xref: /freebsd/sys/kern/kern_sysctl.c (revision 0957b409a90fd597c1e9124cbaf3edd2b488f4ac)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Mike Karels at Berkeley Software Design, Inc.
9  *
10  * Quite extensively rewritten by Poul-Henning Kamp of the FreeBSD
11  * project, to make these variables more userfriendly.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. Neither the name of the University nor the names of its contributors
22  *    may be used to endorse or promote products derived from this software
23  *    without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  *
37  *	@(#)kern_sysctl.c	8.4 (Berkeley) 4/14/94
38  */
39 
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
42 
43 #include "opt_capsicum.h"
44 #include "opt_ktrace.h"
45 
46 #include <sys/param.h>
47 #include <sys/fail.h>
48 #include <sys/systm.h>
49 #include <sys/capsicum.h>
50 #include <sys/kernel.h>
51 #include <sys/sysctl.h>
52 #include <sys/malloc.h>
53 #include <sys/priv.h>
54 #include <sys/proc.h>
55 #include <sys/jail.h>
56 #include <sys/lock.h>
57 #include <sys/mutex.h>
58 #include <sys/rmlock.h>
59 #include <sys/sbuf.h>
60 #include <sys/sx.h>
61 #include <sys/sysproto.h>
62 #include <sys/uio.h>
63 #ifdef KTRACE
64 #include <sys/ktrace.h>
65 #endif
66 
67 #include <net/vnet.h>
68 
69 #include <security/mac/mac_framework.h>
70 
71 #include <vm/vm.h>
72 #include <vm/vm_extern.h>
73 
74 static MALLOC_DEFINE(M_SYSCTL, "sysctl", "sysctl internal magic");
75 static MALLOC_DEFINE(M_SYSCTLOID, "sysctloid", "sysctl dynamic oids");
76 static MALLOC_DEFINE(M_SYSCTLTMP, "sysctltmp", "sysctl temp output buffer");
77 
78 /*
79  * The sysctllock protects the MIB tree.  It also protects sysctl
80  * contexts used with dynamic sysctls.  The sysctl_register_oid() and
81  * sysctl_unregister_oid() routines require the sysctllock to already
82  * be held, so the sysctl_wlock() and sysctl_wunlock() routines are
83  * provided for the few places in the kernel which need to use that
84  * API rather than using the dynamic API.  Use of the dynamic API is
85  * strongly encouraged for most code.
86  *
87  * The sysctlmemlock is used to limit the amount of user memory wired for
88  * sysctl requests.  This is implemented by serializing any userland
89  * sysctl requests larger than a single page via an exclusive lock.
90  */
91 static struct rmlock sysctllock;
92 static struct sx __exclusive_cache_line sysctlmemlock;
93 
94 #define	SYSCTL_WLOCK()		rm_wlock(&sysctllock)
95 #define	SYSCTL_WUNLOCK()	rm_wunlock(&sysctllock)
96 #define	SYSCTL_RLOCK(tracker)	rm_rlock(&sysctllock, (tracker))
97 #define	SYSCTL_RUNLOCK(tracker)	rm_runlock(&sysctllock, (tracker))
98 #define	SYSCTL_WLOCKED()	rm_wowned(&sysctllock)
99 #define	SYSCTL_ASSERT_LOCKED()	rm_assert(&sysctllock, RA_LOCKED)
100 #define	SYSCTL_ASSERT_WLOCKED()	rm_assert(&sysctllock, RA_WLOCKED)
101 #define	SYSCTL_ASSERT_RLOCKED()	rm_assert(&sysctllock, RA_RLOCKED)
102 #define	SYSCTL_INIT()		rm_init_flags(&sysctllock, "sysctl lock", \
103 				    RM_SLEEPABLE)
104 #define	SYSCTL_SLEEP(ch, wmesg, timo)					\
105 				rm_sleep(ch, &sysctllock, 0, wmesg, timo)
106 
107 static int sysctl_root(SYSCTL_HANDLER_ARGS);
108 
109 /* Root list */
110 struct sysctl_oid_list sysctl__children = SLIST_HEAD_INITIALIZER(&sysctl__children);
111 
112 static int	sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del,
113 		    int recurse);
114 static int	sysctl_old_kernel(struct sysctl_req *, const void *, size_t);
115 static int	sysctl_new_kernel(struct sysctl_req *, void *, size_t);
116 
117 static struct sysctl_oid *
118 sysctl_find_oidname(const char *name, struct sysctl_oid_list *list)
119 {
120 	struct sysctl_oid *oidp;
121 
122 	SYSCTL_ASSERT_LOCKED();
123 	SLIST_FOREACH(oidp, list, oid_link) {
124 		if (strcmp(oidp->oid_name, name) == 0) {
125 			return (oidp);
126 		}
127 	}
128 	return (NULL);
129 }
130 
131 /*
132  * Initialization of the MIB tree.
133  *
134  * Order by number in each list.
135  */
136 void
137 sysctl_wlock(void)
138 {
139 
140 	SYSCTL_WLOCK();
141 }
142 
143 void
144 sysctl_wunlock(void)
145 {
146 
147 	SYSCTL_WUNLOCK();
148 }
149 
150 static int
151 sysctl_root_handler_locked(struct sysctl_oid *oid, void *arg1, intmax_t arg2,
152     struct sysctl_req *req, struct rm_priotracker *tracker)
153 {
154 	int error;
155 
156 	if (oid->oid_kind & CTLFLAG_DYN)
157 		atomic_add_int(&oid->oid_running, 1);
158 
159 	if (tracker != NULL)
160 		SYSCTL_RUNLOCK(tracker);
161 	else
162 		SYSCTL_WUNLOCK();
163 
164 	if (!(oid->oid_kind & CTLFLAG_MPSAFE))
165 		mtx_lock(&Giant);
166 	error = oid->oid_handler(oid, arg1, arg2, req);
167 	if (!(oid->oid_kind & CTLFLAG_MPSAFE))
168 		mtx_unlock(&Giant);
169 
170 	KFAIL_POINT_ERROR(_debug_fail_point, sysctl_running, error);
171 
172 	if (tracker != NULL)
173 		SYSCTL_RLOCK(tracker);
174 	else
175 		SYSCTL_WLOCK();
176 
177 	if (oid->oid_kind & CTLFLAG_DYN) {
178 		if (atomic_fetchadd_int(&oid->oid_running, -1) == 1 &&
179 		    (oid->oid_kind & CTLFLAG_DYING) != 0)
180 			wakeup(&oid->oid_running);
181 	}
182 
183 	return (error);
184 }
185 
186 static void
187 sysctl_load_tunable_by_oid_locked(struct sysctl_oid *oidp)
188 {
189 	struct sysctl_req req;
190 	struct sysctl_oid *curr;
191 	char *penv = NULL;
192 	char path[96];
193 	ssize_t rem = sizeof(path);
194 	ssize_t len;
195 	uint8_t data[512] __aligned(sizeof(uint64_t));
196 	int size;
197 	int error;
198 
199 	path[--rem] = 0;
200 
201 	for (curr = oidp; curr != NULL; curr = SYSCTL_PARENT(curr)) {
202 		len = strlen(curr->oid_name);
203 		rem -= len;
204 		if (curr != oidp)
205 			rem -= 1;
206 		if (rem < 0) {
207 			printf("OID path exceeds %d bytes\n", (int)sizeof(path));
208 			return;
209 		}
210 		memcpy(path + rem, curr->oid_name, len);
211 		if (curr != oidp)
212 			path[rem + len] = '.';
213 	}
214 
215 	memset(&req, 0, sizeof(req));
216 
217 	req.td = curthread;
218 	req.oldfunc = sysctl_old_kernel;
219 	req.newfunc = sysctl_new_kernel;
220 	req.lock = REQ_UNWIRED;
221 
222 	switch (oidp->oid_kind & CTLTYPE) {
223 	case CTLTYPE_INT:
224 		if (getenv_array(path + rem, data, sizeof(data), &size,
225 		    sizeof(int), GETENV_SIGNED) == 0)
226 			return;
227 		req.newlen = size;
228 		req.newptr = data;
229 		break;
230 	case CTLTYPE_UINT:
231 		if (getenv_array(path + rem, data, sizeof(data), &size,
232 		    sizeof(int), GETENV_UNSIGNED) == 0)
233 			return;
234 		req.newlen = size;
235 		req.newptr = data;
236 		break;
237 	case CTLTYPE_LONG:
238 		if (getenv_array(path + rem, data, sizeof(data), &size,
239 		    sizeof(long), GETENV_SIGNED) == 0)
240 			return;
241 		req.newlen = size;
242 		req.newptr = data;
243 		break;
244 	case CTLTYPE_ULONG:
245 		if (getenv_array(path + rem, data, sizeof(data), &size,
246 		    sizeof(long), GETENV_UNSIGNED) == 0)
247 			return;
248 		req.newlen = size;
249 		req.newptr = data;
250 		break;
251 	case CTLTYPE_S8:
252 		if (getenv_array(path + rem, data, sizeof(data), &size,
253 		    sizeof(int8_t), GETENV_SIGNED) == 0)
254 			return;
255 		req.newlen = size;
256 		req.newptr = data;
257 		break;
258 	case CTLTYPE_S16:
259 		if (getenv_array(path + rem, data, sizeof(data), &size,
260 		    sizeof(int16_t), GETENV_SIGNED) == 0)
261 			return;
262 		req.newlen = size;
263 		req.newptr = data;
264 		break;
265 	case CTLTYPE_S32:
266 		if (getenv_array(path + rem, data, sizeof(data), &size,
267 		    sizeof(int32_t), GETENV_SIGNED) == 0)
268 			return;
269 		req.newlen = size;
270 		req.newptr = data;
271 		break;
272 	case CTLTYPE_S64:
273 		if (getenv_array(path + rem, data, sizeof(data), &size,
274 		    sizeof(int64_t), GETENV_SIGNED) == 0)
275 			return;
276 		req.newlen = size;
277 		req.newptr = data;
278 		break;
279 	case CTLTYPE_U8:
280 		if (getenv_array(path + rem, data, sizeof(data), &size,
281 		    sizeof(uint8_t), GETENV_UNSIGNED) == 0)
282 			return;
283 		req.newlen = size;
284 		req.newptr = data;
285 		break;
286 	case CTLTYPE_U16:
287 		if (getenv_array(path + rem, data, sizeof(data), &size,
288 		    sizeof(uint16_t), GETENV_UNSIGNED) == 0)
289 			return;
290 		req.newlen = size;
291 		req.newptr = data;
292 		break;
293 	case CTLTYPE_U32:
294 		if (getenv_array(path + rem, data, sizeof(data), &size,
295 		    sizeof(uint32_t), GETENV_UNSIGNED) == 0)
296 			return;
297 		req.newlen = size;
298 		req.newptr = data;
299 		break;
300 	case CTLTYPE_U64:
301 		if (getenv_array(path + rem, data, sizeof(data), &size,
302 		    sizeof(uint64_t), GETENV_UNSIGNED) == 0)
303 			return;
304 		req.newlen = size;
305 		req.newptr = data;
306 		break;
307 	case CTLTYPE_STRING:
308 		penv = kern_getenv(path + rem);
309 		if (penv == NULL)
310 			return;
311 		req.newlen = strlen(penv);
312 		req.newptr = penv;
313 		break;
314 	default:
315 		return;
316 	}
317 	error = sysctl_root_handler_locked(oidp, oidp->oid_arg1,
318 	    oidp->oid_arg2, &req, NULL);
319 	if (error != 0)
320 		printf("Setting sysctl %s failed: %d\n", path + rem, error);
321 	if (penv != NULL)
322 		freeenv(penv);
323 }
324 
325 static int
326 sbuf_printf_drain(void *arg __unused, const char *data, int len)
327 {
328 
329 	return (printf("%.*s", len, data));
330 }
331 
332 /*
333  * Locate the path to a given oid.  Returns the length of the resulting path,
334  * or -1 if the oid was not found.  nodes must have room for CTL_MAXNAME
335  * elements and be NULL initialized.
336  */
337 static int
338 sysctl_search_oid(struct sysctl_oid **nodes, struct sysctl_oid *needle)
339 {
340 	int indx;
341 
342 	SYSCTL_ASSERT_LOCKED();
343 	indx = 0;
344 	while (indx < CTL_MAXNAME && indx >= 0) {
345 		if (nodes[indx] == NULL && indx == 0)
346 			nodes[indx] = SLIST_FIRST(&sysctl__children);
347 		else if (nodes[indx] == NULL)
348 			nodes[indx] = SLIST_FIRST(&nodes[indx - 1]->oid_children);
349 		else
350 			nodes[indx] = SLIST_NEXT(nodes[indx], oid_link);
351 
352 		if (nodes[indx] == needle)
353 			return (indx + 1);
354 
355 		if (nodes[indx] == NULL) {
356 			indx--;
357 			continue;
358 		}
359 
360 		if ((nodes[indx]->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
361 			indx++;
362 			continue;
363 		}
364 	}
365 	return (-1);
366 }
367 
368 static void
369 sysctl_warn_reuse(const char *func, struct sysctl_oid *leaf)
370 {
371 	struct sysctl_oid *nodes[CTL_MAXNAME];
372 	char buf[128];
373 	struct sbuf sb;
374 	int rc, i;
375 
376 	(void)sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN | SBUF_INCLUDENUL);
377 	sbuf_set_drain(&sb, sbuf_printf_drain, NULL);
378 
379 	sbuf_printf(&sb, "%s: can't re-use a leaf (", __func__);
380 
381 	memset(nodes, 0, sizeof(nodes));
382 	rc = sysctl_search_oid(nodes, leaf);
383 	if (rc > 0) {
384 		for (i = 0; i < rc; i++)
385 			sbuf_printf(&sb, "%s%.*s", nodes[i]->oid_name,
386 			    i != (rc - 1), ".");
387 	} else {
388 		sbuf_printf(&sb, "%s", leaf->oid_name);
389 	}
390 	sbuf_printf(&sb, ")!\n");
391 
392 	(void)sbuf_finish(&sb);
393 }
394 
395 #ifdef SYSCTL_DEBUG
396 static int
397 sysctl_reuse_test(SYSCTL_HANDLER_ARGS)
398 {
399 	struct rm_priotracker tracker;
400 
401 	SYSCTL_RLOCK(&tracker);
402 	sysctl_warn_reuse(__func__, oidp);
403 	SYSCTL_RUNLOCK(&tracker);
404 	return (0);
405 }
406 SYSCTL_PROC(_sysctl, 0, reuse_test, CTLTYPE_STRING|CTLFLAG_RD|CTLFLAG_MPSAFE,
407 	0, 0, sysctl_reuse_test, "-", "");
408 #endif
409 
410 void
411 sysctl_register_oid(struct sysctl_oid *oidp)
412 {
413 	struct sysctl_oid_list *parent = oidp->oid_parent;
414 	struct sysctl_oid *p;
415 	struct sysctl_oid *q;
416 	int oid_number;
417 	int timeout = 2;
418 
419 	/*
420 	 * First check if another oid with the same name already
421 	 * exists in the parent's list.
422 	 */
423 	SYSCTL_ASSERT_WLOCKED();
424 	p = sysctl_find_oidname(oidp->oid_name, parent);
425 	if (p != NULL) {
426 		if ((p->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
427 			p->oid_refcnt++;
428 			return;
429 		} else {
430 			sysctl_warn_reuse(__func__, p);
431 			return;
432 		}
433 	}
434 	/* get current OID number */
435 	oid_number = oidp->oid_number;
436 
437 #if (OID_AUTO >= 0)
438 #error "OID_AUTO is expected to be a negative value"
439 #endif
440 	/*
441 	 * Any negative OID number qualifies as OID_AUTO. Valid OID
442 	 * numbers should always be positive.
443 	 *
444 	 * NOTE: DO NOT change the starting value here, change it in
445 	 * <sys/sysctl.h>, and make sure it is at least 256 to
446 	 * accommodate e.g. net.inet.raw as a static sysctl node.
447 	 */
448 	if (oid_number < 0) {
449 		static int newoid;
450 
451 		/*
452 		 * By decrementing the next OID number we spend less
453 		 * time inserting the OIDs into a sorted list.
454 		 */
455 		if (--newoid < CTL_AUTO_START)
456 			newoid = 0x7fffffff;
457 
458 		oid_number = newoid;
459 	}
460 
461 	/*
462 	 * Insert the OID into the parent's list sorted by OID number.
463 	 */
464 retry:
465 	q = NULL;
466 	SLIST_FOREACH(p, parent, oid_link) {
467 		/* check if the current OID number is in use */
468 		if (oid_number == p->oid_number) {
469 			/* get the next valid OID number */
470 			if (oid_number < CTL_AUTO_START ||
471 			    oid_number == 0x7fffffff) {
472 				/* wraparound - restart */
473 				oid_number = CTL_AUTO_START;
474 				/* don't loop forever */
475 				if (!timeout--)
476 					panic("sysctl: Out of OID numbers\n");
477 				goto retry;
478 			} else {
479 				oid_number++;
480 			}
481 		} else if (oid_number < p->oid_number)
482 			break;
483 		q = p;
484 	}
485 	/* check for non-auto OID number collision */
486 	if (oidp->oid_number >= 0 && oidp->oid_number < CTL_AUTO_START &&
487 	    oid_number >= CTL_AUTO_START) {
488 		printf("sysctl: OID number(%d) is already in use for '%s'\n",
489 		    oidp->oid_number, oidp->oid_name);
490 	}
491 	/* update the OID number, if any */
492 	oidp->oid_number = oid_number;
493 	if (q != NULL)
494 		SLIST_INSERT_AFTER(q, oidp, oid_link);
495 	else
496 		SLIST_INSERT_HEAD(parent, oidp, oid_link);
497 
498 	if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE &&
499 #ifdef VIMAGE
500 	    (oidp->oid_kind & CTLFLAG_VNET) == 0 &&
501 #endif
502 	    (oidp->oid_kind & CTLFLAG_TUN) != 0 &&
503 	    (oidp->oid_kind & CTLFLAG_NOFETCH) == 0) {
504 		/* only fetch value once */
505 		oidp->oid_kind |= CTLFLAG_NOFETCH;
506 		/* try to fetch value from kernel environment */
507 		sysctl_load_tunable_by_oid_locked(oidp);
508 	}
509 }
510 
511 void
512 sysctl_register_disabled_oid(struct sysctl_oid *oidp)
513 {
514 
515 	/*
516 	 * Mark the leaf as dormant if it's not to be immediately enabled.
517 	 * We do not disable nodes as they can be shared between modules
518 	 * and it is always safe to access a node.
519 	 */
520 	KASSERT((oidp->oid_kind & CTLFLAG_DORMANT) == 0,
521 	    ("internal flag is set in oid_kind"));
522 	if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
523 		oidp->oid_kind |= CTLFLAG_DORMANT;
524 	sysctl_register_oid(oidp);
525 }
526 
527 void
528 sysctl_enable_oid(struct sysctl_oid *oidp)
529 {
530 
531 	SYSCTL_ASSERT_WLOCKED();
532 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
533 		KASSERT((oidp->oid_kind & CTLFLAG_DORMANT) == 0,
534 		    ("sysctl node is marked as dormant"));
535 		return;
536 	}
537 	KASSERT((oidp->oid_kind & CTLFLAG_DORMANT) != 0,
538 	    ("enabling already enabled sysctl oid"));
539 	oidp->oid_kind &= ~CTLFLAG_DORMANT;
540 }
541 
542 void
543 sysctl_unregister_oid(struct sysctl_oid *oidp)
544 {
545 	struct sysctl_oid *p;
546 	int error;
547 
548 	SYSCTL_ASSERT_WLOCKED();
549 	if (oidp->oid_number == OID_AUTO) {
550 		error = EINVAL;
551 	} else {
552 		error = ENOENT;
553 		SLIST_FOREACH(p, oidp->oid_parent, oid_link) {
554 			if (p == oidp) {
555 				SLIST_REMOVE(oidp->oid_parent, oidp,
556 				    sysctl_oid, oid_link);
557 				error = 0;
558 				break;
559 			}
560 		}
561 	}
562 
563 	/*
564 	 * This can happen when a module fails to register and is
565 	 * being unloaded afterwards.  It should not be a panic()
566 	 * for normal use.
567 	 */
568 	if (error) {
569 		printf("%s: failed(%d) to unregister sysctl(%s)\n",
570 		    __func__, error, oidp->oid_name);
571 	}
572 }
573 
574 /* Initialize a new context to keep track of dynamically added sysctls. */
575 int
576 sysctl_ctx_init(struct sysctl_ctx_list *c)
577 {
578 
579 	if (c == NULL) {
580 		return (EINVAL);
581 	}
582 
583 	/*
584 	 * No locking here, the caller is responsible for not adding
585 	 * new nodes to a context until after this function has
586 	 * returned.
587 	 */
588 	TAILQ_INIT(c);
589 	return (0);
590 }
591 
592 /* Free the context, and destroy all dynamic oids registered in this context */
593 int
594 sysctl_ctx_free(struct sysctl_ctx_list *clist)
595 {
596 	struct sysctl_ctx_entry *e, *e1;
597 	int error;
598 
599 	error = 0;
600 	/*
601 	 * First perform a "dry run" to check if it's ok to remove oids.
602 	 * XXX FIXME
603 	 * XXX This algorithm is a hack. But I don't know any
604 	 * XXX better solution for now...
605 	 */
606 	SYSCTL_WLOCK();
607 	TAILQ_FOREACH(e, clist, link) {
608 		error = sysctl_remove_oid_locked(e->entry, 0, 0);
609 		if (error)
610 			break;
611 	}
612 	/*
613 	 * Restore deregistered entries, either from the end,
614 	 * or from the place where error occurred.
615 	 * e contains the entry that was not unregistered
616 	 */
617 	if (error)
618 		e1 = TAILQ_PREV(e, sysctl_ctx_list, link);
619 	else
620 		e1 = TAILQ_LAST(clist, sysctl_ctx_list);
621 	while (e1 != NULL) {
622 		sysctl_register_oid(e1->entry);
623 		e1 = TAILQ_PREV(e1, sysctl_ctx_list, link);
624 	}
625 	if (error) {
626 		SYSCTL_WUNLOCK();
627 		return(EBUSY);
628 	}
629 	/* Now really delete the entries */
630 	e = TAILQ_FIRST(clist);
631 	while (e != NULL) {
632 		e1 = TAILQ_NEXT(e, link);
633 		error = sysctl_remove_oid_locked(e->entry, 1, 0);
634 		if (error)
635 			panic("sysctl_remove_oid: corrupt tree, entry: %s",
636 			    e->entry->oid_name);
637 		free(e, M_SYSCTLOID);
638 		e = e1;
639 	}
640 	SYSCTL_WUNLOCK();
641 	return (error);
642 }
643 
644 /* Add an entry to the context */
645 struct sysctl_ctx_entry *
646 sysctl_ctx_entry_add(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
647 {
648 	struct sysctl_ctx_entry *e;
649 
650 	SYSCTL_ASSERT_WLOCKED();
651 	if (clist == NULL || oidp == NULL)
652 		return(NULL);
653 	e = malloc(sizeof(struct sysctl_ctx_entry), M_SYSCTLOID, M_WAITOK);
654 	e->entry = oidp;
655 	TAILQ_INSERT_HEAD(clist, e, link);
656 	return (e);
657 }
658 
659 /* Find an entry in the context */
660 struct sysctl_ctx_entry *
661 sysctl_ctx_entry_find(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
662 {
663 	struct sysctl_ctx_entry *e;
664 
665 	SYSCTL_ASSERT_WLOCKED();
666 	if (clist == NULL || oidp == NULL)
667 		return(NULL);
668 	TAILQ_FOREACH(e, clist, link) {
669 		if(e->entry == oidp)
670 			return(e);
671 	}
672 	return (e);
673 }
674 
675 /*
676  * Delete an entry from the context.
677  * NOTE: this function doesn't free oidp! You have to remove it
678  * with sysctl_remove_oid().
679  */
680 int
681 sysctl_ctx_entry_del(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
682 {
683 	struct sysctl_ctx_entry *e;
684 
685 	if (clist == NULL || oidp == NULL)
686 		return (EINVAL);
687 	SYSCTL_WLOCK();
688 	e = sysctl_ctx_entry_find(clist, oidp);
689 	if (e != NULL) {
690 		TAILQ_REMOVE(clist, e, link);
691 		SYSCTL_WUNLOCK();
692 		free(e, M_SYSCTLOID);
693 		return (0);
694 	} else {
695 		SYSCTL_WUNLOCK();
696 		return (ENOENT);
697 	}
698 }
699 
700 /*
701  * Remove dynamically created sysctl trees.
702  * oidp - top of the tree to be removed
703  * del - if 0 - just deregister, otherwise free up entries as well
704  * recurse - if != 0 traverse the subtree to be deleted
705  */
706 int
707 sysctl_remove_oid(struct sysctl_oid *oidp, int del, int recurse)
708 {
709 	int error;
710 
711 	SYSCTL_WLOCK();
712 	error = sysctl_remove_oid_locked(oidp, del, recurse);
713 	SYSCTL_WUNLOCK();
714 	return (error);
715 }
716 
717 int
718 sysctl_remove_name(struct sysctl_oid *parent, const char *name,
719     int del, int recurse)
720 {
721 	struct sysctl_oid *p, *tmp;
722 	int error;
723 
724 	error = ENOENT;
725 	SYSCTL_WLOCK();
726 	SLIST_FOREACH_SAFE(p, SYSCTL_CHILDREN(parent), oid_link, tmp) {
727 		if (strcmp(p->oid_name, name) == 0) {
728 			error = sysctl_remove_oid_locked(p, del, recurse);
729 			break;
730 		}
731 	}
732 	SYSCTL_WUNLOCK();
733 
734 	return (error);
735 }
736 
737 
738 static int
739 sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del, int recurse)
740 {
741 	struct sysctl_oid *p, *tmp;
742 	int error;
743 
744 	SYSCTL_ASSERT_WLOCKED();
745 	if (oidp == NULL)
746 		return(EINVAL);
747 	if ((oidp->oid_kind & CTLFLAG_DYN) == 0) {
748 		printf("Warning: can't remove non-dynamic nodes (%s)!\n",
749 		    oidp->oid_name);
750 		return (EINVAL);
751 	}
752 	/*
753 	 * WARNING: normal method to do this should be through
754 	 * sysctl_ctx_free(). Use recursing as the last resort
755 	 * method to purge your sysctl tree of leftovers...
756 	 * However, if some other code still references these nodes,
757 	 * it will panic.
758 	 */
759 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
760 		if (oidp->oid_refcnt == 1) {
761 			SLIST_FOREACH_SAFE(p,
762 			    SYSCTL_CHILDREN(oidp), oid_link, tmp) {
763 				if (!recurse) {
764 					printf("Warning: failed attempt to "
765 					    "remove oid %s with child %s\n",
766 					    oidp->oid_name, p->oid_name);
767 					return (ENOTEMPTY);
768 				}
769 				error = sysctl_remove_oid_locked(p, del,
770 				    recurse);
771 				if (error)
772 					return (error);
773 			}
774 		}
775 	}
776 	if (oidp->oid_refcnt > 1 ) {
777 		oidp->oid_refcnt--;
778 	} else {
779 		if (oidp->oid_refcnt == 0) {
780 			printf("Warning: bad oid_refcnt=%u (%s)!\n",
781 				oidp->oid_refcnt, oidp->oid_name);
782 			return (EINVAL);
783 		}
784 		sysctl_unregister_oid(oidp);
785 		if (del) {
786 			/*
787 			 * Wait for all threads running the handler to drain.
788 			 * This preserves the previous behavior when the
789 			 * sysctl lock was held across a handler invocation,
790 			 * and is necessary for module unload correctness.
791 			 */
792 			while (oidp->oid_running > 0) {
793 				oidp->oid_kind |= CTLFLAG_DYING;
794 				SYSCTL_SLEEP(&oidp->oid_running, "oidrm", 0);
795 			}
796 			if (oidp->oid_descr)
797 				free(__DECONST(char *, oidp->oid_descr),
798 				    M_SYSCTLOID);
799 			if (oidp->oid_label)
800 				free(__DECONST(char *, oidp->oid_label),
801 				    M_SYSCTLOID);
802 			free(__DECONST(char *, oidp->oid_name), M_SYSCTLOID);
803 			free(oidp, M_SYSCTLOID);
804 		}
805 	}
806 	return (0);
807 }
808 /*
809  * Create new sysctls at run time.
810  * clist may point to a valid context initialized with sysctl_ctx_init().
811  */
812 struct sysctl_oid *
813 sysctl_add_oid(struct sysctl_ctx_list *clist, struct sysctl_oid_list *parent,
814 	int number, const char *name, int kind, void *arg1, intmax_t arg2,
815 	int (*handler)(SYSCTL_HANDLER_ARGS), const char *fmt, const char *descr,
816 	const char *label)
817 {
818 	struct sysctl_oid *oidp;
819 
820 	/* You have to hook up somewhere.. */
821 	if (parent == NULL)
822 		return(NULL);
823 	/* Check if the node already exists, otherwise create it */
824 	SYSCTL_WLOCK();
825 	oidp = sysctl_find_oidname(name, parent);
826 	if (oidp != NULL) {
827 		if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
828 			oidp->oid_refcnt++;
829 			/* Update the context */
830 			if (clist != NULL)
831 				sysctl_ctx_entry_add(clist, oidp);
832 			SYSCTL_WUNLOCK();
833 			return (oidp);
834 		} else {
835 			sysctl_warn_reuse(__func__, oidp);
836 			SYSCTL_WUNLOCK();
837 			return (NULL);
838 		}
839 	}
840 	oidp = malloc(sizeof(struct sysctl_oid), M_SYSCTLOID, M_WAITOK|M_ZERO);
841 	oidp->oid_parent = parent;
842 	SLIST_INIT(&oidp->oid_children);
843 	oidp->oid_number = number;
844 	oidp->oid_refcnt = 1;
845 	oidp->oid_name = strdup(name, M_SYSCTLOID);
846 	oidp->oid_handler = handler;
847 	oidp->oid_kind = CTLFLAG_DYN | kind;
848 	oidp->oid_arg1 = arg1;
849 	oidp->oid_arg2 = arg2;
850 	oidp->oid_fmt = fmt;
851 	if (descr != NULL)
852 		oidp->oid_descr = strdup(descr, M_SYSCTLOID);
853 	if (label != NULL)
854 		oidp->oid_label = strdup(label, M_SYSCTLOID);
855 	/* Update the context, if used */
856 	if (clist != NULL)
857 		sysctl_ctx_entry_add(clist, oidp);
858 	/* Register this oid */
859 	sysctl_register_oid(oidp);
860 	SYSCTL_WUNLOCK();
861 	return (oidp);
862 }
863 
864 /*
865  * Rename an existing oid.
866  */
867 void
868 sysctl_rename_oid(struct sysctl_oid *oidp, const char *name)
869 {
870 	char *newname;
871 	char *oldname;
872 
873 	newname = strdup(name, M_SYSCTLOID);
874 	SYSCTL_WLOCK();
875 	oldname = __DECONST(char *, oidp->oid_name);
876 	oidp->oid_name = newname;
877 	SYSCTL_WUNLOCK();
878 	free(oldname, M_SYSCTLOID);
879 }
880 
881 /*
882  * Reparent an existing oid.
883  */
884 int
885 sysctl_move_oid(struct sysctl_oid *oid, struct sysctl_oid_list *parent)
886 {
887 	struct sysctl_oid *oidp;
888 
889 	SYSCTL_WLOCK();
890 	if (oid->oid_parent == parent) {
891 		SYSCTL_WUNLOCK();
892 		return (0);
893 	}
894 	oidp = sysctl_find_oidname(oid->oid_name, parent);
895 	if (oidp != NULL) {
896 		SYSCTL_WUNLOCK();
897 		return (EEXIST);
898 	}
899 	sysctl_unregister_oid(oid);
900 	oid->oid_parent = parent;
901 	oid->oid_number = OID_AUTO;
902 	sysctl_register_oid(oid);
903 	SYSCTL_WUNLOCK();
904 	return (0);
905 }
906 
907 /*
908  * Register the kernel's oids on startup.
909  */
910 SET_DECLARE(sysctl_set, struct sysctl_oid);
911 
912 static void
913 sysctl_register_all(void *arg)
914 {
915 	struct sysctl_oid **oidp;
916 
917 	sx_init(&sysctlmemlock, "sysctl mem");
918 	SYSCTL_INIT();
919 	SYSCTL_WLOCK();
920 	SET_FOREACH(oidp, sysctl_set)
921 		sysctl_register_oid(*oidp);
922 	SYSCTL_WUNLOCK();
923 }
924 SYSINIT(sysctl, SI_SUB_KMEM, SI_ORDER_FIRST, sysctl_register_all, NULL);
925 
926 /*
927  * "Staff-functions"
928  *
929  * These functions implement a presently undocumented interface
930  * used by the sysctl program to walk the tree, and get the type
931  * so it can print the value.
932  * This interface is under work and consideration, and should probably
933  * be killed with a big axe by the first person who can find the time.
934  * (be aware though, that the proper interface isn't as obvious as it
935  * may seem, there are various conflicting requirements.
936  *
937  * {0,0}	printf the entire MIB-tree.
938  * {0,1,...}	return the name of the "..." OID.
939  * {0,2,...}	return the next OID.
940  * {0,3}	return the OID of the name in "new"
941  * {0,4,...}	return the kind & format info for the "..." OID.
942  * {0,5,...}	return the description of the "..." OID.
943  * {0,6,...}	return the aggregation label of the "..." OID.
944  */
945 
946 #ifdef SYSCTL_DEBUG
947 static void
948 sysctl_sysctl_debug_dump_node(struct sysctl_oid_list *l, int i)
949 {
950 	int k;
951 	struct sysctl_oid *oidp;
952 
953 	SYSCTL_ASSERT_LOCKED();
954 	SLIST_FOREACH(oidp, l, oid_link) {
955 
956 		for (k=0; k<i; k++)
957 			printf(" ");
958 
959 		printf("%d %s ", oidp->oid_number, oidp->oid_name);
960 
961 		printf("%c%c",
962 			oidp->oid_kind & CTLFLAG_RD ? 'R':' ',
963 			oidp->oid_kind & CTLFLAG_WR ? 'W':' ');
964 
965 		if (oidp->oid_handler)
966 			printf(" *Handler");
967 
968 		switch (oidp->oid_kind & CTLTYPE) {
969 			case CTLTYPE_NODE:
970 				printf(" Node\n");
971 				if (!oidp->oid_handler) {
972 					sysctl_sysctl_debug_dump_node(
973 					    SYSCTL_CHILDREN(oidp), i + 2);
974 				}
975 				break;
976 			case CTLTYPE_INT:    printf(" Int\n"); break;
977 			case CTLTYPE_UINT:   printf(" u_int\n"); break;
978 			case CTLTYPE_LONG:   printf(" Long\n"); break;
979 			case CTLTYPE_ULONG:  printf(" u_long\n"); break;
980 			case CTLTYPE_STRING: printf(" String\n"); break;
981 			case CTLTYPE_S8:     printf(" int8_t\n"); break;
982 			case CTLTYPE_S16:    printf(" int16_t\n"); break;
983 			case CTLTYPE_S32:    printf(" int32_t\n"); break;
984 			case CTLTYPE_S64:    printf(" int64_t\n"); break;
985 			case CTLTYPE_U8:     printf(" uint8_t\n"); break;
986 			case CTLTYPE_U16:    printf(" uint16_t\n"); break;
987 			case CTLTYPE_U32:    printf(" uint32_t\n"); break;
988 			case CTLTYPE_U64:    printf(" uint64_t\n"); break;
989 			case CTLTYPE_OPAQUE: printf(" Opaque/struct\n"); break;
990 			default:	     printf("\n");
991 		}
992 
993 	}
994 }
995 
996 static int
997 sysctl_sysctl_debug(SYSCTL_HANDLER_ARGS)
998 {
999 	struct rm_priotracker tracker;
1000 	int error;
1001 
1002 	error = priv_check(req->td, PRIV_SYSCTL_DEBUG);
1003 	if (error)
1004 		return (error);
1005 	SYSCTL_RLOCK(&tracker);
1006 	sysctl_sysctl_debug_dump_node(&sysctl__children, 0);
1007 	SYSCTL_RUNLOCK(&tracker);
1008 	return (ENOENT);
1009 }
1010 
1011 SYSCTL_PROC(_sysctl, 0, debug, CTLTYPE_STRING|CTLFLAG_RD|CTLFLAG_MPSAFE,
1012 	0, 0, sysctl_sysctl_debug, "-", "");
1013 #endif
1014 
1015 static int
1016 sysctl_sysctl_name(SYSCTL_HANDLER_ARGS)
1017 {
1018 	int *name = (int *) arg1;
1019 	u_int namelen = arg2;
1020 	int error = 0;
1021 	struct sysctl_oid *oid;
1022 	struct sysctl_oid_list *lsp = &sysctl__children, *lsp2;
1023 	struct rm_priotracker tracker;
1024 	char buf[10];
1025 
1026 	SYSCTL_RLOCK(&tracker);
1027 	while (namelen) {
1028 		if (!lsp) {
1029 			snprintf(buf,sizeof(buf),"%d",*name);
1030 			if (req->oldidx)
1031 				error = SYSCTL_OUT(req, ".", 1);
1032 			if (!error)
1033 				error = SYSCTL_OUT(req, buf, strlen(buf));
1034 			if (error)
1035 				goto out;
1036 			namelen--;
1037 			name++;
1038 			continue;
1039 		}
1040 		lsp2 = NULL;
1041 		SLIST_FOREACH(oid, lsp, oid_link) {
1042 			if (oid->oid_number != *name)
1043 				continue;
1044 
1045 			if (req->oldidx)
1046 				error = SYSCTL_OUT(req, ".", 1);
1047 			if (!error)
1048 				error = SYSCTL_OUT(req, oid->oid_name,
1049 					strlen(oid->oid_name));
1050 			if (error)
1051 				goto out;
1052 
1053 			namelen--;
1054 			name++;
1055 
1056 			if ((oid->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1057 				break;
1058 
1059 			if (oid->oid_handler)
1060 				break;
1061 
1062 			lsp2 = SYSCTL_CHILDREN(oid);
1063 			break;
1064 		}
1065 		lsp = lsp2;
1066 	}
1067 	error = SYSCTL_OUT(req, "", 1);
1068  out:
1069 	SYSCTL_RUNLOCK(&tracker);
1070 	return (error);
1071 }
1072 
1073 /*
1074  * XXXRW/JA: Shouldn't return name data for nodes that we don't permit in
1075  * capability mode.
1076  */
1077 static SYSCTL_NODE(_sysctl, 1, name, CTLFLAG_RD | CTLFLAG_MPSAFE | CTLFLAG_CAPRD,
1078     sysctl_sysctl_name, "");
1079 
1080 static int
1081 sysctl_sysctl_next_ls(struct sysctl_oid_list *lsp, int *name, u_int namelen,
1082 	int *next, int *len, int level, struct sysctl_oid **oidpp)
1083 {
1084 	struct sysctl_oid *oidp;
1085 
1086 	SYSCTL_ASSERT_LOCKED();
1087 	*len = level;
1088 	SLIST_FOREACH(oidp, lsp, oid_link) {
1089 		*next = oidp->oid_number;
1090 		*oidpp = oidp;
1091 
1092 		if ((oidp->oid_kind & (CTLFLAG_SKIP | CTLFLAG_DORMANT)) != 0)
1093 			continue;
1094 
1095 		if (!namelen) {
1096 			if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1097 				return (0);
1098 			if (oidp->oid_handler)
1099 				/* We really should call the handler here...*/
1100 				return (0);
1101 			lsp = SYSCTL_CHILDREN(oidp);
1102 			if (!sysctl_sysctl_next_ls(lsp, 0, 0, next+1,
1103 				len, level+1, oidpp))
1104 				return (0);
1105 			goto emptynode;
1106 		}
1107 
1108 		if (oidp->oid_number < *name)
1109 			continue;
1110 
1111 		if (oidp->oid_number > *name) {
1112 			if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1113 				return (0);
1114 			if (oidp->oid_handler)
1115 				return (0);
1116 			lsp = SYSCTL_CHILDREN(oidp);
1117 			if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1,
1118 				next+1, len, level+1, oidpp))
1119 				return (0);
1120 			goto next;
1121 		}
1122 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1123 			continue;
1124 
1125 		if (oidp->oid_handler)
1126 			continue;
1127 
1128 		lsp = SYSCTL_CHILDREN(oidp);
1129 		if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1, next+1,
1130 			len, level+1, oidpp))
1131 			return (0);
1132 	next:
1133 		namelen = 1;
1134 	emptynode:
1135 		*len = level;
1136 	}
1137 	return (1);
1138 }
1139 
1140 static int
1141 sysctl_sysctl_next(SYSCTL_HANDLER_ARGS)
1142 {
1143 	int *name = (int *) arg1;
1144 	u_int namelen = arg2;
1145 	int i, j, error;
1146 	struct sysctl_oid *oid;
1147 	struct sysctl_oid_list *lsp = &sysctl__children;
1148 	struct rm_priotracker tracker;
1149 	int newoid[CTL_MAXNAME];
1150 
1151 	SYSCTL_RLOCK(&tracker);
1152 	i = sysctl_sysctl_next_ls(lsp, name, namelen, newoid, &j, 1, &oid);
1153 	SYSCTL_RUNLOCK(&tracker);
1154 	if (i)
1155 		return (ENOENT);
1156 	error = SYSCTL_OUT(req, newoid, j * sizeof (int));
1157 	return (error);
1158 }
1159 
1160 /*
1161  * XXXRW/JA: Shouldn't return next data for nodes that we don't permit in
1162  * capability mode.
1163  */
1164 static SYSCTL_NODE(_sysctl, 2, next, CTLFLAG_RD | CTLFLAG_MPSAFE | CTLFLAG_CAPRD,
1165     sysctl_sysctl_next, "");
1166 
1167 static int
1168 name2oid(char *name, int *oid, int *len, struct sysctl_oid **oidpp)
1169 {
1170 	struct sysctl_oid *oidp;
1171 	struct sysctl_oid_list *lsp = &sysctl__children;
1172 	char *p;
1173 
1174 	SYSCTL_ASSERT_LOCKED();
1175 
1176 	for (*len = 0; *len < CTL_MAXNAME;) {
1177 		p = strsep(&name, ".");
1178 
1179 		oidp = SLIST_FIRST(lsp);
1180 		for (;; oidp = SLIST_NEXT(oidp, oid_link)) {
1181 			if (oidp == NULL)
1182 				return (ENOENT);
1183 			if (strcmp(p, oidp->oid_name) == 0)
1184 				break;
1185 		}
1186 		*oid++ = oidp->oid_number;
1187 		(*len)++;
1188 
1189 		if (name == NULL || *name == '\0') {
1190 			if (oidpp)
1191 				*oidpp = oidp;
1192 			return (0);
1193 		}
1194 
1195 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1196 			break;
1197 
1198 		if (oidp->oid_handler)
1199 			break;
1200 
1201 		lsp = SYSCTL_CHILDREN(oidp);
1202 	}
1203 	return (ENOENT);
1204 }
1205 
1206 static int
1207 sysctl_sysctl_name2oid(SYSCTL_HANDLER_ARGS)
1208 {
1209 	char *p;
1210 	int error, oid[CTL_MAXNAME], len = 0;
1211 	struct sysctl_oid *op = NULL;
1212 	struct rm_priotracker tracker;
1213 	char buf[32];
1214 
1215 	if (!req->newlen)
1216 		return (ENOENT);
1217 	if (req->newlen >= MAXPATHLEN)	/* XXX arbitrary, undocumented */
1218 		return (ENAMETOOLONG);
1219 
1220 	p = buf;
1221 	if (req->newlen >= sizeof(buf))
1222 		p = malloc(req->newlen+1, M_SYSCTL, M_WAITOK);
1223 
1224 	error = SYSCTL_IN(req, p, req->newlen);
1225 	if (error) {
1226 		if (p != buf)
1227 			free(p, M_SYSCTL);
1228 		return (error);
1229 	}
1230 
1231 	p [req->newlen] = '\0';
1232 
1233 	SYSCTL_RLOCK(&tracker);
1234 	error = name2oid(p, oid, &len, &op);
1235 	SYSCTL_RUNLOCK(&tracker);
1236 
1237 	if (p != buf)
1238 		free(p, M_SYSCTL);
1239 
1240 	if (error)
1241 		return (error);
1242 
1243 	error = SYSCTL_OUT(req, oid, len * sizeof *oid);
1244 	return (error);
1245 }
1246 
1247 /*
1248  * XXXRW/JA: Shouldn't return name2oid data for nodes that we don't permit in
1249  * capability mode.
1250  */
1251 SYSCTL_PROC(_sysctl, 3, name2oid,
1252     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_MPSAFE
1253     | CTLFLAG_CAPRW, 0, 0, sysctl_sysctl_name2oid, "I", "");
1254 
1255 static int
1256 sysctl_sysctl_oidfmt(SYSCTL_HANDLER_ARGS)
1257 {
1258 	struct sysctl_oid *oid;
1259 	struct rm_priotracker tracker;
1260 	int error;
1261 
1262 	SYSCTL_RLOCK(&tracker);
1263 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1264 	if (error)
1265 		goto out;
1266 
1267 	if (oid->oid_fmt == NULL) {
1268 		error = ENOENT;
1269 		goto out;
1270 	}
1271 	error = SYSCTL_OUT(req, &oid->oid_kind, sizeof(oid->oid_kind));
1272 	if (error)
1273 		goto out;
1274 	error = SYSCTL_OUT(req, oid->oid_fmt, strlen(oid->oid_fmt) + 1);
1275  out:
1276 	SYSCTL_RUNLOCK(&tracker);
1277 	return (error);
1278 }
1279 
1280 
1281 static SYSCTL_NODE(_sysctl, 4, oidfmt, CTLFLAG_RD|CTLFLAG_MPSAFE|CTLFLAG_CAPRD,
1282     sysctl_sysctl_oidfmt, "");
1283 
1284 static int
1285 sysctl_sysctl_oiddescr(SYSCTL_HANDLER_ARGS)
1286 {
1287 	struct sysctl_oid *oid;
1288 	struct rm_priotracker tracker;
1289 	int error;
1290 
1291 	SYSCTL_RLOCK(&tracker);
1292 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1293 	if (error)
1294 		goto out;
1295 
1296 	if (oid->oid_descr == NULL) {
1297 		error = ENOENT;
1298 		goto out;
1299 	}
1300 	error = SYSCTL_OUT(req, oid->oid_descr, strlen(oid->oid_descr) + 1);
1301  out:
1302 	SYSCTL_RUNLOCK(&tracker);
1303 	return (error);
1304 }
1305 
1306 static SYSCTL_NODE(_sysctl, 5, oiddescr, CTLFLAG_RD|CTLFLAG_MPSAFE|CTLFLAG_CAPRD,
1307     sysctl_sysctl_oiddescr, "");
1308 
1309 static int
1310 sysctl_sysctl_oidlabel(SYSCTL_HANDLER_ARGS)
1311 {
1312 	struct sysctl_oid *oid;
1313 	struct rm_priotracker tracker;
1314 	int error;
1315 
1316 	SYSCTL_RLOCK(&tracker);
1317 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1318 	if (error)
1319 		goto out;
1320 
1321 	if (oid->oid_label == NULL) {
1322 		error = ENOENT;
1323 		goto out;
1324 	}
1325 	error = SYSCTL_OUT(req, oid->oid_label, strlen(oid->oid_label) + 1);
1326  out:
1327 	SYSCTL_RUNLOCK(&tracker);
1328 	return (error);
1329 }
1330 
1331 static SYSCTL_NODE(_sysctl, 6, oidlabel,
1332     CTLFLAG_RD | CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_oidlabel, "");
1333 
1334 /*
1335  * Default "handler" functions.
1336  */
1337 
1338 /*
1339  * Handle a bool.
1340  * Two cases:
1341  *     a variable:  point arg1 at it.
1342  *     a constant:  pass it in arg2.
1343  */
1344 
1345 int
1346 sysctl_handle_bool(SYSCTL_HANDLER_ARGS)
1347 {
1348 	uint8_t temp;
1349 	int error;
1350 
1351 	/*
1352 	 * Attempt to get a coherent snapshot by making a copy of the data.
1353 	 */
1354 	if (arg1)
1355 		temp = *(bool *)arg1 ? 1 : 0;
1356 	else
1357 		temp = arg2 ? 1 : 0;
1358 
1359 	error = SYSCTL_OUT(req, &temp, sizeof(temp));
1360 	if (error || !req->newptr)
1361 		return (error);
1362 
1363 	if (!arg1)
1364 		error = EPERM;
1365 	else {
1366 		error = SYSCTL_IN(req, &temp, sizeof(temp));
1367 		if (!error)
1368 			*(bool *)arg1 = temp ? 1 : 0;
1369 	}
1370 	return (error);
1371 }
1372 
1373 /*
1374  * Handle an int8_t, signed or unsigned.
1375  * Two cases:
1376  *     a variable:  point arg1 at it.
1377  *     a constant:  pass it in arg2.
1378  */
1379 
1380 int
1381 sysctl_handle_8(SYSCTL_HANDLER_ARGS)
1382 {
1383 	int8_t tmpout;
1384 	int error = 0;
1385 
1386 	/*
1387 	 * Attempt to get a coherent snapshot by making a copy of the data.
1388 	 */
1389 	if (arg1)
1390 		tmpout = *(int8_t *)arg1;
1391 	else
1392 		tmpout = arg2;
1393 	error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout));
1394 
1395 	if (error || !req->newptr)
1396 		return (error);
1397 
1398 	if (!arg1)
1399 		error = EPERM;
1400 	else
1401 		error = SYSCTL_IN(req, arg1, sizeof(tmpout));
1402 	return (error);
1403 }
1404 
1405 /*
1406  * Handle an int16_t, signed or unsigned.
1407  * Two cases:
1408  *     a variable:  point arg1 at it.
1409  *     a constant:  pass it in arg2.
1410  */
1411 
1412 int
1413 sysctl_handle_16(SYSCTL_HANDLER_ARGS)
1414 {
1415 	int16_t tmpout;
1416 	int error = 0;
1417 
1418 	/*
1419 	 * Attempt to get a coherent snapshot by making a copy of the data.
1420 	 */
1421 	if (arg1)
1422 		tmpout = *(int16_t *)arg1;
1423 	else
1424 		tmpout = arg2;
1425 	error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout));
1426 
1427 	if (error || !req->newptr)
1428 		return (error);
1429 
1430 	if (!arg1)
1431 		error = EPERM;
1432 	else
1433 		error = SYSCTL_IN(req, arg1, sizeof(tmpout));
1434 	return (error);
1435 }
1436 
1437 /*
1438  * Handle an int32_t, signed or unsigned.
1439  * Two cases:
1440  *     a variable:  point arg1 at it.
1441  *     a constant:  pass it in arg2.
1442  */
1443 
1444 int
1445 sysctl_handle_32(SYSCTL_HANDLER_ARGS)
1446 {
1447 	int32_t tmpout;
1448 	int error = 0;
1449 
1450 	/*
1451 	 * Attempt to get a coherent snapshot by making a copy of the data.
1452 	 */
1453 	if (arg1)
1454 		tmpout = *(int32_t *)arg1;
1455 	else
1456 		tmpout = arg2;
1457 	error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout));
1458 
1459 	if (error || !req->newptr)
1460 		return (error);
1461 
1462 	if (!arg1)
1463 		error = EPERM;
1464 	else
1465 		error = SYSCTL_IN(req, arg1, sizeof(tmpout));
1466 	return (error);
1467 }
1468 
1469 /*
1470  * Handle an int, signed or unsigned.
1471  * Two cases:
1472  *     a variable:  point arg1 at it.
1473  *     a constant:  pass it in arg2.
1474  */
1475 
1476 int
1477 sysctl_handle_int(SYSCTL_HANDLER_ARGS)
1478 {
1479 	int tmpout, error = 0;
1480 
1481 	/*
1482 	 * Attempt to get a coherent snapshot by making a copy of the data.
1483 	 */
1484 	if (arg1)
1485 		tmpout = *(int *)arg1;
1486 	else
1487 		tmpout = arg2;
1488 	error = SYSCTL_OUT(req, &tmpout, sizeof(int));
1489 
1490 	if (error || !req->newptr)
1491 		return (error);
1492 
1493 	if (!arg1)
1494 		error = EPERM;
1495 	else
1496 		error = SYSCTL_IN(req, arg1, sizeof(int));
1497 	return (error);
1498 }
1499 
1500 /*
1501  * Based on on sysctl_handle_int() convert milliseconds into ticks.
1502  * Note: this is used by TCP.
1503  */
1504 
1505 int
1506 sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
1507 {
1508 	int error, s, tt;
1509 
1510 	tt = *(int *)arg1;
1511 	s = (int)((int64_t)tt * 1000 / hz);
1512 
1513 	error = sysctl_handle_int(oidp, &s, 0, req);
1514 	if (error || !req->newptr)
1515 		return (error);
1516 
1517 	tt = (int)((int64_t)s * hz / 1000);
1518 	if (tt < 1)
1519 		return (EINVAL);
1520 
1521 	*(int *)arg1 = tt;
1522 	return (0);
1523 }
1524 
1525 
1526 /*
1527  * Handle a long, signed or unsigned.
1528  * Two cases:
1529  *     a variable:  point arg1 at it.
1530  *     a constant:  pass it in arg2.
1531  */
1532 
1533 int
1534 sysctl_handle_long(SYSCTL_HANDLER_ARGS)
1535 {
1536 	int error = 0;
1537 	long tmplong;
1538 #ifdef SCTL_MASK32
1539 	int tmpint;
1540 #endif
1541 
1542 	/*
1543 	 * Attempt to get a coherent snapshot by making a copy of the data.
1544 	 */
1545 	if (arg1)
1546 		tmplong = *(long *)arg1;
1547 	else
1548 		tmplong = arg2;
1549 #ifdef SCTL_MASK32
1550 	if (req->flags & SCTL_MASK32) {
1551 		tmpint = tmplong;
1552 		error = SYSCTL_OUT(req, &tmpint, sizeof(int));
1553 	} else
1554 #endif
1555 		error = SYSCTL_OUT(req, &tmplong, sizeof(long));
1556 
1557 	if (error || !req->newptr)
1558 		return (error);
1559 
1560 	if (!arg1)
1561 		error = EPERM;
1562 #ifdef SCTL_MASK32
1563 	else if (req->flags & SCTL_MASK32) {
1564 		error = SYSCTL_IN(req, &tmpint, sizeof(int));
1565 		*(long *)arg1 = (long)tmpint;
1566 	}
1567 #endif
1568 	else
1569 		error = SYSCTL_IN(req, arg1, sizeof(long));
1570 	return (error);
1571 }
1572 
1573 /*
1574  * Handle a 64 bit int, signed or unsigned.
1575  * Two cases:
1576  *     a variable:  point arg1 at it.
1577  *     a constant:  pass it in arg2.
1578  */
1579 int
1580 sysctl_handle_64(SYSCTL_HANDLER_ARGS)
1581 {
1582 	int error = 0;
1583 	uint64_t tmpout;
1584 
1585 	/*
1586 	 * Attempt to get a coherent snapshot by making a copy of the data.
1587 	 */
1588 	if (arg1)
1589 		tmpout = *(uint64_t *)arg1;
1590 	else
1591 		tmpout = arg2;
1592 	error = SYSCTL_OUT(req, &tmpout, sizeof(uint64_t));
1593 
1594 	if (error || !req->newptr)
1595 		return (error);
1596 
1597 	if (!arg1)
1598 		error = EPERM;
1599 	else
1600 		error = SYSCTL_IN(req, arg1, sizeof(uint64_t));
1601 	return (error);
1602 }
1603 
1604 /*
1605  * Handle our generic '\0' terminated 'C' string.
1606  * Two cases:
1607  * 	a variable string:  point arg1 at it, arg2 is max length.
1608  * 	a constant string:  point arg1 at it, arg2 is zero.
1609  */
1610 
1611 int
1612 sysctl_handle_string(SYSCTL_HANDLER_ARGS)
1613 {
1614 	size_t outlen;
1615 	int error = 0, ro_string = 0;
1616 
1617 	/*
1618 	 * A zero-length buffer indicates a fixed size read-only
1619 	 * string:
1620 	 */
1621 	if (arg2 == 0) {
1622 		arg2 = strlen((char *)arg1) + 1;
1623 		ro_string = 1;
1624 	}
1625 
1626 	if (req->oldptr != NULL) {
1627 		char *tmparg;
1628 
1629 		if (ro_string) {
1630 			tmparg = arg1;
1631 		} else {
1632 			/* try to make a coherent snapshot of the string */
1633 			tmparg = malloc(arg2, M_SYSCTLTMP, M_WAITOK);
1634 			memcpy(tmparg, arg1, arg2);
1635 		}
1636 
1637 		outlen = strnlen(tmparg, arg2 - 1) + 1;
1638 		error = SYSCTL_OUT(req, tmparg, outlen);
1639 
1640 		if (!ro_string)
1641 			free(tmparg, M_SYSCTLTMP);
1642 	} else {
1643 		outlen = strnlen((char *)arg1, arg2 - 1) + 1;
1644 		error = SYSCTL_OUT(req, NULL, outlen);
1645 	}
1646 	if (error || !req->newptr)
1647 		return (error);
1648 
1649 	if ((req->newlen - req->newidx) >= arg2) {
1650 		error = EINVAL;
1651 	} else {
1652 		arg2 = (req->newlen - req->newidx);
1653 		error = SYSCTL_IN(req, arg1, arg2);
1654 		((char *)arg1)[arg2] = '\0';
1655 	}
1656 	return (error);
1657 }
1658 
1659 /*
1660  * Handle any kind of opaque data.
1661  * arg1 points to it, arg2 is the size.
1662  */
1663 
1664 int
1665 sysctl_handle_opaque(SYSCTL_HANDLER_ARGS)
1666 {
1667 	int error, tries;
1668 	u_int generation;
1669 	struct sysctl_req req2;
1670 
1671 	/*
1672 	 * Attempt to get a coherent snapshot, by using the thread
1673 	 * pre-emption counter updated from within mi_switch() to
1674 	 * determine if we were pre-empted during a bcopy() or
1675 	 * copyout(). Make 3 attempts at doing this before giving up.
1676 	 * If we encounter an error, stop immediately.
1677 	 */
1678 	tries = 0;
1679 	req2 = *req;
1680 retry:
1681 	generation = curthread->td_generation;
1682 	error = SYSCTL_OUT(req, arg1, arg2);
1683 	if (error)
1684 		return (error);
1685 	tries++;
1686 	if (generation != curthread->td_generation && tries < 3) {
1687 		*req = req2;
1688 		goto retry;
1689 	}
1690 
1691 	error = SYSCTL_IN(req, arg1, arg2);
1692 
1693 	return (error);
1694 }
1695 
1696 /*
1697  * Based on on sysctl_handle_int() convert microseconds to a sbintime.
1698  */
1699 int
1700 sysctl_usec_to_sbintime(SYSCTL_HANDLER_ARGS)
1701 {
1702 	int error;
1703 	int64_t tt;
1704 	sbintime_t sb;
1705 
1706 	tt = *(int64_t *)arg1;
1707 	sb = sbttous(tt);
1708 
1709 	error = sysctl_handle_64(oidp, &sb, 0, req);
1710 	if (error || !req->newptr)
1711 		return (error);
1712 
1713 	tt = ustosbt(sb);
1714 	*(int64_t *)arg1 = tt;
1715 
1716 	return (0);
1717 }
1718 
1719 /*
1720  * Based on on sysctl_handle_int() convert milliseconds to a sbintime.
1721  */
1722 int
1723 sysctl_msec_to_sbintime(SYSCTL_HANDLER_ARGS)
1724 {
1725 	int error;
1726 	int64_t tt;
1727 	sbintime_t sb;
1728 
1729 	tt = *(int64_t *)arg1;
1730 	sb = sbttoms(tt);
1731 
1732 	error = sysctl_handle_64(oidp, &sb, 0, req);
1733 	if (error || !req->newptr)
1734 		return (error);
1735 
1736 	tt = mstosbt(sb);
1737 	*(int64_t *)arg1 = tt;
1738 
1739 	return (0);
1740 }
1741 
1742 
1743 /*
1744  * Transfer functions to/from kernel space.
1745  * XXX: rather untested at this point
1746  */
1747 static int
1748 sysctl_old_kernel(struct sysctl_req *req, const void *p, size_t l)
1749 {
1750 	size_t i = 0;
1751 
1752 	if (req->oldptr) {
1753 		i = l;
1754 		if (req->oldlen <= req->oldidx)
1755 			i = 0;
1756 		else
1757 			if (i > req->oldlen - req->oldidx)
1758 				i = req->oldlen - req->oldidx;
1759 		if (i > 0)
1760 			bcopy(p, (char *)req->oldptr + req->oldidx, i);
1761 	}
1762 	req->oldidx += l;
1763 	if (req->oldptr && i != l)
1764 		return (ENOMEM);
1765 	return (0);
1766 }
1767 
1768 static int
1769 sysctl_new_kernel(struct sysctl_req *req, void *p, size_t l)
1770 {
1771 	if (!req->newptr)
1772 		return (0);
1773 	if (req->newlen - req->newidx < l)
1774 		return (EINVAL);
1775 	bcopy((const char *)req->newptr + req->newidx, p, l);
1776 	req->newidx += l;
1777 	return (0);
1778 }
1779 
1780 int
1781 kernel_sysctl(struct thread *td, int *name, u_int namelen, void *old,
1782     size_t *oldlenp, void *new, size_t newlen, size_t *retval, int flags)
1783 {
1784 	int error = 0;
1785 	struct sysctl_req req;
1786 
1787 	bzero(&req, sizeof req);
1788 
1789 	req.td = td;
1790 	req.flags = flags;
1791 
1792 	if (oldlenp) {
1793 		req.oldlen = *oldlenp;
1794 	}
1795 	req.validlen = req.oldlen;
1796 
1797 	if (old) {
1798 		req.oldptr= old;
1799 	}
1800 
1801 	if (new != NULL) {
1802 		req.newlen = newlen;
1803 		req.newptr = new;
1804 	}
1805 
1806 	req.oldfunc = sysctl_old_kernel;
1807 	req.newfunc = sysctl_new_kernel;
1808 	req.lock = REQ_UNWIRED;
1809 
1810 	error = sysctl_root(0, name, namelen, &req);
1811 
1812 	if (req.lock == REQ_WIRED && req.validlen > 0)
1813 		vsunlock(req.oldptr, req.validlen);
1814 
1815 	if (error && error != ENOMEM)
1816 		return (error);
1817 
1818 	if (retval) {
1819 		if (req.oldptr && req.oldidx > req.validlen)
1820 			*retval = req.validlen;
1821 		else
1822 			*retval = req.oldidx;
1823 	}
1824 	return (error);
1825 }
1826 
1827 int
1828 kernel_sysctlbyname(struct thread *td, char *name, void *old, size_t *oldlenp,
1829     void *new, size_t newlen, size_t *retval, int flags)
1830 {
1831         int oid[CTL_MAXNAME];
1832         size_t oidlen, plen;
1833 	int error;
1834 
1835 	oid[0] = 0;		/* sysctl internal magic */
1836 	oid[1] = 3;		/* name2oid */
1837 	oidlen = sizeof(oid);
1838 
1839 	error = kernel_sysctl(td, oid, 2, oid, &oidlen,
1840 	    (void *)name, strlen(name), &plen, flags);
1841 	if (error)
1842 		return (error);
1843 
1844 	error = kernel_sysctl(td, oid, plen / sizeof(int), old, oldlenp,
1845 	    new, newlen, retval, flags);
1846 	return (error);
1847 }
1848 
1849 /*
1850  * Transfer function to/from user space.
1851  */
1852 static int
1853 sysctl_old_user(struct sysctl_req *req, const void *p, size_t l)
1854 {
1855 	size_t i, len, origidx;
1856 	int error;
1857 
1858 	origidx = req->oldidx;
1859 	req->oldidx += l;
1860 	if (req->oldptr == NULL)
1861 		return (0);
1862 	/*
1863 	 * If we have not wired the user supplied buffer and we are currently
1864 	 * holding locks, drop a witness warning, as it's possible that
1865 	 * write operations to the user page can sleep.
1866 	 */
1867 	if (req->lock != REQ_WIRED)
1868 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
1869 		    "sysctl_old_user()");
1870 	i = l;
1871 	len = req->validlen;
1872 	if (len <= origidx)
1873 		i = 0;
1874 	else {
1875 		if (i > len - origidx)
1876 			i = len - origidx;
1877 		if (req->lock == REQ_WIRED) {
1878 			error = copyout_nofault(p, (char *)req->oldptr +
1879 			    origidx, i);
1880 		} else
1881 			error = copyout(p, (char *)req->oldptr + origidx, i);
1882 		if (error != 0)
1883 			return (error);
1884 	}
1885 	if (i < l)
1886 		return (ENOMEM);
1887 	return (0);
1888 }
1889 
1890 static int
1891 sysctl_new_user(struct sysctl_req *req, void *p, size_t l)
1892 {
1893 	int error;
1894 
1895 	if (!req->newptr)
1896 		return (0);
1897 	if (req->newlen - req->newidx < l)
1898 		return (EINVAL);
1899 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
1900 	    "sysctl_new_user()");
1901 	error = copyin((const char *)req->newptr + req->newidx, p, l);
1902 	req->newidx += l;
1903 	return (error);
1904 }
1905 
1906 /*
1907  * Wire the user space destination buffer.  If set to a value greater than
1908  * zero, the len parameter limits the maximum amount of wired memory.
1909  */
1910 int
1911 sysctl_wire_old_buffer(struct sysctl_req *req, size_t len)
1912 {
1913 	int ret;
1914 	size_t wiredlen;
1915 
1916 	wiredlen = (len > 0 && len < req->oldlen) ? len : req->oldlen;
1917 	ret = 0;
1918 	if (req->lock != REQ_WIRED && req->oldptr &&
1919 	    req->oldfunc == sysctl_old_user) {
1920 		if (wiredlen != 0) {
1921 			ret = vslock(req->oldptr, wiredlen);
1922 			if (ret != 0) {
1923 				if (ret != ENOMEM)
1924 					return (ret);
1925 				wiredlen = 0;
1926 			}
1927 		}
1928 		req->lock = REQ_WIRED;
1929 		req->validlen = wiredlen;
1930 	}
1931 	return (0);
1932 }
1933 
1934 int
1935 sysctl_find_oid(int *name, u_int namelen, struct sysctl_oid **noid,
1936     int *nindx, struct sysctl_req *req)
1937 {
1938 	struct sysctl_oid_list *lsp;
1939 	struct sysctl_oid *oid;
1940 	int indx;
1941 
1942 	SYSCTL_ASSERT_LOCKED();
1943 	lsp = &sysctl__children;
1944 	indx = 0;
1945 	while (indx < CTL_MAXNAME) {
1946 		SLIST_FOREACH(oid, lsp, oid_link) {
1947 			if (oid->oid_number == name[indx])
1948 				break;
1949 		}
1950 		if (oid == NULL)
1951 			return (ENOENT);
1952 
1953 		indx++;
1954 		if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1955 			if (oid->oid_handler != NULL || indx == namelen) {
1956 				*noid = oid;
1957 				if (nindx != NULL)
1958 					*nindx = indx;
1959 				KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
1960 				    ("%s found DYING node %p", __func__, oid));
1961 				return (0);
1962 			}
1963 			lsp = SYSCTL_CHILDREN(oid);
1964 		} else if (indx == namelen) {
1965 			if ((oid->oid_kind & CTLFLAG_DORMANT) != 0)
1966 				return (ENOENT);
1967 			*noid = oid;
1968 			if (nindx != NULL)
1969 				*nindx = indx;
1970 			KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
1971 			    ("%s found DYING node %p", __func__, oid));
1972 			return (0);
1973 		} else {
1974 			return (ENOTDIR);
1975 		}
1976 	}
1977 	return (ENOENT);
1978 }
1979 
1980 /*
1981  * Traverse our tree, and find the right node, execute whatever it points
1982  * to, and return the resulting error code.
1983  */
1984 
1985 static int
1986 sysctl_root(SYSCTL_HANDLER_ARGS)
1987 {
1988 	struct sysctl_oid *oid;
1989 	struct rm_priotracker tracker;
1990 	int error, indx, lvl;
1991 
1992 	SYSCTL_RLOCK(&tracker);
1993 
1994 	error = sysctl_find_oid(arg1, arg2, &oid, &indx, req);
1995 	if (error)
1996 		goto out;
1997 
1998 	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1999 		/*
2000 		 * You can't call a sysctl when it's a node, but has
2001 		 * no handler.  Inform the user that it's a node.
2002 		 * The indx may or may not be the same as namelen.
2003 		 */
2004 		if (oid->oid_handler == NULL) {
2005 			error = EISDIR;
2006 			goto out;
2007 		}
2008 	}
2009 
2010 	/* Is this sysctl writable? */
2011 	if (req->newptr && !(oid->oid_kind & CTLFLAG_WR)) {
2012 		error = EPERM;
2013 		goto out;
2014 	}
2015 
2016 	KASSERT(req->td != NULL, ("sysctl_root(): req->td == NULL"));
2017 
2018 #ifdef CAPABILITY_MODE
2019 	/*
2020 	 * If the process is in capability mode, then don't permit reading or
2021 	 * writing unless specifically granted for the node.
2022 	 */
2023 	if (IN_CAPABILITY_MODE(req->td)) {
2024 		if ((req->oldptr && !(oid->oid_kind & CTLFLAG_CAPRD)) ||
2025 		    (req->newptr && !(oid->oid_kind & CTLFLAG_CAPWR))) {
2026 			error = EPERM;
2027 			goto out;
2028 		}
2029 	}
2030 #endif
2031 
2032 	/* Is this sysctl sensitive to securelevels? */
2033 	if (req->newptr && (oid->oid_kind & CTLFLAG_SECURE)) {
2034 		lvl = (oid->oid_kind & CTLMASK_SECURE) >> CTLSHIFT_SECURE;
2035 		error = securelevel_gt(req->td->td_ucred, lvl);
2036 		if (error)
2037 			goto out;
2038 	}
2039 
2040 	/* Is this sysctl writable by only privileged users? */
2041 	if (req->newptr && !(oid->oid_kind & CTLFLAG_ANYBODY)) {
2042 		int priv;
2043 
2044 		if (oid->oid_kind & CTLFLAG_PRISON)
2045 			priv = PRIV_SYSCTL_WRITEJAIL;
2046 #ifdef VIMAGE
2047 		else if ((oid->oid_kind & CTLFLAG_VNET) &&
2048 		     prison_owns_vnet(req->td->td_ucred))
2049 			priv = PRIV_SYSCTL_WRITEJAIL;
2050 #endif
2051 		else
2052 			priv = PRIV_SYSCTL_WRITE;
2053 		error = priv_check(req->td, priv);
2054 		if (error)
2055 			goto out;
2056 	}
2057 
2058 	if (!oid->oid_handler) {
2059 		error = EINVAL;
2060 		goto out;
2061 	}
2062 
2063 	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
2064 		arg1 = (int *)arg1 + indx;
2065 		arg2 -= indx;
2066 	} else {
2067 		arg1 = oid->oid_arg1;
2068 		arg2 = oid->oid_arg2;
2069 	}
2070 #ifdef MAC
2071 	error = mac_system_check_sysctl(req->td->td_ucred, oid, arg1, arg2,
2072 	    req);
2073 	if (error != 0)
2074 		goto out;
2075 #endif
2076 #ifdef VIMAGE
2077 	if ((oid->oid_kind & CTLFLAG_VNET) && arg1 != NULL)
2078 		arg1 = (void *)(curvnet->vnet_data_base + (uintptr_t)arg1);
2079 #endif
2080 	error = sysctl_root_handler_locked(oid, arg1, arg2, req, &tracker);
2081 
2082 out:
2083 	SYSCTL_RUNLOCK(&tracker);
2084 	return (error);
2085 }
2086 
2087 #ifndef _SYS_SYSPROTO_H_
2088 struct sysctl_args {
2089 	int	*name;
2090 	u_int	namelen;
2091 	void	*old;
2092 	size_t	*oldlenp;
2093 	void	*new;
2094 	size_t	newlen;
2095 };
2096 #endif
2097 int
2098 sys___sysctl(struct thread *td, struct sysctl_args *uap)
2099 {
2100 	int error, i, name[CTL_MAXNAME];
2101 	size_t j;
2102 
2103 	if (uap->namelen > CTL_MAXNAME || uap->namelen < 2)
2104 		return (EINVAL);
2105 
2106  	error = copyin(uap->name, &name, uap->namelen * sizeof(int));
2107  	if (error)
2108 		return (error);
2109 
2110 	error = userland_sysctl(td, name, uap->namelen,
2111 		uap->old, uap->oldlenp, 0,
2112 		uap->new, uap->newlen, &j, 0);
2113 	if (error && error != ENOMEM)
2114 		return (error);
2115 	if (uap->oldlenp) {
2116 		i = copyout(&j, uap->oldlenp, sizeof(j));
2117 		if (i)
2118 			return (i);
2119 	}
2120 	return (error);
2121 }
2122 
2123 /*
2124  * This is used from various compatibility syscalls too.  That's why name
2125  * must be in kernel space.
2126  */
2127 int
2128 userland_sysctl(struct thread *td, int *name, u_int namelen, void *old,
2129     size_t *oldlenp, int inkernel, const void *new, size_t newlen,
2130     size_t *retval, int flags)
2131 {
2132 	int error = 0, memlocked;
2133 	struct sysctl_req req;
2134 
2135 	bzero(&req, sizeof req);
2136 
2137 	req.td = td;
2138 	req.flags = flags;
2139 
2140 	if (oldlenp) {
2141 		if (inkernel) {
2142 			req.oldlen = *oldlenp;
2143 		} else {
2144 			error = copyin(oldlenp, &req.oldlen, sizeof(*oldlenp));
2145 			if (error)
2146 				return (error);
2147 		}
2148 	}
2149 	req.validlen = req.oldlen;
2150 	req.oldptr = old;
2151 
2152 	if (new != NULL) {
2153 		req.newlen = newlen;
2154 		req.newptr = new;
2155 	}
2156 
2157 	req.oldfunc = sysctl_old_user;
2158 	req.newfunc = sysctl_new_user;
2159 	req.lock = REQ_UNWIRED;
2160 
2161 #ifdef KTRACE
2162 	if (KTRPOINT(curthread, KTR_SYSCTL))
2163 		ktrsysctl(name, namelen);
2164 #endif
2165 	memlocked = 0;
2166 	if (req.oldptr && req.oldlen > 4 * PAGE_SIZE) {
2167 		memlocked = 1;
2168 		sx_xlock(&sysctlmemlock);
2169 	}
2170 	CURVNET_SET(TD_TO_VNET(td));
2171 
2172 	for (;;) {
2173 		req.oldidx = 0;
2174 		req.newidx = 0;
2175 		error = sysctl_root(0, name, namelen, &req);
2176 		if (error != EAGAIN)
2177 			break;
2178 		kern_yield(PRI_USER);
2179 	}
2180 
2181 	CURVNET_RESTORE();
2182 
2183 	if (req.lock == REQ_WIRED && req.validlen > 0)
2184 		vsunlock(req.oldptr, req.validlen);
2185 	if (memlocked)
2186 		sx_xunlock(&sysctlmemlock);
2187 
2188 	if (error && error != ENOMEM)
2189 		return (error);
2190 
2191 	if (retval) {
2192 		if (req.oldptr && req.oldidx > req.validlen)
2193 			*retval = req.validlen;
2194 		else
2195 			*retval = req.oldidx;
2196 	}
2197 	return (error);
2198 }
2199 
2200 /*
2201  * Drain into a sysctl struct.  The user buffer should be wired if a page
2202  * fault would cause issue.
2203  */
2204 static int
2205 sbuf_sysctl_drain(void *arg, const char *data, int len)
2206 {
2207 	struct sysctl_req *req = arg;
2208 	int error;
2209 
2210 	error = SYSCTL_OUT(req, data, len);
2211 	KASSERT(error >= 0, ("Got unexpected negative value %d", error));
2212 	return (error == 0 ? len : -error);
2213 }
2214 
2215 struct sbuf *
2216 sbuf_new_for_sysctl(struct sbuf *s, char *buf, int length,
2217     struct sysctl_req *req)
2218 {
2219 
2220 	/* Supply a default buffer size if none given. */
2221 	if (buf == NULL && length == 0)
2222 		length = 64;
2223 	s = sbuf_new(s, buf, length, SBUF_FIXEDLEN | SBUF_INCLUDENUL);
2224 	sbuf_set_drain(s, sbuf_sysctl_drain, req);
2225 	return (s);
2226 }
2227