xref: /dragonfly/sys/kern/kern_sysctl.c (revision 3a48e5e1)
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 /*
60  * The sysctllock protects the MIB tree.  It also protects sysctl
61  * contexts used with dynamic sysctls.  The sysctl_register_oid() and
62  * sysctl_unregister_oid() routines require the sysctllock to already
63  * be held, so the sysctl_lock() and sysctl_unlock() routines are
64  * provided for the few places in the kernel which need to use that
65  * API rather than using the dynamic API.  Use of the dynamic API is
66  * strongly encouraged for most code.
67  */
68 
69 static int	sysctl_root(SYSCTL_HANDLER_ARGS);
70 static void	sysctl_register_oid_int(struct sysctl_oid *oipd);
71 static void	sysctl_unregister_oid_int(struct sysctl_oid *oipd);
72 
73 struct sysctl_oid_list sysctl__children; /* root list */
74 
75 static int	sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del,
76 		    int recurse);
77 
78 static struct sysctl_oid *
79 sysctl_find_oidname(const char *name, struct sysctl_oid_list *list, int lock)
80 {
81 	struct sysctl_oid *oidp;
82 
83 	SLIST_FOREACH(oidp, list, oid_link) {
84 		if (strcmp(oidp->oid_name, name) == 0) {
85 			break;
86 		}
87 	}
88 	return (oidp);
89 }
90 
91 /*
92  * Initialization of the MIB tree.
93  *
94  * Order by number in each list.
95  */
96 
97 void
98 sysctl_register_oid(struct sysctl_oid *oidp)
99 {
100 	SYSCTL_XLOCK();
101 	sysctl_register_oid_int(oidp);
102 	SYSCTL_XUNLOCK();
103 }
104 
105 static void
106 sysctl_register_oid_int(struct sysctl_oid *oidp)
107 {
108 	struct sysctl_oid_list *parent = oidp->oid_parent;
109 	struct sysctl_oid *p;
110 	struct sysctl_oid *q;
111 
112 	/*
113 	 * Finish initialization from sysctl_set or add.
114 	 */
115 	lockinit(&oidp->oid_lock, "oidlk", 0, LK_CANRECURSE);
116 
117 	/*
118 	 * First check if another oid with the same name already
119 	 * exists in the parent's list.
120 	 */
121 	p = sysctl_find_oidname(oidp->oid_name, parent, 0);
122 	if (p != NULL) {
123 		if ((p->oid_kind & CTLTYPE) == CTLTYPE_NODE)
124 			p->oid_refcnt++;
125 		else
126 			kprintf("can't re-use a leaf (%s)!\n", p->oid_name);
127 		return;
128 	}
129 
130 	/*
131 	 * If this oid has a number OID_AUTO, give it a number which
132 	 * is greater than any current oid.  Make sure it is at least
133 	 * 256 to leave space for pre-assigned oid numbers.
134 	 */
135 	if (oidp->oid_number == OID_AUTO) {
136 		int newoid = 0x100;	/* minimum AUTO oid */
137 
138 		/*
139 		 * Adjust based on highest oid in parent list
140 		 */
141 		SLIST_FOREACH(p, parent, oid_link) {
142 			if (newoid <= p->oid_number)
143 				newoid = p->oid_number + 1;
144 		}
145 		oidp->oid_number = newoid;
146 	}
147 
148 	/*
149 	 * Insert the oid into the parent's list in order.
150 	 */
151 	q = NULL;
152 	SLIST_FOREACH(p, parent, oid_link) {
153 		if (oidp->oid_number < p->oid_number)
154 			break;
155 		q = p;
156 	}
157 	if (q)
158 		SLIST_INSERT_AFTER(q, oidp, oid_link);
159 	else
160 		SLIST_INSERT_HEAD(parent, oidp, oid_link);
161 }
162 
163 void
164 sysctl_unregister_oid(struct sysctl_oid *oidp)
165 {
166 	SYSCTL_XLOCK();
167 	sysctl_unregister_oid_int(oidp);
168 	SYSCTL_XUNLOCK();
169 }
170 
171 static void
172 sysctl_unregister_oid_int(struct sysctl_oid *oidp)
173 {
174 	struct sysctl_oid *p;
175 
176 	if (oidp->oid_number == OID_AUTO)
177 		panic("Trying to unregister OID_AUTO entry: %p", oidp);
178 
179 	SLIST_FOREACH(p, oidp->oid_parent, oid_link) {
180 		if (p != oidp)
181 			continue;
182 		SLIST_REMOVE(oidp->oid_parent, oidp, sysctl_oid, oid_link);
183 		return;
184 	}
185 
186 	/*
187 	 * This can happen when a module fails to register and is
188 	 * being unloaded afterwards.  It should not be a panic()
189 	 * for normal use.
190 	 */
191 	kprintf("%s: failed to unregister sysctl\n", __func__);
192 }
193 
194 /* Initialize a new context to keep track of dynamically added sysctls. */
195 int
196 sysctl_ctx_init(struct sysctl_ctx_list *c)
197 {
198 	if (c == NULL)
199 		return(EINVAL);
200 	TAILQ_INIT(c);
201 	return(0);
202 }
203 
204 /* Free the context, and destroy all dynamic oids registered in this context */
205 int
206 sysctl_ctx_free(struct sysctl_ctx_list *clist)
207 {
208 	struct sysctl_ctx_entry *e, *e1;
209 	int error;
210 
211 	error = 0;
212 	/*
213 	 * First perform a "dry run" to check if it's ok to remove oids.
214 	 * XXX FIXME
215 	 * XXX This algorithm is a hack. But I don't know any
216 	 * XXX better solution for now...
217 	 */
218 	SYSCTL_XLOCK();
219 	TAILQ_FOREACH(e, clist, link) {
220 		error = sysctl_remove_oid_locked(e->entry, 0, 0);
221 		if (error)
222 			break;
223 	}
224 	/*
225 	 * Restore deregistered entries, either from the end,
226 	 * or from the place where error occured.
227 	 * e contains the entry that was not unregistered
228 	 */
229 	if (error)
230 		e1 = TAILQ_PREV(e, sysctl_ctx_list, link);
231 	else
232 		e1 = TAILQ_LAST(clist, sysctl_ctx_list);
233 	while (e1 != NULL) {
234 		sysctl_register_oid(e1->entry);
235 		e1 = TAILQ_PREV(e1, sysctl_ctx_list, link);
236 	}
237 	if (error) {
238 		SYSCTL_XUNLOCK();
239 		return(EBUSY);
240 	}
241 	/* Now really delete the entries */
242 	e = TAILQ_FIRST(clist);
243 	while (e != NULL) {
244 		e1 = TAILQ_NEXT(e, link);
245 		error = sysctl_remove_oid_locked(e->entry, 1, 0);
246 		if (error)
247 			panic("sysctl_remove_oid: corrupt tree, entry: %s",
248 			    e->entry->oid_name);
249 		kfree(e, M_SYSCTLOID);
250 		e = e1;
251 	}
252 	SYSCTL_XUNLOCK();
253 	return (error);
254 }
255 
256 /* Add an entry to the context */
257 struct sysctl_ctx_entry *
258 sysctl_ctx_entry_add(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
259 {
260 	struct sysctl_ctx_entry *e;
261 
262 	SYSCTL_ASSERT_LOCKED();
263 	if (clist == NULL || oidp == NULL)
264 		return(NULL);
265 	e = kmalloc(sizeof(struct sysctl_ctx_entry), M_SYSCTLOID, M_WAITOK);
266 	e->entry = oidp;
267 	TAILQ_INSERT_HEAD(clist, e, link);
268 	return (e);
269 }
270 
271 /* Find an entry in the context */
272 struct sysctl_ctx_entry *
273 sysctl_ctx_entry_find(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
274 {
275 	struct sysctl_ctx_entry *e;
276 
277 	SYSCTL_ASSERT_LOCKED();
278 	if (clist == NULL || oidp == NULL)
279 		return(NULL);
280 	TAILQ_FOREACH(e, clist, link) {
281 		if(e->entry == oidp)
282 			return(e);
283 	}
284 	return (e);
285 }
286 
287 /*
288  * Delete an entry from the context.
289  * NOTE: this function doesn't free oidp! You have to remove it
290  * with sysctl_remove_oid().
291  */
292 int
293 sysctl_ctx_entry_del(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
294 {
295 	struct sysctl_ctx_entry *e;
296 
297 	if (clist == NULL || oidp == NULL)
298 		return (EINVAL);
299 	SYSCTL_XLOCK();
300 	e = sysctl_ctx_entry_find(clist, oidp);
301 	if (e != NULL) {
302 		TAILQ_REMOVE(clist, e, link);
303 		SYSCTL_XUNLOCK();
304 		kfree(e, M_SYSCTLOID);
305 		return (0);
306 	} else {
307 		SYSCTL_XUNLOCK();
308 		return (ENOENT);
309 	}
310 }
311 
312 /*
313  * Remove dynamically created sysctl trees.
314  * oidp - top of the tree to be removed
315  * del - if 0 - just deregister, otherwise free up entries as well
316  * recurse - if != 0 traverse the subtree to be deleted
317  */
318 int
319 sysctl_remove_oid(struct sysctl_oid *oidp, int del, int recurse)
320 {
321 	int error;
322 
323 	SYSCTL_XLOCK();
324 	error = sysctl_remove_oid_locked(oidp, del, recurse);
325 	SYSCTL_XUNLOCK();
326 	return (error);
327 }
328 
329 static int
330 sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del, int recurse)
331 {
332 	struct sysctl_oid *p, *tmp;
333 	int error;
334 
335 	SYSCTL_ASSERT_LOCKED();
336 	if (oidp == NULL)
337 		return(EINVAL);
338 	if ((oidp->oid_kind & CTLFLAG_DYN) == 0) {
339 		kprintf("can't remove non-dynamic nodes!\n");
340 		return (EINVAL);
341 	}
342 	/*
343 	 * WARNING: normal method to do this should be through
344 	 * sysctl_ctx_free(). Use recursing as the last resort
345 	 * method to purge your sysctl tree of leftovers...
346 	 * However, if some other code still references these nodes,
347 	 * it will panic.
348 	 */
349 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
350 		if (oidp->oid_refcnt == 1) {
351 			SLIST_FOREACH_MUTABLE(p,
352 			    SYSCTL_CHILDREN(oidp), oid_link, tmp) {
353 				if (!recurse) {
354 					kprintf("Warning: failed attempt to "
355 					    "remove oid %s with child %s\n",
356 					    oidp->oid_name, p->oid_name);
357 					return (ENOTEMPTY);
358 				}
359 				error = sysctl_remove_oid_locked(p, del,
360 								 recurse);
361 				if (error)
362 					return (error);
363 			}
364 			if (del)
365 				kfree(SYSCTL_CHILDREN(oidp), M_SYSCTLOID);
366 		}
367 	}
368 	if (oidp->oid_refcnt > 1 ) {
369 		oidp->oid_refcnt--;
370 	} else {
371 		if (oidp->oid_refcnt == 0) {
372 			kprintf("Warning: bad oid_refcnt=%u (%s)!\n",
373 				oidp->oid_refcnt, oidp->oid_name);
374 			return (EINVAL);
375 		}
376 		sysctl_unregister_oid(oidp);
377 		if (del) {
378 			/*
379 			 * Wait for all threads running the handler to drain.
380 			 * This preserves the previous behavior when the
381 			 * sysctl lock was held across a handler invocation,
382 			 * and is necessary for module unload correctness.
383 			 */
384 			while (oidp->oid_running > 0) {
385 				oidp->oid_kind |= CTLFLAG_DYING;
386 				tsleep_interlock(&oidp->oid_running, 0);
387 				SYSCTL_XUNLOCK();
388 				tsleep(&oidp->oid_running, PINTERLOCKED,
389 				       "oidrm", 0);
390 				SYSCTL_XLOCK();
391 			}
392 			if (oidp->oid_descr)
393 				kfree(__DECONST(char *, oidp->oid_descr),
394 				    M_SYSCTLOID);
395 			kfree(__DECONST(char *, oidp->oid_name), M_SYSCTLOID);
396 			lockuninit(&oidp->oid_lock);
397 			kfree(oidp, M_SYSCTLOID);
398 		}
399 	}
400 	return (0);
401 }
402 
403 int
404 sysctl_remove_name(struct sysctl_oid *parent, const char *name,
405     int del, int recurse)
406 {
407 	struct sysctl_oid *p, *tmp;
408 	int error;
409 
410 	error = ENOENT;
411 	SYSCTL_XLOCK();
412 	SLIST_FOREACH_MUTABLE(p, SYSCTL_CHILDREN(parent), oid_link, tmp) {
413 		if (strcmp(p->oid_name, name) == 0) {
414 			error = sysctl_remove_oid_locked(p, del, recurse);
415 			break;
416 		}
417 	}
418 	SYSCTL_XUNLOCK();
419 
420 	return (error);
421 }
422 
423 /*
424  * Create new sysctls at run time.
425  * clist may point to a valid context initialized with sysctl_ctx_init().
426  */
427 struct sysctl_oid *
428 sysctl_add_oid(struct sysctl_ctx_list *clist, struct sysctl_oid_list *parent,
429 	int number, const char *name, int kind, void *arg1, int arg2,
430 	int (*handler)(SYSCTL_HANDLER_ARGS), const char *fmt, const char *descr)
431 {
432 	struct sysctl_oid *oidp;
433 	ssize_t len;
434 	char *newname;
435 
436 	/* You have to hook up somewhere.. */
437 	if (parent == NULL)
438 		return(NULL);
439 	SYSCTL_XLOCK();
440 	/* Check if the node already exists, otherwise create it */
441 	oidp = sysctl_find_oidname(name, parent, 0);
442 	if (oidp != NULL) {
443 		if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
444 			oidp->oid_refcnt++;
445 			/* Update the context */
446 			if (clist != NULL)
447 				sysctl_ctx_entry_add(clist, oidp);
448 			SYSCTL_XUNLOCK();
449 			return (oidp);
450 		} else {
451 			kprintf("can't re-use a leaf (%s)!\n", name);
452 			SYSCTL_XUNLOCK();
453 			return (NULL);
454 		}
455 	}
456 	oidp = kmalloc(sizeof(struct sysctl_oid), M_SYSCTLOID,
457 		       M_WAITOK | M_ZERO);
458 	oidp->oid_parent = parent;
459 	SLIST_NEXT(oidp, oid_link) = NULL;
460 	oidp->oid_number = number;
461 	oidp->oid_refcnt = 1;
462 	len = strlen(name);
463 	newname = kmalloc(len + 1, M_SYSCTLOID, M_WAITOK);
464 	bcopy(name, newname, len + 1);
465 	newname[len] = '\0';
466 	oidp->oid_name = newname;
467 	oidp->oid_handler = handler;
468 	oidp->oid_kind = CTLFLAG_DYN | kind;
469 	if ((kind & CTLTYPE) == CTLTYPE_NODE) {
470 		struct sysctl_oid_list *children;
471 
472 		/* Allocate space for children */
473 		children = kmalloc(sizeof(*children), M_SYSCTLOID, M_WAITOK);
474 		SYSCTL_SET_CHILDREN(oidp, children);
475 		SLIST_INIT(children);
476 	} else {
477 		oidp->oid_arg1 = arg1;
478 		oidp->oid_arg2 = arg2;
479 	}
480 	oidp->oid_fmt = fmt;
481 	if (descr) {
482 		int len = strlen(descr) + 1;
483 		oidp->oid_descr = kmalloc(len, M_SYSCTLOID, M_WAITOK);
484 		strcpy((char *)(uintptr_t)(const void *)oidp->oid_descr, descr);
485 	};
486 	/* Update the context, if used */
487 	if (clist != NULL)
488 		sysctl_ctx_entry_add(clist, oidp);
489 	/* Register this oid */
490 	sysctl_register_oid_int(oidp);
491 	SYSCTL_XUNLOCK();
492 	return (oidp);
493 }
494 
495 /*
496  * Rename an existing oid.
497  */
498 void
499 sysctl_rename_oid(struct sysctl_oid *oidp, const char *name)
500 {
501 	char *newname;
502 	char *oldname;
503 
504 	newname = kstrdup(name, M_SYSCTLOID);
505 	SYSCTL_XLOCK();
506 	oldname = __DECONST(char *, oidp->oid_name);
507 	oidp->oid_name = newname;
508 	SYSCTL_XUNLOCK();
509 	kfree(oldname, M_SYSCTLOID);
510 }
511 
512 /*
513  * Register the kernel's oids on startup.
514  */
515 SET_DECLARE(sysctl_set, struct sysctl_oid);
516 
517 static void
518 sysctl_register_all(void *arg)
519 {
520 	struct sysctl_oid **oidp;
521 
522 	SYSCTL_XLOCK();
523 	SET_FOREACH(oidp, sysctl_set)
524 		sysctl_register_oid(*oidp);
525 	SYSCTL_XUNLOCK();
526 }
527 SYSINIT(sysctl, SI_BOOT1_POST, SI_ORDER_ANY, sysctl_register_all, 0);
528 
529 #ifdef SYSCTL_DEBUG
530 /*
531  * "Staff-functions"
532  *
533  * These functions implement a presently undocumented interface
534  * used by the sysctl program to walk the tree, and get the type
535  * so it can print the value.
536  * This interface is under work and consideration, and should probably
537  * be killed with a big axe by the first person who can find the time.
538  * (be aware though, that the proper interface isn't as obvious as it
539  * may seem, there are various conflicting requirements.
540  *
541  * {CTL_SYSCTL, CTL_SYSCTL_DEBUG}		kprintf the entire MIB-tree.
542  * {CTL_SYSCTL, CTL_SYSCTL_NAME, ...}		return the name of the "..."
543  *						OID.
544  * {CTL_SYSCTL, CTL_SYSCTL_NEXT, ...}		return the next OID.
545  * {CTL_SYSCTL, CTL_SYSCTL_NAME2OID}		return the OID of the name in
546  *						"new"
547  * {CTL_SYSCTL, CTL_SYSCTL_OIDFMT, ...}		return the kind & format info
548  *						for the "..." OID.
549  * {CTL_SYSCTL, CTL_SYSCTL_OIDDESCR, ...}	return the description of the
550  *						"..." OID.
551  */
552 
553 static void
554 sysctl_sysctl_debug_dump_node(struct sysctl_oid_list *l, int i)
555 {
556 	int k;
557 	struct sysctl_oid *oidp;
558 
559 	SLIST_FOREACH(oidp, l, oid_link) {
560 
561 		for (k=0; k<i; k++)
562 			kprintf(" ");
563 
564 		kprintf("%d %s ", oidp->oid_number, oidp->oid_name);
565 
566 		kprintf("%c%c",
567 			oidp->oid_kind & CTLFLAG_RD ? 'R':' ',
568 			oidp->oid_kind & CTLFLAG_WR ? 'W':' ');
569 
570 		if (oidp->oid_handler)
571 			kprintf(" *Handler");
572 
573 		switch (oidp->oid_kind & CTLTYPE) {
574 		case CTLTYPE_NODE:
575 			kprintf(" Node\n");
576 			if (!oidp->oid_handler) {
577 				sysctl_sysctl_debug_dump_node(
578 					oidp->oid_arg1, i+2);
579 			}
580 			break;
581 		case CTLTYPE_INT:
582 			kprintf(" Int\n");
583 			break;
584 		case CTLTYPE_UINT:
585 			kprintf(" u_int\n");
586 			break;
587 		case CTLTYPE_LONG:
588 			kprintf(" Long\n");
589 			break;
590 		case CTLTYPE_ULONG:
591 			kprintf(" u_long\n");
592 			break;
593 		case CTLTYPE_STRING:
594 			kprintf(" String\n");
595 			break;
596 		case CTLTYPE_S8:
597 			kprintf(" int8_t\n");
598 			break;
599 		case CTLTYPE_S16:
600 			kprintf(" int16_t\n");
601 			break;
602 		case CTLTYPE_S32:
603 			kprintf(" int32_t\n");
604 			break;
605 		case CTLTYPE_S64:
606 			kprintf(" int64_t\n");
607 			break;
608 		case CTLTYPE_U8:
609 			kprintf(" uint8_t\n");
610 			break;
611 		case CTLTYPE_U16:
612 			kprintf(" uint16_t\n");
613 			break;
614 		case CTLTYPE_U32:
615 			kprintf(" uint32_t\n");
616 			break;
617 		case CTLTYPE_U64:
618 			kprintf(" uint64_t\n");
619 			break;
620 		case CTLTYPE_BIT32(0):
621 			kprintf(" Int\n");
622 			break;
623 		case CTLTYPE_BIT64(0):
624 			kprintf(" Int\n");
625 			break;
626 		case CTLTYPE_OPAQUE:
627 			kprintf(" Opaque/struct\n");
628 			break;
629 		default:
630 			kprintf("\n");
631 			break;
632 		}
633 
634 	}
635 }
636 
637 static int
638 sysctl_sysctl_debug(SYSCTL_HANDLER_ARGS)
639 {
640 	int error;
641 
642 	error = priv_check(req->td, PRIV_SYSCTL_DEBUG);
643 	if (error)
644 		return (error);
645 	sysctl_sysctl_debug_dump_node(&sysctl__children, 0);
646 
647 	return (ENOENT);
648 }
649 
650 SYSCTL_PROC(_sysctl, CTL_SYSCTL_DEBUG, debug, CTLTYPE_STRING | CTLFLAG_RD,
651 	    0, 0, sysctl_sysctl_debug, "-", "");
652 #endif /* SYSCTL_DEBUG */
653 
654 static int
655 sysctl_sysctl_name(SYSCTL_HANDLER_ARGS)
656 {
657 	int *name = (int *) arg1;
658 	u_int namelen = arg2;
659 	int error = 0;
660 	struct sysctl_oid *oid;
661 	struct sysctl_oid_list *lsp = &sysctl__children, *lsp2;
662 	char buf[10];
663 
664 	while (namelen) {
665 		if (!lsp) {
666 			ksnprintf(buf, sizeof(buf), "%d",  *name);
667 			if (req->oldidx)
668 				error = SYSCTL_OUT(req, ".", 1);
669 			if (!error)
670 				error = SYSCTL_OUT(req, buf, strlen(buf));
671 			if (error)
672 				goto out;
673 			namelen--;
674 			name++;
675 			continue;
676 		}
677 		lsp2 = NULL;
678 		SLIST_FOREACH(oid, lsp, oid_link) {
679 			if (oid->oid_number != *name)
680 				continue;
681 
682 			if (req->oldidx)
683 				error = SYSCTL_OUT(req, ".", 1);
684 			if (!error)
685 				error = SYSCTL_OUT(req, oid->oid_name,
686 					strlen(oid->oid_name));
687 			if (error)
688 				goto out;
689 
690 			namelen--;
691 			name++;
692 
693 			if ((oid->oid_kind & CTLTYPE) != CTLTYPE_NODE)
694 				break;
695 
696 			if (oid->oid_handler)
697 				break;
698 
699 			lsp2 = SYSCTL_CHILDREN(oid);
700 			break;
701 		}
702 		lsp = lsp2;
703 	}
704 	error = SYSCTL_OUT(req, "", 1);
705  out:
706 	return (error);
707 }
708 
709 SYSCTL_NODE(_sysctl, CTL_SYSCTL_NAME, name, CTLFLAG_RD | CTLFLAG_NOLOCK,
710 	    sysctl_sysctl_name, "");
711 
712 static int
713 sysctl_sysctl_next_ls(struct sysctl_oid_list *lsp, int *name, u_int namelen,
714 	int *next, int *len, int level, struct sysctl_oid **oidpp)
715 {
716 	struct sysctl_oid *oidp;
717 
718 	*len = level;
719 	SLIST_FOREACH(oidp, lsp, oid_link) {
720 		*next = oidp->oid_number;
721 		*oidpp = oidp;
722 
723 		if (oidp->oid_kind & CTLFLAG_SKIP)
724 			continue;
725 
726 		if (!namelen) {
727 			if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
728 				return (0);
729 			if (oidp->oid_handler)
730 				/* We really should call the handler here...*/
731 				return (0);
732 			lsp = SYSCTL_CHILDREN(oidp);
733 			if (!sysctl_sysctl_next_ls(lsp, 0, 0, next+1,
734 				len, level+1, oidpp))
735 				return (0);
736 			goto emptynode;
737 		}
738 
739 		if (oidp->oid_number < *name)
740 			continue;
741 
742 		if (oidp->oid_number > *name) {
743 			if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
744 				return (0);
745 			if (oidp->oid_handler)
746 				return (0);
747 			lsp = SYSCTL_CHILDREN(oidp);
748 			if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1,
749 				next+1, len, level+1, oidpp))
750 				return (0);
751 			goto next;
752 		}
753 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
754 			continue;
755 
756 		if (oidp->oid_handler)
757 			continue;
758 
759 		lsp = SYSCTL_CHILDREN(oidp);
760 		if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1, next+1,
761 			len, level+1, oidpp))
762 			return (0);
763 	next:
764 		namelen = 1;
765 	emptynode:
766 		*len = level;
767 	}
768 	return (1);
769 }
770 
771 static int
772 sysctl_sysctl_next(SYSCTL_HANDLER_ARGS)
773 {
774 	int *name = (int *) arg1;
775 	u_int namelen = arg2;
776 	int i, j, error;
777 	struct sysctl_oid *oid;
778 	struct sysctl_oid_list *lsp = &sysctl__children;
779 	int newoid[CTL_MAXNAME];
780 
781 	i = sysctl_sysctl_next_ls(lsp, name, namelen, newoid, &j, 1, &oid);
782 	if (i)
783 		return ENOENT;
784 	error = SYSCTL_OUT(req, newoid, j * sizeof (int));
785 
786 	return (error);
787 }
788 
789 SYSCTL_NODE(_sysctl, CTL_SYSCTL_NEXT, next, CTLFLAG_RD | CTLFLAG_NOLOCK,
790 	    sysctl_sysctl_next, "");
791 
792 static int
793 name2oid(char *name, int *oid, int *len, struct sysctl_oid **oidpp)
794 {
795 	struct sysctl_oid *oidp;
796 	struct sysctl_oid_list *lsp = &sysctl__children;
797 	char *p;
798 
799 	SYSCTL_ASSERT_LOCKED();
800 
801 	for (*len = 0; *len < CTL_MAXNAME;) {
802 		p = strsep(&name, ".");
803 
804 		oidp = SLIST_FIRST(lsp);
805 		for (;; oidp = SLIST_NEXT(oidp, oid_link)) {
806 			if (oidp == NULL)
807 				return (ENOENT);
808 			if (strcmp(p, oidp->oid_name) == 0)
809 				break;
810 		}
811 		*oid++ = oidp->oid_number;
812 		(*len)++;
813 
814 		if (name == NULL || *name == '\0') {
815 			if (oidpp)
816 				*oidpp = oidp;
817 			return (0);
818 		}
819 
820 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
821 			break;
822 
823 		if (oidp->oid_handler)
824 			break;
825 
826 		lsp = SYSCTL_CHILDREN(oidp);
827 	}
828 	return (ENOENT);
829 }
830 
831 static int
832 sysctl_sysctl_name2oid(SYSCTL_HANDLER_ARGS)
833 {
834 	char *p;
835 	int error, oid[CTL_MAXNAME], len;
836 	struct sysctl_oid *op = NULL;
837 
838 	if (!req->newlen)
839 		return ENOENT;
840 	if (req->newlen >= MAXPATHLEN)	/* XXX arbitrary, undocumented */
841 		return (ENAMETOOLONG);
842 
843 	p = kmalloc(req->newlen+1, M_SYSCTL, M_WAITOK);
844 
845 	error = SYSCTL_IN(req, p, req->newlen);
846 	if (error) {
847 		kfree(p, M_SYSCTL);
848 		return (error);
849 	}
850 
851 	p [req->newlen] = '\0';
852 
853 	error = name2oid(p, oid, &len, &op);
854 
855 	kfree(p, M_SYSCTL);
856 
857 	if (error)
858 		return (error);
859 
860 	error = SYSCTL_OUT(req, oid, len * sizeof *oid);
861 	return (error);
862 }
863 
864 SYSCTL_PROC(_sysctl, CTL_SYSCTL_NAME2OID, name2oid,
865 	    CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_NOLOCK,
866 	    0, 0, sysctl_sysctl_name2oid, "I", "");
867 
868 static int
869 sysctl_sysctl_oidfmt(SYSCTL_HANDLER_ARGS)
870 {
871 	struct sysctl_oid *oid;
872 	int error;
873 
874 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
875 	if (error)
876 		return (error);
877 
878 	if (!oid->oid_fmt)
879 		return (ENOENT);
880 	error = SYSCTL_OUT(req, &oid->oid_kind, sizeof(oid->oid_kind));
881 	if (error)
882 		return (error);
883 	error = SYSCTL_OUT(req, oid->oid_fmt, strlen(oid->oid_fmt) + 1);
884 	return (error);
885 }
886 
887 
888 SYSCTL_NODE(_sysctl, CTL_SYSCTL_OIDFMT, oidfmt, CTLFLAG_RD | CTLFLAG_NOLOCK,
889 	    sysctl_sysctl_oidfmt, "");
890 
891 static int
892 sysctl_sysctl_oiddescr(SYSCTL_HANDLER_ARGS)
893 {
894 	struct sysctl_oid *oid;
895 	int error;
896 
897 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
898 	if (error)
899 		return (error);
900 
901 	if (!oid->oid_descr)
902 		return (ENOENT);
903 	error = SYSCTL_OUT(req, oid->oid_descr, strlen(oid->oid_descr) + 1);
904 	return (error);
905 }
906 
907 SYSCTL_NODE(_sysctl, CTL_SYSCTL_OIDDESCR, oiddescr,
908 	    CTLFLAG_RD | CTLFLAG_NOLOCK,
909 	    sysctl_sysctl_oiddescr, "");
910 
911 /*
912  * Default "handler" functions.
913  */
914 
915 /*
916  * Handle an 8-bit number, signed or unsigned.  arg1 points to it.
917  */
918 
919 int
920 sysctl_handle_8(SYSCTL_HANDLER_ARGS)
921 {
922 	int error = 0;
923 
924 	if (!arg1)
925 		return (EINVAL);
926 	error = SYSCTL_OUT(req, arg1, sizeof(int8_t));
927 
928 	if (error || !req->newptr)
929 		return (error);
930 
931 	error = SYSCTL_IN(req, arg1, sizeof(int8_t));
932 	return (error);
933 }
934 
935 /*
936  * Handle a 16-bit number, signed or unsigned.  arg1 points to it.
937  */
938 
939 int
940 sysctl_handle_16(SYSCTL_HANDLER_ARGS)
941 {
942 	int error = 0;
943 
944 	if (!arg1)
945 		return (EINVAL);
946 	error = SYSCTL_OUT(req, arg1, sizeof(int16_t));
947 
948 	if (error || !req->newptr)
949 		return (error);
950 
951 	error = SYSCTL_IN(req, arg1, sizeof(int16_t));
952 	return (error);
953 }
954 
955 /*
956  * Handle a 32-bit number, signed or unsigned.  arg1 points to it.
957  */
958 
959 int
960 sysctl_handle_32(SYSCTL_HANDLER_ARGS)
961 {
962 	int error = 0;
963 
964 	if (!arg1)
965 		return (EINVAL);
966 	error = SYSCTL_OUT(req, arg1, sizeof(int32_t));
967 
968 	if (error || !req->newptr)
969 		return (error);
970 
971 	error = SYSCTL_IN(req, arg1, sizeof(int32_t));
972 	return (error);
973 }
974 
975 /*
976  * Handle a 64-bit number, signed or unsigned.  arg1 points to it.
977  */
978 
979 int
980 sysctl_handle_64(SYSCTL_HANDLER_ARGS)
981 {
982 	int error = 0;
983 
984 	if (!arg1)
985 		return (EINVAL);
986 	error = SYSCTL_OUT(req, arg1, sizeof(int64_t));
987 
988 	if (error || !req->newptr)
989 		return (error);
990 
991 	error = SYSCTL_IN(req, arg1, sizeof(int64_t));
992 	return (error);
993 }
994 
995 /*
996  * Handle an int, signed or unsigned.
997  * Two cases:
998  *     a variable:  point arg1 at it.
999  *     a constant:  pass it in arg2.
1000  */
1001 
1002 int
1003 sysctl_handle_int(SYSCTL_HANDLER_ARGS)
1004 {
1005 	int error = 0;
1006 
1007 	if (arg1)
1008 		error = SYSCTL_OUT(req, arg1, sizeof(int));
1009 	else
1010 		error = SYSCTL_OUT(req, &arg2, sizeof(int));
1011 
1012 	if (error || !req->newptr)
1013 		return (error);
1014 
1015 	if (!arg1)
1016 		error = EPERM;
1017 	else
1018 		error = SYSCTL_IN(req, arg1, sizeof(int));
1019 	return (error);
1020 }
1021 
1022 /*
1023  * Handle a long, signed or unsigned.  arg1 points to it.
1024  */
1025 
1026 int
1027 sysctl_handle_long(SYSCTL_HANDLER_ARGS)
1028 {
1029 	int error = 0;
1030 
1031 	if (!arg1)
1032 		return (EINVAL);
1033 	if (req->oldlen == sizeof(int) &&
1034 	    *(long *)arg1 >= INT_MIN &&
1035 	    *(long *)arg1 <= INT_MAX) {
1036 		/*
1037 		 * Backwards compatibility for read-only fields promoted
1038 		 * from int to long.  Allow userland to request the field
1039 		 * as an integer if the value is in-range.
1040 		 */
1041 		int val = (int)*(long *)arg1;
1042 		error = SYSCTL_OUT(req, &val, sizeof(int));
1043 	} else {
1044 		/*
1045 		 * Normal operation fo a long
1046 		 */
1047 		error = SYSCTL_OUT(req, arg1, sizeof(long));
1048 	}
1049 
1050 	if (error || !req->newptr)
1051 		return (error);
1052 
1053 	error = SYSCTL_IN(req, arg1, sizeof(long));
1054 
1055 	return (error);
1056 }
1057 
1058 /*
1059  * Handle a quad, signed or unsigned.  arg1 points to it.
1060  */
1061 
1062 int
1063 sysctl_handle_quad(SYSCTL_HANDLER_ARGS)
1064 {
1065 	int error = 0;
1066 
1067 	if (!arg1)
1068 		return (EINVAL);
1069 	error = SYSCTL_OUT(req, arg1, sizeof(quad_t));
1070 
1071 	if (error || !req->newptr)
1072 		return (error);
1073 
1074 	error = SYSCTL_IN(req, arg1, sizeof(quad_t));
1075 	return (error);
1076 }
1077 
1078 /*
1079  * Handle an bit in a 32-bit field, pass and return an 'int'
1080  * Two cases:
1081  *     a variable:  point arg1 at it.
1082  *     a constant:  pass it in arg2.
1083  */
1084 
1085 int
1086 sysctl_handle_bit32(SYSCTL_HANDLER_ARGS)
1087 {
1088 	int error = 0;
1089 	uint32_t mask;
1090 	int v;
1091 	int bit;
1092 
1093 	bit = (oidp->oid_kind & CTLMASK_BITFLD) >> CTLSHIFT_BITFLD;
1094 	mask = arg1 ? *(uint32_t *)arg1 : (uint32_t)arg2;
1095 	v = (mask & (1U << bit)) ? 1 : 0;
1096 	error = SYSCTL_OUT(req, &v, sizeof(int));
1097 
1098 	if (error || !req->newptr)
1099 		return (error);
1100 
1101 	if (!arg1) {
1102 		error = EPERM;
1103 	} else {
1104 		error = SYSCTL_IN(req, &v, sizeof(int));
1105 		if (error == 0) {
1106 			if (v)
1107 				atomic_set_int((uint32_t *)arg1, 1U << bit);
1108 			else
1109 				atomic_clear_int((uint32_t *)arg1, 1U << bit);
1110 		}
1111 	}
1112 	return (error);
1113 }
1114 
1115 /*
1116  * Handle an bit in a 64-bit field, pass and return an 'int'
1117  * Two cases:
1118  *     a variable:  point arg1 at it.
1119  *     a constant:  pass it in arg2.  (NOTE: arg2 is only 32bits)
1120  */
1121 
1122 int
1123 sysctl_handle_bit64(SYSCTL_HANDLER_ARGS)
1124 {
1125 	int error = 0;
1126 	uint64_t mask;
1127 	int v;
1128 	int bit;
1129 
1130 	bit = (oidp->oid_kind & CTLMASK_BITFLD) >> CTLSHIFT_BITFLD;
1131 	mask = arg1 ? *(uint64_t *)arg1 : (uint64_t)(uint32_t)arg2;
1132 	v = (mask & (1LU << bit)) ? 1 : 0;
1133 	error = SYSCTL_OUT(req, &v, sizeof(int));
1134 
1135 	if (error || !req->newptr)
1136 		return (error);
1137 
1138 	if (!arg1) {
1139 		error = EPERM;
1140 	} else {
1141 		error = SYSCTL_IN(req, &v, sizeof(int));
1142 		if (error == 0) {
1143 			if (v)
1144 				atomic_set_long((uint64_t *)arg1, 1LU << bit);
1145 			else
1146 				atomic_clear_long((uint64_t *)arg1, 1LU << bit);
1147 		}
1148 	}
1149 	return (error);
1150 }
1151 
1152 /*
1153  * Handle our generic '\0' terminated 'C' string.
1154  * Two cases:
1155  *	a variable string:  point arg1 at it, arg2 is max length.
1156  *	a constant string:  point arg1 at it, arg2 is zero.
1157  */
1158 
1159 int
1160 sysctl_handle_string(SYSCTL_HANDLER_ARGS)
1161 {
1162 	int error=0;
1163 
1164 	error = SYSCTL_OUT(req, arg1, strlen((char *)arg1)+1);
1165 
1166 	if (error || !req->newptr)
1167 		return (error);
1168 
1169 	if ((req->newlen - req->newidx) >= arg2) {
1170 		error = EINVAL;
1171 	} else {
1172 		arg2 = (req->newlen - req->newidx);
1173 		error = SYSCTL_IN(req, arg1, arg2);
1174 		((char *)arg1)[arg2] = '\0';
1175 	}
1176 
1177 	return (error);
1178 }
1179 
1180 /*
1181  * Handle any kind of opaque data.
1182  * arg1 points to it, arg2 is the size.
1183  */
1184 
1185 int
1186 sysctl_handle_opaque(SYSCTL_HANDLER_ARGS)
1187 {
1188 	int error;
1189 
1190 	error = SYSCTL_OUT(req, arg1, arg2);
1191 
1192 	if (error || !req->newptr)
1193 		return (error);
1194 
1195 	error = SYSCTL_IN(req, arg1, arg2);
1196 
1197 	return (error);
1198 }
1199 
1200 /*
1201  * Transfer functions to/from kernel space.
1202  * XXX: rather untested at this point
1203  */
1204 static int
1205 sysctl_old_kernel(struct sysctl_req *req, const void *p, size_t l)
1206 {
1207 	size_t i = 0;
1208 
1209 	if (req->oldptr) {
1210 		i = l;
1211 		if (i > req->oldlen - req->oldidx)
1212 			i = req->oldlen - req->oldidx;
1213 		if (i > 0)
1214 			bcopy(p, (char *)req->oldptr + req->oldidx, i);
1215 	}
1216 	req->oldidx += l;
1217 	if (req->oldptr && i != l)
1218 		return (ENOMEM);
1219 	return (0);
1220 }
1221 
1222 static int
1223 sysctl_new_kernel(struct sysctl_req *req, void *p, size_t l)
1224 {
1225 
1226 	if (!req->newptr)
1227 		return 0;
1228 	if (req->newlen - req->newidx < l)
1229 		return (EINVAL);
1230 	bcopy((char *)req->newptr + req->newidx, p, l);
1231 	req->newidx += l;
1232 	return (0);
1233 }
1234 
1235 int
1236 kernel_sysctl(int *name, u_int namelen,
1237 	      void *old, size_t *oldlenp,
1238 	      void *new, size_t newlen, size_t *retval)
1239 {
1240 	int error = 0;
1241 	struct sysctl_req req;
1242 
1243 	bzero(&req, sizeof req);
1244 
1245 	req.td = curthread;
1246 
1247 	if (oldlenp) {
1248 		req.oldlen = *oldlenp;
1249 	}
1250 	req.validlen = req.oldlen;
1251 
1252 	if (old) {
1253 		req.oldptr= old;
1254 	}
1255 
1256 	if (new != NULL) {
1257 		req.newlen = newlen;
1258 		req.newptr = new;
1259 	}
1260 
1261 	req.oldfunc = sysctl_old_kernel;
1262 	req.newfunc = sysctl_new_kernel;
1263 #if 0
1264 	req.lock = REQ_UNWIRED;
1265 #endif
1266 
1267 	SYSCTL_SLOCK();
1268 	error = sysctl_root(0, name, namelen, &req);
1269 	SYSCTL_SUNLOCK();
1270 
1271 #if 0
1272 	if (req.lock == REQ_WIRED && req.validlen > 0)
1273 		vsunlock(req.oldptr, req.validlen);
1274 #endif
1275 
1276 	if (error && error != ENOMEM)
1277 		return (error);
1278 
1279 	if (retval) {
1280 		if (req.oldptr && req.oldidx > req.validlen)
1281 			*retval = req.validlen;
1282 		else
1283 			*retval = req.oldidx;
1284 	}
1285 	return (error);
1286 }
1287 
1288 int
1289 kernel_sysctlbyname(char *name,
1290 		    void *old, size_t *oldlenp,
1291 		    void *new, size_t newlen, size_t *retval)
1292 {
1293         int oid[CTL_MAXNAME];
1294         size_t oidlen, plen;
1295 	int error;
1296 
1297 	oid[0] = CTL_SYSCTL;
1298 	oid[1] = CTL_SYSCTL_NAME2OID;
1299 	oidlen = sizeof(oid);
1300 
1301 	error = kernel_sysctl(oid, 2, oid, &oidlen, name, strlen(name), &plen);
1302 	if (error)
1303 		return (error);
1304 
1305 	error = kernel_sysctl(oid, plen / sizeof(int), old, oldlenp,
1306 	    new, newlen, retval);
1307 	return (error);
1308 }
1309 
1310 /*
1311  * Transfer function to/from user space.
1312  */
1313 static int
1314 sysctl_old_user(struct sysctl_req *req, const void *p, size_t l)
1315 {
1316 	int error = 0;
1317 	size_t i = 0;
1318 
1319 #if 0
1320 	if (req->lock == 1 && req->oldptr) {
1321 		vslock(req->oldptr, req->oldlen);
1322 		req->lock = 2;
1323 	}
1324 #endif
1325 	if (req->oldptr) {
1326 		i = l;
1327 		if (i > req->oldlen - req->oldidx)
1328 			i = req->oldlen - req->oldidx;
1329 		if (i > 0)
1330 			error = copyout(p, (char *)req->oldptr + req->oldidx,
1331 					i);
1332 	}
1333 	req->oldidx += l;
1334 	if (error)
1335 		return (error);
1336 	if (req->oldptr && i < l)
1337 		return (ENOMEM);
1338 	return (0);
1339 }
1340 
1341 static int
1342 sysctl_new_user(struct sysctl_req *req, void *p, size_t l)
1343 {
1344 	int error;
1345 
1346 	if (!req->newptr)
1347 		return 0;
1348 	if (req->newlen - req->newidx < l)
1349 		return (EINVAL);
1350 	error = copyin((char *)req->newptr + req->newidx, p, l);
1351 	req->newidx += l;
1352 	return (error);
1353 }
1354 
1355 int
1356 sysctl_find_oid(int *name, u_int namelen, struct sysctl_oid **noid,
1357 		int *nindx, struct sysctl_req *req)
1358 {
1359 	struct sysctl_oid_list *lsp;
1360 	struct sysctl_oid *oid;
1361 	int indx;
1362 
1363 	lsp = &sysctl__children;
1364 	indx = 0;
1365 	while (indx < CTL_MAXNAME) {
1366 		SLIST_FOREACH(oid, lsp, oid_link) {
1367 			if (oid->oid_number == name[indx])
1368 				break;
1369 		}
1370 		if (oid == NULL)
1371 			return (ENOENT);
1372 
1373 		indx++;
1374 		if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1375 			if (oid->oid_handler != NULL || indx == namelen) {
1376 				*noid = oid;
1377 				if (nindx != NULL)
1378 					*nindx = indx;
1379 				KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
1380 				    ("%s found DYING node %p", __func__, oid));
1381 				return (0);
1382 			}
1383 			lsp = SYSCTL_CHILDREN(oid);
1384 		} else if (indx == namelen) {
1385 			*noid = oid;
1386 			if (nindx != NULL)
1387 				*nindx = indx;
1388 			KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
1389 			    ("%s found DYING node %p", __func__, oid));
1390 			return (0);
1391 		} else {
1392 			return (ENOTDIR);
1393 		}
1394 	}
1395 	return (ENOENT);
1396 }
1397 
1398 /*
1399  * Traverse our tree, and find the right node, execute whatever it points
1400  * to, and return the resulting error code.
1401  */
1402 static int
1403 sysctl_root(SYSCTL_HANDLER_ARGS)
1404 {
1405 	struct thread *td = req->td;
1406 	struct proc *p = td ? td->td_proc : NULL;
1407 	struct sysctl_oid *oid;
1408 	int error, indx;
1409 	int lktype;
1410 
1411 	error = sysctl_find_oid(arg1, arg2, &oid, &indx, req);
1412 	if (error)
1413 		return (error);
1414 
1415 	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1416 		/*
1417 		 * You can't call a sysctl when it's a node, but has
1418 		 * no handler.  Inform the user that it's a node.
1419 		 * The indx may or may not be the same as namelen.
1420 		 */
1421 		if (oid->oid_handler == NULL)
1422 			return (EISDIR);
1423 	}
1424 
1425 	/* If writing isn't allowed */
1426 	if (req->newptr && (!(oid->oid_kind & CTLFLAG_WR) ||
1427 	    ((oid->oid_kind & CTLFLAG_SECURE) && securelevel > 0)))
1428 		return (EPERM);
1429 
1430 	/* Most likely only root can write */
1431 	if (!(oid->oid_kind & CTLFLAG_ANYBODY) && req->newptr && p &&
1432 	    (error = priv_check_cred(td->td_ucred,
1433 	     (oid->oid_kind & CTLFLAG_PRISON) ? PRIV_SYSCTL_WRITEJAIL :
1434 	                                        PRIV_SYSCTL_WRITE, 0)))
1435 		return (error);
1436 
1437 	if (oid->oid_handler == NULL)
1438 		return EINVAL;
1439 
1440 	/*
1441 	 * Default oid locking is exclusive when modifying (newptr),
1442 	 * shared otherwise, unless overridden with a control flag.
1443 	 */
1444 	if ((oid->oid_kind & CTLFLAG_NOLOCK) == 0) {
1445 		lktype = (req->newptr != NULL) ? LK_EXCLUSIVE : LK_SHARED;
1446 		if (oid->oid_kind & CTLFLAG_SHLOCK)
1447 			lktype = LK_SHARED;
1448 		if (oid->oid_kind & CTLFLAG_EXLOCK)
1449 			lktype = LK_EXCLUSIVE;
1450 #if 1
1451 		lockmgr(&oid->oid_lock, lktype);
1452 #else
1453 		/* DEBUGGING */
1454 		if (lockmgr(&oid->oid_lock, lktype | LK_SLEEPFAIL)) {
1455 			kprintf("%s\n", oid->oid_name);
1456 			lockmgr(&oid->oid_lock, lktype);
1457 		}
1458 #endif
1459 	}
1460 
1461 	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE)
1462 		error = oid->oid_handler(oid, (int *)arg1 + indx, arg2 - indx,
1463 					 req);
1464 	else
1465 		error = oid->oid_handler(oid, oid->oid_arg1, oid->oid_arg2,
1466 					 req);
1467 
1468 	if ((oid->oid_kind & CTLFLAG_NOLOCK) == 0)
1469 		lockmgr(&oid->oid_lock, LK_RELEASE);
1470 	return (error);
1471 }
1472 
1473 int
1474 sys___sysctl(struct sysmsg *sysmsg, const struct sysctl_args *uap)
1475 {
1476 	int error, i, name[CTL_MAXNAME];
1477 	size_t j;
1478 
1479 	if (uap->namelen > CTL_MAXNAME || uap->namelen < 2)
1480 		return (EINVAL);
1481 
1482 	error = copyin(uap->name, &name, uap->namelen * sizeof(int));
1483 	if (error)
1484 		return (error);
1485 
1486 	error = userland_sysctl(name, uap->namelen,
1487 				uap->old, uap->oldlenp, 0,
1488 				uap->new, uap->newlen, &j);
1489 	if (error && error != ENOMEM)
1490 		return (error);
1491 	if (uap->oldlenp) {
1492 		i = copyout(&j, uap->oldlenp, sizeof(j));
1493 		if (i)
1494 			return (i);
1495 	}
1496 	return (error);
1497 }
1498 
1499 /*
1500  * This is used from various compatibility syscalls too.  That's why name
1501  * must be in kernel space.
1502  */
1503 int
1504 userland_sysctl(int *name, u_int namelen,
1505 		void *old, size_t *oldlenp, int inkernel,
1506 		void *new, size_t newlen, size_t *retval)
1507 {
1508 	int error = 0;
1509 	struct sysctl_req req;
1510 
1511 	bzero(&req, sizeof req);
1512 
1513 	req.td = curthread;
1514 	req.flags = 0;
1515 
1516 	if (oldlenp) {
1517 		if (inkernel) {
1518 			req.oldlen = *oldlenp;
1519 		} else {
1520 			error = copyin(oldlenp, &req.oldlen, sizeof(*oldlenp));
1521 			if (error)
1522 				return (error);
1523 		}
1524 	}
1525 	req.validlen = req.oldlen;
1526 
1527 	/*
1528 	 * NOTE: User supplied buffers are not guaranteed to be good,
1529 	 *	 the sysctl copyins and copyouts can fail.
1530 	 */
1531 	if (old)
1532 		req.oldptr= old;
1533 
1534 	if (new != NULL) {
1535 		req.newlen = newlen;
1536 		req.newptr = new;
1537 	}
1538 
1539 	req.oldfunc = sysctl_old_user;
1540 	req.newfunc = sysctl_new_user;
1541 #if 0
1542 	req.lock = REQ_UNWIRED;
1543 #endif
1544 
1545 #ifdef KTRACE
1546 	if (KTRPOINT(curthread, KTR_SYSCTL))
1547 		ktrsysctl(name, namelen);
1548 #endif
1549 
1550 	for (;;) {
1551 		req.oldidx = 0;
1552 		req.newidx = 0;
1553 		SYSCTL_SLOCK();
1554 		error = sysctl_root(0, name, namelen, &req);
1555 		SYSCTL_SUNLOCK();
1556 		if (error != EAGAIN)
1557 			break;
1558 		lwkt_yield();
1559 	}
1560 
1561 #if 0
1562 	if (req.lock == REQ_WIRED && req.validlen > 0)
1563 		vsunlock(req.oldptr, req.validlen);
1564 #endif
1565 	if (error && error != ENOMEM)
1566 		return (error);
1567 
1568 	if (retval) {
1569 		if (req.oldptr && req.oldidx > req.validlen)
1570 			*retval = req.validlen;
1571 		else
1572 			*retval = req.oldidx;
1573 	}
1574 	return (error);
1575 }
1576 
1577 int
1578 sysctl_int_range(SYSCTL_HANDLER_ARGS, int low, int high)
1579 {
1580 	int error, value;
1581 
1582 	value = *(int *)arg1;
1583 	error = sysctl_handle_int(oidp, &value, 0, req);
1584 	if (error || !req->newptr)
1585 		return (error);
1586 	if (value < low || value > high)
1587 		return (EINVAL);
1588 	*(int *)arg1 = value;
1589 	return (0);
1590 }
1591 
1592 /*
1593  * Drain into a sysctl struct.  The user buffer should be wired if a page
1594  * fault would cause issue.
1595  */
1596 static int
1597 sbuf_sysctl_drain(void *arg, const char *data, int len)
1598 {
1599 	struct sysctl_req *req = arg;
1600 	int error;
1601 
1602 	error = SYSCTL_OUT(req, data, len);
1603 	KASSERT(error >= 0, ("Got unexpected negative value %d", error));
1604 	return (error == 0 ? len : -error);
1605 }
1606 
1607 struct sbuf *
1608 sbuf_new_for_sysctl(struct sbuf *s, char *buf, int length,
1609     struct sysctl_req *req)
1610 {
1611 
1612 	s = sbuf_new(s, buf, length, SBUF_FIXEDLEN);
1613 	sbuf_set_drain(s, sbuf_sysctl_drain, req);
1614 	return (s);
1615 }
1616 
1617 /*
1618  * The exclusive sysctl lock only protects its topology, and is
1619  * very expensive, but allows us to use a pcpu shared lock for
1620  * critical path accesses.
1621  */
1622 void
1623 _sysctl_xlock(void)
1624 {
1625 	globaldata_t gd;
1626 	int i;
1627 
1628 	for (i = 0; i < ncpus; ++i) {
1629 		gd = globaldata_find(i);
1630 		lockmgr(&gd->gd_sysctllock, LK_EXCLUSIVE);
1631 	}
1632 }
1633 
1634 void
1635 _sysctl_xunlock(void)
1636 {
1637 	globaldata_t gd;
1638 	int i;
1639 
1640 	for (i = 0; i < ncpus; ++i) {
1641 		gd = globaldata_find(i);
1642 		lockmgr(&gd->gd_sysctllock, LK_RELEASE);
1643 	}
1644 }
1645