xref: /freebsd/sys/kern/kern_sysctl.c (revision 84823cc70824c8d842f503d8c2e6d7b0c2d95b61)
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 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 	SLIST_FOREACH(oidp, list, oid_link) {
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 and be NULL initialized.
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 	while (indx < CTL_MAXNAME && indx >= 0) {
358 		if (nodes[indx] == NULL && indx == 0)
359 			nodes[indx] = SLIST_FIRST(&sysctl__children);
360 		else if (nodes[indx] == NULL)
361 			nodes[indx] = SLIST_FIRST(&nodes[indx - 1]->oid_children);
362 		else
363 			nodes[indx] = SLIST_NEXT(nodes[indx], oid_link);
364 
365 		if (nodes[indx] == needle)
366 			return (indx + 1);
367 
368 		if (nodes[indx] == NULL) {
369 			indx--;
370 			continue;
371 		}
372 
373 		if ((nodes[indx]->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
374 			indx++;
375 			continue;
376 		}
377 	}
378 	return (-1);
379 }
380 
381 static void
382 sysctl_warn_reuse(const char *func, struct sysctl_oid *leaf)
383 {
384 	struct sysctl_oid *nodes[CTL_MAXNAME];
385 	char buf[128];
386 	struct sbuf sb;
387 	int rc, i;
388 
389 	(void)sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN | SBUF_INCLUDENUL);
390 	sbuf_set_drain(&sb, sbuf_printf_drain, NULL);
391 
392 	sbuf_printf(&sb, "%s: can't re-use a leaf (", __func__);
393 
394 	memset(nodes, 0, sizeof(nodes));
395 	rc = sysctl_search_oid(nodes, leaf);
396 	if (rc > 0) {
397 		for (i = 0; i < rc; i++)
398 			sbuf_printf(&sb, "%s%.*s", nodes[i]->oid_name,
399 			    i != (rc - 1), ".");
400 	} else {
401 		sbuf_printf(&sb, "%s", leaf->oid_name);
402 	}
403 	sbuf_printf(&sb, ")!\n");
404 
405 	(void)sbuf_finish(&sb);
406 }
407 
408 #ifdef SYSCTL_DEBUG
409 static int
410 sysctl_reuse_test(SYSCTL_HANDLER_ARGS)
411 {
412 	struct rm_priotracker tracker;
413 
414 	SYSCTL_RLOCK(&tracker);
415 	sysctl_warn_reuse(__func__, oidp);
416 	SYSCTL_RUNLOCK(&tracker);
417 	return (0);
418 }
419 SYSCTL_PROC(_sysctl, OID_AUTO, reuse_test,
420     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 0, sysctl_reuse_test, "-",
421     "");
422 #endif
423 
424 void
425 sysctl_register_oid(struct sysctl_oid *oidp)
426 {
427 	struct sysctl_oid_list *parent = oidp->oid_parent;
428 	struct sysctl_oid *p;
429 	struct sysctl_oid *q;
430 	int oid_number;
431 	int timeout = 2;
432 
433 	/*
434 	 * First check if another oid with the same name already
435 	 * exists in the parent's list.
436 	 */
437 	SYSCTL_ASSERT_WLOCKED();
438 	p = sysctl_find_oidname(oidp->oid_name, parent);
439 	if (p != NULL) {
440 		if ((p->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
441 			p->oid_refcnt++;
442 			return;
443 		} else {
444 			sysctl_warn_reuse(__func__, p);
445 			return;
446 		}
447 	}
448 	/* get current OID number */
449 	oid_number = oidp->oid_number;
450 
451 #if (OID_AUTO >= 0)
452 #error "OID_AUTO is expected to be a negative value"
453 #endif
454 	/*
455 	 * Any negative OID number qualifies as OID_AUTO. Valid OID
456 	 * numbers should always be positive.
457 	 *
458 	 * NOTE: DO NOT change the starting value here, change it in
459 	 * <sys/sysctl.h>, and make sure it is at least 256 to
460 	 * accommodate e.g. net.inet.raw as a static sysctl node.
461 	 */
462 	if (oid_number < 0) {
463 		static int newoid;
464 
465 		/*
466 		 * By decrementing the next OID number we spend less
467 		 * time inserting the OIDs into a sorted list.
468 		 */
469 		if (--newoid < CTL_AUTO_START)
470 			newoid = 0x7fffffff;
471 
472 		oid_number = newoid;
473 	}
474 
475 	/*
476 	 * Insert the OID into the parent's list sorted by OID number.
477 	 */
478 retry:
479 	q = NULL;
480 	SLIST_FOREACH(p, parent, oid_link) {
481 		/* check if the current OID number is in use */
482 		if (oid_number == p->oid_number) {
483 			/* get the next valid OID number */
484 			if (oid_number < CTL_AUTO_START ||
485 			    oid_number == 0x7fffffff) {
486 				/* wraparound - restart */
487 				oid_number = CTL_AUTO_START;
488 				/* don't loop forever */
489 				if (!timeout--)
490 					panic("sysctl: Out of OID numbers\n");
491 				goto retry;
492 			} else {
493 				oid_number++;
494 			}
495 		} else if (oid_number < p->oid_number)
496 			break;
497 		q = p;
498 	}
499 	/* check for non-auto OID number collision */
500 	if (oidp->oid_number >= 0 && oidp->oid_number < CTL_AUTO_START &&
501 	    oid_number >= CTL_AUTO_START) {
502 		printf("sysctl: OID number(%d) is already in use for '%s'\n",
503 		    oidp->oid_number, oidp->oid_name);
504 	}
505 	/* update the OID number, if any */
506 	oidp->oid_number = oid_number;
507 	if (q != NULL)
508 		SLIST_INSERT_AFTER(q, oidp, oid_link);
509 	else
510 		SLIST_INSERT_HEAD(parent, oidp, oid_link);
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 	struct sysctl_oid *p;
560 	int error;
561 
562 	SYSCTL_ASSERT_WLOCKED();
563 	if (oidp->oid_number == OID_AUTO) {
564 		error = EINVAL;
565 	} else {
566 		error = ENOENT;
567 		SLIST_FOREACH(p, oidp->oid_parent, oid_link) {
568 			if (p == oidp) {
569 				SLIST_REMOVE(oidp->oid_parent, oidp,
570 				    sysctl_oid, oid_link);
571 				error = 0;
572 				break;
573 			}
574 		}
575 	}
576 
577 	/*
578 	 * This can happen when a module fails to register and is
579 	 * being unloaded afterwards.  It should not be a panic()
580 	 * for normal use.
581 	 */
582 	if (error) {
583 		printf("%s: failed(%d) to unregister sysctl(%s)\n",
584 		    __func__, error, oidp->oid_name);
585 	}
586 }
587 
588 /* Initialize a new context to keep track of dynamically added sysctls. */
589 int
590 sysctl_ctx_init(struct sysctl_ctx_list *c)
591 {
592 
593 	if (c == NULL) {
594 		return (EINVAL);
595 	}
596 
597 	/*
598 	 * No locking here, the caller is responsible for not adding
599 	 * new nodes to a context until after this function has
600 	 * returned.
601 	 */
602 	TAILQ_INIT(c);
603 	return (0);
604 }
605 
606 /* Free the context, and destroy all dynamic oids registered in this context */
607 int
608 sysctl_ctx_free(struct sysctl_ctx_list *clist)
609 {
610 	struct sysctl_ctx_entry *e, *e1;
611 	int error;
612 
613 	error = 0;
614 	/*
615 	 * First perform a "dry run" to check if it's ok to remove oids.
616 	 * XXX FIXME
617 	 * XXX This algorithm is a hack. But I don't know any
618 	 * XXX better solution for now...
619 	 */
620 	SYSCTL_WLOCK();
621 	TAILQ_FOREACH(e, clist, link) {
622 		error = sysctl_remove_oid_locked(e->entry, 0, 0);
623 		if (error)
624 			break;
625 	}
626 	/*
627 	 * Restore deregistered entries, either from the end,
628 	 * or from the place where error occurred.
629 	 * e contains the entry that was not unregistered
630 	 */
631 	if (error)
632 		e1 = TAILQ_PREV(e, sysctl_ctx_list, link);
633 	else
634 		e1 = TAILQ_LAST(clist, sysctl_ctx_list);
635 	while (e1 != NULL) {
636 		sysctl_register_oid(e1->entry);
637 		e1 = TAILQ_PREV(e1, sysctl_ctx_list, link);
638 	}
639 	if (error) {
640 		SYSCTL_WUNLOCK();
641 		return(EBUSY);
642 	}
643 	/* Now really delete the entries */
644 	e = TAILQ_FIRST(clist);
645 	while (e != NULL) {
646 		e1 = TAILQ_NEXT(e, link);
647 		error = sysctl_remove_oid_locked(e->entry, 1, 0);
648 		if (error)
649 			panic("sysctl_remove_oid: corrupt tree, entry: %s",
650 			    e->entry->oid_name);
651 		free(e, M_SYSCTLOID);
652 		e = e1;
653 	}
654 	SYSCTL_WUNLOCK();
655 	return (error);
656 }
657 
658 /* Add an entry to the context */
659 struct sysctl_ctx_entry *
660 sysctl_ctx_entry_add(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
661 {
662 	struct sysctl_ctx_entry *e;
663 
664 	SYSCTL_ASSERT_WLOCKED();
665 	if (clist == NULL || oidp == NULL)
666 		return(NULL);
667 	e = malloc(sizeof(struct sysctl_ctx_entry), M_SYSCTLOID, M_WAITOK);
668 	e->entry = oidp;
669 	TAILQ_INSERT_HEAD(clist, e, link);
670 	return (e);
671 }
672 
673 /* Find an entry in the context */
674 struct sysctl_ctx_entry *
675 sysctl_ctx_entry_find(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
676 {
677 	struct sysctl_ctx_entry *e;
678 
679 	SYSCTL_ASSERT_WLOCKED();
680 	if (clist == NULL || oidp == NULL)
681 		return(NULL);
682 	TAILQ_FOREACH(e, clist, link) {
683 		if (e->entry == oidp)
684 			return(e);
685 	}
686 	return (e);
687 }
688 
689 /*
690  * Delete an entry from the context.
691  * NOTE: this function doesn't free oidp! You have to remove it
692  * with sysctl_remove_oid().
693  */
694 int
695 sysctl_ctx_entry_del(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
696 {
697 	struct sysctl_ctx_entry *e;
698 
699 	if (clist == NULL || oidp == NULL)
700 		return (EINVAL);
701 	SYSCTL_WLOCK();
702 	e = sysctl_ctx_entry_find(clist, oidp);
703 	if (e != NULL) {
704 		TAILQ_REMOVE(clist, e, link);
705 		SYSCTL_WUNLOCK();
706 		free(e, M_SYSCTLOID);
707 		return (0);
708 	} else {
709 		SYSCTL_WUNLOCK();
710 		return (ENOENT);
711 	}
712 }
713 
714 /*
715  * Remove dynamically created sysctl trees.
716  * oidp - top of the tree to be removed
717  * del - if 0 - just deregister, otherwise free up entries as well
718  * recurse - if != 0 traverse the subtree to be deleted
719  */
720 int
721 sysctl_remove_oid(struct sysctl_oid *oidp, int del, int recurse)
722 {
723 	int error;
724 
725 	SYSCTL_WLOCK();
726 	error = sysctl_remove_oid_locked(oidp, del, recurse);
727 	SYSCTL_WUNLOCK();
728 	return (error);
729 }
730 
731 int
732 sysctl_remove_name(struct sysctl_oid *parent, const char *name,
733     int del, int recurse)
734 {
735 	struct sysctl_oid *p, *tmp;
736 	int error;
737 
738 	error = ENOENT;
739 	SYSCTL_WLOCK();
740 	SLIST_FOREACH_SAFE(p, SYSCTL_CHILDREN(parent), oid_link, tmp) {
741 		if (strcmp(p->oid_name, name) == 0) {
742 			error = sysctl_remove_oid_locked(p, del, recurse);
743 			break;
744 		}
745 	}
746 	SYSCTL_WUNLOCK();
747 
748 	return (error);
749 }
750 
751 /*
752  * Duplicate the provided string, escaping any illegal characters.  The result
753  * must be freed when no longer in use.
754  *
755  * The list of illegal characters is ".".
756  */
757 static char*
758 sysctl_escape_name(const char* orig)
759 {
760 	int i, s = 0, d = 0, nillegals = 0;
761 	char *new;
762 
763 	/* First count the number of illegal characters */
764 	for (i = 0; orig[i] != '\0'; i++) {
765 		if (orig[i] == '.')
766 			nillegals++;
767 	}
768 
769 	/* Allocate storage for new string */
770 	new = malloc(i + 2 * nillegals + 1, M_SYSCTLOID, M_WAITOK);
771 
772 	/* Copy the name, escaping characters as we go */
773 	while (orig[s] != '\0') {
774 		if (orig[s] == '.') {
775 			/* %25 is the hexadecimal representation of '.' */
776 			new[d++] = '%';
777 			new[d++] = '2';
778 			new[d++] = '5';
779 			s++;
780 		} else {
781 			new[d++] = orig[s++];
782 		}
783 	}
784 
785 	/* Finally, nul-terminate */
786 	new[d] = '\0';
787 
788 	return (new);
789 }
790 
791 static int
792 sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del, int recurse)
793 {
794 	struct sysctl_oid *p, *tmp;
795 	int error;
796 
797 	SYSCTL_ASSERT_WLOCKED();
798 	if (oidp == NULL)
799 		return(EINVAL);
800 	if ((oidp->oid_kind & CTLFLAG_DYN) == 0) {
801 		printf("Warning: can't remove non-dynamic nodes (%s)!\n",
802 		    oidp->oid_name);
803 		return (EINVAL);
804 	}
805 	/*
806 	 * WARNING: normal method to do this should be through
807 	 * sysctl_ctx_free(). Use recursing as the last resort
808 	 * method to purge your sysctl tree of leftovers...
809 	 * However, if some other code still references these nodes,
810 	 * it will panic.
811 	 */
812 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
813 		if (oidp->oid_refcnt == 1) {
814 			SLIST_FOREACH_SAFE(p,
815 			    SYSCTL_CHILDREN(oidp), oid_link, tmp) {
816 				if (!recurse) {
817 					printf("Warning: failed attempt to "
818 					    "remove oid %s with child %s\n",
819 					    oidp->oid_name, p->oid_name);
820 					return (ENOTEMPTY);
821 				}
822 				error = sysctl_remove_oid_locked(p, del,
823 				    recurse);
824 				if (error)
825 					return (error);
826 			}
827 		}
828 	}
829 	if (oidp->oid_refcnt > 1 ) {
830 		oidp->oid_refcnt--;
831 	} else {
832 		if (oidp->oid_refcnt == 0) {
833 			printf("Warning: bad oid_refcnt=%u (%s)!\n",
834 				oidp->oid_refcnt, oidp->oid_name);
835 			return (EINVAL);
836 		}
837 		sysctl_unregister_oid(oidp);
838 		if (del) {
839 			/*
840 			 * Wait for all threads running the handler to drain.
841 			 * This preserves the previous behavior when the
842 			 * sysctl lock was held across a handler invocation,
843 			 * and is necessary for module unload correctness.
844 			 */
845 			while (oidp->oid_running > 0) {
846 				oidp->oid_kind |= CTLFLAG_DYING;
847 				SYSCTL_SLEEP(&oidp->oid_running, "oidrm", 0);
848 			}
849 			if (oidp->oid_descr)
850 				free(__DECONST(char *, oidp->oid_descr),
851 				    M_SYSCTLOID);
852 			if (oidp->oid_label)
853 				free(__DECONST(char *, oidp->oid_label),
854 				    M_SYSCTLOID);
855 			free(__DECONST(char *, oidp->oid_name), M_SYSCTLOID);
856 			free(oidp, M_SYSCTLOID);
857 		}
858 	}
859 	return (0);
860 }
861 /*
862  * Create new sysctls at run time.
863  * clist may point to a valid context initialized with sysctl_ctx_init().
864  */
865 struct sysctl_oid *
866 sysctl_add_oid(struct sysctl_ctx_list *clist, struct sysctl_oid_list *parent,
867 	int number, const char *name, int kind, void *arg1, intmax_t arg2,
868 	int (*handler)(SYSCTL_HANDLER_ARGS), const char *fmt, const char *descr,
869 	const char *label)
870 {
871 	struct sysctl_oid *oidp;
872 	char *escaped;
873 
874 	/* You have to hook up somewhere.. */
875 	if (parent == NULL)
876 		return(NULL);
877 	escaped = sysctl_escape_name(name);
878 	/* Check if the node already exists, otherwise create it */
879 	SYSCTL_WLOCK();
880 	oidp = sysctl_find_oidname(escaped, parent);
881 	if (oidp != NULL) {
882 		free(escaped, M_SYSCTLOID);
883 		if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
884 			oidp->oid_refcnt++;
885 			/* Update the context */
886 			if (clist != NULL)
887 				sysctl_ctx_entry_add(clist, oidp);
888 			SYSCTL_WUNLOCK();
889 			return (oidp);
890 		} else {
891 			sysctl_warn_reuse(__func__, oidp);
892 			SYSCTL_WUNLOCK();
893 			return (NULL);
894 		}
895 	}
896 	oidp = malloc(sizeof(struct sysctl_oid), M_SYSCTLOID, M_WAITOK|M_ZERO);
897 	oidp->oid_parent = parent;
898 	SLIST_INIT(&oidp->oid_children);
899 	oidp->oid_number = number;
900 	oidp->oid_refcnt = 1;
901 	oidp->oid_name = escaped;
902 	oidp->oid_handler = handler;
903 	oidp->oid_kind = CTLFLAG_DYN | kind;
904 	oidp->oid_arg1 = arg1;
905 	oidp->oid_arg2 = arg2;
906 	oidp->oid_fmt = fmt;
907 	if (descr != NULL)
908 		oidp->oid_descr = strdup(descr, M_SYSCTLOID);
909 	if (label != NULL)
910 		oidp->oid_label = strdup(label, M_SYSCTLOID);
911 	/* Update the context, if used */
912 	if (clist != NULL)
913 		sysctl_ctx_entry_add(clist, oidp);
914 	/* Register this oid */
915 	sysctl_register_oid(oidp);
916 	SYSCTL_WUNLOCK();
917 	return (oidp);
918 }
919 
920 /*
921  * Rename an existing oid.
922  */
923 void
924 sysctl_rename_oid(struct sysctl_oid *oidp, const char *name)
925 {
926 	char *newname;
927 	char *oldname;
928 
929 	newname = strdup(name, M_SYSCTLOID);
930 	SYSCTL_WLOCK();
931 	oldname = __DECONST(char *, oidp->oid_name);
932 	oidp->oid_name = newname;
933 	SYSCTL_WUNLOCK();
934 	free(oldname, M_SYSCTLOID);
935 }
936 
937 /*
938  * Reparent an existing oid.
939  */
940 int
941 sysctl_move_oid(struct sysctl_oid *oid, struct sysctl_oid_list *parent)
942 {
943 	struct sysctl_oid *oidp;
944 
945 	SYSCTL_WLOCK();
946 	if (oid->oid_parent == parent) {
947 		SYSCTL_WUNLOCK();
948 		return (0);
949 	}
950 	oidp = sysctl_find_oidname(oid->oid_name, parent);
951 	if (oidp != NULL) {
952 		SYSCTL_WUNLOCK();
953 		return (EEXIST);
954 	}
955 	sysctl_unregister_oid(oid);
956 	oid->oid_parent = parent;
957 	oid->oid_number = OID_AUTO;
958 	sysctl_register_oid(oid);
959 	SYSCTL_WUNLOCK();
960 	return (0);
961 }
962 
963 /*
964  * Register the kernel's oids on startup.
965  */
966 SET_DECLARE(sysctl_set, struct sysctl_oid);
967 
968 static void
969 sysctl_register_all(void *arg)
970 {
971 	struct sysctl_oid **oidp;
972 
973 	sx_init(&sysctlmemlock, "sysctl mem");
974 	sx_init(&sysctlstringlock, "sysctl string handler");
975 	SYSCTL_INIT();
976 	SYSCTL_WLOCK();
977 	SET_FOREACH(oidp, sysctl_set)
978 		sysctl_register_oid(*oidp);
979 	SYSCTL_WUNLOCK();
980 }
981 SYSINIT(sysctl, SI_SUB_KMEM, SI_ORDER_FIRST, sysctl_register_all, NULL);
982 
983 /*
984  * "Staff-functions"
985  *
986  * These functions implement a presently undocumented interface
987  * used by the sysctl program to walk the tree, and get the type
988  * so it can print the value.
989  * This interface is under work and consideration, and should probably
990  * be killed with a big axe by the first person who can find the time.
991  * (be aware though, that the proper interface isn't as obvious as it
992  * may seem, there are various conflicting requirements.
993  *
994  * {CTL_SYSCTL, CTL_SYSCTL_DEBUG}		printf the entire MIB-tree.
995  * {CTL_SYSCTL, CTL_SYSCTL_NAME, ...}		return the name of the "..."
996  *						OID.
997  * {CTL_SYSCTL, CTL_SYSCTL_NEXT, ...}		return the next OID, honoring
998  *						CTLFLAG_SKIP.
999  * {CTL_SYSCTL, CTL_SYSCTL_NAME2OID}		return the OID of the name in
1000  *						"new"
1001  * {CTL_SYSCTL, CTL_SYSCTL_OIDFMT, ...}		return the kind & format info
1002  *						for the "..." OID.
1003  * {CTL_SYSCTL, CTL_SYSCTL_OIDDESCR, ...}	return the description of the
1004  *						"..." OID.
1005  * {CTL_SYSCTL, CTL_SYSCTL_OIDLABEL, ...}	return the aggregation label of
1006  *						the "..." OID.
1007  * {CTL_SYSCTL, CTL_SYSCTL_NEXTNOSKIP, ...}	return the next OID, ignoring
1008  *						CTLFLAG_SKIP.
1009  */
1010 
1011 #ifdef SYSCTL_DEBUG
1012 static void
1013 sysctl_sysctl_debug_dump_node(struct sysctl_oid_list *l, int i)
1014 {
1015 	int k;
1016 	struct sysctl_oid *oidp;
1017 
1018 	SYSCTL_ASSERT_LOCKED();
1019 	SLIST_FOREACH(oidp, l, oid_link) {
1020 		for (k=0; k<i; k++)
1021 			printf(" ");
1022 
1023 		printf("%d %s ", oidp->oid_number, oidp->oid_name);
1024 
1025 		printf("%c%c",
1026 			oidp->oid_kind & CTLFLAG_RD ? 'R':' ',
1027 			oidp->oid_kind & CTLFLAG_WR ? 'W':' ');
1028 
1029 		if (oidp->oid_handler)
1030 			printf(" *Handler");
1031 
1032 		switch (oidp->oid_kind & CTLTYPE) {
1033 			case CTLTYPE_NODE:
1034 				printf(" Node\n");
1035 				if (!oidp->oid_handler) {
1036 					sysctl_sysctl_debug_dump_node(
1037 					    SYSCTL_CHILDREN(oidp), i + 2);
1038 				}
1039 				break;
1040 			case CTLTYPE_INT:    printf(" Int\n"); break;
1041 			case CTLTYPE_UINT:   printf(" u_int\n"); break;
1042 			case CTLTYPE_LONG:   printf(" Long\n"); break;
1043 			case CTLTYPE_ULONG:  printf(" u_long\n"); break;
1044 			case CTLTYPE_STRING: printf(" String\n"); break;
1045 			case CTLTYPE_S8:     printf(" int8_t\n"); break;
1046 			case CTLTYPE_S16:    printf(" int16_t\n"); break;
1047 			case CTLTYPE_S32:    printf(" int32_t\n"); break;
1048 			case CTLTYPE_S64:    printf(" int64_t\n"); break;
1049 			case CTLTYPE_U8:     printf(" uint8_t\n"); break;
1050 			case CTLTYPE_U16:    printf(" uint16_t\n"); break;
1051 			case CTLTYPE_U32:    printf(" uint32_t\n"); break;
1052 			case CTLTYPE_U64:    printf(" uint64_t\n"); break;
1053 			case CTLTYPE_OPAQUE: printf(" Opaque/struct\n"); break;
1054 			default:	     printf("\n");
1055 		}
1056 	}
1057 }
1058 
1059 static int
1060 sysctl_sysctl_debug(SYSCTL_HANDLER_ARGS)
1061 {
1062 	struct rm_priotracker tracker;
1063 	int error;
1064 
1065 	error = priv_check(req->td, PRIV_SYSCTL_DEBUG);
1066 	if (error)
1067 		return (error);
1068 	SYSCTL_RLOCK(&tracker);
1069 	sysctl_sysctl_debug_dump_node(&sysctl__children, 0);
1070 	SYSCTL_RUNLOCK(&tracker);
1071 	return (ENOENT);
1072 }
1073 
1074 SYSCTL_PROC(_sysctl, CTL_SYSCTL_DEBUG, debug, CTLTYPE_STRING | CTLFLAG_RD |
1075     CTLFLAG_MPSAFE, 0, 0, sysctl_sysctl_debug, "-", "");
1076 #endif
1077 
1078 static int
1079 sysctl_sysctl_name(SYSCTL_HANDLER_ARGS)
1080 {
1081 	int *name = (int *) arg1;
1082 	u_int namelen = arg2;
1083 	int error;
1084 	struct sysctl_oid *oid;
1085 	struct sysctl_oid_list *lsp = &sysctl__children, *lsp2;
1086 	struct rm_priotracker tracker;
1087 	char buf[10];
1088 
1089 	error = sysctl_wire_old_buffer(req, 0);
1090 	if (error)
1091 		return (error);
1092 
1093 	SYSCTL_RLOCK(&tracker);
1094 	while (namelen) {
1095 		if (!lsp) {
1096 			snprintf(buf,sizeof(buf),"%d",*name);
1097 			if (req->oldidx)
1098 				error = SYSCTL_OUT(req, ".", 1);
1099 			if (!error)
1100 				error = SYSCTL_OUT(req, buf, strlen(buf));
1101 			if (error)
1102 				goto out;
1103 			namelen--;
1104 			name++;
1105 			continue;
1106 		}
1107 		lsp2 = NULL;
1108 		SLIST_FOREACH(oid, lsp, oid_link) {
1109 			if (oid->oid_number != *name)
1110 				continue;
1111 
1112 			if (req->oldidx)
1113 				error = SYSCTL_OUT(req, ".", 1);
1114 			if (!error)
1115 				error = SYSCTL_OUT(req, oid->oid_name,
1116 					strlen(oid->oid_name));
1117 			if (error)
1118 				goto out;
1119 
1120 			namelen--;
1121 			name++;
1122 
1123 			if ((oid->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1124 				break;
1125 
1126 			if (oid->oid_handler)
1127 				break;
1128 
1129 			lsp2 = SYSCTL_CHILDREN(oid);
1130 			break;
1131 		}
1132 		lsp = lsp2;
1133 	}
1134 	error = SYSCTL_OUT(req, "", 1);
1135  out:
1136 	SYSCTL_RUNLOCK(&tracker);
1137 	return (error);
1138 }
1139 
1140 /*
1141  * XXXRW/JA: Shouldn't return name data for nodes that we don't permit in
1142  * capability mode.
1143  */
1144 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_NAME, name, CTLFLAG_RD |
1145     CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_name, "");
1146 
1147 enum sysctl_iter_action {
1148 	ITER_SIBLINGS,	/* Not matched, continue iterating siblings */
1149 	ITER_CHILDREN,	/* Node has children we need to iterate over them */
1150 	ITER_FOUND,	/* Matching node was found */
1151 };
1152 
1153 /*
1154  * Tries to find the next node for @name and @namelen.
1155  *
1156  * Returns next action to take.
1157  */
1158 static enum sysctl_iter_action
1159 sysctl_sysctl_next_node(struct sysctl_oid *oidp, int *name, unsigned int namelen,
1160     bool honor_skip)
1161 {
1162 
1163 	if ((oidp->oid_kind & CTLFLAG_DORMANT) != 0)
1164 		return (ITER_SIBLINGS);
1165 
1166 	if (honor_skip && (oidp->oid_kind & CTLFLAG_SKIP) != 0)
1167 		return (ITER_SIBLINGS);
1168 
1169 	if (namelen == 0) {
1170 		/*
1171 		 * We have reached a node with a full name match and are
1172 		 * looking for the next oid in its children.
1173 		 *
1174 		 * For CTL_SYSCTL_NEXTNOSKIP we are done.
1175 		 *
1176 		 * For CTL_SYSCTL_NEXT we skip CTLTYPE_NODE (unless it
1177 		 * has a handler) and move on to the children.
1178 		 */
1179 		if (!honor_skip)
1180 			return (ITER_FOUND);
1181 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1182 			return (ITER_FOUND);
1183 		/* If node does not have an iterator, treat it as leaf */
1184 		if (oidp->oid_handler)
1185 			return (ITER_FOUND);
1186 
1187 		/* Report oid as a node to iterate */
1188 		return (ITER_CHILDREN);
1189 	}
1190 
1191 	/*
1192 	 * No match yet. Continue seeking the given name.
1193 	 *
1194 	 * We are iterating in order by oid_number, so skip oids lower
1195 	 * than the one we are looking for.
1196 	 *
1197 	 * When the current oid_number is higher than the one we seek,
1198 	 * that means we have reached the next oid in the sequence and
1199 	 * should return it.
1200 	 *
1201 	 * If the oid_number matches the name at this level then we
1202 	 * have to find a node to continue searching at the next level.
1203 	 */
1204 	if (oidp->oid_number < *name)
1205 		return (ITER_SIBLINGS);
1206 	if (oidp->oid_number > *name) {
1207 		/*
1208 		 * We have reached the next oid.
1209 		 *
1210 		 * For CTL_SYSCTL_NEXTNOSKIP we are done.
1211 		 *
1212 		 * For CTL_SYSCTL_NEXT we skip CTLTYPE_NODE (unless it
1213 		 * has a handler) and move on to the children.
1214 		 */
1215 		if (!honor_skip)
1216 			return (ITER_FOUND);
1217 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1218 			return (ITER_FOUND);
1219 		/* If node does not have an iterator, treat it as leaf */
1220 		if (oidp->oid_handler)
1221 			return (ITER_FOUND);
1222 		return (ITER_CHILDREN);
1223 	}
1224 
1225 	/* match at a current level */
1226 	if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1227 		return (ITER_SIBLINGS);
1228 	if (oidp->oid_handler)
1229 		return (ITER_SIBLINGS);
1230 
1231 	return (ITER_CHILDREN);
1232 }
1233 
1234 /*
1235  * Recursively walk the sysctl subtree at lsp until we find the given name.
1236  * Returns true and fills in next oid data in @next and @len if oid is found.
1237  */
1238 static bool
1239 sysctl_sysctl_next_action(struct sysctl_oid_list *lsp, int *name, u_int namelen,
1240     int *next, int *len, int level, bool honor_skip)
1241 {
1242 	struct sysctl_oid *oidp;
1243 	bool success = false;
1244 	enum sysctl_iter_action action;
1245 
1246 	SYSCTL_ASSERT_LOCKED();
1247 	SLIST_FOREACH(oidp, lsp, oid_link) {
1248 		action = sysctl_sysctl_next_node(oidp, name, namelen, honor_skip);
1249 		if (action == ITER_SIBLINGS)
1250 			continue;
1251 		if (action == ITER_FOUND) {
1252 			success = true;
1253 			break;
1254 		}
1255 		KASSERT((action== ITER_CHILDREN), ("ret(%d)!=ITER_CHILDREN", action));
1256 
1257 		lsp = SYSCTL_CHILDREN(oidp);
1258 		if (namelen == 0) {
1259 			success = sysctl_sysctl_next_action(lsp, NULL, 0,
1260 			    next + 1, len, level + 1, honor_skip);
1261 		} else {
1262 			success = sysctl_sysctl_next_action(lsp, name + 1, namelen - 1,
1263 			    next + 1, len, level + 1, honor_skip);
1264 			if (!success) {
1265 
1266 				/*
1267 				 * We maintain the invariant that current node oid
1268 				 * is >= the oid provided in @name.
1269 				 * As there are no usable children at this node,
1270 				 *  current node oid is strictly > than the requested
1271 				 *  oid.
1272 				 * Hence, reduce namelen to 0 to allow for picking first
1273 				 *  nodes/leafs in the next node in list.
1274 				 */
1275 				namelen = 0;
1276 			}
1277 		}
1278 		if (success)
1279 			break;
1280 	}
1281 
1282 	if (success) {
1283 		*next = oidp->oid_number;
1284 		if (level > *len)
1285 			*len = level;
1286 	}
1287 
1288 	return (success);
1289 }
1290 
1291 static int
1292 sysctl_sysctl_next(SYSCTL_HANDLER_ARGS)
1293 {
1294 	int *name = (int *) arg1;
1295 	u_int namelen = arg2;
1296 	int len, error;
1297 	bool success;
1298 	struct sysctl_oid_list *lsp = &sysctl__children;
1299 	struct rm_priotracker tracker;
1300 	int next[CTL_MAXNAME];
1301 
1302 	len = 0;
1303 	SYSCTL_RLOCK(&tracker);
1304 	success = sysctl_sysctl_next_action(lsp, name, namelen, next, &len, 1,
1305 	    oidp->oid_number == CTL_SYSCTL_NEXT);
1306 	SYSCTL_RUNLOCK(&tracker);
1307 	if (!success)
1308 		return (ENOENT);
1309 	error = SYSCTL_OUT(req, next, len * sizeof (int));
1310 	return (error);
1311 }
1312 
1313 /*
1314  * XXXRW/JA: Shouldn't return next data for nodes that we don't permit in
1315  * capability mode.
1316  */
1317 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_NEXT, next, CTLFLAG_RD |
1318     CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_next, "");
1319 
1320 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_NEXTNOSKIP, nextnoskip, CTLFLAG_RD |
1321     CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_next, "");
1322 
1323 static int
1324 name2oid(char *name, int *oid, int *len, struct sysctl_oid **oidpp)
1325 {
1326 	struct sysctl_oid *oidp;
1327 	struct sysctl_oid_list *lsp = &sysctl__children;
1328 	char *p;
1329 
1330 	SYSCTL_ASSERT_LOCKED();
1331 
1332 	for (*len = 0; *len < CTL_MAXNAME;) {
1333 		p = strsep(&name, ".");
1334 
1335 		oidp = SLIST_FIRST(lsp);
1336 		for (;; oidp = SLIST_NEXT(oidp, oid_link)) {
1337 			if (oidp == NULL)
1338 				return (ENOENT);
1339 			if (strcmp(p, oidp->oid_name) == 0)
1340 				break;
1341 		}
1342 		*oid++ = oidp->oid_number;
1343 		(*len)++;
1344 
1345 		if (name == NULL || *name == '\0') {
1346 			if (oidpp)
1347 				*oidpp = oidp;
1348 			return (0);
1349 		}
1350 
1351 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1352 			break;
1353 
1354 		if (oidp->oid_handler)
1355 			break;
1356 
1357 		lsp = SYSCTL_CHILDREN(oidp);
1358 	}
1359 	return (ENOENT);
1360 }
1361 
1362 static int
1363 sysctl_sysctl_name2oid(SYSCTL_HANDLER_ARGS)
1364 {
1365 	char *p;
1366 	int error, oid[CTL_MAXNAME], len = 0;
1367 	struct sysctl_oid *op = NULL;
1368 	struct rm_priotracker tracker;
1369 	char buf[32];
1370 
1371 	if (!req->newlen)
1372 		return (ENOENT);
1373 	if (req->newlen >= MAXPATHLEN)	/* XXX arbitrary, undocumented */
1374 		return (ENAMETOOLONG);
1375 
1376 	p = buf;
1377 	if (req->newlen >= sizeof(buf))
1378 		p = malloc(req->newlen+1, M_SYSCTL, M_WAITOK);
1379 
1380 	error = SYSCTL_IN(req, p, req->newlen);
1381 	if (error) {
1382 		if (p != buf)
1383 			free(p, M_SYSCTL);
1384 		return (error);
1385 	}
1386 
1387 	p [req->newlen] = '\0';
1388 
1389 	SYSCTL_RLOCK(&tracker);
1390 	error = name2oid(p, oid, &len, &op);
1391 	SYSCTL_RUNLOCK(&tracker);
1392 
1393 	if (p != buf)
1394 		free(p, M_SYSCTL);
1395 
1396 	if (error)
1397 		return (error);
1398 
1399 	error = SYSCTL_OUT(req, oid, len * sizeof *oid);
1400 	return (error);
1401 }
1402 
1403 /*
1404  * XXXRW/JA: Shouldn't return name2oid data for nodes that we don't permit in
1405  * capability mode.
1406  */
1407 SYSCTL_PROC(_sysctl, CTL_SYSCTL_NAME2OID, name2oid, CTLTYPE_INT | CTLFLAG_RW |
1408     CTLFLAG_ANYBODY | CTLFLAG_MPSAFE | CTLFLAG_CAPRW, 0, 0,
1409     sysctl_sysctl_name2oid, "I", "");
1410 
1411 static int
1412 sysctl_sysctl_oidfmt(SYSCTL_HANDLER_ARGS)
1413 {
1414 	struct sysctl_oid *oid;
1415 	struct rm_priotracker tracker;
1416 	int error;
1417 
1418 	error = sysctl_wire_old_buffer(req, 0);
1419 	if (error)
1420 		return (error);
1421 
1422 	SYSCTL_RLOCK(&tracker);
1423 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1424 	if (error)
1425 		goto out;
1426 
1427 	if (oid->oid_fmt == NULL) {
1428 		error = ENOENT;
1429 		goto out;
1430 	}
1431 	error = SYSCTL_OUT(req, &oid->oid_kind, sizeof(oid->oid_kind));
1432 	if (error)
1433 		goto out;
1434 	error = SYSCTL_OUT(req, oid->oid_fmt, strlen(oid->oid_fmt) + 1);
1435  out:
1436 	SYSCTL_RUNLOCK(&tracker);
1437 	return (error);
1438 }
1439 
1440 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_OIDFMT, oidfmt, CTLFLAG_RD |
1441     CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_oidfmt, "");
1442 
1443 static int
1444 sysctl_sysctl_oiddescr(SYSCTL_HANDLER_ARGS)
1445 {
1446 	struct sysctl_oid *oid;
1447 	struct rm_priotracker tracker;
1448 	int error;
1449 
1450 	error = sysctl_wire_old_buffer(req, 0);
1451 	if (error)
1452 		return (error);
1453 
1454 	SYSCTL_RLOCK(&tracker);
1455 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1456 	if (error)
1457 		goto out;
1458 
1459 	if (oid->oid_descr == NULL) {
1460 		error = ENOENT;
1461 		goto out;
1462 	}
1463 	error = SYSCTL_OUT(req, oid->oid_descr, strlen(oid->oid_descr) + 1);
1464  out:
1465 	SYSCTL_RUNLOCK(&tracker);
1466 	return (error);
1467 }
1468 
1469 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_OIDDESCR, oiddescr, CTLFLAG_RD |
1470     CTLFLAG_MPSAFE|CTLFLAG_CAPRD, sysctl_sysctl_oiddescr, "");
1471 
1472 static int
1473 sysctl_sysctl_oidlabel(SYSCTL_HANDLER_ARGS)
1474 {
1475 	struct sysctl_oid *oid;
1476 	struct rm_priotracker tracker;
1477 	int error;
1478 
1479 	error = sysctl_wire_old_buffer(req, 0);
1480 	if (error)
1481 		return (error);
1482 
1483 	SYSCTL_RLOCK(&tracker);
1484 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1485 	if (error)
1486 		goto out;
1487 
1488 	if (oid->oid_label == NULL) {
1489 		error = ENOENT;
1490 		goto out;
1491 	}
1492 	error = SYSCTL_OUT(req, oid->oid_label, strlen(oid->oid_label) + 1);
1493  out:
1494 	SYSCTL_RUNLOCK(&tracker);
1495 	return (error);
1496 }
1497 
1498 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_OIDLABEL, oidlabel, CTLFLAG_RD |
1499     CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_oidlabel, "");
1500 
1501 /*
1502  * Default "handler" functions.
1503  */
1504 
1505 /*
1506  * Handle a bool.
1507  * Two cases:
1508  *     a variable:  point arg1 at it.
1509  *     a constant:  pass it in arg2.
1510  */
1511 
1512 int
1513 sysctl_handle_bool(SYSCTL_HANDLER_ARGS)
1514 {
1515 	uint8_t temp;
1516 	int error;
1517 
1518 	/*
1519 	 * Attempt to get a coherent snapshot by making a copy of the data.
1520 	 */
1521 	if (arg1)
1522 		temp = *(bool *)arg1 ? 1 : 0;
1523 	else
1524 		temp = arg2 ? 1 : 0;
1525 
1526 	error = SYSCTL_OUT(req, &temp, sizeof(temp));
1527 	if (error || !req->newptr)
1528 		return (error);
1529 
1530 	if (!arg1)
1531 		error = EPERM;
1532 	else {
1533 		error = SYSCTL_IN(req, &temp, sizeof(temp));
1534 		if (!error)
1535 			*(bool *)arg1 = temp ? 1 : 0;
1536 	}
1537 	return (error);
1538 }
1539 
1540 /*
1541  * Handle an int8_t, signed or unsigned.
1542  * Two cases:
1543  *     a variable:  point arg1 at it.
1544  *     a constant:  pass it in arg2.
1545  */
1546 
1547 int
1548 sysctl_handle_8(SYSCTL_HANDLER_ARGS)
1549 {
1550 	int8_t tmpout;
1551 	int error = 0;
1552 
1553 	/*
1554 	 * Attempt to get a coherent snapshot by making a copy of the data.
1555 	 */
1556 	if (arg1)
1557 		tmpout = *(int8_t *)arg1;
1558 	else
1559 		tmpout = arg2;
1560 	error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout));
1561 
1562 	if (error || !req->newptr)
1563 		return (error);
1564 
1565 	if (!arg1)
1566 		error = EPERM;
1567 	else
1568 		error = SYSCTL_IN(req, arg1, sizeof(tmpout));
1569 	return (error);
1570 }
1571 
1572 /*
1573  * Handle an int16_t, signed or unsigned.
1574  * Two cases:
1575  *     a variable:  point arg1 at it.
1576  *     a constant:  pass it in arg2.
1577  */
1578 
1579 int
1580 sysctl_handle_16(SYSCTL_HANDLER_ARGS)
1581 {
1582 	int16_t tmpout;
1583 	int error = 0;
1584 
1585 	/*
1586 	 * Attempt to get a coherent snapshot by making a copy of the data.
1587 	 */
1588 	if (arg1)
1589 		tmpout = *(int16_t *)arg1;
1590 	else
1591 		tmpout = arg2;
1592 	error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout));
1593 
1594 	if (error || !req->newptr)
1595 		return (error);
1596 
1597 	if (!arg1)
1598 		error = EPERM;
1599 	else
1600 		error = SYSCTL_IN(req, arg1, sizeof(tmpout));
1601 	return (error);
1602 }
1603 
1604 /*
1605  * Handle an int32_t, signed or unsigned.
1606  * Two cases:
1607  *     a variable:  point arg1 at it.
1608  *     a constant:  pass it in arg2.
1609  */
1610 
1611 int
1612 sysctl_handle_32(SYSCTL_HANDLER_ARGS)
1613 {
1614 	int32_t tmpout;
1615 	int error = 0;
1616 
1617 	/*
1618 	 * Attempt to get a coherent snapshot by making a copy of the data.
1619 	 */
1620 	if (arg1)
1621 		tmpout = *(int32_t *)arg1;
1622 	else
1623 		tmpout = arg2;
1624 	error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout));
1625 
1626 	if (error || !req->newptr)
1627 		return (error);
1628 
1629 	if (!arg1)
1630 		error = EPERM;
1631 	else
1632 		error = SYSCTL_IN(req, arg1, sizeof(tmpout));
1633 	return (error);
1634 }
1635 
1636 /*
1637  * Handle an int, signed or unsigned.
1638  * Two cases:
1639  *     a variable:  point arg1 at it.
1640  *     a constant:  pass it in arg2.
1641  */
1642 
1643 int
1644 sysctl_handle_int(SYSCTL_HANDLER_ARGS)
1645 {
1646 	int tmpout, error = 0;
1647 
1648 	/*
1649 	 * Attempt to get a coherent snapshot by making a copy of the data.
1650 	 */
1651 	if (arg1)
1652 		tmpout = *(int *)arg1;
1653 	else
1654 		tmpout = arg2;
1655 	error = SYSCTL_OUT(req, &tmpout, sizeof(int));
1656 
1657 	if (error || !req->newptr)
1658 		return (error);
1659 
1660 	if (!arg1)
1661 		error = EPERM;
1662 	else
1663 		error = SYSCTL_IN(req, arg1, sizeof(int));
1664 	return (error);
1665 }
1666 
1667 /*
1668  * Based on on sysctl_handle_int() convert milliseconds into ticks.
1669  * Note: this is used by TCP.
1670  */
1671 
1672 int
1673 sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
1674 {
1675 	int error, s, tt;
1676 
1677 	tt = *(int *)arg1;
1678 	s = (int)((int64_t)tt * 1000 / hz);
1679 
1680 	error = sysctl_handle_int(oidp, &s, 0, req);
1681 	if (error || !req->newptr)
1682 		return (error);
1683 
1684 	tt = (int)((int64_t)s * hz / 1000);
1685 	if (tt < 1)
1686 		return (EINVAL);
1687 
1688 	*(int *)arg1 = tt;
1689 	return (0);
1690 }
1691 
1692 /*
1693  * Handle a long, signed or unsigned.
1694  * Two cases:
1695  *     a variable:  point arg1 at it.
1696  *     a constant:  pass it in arg2.
1697  */
1698 
1699 int
1700 sysctl_handle_long(SYSCTL_HANDLER_ARGS)
1701 {
1702 	int error = 0;
1703 	long tmplong;
1704 #ifdef SCTL_MASK32
1705 	int tmpint;
1706 #endif
1707 
1708 	/*
1709 	 * Attempt to get a coherent snapshot by making a copy of the data.
1710 	 */
1711 	if (arg1)
1712 		tmplong = *(long *)arg1;
1713 	else
1714 		tmplong = arg2;
1715 #ifdef SCTL_MASK32
1716 	if (req->flags & SCTL_MASK32) {
1717 		tmpint = tmplong;
1718 		error = SYSCTL_OUT(req, &tmpint, sizeof(int));
1719 	} else
1720 #endif
1721 		error = SYSCTL_OUT(req, &tmplong, sizeof(long));
1722 
1723 	if (error || !req->newptr)
1724 		return (error);
1725 
1726 	if (!arg1)
1727 		error = EPERM;
1728 #ifdef SCTL_MASK32
1729 	else if (req->flags & SCTL_MASK32) {
1730 		error = SYSCTL_IN(req, &tmpint, sizeof(int));
1731 		*(long *)arg1 = (long)tmpint;
1732 	}
1733 #endif
1734 	else
1735 		error = SYSCTL_IN(req, arg1, sizeof(long));
1736 	return (error);
1737 }
1738 
1739 /*
1740  * Handle a 64 bit int, signed or unsigned.
1741  * Two cases:
1742  *     a variable:  point arg1 at it.
1743  *     a constant:  pass it in arg2.
1744  */
1745 int
1746 sysctl_handle_64(SYSCTL_HANDLER_ARGS)
1747 {
1748 	int error = 0;
1749 	uint64_t tmpout;
1750 
1751 	/*
1752 	 * Attempt to get a coherent snapshot by making a copy of the data.
1753 	 */
1754 	if (arg1)
1755 		tmpout = *(uint64_t *)arg1;
1756 	else
1757 		tmpout = arg2;
1758 	error = SYSCTL_OUT(req, &tmpout, sizeof(uint64_t));
1759 
1760 	if (error || !req->newptr)
1761 		return (error);
1762 
1763 	if (!arg1)
1764 		error = EPERM;
1765 	else
1766 		error = SYSCTL_IN(req, arg1, sizeof(uint64_t));
1767 	return (error);
1768 }
1769 
1770 /*
1771  * Handle our generic '\0' terminated 'C' string.
1772  * Two cases:
1773  * 	a variable string:  point arg1 at it, arg2 is max length.
1774  * 	a constant string:  point arg1 at it, arg2 is zero.
1775  */
1776 
1777 int
1778 sysctl_handle_string(SYSCTL_HANDLER_ARGS)
1779 {
1780 	char *tmparg;
1781 	size_t outlen;
1782 	int error = 0, ro_string = 0;
1783 
1784 	/*
1785 	 * If the sysctl isn't writable and isn't a preallocated tunable that
1786 	 * can be modified by kenv(2), microoptimise and treat it as a
1787 	 * read-only string.
1788 	 * A zero-length buffer indicates a fixed size read-only
1789 	 * string.  In ddb, don't worry about trying to make a malloced
1790 	 * snapshot.
1791 	 */
1792 	if ((oidp->oid_kind & (CTLFLAG_WR | CTLFLAG_TUN)) == 0 ||
1793 	    arg2 == 0 || kdb_active) {
1794 		arg2 = strlen((char *)arg1) + 1;
1795 		ro_string = 1;
1796 	}
1797 
1798 	if (req->oldptr != NULL) {
1799 		if (ro_string) {
1800 			tmparg = arg1;
1801 			outlen = strlen(tmparg) + 1;
1802 		} else {
1803 			tmparg = malloc(arg2, M_SYSCTLTMP, M_WAITOK);
1804 			sx_slock(&sysctlstringlock);
1805 			memcpy(tmparg, arg1, arg2);
1806 			sx_sunlock(&sysctlstringlock);
1807 			outlen = strlen(tmparg) + 1;
1808 		}
1809 
1810 		error = SYSCTL_OUT(req, tmparg, outlen);
1811 
1812 		if (!ro_string)
1813 			free(tmparg, M_SYSCTLTMP);
1814 	} else {
1815 		if (!ro_string)
1816 			sx_slock(&sysctlstringlock);
1817 		outlen = strlen((char *)arg1) + 1;
1818 		if (!ro_string)
1819 			sx_sunlock(&sysctlstringlock);
1820 		error = SYSCTL_OUT(req, NULL, outlen);
1821 	}
1822 	if (error || !req->newptr)
1823 		return (error);
1824 
1825 	if (req->newlen - req->newidx >= arg2 ||
1826 	    req->newlen - req->newidx < 0) {
1827 		error = EINVAL;
1828 	} else if (req->newlen - req->newidx == 0) {
1829 		sx_xlock(&sysctlstringlock);
1830 		((char *)arg1)[0] = '\0';
1831 		sx_xunlock(&sysctlstringlock);
1832 	} else if (req->newfunc == sysctl_new_kernel) {
1833 		arg2 = req->newlen - req->newidx;
1834 		sx_xlock(&sysctlstringlock);
1835 		error = SYSCTL_IN(req, arg1, arg2);
1836 		if (error == 0) {
1837 			((char *)arg1)[arg2] = '\0';
1838 			req->newidx += arg2;
1839 		}
1840 		sx_xunlock(&sysctlstringlock);
1841 	} else {
1842 		arg2 = req->newlen - req->newidx;
1843 		tmparg = malloc(arg2, M_SYSCTLTMP, M_WAITOK);
1844 
1845 		error = SYSCTL_IN(req, tmparg, arg2);
1846 		if (error) {
1847 			free(tmparg, M_SYSCTLTMP);
1848 			return (error);
1849 		}
1850 
1851 		sx_xlock(&sysctlstringlock);
1852 		memcpy(arg1, tmparg, arg2);
1853 		((char *)arg1)[arg2] = '\0';
1854 		sx_xunlock(&sysctlstringlock);
1855 		free(tmparg, M_SYSCTLTMP);
1856 		req->newidx += arg2;
1857 	}
1858 	return (error);
1859 }
1860 
1861 /*
1862  * Handle any kind of opaque data.
1863  * arg1 points to it, arg2 is the size.
1864  */
1865 
1866 int
1867 sysctl_handle_opaque(SYSCTL_HANDLER_ARGS)
1868 {
1869 	int error, tries;
1870 	u_int generation;
1871 	struct sysctl_req req2;
1872 
1873 	/*
1874 	 * Attempt to get a coherent snapshot, by using the thread
1875 	 * pre-emption counter updated from within mi_switch() to
1876 	 * determine if we were pre-empted during a bcopy() or
1877 	 * copyout(). Make 3 attempts at doing this before giving up.
1878 	 * If we encounter an error, stop immediately.
1879 	 */
1880 	tries = 0;
1881 	req2 = *req;
1882 retry:
1883 	generation = curthread->td_generation;
1884 	error = SYSCTL_OUT(req, arg1, arg2);
1885 	if (error)
1886 		return (error);
1887 	tries++;
1888 	if (generation != curthread->td_generation && tries < 3) {
1889 		*req = req2;
1890 		goto retry;
1891 	}
1892 
1893 	error = SYSCTL_IN(req, arg1, arg2);
1894 
1895 	return (error);
1896 }
1897 
1898 /*
1899  * Based on on sysctl_handle_int() convert microseconds to a sbintime.
1900  */
1901 int
1902 sysctl_usec_to_sbintime(SYSCTL_HANDLER_ARGS)
1903 {
1904 	int error;
1905 	int64_t tt;
1906 	sbintime_t sb;
1907 
1908 	tt = *(int64_t *)arg1;
1909 	sb = sbttous(tt);
1910 
1911 	error = sysctl_handle_64(oidp, &sb, 0, req);
1912 	if (error || !req->newptr)
1913 		return (error);
1914 
1915 	tt = ustosbt(sb);
1916 	*(int64_t *)arg1 = tt;
1917 
1918 	return (0);
1919 }
1920 
1921 /*
1922  * Based on on sysctl_handle_int() convert milliseconds to a sbintime.
1923  */
1924 int
1925 sysctl_msec_to_sbintime(SYSCTL_HANDLER_ARGS)
1926 {
1927 	int error;
1928 	int64_t tt;
1929 	sbintime_t sb;
1930 
1931 	tt = *(int64_t *)arg1;
1932 	sb = sbttoms(tt);
1933 
1934 	error = sysctl_handle_64(oidp, &sb, 0, req);
1935 	if (error || !req->newptr)
1936 		return (error);
1937 
1938 	tt = mstosbt(sb);
1939 	*(int64_t *)arg1 = tt;
1940 
1941 	return (0);
1942 }
1943 
1944 /*
1945  * Convert seconds to a struct timeval.  Intended for use with
1946  * intervals and thus does not permit negative seconds.
1947  */
1948 int
1949 sysctl_sec_to_timeval(SYSCTL_HANDLER_ARGS)
1950 {
1951 	struct timeval *tv;
1952 	int error, secs;
1953 
1954 	tv = arg1;
1955 	secs = tv->tv_sec;
1956 
1957 	error = sysctl_handle_int(oidp, &secs, 0, req);
1958 	if (error || req->newptr == NULL)
1959 		return (error);
1960 
1961 	if (secs < 0)
1962 		return (EINVAL);
1963 	tv->tv_sec = secs;
1964 
1965 	return (0);
1966 }
1967 
1968 /*
1969  * Transfer functions to/from kernel space.
1970  * XXX: rather untested at this point
1971  */
1972 static int
1973 sysctl_old_kernel(struct sysctl_req *req, const void *p, size_t l)
1974 {
1975 	size_t i = 0;
1976 
1977 	if (req->oldptr) {
1978 		i = l;
1979 		if (req->oldlen <= req->oldidx)
1980 			i = 0;
1981 		else
1982 			if (i > req->oldlen - req->oldidx)
1983 				i = req->oldlen - req->oldidx;
1984 		if (i > 0)
1985 			bcopy(p, (char *)req->oldptr + req->oldidx, i);
1986 	}
1987 	req->oldidx += l;
1988 	if (req->oldptr && i != l)
1989 		return (ENOMEM);
1990 	return (0);
1991 }
1992 
1993 static int
1994 sysctl_new_kernel(struct sysctl_req *req, void *p, size_t l)
1995 {
1996 	if (!req->newptr)
1997 		return (0);
1998 	if (req->newlen - req->newidx < l)
1999 		return (EINVAL);
2000 	bcopy((const char *)req->newptr + req->newidx, p, l);
2001 	req->newidx += l;
2002 	return (0);
2003 }
2004 
2005 int
2006 kernel_sysctl(struct thread *td, int *name, u_int namelen, void *old,
2007     size_t *oldlenp, void *new, size_t newlen, size_t *retval, int flags)
2008 {
2009 	int error = 0;
2010 	struct sysctl_req req;
2011 
2012 	bzero(&req, sizeof req);
2013 
2014 	req.td = td;
2015 	req.flags = flags;
2016 
2017 	if (oldlenp) {
2018 		req.oldlen = *oldlenp;
2019 	}
2020 	req.validlen = req.oldlen;
2021 
2022 	if (old) {
2023 		req.oldptr= old;
2024 	}
2025 
2026 	if (new != NULL) {
2027 		req.newlen = newlen;
2028 		req.newptr = new;
2029 	}
2030 
2031 	req.oldfunc = sysctl_old_kernel;
2032 	req.newfunc = sysctl_new_kernel;
2033 	req.lock = REQ_UNWIRED;
2034 
2035 	error = sysctl_root(0, name, namelen, &req);
2036 
2037 	if (req.lock == REQ_WIRED && req.validlen > 0)
2038 		vsunlock(req.oldptr, req.validlen);
2039 
2040 	if (error && error != ENOMEM)
2041 		return (error);
2042 
2043 	if (retval) {
2044 		if (req.oldptr && req.oldidx > req.validlen)
2045 			*retval = req.validlen;
2046 		else
2047 			*retval = req.oldidx;
2048 	}
2049 	return (error);
2050 }
2051 
2052 int
2053 kernel_sysctlbyname(struct thread *td, char *name, void *old, size_t *oldlenp,
2054     void *new, size_t newlen, size_t *retval, int flags)
2055 {
2056         int oid[CTL_MAXNAME];
2057         size_t oidlen, plen;
2058 	int error;
2059 
2060 	oid[0] = CTL_SYSCTL;
2061 	oid[1] = CTL_SYSCTL_NAME2OID;
2062 	oidlen = sizeof(oid);
2063 
2064 	error = kernel_sysctl(td, oid, 2, oid, &oidlen,
2065 	    (void *)name, strlen(name), &plen, flags);
2066 	if (error)
2067 		return (error);
2068 
2069 	error = kernel_sysctl(td, oid, plen / sizeof(int), old, oldlenp,
2070 	    new, newlen, retval, flags);
2071 	return (error);
2072 }
2073 
2074 /*
2075  * Transfer function to/from user space.
2076  */
2077 static int
2078 sysctl_old_user(struct sysctl_req *req, const void *p, size_t l)
2079 {
2080 	size_t i, len, origidx;
2081 	int error;
2082 
2083 	origidx = req->oldidx;
2084 	req->oldidx += l;
2085 	if (req->oldptr == NULL)
2086 		return (0);
2087 	/*
2088 	 * If we have not wired the user supplied buffer and we are currently
2089 	 * holding locks, drop a witness warning, as it's possible that
2090 	 * write operations to the user page can sleep.
2091 	 */
2092 	if (req->lock != REQ_WIRED)
2093 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
2094 		    "sysctl_old_user()");
2095 	i = l;
2096 	len = req->validlen;
2097 	if (len <= origidx)
2098 		i = 0;
2099 	else {
2100 		if (i > len - origidx)
2101 			i = len - origidx;
2102 		if (req->lock == REQ_WIRED) {
2103 			error = copyout_nofault(p, (char *)req->oldptr +
2104 			    origidx, i);
2105 		} else
2106 			error = copyout(p, (char *)req->oldptr + origidx, i);
2107 		if (error != 0)
2108 			return (error);
2109 	}
2110 	if (i < l)
2111 		return (ENOMEM);
2112 	return (0);
2113 }
2114 
2115 static int
2116 sysctl_new_user(struct sysctl_req *req, void *p, size_t l)
2117 {
2118 	int error;
2119 
2120 	if (!req->newptr)
2121 		return (0);
2122 	if (req->newlen - req->newidx < l)
2123 		return (EINVAL);
2124 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
2125 	    "sysctl_new_user()");
2126 	error = copyin((const char *)req->newptr + req->newidx, p, l);
2127 	req->newidx += l;
2128 	return (error);
2129 }
2130 
2131 /*
2132  * Wire the user space destination buffer.  If set to a value greater than
2133  * zero, the len parameter limits the maximum amount of wired memory.
2134  */
2135 int
2136 sysctl_wire_old_buffer(struct sysctl_req *req, size_t len)
2137 {
2138 	int ret;
2139 	size_t wiredlen;
2140 
2141 	wiredlen = (len > 0 && len < req->oldlen) ? len : req->oldlen;
2142 	ret = 0;
2143 	if (req->lock != REQ_WIRED && req->oldptr &&
2144 	    req->oldfunc == sysctl_old_user) {
2145 		if (wiredlen != 0) {
2146 			ret = vslock(req->oldptr, wiredlen);
2147 			if (ret != 0) {
2148 				if (ret != ENOMEM)
2149 					return (ret);
2150 				wiredlen = 0;
2151 			}
2152 		}
2153 		req->lock = REQ_WIRED;
2154 		req->validlen = wiredlen;
2155 	}
2156 	return (0);
2157 }
2158 
2159 int
2160 sysctl_find_oid(int *name, u_int namelen, struct sysctl_oid **noid,
2161     int *nindx, struct sysctl_req *req)
2162 {
2163 	struct sysctl_oid_list *lsp;
2164 	struct sysctl_oid *oid;
2165 	int indx;
2166 
2167 	SYSCTL_ASSERT_LOCKED();
2168 	lsp = &sysctl__children;
2169 	indx = 0;
2170 	while (indx < CTL_MAXNAME) {
2171 		SLIST_FOREACH(oid, lsp, oid_link) {
2172 			if (oid->oid_number == name[indx])
2173 				break;
2174 		}
2175 		if (oid == NULL)
2176 			return (ENOENT);
2177 
2178 		indx++;
2179 		if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
2180 			if (oid->oid_handler != NULL || indx == namelen) {
2181 				*noid = oid;
2182 				if (nindx != NULL)
2183 					*nindx = indx;
2184 				KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
2185 				    ("%s found DYING node %p", __func__, oid));
2186 				return (0);
2187 			}
2188 			lsp = SYSCTL_CHILDREN(oid);
2189 		} else if (indx == namelen) {
2190 			if ((oid->oid_kind & CTLFLAG_DORMANT) != 0)
2191 				return (ENOENT);
2192 			*noid = oid;
2193 			if (nindx != NULL)
2194 				*nindx = indx;
2195 			KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
2196 			    ("%s found DYING node %p", __func__, oid));
2197 			return (0);
2198 		} else {
2199 			return (ENOTDIR);
2200 		}
2201 	}
2202 	return (ENOENT);
2203 }
2204 
2205 /*
2206  * Traverse our tree, and find the right node, execute whatever it points
2207  * to, and return the resulting error code.
2208  */
2209 
2210 static int
2211 sysctl_root(SYSCTL_HANDLER_ARGS)
2212 {
2213 	struct sysctl_oid *oid;
2214 	struct rm_priotracker tracker;
2215 	int error, indx, lvl;
2216 
2217 	SYSCTL_RLOCK(&tracker);
2218 
2219 	error = sysctl_find_oid(arg1, arg2, &oid, &indx, req);
2220 	if (error)
2221 		goto out;
2222 
2223 	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
2224 		/*
2225 		 * You can't call a sysctl when it's a node, but has
2226 		 * no handler.  Inform the user that it's a node.
2227 		 * The indx may or may not be the same as namelen.
2228 		 */
2229 		if (oid->oid_handler == NULL) {
2230 			error = EISDIR;
2231 			goto out;
2232 		}
2233 	}
2234 
2235 	/* Is this sysctl writable? */
2236 	if (req->newptr && !(oid->oid_kind & CTLFLAG_WR)) {
2237 		error = EPERM;
2238 		goto out;
2239 	}
2240 
2241 	KASSERT(req->td != NULL, ("sysctl_root(): req->td == NULL"));
2242 
2243 #ifdef CAPABILITY_MODE
2244 	/*
2245 	 * If the process is in capability mode, then don't permit reading or
2246 	 * writing unless specifically granted for the node.
2247 	 */
2248 	if (IN_CAPABILITY_MODE(req->td)) {
2249 		if ((req->oldptr && !(oid->oid_kind & CTLFLAG_CAPRD)) ||
2250 		    (req->newptr && !(oid->oid_kind & CTLFLAG_CAPWR))) {
2251 			error = EPERM;
2252 			goto out;
2253 		}
2254 	}
2255 #endif
2256 
2257 	/* Is this sysctl sensitive to securelevels? */
2258 	if (req->newptr && (oid->oid_kind & CTLFLAG_SECURE)) {
2259 		lvl = (oid->oid_kind & CTLMASK_SECURE) >> CTLSHIFT_SECURE;
2260 		error = securelevel_gt(req->td->td_ucred, lvl);
2261 		if (error)
2262 			goto out;
2263 	}
2264 
2265 	/* Is this sysctl writable by only privileged users? */
2266 	if (req->newptr && !(oid->oid_kind & CTLFLAG_ANYBODY)) {
2267 		int priv;
2268 
2269 		if (oid->oid_kind & CTLFLAG_PRISON)
2270 			priv = PRIV_SYSCTL_WRITEJAIL;
2271 #ifdef VIMAGE
2272 		else if ((oid->oid_kind & CTLFLAG_VNET) &&
2273 		     prison_owns_vnet(req->td->td_ucred))
2274 			priv = PRIV_SYSCTL_WRITEJAIL;
2275 #endif
2276 		else
2277 			priv = PRIV_SYSCTL_WRITE;
2278 		error = priv_check(req->td, priv);
2279 		if (error)
2280 			goto out;
2281 	}
2282 
2283 	if (!oid->oid_handler) {
2284 		error = EINVAL;
2285 		goto out;
2286 	}
2287 
2288 	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
2289 		arg1 = (int *)arg1 + indx;
2290 		arg2 -= indx;
2291 	} else {
2292 		arg1 = oid->oid_arg1;
2293 		arg2 = oid->oid_arg2;
2294 	}
2295 #ifdef MAC
2296 	error = mac_system_check_sysctl(req->td->td_ucred, oid, arg1, arg2,
2297 	    req);
2298 	if (error != 0)
2299 		goto out;
2300 #endif
2301 #ifdef VIMAGE
2302 	if ((oid->oid_kind & CTLFLAG_VNET) && arg1 != NULL)
2303 		arg1 = (void *)(curvnet->vnet_data_base + (uintptr_t)arg1);
2304 #endif
2305 	error = sysctl_root_handler_locked(oid, arg1, arg2, req, &tracker);
2306 
2307 out:
2308 	SYSCTL_RUNLOCK(&tracker);
2309 	return (error);
2310 }
2311 
2312 #ifndef _SYS_SYSPROTO_H_
2313 struct __sysctl_args {
2314 	int	*name;
2315 	u_int	namelen;
2316 	void	*old;
2317 	size_t	*oldlenp;
2318 	void	*new;
2319 	size_t	newlen;
2320 };
2321 #endif
2322 int
2323 sys___sysctl(struct thread *td, struct __sysctl_args *uap)
2324 {
2325 	int error, i, name[CTL_MAXNAME];
2326 	size_t j;
2327 
2328 	if (uap->namelen > CTL_MAXNAME || uap->namelen < 2)
2329 		return (EINVAL);
2330 
2331  	error = copyin(uap->name, &name, uap->namelen * sizeof(int));
2332  	if (error)
2333 		return (error);
2334 
2335 	error = userland_sysctl(td, name, uap->namelen,
2336 		uap->old, uap->oldlenp, 0,
2337 		uap->new, uap->newlen, &j, 0);
2338 	if (error && error != ENOMEM)
2339 		return (error);
2340 	if (uap->oldlenp) {
2341 		i = copyout(&j, uap->oldlenp, sizeof(j));
2342 		if (i)
2343 			return (i);
2344 	}
2345 	return (error);
2346 }
2347 
2348 int
2349 kern___sysctlbyname(struct thread *td, const char *oname, size_t namelen,
2350     void *old, size_t *oldlenp, void *new, size_t newlen, size_t *retval,
2351     int flags, bool inkernel)
2352 {
2353 	int oid[CTL_MAXNAME];
2354 	char namebuf[16];
2355 	char *name;
2356 	size_t oidlen;
2357 	int error;
2358 
2359 	if (namelen > MAXPATHLEN || namelen == 0)
2360 		return (EINVAL);
2361 	name = namebuf;
2362 	if (namelen > sizeof(namebuf))
2363 		name = malloc(namelen, M_SYSCTL, M_WAITOK);
2364 	error = copyin(oname, name, namelen);
2365 	if (error != 0)
2366 		goto out;
2367 
2368 	oid[0] = CTL_SYSCTL;
2369 	oid[1] = CTL_SYSCTL_NAME2OID;
2370 	oidlen = sizeof(oid);
2371 	error = kernel_sysctl(td, oid, 2, oid, &oidlen, (void *)name, namelen,
2372 	    retval, flags);
2373 	if (error != 0)
2374 		goto out;
2375 	error = userland_sysctl(td, oid, *retval / sizeof(int), old, oldlenp,
2376 	    inkernel, new, newlen, retval, flags);
2377 
2378 out:
2379 	if (namelen > sizeof(namebuf))
2380 		free(name, M_SYSCTL);
2381 	return (error);
2382 }
2383 
2384 #ifndef	_SYS_SYSPROTO_H_
2385 struct __sysctlbyname_args {
2386 	const char	*name;
2387 	size_t	namelen;
2388 	void	*old;
2389 	size_t	*oldlenp;
2390 	void	*new;
2391 	size_t	newlen;
2392 };
2393 #endif
2394 int
2395 sys___sysctlbyname(struct thread *td, struct __sysctlbyname_args *uap)
2396 {
2397 	size_t rv;
2398 	int error;
2399 
2400 	error = kern___sysctlbyname(td, uap->name, uap->namelen, uap->old,
2401 	    uap->oldlenp, uap->new, uap->newlen, &rv, 0, 0);
2402 	if (error != 0)
2403 		return (error);
2404 	if (uap->oldlenp != NULL)
2405 		error = copyout(&rv, uap->oldlenp, sizeof(rv));
2406 
2407 	return (error);
2408 }
2409 
2410 /*
2411  * This is used from various compatibility syscalls too.  That's why name
2412  * must be in kernel space.
2413  */
2414 int
2415 userland_sysctl(struct thread *td, int *name, u_int namelen, void *old,
2416     size_t *oldlenp, int inkernel, const void *new, size_t newlen,
2417     size_t *retval, int flags)
2418 {
2419 	int error = 0, memlocked;
2420 	struct sysctl_req req;
2421 
2422 	bzero(&req, sizeof req);
2423 
2424 	req.td = td;
2425 	req.flags = flags;
2426 
2427 	if (oldlenp) {
2428 		if (inkernel) {
2429 			req.oldlen = *oldlenp;
2430 		} else {
2431 			error = copyin(oldlenp, &req.oldlen, sizeof(*oldlenp));
2432 			if (error)
2433 				return (error);
2434 		}
2435 	}
2436 	req.validlen = req.oldlen;
2437 	req.oldptr = old;
2438 
2439 	if (new != NULL) {
2440 		req.newlen = newlen;
2441 		req.newptr = new;
2442 	}
2443 
2444 	req.oldfunc = sysctl_old_user;
2445 	req.newfunc = sysctl_new_user;
2446 	req.lock = REQ_UNWIRED;
2447 
2448 #ifdef KTRACE
2449 	if (KTRPOINT(curthread, KTR_SYSCTL))
2450 		ktrsysctl(name, namelen);
2451 #endif
2452 	memlocked = 0;
2453 	if (req.oldptr && req.oldlen > 4 * PAGE_SIZE) {
2454 		memlocked = 1;
2455 		sx_xlock(&sysctlmemlock);
2456 	}
2457 	CURVNET_SET(TD_TO_VNET(td));
2458 
2459 	for (;;) {
2460 		req.oldidx = 0;
2461 		req.newidx = 0;
2462 		error = sysctl_root(0, name, namelen, &req);
2463 		if (error != EAGAIN)
2464 			break;
2465 		kern_yield(PRI_USER);
2466 	}
2467 
2468 	CURVNET_RESTORE();
2469 
2470 	if (req.lock == REQ_WIRED && req.validlen > 0)
2471 		vsunlock(req.oldptr, req.validlen);
2472 	if (memlocked)
2473 		sx_xunlock(&sysctlmemlock);
2474 
2475 	if (error && error != ENOMEM)
2476 		return (error);
2477 
2478 	if (retval) {
2479 		if (req.oldptr && req.oldidx > req.validlen)
2480 			*retval = req.validlen;
2481 		else
2482 			*retval = req.oldidx;
2483 	}
2484 	return (error);
2485 }
2486 
2487 /*
2488  * Drain into a sysctl struct.  The user buffer should be wired if a page
2489  * fault would cause issue.
2490  */
2491 static int
2492 sbuf_sysctl_drain(void *arg, const char *data, int len)
2493 {
2494 	struct sysctl_req *req = arg;
2495 	int error;
2496 
2497 	error = SYSCTL_OUT(req, data, len);
2498 	KASSERT(error >= 0, ("Got unexpected negative value %d", error));
2499 	return (error == 0 ? len : -error);
2500 }
2501 
2502 struct sbuf *
2503 sbuf_new_for_sysctl(struct sbuf *s, char *buf, int length,
2504     struct sysctl_req *req)
2505 {
2506 
2507 	/* Supply a default buffer size if none given. */
2508 	if (buf == NULL && length == 0)
2509 		length = 64;
2510 	s = sbuf_new(s, buf, length, SBUF_FIXEDLEN | SBUF_INCLUDENUL);
2511 	sbuf_set_drain(s, sbuf_sysctl_drain, req);
2512 	return (s);
2513 }
2514 
2515 #ifdef DDB
2516 
2517 /* The current OID the debugger is working with */
2518 static struct sysctl_oid *g_ddb_oid;
2519 
2520 /* The current flags specified by the user */
2521 static int g_ddb_sysctl_flags;
2522 
2523 /* Check to see if the last sysctl printed */
2524 static int g_ddb_sysctl_printed;
2525 
2526 static const int ctl_sign[CTLTYPE+1] = {
2527 	[CTLTYPE_INT] = 1,
2528 	[CTLTYPE_LONG] = 1,
2529 	[CTLTYPE_S8] = 1,
2530 	[CTLTYPE_S16] = 1,
2531 	[CTLTYPE_S32] = 1,
2532 	[CTLTYPE_S64] = 1,
2533 };
2534 
2535 static const int ctl_size[CTLTYPE+1] = {
2536 	[CTLTYPE_INT] = sizeof(int),
2537 	[CTLTYPE_UINT] = sizeof(u_int),
2538 	[CTLTYPE_LONG] = sizeof(long),
2539 	[CTLTYPE_ULONG] = sizeof(u_long),
2540 	[CTLTYPE_S8] = sizeof(int8_t),
2541 	[CTLTYPE_S16] = sizeof(int16_t),
2542 	[CTLTYPE_S32] = sizeof(int32_t),
2543 	[CTLTYPE_S64] = sizeof(int64_t),
2544 	[CTLTYPE_U8] = sizeof(uint8_t),
2545 	[CTLTYPE_U16] = sizeof(uint16_t),
2546 	[CTLTYPE_U32] = sizeof(uint32_t),
2547 	[CTLTYPE_U64] = sizeof(uint64_t),
2548 };
2549 
2550 #define DB_SYSCTL_NAME_ONLY	0x001	/* Compare with -N */
2551 #define DB_SYSCTL_VALUE_ONLY	0x002	/* Compare with -n */
2552 #define DB_SYSCTL_OPAQUE	0x004	/* Compare with -o */
2553 #define DB_SYSCTL_HEX		0x008	/* Compare with -x */
2554 
2555 #define DB_SYSCTL_SAFE_ONLY	0x100	/* Only simple types */
2556 
2557 static const char db_sysctl_modifs[] = {
2558 	'N', 'n', 'o', 'x',
2559 };
2560 
2561 static const int db_sysctl_modif_values[] = {
2562 	DB_SYSCTL_NAME_ONLY, DB_SYSCTL_VALUE_ONLY,
2563 	DB_SYSCTL_OPAQUE, DB_SYSCTL_HEX,
2564 };
2565 
2566 /* Handlers considered safe to print while recursing */
2567 static int (* const db_safe_handlers[])(SYSCTL_HANDLER_ARGS) = {
2568 	sysctl_handle_bool,
2569 	sysctl_handle_8,
2570 	sysctl_handle_16,
2571 	sysctl_handle_32,
2572 	sysctl_handle_64,
2573 	sysctl_handle_int,
2574 	sysctl_handle_long,
2575 	sysctl_handle_string,
2576 	sysctl_handle_opaque,
2577 };
2578 
2579 /*
2580  * Use in place of sysctl_old_kernel to print sysctl values.
2581  *
2582  * Compare to the output handling in show_var from sbin/sysctl/sysctl.c
2583  */
2584 static int
2585 sysctl_old_ddb(struct sysctl_req *req, const void *ptr, size_t len)
2586 {
2587 	const u_char *val, *p;
2588 	const char *sep1;
2589 	size_t intlen, slen;
2590 	uintmax_t umv;
2591 	intmax_t mv;
2592 	int sign, ctltype, hexlen, xflag, error;
2593 
2594 	/* Suppress false-positive GCC uninitialized variable warnings */
2595 	mv = 0;
2596 	umv = 0;
2597 
2598 	slen = len;
2599 	val = p = ptr;
2600 
2601 	if (ptr == NULL) {
2602 		error = 0;
2603 		goto out;
2604 	}
2605 
2606 	/* We are going to print */
2607 	g_ddb_sysctl_printed = 1;
2608 
2609 	xflag = g_ddb_sysctl_flags & DB_SYSCTL_HEX;
2610 
2611 	ctltype = (g_ddb_oid->oid_kind & CTLTYPE);
2612 	sign = ctl_sign[ctltype];
2613 	intlen = ctl_size[ctltype];
2614 
2615 	switch (ctltype) {
2616 	case CTLTYPE_NODE:
2617 	case CTLTYPE_STRING:
2618 		db_printf("%.*s", (int) len, (const char *) p);
2619 		error = 0;
2620 		goto out;
2621 
2622 	case CTLTYPE_INT:
2623 	case CTLTYPE_UINT:
2624 	case CTLTYPE_LONG:
2625 	case CTLTYPE_ULONG:
2626 	case CTLTYPE_S8:
2627 	case CTLTYPE_S16:
2628 	case CTLTYPE_S32:
2629 	case CTLTYPE_S64:
2630 	case CTLTYPE_U8:
2631 	case CTLTYPE_U16:
2632 	case CTLTYPE_U32:
2633 	case CTLTYPE_U64:
2634 		hexlen = 2 + (intlen * CHAR_BIT + 3) / 4;
2635 		sep1 = "";
2636 		while (len >= intlen) {
2637 			switch (ctltype) {
2638 			case CTLTYPE_INT:
2639 			case CTLTYPE_UINT:
2640 				umv = *(const u_int *)p;
2641 				mv = *(const int *)p;
2642 				break;
2643 			case CTLTYPE_LONG:
2644 			case CTLTYPE_ULONG:
2645 				umv = *(const u_long *)p;
2646 				mv = *(const long *)p;
2647 				break;
2648 			case CTLTYPE_S8:
2649 			case CTLTYPE_U8:
2650 				umv = *(const uint8_t *)p;
2651 				mv = *(const int8_t *)p;
2652 				break;
2653 			case CTLTYPE_S16:
2654 			case CTLTYPE_U16:
2655 				umv = *(const uint16_t *)p;
2656 				mv = *(const int16_t *)p;
2657 				break;
2658 			case CTLTYPE_S32:
2659 			case CTLTYPE_U32:
2660 				umv = *(const uint32_t *)p;
2661 				mv = *(const int32_t *)p;
2662 				break;
2663 			case CTLTYPE_S64:
2664 			case CTLTYPE_U64:
2665 				umv = *(const uint64_t *)p;
2666 				mv = *(const int64_t *)p;
2667 				break;
2668 			}
2669 
2670 			db_printf("%s", sep1);
2671 			if (xflag)
2672 				db_printf("%#0*jx", hexlen, umv);
2673 			else if (!sign)
2674 				db_printf("%ju", umv);
2675 			else if (g_ddb_oid->oid_fmt[1] == 'K') {
2676 				/* Kelvins are currently unsupported. */
2677 				error = EOPNOTSUPP;
2678 				goto out;
2679 			} else
2680 				db_printf("%jd", mv);
2681 
2682 			sep1 = " ";
2683 			len -= intlen;
2684 			p += intlen;
2685 		}
2686 		error = 0;
2687 		goto out;
2688 
2689 	case CTLTYPE_OPAQUE:
2690 		/* TODO: Support struct functions. */
2691 
2692 		/* FALLTHROUGH */
2693 	default:
2694 		db_printf("Format:%s Length:%zu Dump:0x",
2695 		    g_ddb_oid->oid_fmt, len);
2696 		while (len-- && (xflag || p < val + 16))
2697 			db_printf("%02x", *p++);
2698 		if (!xflag && len > 16)
2699 			db_printf("...");
2700 		error = 0;
2701 		goto out;
2702 	}
2703 
2704 out:
2705 	req->oldidx += slen;
2706 	return (error);
2707 }
2708 
2709 /*
2710  * Avoid setting new sysctl values from the debugger
2711  */
2712 static int
2713 sysctl_new_ddb(struct sysctl_req *req, void *p, size_t l)
2714 {
2715 
2716 	if (!req->newptr)
2717 		return (0);
2718 
2719 	/* Changing sysctls from the debugger is currently unsupported */
2720 	return (EPERM);
2721 }
2722 
2723 /*
2724  * Run a sysctl handler with the DDB oldfunc and newfunc attached.
2725  * Instead of copying any output to a buffer we'll dump it right to
2726  * the console.
2727  */
2728 static int
2729 db_sysctl(struct sysctl_oid *oidp, int *name, u_int namelen,
2730     void *old, size_t *oldlenp, size_t *retval, int flags)
2731 {
2732 	struct sysctl_req req;
2733 	int error;
2734 
2735 	/* Setup the request */
2736 	bzero(&req, sizeof req);
2737 	req.td = kdb_thread;
2738 	req.oldfunc = sysctl_old_ddb;
2739 	req.newfunc = sysctl_new_ddb;
2740 	req.lock = REQ_UNWIRED;
2741 	if (oldlenp) {
2742 		req.oldlen = *oldlenp;
2743 	}
2744 	req.validlen = req.oldlen;
2745 	if (old) {
2746 		req.oldptr = old;
2747 	}
2748 
2749 	/* Setup our globals for sysctl_old_ddb */
2750 	g_ddb_oid = oidp;
2751 	g_ddb_sysctl_flags = flags;
2752 	g_ddb_sysctl_printed = 0;
2753 
2754 	error = sysctl_root(0, name, namelen, &req);
2755 
2756 	/* Reset globals */
2757 	g_ddb_oid = NULL;
2758 	g_ddb_sysctl_flags = 0;
2759 
2760 	if (retval) {
2761 		if (req.oldptr && req.oldidx > req.validlen)
2762 			*retval = req.validlen;
2763 		else
2764 			*retval = req.oldidx;
2765 	}
2766 	return (error);
2767 }
2768 
2769 /*
2770  * Show a sysctl's name
2771  */
2772 static void
2773 db_show_oid_name(int *oid, size_t nlen)
2774 {
2775 	struct sysctl_oid *oidp;
2776 	int qoid[CTL_MAXNAME+2];
2777 	int error;
2778 
2779 	qoid[0] = 0;
2780 	memcpy(qoid + 2, oid, nlen * sizeof(int));
2781 	qoid[1] = 1;
2782 
2783 	error = sysctl_find_oid(qoid, nlen + 2, &oidp, NULL, NULL);
2784 	if (error)
2785 		db_error("sysctl name oid");
2786 
2787 	error = db_sysctl(oidp, qoid, nlen + 2, NULL, NULL, NULL, 0);
2788 	if (error)
2789 		db_error("sysctl name");
2790 }
2791 
2792 /*
2793  * Check to see if an OID is safe to print from ddb.
2794  */
2795 static bool
2796 db_oid_safe(const struct sysctl_oid *oidp)
2797 {
2798 	for (unsigned int i = 0; i < nitems(db_safe_handlers); ++i) {
2799 		if (oidp->oid_handler == db_safe_handlers[i])
2800 			return (true);
2801 	}
2802 
2803 	return (false);
2804 }
2805 
2806 /*
2807  * Show a sysctl at a specific OID
2808  * Compare to the input handling in show_var from sbin/sysctl/sysctl.c
2809  */
2810 static int
2811 db_show_oid(struct sysctl_oid *oidp, int *oid, size_t nlen, int flags)
2812 {
2813 	int error, xflag, oflag, Nflag, nflag;
2814 	size_t len;
2815 
2816 	xflag = flags & DB_SYSCTL_HEX;
2817 	oflag = flags & DB_SYSCTL_OPAQUE;
2818 	nflag = flags & DB_SYSCTL_VALUE_ONLY;
2819 	Nflag = flags & DB_SYSCTL_NAME_ONLY;
2820 
2821 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_OPAQUE &&
2822 	    (!xflag && !oflag))
2823 		return (0);
2824 
2825 	if (Nflag) {
2826 		db_show_oid_name(oid, nlen);
2827 		error = 0;
2828 		goto out;
2829 	}
2830 
2831 	if (!nflag) {
2832 		db_show_oid_name(oid, nlen);
2833 		db_printf(": ");
2834 	}
2835 
2836 	if ((flags & DB_SYSCTL_SAFE_ONLY) && !db_oid_safe(oidp)) {
2837 		db_printf("Skipping, unsafe to print while recursing.");
2838 		error = 0;
2839 		goto out;
2840 	}
2841 
2842 	/* Try once, and ask about the size */
2843 	len = 0;
2844 	error = db_sysctl(oidp, oid, nlen,
2845 	    NULL, NULL, &len, flags);
2846 	if (error)
2847 		goto out;
2848 
2849 	if (!g_ddb_sysctl_printed)
2850 		/* Lie about the size */
2851 		error = db_sysctl(oidp, oid, nlen,
2852 		    (void *) 1, &len, NULL, flags);
2853 
2854 out:
2855 	db_printf("\n");
2856 	return (error);
2857 }
2858 
2859 /*
2860  * Show all sysctls under a specific OID
2861  * Compare to sysctl_all from sbin/sysctl/sysctl.c
2862  */
2863 static int
2864 db_show_sysctl_all(int *oid, size_t len, int flags)
2865 {
2866 	struct sysctl_oid *oidp;
2867 	int name1[CTL_MAXNAME + 2], name2[CTL_MAXNAME + 2];
2868 	size_t l1, l2;
2869 
2870 	name1[0] = CTL_SYSCTL;
2871 	name1[1] = CTL_SYSCTL_NEXT;
2872 	l1 = 2;
2873 	if (len) {
2874 		memcpy(name1 + 2, oid, len * sizeof(int));
2875 		l1 += len;
2876 	} else {
2877 		name1[2] = CTL_KERN;
2878 		l1++;
2879 	}
2880 	for (;;) {
2881 		int i, error;
2882 
2883 		l2 = sizeof(name2);
2884 		error = kernel_sysctl(kdb_thread, name1, l1,
2885 		    name2, &l2, NULL, 0, &l2, 0);
2886 		if (error != 0) {
2887 			if (error == ENOENT)
2888 				return (0);
2889 			else
2890 				db_error("sysctl(next)");
2891 		}
2892 
2893 		l2 /= sizeof(int);
2894 
2895 		if (l2 < (unsigned int)len)
2896 			return (0);
2897 
2898 		for (i = 0; i < len; i++)
2899 			if (name2[i] != oid[i])
2900 				return (0);
2901 
2902 		/* Find the OID in question */
2903 		error = sysctl_find_oid(name2, l2, &oidp, NULL, NULL);
2904 		if (error)
2905 			return (error);
2906 
2907 		i = db_show_oid(oidp, name2, l2, flags | DB_SYSCTL_SAFE_ONLY);
2908 
2909 		if (db_pager_quit)
2910 			return (0);
2911 
2912 		memcpy(name1+2, name2, l2 * sizeof(int));
2913 		l1 = 2 + l2;
2914 	}
2915 }
2916 
2917 /*
2918  * Show a sysctl by its user facing string
2919  */
2920 static int
2921 db_sysctlbyname(char *name, int flags)
2922 {
2923 	struct sysctl_oid *oidp;
2924 	int oid[CTL_MAXNAME];
2925 	int error, nlen;
2926 
2927 	error = name2oid(name, oid, &nlen, &oidp);
2928 	if (error) {
2929 		return (error);
2930 	}
2931 
2932 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
2933 		db_show_sysctl_all(oid, nlen, flags);
2934 	} else {
2935 		error = db_show_oid(oidp, oid, nlen, flags);
2936 	}
2937 
2938 	return (error);
2939 }
2940 
2941 static void
2942 db_sysctl_cmd_usage(void)
2943 {
2944 	db_printf(
2945 	    " sysctl [/Nnox] <sysctl>					    \n"
2946 	    "								    \n"
2947 	    " <sysctl> The name of the sysctl to show.			    \n"
2948 	    "								    \n"
2949 	    " Show a sysctl by hooking into SYSCTL_IN and SYSCTL_OUT.	    \n"
2950 	    " This will work for most sysctls, but should not be used	    \n"
2951 	    " with sysctls that are known to malloc.			    \n"
2952 	    "								    \n"
2953 	    " While recursing any \"unsafe\" sysctls will be skipped.	    \n"
2954 	    " Call sysctl directly on the sysctl to try printing the	    \n"
2955 	    " skipped sysctl. This is unsafe and may make the ddb	    \n"
2956 	    " session unusable.						    \n"
2957 	    "								    \n"
2958 	    " Arguments:						    \n"
2959 	    "	/N	Display only the name of the sysctl.		    \n"
2960 	    "	/n	Display only the value of the sysctl.		    \n"
2961 	    "	/o	Display opaque values.				    \n"
2962 	    "	/x	Display the sysctl in hex.			    \n"
2963 	    "								    \n"
2964 	    "For example:						    \n"
2965 	    "sysctl vm.v_free_min					    \n"
2966 	    "vn.v_free_min: 12669					    \n"
2967 	    );
2968 }
2969 
2970 /*
2971  * Show a specific sysctl similar to sysctl (8).
2972  */
2973 DB_COMMAND_FLAGS(sysctl, db_sysctl_cmd, CS_OWN)
2974 {
2975 	char name[TOK_STRING_SIZE];
2976 	int error, i, t, flags;
2977 
2978 	/* Parse the modifiers */
2979 	t = db_read_token();
2980 	if (t == tSLASH || t == tMINUS) {
2981 		t = db_read_token();
2982 		if (t != tIDENT) {
2983 			db_printf("Bad modifier\n");
2984 			error = EINVAL;
2985 			goto out;
2986 		}
2987 		db_strcpy(modif, db_tok_string);
2988 	}
2989 	else {
2990 		db_unread_token(t);
2991 		modif[0] = '\0';
2992 	}
2993 
2994 	flags = 0;
2995 	for (i = 0; i < nitems(db_sysctl_modifs); i++) {
2996 		if (strchr(modif, db_sysctl_modifs[i])) {
2997 			flags |= db_sysctl_modif_values[i];
2998 		}
2999 	}
3000 
3001 	/* Parse the sysctl names */
3002 	t = db_read_token();
3003 	if (t != tIDENT) {
3004 		db_printf("Need sysctl name\n");
3005 		error = EINVAL;
3006 		goto out;
3007 	}
3008 
3009 	/* Copy the name into a temporary buffer */
3010 	db_strcpy(name, db_tok_string);
3011 
3012 	/* Ensure there is no trailing cruft */
3013 	t = db_read_token();
3014 	if (t != tEOL) {
3015 		db_printf("Unexpected sysctl argument\n");
3016 		error = EINVAL;
3017 		goto out;
3018 	}
3019 
3020 	error = db_sysctlbyname(name, flags);
3021 	if (error == ENOENT) {
3022 		db_printf("unknown oid: '%s'\n", db_tok_string);
3023 		goto out;
3024 	} else if (error) {
3025 		db_printf("%s: error: %d\n", db_tok_string, error);
3026 		goto out;
3027 	}
3028 
3029 out:
3030 	/* Ensure we eat all of our text */
3031 	db_flush_lex();
3032 
3033 	if (error == EINVAL) {
3034 		db_sysctl_cmd_usage();
3035 	}
3036 }
3037 
3038 #endif /* DDB */
3039