xref: /freebsd/sys/kern/kern_sysctl.c (revision 908e960ea6343acd9515d89d5d5696f9d8bf090c)
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  * 4. 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  */
37 
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
40 
41 #include "opt_compat.h"
42 #include "opt_ktrace.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/kernel.h>
47 #include <sys/sysctl.h>
48 #include <sys/malloc.h>
49 #include <sys/priv.h>
50 #include <sys/proc.h>
51 #include <sys/lock.h>
52 #include <sys/mutex.h>
53 #include <sys/sx.h>
54 #include <sys/sysproto.h>
55 #include <sys/uio.h>
56 #include <sys/vimage.h>
57 #ifdef KTRACE
58 #include <sys/ktrace.h>
59 #endif
60 
61 #include <security/mac/mac_framework.h>
62 
63 #include <vm/vm.h>
64 #include <vm/vm_extern.h>
65 
66 static MALLOC_DEFINE(M_SYSCTL, "sysctl", "sysctl internal magic");
67 static MALLOC_DEFINE(M_SYSCTLOID, "sysctloid", "sysctl dynamic oids");
68 static MALLOC_DEFINE(M_SYSCTLTMP, "sysctltmp", "sysctl temp output buffer");
69 
70 /*
71  * The sysctllock protects the MIB tree.  It also protects sysctl
72  * contexts used with dynamic sysctls.  The sysctl_register_oid() and
73  * sysctl_unregister_oid() routines require the sysctllock to already
74  * be held, so the sysctl_lock() and sysctl_unlock() routines are
75  * provided for the few places in the kernel which need to use that
76  * API rather than using the dynamic API.  Use of the dynamic API is
77  * strongly encouraged for most code.
78  *
79  * The sysctlmemlock is used to limit the amount of user memory wired for
80  * sysctl requests.  This is implemented by serializing any userland
81  * sysctl requests larger than a single page via an exclusive lock.
82  */
83 static struct sx sysctllock;
84 static struct sx sysctlmemlock;
85 
86 #define	SYSCTL_SLOCK()		sx_slock(&sysctllock)
87 #define	SYSCTL_SUNLOCK()	sx_sunlock(&sysctllock)
88 #define	SYSCTL_XLOCK()		sx_xlock(&sysctllock)
89 #define	SYSCTL_XUNLOCK()	sx_xunlock(&sysctllock)
90 #define	SYSCTL_ASSERT_XLOCKED()	sx_assert(&sysctllock, SA_XLOCKED)
91 #define	SYSCTL_ASSERT_LOCKED()	sx_assert(&sysctllock, SA_LOCKED)
92 #define	SYSCTL_INIT()		sx_init(&sysctllock, "sysctl lock")
93 
94 static int sysctl_root(SYSCTL_HANDLER_ARGS);
95 
96 struct sysctl_oid_list sysctl__children; /* root list */
97 
98 static int	sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del,
99 		    int recurse);
100 
101 static struct sysctl_oid *
102 sysctl_find_oidname(const char *name, struct sysctl_oid_list *list)
103 {
104 	struct sysctl_oid *oidp;
105 
106 	SYSCTL_ASSERT_LOCKED();
107 	SLIST_FOREACH(oidp, list, oid_link) {
108 		if (strcmp(oidp->oid_name, name) == 0) {
109 			return (oidp);
110 		}
111 	}
112 	return (NULL);
113 }
114 
115 /*
116  * Initialization of the MIB tree.
117  *
118  * Order by number in each list.
119  */
120 void
121 sysctl_lock(void)
122 {
123 
124 	SYSCTL_XLOCK();
125 }
126 
127 void
128 sysctl_unlock(void)
129 {
130 
131 	SYSCTL_XUNLOCK();
132 }
133 
134 void
135 sysctl_register_oid(struct sysctl_oid *oidp)
136 {
137 	struct sysctl_oid_list *parent = oidp->oid_parent;
138 	struct sysctl_oid *p;
139 	struct sysctl_oid *q;
140 
141 	/*
142 	 * First check if another oid with the same name already
143 	 * exists in the parent's list.
144 	 */
145 	SYSCTL_ASSERT_XLOCKED();
146 	p = sysctl_find_oidname(oidp->oid_name, parent);
147 	if (p != NULL) {
148 		if ((p->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
149 			p->oid_refcnt++;
150 			return;
151 		} else {
152 			printf("can't re-use a leaf (%s)!\n", p->oid_name);
153 			return;
154 		}
155 	}
156 	/*
157 	 * If this oid has a number OID_AUTO, give it a number which
158 	 * is greater than any current oid.
159 	 * NOTE: DO NOT change the starting value here, change it in
160 	 * <sys/sysctl.h>, and make sure it is at least 256 to
161 	 * accomodate e.g. net.inet.raw as a static sysctl node.
162 	 */
163 	if (oidp->oid_number == OID_AUTO) {
164 		static int newoid = CTL_AUTO_START;
165 
166 		oidp->oid_number = newoid++;
167 		if (newoid == 0x7fffffff)
168 			panic("out of oids");
169 	}
170 #if 0
171 	else if (oidp->oid_number >= CTL_AUTO_START) {
172 		/* do not panic; this happens when unregistering sysctl sets */
173 		printf("static sysctl oid too high: %d", oidp->oid_number);
174 	}
175 #endif
176 
177 	/*
178 	 * Insert the oid into the parent's list in order.
179 	 */
180 	q = NULL;
181 	SLIST_FOREACH(p, parent, oid_link) {
182 		if (oidp->oid_number < p->oid_number)
183 			break;
184 		q = p;
185 	}
186 	if (q)
187 		SLIST_INSERT_AFTER(q, oidp, oid_link);
188 	else
189 		SLIST_INSERT_HEAD(parent, oidp, oid_link);
190 }
191 
192 void
193 sysctl_unregister_oid(struct sysctl_oid *oidp)
194 {
195 	struct sysctl_oid *p;
196 	int error;
197 
198 	SYSCTL_ASSERT_XLOCKED();
199 	error = ENOENT;
200 	if (oidp->oid_number == OID_AUTO) {
201 		error = EINVAL;
202 	} else {
203 		SLIST_FOREACH(p, oidp->oid_parent, oid_link) {
204 			if (p == oidp) {
205 				SLIST_REMOVE(oidp->oid_parent, oidp,
206 				    sysctl_oid, oid_link);
207 				error = 0;
208 				break;
209 			}
210 		}
211 	}
212 
213 	/*
214 	 * This can happen when a module fails to register and is
215 	 * being unloaded afterwards.  It should not be a panic()
216 	 * for normal use.
217 	 */
218 	if (error)
219 		printf("%s: failed to unregister sysctl\n", __func__);
220 }
221 
222 /* Initialize a new context to keep track of dynamically added sysctls. */
223 int
224 sysctl_ctx_init(struct sysctl_ctx_list *c)
225 {
226 
227 	if (c == NULL) {
228 		return (EINVAL);
229 	}
230 
231 	/*
232 	 * No locking here, the caller is responsible for not adding
233 	 * new nodes to a context until after this function has
234 	 * returned.
235 	 */
236 	TAILQ_INIT(c);
237 	return (0);
238 }
239 
240 /* Free the context, and destroy all dynamic oids registered in this context */
241 int
242 sysctl_ctx_free(struct sysctl_ctx_list *clist)
243 {
244 	struct sysctl_ctx_entry *e, *e1;
245 	int error;
246 
247 	error = 0;
248 	/*
249 	 * First perform a "dry run" to check if it's ok to remove oids.
250 	 * XXX FIXME
251 	 * XXX This algorithm is a hack. But I don't know any
252 	 * XXX better solution for now...
253 	 */
254 	SYSCTL_XLOCK();
255 	TAILQ_FOREACH(e, clist, link) {
256 		error = sysctl_remove_oid_locked(e->entry, 0, 0);
257 		if (error)
258 			break;
259 	}
260 	/*
261 	 * Restore deregistered entries, either from the end,
262 	 * or from the place where error occured.
263 	 * e contains the entry that was not unregistered
264 	 */
265 	if (error)
266 		e1 = TAILQ_PREV(e, sysctl_ctx_list, link);
267 	else
268 		e1 = TAILQ_LAST(clist, sysctl_ctx_list);
269 	while (e1 != NULL) {
270 		sysctl_register_oid(e1->entry);
271 		e1 = TAILQ_PREV(e1, sysctl_ctx_list, link);
272 	}
273 	if (error) {
274 		SYSCTL_XUNLOCK();
275 		return(EBUSY);
276 	}
277 	/* Now really delete the entries */
278 	e = TAILQ_FIRST(clist);
279 	while (e != NULL) {
280 		e1 = TAILQ_NEXT(e, link);
281 		error = sysctl_remove_oid_locked(e->entry, 1, 0);
282 		if (error)
283 			panic("sysctl_remove_oid: corrupt tree, entry: %s",
284 			    e->entry->oid_name);
285 		free(e, M_SYSCTLOID);
286 		e = e1;
287 	}
288 	SYSCTL_XUNLOCK();
289 	return (error);
290 }
291 
292 /* Add an entry to the context */
293 struct sysctl_ctx_entry *
294 sysctl_ctx_entry_add(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
295 {
296 	struct sysctl_ctx_entry *e;
297 
298 	SYSCTL_ASSERT_XLOCKED();
299 	if (clist == NULL || oidp == NULL)
300 		return(NULL);
301 	e = malloc(sizeof(struct sysctl_ctx_entry), M_SYSCTLOID, M_WAITOK);
302 	e->entry = oidp;
303 	TAILQ_INSERT_HEAD(clist, e, link);
304 	return (e);
305 }
306 
307 /* Find an entry in the context */
308 struct sysctl_ctx_entry *
309 sysctl_ctx_entry_find(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
310 {
311 	struct sysctl_ctx_entry *e;
312 
313 	SYSCTL_ASSERT_LOCKED();
314 	if (clist == NULL || oidp == NULL)
315 		return(NULL);
316 	TAILQ_FOREACH(e, clist, link) {
317 		if(e->entry == oidp)
318 			return(e);
319 	}
320 	return (e);
321 }
322 
323 /*
324  * Delete an entry from the context.
325  * NOTE: this function doesn't free oidp! You have to remove it
326  * with sysctl_remove_oid().
327  */
328 int
329 sysctl_ctx_entry_del(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
330 {
331 	struct sysctl_ctx_entry *e;
332 
333 	if (clist == NULL || oidp == NULL)
334 		return (EINVAL);
335 	SYSCTL_XLOCK();
336 	e = sysctl_ctx_entry_find(clist, oidp);
337 	if (e != NULL) {
338 		TAILQ_REMOVE(clist, e, link);
339 		SYSCTL_XUNLOCK();
340 		free(e, M_SYSCTLOID);
341 		return (0);
342 	} else {
343 		SYSCTL_XUNLOCK();
344 		return (ENOENT);
345 	}
346 }
347 
348 /*
349  * Remove dynamically created sysctl trees.
350  * oidp - top of the tree to be removed
351  * del - if 0 - just deregister, otherwise free up entries as well
352  * recurse - if != 0 traverse the subtree to be deleted
353  */
354 int
355 sysctl_remove_oid(struct sysctl_oid *oidp, int del, int recurse)
356 {
357 	int error;
358 
359 	SYSCTL_XLOCK();
360 	error = sysctl_remove_oid_locked(oidp, del, recurse);
361 	SYSCTL_XUNLOCK();
362 	return (error);
363 }
364 
365 static int
366 sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del, int recurse)
367 {
368 	struct sysctl_oid *p;
369 	int error;
370 
371 	SYSCTL_ASSERT_XLOCKED();
372 	if (oidp == NULL)
373 		return(EINVAL);
374 	if ((oidp->oid_kind & CTLFLAG_DYN) == 0) {
375 		printf("can't remove non-dynamic nodes!\n");
376 		return (EINVAL);
377 	}
378 	/*
379 	 * WARNING: normal method to do this should be through
380 	 * sysctl_ctx_free(). Use recursing as the last resort
381 	 * method to purge your sysctl tree of leftovers...
382 	 * However, if some other code still references these nodes,
383 	 * it will panic.
384 	 */
385 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
386 		if (oidp->oid_refcnt == 1) {
387 			SLIST_FOREACH(p, SYSCTL_CHILDREN(oidp), oid_link) {
388 				if (!recurse)
389 					return (ENOTEMPTY);
390 				error = sysctl_remove_oid_locked(p, del,
391 				    recurse);
392 				if (error)
393 					return (error);
394 			}
395 			if (del)
396 				free(SYSCTL_CHILDREN(oidp), M_SYSCTLOID);
397 		}
398 	}
399 	if (oidp->oid_refcnt > 1 ) {
400 		oidp->oid_refcnt--;
401 	} else {
402 		if (oidp->oid_refcnt == 0) {
403 			printf("Warning: bad oid_refcnt=%u (%s)!\n",
404 				oidp->oid_refcnt, oidp->oid_name);
405 			return (EINVAL);
406 		}
407 		sysctl_unregister_oid(oidp);
408 		if (del) {
409 			if (oidp->oid_descr)
410 				free((void *)(uintptr_t)(const void *)oidp->oid_descr, M_SYSCTLOID);
411 			free((void *)(uintptr_t)(const void *)oidp->oid_name,
412 			     M_SYSCTLOID);
413 			free(oidp, M_SYSCTLOID);
414 		}
415 	}
416 	return (0);
417 }
418 
419 /*
420  * Create new sysctls at run time.
421  * clist may point to a valid context initialized with sysctl_ctx_init().
422  */
423 struct sysctl_oid *
424 sysctl_add_oid(struct sysctl_ctx_list *clist, struct sysctl_oid_list *parent,
425 	int number, const char *name, int kind, void *arg1, int arg2,
426 	int (*handler)(SYSCTL_HANDLER_ARGS), const char *fmt, const char *descr)
427 {
428 	struct sysctl_oid *oidp;
429 	ssize_t len;
430 	char *newname;
431 
432 	/* You have to hook up somewhere.. */
433 	if (parent == NULL)
434 		return(NULL);
435 	/* Check if the node already exists, otherwise create it */
436 	SYSCTL_XLOCK();
437 	oidp = sysctl_find_oidname(name, parent);
438 	if (oidp != NULL) {
439 		if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
440 			oidp->oid_refcnt++;
441 			/* Update the context */
442 			if (clist != NULL)
443 				sysctl_ctx_entry_add(clist, oidp);
444 			SYSCTL_XUNLOCK();
445 			return (oidp);
446 		} else {
447 			SYSCTL_XUNLOCK();
448 			printf("can't re-use a leaf (%s)!\n", name);
449 			return (NULL);
450 		}
451 	}
452 	oidp = malloc(sizeof(struct sysctl_oid), M_SYSCTLOID, M_WAITOK|M_ZERO);
453 	oidp->oid_parent = parent;
454 	SLIST_NEXT(oidp, oid_link) = NULL;
455 	oidp->oid_number = number;
456 	oidp->oid_refcnt = 1;
457 	len = strlen(name);
458 	newname = malloc(len + 1, M_SYSCTLOID, M_WAITOK);
459 	bcopy(name, newname, len + 1);
460 	newname[len] = '\0';
461 	oidp->oid_name = newname;
462 	oidp->oid_handler = handler;
463 	oidp->oid_kind = CTLFLAG_DYN | kind;
464 	if ((kind & CTLTYPE) == CTLTYPE_NODE) {
465 		/* Allocate space for children */
466 		SYSCTL_CHILDREN_SET(oidp, malloc(sizeof(struct sysctl_oid_list),
467 		    M_SYSCTLOID, M_WAITOK));
468 		SLIST_INIT(SYSCTL_CHILDREN(oidp));
469 	} else {
470 		oidp->oid_arg1 = arg1;
471 		oidp->oid_arg2 = arg2;
472 	}
473 	oidp->oid_fmt = fmt;
474 	if (descr) {
475 		int len = strlen(descr) + 1;
476 		oidp->oid_descr = malloc(len, M_SYSCTLOID, M_WAITOK);
477 		if (oidp->oid_descr)
478 			strcpy((char *)(uintptr_t)(const void *)oidp->oid_descr, descr);
479 	}
480 	/* Update the context, if used */
481 	if (clist != NULL)
482 		sysctl_ctx_entry_add(clist, oidp);
483 	/* Register this oid */
484 	sysctl_register_oid(oidp);
485 	SYSCTL_XUNLOCK();
486 	return (oidp);
487 }
488 
489 /*
490  * Rename an existing oid.
491  */
492 void
493 sysctl_rename_oid(struct sysctl_oid *oidp, const char *name)
494 {
495 	ssize_t len;
496 	char *newname;
497 	void *oldname;
498 
499 	len = strlen(name);
500 	newname = malloc(len + 1, M_SYSCTLOID, M_WAITOK);
501 	bcopy(name, newname, len + 1);
502 	newname[len] = '\0';
503 	SYSCTL_XLOCK();
504 	oldname = (void *)(uintptr_t)(const void *)oidp->oid_name;
505 	oidp->oid_name = newname;
506 	SYSCTL_XUNLOCK();
507 	free(oldname, M_SYSCTLOID);
508 }
509 
510 /*
511  * Reparent an existing oid.
512  */
513 int
514 sysctl_move_oid(struct sysctl_oid *oid, struct sysctl_oid_list *parent)
515 {
516 	struct sysctl_oid *oidp;
517 
518 	SYSCTL_XLOCK();
519 	if (oid->oid_parent == parent) {
520 		SYSCTL_XUNLOCK();
521 		return (0);
522 	}
523 	oidp = sysctl_find_oidname(oid->oid_name, parent);
524 	if (oidp != NULL) {
525 		SYSCTL_XUNLOCK();
526 		return (EEXIST);
527 	}
528 	sysctl_unregister_oid(oid);
529 	oid->oid_parent = parent;
530 	oid->oid_number = OID_AUTO;
531 	sysctl_register_oid(oid);
532 	SYSCTL_XUNLOCK();
533 	return (0);
534 }
535 
536 /*
537  * Register the kernel's oids on startup.
538  */
539 SET_DECLARE(sysctl_set, struct sysctl_oid);
540 
541 static void
542 sysctl_register_all(void *arg)
543 {
544 	struct sysctl_oid **oidp;
545 
546 	sx_init(&sysctlmemlock, "sysctl mem");
547 	SYSCTL_INIT();
548 	SYSCTL_XLOCK();
549 	SET_FOREACH(oidp, sysctl_set)
550 		sysctl_register_oid(*oidp);
551 	SYSCTL_XUNLOCK();
552 }
553 SYSINIT(sysctl, SI_SUB_KMEM, SI_ORDER_ANY, sysctl_register_all, 0);
554 
555 /*
556  * "Staff-functions"
557  *
558  * These functions implement a presently undocumented interface
559  * used by the sysctl program to walk the tree, and get the type
560  * so it can print the value.
561  * This interface is under work and consideration, and should probably
562  * be killed with a big axe by the first person who can find the time.
563  * (be aware though, that the proper interface isn't as obvious as it
564  * may seem, there are various conflicting requirements.
565  *
566  * {0,0}	printf the entire MIB-tree.
567  * {0,1,...}	return the name of the "..." OID.
568  * {0,2,...}	return the next OID.
569  * {0,3}	return the OID of the name in "new"
570  * {0,4,...}	return the kind & format info for the "..." OID.
571  * {0,5,...}	return the description the "..." OID.
572  */
573 
574 #ifdef SYSCTL_DEBUG
575 static void
576 sysctl_sysctl_debug_dump_node(struct sysctl_oid_list *l, int i)
577 {
578 	int k;
579 	struct sysctl_oid *oidp;
580 
581 	SYSCTL_ASSERT_LOCKED();
582 	SLIST_FOREACH(oidp, l, oid_link) {
583 
584 		for (k=0; k<i; k++)
585 			printf(" ");
586 
587 		printf("%d %s ", oidp->oid_number, oidp->oid_name);
588 
589 		printf("%c%c",
590 			oidp->oid_kind & CTLFLAG_RD ? 'R':' ',
591 			oidp->oid_kind & CTLFLAG_WR ? 'W':' ');
592 
593 		if (oidp->oid_handler)
594 			printf(" *Handler");
595 
596 		switch (oidp->oid_kind & CTLTYPE) {
597 			case CTLTYPE_NODE:
598 				printf(" Node\n");
599 				if (!oidp->oid_handler) {
600 					sysctl_sysctl_debug_dump_node(
601 						oidp->oid_arg1, i+2);
602 				}
603 				break;
604 			case CTLTYPE_INT:    printf(" Int\n"); break;
605 			case CTLTYPE_STRING: printf(" String\n"); break;
606 			case CTLTYPE_QUAD:   printf(" Quad\n"); break;
607 			case CTLTYPE_OPAQUE: printf(" Opaque/struct\n"); break;
608 			default:	     printf("\n");
609 		}
610 
611 	}
612 }
613 
614 static int
615 sysctl_sysctl_debug(SYSCTL_HANDLER_ARGS)
616 {
617 	int error;
618 
619 	error = priv_check(req->td, PRIV_SYSCTL_DEBUG);
620 	if (error)
621 		return (error);
622 	sysctl_sysctl_debug_dump_node(&sysctl__children, 0);
623 	return (ENOENT);
624 }
625 
626 SYSCTL_PROC(_sysctl, 0, debug, CTLTYPE_STRING|CTLFLAG_RD,
627 	0, 0, sysctl_sysctl_debug, "-", "");
628 #endif
629 
630 static int
631 sysctl_sysctl_name(SYSCTL_HANDLER_ARGS)
632 {
633 	int *name = (int *) arg1;
634 	u_int namelen = arg2;
635 	int error = 0;
636 	struct sysctl_oid *oid;
637 	struct sysctl_oid_list *lsp = &sysctl__children, *lsp2;
638 	char buf[10];
639 
640 	SYSCTL_ASSERT_LOCKED();
641 	while (namelen) {
642 		if (!lsp) {
643 			snprintf(buf,sizeof(buf),"%d",*name);
644 			if (req->oldidx)
645 				error = SYSCTL_OUT(req, ".", 1);
646 			if (!error)
647 				error = SYSCTL_OUT(req, buf, strlen(buf));
648 			if (error)
649 				return (error);
650 			namelen--;
651 			name++;
652 			continue;
653 		}
654 		lsp2 = 0;
655 		SLIST_FOREACH(oid, lsp, oid_link) {
656 			if (oid->oid_number != *name)
657 				continue;
658 
659 			if (req->oldidx)
660 				error = SYSCTL_OUT(req, ".", 1);
661 			if (!error)
662 				error = SYSCTL_OUT(req, oid->oid_name,
663 					strlen(oid->oid_name));
664 			if (error)
665 				return (error);
666 
667 			namelen--;
668 			name++;
669 
670 			if ((oid->oid_kind & CTLTYPE) != CTLTYPE_NODE)
671 				break;
672 
673 			if (oid->oid_handler)
674 				break;
675 
676 			lsp2 = (struct sysctl_oid_list *)oid->oid_arg1;
677 			break;
678 		}
679 		lsp = lsp2;
680 	}
681 	return (SYSCTL_OUT(req, "", 1));
682 }
683 
684 static SYSCTL_NODE(_sysctl, 1, name, CTLFLAG_RD, sysctl_sysctl_name, "");
685 
686 static int
687 sysctl_sysctl_next_ls(struct sysctl_oid_list *lsp, int *name, u_int namelen,
688 	int *next, int *len, int level, struct sysctl_oid **oidpp)
689 {
690 	struct sysctl_oid *oidp;
691 
692 	SYSCTL_ASSERT_LOCKED();
693 	*len = level;
694 	SLIST_FOREACH(oidp, lsp, oid_link) {
695 		*next = oidp->oid_number;
696 		*oidpp = oidp;
697 
698 		if (oidp->oid_kind & CTLFLAG_SKIP)
699 			continue;
700 
701 		if (!namelen) {
702 			if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
703 				return (0);
704 			if (oidp->oid_handler)
705 				/* We really should call the handler here...*/
706 				return (0);
707 			lsp = (struct sysctl_oid_list *)oidp->oid_arg1;
708 			if (!sysctl_sysctl_next_ls(lsp, 0, 0, next+1,
709 				len, level+1, oidpp))
710 				return (0);
711 			goto emptynode;
712 		}
713 
714 		if (oidp->oid_number < *name)
715 			continue;
716 
717 		if (oidp->oid_number > *name) {
718 			if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
719 				return (0);
720 			if (oidp->oid_handler)
721 				return (0);
722 			lsp = (struct sysctl_oid_list *)oidp->oid_arg1;
723 			if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1,
724 				next+1, len, level+1, oidpp))
725 				return (0);
726 			goto next;
727 		}
728 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
729 			continue;
730 
731 		if (oidp->oid_handler)
732 			continue;
733 
734 		lsp = (struct sysctl_oid_list *)oidp->oid_arg1;
735 		if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1, next+1,
736 			len, level+1, oidpp))
737 			return (0);
738 	next:
739 		namelen = 1;
740 	emptynode:
741 		*len = level;
742 	}
743 	return (1);
744 }
745 
746 static int
747 sysctl_sysctl_next(SYSCTL_HANDLER_ARGS)
748 {
749 	int *name = (int *) arg1;
750 	u_int namelen = arg2;
751 	int i, j, error;
752 	struct sysctl_oid *oid;
753 	struct sysctl_oid_list *lsp = &sysctl__children;
754 	int newoid[CTL_MAXNAME];
755 
756 	i = sysctl_sysctl_next_ls(lsp, name, namelen, newoid, &j, 1, &oid);
757 	if (i)
758 		return (ENOENT);
759 	error = SYSCTL_OUT(req, newoid, j * sizeof (int));
760 	return (error);
761 }
762 
763 static SYSCTL_NODE(_sysctl, 2, next, CTLFLAG_RD, sysctl_sysctl_next, "");
764 
765 static int
766 name2oid(char *name, int *oid, int *len, struct sysctl_oid **oidpp)
767 {
768 	int i;
769 	struct sysctl_oid *oidp;
770 	struct sysctl_oid_list *lsp = &sysctl__children;
771 	char *p;
772 
773 	SYSCTL_ASSERT_LOCKED();
774 
775 	if (!*name)
776 		return (ENOENT);
777 
778 	p = name + strlen(name) - 1 ;
779 	if (*p == '.')
780 		*p = '\0';
781 
782 	*len = 0;
783 
784 	for (p = name; *p && *p != '.'; p++)
785 		;
786 	i = *p;
787 	if (i == '.')
788 		*p = '\0';
789 
790 	oidp = SLIST_FIRST(lsp);
791 
792 	while (oidp && *len < CTL_MAXNAME) {
793 		if (strcmp(name, oidp->oid_name)) {
794 			oidp = SLIST_NEXT(oidp, oid_link);
795 			continue;
796 		}
797 		*oid++ = oidp->oid_number;
798 		(*len)++;
799 
800 		if (!i) {
801 			if (oidpp)
802 				*oidpp = oidp;
803 			return (0);
804 		}
805 
806 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
807 			break;
808 
809 		if (oidp->oid_handler)
810 			break;
811 
812 		lsp = (struct sysctl_oid_list *)oidp->oid_arg1;
813 		oidp = SLIST_FIRST(lsp);
814 		name = p+1;
815 		for (p = name; *p && *p != '.'; p++)
816 				;
817 		i = *p;
818 		if (i == '.')
819 			*p = '\0';
820 	}
821 	return (ENOENT);
822 }
823 
824 static int
825 sysctl_sysctl_name2oid(SYSCTL_HANDLER_ARGS)
826 {
827 	char *p;
828 	int error, oid[CTL_MAXNAME], len;
829 	struct sysctl_oid *op = 0;
830 
831 	SYSCTL_ASSERT_LOCKED();
832 
833 	if (!req->newlen)
834 		return (ENOENT);
835 	if (req->newlen >= MAXPATHLEN)	/* XXX arbitrary, undocumented */
836 		return (ENAMETOOLONG);
837 
838 	p = malloc(req->newlen+1, M_SYSCTL, M_WAITOK);
839 
840 	error = SYSCTL_IN(req, p, req->newlen);
841 	if (error) {
842 		free(p, M_SYSCTL);
843 		return (error);
844 	}
845 
846 	p [req->newlen] = '\0';
847 
848 	error = name2oid(p, oid, &len, &op);
849 
850 	free(p, M_SYSCTL);
851 
852 	if (error)
853 		return (error);
854 
855 	error = SYSCTL_OUT(req, oid, len * sizeof *oid);
856 	return (error);
857 }
858 
859 SYSCTL_PROC(_sysctl, 3, name2oid, CTLFLAG_RW|CTLFLAG_ANYBODY|CTLFLAG_MPSAFE,
860     0, 0, sysctl_sysctl_name2oid, "I", "");
861 
862 static int
863 sysctl_sysctl_oidfmt(SYSCTL_HANDLER_ARGS)
864 {
865 	struct sysctl_oid *oid;
866 	int error;
867 
868 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
869 	if (error)
870 		return (error);
871 
872 	if (!oid->oid_fmt)
873 		return (ENOENT);
874 	error = SYSCTL_OUT(req, &oid->oid_kind, sizeof(oid->oid_kind));
875 	if (error)
876 		return (error);
877 	error = SYSCTL_OUT(req, oid->oid_fmt, strlen(oid->oid_fmt) + 1);
878 	return (error);
879 }
880 
881 
882 static SYSCTL_NODE(_sysctl, 4, oidfmt, CTLFLAG_RD|CTLFLAG_MPSAFE,
883     sysctl_sysctl_oidfmt, "");
884 
885 static int
886 sysctl_sysctl_oiddescr(SYSCTL_HANDLER_ARGS)
887 {
888 	struct sysctl_oid *oid;
889 	int error;
890 
891 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
892 	if (error)
893 		return (error);
894 
895 	if (!oid->oid_descr)
896 		return (ENOENT);
897 	error = SYSCTL_OUT(req, oid->oid_descr, strlen(oid->oid_descr) + 1);
898 	return (error);
899 }
900 
901 static SYSCTL_NODE(_sysctl, 5, oiddescr, CTLFLAG_RD, sysctl_sysctl_oiddescr, "");
902 
903 /*
904  * Default "handler" functions.
905  */
906 
907 /*
908  * Handle an int, signed or unsigned.
909  * Two cases:
910  *     a variable:  point arg1 at it.
911  *     a constant:  pass it in arg2.
912  */
913 
914 int
915 sysctl_handle_int(SYSCTL_HANDLER_ARGS)
916 {
917 	int tmpout, error = 0;
918 
919 	/*
920 	 * Attempt to get a coherent snapshot by making a copy of the data.
921 	 */
922 	if (arg1)
923 		tmpout = *(int *)arg1;
924 	else
925 		tmpout = arg2;
926 	error = SYSCTL_OUT(req, &tmpout, sizeof(int));
927 
928 	if (error || !req->newptr)
929 		return (error);
930 
931 	if (!arg1)
932 		error = EPERM;
933 	else
934 		error = SYSCTL_IN(req, arg1, sizeof(int));
935 	return (error);
936 }
937 
938 #ifdef VIMAGE
939 int
940 sysctl_handle_v_int(SYSCTL_HANDLER_ARGS)
941 {
942 	int tmpout, error = 0;
943 
944 	SYSCTL_RESOLVE_V_ARG1();
945 
946 	/*
947 	 * Attempt to get a coherent snapshot by making a copy of the data.
948 	 */
949 	tmpout = *(int *)arg1;
950 	error = SYSCTL_OUT(req, &tmpout, sizeof(int));
951 
952 	if (error || !req->newptr)
953 		return (error);
954 
955 	if (!arg1)
956 		error = EPERM;
957 	else
958 		error = SYSCTL_IN(req, arg1, sizeof(int));
959 	return (error);
960 }
961 #endif
962 
963 /*
964  * Based on on sysctl_handle_int() convert milliseconds into ticks.
965  */
966 
967 int
968 sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
969 {
970 	int error, s, tt;
971 
972 	SYSCTL_RESOLVE_V_ARG1();
973 
974 	tt = *(int *)arg1;
975 	s = (int)((int64_t)tt * 1000 / hz);
976 
977 	error = sysctl_handle_int(oidp, &s, 0, req);
978 	if (error || !req->newptr)
979 		return (error);
980 
981 	tt = (int)((int64_t)s * hz / 1000);
982 	if (tt < 1)
983 		return (EINVAL);
984 
985 	*(int *)arg1 = tt;
986 	return (0);
987 }
988 
989 
990 /*
991  * Handle a long, signed or unsigned.  arg1 points to it.
992  */
993 
994 int
995 sysctl_handle_long(SYSCTL_HANDLER_ARGS)
996 {
997 	int error = 0;
998 	long tmplong;
999 #ifdef SCTL_MASK32
1000 	int tmpint;
1001 #endif
1002 
1003 	/*
1004 	 * Attempt to get a coherent snapshot by making a copy of the data.
1005 	 */
1006 	if (!arg1)
1007 		return (EINVAL);
1008 	tmplong = *(long *)arg1;
1009 #ifdef SCTL_MASK32
1010 	if (req->flags & SCTL_MASK32) {
1011 		tmpint = tmplong;
1012 		error = SYSCTL_OUT(req, &tmpint, sizeof(int));
1013 	} else
1014 #endif
1015 		error = SYSCTL_OUT(req, &tmplong, sizeof(long));
1016 
1017 	if (error || !req->newptr)
1018 		return (error);
1019 
1020 #ifdef SCTL_MASK32
1021 	if (req->flags & SCTL_MASK32) {
1022 		error = SYSCTL_IN(req, &tmpint, sizeof(int));
1023 		*(long *)arg1 = (long)tmpint;
1024 	} else
1025 #endif
1026 		error = SYSCTL_IN(req, arg1, sizeof(long));
1027 	return (error);
1028 }
1029 
1030 /*
1031  * Handle a 64 bit int, signed or unsigned.  arg1 points to it.
1032  */
1033 
1034 int
1035 sysctl_handle_quad(SYSCTL_HANDLER_ARGS)
1036 {
1037 	int error = 0;
1038 	uint64_t tmpout;
1039 
1040 	/*
1041 	 * Attempt to get a coherent snapshot by making a copy of the data.
1042 	 */
1043 	if (!arg1)
1044 		return (EINVAL);
1045 	tmpout = *(uint64_t *)arg1;
1046 	error = SYSCTL_OUT(req, &tmpout, sizeof(uint64_t));
1047 
1048 	if (error || !req->newptr)
1049 		return (error);
1050 
1051 	error = SYSCTL_IN(req, arg1, sizeof(uint64_t));
1052 	return (error);
1053 }
1054 
1055 /*
1056  * Handle our generic '\0' terminated 'C' string.
1057  * Two cases:
1058  * 	a variable string:  point arg1 at it, arg2 is max length.
1059  * 	a constant string:  point arg1 at it, arg2 is zero.
1060  */
1061 
1062 int
1063 sysctl_handle_string(SYSCTL_HANDLER_ARGS)
1064 {
1065 	int error=0;
1066 	char *tmparg;
1067 	size_t outlen;
1068 
1069 	/*
1070 	 * Attempt to get a coherent snapshot by copying to a
1071 	 * temporary kernel buffer.
1072 	 */
1073 retry:
1074 	outlen = strlen((char *)arg1)+1;
1075 	tmparg = malloc(outlen, M_SYSCTLTMP, M_WAITOK);
1076 
1077 	if (strlcpy(tmparg, (char *)arg1, outlen) >= outlen) {
1078 		free(tmparg, M_SYSCTLTMP);
1079 		goto retry;
1080 	}
1081 
1082 	error = SYSCTL_OUT(req, tmparg, outlen);
1083 	free(tmparg, M_SYSCTLTMP);
1084 
1085 	if (error || !req->newptr)
1086 		return (error);
1087 
1088 	if ((req->newlen - req->newidx) >= arg2) {
1089 		error = EINVAL;
1090 	} else {
1091 		arg2 = (req->newlen - req->newidx);
1092 		error = SYSCTL_IN(req, arg1, arg2);
1093 		((char *)arg1)[arg2] = '\0';
1094 	}
1095 
1096 	return (error);
1097 }
1098 
1099 #ifdef VIMAGE
1100 int
1101 sysctl_handle_v_string(SYSCTL_HANDLER_ARGS)
1102 {
1103 	int error=0;
1104 	char *tmparg;
1105 	size_t outlen;
1106 
1107 	SYSCTL_RESOLVE_V_ARG1();
1108 
1109 	/*
1110 	 * Attempt to get a coherent snapshot by copying to a
1111 	 * temporary kernel buffer.
1112 	 */
1113 retry:
1114 	outlen = strlen((char *)arg1)+1;
1115 	tmparg = malloc(outlen, M_SYSCTLTMP, M_WAITOK);
1116 
1117 	if (strlcpy(tmparg, (char *)arg1, outlen) >= outlen) {
1118 		free(tmparg, M_SYSCTLTMP);
1119 		goto retry;
1120 	}
1121 
1122 	error = SYSCTL_OUT(req, tmparg, outlen);
1123 	free(tmparg, M_SYSCTLTMP);
1124 
1125 	if (error || !req->newptr)
1126 		return (error);
1127 
1128 	if ((req->newlen - req->newidx) >= arg2) {
1129 		error = EINVAL;
1130 	} else {
1131 		arg2 = (req->newlen - req->newidx);
1132 		error = SYSCTL_IN(req, arg1, arg2);
1133 		((char *)arg1)[arg2] = '\0';
1134 	}
1135 
1136 	return (error);
1137 }
1138 #endif
1139 
1140 /*
1141  * Handle any kind of opaque data.
1142  * arg1 points to it, arg2 is the size.
1143  */
1144 
1145 int
1146 sysctl_handle_opaque(SYSCTL_HANDLER_ARGS)
1147 {
1148 	int error, tries;
1149 	u_int generation;
1150 	struct sysctl_req req2;
1151 
1152 	/*
1153 	 * Attempt to get a coherent snapshot, by using the thread
1154 	 * pre-emption counter updated from within mi_switch() to
1155 	 * determine if we were pre-empted during a bcopy() or
1156 	 * copyout(). Make 3 attempts at doing this before giving up.
1157 	 * If we encounter an error, stop immediately.
1158 	 */
1159 	tries = 0;
1160 	req2 = *req;
1161 retry:
1162 	generation = curthread->td_generation;
1163 	error = SYSCTL_OUT(req, arg1, arg2);
1164 	if (error)
1165 		return (error);
1166 	tries++;
1167 	if (generation != curthread->td_generation && tries < 3) {
1168 		*req = req2;
1169 		goto retry;
1170 	}
1171 
1172 	error = SYSCTL_IN(req, arg1, arg2);
1173 
1174 	return (error);
1175 }
1176 
1177 #ifdef VIMAGE
1178 int
1179 sysctl_handle_v_opaque(SYSCTL_HANDLER_ARGS)
1180 {
1181 	int error, tries;
1182 	u_int generation;
1183 	struct sysctl_req req2;
1184 
1185 	SYSCTL_RESOLVE_V_ARG1();
1186 
1187 	tries = 0;
1188 	req2 = *req;
1189 retry:
1190 	generation = curthread->td_generation;
1191 	error = SYSCTL_OUT(req, arg1, arg2);
1192 	if (error)
1193 		return (error);
1194 	tries++;
1195 	if (generation != curthread->td_generation && tries < 3) {
1196 		*req = req2;
1197 		goto retry;
1198 	}
1199 
1200 	error = SYSCTL_IN(req, arg1, arg2);
1201 
1202 	return (error);
1203 }
1204 #endif
1205 
1206 /*
1207  * Transfer functions to/from kernel space.
1208  * XXX: rather untested at this point
1209  */
1210 static int
1211 sysctl_old_kernel(struct sysctl_req *req, const void *p, size_t l)
1212 {
1213 	size_t i = 0;
1214 
1215 	if (req->oldptr) {
1216 		i = l;
1217 		if (req->oldlen <= req->oldidx)
1218 			i = 0;
1219 		else
1220 			if (i > req->oldlen - req->oldidx)
1221 				i = req->oldlen - req->oldidx;
1222 		if (i > 0)
1223 			bcopy(p, (char *)req->oldptr + req->oldidx, i);
1224 	}
1225 	req->oldidx += l;
1226 	if (req->oldptr && i != l)
1227 		return (ENOMEM);
1228 	return (0);
1229 }
1230 
1231 static int
1232 sysctl_new_kernel(struct sysctl_req *req, void *p, size_t l)
1233 {
1234 	if (!req->newptr)
1235 		return (0);
1236 	if (req->newlen - req->newidx < l)
1237 		return (EINVAL);
1238 	bcopy((char *)req->newptr + req->newidx, p, l);
1239 	req->newidx += l;
1240 	return (0);
1241 }
1242 
1243 int
1244 kernel_sysctl(struct thread *td, int *name, u_int namelen, void *old,
1245     size_t *oldlenp, void *new, size_t newlen, size_t *retval, int flags)
1246 {
1247 	int error = 0;
1248 	struct sysctl_req req;
1249 
1250 	bzero(&req, sizeof req);
1251 
1252 	req.td = td;
1253 	req.flags = flags;
1254 
1255 	if (oldlenp) {
1256 		req.oldlen = *oldlenp;
1257 	}
1258 	req.validlen = req.oldlen;
1259 
1260 	if (old) {
1261 		req.oldptr= old;
1262 	}
1263 
1264 	if (new != NULL) {
1265 		req.newlen = newlen;
1266 		req.newptr = new;
1267 	}
1268 
1269 	req.oldfunc = sysctl_old_kernel;
1270 	req.newfunc = sysctl_new_kernel;
1271 	req.lock = REQ_LOCKED;
1272 
1273 	SYSCTL_SLOCK();
1274 	error = sysctl_root(0, name, namelen, &req);
1275 	SYSCTL_SUNLOCK();
1276 
1277 	if (req.lock == REQ_WIRED && req.validlen > 0)
1278 		vsunlock(req.oldptr, req.validlen);
1279 
1280 	if (error && error != ENOMEM)
1281 		return (error);
1282 
1283 	if (retval) {
1284 		if (req.oldptr && req.oldidx > req.validlen)
1285 			*retval = req.validlen;
1286 		else
1287 			*retval = req.oldidx;
1288 	}
1289 	return (error);
1290 }
1291 
1292 int
1293 kernel_sysctlbyname(struct thread *td, char *name, void *old, size_t *oldlenp,
1294     void *new, size_t newlen, size_t *retval, int flags)
1295 {
1296         int oid[CTL_MAXNAME];
1297         size_t oidlen, plen;
1298 	int error;
1299 
1300 	oid[0] = 0;		/* sysctl internal magic */
1301 	oid[1] = 3;		/* name2oid */
1302 	oidlen = sizeof(oid);
1303 
1304 	error = kernel_sysctl(td, oid, 2, oid, &oidlen,
1305 	    (void *)name, strlen(name), &plen, flags);
1306 	if (error)
1307 		return (error);
1308 
1309 	error = kernel_sysctl(td, oid, plen / sizeof(int), old, oldlenp,
1310 	    new, newlen, retval, flags);
1311 	return (error);
1312 }
1313 
1314 /*
1315  * Transfer function to/from user space.
1316  */
1317 static int
1318 sysctl_old_user(struct sysctl_req *req, const void *p, size_t l)
1319 {
1320 	int error = 0;
1321 	size_t i, len, origidx;
1322 
1323 	origidx = req->oldidx;
1324 	req->oldidx += l;
1325 	if (req->oldptr == NULL)
1326 		return (0);
1327 	/*
1328 	 * If we have not wired the user supplied buffer and we are currently
1329 	 * holding locks, drop a witness warning, as it's possible that
1330 	 * write operations to the user page can sleep.
1331 	 */
1332 	if (req->lock != REQ_WIRED)
1333 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
1334 		    "sysctl_old_user()");
1335 	i = l;
1336 	len = req->validlen;
1337 	if (len <= origidx)
1338 		i = 0;
1339 	else {
1340 		if (i > len - origidx)
1341 			i = len - origidx;
1342 		error = copyout(p, (char *)req->oldptr + origidx, i);
1343 	}
1344 	if (error)
1345 		return (error);
1346 	if (i < l)
1347 		return (ENOMEM);
1348 	return (0);
1349 }
1350 
1351 static int
1352 sysctl_new_user(struct sysctl_req *req, void *p, size_t l)
1353 {
1354 	int error;
1355 
1356 	if (!req->newptr)
1357 		return (0);
1358 	if (req->newlen - req->newidx < l)
1359 		return (EINVAL);
1360 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
1361 	    "sysctl_new_user()");
1362 	error = copyin((char *)req->newptr + req->newidx, p, l);
1363 	req->newidx += l;
1364 	return (error);
1365 }
1366 
1367 /*
1368  * Wire the user space destination buffer.  If set to a value greater than
1369  * zero, the len parameter limits the maximum amount of wired memory.
1370  */
1371 int
1372 sysctl_wire_old_buffer(struct sysctl_req *req, size_t len)
1373 {
1374 	int ret;
1375 	size_t wiredlen;
1376 
1377 	wiredlen = (len > 0 && len < req->oldlen) ? len : req->oldlen;
1378 	ret = 0;
1379 	if (req->lock == REQ_LOCKED && req->oldptr &&
1380 	    req->oldfunc == sysctl_old_user) {
1381 		if (wiredlen != 0) {
1382 			ret = vslock(req->oldptr, wiredlen);
1383 			if (ret != 0) {
1384 				if (ret != ENOMEM)
1385 					return (ret);
1386 				wiredlen = 0;
1387 			}
1388 		}
1389 		req->lock = REQ_WIRED;
1390 		req->validlen = wiredlen;
1391 	}
1392 	return (0);
1393 }
1394 
1395 int
1396 sysctl_find_oid(int *name, u_int namelen, struct sysctl_oid **noid,
1397     int *nindx, struct sysctl_req *req)
1398 {
1399 	struct sysctl_oid *oid;
1400 	int indx;
1401 
1402 	SYSCTL_ASSERT_LOCKED();
1403 	oid = SLIST_FIRST(&sysctl__children);
1404 	indx = 0;
1405 	while (oid && indx < CTL_MAXNAME) {
1406 		if (oid->oid_number == name[indx]) {
1407 			indx++;
1408 			if (oid->oid_kind & CTLFLAG_NOLOCK)
1409 				req->lock = REQ_UNLOCKED;
1410 			if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1411 				if (oid->oid_handler != NULL ||
1412 				    indx == namelen) {
1413 					*noid = oid;
1414 					if (nindx != NULL)
1415 						*nindx = indx;
1416 					return (0);
1417 				}
1418 				oid = SLIST_FIRST(
1419 				    (struct sysctl_oid_list *)oid->oid_arg1);
1420 			} else if (indx == namelen) {
1421 				*noid = oid;
1422 				if (nindx != NULL)
1423 					*nindx = indx;
1424 				return (0);
1425 			} else {
1426 				return (ENOTDIR);
1427 			}
1428 		} else {
1429 			oid = SLIST_NEXT(oid, oid_link);
1430 		}
1431 	}
1432 	return (ENOENT);
1433 }
1434 
1435 /*
1436  * Traverse our tree, and find the right node, execute whatever it points
1437  * to, and return the resulting error code.
1438  */
1439 
1440 static int
1441 sysctl_root(SYSCTL_HANDLER_ARGS)
1442 {
1443 	struct sysctl_oid *oid;
1444 	int error, indx, lvl;
1445 
1446 	SYSCTL_ASSERT_LOCKED();
1447 
1448 	error = sysctl_find_oid(arg1, arg2, &oid, &indx, req);
1449 	if (error)
1450 		return (error);
1451 
1452 	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1453 		/*
1454 		 * You can't call a sysctl when it's a node, but has
1455 		 * no handler.  Inform the user that it's a node.
1456 		 * The indx may or may not be the same as namelen.
1457 		 */
1458 		if (oid->oid_handler == NULL)
1459 			return (EISDIR);
1460 	}
1461 
1462 	/* Is this sysctl writable? */
1463 	if (req->newptr && !(oid->oid_kind & CTLFLAG_WR))
1464 		return (EPERM);
1465 
1466 	KASSERT(req->td != NULL, ("sysctl_root(): req->td == NULL"));
1467 
1468 	/* Is this sysctl sensitive to securelevels? */
1469 	if (req->newptr && (oid->oid_kind & CTLFLAG_SECURE)) {
1470 		lvl = (oid->oid_kind & CTLMASK_SECURE) >> CTLSHIFT_SECURE;
1471 		error = securelevel_gt(req->td->td_ucred, lvl);
1472 		if (error)
1473 			return (error);
1474 	}
1475 
1476 	/* Is this sysctl writable by only privileged users? */
1477 	if (req->newptr && !(oid->oid_kind & CTLFLAG_ANYBODY)) {
1478 		if (oid->oid_kind & CTLFLAG_PRISON)
1479 			error = priv_check(req->td, PRIV_SYSCTL_WRITEJAIL);
1480 		else
1481 			error = priv_check(req->td, PRIV_SYSCTL_WRITE);
1482 		if (error)
1483 			return (error);
1484 	}
1485 
1486 	if (!oid->oid_handler)
1487 		return (EINVAL);
1488 
1489 	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1490 		arg1 = (int *)arg1 + indx;
1491 		arg2 -= indx;
1492 	} else {
1493 		arg1 = oid->oid_arg1;
1494 		arg2 = oid->oid_arg2;
1495 	}
1496 #ifdef MAC
1497 	error = mac_system_check_sysctl(req->td->td_ucred, oid, arg1, arg2,
1498 	    req);
1499 	if (error != 0)
1500 		return (error);
1501 #endif
1502 	if (!(oid->oid_kind & CTLFLAG_MPSAFE))
1503 		mtx_lock(&Giant);
1504 	error = oid->oid_handler(oid, arg1, arg2, req);
1505 	if (!(oid->oid_kind & CTLFLAG_MPSAFE))
1506 		mtx_unlock(&Giant);
1507 
1508 	return (error);
1509 }
1510 
1511 #ifndef _SYS_SYSPROTO_H_
1512 struct sysctl_args {
1513 	int	*name;
1514 	u_int	namelen;
1515 	void	*old;
1516 	size_t	*oldlenp;
1517 	void	*new;
1518 	size_t	newlen;
1519 };
1520 #endif
1521 int
1522 __sysctl(struct thread *td, struct sysctl_args *uap)
1523 {
1524 	int error, i, name[CTL_MAXNAME];
1525 	size_t j;
1526 
1527 	if (uap->namelen > CTL_MAXNAME || uap->namelen < 2)
1528 		return (EINVAL);
1529 
1530  	error = copyin(uap->name, &name, uap->namelen * sizeof(int));
1531  	if (error)
1532 		return (error);
1533 
1534 	error = userland_sysctl(td, name, uap->namelen,
1535 		uap->old, uap->oldlenp, 0,
1536 		uap->new, uap->newlen, &j, 0);
1537 	if (error && error != ENOMEM)
1538 		return (error);
1539 	if (uap->oldlenp) {
1540 		i = copyout(&j, uap->oldlenp, sizeof(j));
1541 		if (i)
1542 			return (i);
1543 	}
1544 	return (error);
1545 }
1546 
1547 /*
1548  * This is used from various compatibility syscalls too.  That's why name
1549  * must be in kernel space.
1550  */
1551 int
1552 userland_sysctl(struct thread *td, int *name, u_int namelen, void *old,
1553     size_t *oldlenp, int inkernel, void *new, size_t newlen, size_t *retval,
1554     int flags)
1555 {
1556 	int error = 0, memlocked;
1557 	struct sysctl_req req;
1558 
1559 	bzero(&req, sizeof req);
1560 
1561 	req.td = td;
1562 	req.flags = flags;
1563 
1564 	if (oldlenp) {
1565 		if (inkernel) {
1566 			req.oldlen = *oldlenp;
1567 		} else {
1568 			error = copyin(oldlenp, &req.oldlen, sizeof(*oldlenp));
1569 			if (error)
1570 				return (error);
1571 		}
1572 	}
1573 	req.validlen = req.oldlen;
1574 
1575 	if (old) {
1576 		if (!useracc(old, req.oldlen, VM_PROT_WRITE))
1577 			return (EFAULT);
1578 		req.oldptr= old;
1579 	}
1580 
1581 	if (new != NULL) {
1582 		if (!useracc(new, newlen, VM_PROT_READ))
1583 			return (EFAULT);
1584 		req.newlen = newlen;
1585 		req.newptr = new;
1586 	}
1587 
1588 	req.oldfunc = sysctl_old_user;
1589 	req.newfunc = sysctl_new_user;
1590 	req.lock = REQ_LOCKED;
1591 
1592 #ifdef KTRACE
1593 	if (KTRPOINT(curthread, KTR_SYSCTL))
1594 		ktrsysctl(name, namelen);
1595 #endif
1596 
1597 	if (req.oldlen > PAGE_SIZE) {
1598 		memlocked = 1;
1599 		sx_xlock(&sysctlmemlock);
1600 	} else
1601 		memlocked = 0;
1602 	CURVNET_SET(TD_TO_VNET(curthread));
1603 
1604 	for (;;) {
1605 		req.oldidx = 0;
1606 		req.newidx = 0;
1607 		SYSCTL_SLOCK();
1608 		error = sysctl_root(0, name, namelen, &req);
1609 		SYSCTL_SUNLOCK();
1610 		if (error != EAGAIN)
1611 			break;
1612 		uio_yield();
1613 	}
1614 
1615 	CURVNET_RESTORE();
1616 
1617 	if (req.lock == REQ_WIRED && req.validlen > 0)
1618 		vsunlock(req.oldptr, req.validlen);
1619 	if (memlocked)
1620 		sx_xunlock(&sysctlmemlock);
1621 
1622 	if (error && error != ENOMEM)
1623 		return (error);
1624 
1625 	if (retval) {
1626 		if (req.oldptr && req.oldidx > req.validlen)
1627 			*retval = req.validlen;
1628 		else
1629 			*retval = req.oldidx;
1630 	}
1631 	return (error);
1632 }
1633