMEM)
1280 		return (error);
1281 
1282 	if (retval) {
1283 		if (req.oldptr && req.oldidx > req.validlen)
1284 			*retval = req.validlen;
1285 		else
1286 			*retval = req.oldidx;
1287 	}
1288 	return (error);
1289 }
1290 
1291 int
1292 kernel_sysctlbyname(char *name,
1293 		    void *old, size_t *oldlenp,
1294 		    void *new, size_t newlen, size_t *retval)
1295 {
1296         int oid[CTL_MAXNAME];
1297         size_t oidlen, plen;
1298 	int error;
1299 
1300 	oid[0] = CTL_SYSCTL;
1301 	oid[1] = CTL_SYSCTL_NAME2OID;
1302 	oidlen = sizeof(oid);
1303 
1304 	error = kernel_sysctl(oid, 2, oid, &oidlen, name, strlen(name), &plen);
1305 	if (error)
1306 		return (error);
1307 
1308 	error = kernel_sysctl(oid, plen / sizeof(int), old, oldlenp,
1309 	    new, newlen, retval);
1310 	return (error);
1311 }
1312 
1313 /*
1314  * Transfer function to/from user space.
1315  */
1316 static int
1317 sysctl_old_user(struct sysctl_req *req, const void *p, size_t l)
1318 {
1319 	int error = 0;
1320 	size_t i = 0;
1321 
1322 #if 0
1323 	if (req->lock == 1 && req->oldptr) {
1324 		vslock(req->oldptr, req->oldlen);
1325 		req->lock = 2;
1326 	}
1327 #endif
1328 	if (req->oldptr) {
1329 		i = l;
1330 		if (i > req->oldlen - req->oldidx)
1331 			i = req->oldlen - req->oldidx;
1332 		if (i > 0)
1333 			error = copyout(p, (char *)req->oldptr + req->oldidx,
1334 					i);
1335 	}
1336 	req->oldidx += l;
1337 	if (error)
1338 		return (error);
1339 	if (req->oldptr && i < l)
1340 		return (ENOMEM);
1341 	return (0);
1342 }
1343 
1344 static int
1345 sysctl_new_user(struct sysctl_req *req, void *p, size_t l)
1346 {
1347 	int error;
1348 
1349 	if (!req->newptr)
1350 		return 0;
1351 	if (req->newlen - req->newidx < l)
1352 		return (EINVAL);
1353 	error = copyin((char *)req->newptr + req->newidx, p, l);
1354 	req->newidx += l;
1355 	return (error);
1356 }
1357 
1358 int
1359 sysctl_find_oid(int *name, u_int namelen, struct sysctl_oid **noid,
1360 		int *nindx, struct sysctl_req *req)
1361 {
1362 	struct sysctl_oid_list *lsp;
1363 	struct sysctl_oid *oid;
1364 	int indx;
1365 
1366 	lsp = &sysctl__children;
1367 	indx = 0;
1368 	while (indx < CTL_MAXNAME) {
1369 		SLIST_FOREACH(oid, lsp, oid_link) {
1370 			if (oid->oid_number == name[indx])
1371 				break;
1372 		}
1373 		if (oid == NULL)
1374 			return (ENOENT);
1375 
1376 		indx++;
1377 		if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1378 			if (oid->oid_handler != NULL || indx == namelen) {
1379 				*noid = oid;
1380 				if (nindx != NULL)
1381 					*nindx = indx;
1382 				KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
1383 				    ("%s found DYING node %p", __func__, oid));
1384 				return (0);
1385 			}
1386 			lsp = SYSCTL_CHILDREN(oid);
1387 		} else if (indx == namelen) {
1388 			*noid = oid;
1389 			if (nindx != NULL)
1390 				*nindx = indx;
1391 			KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
1392 			    ("%s found DYING node %p", __func__, oid));
1393 			return (0);
1394 		} else {
1395 			return (ENOTDIR);
1396 		}
1397 	}
1398 	return (ENOENT);
1399 }
1400 
1401 /*
1402  * Traverse our tree, and find the right node, execute whatever it points
1403  * to, and return the resulting error code.
1404  */
1405 static int
1406 sysctl_root(SYSCTL_HANDLER_ARGS)
1407 {
1408 	struct thread *td = req->td;
1409 	struct proc *p = td ? td->td_proc : NULL;
1410 	struct sysctl_oid *oid;
1411 	int error, indx;
1412 	int lktype;
1413 
1414 	error = sysctl_find_oid(arg1, arg2, &oid, &indx, req);
1415 	if (error)
1416 		return (error);
1417 
1418 	if (sysctl_debugx & 1) {
1419 		kprintf("pid %d oid %p %s\n",
1420 			(p ? p->p_pid : -1), oid, oid->oid_name);
1421 	}
1422 
1423 
1424 	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1425 		/*
1426 		 * You can't call a sysctl when it's a node, but has
1427 		 * no handler.  Inform the user that it's a node.
1428 		 * The indx may or may not be the same as namelen.
1429 		 */
1430 		if (oid->oid_handler == NULL)
1431 			return (EISDIR);
1432 	}
1433 
1434 	/* If writing isn't allowed */
1435 	if (req->newptr && (!(oid->oid_kind & CTLFLAG_WR) ||
1436 	    ((oid->oid_kind & CTLFLAG_SECURE) && securelevel > 0)))
1437 		return (EPERM);
1438 
1439 	/* Most likely only root can write */
1440 	if (!(oid->oid_kind & CTLFLAG_ANYBODY) && req->newptr && p &&
1441 	    (error = priv_check_cred(td->td_ucred,
1442 	     (oid->oid_kind & CTLFLAG_PRISON) ? PRIV_SYSCTL_WRITEJAIL :
1443 	                                        PRIV_SYSCTL_WRITE, 0)))
1444 		return (error);
1445 
1446 	if (oid->oid_handler == NULL)
1447 		return EINVAL;
1448 
1449 	/*
1450 	 * Default oid locking is exclusive when modifying (newptr),
1451 	 * shared otherwise, unless overridden with a control flag.
1452 	 */
1453 	if ((oid->oid_kind & CTLFLAG_NOLOCK) == 0) {
1454 		lktype = (req->newptr != NULL) ? LK_EXCLUSIVE : LK_SHARED;
1455 		if (oid->oid_kind & CTLFLAG_SHLOCK)
1456 			lktype = LK_SHARED;
1457 		if (oid->oid_kind & CTLFLAG_EXLOCK)
1458 			lktype = LK_EXCLUSIVE;
1459 #if 1
1460 		lockmgr(&oid->oid_lock, lktype);
1461 #else
1462 		/* DEBUGGING */
1463 		if (lockmgr(&oid->oid_lock, lktype | LK_SLEEPFAIL)) {
1464 			kprintf("%s\n", oid->oid_name);
1465 			lockmgr(&oid->oid_lock, lktype);
1466 		}
1467 #endif
1468 	}
1469 
1470 	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE)
1471 		error = oid->oid_handler(oid, (int *)arg1 + indx, arg2 - indx,
1472 					 req);
1473 	else
1474 		error = oid->oid_handler(oid, oid->oid_arg1, oid->oid_arg2,
1475 					 req);
1476 
1477 	if ((oid->oid_kind & CTLFLAG_NOLOCK) == 0)
1478 		lockmgr(&oid->oid_lock, LK_RELEASE);
1479 	return (error);
1480 }
1481 
1482 int
1483 sys___sysctl(struct sysmsg *sysmsg, const struct sysctl_args *uap)
1484 {
1485 	int error, i, name[CTL_MAXNAME];
1486 	size_t j;
1487 
1488 	if (uap->namelen > CTL_MAXNAME || uap->namelen < 2)
1489 		return (EINVAL);
1490 
1491 	error = copyin(uap->name, &name, uap->namelen * sizeof(int));
1492 	if (error)
1493 		return (error);
1494 
1495 	error = userland_sysctl(name, uap->namelen,
1496 				uap->old, uap->oldlenp, 0,
1497 				uap->new, uap->newlen, &j);
1498 	if (error && error != ENOMEM)
1499 		return (error);
1500 	if (uap->oldlenp) {
1501 		i = copyout(&j, uap->oldlenp, sizeof(j));
1502 		if (i)
1503 			return (i);
1504 	}
1505 	return (error);
1506 }
1507 
1508 /*
1509  * This is used from various compatibility syscalls too.  That's why name
1510  * must be in kernel space.
1511  */
1512 int
1513 userland_sysctl(int *name, u_int namelen,
1514 		void *old, size_t *oldlenp, int inkernel,
1515 		void *new, size_t newlen, size_t *retval)
1516 {
1517 	int error = 0;
1518 	struct sysctl_req req;
1519 
1520 	bzero(&req, sizeof req);
1521 
1522 	req.td = curthread;
1523 	req.flags = 0;
1524 
1525 	if (oldlenp) {
1526 		if (inkernel) {
1527 			req.oldlen = *oldlenp;
1528 		} else {
1529 			error = copyin(oldlenp, &req.oldlen, sizeof(*oldlenp));
1530 			if (error)
1531 				return (error);
1532 		}
1533 	}
1534 	req.validlen = req.oldlen;
1535 
1536 	/*
1537 	 * NOTE: User supplied buffers are not guaranteed to be good,
1538 	 *	 the sysctl copyins and copyouts can fail.
1539 	 */
1540 	if (old)
1541 		req.oldptr= old;
1542 
1543 	if (new != NULL) {
1544 		req.newlen = newlen;
1545 		req.newptr = new;
1546 	}
1547 
1548 	req.oldfunc = sysctl_old_user;
1549 	req.newfunc = sysctl_new_user;
1550 #if 0
1551 	req.lock = REQ_UNWIRED;
1552 #endif
1553 
1554 #ifdef KTRACE
1555 	if (KTRPOINT(curthread, KTR_SYSCTL))
1556 		ktrsysctl(name, namelen);
1557 #endif
1558 
1559 	for (;;) {
1560 		req.oldidx = 0;
1561 		req.newidx = 0;
1562 		SYSCTL_SLOCK();
1563 		error = sysctl_root(0, name, namelen, &req);
1564 		SYSCTL_SUNLOCK();
1565 		if (error != EAGAIN)
1566 			break;
1567 		lwkt_yield();
1568 	}
1569 
1570 #if 0
1571 	if (req.lock == REQ_WIRED && req.validlen > 0)
1572 		vsunlock(req.oldptr, req.validlen);
1573 #endif
1574 	if (error && error != ENOMEM)
1575 		return (error);
1576 
1577 	if (retval) {
1578 		if (req.oldptr && req.oldidx > req.validlen)
1579 			*retval = req.validlen;
1580 		else
1581 			*retval = req.oldidx;
1582 	}
1583 	return (error);
1584 }
1585 
1586 int
1587 sysctl_int_range(SYSCTL_HANDLER_ARGS, int low, int high)
1588 {
1589 	int error, value;
1590 
1591 	value = *(int *)arg1;
1592 	error = sysctl_handle_int(oidp, &value, 0, req);
1593 	if (error || !req->newptr)
1594 		return (error);
1595 	if (value < low || value > high)
1596 		return (EINVAL);
1597 	*(int *)arg1 = value;
1598 	return (0);
1599 }
1600 
1601 /*
1602  * Drain into a sysctl struct.  The user buffer should be wired if a page
1603  * fault would cause issue.
1604  */
1605 static int
1606 sbuf_sysctl_drain(void *arg, const char *data, int len)
1607 {
1608 	struct sysctl_req *req = arg;
1609 	int error;
1610 
1611 	error = SYSCTL_OUT(req, data, len);
1612 	KASSERT(error >= 0, ("Got unexpected negative value %d", error));
1613 	return (error == 0 ? len : -error);
1614 }
1615 
1616 struct sbuf *
1617 sbuf_new_for_sysctl(struct sbuf *s, char *buf, int length,
1618     struct sysctl_req *req)
1619 {
1620 
1621 	s = sbuf_new(s, buf, length, SBUF_FIXEDLEN);
1622 	sbuf_set_drain(s, sbuf_sysctl_drain, req);
1623 	return (s);
1624 }
1625 
1626 /*
1627  * The exclusive sysctl lock only protects its topology, and is
1628  * very expensive, but allows us to use a pcpu shared lock for
1629  * critical path accesses.
1630  */
1631 void
1632 _sysctl_xlock(void)
1633 {
1634 	globaldata_t gd;
1635 	int i;
1636 
1637 	for (i = 0; i < ncpus; ++i) {
1638 		gd = globaldata_find(i);
1639 		lockmgr(&gd->gd_sysctllock, LK_EXCLUSIVE);
1640 	}
1641 }
1642 
1643 void
1644 _sysctl_xunlock(void)
1645 {
1646 	globaldata_t gd;
1647 	int i;
1648 
1649 	for (i = 0; i < ncpus; ++i) {
1650 		gd = globaldata_find(i);
1651 		lockmgr(&gd->gd_sysctllock, LK_RELEASE);
1652 	}
1653 }
1654