xref: /freebsd/sys/kern/kern_sysctl.c (revision 094fc1ed0f2627525c7b0342efcbad5be7a8546a)
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  */
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/capsicum.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/rmlock.h>
58 #include <sys/sbuf.h>
59 #include <sys/sx.h>
60 #include <sys/sysproto.h>
61 #include <sys/uio.h>
62 #ifdef KTRACE
63 #include <sys/ktrace.h>
64 #endif
65 
66 #include <net/vnet.h>
67 
68 #include <security/mac/mac_framework.h>
69 
70 #include <vm/vm.h>
71 #include <vm/vm_extern.h>
72 
73 static MALLOC_DEFINE(M_SYSCTL, "sysctl", "sysctl internal magic");
74 static MALLOC_DEFINE(M_SYSCTLOID, "sysctloid", "sysctl dynamic oids");
75 static MALLOC_DEFINE(M_SYSCTLTMP, "sysctltmp", "sysctl temp output buffer");
76 
77 /*
78  * The sysctllock protects the MIB tree.  It also protects sysctl
79  * contexts used with dynamic sysctls.  The sysctl_register_oid() and
80  * sysctl_unregister_oid() routines require the sysctllock to already
81  * be held, so the sysctl_wlock() and sysctl_wunlock() routines are
82  * provided for the few places in the kernel which need to use that
83  * API rather than using the dynamic API.  Use of the dynamic API is
84  * strongly encouraged for most code.
85  *
86  * The sysctlmemlock is used to limit the amount of user memory wired for
87  * sysctl requests.  This is implemented by serializing any userland
88  * sysctl requests larger than a single page via an exclusive lock.
89  */
90 static struct rmlock sysctllock;
91 static struct sx __exclusive_cache_line sysctlmemlock;
92 
93 #define	SYSCTL_WLOCK()		rm_wlock(&sysctllock)
94 #define	SYSCTL_WUNLOCK()	rm_wunlock(&sysctllock)
95 #define	SYSCTL_RLOCK(tracker)	rm_rlock(&sysctllock, (tracker))
96 #define	SYSCTL_RUNLOCK(tracker)	rm_runlock(&sysctllock, (tracker))
97 #define	SYSCTL_WLOCKED()	rm_wowned(&sysctllock)
98 #define	SYSCTL_ASSERT_LOCKED()	rm_assert(&sysctllock, RA_LOCKED)
99 #define	SYSCTL_ASSERT_WLOCKED()	rm_assert(&sysctllock, RA_WLOCKED)
100 #define	SYSCTL_ASSERT_RLOCKED()	rm_assert(&sysctllock, RA_RLOCKED)
101 #define	SYSCTL_INIT()		rm_init_flags(&sysctllock, "sysctl lock", \
102 				    RM_SLEEPABLE)
103 #define	SYSCTL_SLEEP(ch, wmesg, timo)					\
104 				rm_sleep(ch, &sysctllock, 0, wmesg, timo)
105 
106 static int sysctl_root(SYSCTL_HANDLER_ARGS);
107 
108 /* Root list */
109 struct sysctl_oid_list sysctl__children = SLIST_HEAD_INITIALIZER(&sysctl__children);
110 
111 static int	sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del,
112 		    int recurse);
113 static int	sysctl_old_kernel(struct sysctl_req *, const void *, size_t);
114 static int	sysctl_new_kernel(struct sysctl_req *, void *, size_t);
115 
116 static struct sysctl_oid *
117 sysctl_find_oidname(const char *name, struct sysctl_oid_list *list)
118 {
119 	struct sysctl_oid *oidp;
120 
121 	SYSCTL_ASSERT_LOCKED();
122 	SLIST_FOREACH(oidp, list, oid_link) {
123 		if (strcmp(oidp->oid_name, name) == 0) {
124 			return (oidp);
125 		}
126 	}
127 	return (NULL);
128 }
129 
130 /*
131  * Initialization of the MIB tree.
132  *
133  * Order by number in each list.
134  */
135 void
136 sysctl_wlock(void)
137 {
138 
139 	SYSCTL_WLOCK();
140 }
141 
142 void
143 sysctl_wunlock(void)
144 {
145 
146 	SYSCTL_WUNLOCK();
147 }
148 
149 static int
150 sysctl_root_handler_locked(struct sysctl_oid *oid, void *arg1, intmax_t arg2,
151     struct sysctl_req *req, struct rm_priotracker *tracker)
152 {
153 	int error;
154 
155 	if (oid->oid_kind & CTLFLAG_DYN)
156 		atomic_add_int(&oid->oid_running, 1);
157 
158 	if (tracker != NULL)
159 		SYSCTL_RUNLOCK(tracker);
160 	else
161 		SYSCTL_WUNLOCK();
162 
163 	if (!(oid->oid_kind & CTLFLAG_MPSAFE))
164 		mtx_lock(&Giant);
165 	error = oid->oid_handler(oid, arg1, arg2, req);
166 	if (!(oid->oid_kind & CTLFLAG_MPSAFE))
167 		mtx_unlock(&Giant);
168 
169 	KFAIL_POINT_ERROR(_debug_fail_point, sysctl_running, error);
170 
171 	if (tracker != NULL)
172 		SYSCTL_RLOCK(tracker);
173 	else
174 		SYSCTL_WLOCK();
175 
176 	if (oid->oid_kind & CTLFLAG_DYN) {
177 		if (atomic_fetchadd_int(&oid->oid_running, -1) == 1 &&
178 		    (oid->oid_kind & CTLFLAG_DYING) != 0)
179 			wakeup(&oid->oid_running);
180 	}
181 
182 	return (error);
183 }
184 
185 static void
186 sysctl_load_tunable_by_oid_locked(struct sysctl_oid *oidp)
187 {
188 	struct sysctl_req req;
189 	struct sysctl_oid *curr;
190 	char *penv = NULL;
191 	char path[64];
192 	ssize_t rem = sizeof(path);
193 	ssize_t len;
194 	uint8_t val_8;
195 	uint16_t val_16;
196 	uint32_t val_32;
197 	int val_int;
198 	long val_long;
199 	int64_t val_64;
200 	quad_t val_quad;
201 	int error;
202 
203 	path[--rem] = 0;
204 
205 	for (curr = oidp; curr != NULL; curr = SYSCTL_PARENT(curr)) {
206 		len = strlen(curr->oid_name);
207 		rem -= len;
208 		if (curr != oidp)
209 			rem -= 1;
210 		if (rem < 0) {
211 			printf("OID path exceeds %d bytes\n", (int)sizeof(path));
212 			return;
213 		}
214 		memcpy(path + rem, curr->oid_name, len);
215 		if (curr != oidp)
216 			path[rem + len] = '.';
217 	}
218 
219 	memset(&req, 0, sizeof(req));
220 
221 	req.td = curthread;
222 	req.oldfunc = sysctl_old_kernel;
223 	req.newfunc = sysctl_new_kernel;
224 	req.lock = REQ_UNWIRED;
225 
226 	switch (oidp->oid_kind & CTLTYPE) {
227 	case CTLTYPE_INT:
228 		if (getenv_int(path + rem, &val_int) == 0)
229 			return;
230 		req.newlen = sizeof(val_int);
231 		req.newptr = &val_int;
232 		break;
233 	case CTLTYPE_UINT:
234 		if (getenv_uint(path + rem, (unsigned int *)&val_int) == 0)
235 			return;
236 		req.newlen = sizeof(val_int);
237 		req.newptr = &val_int;
238 		break;
239 	case CTLTYPE_LONG:
240 		if (getenv_long(path + rem, &val_long) == 0)
241 			return;
242 		req.newlen = sizeof(val_long);
243 		req.newptr = &val_long;
244 		break;
245 	case CTLTYPE_ULONG:
246 		if (getenv_ulong(path + rem, (unsigned long *)&val_long) == 0)
247 			return;
248 		req.newlen = sizeof(val_long);
249 		req.newptr = &val_long;
250 		break;
251 	case CTLTYPE_S8:
252 		if (getenv_int(path + rem, &val_int) == 0)
253 			return;
254 		val_8 = val_int;
255 		req.newlen = sizeof(val_8);
256 		req.newptr = &val_8;
257 		break;
258 	case CTLTYPE_S16:
259 		if (getenv_int(path + rem, &val_int) == 0)
260 			return;
261 		val_16 = val_int;
262 		req.newlen = sizeof(val_16);
263 		req.newptr = &val_16;
264 		break;
265 	case CTLTYPE_S32:
266 		if (getenv_long(path + rem, &val_long) == 0)
267 			return;
268 		val_32 = val_long;
269 		req.newlen = sizeof(val_32);
270 		req.newptr = &val_32;
271 		break;
272 	case CTLTYPE_S64:
273 		if (getenv_quad(path + rem, &val_quad) == 0)
274 			return;
275 		val_64 = val_quad;
276 		req.newlen = sizeof(val_64);
277 		req.newptr = &val_64;
278 		break;
279 	case CTLTYPE_U8:
280 		if (getenv_uint(path + rem, (unsigned int *)&val_int) == 0)
281 			return;
282 		val_8 = val_int;
283 		req.newlen = sizeof(val_8);
284 		req.newptr = &val_8;
285 		break;
286 	case CTLTYPE_U16:
287 		if (getenv_uint(path + rem, (unsigned int *)&val_int) == 0)
288 			return;
289 		val_16 = val_int;
290 		req.newlen = sizeof(val_16);
291 		req.newptr = &val_16;
292 		break;
293 	case CTLTYPE_U32:
294 		if (getenv_ulong(path + rem, (unsigned long *)&val_long) == 0)
295 			return;
296 		val_32 = val_long;
297 		req.newlen = sizeof(val_32);
298 		req.newptr = &val_32;
299 		break;
300 	case CTLTYPE_U64:
301 		/* XXX there is no getenv_uquad() */
302 		if (getenv_quad(path + rem, &val_quad) == 0)
303 			return;
304 		val_64 = val_quad;
305 		req.newlen = sizeof(val_64);
306 		req.newptr = &val_64;
307 		break;
308 	case CTLTYPE_STRING:
309 		penv = kern_getenv(path + rem);
310 		if (penv == NULL)
311 			return;
312 		req.newlen = strlen(penv);
313 		req.newptr = penv;
314 		break;
315 	default:
316 		return;
317 	}
318 	error = sysctl_root_handler_locked(oidp, oidp->oid_arg1,
319 	    oidp->oid_arg2, &req, NULL);
320 	if (error != 0)
321 		printf("Setting sysctl %s failed: %d\n", path + rem, error);
322 	if (penv != NULL)
323 		freeenv(penv);
324 }
325 
326 static int
327 sbuf_printf_drain(void *arg __unused, const char *data, int len)
328 {
329 
330 	return (printf("%.*s", len, data));
331 }
332 
333 /*
334  * Locate the path to a given oid.  Returns the length of the resulting path,
335  * or -1 if the oid was not found.  nodes must have room for CTL_MAXNAME
336  * elements and be NULL initialized.
337  */
338 static int
339 sysctl_search_oid(struct sysctl_oid **nodes, struct sysctl_oid *needle)
340 {
341 	int indx;
342 
343 	SYSCTL_ASSERT_LOCKED();
344 	indx = 0;
345 	while (indx < CTL_MAXNAME && indx >= 0) {
346 		if (nodes[indx] == NULL && indx == 0)
347 			nodes[indx] = SLIST_FIRST(&sysctl__children);
348 		else if (nodes[indx] == NULL)
349 			nodes[indx] = SLIST_FIRST(&nodes[indx - 1]->oid_children);
350 		else
351 			nodes[indx] = SLIST_NEXT(nodes[indx], oid_link);
352 
353 		if (nodes[indx] == needle)
354 			return (indx + 1);
355 
356 		if (nodes[indx] == NULL) {
357 			indx--;
358 			continue;
359 		}
360 
361 		if ((nodes[indx]->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
362 			indx++;
363 			continue;
364 		}
365 	}
366 	return (-1);
367 }
368 
369 static void
370 sysctl_warn_reuse(const char *func, struct sysctl_oid *leaf)
371 {
372 	struct sysctl_oid *nodes[CTL_MAXNAME];
373 	char buf[128];
374 	struct sbuf sb;
375 	int rc, i;
376 
377 	(void)sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN | SBUF_INCLUDENUL);
378 	sbuf_set_drain(&sb, sbuf_printf_drain, NULL);
379 
380 	sbuf_printf(&sb, "%s: can't re-use a leaf (", __func__);
381 
382 	memset(nodes, 0, sizeof(nodes));
383 	rc = sysctl_search_oid(nodes, leaf);
384 	if (rc > 0) {
385 		for (i = 0; i < rc; i++)
386 			sbuf_printf(&sb, "%s%.*s", nodes[i]->oid_name,
387 			    i != (rc - 1), ".");
388 	} else {
389 		sbuf_printf(&sb, "%s", leaf->oid_name);
390 	}
391 	sbuf_printf(&sb, ")!\n");
392 
393 	(void)sbuf_finish(&sb);
394 }
395 
396 #ifdef SYSCTL_DEBUG
397 static int
398 sysctl_reuse_test(SYSCTL_HANDLER_ARGS)
399 {
400 	struct rm_priotracker tracker;
401 
402 	SYSCTL_RLOCK(&tracker);
403 	sysctl_warn_reuse(__func__, oidp);
404 	SYSCTL_RUNLOCK(&tracker);
405 	return (0);
406 }
407 SYSCTL_PROC(_sysctl, 0, reuse_test, CTLTYPE_STRING|CTLFLAG_RD|CTLFLAG_MPSAFE,
408 	0, 0, sysctl_reuse_test, "-", "");
409 #endif
410 
411 void
412 sysctl_register_oid(struct sysctl_oid *oidp)
413 {
414 	struct sysctl_oid_list *parent = oidp->oid_parent;
415 	struct sysctl_oid *p;
416 	struct sysctl_oid *q;
417 	int oid_number;
418 	int timeout = 2;
419 
420 	/*
421 	 * First check if another oid with the same name already
422 	 * exists in the parent's list.
423 	 */
424 	SYSCTL_ASSERT_WLOCKED();
425 	p = sysctl_find_oidname(oidp->oid_name, parent);
426 	if (p != NULL) {
427 		if ((p->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
428 			p->oid_refcnt++;
429 			return;
430 		} else {
431 			sysctl_warn_reuse(__func__, p);
432 			return;
433 		}
434 	}
435 	/* get current OID number */
436 	oid_number = oidp->oid_number;
437 
438 #if (OID_AUTO >= 0)
439 #error "OID_AUTO is expected to be a negative value"
440 #endif
441 	/*
442 	 * Any negative OID number qualifies as OID_AUTO. Valid OID
443 	 * numbers should always be positive.
444 	 *
445 	 * NOTE: DO NOT change the starting value here, change it in
446 	 * <sys/sysctl.h>, and make sure it is at least 256 to
447 	 * accommodate e.g. net.inet.raw as a static sysctl node.
448 	 */
449 	if (oid_number < 0) {
450 		static int newoid;
451 
452 		/*
453 		 * By decrementing the next OID number we spend less
454 		 * time inserting the OIDs into a sorted list.
455 		 */
456 		if (--newoid < CTL_AUTO_START)
457 			newoid = 0x7fffffff;
458 
459 		oid_number = newoid;
460 	}
461 
462 	/*
463 	 * Insert the OID into the parent's list sorted by OID number.
464 	 */
465 retry:
466 	q = NULL;
467 	SLIST_FOREACH(p, parent, oid_link) {
468 		/* check if the current OID number is in use */
469 		if (oid_number == p->oid_number) {
470 			/* get the next valid OID number */
471 			if (oid_number < CTL_AUTO_START ||
472 			    oid_number == 0x7fffffff) {
473 				/* wraparound - restart */
474 				oid_number = CTL_AUTO_START;
475 				/* don't loop forever */
476 				if (!timeout--)
477 					panic("sysctl: Out of OID numbers\n");
478 				goto retry;
479 			} else {
480 				oid_number++;
481 			}
482 		} else if (oid_number < p->oid_number)
483 			break;
484 		q = p;
485 	}
486 	/* check for non-auto OID number collision */
487 	if (oidp->oid_number >= 0 && oidp->oid_number < CTL_AUTO_START &&
488 	    oid_number >= CTL_AUTO_START) {
489 		printf("sysctl: OID number(%d) is already in use for '%s'\n",
490 		    oidp->oid_number, oidp->oid_name);
491 	}
492 	/* update the OID number, if any */
493 	oidp->oid_number = oid_number;
494 	if (q != NULL)
495 		SLIST_INSERT_AFTER(q, oidp, oid_link);
496 	else
497 		SLIST_INSERT_HEAD(parent, oidp, oid_link);
498 
499 	if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE &&
500 #ifdef VIMAGE
501 	    (oidp->oid_kind & CTLFLAG_VNET) == 0 &&
502 #endif
503 	    (oidp->oid_kind & CTLFLAG_TUN) != 0 &&
504 	    (oidp->oid_kind & CTLFLAG_NOFETCH) == 0) {
505 		/* only fetch value once */
506 		oidp->oid_kind |= CTLFLAG_NOFETCH;
507 		/* try to fetch value from kernel environment */
508 		sysctl_load_tunable_by_oid_locked(oidp);
509 	}
510 }
511 
512 void
513 sysctl_register_disabled_oid(struct sysctl_oid *oidp)
514 {
515 
516 	/*
517 	 * Mark the leaf as dormant if it's not to be immediately enabled.
518 	 * We do not disable nodes as they can be shared between modules
519 	 * and it is always safe to access a node.
520 	 */
521 	KASSERT((oidp->oid_kind & CTLFLAG_DORMANT) == 0,
522 	    ("internal flag is set in oid_kind"));
523 	if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
524 		oidp->oid_kind |= CTLFLAG_DORMANT;
525 	sysctl_register_oid(oidp);
526 }
527 
528 void
529 sysctl_enable_oid(struct sysctl_oid *oidp)
530 {
531 
532 	SYSCTL_ASSERT_WLOCKED();
533 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
534 		KASSERT((oidp->oid_kind & CTLFLAG_DORMANT) == 0,
535 		    ("sysctl node is marked as dormant"));
536 		return;
537 	}
538 	KASSERT((oidp->oid_kind & CTLFLAG_DORMANT) != 0,
539 	    ("enabling already enabled sysctl oid"));
540 	oidp->oid_kind &= ~CTLFLAG_DORMANT;
541 }
542 
543 void
544 sysctl_unregister_oid(struct sysctl_oid *oidp)
545 {
546 	struct sysctl_oid *p;
547 	int error;
548 
549 	SYSCTL_ASSERT_WLOCKED();
550 	error = ENOENT;
551 	if (oidp->oid_number == OID_AUTO) {
552 		error = EINVAL;
553 	} else {
554 		SLIST_FOREACH(p, oidp->oid_parent, oid_link) {
555 			if (p == oidp) {
556 				SLIST_REMOVE(oidp->oid_parent, oidp,
557 				    sysctl_oid, oid_link);
558 				error = 0;
559 				break;
560 			}
561 		}
562 	}
563 
564 	/*
565 	 * This can happen when a module fails to register and is
566 	 * being unloaded afterwards.  It should not be a panic()
567 	 * for normal use.
568 	 */
569 	if (error)
570 		printf("%s: failed to unregister sysctl\n", __func__);
571 }
572 
573 /* Initialize a new context to keep track of dynamically added sysctls. */
574 int
575 sysctl_ctx_init(struct sysctl_ctx_list *c)
576 {
577 
578 	if (c == NULL) {
579 		return (EINVAL);
580 	}
581 
582 	/*
583 	 * No locking here, the caller is responsible for not adding
584 	 * new nodes to a context until after this function has
585 	 * returned.
586 	 */
587 	TAILQ_INIT(c);
588 	return (0);
589 }
590 
591 /* Free the context, and destroy all dynamic oids registered in this context */
592 int
593 sysctl_ctx_free(struct sysctl_ctx_list *clist)
594 {
595 	struct sysctl_ctx_entry *e, *e1;
596 	int error;
597 
598 	error = 0;
599 	/*
600 	 * First perform a "dry run" to check if it's ok to remove oids.
601 	 * XXX FIXME
602 	 * XXX This algorithm is a hack. But I don't know any
603 	 * XXX better solution for now...
604 	 */
605 	SYSCTL_WLOCK();
606 	TAILQ_FOREACH(e, clist, link) {
607 		error = sysctl_remove_oid_locked(e->entry, 0, 0);
608 		if (error)
609 			break;
610 	}
611 	/*
612 	 * Restore deregistered entries, either from the end,
613 	 * or from the place where error occurred.
614 	 * e contains the entry that was not unregistered
615 	 */
616 	if (error)
617 		e1 = TAILQ_PREV(e, sysctl_ctx_list, link);
618 	else
619 		e1 = TAILQ_LAST(clist, sysctl_ctx_list);
620 	while (e1 != NULL) {
621 		sysctl_register_oid(e1->entry);
622 		e1 = TAILQ_PREV(e1, sysctl_ctx_list, link);
623 	}
624 	if (error) {
625 		SYSCTL_WUNLOCK();
626 		return(EBUSY);
627 	}
628 	/* Now really delete the entries */
629 	e = TAILQ_FIRST(clist);
630 	while (e != NULL) {
631 		e1 = TAILQ_NEXT(e, link);
632 		error = sysctl_remove_oid_locked(e->entry, 1, 0);
633 		if (error)
634 			panic("sysctl_remove_oid: corrupt tree, entry: %s",
635 			    e->entry->oid_name);
636 		free(e, M_SYSCTLOID);
637 		e = e1;
638 	}
639 	SYSCTL_WUNLOCK();
640 	return (error);
641 }
642 
643 /* Add an entry to the context */
644 struct sysctl_ctx_entry *
645 sysctl_ctx_entry_add(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
646 {
647 	struct sysctl_ctx_entry *e;
648 
649 	SYSCTL_ASSERT_WLOCKED();
650 	if (clist == NULL || oidp == NULL)
651 		return(NULL);
652 	e = malloc(sizeof(struct sysctl_ctx_entry), M_SYSCTLOID, M_WAITOK);
653 	e->entry = oidp;
654 	TAILQ_INSERT_HEAD(clist, e, link);
655 	return (e);
656 }
657 
658 /* Find an entry in the context */
659 struct sysctl_ctx_entry *
660 sysctl_ctx_entry_find(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
661 {
662 	struct sysctl_ctx_entry *e;
663 
664 	SYSCTL_ASSERT_WLOCKED();
665 	if (clist == NULL || oidp == NULL)
666 		return(NULL);
667 	TAILQ_FOREACH(e, clist, link) {
668 		if(e->entry == oidp)
669 			return(e);
670 	}
671 	return (e);
672 }
673 
674 /*
675  * Delete an entry from the context.
676  * NOTE: this function doesn't free oidp! You have to remove it
677  * with sysctl_remove_oid().
678  */
679 int
680 sysctl_ctx_entry_del(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
681 {
682 	struct sysctl_ctx_entry *e;
683 
684 	if (clist == NULL || oidp == NULL)
685 		return (EINVAL);
686 	SYSCTL_WLOCK();
687 	e = sysctl_ctx_entry_find(clist, oidp);
688 	if (e != NULL) {
689 		TAILQ_REMOVE(clist, e, link);
690 		SYSCTL_WUNLOCK();
691 		free(e, M_SYSCTLOID);
692 		return (0);
693 	} else {
694 		SYSCTL_WUNLOCK();
695 		return (ENOENT);
696 	}
697 }
698 
699 /*
700  * Remove dynamically created sysctl trees.
701  * oidp - top of the tree to be removed
702  * del - if 0 - just deregister, otherwise free up entries as well
703  * recurse - if != 0 traverse the subtree to be deleted
704  */
705 int
706 sysctl_remove_oid(struct sysctl_oid *oidp, int del, int recurse)
707 {
708 	int error;
709 
710 	SYSCTL_WLOCK();
711 	error = sysctl_remove_oid_locked(oidp, del, recurse);
712 	SYSCTL_WUNLOCK();
713 	return (error);
714 }
715 
716 int
717 sysctl_remove_name(struct sysctl_oid *parent, const char *name,
718     int del, int recurse)
719 {
720 	struct sysctl_oid *p, *tmp;
721 	int error;
722 
723 	error = ENOENT;
724 	SYSCTL_WLOCK();
725 	SLIST_FOREACH_SAFE(p, SYSCTL_CHILDREN(parent), oid_link, tmp) {
726 		if (strcmp(p->oid_name, name) == 0) {
727 			error = sysctl_remove_oid_locked(p, del, recurse);
728 			break;
729 		}
730 	}
731 	SYSCTL_WUNLOCK();
732 
733 	return (error);
734 }
735 
736 
737 static int
738 sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del, int recurse)
739 {
740 	struct sysctl_oid *p, *tmp;
741 	int error;
742 
743 	SYSCTL_ASSERT_WLOCKED();
744 	if (oidp == NULL)
745 		return(EINVAL);
746 	if ((oidp->oid_kind & CTLFLAG_DYN) == 0) {
747 		printf("Warning: can't remove non-dynamic nodes (%s)!\n",
748 		    oidp->oid_name);
749 		return (EINVAL);
750 	}
751 	/*
752 	 * WARNING: normal method to do this should be through
753 	 * sysctl_ctx_free(). Use recursing as the last resort
754 	 * method to purge your sysctl tree of leftovers...
755 	 * However, if some other code still references these nodes,
756 	 * it will panic.
757 	 */
758 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
759 		if (oidp->oid_refcnt == 1) {
760 			SLIST_FOREACH_SAFE(p,
761 			    SYSCTL_CHILDREN(oidp), oid_link, tmp) {
762 				if (!recurse) {
763 					printf("Warning: failed attempt to "
764 					    "remove oid %s with child %s\n",
765 					    oidp->oid_name, p->oid_name);
766 					return (ENOTEMPTY);
767 				}
768 				error = sysctl_remove_oid_locked(p, del,
769 				    recurse);
770 				if (error)
771 					return (error);
772 			}
773 		}
774 	}
775 	if (oidp->oid_refcnt > 1 ) {
776 		oidp->oid_refcnt--;
777 	} else {
778 		if (oidp->oid_refcnt == 0) {
779 			printf("Warning: bad oid_refcnt=%u (%s)!\n",
780 				oidp->oid_refcnt, oidp->oid_name);
781 			return (EINVAL);
782 		}
783 		sysctl_unregister_oid(oidp);
784 		if (del) {
785 			/*
786 			 * Wait for all threads running the handler to drain.
787 			 * This preserves the previous behavior when the
788 			 * sysctl lock was held across a handler invocation,
789 			 * and is necessary for module unload correctness.
790 			 */
791 			while (oidp->oid_running > 0) {
792 				oidp->oid_kind |= CTLFLAG_DYING;
793 				SYSCTL_SLEEP(&oidp->oid_running, "oidrm", 0);
794 			}
795 			if (oidp->oid_descr)
796 				free(__DECONST(char *, oidp->oid_descr),
797 				    M_SYSCTLOID);
798 			if (oidp->oid_label)
799 				free(__DECONST(char *, oidp->oid_label),
800 				    M_SYSCTLOID);
801 			free(__DECONST(char *, oidp->oid_name), M_SYSCTLOID);
802 			free(oidp, M_SYSCTLOID);
803 		}
804 	}
805 	return (0);
806 }
807 /*
808  * Create new sysctls at run time.
809  * clist may point to a valid context initialized with sysctl_ctx_init().
810  */
811 struct sysctl_oid *
812 sysctl_add_oid(struct sysctl_ctx_list *clist, struct sysctl_oid_list *parent,
813 	int number, const char *name, int kind, void *arg1, intmax_t arg2,
814 	int (*handler)(SYSCTL_HANDLER_ARGS), const char *fmt, const char *descr,
815 	const char *label)
816 {
817 	struct sysctl_oid *oidp;
818 
819 	/* You have to hook up somewhere.. */
820 	if (parent == NULL)
821 		return(NULL);
822 	/* Check if the node already exists, otherwise create it */
823 	SYSCTL_WLOCK();
824 	oidp = sysctl_find_oidname(name, parent);
825 	if (oidp != NULL) {
826 		if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
827 			oidp->oid_refcnt++;
828 			/* Update the context */
829 			if (clist != NULL)
830 				sysctl_ctx_entry_add(clist, oidp);
831 			SYSCTL_WUNLOCK();
832 			return (oidp);
833 		} else {
834 			sysctl_warn_reuse(__func__, oidp);
835 			SYSCTL_WUNLOCK();
836 			return (NULL);
837 		}
838 	}
839 	oidp = malloc(sizeof(struct sysctl_oid), M_SYSCTLOID, M_WAITOK|M_ZERO);
840 	oidp->oid_parent = parent;
841 	SLIST_INIT(&oidp->oid_children);
842 	oidp->oid_number = number;
843 	oidp->oid_refcnt = 1;
844 	oidp->oid_name = strdup(name, M_SYSCTLOID);
845 	oidp->oid_handler = handler;
846 	oidp->oid_kind = CTLFLAG_DYN | kind;
847 	oidp->oid_arg1 = arg1;
848 	oidp->oid_arg2 = arg2;
849 	oidp->oid_fmt = fmt;
850 	if (descr != NULL)
851 		oidp->oid_descr = strdup(descr, M_SYSCTLOID);
852 	if (label != NULL)
853 		oidp->oid_label = strdup(label, M_SYSCTLOID);
854 	/* Update the context, if used */
855 	if (clist != NULL)
856 		sysctl_ctx_entry_add(clist, oidp);
857 	/* Register this oid */
858 	sysctl_register_oid(oidp);
859 	SYSCTL_WUNLOCK();
860 	return (oidp);
861 }
862 
863 /*
864  * Rename an existing oid.
865  */
866 void
867 sysctl_rename_oid(struct sysctl_oid *oidp, const char *name)
868 {
869 	char *newname;
870 	char *oldname;
871 
872 	newname = strdup(name, M_SYSCTLOID);
873 	SYSCTL_WLOCK();
874 	oldname = __DECONST(char *, oidp->oid_name);
875 	oidp->oid_name = newname;
876 	SYSCTL_WUNLOCK();
877 	free(oldname, M_SYSCTLOID);
878 }
879 
880 /*
881  * Reparent an existing oid.
882  */
883 int
884 sysctl_move_oid(struct sysctl_oid *oid, struct sysctl_oid_list *parent)
885 {
886 	struct sysctl_oid *oidp;
887 
888 	SYSCTL_WLOCK();
889 	if (oid->oid_parent == parent) {
890 		SYSCTL_WUNLOCK();
891 		return (0);
892 	}
893 	oidp = sysctl_find_oidname(oid->oid_name, parent);
894 	if (oidp != NULL) {
895 		SYSCTL_WUNLOCK();
896 		return (EEXIST);
897 	}
898 	sysctl_unregister_oid(oid);
899 	oid->oid_parent = parent;
900 	oid->oid_number = OID_AUTO;
901 	sysctl_register_oid(oid);
902 	SYSCTL_WUNLOCK();
903 	return (0);
904 }
905 
906 /*
907  * Register the kernel's oids on startup.
908  */
909 SET_DECLARE(sysctl_set, struct sysctl_oid);
910 
911 static void
912 sysctl_register_all(void *arg)
913 {
914 	struct sysctl_oid **oidp;
915 
916 	sx_init(&sysctlmemlock, "sysctl mem");
917 	SYSCTL_INIT();
918 	SYSCTL_WLOCK();
919 	SET_FOREACH(oidp, sysctl_set)
920 		sysctl_register_oid(*oidp);
921 	SYSCTL_WUNLOCK();
922 }
923 SYSINIT(sysctl, SI_SUB_KMEM, SI_ORDER_FIRST, sysctl_register_all, 0);
924 
925 /*
926  * "Staff-functions"
927  *
928  * These functions implement a presently undocumented interface
929  * used by the sysctl program to walk the tree, and get the type
930  * so it can print the value.
931  * This interface is under work and consideration, and should probably
932  * be killed with a big axe by the first person who can find the time.
933  * (be aware though, that the proper interface isn't as obvious as it
934  * may seem, there are various conflicting requirements.
935  *
936  * {0,0}	printf the entire MIB-tree.
937  * {0,1,...}	return the name of the "..." OID.
938  * {0,2,...}	return the next OID.
939  * {0,3}	return the OID of the name in "new"
940  * {0,4,...}	return the kind & format info for the "..." OID.
941  * {0,5,...}	return the description of the "..." OID.
942  * {0,6,...}	return the aggregation label of the "..." OID.
943  */
944 
945 #ifdef SYSCTL_DEBUG
946 static void
947 sysctl_sysctl_debug_dump_node(struct sysctl_oid_list *l, int i)
948 {
949 	int k;
950 	struct sysctl_oid *oidp;
951 
952 	SYSCTL_ASSERT_LOCKED();
953 	SLIST_FOREACH(oidp, l, oid_link) {
954 
955 		for (k=0; k<i; k++)
956 			printf(" ");
957 
958 		printf("%d %s ", oidp->oid_number, oidp->oid_name);
959 
960 		printf("%c%c",
961 			oidp->oid_kind & CTLFLAG_RD ? 'R':' ',
962 			oidp->oid_kind & CTLFLAG_WR ? 'W':' ');
963 
964 		if (oidp->oid_handler)
965 			printf(" *Handler");
966 
967 		switch (oidp->oid_kind & CTLTYPE) {
968 			case CTLTYPE_NODE:
969 				printf(" Node\n");
970 				if (!oidp->oid_handler) {
971 					sysctl_sysctl_debug_dump_node(
972 					    SYSCTL_CHILDREN(oidp), i + 2);
973 				}
974 				break;
975 			case CTLTYPE_INT:    printf(" Int\n"); break;
976 			case CTLTYPE_UINT:   printf(" u_int\n"); break;
977 			case CTLTYPE_LONG:   printf(" Long\n"); break;
978 			case CTLTYPE_ULONG:  printf(" u_long\n"); break;
979 			case CTLTYPE_STRING: printf(" String\n"); break;
980 			case CTLTYPE_S8:     printf(" int8_t\n"); break;
981 			case CTLTYPE_S16:    printf(" int16_t\n"); break;
982 			case CTLTYPE_S32:    printf(" int32_t\n"); break;
983 			case CTLTYPE_S64:    printf(" int64_t\n"); break;
984 			case CTLTYPE_U8:     printf(" uint8_t\n"); break;
985 			case CTLTYPE_U16:    printf(" uint16_t\n"); break;
986 			case CTLTYPE_U32:    printf(" uint32_t\n"); break;
987 			case CTLTYPE_U64:    printf(" uint64_t\n"); break;
988 			case CTLTYPE_OPAQUE: printf(" Opaque/struct\n"); break;
989 			default:	     printf("\n");
990 		}
991 
992 	}
993 }
994 
995 static int
996 sysctl_sysctl_debug(SYSCTL_HANDLER_ARGS)
997 {
998 	struct rm_priotracker tracker;
999 	int error;
1000 
1001 	error = priv_check(req->td, PRIV_SYSCTL_DEBUG);
1002 	if (error)
1003 		return (error);
1004 	SYSCTL_RLOCK(&tracker);
1005 	sysctl_sysctl_debug_dump_node(&sysctl__children, 0);
1006 	SYSCTL_RUNLOCK(&tracker);
1007 	return (ENOENT);
1008 }
1009 
1010 SYSCTL_PROC(_sysctl, 0, debug, CTLTYPE_STRING|CTLFLAG_RD|CTLFLAG_MPSAFE,
1011 	0, 0, sysctl_sysctl_debug, "-", "");
1012 #endif
1013 
1014 static int
1015 sysctl_sysctl_name(SYSCTL_HANDLER_ARGS)
1016 {
1017 	int *name = (int *) arg1;
1018 	u_int namelen = arg2;
1019 	int error = 0;
1020 	struct sysctl_oid *oid;
1021 	struct sysctl_oid_list *lsp = &sysctl__children, *lsp2;
1022 	struct rm_priotracker tracker;
1023 	char buf[10];
1024 
1025 	SYSCTL_RLOCK(&tracker);
1026 	while (namelen) {
1027 		if (!lsp) {
1028 			snprintf(buf,sizeof(buf),"%d",*name);
1029 			if (req->oldidx)
1030 				error = SYSCTL_OUT(req, ".", 1);
1031 			if (!error)
1032 				error = SYSCTL_OUT(req, buf, strlen(buf));
1033 			if (error)
1034 				goto out;
1035 			namelen--;
1036 			name++;
1037 			continue;
1038 		}
1039 		lsp2 = NULL;
1040 		SLIST_FOREACH(oid, lsp, oid_link) {
1041 			if (oid->oid_number != *name)
1042 				continue;
1043 
1044 			if (req->oldidx)
1045 				error = SYSCTL_OUT(req, ".", 1);
1046 			if (!error)
1047 				error = SYSCTL_OUT(req, oid->oid_name,
1048 					strlen(oid->oid_name));
1049 			if (error)
1050 				goto out;
1051 
1052 			namelen--;
1053 			name++;
1054 
1055 			if ((oid->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1056 				break;
1057 
1058 			if (oid->oid_handler)
1059 				break;
1060 
1061 			lsp2 = SYSCTL_CHILDREN(oid);
1062 			break;
1063 		}
1064 		lsp = lsp2;
1065 	}
1066 	error = SYSCTL_OUT(req, "", 1);
1067  out:
1068 	SYSCTL_RUNLOCK(&tracker);
1069 	return (error);
1070 }
1071 
1072 /*
1073  * XXXRW/JA: Shouldn't return name data for nodes that we don't permit in
1074  * capability mode.
1075  */
1076 static SYSCTL_NODE(_sysctl, 1, name, CTLFLAG_RD | CTLFLAG_MPSAFE | CTLFLAG_CAPRD,
1077     sysctl_sysctl_name, "");
1078 
1079 static int
1080 sysctl_sysctl_next_ls(struct sysctl_oid_list *lsp, int *name, u_int namelen,
1081 	int *next, int *len, int level, struct sysctl_oid **oidpp)
1082 {
1083 	struct sysctl_oid *oidp;
1084 
1085 	SYSCTL_ASSERT_LOCKED();
1086 	*len = level;
1087 	SLIST_FOREACH(oidp, lsp, oid_link) {
1088 		*next = oidp->oid_number;
1089 		*oidpp = oidp;
1090 
1091 		if ((oidp->oid_kind & (CTLFLAG_SKIP | CTLFLAG_DORMANT)) != 0)
1092 			continue;
1093 
1094 		if (!namelen) {
1095 			if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1096 				return (0);
1097 			if (oidp->oid_handler)
1098 				/* We really should call the handler here...*/
1099 				return (0);
1100 			lsp = SYSCTL_CHILDREN(oidp);
1101 			if (!sysctl_sysctl_next_ls(lsp, 0, 0, next+1,
1102 				len, level+1, oidpp))
1103 				return (0);
1104 			goto emptynode;
1105 		}
1106 
1107 		if (oidp->oid_number < *name)
1108 			continue;
1109 
1110 		if (oidp->oid_number > *name) {
1111 			if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1112 				return (0);
1113 			if (oidp->oid_handler)
1114 				return (0);
1115 			lsp = SYSCTL_CHILDREN(oidp);
1116 			if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1,
1117 				next+1, len, level+1, oidpp))
1118 				return (0);
1119 			goto next;
1120 		}
1121 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1122 			continue;
1123 
1124 		if (oidp->oid_handler)
1125 			continue;
1126 
1127 		lsp = SYSCTL_CHILDREN(oidp);
1128 		if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1, next+1,
1129 			len, level+1, oidpp))
1130 			return (0);
1131 	next:
1132 		namelen = 1;
1133 	emptynode:
1134 		*len = level;
1135 	}
1136 	return (1);
1137 }
1138 
1139 static int
1140 sysctl_sysctl_next(SYSCTL_HANDLER_ARGS)
1141 {
1142 	int *name = (int *) arg1;
1143 	u_int namelen = arg2;
1144 	int i, j, error;
1145 	struct sysctl_oid *oid;
1146 	struct sysctl_oid_list *lsp = &sysctl__children;
1147 	struct rm_priotracker tracker;
1148 	int newoid[CTL_MAXNAME];
1149 
1150 	SYSCTL_RLOCK(&tracker);
1151 	i = sysctl_sysctl_next_ls(lsp, name, namelen, newoid, &j, 1, &oid);
1152 	SYSCTL_RUNLOCK(&tracker);
1153 	if (i)
1154 		return (ENOENT);
1155 	error = SYSCTL_OUT(req, newoid, j * sizeof (int));
1156 	return (error);
1157 }
1158 
1159 /*
1160  * XXXRW/JA: Shouldn't return next data for nodes that we don't permit in
1161  * capability mode.
1162  */
1163 static SYSCTL_NODE(_sysctl, 2, next, CTLFLAG_RD | CTLFLAG_MPSAFE | CTLFLAG_CAPRD,
1164     sysctl_sysctl_next, "");
1165 
1166 static int
1167 name2oid(char *name, int *oid, int *len, struct sysctl_oid **oidpp)
1168 {
1169 	struct sysctl_oid *oidp;
1170 	struct sysctl_oid_list *lsp = &sysctl__children;
1171 	char *p;
1172 
1173 	SYSCTL_ASSERT_LOCKED();
1174 
1175 	for (*len = 0; *len < CTL_MAXNAME;) {
1176 		p = strsep(&name, ".");
1177 
1178 		oidp = SLIST_FIRST(lsp);
1179 		for (;; oidp = SLIST_NEXT(oidp, oid_link)) {
1180 			if (oidp == NULL)
1181 				return (ENOENT);
1182 			if (strcmp(p, oidp->oid_name) == 0)
1183 				break;
1184 		}
1185 		*oid++ = oidp->oid_number;
1186 		(*len)++;
1187 
1188 		if (name == NULL || *name == '\0') {
1189 			if (oidpp)
1190 				*oidpp = oidp;
1191 			return (0);
1192 		}
1193 
1194 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1195 			break;
1196 
1197 		if (oidp->oid_handler)
1198 			break;
1199 
1200 		lsp = SYSCTL_CHILDREN(oidp);
1201 	}
1202 	return (ENOENT);
1203 }
1204 
1205 static int
1206 sysctl_sysctl_name2oid(SYSCTL_HANDLER_ARGS)
1207 {
1208 	char *p;
1209 	int error, oid[CTL_MAXNAME], len = 0;
1210 	struct sysctl_oid *op = NULL;
1211 	struct rm_priotracker tracker;
1212 
1213 	if (!req->newlen)
1214 		return (ENOENT);
1215 	if (req->newlen >= MAXPATHLEN)	/* XXX arbitrary, undocumented */
1216 		return (ENAMETOOLONG);
1217 
1218 	p = malloc(req->newlen+1, M_SYSCTL, M_WAITOK);
1219 
1220 	error = SYSCTL_IN(req, p, req->newlen);
1221 	if (error) {
1222 		free(p, M_SYSCTL);
1223 		return (error);
1224 	}
1225 
1226 	p [req->newlen] = '\0';
1227 
1228 	SYSCTL_RLOCK(&tracker);
1229 	error = name2oid(p, oid, &len, &op);
1230 	SYSCTL_RUNLOCK(&tracker);
1231 
1232 	free(p, M_SYSCTL);
1233 
1234 	if (error)
1235 		return (error);
1236 
1237 	error = SYSCTL_OUT(req, oid, len * sizeof *oid);
1238 	return (error);
1239 }
1240 
1241 /*
1242  * XXXRW/JA: Shouldn't return name2oid data for nodes that we don't permit in
1243  * capability mode.
1244  */
1245 SYSCTL_PROC(_sysctl, 3, name2oid,
1246     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_MPSAFE
1247     | CTLFLAG_CAPRW, 0, 0, sysctl_sysctl_name2oid, "I", "");
1248 
1249 static int
1250 sysctl_sysctl_oidfmt(SYSCTL_HANDLER_ARGS)
1251 {
1252 	struct sysctl_oid *oid;
1253 	struct rm_priotracker tracker;
1254 	int error;
1255 
1256 	SYSCTL_RLOCK(&tracker);
1257 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1258 	if (error)
1259 		goto out;
1260 
1261 	if (oid->oid_fmt == NULL) {
1262 		error = ENOENT;
1263 		goto out;
1264 	}
1265 	error = SYSCTL_OUT(req, &oid->oid_kind, sizeof(oid->oid_kind));
1266 	if (error)
1267 		goto out;
1268 	error = SYSCTL_OUT(req, oid->oid_fmt, strlen(oid->oid_fmt) + 1);
1269  out:
1270 	SYSCTL_RUNLOCK(&tracker);
1271 	return (error);
1272 }
1273 
1274 
1275 static SYSCTL_NODE(_sysctl, 4, oidfmt, CTLFLAG_RD|CTLFLAG_MPSAFE|CTLFLAG_CAPRD,
1276     sysctl_sysctl_oidfmt, "");
1277 
1278 static int
1279 sysctl_sysctl_oiddescr(SYSCTL_HANDLER_ARGS)
1280 {
1281 	struct sysctl_oid *oid;
1282 	struct rm_priotracker tracker;
1283 	int error;
1284 
1285 	SYSCTL_RLOCK(&tracker);
1286 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1287 	if (error)
1288 		goto out;
1289 
1290 	if (oid->oid_descr == NULL) {
1291 		error = ENOENT;
1292 		goto out;
1293 	}
1294 	error = SYSCTL_OUT(req, oid->oid_descr, strlen(oid->oid_descr) + 1);
1295  out:
1296 	SYSCTL_RUNLOCK(&tracker);
1297 	return (error);
1298 }
1299 
1300 static SYSCTL_NODE(_sysctl, 5, oiddescr, CTLFLAG_RD|CTLFLAG_MPSAFE|CTLFLAG_CAPRD,
1301     sysctl_sysctl_oiddescr, "");
1302 
1303 static int
1304 sysctl_sysctl_oidlabel(SYSCTL_HANDLER_ARGS)
1305 {
1306 	struct sysctl_oid *oid;
1307 	struct rm_priotracker tracker;
1308 	int error;
1309 
1310 	SYSCTL_RLOCK(&tracker);
1311 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1312 	if (error)
1313 		goto out;
1314 
1315 	if (oid->oid_label == NULL) {
1316 		error = ENOENT;
1317 		goto out;
1318 	}
1319 	error = SYSCTL_OUT(req, oid->oid_label, strlen(oid->oid_label) + 1);
1320  out:
1321 	SYSCTL_RUNLOCK(&tracker);
1322 	return (error);
1323 }
1324 
1325 static SYSCTL_NODE(_sysctl, 6, oidlabel,
1326     CTLFLAG_RD | CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_oidlabel, "");
1327 
1328 /*
1329  * Default "handler" functions.
1330  */
1331 
1332 /*
1333  * Handle a bool.
1334  * Two cases:
1335  *     a variable:  point arg1 at it.
1336  *     a constant:  pass it in arg2.
1337  */
1338 
1339 int
1340 sysctl_handle_bool(SYSCTL_HANDLER_ARGS)
1341 {
1342 	uint8_t temp;
1343 	int error;
1344 
1345 	/*
1346 	 * Attempt to get a coherent snapshot by making a copy of the data.
1347 	 */
1348 	if (arg1)
1349 		temp = *(bool *)arg1 ? 1 : 0;
1350 	else
1351 		temp = arg2 ? 1 : 0;
1352 
1353 	error = SYSCTL_OUT(req, &temp, sizeof(temp));
1354 	if (error || !req->newptr)
1355 		return (error);
1356 
1357 	if (!arg1)
1358 		error = EPERM;
1359 	else {
1360 		error = SYSCTL_IN(req, &temp, sizeof(temp));
1361 		if (!error)
1362 			*(bool *)arg1 = temp ? 1 : 0;
1363 	}
1364 	return (error);
1365 }
1366 
1367 /*
1368  * Handle an int8_t, signed or unsigned.
1369  * Two cases:
1370  *     a variable:  point arg1 at it.
1371  *     a constant:  pass it in arg2.
1372  */
1373 
1374 int
1375 sysctl_handle_8(SYSCTL_HANDLER_ARGS)
1376 {
1377 	int8_t tmpout;
1378 	int error = 0;
1379 
1380 	/*
1381 	 * Attempt to get a coherent snapshot by making a copy of the data.
1382 	 */
1383 	if (arg1)
1384 		tmpout = *(int8_t *)arg1;
1385 	else
1386 		tmpout = arg2;
1387 	error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout));
1388 
1389 	if (error || !req->newptr)
1390 		return (error);
1391 
1392 	if (!arg1)
1393 		error = EPERM;
1394 	else
1395 		error = SYSCTL_IN(req, arg1, sizeof(tmpout));
1396 	return (error);
1397 }
1398 
1399 /*
1400  * Handle an int16_t, signed or unsigned.
1401  * Two cases:
1402  *     a variable:  point arg1 at it.
1403  *     a constant:  pass it in arg2.
1404  */
1405 
1406 int
1407 sysctl_handle_16(SYSCTL_HANDLER_ARGS)
1408 {
1409 	int16_t tmpout;
1410 	int error = 0;
1411 
1412 	/*
1413 	 * Attempt to get a coherent snapshot by making a copy of the data.
1414 	 */
1415 	if (arg1)
1416 		tmpout = *(int16_t *)arg1;
1417 	else
1418 		tmpout = arg2;
1419 	error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout));
1420 
1421 	if (error || !req->newptr)
1422 		return (error);
1423 
1424 	if (!arg1)
1425 		error = EPERM;
1426 	else
1427 		error = SYSCTL_IN(req, arg1, sizeof(tmpout));
1428 	return (error);
1429 }
1430 
1431 /*
1432  * Handle an int32_t, signed or unsigned.
1433  * Two cases:
1434  *     a variable:  point arg1 at it.
1435  *     a constant:  pass it in arg2.
1436  */
1437 
1438 int
1439 sysctl_handle_32(SYSCTL_HANDLER_ARGS)
1440 {
1441 	int32_t tmpout;
1442 	int error = 0;
1443 
1444 	/*
1445 	 * Attempt to get a coherent snapshot by making a copy of the data.
1446 	 */
1447 	if (arg1)
1448 		tmpout = *(int32_t *)arg1;
1449 	else
1450 		tmpout = arg2;
1451 	error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout));
1452 
1453 	if (error || !req->newptr)
1454 		return (error);
1455 
1456 	if (!arg1)
1457 		error = EPERM;
1458 	else
1459 		error = SYSCTL_IN(req, arg1, sizeof(tmpout));
1460 	return (error);
1461 }
1462 
1463 /*
1464  * Handle an int, signed or unsigned.
1465  * Two cases:
1466  *     a variable:  point arg1 at it.
1467  *     a constant:  pass it in arg2.
1468  */
1469 
1470 int
1471 sysctl_handle_int(SYSCTL_HANDLER_ARGS)
1472 {
1473 	int tmpout, error = 0;
1474 
1475 	/*
1476 	 * Attempt to get a coherent snapshot by making a copy of the data.
1477 	 */
1478 	if (arg1)
1479 		tmpout = *(int *)arg1;
1480 	else
1481 		tmpout = arg2;
1482 	error = SYSCTL_OUT(req, &tmpout, sizeof(int));
1483 
1484 	if (error || !req->newptr)
1485 		return (error);
1486 
1487 	if (!arg1)
1488 		error = EPERM;
1489 	else
1490 		error = SYSCTL_IN(req, arg1, sizeof(int));
1491 	return (error);
1492 }
1493 
1494 /*
1495  * Based on on sysctl_handle_int() convert milliseconds into ticks.
1496  * Note: this is used by TCP.
1497  */
1498 
1499 int
1500 sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
1501 {
1502 	int error, s, tt;
1503 
1504 	tt = *(int *)arg1;
1505 	s = (int)((int64_t)tt * 1000 / hz);
1506 
1507 	error = sysctl_handle_int(oidp, &s, 0, req);
1508 	if (error || !req->newptr)
1509 		return (error);
1510 
1511 	tt = (int)((int64_t)s * hz / 1000);
1512 	if (tt < 1)
1513 		return (EINVAL);
1514 
1515 	*(int *)arg1 = tt;
1516 	return (0);
1517 }
1518 
1519 
1520 /*
1521  * Handle a long, signed or unsigned.
1522  * Two cases:
1523  *     a variable:  point arg1 at it.
1524  *     a constant:  pass it in arg2.
1525  */
1526 
1527 int
1528 sysctl_handle_long(SYSCTL_HANDLER_ARGS)
1529 {
1530 	int error = 0;
1531 	long tmplong;
1532 #ifdef SCTL_MASK32
1533 	int tmpint;
1534 #endif
1535 
1536 	/*
1537 	 * Attempt to get a coherent snapshot by making a copy of the data.
1538 	 */
1539 	if (arg1)
1540 		tmplong = *(long *)arg1;
1541 	else
1542 		tmplong = arg2;
1543 #ifdef SCTL_MASK32
1544 	if (req->flags & SCTL_MASK32) {
1545 		tmpint = tmplong;
1546 		error = SYSCTL_OUT(req, &tmpint, sizeof(int));
1547 	} else
1548 #endif
1549 		error = SYSCTL_OUT(req, &tmplong, sizeof(long));
1550 
1551 	if (error || !req->newptr)
1552 		return (error);
1553 
1554 	if (!arg1)
1555 		error = EPERM;
1556 #ifdef SCTL_MASK32
1557 	else if (req->flags & SCTL_MASK32) {
1558 		error = SYSCTL_IN(req, &tmpint, sizeof(int));
1559 		*(long *)arg1 = (long)tmpint;
1560 	}
1561 #endif
1562 	else
1563 		error = SYSCTL_IN(req, arg1, sizeof(long));
1564 	return (error);
1565 }
1566 
1567 /*
1568  * Handle a 64 bit int, signed or unsigned.
1569  * Two cases:
1570  *     a variable:  point arg1 at it.
1571  *     a constant:  pass it in arg2.
1572  */
1573 int
1574 sysctl_handle_64(SYSCTL_HANDLER_ARGS)
1575 {
1576 	int error = 0;
1577 	uint64_t tmpout;
1578 
1579 	/*
1580 	 * Attempt to get a coherent snapshot by making a copy of the data.
1581 	 */
1582 	if (arg1)
1583 		tmpout = *(uint64_t *)arg1;
1584 	else
1585 		tmpout = arg2;
1586 	error = SYSCTL_OUT(req, &tmpout, sizeof(uint64_t));
1587 
1588 	if (error || !req->newptr)
1589 		return (error);
1590 
1591 	if (!arg1)
1592 		error = EPERM;
1593 	else
1594 		error = SYSCTL_IN(req, arg1, sizeof(uint64_t));
1595 	return (error);
1596 }
1597 
1598 /*
1599  * Handle our generic '\0' terminated 'C' string.
1600  * Two cases:
1601  * 	a variable string:  point arg1 at it, arg2 is max length.
1602  * 	a constant string:  point arg1 at it, arg2 is zero.
1603  */
1604 
1605 int
1606 sysctl_handle_string(SYSCTL_HANDLER_ARGS)
1607 {
1608 	size_t outlen;
1609 	int error = 0, ro_string = 0;
1610 
1611 	/*
1612 	 * A zero-length buffer indicates a fixed size read-only
1613 	 * string:
1614 	 */
1615 	if (arg2 == 0) {
1616 		arg2 = strlen((char *)arg1) + 1;
1617 		ro_string = 1;
1618 	}
1619 
1620 	if (req->oldptr != NULL) {
1621 		char *tmparg;
1622 
1623 		if (ro_string) {
1624 			tmparg = arg1;
1625 		} else {
1626 			/* try to make a coherent snapshot of the string */
1627 			tmparg = malloc(arg2, M_SYSCTLTMP, M_WAITOK);
1628 			memcpy(tmparg, arg1, arg2);
1629 		}
1630 
1631 		outlen = strnlen(tmparg, arg2 - 1) + 1;
1632 		error = SYSCTL_OUT(req, tmparg, outlen);
1633 
1634 		if (!ro_string)
1635 			free(tmparg, M_SYSCTLTMP);
1636 	} else {
1637 		outlen = strnlen((char *)arg1, arg2 - 1) + 1;
1638 		error = SYSCTL_OUT(req, NULL, outlen);
1639 	}
1640 	if (error || !req->newptr)
1641 		return (error);
1642 
1643 	if ((req->newlen - req->newidx) >= arg2) {
1644 		error = EINVAL;
1645 	} else {
1646 		arg2 = (req->newlen - req->newidx);
1647 		error = SYSCTL_IN(req, arg1, arg2);
1648 		((char *)arg1)[arg2] = '\0';
1649 	}
1650 	return (error);
1651 }
1652 
1653 /*
1654  * Handle any kind of opaque data.
1655  * arg1 points to it, arg2 is the size.
1656  */
1657 
1658 int
1659 sysctl_handle_opaque(SYSCTL_HANDLER_ARGS)
1660 {
1661 	int error, tries;
1662 	u_int generation;
1663 	struct sysctl_req req2;
1664 
1665 	/*
1666 	 * Attempt to get a coherent snapshot, by using the thread
1667 	 * pre-emption counter updated from within mi_switch() to
1668 	 * determine if we were pre-empted during a bcopy() or
1669 	 * copyout(). Make 3 attempts at doing this before giving up.
1670 	 * If we encounter an error, stop immediately.
1671 	 */
1672 	tries = 0;
1673 	req2 = *req;
1674 retry:
1675 	generation = curthread->td_generation;
1676 	error = SYSCTL_OUT(req, arg1, arg2);
1677 	if (error)
1678 		return (error);
1679 	tries++;
1680 	if (generation != curthread->td_generation && tries < 3) {
1681 		*req = req2;
1682 		goto retry;
1683 	}
1684 
1685 	error = SYSCTL_IN(req, arg1, arg2);
1686 
1687 	return (error);
1688 }
1689 
1690 /*
1691  * Transfer functions to/from kernel space.
1692  * XXX: rather untested at this point
1693  */
1694 static int
1695 sysctl_old_kernel(struct sysctl_req *req, const void *p, size_t l)
1696 {
1697 	size_t i = 0;
1698 
1699 	if (req->oldptr) {
1700 		i = l;
1701 		if (req->oldlen <= req->oldidx)
1702 			i = 0;
1703 		else
1704 			if (i > req->oldlen - req->oldidx)
1705 				i = req->oldlen - req->oldidx;
1706 		if (i > 0)
1707 			bcopy(p, (char *)req->oldptr + req->oldidx, i);
1708 	}
1709 	req->oldidx += l;
1710 	if (req->oldptr && i != l)
1711 		return (ENOMEM);
1712 	return (0);
1713 }
1714 
1715 static int
1716 sysctl_new_kernel(struct sysctl_req *req, void *p, size_t l)
1717 {
1718 	if (!req->newptr)
1719 		return (0);
1720 	if (req->newlen - req->newidx < l)
1721 		return (EINVAL);
1722 	bcopy((char *)req->newptr + req->newidx, p, l);
1723 	req->newidx += l;
1724 	return (0);
1725 }
1726 
1727 int
1728 kernel_sysctl(struct thread *td, int *name, u_int namelen, void *old,
1729     size_t *oldlenp, void *new, size_t newlen, size_t *retval, int flags)
1730 {
1731 	int error = 0;
1732 	struct sysctl_req req;
1733 
1734 	bzero(&req, sizeof req);
1735 
1736 	req.td = td;
1737 	req.flags = flags;
1738 
1739 	if (oldlenp) {
1740 		req.oldlen = *oldlenp;
1741 	}
1742 	req.validlen = req.oldlen;
1743 
1744 	if (old) {
1745 		req.oldptr= old;
1746 	}
1747 
1748 	if (new != NULL) {
1749 		req.newlen = newlen;
1750 		req.newptr = new;
1751 	}
1752 
1753 	req.oldfunc = sysctl_old_kernel;
1754 	req.newfunc = sysctl_new_kernel;
1755 	req.lock = REQ_UNWIRED;
1756 
1757 	error = sysctl_root(0, name, namelen, &req);
1758 
1759 	if (req.lock == REQ_WIRED && req.validlen > 0)
1760 		vsunlock(req.oldptr, req.validlen);
1761 
1762 	if (error && error != ENOMEM)
1763 		return (error);
1764 
1765 	if (retval) {
1766 		if (req.oldptr && req.oldidx > req.validlen)
1767 			*retval = req.validlen;
1768 		else
1769 			*retval = req.oldidx;
1770 	}
1771 	return (error);
1772 }
1773 
1774 int
1775 kernel_sysctlbyname(struct thread *td, char *name, void *old, size_t *oldlenp,
1776     void *new, size_t newlen, size_t *retval, int flags)
1777 {
1778         int oid[CTL_MAXNAME];
1779         size_t oidlen, plen;
1780 	int error;
1781 
1782 	oid[0] = 0;		/* sysctl internal magic */
1783 	oid[1] = 3;		/* name2oid */
1784 	oidlen = sizeof(oid);
1785 
1786 	error = kernel_sysctl(td, oid, 2, oid, &oidlen,
1787 	    (void *)name, strlen(name), &plen, flags);
1788 	if (error)
1789 		return (error);
1790 
1791 	error = kernel_sysctl(td, oid, plen / sizeof(int), old, oldlenp,
1792 	    new, newlen, retval, flags);
1793 	return (error);
1794 }
1795 
1796 /*
1797  * Transfer function to/from user space.
1798  */
1799 static int
1800 sysctl_old_user(struct sysctl_req *req, const void *p, size_t l)
1801 {
1802 	size_t i, len, origidx;
1803 	int error;
1804 
1805 	origidx = req->oldidx;
1806 	req->oldidx += l;
1807 	if (req->oldptr == NULL)
1808 		return (0);
1809 	/*
1810 	 * If we have not wired the user supplied buffer and we are currently
1811 	 * holding locks, drop a witness warning, as it's possible that
1812 	 * write operations to the user page can sleep.
1813 	 */
1814 	if (req->lock != REQ_WIRED)
1815 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
1816 		    "sysctl_old_user()");
1817 	i = l;
1818 	len = req->validlen;
1819 	if (len <= origidx)
1820 		i = 0;
1821 	else {
1822 		if (i > len - origidx)
1823 			i = len - origidx;
1824 		if (req->lock == REQ_WIRED) {
1825 			error = copyout_nofault(p, (char *)req->oldptr +
1826 			    origidx, i);
1827 		} else
1828 			error = copyout(p, (char *)req->oldptr + origidx, i);
1829 		if (error != 0)
1830 			return (error);
1831 	}
1832 	if (i < l)
1833 		return (ENOMEM);
1834 	return (0);
1835 }
1836 
1837 static int
1838 sysctl_new_user(struct sysctl_req *req, void *p, size_t l)
1839 {
1840 	int error;
1841 
1842 	if (!req->newptr)
1843 		return (0);
1844 	if (req->newlen - req->newidx < l)
1845 		return (EINVAL);
1846 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
1847 	    "sysctl_new_user()");
1848 	error = copyin((char *)req->newptr + req->newidx, p, l);
1849 	req->newidx += l;
1850 	return (error);
1851 }
1852 
1853 /*
1854  * Wire the user space destination buffer.  If set to a value greater than
1855  * zero, the len parameter limits the maximum amount of wired memory.
1856  */
1857 int
1858 sysctl_wire_old_buffer(struct sysctl_req *req, size_t len)
1859 {
1860 	int ret;
1861 	size_t wiredlen;
1862 
1863 	wiredlen = (len > 0 && len < req->oldlen) ? len : req->oldlen;
1864 	ret = 0;
1865 	if (req->lock != REQ_WIRED && req->oldptr &&
1866 	    req->oldfunc == sysctl_old_user) {
1867 		if (wiredlen != 0) {
1868 			ret = vslock(req->oldptr, wiredlen);
1869 			if (ret != 0) {
1870 				if (ret != ENOMEM)
1871 					return (ret);
1872 				wiredlen = 0;
1873 			}
1874 		}
1875 		req->lock = REQ_WIRED;
1876 		req->validlen = wiredlen;
1877 	}
1878 	return (0);
1879 }
1880 
1881 int
1882 sysctl_find_oid(int *name, u_int namelen, struct sysctl_oid **noid,
1883     int *nindx, struct sysctl_req *req)
1884 {
1885 	struct sysctl_oid_list *lsp;
1886 	struct sysctl_oid *oid;
1887 	int indx;
1888 
1889 	SYSCTL_ASSERT_LOCKED();
1890 	lsp = &sysctl__children;
1891 	indx = 0;
1892 	while (indx < CTL_MAXNAME) {
1893 		SLIST_FOREACH(oid, lsp, oid_link) {
1894 			if (oid->oid_number == name[indx])
1895 				break;
1896 		}
1897 		if (oid == NULL)
1898 			return (ENOENT);
1899 
1900 		indx++;
1901 		if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1902 			if (oid->oid_handler != NULL || indx == namelen) {
1903 				*noid = oid;
1904 				if (nindx != NULL)
1905 					*nindx = indx;
1906 				KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
1907 				    ("%s found DYING node %p", __func__, oid));
1908 				return (0);
1909 			}
1910 			lsp = SYSCTL_CHILDREN(oid);
1911 		} else if (indx == namelen) {
1912 			if ((oid->oid_kind & CTLFLAG_DORMANT) != 0)
1913 				return (ENOENT);
1914 			*noid = oid;
1915 			if (nindx != NULL)
1916 				*nindx = indx;
1917 			KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
1918 			    ("%s found DYING node %p", __func__, oid));
1919 			return (0);
1920 		} else {
1921 			return (ENOTDIR);
1922 		}
1923 	}
1924 	return (ENOENT);
1925 }
1926 
1927 /*
1928  * Traverse our tree, and find the right node, execute whatever it points
1929  * to, and return the resulting error code.
1930  */
1931 
1932 static int
1933 sysctl_root(SYSCTL_HANDLER_ARGS)
1934 {
1935 	struct sysctl_oid *oid;
1936 	struct rm_priotracker tracker;
1937 	int error, indx, lvl;
1938 
1939 	SYSCTL_RLOCK(&tracker);
1940 
1941 	error = sysctl_find_oid(arg1, arg2, &oid, &indx, req);
1942 	if (error)
1943 		goto out;
1944 
1945 	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1946 		/*
1947 		 * You can't call a sysctl when it's a node, but has
1948 		 * no handler.  Inform the user that it's a node.
1949 		 * The indx may or may not be the same as namelen.
1950 		 */
1951 		if (oid->oid_handler == NULL) {
1952 			error = EISDIR;
1953 			goto out;
1954 		}
1955 	}
1956 
1957 	/* Is this sysctl writable? */
1958 	if (req->newptr && !(oid->oid_kind & CTLFLAG_WR)) {
1959 		error = EPERM;
1960 		goto out;
1961 	}
1962 
1963 	KASSERT(req->td != NULL, ("sysctl_root(): req->td == NULL"));
1964 
1965 #ifdef CAPABILITY_MODE
1966 	/*
1967 	 * If the process is in capability mode, then don't permit reading or
1968 	 * writing unless specifically granted for the node.
1969 	 */
1970 	if (IN_CAPABILITY_MODE(req->td)) {
1971 		if ((req->oldptr && !(oid->oid_kind & CTLFLAG_CAPRD)) ||
1972 		    (req->newptr && !(oid->oid_kind & CTLFLAG_CAPWR))) {
1973 			error = EPERM;
1974 			goto out;
1975 		}
1976 	}
1977 #endif
1978 
1979 	/* Is this sysctl sensitive to securelevels? */
1980 	if (req->newptr && (oid->oid_kind & CTLFLAG_SECURE)) {
1981 		lvl = (oid->oid_kind & CTLMASK_SECURE) >> CTLSHIFT_SECURE;
1982 		error = securelevel_gt(req->td->td_ucred, lvl);
1983 		if (error)
1984 			goto out;
1985 	}
1986 
1987 	/* Is this sysctl writable by only privileged users? */
1988 	if (req->newptr && !(oid->oid_kind & CTLFLAG_ANYBODY)) {
1989 		int priv;
1990 
1991 		if (oid->oid_kind & CTLFLAG_PRISON)
1992 			priv = PRIV_SYSCTL_WRITEJAIL;
1993 #ifdef VIMAGE
1994 		else if ((oid->oid_kind & CTLFLAG_VNET) &&
1995 		     prison_owns_vnet(req->td->td_ucred))
1996 			priv = PRIV_SYSCTL_WRITEJAIL;
1997 #endif
1998 		else
1999 			priv = PRIV_SYSCTL_WRITE;
2000 		error = priv_check(req->td, priv);
2001 		if (error)
2002 			goto out;
2003 	}
2004 
2005 	if (!oid->oid_handler) {
2006 		error = EINVAL;
2007 		goto out;
2008 	}
2009 
2010 	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
2011 		arg1 = (int *)arg1 + indx;
2012 		arg2 -= indx;
2013 	} else {
2014 		arg1 = oid->oid_arg1;
2015 		arg2 = oid->oid_arg2;
2016 	}
2017 #ifdef MAC
2018 	error = mac_system_check_sysctl(req->td->td_ucred, oid, arg1, arg2,
2019 	    req);
2020 	if (error != 0)
2021 		goto out;
2022 #endif
2023 #ifdef VIMAGE
2024 	if ((oid->oid_kind & CTLFLAG_VNET) && arg1 != NULL)
2025 		arg1 = (void *)(curvnet->vnet_data_base + (uintptr_t)arg1);
2026 #endif
2027 	error = sysctl_root_handler_locked(oid, arg1, arg2, req, &tracker);
2028 
2029 out:
2030 	SYSCTL_RUNLOCK(&tracker);
2031 	return (error);
2032 }
2033 
2034 #ifndef _SYS_SYSPROTO_H_
2035 struct sysctl_args {
2036 	int	*name;
2037 	u_int	namelen;
2038 	void	*old;
2039 	size_t	*oldlenp;
2040 	void	*new;
2041 	size_t	newlen;
2042 };
2043 #endif
2044 int
2045 sys___sysctl(struct thread *td, struct sysctl_args *uap)
2046 {
2047 	int error, i, name[CTL_MAXNAME];
2048 	size_t j;
2049 
2050 	if (uap->namelen > CTL_MAXNAME || uap->namelen < 2)
2051 		return (EINVAL);
2052 
2053  	error = copyin(uap->name, &name, uap->namelen * sizeof(int));
2054  	if (error)
2055 		return (error);
2056 
2057 	error = userland_sysctl(td, name, uap->namelen,
2058 		uap->old, uap->oldlenp, 0,
2059 		uap->new, uap->newlen, &j, 0);
2060 	if (error && error != ENOMEM)
2061 		return (error);
2062 	if (uap->oldlenp) {
2063 		i = copyout(&j, uap->oldlenp, sizeof(j));
2064 		if (i)
2065 			return (i);
2066 	}
2067 	return (error);
2068 }
2069 
2070 /*
2071  * This is used from various compatibility syscalls too.  That's why name
2072  * must be in kernel space.
2073  */
2074 int
2075 userland_sysctl(struct thread *td, int *name, u_int namelen, void *old,
2076     size_t *oldlenp, int inkernel, void *new, size_t newlen, size_t *retval,
2077     int flags)
2078 {
2079 	int error = 0, memlocked;
2080 	struct sysctl_req req;
2081 
2082 	bzero(&req, sizeof req);
2083 
2084 	req.td = td;
2085 	req.flags = flags;
2086 
2087 	if (oldlenp) {
2088 		if (inkernel) {
2089 			req.oldlen = *oldlenp;
2090 		} else {
2091 			error = copyin(oldlenp, &req.oldlen, sizeof(*oldlenp));
2092 			if (error)
2093 				return (error);
2094 		}
2095 	}
2096 	req.validlen = req.oldlen;
2097 	req.oldptr = old;
2098 
2099 	if (new != NULL) {
2100 		req.newlen = newlen;
2101 		req.newptr = new;
2102 	}
2103 
2104 	req.oldfunc = sysctl_old_user;
2105 	req.newfunc = sysctl_new_user;
2106 	req.lock = REQ_UNWIRED;
2107 
2108 #ifdef KTRACE
2109 	if (KTRPOINT(curthread, KTR_SYSCTL))
2110 		ktrsysctl(name, namelen);
2111 #endif
2112 
2113 	if (req.oldptr && req.oldlen > PAGE_SIZE) {
2114 		memlocked = 1;
2115 		sx_xlock(&sysctlmemlock);
2116 	} else
2117 		memlocked = 0;
2118 	CURVNET_SET(TD_TO_VNET(td));
2119 
2120 	for (;;) {
2121 		req.oldidx = 0;
2122 		req.newidx = 0;
2123 		error = sysctl_root(0, name, namelen, &req);
2124 		if (error != EAGAIN)
2125 			break;
2126 		kern_yield(PRI_USER);
2127 	}
2128 
2129 	CURVNET_RESTORE();
2130 
2131 	if (req.lock == REQ_WIRED && req.validlen > 0)
2132 		vsunlock(req.oldptr, req.validlen);
2133 	if (memlocked)
2134 		sx_xunlock(&sysctlmemlock);
2135 
2136 	if (error && error != ENOMEM)
2137 		return (error);
2138 
2139 	if (retval) {
2140 		if (req.oldptr && req.oldidx > req.validlen)
2141 			*retval = req.validlen;
2142 		else
2143 			*retval = req.oldidx;
2144 	}
2145 	return (error);
2146 }
2147 
2148 /*
2149  * Drain into a sysctl struct.  The user buffer should be wired if a page
2150  * fault would cause issue.
2151  */
2152 static int
2153 sbuf_sysctl_drain(void *arg, const char *data, int len)
2154 {
2155 	struct sysctl_req *req = arg;
2156 	int error;
2157 
2158 	error = SYSCTL_OUT(req, data, len);
2159 	KASSERT(error >= 0, ("Got unexpected negative value %d", error));
2160 	return (error == 0 ? len : -error);
2161 }
2162 
2163 struct sbuf *
2164 sbuf_new_for_sysctl(struct sbuf *s, char *buf, int length,
2165     struct sysctl_req *req)
2166 {
2167 
2168 	/* Supply a default buffer size if none given. */
2169 	if (buf == NULL && length == 0)
2170 		length = 64;
2171 	s = sbuf_new(s, buf, length, SBUF_FIXEDLEN | SBUF_INCLUDENUL);
2172 	sbuf_set_drain(s, sbuf_sysctl_drain, req);
2173 	return (s);
2174 }
2175