xref: /freebsd/sys/kern/kern_sysctl.c (revision e14ddd1f16e7e5788392c50de21ea7c927e0690c)
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_capsicum.h"
42 #include "opt_compat.h"
43 #include "opt_ktrace.h"
44 
45 #include <sys/param.h>
46 #include <sys/fail.h>
47 #include <sys/systm.h>
48 #include <sys/capability.h>
49 #include <sys/kernel.h>
50 #include <sys/sysctl.h>
51 #include <sys/malloc.h>
52 #include <sys/priv.h>
53 #include <sys/proc.h>
54 #include <sys/jail.h>
55 #include <sys/lock.h>
56 #include <sys/mutex.h>
57 #include <sys/sbuf.h>
58 #include <sys/sx.h>
59 #include <sys/sysproto.h>
60 #include <sys/uio.h>
61 #ifdef KTRACE
62 #include <sys/ktrace.h>
63 #endif
64 
65 #include <net/vnet.h>
66 
67 #include <security/mac/mac_framework.h>
68 
69 #include <vm/vm.h>
70 #include <vm/vm_extern.h>
71 
72 static MALLOC_DEFINE(M_SYSCTL, "sysctl", "sysctl internal magic");
73 static MALLOC_DEFINE(M_SYSCTLOID, "sysctloid", "sysctl dynamic oids");
74 static MALLOC_DEFINE(M_SYSCTLTMP, "sysctltmp", "sysctl temp output buffer");
75 
76 /*
77  * The sysctllock protects the MIB tree.  It also protects sysctl
78  * contexts used with dynamic sysctls.  The sysctl_register_oid() and
79  * sysctl_unregister_oid() routines require the sysctllock to already
80  * be held, so the sysctl_lock() and sysctl_unlock() routines are
81  * provided for the few places in the kernel which need to use that
82  * API rather than using the dynamic API.  Use of the dynamic API is
83  * strongly encouraged for most code.
84  *
85  * The sysctlmemlock is used to limit the amount of user memory wired for
86  * sysctl requests.  This is implemented by serializing any userland
87  * sysctl requests larger than a single page via an exclusive lock.
88  */
89 static struct sx sysctllock;
90 static struct sx sysctlmemlock;
91 
92 #define	SYSCTL_XLOCK()		sx_xlock(&sysctllock)
93 #define	SYSCTL_XUNLOCK()	sx_xunlock(&sysctllock)
94 #define	SYSCTL_ASSERT_XLOCKED()	sx_assert(&sysctllock, SA_XLOCKED)
95 #define	SYSCTL_INIT()		sx_init(&sysctllock, "sysctl lock")
96 #define	SYSCTL_SLEEP(ch, wmesg, timo)					\
97 				sx_sleep(ch, &sysctllock, 0, wmesg, timo)
98 
99 static int sysctl_root(SYSCTL_HANDLER_ARGS);
100 
101 struct sysctl_oid_list sysctl__children; /* root list */
102 
103 static int	sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del,
104 		    int recurse);
105 
106 static struct sysctl_oid *
107 sysctl_find_oidname(const char *name, struct sysctl_oid_list *list)
108 {
109 	struct sysctl_oid *oidp;
110 
111 	SYSCTL_ASSERT_XLOCKED();
112 	SLIST_FOREACH(oidp, list, oid_link) {
113 		if (strcmp(oidp->oid_name, name) == 0) {
114 			return (oidp);
115 		}
116 	}
117 	return (NULL);
118 }
119 
120 /*
121  * Initialization of the MIB tree.
122  *
123  * Order by number in each list.
124  */
125 void
126 sysctl_lock(void)
127 {
128 
129 	SYSCTL_XLOCK();
130 }
131 
132 void
133 sysctl_unlock(void)
134 {
135 
136 	SYSCTL_XUNLOCK();
137 }
138 
139 void
140 sysctl_register_oid(struct sysctl_oid *oidp)
141 {
142 	struct sysctl_oid_list *parent = oidp->oid_parent;
143 	struct sysctl_oid *p;
144 	struct sysctl_oid *q;
145 
146 	/*
147 	 * First check if another oid with the same name already
148 	 * exists in the parent's list.
149 	 */
150 	SYSCTL_ASSERT_XLOCKED();
151 	p = sysctl_find_oidname(oidp->oid_name, parent);
152 	if (p != NULL) {
153 		if ((p->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
154 			p->oid_refcnt++;
155 			return;
156 		} else {
157 			printf("can't re-use a leaf (%s)!\n", p->oid_name);
158 			return;
159 		}
160 	}
161 	/*
162 	 * If this oid has a number OID_AUTO, give it a number which
163 	 * is greater than any current oid.
164 	 * NOTE: DO NOT change the starting value here, change it in
165 	 * <sys/sysctl.h>, and make sure it is at least 256 to
166 	 * accomodate e.g. net.inet.raw as a static sysctl node.
167 	 */
168 	if (oidp->oid_number == OID_AUTO) {
169 		static int newoid = CTL_AUTO_START;
170 
171 		oidp->oid_number = newoid++;
172 		if (newoid == 0x7fffffff)
173 			panic("out of oids");
174 	}
175 #if 0
176 	else if (oidp->oid_number >= CTL_AUTO_START) {
177 		/* do not panic; this happens when unregistering sysctl sets */
178 		printf("static sysctl oid too high: %d", oidp->oid_number);
179 	}
180 #endif
181 
182 	/*
183 	 * Insert the oid into the parent's list in order.
184 	 */
185 	q = NULL;
186 	SLIST_FOREACH(p, parent, oid_link) {
187 		if (oidp->oid_number < p->oid_number)
188 			break;
189 		q = p;
190 	}
191 	if (q)
192 		SLIST_INSERT_AFTER(q, oidp, oid_link);
193 	else
194 		SLIST_INSERT_HEAD(parent, oidp, oid_link);
195 }
196 
197 void
198 sysctl_unregister_oid(struct sysctl_oid *oidp)
199 {
200 	struct sysctl_oid *p;
201 	int error;
202 
203 	SYSCTL_ASSERT_XLOCKED();
204 	error = ENOENT;
205 	if (oidp->oid_number == OID_AUTO) {
206 		error = EINVAL;
207 	} else {
208 		SLIST_FOREACH(p, oidp->oid_parent, oid_link) {
209 			if (p == oidp) {
210 				SLIST_REMOVE(oidp->oid_parent, oidp,
211 				    sysctl_oid, oid_link);
212 				error = 0;
213 				break;
214 			}
215 		}
216 	}
217 
218 	/*
219 	 * This can happen when a module fails to register and is
220 	 * being unloaded afterwards.  It should not be a panic()
221 	 * for normal use.
222 	 */
223 	if (error)
224 		printf("%s: failed to unregister sysctl\n", __func__);
225 }
226 
227 /* Initialize a new context to keep track of dynamically added sysctls. */
228 int
229 sysctl_ctx_init(struct sysctl_ctx_list *c)
230 {
231 
232 	if (c == NULL) {
233 		return (EINVAL);
234 	}
235 
236 	/*
237 	 * No locking here, the caller is responsible for not adding
238 	 * new nodes to a context until after this function has
239 	 * returned.
240 	 */
241 	TAILQ_INIT(c);
242 	return (0);
243 }
244 
245 /* Free the context, and destroy all dynamic oids registered in this context */
246 int
247 sysctl_ctx_free(struct sysctl_ctx_list *clist)
248 {
249 	struct sysctl_ctx_entry *e, *e1;
250 	int error;
251 
252 	error = 0;
253 	/*
254 	 * First perform a "dry run" to check if it's ok to remove oids.
255 	 * XXX FIXME
256 	 * XXX This algorithm is a hack. But I don't know any
257 	 * XXX better solution for now...
258 	 */
259 	SYSCTL_XLOCK();
260 	TAILQ_FOREACH(e, clist, link) {
261 		error = sysctl_remove_oid_locked(e->entry, 0, 0);
262 		if (error)
263 			break;
264 	}
265 	/*
266 	 * Restore deregistered entries, either from the end,
267 	 * or from the place where error occured.
268 	 * e contains the entry that was not unregistered
269 	 */
270 	if (error)
271 		e1 = TAILQ_PREV(e, sysctl_ctx_list, link);
272 	else
273 		e1 = TAILQ_LAST(clist, sysctl_ctx_list);
274 	while (e1 != NULL) {
275 		sysctl_register_oid(e1->entry);
276 		e1 = TAILQ_PREV(e1, sysctl_ctx_list, link);
277 	}
278 	if (error) {
279 		SYSCTL_XUNLOCK();
280 		return(EBUSY);
281 	}
282 	/* Now really delete the entries */
283 	e = TAILQ_FIRST(clist);
284 	while (e != NULL) {
285 		e1 = TAILQ_NEXT(e, link);
286 		error = sysctl_remove_oid_locked(e->entry, 1, 0);
287 		if (error)
288 			panic("sysctl_remove_oid: corrupt tree, entry: %s",
289 			    e->entry->oid_name);
290 		free(e, M_SYSCTLOID);
291 		e = e1;
292 	}
293 	SYSCTL_XUNLOCK();
294 	return (error);
295 }
296 
297 /* Add an entry to the context */
298 struct sysctl_ctx_entry *
299 sysctl_ctx_entry_add(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
300 {
301 	struct sysctl_ctx_entry *e;
302 
303 	SYSCTL_ASSERT_XLOCKED();
304 	if (clist == NULL || oidp == NULL)
305 		return(NULL);
306 	e = malloc(sizeof(struct sysctl_ctx_entry), M_SYSCTLOID, M_WAITOK);
307 	e->entry = oidp;
308 	TAILQ_INSERT_HEAD(clist, e, link);
309 	return (e);
310 }
311 
312 /* Find an entry in the context */
313 struct sysctl_ctx_entry *
314 sysctl_ctx_entry_find(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
315 {
316 	struct sysctl_ctx_entry *e;
317 
318 	SYSCTL_ASSERT_XLOCKED();
319 	if (clist == NULL || oidp == NULL)
320 		return(NULL);
321 	TAILQ_FOREACH(e, clist, link) {
322 		if(e->entry == oidp)
323 			return(e);
324 	}
325 	return (e);
326 }
327 
328 /*
329  * Delete an entry from the context.
330  * NOTE: this function doesn't free oidp! You have to remove it
331  * with sysctl_remove_oid().
332  */
333 int
334 sysctl_ctx_entry_del(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
335 {
336 	struct sysctl_ctx_entry *e;
337 
338 	if (clist == NULL || oidp == NULL)
339 		return (EINVAL);
340 	SYSCTL_XLOCK();
341 	e = sysctl_ctx_entry_find(clist, oidp);
342 	if (e != NULL) {
343 		TAILQ_REMOVE(clist, e, link);
344 		SYSCTL_XUNLOCK();
345 		free(e, M_SYSCTLOID);
346 		return (0);
347 	} else {
348 		SYSCTL_XUNLOCK();
349 		return (ENOENT);
350 	}
351 }
352 
353 /*
354  * Remove dynamically created sysctl trees.
355  * oidp - top of the tree to be removed
356  * del - if 0 - just deregister, otherwise free up entries as well
357  * recurse - if != 0 traverse the subtree to be deleted
358  */
359 int
360 sysctl_remove_oid(struct sysctl_oid *oidp, int del, int recurse)
361 {
362 	int error;
363 
364 	SYSCTL_XLOCK();
365 	error = sysctl_remove_oid_locked(oidp, del, recurse);
366 	SYSCTL_XUNLOCK();
367 	return (error);
368 }
369 
370 int
371 sysctl_remove_name(struct sysctl_oid *parent, const char *name,
372     int del, int recurse)
373 {
374 	struct sysctl_oid *p, *tmp;
375 	int error;
376 
377 	error = ENOENT;
378 	SYSCTL_XLOCK();
379 	SLIST_FOREACH_SAFE(p, SYSCTL_CHILDREN(parent), oid_link, tmp) {
380 		if (strcmp(p->oid_name, name) == 0) {
381 			error = sysctl_remove_oid_locked(p, del, recurse);
382 			break;
383 		}
384 	}
385 	SYSCTL_XUNLOCK();
386 
387 	return (error);
388 }
389 
390 
391 static int
392 sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del, int recurse)
393 {
394 	struct sysctl_oid *p, *tmp;
395 	int error;
396 
397 	SYSCTL_ASSERT_XLOCKED();
398 	if (oidp == NULL)
399 		return(EINVAL);
400 	if ((oidp->oid_kind & CTLFLAG_DYN) == 0) {
401 		printf("can't remove non-dynamic nodes!\n");
402 		return (EINVAL);
403 	}
404 	/*
405 	 * WARNING: normal method to do this should be through
406 	 * sysctl_ctx_free(). Use recursing as the last resort
407 	 * method to purge your sysctl tree of leftovers...
408 	 * However, if some other code still references these nodes,
409 	 * it will panic.
410 	 */
411 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
412 		if (oidp->oid_refcnt == 1) {
413 			SLIST_FOREACH_SAFE(p,
414 			    SYSCTL_CHILDREN(oidp), oid_link, tmp) {
415 				if (!recurse)
416 					return (ENOTEMPTY);
417 				error = sysctl_remove_oid_locked(p, del,
418 				    recurse);
419 				if (error)
420 					return (error);
421 			}
422 			if (del)
423 				free(SYSCTL_CHILDREN(oidp), M_SYSCTLOID);
424 		}
425 	}
426 	if (oidp->oid_refcnt > 1 ) {
427 		oidp->oid_refcnt--;
428 	} else {
429 		if (oidp->oid_refcnt == 0) {
430 			printf("Warning: bad oid_refcnt=%u (%s)!\n",
431 				oidp->oid_refcnt, oidp->oid_name);
432 			return (EINVAL);
433 		}
434 		sysctl_unregister_oid(oidp);
435 		if (del) {
436 			/*
437 			 * Wait for all threads running the handler to drain.
438 			 * This preserves the previous behavior when the
439 			 * sysctl lock was held across a handler invocation,
440 			 * and is necessary for module unload correctness.
441 			 */
442 			while (oidp->oid_running > 0) {
443 				oidp->oid_kind |= CTLFLAG_DYING;
444 				SYSCTL_SLEEP(&oidp->oid_running, "oidrm", 0);
445 			}
446 			if (oidp->oid_descr)
447 				free(__DECONST(char *, oidp->oid_descr),
448 				    M_SYSCTLOID);
449 			free(__DECONST(char *, oidp->oid_name), M_SYSCTLOID);
450 			free(oidp, M_SYSCTLOID);
451 		}
452 	}
453 	return (0);
454 }
455 /*
456  * Create new sysctls at run time.
457  * clist may point to a valid context initialized with sysctl_ctx_init().
458  */
459 struct sysctl_oid *
460 sysctl_add_oid(struct sysctl_ctx_list *clist, struct sysctl_oid_list *parent,
461 	int number, const char *name, int kind, void *arg1, intptr_t arg2,
462 	int (*handler)(SYSCTL_HANDLER_ARGS), const char *fmt, const char *descr)
463 {
464 	struct sysctl_oid *oidp;
465 
466 	/* You have to hook up somewhere.. */
467 	if (parent == NULL)
468 		return(NULL);
469 	/* Check if the node already exists, otherwise create it */
470 	SYSCTL_XLOCK();
471 	oidp = sysctl_find_oidname(name, parent);
472 	if (oidp != NULL) {
473 		if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
474 			oidp->oid_refcnt++;
475 			/* Update the context */
476 			if (clist != NULL)
477 				sysctl_ctx_entry_add(clist, oidp);
478 			SYSCTL_XUNLOCK();
479 			return (oidp);
480 		} else {
481 			SYSCTL_XUNLOCK();
482 			printf("can't re-use a leaf (%s)!\n", name);
483 			return (NULL);
484 		}
485 	}
486 	oidp = malloc(sizeof(struct sysctl_oid), M_SYSCTLOID, M_WAITOK|M_ZERO);
487 	oidp->oid_parent = parent;
488 	SLIST_NEXT(oidp, oid_link) = NULL;
489 	oidp->oid_number = number;
490 	oidp->oid_refcnt = 1;
491 	oidp->oid_name = strdup(name, M_SYSCTLOID);
492 	oidp->oid_handler = handler;
493 	oidp->oid_kind = CTLFLAG_DYN | kind;
494 	if ((kind & CTLTYPE) == CTLTYPE_NODE) {
495 		/* Allocate space for children */
496 		SYSCTL_CHILDREN_SET(oidp, malloc(sizeof(struct sysctl_oid_list),
497 		    M_SYSCTLOID, M_WAITOK));
498 		SLIST_INIT(SYSCTL_CHILDREN(oidp));
499 		oidp->oid_arg2 = arg2;
500 	} else {
501 		oidp->oid_arg1 = arg1;
502 		oidp->oid_arg2 = arg2;
503 	}
504 	oidp->oid_fmt = fmt;
505 	if (descr)
506 		oidp->oid_descr = strdup(descr, M_SYSCTLOID);
507 	/* Update the context, if used */
508 	if (clist != NULL)
509 		sysctl_ctx_entry_add(clist, oidp);
510 	/* Register this oid */
511 	sysctl_register_oid(oidp);
512 	SYSCTL_XUNLOCK();
513 	return (oidp);
514 }
515 
516 /*
517  * Rename an existing oid.
518  */
519 void
520 sysctl_rename_oid(struct sysctl_oid *oidp, const char *name)
521 {
522 	char *newname;
523 	char *oldname;
524 
525 	newname = strdup(name, M_SYSCTLOID);
526 	SYSCTL_XLOCK();
527 	oldname = __DECONST(char *, oidp->oid_name);
528 	oidp->oid_name = newname;
529 	SYSCTL_XUNLOCK();
530 	free(oldname, M_SYSCTLOID);
531 }
532 
533 /*
534  * Reparent an existing oid.
535  */
536 int
537 sysctl_move_oid(struct sysctl_oid *oid, struct sysctl_oid_list *parent)
538 {
539 	struct sysctl_oid *oidp;
540 
541 	SYSCTL_XLOCK();
542 	if (oid->oid_parent == parent) {
543 		SYSCTL_XUNLOCK();
544 		return (0);
545 	}
546 	oidp = sysctl_find_oidname(oid->oid_name, parent);
547 	if (oidp != NULL) {
548 		SYSCTL_XUNLOCK();
549 		return (EEXIST);
550 	}
551 	sysctl_unregister_oid(oid);
552 	oid->oid_parent = parent;
553 	oid->oid_number = OID_AUTO;
554 	sysctl_register_oid(oid);
555 	SYSCTL_XUNLOCK();
556 	return (0);
557 }
558 
559 /*
560  * Register the kernel's oids on startup.
561  */
562 SET_DECLARE(sysctl_set, struct sysctl_oid);
563 
564 static void
565 sysctl_register_all(void *arg)
566 {
567 	struct sysctl_oid **oidp;
568 
569 	sx_init(&sysctlmemlock, "sysctl mem");
570 	SYSCTL_INIT();
571 	SYSCTL_XLOCK();
572 	SET_FOREACH(oidp, sysctl_set)
573 		sysctl_register_oid(*oidp);
574 	SYSCTL_XUNLOCK();
575 }
576 SYSINIT(sysctl, SI_SUB_KMEM, SI_ORDER_ANY, sysctl_register_all, 0);
577 
578 /*
579  * "Staff-functions"
580  *
581  * These functions implement a presently undocumented interface
582  * used by the sysctl program to walk the tree, and get the type
583  * so it can print the value.
584  * This interface is under work and consideration, and should probably
585  * be killed with a big axe by the first person who can find the time.
586  * (be aware though, that the proper interface isn't as obvious as it
587  * may seem, there are various conflicting requirements.
588  *
589  * {0,0}	printf the entire MIB-tree.
590  * {0,1,...}	return the name of the "..." OID.
591  * {0,2,...}	return the next OID.
592  * {0,3}	return the OID of the name in "new"
593  * {0,4,...}	return the kind & format info for the "..." OID.
594  * {0,5,...}	return the description the "..." OID.
595  */
596 
597 #ifdef SYSCTL_DEBUG
598 static void
599 sysctl_sysctl_debug_dump_node(struct sysctl_oid_list *l, int i)
600 {
601 	int k;
602 	struct sysctl_oid *oidp;
603 
604 	SYSCTL_ASSERT_XLOCKED();
605 	SLIST_FOREACH(oidp, l, oid_link) {
606 
607 		for (k=0; k<i; k++)
608 			printf(" ");
609 
610 		printf("%d %s ", oidp->oid_number, oidp->oid_name);
611 
612 		printf("%c%c",
613 			oidp->oid_kind & CTLFLAG_RD ? 'R':' ',
614 			oidp->oid_kind & CTLFLAG_WR ? 'W':' ');
615 
616 		if (oidp->oid_handler)
617 			printf(" *Handler");
618 
619 		switch (oidp->oid_kind & CTLTYPE) {
620 			case CTLTYPE_NODE:
621 				printf(" Node\n");
622 				if (!oidp->oid_handler) {
623 					sysctl_sysctl_debug_dump_node(
624 						oidp->oid_arg1, i+2);
625 				}
626 				break;
627 			case CTLTYPE_INT:    printf(" Int\n"); break;
628 			case CTLTYPE_UINT:   printf(" u_int\n"); break;
629 			case CTLTYPE_LONG:   printf(" Long\n"); break;
630 			case CTLTYPE_ULONG:  printf(" u_long\n"); break;
631 			case CTLTYPE_STRING: printf(" String\n"); break;
632 			case CTLTYPE_U64:    printf(" uint64_t\n"); break;
633 			case CTLTYPE_S64:    printf(" int64_t\n"); break;
634 			case CTLTYPE_OPAQUE: printf(" Opaque/struct\n"); break;
635 			default:	     printf("\n");
636 		}
637 
638 	}
639 }
640 
641 static int
642 sysctl_sysctl_debug(SYSCTL_HANDLER_ARGS)
643 {
644 	int error;
645 
646 	error = priv_check(req->td, PRIV_SYSCTL_DEBUG);
647 	if (error)
648 		return (error);
649 	SYSCTL_XLOCK();
650 	sysctl_sysctl_debug_dump_node(&sysctl__children, 0);
651 	SYSCTL_XUNLOCK();
652 	return (ENOENT);
653 }
654 
655 SYSCTL_PROC(_sysctl, 0, debug, CTLTYPE_STRING|CTLFLAG_RD,
656 	0, 0, sysctl_sysctl_debug, "-", "");
657 #endif
658 
659 static int
660 sysctl_sysctl_name(SYSCTL_HANDLER_ARGS)
661 {
662 	int *name = (int *) arg1;
663 	u_int namelen = arg2;
664 	int error = 0;
665 	struct sysctl_oid *oid;
666 	struct sysctl_oid_list *lsp = &sysctl__children, *lsp2;
667 	char buf[10];
668 
669 	SYSCTL_XLOCK();
670 	while (namelen) {
671 		if (!lsp) {
672 			snprintf(buf,sizeof(buf),"%d",*name);
673 			if (req->oldidx)
674 				error = SYSCTL_OUT(req, ".", 1);
675 			if (!error)
676 				error = SYSCTL_OUT(req, buf, strlen(buf));
677 			if (error)
678 				goto out;
679 			namelen--;
680 			name++;
681 			continue;
682 		}
683 		lsp2 = 0;
684 		SLIST_FOREACH(oid, lsp, oid_link) {
685 			if (oid->oid_number != *name)
686 				continue;
687 
688 			if (req->oldidx)
689 				error = SYSCTL_OUT(req, ".", 1);
690 			if (!error)
691 				error = SYSCTL_OUT(req, oid->oid_name,
692 					strlen(oid->oid_name));
693 			if (error)
694 				goto out;
695 
696 			namelen--;
697 			name++;
698 
699 			if ((oid->oid_kind & CTLTYPE) != CTLTYPE_NODE)
700 				break;
701 
702 			if (oid->oid_handler)
703 				break;
704 
705 			lsp2 = SYSCTL_CHILDREN(oid);
706 			break;
707 		}
708 		lsp = lsp2;
709 	}
710 	error = SYSCTL_OUT(req, "", 1);
711  out:
712 	SYSCTL_XUNLOCK();
713 	return (error);
714 }
715 
716 /*
717  * XXXRW/JA: Shouldn't return name data for nodes that we don't permit in
718  * capability mode.
719  */
720 static SYSCTL_NODE(_sysctl, 1, name, CTLFLAG_RD | CTLFLAG_CAPRD,
721     sysctl_sysctl_name, "");
722 
723 static int
724 sysctl_sysctl_next_ls(struct sysctl_oid_list *lsp, int *name, u_int namelen,
725 	int *next, int *len, int level, struct sysctl_oid **oidpp)
726 {
727 	struct sysctl_oid *oidp;
728 
729 	SYSCTL_ASSERT_XLOCKED();
730 	*len = level;
731 	SLIST_FOREACH(oidp, lsp, oid_link) {
732 		*next = oidp->oid_number;
733 		*oidpp = oidp;
734 
735 		if (oidp->oid_kind & CTLFLAG_SKIP)
736 			continue;
737 
738 		if (!namelen) {
739 			if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
740 				return (0);
741 			if (oidp->oid_handler)
742 				/* We really should call the handler here...*/
743 				return (0);
744 			lsp = SYSCTL_CHILDREN(oidp);
745 			if (!sysctl_sysctl_next_ls(lsp, 0, 0, next+1,
746 				len, level+1, oidpp))
747 				return (0);
748 			goto emptynode;
749 		}
750 
751 		if (oidp->oid_number < *name)
752 			continue;
753 
754 		if (oidp->oid_number > *name) {
755 			if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
756 				return (0);
757 			if (oidp->oid_handler)
758 				return (0);
759 			lsp = SYSCTL_CHILDREN(oidp);
760 			if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1,
761 				next+1, len, level+1, oidpp))
762 				return (0);
763 			goto next;
764 		}
765 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
766 			continue;
767 
768 		if (oidp->oid_handler)
769 			continue;
770 
771 		lsp = SYSCTL_CHILDREN(oidp);
772 		if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1, next+1,
773 			len, level+1, oidpp))
774 			return (0);
775 	next:
776 		namelen = 1;
777 	emptynode:
778 		*len = level;
779 	}
780 	return (1);
781 }
782 
783 static int
784 sysctl_sysctl_next(SYSCTL_HANDLER_ARGS)
785 {
786 	int *name = (int *) arg1;
787 	u_int namelen = arg2;
788 	int i, j, error;
789 	struct sysctl_oid *oid;
790 	struct sysctl_oid_list *lsp = &sysctl__children;
791 	int newoid[CTL_MAXNAME];
792 
793 	SYSCTL_XLOCK();
794 	i = sysctl_sysctl_next_ls(lsp, name, namelen, newoid, &j, 1, &oid);
795 	SYSCTL_XUNLOCK();
796 	if (i)
797 		return (ENOENT);
798 	error = SYSCTL_OUT(req, newoid, j * sizeof (int));
799 	return (error);
800 }
801 
802 /*
803  * XXXRW/JA: Shouldn't return next data for nodes that we don't permit in
804  * capability mode.
805  */
806 static SYSCTL_NODE(_sysctl, 2, next, CTLFLAG_RD | CTLFLAG_CAPRD,
807     sysctl_sysctl_next, "");
808 
809 static int
810 name2oid(char *name, int *oid, int *len, struct sysctl_oid **oidpp)
811 {
812 	struct sysctl_oid *oidp;
813 	struct sysctl_oid_list *lsp = &sysctl__children;
814 	char *p;
815 
816 	SYSCTL_ASSERT_XLOCKED();
817 
818 	for (*len = 0; *len < CTL_MAXNAME;) {
819 		p = strsep(&name, ".");
820 
821 		oidp = SLIST_FIRST(lsp);
822 		for (;; oidp = SLIST_NEXT(oidp, oid_link)) {
823 			if (oidp == NULL)
824 				return (ENOENT);
825 			if (strcmp(p, oidp->oid_name) == 0)
826 				break;
827 		}
828 		*oid++ = oidp->oid_number;
829 		(*len)++;
830 
831 		if (name == NULL || *name == '\0') {
832 			if (oidpp)
833 				*oidpp = oidp;
834 			return (0);
835 		}
836 
837 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
838 			break;
839 
840 		if (oidp->oid_handler)
841 			break;
842 
843 		lsp = SYSCTL_CHILDREN(oidp);
844 	}
845 	return (ENOENT);
846 }
847 
848 static int
849 sysctl_sysctl_name2oid(SYSCTL_HANDLER_ARGS)
850 {
851 	char *p;
852 	int error, oid[CTL_MAXNAME], len = 0;
853 	struct sysctl_oid *op = 0;
854 
855 	if (!req->newlen)
856 		return (ENOENT);
857 	if (req->newlen >= MAXPATHLEN)	/* XXX arbitrary, undocumented */
858 		return (ENAMETOOLONG);
859 
860 	p = malloc(req->newlen+1, M_SYSCTL, M_WAITOK);
861 
862 	error = SYSCTL_IN(req, p, req->newlen);
863 	if (error) {
864 		free(p, M_SYSCTL);
865 		return (error);
866 	}
867 
868 	p [req->newlen] = '\0';
869 
870 	SYSCTL_XLOCK();
871 	error = name2oid(p, oid, &len, &op);
872 	SYSCTL_XUNLOCK();
873 
874 	free(p, M_SYSCTL);
875 
876 	if (error)
877 		return (error);
878 
879 	error = SYSCTL_OUT(req, oid, len * sizeof *oid);
880 	return (error);
881 }
882 
883 /*
884  * XXXRW/JA: Shouldn't return name2oid data for nodes that we don't permit in
885  * capability mode.
886  */
887 SYSCTL_PROC(_sysctl, 3, name2oid,
888     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_MPSAFE
889     | CTLFLAG_CAPRW, 0, 0, sysctl_sysctl_name2oid, "I", "");
890 
891 static int
892 sysctl_sysctl_oidfmt(SYSCTL_HANDLER_ARGS)
893 {
894 	struct sysctl_oid *oid;
895 	int error;
896 
897 	SYSCTL_XLOCK();
898 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
899 	if (error)
900 		goto out;
901 
902 	if (oid->oid_fmt == NULL) {
903 		error = ENOENT;
904 		goto out;
905 	}
906 	error = SYSCTL_OUT(req, &oid->oid_kind, sizeof(oid->oid_kind));
907 	if (error)
908 		goto out;
909 	error = SYSCTL_OUT(req, oid->oid_fmt, strlen(oid->oid_fmt) + 1);
910  out:
911 	SYSCTL_XUNLOCK();
912 	return (error);
913 }
914 
915 
916 static SYSCTL_NODE(_sysctl, 4, oidfmt, CTLFLAG_RD|CTLFLAG_MPSAFE|CTLFLAG_CAPRD,
917     sysctl_sysctl_oidfmt, "");
918 
919 static int
920 sysctl_sysctl_oiddescr(SYSCTL_HANDLER_ARGS)
921 {
922 	struct sysctl_oid *oid;
923 	int error;
924 
925 	SYSCTL_XLOCK();
926 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
927 	if (error)
928 		goto out;
929 
930 	if (oid->oid_descr == NULL) {
931 		error = ENOENT;
932 		goto out;
933 	}
934 	error = SYSCTL_OUT(req, oid->oid_descr, strlen(oid->oid_descr) + 1);
935  out:
936 	SYSCTL_XUNLOCK();
937 	return (error);
938 }
939 
940 static SYSCTL_NODE(_sysctl, 5, oiddescr, CTLFLAG_RD|CTLFLAG_CAPRD,
941     sysctl_sysctl_oiddescr, "");
942 
943 /*
944  * Default "handler" functions.
945  */
946 
947 /*
948  * Handle an int, signed or unsigned.
949  * Two cases:
950  *     a variable:  point arg1 at it.
951  *     a constant:  pass it in arg2.
952  */
953 
954 int
955 sysctl_handle_int(SYSCTL_HANDLER_ARGS)
956 {
957 	int tmpout, error = 0;
958 
959 	/*
960 	 * Attempt to get a coherent snapshot by making a copy of the data.
961 	 */
962 	if (arg1)
963 		tmpout = *(int *)arg1;
964 	else
965 		tmpout = arg2;
966 	error = SYSCTL_OUT(req, &tmpout, sizeof(int));
967 
968 	if (error || !req->newptr)
969 		return (error);
970 
971 	if (!arg1)
972 		error = EPERM;
973 	else
974 		error = SYSCTL_IN(req, arg1, sizeof(int));
975 	return (error);
976 }
977 
978 /*
979  * Based on on sysctl_handle_int() convert milliseconds into ticks.
980  * Note: this is used by TCP.
981  */
982 
983 int
984 sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
985 {
986 	int error, s, tt;
987 
988 	tt = *(int *)arg1;
989 	s = (int)((int64_t)tt * 1000 / hz);
990 
991 	error = sysctl_handle_int(oidp, &s, 0, req);
992 	if (error || !req->newptr)
993 		return (error);
994 
995 	tt = (int)((int64_t)s * hz / 1000);
996 	if (tt < 1)
997 		return (EINVAL);
998 
999 	*(int *)arg1 = tt;
1000 	return (0);
1001 }
1002 
1003 
1004 /*
1005  * Handle a long, signed or unsigned.
1006  * Two cases:
1007  *     a variable:  point arg1 at it.
1008  *     a constant:  pass it in arg2.
1009  */
1010 
1011 int
1012 sysctl_handle_long(SYSCTL_HANDLER_ARGS)
1013 {
1014 	int error = 0;
1015 	long tmplong;
1016 #ifdef SCTL_MASK32
1017 	int tmpint;
1018 #endif
1019 
1020 	/*
1021 	 * Attempt to get a coherent snapshot by making a copy of the data.
1022 	 */
1023 	if (arg1)
1024 		tmplong = *(long *)arg1;
1025 	else
1026 		tmplong = arg2;
1027 #ifdef SCTL_MASK32
1028 	if (req->flags & SCTL_MASK32) {
1029 		tmpint = tmplong;
1030 		error = SYSCTL_OUT(req, &tmpint, sizeof(int));
1031 	} else
1032 #endif
1033 		error = SYSCTL_OUT(req, &tmplong, sizeof(long));
1034 
1035 	if (error || !req->newptr)
1036 		return (error);
1037 
1038 	if (!arg1)
1039 		error = EPERM;
1040 #ifdef SCTL_MASK32
1041 	else if (req->flags & SCTL_MASK32) {
1042 		error = SYSCTL_IN(req, &tmpint, sizeof(int));
1043 		*(long *)arg1 = (long)tmpint;
1044 	}
1045 #endif
1046 	else
1047 		error = SYSCTL_IN(req, arg1, sizeof(long));
1048 	return (error);
1049 }
1050 
1051 /*
1052  * Handle a 64 bit int, signed or unsigned.
1053  * Two cases:
1054  *     a variable:  point arg1 at it.
1055  *     a constant:  pass it in arg2.
1056  */
1057 int
1058 sysctl_handle_64(SYSCTL_HANDLER_ARGS)
1059 {
1060 	int error = 0;
1061 	uint64_t tmpout;
1062 
1063 	/*
1064 	 * Attempt to get a coherent snapshot by making a copy of the data.
1065 	 */
1066 	if (arg1)
1067 		tmpout = *(uint64_t *)arg1;
1068 	else
1069 		tmpout = arg2;
1070 	error = SYSCTL_OUT(req, &tmpout, sizeof(uint64_t));
1071 
1072 	if (error || !req->newptr)
1073 		return (error);
1074 
1075 	if (!arg1)
1076 		error = EPERM;
1077 	else
1078 		error = SYSCTL_IN(req, arg1, sizeof(uint64_t));
1079 	return (error);
1080 }
1081 
1082 /*
1083  * Handle our generic '\0' terminated 'C' string.
1084  * Two cases:
1085  * 	a variable string:  point arg1 at it, arg2 is max length.
1086  * 	a constant string:  point arg1 at it, arg2 is zero.
1087  */
1088 
1089 int
1090 sysctl_handle_string(SYSCTL_HANDLER_ARGS)
1091 {
1092 	int error=0;
1093 	char *tmparg;
1094 	size_t outlen;
1095 
1096 	/*
1097 	 * Attempt to get a coherent snapshot by copying to a
1098 	 * temporary kernel buffer.
1099 	 */
1100 retry:
1101 	outlen = strlen((char *)arg1)+1;
1102 	tmparg = malloc(outlen, M_SYSCTLTMP, M_WAITOK);
1103 
1104 	if (strlcpy(tmparg, (char *)arg1, outlen) >= outlen) {
1105 		free(tmparg, M_SYSCTLTMP);
1106 		goto retry;
1107 	}
1108 
1109 	error = SYSCTL_OUT(req, tmparg, outlen);
1110 	free(tmparg, M_SYSCTLTMP);
1111 
1112 	if (error || !req->newptr)
1113 		return (error);
1114 
1115 	if ((req->newlen - req->newidx) >= arg2) {
1116 		error = EINVAL;
1117 	} else {
1118 		arg2 = (req->newlen - req->newidx);
1119 		error = SYSCTL_IN(req, arg1, arg2);
1120 		((char *)arg1)[arg2] = '\0';
1121 	}
1122 
1123 	return (error);
1124 }
1125 
1126 /*
1127  * Handle any kind of opaque data.
1128  * arg1 points to it, arg2 is the size.
1129  */
1130 
1131 int
1132 sysctl_handle_opaque(SYSCTL_HANDLER_ARGS)
1133 {
1134 	int error, tries;
1135 	u_int generation;
1136 	struct sysctl_req req2;
1137 
1138 	/*
1139 	 * Attempt to get a coherent snapshot, by using the thread
1140 	 * pre-emption counter updated from within mi_switch() to
1141 	 * determine if we were pre-empted during a bcopy() or
1142 	 * copyout(). Make 3 attempts at doing this before giving up.
1143 	 * If we encounter an error, stop immediately.
1144 	 */
1145 	tries = 0;
1146 	req2 = *req;
1147 retry:
1148 	generation = curthread->td_generation;
1149 	error = SYSCTL_OUT(req, arg1, arg2);
1150 	if (error)
1151 		return (error);
1152 	tries++;
1153 	if (generation != curthread->td_generation && tries < 3) {
1154 		*req = req2;
1155 		goto retry;
1156 	}
1157 
1158 	error = SYSCTL_IN(req, arg1, arg2);
1159 
1160 	return (error);
1161 }
1162 
1163 /*
1164  * Transfer functions to/from kernel space.
1165  * XXX: rather untested at this point
1166  */
1167 static int
1168 sysctl_old_kernel(struct sysctl_req *req, const void *p, size_t l)
1169 {
1170 	size_t i = 0;
1171 
1172 	if (req->oldptr) {
1173 		i = l;
1174 		if (req->oldlen <= req->oldidx)
1175 			i = 0;
1176 		else
1177 			if (i > req->oldlen - req->oldidx)
1178 				i = req->oldlen - req->oldidx;
1179 		if (i > 0)
1180 			bcopy(p, (char *)req->oldptr + req->oldidx, i);
1181 	}
1182 	req->oldidx += l;
1183 	if (req->oldptr && i != l)
1184 		return (ENOMEM);
1185 	return (0);
1186 }
1187 
1188 static int
1189 sysctl_new_kernel(struct sysctl_req *req, void *p, size_t l)
1190 {
1191 	if (!req->newptr)
1192 		return (0);
1193 	if (req->newlen - req->newidx < l)
1194 		return (EINVAL);
1195 	bcopy((char *)req->newptr + req->newidx, p, l);
1196 	req->newidx += l;
1197 	return (0);
1198 }
1199 
1200 int
1201 kernel_sysctl(struct thread *td, int *name, u_int namelen, void *old,
1202     size_t *oldlenp, void *new, size_t newlen, size_t *retval, int flags)
1203 {
1204 	int error = 0;
1205 	struct sysctl_req req;
1206 
1207 	bzero(&req, sizeof req);
1208 
1209 	req.td = td;
1210 	req.flags = flags;
1211 
1212 	if (oldlenp) {
1213 		req.oldlen = *oldlenp;
1214 	}
1215 	req.validlen = req.oldlen;
1216 
1217 	if (old) {
1218 		req.oldptr= old;
1219 	}
1220 
1221 	if (new != NULL) {
1222 		req.newlen = newlen;
1223 		req.newptr = new;
1224 	}
1225 
1226 	req.oldfunc = sysctl_old_kernel;
1227 	req.newfunc = sysctl_new_kernel;
1228 	req.lock = REQ_UNWIRED;
1229 
1230 	SYSCTL_XLOCK();
1231 	error = sysctl_root(0, name, namelen, &req);
1232 	SYSCTL_XUNLOCK();
1233 
1234 	if (req.lock == REQ_WIRED && req.validlen > 0)
1235 		vsunlock(req.oldptr, req.validlen);
1236 
1237 	if (error && error != ENOMEM)
1238 		return (error);
1239 
1240 	if (retval) {
1241 		if (req.oldptr && req.oldidx > req.validlen)
1242 			*retval = req.validlen;
1243 		else
1244 			*retval = req.oldidx;
1245 	}
1246 	return (error);
1247 }
1248 
1249 int
1250 kernel_sysctlbyname(struct thread *td, char *name, void *old, size_t *oldlenp,
1251     void *new, size_t newlen, size_t *retval, int flags)
1252 {
1253         int oid[CTL_MAXNAME];
1254         size_t oidlen, plen;
1255 	int error;
1256 
1257 	oid[0] = 0;		/* sysctl internal magic */
1258 	oid[1] = 3;		/* name2oid */
1259 	oidlen = sizeof(oid);
1260 
1261 	error = kernel_sysctl(td, oid, 2, oid, &oidlen,
1262 	    (void *)name, strlen(name), &plen, flags);
1263 	if (error)
1264 		return (error);
1265 
1266 	error = kernel_sysctl(td, oid, plen / sizeof(int), old, oldlenp,
1267 	    new, newlen, retval, flags);
1268 	return (error);
1269 }
1270 
1271 /*
1272  * Transfer function to/from user space.
1273  */
1274 static int
1275 sysctl_old_user(struct sysctl_req *req, const void *p, size_t l)
1276 {
1277 	size_t i, len, origidx;
1278 	int error;
1279 
1280 	origidx = req->oldidx;
1281 	req->oldidx += l;
1282 	if (req->oldptr == NULL)
1283 		return (0);
1284 	/*
1285 	 * If we have not wired the user supplied buffer and we are currently
1286 	 * holding locks, drop a witness warning, as it's possible that
1287 	 * write operations to the user page can sleep.
1288 	 */
1289 	if (req->lock != REQ_WIRED)
1290 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
1291 		    "sysctl_old_user()");
1292 	i = l;
1293 	len = req->validlen;
1294 	if (len <= origidx)
1295 		i = 0;
1296 	else {
1297 		if (i > len - origidx)
1298 			i = len - origidx;
1299 		if (req->lock == REQ_WIRED) {
1300 			error = copyout_nofault(p, (char *)req->oldptr +
1301 			    origidx, i);
1302 		} else
1303 			error = copyout(p, (char *)req->oldptr + origidx, i);
1304 		if (error != 0)
1305 			return (error);
1306 	}
1307 	if (i < l)
1308 		return (ENOMEM);
1309 	return (0);
1310 }
1311 
1312 static int
1313 sysctl_new_user(struct sysctl_req *req, void *p, size_t l)
1314 {
1315 	int error;
1316 
1317 	if (!req->newptr)
1318 		return (0);
1319 	if (req->newlen - req->newidx < l)
1320 		return (EINVAL);
1321 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
1322 	    "sysctl_new_user()");
1323 	error = copyin((char *)req->newptr + req->newidx, p, l);
1324 	req->newidx += l;
1325 	return (error);
1326 }
1327 
1328 /*
1329  * Wire the user space destination buffer.  If set to a value greater than
1330  * zero, the len parameter limits the maximum amount of wired memory.
1331  */
1332 int
1333 sysctl_wire_old_buffer(struct sysctl_req *req, size_t len)
1334 {
1335 	int ret;
1336 	size_t wiredlen;
1337 
1338 	wiredlen = (len > 0 && len < req->oldlen) ? len : req->oldlen;
1339 	ret = 0;
1340 	if (req->lock != REQ_WIRED && req->oldptr &&
1341 	    req->oldfunc == sysctl_old_user) {
1342 		if (wiredlen != 0) {
1343 			ret = vslock(req->oldptr, wiredlen);
1344 			if (ret != 0) {
1345 				if (ret != ENOMEM)
1346 					return (ret);
1347 				wiredlen = 0;
1348 			}
1349 		}
1350 		req->lock = REQ_WIRED;
1351 		req->validlen = wiredlen;
1352 	}
1353 	return (0);
1354 }
1355 
1356 int
1357 sysctl_find_oid(int *name, u_int namelen, struct sysctl_oid **noid,
1358     int *nindx, struct sysctl_req *req)
1359 {
1360 	struct sysctl_oid_list *lsp;
1361 	struct sysctl_oid *oid;
1362 	int indx;
1363 
1364 	SYSCTL_ASSERT_XLOCKED();
1365 	lsp = &sysctl__children;
1366 	indx = 0;
1367 	while (indx < CTL_MAXNAME) {
1368 		SLIST_FOREACH(oid, lsp, oid_link) {
1369 			if (oid->oid_number == name[indx])
1370 				break;
1371 		}
1372 		if (oid == NULL)
1373 			return (ENOENT);
1374 
1375 		indx++;
1376 		if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1377 			if (oid->oid_handler != NULL || indx == namelen) {
1378 				*noid = oid;
1379 				if (nindx != NULL)
1380 					*nindx = indx;
1381 				KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
1382 				    ("%s found DYING node %p", __func__, oid));
1383 				return (0);
1384 			}
1385 			lsp = SYSCTL_CHILDREN(oid);
1386 		} else if (indx == namelen) {
1387 			*noid = oid;
1388 			if (nindx != NULL)
1389 				*nindx = indx;
1390 			KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
1391 			    ("%s found DYING node %p", __func__, oid));
1392 			return (0);
1393 		} else {
1394 			return (ENOTDIR);
1395 		}
1396 	}
1397 	return (ENOENT);
1398 }
1399 
1400 /*
1401  * Traverse our tree, and find the right node, execute whatever it points
1402  * to, and return the resulting error code.
1403  */
1404 
1405 static int
1406 sysctl_root(SYSCTL_HANDLER_ARGS)
1407 {
1408 	struct sysctl_oid *oid;
1409 	int error, indx, lvl;
1410 
1411 	SYSCTL_ASSERT_XLOCKED();
1412 
1413 	error = sysctl_find_oid(arg1, arg2, &oid, &indx, req);
1414 	if (error)
1415 		return (error);
1416 
1417 	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1418 		/*
1419 		 * You can't call a sysctl when it's a node, but has
1420 		 * no handler.  Inform the user that it's a node.
1421 		 * The indx may or may not be the same as namelen.
1422 		 */
1423 		if (oid->oid_handler == NULL)
1424 			return (EISDIR);
1425 	}
1426 
1427 	/* Is this sysctl writable? */
1428 	if (req->newptr && !(oid->oid_kind & CTLFLAG_WR))
1429 		return (EPERM);
1430 
1431 	KASSERT(req->td != NULL, ("sysctl_root(): req->td == NULL"));
1432 
1433 #ifdef CAPABILITY_MODE
1434 	/*
1435 	 * If the process is in capability mode, then don't permit reading or
1436 	 * writing unless specifically granted for the node.
1437 	 */
1438 	if (IN_CAPABILITY_MODE(req->td)) {
1439 		if (req->oldptr && !(oid->oid_kind & CTLFLAG_CAPRD))
1440 			return (EPERM);
1441 		if (req->newptr && !(oid->oid_kind & CTLFLAG_CAPWR))
1442 			return (EPERM);
1443 	}
1444 #endif
1445 
1446 	/* Is this sysctl sensitive to securelevels? */
1447 	if (req->newptr && (oid->oid_kind & CTLFLAG_SECURE)) {
1448 		lvl = (oid->oid_kind & CTLMASK_SECURE) >> CTLSHIFT_SECURE;
1449 		error = securelevel_gt(req->td->td_ucred, lvl);
1450 		if (error)
1451 			return (error);
1452 	}
1453 
1454 	/* Is this sysctl writable by only privileged users? */
1455 	if (req->newptr && !(oid->oid_kind & CTLFLAG_ANYBODY)) {
1456 		int priv;
1457 
1458 		if (oid->oid_kind & CTLFLAG_PRISON)
1459 			priv = PRIV_SYSCTL_WRITEJAIL;
1460 #ifdef VIMAGE
1461 		else if ((oid->oid_kind & CTLFLAG_VNET) &&
1462 		     prison_owns_vnet(req->td->td_ucred))
1463 			priv = PRIV_SYSCTL_WRITEJAIL;
1464 #endif
1465 		else
1466 			priv = PRIV_SYSCTL_WRITE;
1467 		error = priv_check(req->td, priv);
1468 		if (error)
1469 			return (error);
1470 	}
1471 
1472 	if (!oid->oid_handler)
1473 		return (EINVAL);
1474 
1475 	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1476 		arg1 = (int *)arg1 + indx;
1477 		arg2 -= indx;
1478 	} else {
1479 		arg1 = oid->oid_arg1;
1480 		arg2 = oid->oid_arg2;
1481 	}
1482 #ifdef MAC
1483 	error = mac_system_check_sysctl(req->td->td_ucred, oid, arg1, arg2,
1484 	    req);
1485 	if (error != 0)
1486 		return (error);
1487 #endif
1488 	oid->oid_running++;
1489 	SYSCTL_XUNLOCK();
1490 
1491 	if (!(oid->oid_kind & CTLFLAG_MPSAFE))
1492 		mtx_lock(&Giant);
1493 	error = oid->oid_handler(oid, arg1, arg2, req);
1494 	if (!(oid->oid_kind & CTLFLAG_MPSAFE))
1495 		mtx_unlock(&Giant);
1496 
1497 	KFAIL_POINT_ERROR(_debug_fail_point, sysctl_running, error);
1498 
1499 	SYSCTL_XLOCK();
1500 	oid->oid_running--;
1501 	if (oid->oid_running == 0 && (oid->oid_kind & CTLFLAG_DYING) != 0)
1502 		wakeup(&oid->oid_running);
1503 	return (error);
1504 }
1505 
1506 #ifndef _SYS_SYSPROTO_H_
1507 struct sysctl_args {
1508 	int	*name;
1509 	u_int	namelen;
1510 	void	*old;
1511 	size_t	*oldlenp;
1512 	void	*new;
1513 	size_t	newlen;
1514 };
1515 #endif
1516 int
1517 sys___sysctl(struct thread *td, struct sysctl_args *uap)
1518 {
1519 	int error, i, name[CTL_MAXNAME];
1520 	size_t j;
1521 
1522 	if (uap->namelen > CTL_MAXNAME || uap->namelen < 2)
1523 		return (EINVAL);
1524 
1525  	error = copyin(uap->name, &name, uap->namelen * sizeof(int));
1526  	if (error)
1527 		return (error);
1528 
1529 	error = userland_sysctl(td, name, uap->namelen,
1530 		uap->old, uap->oldlenp, 0,
1531 		uap->new, uap->newlen, &j, 0);
1532 	if (error && error != ENOMEM)
1533 		return (error);
1534 	if (uap->oldlenp) {
1535 		i = copyout(&j, uap->oldlenp, sizeof(j));
1536 		if (i)
1537 			return (i);
1538 	}
1539 	return (error);
1540 }
1541 
1542 /*
1543  * This is used from various compatibility syscalls too.  That's why name
1544  * must be in kernel space.
1545  */
1546 int
1547 userland_sysctl(struct thread *td, int *name, u_int namelen, void *old,
1548     size_t *oldlenp, int inkernel, void *new, size_t newlen, size_t *retval,
1549     int flags)
1550 {
1551 	int error = 0, memlocked;
1552 	struct sysctl_req req;
1553 
1554 	bzero(&req, sizeof req);
1555 
1556 	req.td = td;
1557 	req.flags = flags;
1558 
1559 	if (oldlenp) {
1560 		if (inkernel) {
1561 			req.oldlen = *oldlenp;
1562 		} else {
1563 			error = copyin(oldlenp, &req.oldlen, sizeof(*oldlenp));
1564 			if (error)
1565 				return (error);
1566 		}
1567 	}
1568 	req.validlen = req.oldlen;
1569 
1570 	if (old) {
1571 		if (!useracc(old, req.oldlen, VM_PROT_WRITE))
1572 			return (EFAULT);
1573 		req.oldptr= old;
1574 	}
1575 
1576 	if (new != NULL) {
1577 		if (!useracc(new, newlen, VM_PROT_READ))
1578 			return (EFAULT);
1579 		req.newlen = newlen;
1580 		req.newptr = new;
1581 	}
1582 
1583 	req.oldfunc = sysctl_old_user;
1584 	req.newfunc = sysctl_new_user;
1585 	req.lock = REQ_UNWIRED;
1586 
1587 #ifdef KTRACE
1588 	if (KTRPOINT(curthread, KTR_SYSCTL))
1589 		ktrsysctl(name, namelen);
1590 #endif
1591 
1592 	if (req.oldlen > PAGE_SIZE) {
1593 		memlocked = 1;
1594 		sx_xlock(&sysctlmemlock);
1595 	} else
1596 		memlocked = 0;
1597 	CURVNET_SET(TD_TO_VNET(td));
1598 
1599 	for (;;) {
1600 		req.oldidx = 0;
1601 		req.newidx = 0;
1602 		SYSCTL_XLOCK();
1603 		error = sysctl_root(0, name, namelen, &req);
1604 		SYSCTL_XUNLOCK();
1605 		if (error != EAGAIN)
1606 			break;
1607 		kern_yield(PRI_USER);
1608 	}
1609 
1610 	CURVNET_RESTORE();
1611 
1612 	if (req.lock == REQ_WIRED && req.validlen > 0)
1613 		vsunlock(req.oldptr, req.validlen);
1614 	if (memlocked)
1615 		sx_xunlock(&sysctlmemlock);
1616 
1617 	if (error && error != ENOMEM)
1618 		return (error);
1619 
1620 	if (retval) {
1621 		if (req.oldptr && req.oldidx > req.validlen)
1622 			*retval = req.validlen;
1623 		else
1624 			*retval = req.oldidx;
1625 	}
1626 	return (error);
1627 }
1628 
1629 /*
1630  * Drain into a sysctl struct.  The user buffer should be wired if a page
1631  * fault would cause issue.
1632  */
1633 static int
1634 sbuf_sysctl_drain(void *arg, const char *data, int len)
1635 {
1636 	struct sysctl_req *req = arg;
1637 	int error;
1638 
1639 	error = SYSCTL_OUT(req, data, len);
1640 	KASSERT(error >= 0, ("Got unexpected negative value %d", error));
1641 	return (error == 0 ? len : -error);
1642 }
1643 
1644 struct sbuf *
1645 sbuf_new_for_sysctl(struct sbuf *s, char *buf, int length,
1646     struct sysctl_req *req)
1647 {
1648 
1649 	s = sbuf_new(s, buf, length, SBUF_FIXEDLEN);
1650 	sbuf_set_drain(s, sbuf_sysctl_drain, req);
1651 	return (s);
1652 }
1653