xref: /freebsd/sys/kern/kern_sysctl.c (revision 73577bf01de5c4677dc54d97f93310286c254780)
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_ddb.h"
45 #include "opt_ktrace.h"
46 #include "opt_sysctl.h"
47 
48 #include <sys/param.h>
49 #include <sys/fail.h>
50 #include <sys/systm.h>
51 #include <sys/capsicum.h>
52 #include <sys/kernel.h>
53 #include <sys/limits.h>
54 #include <sys/sysctl.h>
55 #include <sys/malloc.h>
56 #include <sys/priv.h>
57 #include <sys/proc.h>
58 #include <sys/jail.h>
59 #include <sys/kdb.h>
60 #include <sys/lock.h>
61 #include <sys/mutex.h>
62 #include <sys/rmlock.h>
63 #include <sys/sbuf.h>
64 #include <sys/sx.h>
65 #include <sys/sysproto.h>
66 #include <sys/uio.h>
67 #ifdef KTRACE
68 #include <sys/ktrace.h>
69 #endif
70 
71 #ifdef DDB
72 #include <ddb/ddb.h>
73 #include <ddb/db_lex.h>
74 #endif
75 
76 #include <net/vnet.h>
77 
78 #include <security/mac/mac_framework.h>
79 
80 #include <vm/vm.h>
81 #include <vm/vm_extern.h>
82 
83 static MALLOC_DEFINE(M_SYSCTL, "sysctl", "sysctl internal magic");
84 static MALLOC_DEFINE(M_SYSCTLOID, "sysctloid", "sysctl dynamic oids");
85 static MALLOC_DEFINE(M_SYSCTLTMP, "sysctltmp", "sysctl temp output buffer");
86 
87 /*
88  * The sysctllock protects the MIB tree.  It also protects sysctl
89  * contexts used with dynamic sysctls.  The sysctl_register_oid() and
90  * sysctl_unregister_oid() routines require the sysctllock to already
91  * be held, so the sysctl_wlock() and sysctl_wunlock() routines are
92  * provided for the few places in the kernel which need to use that
93  * API rather than using the dynamic API.  Use of the dynamic API is
94  * strongly encouraged for most code.
95  *
96  * The sysctlmemlock is used to limit the amount of user memory wired for
97  * sysctl requests.  This is implemented by serializing any userland
98  * sysctl requests larger than a single page via an exclusive lock.
99  *
100  * The sysctlstringlock is used to protect concurrent access to writable
101  * string nodes in sysctl_handle_string().
102  */
103 static struct rmlock sysctllock;
104 static struct sx __exclusive_cache_line sysctlmemlock;
105 static struct sx sysctlstringlock;
106 
107 #define	SYSCTL_WLOCK()		rm_wlock(&sysctllock)
108 #define	SYSCTL_WUNLOCK()	rm_wunlock(&sysctllock)
109 #define	SYSCTL_RLOCK(tracker)	rm_rlock(&sysctllock, (tracker))
110 #define	SYSCTL_RUNLOCK(tracker)	rm_runlock(&sysctllock, (tracker))
111 #define	SYSCTL_WLOCKED()	rm_wowned(&sysctllock)
112 #define	SYSCTL_ASSERT_LOCKED()	rm_assert(&sysctllock, RA_LOCKED)
113 #define	SYSCTL_ASSERT_WLOCKED()	rm_assert(&sysctllock, RA_WLOCKED)
114 #define	SYSCTL_ASSERT_RLOCKED()	rm_assert(&sysctllock, RA_RLOCKED)
115 #define	SYSCTL_INIT()		rm_init_flags(&sysctllock, "sysctl lock", \
116 				    RM_SLEEPABLE)
117 #define	SYSCTL_SLEEP(ch, wmesg, timo)					\
118 				rm_sleep(ch, &sysctllock, 0, wmesg, timo)
119 
120 static int sysctl_root(SYSCTL_HANDLER_ARGS);
121 
122 /* Root list */
123 struct sysctl_oid_list sysctl__children = SLIST_HEAD_INITIALIZER(&sysctl__children);
124 
125 static int	sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del,
126 		    int recurse);
127 static int	sysctl_old_kernel(struct sysctl_req *, const void *, size_t);
128 static int	sysctl_new_kernel(struct sysctl_req *, void *, size_t);
129 
130 static struct sysctl_oid *
131 sysctl_find_oidname(const char *name, struct sysctl_oid_list *list)
132 {
133 	struct sysctl_oid *oidp;
134 
135 	SYSCTL_ASSERT_LOCKED();
136 	SLIST_FOREACH(oidp, list, oid_link) {
137 		if (strcmp(oidp->oid_name, name) == 0) {
138 			return (oidp);
139 		}
140 	}
141 	return (NULL);
142 }
143 
144 /*
145  * Initialization of the MIB tree.
146  *
147  * Order by number in each list.
148  */
149 void
150 sysctl_wlock(void)
151 {
152 
153 	SYSCTL_WLOCK();
154 }
155 
156 void
157 sysctl_wunlock(void)
158 {
159 
160 	SYSCTL_WUNLOCK();
161 }
162 
163 static int
164 sysctl_root_handler_locked(struct sysctl_oid *oid, void *arg1, intmax_t arg2,
165     struct sysctl_req *req, struct rm_priotracker *tracker)
166 {
167 	int error;
168 
169 	if (oid->oid_kind & CTLFLAG_DYN)
170 		atomic_add_int(&oid->oid_running, 1);
171 
172 	if (tracker != NULL)
173 		SYSCTL_RUNLOCK(tracker);
174 	else
175 		SYSCTL_WUNLOCK();
176 
177 	/*
178 	 * Treat set CTLFLAG_NEEDGIANT and unset CTLFLAG_MPSAFE flags the same,
179 	 * untill we're ready to remove all traces of Giant from sysctl(9).
180 	 */
181 	if ((oid->oid_kind & CTLFLAG_NEEDGIANT) ||
182 	    (!(oid->oid_kind & CTLFLAG_MPSAFE)))
183 		mtx_lock(&Giant);
184 	error = oid->oid_handler(oid, arg1, arg2, req);
185 	if ((oid->oid_kind & CTLFLAG_NEEDGIANT) ||
186 	    (!(oid->oid_kind & CTLFLAG_MPSAFE)))
187 		mtx_unlock(&Giant);
188 
189 	KFAIL_POINT_ERROR(_debug_fail_point, sysctl_running, error);
190 
191 	if (tracker != NULL)
192 		SYSCTL_RLOCK(tracker);
193 	else
194 		SYSCTL_WLOCK();
195 
196 	if (oid->oid_kind & CTLFLAG_DYN) {
197 		if (atomic_fetchadd_int(&oid->oid_running, -1) == 1 &&
198 		    (oid->oid_kind & CTLFLAG_DYING) != 0)
199 			wakeup(&oid->oid_running);
200 	}
201 
202 	return (error);
203 }
204 
205 static void
206 sysctl_load_tunable_by_oid_locked(struct sysctl_oid *oidp)
207 {
208 	struct sysctl_req req;
209 	struct sysctl_oid *curr;
210 	char *penv = NULL;
211 	char path[96];
212 	ssize_t rem = sizeof(path);
213 	ssize_t len;
214 	uint8_t data[512] __aligned(sizeof(uint64_t));
215 	int size;
216 	int error;
217 
218 	path[--rem] = 0;
219 
220 	for (curr = oidp; curr != NULL; curr = SYSCTL_PARENT(curr)) {
221 		len = strlen(curr->oid_name);
222 		rem -= len;
223 		if (curr != oidp)
224 			rem -= 1;
225 		if (rem < 0) {
226 			printf("OID path exceeds %d bytes\n", (int)sizeof(path));
227 			return;
228 		}
229 		memcpy(path + rem, curr->oid_name, len);
230 		if (curr != oidp)
231 			path[rem + len] = '.';
232 	}
233 
234 	memset(&req, 0, sizeof(req));
235 
236 	req.td = curthread;
237 	req.oldfunc = sysctl_old_kernel;
238 	req.newfunc = sysctl_new_kernel;
239 	req.lock = REQ_UNWIRED;
240 
241 	switch (oidp->oid_kind & CTLTYPE) {
242 	case CTLTYPE_INT:
243 		if (getenv_array(path + rem, data, sizeof(data), &size,
244 		    sizeof(int), GETENV_SIGNED) == 0)
245 			return;
246 		req.newlen = size;
247 		req.newptr = data;
248 		break;
249 	case CTLTYPE_UINT:
250 		if (getenv_array(path + rem, data, sizeof(data), &size,
251 		    sizeof(int), GETENV_UNSIGNED) == 0)
252 			return;
253 		req.newlen = size;
254 		req.newptr = data;
255 		break;
256 	case CTLTYPE_LONG:
257 		if (getenv_array(path + rem, data, sizeof(data), &size,
258 		    sizeof(long), GETENV_SIGNED) == 0)
259 			return;
260 		req.newlen = size;
261 		req.newptr = data;
262 		break;
263 	case CTLTYPE_ULONG:
264 		if (getenv_array(path + rem, data, sizeof(data), &size,
265 		    sizeof(long), GETENV_UNSIGNED) == 0)
266 			return;
267 		req.newlen = size;
268 		req.newptr = data;
269 		break;
270 	case CTLTYPE_S8:
271 		if (getenv_array(path + rem, data, sizeof(data), &size,
272 		    sizeof(int8_t), GETENV_SIGNED) == 0)
273 			return;
274 		req.newlen = size;
275 		req.newptr = data;
276 		break;
277 	case CTLTYPE_S16:
278 		if (getenv_array(path + rem, data, sizeof(data), &size,
279 		    sizeof(int16_t), GETENV_SIGNED) == 0)
280 			return;
281 		req.newlen = size;
282 		req.newptr = data;
283 		break;
284 	case CTLTYPE_S32:
285 		if (getenv_array(path + rem, data, sizeof(data), &size,
286 		    sizeof(int32_t), GETENV_SIGNED) == 0)
287 			return;
288 		req.newlen = size;
289 		req.newptr = data;
290 		break;
291 	case CTLTYPE_S64:
292 		if (getenv_array(path + rem, data, sizeof(data), &size,
293 		    sizeof(int64_t), GETENV_SIGNED) == 0)
294 			return;
295 		req.newlen = size;
296 		req.newptr = data;
297 		break;
298 	case CTLTYPE_U8:
299 		if (getenv_array(path + rem, data, sizeof(data), &size,
300 		    sizeof(uint8_t), GETENV_UNSIGNED) == 0)
301 			return;
302 		req.newlen = size;
303 		req.newptr = data;
304 		break;
305 	case CTLTYPE_U16:
306 		if (getenv_array(path + rem, data, sizeof(data), &size,
307 		    sizeof(uint16_t), GETENV_UNSIGNED) == 0)
308 			return;
309 		req.newlen = size;
310 		req.newptr = data;
311 		break;
312 	case CTLTYPE_U32:
313 		if (getenv_array(path + rem, data, sizeof(data), &size,
314 		    sizeof(uint32_t), GETENV_UNSIGNED) == 0)
315 			return;
316 		req.newlen = size;
317 		req.newptr = data;
318 		break;
319 	case CTLTYPE_U64:
320 		if (getenv_array(path + rem, data, sizeof(data), &size,
321 		    sizeof(uint64_t), GETENV_UNSIGNED) == 0)
322 			return;
323 		req.newlen = size;
324 		req.newptr = data;
325 		break;
326 	case CTLTYPE_STRING:
327 		penv = kern_getenv(path + rem);
328 		if (penv == NULL)
329 			return;
330 		req.newlen = strlen(penv);
331 		req.newptr = penv;
332 		break;
333 	default:
334 		return;
335 	}
336 	error = sysctl_root_handler_locked(oidp, oidp->oid_arg1,
337 	    oidp->oid_arg2, &req, NULL);
338 	if (error != 0)
339 		printf("Setting sysctl %s failed: %d\n", path + rem, error);
340 	if (penv != NULL)
341 		freeenv(penv);
342 }
343 
344 /*
345  * Locate the path to a given oid.  Returns the length of the resulting path,
346  * or -1 if the oid was not found.  nodes must have room for CTL_MAXNAME
347  * elements and be NULL initialized.
348  */
349 static int
350 sysctl_search_oid(struct sysctl_oid **nodes, struct sysctl_oid *needle)
351 {
352 	int indx;
353 
354 	SYSCTL_ASSERT_LOCKED();
355 	indx = 0;
356 	while (indx < CTL_MAXNAME && indx >= 0) {
357 		if (nodes[indx] == NULL && indx == 0)
358 			nodes[indx] = SLIST_FIRST(&sysctl__children);
359 		else if (nodes[indx] == NULL)
360 			nodes[indx] = SLIST_FIRST(&nodes[indx - 1]->oid_children);
361 		else
362 			nodes[indx] = SLIST_NEXT(nodes[indx], oid_link);
363 
364 		if (nodes[indx] == needle)
365 			return (indx + 1);
366 
367 		if (nodes[indx] == NULL) {
368 			indx--;
369 			continue;
370 		}
371 
372 		if ((nodes[indx]->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
373 			indx++;
374 			continue;
375 		}
376 	}
377 	return (-1);
378 }
379 
380 static void
381 sysctl_warn_reuse(const char *func, struct sysctl_oid *leaf)
382 {
383 	struct sysctl_oid *nodes[CTL_MAXNAME];
384 	char buf[128];
385 	struct sbuf sb;
386 	int rc, i;
387 
388 	(void)sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN | SBUF_INCLUDENUL);
389 	sbuf_set_drain(&sb, sbuf_printf_drain, NULL);
390 
391 	sbuf_printf(&sb, "%s: can't re-use a leaf (", __func__);
392 
393 	memset(nodes, 0, sizeof(nodes));
394 	rc = sysctl_search_oid(nodes, leaf);
395 	if (rc > 0) {
396 		for (i = 0; i < rc; i++)
397 			sbuf_printf(&sb, "%s%.*s", nodes[i]->oid_name,
398 			    i != (rc - 1), ".");
399 	} else {
400 		sbuf_printf(&sb, "%s", leaf->oid_name);
401 	}
402 	sbuf_printf(&sb, ")!\n");
403 
404 	(void)sbuf_finish(&sb);
405 }
406 
407 #ifdef SYSCTL_DEBUG
408 static int
409 sysctl_reuse_test(SYSCTL_HANDLER_ARGS)
410 {
411 	struct rm_priotracker tracker;
412 
413 	SYSCTL_RLOCK(&tracker);
414 	sysctl_warn_reuse(__func__, oidp);
415 	SYSCTL_RUNLOCK(&tracker);
416 	return (0);
417 }
418 SYSCTL_PROC(_sysctl, OID_AUTO, reuse_test,
419     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 0, sysctl_reuse_test, "-",
420     "");
421 #endif
422 
423 void
424 sysctl_register_oid(struct sysctl_oid *oidp)
425 {
426 	struct sysctl_oid_list *parent = oidp->oid_parent;
427 	struct sysctl_oid *p;
428 	struct sysctl_oid *q;
429 	int oid_number;
430 	int timeout = 2;
431 
432 	/*
433 	 * First check if another oid with the same name already
434 	 * exists in the parent's list.
435 	 */
436 	SYSCTL_ASSERT_WLOCKED();
437 	p = sysctl_find_oidname(oidp->oid_name, parent);
438 	if (p != NULL) {
439 		if ((p->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
440 			p->oid_refcnt++;
441 			return;
442 		} else {
443 			sysctl_warn_reuse(__func__, p);
444 			return;
445 		}
446 	}
447 	/* get current OID number */
448 	oid_number = oidp->oid_number;
449 
450 #if (OID_AUTO >= 0)
451 #error "OID_AUTO is expected to be a negative value"
452 #endif
453 	/*
454 	 * Any negative OID number qualifies as OID_AUTO. Valid OID
455 	 * numbers should always be positive.
456 	 *
457 	 * NOTE: DO NOT change the starting value here, change it in
458 	 * <sys/sysctl.h>, and make sure it is at least 256 to
459 	 * accommodate e.g. net.inet.raw as a static sysctl node.
460 	 */
461 	if (oid_number < 0) {
462 		static int newoid;
463 
464 		/*
465 		 * By decrementing the next OID number we spend less
466 		 * time inserting the OIDs into a sorted list.
467 		 */
468 		if (--newoid < CTL_AUTO_START)
469 			newoid = 0x7fffffff;
470 
471 		oid_number = newoid;
472 	}
473 
474 	/*
475 	 * Insert the OID into the parent's list sorted by OID number.
476 	 */
477 retry:
478 	q = NULL;
479 	SLIST_FOREACH(p, parent, oid_link) {
480 		/* check if the current OID number is in use */
481 		if (oid_number == p->oid_number) {
482 			/* get the next valid OID number */
483 			if (oid_number < CTL_AUTO_START ||
484 			    oid_number == 0x7fffffff) {
485 				/* wraparound - restart */
486 				oid_number = CTL_AUTO_START;
487 				/* don't loop forever */
488 				if (!timeout--)
489 					panic("sysctl: Out of OID numbers\n");
490 				goto retry;
491 			} else {
492 				oid_number++;
493 			}
494 		} else if (oid_number < p->oid_number)
495 			break;
496 		q = p;
497 	}
498 	/* check for non-auto OID number collision */
499 	if (oidp->oid_number >= 0 && oidp->oid_number < CTL_AUTO_START &&
500 	    oid_number >= CTL_AUTO_START) {
501 		printf("sysctl: OID number(%d) is already in use for '%s'\n",
502 		    oidp->oid_number, oidp->oid_name);
503 	}
504 	/* update the OID number, if any */
505 	oidp->oid_number = oid_number;
506 	if (q != NULL)
507 		SLIST_INSERT_AFTER(q, oidp, oid_link);
508 	else
509 		SLIST_INSERT_HEAD(parent, oidp, oid_link);
510 
511 	if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE &&
512 #ifdef VIMAGE
513 	    (oidp->oid_kind & CTLFLAG_VNET) == 0 &&
514 #endif
515 	    (oidp->oid_kind & CTLFLAG_TUN) != 0 &&
516 	    (oidp->oid_kind & CTLFLAG_NOFETCH) == 0) {
517 		/* only fetch value once */
518 		oidp->oid_kind |= CTLFLAG_NOFETCH;
519 		/* try to fetch value from kernel environment */
520 		sysctl_load_tunable_by_oid_locked(oidp);
521 	}
522 }
523 
524 void
525 sysctl_register_disabled_oid(struct sysctl_oid *oidp)
526 {
527 
528 	/*
529 	 * Mark the leaf as dormant if it's not to be immediately enabled.
530 	 * We do not disable nodes as they can be shared between modules
531 	 * and it is always safe to access a node.
532 	 */
533 	KASSERT((oidp->oid_kind & CTLFLAG_DORMANT) == 0,
534 	    ("internal flag is set in oid_kind"));
535 	if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
536 		oidp->oid_kind |= CTLFLAG_DORMANT;
537 	sysctl_register_oid(oidp);
538 }
539 
540 void
541 sysctl_enable_oid(struct sysctl_oid *oidp)
542 {
543 
544 	SYSCTL_ASSERT_WLOCKED();
545 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
546 		KASSERT((oidp->oid_kind & CTLFLAG_DORMANT) == 0,
547 		    ("sysctl node is marked as dormant"));
548 		return;
549 	}
550 	KASSERT((oidp->oid_kind & CTLFLAG_DORMANT) != 0,
551 	    ("enabling already enabled sysctl oid"));
552 	oidp->oid_kind &= ~CTLFLAG_DORMANT;
553 }
554 
555 void
556 sysctl_unregister_oid(struct sysctl_oid *oidp)
557 {
558 	struct sysctl_oid *p;
559 	int error;
560 
561 	SYSCTL_ASSERT_WLOCKED();
562 	if (oidp->oid_number == OID_AUTO) {
563 		error = EINVAL;
564 	} else {
565 		error = ENOENT;
566 		SLIST_FOREACH(p, oidp->oid_parent, oid_link) {
567 			if (p == oidp) {
568 				SLIST_REMOVE(oidp->oid_parent, oidp,
569 				    sysctl_oid, oid_link);
570 				error = 0;
571 				break;
572 			}
573 		}
574 	}
575 
576 	/*
577 	 * This can happen when a module fails to register and is
578 	 * being unloaded afterwards.  It should not be a panic()
579 	 * for normal use.
580 	 */
581 	if (error) {
582 		printf("%s: failed(%d) to unregister sysctl(%s)\n",
583 		    __func__, error, oidp->oid_name);
584 	}
585 }
586 
587 /* Initialize a new context to keep track of dynamically added sysctls. */
588 int
589 sysctl_ctx_init(struct sysctl_ctx_list *c)
590 {
591 
592 	if (c == NULL) {
593 		return (EINVAL);
594 	}
595 
596 	/*
597 	 * No locking here, the caller is responsible for not adding
598 	 * new nodes to a context until after this function has
599 	 * returned.
600 	 */
601 	TAILQ_INIT(c);
602 	return (0);
603 }
604 
605 /* Free the context, and destroy all dynamic oids registered in this context */
606 int
607 sysctl_ctx_free(struct sysctl_ctx_list *clist)
608 {
609 	struct sysctl_ctx_entry *e, *e1;
610 	int error;
611 
612 	error = 0;
613 	/*
614 	 * First perform a "dry run" to check if it's ok to remove oids.
615 	 * XXX FIXME
616 	 * XXX This algorithm is a hack. But I don't know any
617 	 * XXX better solution for now...
618 	 */
619 	SYSCTL_WLOCK();
620 	TAILQ_FOREACH(e, clist, link) {
621 		error = sysctl_remove_oid_locked(e->entry, 0, 0);
622 		if (error)
623 			break;
624 	}
625 	/*
626 	 * Restore deregistered entries, either from the end,
627 	 * or from the place where error occurred.
628 	 * e contains the entry that was not unregistered
629 	 */
630 	if (error)
631 		e1 = TAILQ_PREV(e, sysctl_ctx_list, link);
632 	else
633 		e1 = TAILQ_LAST(clist, sysctl_ctx_list);
634 	while (e1 != NULL) {
635 		sysctl_register_oid(e1->entry);
636 		e1 = TAILQ_PREV(e1, sysctl_ctx_list, link);
637 	}
638 	if (error) {
639 		SYSCTL_WUNLOCK();
640 		return(EBUSY);
641 	}
642 	/* Now really delete the entries */
643 	e = TAILQ_FIRST(clist);
644 	while (e != NULL) {
645 		e1 = TAILQ_NEXT(e, link);
646 		error = sysctl_remove_oid_locked(e->entry, 1, 0);
647 		if (error)
648 			panic("sysctl_remove_oid: corrupt tree, entry: %s",
649 			    e->entry->oid_name);
650 		free(e, M_SYSCTLOID);
651 		e = e1;
652 	}
653 	SYSCTL_WUNLOCK();
654 	return (error);
655 }
656 
657 /* Add an entry to the context */
658 struct sysctl_ctx_entry *
659 sysctl_ctx_entry_add(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
660 {
661 	struct sysctl_ctx_entry *e;
662 
663 	SYSCTL_ASSERT_WLOCKED();
664 	if (clist == NULL || oidp == NULL)
665 		return(NULL);
666 	e = malloc(sizeof(struct sysctl_ctx_entry), M_SYSCTLOID, M_WAITOK);
667 	e->entry = oidp;
668 	TAILQ_INSERT_HEAD(clist, e, link);
669 	return (e);
670 }
671 
672 /* Find an entry in the context */
673 struct sysctl_ctx_entry *
674 sysctl_ctx_entry_find(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
675 {
676 	struct sysctl_ctx_entry *e;
677 
678 	SYSCTL_ASSERT_WLOCKED();
679 	if (clist == NULL || oidp == NULL)
680 		return(NULL);
681 	TAILQ_FOREACH(e, clist, link) {
682 		if(e->entry == oidp)
683 			return(e);
684 	}
685 	return (e);
686 }
687 
688 /*
689  * Delete an entry from the context.
690  * NOTE: this function doesn't free oidp! You have to remove it
691  * with sysctl_remove_oid().
692  */
693 int
694 sysctl_ctx_entry_del(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
695 {
696 	struct sysctl_ctx_entry *e;
697 
698 	if (clist == NULL || oidp == NULL)
699 		return (EINVAL);
700 	SYSCTL_WLOCK();
701 	e = sysctl_ctx_entry_find(clist, oidp);
702 	if (e != NULL) {
703 		TAILQ_REMOVE(clist, e, link);
704 		SYSCTL_WUNLOCK();
705 		free(e, M_SYSCTLOID);
706 		return (0);
707 	} else {
708 		SYSCTL_WUNLOCK();
709 		return (ENOENT);
710 	}
711 }
712 
713 /*
714  * Remove dynamically created sysctl trees.
715  * oidp - top of the tree to be removed
716  * del - if 0 - just deregister, otherwise free up entries as well
717  * recurse - if != 0 traverse the subtree to be deleted
718  */
719 int
720 sysctl_remove_oid(struct sysctl_oid *oidp, int del, int recurse)
721 {
722 	int error;
723 
724 	SYSCTL_WLOCK();
725 	error = sysctl_remove_oid_locked(oidp, del, recurse);
726 	SYSCTL_WUNLOCK();
727 	return (error);
728 }
729 
730 int
731 sysctl_remove_name(struct sysctl_oid *parent, const char *name,
732     int del, int recurse)
733 {
734 	struct sysctl_oid *p, *tmp;
735 	int error;
736 
737 	error = ENOENT;
738 	SYSCTL_WLOCK();
739 	SLIST_FOREACH_SAFE(p, SYSCTL_CHILDREN(parent), oid_link, tmp) {
740 		if (strcmp(p->oid_name, name) == 0) {
741 			error = sysctl_remove_oid_locked(p, del, recurse);
742 			break;
743 		}
744 	}
745 	SYSCTL_WUNLOCK();
746 
747 	return (error);
748 }
749 
750 static int
751 sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del, int recurse)
752 {
753 	struct sysctl_oid *p, *tmp;
754 	int error;
755 
756 	SYSCTL_ASSERT_WLOCKED();
757 	if (oidp == NULL)
758 		return(EINVAL);
759 	if ((oidp->oid_kind & CTLFLAG_DYN) == 0) {
760 		printf("Warning: can't remove non-dynamic nodes (%s)!\n",
761 		    oidp->oid_name);
762 		return (EINVAL);
763 	}
764 	/*
765 	 * WARNING: normal method to do this should be through
766 	 * sysctl_ctx_free(). Use recursing as the last resort
767 	 * method to purge your sysctl tree of leftovers...
768 	 * However, if some other code still references these nodes,
769 	 * it will panic.
770 	 */
771 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
772 		if (oidp->oid_refcnt == 1) {
773 			SLIST_FOREACH_SAFE(p,
774 			    SYSCTL_CHILDREN(oidp), oid_link, tmp) {
775 				if (!recurse) {
776 					printf("Warning: failed attempt to "
777 					    "remove oid %s with child %s\n",
778 					    oidp->oid_name, p->oid_name);
779 					return (ENOTEMPTY);
780 				}
781 				error = sysctl_remove_oid_locked(p, del,
782 				    recurse);
783 				if (error)
784 					return (error);
785 			}
786 		}
787 	}
788 	if (oidp->oid_refcnt > 1 ) {
789 		oidp->oid_refcnt--;
790 	} else {
791 		if (oidp->oid_refcnt == 0) {
792 			printf("Warning: bad oid_refcnt=%u (%s)!\n",
793 				oidp->oid_refcnt, oidp->oid_name);
794 			return (EINVAL);
795 		}
796 		sysctl_unregister_oid(oidp);
797 		if (del) {
798 			/*
799 			 * Wait for all threads running the handler to drain.
800 			 * This preserves the previous behavior when the
801 			 * sysctl lock was held across a handler invocation,
802 			 * and is necessary for module unload correctness.
803 			 */
804 			while (oidp->oid_running > 0) {
805 				oidp->oid_kind |= CTLFLAG_DYING;
806 				SYSCTL_SLEEP(&oidp->oid_running, "oidrm", 0);
807 			}
808 			if (oidp->oid_descr)
809 				free(__DECONST(char *, oidp->oid_descr),
810 				    M_SYSCTLOID);
811 			if (oidp->oid_label)
812 				free(__DECONST(char *, oidp->oid_label),
813 				    M_SYSCTLOID);
814 			free(__DECONST(char *, oidp->oid_name), M_SYSCTLOID);
815 			free(oidp, M_SYSCTLOID);
816 		}
817 	}
818 	return (0);
819 }
820 /*
821  * Create new sysctls at run time.
822  * clist may point to a valid context initialized with sysctl_ctx_init().
823  */
824 struct sysctl_oid *
825 sysctl_add_oid(struct sysctl_ctx_list *clist, struct sysctl_oid_list *parent,
826 	int number, const char *name, int kind, void *arg1, intmax_t arg2,
827 	int (*handler)(SYSCTL_HANDLER_ARGS), const char *fmt, const char *descr,
828 	const char *label)
829 {
830 	struct sysctl_oid *oidp;
831 
832 	/* You have to hook up somewhere.. */
833 	if (parent == NULL)
834 		return(NULL);
835 	/* Check if the node already exists, otherwise create it */
836 	SYSCTL_WLOCK();
837 	oidp = sysctl_find_oidname(name, parent);
838 	if (oidp != NULL) {
839 		if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
840 			oidp->oid_refcnt++;
841 			/* Update the context */
842 			if (clist != NULL)
843 				sysctl_ctx_entry_add(clist, oidp);
844 			SYSCTL_WUNLOCK();
845 			return (oidp);
846 		} else {
847 			sysctl_warn_reuse(__func__, oidp);
848 			SYSCTL_WUNLOCK();
849 			return (NULL);
850 		}
851 	}
852 	oidp = malloc(sizeof(struct sysctl_oid), M_SYSCTLOID, M_WAITOK|M_ZERO);
853 	oidp->oid_parent = parent;
854 	SLIST_INIT(&oidp->oid_children);
855 	oidp->oid_number = number;
856 	oidp->oid_refcnt = 1;
857 	oidp->oid_name = strdup(name, M_SYSCTLOID);
858 	oidp->oid_handler = handler;
859 	oidp->oid_kind = CTLFLAG_DYN | kind;
860 	oidp->oid_arg1 = arg1;
861 	oidp->oid_arg2 = arg2;
862 	oidp->oid_fmt = fmt;
863 	if (descr != NULL)
864 		oidp->oid_descr = strdup(descr, M_SYSCTLOID);
865 	if (label != NULL)
866 		oidp->oid_label = strdup(label, M_SYSCTLOID);
867 	/* Update the context, if used */
868 	if (clist != NULL)
869 		sysctl_ctx_entry_add(clist, oidp);
870 	/* Register this oid */
871 	sysctl_register_oid(oidp);
872 	SYSCTL_WUNLOCK();
873 	return (oidp);
874 }
875 
876 /*
877  * Rename an existing oid.
878  */
879 void
880 sysctl_rename_oid(struct sysctl_oid *oidp, const char *name)
881 {
882 	char *newname;
883 	char *oldname;
884 
885 	newname = strdup(name, M_SYSCTLOID);
886 	SYSCTL_WLOCK();
887 	oldname = __DECONST(char *, oidp->oid_name);
888 	oidp->oid_name = newname;
889 	SYSCTL_WUNLOCK();
890 	free(oldname, M_SYSCTLOID);
891 }
892 
893 /*
894  * Reparent an existing oid.
895  */
896 int
897 sysctl_move_oid(struct sysctl_oid *oid, struct sysctl_oid_list *parent)
898 {
899 	struct sysctl_oid *oidp;
900 
901 	SYSCTL_WLOCK();
902 	if (oid->oid_parent == parent) {
903 		SYSCTL_WUNLOCK();
904 		return (0);
905 	}
906 	oidp = sysctl_find_oidname(oid->oid_name, parent);
907 	if (oidp != NULL) {
908 		SYSCTL_WUNLOCK();
909 		return (EEXIST);
910 	}
911 	sysctl_unregister_oid(oid);
912 	oid->oid_parent = parent;
913 	oid->oid_number = OID_AUTO;
914 	sysctl_register_oid(oid);
915 	SYSCTL_WUNLOCK();
916 	return (0);
917 }
918 
919 /*
920  * Register the kernel's oids on startup.
921  */
922 SET_DECLARE(sysctl_set, struct sysctl_oid);
923 
924 static void
925 sysctl_register_all(void *arg)
926 {
927 	struct sysctl_oid **oidp;
928 
929 	sx_init(&sysctlmemlock, "sysctl mem");
930 	sx_init(&sysctlstringlock, "sysctl string handler");
931 	SYSCTL_INIT();
932 	SYSCTL_WLOCK();
933 	SET_FOREACH(oidp, sysctl_set)
934 		sysctl_register_oid(*oidp);
935 	SYSCTL_WUNLOCK();
936 }
937 SYSINIT(sysctl, SI_SUB_KMEM, SI_ORDER_FIRST, sysctl_register_all, NULL);
938 
939 /*
940  * "Staff-functions"
941  *
942  * These functions implement a presently undocumented interface
943  * used by the sysctl program to walk the tree, and get the type
944  * so it can print the value.
945  * This interface is under work and consideration, and should probably
946  * be killed with a big axe by the first person who can find the time.
947  * (be aware though, that the proper interface isn't as obvious as it
948  * may seem, there are various conflicting requirements.
949  *
950  * {CTL_SYSCTL, CTL_SYSCTL_DEBUG}		printf the entire MIB-tree.
951  * {CTL_SYSCTL, CTL_SYSCTL_NAME, ...}		return the name of the "..."
952  *						OID.
953  * {CTL_SYSCTL, CTL_SYSCTL_NEXT, ...}		return the next OID, honoring
954  *						CTLFLAG_SKIP.
955  * {CTL_SYSCTL, CTL_SYSCTL_NAME2OID}		return the OID of the name in
956  *						"new"
957  * {CTL_SYSCTL, CTL_SYSCTL_OIDFMT, ...}		return the kind & format info
958  *						for the "..." OID.
959  * {CTL_SYSCTL, CTL_SYSCTL_OIDDESCR, ...}	return the description of the
960  *						"..." OID.
961  * {CTL_SYSCTL, CTL_SYSCTL_OIDLABEL, ...}	return the aggregation label of
962  *						the "..." OID.
963  * {CTL_SYSCTL, CTL_SYSCTL_NEXTNOSKIP, ...}	return the next OID, ignoring
964  *						CTLFLAG_SKIP.
965  */
966 
967 #ifdef SYSCTL_DEBUG
968 static void
969 sysctl_sysctl_debug_dump_node(struct sysctl_oid_list *l, int i)
970 {
971 	int k;
972 	struct sysctl_oid *oidp;
973 
974 	SYSCTL_ASSERT_LOCKED();
975 	SLIST_FOREACH(oidp, l, oid_link) {
976 		for (k=0; k<i; k++)
977 			printf(" ");
978 
979 		printf("%d %s ", oidp->oid_number, oidp->oid_name);
980 
981 		printf("%c%c",
982 			oidp->oid_kind & CTLFLAG_RD ? 'R':' ',
983 			oidp->oid_kind & CTLFLAG_WR ? 'W':' ');
984 
985 		if (oidp->oid_handler)
986 			printf(" *Handler");
987 
988 		switch (oidp->oid_kind & CTLTYPE) {
989 			case CTLTYPE_NODE:
990 				printf(" Node\n");
991 				if (!oidp->oid_handler) {
992 					sysctl_sysctl_debug_dump_node(
993 					    SYSCTL_CHILDREN(oidp), i + 2);
994 				}
995 				break;
996 			case CTLTYPE_INT:    printf(" Int\n"); break;
997 			case CTLTYPE_UINT:   printf(" u_int\n"); break;
998 			case CTLTYPE_LONG:   printf(" Long\n"); break;
999 			case CTLTYPE_ULONG:  printf(" u_long\n"); break;
1000 			case CTLTYPE_STRING: printf(" String\n"); break;
1001 			case CTLTYPE_S8:     printf(" int8_t\n"); break;
1002 			case CTLTYPE_S16:    printf(" int16_t\n"); break;
1003 			case CTLTYPE_S32:    printf(" int32_t\n"); break;
1004 			case CTLTYPE_S64:    printf(" int64_t\n"); break;
1005 			case CTLTYPE_U8:     printf(" uint8_t\n"); break;
1006 			case CTLTYPE_U16:    printf(" uint16_t\n"); break;
1007 			case CTLTYPE_U32:    printf(" uint32_t\n"); break;
1008 			case CTLTYPE_U64:    printf(" uint64_t\n"); break;
1009 			case CTLTYPE_OPAQUE: printf(" Opaque/struct\n"); break;
1010 			default:	     printf("\n");
1011 		}
1012 	}
1013 }
1014 
1015 static int
1016 sysctl_sysctl_debug(SYSCTL_HANDLER_ARGS)
1017 {
1018 	struct rm_priotracker tracker;
1019 	int error;
1020 
1021 	error = priv_check(req->td, PRIV_SYSCTL_DEBUG);
1022 	if (error)
1023 		return (error);
1024 	SYSCTL_RLOCK(&tracker);
1025 	sysctl_sysctl_debug_dump_node(&sysctl__children, 0);
1026 	SYSCTL_RUNLOCK(&tracker);
1027 	return (ENOENT);
1028 }
1029 
1030 SYSCTL_PROC(_sysctl, CTL_SYSCTL_DEBUG, debug, CTLTYPE_STRING | CTLFLAG_RD |
1031     CTLFLAG_MPSAFE, 0, 0, sysctl_sysctl_debug, "-", "");
1032 #endif
1033 
1034 static int
1035 sysctl_sysctl_name(SYSCTL_HANDLER_ARGS)
1036 {
1037 	int *name = (int *) arg1;
1038 	u_int namelen = arg2;
1039 	int error;
1040 	struct sysctl_oid *oid;
1041 	struct sysctl_oid_list *lsp = &sysctl__children, *lsp2;
1042 	struct rm_priotracker tracker;
1043 	char buf[10];
1044 
1045 	error = sysctl_wire_old_buffer(req, 0);
1046 	if (error)
1047 		return (error);
1048 
1049 	SYSCTL_RLOCK(&tracker);
1050 	while (namelen) {
1051 		if (!lsp) {
1052 			snprintf(buf,sizeof(buf),"%d",*name);
1053 			if (req->oldidx)
1054 				error = SYSCTL_OUT(req, ".", 1);
1055 			if (!error)
1056 				error = SYSCTL_OUT(req, buf, strlen(buf));
1057 			if (error)
1058 				goto out;
1059 			namelen--;
1060 			name++;
1061 			continue;
1062 		}
1063 		lsp2 = NULL;
1064 		SLIST_FOREACH(oid, lsp, oid_link) {
1065 			if (oid->oid_number != *name)
1066 				continue;
1067 
1068 			if (req->oldidx)
1069 				error = SYSCTL_OUT(req, ".", 1);
1070 			if (!error)
1071 				error = SYSCTL_OUT(req, oid->oid_name,
1072 					strlen(oid->oid_name));
1073 			if (error)
1074 				goto out;
1075 
1076 			namelen--;
1077 			name++;
1078 
1079 			if ((oid->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1080 				break;
1081 
1082 			if (oid->oid_handler)
1083 				break;
1084 
1085 			lsp2 = SYSCTL_CHILDREN(oid);
1086 			break;
1087 		}
1088 		lsp = lsp2;
1089 	}
1090 	error = SYSCTL_OUT(req, "", 1);
1091  out:
1092 	SYSCTL_RUNLOCK(&tracker);
1093 	return (error);
1094 }
1095 
1096 /*
1097  * XXXRW/JA: Shouldn't return name data for nodes that we don't permit in
1098  * capability mode.
1099  */
1100 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_NAME, name, CTLFLAG_RD |
1101     CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_name, "");
1102 
1103 /*
1104  * Walk the sysctl subtree at lsp until we find the given name,
1105  * and return the next name in order by oid_number.
1106  */
1107 static int
1108 sysctl_sysctl_next_ls(struct sysctl_oid_list *lsp, int *name, u_int namelen,
1109     int *next, int *len, int level, bool honor_skip)
1110 {
1111 	struct sysctl_oid *oidp;
1112 
1113 	SYSCTL_ASSERT_LOCKED();
1114 	*len = level;
1115 	SLIST_FOREACH(oidp, lsp, oid_link) {
1116 		*next = oidp->oid_number;
1117 
1118 		if ((oidp->oid_kind & CTLFLAG_DORMANT) != 0)
1119 			continue;
1120 
1121 		if (honor_skip && (oidp->oid_kind & CTLFLAG_SKIP) != 0)
1122 			continue;
1123 
1124 		if (namelen == 0) {
1125 			/*
1126 			 * We have reached a node with a full name match and are
1127 			 * looking for the next oid in its children.
1128 			 *
1129 			 * For CTL_SYSCTL_NEXTNOSKIP we are done.
1130 			 *
1131 			 * For CTL_SYSCTL_NEXT we skip CTLTYPE_NODE (unless it
1132 			 * has a handler) and move on to the children.
1133 			 */
1134 			if (!honor_skip)
1135 				return (0);
1136 			if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1137 				return (0);
1138 			if (oidp->oid_handler)
1139 				return (0);
1140 			lsp = SYSCTL_CHILDREN(oidp);
1141 			if (!sysctl_sysctl_next_ls(lsp, NULL, 0, next + 1, len,
1142 			    level + 1, honor_skip))
1143 				return (0);
1144 			/*
1145 			 * There were no useable children in this node.
1146 			 * Continue searching for the next oid at this level.
1147 			 */
1148 			goto emptynode;
1149 		}
1150 
1151 		/*
1152 		 * No match yet. Continue seeking the given name.
1153 		 *
1154 		 * We are iterating in order by oid_number, so skip oids lower
1155 		 * than the one we are looking for.
1156 		 *
1157 		 * When the current oid_number is higher than the one we seek,
1158 		 * that means we have reached the next oid in the sequence and
1159 		 * should return it.
1160 		 *
1161 		 * If the oid_number matches the name at this level then we
1162 		 * have to find a node to continue searching at the next level.
1163 		 */
1164 		if (oidp->oid_number < *name)
1165 			continue;
1166 		if (oidp->oid_number > *name) {
1167 			/*
1168 			 * We have reached the next oid.
1169 			 *
1170 			 * For CTL_SYSCTL_NEXTNOSKIP we are done.
1171 			 *
1172 			 * For CTL_SYSCTL_NEXT we skip CTLTYPE_NODE (unless it
1173 			 * has a handler) and move on to the children.
1174 			 */
1175 			if (!honor_skip)
1176 				return (0);
1177 			if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1178 				return (0);
1179 			if (oidp->oid_handler)
1180 				return (0);
1181 			lsp = SYSCTL_CHILDREN(oidp);
1182 			if (!sysctl_sysctl_next_ls(lsp, name + 1, namelen - 1,
1183 			    next + 1, len, level + 1, honor_skip))
1184 				return (0);
1185 			goto next;
1186 		}
1187 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1188 			continue;
1189 		if (oidp->oid_handler)
1190 			continue;
1191 		lsp = SYSCTL_CHILDREN(oidp);
1192 		if (!sysctl_sysctl_next_ls(lsp, name + 1, namelen - 1,
1193 		    next + 1, len, level + 1, honor_skip))
1194 			return (0);
1195 	next:
1196 		/*
1197 		 * There were no useable children in this node.
1198 		 * Continue searching for the next oid at the root level.
1199 		 */
1200 		namelen = 1;
1201 	emptynode:
1202 		/* Reset len in case a failed recursive call changed it. */
1203 		*len = level;
1204 	}
1205 	return (ENOENT);
1206 }
1207 
1208 static int
1209 sysctl_sysctl_next(SYSCTL_HANDLER_ARGS)
1210 {
1211 	int *name = (int *) arg1;
1212 	u_int namelen = arg2;
1213 	int len, error;
1214 	struct sysctl_oid_list *lsp = &sysctl__children;
1215 	struct rm_priotracker tracker;
1216 	int next[CTL_MAXNAME];
1217 
1218 	SYSCTL_RLOCK(&tracker);
1219 	error = sysctl_sysctl_next_ls(lsp, name, namelen, next, &len, 1,
1220 	    oidp->oid_number == CTL_SYSCTL_NEXT);
1221 	SYSCTL_RUNLOCK(&tracker);
1222 	if (error)
1223 		return (error);
1224 	error = SYSCTL_OUT(req, next, len * sizeof (int));
1225 	return (error);
1226 }
1227 
1228 /*
1229  * XXXRW/JA: Shouldn't return next data for nodes that we don't permit in
1230  * capability mode.
1231  */
1232 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_NEXT, next, CTLFLAG_RD |
1233     CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_next, "");
1234 
1235 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_NEXTNOSKIP, nextnoskip, CTLFLAG_RD |
1236     CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_next, "");
1237 
1238 static int
1239 name2oid(char *name, int *oid, int *len, struct sysctl_oid **oidpp)
1240 {
1241 	struct sysctl_oid *oidp;
1242 	struct sysctl_oid_list *lsp = &sysctl__children;
1243 	char *p;
1244 
1245 	SYSCTL_ASSERT_LOCKED();
1246 
1247 	for (*len = 0; *len < CTL_MAXNAME;) {
1248 		p = strsep(&name, ".");
1249 
1250 		oidp = SLIST_FIRST(lsp);
1251 		for (;; oidp = SLIST_NEXT(oidp, oid_link)) {
1252 			if (oidp == NULL)
1253 				return (ENOENT);
1254 			if (strcmp(p, oidp->oid_name) == 0)
1255 				break;
1256 		}
1257 		*oid++ = oidp->oid_number;
1258 		(*len)++;
1259 
1260 		if (name == NULL || *name == '\0') {
1261 			if (oidpp)
1262 				*oidpp = oidp;
1263 			return (0);
1264 		}
1265 
1266 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1267 			break;
1268 
1269 		if (oidp->oid_handler)
1270 			break;
1271 
1272 		lsp = SYSCTL_CHILDREN(oidp);
1273 	}
1274 	return (ENOENT);
1275 }
1276 
1277 static int
1278 sysctl_sysctl_name2oid(SYSCTL_HANDLER_ARGS)
1279 {
1280 	char *p;
1281 	int error, oid[CTL_MAXNAME], len = 0;
1282 	struct sysctl_oid *op = NULL;
1283 	struct rm_priotracker tracker;
1284 	char buf[32];
1285 
1286 	if (!req->newlen)
1287 		return (ENOENT);
1288 	if (req->newlen >= MAXPATHLEN)	/* XXX arbitrary, undocumented */
1289 		return (ENAMETOOLONG);
1290 
1291 	p = buf;
1292 	if (req->newlen >= sizeof(buf))
1293 		p = malloc(req->newlen+1, M_SYSCTL, M_WAITOK);
1294 
1295 	error = SYSCTL_IN(req, p, req->newlen);
1296 	if (error) {
1297 		if (p != buf)
1298 			free(p, M_SYSCTL);
1299 		return (error);
1300 	}
1301 
1302 	p [req->newlen] = '\0';
1303 
1304 	SYSCTL_RLOCK(&tracker);
1305 	error = name2oid(p, oid, &len, &op);
1306 	SYSCTL_RUNLOCK(&tracker);
1307 
1308 	if (p != buf)
1309 		free(p, M_SYSCTL);
1310 
1311 	if (error)
1312 		return (error);
1313 
1314 	error = SYSCTL_OUT(req, oid, len * sizeof *oid);
1315 	return (error);
1316 }
1317 
1318 /*
1319  * XXXRW/JA: Shouldn't return name2oid data for nodes that we don't permit in
1320  * capability mode.
1321  */
1322 SYSCTL_PROC(_sysctl, CTL_SYSCTL_NAME2OID, name2oid, CTLTYPE_INT | CTLFLAG_RW |
1323     CTLFLAG_ANYBODY | CTLFLAG_MPSAFE | CTLFLAG_CAPRW, 0, 0,
1324     sysctl_sysctl_name2oid, "I", "");
1325 
1326 static int
1327 sysctl_sysctl_oidfmt(SYSCTL_HANDLER_ARGS)
1328 {
1329 	struct sysctl_oid *oid;
1330 	struct rm_priotracker tracker;
1331 	int error;
1332 
1333 	error = sysctl_wire_old_buffer(req, 0);
1334 	if (error)
1335 		return (error);
1336 
1337 	SYSCTL_RLOCK(&tracker);
1338 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1339 	if (error)
1340 		goto out;
1341 
1342 	if (oid->oid_fmt == NULL) {
1343 		error = ENOENT;
1344 		goto out;
1345 	}
1346 	error = SYSCTL_OUT(req, &oid->oid_kind, sizeof(oid->oid_kind));
1347 	if (error)
1348 		goto out;
1349 	error = SYSCTL_OUT(req, oid->oid_fmt, strlen(oid->oid_fmt) + 1);
1350  out:
1351 	SYSCTL_RUNLOCK(&tracker);
1352 	return (error);
1353 }
1354 
1355 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_OIDFMT, oidfmt, CTLFLAG_RD |
1356     CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_oidfmt, "");
1357 
1358 static int
1359 sysctl_sysctl_oiddescr(SYSCTL_HANDLER_ARGS)
1360 {
1361 	struct sysctl_oid *oid;
1362 	struct rm_priotracker tracker;
1363 	int error;
1364 
1365 	error = sysctl_wire_old_buffer(req, 0);
1366 	if (error)
1367 		return (error);
1368 
1369 	SYSCTL_RLOCK(&tracker);
1370 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1371 	if (error)
1372 		goto out;
1373 
1374 	if (oid->oid_descr == NULL) {
1375 		error = ENOENT;
1376 		goto out;
1377 	}
1378 	error = SYSCTL_OUT(req, oid->oid_descr, strlen(oid->oid_descr) + 1);
1379  out:
1380 	SYSCTL_RUNLOCK(&tracker);
1381 	return (error);
1382 }
1383 
1384 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_OIDDESCR, oiddescr, CTLFLAG_RD |
1385     CTLFLAG_MPSAFE|CTLFLAG_CAPRD, sysctl_sysctl_oiddescr, "");
1386 
1387 static int
1388 sysctl_sysctl_oidlabel(SYSCTL_HANDLER_ARGS)
1389 {
1390 	struct sysctl_oid *oid;
1391 	struct rm_priotracker tracker;
1392 	int error;
1393 
1394 	error = sysctl_wire_old_buffer(req, 0);
1395 	if (error)
1396 		return (error);
1397 
1398 	SYSCTL_RLOCK(&tracker);
1399 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1400 	if (error)
1401 		goto out;
1402 
1403 	if (oid->oid_label == NULL) {
1404 		error = ENOENT;
1405 		goto out;
1406 	}
1407 	error = SYSCTL_OUT(req, oid->oid_label, strlen(oid->oid_label) + 1);
1408  out:
1409 	SYSCTL_RUNLOCK(&tracker);
1410 	return (error);
1411 }
1412 
1413 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_OIDLABEL, oidlabel, CTLFLAG_RD |
1414     CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_oidlabel, "");
1415 
1416 /*
1417  * Default "handler" functions.
1418  */
1419 
1420 /*
1421  * Handle a bool.
1422  * Two cases:
1423  *     a variable:  point arg1 at it.
1424  *     a constant:  pass it in arg2.
1425  */
1426 
1427 int
1428 sysctl_handle_bool(SYSCTL_HANDLER_ARGS)
1429 {
1430 	uint8_t temp;
1431 	int error;
1432 
1433 	/*
1434 	 * Attempt to get a coherent snapshot by making a copy of the data.
1435 	 */
1436 	if (arg1)
1437 		temp = *(bool *)arg1 ? 1 : 0;
1438 	else
1439 		temp = arg2 ? 1 : 0;
1440 
1441 	error = SYSCTL_OUT(req, &temp, sizeof(temp));
1442 	if (error || !req->newptr)
1443 		return (error);
1444 
1445 	if (!arg1)
1446 		error = EPERM;
1447 	else {
1448 		error = SYSCTL_IN(req, &temp, sizeof(temp));
1449 		if (!error)
1450 			*(bool *)arg1 = temp ? 1 : 0;
1451 	}
1452 	return (error);
1453 }
1454 
1455 /*
1456  * Handle an int8_t, signed or unsigned.
1457  * Two cases:
1458  *     a variable:  point arg1 at it.
1459  *     a constant:  pass it in arg2.
1460  */
1461 
1462 int
1463 sysctl_handle_8(SYSCTL_HANDLER_ARGS)
1464 {
1465 	int8_t tmpout;
1466 	int error = 0;
1467 
1468 	/*
1469 	 * Attempt to get a coherent snapshot by making a copy of the data.
1470 	 */
1471 	if (arg1)
1472 		tmpout = *(int8_t *)arg1;
1473 	else
1474 		tmpout = arg2;
1475 	error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout));
1476 
1477 	if (error || !req->newptr)
1478 		return (error);
1479 
1480 	if (!arg1)
1481 		error = EPERM;
1482 	else
1483 		error = SYSCTL_IN(req, arg1, sizeof(tmpout));
1484 	return (error);
1485 }
1486 
1487 /*
1488  * Handle an int16_t, signed or unsigned.
1489  * Two cases:
1490  *     a variable:  point arg1 at it.
1491  *     a constant:  pass it in arg2.
1492  */
1493 
1494 int
1495 sysctl_handle_16(SYSCTL_HANDLER_ARGS)
1496 {
1497 	int16_t tmpout;
1498 	int error = 0;
1499 
1500 	/*
1501 	 * Attempt to get a coherent snapshot by making a copy of the data.
1502 	 */
1503 	if (arg1)
1504 		tmpout = *(int16_t *)arg1;
1505 	else
1506 		tmpout = arg2;
1507 	error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout));
1508 
1509 	if (error || !req->newptr)
1510 		return (error);
1511 
1512 	if (!arg1)
1513 		error = EPERM;
1514 	else
1515 		error = SYSCTL_IN(req, arg1, sizeof(tmpout));
1516 	return (error);
1517 }
1518 
1519 /*
1520  * Handle an int32_t, signed or unsigned.
1521  * Two cases:
1522  *     a variable:  point arg1 at it.
1523  *     a constant:  pass it in arg2.
1524  */
1525 
1526 int
1527 sysctl_handle_32(SYSCTL_HANDLER_ARGS)
1528 {
1529 	int32_t tmpout;
1530 	int error = 0;
1531 
1532 	/*
1533 	 * Attempt to get a coherent snapshot by making a copy of the data.
1534 	 */
1535 	if (arg1)
1536 		tmpout = *(int32_t *)arg1;
1537 	else
1538 		tmpout = arg2;
1539 	error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout));
1540 
1541 	if (error || !req->newptr)
1542 		return (error);
1543 
1544 	if (!arg1)
1545 		error = EPERM;
1546 	else
1547 		error = SYSCTL_IN(req, arg1, sizeof(tmpout));
1548 	return (error);
1549 }
1550 
1551 /*
1552  * Handle an int, signed or unsigned.
1553  * Two cases:
1554  *     a variable:  point arg1 at it.
1555  *     a constant:  pass it in arg2.
1556  */
1557 
1558 int
1559 sysctl_handle_int(SYSCTL_HANDLER_ARGS)
1560 {
1561 	int tmpout, error = 0;
1562 
1563 	/*
1564 	 * Attempt to get a coherent snapshot by making a copy of the data.
1565 	 */
1566 	if (arg1)
1567 		tmpout = *(int *)arg1;
1568 	else
1569 		tmpout = arg2;
1570 	error = SYSCTL_OUT(req, &tmpout, sizeof(int));
1571 
1572 	if (error || !req->newptr)
1573 		return (error);
1574 
1575 	if (!arg1)
1576 		error = EPERM;
1577 	else
1578 		error = SYSCTL_IN(req, arg1, sizeof(int));
1579 	return (error);
1580 }
1581 
1582 /*
1583  * Based on on sysctl_handle_int() convert milliseconds into ticks.
1584  * Note: this is used by TCP.
1585  */
1586 
1587 int
1588 sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
1589 {
1590 	int error, s, tt;
1591 
1592 	tt = *(int *)arg1;
1593 	s = (int)((int64_t)tt * 1000 / hz);
1594 
1595 	error = sysctl_handle_int(oidp, &s, 0, req);
1596 	if (error || !req->newptr)
1597 		return (error);
1598 
1599 	tt = (int)((int64_t)s * hz / 1000);
1600 	if (tt < 1)
1601 		return (EINVAL);
1602 
1603 	*(int *)arg1 = tt;
1604 	return (0);
1605 }
1606 
1607 /*
1608  * Handle a long, signed or unsigned.
1609  * Two cases:
1610  *     a variable:  point arg1 at it.
1611  *     a constant:  pass it in arg2.
1612  */
1613 
1614 int
1615 sysctl_handle_long(SYSCTL_HANDLER_ARGS)
1616 {
1617 	int error = 0;
1618 	long tmplong;
1619 #ifdef SCTL_MASK32
1620 	int tmpint;
1621 #endif
1622 
1623 	/*
1624 	 * Attempt to get a coherent snapshot by making a copy of the data.
1625 	 */
1626 	if (arg1)
1627 		tmplong = *(long *)arg1;
1628 	else
1629 		tmplong = arg2;
1630 #ifdef SCTL_MASK32
1631 	if (req->flags & SCTL_MASK32) {
1632 		tmpint = tmplong;
1633 		error = SYSCTL_OUT(req, &tmpint, sizeof(int));
1634 	} else
1635 #endif
1636 		error = SYSCTL_OUT(req, &tmplong, sizeof(long));
1637 
1638 	if (error || !req->newptr)
1639 		return (error);
1640 
1641 	if (!arg1)
1642 		error = EPERM;
1643 #ifdef SCTL_MASK32
1644 	else if (req->flags & SCTL_MASK32) {
1645 		error = SYSCTL_IN(req, &tmpint, sizeof(int));
1646 		*(long *)arg1 = (long)tmpint;
1647 	}
1648 #endif
1649 	else
1650 		error = SYSCTL_IN(req, arg1, sizeof(long));
1651 	return (error);
1652 }
1653 
1654 /*
1655  * Handle a 64 bit int, signed or unsigned.
1656  * Two cases:
1657  *     a variable:  point arg1 at it.
1658  *     a constant:  pass it in arg2.
1659  */
1660 int
1661 sysctl_handle_64(SYSCTL_HANDLER_ARGS)
1662 {
1663 	int error = 0;
1664 	uint64_t tmpout;
1665 
1666 	/*
1667 	 * Attempt to get a coherent snapshot by making a copy of the data.
1668 	 */
1669 	if (arg1)
1670 		tmpout = *(uint64_t *)arg1;
1671 	else
1672 		tmpout = arg2;
1673 	error = SYSCTL_OUT(req, &tmpout, sizeof(uint64_t));
1674 
1675 	if (error || !req->newptr)
1676 		return (error);
1677 
1678 	if (!arg1)
1679 		error = EPERM;
1680 	else
1681 		error = SYSCTL_IN(req, arg1, sizeof(uint64_t));
1682 	return (error);
1683 }
1684 
1685 /*
1686  * Handle our generic '\0' terminated 'C' string.
1687  * Two cases:
1688  * 	a variable string:  point arg1 at it, arg2 is max length.
1689  * 	a constant string:  point arg1 at it, arg2 is zero.
1690  */
1691 
1692 int
1693 sysctl_handle_string(SYSCTL_HANDLER_ARGS)
1694 {
1695 	char *tmparg;
1696 	size_t outlen;
1697 	int error = 0, ro_string = 0;
1698 
1699 	/*
1700 	 * If the sysctl isn't writable and isn't a preallocated tunable that
1701 	 * can be modified by kenv(2), microoptimise and treat it as a
1702 	 * read-only string.
1703 	 * A zero-length buffer indicates a fixed size read-only
1704 	 * string.  In ddb, don't worry about trying to make a malloced
1705 	 * snapshot.
1706 	 */
1707 	if ((oidp->oid_kind & (CTLFLAG_WR | CTLFLAG_TUN)) == 0 ||
1708 	    arg2 == 0 || kdb_active) {
1709 		arg2 = strlen((char *)arg1) + 1;
1710 		ro_string = 1;
1711 	}
1712 
1713 	if (req->oldptr != NULL) {
1714 		if (ro_string) {
1715 			tmparg = arg1;
1716 			outlen = strlen(tmparg) + 1;
1717 		} else {
1718 			tmparg = malloc(arg2, M_SYSCTLTMP, M_WAITOK);
1719 			sx_slock(&sysctlstringlock);
1720 			memcpy(tmparg, arg1, arg2);
1721 			sx_sunlock(&sysctlstringlock);
1722 			outlen = strlen(tmparg) + 1;
1723 		}
1724 
1725 		error = SYSCTL_OUT(req, tmparg, outlen);
1726 
1727 		if (!ro_string)
1728 			free(tmparg, M_SYSCTLTMP);
1729 	} else {
1730 		if (!ro_string)
1731 			sx_slock(&sysctlstringlock);
1732 		outlen = strlen((char *)arg1) + 1;
1733 		if (!ro_string)
1734 			sx_sunlock(&sysctlstringlock);
1735 		error = SYSCTL_OUT(req, NULL, outlen);
1736 	}
1737 	if (error || !req->newptr)
1738 		return (error);
1739 
1740 	if (req->newlen - req->newidx >= arg2 ||
1741 	    req->newlen - req->newidx < 0) {
1742 		error = EINVAL;
1743 	} else if (req->newlen - req->newidx == 0) {
1744 		sx_xlock(&sysctlstringlock);
1745 		((char *)arg1)[0] = '\0';
1746 		sx_xunlock(&sysctlstringlock);
1747 	} else {
1748 		arg2 = req->newlen - req->newidx;
1749 		tmparg = malloc(arg2, M_SYSCTLTMP, M_WAITOK);
1750 
1751 		error = SYSCTL_IN(req, tmparg, arg2);
1752 		if (error) {
1753 			free(tmparg, M_SYSCTLTMP);
1754 			return (error);
1755 		}
1756 
1757 		sx_xlock(&sysctlstringlock);
1758 		memcpy(arg1, tmparg, arg2);
1759 		((char *)arg1)[arg2] = '\0';
1760 		sx_xunlock(&sysctlstringlock);
1761 		free(tmparg, M_SYSCTLTMP);
1762 		req->newidx += arg2;
1763 	}
1764 	return (error);
1765 }
1766 
1767 /*
1768  * Handle any kind of opaque data.
1769  * arg1 points to it, arg2 is the size.
1770  */
1771 
1772 int
1773 sysctl_handle_opaque(SYSCTL_HANDLER_ARGS)
1774 {
1775 	int error, tries;
1776 	u_int generation;
1777 	struct sysctl_req req2;
1778 
1779 	/*
1780 	 * Attempt to get a coherent snapshot, by using the thread
1781 	 * pre-emption counter updated from within mi_switch() to
1782 	 * determine if we were pre-empted during a bcopy() or
1783 	 * copyout(). Make 3 attempts at doing this before giving up.
1784 	 * If we encounter an error, stop immediately.
1785 	 */
1786 	tries = 0;
1787 	req2 = *req;
1788 retry:
1789 	generation = curthread->td_generation;
1790 	error = SYSCTL_OUT(req, arg1, arg2);
1791 	if (error)
1792 		return (error);
1793 	tries++;
1794 	if (generation != curthread->td_generation && tries < 3) {
1795 		*req = req2;
1796 		goto retry;
1797 	}
1798 
1799 	error = SYSCTL_IN(req, arg1, arg2);
1800 
1801 	return (error);
1802 }
1803 
1804 /*
1805  * Based on on sysctl_handle_int() convert microseconds to a sbintime.
1806  */
1807 int
1808 sysctl_usec_to_sbintime(SYSCTL_HANDLER_ARGS)
1809 {
1810 	int error;
1811 	int64_t tt;
1812 	sbintime_t sb;
1813 
1814 	tt = *(int64_t *)arg1;
1815 	sb = sbttous(tt);
1816 
1817 	error = sysctl_handle_64(oidp, &sb, 0, req);
1818 	if (error || !req->newptr)
1819 		return (error);
1820 
1821 	tt = ustosbt(sb);
1822 	*(int64_t *)arg1 = tt;
1823 
1824 	return (0);
1825 }
1826 
1827 /*
1828  * Based on on sysctl_handle_int() convert milliseconds to a sbintime.
1829  */
1830 int
1831 sysctl_msec_to_sbintime(SYSCTL_HANDLER_ARGS)
1832 {
1833 	int error;
1834 	int64_t tt;
1835 	sbintime_t sb;
1836 
1837 	tt = *(int64_t *)arg1;
1838 	sb = sbttoms(tt);
1839 
1840 	error = sysctl_handle_64(oidp, &sb, 0, req);
1841 	if (error || !req->newptr)
1842 		return (error);
1843 
1844 	tt = mstosbt(sb);
1845 	*(int64_t *)arg1 = tt;
1846 
1847 	return (0);
1848 }
1849 
1850 /*
1851  * Convert seconds to a struct timeval.  Intended for use with
1852  * intervals and thus does not permit negative seconds.
1853  */
1854 int
1855 sysctl_sec_to_timeval(SYSCTL_HANDLER_ARGS)
1856 {
1857 	struct timeval *tv;
1858 	int error, secs;
1859 
1860 	tv = arg1;
1861 	secs = tv->tv_sec;
1862 
1863 	error = sysctl_handle_int(oidp, &secs, 0, req);
1864 	if (error || req->newptr == NULL)
1865 		return (error);
1866 
1867 	if (secs < 0)
1868 		return (EINVAL);
1869 	tv->tv_sec = secs;
1870 
1871 	return (0);
1872 }
1873 
1874 /*
1875  * Transfer functions to/from kernel space.
1876  * XXX: rather untested at this point
1877  */
1878 static int
1879 sysctl_old_kernel(struct sysctl_req *req, const void *p, size_t l)
1880 {
1881 	size_t i = 0;
1882 
1883 	if (req->oldptr) {
1884 		i = l;
1885 		if (req->oldlen <= req->oldidx)
1886 			i = 0;
1887 		else
1888 			if (i > req->oldlen - req->oldidx)
1889 				i = req->oldlen - req->oldidx;
1890 		if (i > 0)
1891 			bcopy(p, (char *)req->oldptr + req->oldidx, i);
1892 	}
1893 	req->oldidx += l;
1894 	if (req->oldptr && i != l)
1895 		return (ENOMEM);
1896 	return (0);
1897 }
1898 
1899 static int
1900 sysctl_new_kernel(struct sysctl_req *req, void *p, size_t l)
1901 {
1902 	if (!req->newptr)
1903 		return (0);
1904 	if (req->newlen - req->newidx < l)
1905 		return (EINVAL);
1906 	bcopy((const char *)req->newptr + req->newidx, p, l);
1907 	req->newidx += l;
1908 	return (0);
1909 }
1910 
1911 int
1912 kernel_sysctl(struct thread *td, int *name, u_int namelen, void *old,
1913     size_t *oldlenp, void *new, size_t newlen, size_t *retval, int flags)
1914 {
1915 	int error = 0;
1916 	struct sysctl_req req;
1917 
1918 	bzero(&req, sizeof req);
1919 
1920 	req.td = td;
1921 	req.flags = flags;
1922 
1923 	if (oldlenp) {
1924 		req.oldlen = *oldlenp;
1925 	}
1926 	req.validlen = req.oldlen;
1927 
1928 	if (old) {
1929 		req.oldptr= old;
1930 	}
1931 
1932 	if (new != NULL) {
1933 		req.newlen = newlen;
1934 		req.newptr = new;
1935 	}
1936 
1937 	req.oldfunc = sysctl_old_kernel;
1938 	req.newfunc = sysctl_new_kernel;
1939 	req.lock = REQ_UNWIRED;
1940 
1941 	error = sysctl_root(0, name, namelen, &req);
1942 
1943 	if (req.lock == REQ_WIRED && req.validlen > 0)
1944 		vsunlock(req.oldptr, req.validlen);
1945 
1946 	if (error && error != ENOMEM)
1947 		return (error);
1948 
1949 	if (retval) {
1950 		if (req.oldptr && req.oldidx > req.validlen)
1951 			*retval = req.validlen;
1952 		else
1953 			*retval = req.oldidx;
1954 	}
1955 	return (error);
1956 }
1957 
1958 int
1959 kernel_sysctlbyname(struct thread *td, char *name, void *old, size_t *oldlenp,
1960     void *new, size_t newlen, size_t *retval, int flags)
1961 {
1962         int oid[CTL_MAXNAME];
1963         size_t oidlen, plen;
1964 	int error;
1965 
1966 	oid[0] = CTL_SYSCTL;
1967 	oid[1] = CTL_SYSCTL_NAME2OID;
1968 	oidlen = sizeof(oid);
1969 
1970 	error = kernel_sysctl(td, oid, 2, oid, &oidlen,
1971 	    (void *)name, strlen(name), &plen, flags);
1972 	if (error)
1973 		return (error);
1974 
1975 	error = kernel_sysctl(td, oid, plen / sizeof(int), old, oldlenp,
1976 	    new, newlen, retval, flags);
1977 	return (error);
1978 }
1979 
1980 /*
1981  * Transfer function to/from user space.
1982  */
1983 static int
1984 sysctl_old_user(struct sysctl_req *req, const void *p, size_t l)
1985 {
1986 	size_t i, len, origidx;
1987 	int error;
1988 
1989 	origidx = req->oldidx;
1990 	req->oldidx += l;
1991 	if (req->oldptr == NULL)
1992 		return (0);
1993 	/*
1994 	 * If we have not wired the user supplied buffer and we are currently
1995 	 * holding locks, drop a witness warning, as it's possible that
1996 	 * write operations to the user page can sleep.
1997 	 */
1998 	if (req->lock != REQ_WIRED)
1999 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
2000 		    "sysctl_old_user()");
2001 	i = l;
2002 	len = req->validlen;
2003 	if (len <= origidx)
2004 		i = 0;
2005 	else {
2006 		if (i > len - origidx)
2007 			i = len - origidx;
2008 		if (req->lock == REQ_WIRED) {
2009 			error = copyout_nofault(p, (char *)req->oldptr +
2010 			    origidx, i);
2011 		} else
2012 			error = copyout(p, (char *)req->oldptr + origidx, i);
2013 		if (error != 0)
2014 			return (error);
2015 	}
2016 	if (i < l)
2017 		return (ENOMEM);
2018 	return (0);
2019 }
2020 
2021 static int
2022 sysctl_new_user(struct sysctl_req *req, void *p, size_t l)
2023 {
2024 	int error;
2025 
2026 	if (!req->newptr)
2027 		return (0);
2028 	if (req->newlen - req->newidx < l)
2029 		return (EINVAL);
2030 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
2031 	    "sysctl_new_user()");
2032 	error = copyin((const char *)req->newptr + req->newidx, p, l);
2033 	req->newidx += l;
2034 	return (error);
2035 }
2036 
2037 /*
2038  * Wire the user space destination buffer.  If set to a value greater than
2039  * zero, the len parameter limits the maximum amount of wired memory.
2040  */
2041 int
2042 sysctl_wire_old_buffer(struct sysctl_req *req, size_t len)
2043 {
2044 	int ret;
2045 	size_t wiredlen;
2046 
2047 	wiredlen = (len > 0 && len < req->oldlen) ? len : req->oldlen;
2048 	ret = 0;
2049 	if (req->lock != REQ_WIRED && req->oldptr &&
2050 	    req->oldfunc == sysctl_old_user) {
2051 		if (wiredlen != 0) {
2052 			ret = vslock(req->oldptr, wiredlen);
2053 			if (ret != 0) {
2054 				if (ret != ENOMEM)
2055 					return (ret);
2056 				wiredlen = 0;
2057 			}
2058 		}
2059 		req->lock = REQ_WIRED;
2060 		req->validlen = wiredlen;
2061 	}
2062 	return (0);
2063 }
2064 
2065 int
2066 sysctl_find_oid(int *name, u_int namelen, struct sysctl_oid **noid,
2067     int *nindx, struct sysctl_req *req)
2068 {
2069 	struct sysctl_oid_list *lsp;
2070 	struct sysctl_oid *oid;
2071 	int indx;
2072 
2073 	SYSCTL_ASSERT_LOCKED();
2074 	lsp = &sysctl__children;
2075 	indx = 0;
2076 	while (indx < CTL_MAXNAME) {
2077 		SLIST_FOREACH(oid, lsp, oid_link) {
2078 			if (oid->oid_number == name[indx])
2079 				break;
2080 		}
2081 		if (oid == NULL)
2082 			return (ENOENT);
2083 
2084 		indx++;
2085 		if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
2086 			if (oid->oid_handler != NULL || indx == namelen) {
2087 				*noid = oid;
2088 				if (nindx != NULL)
2089 					*nindx = indx;
2090 				KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
2091 				    ("%s found DYING node %p", __func__, oid));
2092 				return (0);
2093 			}
2094 			lsp = SYSCTL_CHILDREN(oid);
2095 		} else if (indx == namelen) {
2096 			if ((oid->oid_kind & CTLFLAG_DORMANT) != 0)
2097 				return (ENOENT);
2098 			*noid = oid;
2099 			if (nindx != NULL)
2100 				*nindx = indx;
2101 			KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
2102 			    ("%s found DYING node %p", __func__, oid));
2103 			return (0);
2104 		} else {
2105 			return (ENOTDIR);
2106 		}
2107 	}
2108 	return (ENOENT);
2109 }
2110 
2111 /*
2112  * Traverse our tree, and find the right node, execute whatever it points
2113  * to, and return the resulting error code.
2114  */
2115 
2116 static int
2117 sysctl_root(SYSCTL_HANDLER_ARGS)
2118 {
2119 	struct sysctl_oid *oid;
2120 	struct rm_priotracker tracker;
2121 	int error, indx, lvl;
2122 
2123 	SYSCTL_RLOCK(&tracker);
2124 
2125 	error = sysctl_find_oid(arg1, arg2, &oid, &indx, req);
2126 	if (error)
2127 		goto out;
2128 
2129 	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
2130 		/*
2131 		 * You can't call a sysctl when it's a node, but has
2132 		 * no handler.  Inform the user that it's a node.
2133 		 * The indx may or may not be the same as namelen.
2134 		 */
2135 		if (oid->oid_handler == NULL) {
2136 			error = EISDIR;
2137 			goto out;
2138 		}
2139 	}
2140 
2141 	/* Is this sysctl writable? */
2142 	if (req->newptr && !(oid->oid_kind & CTLFLAG_WR)) {
2143 		error = EPERM;
2144 		goto out;
2145 	}
2146 
2147 	KASSERT(req->td != NULL, ("sysctl_root(): req->td == NULL"));
2148 
2149 #ifdef CAPABILITY_MODE
2150 	/*
2151 	 * If the process is in capability mode, then don't permit reading or
2152 	 * writing unless specifically granted for the node.
2153 	 */
2154 	if (IN_CAPABILITY_MODE(req->td)) {
2155 		if ((req->oldptr && !(oid->oid_kind & CTLFLAG_CAPRD)) ||
2156 		    (req->newptr && !(oid->oid_kind & CTLFLAG_CAPWR))) {
2157 			error = EPERM;
2158 			goto out;
2159 		}
2160 	}
2161 #endif
2162 
2163 	/* Is this sysctl sensitive to securelevels? */
2164 	if (req->newptr && (oid->oid_kind & CTLFLAG_SECURE)) {
2165 		lvl = (oid->oid_kind & CTLMASK_SECURE) >> CTLSHIFT_SECURE;
2166 		error = securelevel_gt(req->td->td_ucred, lvl);
2167 		if (error)
2168 			goto out;
2169 	}
2170 
2171 	/* Is this sysctl writable by only privileged users? */
2172 	if (req->newptr && !(oid->oid_kind & CTLFLAG_ANYBODY)) {
2173 		int priv;
2174 
2175 		if (oid->oid_kind & CTLFLAG_PRISON)
2176 			priv = PRIV_SYSCTL_WRITEJAIL;
2177 #ifdef VIMAGE
2178 		else if ((oid->oid_kind & CTLFLAG_VNET) &&
2179 		     prison_owns_vnet(req->td->td_ucred))
2180 			priv = PRIV_SYSCTL_WRITEJAIL;
2181 #endif
2182 		else
2183 			priv = PRIV_SYSCTL_WRITE;
2184 		error = priv_check(req->td, priv);
2185 		if (error)
2186 			goto out;
2187 	}
2188 
2189 	if (!oid->oid_handler) {
2190 		error = EINVAL;
2191 		goto out;
2192 	}
2193 
2194 	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
2195 		arg1 = (int *)arg1 + indx;
2196 		arg2 -= indx;
2197 	} else {
2198 		arg1 = oid->oid_arg1;
2199 		arg2 = oid->oid_arg2;
2200 	}
2201 #ifdef MAC
2202 	error = mac_system_check_sysctl(req->td->td_ucred, oid, arg1, arg2,
2203 	    req);
2204 	if (error != 0)
2205 		goto out;
2206 #endif
2207 #ifdef VIMAGE
2208 	if ((oid->oid_kind & CTLFLAG_VNET) && arg1 != NULL)
2209 		arg1 = (void *)(curvnet->vnet_data_base + (uintptr_t)arg1);
2210 #endif
2211 	error = sysctl_root_handler_locked(oid, arg1, arg2, req, &tracker);
2212 
2213 out:
2214 	SYSCTL_RUNLOCK(&tracker);
2215 	return (error);
2216 }
2217 
2218 #ifndef _SYS_SYSPROTO_H_
2219 struct sysctl_args {
2220 	int	*name;
2221 	u_int	namelen;
2222 	void	*old;
2223 	size_t	*oldlenp;
2224 	void	*new;
2225 	size_t	newlen;
2226 };
2227 #endif
2228 int
2229 sys___sysctl(struct thread *td, struct sysctl_args *uap)
2230 {
2231 	int error, i, name[CTL_MAXNAME];
2232 	size_t j;
2233 
2234 	if (uap->namelen > CTL_MAXNAME || uap->namelen < 2)
2235 		return (EINVAL);
2236 
2237  	error = copyin(uap->name, &name, uap->namelen * sizeof(int));
2238  	if (error)
2239 		return (error);
2240 
2241 	error = userland_sysctl(td, name, uap->namelen,
2242 		uap->old, uap->oldlenp, 0,
2243 		uap->new, uap->newlen, &j, 0);
2244 	if (error && error != ENOMEM)
2245 		return (error);
2246 	if (uap->oldlenp) {
2247 		i = copyout(&j, uap->oldlenp, sizeof(j));
2248 		if (i)
2249 			return (i);
2250 	}
2251 	return (error);
2252 }
2253 
2254 int
2255 kern___sysctlbyname(struct thread *td, const char *oname, size_t namelen,
2256     void *old, size_t *oldlenp, void *new, size_t newlen, size_t *retval,
2257     int flags, bool inkernel)
2258 {
2259 	int oid[CTL_MAXNAME];
2260 	char namebuf[16];
2261 	char *name;
2262 	size_t oidlen;
2263 	int error;
2264 
2265 	if (namelen > MAXPATHLEN || namelen == 0)
2266 		return (EINVAL);
2267 	name = namebuf;
2268 	if (namelen > sizeof(namebuf))
2269 		name = malloc(namelen, M_SYSCTL, M_WAITOK);
2270 	error = copyin(oname, name, namelen);
2271 	if (error != 0)
2272 		goto out;
2273 
2274 	oid[0] = CTL_SYSCTL;
2275 	oid[1] = CTL_SYSCTL_NAME2OID;
2276 	oidlen = sizeof(oid);
2277 	error = kernel_sysctl(td, oid, 2, oid, &oidlen, (void *)name, namelen,
2278 	    retval, flags);
2279 	if (error != 0)
2280 		goto out;
2281 	error = userland_sysctl(td, oid, *retval / sizeof(int), old, oldlenp,
2282 	    inkernel, new, newlen, retval, flags);
2283 
2284 out:
2285 	if (namelen > sizeof(namebuf))
2286 		free(name, M_SYSCTL);
2287 	return (error);
2288 }
2289 
2290 #ifndef	_SYS_SYSPROTO_H_
2291 struct __sysctlbyname_args {
2292 	const char	*name;
2293 	size_t	namelen;
2294 	void	*old;
2295 	size_t	*oldlenp;
2296 	void	*new;
2297 	size_t	newlen;
2298 };
2299 #endif
2300 int
2301 sys___sysctlbyname(struct thread *td, struct __sysctlbyname_args *uap)
2302 {
2303 	size_t rv;
2304 	int error;
2305 
2306 	error = kern___sysctlbyname(td, uap->name, uap->namelen, uap->old,
2307 	    uap->oldlenp, uap->new, uap->newlen, &rv, 0, 0);
2308 	if (error != 0)
2309 		return (error);
2310 	if (uap->oldlenp != NULL)
2311 		error = copyout(&rv, uap->oldlenp, sizeof(rv));
2312 
2313 	return (error);
2314 }
2315 
2316 /*
2317  * This is used from various compatibility syscalls too.  That's why name
2318  * must be in kernel space.
2319  */
2320 int
2321 userland_sysctl(struct thread *td, int *name, u_int namelen, void *old,
2322     size_t *oldlenp, int inkernel, const void *new, size_t newlen,
2323     size_t *retval, int flags)
2324 {
2325 	int error = 0, memlocked;
2326 	struct sysctl_req req;
2327 
2328 	bzero(&req, sizeof req);
2329 
2330 	req.td = td;
2331 	req.flags = flags;
2332 
2333 	if (oldlenp) {
2334 		if (inkernel) {
2335 			req.oldlen = *oldlenp;
2336 		} else {
2337 			error = copyin(oldlenp, &req.oldlen, sizeof(*oldlenp));
2338 			if (error)
2339 				return (error);
2340 		}
2341 	}
2342 	req.validlen = req.oldlen;
2343 	req.oldptr = old;
2344 
2345 	if (new != NULL) {
2346 		req.newlen = newlen;
2347 		req.newptr = new;
2348 	}
2349 
2350 	req.oldfunc = sysctl_old_user;
2351 	req.newfunc = sysctl_new_user;
2352 	req.lock = REQ_UNWIRED;
2353 
2354 #ifdef KTRACE
2355 	if (KTRPOINT(curthread, KTR_SYSCTL))
2356 		ktrsysctl(name, namelen);
2357 #endif
2358 	memlocked = 0;
2359 	if (req.oldptr && req.oldlen > 4 * PAGE_SIZE) {
2360 		memlocked = 1;
2361 		sx_xlock(&sysctlmemlock);
2362 	}
2363 	CURVNET_SET(TD_TO_VNET(td));
2364 
2365 	for (;;) {
2366 		req.oldidx = 0;
2367 		req.newidx = 0;
2368 		error = sysctl_root(0, name, namelen, &req);
2369 		if (error != EAGAIN)
2370 			break;
2371 		kern_yield(PRI_USER);
2372 	}
2373 
2374 	CURVNET_RESTORE();
2375 
2376 	if (req.lock == REQ_WIRED && req.validlen > 0)
2377 		vsunlock(req.oldptr, req.validlen);
2378 	if (memlocked)
2379 		sx_xunlock(&sysctlmemlock);
2380 
2381 	if (error && error != ENOMEM)
2382 		return (error);
2383 
2384 	if (retval) {
2385 		if (req.oldptr && req.oldidx > req.validlen)
2386 			*retval = req.validlen;
2387 		else
2388 			*retval = req.oldidx;
2389 	}
2390 	return (error);
2391 }
2392 
2393 /*
2394  * Drain into a sysctl struct.  The user buffer should be wired if a page
2395  * fault would cause issue.
2396  */
2397 static int
2398 sbuf_sysctl_drain(void *arg, const char *data, int len)
2399 {
2400 	struct sysctl_req *req = arg;
2401 	int error;
2402 
2403 	error = SYSCTL_OUT(req, data, len);
2404 	KASSERT(error >= 0, ("Got unexpected negative value %d", error));
2405 	return (error == 0 ? len : -error);
2406 }
2407 
2408 struct sbuf *
2409 sbuf_new_for_sysctl(struct sbuf *s, char *buf, int length,
2410     struct sysctl_req *req)
2411 {
2412 
2413 	/* Supply a default buffer size if none given. */
2414 	if (buf == NULL && length == 0)
2415 		length = 64;
2416 	s = sbuf_new(s, buf, length, SBUF_FIXEDLEN | SBUF_INCLUDENUL);
2417 	sbuf_set_drain(s, sbuf_sysctl_drain, req);
2418 	return (s);
2419 }
2420 
2421 #ifdef DDB
2422 
2423 /* The current OID the debugger is working with */
2424 static struct sysctl_oid *g_ddb_oid;
2425 
2426 /* The current flags specified by the user */
2427 static int g_ddb_sysctl_flags;
2428 
2429 /* Check to see if the last sysctl printed */
2430 static int g_ddb_sysctl_printed;
2431 
2432 static const int ctl_sign[CTLTYPE+1] = {
2433 	[CTLTYPE_INT] = 1,
2434 	[CTLTYPE_LONG] = 1,
2435 	[CTLTYPE_S8] = 1,
2436 	[CTLTYPE_S16] = 1,
2437 	[CTLTYPE_S32] = 1,
2438 	[CTLTYPE_S64] = 1,
2439 };
2440 
2441 static const int ctl_size[CTLTYPE+1] = {
2442 	[CTLTYPE_INT] = sizeof(int),
2443 	[CTLTYPE_UINT] = sizeof(u_int),
2444 	[CTLTYPE_LONG] = sizeof(long),
2445 	[CTLTYPE_ULONG] = sizeof(u_long),
2446 	[CTLTYPE_S8] = sizeof(int8_t),
2447 	[CTLTYPE_S16] = sizeof(int16_t),
2448 	[CTLTYPE_S32] = sizeof(int32_t),
2449 	[CTLTYPE_S64] = sizeof(int64_t),
2450 	[CTLTYPE_U8] = sizeof(uint8_t),
2451 	[CTLTYPE_U16] = sizeof(uint16_t),
2452 	[CTLTYPE_U32] = sizeof(uint32_t),
2453 	[CTLTYPE_U64] = sizeof(uint64_t),
2454 };
2455 
2456 #define DB_SYSCTL_NAME_ONLY	0x001	/* Compare with -N */
2457 #define DB_SYSCTL_VALUE_ONLY	0x002	/* Compare with -n */
2458 #define DB_SYSCTL_OPAQUE	0x004	/* Compare with -o */
2459 #define DB_SYSCTL_HEX		0x008	/* Compare with -x */
2460 
2461 #define DB_SYSCTL_SAFE_ONLY	0x100	/* Only simple types */
2462 
2463 static const char db_sysctl_modifs[] = {
2464 	'N', 'n', 'o', 'x',
2465 };
2466 
2467 static const int db_sysctl_modif_values[] = {
2468 	DB_SYSCTL_NAME_ONLY, DB_SYSCTL_VALUE_ONLY,
2469 	DB_SYSCTL_OPAQUE, DB_SYSCTL_HEX,
2470 };
2471 
2472 /* Handlers considered safe to print while recursing */
2473 static int (* const db_safe_handlers[])(SYSCTL_HANDLER_ARGS) = {
2474 	sysctl_handle_bool,
2475 	sysctl_handle_8,
2476 	sysctl_handle_16,
2477 	sysctl_handle_32,
2478 	sysctl_handle_64,
2479 	sysctl_handle_int,
2480 	sysctl_handle_long,
2481 	sysctl_handle_string,
2482 	sysctl_handle_opaque,
2483 };
2484 
2485 /*
2486  * Use in place of sysctl_old_kernel to print sysctl values.
2487  *
2488  * Compare to the output handling in show_var from sbin/sysctl/sysctl.c
2489  */
2490 static int
2491 sysctl_old_ddb(struct sysctl_req *req, const void *ptr, size_t len)
2492 {
2493 	const u_char *val, *p;
2494 	const char *sep1;
2495 	size_t intlen, slen;
2496 	uintmax_t umv;
2497 	intmax_t mv;
2498 	int sign, ctltype, hexlen, xflag, error;
2499 
2500 	/* Suppress false-positive GCC uninitialized variable warnings */
2501 	mv = 0;
2502 	umv = 0;
2503 
2504 	slen = len;
2505 	val = p = ptr;
2506 
2507 	if (ptr == NULL) {
2508 		error = 0;
2509 		goto out;
2510 	}
2511 
2512 	/* We are going to print */
2513 	g_ddb_sysctl_printed = 1;
2514 
2515 	xflag = g_ddb_sysctl_flags & DB_SYSCTL_HEX;
2516 
2517 	ctltype = (g_ddb_oid->oid_kind & CTLTYPE);
2518 	sign = ctl_sign[ctltype];
2519 	intlen = ctl_size[ctltype];
2520 
2521 	switch (ctltype) {
2522 	case CTLTYPE_NODE:
2523 	case CTLTYPE_STRING:
2524 		db_printf("%.*s", (int) len, (const char *) p);
2525 		error = 0;
2526 		goto out;
2527 
2528 	case CTLTYPE_INT:
2529 	case CTLTYPE_UINT:
2530 	case CTLTYPE_LONG:
2531 	case CTLTYPE_ULONG:
2532 	case CTLTYPE_S8:
2533 	case CTLTYPE_S16:
2534 	case CTLTYPE_S32:
2535 	case CTLTYPE_S64:
2536 	case CTLTYPE_U8:
2537 	case CTLTYPE_U16:
2538 	case CTLTYPE_U32:
2539 	case CTLTYPE_U64:
2540 		hexlen = 2 + (intlen * CHAR_BIT + 3) / 4;
2541 		sep1 = "";
2542 		while (len >= intlen) {
2543 			switch (ctltype) {
2544 			case CTLTYPE_INT:
2545 			case CTLTYPE_UINT:
2546 				umv = *(const u_int *)p;
2547 				mv = *(const int *)p;
2548 				break;
2549 			case CTLTYPE_LONG:
2550 			case CTLTYPE_ULONG:
2551 				umv = *(const u_long *)p;
2552 				mv = *(const long *)p;
2553 				break;
2554 			case CTLTYPE_S8:
2555 			case CTLTYPE_U8:
2556 				umv = *(const uint8_t *)p;
2557 				mv = *(const int8_t *)p;
2558 				break;
2559 			case CTLTYPE_S16:
2560 			case CTLTYPE_U16:
2561 				umv = *(const uint16_t *)p;
2562 				mv = *(const int16_t *)p;
2563 				break;
2564 			case CTLTYPE_S32:
2565 			case CTLTYPE_U32:
2566 				umv = *(const uint32_t *)p;
2567 				mv = *(const int32_t *)p;
2568 				break;
2569 			case CTLTYPE_S64:
2570 			case CTLTYPE_U64:
2571 				umv = *(const uint64_t *)p;
2572 				mv = *(const int64_t *)p;
2573 				break;
2574 			}
2575 
2576 			db_printf("%s", sep1);
2577 			if (xflag)
2578 				db_printf("%#0*jx", hexlen, umv);
2579 			else if (!sign)
2580 				db_printf("%ju", umv);
2581 			else if (g_ddb_oid->oid_fmt[1] == 'K') {
2582 				/* Kelvins are currently unsupported. */
2583 				error = EOPNOTSUPP;
2584 				goto out;
2585 			} else
2586 				db_printf("%jd", mv);
2587 
2588 			sep1 = " ";
2589 			len -= intlen;
2590 			p += intlen;
2591 		}
2592 		error = 0;
2593 		goto out;
2594 
2595 	case CTLTYPE_OPAQUE:
2596 		/* TODO: Support struct functions. */
2597 
2598 		/* FALLTHROUGH */
2599 	default:
2600 		db_printf("Format:%s Length:%zu Dump:0x",
2601 		    g_ddb_oid->oid_fmt, len);
2602 		while (len-- && (xflag || p < val + 16))
2603 			db_printf("%02x", *p++);
2604 		if (!xflag && len > 16)
2605 			db_printf("...");
2606 		error = 0;
2607 		goto out;
2608 	}
2609 
2610 out:
2611 	req->oldidx += slen;
2612 	return (error);
2613 }
2614 
2615 /*
2616  * Avoid setting new sysctl values from the debugger
2617  */
2618 static int
2619 sysctl_new_ddb(struct sysctl_req *req, void *p, size_t l)
2620 {
2621 
2622 	if (!req->newptr)
2623 		return (0);
2624 
2625 	/* Changing sysctls from the debugger is currently unsupported */
2626 	return (EPERM);
2627 }
2628 
2629 /*
2630  * Run a sysctl handler with the DDB oldfunc and newfunc attached.
2631  * Instead of copying any output to a buffer we'll dump it right to
2632  * the console.
2633  */
2634 static int
2635 db_sysctl(struct sysctl_oid *oidp, int *name, u_int namelen,
2636     void *old, size_t *oldlenp, size_t *retval, int flags)
2637 {
2638 	struct sysctl_req req;
2639 	int error;
2640 
2641 	/* Setup the request */
2642 	bzero(&req, sizeof req);
2643 	req.td = kdb_thread;
2644 	req.oldfunc = sysctl_old_ddb;
2645 	req.newfunc = sysctl_new_ddb;
2646 	req.lock = REQ_UNWIRED;
2647 	if (oldlenp) {
2648 		req.oldlen = *oldlenp;
2649 	}
2650 	req.validlen = req.oldlen;
2651 	if (old) {
2652 		req.oldptr = old;
2653 	}
2654 
2655 	/* Setup our globals for sysctl_old_ddb */
2656 	g_ddb_oid = oidp;
2657 	g_ddb_sysctl_flags = flags;
2658 	g_ddb_sysctl_printed = 0;
2659 
2660 	error = sysctl_root(0, name, namelen, &req);
2661 
2662 	/* Reset globals */
2663 	g_ddb_oid = NULL;
2664 	g_ddb_sysctl_flags = 0;
2665 
2666 	if (retval) {
2667 		if (req.oldptr && req.oldidx > req.validlen)
2668 			*retval = req.validlen;
2669 		else
2670 			*retval = req.oldidx;
2671 	}
2672 	return (error);
2673 }
2674 
2675 /*
2676  * Show a sysctl's name
2677  */
2678 static void
2679 db_show_oid_name(int *oid, size_t nlen)
2680 {
2681 	struct sysctl_oid *oidp;
2682 	int qoid[CTL_MAXNAME+2];
2683 	int error;
2684 
2685 	qoid[0] = 0;
2686 	memcpy(qoid + 2, oid, nlen * sizeof(int));
2687 	qoid[1] = 1;
2688 
2689 	error = sysctl_find_oid(qoid, nlen + 2, &oidp, NULL, NULL);
2690 	if (error)
2691 		db_error("sysctl name oid");
2692 
2693 	error = db_sysctl(oidp, qoid, nlen + 2, NULL, NULL, NULL, 0);
2694 	if (error)
2695 		db_error("sysctl name");
2696 }
2697 
2698 /*
2699  * Check to see if an OID is safe to print from ddb.
2700  */
2701 static bool
2702 db_oid_safe(const struct sysctl_oid *oidp)
2703 {
2704 	for (unsigned int i = 0; i < nitems(db_safe_handlers); ++i) {
2705 		if (oidp->oid_handler == db_safe_handlers[i])
2706 			return (true);
2707 	}
2708 
2709 	return (false);
2710 }
2711 
2712 /*
2713  * Show a sysctl at a specific OID
2714  * Compare to the input handling in show_var from sbin/sysctl/sysctl.c
2715  */
2716 static int
2717 db_show_oid(struct sysctl_oid *oidp, int *oid, size_t nlen, int flags)
2718 {
2719 	int error, xflag, oflag, Nflag, nflag;
2720 	size_t len;
2721 
2722 	xflag = flags & DB_SYSCTL_HEX;
2723 	oflag = flags & DB_SYSCTL_OPAQUE;
2724 	nflag = flags & DB_SYSCTL_VALUE_ONLY;
2725 	Nflag = flags & DB_SYSCTL_NAME_ONLY;
2726 
2727 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_OPAQUE &&
2728 	    (!xflag && !oflag))
2729 		return (0);
2730 
2731 	if (Nflag) {
2732 		db_show_oid_name(oid, nlen);
2733 		error = 0;
2734 		goto out;
2735 	}
2736 
2737 	if (!nflag) {
2738 		db_show_oid_name(oid, nlen);
2739 		db_printf(": ");
2740 	}
2741 
2742 	if ((flags & DB_SYSCTL_SAFE_ONLY) && !db_oid_safe(oidp)) {
2743 		db_printf("Skipping, unsafe to print while recursing.");
2744 		error = 0;
2745 		goto out;
2746 	}
2747 
2748 	/* Try once, and ask about the size */
2749 	len = 0;
2750 	error = db_sysctl(oidp, oid, nlen,
2751 	    NULL, NULL, &len, flags);
2752 	if (error)
2753 		goto out;
2754 
2755 	if (!g_ddb_sysctl_printed)
2756 		/* Lie about the size */
2757 		error = db_sysctl(oidp, oid, nlen,
2758 		    (void *) 1, &len, NULL, flags);
2759 
2760 out:
2761 	db_printf("\n");
2762 	return (error);
2763 }
2764 
2765 /*
2766  * Show all sysctls under a specific OID
2767  * Compare to sysctl_all from sbin/sysctl/sysctl.c
2768  */
2769 static int
2770 db_show_sysctl_all(int *oid, size_t len, int flags)
2771 {
2772 	struct sysctl_oid *oidp;
2773 	int name1[CTL_MAXNAME + 2], name2[CTL_MAXNAME + 2];
2774 	size_t l1, l2;
2775 
2776 	name1[0] = CTL_SYSCTL;
2777 	name1[1] = CTL_SYSCTL_NEXT;
2778 	l1 = 2;
2779 	if (len) {
2780 		memcpy(name1 + 2, oid, len * sizeof(int));
2781 		l1 += len;
2782 	} else {
2783 		name1[2] = CTL_KERN;
2784 		l1++;
2785 	}
2786 	for (;;) {
2787 		int i, error;
2788 
2789 		l2 = sizeof(name2);
2790 		error = kernel_sysctl(kdb_thread, name1, l1,
2791 		    name2, &l2, NULL, 0, &l2, 0);
2792 		if (error != 0) {
2793 			if (error == ENOENT)
2794 				return (0);
2795 			else
2796 				db_error("sysctl(next)");
2797 		}
2798 
2799 		l2 /= sizeof(int);
2800 
2801 		if (l2 < (unsigned int)len)
2802 			return (0);
2803 
2804 		for (i = 0; i < len; i++)
2805 			if (name2[i] != oid[i])
2806 				return (0);
2807 
2808 		/* Find the OID in question */
2809 		error = sysctl_find_oid(name2, l2, &oidp, NULL, NULL);
2810 		if (error)
2811 			return (error);
2812 
2813 		i = db_show_oid(oidp, name2, l2, flags | DB_SYSCTL_SAFE_ONLY);
2814 
2815 		if (db_pager_quit)
2816 			return (0);
2817 
2818 		memcpy(name1+2, name2, l2 * sizeof(int));
2819 		l1 = 2 + l2;
2820 	}
2821 }
2822 
2823 /*
2824  * Show a sysctl by its user facing string
2825  */
2826 static int
2827 db_sysctlbyname(char *name, int flags)
2828 {
2829 	struct sysctl_oid *oidp;
2830 	int oid[CTL_MAXNAME];
2831 	int error, nlen;
2832 
2833 	error = name2oid(name, oid, &nlen, &oidp);
2834 	if (error) {
2835 		return (error);
2836 	}
2837 
2838 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
2839 		db_show_sysctl_all(oid, nlen, flags);
2840 	} else {
2841 		error = db_show_oid(oidp, oid, nlen, flags);
2842 	}
2843 
2844 	return (error);
2845 }
2846 
2847 static void
2848 db_sysctl_cmd_usage(void)
2849 {
2850 	db_printf(
2851 	    " sysctl [/Nnox] <sysctl>					    \n"
2852 	    "								    \n"
2853 	    " <sysctl> The name of the sysctl to show.			    \n"
2854 	    "								    \n"
2855 	    " Show a sysctl by hooking into SYSCTL_IN and SYSCTL_OUT.	    \n"
2856 	    " This will work for most sysctls, but should not be used	    \n"
2857 	    " with sysctls that are known to malloc.			    \n"
2858 	    "								    \n"
2859 	    " While recursing any \"unsafe\" sysctls will be skipped.	    \n"
2860 	    " Call sysctl directly on the sysctl to try printing the	    \n"
2861 	    " skipped sysctl. This is unsafe and may make the ddb	    \n"
2862 	    " session unusable.						    \n"
2863 	    "								    \n"
2864 	    " Arguments:						    \n"
2865 	    "	/N	Display only the name of the sysctl.		    \n"
2866 	    "	/n	Display only the value of the sysctl.		    \n"
2867 	    "	/o	Display opaque values.				    \n"
2868 	    "	/x	Display the sysctl in hex.			    \n"
2869 	    "								    \n"
2870 	    "For example:						    \n"
2871 	    "sysctl vm.v_free_min					    \n"
2872 	    "vn.v_free_min: 12669					    \n"
2873 	    );
2874 }
2875 
2876 /*
2877  * Show a specific sysctl similar to sysctl (8).
2878  */
2879 DB_FUNC(sysctl, db_sysctl_cmd, db_cmd_table, CS_OWN, NULL)
2880 {
2881 	char name[TOK_STRING_SIZE];
2882 	int error, i, t, flags;
2883 
2884 	/* Parse the modifiers */
2885 	t = db_read_token();
2886 	if (t == tSLASH || t == tMINUS) {
2887 		t = db_read_token();
2888 		if (t != tIDENT) {
2889 			db_printf("Bad modifier\n");
2890 			error = EINVAL;
2891 			goto out;
2892 		}
2893 		db_strcpy(modif, db_tok_string);
2894 	}
2895 	else {
2896 		db_unread_token(t);
2897 		modif[0] = '\0';
2898 	}
2899 
2900 	flags = 0;
2901 	for (i = 0; i < nitems(db_sysctl_modifs); i++) {
2902 		if (strchr(modif, db_sysctl_modifs[i])) {
2903 			flags |= db_sysctl_modif_values[i];
2904 		}
2905 	}
2906 
2907 	/* Parse the sysctl names */
2908 	t = db_read_token();
2909 	if (t != tIDENT) {
2910 		db_printf("Need sysctl name\n");
2911 		error = EINVAL;
2912 		goto out;
2913 	}
2914 
2915 	/* Copy the name into a temporary buffer */
2916 	db_strcpy(name, db_tok_string);
2917 
2918 	/* Ensure there is no trailing cruft */
2919 	t = db_read_token();
2920 	if (t != tEOL) {
2921 		db_printf("Unexpected sysctl argument\n");
2922 		error = EINVAL;
2923 		goto out;
2924 	}
2925 
2926 	error = db_sysctlbyname(name, flags);
2927 	if (error == ENOENT) {
2928 		db_printf("unknown oid: '%s'\n", db_tok_string);
2929 		goto out;
2930 	} else if (error) {
2931 		db_printf("%s: error: %d\n", db_tok_string, error);
2932 		goto out;
2933 	}
2934 
2935 out:
2936 	/* Ensure we eat all of our text */
2937 	db_flush_lex();
2938 
2939 	if (error == EINVAL) {
2940 		db_sysctl_cmd_usage();
2941 	}
2942 }
2943 
2944 #endif /* DDB */
2945