xref: /freebsd/sys/kern/kern_sysctl.c (revision 22cf89c938886d14f5796fc49f9f020c23ea8eaf)
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 #include "opt_capsicum.h"
42 #include "opt_ddb.h"
43 #include "opt_ktrace.h"
44 #include "opt_sysctl.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/limits.h>
52 #include <sys/sysctl.h>
53 #include <sys/malloc.h>
54 #include <sys/priv.h>
55 #include <sys/proc.h>
56 #include <sys/jail.h>
57 #include <sys/kdb.h>
58 #include <sys/lock.h>
59 #include <sys/mutex.h>
60 #include <sys/rmlock.h>
61 #include <sys/sbuf.h>
62 #include <sys/sx.h>
63 #include <sys/sysproto.h>
64 #include <sys/uio.h>
65 #ifdef KTRACE
66 #include <sys/ktrace.h>
67 #endif
68 
69 #ifdef DDB
70 #include <ddb/ddb.h>
71 #include <ddb/db_lex.h>
72 #endif
73 
74 #include <net/vnet.h>
75 
76 #include <security/mac/mac_framework.h>
77 
78 #include <vm/vm.h>
79 #include <vm/vm_extern.h>
80 
81 static MALLOC_DEFINE(M_SYSCTL, "sysctl", "sysctl internal magic");
82 static MALLOC_DEFINE(M_SYSCTLOID, "sysctloid", "sysctl dynamic oids");
83 static MALLOC_DEFINE(M_SYSCTLTMP, "sysctltmp", "sysctl temp output buffer");
84 
85 RB_GENERATE(sysctl_oid_list, sysctl_oid, oid_link, cmp_sysctl_oid);
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 = RB_INITIALIZER(&sysctl__children);
124 
125 static char*	sysctl_escape_name(const char*);
126 static int	sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del,
127 		    int recurse);
128 static int	sysctl_old_kernel(struct sysctl_req *, const void *, size_t);
129 static int	sysctl_new_kernel(struct sysctl_req *, void *, size_t);
130 
131 static struct sysctl_oid *
132 sysctl_find_oidname(const char *name, struct sysctl_oid_list *list)
133 {
134 	struct sysctl_oid *oidp;
135 
136 	SYSCTL_ASSERT_LOCKED();
137 	SYSCTL_FOREACH(oidp, list) {
138 		if (strcmp(oidp->oid_name, name) == 0) {
139 			return (oidp);
140 		}
141 	}
142 	return (NULL);
143 }
144 
145 /*
146  * Initialization of the MIB tree.
147  *
148  * Order by number in each list.
149  */
150 void
151 sysctl_wlock(void)
152 {
153 
154 	SYSCTL_WLOCK();
155 }
156 
157 void
158 sysctl_wunlock(void)
159 {
160 
161 	SYSCTL_WUNLOCK();
162 }
163 
164 static int
165 sysctl_root_handler_locked(struct sysctl_oid *oid, void *arg1, intmax_t arg2,
166     struct sysctl_req *req, struct rm_priotracker *tracker)
167 {
168 	int error;
169 
170 	if (oid->oid_kind & CTLFLAG_DYN)
171 		atomic_add_int(&oid->oid_running, 1);
172 
173 	if (tracker != NULL)
174 		SYSCTL_RUNLOCK(tracker);
175 	else
176 		SYSCTL_WUNLOCK();
177 
178 	/*
179 	 * Treat set CTLFLAG_NEEDGIANT and unset CTLFLAG_MPSAFE flags the same,
180 	 * untill we're ready to remove all traces of Giant from sysctl(9).
181 	 */
182 	if ((oid->oid_kind & CTLFLAG_NEEDGIANT) ||
183 	    (!(oid->oid_kind & CTLFLAG_MPSAFE)))
184 		mtx_lock(&Giant);
185 	error = oid->oid_handler(oid, arg1, arg2, req);
186 	if ((oid->oid_kind & CTLFLAG_NEEDGIANT) ||
187 	    (!(oid->oid_kind & CTLFLAG_MPSAFE)))
188 		mtx_unlock(&Giant);
189 
190 	KFAIL_POINT_ERROR(_debug_fail_point, sysctl_running, error);
191 
192 	if (tracker != NULL)
193 		SYSCTL_RLOCK(tracker);
194 	else
195 		SYSCTL_WLOCK();
196 
197 	if (oid->oid_kind & CTLFLAG_DYN) {
198 		if (atomic_fetchadd_int(&oid->oid_running, -1) == 1 &&
199 		    (oid->oid_kind & CTLFLAG_DYING) != 0)
200 			wakeup(&oid->oid_running);
201 	}
202 
203 	return (error);
204 }
205 
206 static void
207 sysctl_load_tunable_by_oid_locked(struct sysctl_oid *oidp)
208 {
209 	struct sysctl_req req;
210 	struct sysctl_oid *curr;
211 	char *penv = NULL;
212 	char path[96];
213 	ssize_t rem = sizeof(path);
214 	ssize_t len;
215 	uint8_t data[512] __aligned(sizeof(uint64_t));
216 	int size;
217 	int error;
218 
219 	path[--rem] = 0;
220 
221 	for (curr = oidp; curr != NULL; curr = SYSCTL_PARENT(curr)) {
222 		len = strlen(curr->oid_name);
223 		rem -= len;
224 		if (curr != oidp)
225 			rem -= 1;
226 		if (rem < 0) {
227 			printf("OID path exceeds %d bytes\n", (int)sizeof(path));
228 			return;
229 		}
230 		memcpy(path + rem, curr->oid_name, len);
231 		if (curr != oidp)
232 			path[rem + len] = '.';
233 	}
234 
235 	memset(&req, 0, sizeof(req));
236 
237 	req.td = curthread;
238 	req.oldfunc = sysctl_old_kernel;
239 	req.newfunc = sysctl_new_kernel;
240 	req.lock = REQ_UNWIRED;
241 
242 	switch (oidp->oid_kind & CTLTYPE) {
243 	case CTLTYPE_INT:
244 		if (getenv_array(path + rem, data, sizeof(data), &size,
245 		    sizeof(int), GETENV_SIGNED) == 0)
246 			return;
247 		req.newlen = size;
248 		req.newptr = data;
249 		break;
250 	case CTLTYPE_UINT:
251 		if (getenv_array(path + rem, data, sizeof(data), &size,
252 		    sizeof(int), GETENV_UNSIGNED) == 0)
253 			return;
254 		req.newlen = size;
255 		req.newptr = data;
256 		break;
257 	case CTLTYPE_LONG:
258 		if (getenv_array(path + rem, data, sizeof(data), &size,
259 		    sizeof(long), GETENV_SIGNED) == 0)
260 			return;
261 		req.newlen = size;
262 		req.newptr = data;
263 		break;
264 	case CTLTYPE_ULONG:
265 		if (getenv_array(path + rem, data, sizeof(data), &size,
266 		    sizeof(long), GETENV_UNSIGNED) == 0)
267 			return;
268 		req.newlen = size;
269 		req.newptr = data;
270 		break;
271 	case CTLTYPE_S8:
272 		if (getenv_array(path + rem, data, sizeof(data), &size,
273 		    sizeof(int8_t), GETENV_SIGNED) == 0)
274 			return;
275 		req.newlen = size;
276 		req.newptr = data;
277 		break;
278 	case CTLTYPE_S16:
279 		if (getenv_array(path + rem, data, sizeof(data), &size,
280 		    sizeof(int16_t), GETENV_SIGNED) == 0)
281 			return;
282 		req.newlen = size;
283 		req.newptr = data;
284 		break;
285 	case CTLTYPE_S32:
286 		if (getenv_array(path + rem, data, sizeof(data), &size,
287 		    sizeof(int32_t), GETENV_SIGNED) == 0)
288 			return;
289 		req.newlen = size;
290 		req.newptr = data;
291 		break;
292 	case CTLTYPE_S64:
293 		if (getenv_array(path + rem, data, sizeof(data), &size,
294 		    sizeof(int64_t), GETENV_SIGNED) == 0)
295 			return;
296 		req.newlen = size;
297 		req.newptr = data;
298 		break;
299 	case CTLTYPE_U8:
300 		if (getenv_array(path + rem, data, sizeof(data), &size,
301 		    sizeof(uint8_t), GETENV_UNSIGNED) == 0)
302 			return;
303 		req.newlen = size;
304 		req.newptr = data;
305 		break;
306 	case CTLTYPE_U16:
307 		if (getenv_array(path + rem, data, sizeof(data), &size,
308 		    sizeof(uint16_t), GETENV_UNSIGNED) == 0)
309 			return;
310 		req.newlen = size;
311 		req.newptr = data;
312 		break;
313 	case CTLTYPE_U32:
314 		if (getenv_array(path + rem, data, sizeof(data), &size,
315 		    sizeof(uint32_t), GETENV_UNSIGNED) == 0)
316 			return;
317 		req.newlen = size;
318 		req.newptr = data;
319 		break;
320 	case CTLTYPE_U64:
321 		if (getenv_array(path + rem, data, sizeof(data), &size,
322 		    sizeof(uint64_t), GETENV_UNSIGNED) == 0)
323 			return;
324 		req.newlen = size;
325 		req.newptr = data;
326 		break;
327 	case CTLTYPE_STRING:
328 		penv = kern_getenv(path + rem);
329 		if (penv == NULL)
330 			return;
331 		req.newlen = strlen(penv);
332 		req.newptr = penv;
333 		break;
334 	default:
335 		return;
336 	}
337 	error = sysctl_root_handler_locked(oidp, oidp->oid_arg1,
338 	    oidp->oid_arg2, &req, NULL);
339 	if (error != 0)
340 		printf("Setting sysctl %s failed: %d\n", path + rem, error);
341 	if (penv != NULL)
342 		freeenv(penv);
343 }
344 
345 /*
346  * Locate the path to a given oid.  Returns the length of the resulting path,
347  * or -1 if the oid was not found.  nodes must have room for CTL_MAXNAME
348  * elements.
349  */
350 static int
351 sysctl_search_oid(struct sysctl_oid **nodes, struct sysctl_oid *needle)
352 {
353 	int indx;
354 
355 	SYSCTL_ASSERT_LOCKED();
356 	indx = 0;
357 	/*
358 	 * Do a depth-first search of the oid tree, looking for 'needle'. Start
359 	 * with the first child of the root.
360 	 */
361 	nodes[indx] = RB_MIN(sysctl_oid_list, &sysctl__children);
362 	for (;;) {
363 		if (nodes[indx] == needle)
364 			return (indx + 1);
365 
366 		if (nodes[indx] == NULL) {
367 			/* Node has no more siblings, so back up to parent. */
368 			if (indx-- == 0) {
369 				/* Retreat to root, so give up. */
370 				break;
371 			}
372 		} else if ((nodes[indx]->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
373 			/* Node has children. */
374 			if (++indx == CTL_MAXNAME) {
375 				/* Max search depth reached, so give up. */
376 				break;
377 			}
378 			/* Start with the first child. */
379 			nodes[indx] = RB_MIN(sysctl_oid_list,
380 			    &nodes[indx - 1]->oid_children);
381 			continue;
382 		}
383 		/* Consider next sibling. */
384 		nodes[indx] = RB_NEXT(sysctl_oid_list, NULL, nodes[indx]);
385 	}
386 	return (-1);
387 }
388 
389 static void
390 sysctl_warn_reuse(const char *func, struct sysctl_oid *leaf)
391 {
392 	struct sysctl_oid *nodes[CTL_MAXNAME];
393 	char buf[128];
394 	struct sbuf sb;
395 	int rc, i;
396 
397 	(void)sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN | SBUF_INCLUDENUL);
398 	sbuf_set_drain(&sb, sbuf_printf_drain, NULL);
399 
400 	sbuf_printf(&sb, "%s: can't re-use a leaf (", __func__);
401 
402 	rc = sysctl_search_oid(nodes, leaf);
403 	if (rc > 0) {
404 		for (i = 0; i < rc; i++)
405 			sbuf_printf(&sb, "%s%.*s", nodes[i]->oid_name,
406 			    i != (rc - 1), ".");
407 	} else {
408 		sbuf_printf(&sb, "%s", leaf->oid_name);
409 	}
410 	sbuf_printf(&sb, ")!\n");
411 
412 	(void)sbuf_finish(&sb);
413 }
414 
415 #ifdef SYSCTL_DEBUG
416 static int
417 sysctl_reuse_test(SYSCTL_HANDLER_ARGS)
418 {
419 	struct rm_priotracker tracker;
420 
421 	SYSCTL_RLOCK(&tracker);
422 	sysctl_warn_reuse(__func__, oidp);
423 	SYSCTL_RUNLOCK(&tracker);
424 	return (0);
425 }
426 SYSCTL_PROC(_sysctl, OID_AUTO, reuse_test,
427     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 0, sysctl_reuse_test, "-",
428     "");
429 #endif
430 
431 void
432 sysctl_register_oid(struct sysctl_oid *oidp)
433 {
434 	struct sysctl_oid_list *parent = oidp->oid_parent;
435 	struct sysctl_oid *p, key;
436 	int oid_number;
437 	int timeout = 2;
438 
439 	/*
440 	 * First check if another oid with the same name already
441 	 * exists in the parent's list.
442 	 */
443 	SYSCTL_ASSERT_WLOCKED();
444 	p = sysctl_find_oidname(oidp->oid_name, parent);
445 	if (p != NULL) {
446 		if ((p->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
447 			p->oid_refcnt++;
448 			return;
449 		} else {
450 			sysctl_warn_reuse(__func__, p);
451 			return;
452 		}
453 	}
454 	/* get current OID number */
455 	oid_number = oidp->oid_number;
456 
457 #if (OID_AUTO >= 0)
458 #error "OID_AUTO is expected to be a negative value"
459 #endif
460 	/*
461 	 * Any negative OID number qualifies as OID_AUTO. Valid OID
462 	 * numbers should always be positive.
463 	 *
464 	 * NOTE: DO NOT change the starting value here, change it in
465 	 * <sys/sysctl.h>, and make sure it is at least 256 to
466 	 * accommodate e.g. net.inet.raw as a static sysctl node.
467 	 */
468 	if (oid_number < 0) {
469 		static int newoid;
470 
471 		/*
472 		 * By decrementing the next OID number we spend less
473 		 * time inserting the OIDs into a sorted list.
474 		 */
475 		if (--newoid < CTL_AUTO_START)
476 			newoid = 0x7fffffff;
477 
478 		oid_number = newoid;
479 	}
480 
481 	/*
482 	 * Insert the OID into the parent's list sorted by OID number.
483 	 */
484 	key.oid_number = oid_number;
485 	p = RB_NFIND(sysctl_oid_list, parent, &key);
486 	while (p != NULL && oid_number == p->oid_number) {
487 		/* get the next valid OID number */
488 		if (oid_number < CTL_AUTO_START ||
489 		    oid_number == 0x7fffffff) {
490 			/* wraparound - restart */
491 			oid_number = CTL_AUTO_START;
492 			/* don't loop forever */
493 			if (!timeout--)
494 				panic("sysctl: Out of OID numbers\n");
495 			key.oid_number = oid_number;
496 			p = RB_NFIND(sysctl_oid_list, parent, &key);
497 			continue;
498 		}
499 		p = RB_NEXT(sysctl_oid_list, NULL, p);
500 		oid_number++;
501 	}
502 	/* check for non-auto OID number collision */
503 	if (oidp->oid_number >= 0 && oidp->oid_number < CTL_AUTO_START &&
504 	    oid_number >= CTL_AUTO_START) {
505 		printf("sysctl: OID number(%d) is already in use for '%s'\n",
506 		    oidp->oid_number, oidp->oid_name);
507 	}
508 	/* update the OID number, if any */
509 	oidp->oid_number = oid_number;
510 	RB_INSERT(sysctl_oid_list, parent, oidp);
511 
512 	if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE &&
513 #ifdef VIMAGE
514 	    (oidp->oid_kind & CTLFLAG_VNET) == 0 &&
515 #endif
516 	    (oidp->oid_kind & CTLFLAG_TUN) != 0 &&
517 	    (oidp->oid_kind & CTLFLAG_NOFETCH) == 0) {
518 		/* only fetch value once */
519 		oidp->oid_kind |= CTLFLAG_NOFETCH;
520 		/* try to fetch value from kernel environment */
521 		sysctl_load_tunable_by_oid_locked(oidp);
522 	}
523 }
524 
525 void
526 sysctl_register_disabled_oid(struct sysctl_oid *oidp)
527 {
528 
529 	/*
530 	 * Mark the leaf as dormant if it's not to be immediately enabled.
531 	 * We do not disable nodes as they can be shared between modules
532 	 * and it is always safe to access a node.
533 	 */
534 	KASSERT((oidp->oid_kind & CTLFLAG_DORMANT) == 0,
535 	    ("internal flag is set in oid_kind"));
536 	if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
537 		oidp->oid_kind |= CTLFLAG_DORMANT;
538 	sysctl_register_oid(oidp);
539 }
540 
541 void
542 sysctl_enable_oid(struct sysctl_oid *oidp)
543 {
544 
545 	SYSCTL_ASSERT_WLOCKED();
546 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
547 		KASSERT((oidp->oid_kind & CTLFLAG_DORMANT) == 0,
548 		    ("sysctl node is marked as dormant"));
549 		return;
550 	}
551 	KASSERT((oidp->oid_kind & CTLFLAG_DORMANT) != 0,
552 	    ("enabling already enabled sysctl oid"));
553 	oidp->oid_kind &= ~CTLFLAG_DORMANT;
554 }
555 
556 void
557 sysctl_unregister_oid(struct sysctl_oid *oidp)
558 {
559 	int error;
560 
561 	SYSCTL_ASSERT_WLOCKED();
562 	if (oidp->oid_number == OID_AUTO) {
563 		error = EINVAL;
564 	} else {
565 		error = ENOENT;
566 		if (RB_REMOVE(sysctl_oid_list, oidp->oid_parent, oidp))
567 			error = 0;
568 	}
569 
570 	/*
571 	 * This can happen when a module fails to register and is
572 	 * being unloaded afterwards.  It should not be a panic()
573 	 * for normal use.
574 	 */
575 	if (error) {
576 		printf("%s: failed(%d) to unregister sysctl(%s)\n",
577 		    __func__, error, oidp->oid_name);
578 	}
579 }
580 
581 /* Initialize a new context to keep track of dynamically added sysctls. */
582 int
583 sysctl_ctx_init(struct sysctl_ctx_list *c)
584 {
585 
586 	if (c == NULL) {
587 		return (EINVAL);
588 	}
589 
590 	/*
591 	 * No locking here, the caller is responsible for not adding
592 	 * new nodes to a context until after this function has
593 	 * returned.
594 	 */
595 	TAILQ_INIT(c);
596 	return (0);
597 }
598 
599 /* Free the context, and destroy all dynamic oids registered in this context */
600 int
601 sysctl_ctx_free(struct sysctl_ctx_list *clist)
602 {
603 	struct sysctl_ctx_entry *e, *e1;
604 	int error;
605 
606 	error = 0;
607 	/*
608 	 * First perform a "dry run" to check if it's ok to remove oids.
609 	 * XXX FIXME
610 	 * XXX This algorithm is a hack. But I don't know any
611 	 * XXX better solution for now...
612 	 */
613 	SYSCTL_WLOCK();
614 	TAILQ_FOREACH(e, clist, link) {
615 		error = sysctl_remove_oid_locked(e->entry, 0, 0);
616 		if (error)
617 			break;
618 	}
619 	/*
620 	 * Restore deregistered entries, either from the end,
621 	 * or from the place where error occurred.
622 	 * e contains the entry that was not unregistered
623 	 */
624 	if (error)
625 		e1 = TAILQ_PREV(e, sysctl_ctx_list, link);
626 	else
627 		e1 = TAILQ_LAST(clist, sysctl_ctx_list);
628 	while (e1 != NULL) {
629 		sysctl_register_oid(e1->entry);
630 		e1 = TAILQ_PREV(e1, sysctl_ctx_list, link);
631 	}
632 	if (error) {
633 		SYSCTL_WUNLOCK();
634 		return(EBUSY);
635 	}
636 	/* Now really delete the entries */
637 	e = TAILQ_FIRST(clist);
638 	while (e != NULL) {
639 		e1 = TAILQ_NEXT(e, link);
640 		error = sysctl_remove_oid_locked(e->entry, 1, 0);
641 		if (error)
642 			panic("sysctl_remove_oid: corrupt tree, entry: %s",
643 			    e->entry->oid_name);
644 		free(e, M_SYSCTLOID);
645 		e = e1;
646 	}
647 	SYSCTL_WUNLOCK();
648 	return (error);
649 }
650 
651 /* Add an entry to the context */
652 struct sysctl_ctx_entry *
653 sysctl_ctx_entry_add(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
654 {
655 	struct sysctl_ctx_entry *e;
656 
657 	SYSCTL_ASSERT_WLOCKED();
658 	if (clist == NULL || oidp == NULL)
659 		return(NULL);
660 	e = malloc(sizeof(struct sysctl_ctx_entry), M_SYSCTLOID, M_WAITOK);
661 	e->entry = oidp;
662 	TAILQ_INSERT_HEAD(clist, e, link);
663 	return (e);
664 }
665 
666 /* Find an entry in the context */
667 struct sysctl_ctx_entry *
668 sysctl_ctx_entry_find(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
669 {
670 	struct sysctl_ctx_entry *e;
671 
672 	SYSCTL_ASSERT_WLOCKED();
673 	if (clist == NULL || oidp == NULL)
674 		return(NULL);
675 	TAILQ_FOREACH(e, clist, link) {
676 		if (e->entry == oidp)
677 			return(e);
678 	}
679 	return (e);
680 }
681 
682 /*
683  * Delete an entry from the context.
684  * NOTE: this function doesn't free oidp! You have to remove it
685  * with sysctl_remove_oid().
686  */
687 int
688 sysctl_ctx_entry_del(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
689 {
690 	struct sysctl_ctx_entry *e;
691 
692 	if (clist == NULL || oidp == NULL)
693 		return (EINVAL);
694 	SYSCTL_WLOCK();
695 	e = sysctl_ctx_entry_find(clist, oidp);
696 	if (e != NULL) {
697 		TAILQ_REMOVE(clist, e, link);
698 		SYSCTL_WUNLOCK();
699 		free(e, M_SYSCTLOID);
700 		return (0);
701 	} else {
702 		SYSCTL_WUNLOCK();
703 		return (ENOENT);
704 	}
705 }
706 
707 /*
708  * Remove dynamically created sysctl trees.
709  * oidp - top of the tree to be removed
710  * del - if 0 - just deregister, otherwise free up entries as well
711  * recurse - if != 0 traverse the subtree to be deleted
712  */
713 int
714 sysctl_remove_oid(struct sysctl_oid *oidp, int del, int recurse)
715 {
716 	int error;
717 
718 	SYSCTL_WLOCK();
719 	error = sysctl_remove_oid_locked(oidp, del, recurse);
720 	SYSCTL_WUNLOCK();
721 	return (error);
722 }
723 
724 int
725 sysctl_remove_name(struct sysctl_oid *parent, const char *name,
726     int del, int recurse)
727 {
728 	struct sysctl_oid *p;
729 	int error;
730 
731 	error = ENOENT;
732 	SYSCTL_WLOCK();
733 	p = sysctl_find_oidname(name, &parent->oid_children);
734 	if (p)
735 		error = sysctl_remove_oid_locked(p, del, recurse);
736 	SYSCTL_WUNLOCK();
737 
738 	return (error);
739 }
740 
741 /*
742  * Duplicate the provided string, escaping any illegal characters.  The result
743  * must be freed when no longer in use.
744  *
745  * The list of illegal characters is ".".
746  */
747 static char*
748 sysctl_escape_name(const char* orig)
749 {
750 	int i, s = 0, d = 0, nillegals = 0;
751 	char *new;
752 
753 	/* First count the number of illegal characters */
754 	for (i = 0; orig[i] != '\0'; i++) {
755 		if (orig[i] == '.')
756 			nillegals++;
757 	}
758 
759 	/* Allocate storage for new string */
760 	new = malloc(i + 2 * nillegals + 1, M_SYSCTLOID, M_WAITOK);
761 
762 	/* Copy the name, escaping characters as we go */
763 	while (orig[s] != '\0') {
764 		if (orig[s] == '.') {
765 			/* %25 is the hexadecimal representation of '.' */
766 			new[d++] = '%';
767 			new[d++] = '2';
768 			new[d++] = '5';
769 			s++;
770 		} else {
771 			new[d++] = orig[s++];
772 		}
773 	}
774 
775 	/* Finally, nul-terminate */
776 	new[d] = '\0';
777 
778 	return (new);
779 }
780 
781 static int
782 sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del, int recurse)
783 {
784 	struct sysctl_oid *p, *tmp;
785 	int error;
786 
787 	SYSCTL_ASSERT_WLOCKED();
788 	if (oidp == NULL)
789 		return(EINVAL);
790 	if ((oidp->oid_kind & CTLFLAG_DYN) == 0) {
791 		printf("Warning: can't remove non-dynamic nodes (%s)!\n",
792 		    oidp->oid_name);
793 		return (EINVAL);
794 	}
795 	/*
796 	 * WARNING: normal method to do this should be through
797 	 * sysctl_ctx_free(). Use recursing as the last resort
798 	 * method to purge your sysctl tree of leftovers...
799 	 * However, if some other code still references these nodes,
800 	 * it will panic.
801 	 */
802 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
803 		if (oidp->oid_refcnt == 1) {
804 			for(p = RB_MIN(sysctl_oid_list, &oidp->oid_children);
805 			    p != NULL; p = tmp) {
806 				if (!recurse) {
807 					printf("Warning: failed attempt to "
808 					    "remove oid %s with child %s\n",
809 					    oidp->oid_name, p->oid_name);
810 					return (ENOTEMPTY);
811 				}
812 				tmp = RB_NEXT(sysctl_oid_list,
813 				    &oidp->oid_children, p);
814 				error = sysctl_remove_oid_locked(p, del,
815 				    recurse);
816 				if (error)
817 					return (error);
818 			}
819 		}
820 	}
821 	if (oidp->oid_refcnt > 1 ) {
822 		oidp->oid_refcnt--;
823 	} else {
824 		if (oidp->oid_refcnt == 0) {
825 			printf("Warning: bad oid_refcnt=%u (%s)!\n",
826 				oidp->oid_refcnt, oidp->oid_name);
827 			return (EINVAL);
828 		}
829 		sysctl_unregister_oid(oidp);
830 		if (del) {
831 			/*
832 			 * Wait for all threads running the handler to drain.
833 			 * This preserves the previous behavior when the
834 			 * sysctl lock was held across a handler invocation,
835 			 * and is necessary for module unload correctness.
836 			 */
837 			while (oidp->oid_running > 0) {
838 				oidp->oid_kind |= CTLFLAG_DYING;
839 				SYSCTL_SLEEP(&oidp->oid_running, "oidrm", 0);
840 			}
841 			if (oidp->oid_descr)
842 				free(__DECONST(char *, oidp->oid_descr),
843 				    M_SYSCTLOID);
844 			if (oidp->oid_label)
845 				free(__DECONST(char *, oidp->oid_label),
846 				    M_SYSCTLOID);
847 			free(__DECONST(char *, oidp->oid_name), M_SYSCTLOID);
848 			free(oidp, M_SYSCTLOID);
849 		}
850 	}
851 	return (0);
852 }
853 /*
854  * Create new sysctls at run time.
855  * clist may point to a valid context initialized with sysctl_ctx_init().
856  */
857 struct sysctl_oid *
858 sysctl_add_oid(struct sysctl_ctx_list *clist, struct sysctl_oid_list *parent,
859 	int number, const char *name, int kind, void *arg1, intmax_t arg2,
860 	int (*handler)(SYSCTL_HANDLER_ARGS), const char *fmt, const char *descr,
861 	const char *label)
862 {
863 	struct sysctl_oid *oidp;
864 	char *escaped;
865 
866 	/* You have to hook up somewhere.. */
867 	if (parent == NULL)
868 		return(NULL);
869 	escaped = sysctl_escape_name(name);
870 	/* Check if the node already exists, otherwise create it */
871 	SYSCTL_WLOCK();
872 	oidp = sysctl_find_oidname(escaped, parent);
873 	if (oidp != NULL) {
874 		free(escaped, M_SYSCTLOID);
875 		if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
876 			oidp->oid_refcnt++;
877 			/* Update the context */
878 			if (clist != NULL)
879 				sysctl_ctx_entry_add(clist, oidp);
880 			SYSCTL_WUNLOCK();
881 			return (oidp);
882 		} else {
883 			sysctl_warn_reuse(__func__, oidp);
884 			SYSCTL_WUNLOCK();
885 			return (NULL);
886 		}
887 	}
888 	oidp = malloc(sizeof(struct sysctl_oid), M_SYSCTLOID, M_WAITOK|M_ZERO);
889 	oidp->oid_parent = parent;
890 	RB_INIT(&oidp->oid_children);
891 	oidp->oid_number = number;
892 	oidp->oid_refcnt = 1;
893 	oidp->oid_name = escaped;
894 	oidp->oid_handler = handler;
895 	oidp->oid_kind = CTLFLAG_DYN | kind;
896 	oidp->oid_arg1 = arg1;
897 	oidp->oid_arg2 = arg2;
898 	oidp->oid_fmt = fmt;
899 	if (descr != NULL)
900 		oidp->oid_descr = strdup(descr, M_SYSCTLOID);
901 	if (label != NULL)
902 		oidp->oid_label = strdup(label, M_SYSCTLOID);
903 	/* Update the context, if used */
904 	if (clist != NULL)
905 		sysctl_ctx_entry_add(clist, oidp);
906 	/* Register this oid */
907 	sysctl_register_oid(oidp);
908 	SYSCTL_WUNLOCK();
909 	return (oidp);
910 }
911 
912 /*
913  * Rename an existing oid.
914  */
915 void
916 sysctl_rename_oid(struct sysctl_oid *oidp, const char *name)
917 {
918 	char *newname;
919 	char *oldname;
920 
921 	newname = strdup(name, M_SYSCTLOID);
922 	SYSCTL_WLOCK();
923 	oldname = __DECONST(char *, oidp->oid_name);
924 	oidp->oid_name = newname;
925 	SYSCTL_WUNLOCK();
926 	free(oldname, M_SYSCTLOID);
927 }
928 
929 /*
930  * Reparent an existing oid.
931  */
932 int
933 sysctl_move_oid(struct sysctl_oid *oid, struct sysctl_oid_list *parent)
934 {
935 	struct sysctl_oid *oidp;
936 
937 	SYSCTL_WLOCK();
938 	if (oid->oid_parent == parent) {
939 		SYSCTL_WUNLOCK();
940 		return (0);
941 	}
942 	oidp = sysctl_find_oidname(oid->oid_name, parent);
943 	if (oidp != NULL) {
944 		SYSCTL_WUNLOCK();
945 		return (EEXIST);
946 	}
947 	sysctl_unregister_oid(oid);
948 	oid->oid_parent = parent;
949 	oid->oid_number = OID_AUTO;
950 	sysctl_register_oid(oid);
951 	SYSCTL_WUNLOCK();
952 	return (0);
953 }
954 
955 /*
956  * Register the kernel's oids on startup.
957  */
958 SET_DECLARE(sysctl_set, struct sysctl_oid);
959 
960 static void
961 sysctl_register_all(void *arg)
962 {
963 	struct sysctl_oid **oidp;
964 
965 	sx_init(&sysctlmemlock, "sysctl mem");
966 	sx_init(&sysctlstringlock, "sysctl string handler");
967 	SYSCTL_INIT();
968 	SYSCTL_WLOCK();
969 	SET_FOREACH(oidp, sysctl_set)
970 		sysctl_register_oid(*oidp);
971 	SYSCTL_WUNLOCK();
972 }
973 SYSINIT(sysctl, SI_SUB_KMEM, SI_ORDER_FIRST, sysctl_register_all, NULL);
974 
975 /*
976  * "Staff-functions"
977  *
978  * These functions implement a presently undocumented interface
979  * used by the sysctl program to walk the tree, and get the type
980  * so it can print the value.
981  * This interface is under work and consideration, and should probably
982  * be killed with a big axe by the first person who can find the time.
983  * (be aware though, that the proper interface isn't as obvious as it
984  * may seem, there are various conflicting requirements.
985  *
986  * {CTL_SYSCTL, CTL_SYSCTL_DEBUG}		printf the entire MIB-tree.
987  * {CTL_SYSCTL, CTL_SYSCTL_NAME, ...}		return the name of the "..."
988  *						OID.
989  * {CTL_SYSCTL, CTL_SYSCTL_NEXT, ...}		return the next OID, honoring
990  *						CTLFLAG_SKIP.
991  * {CTL_SYSCTL, CTL_SYSCTL_NAME2OID}		return the OID of the name in
992  *						"new"
993  * {CTL_SYSCTL, CTL_SYSCTL_OIDFMT, ...}		return the kind & format info
994  *						for the "..." OID.
995  * {CTL_SYSCTL, CTL_SYSCTL_OIDDESCR, ...}	return the description of the
996  *						"..." OID.
997  * {CTL_SYSCTL, CTL_SYSCTL_OIDLABEL, ...}	return the aggregation label of
998  *						the "..." OID.
999  * {CTL_SYSCTL, CTL_SYSCTL_NEXTNOSKIP, ...}	return the next OID, ignoring
1000  *						CTLFLAG_SKIP.
1001  */
1002 
1003 #ifdef SYSCTL_DEBUG
1004 static void
1005 sysctl_sysctl_debug_dump_node(struct sysctl_oid_list *l, int i)
1006 {
1007 	int k;
1008 	struct sysctl_oid *oidp;
1009 
1010 	SYSCTL_ASSERT_LOCKED();
1011 	SYSCTL_FOREACH(oidp, l) {
1012 		for (k=0; k<i; k++)
1013 			printf(" ");
1014 
1015 		printf("%d %s ", oidp->oid_number, oidp->oid_name);
1016 
1017 		printf("%c%c",
1018 			oidp->oid_kind & CTLFLAG_RD ? 'R':' ',
1019 			oidp->oid_kind & CTLFLAG_WR ? 'W':' ');
1020 
1021 		if (oidp->oid_handler)
1022 			printf(" *Handler");
1023 
1024 		switch (oidp->oid_kind & CTLTYPE) {
1025 			case CTLTYPE_NODE:
1026 				printf(" Node\n");
1027 				if (!oidp->oid_handler) {
1028 					sysctl_sysctl_debug_dump_node(
1029 					    SYSCTL_CHILDREN(oidp), i + 2);
1030 				}
1031 				break;
1032 			case CTLTYPE_INT:    printf(" Int\n"); break;
1033 			case CTLTYPE_UINT:   printf(" u_int\n"); break;
1034 			case CTLTYPE_LONG:   printf(" Long\n"); break;
1035 			case CTLTYPE_ULONG:  printf(" u_long\n"); break;
1036 			case CTLTYPE_STRING: printf(" String\n"); break;
1037 			case CTLTYPE_S8:     printf(" int8_t\n"); break;
1038 			case CTLTYPE_S16:    printf(" int16_t\n"); break;
1039 			case CTLTYPE_S32:    printf(" int32_t\n"); break;
1040 			case CTLTYPE_S64:    printf(" int64_t\n"); break;
1041 			case CTLTYPE_U8:     printf(" uint8_t\n"); break;
1042 			case CTLTYPE_U16:    printf(" uint16_t\n"); break;
1043 			case CTLTYPE_U32:    printf(" uint32_t\n"); break;
1044 			case CTLTYPE_U64:    printf(" uint64_t\n"); break;
1045 			case CTLTYPE_OPAQUE: printf(" Opaque/struct\n"); break;
1046 			default:	     printf("\n");
1047 		}
1048 	}
1049 }
1050 
1051 static int
1052 sysctl_sysctl_debug(SYSCTL_HANDLER_ARGS)
1053 {
1054 	struct rm_priotracker tracker;
1055 	int error;
1056 
1057 	error = priv_check(req->td, PRIV_SYSCTL_DEBUG);
1058 	if (error)
1059 		return (error);
1060 	SYSCTL_RLOCK(&tracker);
1061 	sysctl_sysctl_debug_dump_node(&sysctl__children, 0);
1062 	SYSCTL_RUNLOCK(&tracker);
1063 	return (ENOENT);
1064 }
1065 
1066 SYSCTL_PROC(_sysctl, CTL_SYSCTL_DEBUG, debug, CTLTYPE_STRING | CTLFLAG_RD |
1067     CTLFLAG_MPSAFE, 0, 0, sysctl_sysctl_debug, "-", "");
1068 #endif
1069 
1070 static int
1071 sysctl_sysctl_name(SYSCTL_HANDLER_ARGS)
1072 {
1073 	int *name = (int *) arg1;
1074 	u_int namelen = arg2;
1075 	int error;
1076 	struct sysctl_oid *oid, key;
1077 	struct sysctl_oid_list *lsp = &sysctl__children, *lsp2;
1078 	struct rm_priotracker tracker;
1079 	char buf[10];
1080 
1081 	error = sysctl_wire_old_buffer(req, 0);
1082 	if (error)
1083 		return (error);
1084 
1085 	SYSCTL_RLOCK(&tracker);
1086 	while (namelen) {
1087 		if (!lsp) {
1088 			snprintf(buf,sizeof(buf),"%d",*name);
1089 			if (req->oldidx)
1090 				error = SYSCTL_OUT(req, ".", 1);
1091 			if (!error)
1092 				error = SYSCTL_OUT(req, buf, strlen(buf));
1093 			if (error)
1094 				goto out;
1095 			namelen--;
1096 			name++;
1097 			continue;
1098 		}
1099 		lsp2 = NULL;
1100 		key.oid_number = *name;
1101 		oid = RB_FIND(sysctl_oid_list, lsp, &key);
1102 		if (oid) {
1103 			if (req->oldidx)
1104 				error = SYSCTL_OUT(req, ".", 1);
1105 			if (!error)
1106 				error = SYSCTL_OUT(req, oid->oid_name,
1107 					strlen(oid->oid_name));
1108 			if (error)
1109 				goto out;
1110 
1111 			namelen--;
1112 			name++;
1113 
1114 			if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE &&
1115 				!oid->oid_handler)
1116 				lsp2 = SYSCTL_CHILDREN(oid);
1117 		}
1118 		lsp = lsp2;
1119 	}
1120 	error = SYSCTL_OUT(req, "", 1);
1121  out:
1122 	SYSCTL_RUNLOCK(&tracker);
1123 	return (error);
1124 }
1125 
1126 /*
1127  * XXXRW/JA: Shouldn't return name data for nodes that we don't permit in
1128  * capability mode.
1129  */
1130 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_NAME, name, CTLFLAG_RD |
1131     CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_name, "");
1132 
1133 enum sysctl_iter_action {
1134 	ITER_SIBLINGS,	/* Not matched, continue iterating siblings */
1135 	ITER_CHILDREN,	/* Node has children we need to iterate over them */
1136 	ITER_FOUND,	/* Matching node was found */
1137 };
1138 
1139 /*
1140  * Tries to find the next node for @name and @namelen.
1141  *
1142  * Returns next action to take.
1143  */
1144 static enum sysctl_iter_action
1145 sysctl_sysctl_next_node(struct sysctl_oid *oidp, int *name, unsigned int namelen,
1146     bool honor_skip)
1147 {
1148 
1149 	if ((oidp->oid_kind & CTLFLAG_DORMANT) != 0)
1150 		return (ITER_SIBLINGS);
1151 
1152 	if (honor_skip && (oidp->oid_kind & CTLFLAG_SKIP) != 0)
1153 		return (ITER_SIBLINGS);
1154 
1155 	if (namelen == 0) {
1156 		/*
1157 		 * We have reached a node with a full name match and are
1158 		 * looking for the next oid in its children.
1159 		 *
1160 		 * For CTL_SYSCTL_NEXTNOSKIP we are done.
1161 		 *
1162 		 * For CTL_SYSCTL_NEXT we skip CTLTYPE_NODE (unless it
1163 		 * has a handler) and move on to the children.
1164 		 */
1165 		if (!honor_skip)
1166 			return (ITER_FOUND);
1167 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1168 			return (ITER_FOUND);
1169 		/* If node does not have an iterator, treat it as leaf */
1170 		if (oidp->oid_handler)
1171 			return (ITER_FOUND);
1172 
1173 		/* Report oid as a node to iterate */
1174 		return (ITER_CHILDREN);
1175 	}
1176 
1177 	/*
1178 	 * No match yet. Continue seeking the given name.
1179 	 *
1180 	 * We are iterating in order by oid_number, so skip oids lower
1181 	 * than the one we are looking for.
1182 	 *
1183 	 * When the current oid_number is higher than the one we seek,
1184 	 * that means we have reached the next oid in the sequence and
1185 	 * should return it.
1186 	 *
1187 	 * If the oid_number matches the name at this level then we
1188 	 * have to find a node to continue searching at the next level.
1189 	 */
1190 	if (oidp->oid_number < *name)
1191 		return (ITER_SIBLINGS);
1192 	if (oidp->oid_number > *name) {
1193 		/*
1194 		 * We have reached the next oid.
1195 		 *
1196 		 * For CTL_SYSCTL_NEXTNOSKIP we are done.
1197 		 *
1198 		 * For CTL_SYSCTL_NEXT we skip CTLTYPE_NODE (unless it
1199 		 * has a handler) and move on to the children.
1200 		 */
1201 		if (!honor_skip)
1202 			return (ITER_FOUND);
1203 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1204 			return (ITER_FOUND);
1205 		/* If node does not have an iterator, treat it as leaf */
1206 		if (oidp->oid_handler)
1207 			return (ITER_FOUND);
1208 		return (ITER_CHILDREN);
1209 	}
1210 
1211 	/* match at a current level */
1212 	if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1213 		return (ITER_SIBLINGS);
1214 	if (oidp->oid_handler)
1215 		return (ITER_SIBLINGS);
1216 
1217 	return (ITER_CHILDREN);
1218 }
1219 
1220 /*
1221  * Recursively walk the sysctl subtree at lsp until we find the given name.
1222  * Returns true and fills in next oid data in @next and @len if oid is found.
1223  */
1224 static bool
1225 sysctl_sysctl_next_action(struct sysctl_oid_list *lsp, int *name, u_int namelen,
1226     int *next, int *len, int level, bool honor_skip)
1227 {
1228 	struct sysctl_oid_list *next_lsp;
1229 	struct sysctl_oid *oidp = NULL, key;
1230 	bool success = false;
1231 	enum sysctl_iter_action action;
1232 
1233 	SYSCTL_ASSERT_LOCKED();
1234 	/*
1235 	 * Start the search at the requested oid.  But if not found, then scan
1236 	 * through all children.
1237 	 */
1238 	if (namelen > 0) {
1239 		key.oid_number = *name;
1240 		oidp = RB_FIND(sysctl_oid_list, lsp, &key);
1241 	}
1242 	if (!oidp)
1243 		oidp = RB_MIN(sysctl_oid_list, lsp);
1244 	for(; oidp != NULL; oidp = RB_NEXT(sysctl_oid_list, lsp, oidp)) {
1245 		action = sysctl_sysctl_next_node(oidp, name, namelen,
1246 		    honor_skip);
1247 		if (action == ITER_SIBLINGS)
1248 			continue;
1249 		if (action == ITER_FOUND) {
1250 			success = true;
1251 			break;
1252 		}
1253 		KASSERT((action== ITER_CHILDREN), ("ret(%d)!=ITER_CHILDREN", action));
1254 
1255 		next_lsp = SYSCTL_CHILDREN(oidp);
1256 		if (namelen == 0) {
1257 			success = sysctl_sysctl_next_action(next_lsp, NULL, 0,
1258 			    next + 1, len, level + 1, honor_skip);
1259 		} else {
1260 			success = sysctl_sysctl_next_action(next_lsp, name + 1,
1261 			    namelen - 1, next + 1, len, level + 1, honor_skip);
1262 			if (!success) {
1263 
1264 				/*
1265 				 * We maintain the invariant that current node oid
1266 				 * is >= the oid provided in @name.
1267 				 * As there are no usable children at this node,
1268 				 *  current node oid is strictly > than the requested
1269 				 *  oid.
1270 				 * Hence, reduce namelen to 0 to allow for picking first
1271 				 *  nodes/leafs in the next node in list.
1272 				 */
1273 				namelen = 0;
1274 			}
1275 		}
1276 		if (success)
1277 			break;
1278 	}
1279 
1280 	if (success) {
1281 		*next = oidp->oid_number;
1282 		if (level > *len)
1283 			*len = level;
1284 	}
1285 
1286 	return (success);
1287 }
1288 
1289 static int
1290 sysctl_sysctl_next(SYSCTL_HANDLER_ARGS)
1291 {
1292 	int *name = (int *) arg1;
1293 	u_int namelen = arg2;
1294 	int len, error;
1295 	bool success;
1296 	struct sysctl_oid_list *lsp = &sysctl__children;
1297 	struct rm_priotracker tracker;
1298 	int next[CTL_MAXNAME];
1299 
1300 	len = 0;
1301 	SYSCTL_RLOCK(&tracker);
1302 	success = sysctl_sysctl_next_action(lsp, name, namelen, next, &len, 1,
1303 	    oidp->oid_number == CTL_SYSCTL_NEXT);
1304 	SYSCTL_RUNLOCK(&tracker);
1305 	if (!success)
1306 		return (ENOENT);
1307 	error = SYSCTL_OUT(req, next, len * sizeof (int));
1308 	return (error);
1309 }
1310 
1311 /*
1312  * XXXRW/JA: Shouldn't return next data for nodes that we don't permit in
1313  * capability mode.
1314  */
1315 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_NEXT, next, CTLFLAG_RD |
1316     CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_next, "");
1317 
1318 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_NEXTNOSKIP, nextnoskip, CTLFLAG_RD |
1319     CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_next, "");
1320 
1321 static int
1322 name2oid(char *name, int *oid, int *len, struct sysctl_oid **oidpp)
1323 {
1324 	struct sysctl_oid *oidp;
1325 	struct sysctl_oid_list *lsp = &sysctl__children;
1326 
1327 	SYSCTL_ASSERT_LOCKED();
1328 
1329 	for (*len = 0; *len < CTL_MAXNAME;) {
1330 		oidp = sysctl_find_oidname(strsep(&name, "."), lsp);
1331 		if (oidp == NULL)
1332 			return (ENOENT);
1333 		*oid++ = oidp->oid_number;
1334 		(*len)++;
1335 
1336 		if (name == NULL || *name == '\0') {
1337 			if (oidpp)
1338 				*oidpp = oidp;
1339 			return (0);
1340 		}
1341 
1342 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1343 			break;
1344 
1345 		if (oidp->oid_handler)
1346 			break;
1347 
1348 		lsp = SYSCTL_CHILDREN(oidp);
1349 	}
1350 	return (ENOENT);
1351 }
1352 
1353 static int
1354 sysctl_sysctl_name2oid(SYSCTL_HANDLER_ARGS)
1355 {
1356 	char *p;
1357 	int error, oid[CTL_MAXNAME], len = 0;
1358 	struct sysctl_oid *op = NULL;
1359 	struct rm_priotracker tracker;
1360 	char buf[32];
1361 
1362 	if (!req->newlen)
1363 		return (ENOENT);
1364 	if (req->newlen >= MAXPATHLEN)	/* XXX arbitrary, undocumented */
1365 		return (ENAMETOOLONG);
1366 
1367 	p = buf;
1368 	if (req->newlen >= sizeof(buf))
1369 		p = malloc(req->newlen+1, M_SYSCTL, M_WAITOK);
1370 
1371 	error = SYSCTL_IN(req, p, req->newlen);
1372 	if (error) {
1373 		if (p != buf)
1374 			free(p, M_SYSCTL);
1375 		return (error);
1376 	}
1377 
1378 	p [req->newlen] = '\0';
1379 
1380 	SYSCTL_RLOCK(&tracker);
1381 	error = name2oid(p, oid, &len, &op);
1382 	SYSCTL_RUNLOCK(&tracker);
1383 
1384 	if (p != buf)
1385 		free(p, M_SYSCTL);
1386 
1387 	if (error)
1388 		return (error);
1389 
1390 	error = SYSCTL_OUT(req, oid, len * sizeof *oid);
1391 	return (error);
1392 }
1393 
1394 /*
1395  * XXXRW/JA: Shouldn't return name2oid data for nodes that we don't permit in
1396  * capability mode.
1397  */
1398 SYSCTL_PROC(_sysctl, CTL_SYSCTL_NAME2OID, name2oid, CTLTYPE_INT | CTLFLAG_RW |
1399     CTLFLAG_ANYBODY | CTLFLAG_MPSAFE | CTLFLAG_CAPRW, 0, 0,
1400     sysctl_sysctl_name2oid, "I", "");
1401 
1402 static int
1403 sysctl_sysctl_oidfmt(SYSCTL_HANDLER_ARGS)
1404 {
1405 	struct sysctl_oid *oid;
1406 	struct rm_priotracker tracker;
1407 	int error;
1408 
1409 	error = sysctl_wire_old_buffer(req, 0);
1410 	if (error)
1411 		return (error);
1412 
1413 	SYSCTL_RLOCK(&tracker);
1414 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1415 	if (error)
1416 		goto out;
1417 
1418 	if (oid->oid_fmt == NULL) {
1419 		error = ENOENT;
1420 		goto out;
1421 	}
1422 	error = SYSCTL_OUT(req, &oid->oid_kind, sizeof(oid->oid_kind));
1423 	if (error)
1424 		goto out;
1425 	error = SYSCTL_OUT(req, oid->oid_fmt, strlen(oid->oid_fmt) + 1);
1426  out:
1427 	SYSCTL_RUNLOCK(&tracker);
1428 	return (error);
1429 }
1430 
1431 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_OIDFMT, oidfmt, CTLFLAG_RD |
1432     CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_oidfmt, "");
1433 
1434 static int
1435 sysctl_sysctl_oiddescr(SYSCTL_HANDLER_ARGS)
1436 {
1437 	struct sysctl_oid *oid;
1438 	struct rm_priotracker tracker;
1439 	int error;
1440 
1441 	error = sysctl_wire_old_buffer(req, 0);
1442 	if (error)
1443 		return (error);
1444 
1445 	SYSCTL_RLOCK(&tracker);
1446 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1447 	if (error)
1448 		goto out;
1449 
1450 	if (oid->oid_descr == NULL) {
1451 		error = ENOENT;
1452 		goto out;
1453 	}
1454 	error = SYSCTL_OUT(req, oid->oid_descr, strlen(oid->oid_descr) + 1);
1455  out:
1456 	SYSCTL_RUNLOCK(&tracker);
1457 	return (error);
1458 }
1459 
1460 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_OIDDESCR, oiddescr, CTLFLAG_RD |
1461     CTLFLAG_MPSAFE|CTLFLAG_CAPRD, sysctl_sysctl_oiddescr, "");
1462 
1463 static int
1464 sysctl_sysctl_oidlabel(SYSCTL_HANDLER_ARGS)
1465 {
1466 	struct sysctl_oid *oid;
1467 	struct rm_priotracker tracker;
1468 	int error;
1469 
1470 	error = sysctl_wire_old_buffer(req, 0);
1471 	if (error)
1472 		return (error);
1473 
1474 	SYSCTL_RLOCK(&tracker);
1475 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1476 	if (error)
1477 		goto out;
1478 
1479 	if (oid->oid_label == NULL) {
1480 		error = ENOENT;
1481 		goto out;
1482 	}
1483 	error = SYSCTL_OUT(req, oid->oid_label, strlen(oid->oid_label) + 1);
1484  out:
1485 	SYSCTL_RUNLOCK(&tracker);
1486 	return (error);
1487 }
1488 
1489 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_OIDLABEL, oidlabel, CTLFLAG_RD |
1490     CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_oidlabel, "");
1491 
1492 /*
1493  * Default "handler" functions.
1494  */
1495 
1496 /*
1497  * Handle a bool.
1498  * Two cases:
1499  *     a variable:  point arg1 at it.
1500  *     a constant:  pass it in arg2.
1501  */
1502 
1503 int
1504 sysctl_handle_bool(SYSCTL_HANDLER_ARGS)
1505 {
1506 	uint8_t temp;
1507 	int error;
1508 
1509 	/*
1510 	 * Attempt to get a coherent snapshot by making a copy of the data.
1511 	 */
1512 	if (arg1)
1513 		temp = *(bool *)arg1 ? 1 : 0;
1514 	else
1515 		temp = arg2 ? 1 : 0;
1516 
1517 	error = SYSCTL_OUT(req, &temp, sizeof(temp));
1518 	if (error || !req->newptr)
1519 		return (error);
1520 
1521 	if (!arg1)
1522 		error = EPERM;
1523 	else {
1524 		error = SYSCTL_IN(req, &temp, sizeof(temp));
1525 		if (!error)
1526 			*(bool *)arg1 = temp ? 1 : 0;
1527 	}
1528 	return (error);
1529 }
1530 
1531 /*
1532  * Handle an int8_t, signed or unsigned.
1533  * Two cases:
1534  *     a variable:  point arg1 at it.
1535  *     a constant:  pass it in arg2.
1536  */
1537 
1538 int
1539 sysctl_handle_8(SYSCTL_HANDLER_ARGS)
1540 {
1541 	int8_t tmpout;
1542 	int error = 0;
1543 
1544 	/*
1545 	 * Attempt to get a coherent snapshot by making a copy of the data.
1546 	 */
1547 	if (arg1)
1548 		tmpout = *(int8_t *)arg1;
1549 	else
1550 		tmpout = arg2;
1551 	error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout));
1552 
1553 	if (error || !req->newptr)
1554 		return (error);
1555 
1556 	if (!arg1)
1557 		error = EPERM;
1558 	else
1559 		error = SYSCTL_IN(req, arg1, sizeof(tmpout));
1560 	return (error);
1561 }
1562 
1563 /*
1564  * Handle an int16_t, signed or unsigned.
1565  * Two cases:
1566  *     a variable:  point arg1 at it.
1567  *     a constant:  pass it in arg2.
1568  */
1569 
1570 int
1571 sysctl_handle_16(SYSCTL_HANDLER_ARGS)
1572 {
1573 	int16_t tmpout;
1574 	int error = 0;
1575 
1576 	/*
1577 	 * Attempt to get a coherent snapshot by making a copy of the data.
1578 	 */
1579 	if (arg1)
1580 		tmpout = *(int16_t *)arg1;
1581 	else
1582 		tmpout = arg2;
1583 	error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout));
1584 
1585 	if (error || !req->newptr)
1586 		return (error);
1587 
1588 	if (!arg1)
1589 		error = EPERM;
1590 	else
1591 		error = SYSCTL_IN(req, arg1, sizeof(tmpout));
1592 	return (error);
1593 }
1594 
1595 /*
1596  * Handle an int32_t, signed or unsigned.
1597  * Two cases:
1598  *     a variable:  point arg1 at it.
1599  *     a constant:  pass it in arg2.
1600  */
1601 
1602 int
1603 sysctl_handle_32(SYSCTL_HANDLER_ARGS)
1604 {
1605 	int32_t tmpout;
1606 	int error = 0;
1607 
1608 	/*
1609 	 * Attempt to get a coherent snapshot by making a copy of the data.
1610 	 */
1611 	if (arg1)
1612 		tmpout = *(int32_t *)arg1;
1613 	else
1614 		tmpout = arg2;
1615 	error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout));
1616 
1617 	if (error || !req->newptr)
1618 		return (error);
1619 
1620 	if (!arg1)
1621 		error = EPERM;
1622 	else
1623 		error = SYSCTL_IN(req, arg1, sizeof(tmpout));
1624 	return (error);
1625 }
1626 
1627 /*
1628  * Handle an int, signed or unsigned.
1629  * Two cases:
1630  *     a variable:  point arg1 at it.
1631  *     a constant:  pass it in arg2.
1632  */
1633 
1634 int
1635 sysctl_handle_int(SYSCTL_HANDLER_ARGS)
1636 {
1637 	int tmpout, error = 0;
1638 
1639 	/*
1640 	 * Attempt to get a coherent snapshot by making a copy of the data.
1641 	 */
1642 	if (arg1)
1643 		tmpout = *(int *)arg1;
1644 	else
1645 		tmpout = arg2;
1646 	error = SYSCTL_OUT(req, &tmpout, sizeof(int));
1647 
1648 	if (error || !req->newptr)
1649 		return (error);
1650 
1651 	if (!arg1)
1652 		error = EPERM;
1653 	else
1654 		error = SYSCTL_IN(req, arg1, sizeof(int));
1655 	return (error);
1656 }
1657 
1658 /*
1659  * Based on sysctl_handle_int() convert milliseconds into ticks.
1660  * Note: this is used by TCP.
1661  */
1662 
1663 int
1664 sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
1665 {
1666 	int error, s, tt;
1667 
1668 	tt = *(int *)arg1;
1669 	s = (int)((int64_t)tt * 1000 / hz);
1670 
1671 	error = sysctl_handle_int(oidp, &s, 0, req);
1672 	if (error || !req->newptr)
1673 		return (error);
1674 
1675 	tt = (int)((int64_t)s * hz / 1000);
1676 	if (tt < 1)
1677 		return (EINVAL);
1678 
1679 	*(int *)arg1 = tt;
1680 	return (0);
1681 }
1682 
1683 /*
1684  * Handle a long, signed or unsigned.
1685  * Two cases:
1686  *     a variable:  point arg1 at it.
1687  *     a constant:  pass it in arg2.
1688  */
1689 
1690 int
1691 sysctl_handle_long(SYSCTL_HANDLER_ARGS)
1692 {
1693 	int error = 0;
1694 	long tmplong;
1695 #ifdef SCTL_MASK32
1696 	int tmpint;
1697 #endif
1698 
1699 	/*
1700 	 * Attempt to get a coherent snapshot by making a copy of the data.
1701 	 */
1702 	if (arg1)
1703 		tmplong = *(long *)arg1;
1704 	else
1705 		tmplong = arg2;
1706 #ifdef SCTL_MASK32
1707 	if (req->flags & SCTL_MASK32) {
1708 		tmpint = tmplong;
1709 		error = SYSCTL_OUT(req, &tmpint, sizeof(int));
1710 	} else
1711 #endif
1712 		error = SYSCTL_OUT(req, &tmplong, sizeof(long));
1713 
1714 	if (error || !req->newptr)
1715 		return (error);
1716 
1717 	if (!arg1)
1718 		error = EPERM;
1719 #ifdef SCTL_MASK32
1720 	else if (req->flags & SCTL_MASK32) {
1721 		error = SYSCTL_IN(req, &tmpint, sizeof(int));
1722 		*(long *)arg1 = (long)tmpint;
1723 	}
1724 #endif
1725 	else
1726 		error = SYSCTL_IN(req, arg1, sizeof(long));
1727 	return (error);
1728 }
1729 
1730 /*
1731  * Handle a 64 bit int, signed or unsigned.
1732  * Two cases:
1733  *     a variable:  point arg1 at it.
1734  *     a constant:  pass it in arg2.
1735  */
1736 int
1737 sysctl_handle_64(SYSCTL_HANDLER_ARGS)
1738 {
1739 	int error = 0;
1740 	uint64_t tmpout;
1741 
1742 	/*
1743 	 * Attempt to get a coherent snapshot by making a copy of the data.
1744 	 */
1745 	if (arg1)
1746 		tmpout = *(uint64_t *)arg1;
1747 	else
1748 		tmpout = arg2;
1749 	error = SYSCTL_OUT(req, &tmpout, sizeof(uint64_t));
1750 
1751 	if (error || !req->newptr)
1752 		return (error);
1753 
1754 	if (!arg1)
1755 		error = EPERM;
1756 	else
1757 		error = SYSCTL_IN(req, arg1, sizeof(uint64_t));
1758 	return (error);
1759 }
1760 
1761 /*
1762  * Handle our generic '\0' terminated 'C' string.
1763  * Two cases:
1764  * 	a variable string:  point arg1 at it, arg2 is max length.
1765  * 	a constant string:  point arg1 at it, arg2 is zero.
1766  */
1767 
1768 int
1769 sysctl_handle_string(SYSCTL_HANDLER_ARGS)
1770 {
1771 	char *tmparg;
1772 	size_t outlen;
1773 	int error = 0, ro_string = 0;
1774 
1775 	/*
1776 	 * If the sysctl isn't writable and isn't a preallocated tunable that
1777 	 * can be modified by kenv(2), microoptimise and treat it as a
1778 	 * read-only string.
1779 	 * A zero-length buffer indicates a fixed size read-only
1780 	 * string.  In ddb, don't worry about trying to make a malloced
1781 	 * snapshot.
1782 	 */
1783 	if ((oidp->oid_kind & (CTLFLAG_WR | CTLFLAG_TUN)) == 0 ||
1784 	    arg2 == 0 || kdb_active) {
1785 		arg2 = strlen((char *)arg1) + 1;
1786 		ro_string = 1;
1787 	}
1788 
1789 	if (req->oldptr != NULL) {
1790 		if (ro_string) {
1791 			tmparg = arg1;
1792 			outlen = strlen(tmparg) + 1;
1793 		} else {
1794 			tmparg = malloc(arg2, M_SYSCTLTMP, M_WAITOK);
1795 			sx_slock(&sysctlstringlock);
1796 			memcpy(tmparg, arg1, arg2);
1797 			sx_sunlock(&sysctlstringlock);
1798 			outlen = strlen(tmparg) + 1;
1799 		}
1800 
1801 		error = SYSCTL_OUT(req, tmparg, outlen);
1802 
1803 		if (!ro_string)
1804 			free(tmparg, M_SYSCTLTMP);
1805 	} else {
1806 		if (!ro_string)
1807 			sx_slock(&sysctlstringlock);
1808 		outlen = strlen((char *)arg1) + 1;
1809 		if (!ro_string)
1810 			sx_sunlock(&sysctlstringlock);
1811 		error = SYSCTL_OUT(req, NULL, outlen);
1812 	}
1813 	if (error || !req->newptr)
1814 		return (error);
1815 
1816 	if (req->newlen - req->newidx >= arg2 ||
1817 	    req->newlen - req->newidx < 0) {
1818 		error = EINVAL;
1819 	} else if (req->newlen - req->newidx == 0) {
1820 		sx_xlock(&sysctlstringlock);
1821 		((char *)arg1)[0] = '\0';
1822 		sx_xunlock(&sysctlstringlock);
1823 	} else if (req->newfunc == sysctl_new_kernel) {
1824 		arg2 = req->newlen - req->newidx;
1825 		sx_xlock(&sysctlstringlock);
1826 		error = SYSCTL_IN(req, arg1, arg2);
1827 		if (error == 0) {
1828 			((char *)arg1)[arg2] = '\0';
1829 			req->newidx += arg2;
1830 		}
1831 		sx_xunlock(&sysctlstringlock);
1832 	} else {
1833 		arg2 = req->newlen - req->newidx;
1834 		tmparg = malloc(arg2, M_SYSCTLTMP, M_WAITOK);
1835 
1836 		error = SYSCTL_IN(req, tmparg, arg2);
1837 		if (error) {
1838 			free(tmparg, M_SYSCTLTMP);
1839 			return (error);
1840 		}
1841 
1842 		sx_xlock(&sysctlstringlock);
1843 		memcpy(arg1, tmparg, arg2);
1844 		((char *)arg1)[arg2] = '\0';
1845 		sx_xunlock(&sysctlstringlock);
1846 		free(tmparg, M_SYSCTLTMP);
1847 		req->newidx += arg2;
1848 	}
1849 	return (error);
1850 }
1851 
1852 /*
1853  * Handle any kind of opaque data.
1854  * arg1 points to it, arg2 is the size.
1855  */
1856 
1857 int
1858 sysctl_handle_opaque(SYSCTL_HANDLER_ARGS)
1859 {
1860 	int error, tries;
1861 	u_int generation;
1862 	struct sysctl_req req2;
1863 
1864 	/*
1865 	 * Attempt to get a coherent snapshot, by using the thread
1866 	 * pre-emption counter updated from within mi_switch() to
1867 	 * determine if we were pre-empted during a bcopy() or
1868 	 * copyout(). Make 3 attempts at doing this before giving up.
1869 	 * If we encounter an error, stop immediately.
1870 	 */
1871 	tries = 0;
1872 	req2 = *req;
1873 retry:
1874 	generation = curthread->td_generation;
1875 	error = SYSCTL_OUT(req, arg1, arg2);
1876 	if (error)
1877 		return (error);
1878 	tries++;
1879 	if (generation != curthread->td_generation && tries < 3) {
1880 		*req = req2;
1881 		goto retry;
1882 	}
1883 
1884 	error = SYSCTL_IN(req, arg1, arg2);
1885 
1886 	return (error);
1887 }
1888 
1889 /*
1890  * Based on sysctl_handle_64() convert microseconds to a sbintime.
1891  */
1892 int
1893 sysctl_usec_to_sbintime(SYSCTL_HANDLER_ARGS)
1894 {
1895 	int error;
1896 	int64_t usec;
1897 
1898 	usec = sbttous(*(sbintime_t *)arg1);
1899 
1900 	error = sysctl_handle_64(oidp, &usec, 0, req);
1901 	if (error || !req->newptr)
1902 		return (error);
1903 
1904 	*(sbintime_t *)arg1 = ustosbt(usec);
1905 
1906 	return (0);
1907 }
1908 
1909 /*
1910  * Based on sysctl_handle_64() convert milliseconds to a sbintime.
1911  */
1912 int
1913 sysctl_msec_to_sbintime(SYSCTL_HANDLER_ARGS)
1914 {
1915 	int error;
1916 	int64_t msec;
1917 
1918 	msec = sbttoms(*(sbintime_t *)arg1);
1919 
1920 	error = sysctl_handle_64(oidp, &msec, 0, req);
1921 	if (error || !req->newptr)
1922 		return (error);
1923 
1924 	*(sbintime_t *)arg1 = mstosbt(msec);
1925 
1926 	return (0);
1927 }
1928 
1929 /*
1930  * Convert seconds to a struct timeval.  Intended for use with
1931  * intervals and thus does not permit negative seconds.
1932  */
1933 int
1934 sysctl_sec_to_timeval(SYSCTL_HANDLER_ARGS)
1935 {
1936 	struct timeval *tv;
1937 	int error, secs;
1938 
1939 	tv = arg1;
1940 	secs = tv->tv_sec;
1941 
1942 	error = sysctl_handle_int(oidp, &secs, 0, req);
1943 	if (error || req->newptr == NULL)
1944 		return (error);
1945 
1946 	if (secs < 0)
1947 		return (EINVAL);
1948 	tv->tv_sec = secs;
1949 
1950 	return (0);
1951 }
1952 
1953 /*
1954  * Transfer functions to/from kernel space.
1955  * XXX: rather untested at this point
1956  */
1957 static int
1958 sysctl_old_kernel(struct sysctl_req *req, const void *p, size_t l)
1959 {
1960 	size_t i = 0;
1961 
1962 	if (req->oldptr) {
1963 		i = l;
1964 		if (req->oldlen <= req->oldidx)
1965 			i = 0;
1966 		else
1967 			if (i > req->oldlen - req->oldidx)
1968 				i = req->oldlen - req->oldidx;
1969 		if (i > 0)
1970 			bcopy(p, (char *)req->oldptr + req->oldidx, i);
1971 	}
1972 	req->oldidx += l;
1973 	if (req->oldptr && i != l)
1974 		return (ENOMEM);
1975 	return (0);
1976 }
1977 
1978 static int
1979 sysctl_new_kernel(struct sysctl_req *req, void *p, size_t l)
1980 {
1981 	if (!req->newptr)
1982 		return (0);
1983 	if (req->newlen - req->newidx < l)
1984 		return (EINVAL);
1985 	bcopy((const char *)req->newptr + req->newidx, p, l);
1986 	req->newidx += l;
1987 	return (0);
1988 }
1989 
1990 int
1991 kernel_sysctl(struct thread *td, int *name, u_int namelen, void *old,
1992     size_t *oldlenp, void *new, size_t newlen, size_t *retval, int flags)
1993 {
1994 	int error = 0;
1995 	struct sysctl_req req;
1996 
1997 	bzero(&req, sizeof req);
1998 
1999 	req.td = td;
2000 	req.flags = flags;
2001 
2002 	if (oldlenp) {
2003 		req.oldlen = *oldlenp;
2004 	}
2005 	req.validlen = req.oldlen;
2006 
2007 	if (old) {
2008 		req.oldptr= old;
2009 	}
2010 
2011 	if (new != NULL) {
2012 		req.newlen = newlen;
2013 		req.newptr = new;
2014 	}
2015 
2016 	req.oldfunc = sysctl_old_kernel;
2017 	req.newfunc = sysctl_new_kernel;
2018 	req.lock = REQ_UNWIRED;
2019 
2020 	error = sysctl_root(0, name, namelen, &req);
2021 
2022 	if (req.lock == REQ_WIRED && req.validlen > 0)
2023 		vsunlock(req.oldptr, req.validlen);
2024 
2025 	if (error && error != ENOMEM)
2026 		return (error);
2027 
2028 	if (retval) {
2029 		if (req.oldptr && req.oldidx > req.validlen)
2030 			*retval = req.validlen;
2031 		else
2032 			*retval = req.oldidx;
2033 	}
2034 	return (error);
2035 }
2036 
2037 int
2038 kernel_sysctlbyname(struct thread *td, char *name, void *old, size_t *oldlenp,
2039     void *new, size_t newlen, size_t *retval, int flags)
2040 {
2041         int oid[CTL_MAXNAME];
2042         size_t oidlen, plen;
2043 	int error;
2044 
2045 	oid[0] = CTL_SYSCTL;
2046 	oid[1] = CTL_SYSCTL_NAME2OID;
2047 	oidlen = sizeof(oid);
2048 
2049 	error = kernel_sysctl(td, oid, 2, oid, &oidlen,
2050 	    (void *)name, strlen(name), &plen, flags);
2051 	if (error)
2052 		return (error);
2053 
2054 	error = kernel_sysctl(td, oid, plen / sizeof(int), old, oldlenp,
2055 	    new, newlen, retval, flags);
2056 	return (error);
2057 }
2058 
2059 /*
2060  * Transfer function to/from user space.
2061  */
2062 static int
2063 sysctl_old_user(struct sysctl_req *req, const void *p, size_t l)
2064 {
2065 	size_t i, len, origidx;
2066 	int error;
2067 
2068 	origidx = req->oldidx;
2069 	req->oldidx += l;
2070 	if (req->oldptr == NULL)
2071 		return (0);
2072 	/*
2073 	 * If we have not wired the user supplied buffer and we are currently
2074 	 * holding locks, drop a witness warning, as it's possible that
2075 	 * write operations to the user page can sleep.
2076 	 */
2077 	if (req->lock != REQ_WIRED)
2078 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
2079 		    "sysctl_old_user()");
2080 	i = l;
2081 	len = req->validlen;
2082 	if (len <= origidx)
2083 		i = 0;
2084 	else {
2085 		if (i > len - origidx)
2086 			i = len - origidx;
2087 		if (req->lock == REQ_WIRED) {
2088 			error = copyout_nofault(p, (char *)req->oldptr +
2089 			    origidx, i);
2090 		} else
2091 			error = copyout(p, (char *)req->oldptr + origidx, i);
2092 		if (error != 0)
2093 			return (error);
2094 	}
2095 	if (i < l)
2096 		return (ENOMEM);
2097 	return (0);
2098 }
2099 
2100 static int
2101 sysctl_new_user(struct sysctl_req *req, void *p, size_t l)
2102 {
2103 	int error;
2104 
2105 	if (!req->newptr)
2106 		return (0);
2107 	if (req->newlen - req->newidx < l)
2108 		return (EINVAL);
2109 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
2110 	    "sysctl_new_user()");
2111 	error = copyin((const char *)req->newptr + req->newidx, p, l);
2112 	req->newidx += l;
2113 	return (error);
2114 }
2115 
2116 /*
2117  * Wire the user space destination buffer.  If set to a value greater than
2118  * zero, the len parameter limits the maximum amount of wired memory.
2119  */
2120 int
2121 sysctl_wire_old_buffer(struct sysctl_req *req, size_t len)
2122 {
2123 	int ret;
2124 	size_t wiredlen;
2125 
2126 	wiredlen = (len > 0 && len < req->oldlen) ? len : req->oldlen;
2127 	ret = 0;
2128 	if (req->lock != REQ_WIRED && req->oldptr &&
2129 	    req->oldfunc == sysctl_old_user) {
2130 		if (wiredlen != 0) {
2131 			ret = vslock(req->oldptr, wiredlen);
2132 			if (ret != 0) {
2133 				if (ret != ENOMEM)
2134 					return (ret);
2135 				wiredlen = 0;
2136 			}
2137 		}
2138 		req->lock = REQ_WIRED;
2139 		req->validlen = wiredlen;
2140 	}
2141 	return (0);
2142 }
2143 
2144 int
2145 sysctl_find_oid(int *name, u_int namelen, struct sysctl_oid **noid,
2146     int *nindx, struct sysctl_req *req)
2147 {
2148 	struct sysctl_oid_list *lsp;
2149 	struct sysctl_oid *oid;
2150 	struct sysctl_oid key;
2151 	int indx;
2152 
2153 	SYSCTL_ASSERT_LOCKED();
2154 	lsp = &sysctl__children;
2155 	indx = 0;
2156 	while (indx < CTL_MAXNAME) {
2157 		key.oid_number = name[indx];
2158 		oid = RB_FIND(sysctl_oid_list, lsp, &key);
2159 		if (oid == NULL)
2160 			return (ENOENT);
2161 
2162 		indx++;
2163 		if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
2164 			if (oid->oid_handler != NULL || indx == namelen) {
2165 				*noid = oid;
2166 				if (nindx != NULL)
2167 					*nindx = indx;
2168 				KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
2169 				    ("%s found DYING node %p", __func__, oid));
2170 				return (0);
2171 			}
2172 			lsp = SYSCTL_CHILDREN(oid);
2173 		} else if (indx == namelen) {
2174 			if ((oid->oid_kind & CTLFLAG_DORMANT) != 0)
2175 				return (ENOENT);
2176 			*noid = oid;
2177 			if (nindx != NULL)
2178 				*nindx = indx;
2179 			KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
2180 			    ("%s found DYING node %p", __func__, oid));
2181 			return (0);
2182 		} else {
2183 			return (ENOTDIR);
2184 		}
2185 	}
2186 	return (ENOENT);
2187 }
2188 
2189 /*
2190  * Traverse our tree, and find the right node, execute whatever it points
2191  * to, and return the resulting error code.
2192  */
2193 
2194 static int
2195 sysctl_root(SYSCTL_HANDLER_ARGS)
2196 {
2197 	struct sysctl_oid *oid;
2198 	struct rm_priotracker tracker;
2199 	int error, indx, lvl;
2200 
2201 	SYSCTL_RLOCK(&tracker);
2202 
2203 	error = sysctl_find_oid(arg1, arg2, &oid, &indx, req);
2204 	if (error)
2205 		goto out;
2206 
2207 	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
2208 		/*
2209 		 * You can't call a sysctl when it's a node, but has
2210 		 * no handler.  Inform the user that it's a node.
2211 		 * The indx may or may not be the same as namelen.
2212 		 */
2213 		if (oid->oid_handler == NULL) {
2214 			error = EISDIR;
2215 			goto out;
2216 		}
2217 	}
2218 
2219 	/* Is this sysctl writable? */
2220 	if (req->newptr && !(oid->oid_kind & CTLFLAG_WR)) {
2221 		error = EPERM;
2222 		goto out;
2223 	}
2224 
2225 	KASSERT(req->td != NULL, ("sysctl_root(): req->td == NULL"));
2226 
2227 #ifdef CAPABILITY_MODE
2228 	/*
2229 	 * If the process is in capability mode, then don't permit reading or
2230 	 * writing unless specifically granted for the node.
2231 	 */
2232 	if (IN_CAPABILITY_MODE(req->td)) {
2233 		if ((req->oldptr && !(oid->oid_kind & CTLFLAG_CAPRD)) ||
2234 		    (req->newptr && !(oid->oid_kind & CTLFLAG_CAPWR))) {
2235 			error = EPERM;
2236 			goto out;
2237 		}
2238 	}
2239 #endif
2240 
2241 	/* Is this sysctl sensitive to securelevels? */
2242 	if (req->newptr && (oid->oid_kind & CTLFLAG_SECURE)) {
2243 		lvl = (oid->oid_kind & CTLMASK_SECURE) >> CTLSHIFT_SECURE;
2244 		error = securelevel_gt(req->td->td_ucred, lvl);
2245 		if (error)
2246 			goto out;
2247 	}
2248 
2249 	/* Is this sysctl writable by only privileged users? */
2250 	if (req->newptr && !(oid->oid_kind & CTLFLAG_ANYBODY)) {
2251 		int priv;
2252 
2253 		if (oid->oid_kind & CTLFLAG_PRISON)
2254 			priv = PRIV_SYSCTL_WRITEJAIL;
2255 #ifdef VIMAGE
2256 		else if ((oid->oid_kind & CTLFLAG_VNET) &&
2257 		     prison_owns_vnet(req->td->td_ucred))
2258 			priv = PRIV_SYSCTL_WRITEJAIL;
2259 #endif
2260 		else
2261 			priv = PRIV_SYSCTL_WRITE;
2262 		error = priv_check(req->td, priv);
2263 		if (error)
2264 			goto out;
2265 	}
2266 
2267 	if (!oid->oid_handler) {
2268 		error = EINVAL;
2269 		goto out;
2270 	}
2271 
2272 	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
2273 		arg1 = (int *)arg1 + indx;
2274 		arg2 -= indx;
2275 	} else {
2276 		arg1 = oid->oid_arg1;
2277 		arg2 = oid->oid_arg2;
2278 	}
2279 #ifdef MAC
2280 	error = mac_system_check_sysctl(req->td->td_ucred, oid, arg1, arg2,
2281 	    req);
2282 	if (error != 0)
2283 		goto out;
2284 #endif
2285 #ifdef VIMAGE
2286 	if ((oid->oid_kind & CTLFLAG_VNET) && arg1 != NULL)
2287 		arg1 = (void *)(curvnet->vnet_data_base + (uintptr_t)arg1);
2288 #endif
2289 	error = sysctl_root_handler_locked(oid, arg1, arg2, req, &tracker);
2290 
2291 out:
2292 	SYSCTL_RUNLOCK(&tracker);
2293 	return (error);
2294 }
2295 
2296 #ifndef _SYS_SYSPROTO_H_
2297 struct __sysctl_args {
2298 	int	*name;
2299 	u_int	namelen;
2300 	void	*old;
2301 	size_t	*oldlenp;
2302 	void	*new;
2303 	size_t	newlen;
2304 };
2305 #endif
2306 int
2307 sys___sysctl(struct thread *td, struct __sysctl_args *uap)
2308 {
2309 	int error, i, name[CTL_MAXNAME];
2310 	size_t j;
2311 
2312 	if (uap->namelen > CTL_MAXNAME || uap->namelen < 2)
2313 		return (EINVAL);
2314 
2315  	error = copyin(uap->name, &name, uap->namelen * sizeof(int));
2316  	if (error)
2317 		return (error);
2318 
2319 	error = userland_sysctl(td, name, uap->namelen,
2320 		uap->old, uap->oldlenp, 0,
2321 		uap->new, uap->newlen, &j, 0);
2322 	if (error && error != ENOMEM)
2323 		return (error);
2324 	if (uap->oldlenp) {
2325 		i = copyout(&j, uap->oldlenp, sizeof(j));
2326 		if (i)
2327 			return (i);
2328 	}
2329 	return (error);
2330 }
2331 
2332 int
2333 kern___sysctlbyname(struct thread *td, const char *oname, size_t namelen,
2334     void *old, size_t *oldlenp, void *new, size_t newlen, size_t *retval,
2335     int flags, bool inkernel)
2336 {
2337 	int oid[CTL_MAXNAME];
2338 	char namebuf[16];
2339 	char *name;
2340 	size_t oidlen;
2341 	int error;
2342 
2343 	if (namelen > MAXPATHLEN || namelen == 0)
2344 		return (EINVAL);
2345 	name = namebuf;
2346 	if (namelen > sizeof(namebuf))
2347 		name = malloc(namelen, M_SYSCTL, M_WAITOK);
2348 	error = copyin(oname, name, namelen);
2349 	if (error != 0)
2350 		goto out;
2351 
2352 	oid[0] = CTL_SYSCTL;
2353 	oid[1] = CTL_SYSCTL_NAME2OID;
2354 	oidlen = sizeof(oid);
2355 	error = kernel_sysctl(td, oid, 2, oid, &oidlen, (void *)name, namelen,
2356 	    retval, flags);
2357 	if (error != 0)
2358 		goto out;
2359 	error = userland_sysctl(td, oid, *retval / sizeof(int), old, oldlenp,
2360 	    inkernel, new, newlen, retval, flags);
2361 
2362 out:
2363 	if (namelen > sizeof(namebuf))
2364 		free(name, M_SYSCTL);
2365 	return (error);
2366 }
2367 
2368 #ifndef	_SYS_SYSPROTO_H_
2369 struct __sysctlbyname_args {
2370 	const char	*name;
2371 	size_t	namelen;
2372 	void	*old;
2373 	size_t	*oldlenp;
2374 	void	*new;
2375 	size_t	newlen;
2376 };
2377 #endif
2378 int
2379 sys___sysctlbyname(struct thread *td, struct __sysctlbyname_args *uap)
2380 {
2381 	size_t rv;
2382 	int error;
2383 
2384 	error = kern___sysctlbyname(td, uap->name, uap->namelen, uap->old,
2385 	    uap->oldlenp, uap->new, uap->newlen, &rv, 0, 0);
2386 	if (error != 0)
2387 		return (error);
2388 	if (uap->oldlenp != NULL)
2389 		error = copyout(&rv, uap->oldlenp, sizeof(rv));
2390 
2391 	return (error);
2392 }
2393 
2394 /*
2395  * This is used from various compatibility syscalls too.  That's why name
2396  * must be in kernel space.
2397  */
2398 int
2399 userland_sysctl(struct thread *td, int *name, u_int namelen, void *old,
2400     size_t *oldlenp, int inkernel, const void *new, size_t newlen,
2401     size_t *retval, int flags)
2402 {
2403 	int error = 0, memlocked;
2404 	struct sysctl_req req;
2405 
2406 	bzero(&req, sizeof req);
2407 
2408 	req.td = td;
2409 	req.flags = flags;
2410 
2411 	if (oldlenp) {
2412 		if (inkernel) {
2413 			req.oldlen = *oldlenp;
2414 		} else {
2415 			error = copyin(oldlenp, &req.oldlen, sizeof(*oldlenp));
2416 			if (error)
2417 				return (error);
2418 		}
2419 	}
2420 	req.validlen = req.oldlen;
2421 	req.oldptr = old;
2422 
2423 	if (new != NULL) {
2424 		req.newlen = newlen;
2425 		req.newptr = new;
2426 	}
2427 
2428 	req.oldfunc = sysctl_old_user;
2429 	req.newfunc = sysctl_new_user;
2430 	req.lock = REQ_UNWIRED;
2431 
2432 #ifdef KTRACE
2433 	if (KTRPOINT(curthread, KTR_SYSCTL))
2434 		ktrsysctl(name, namelen);
2435 #endif
2436 	memlocked = 0;
2437 	if (req.oldptr && req.oldlen > 4 * PAGE_SIZE) {
2438 		memlocked = 1;
2439 		sx_xlock(&sysctlmemlock);
2440 	}
2441 	CURVNET_SET(TD_TO_VNET(td));
2442 
2443 	for (;;) {
2444 		req.oldidx = 0;
2445 		req.newidx = 0;
2446 		error = sysctl_root(0, name, namelen, &req);
2447 		if (error != EAGAIN)
2448 			break;
2449 		kern_yield(PRI_USER);
2450 	}
2451 
2452 	CURVNET_RESTORE();
2453 
2454 	if (req.lock == REQ_WIRED && req.validlen > 0)
2455 		vsunlock(req.oldptr, req.validlen);
2456 	if (memlocked)
2457 		sx_xunlock(&sysctlmemlock);
2458 
2459 	if (error && error != ENOMEM)
2460 		return (error);
2461 
2462 	if (retval) {
2463 		if (req.oldptr && req.oldidx > req.validlen)
2464 			*retval = req.validlen;
2465 		else
2466 			*retval = req.oldidx;
2467 	}
2468 	return (error);
2469 }
2470 
2471 /*
2472  * Drain into a sysctl struct.  The user buffer should be wired if a page
2473  * fault would cause issue.
2474  */
2475 static int
2476 sbuf_sysctl_drain(void *arg, const char *data, int len)
2477 {
2478 	struct sysctl_req *req = arg;
2479 	int error;
2480 
2481 	error = SYSCTL_OUT(req, data, len);
2482 	KASSERT(error >= 0, ("Got unexpected negative value %d", error));
2483 	return (error == 0 ? len : -error);
2484 }
2485 
2486 struct sbuf *
2487 sbuf_new_for_sysctl(struct sbuf *s, char *buf, int length,
2488     struct sysctl_req *req)
2489 {
2490 
2491 	/* Supply a default buffer size if none given. */
2492 	if (buf == NULL && length == 0)
2493 		length = 64;
2494 	s = sbuf_new(s, buf, length, SBUF_FIXEDLEN | SBUF_INCLUDENUL);
2495 	sbuf_set_drain(s, sbuf_sysctl_drain, req);
2496 	return (s);
2497 }
2498 
2499 #ifdef DDB
2500 
2501 /* The current OID the debugger is working with */
2502 static struct sysctl_oid *g_ddb_oid;
2503 
2504 /* The current flags specified by the user */
2505 static int g_ddb_sysctl_flags;
2506 
2507 /* Check to see if the last sysctl printed */
2508 static int g_ddb_sysctl_printed;
2509 
2510 static const int ctl_sign[CTLTYPE+1] = {
2511 	[CTLTYPE_INT] = 1,
2512 	[CTLTYPE_LONG] = 1,
2513 	[CTLTYPE_S8] = 1,
2514 	[CTLTYPE_S16] = 1,
2515 	[CTLTYPE_S32] = 1,
2516 	[CTLTYPE_S64] = 1,
2517 };
2518 
2519 static const int ctl_size[CTLTYPE+1] = {
2520 	[CTLTYPE_INT] = sizeof(int),
2521 	[CTLTYPE_UINT] = sizeof(u_int),
2522 	[CTLTYPE_LONG] = sizeof(long),
2523 	[CTLTYPE_ULONG] = sizeof(u_long),
2524 	[CTLTYPE_S8] = sizeof(int8_t),
2525 	[CTLTYPE_S16] = sizeof(int16_t),
2526 	[CTLTYPE_S32] = sizeof(int32_t),
2527 	[CTLTYPE_S64] = sizeof(int64_t),
2528 	[CTLTYPE_U8] = sizeof(uint8_t),
2529 	[CTLTYPE_U16] = sizeof(uint16_t),
2530 	[CTLTYPE_U32] = sizeof(uint32_t),
2531 	[CTLTYPE_U64] = sizeof(uint64_t),
2532 };
2533 
2534 #define DB_SYSCTL_NAME_ONLY	0x001	/* Compare with -N */
2535 #define DB_SYSCTL_VALUE_ONLY	0x002	/* Compare with -n */
2536 #define DB_SYSCTL_OPAQUE	0x004	/* Compare with -o */
2537 #define DB_SYSCTL_HEX		0x008	/* Compare with -x */
2538 
2539 #define DB_SYSCTL_SAFE_ONLY	0x100	/* Only simple types */
2540 
2541 static const char db_sysctl_modifs[] = {
2542 	'N', 'n', 'o', 'x',
2543 };
2544 
2545 static const int db_sysctl_modif_values[] = {
2546 	DB_SYSCTL_NAME_ONLY, DB_SYSCTL_VALUE_ONLY,
2547 	DB_SYSCTL_OPAQUE, DB_SYSCTL_HEX,
2548 };
2549 
2550 /* Handlers considered safe to print while recursing */
2551 static int (* const db_safe_handlers[])(SYSCTL_HANDLER_ARGS) = {
2552 	sysctl_handle_bool,
2553 	sysctl_handle_8,
2554 	sysctl_handle_16,
2555 	sysctl_handle_32,
2556 	sysctl_handle_64,
2557 	sysctl_handle_int,
2558 	sysctl_handle_long,
2559 	sysctl_handle_string,
2560 	sysctl_handle_opaque,
2561 };
2562 
2563 /*
2564  * Use in place of sysctl_old_kernel to print sysctl values.
2565  *
2566  * Compare to the output handling in show_var from sbin/sysctl/sysctl.c
2567  */
2568 static int
2569 sysctl_old_ddb(struct sysctl_req *req, const void *ptr, size_t len)
2570 {
2571 	const u_char *val, *p;
2572 	const char *sep1;
2573 	size_t intlen, slen;
2574 	uintmax_t umv;
2575 	intmax_t mv;
2576 	int sign, ctltype, hexlen, xflag, error;
2577 
2578 	/* Suppress false-positive GCC uninitialized variable warnings */
2579 	mv = 0;
2580 	umv = 0;
2581 
2582 	slen = len;
2583 	val = p = ptr;
2584 
2585 	if (ptr == NULL) {
2586 		error = 0;
2587 		goto out;
2588 	}
2589 
2590 	/* We are going to print */
2591 	g_ddb_sysctl_printed = 1;
2592 
2593 	xflag = g_ddb_sysctl_flags & DB_SYSCTL_HEX;
2594 
2595 	ctltype = (g_ddb_oid->oid_kind & CTLTYPE);
2596 	sign = ctl_sign[ctltype];
2597 	intlen = ctl_size[ctltype];
2598 
2599 	switch (ctltype) {
2600 	case CTLTYPE_NODE:
2601 	case CTLTYPE_STRING:
2602 		db_printf("%.*s", (int) len, (const char *) p);
2603 		error = 0;
2604 		goto out;
2605 
2606 	case CTLTYPE_INT:
2607 	case CTLTYPE_UINT:
2608 	case CTLTYPE_LONG:
2609 	case CTLTYPE_ULONG:
2610 	case CTLTYPE_S8:
2611 	case CTLTYPE_S16:
2612 	case CTLTYPE_S32:
2613 	case CTLTYPE_S64:
2614 	case CTLTYPE_U8:
2615 	case CTLTYPE_U16:
2616 	case CTLTYPE_U32:
2617 	case CTLTYPE_U64:
2618 		hexlen = 2 + (intlen * CHAR_BIT + 3) / 4;
2619 		sep1 = "";
2620 		while (len >= intlen) {
2621 			switch (ctltype) {
2622 			case CTLTYPE_INT:
2623 			case CTLTYPE_UINT:
2624 				umv = *(const u_int *)p;
2625 				mv = *(const int *)p;
2626 				break;
2627 			case CTLTYPE_LONG:
2628 			case CTLTYPE_ULONG:
2629 				umv = *(const u_long *)p;
2630 				mv = *(const long *)p;
2631 				break;
2632 			case CTLTYPE_S8:
2633 			case CTLTYPE_U8:
2634 				umv = *(const uint8_t *)p;
2635 				mv = *(const int8_t *)p;
2636 				break;
2637 			case CTLTYPE_S16:
2638 			case CTLTYPE_U16:
2639 				umv = *(const uint16_t *)p;
2640 				mv = *(const int16_t *)p;
2641 				break;
2642 			case CTLTYPE_S32:
2643 			case CTLTYPE_U32:
2644 				umv = *(const uint32_t *)p;
2645 				mv = *(const int32_t *)p;
2646 				break;
2647 			case CTLTYPE_S64:
2648 			case CTLTYPE_U64:
2649 				umv = *(const uint64_t *)p;
2650 				mv = *(const int64_t *)p;
2651 				break;
2652 			}
2653 
2654 			db_printf("%s", sep1);
2655 			if (xflag)
2656 				db_printf("%#0*jx", hexlen, umv);
2657 			else if (!sign)
2658 				db_printf("%ju", umv);
2659 			else if (g_ddb_oid->oid_fmt[1] == 'K') {
2660 				/* Kelvins are currently unsupported. */
2661 				error = EOPNOTSUPP;
2662 				goto out;
2663 			} else
2664 				db_printf("%jd", mv);
2665 
2666 			sep1 = " ";
2667 			len -= intlen;
2668 			p += intlen;
2669 		}
2670 		error = 0;
2671 		goto out;
2672 
2673 	case CTLTYPE_OPAQUE:
2674 		/* TODO: Support struct functions. */
2675 
2676 		/* FALLTHROUGH */
2677 	default:
2678 		db_printf("Format:%s Length:%zu Dump:0x",
2679 		    g_ddb_oid->oid_fmt, len);
2680 		while (len-- && (xflag || p < val + 16))
2681 			db_printf("%02x", *p++);
2682 		if (!xflag && len > 16)
2683 			db_printf("...");
2684 		error = 0;
2685 		goto out;
2686 	}
2687 
2688 out:
2689 	req->oldidx += slen;
2690 	return (error);
2691 }
2692 
2693 /*
2694  * Avoid setting new sysctl values from the debugger
2695  */
2696 static int
2697 sysctl_new_ddb(struct sysctl_req *req, void *p, size_t l)
2698 {
2699 
2700 	if (!req->newptr)
2701 		return (0);
2702 
2703 	/* Changing sysctls from the debugger is currently unsupported */
2704 	return (EPERM);
2705 }
2706 
2707 /*
2708  * Run a sysctl handler with the DDB oldfunc and newfunc attached.
2709  * Instead of copying any output to a buffer we'll dump it right to
2710  * the console.
2711  */
2712 static int
2713 db_sysctl(struct sysctl_oid *oidp, int *name, u_int namelen,
2714     void *old, size_t *oldlenp, size_t *retval, int flags)
2715 {
2716 	struct sysctl_req req;
2717 	int error;
2718 
2719 	/* Setup the request */
2720 	bzero(&req, sizeof req);
2721 	req.td = kdb_thread;
2722 	req.oldfunc = sysctl_old_ddb;
2723 	req.newfunc = sysctl_new_ddb;
2724 	req.lock = REQ_UNWIRED;
2725 	if (oldlenp) {
2726 		req.oldlen = *oldlenp;
2727 	}
2728 	req.validlen = req.oldlen;
2729 	if (old) {
2730 		req.oldptr = old;
2731 	}
2732 
2733 	/* Setup our globals for sysctl_old_ddb */
2734 	g_ddb_oid = oidp;
2735 	g_ddb_sysctl_flags = flags;
2736 	g_ddb_sysctl_printed = 0;
2737 
2738 	error = sysctl_root(0, name, namelen, &req);
2739 
2740 	/* Reset globals */
2741 	g_ddb_oid = NULL;
2742 	g_ddb_sysctl_flags = 0;
2743 
2744 	if (retval) {
2745 		if (req.oldptr && req.oldidx > req.validlen)
2746 			*retval = req.validlen;
2747 		else
2748 			*retval = req.oldidx;
2749 	}
2750 	return (error);
2751 }
2752 
2753 /*
2754  * Show a sysctl's name
2755  */
2756 static void
2757 db_show_oid_name(int *oid, size_t nlen)
2758 {
2759 	struct sysctl_oid *oidp;
2760 	int qoid[CTL_MAXNAME + 2];
2761 	int error;
2762 
2763 	qoid[0] = CTL_SYSCTL;
2764 	qoid[1] = CTL_SYSCTL_NAME;
2765 	memcpy(qoid + 2, oid, nlen * sizeof(int));
2766 
2767 	error = sysctl_find_oid(qoid, nlen + 2, &oidp, NULL, NULL);
2768 	if (error)
2769 		db_error("sysctl name oid");
2770 
2771 	error = db_sysctl(oidp, qoid, nlen + 2, NULL, NULL, NULL, 0);
2772 	if (error)
2773 		db_error("sysctl name");
2774 }
2775 
2776 /*
2777  * Check to see if an OID is safe to print from ddb.
2778  */
2779 static bool
2780 db_oid_safe(const struct sysctl_oid *oidp)
2781 {
2782 	for (unsigned int i = 0; i < nitems(db_safe_handlers); ++i) {
2783 		if (oidp->oid_handler == db_safe_handlers[i])
2784 			return (true);
2785 	}
2786 
2787 	return (false);
2788 }
2789 
2790 /*
2791  * Show a sysctl at a specific OID
2792  * Compare to the input handling in show_var from sbin/sysctl/sysctl.c
2793  */
2794 static int
2795 db_show_oid(struct sysctl_oid *oidp, int *oid, size_t nlen, int flags)
2796 {
2797 	int error, xflag, oflag, Nflag, nflag;
2798 	size_t len;
2799 
2800 	xflag = flags & DB_SYSCTL_HEX;
2801 	oflag = flags & DB_SYSCTL_OPAQUE;
2802 	nflag = flags & DB_SYSCTL_VALUE_ONLY;
2803 	Nflag = flags & DB_SYSCTL_NAME_ONLY;
2804 
2805 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_OPAQUE &&
2806 	    (!xflag && !oflag))
2807 		return (0);
2808 
2809 	if (Nflag) {
2810 		db_show_oid_name(oid, nlen);
2811 		error = 0;
2812 		goto out;
2813 	}
2814 
2815 	if (!nflag) {
2816 		db_show_oid_name(oid, nlen);
2817 		db_printf(": ");
2818 	}
2819 
2820 	if ((flags & DB_SYSCTL_SAFE_ONLY) && !db_oid_safe(oidp)) {
2821 		db_printf("Skipping, unsafe to print while recursing.");
2822 		error = 0;
2823 		goto out;
2824 	}
2825 
2826 	/* Try once, and ask about the size */
2827 	len = 0;
2828 	error = db_sysctl(oidp, oid, nlen,
2829 	    NULL, NULL, &len, flags);
2830 	if (error)
2831 		goto out;
2832 
2833 	if (!g_ddb_sysctl_printed)
2834 		/* Lie about the size */
2835 		error = db_sysctl(oidp, oid, nlen,
2836 		    (void *) 1, &len, NULL, flags);
2837 
2838 out:
2839 	db_printf("\n");
2840 	return (error);
2841 }
2842 
2843 /*
2844  * Show all sysctls under a specific OID
2845  * Compare to sysctl_all from sbin/sysctl/sysctl.c
2846  */
2847 static int
2848 db_show_sysctl_all(int *oid, size_t len, int flags)
2849 {
2850 	struct sysctl_oid *oidp;
2851 	int qoid[CTL_MAXNAME + 2], next[CTL_MAXNAME];
2852 	size_t nlen;
2853 
2854 	qoid[0] = CTL_SYSCTL;
2855 	qoid[1] = CTL_SYSCTL_NEXT;
2856 	if (len) {
2857 		nlen = len;
2858 		memcpy(&qoid[2], oid, nlen * sizeof(int));
2859 	} else {
2860 		nlen = 1;
2861 		qoid[2] = CTL_KERN;
2862 	}
2863 	for (;;) {
2864 		int error;
2865 		size_t nextsize = sizeof(next);
2866 
2867 		error = kernel_sysctl(kdb_thread, qoid, nlen + 2,
2868 		    next, &nextsize, NULL, 0, &nlen, 0);
2869 		if (error != 0) {
2870 			if (error == ENOENT)
2871 				return (0);
2872 			else
2873 				db_error("sysctl(next)");
2874 		}
2875 
2876 		nlen /= sizeof(int);
2877 
2878 		if (nlen < (unsigned int)len)
2879 			return (0);
2880 
2881 		if (memcmp(&oid[0], &next[0], len * sizeof(int)) != 0)
2882 			return (0);
2883 
2884 		/* Find the OID in question */
2885 		error = sysctl_find_oid(next, nlen, &oidp, NULL, NULL);
2886 		if (error)
2887 			return (error);
2888 
2889 		(void)db_show_oid(oidp, next, nlen, flags | DB_SYSCTL_SAFE_ONLY);
2890 
2891 		if (db_pager_quit)
2892 			return (0);
2893 
2894 		memcpy(&qoid[2 + len], &next[len], (nlen - len) * sizeof(int));
2895 	}
2896 }
2897 
2898 /*
2899  * Show a sysctl by its user facing string
2900  */
2901 static int
2902 db_sysctlbyname(char *name, int flags)
2903 {
2904 	struct sysctl_oid *oidp;
2905 	int oid[CTL_MAXNAME];
2906 	int error, nlen;
2907 
2908 	error = name2oid(name, oid, &nlen, &oidp);
2909 	if (error) {
2910 		return (error);
2911 	}
2912 
2913 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
2914 		db_show_sysctl_all(oid, nlen, flags);
2915 	} else {
2916 		error = db_show_oid(oidp, oid, nlen, flags);
2917 	}
2918 
2919 	return (error);
2920 }
2921 
2922 static void
2923 db_sysctl_cmd_usage(void)
2924 {
2925 	db_printf(
2926 	    " sysctl [/Nnox] <sysctl>					    \n"
2927 	    "								    \n"
2928 	    " <sysctl> The name of the sysctl to show.			    \n"
2929 	    "								    \n"
2930 	    " Show a sysctl by hooking into SYSCTL_IN and SYSCTL_OUT.	    \n"
2931 	    " This will work for most sysctls, but should not be used	    \n"
2932 	    " with sysctls that are known to malloc.			    \n"
2933 	    "								    \n"
2934 	    " While recursing any \"unsafe\" sysctls will be skipped.	    \n"
2935 	    " Call sysctl directly on the sysctl to try printing the	    \n"
2936 	    " skipped sysctl. This is unsafe and may make the ddb	    \n"
2937 	    " session unusable.						    \n"
2938 	    "								    \n"
2939 	    " Arguments:						    \n"
2940 	    "	/N	Display only the name of the sysctl.		    \n"
2941 	    "	/n	Display only the value of the sysctl.		    \n"
2942 	    "	/o	Display opaque values.				    \n"
2943 	    "	/x	Display the sysctl in hex.			    \n"
2944 	    "								    \n"
2945 	    "For example:						    \n"
2946 	    "sysctl vm.v_free_min					    \n"
2947 	    "vn.v_free_min: 12669					    \n"
2948 	    );
2949 }
2950 
2951 /*
2952  * Show a specific sysctl similar to sysctl (8).
2953  */
2954 DB_COMMAND_FLAGS(sysctl, db_sysctl_cmd, CS_OWN)
2955 {
2956 	char name[TOK_STRING_SIZE];
2957 	int error, i, t, flags;
2958 
2959 	/* Parse the modifiers */
2960 	t = db_read_token();
2961 	if (t == tSLASH || t == tMINUS) {
2962 		t = db_read_token();
2963 		if (t != tIDENT) {
2964 			db_printf("Bad modifier\n");
2965 			error = EINVAL;
2966 			goto out;
2967 		}
2968 		db_strcpy(modif, db_tok_string);
2969 	}
2970 	else {
2971 		db_unread_token(t);
2972 		modif[0] = '\0';
2973 	}
2974 
2975 	flags = 0;
2976 	for (i = 0; i < nitems(db_sysctl_modifs); i++) {
2977 		if (strchr(modif, db_sysctl_modifs[i])) {
2978 			flags |= db_sysctl_modif_values[i];
2979 		}
2980 	}
2981 
2982 	/* Parse the sysctl names */
2983 	t = db_read_token();
2984 	if (t != tIDENT) {
2985 		db_printf("Need sysctl name\n");
2986 		error = EINVAL;
2987 		goto out;
2988 	}
2989 
2990 	/* Copy the name into a temporary buffer */
2991 	db_strcpy(name, db_tok_string);
2992 
2993 	/* Ensure there is no trailing cruft */
2994 	t = db_read_token();
2995 	if (t != tEOL) {
2996 		db_printf("Unexpected sysctl argument\n");
2997 		error = EINVAL;
2998 		goto out;
2999 	}
3000 
3001 	error = db_sysctlbyname(name, flags);
3002 	if (error == ENOENT) {
3003 		db_printf("unknown oid: '%s'\n", db_tok_string);
3004 		goto out;
3005 	} else if (error) {
3006 		db_printf("%s: error: %d\n", db_tok_string, error);
3007 		goto out;
3008 	}
3009 
3010 out:
3011 	/* Ensure we eat all of our text */
3012 	db_flush_lex();
3013 
3014 	if (error == EINVAL) {
3015 		db_sysctl_cmd_usage();
3016 	}
3017 }
3018 
3019 #endif /* DDB */
3020