xref: /freebsd/sys/kern/kern_sysctl.c (revision de56ac880999cc5460f68096ff0ad1b58a3433ce)
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 RB_GENERATE(sysctl_oid_list, sysctl_oid, oid_link, cmp_sysctl_oid);
88 
89 /*
90  * The sysctllock protects the MIB tree.  It also protects sysctl
91  * contexts used with dynamic sysctls.  The sysctl_register_oid() and
92  * sysctl_unregister_oid() routines require the sysctllock to already
93  * be held, so the sysctl_wlock() and sysctl_wunlock() routines are
94  * provided for the few places in the kernel which need to use that
95  * API rather than using the dynamic API.  Use of the dynamic API is
96  * strongly encouraged for most code.
97  *
98  * The sysctlmemlock is used to limit the amount of user memory wired for
99  * sysctl requests.  This is implemented by serializing any userland
100  * sysctl requests larger than a single page via an exclusive lock.
101  *
102  * The sysctlstringlock is used to protect concurrent access to writable
103  * string nodes in sysctl_handle_string().
104  */
105 static struct rmlock sysctllock;
106 static struct sx __exclusive_cache_line sysctlmemlock;
107 static struct sx sysctlstringlock;
108 
109 #define	SYSCTL_WLOCK()		rm_wlock(&sysctllock)
110 #define	SYSCTL_WUNLOCK()	rm_wunlock(&sysctllock)
111 #define	SYSCTL_RLOCK(tracker)	rm_rlock(&sysctllock, (tracker))
112 #define	SYSCTL_RUNLOCK(tracker)	rm_runlock(&sysctllock, (tracker))
113 #define	SYSCTL_WLOCKED()	rm_wowned(&sysctllock)
114 #define	SYSCTL_ASSERT_LOCKED()	rm_assert(&sysctllock, RA_LOCKED)
115 #define	SYSCTL_ASSERT_WLOCKED()	rm_assert(&sysctllock, RA_WLOCKED)
116 #define	SYSCTL_ASSERT_RLOCKED()	rm_assert(&sysctllock, RA_RLOCKED)
117 #define	SYSCTL_INIT()		rm_init_flags(&sysctllock, "sysctl lock", \
118 				    RM_SLEEPABLE)
119 #define	SYSCTL_SLEEP(ch, wmesg, timo)					\
120 				rm_sleep(ch, &sysctllock, 0, wmesg, timo)
121 
122 static int sysctl_root(SYSCTL_HANDLER_ARGS);
123 
124 /* Root list */
125 struct sysctl_oid_list sysctl__children = RB_INITIALIZER(&sysctl__children);
126 
127 static char*	sysctl_escape_name(const char*);
128 static int	sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del,
129 		    int recurse);
130 static int	sysctl_old_kernel(struct sysctl_req *, const void *, size_t);
131 static int	sysctl_new_kernel(struct sysctl_req *, void *, size_t);
132 
133 static struct sysctl_oid *
134 sysctl_find_oidname(const char *name, struct sysctl_oid_list *list)
135 {
136 	struct sysctl_oid *oidp;
137 
138 	SYSCTL_ASSERT_LOCKED();
139 	SYSCTL_FOREACH(oidp, list) {
140 		if (strcmp(oidp->oid_name, name) == 0) {
141 			return (oidp);
142 		}
143 	}
144 	return (NULL);
145 }
146 
147 /*
148  * Initialization of the MIB tree.
149  *
150  * Order by number in each list.
151  */
152 void
153 sysctl_wlock(void)
154 {
155 
156 	SYSCTL_WLOCK();
157 }
158 
159 void
160 sysctl_wunlock(void)
161 {
162 
163 	SYSCTL_WUNLOCK();
164 }
165 
166 static int
167 sysctl_root_handler_locked(struct sysctl_oid *oid, void *arg1, intmax_t arg2,
168     struct sysctl_req *req, struct rm_priotracker *tracker)
169 {
170 	int error;
171 
172 	if (oid->oid_kind & CTLFLAG_DYN)
173 		atomic_add_int(&oid->oid_running, 1);
174 
175 	if (tracker != NULL)
176 		SYSCTL_RUNLOCK(tracker);
177 	else
178 		SYSCTL_WUNLOCK();
179 
180 	/*
181 	 * Treat set CTLFLAG_NEEDGIANT and unset CTLFLAG_MPSAFE flags the same,
182 	 * untill we're ready to remove all traces of Giant from sysctl(9).
183 	 */
184 	if ((oid->oid_kind & CTLFLAG_NEEDGIANT) ||
185 	    (!(oid->oid_kind & CTLFLAG_MPSAFE)))
186 		mtx_lock(&Giant);
187 	error = oid->oid_handler(oid, arg1, arg2, req);
188 	if ((oid->oid_kind & CTLFLAG_NEEDGIANT) ||
189 	    (!(oid->oid_kind & CTLFLAG_MPSAFE)))
190 		mtx_unlock(&Giant);
191 
192 	KFAIL_POINT_ERROR(_debug_fail_point, sysctl_running, error);
193 
194 	if (tracker != NULL)
195 		SYSCTL_RLOCK(tracker);
196 	else
197 		SYSCTL_WLOCK();
198 
199 	if (oid->oid_kind & CTLFLAG_DYN) {
200 		if (atomic_fetchadd_int(&oid->oid_running, -1) == 1 &&
201 		    (oid->oid_kind & CTLFLAG_DYING) != 0)
202 			wakeup(&oid->oid_running);
203 	}
204 
205 	return (error);
206 }
207 
208 static void
209 sysctl_load_tunable_by_oid_locked(struct sysctl_oid *oidp)
210 {
211 	struct sysctl_req req;
212 	struct sysctl_oid *curr;
213 	char *penv = NULL;
214 	char path[96];
215 	ssize_t rem = sizeof(path);
216 	ssize_t len;
217 	uint8_t data[512] __aligned(sizeof(uint64_t));
218 	int size;
219 	int error;
220 
221 	path[--rem] = 0;
222 
223 	for (curr = oidp; curr != NULL; curr = SYSCTL_PARENT(curr)) {
224 		len = strlen(curr->oid_name);
225 		rem -= len;
226 		if (curr != oidp)
227 			rem -= 1;
228 		if (rem < 0) {
229 			printf("OID path exceeds %d bytes\n", (int)sizeof(path));
230 			return;
231 		}
232 		memcpy(path + rem, curr->oid_name, len);
233 		if (curr != oidp)
234 			path[rem + len] = '.';
235 	}
236 
237 	memset(&req, 0, sizeof(req));
238 
239 	req.td = curthread;
240 	req.oldfunc = sysctl_old_kernel;
241 	req.newfunc = sysctl_new_kernel;
242 	req.lock = REQ_UNWIRED;
243 
244 	switch (oidp->oid_kind & CTLTYPE) {
245 	case CTLTYPE_INT:
246 		if (getenv_array(path + rem, data, sizeof(data), &size,
247 		    sizeof(int), GETENV_SIGNED) == 0)
248 			return;
249 		req.newlen = size;
250 		req.newptr = data;
251 		break;
252 	case CTLTYPE_UINT:
253 		if (getenv_array(path + rem, data, sizeof(data), &size,
254 		    sizeof(int), GETENV_UNSIGNED) == 0)
255 			return;
256 		req.newlen = size;
257 		req.newptr = data;
258 		break;
259 	case CTLTYPE_LONG:
260 		if (getenv_array(path + rem, data, sizeof(data), &size,
261 		    sizeof(long), GETENV_SIGNED) == 0)
262 			return;
263 		req.newlen = size;
264 		req.newptr = data;
265 		break;
266 	case CTLTYPE_ULONG:
267 		if (getenv_array(path + rem, data, sizeof(data), &size,
268 		    sizeof(long), GETENV_UNSIGNED) == 0)
269 			return;
270 		req.newlen = size;
271 		req.newptr = data;
272 		break;
273 	case CTLTYPE_S8:
274 		if (getenv_array(path + rem, data, sizeof(data), &size,
275 		    sizeof(int8_t), GETENV_SIGNED) == 0)
276 			return;
277 		req.newlen = size;
278 		req.newptr = data;
279 		break;
280 	case CTLTYPE_S16:
281 		if (getenv_array(path + rem, data, sizeof(data), &size,
282 		    sizeof(int16_t), GETENV_SIGNED) == 0)
283 			return;
284 		req.newlen = size;
285 		req.newptr = data;
286 		break;
287 	case CTLTYPE_S32:
288 		if (getenv_array(path + rem, data, sizeof(data), &size,
289 		    sizeof(int32_t), GETENV_SIGNED) == 0)
290 			return;
291 		req.newlen = size;
292 		req.newptr = data;
293 		break;
294 	case CTLTYPE_S64:
295 		if (getenv_array(path + rem, data, sizeof(data), &size,
296 		    sizeof(int64_t), GETENV_SIGNED) == 0)
297 			return;
298 		req.newlen = size;
299 		req.newptr = data;
300 		break;
301 	case CTLTYPE_U8:
302 		if (getenv_array(path + rem, data, sizeof(data), &size,
303 		    sizeof(uint8_t), GETENV_UNSIGNED) == 0)
304 			return;
305 		req.newlen = size;
306 		req.newptr = data;
307 		break;
308 	case CTLTYPE_U16:
309 		if (getenv_array(path + rem, data, sizeof(data), &size,
310 		    sizeof(uint16_t), GETENV_UNSIGNED) == 0)
311 			return;
312 		req.newlen = size;
313 		req.newptr = data;
314 		break;
315 	case CTLTYPE_U32:
316 		if (getenv_array(path + rem, data, sizeof(data), &size,
317 		    sizeof(uint32_t), GETENV_UNSIGNED) == 0)
318 			return;
319 		req.newlen = size;
320 		req.newptr = data;
321 		break;
322 	case CTLTYPE_U64:
323 		if (getenv_array(path + rem, data, sizeof(data), &size,
324 		    sizeof(uint64_t), GETENV_UNSIGNED) == 0)
325 			return;
326 		req.newlen = size;
327 		req.newptr = data;
328 		break;
329 	case CTLTYPE_STRING:
330 		penv = kern_getenv(path + rem);
331 		if (penv == NULL)
332 			return;
333 		req.newlen = strlen(penv);
334 		req.newptr = penv;
335 		break;
336 	default:
337 		return;
338 	}
339 	error = sysctl_root_handler_locked(oidp, oidp->oid_arg1,
340 	    oidp->oid_arg2, &req, NULL);
341 	if (error != 0)
342 		printf("Setting sysctl %s failed: %d\n", path + rem, error);
343 	if (penv != NULL)
344 		freeenv(penv);
345 }
346 
347 /*
348  * Locate the path to a given oid.  Returns the length of the resulting path,
349  * or -1 if the oid was not found.  nodes must have room for CTL_MAXNAME
350  * elements and be NULL initialized.
351  */
352 static int
353 sysctl_search_oid(struct sysctl_oid **nodes, struct sysctl_oid *needle)
354 {
355 	int indx;
356 
357 	SYSCTL_ASSERT_LOCKED();
358 	indx = 0;
359 	/*
360 	 * Do a depth-first search of the oid tree, looking for 'needle'. Start
361 	 * with the first child of the root.
362 	 */
363 	nodes[indx] = RB_MIN(sysctl_oid_list, &sysctl__children);
364 	for (;;) {
365 		if (nodes[indx] == needle)
366 			return (indx + 1);
367 
368 		if (nodes[indx] == NULL) {
369 			/* Node has no more siblings, so back up to parent. */
370 			if (indx-- == 0) {
371 				/* Retreat to root, so give up. */
372 				break;
373 			}
374 		} else if ((nodes[indx]->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
375 			/* Node has children. */
376 			if (++indx == CTL_MAXNAME) {
377 				/* Max search depth reached, so give up. */
378 				break;
379 			}
380 			/* Start with the first child. */
381 			nodes[indx] = RB_MIN(sysctl_oid_list,
382 			    &nodes[indx - 1]->oid_children);
383 			continue;
384 		}
385 		/* Consider next sibling. */
386 		nodes[indx] = RB_NEXT(sysctl_oid_list, NULL, nodes[indx]);
387 	}
388 	return (-1);
389 }
390 
391 static void
392 sysctl_warn_reuse(const char *func, struct sysctl_oid *leaf)
393 {
394 	struct sysctl_oid *nodes[CTL_MAXNAME];
395 	char buf[128];
396 	struct sbuf sb;
397 	int rc, i;
398 
399 	(void)sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN | SBUF_INCLUDENUL);
400 	sbuf_set_drain(&sb, sbuf_printf_drain, NULL);
401 
402 	sbuf_printf(&sb, "%s: can't re-use a leaf (", __func__);
403 
404 	memset(nodes, 0, sizeof(nodes));
405 	rc = sysctl_search_oid(nodes, leaf);
406 	if (rc > 0) {
407 		for (i = 0; i < rc; i++)
408 			sbuf_printf(&sb, "%s%.*s", nodes[i]->oid_name,
409 			    i != (rc - 1), ".");
410 	} else {
411 		sbuf_printf(&sb, "%s", leaf->oid_name);
412 	}
413 	sbuf_printf(&sb, ")!\n");
414 
415 	(void)sbuf_finish(&sb);
416 }
417 
418 #ifdef SYSCTL_DEBUG
419 static int
420 sysctl_reuse_test(SYSCTL_HANDLER_ARGS)
421 {
422 	struct rm_priotracker tracker;
423 
424 	SYSCTL_RLOCK(&tracker);
425 	sysctl_warn_reuse(__func__, oidp);
426 	SYSCTL_RUNLOCK(&tracker);
427 	return (0);
428 }
429 SYSCTL_PROC(_sysctl, OID_AUTO, reuse_test,
430     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 0, sysctl_reuse_test, "-",
431     "");
432 #endif
433 
434 void
435 sysctl_register_oid(struct sysctl_oid *oidp)
436 {
437 	struct sysctl_oid_list *parent = oidp->oid_parent;
438 	struct sysctl_oid *p, key;
439 	int oid_number;
440 	int timeout = 2;
441 
442 	/*
443 	 * First check if another oid with the same name already
444 	 * exists in the parent's list.
445 	 */
446 	SYSCTL_ASSERT_WLOCKED();
447 	p = sysctl_find_oidname(oidp->oid_name, parent);
448 	if (p != NULL) {
449 		if ((p->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
450 			p->oid_refcnt++;
451 			return;
452 		} else {
453 			sysctl_warn_reuse(__func__, p);
454 			return;
455 		}
456 	}
457 	/* get current OID number */
458 	oid_number = oidp->oid_number;
459 
460 #if (OID_AUTO >= 0)
461 #error "OID_AUTO is expected to be a negative value"
462 #endif
463 	/*
464 	 * Any negative OID number qualifies as OID_AUTO. Valid OID
465 	 * numbers should always be positive.
466 	 *
467 	 * NOTE: DO NOT change the starting value here, change it in
468 	 * <sys/sysctl.h>, and make sure it is at least 256 to
469 	 * accommodate e.g. net.inet.raw as a static sysctl node.
470 	 */
471 	if (oid_number < 0) {
472 		static int newoid;
473 
474 		/*
475 		 * By decrementing the next OID number we spend less
476 		 * time inserting the OIDs into a sorted list.
477 		 */
478 		if (--newoid < CTL_AUTO_START)
479 			newoid = 0x7fffffff;
480 
481 		oid_number = newoid;
482 	}
483 
484 	/*
485 	 * Insert the OID into the parent's list sorted by OID number.
486 	 */
487 	key.oid_number = oid_number;
488 	p = RB_NFIND(sysctl_oid_list, parent, &key);
489 	while (p != NULL && oid_number == p->oid_number) {
490 		/* get the next valid OID number */
491 		if (oid_number < CTL_AUTO_START ||
492 		    oid_number == 0x7fffffff) {
493 			/* wraparound - restart */
494 			oid_number = CTL_AUTO_START;
495 			/* don't loop forever */
496 			if (!timeout--)
497 				panic("sysctl: Out of OID numbers\n");
498 			key.oid_number = oid_number;
499 			p = RB_NFIND(sysctl_oid_list, parent, &key);
500 			continue;
501 		}
502 		p = RB_NEXT(sysctl_oid_list, NULL, p);
503 		oid_number++;
504 	}
505 	/* check for non-auto OID number collision */
506 	if (oidp->oid_number >= 0 && oidp->oid_number < CTL_AUTO_START &&
507 	    oid_number >= CTL_AUTO_START) {
508 		printf("sysctl: OID number(%d) is already in use for '%s'\n",
509 		    oidp->oid_number, oidp->oid_name);
510 	}
511 	/* update the OID number, if any */
512 	oidp->oid_number = oid_number;
513 	RB_INSERT(sysctl_oid_list, parent, oidp);
514 
515 	if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE &&
516 #ifdef VIMAGE
517 	    (oidp->oid_kind & CTLFLAG_VNET) == 0 &&
518 #endif
519 	    (oidp->oid_kind & CTLFLAG_TUN) != 0 &&
520 	    (oidp->oid_kind & CTLFLAG_NOFETCH) == 0) {
521 		/* only fetch value once */
522 		oidp->oid_kind |= CTLFLAG_NOFETCH;
523 		/* try to fetch value from kernel environment */
524 		sysctl_load_tunable_by_oid_locked(oidp);
525 	}
526 }
527 
528 void
529 sysctl_register_disabled_oid(struct sysctl_oid *oidp)
530 {
531 
532 	/*
533 	 * Mark the leaf as dormant if it's not to be immediately enabled.
534 	 * We do not disable nodes as they can be shared between modules
535 	 * and it is always safe to access a node.
536 	 */
537 	KASSERT((oidp->oid_kind & CTLFLAG_DORMANT) == 0,
538 	    ("internal flag is set in oid_kind"));
539 	if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
540 		oidp->oid_kind |= CTLFLAG_DORMANT;
541 	sysctl_register_oid(oidp);
542 }
543 
544 void
545 sysctl_enable_oid(struct sysctl_oid *oidp)
546 {
547 
548 	SYSCTL_ASSERT_WLOCKED();
549 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
550 		KASSERT((oidp->oid_kind & CTLFLAG_DORMANT) == 0,
551 		    ("sysctl node is marked as dormant"));
552 		return;
553 	}
554 	KASSERT((oidp->oid_kind & CTLFLAG_DORMANT) != 0,
555 	    ("enabling already enabled sysctl oid"));
556 	oidp->oid_kind &= ~CTLFLAG_DORMANT;
557 }
558 
559 void
560 sysctl_unregister_oid(struct sysctl_oid *oidp)
561 {
562 	int error;
563 
564 	SYSCTL_ASSERT_WLOCKED();
565 	if (oidp->oid_number == OID_AUTO) {
566 		error = EINVAL;
567 	} else {
568 		error = ENOENT;
569 		if (RB_REMOVE(sysctl_oid_list, oidp->oid_parent, oidp))
570 			error = 0;
571 	}
572 
573 	/*
574 	 * This can happen when a module fails to register and is
575 	 * being unloaded afterwards.  It should not be a panic()
576 	 * for normal use.
577 	 */
578 	if (error) {
579 		printf("%s: failed(%d) to unregister sysctl(%s)\n",
580 		    __func__, error, oidp->oid_name);
581 	}
582 }
583 
584 /* Initialize a new context to keep track of dynamically added sysctls. */
585 int
586 sysctl_ctx_init(struct sysctl_ctx_list *c)
587 {
588 
589 	if (c == NULL) {
590 		return (EINVAL);
591 	}
592 
593 	/*
594 	 * No locking here, the caller is responsible for not adding
595 	 * new nodes to a context until after this function has
596 	 * returned.
597 	 */
598 	TAILQ_INIT(c);
599 	return (0);
600 }
601 
602 /* Free the context, and destroy all dynamic oids registered in this context */
603 int
604 sysctl_ctx_free(struct sysctl_ctx_list *clist)
605 {
606 	struct sysctl_ctx_entry *e, *e1;
607 	int error;
608 
609 	error = 0;
610 	/*
611 	 * First perform a "dry run" to check if it's ok to remove oids.
612 	 * XXX FIXME
613 	 * XXX This algorithm is a hack. But I don't know any
614 	 * XXX better solution for now...
615 	 */
616 	SYSCTL_WLOCK();
617 	TAILQ_FOREACH(e, clist, link) {
618 		error = sysctl_remove_oid_locked(e->entry, 0, 0);
619 		if (error)
620 			break;
621 	}
622 	/*
623 	 * Restore deregistered entries, either from the end,
624 	 * or from the place where error occurred.
625 	 * e contains the entry that was not unregistered
626 	 */
627 	if (error)
628 		e1 = TAILQ_PREV(e, sysctl_ctx_list, link);
629 	else
630 		e1 = TAILQ_LAST(clist, sysctl_ctx_list);
631 	while (e1 != NULL) {
632 		sysctl_register_oid(e1->entry);
633 		e1 = TAILQ_PREV(e1, sysctl_ctx_list, link);
634 	}
635 	if (error) {
636 		SYSCTL_WUNLOCK();
637 		return(EBUSY);
638 	}
639 	/* Now really delete the entries */
640 	e = TAILQ_FIRST(clist);
641 	while (e != NULL) {
642 		e1 = TAILQ_NEXT(e, link);
643 		error = sysctl_remove_oid_locked(e->entry, 1, 0);
644 		if (error)
645 			panic("sysctl_remove_oid: corrupt tree, entry: %s",
646 			    e->entry->oid_name);
647 		free(e, M_SYSCTLOID);
648 		e = e1;
649 	}
650 	SYSCTL_WUNLOCK();
651 	return (error);
652 }
653 
654 /* Add an entry to the context */
655 struct sysctl_ctx_entry *
656 sysctl_ctx_entry_add(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
657 {
658 	struct sysctl_ctx_entry *e;
659 
660 	SYSCTL_ASSERT_WLOCKED();
661 	if (clist == NULL || oidp == NULL)
662 		return(NULL);
663 	e = malloc(sizeof(struct sysctl_ctx_entry), M_SYSCTLOID, M_WAITOK);
664 	e->entry = oidp;
665 	TAILQ_INSERT_HEAD(clist, e, link);
666 	return (e);
667 }
668 
669 /* Find an entry in the context */
670 struct sysctl_ctx_entry *
671 sysctl_ctx_entry_find(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
672 {
673 	struct sysctl_ctx_entry *e;
674 
675 	SYSCTL_ASSERT_WLOCKED();
676 	if (clist == NULL || oidp == NULL)
677 		return(NULL);
678 	TAILQ_FOREACH(e, clist, link) {
679 		if (e->entry == oidp)
680 			return(e);
681 	}
682 	return (e);
683 }
684 
685 /*
686  * Delete an entry from the context.
687  * NOTE: this function doesn't free oidp! You have to remove it
688  * with sysctl_remove_oid().
689  */
690 int
691 sysctl_ctx_entry_del(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
692 {
693 	struct sysctl_ctx_entry *e;
694 
695 	if (clist == NULL || oidp == NULL)
696 		return (EINVAL);
697 	SYSCTL_WLOCK();
698 	e = sysctl_ctx_entry_find(clist, oidp);
699 	if (e != NULL) {
700 		TAILQ_REMOVE(clist, e, link);
701 		SYSCTL_WUNLOCK();
702 		free(e, M_SYSCTLOID);
703 		return (0);
704 	} else {
705 		SYSCTL_WUNLOCK();
706 		return (ENOENT);
707 	}
708 }
709 
710 /*
711  * Remove dynamically created sysctl trees.
712  * oidp - top of the tree to be removed
713  * del - if 0 - just deregister, otherwise free up entries as well
714  * recurse - if != 0 traverse the subtree to be deleted
715  */
716 int
717 sysctl_remove_oid(struct sysctl_oid *oidp, int del, int recurse)
718 {
719 	int error;
720 
721 	SYSCTL_WLOCK();
722 	error = sysctl_remove_oid_locked(oidp, del, recurse);
723 	SYSCTL_WUNLOCK();
724 	return (error);
725 }
726 
727 int
728 sysctl_remove_name(struct sysctl_oid *parent, const char *name,
729     int del, int recurse)
730 {
731 	struct sysctl_oid *p;
732 	int error;
733 
734 	error = ENOENT;
735 	SYSCTL_WLOCK();
736 	p = sysctl_find_oidname(name, &parent->oid_children);
737 	if (p)
738 		error = sysctl_remove_oid_locked(p, del, recurse);
739 	SYSCTL_WUNLOCK();
740 
741 	return (error);
742 }
743 
744 /*
745  * Duplicate the provided string, escaping any illegal characters.  The result
746  * must be freed when no longer in use.
747  *
748  * The list of illegal characters is ".".
749  */
750 static char*
751 sysctl_escape_name(const char* orig)
752 {
753 	int i, s = 0, d = 0, nillegals = 0;
754 	char *new;
755 
756 	/* First count the number of illegal characters */
757 	for (i = 0; orig[i] != '\0'; i++) {
758 		if (orig[i] == '.')
759 			nillegals++;
760 	}
761 
762 	/* Allocate storage for new string */
763 	new = malloc(i + 2 * nillegals + 1, M_SYSCTLOID, M_WAITOK);
764 
765 	/* Copy the name, escaping characters as we go */
766 	while (orig[s] != '\0') {
767 		if (orig[s] == '.') {
768 			/* %25 is the hexadecimal representation of '.' */
769 			new[d++] = '%';
770 			new[d++] = '2';
771 			new[d++] = '5';
772 			s++;
773 		} else {
774 			new[d++] = orig[s++];
775 		}
776 	}
777 
778 	/* Finally, nul-terminate */
779 	new[d] = '\0';
780 
781 	return (new);
782 }
783 
784 static int
785 sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del, int recurse)
786 {
787 	struct sysctl_oid *p, *tmp;
788 	int error;
789 
790 	SYSCTL_ASSERT_WLOCKED();
791 	if (oidp == NULL)
792 		return(EINVAL);
793 	if ((oidp->oid_kind & CTLFLAG_DYN) == 0) {
794 		printf("Warning: can't remove non-dynamic nodes (%s)!\n",
795 		    oidp->oid_name);
796 		return (EINVAL);
797 	}
798 	/*
799 	 * WARNING: normal method to do this should be through
800 	 * sysctl_ctx_free(). Use recursing as the last resort
801 	 * method to purge your sysctl tree of leftovers...
802 	 * However, if some other code still references these nodes,
803 	 * it will panic.
804 	 */
805 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
806 		if (oidp->oid_refcnt == 1) {
807 			for(p = RB_MIN(sysctl_oid_list, &oidp->oid_children);
808 			    p != NULL; p = tmp) {
809 				if (!recurse) {
810 					printf("Warning: failed attempt to "
811 					    "remove oid %s with child %s\n",
812 					    oidp->oid_name, p->oid_name);
813 					return (ENOTEMPTY);
814 				}
815 				tmp = RB_NEXT(sysctl_oid_list,
816 				    &oidp->oid_children, p);
817 				error = sysctl_remove_oid_locked(p, del,
818 				    recurse);
819 				if (error)
820 					return (error);
821 			}
822 		}
823 	}
824 	if (oidp->oid_refcnt > 1 ) {
825 		oidp->oid_refcnt--;
826 	} else {
827 		if (oidp->oid_refcnt == 0) {
828 			printf("Warning: bad oid_refcnt=%u (%s)!\n",
829 				oidp->oid_refcnt, oidp->oid_name);
830 			return (EINVAL);
831 		}
832 		sysctl_unregister_oid(oidp);
833 		if (del) {
834 			/*
835 			 * Wait for all threads running the handler to drain.
836 			 * This preserves the previous behavior when the
837 			 * sysctl lock was held across a handler invocation,
838 			 * and is necessary for module unload correctness.
839 			 */
840 			while (oidp->oid_running > 0) {
841 				oidp->oid_kind |= CTLFLAG_DYING;
842 				SYSCTL_SLEEP(&oidp->oid_running, "oidrm", 0);
843 			}
844 			if (oidp->oid_descr)
845 				free(__DECONST(char *, oidp->oid_descr),
846 				    M_SYSCTLOID);
847 			if (oidp->oid_label)
848 				free(__DECONST(char *, oidp->oid_label),
849 				    M_SYSCTLOID);
850 			free(__DECONST(char *, oidp->oid_name), M_SYSCTLOID);
851 			free(oidp, M_SYSCTLOID);
852 		}
853 	}
854 	return (0);
855 }
856 /*
857  * Create new sysctls at run time.
858  * clist may point to a valid context initialized with sysctl_ctx_init().
859  */
860 struct sysctl_oid *
861 sysctl_add_oid(struct sysctl_ctx_list *clist, struct sysctl_oid_list *parent,
862 	int number, const char *name, int kind, void *arg1, intmax_t arg2,
863 	int (*handler)(SYSCTL_HANDLER_ARGS), const char *fmt, const char *descr,
864 	const char *label)
865 {
866 	struct sysctl_oid *oidp;
867 	char *escaped;
868 
869 	/* You have to hook up somewhere.. */
870 	if (parent == NULL)
871 		return(NULL);
872 	escaped = sysctl_escape_name(name);
873 	/* Check if the node already exists, otherwise create it */
874 	SYSCTL_WLOCK();
875 	oidp = sysctl_find_oidname(escaped, parent);
876 	if (oidp != NULL) {
877 		free(escaped, M_SYSCTLOID);
878 		if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
879 			oidp->oid_refcnt++;
880 			/* Update the context */
881 			if (clist != NULL)
882 				sysctl_ctx_entry_add(clist, oidp);
883 			SYSCTL_WUNLOCK();
884 			return (oidp);
885 		} else {
886 			sysctl_warn_reuse(__func__, oidp);
887 			SYSCTL_WUNLOCK();
888 			return (NULL);
889 		}
890 	}
891 	oidp = malloc(sizeof(struct sysctl_oid), M_SYSCTLOID, M_WAITOK|M_ZERO);
892 	oidp->oid_parent = parent;
893 	RB_INIT(&oidp->oid_children);
894 	oidp->oid_number = number;
895 	oidp->oid_refcnt = 1;
896 	oidp->oid_name = escaped;
897 	oidp->oid_handler = handler;
898 	oidp->oid_kind = CTLFLAG_DYN | kind;
899 	oidp->oid_arg1 = arg1;
900 	oidp->oid_arg2 = arg2;
901 	oidp->oid_fmt = fmt;
902 	if (descr != NULL)
903 		oidp->oid_descr = strdup(descr, M_SYSCTLOID);
904 	if (label != NULL)
905 		oidp->oid_label = strdup(label, M_SYSCTLOID);
906 	/* Update the context, if used */
907 	if (clist != NULL)
908 		sysctl_ctx_entry_add(clist, oidp);
909 	/* Register this oid */
910 	sysctl_register_oid(oidp);
911 	SYSCTL_WUNLOCK();
912 	return (oidp);
913 }
914 
915 /*
916  * Rename an existing oid.
917  */
918 void
919 sysctl_rename_oid(struct sysctl_oid *oidp, const char *name)
920 {
921 	char *newname;
922 	char *oldname;
923 
924 	newname = strdup(name, M_SYSCTLOID);
925 	SYSCTL_WLOCK();
926 	oldname = __DECONST(char *, oidp->oid_name);
927 	oidp->oid_name = newname;
928 	SYSCTL_WUNLOCK();
929 	free(oldname, M_SYSCTLOID);
930 }
931 
932 /*
933  * Reparent an existing oid.
934  */
935 int
936 sysctl_move_oid(struct sysctl_oid *oid, struct sysctl_oid_list *parent)
937 {
938 	struct sysctl_oid *oidp;
939 
940 	SYSCTL_WLOCK();
941 	if (oid->oid_parent == parent) {
942 		SYSCTL_WUNLOCK();
943 		return (0);
944 	}
945 	oidp = sysctl_find_oidname(oid->oid_name, parent);
946 	if (oidp != NULL) {
947 		SYSCTL_WUNLOCK();
948 		return (EEXIST);
949 	}
950 	sysctl_unregister_oid(oid);
951 	oid->oid_parent = parent;
952 	oid->oid_number = OID_AUTO;
953 	sysctl_register_oid(oid);
954 	SYSCTL_WUNLOCK();
955 	return (0);
956 }
957 
958 /*
959  * Register the kernel's oids on startup.
960  */
961 SET_DECLARE(sysctl_set, struct sysctl_oid);
962 
963 static void
964 sysctl_register_all(void *arg)
965 {
966 	struct sysctl_oid **oidp;
967 
968 	sx_init(&sysctlmemlock, "sysctl mem");
969 	sx_init(&sysctlstringlock, "sysctl string handler");
970 	SYSCTL_INIT();
971 	SYSCTL_WLOCK();
972 	SET_FOREACH(oidp, sysctl_set)
973 		sysctl_register_oid(*oidp);
974 	SYSCTL_WUNLOCK();
975 }
976 SYSINIT(sysctl, SI_SUB_KMEM, SI_ORDER_FIRST, sysctl_register_all, NULL);
977 
978 /*
979  * "Staff-functions"
980  *
981  * These functions implement a presently undocumented interface
982  * used by the sysctl program to walk the tree, and get the type
983  * so it can print the value.
984  * This interface is under work and consideration, and should probably
985  * be killed with a big axe by the first person who can find the time.
986  * (be aware though, that the proper interface isn't as obvious as it
987  * may seem, there are various conflicting requirements.
988  *
989  * {CTL_SYSCTL, CTL_SYSCTL_DEBUG}		printf the entire MIB-tree.
990  * {CTL_SYSCTL, CTL_SYSCTL_NAME, ...}		return the name of the "..."
991  *						OID.
992  * {CTL_SYSCTL, CTL_SYSCTL_NEXT, ...}		return the next OID, honoring
993  *						CTLFLAG_SKIP.
994  * {CTL_SYSCTL, CTL_SYSCTL_NAME2OID}		return the OID of the name in
995  *						"new"
996  * {CTL_SYSCTL, CTL_SYSCTL_OIDFMT, ...}		return the kind & format info
997  *						for the "..." OID.
998  * {CTL_SYSCTL, CTL_SYSCTL_OIDDESCR, ...}	return the description of the
999  *						"..." OID.
1000  * {CTL_SYSCTL, CTL_SYSCTL_OIDLABEL, ...}	return the aggregation label of
1001  *						the "..." OID.
1002  * {CTL_SYSCTL, CTL_SYSCTL_NEXTNOSKIP, ...}	return the next OID, ignoring
1003  *						CTLFLAG_SKIP.
1004  */
1005 
1006 #ifdef SYSCTL_DEBUG
1007 static void
1008 sysctl_sysctl_debug_dump_node(struct sysctl_oid_list *l, int i)
1009 {
1010 	int k;
1011 	struct sysctl_oid *oidp;
1012 
1013 	SYSCTL_ASSERT_LOCKED();
1014 	SYSCTL_FOREACH(oidp, l) {
1015 		for (k=0; k<i; k++)
1016 			printf(" ");
1017 
1018 		printf("%d %s ", oidp->oid_number, oidp->oid_name);
1019 
1020 		printf("%c%c",
1021 			oidp->oid_kind & CTLFLAG_RD ? 'R':' ',
1022 			oidp->oid_kind & CTLFLAG_WR ? 'W':' ');
1023 
1024 		if (oidp->oid_handler)
1025 			printf(" *Handler");
1026 
1027 		switch (oidp->oid_kind & CTLTYPE) {
1028 			case CTLTYPE_NODE:
1029 				printf(" Node\n");
1030 				if (!oidp->oid_handler) {
1031 					sysctl_sysctl_debug_dump_node(
1032 					    SYSCTL_CHILDREN(oidp), i + 2);
1033 				}
1034 				break;
1035 			case CTLTYPE_INT:    printf(" Int\n"); break;
1036 			case CTLTYPE_UINT:   printf(" u_int\n"); break;
1037 			case CTLTYPE_LONG:   printf(" Long\n"); break;
1038 			case CTLTYPE_ULONG:  printf(" u_long\n"); break;
1039 			case CTLTYPE_STRING: printf(" String\n"); break;
1040 			case CTLTYPE_S8:     printf(" int8_t\n"); break;
1041 			case CTLTYPE_S16:    printf(" int16_t\n"); break;
1042 			case CTLTYPE_S32:    printf(" int32_t\n"); break;
1043 			case CTLTYPE_S64:    printf(" int64_t\n"); break;
1044 			case CTLTYPE_U8:     printf(" uint8_t\n"); break;
1045 			case CTLTYPE_U16:    printf(" uint16_t\n"); break;
1046 			case CTLTYPE_U32:    printf(" uint32_t\n"); break;
1047 			case CTLTYPE_U64:    printf(" uint64_t\n"); break;
1048 			case CTLTYPE_OPAQUE: printf(" Opaque/struct\n"); break;
1049 			default:	     printf("\n");
1050 		}
1051 	}
1052 }
1053 
1054 static int
1055 sysctl_sysctl_debug(SYSCTL_HANDLER_ARGS)
1056 {
1057 	struct rm_priotracker tracker;
1058 	int error;
1059 
1060 	error = priv_check(req->td, PRIV_SYSCTL_DEBUG);
1061 	if (error)
1062 		return (error);
1063 	SYSCTL_RLOCK(&tracker);
1064 	sysctl_sysctl_debug_dump_node(&sysctl__children, 0);
1065 	SYSCTL_RUNLOCK(&tracker);
1066 	return (ENOENT);
1067 }
1068 
1069 SYSCTL_PROC(_sysctl, CTL_SYSCTL_DEBUG, debug, CTLTYPE_STRING | CTLFLAG_RD |
1070     CTLFLAG_MPSAFE, 0, 0, sysctl_sysctl_debug, "-", "");
1071 #endif
1072 
1073 static int
1074 sysctl_sysctl_name(SYSCTL_HANDLER_ARGS)
1075 {
1076 	int *name = (int *) arg1;
1077 	u_int namelen = arg2;
1078 	int error;
1079 	struct sysctl_oid *oid, key;
1080 	struct sysctl_oid_list *lsp = &sysctl__children, *lsp2;
1081 	struct rm_priotracker tracker;
1082 	char buf[10];
1083 
1084 	error = sysctl_wire_old_buffer(req, 0);
1085 	if (error)
1086 		return (error);
1087 
1088 	SYSCTL_RLOCK(&tracker);
1089 	while (namelen) {
1090 		if (!lsp) {
1091 			snprintf(buf,sizeof(buf),"%d",*name);
1092 			if (req->oldidx)
1093 				error = SYSCTL_OUT(req, ".", 1);
1094 			if (!error)
1095 				error = SYSCTL_OUT(req, buf, strlen(buf));
1096 			if (error)
1097 				goto out;
1098 			namelen--;
1099 			name++;
1100 			continue;
1101 		}
1102 		lsp2 = NULL;
1103 		key.oid_number = *name;
1104 		oid = RB_FIND(sysctl_oid_list, lsp, &key);
1105 		if (oid) {
1106 			if (req->oldidx)
1107 				error = SYSCTL_OUT(req, ".", 1);
1108 			if (!error)
1109 				error = SYSCTL_OUT(req, oid->oid_name,
1110 					strlen(oid->oid_name));
1111 			if (error)
1112 				goto out;
1113 
1114 			namelen--;
1115 			name++;
1116 
1117 			if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE &&
1118 				!oid->oid_handler)
1119 				lsp2 = SYSCTL_CHILDREN(oid);
1120 		}
1121 		lsp = lsp2;
1122 	}
1123 	error = SYSCTL_OUT(req, "", 1);
1124  out:
1125 	SYSCTL_RUNLOCK(&tracker);
1126 	return (error);
1127 }
1128 
1129 /*
1130  * XXXRW/JA: Shouldn't return name data for nodes that we don't permit in
1131  * capability mode.
1132  */
1133 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_NAME, name, CTLFLAG_RD |
1134     CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_name, "");
1135 
1136 enum sysctl_iter_action {
1137 	ITER_SIBLINGS,	/* Not matched, continue iterating siblings */
1138 	ITER_CHILDREN,	/* Node has children we need to iterate over them */
1139 	ITER_FOUND,	/* Matching node was found */
1140 };
1141 
1142 /*
1143  * Tries to find the next node for @name and @namelen.
1144  *
1145  * Returns next action to take.
1146  */
1147 static enum sysctl_iter_action
1148 sysctl_sysctl_next_node(struct sysctl_oid *oidp, int *name, unsigned int namelen,
1149     bool honor_skip)
1150 {
1151 
1152 	if ((oidp->oid_kind & CTLFLAG_DORMANT) != 0)
1153 		return (ITER_SIBLINGS);
1154 
1155 	if (honor_skip && (oidp->oid_kind & CTLFLAG_SKIP) != 0)
1156 		return (ITER_SIBLINGS);
1157 
1158 	if (namelen == 0) {
1159 		/*
1160 		 * We have reached a node with a full name match and are
1161 		 * looking for the next oid in its children.
1162 		 *
1163 		 * For CTL_SYSCTL_NEXTNOSKIP we are done.
1164 		 *
1165 		 * For CTL_SYSCTL_NEXT we skip CTLTYPE_NODE (unless it
1166 		 * has a handler) and move on to the children.
1167 		 */
1168 		if (!honor_skip)
1169 			return (ITER_FOUND);
1170 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1171 			return (ITER_FOUND);
1172 		/* If node does not have an iterator, treat it as leaf */
1173 		if (oidp->oid_handler)
1174 			return (ITER_FOUND);
1175 
1176 		/* Report oid as a node to iterate */
1177 		return (ITER_CHILDREN);
1178 	}
1179 
1180 	/*
1181 	 * No match yet. Continue seeking the given name.
1182 	 *
1183 	 * We are iterating in order by oid_number, so skip oids lower
1184 	 * than the one we are looking for.
1185 	 *
1186 	 * When the current oid_number is higher than the one we seek,
1187 	 * that means we have reached the next oid in the sequence and
1188 	 * should return it.
1189 	 *
1190 	 * If the oid_number matches the name at this level then we
1191 	 * have to find a node to continue searching at the next level.
1192 	 */
1193 	if (oidp->oid_number < *name)
1194 		return (ITER_SIBLINGS);
1195 	if (oidp->oid_number > *name) {
1196 		/*
1197 		 * We have reached the next oid.
1198 		 *
1199 		 * For CTL_SYSCTL_NEXTNOSKIP we are done.
1200 		 *
1201 		 * For CTL_SYSCTL_NEXT we skip CTLTYPE_NODE (unless it
1202 		 * has a handler) and move on to the children.
1203 		 */
1204 		if (!honor_skip)
1205 			return (ITER_FOUND);
1206 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1207 			return (ITER_FOUND);
1208 		/* If node does not have an iterator, treat it as leaf */
1209 		if (oidp->oid_handler)
1210 			return (ITER_FOUND);
1211 		return (ITER_CHILDREN);
1212 	}
1213 
1214 	/* match at a current level */
1215 	if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1216 		return (ITER_SIBLINGS);
1217 	if (oidp->oid_handler)
1218 		return (ITER_SIBLINGS);
1219 
1220 	return (ITER_CHILDREN);
1221 }
1222 
1223 /*
1224  * Recursively walk the sysctl subtree at lsp until we find the given name.
1225  * Returns true and fills in next oid data in @next and @len if oid is found.
1226  */
1227 static bool
1228 sysctl_sysctl_next_action(struct sysctl_oid_list *lsp, int *name, u_int namelen,
1229     int *next, int *len, int level, bool honor_skip)
1230 {
1231 	struct sysctl_oid_list *next_lsp;
1232 	struct sysctl_oid *oidp = NULL, key;
1233 	bool success = false;
1234 	enum sysctl_iter_action action;
1235 
1236 	SYSCTL_ASSERT_LOCKED();
1237 	/*
1238 	 * Start the search at the requested oid.  But if not found, then scan
1239 	 * through all children.
1240 	 */
1241 	if (namelen > 0) {
1242 		key.oid_number = *name;
1243 		oidp = RB_FIND(sysctl_oid_list, lsp, &key);
1244 	}
1245 	if (!oidp)
1246 		oidp = RB_MIN(sysctl_oid_list, lsp);
1247 	for(; oidp != NULL; oidp = RB_NEXT(sysctl_oid_list, lsp, oidp)) {
1248 		action = sysctl_sysctl_next_node(oidp, name, namelen,
1249 		    honor_skip);
1250 		if (action == ITER_SIBLINGS)
1251 			continue;
1252 		if (action == ITER_FOUND) {
1253 			success = true;
1254 			break;
1255 		}
1256 		KASSERT((action== ITER_CHILDREN), ("ret(%d)!=ITER_CHILDREN", action));
1257 
1258 		next_lsp = SYSCTL_CHILDREN(oidp);
1259 		if (namelen == 0) {
1260 			success = sysctl_sysctl_next_action(next_lsp, NULL, 0,
1261 			    next + 1, len, level + 1, honor_skip);
1262 		} else {
1263 			success = sysctl_sysctl_next_action(next_lsp, name + 1,
1264 			    namelen - 1, next + 1, len, level + 1, honor_skip);
1265 			if (!success) {
1266 
1267 				/*
1268 				 * We maintain the invariant that current node oid
1269 				 * is >= the oid provided in @name.
1270 				 * As there are no usable children at this node,
1271 				 *  current node oid is strictly > than the requested
1272 				 *  oid.
1273 				 * Hence, reduce namelen to 0 to allow for picking first
1274 				 *  nodes/leafs in the next node in list.
1275 				 */
1276 				namelen = 0;
1277 			}
1278 		}
1279 		if (success)
1280 			break;
1281 	}
1282 
1283 	if (success) {
1284 		*next = oidp->oid_number;
1285 		if (level > *len)
1286 			*len = level;
1287 	}
1288 
1289 	return (success);
1290 }
1291 
1292 static int
1293 sysctl_sysctl_next(SYSCTL_HANDLER_ARGS)
1294 {
1295 	int *name = (int *) arg1;
1296 	u_int namelen = arg2;
1297 	int len, error;
1298 	bool success;
1299 	struct sysctl_oid_list *lsp = &sysctl__children;
1300 	struct rm_priotracker tracker;
1301 	int next[CTL_MAXNAME];
1302 
1303 	len = 0;
1304 	SYSCTL_RLOCK(&tracker);
1305 	success = sysctl_sysctl_next_action(lsp, name, namelen, next, &len, 1,
1306 	    oidp->oid_number == CTL_SYSCTL_NEXT);
1307 	SYSCTL_RUNLOCK(&tracker);
1308 	if (!success)
1309 		return (ENOENT);
1310 	error = SYSCTL_OUT(req, next, len * sizeof (int));
1311 	return (error);
1312 }
1313 
1314 /*
1315  * XXXRW/JA: Shouldn't return next data for nodes that we don't permit in
1316  * capability mode.
1317  */
1318 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_NEXT, next, CTLFLAG_RD |
1319     CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_next, "");
1320 
1321 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_NEXTNOSKIP, nextnoskip, CTLFLAG_RD |
1322     CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_next, "");
1323 
1324 static int
1325 name2oid(char *name, int *oid, int *len, struct sysctl_oid **oidpp)
1326 {
1327 	struct sysctl_oid *oidp;
1328 	struct sysctl_oid_list *lsp = &sysctl__children;
1329 
1330 	SYSCTL_ASSERT_LOCKED();
1331 
1332 	for (*len = 0; *len < CTL_MAXNAME;) {
1333 		oidp = sysctl_find_oidname(strsep(&name, "."), lsp);
1334 		if (oidp == NULL)
1335 			return (ENOENT);
1336 		*oid++ = oidp->oid_number;
1337 		(*len)++;
1338 
1339 		if (name == NULL || *name == '\0') {
1340 			if (oidpp)
1341 				*oidpp = oidp;
1342 			return (0);
1343 		}
1344 
1345 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1346 			break;
1347 
1348 		if (oidp->oid_handler)
1349 			break;
1350 
1351 		lsp = SYSCTL_CHILDREN(oidp);
1352 	}
1353 	return (ENOENT);
1354 }
1355 
1356 static int
1357 sysctl_sysctl_name2oid(SYSCTL_HANDLER_ARGS)
1358 {
1359 	char *p;
1360 	int error, oid[CTL_MAXNAME], len = 0;
1361 	struct sysctl_oid *op = NULL;
1362 	struct rm_priotracker tracker;
1363 	char buf[32];
1364 
1365 	if (!req->newlen)
1366 		return (ENOENT);
1367 	if (req->newlen >= MAXPATHLEN)	/* XXX arbitrary, undocumented */
1368 		return (ENAMETOOLONG);
1369 
1370 	p = buf;
1371 	if (req->newlen >= sizeof(buf))
1372 		p = malloc(req->newlen+1, M_SYSCTL, M_WAITOK);
1373 
1374 	error = SYSCTL_IN(req, p, req->newlen);
1375 	if (error) {
1376 		if (p != buf)
1377 			free(p, M_SYSCTL);
1378 		return (error);
1379 	}
1380 
1381 	p [req->newlen] = '\0';
1382 
1383 	SYSCTL_RLOCK(&tracker);
1384 	error = name2oid(p, oid, &len, &op);
1385 	SYSCTL_RUNLOCK(&tracker);
1386 
1387 	if (p != buf)
1388 		free(p, M_SYSCTL);
1389 
1390 	if (error)
1391 		return (error);
1392 
1393 	error = SYSCTL_OUT(req, oid, len * sizeof *oid);
1394 	return (error);
1395 }
1396 
1397 /*
1398  * XXXRW/JA: Shouldn't return name2oid data for nodes that we don't permit in
1399  * capability mode.
1400  */
1401 SYSCTL_PROC(_sysctl, CTL_SYSCTL_NAME2OID, name2oid, CTLTYPE_INT | CTLFLAG_RW |
1402     CTLFLAG_ANYBODY | CTLFLAG_MPSAFE | CTLFLAG_CAPRW, 0, 0,
1403     sysctl_sysctl_name2oid, "I", "");
1404 
1405 static int
1406 sysctl_sysctl_oidfmt(SYSCTL_HANDLER_ARGS)
1407 {
1408 	struct sysctl_oid *oid;
1409 	struct rm_priotracker tracker;
1410 	int error;
1411 
1412 	error = sysctl_wire_old_buffer(req, 0);
1413 	if (error)
1414 		return (error);
1415 
1416 	SYSCTL_RLOCK(&tracker);
1417 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1418 	if (error)
1419 		goto out;
1420 
1421 	if (oid->oid_fmt == NULL) {
1422 		error = ENOENT;
1423 		goto out;
1424 	}
1425 	error = SYSCTL_OUT(req, &oid->oid_kind, sizeof(oid->oid_kind));
1426 	if (error)
1427 		goto out;
1428 	error = SYSCTL_OUT(req, oid->oid_fmt, strlen(oid->oid_fmt) + 1);
1429  out:
1430 	SYSCTL_RUNLOCK(&tracker);
1431 	return (error);
1432 }
1433 
1434 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_OIDFMT, oidfmt, CTLFLAG_RD |
1435     CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_oidfmt, "");
1436 
1437 static int
1438 sysctl_sysctl_oiddescr(SYSCTL_HANDLER_ARGS)
1439 {
1440 	struct sysctl_oid *oid;
1441 	struct rm_priotracker tracker;
1442 	int error;
1443 
1444 	error = sysctl_wire_old_buffer(req, 0);
1445 	if (error)
1446 		return (error);
1447 
1448 	SYSCTL_RLOCK(&tracker);
1449 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1450 	if (error)
1451 		goto out;
1452 
1453 	if (oid->oid_descr == NULL) {
1454 		error = ENOENT;
1455 		goto out;
1456 	}
1457 	error = SYSCTL_OUT(req, oid->oid_descr, strlen(oid->oid_descr) + 1);
1458  out:
1459 	SYSCTL_RUNLOCK(&tracker);
1460 	return (error);
1461 }
1462 
1463 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_OIDDESCR, oiddescr, CTLFLAG_RD |
1464     CTLFLAG_MPSAFE|CTLFLAG_CAPRD, sysctl_sysctl_oiddescr, "");
1465 
1466 static int
1467 sysctl_sysctl_oidlabel(SYSCTL_HANDLER_ARGS)
1468 {
1469 	struct sysctl_oid *oid;
1470 	struct rm_priotracker tracker;
1471 	int error;
1472 
1473 	error = sysctl_wire_old_buffer(req, 0);
1474 	if (error)
1475 		return (error);
1476 
1477 	SYSCTL_RLOCK(&tracker);
1478 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1479 	if (error)
1480 		goto out;
1481 
1482 	if (oid->oid_label == NULL) {
1483 		error = ENOENT;
1484 		goto out;
1485 	}
1486 	error = SYSCTL_OUT(req, oid->oid_label, strlen(oid->oid_label) + 1);
1487  out:
1488 	SYSCTL_RUNLOCK(&tracker);
1489 	return (error);
1490 }
1491 
1492 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_OIDLABEL, oidlabel, CTLFLAG_RD |
1493     CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_oidlabel, "");
1494 
1495 /*
1496  * Default "handler" functions.
1497  */
1498 
1499 /*
1500  * Handle a bool.
1501  * Two cases:
1502  *     a variable:  point arg1 at it.
1503  *     a constant:  pass it in arg2.
1504  */
1505 
1506 int
1507 sysctl_handle_bool(SYSCTL_HANDLER_ARGS)
1508 {
1509 	uint8_t temp;
1510 	int error;
1511 
1512 	/*
1513 	 * Attempt to get a coherent snapshot by making a copy of the data.
1514 	 */
1515 	if (arg1)
1516 		temp = *(bool *)arg1 ? 1 : 0;
1517 	else
1518 		temp = arg2 ? 1 : 0;
1519 
1520 	error = SYSCTL_OUT(req, &temp, sizeof(temp));
1521 	if (error || !req->newptr)
1522 		return (error);
1523 
1524 	if (!arg1)
1525 		error = EPERM;
1526 	else {
1527 		error = SYSCTL_IN(req, &temp, sizeof(temp));
1528 		if (!error)
1529 			*(bool *)arg1 = temp ? 1 : 0;
1530 	}
1531 	return (error);
1532 }
1533 
1534 /*
1535  * Handle an int8_t, signed or unsigned.
1536  * Two cases:
1537  *     a variable:  point arg1 at it.
1538  *     a constant:  pass it in arg2.
1539  */
1540 
1541 int
1542 sysctl_handle_8(SYSCTL_HANDLER_ARGS)
1543 {
1544 	int8_t tmpout;
1545 	int error = 0;
1546 
1547 	/*
1548 	 * Attempt to get a coherent snapshot by making a copy of the data.
1549 	 */
1550 	if (arg1)
1551 		tmpout = *(int8_t *)arg1;
1552 	else
1553 		tmpout = arg2;
1554 	error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout));
1555 
1556 	if (error || !req->newptr)
1557 		return (error);
1558 
1559 	if (!arg1)
1560 		error = EPERM;
1561 	else
1562 		error = SYSCTL_IN(req, arg1, sizeof(tmpout));
1563 	return (error);
1564 }
1565 
1566 /*
1567  * Handle an int16_t, signed or unsigned.
1568  * Two cases:
1569  *     a variable:  point arg1 at it.
1570  *     a constant:  pass it in arg2.
1571  */
1572 
1573 int
1574 sysctl_handle_16(SYSCTL_HANDLER_ARGS)
1575 {
1576 	int16_t tmpout;
1577 	int error = 0;
1578 
1579 	/*
1580 	 * Attempt to get a coherent snapshot by making a copy of the data.
1581 	 */
1582 	if (arg1)
1583 		tmpout = *(int16_t *)arg1;
1584 	else
1585 		tmpout = arg2;
1586 	error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout));
1587 
1588 	if (error || !req->newptr)
1589 		return (error);
1590 
1591 	if (!arg1)
1592 		error = EPERM;
1593 	else
1594 		error = SYSCTL_IN(req, arg1, sizeof(tmpout));
1595 	return (error);
1596 }
1597 
1598 /*
1599  * Handle an int32_t, signed or unsigned.
1600  * Two cases:
1601  *     a variable:  point arg1 at it.
1602  *     a constant:  pass it in arg2.
1603  */
1604 
1605 int
1606 sysctl_handle_32(SYSCTL_HANDLER_ARGS)
1607 {
1608 	int32_t tmpout;
1609 	int error = 0;
1610 
1611 	/*
1612 	 * Attempt to get a coherent snapshot by making a copy of the data.
1613 	 */
1614 	if (arg1)
1615 		tmpout = *(int32_t *)arg1;
1616 	else
1617 		tmpout = arg2;
1618 	error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout));
1619 
1620 	if (error || !req->newptr)
1621 		return (error);
1622 
1623 	if (!arg1)
1624 		error = EPERM;
1625 	else
1626 		error = SYSCTL_IN(req, arg1, sizeof(tmpout));
1627 	return (error);
1628 }
1629 
1630 /*
1631  * Handle an int, signed or unsigned.
1632  * Two cases:
1633  *     a variable:  point arg1 at it.
1634  *     a constant:  pass it in arg2.
1635  */
1636 
1637 int
1638 sysctl_handle_int(SYSCTL_HANDLER_ARGS)
1639 {
1640 	int tmpout, error = 0;
1641 
1642 	/*
1643 	 * Attempt to get a coherent snapshot by making a copy of the data.
1644 	 */
1645 	if (arg1)
1646 		tmpout = *(int *)arg1;
1647 	else
1648 		tmpout = arg2;
1649 	error = SYSCTL_OUT(req, &tmpout, sizeof(int));
1650 
1651 	if (error || !req->newptr)
1652 		return (error);
1653 
1654 	if (!arg1)
1655 		error = EPERM;
1656 	else
1657 		error = SYSCTL_IN(req, arg1, sizeof(int));
1658 	return (error);
1659 }
1660 
1661 /*
1662  * Based on on sysctl_handle_int() convert milliseconds into ticks.
1663  * Note: this is used by TCP.
1664  */
1665 
1666 int
1667 sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
1668 {
1669 	int error, s, tt;
1670 
1671 	tt = *(int *)arg1;
1672 	s = (int)((int64_t)tt * 1000 / hz);
1673 
1674 	error = sysctl_handle_int(oidp, &s, 0, req);
1675 	if (error || !req->newptr)
1676 		return (error);
1677 
1678 	tt = (int)((int64_t)s * hz / 1000);
1679 	if (tt < 1)
1680 		return (EINVAL);
1681 
1682 	*(int *)arg1 = tt;
1683 	return (0);
1684 }
1685 
1686 /*
1687  * Handle a long, signed or unsigned.
1688  * Two cases:
1689  *     a variable:  point arg1 at it.
1690  *     a constant:  pass it in arg2.
1691  */
1692 
1693 int
1694 sysctl_handle_long(SYSCTL_HANDLER_ARGS)
1695 {
1696 	int error = 0;
1697 	long tmplong;
1698 #ifdef SCTL_MASK32
1699 	int tmpint;
1700 #endif
1701 
1702 	/*
1703 	 * Attempt to get a coherent snapshot by making a copy of the data.
1704 	 */
1705 	if (arg1)
1706 		tmplong = *(long *)arg1;
1707 	else
1708 		tmplong = arg2;
1709 #ifdef SCTL_MASK32
1710 	if (req->flags & SCTL_MASK32) {
1711 		tmpint = tmplong;
1712 		error = SYSCTL_OUT(req, &tmpint, sizeof(int));
1713 	} else
1714 #endif
1715 		error = SYSCTL_OUT(req, &tmplong, sizeof(long));
1716 
1717 	if (error || !req->newptr)
1718 		return (error);
1719 
1720 	if (!arg1)
1721 		error = EPERM;
1722 #ifdef SCTL_MASK32
1723 	else if (req->flags & SCTL_MASK32) {
1724 		error = SYSCTL_IN(req, &tmpint, sizeof(int));
1725 		*(long *)arg1 = (long)tmpint;
1726 	}
1727 #endif
1728 	else
1729 		error = SYSCTL_IN(req, arg1, sizeof(long));
1730 	return (error);
1731 }
1732 
1733 /*
1734  * Handle a 64 bit int, signed or unsigned.
1735  * Two cases:
1736  *     a variable:  point arg1 at it.
1737  *     a constant:  pass it in arg2.
1738  */
1739 int
1740 sysctl_handle_64(SYSCTL_HANDLER_ARGS)
1741 {
1742 	int error = 0;
1743 	uint64_t tmpout;
1744 
1745 	/*
1746 	 * Attempt to get a coherent snapshot by making a copy of the data.
1747 	 */
1748 	if (arg1)
1749 		tmpout = *(uint64_t *)arg1;
1750 	else
1751 		tmpout = arg2;
1752 	error = SYSCTL_OUT(req, &tmpout, sizeof(uint64_t));
1753 
1754 	if (error || !req->newptr)
1755 		return (error);
1756 
1757 	if (!arg1)
1758 		error = EPERM;
1759 	else
1760 		error = SYSCTL_IN(req, arg1, sizeof(uint64_t));
1761 	return (error);
1762 }
1763 
1764 /*
1765  * Handle our generic '\0' terminated 'C' string.
1766  * Two cases:
1767  * 	a variable string:  point arg1 at it, arg2 is max length.
1768  * 	a constant string:  point arg1 at it, arg2 is zero.
1769  */
1770 
1771 int
1772 sysctl_handle_string(SYSCTL_HANDLER_ARGS)
1773 {
1774 	char *tmparg;
1775 	size_t outlen;
1776 	int error = 0, ro_string = 0;
1777 
1778 	/*
1779 	 * If the sysctl isn't writable and isn't a preallocated tunable that
1780 	 * can be modified by kenv(2), microoptimise and treat it as a
1781 	 * read-only string.
1782 	 * A zero-length buffer indicates a fixed size read-only
1783 	 * string.  In ddb, don't worry about trying to make a malloced
1784 	 * snapshot.
1785 	 */
1786 	if ((oidp->oid_kind & (CTLFLAG_WR | CTLFLAG_TUN)) == 0 ||
1787 	    arg2 == 0 || kdb_active) {
1788 		arg2 = strlen((char *)arg1) + 1;
1789 		ro_string = 1;
1790 	}
1791 
1792 	if (req->oldptr != NULL) {
1793 		if (ro_string) {
1794 			tmparg = arg1;
1795 			outlen = strlen(tmparg) + 1;
1796 		} else {
1797 			tmparg = malloc(arg2, M_SYSCTLTMP, M_WAITOK);
1798 			sx_slock(&sysctlstringlock);
1799 			memcpy(tmparg, arg1, arg2);
1800 			sx_sunlock(&sysctlstringlock);
1801 			outlen = strlen(tmparg) + 1;
1802 		}
1803 
1804 		error = SYSCTL_OUT(req, tmparg, outlen);
1805 
1806 		if (!ro_string)
1807 			free(tmparg, M_SYSCTLTMP);
1808 	} else {
1809 		if (!ro_string)
1810 			sx_slock(&sysctlstringlock);
1811 		outlen = strlen((char *)arg1) + 1;
1812 		if (!ro_string)
1813 			sx_sunlock(&sysctlstringlock);
1814 		error = SYSCTL_OUT(req, NULL, outlen);
1815 	}
1816 	if (error || !req->newptr)
1817 		return (error);
1818 
1819 	if (req->newlen - req->newidx >= arg2 ||
1820 	    req->newlen - req->newidx < 0) {
1821 		error = EINVAL;
1822 	} else if (req->newlen - req->newidx == 0) {
1823 		sx_xlock(&sysctlstringlock);
1824 		((char *)arg1)[0] = '\0';
1825 		sx_xunlock(&sysctlstringlock);
1826 	} else if (req->newfunc == sysctl_new_kernel) {
1827 		arg2 = req->newlen - req->newidx;
1828 		sx_xlock(&sysctlstringlock);
1829 		error = SYSCTL_IN(req, arg1, arg2);
1830 		if (error == 0) {
1831 			((char *)arg1)[arg2] = '\0';
1832 			req->newidx += arg2;
1833 		}
1834 		sx_xunlock(&sysctlstringlock);
1835 	} else {
1836 		arg2 = req->newlen - req->newidx;
1837 		tmparg = malloc(arg2, M_SYSCTLTMP, M_WAITOK);
1838 
1839 		error = SYSCTL_IN(req, tmparg, arg2);
1840 		if (error) {
1841 			free(tmparg, M_SYSCTLTMP);
1842 			return (error);
1843 		}
1844 
1845 		sx_xlock(&sysctlstringlock);
1846 		memcpy(arg1, tmparg, arg2);
1847 		((char *)arg1)[arg2] = '\0';
1848 		sx_xunlock(&sysctlstringlock);
1849 		free(tmparg, M_SYSCTLTMP);
1850 		req->newidx += arg2;
1851 	}
1852 	return (error);
1853 }
1854 
1855 /*
1856  * Handle any kind of opaque data.
1857  * arg1 points to it, arg2 is the size.
1858  */
1859 
1860 int
1861 sysctl_handle_opaque(SYSCTL_HANDLER_ARGS)
1862 {
1863 	int error, tries;
1864 	u_int generation;
1865 	struct sysctl_req req2;
1866 
1867 	/*
1868 	 * Attempt to get a coherent snapshot, by using the thread
1869 	 * pre-emption counter updated from within mi_switch() to
1870 	 * determine if we were pre-empted during a bcopy() or
1871 	 * copyout(). Make 3 attempts at doing this before giving up.
1872 	 * If we encounter an error, stop immediately.
1873 	 */
1874 	tries = 0;
1875 	req2 = *req;
1876 retry:
1877 	generation = curthread->td_generation;
1878 	error = SYSCTL_OUT(req, arg1, arg2);
1879 	if (error)
1880 		return (error);
1881 	tries++;
1882 	if (generation != curthread->td_generation && tries < 3) {
1883 		*req = req2;
1884 		goto retry;
1885 	}
1886 
1887 	error = SYSCTL_IN(req, arg1, arg2);
1888 
1889 	return (error);
1890 }
1891 
1892 /*
1893  * Based on on sysctl_handle_int() convert microseconds to a sbintime.
1894  */
1895 int
1896 sysctl_usec_to_sbintime(SYSCTL_HANDLER_ARGS)
1897 {
1898 	int error;
1899 	int64_t tt;
1900 	sbintime_t sb;
1901 
1902 	tt = *(int64_t *)arg1;
1903 	sb = sbttous(tt);
1904 
1905 	error = sysctl_handle_64(oidp, &sb, 0, req);
1906 	if (error || !req->newptr)
1907 		return (error);
1908 
1909 	tt = ustosbt(sb);
1910 	*(int64_t *)arg1 = tt;
1911 
1912 	return (0);
1913 }
1914 
1915 /*
1916  * Based on on sysctl_handle_int() convert milliseconds to a sbintime.
1917  */
1918 int
1919 sysctl_msec_to_sbintime(SYSCTL_HANDLER_ARGS)
1920 {
1921 	int error;
1922 	int64_t tt;
1923 	sbintime_t sb;
1924 
1925 	tt = *(int64_t *)arg1;
1926 	sb = sbttoms(tt);
1927 
1928 	error = sysctl_handle_64(oidp, &sb, 0, req);
1929 	if (error || !req->newptr)
1930 		return (error);
1931 
1932 	tt = mstosbt(sb);
1933 	*(int64_t *)arg1 = tt;
1934 
1935 	return (0);
1936 }
1937 
1938 /*
1939  * Convert seconds to a struct timeval.  Intended for use with
1940  * intervals and thus does not permit negative seconds.
1941  */
1942 int
1943 sysctl_sec_to_timeval(SYSCTL_HANDLER_ARGS)
1944 {
1945 	struct timeval *tv;
1946 	int error, secs;
1947 
1948 	tv = arg1;
1949 	secs = tv->tv_sec;
1950 
1951 	error = sysctl_handle_int(oidp, &secs, 0, req);
1952 	if (error || req->newptr == NULL)
1953 		return (error);
1954 
1955 	if (secs < 0)
1956 		return (EINVAL);
1957 	tv->tv_sec = secs;
1958 
1959 	return (0);
1960 }
1961 
1962 /*
1963  * Transfer functions to/from kernel space.
1964  * XXX: rather untested at this point
1965  */
1966 static int
1967 sysctl_old_kernel(struct sysctl_req *req, const void *p, size_t l)
1968 {
1969 	size_t i = 0;
1970 
1971 	if (req->oldptr) {
1972 		i = l;
1973 		if (req->oldlen <= req->oldidx)
1974 			i = 0;
1975 		else
1976 			if (i > req->oldlen - req->oldidx)
1977 				i = req->oldlen - req->oldidx;
1978 		if (i > 0)
1979 			bcopy(p, (char *)req->oldptr + req->oldidx, i);
1980 	}
1981 	req->oldidx += l;
1982 	if (req->oldptr && i != l)
1983 		return (ENOMEM);
1984 	return (0);
1985 }
1986 
1987 static int
1988 sysctl_new_kernel(struct sysctl_req *req, void *p, size_t l)
1989 {
1990 	if (!req->newptr)
1991 		return (0);
1992 	if (req->newlen - req->newidx < l)
1993 		return (EINVAL);
1994 	bcopy((const char *)req->newptr + req->newidx, p, l);
1995 	req->newidx += l;
1996 	return (0);
1997 }
1998 
1999 int
2000 kernel_sysctl(struct thread *td, int *name, u_int namelen, void *old,
2001     size_t *oldlenp, void *new, size_t newlen, size_t *retval, int flags)
2002 {
2003 	int error = 0;
2004 	struct sysctl_req req;
2005 
2006 	bzero(&req, sizeof req);
2007 
2008 	req.td = td;
2009 	req.flags = flags;
2010 
2011 	if (oldlenp) {
2012 		req.oldlen = *oldlenp;
2013 	}
2014 	req.validlen = req.oldlen;
2015 
2016 	if (old) {
2017 		req.oldptr= old;
2018 	}
2019 
2020 	if (new != NULL) {
2021 		req.newlen = newlen;
2022 		req.newptr = new;
2023 	}
2024 
2025 	req.oldfunc = sysctl_old_kernel;
2026 	req.newfunc = sysctl_new_kernel;
2027 	req.lock = REQ_UNWIRED;
2028 
2029 	error = sysctl_root(0, name, namelen, &req);
2030 
2031 	if (req.lock == REQ_WIRED && req.validlen > 0)
2032 		vsunlock(req.oldptr, req.validlen);
2033 
2034 	if (error && error != ENOMEM)
2035 		return (error);
2036 
2037 	if (retval) {
2038 		if (req.oldptr && req.oldidx > req.validlen)
2039 			*retval = req.validlen;
2040 		else
2041 			*retval = req.oldidx;
2042 	}
2043 	return (error);
2044 }
2045 
2046 int
2047 kernel_sysctlbyname(struct thread *td, char *name, void *old, size_t *oldlenp,
2048     void *new, size_t newlen, size_t *retval, int flags)
2049 {
2050         int oid[CTL_MAXNAME];
2051         size_t oidlen, plen;
2052 	int error;
2053 
2054 	oid[0] = CTL_SYSCTL;
2055 	oid[1] = CTL_SYSCTL_NAME2OID;
2056 	oidlen = sizeof(oid);
2057 
2058 	error = kernel_sysctl(td, oid, 2, oid, &oidlen,
2059 	    (void *)name, strlen(name), &plen, flags);
2060 	if (error)
2061 		return (error);
2062 
2063 	error = kernel_sysctl(td, oid, plen / sizeof(int), old, oldlenp,
2064 	    new, newlen, retval, flags);
2065 	return (error);
2066 }
2067 
2068 /*
2069  * Transfer function to/from user space.
2070  */
2071 static int
2072 sysctl_old_user(struct sysctl_req *req, const void *p, size_t l)
2073 {
2074 	size_t i, len, origidx;
2075 	int error;
2076 
2077 	origidx = req->oldidx;
2078 	req->oldidx += l;
2079 	if (req->oldptr == NULL)
2080 		return (0);
2081 	/*
2082 	 * If we have not wired the user supplied buffer and we are currently
2083 	 * holding locks, drop a witness warning, as it's possible that
2084 	 * write operations to the user page can sleep.
2085 	 */
2086 	if (req->lock != REQ_WIRED)
2087 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
2088 		    "sysctl_old_user()");
2089 	i = l;
2090 	len = req->validlen;
2091 	if (len <= origidx)
2092 		i = 0;
2093 	else {
2094 		if (i > len - origidx)
2095 			i = len - origidx;
2096 		if (req->lock == REQ_WIRED) {
2097 			error = copyout_nofault(p, (char *)req->oldptr +
2098 			    origidx, i);
2099 		} else
2100 			error = copyout(p, (char *)req->oldptr + origidx, i);
2101 		if (error != 0)
2102 			return (error);
2103 	}
2104 	if (i < l)
2105 		return (ENOMEM);
2106 	return (0);
2107 }
2108 
2109 static int
2110 sysctl_new_user(struct sysctl_req *req, void *p, size_t l)
2111 {
2112 	int error;
2113 
2114 	if (!req->newptr)
2115 		return (0);
2116 	if (req->newlen - req->newidx < l)
2117 		return (EINVAL);
2118 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
2119 	    "sysctl_new_user()");
2120 	error = copyin((const char *)req->newptr + req->newidx, p, l);
2121 	req->newidx += l;
2122 	return (error);
2123 }
2124 
2125 /*
2126  * Wire the user space destination buffer.  If set to a value greater than
2127  * zero, the len parameter limits the maximum amount of wired memory.
2128  */
2129 int
2130 sysctl_wire_old_buffer(struct sysctl_req *req, size_t len)
2131 {
2132 	int ret;
2133 	size_t wiredlen;
2134 
2135 	wiredlen = (len > 0 && len < req->oldlen) ? len : req->oldlen;
2136 	ret = 0;
2137 	if (req->lock != REQ_WIRED && req->oldptr &&
2138 	    req->oldfunc == sysctl_old_user) {
2139 		if (wiredlen != 0) {
2140 			ret = vslock(req->oldptr, wiredlen);
2141 			if (ret != 0) {
2142 				if (ret != ENOMEM)
2143 					return (ret);
2144 				wiredlen = 0;
2145 			}
2146 		}
2147 		req->lock = REQ_WIRED;
2148 		req->validlen = wiredlen;
2149 	}
2150 	return (0);
2151 }
2152 
2153 int
2154 sysctl_find_oid(int *name, u_int namelen, struct sysctl_oid **noid,
2155     int *nindx, struct sysctl_req *req)
2156 {
2157 	struct sysctl_oid_list *lsp;
2158 	struct sysctl_oid *oid;
2159 	struct sysctl_oid key;
2160 	int indx;
2161 
2162 	SYSCTL_ASSERT_LOCKED();
2163 	lsp = &sysctl__children;
2164 	indx = 0;
2165 	while (indx < CTL_MAXNAME) {
2166 		key.oid_number = name[indx];
2167 		oid = RB_FIND(sysctl_oid_list, lsp, &key);
2168 		if (oid == NULL)
2169 			return (ENOENT);
2170 
2171 		indx++;
2172 		if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
2173 			if (oid->oid_handler != NULL || indx == namelen) {
2174 				*noid = oid;
2175 				if (nindx != NULL)
2176 					*nindx = indx;
2177 				KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
2178 				    ("%s found DYING node %p", __func__, oid));
2179 				return (0);
2180 			}
2181 			lsp = SYSCTL_CHILDREN(oid);
2182 		} else if (indx == namelen) {
2183 			if ((oid->oid_kind & CTLFLAG_DORMANT) != 0)
2184 				return (ENOENT);
2185 			*noid = oid;
2186 			if (nindx != NULL)
2187 				*nindx = indx;
2188 			KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
2189 			    ("%s found DYING node %p", __func__, oid));
2190 			return (0);
2191 		} else {
2192 			return (ENOTDIR);
2193 		}
2194 	}
2195 	return (ENOENT);
2196 }
2197 
2198 /*
2199  * Traverse our tree, and find the right node, execute whatever it points
2200  * to, and return the resulting error code.
2201  */
2202 
2203 static int
2204 sysctl_root(SYSCTL_HANDLER_ARGS)
2205 {
2206 	struct sysctl_oid *oid;
2207 	struct rm_priotracker tracker;
2208 	int error, indx, lvl;
2209 
2210 	SYSCTL_RLOCK(&tracker);
2211 
2212 	error = sysctl_find_oid(arg1, arg2, &oid, &indx, req);
2213 	if (error)
2214 		goto out;
2215 
2216 	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
2217 		/*
2218 		 * You can't call a sysctl when it's a node, but has
2219 		 * no handler.  Inform the user that it's a node.
2220 		 * The indx may or may not be the same as namelen.
2221 		 */
2222 		if (oid->oid_handler == NULL) {
2223 			error = EISDIR;
2224 			goto out;
2225 		}
2226 	}
2227 
2228 	/* Is this sysctl writable? */
2229 	if (req->newptr && !(oid->oid_kind & CTLFLAG_WR)) {
2230 		error = EPERM;
2231 		goto out;
2232 	}
2233 
2234 	KASSERT(req->td != NULL, ("sysctl_root(): req->td == NULL"));
2235 
2236 #ifdef CAPABILITY_MODE
2237 	/*
2238 	 * If the process is in capability mode, then don't permit reading or
2239 	 * writing unless specifically granted for the node.
2240 	 */
2241 	if (IN_CAPABILITY_MODE(req->td)) {
2242 		if ((req->oldptr && !(oid->oid_kind & CTLFLAG_CAPRD)) ||
2243 		    (req->newptr && !(oid->oid_kind & CTLFLAG_CAPWR))) {
2244 			error = EPERM;
2245 			goto out;
2246 		}
2247 	}
2248 #endif
2249 
2250 	/* Is this sysctl sensitive to securelevels? */
2251 	if (req->newptr && (oid->oid_kind & CTLFLAG_SECURE)) {
2252 		lvl = (oid->oid_kind & CTLMASK_SECURE) >> CTLSHIFT_SECURE;
2253 		error = securelevel_gt(req->td->td_ucred, lvl);
2254 		if (error)
2255 			goto out;
2256 	}
2257 
2258 	/* Is this sysctl writable by only privileged users? */
2259 	if (req->newptr && !(oid->oid_kind & CTLFLAG_ANYBODY)) {
2260 		int priv;
2261 
2262 		if (oid->oid_kind & CTLFLAG_PRISON)
2263 			priv = PRIV_SYSCTL_WRITEJAIL;
2264 #ifdef VIMAGE
2265 		else if ((oid->oid_kind & CTLFLAG_VNET) &&
2266 		     prison_owns_vnet(req->td->td_ucred))
2267 			priv = PRIV_SYSCTL_WRITEJAIL;
2268 #endif
2269 		else
2270 			priv = PRIV_SYSCTL_WRITE;
2271 		error = priv_check(req->td, priv);
2272 		if (error)
2273 			goto out;
2274 	}
2275 
2276 	if (!oid->oid_handler) {
2277 		error = EINVAL;
2278 		goto out;
2279 	}
2280 
2281 	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
2282 		arg1 = (int *)arg1 + indx;
2283 		arg2 -= indx;
2284 	} else {
2285 		arg1 = oid->oid_arg1;
2286 		arg2 = oid->oid_arg2;
2287 	}
2288 #ifdef MAC
2289 	error = mac_system_check_sysctl(req->td->td_ucred, oid, arg1, arg2,
2290 	    req);
2291 	if (error != 0)
2292 		goto out;
2293 #endif
2294 #ifdef VIMAGE
2295 	if ((oid->oid_kind & CTLFLAG_VNET) && arg1 != NULL)
2296 		arg1 = (void *)(curvnet->vnet_data_base + (uintptr_t)arg1);
2297 #endif
2298 	error = sysctl_root_handler_locked(oid, arg1, arg2, req, &tracker);
2299 
2300 out:
2301 	SYSCTL_RUNLOCK(&tracker);
2302 	return (error);
2303 }
2304 
2305 #ifndef _SYS_SYSPROTO_H_
2306 struct __sysctl_args {
2307 	int	*name;
2308 	u_int	namelen;
2309 	void	*old;
2310 	size_t	*oldlenp;
2311 	void	*new;
2312 	size_t	newlen;
2313 };
2314 #endif
2315 int
2316 sys___sysctl(struct thread *td, struct __sysctl_args *uap)
2317 {
2318 	int error, i, name[CTL_MAXNAME];
2319 	size_t j;
2320 
2321 	if (uap->namelen > CTL_MAXNAME || uap->namelen < 2)
2322 		return (EINVAL);
2323 
2324  	error = copyin(uap->name, &name, uap->namelen * sizeof(int));
2325  	if (error)
2326 		return (error);
2327 
2328 	error = userland_sysctl(td, name, uap->namelen,
2329 		uap->old, uap->oldlenp, 0,
2330 		uap->new, uap->newlen, &j, 0);
2331 	if (error && error != ENOMEM)
2332 		return (error);
2333 	if (uap->oldlenp) {
2334 		i = copyout(&j, uap->oldlenp, sizeof(j));
2335 		if (i)
2336 			return (i);
2337 	}
2338 	return (error);
2339 }
2340 
2341 int
2342 kern___sysctlbyname(struct thread *td, const char *oname, size_t namelen,
2343     void *old, size_t *oldlenp, void *new, size_t newlen, size_t *retval,
2344     int flags, bool inkernel)
2345 {
2346 	int oid[CTL_MAXNAME];
2347 	char namebuf[16];
2348 	char *name;
2349 	size_t oidlen;
2350 	int error;
2351 
2352 	if (namelen > MAXPATHLEN || namelen == 0)
2353 		return (EINVAL);
2354 	name = namebuf;
2355 	if (namelen > sizeof(namebuf))
2356 		name = malloc(namelen, M_SYSCTL, M_WAITOK);
2357 	error = copyin(oname, name, namelen);
2358 	if (error != 0)
2359 		goto out;
2360 
2361 	oid[0] = CTL_SYSCTL;
2362 	oid[1] = CTL_SYSCTL_NAME2OID;
2363 	oidlen = sizeof(oid);
2364 	error = kernel_sysctl(td, oid, 2, oid, &oidlen, (void *)name, namelen,
2365 	    retval, flags);
2366 	if (error != 0)
2367 		goto out;
2368 	error = userland_sysctl(td, oid, *retval / sizeof(int), old, oldlenp,
2369 	    inkernel, new, newlen, retval, flags);
2370 
2371 out:
2372 	if (namelen > sizeof(namebuf))
2373 		free(name, M_SYSCTL);
2374 	return (error);
2375 }
2376 
2377 #ifndef	_SYS_SYSPROTO_H_
2378 struct __sysctlbyname_args {
2379 	const char	*name;
2380 	size_t	namelen;
2381 	void	*old;
2382 	size_t	*oldlenp;
2383 	void	*new;
2384 	size_t	newlen;
2385 };
2386 #endif
2387 int
2388 sys___sysctlbyname(struct thread *td, struct __sysctlbyname_args *uap)
2389 {
2390 	size_t rv;
2391 	int error;
2392 
2393 	error = kern___sysctlbyname(td, uap->name, uap->namelen, uap->old,
2394 	    uap->oldlenp, uap->new, uap->newlen, &rv, 0, 0);
2395 	if (error != 0)
2396 		return (error);
2397 	if (uap->oldlenp != NULL)
2398 		error = copyout(&rv, uap->oldlenp, sizeof(rv));
2399 
2400 	return (error);
2401 }
2402 
2403 /*
2404  * This is used from various compatibility syscalls too.  That's why name
2405  * must be in kernel space.
2406  */
2407 int
2408 userland_sysctl(struct thread *td, int *name, u_int namelen, void *old,
2409     size_t *oldlenp, int inkernel, const void *new, size_t newlen,
2410     size_t *retval, int flags)
2411 {
2412 	int error = 0, memlocked;
2413 	struct sysctl_req req;
2414 
2415 	bzero(&req, sizeof req);
2416 
2417 	req.td = td;
2418 	req.flags = flags;
2419 
2420 	if (oldlenp) {
2421 		if (inkernel) {
2422 			req.oldlen = *oldlenp;
2423 		} else {
2424 			error = copyin(oldlenp, &req.oldlen, sizeof(*oldlenp));
2425 			if (error)
2426 				return (error);
2427 		}
2428 	}
2429 	req.validlen = req.oldlen;
2430 	req.oldptr = old;
2431 
2432 	if (new != NULL) {
2433 		req.newlen = newlen;
2434 		req.newptr = new;
2435 	}
2436 
2437 	req.oldfunc = sysctl_old_user;
2438 	req.newfunc = sysctl_new_user;
2439 	req.lock = REQ_UNWIRED;
2440 
2441 #ifdef KTRACE
2442 	if (KTRPOINT(curthread, KTR_SYSCTL))
2443 		ktrsysctl(name, namelen);
2444 #endif
2445 	memlocked = 0;
2446 	if (req.oldptr && req.oldlen > 4 * PAGE_SIZE) {
2447 		memlocked = 1;
2448 		sx_xlock(&sysctlmemlock);
2449 	}
2450 	CURVNET_SET(TD_TO_VNET(td));
2451 
2452 	for (;;) {
2453 		req.oldidx = 0;
2454 		req.newidx = 0;
2455 		error = sysctl_root(0, name, namelen, &req);
2456 		if (error != EAGAIN)
2457 			break;
2458 		kern_yield(PRI_USER);
2459 	}
2460 
2461 	CURVNET_RESTORE();
2462 
2463 	if (req.lock == REQ_WIRED && req.validlen > 0)
2464 		vsunlock(req.oldptr, req.validlen);
2465 	if (memlocked)
2466 		sx_xunlock(&sysctlmemlock);
2467 
2468 	if (error && error != ENOMEM)
2469 		return (error);
2470 
2471 	if (retval) {
2472 		if (req.oldptr && req.oldidx > req.validlen)
2473 			*retval = req.validlen;
2474 		else
2475 			*retval = req.oldidx;
2476 	}
2477 	return (error);
2478 }
2479 
2480 /*
2481  * Drain into a sysctl struct.  The user buffer should be wired if a page
2482  * fault would cause issue.
2483  */
2484 static int
2485 sbuf_sysctl_drain(void *arg, const char *data, int len)
2486 {
2487 	struct sysctl_req *req = arg;
2488 	int error;
2489 
2490 	error = SYSCTL_OUT(req, data, len);
2491 	KASSERT(error >= 0, ("Got unexpected negative value %d", error));
2492 	return (error == 0 ? len : -error);
2493 }
2494 
2495 struct sbuf *
2496 sbuf_new_for_sysctl(struct sbuf *s, char *buf, int length,
2497     struct sysctl_req *req)
2498 {
2499 
2500 	/* Supply a default buffer size if none given. */
2501 	if (buf == NULL && length == 0)
2502 		length = 64;
2503 	s = sbuf_new(s, buf, length, SBUF_FIXEDLEN | SBUF_INCLUDENUL);
2504 	sbuf_set_drain(s, sbuf_sysctl_drain, req);
2505 	return (s);
2506 }
2507 
2508 #ifdef DDB
2509 
2510 /* The current OID the debugger is working with */
2511 static struct sysctl_oid *g_ddb_oid;
2512 
2513 /* The current flags specified by the user */
2514 static int g_ddb_sysctl_flags;
2515 
2516 /* Check to see if the last sysctl printed */
2517 static int g_ddb_sysctl_printed;
2518 
2519 static const int ctl_sign[CTLTYPE+1] = {
2520 	[CTLTYPE_INT] = 1,
2521 	[CTLTYPE_LONG] = 1,
2522 	[CTLTYPE_S8] = 1,
2523 	[CTLTYPE_S16] = 1,
2524 	[CTLTYPE_S32] = 1,
2525 	[CTLTYPE_S64] = 1,
2526 };
2527 
2528 static const int ctl_size[CTLTYPE+1] = {
2529 	[CTLTYPE_INT] = sizeof(int),
2530 	[CTLTYPE_UINT] = sizeof(u_int),
2531 	[CTLTYPE_LONG] = sizeof(long),
2532 	[CTLTYPE_ULONG] = sizeof(u_long),
2533 	[CTLTYPE_S8] = sizeof(int8_t),
2534 	[CTLTYPE_S16] = sizeof(int16_t),
2535 	[CTLTYPE_S32] = sizeof(int32_t),
2536 	[CTLTYPE_S64] = sizeof(int64_t),
2537 	[CTLTYPE_U8] = sizeof(uint8_t),
2538 	[CTLTYPE_U16] = sizeof(uint16_t),
2539 	[CTLTYPE_U32] = sizeof(uint32_t),
2540 	[CTLTYPE_U64] = sizeof(uint64_t),
2541 };
2542 
2543 #define DB_SYSCTL_NAME_ONLY	0x001	/* Compare with -N */
2544 #define DB_SYSCTL_VALUE_ONLY	0x002	/* Compare with -n */
2545 #define DB_SYSCTL_OPAQUE	0x004	/* Compare with -o */
2546 #define DB_SYSCTL_HEX		0x008	/* Compare with -x */
2547 
2548 #define DB_SYSCTL_SAFE_ONLY	0x100	/* Only simple types */
2549 
2550 static const char db_sysctl_modifs[] = {
2551 	'N', 'n', 'o', 'x',
2552 };
2553 
2554 static const int db_sysctl_modif_values[] = {
2555 	DB_SYSCTL_NAME_ONLY, DB_SYSCTL_VALUE_ONLY,
2556 	DB_SYSCTL_OPAQUE, DB_SYSCTL_HEX,
2557 };
2558 
2559 /* Handlers considered safe to print while recursing */
2560 static int (* const db_safe_handlers[])(SYSCTL_HANDLER_ARGS) = {
2561 	sysctl_handle_bool,
2562 	sysctl_handle_8,
2563 	sysctl_handle_16,
2564 	sysctl_handle_32,
2565 	sysctl_handle_64,
2566 	sysctl_handle_int,
2567 	sysctl_handle_long,
2568 	sysctl_handle_string,
2569 	sysctl_handle_opaque,
2570 };
2571 
2572 /*
2573  * Use in place of sysctl_old_kernel to print sysctl values.
2574  *
2575  * Compare to the output handling in show_var from sbin/sysctl/sysctl.c
2576  */
2577 static int
2578 sysctl_old_ddb(struct sysctl_req *req, const void *ptr, size_t len)
2579 {
2580 	const u_char *val, *p;
2581 	const char *sep1;
2582 	size_t intlen, slen;
2583 	uintmax_t umv;
2584 	intmax_t mv;
2585 	int sign, ctltype, hexlen, xflag, error;
2586 
2587 	/* Suppress false-positive GCC uninitialized variable warnings */
2588 	mv = 0;
2589 	umv = 0;
2590 
2591 	slen = len;
2592 	val = p = ptr;
2593 
2594 	if (ptr == NULL) {
2595 		error = 0;
2596 		goto out;
2597 	}
2598 
2599 	/* We are going to print */
2600 	g_ddb_sysctl_printed = 1;
2601 
2602 	xflag = g_ddb_sysctl_flags & DB_SYSCTL_HEX;
2603 
2604 	ctltype = (g_ddb_oid->oid_kind & CTLTYPE);
2605 	sign = ctl_sign[ctltype];
2606 	intlen = ctl_size[ctltype];
2607 
2608 	switch (ctltype) {
2609 	case CTLTYPE_NODE:
2610 	case CTLTYPE_STRING:
2611 		db_printf("%.*s", (int) len, (const char *) p);
2612 		error = 0;
2613 		goto out;
2614 
2615 	case CTLTYPE_INT:
2616 	case CTLTYPE_UINT:
2617 	case CTLTYPE_LONG:
2618 	case CTLTYPE_ULONG:
2619 	case CTLTYPE_S8:
2620 	case CTLTYPE_S16:
2621 	case CTLTYPE_S32:
2622 	case CTLTYPE_S64:
2623 	case CTLTYPE_U8:
2624 	case CTLTYPE_U16:
2625 	case CTLTYPE_U32:
2626 	case CTLTYPE_U64:
2627 		hexlen = 2 + (intlen * CHAR_BIT + 3) / 4;
2628 		sep1 = "";
2629 		while (len >= intlen) {
2630 			switch (ctltype) {
2631 			case CTLTYPE_INT:
2632 			case CTLTYPE_UINT:
2633 				umv = *(const u_int *)p;
2634 				mv = *(const int *)p;
2635 				break;
2636 			case CTLTYPE_LONG:
2637 			case CTLTYPE_ULONG:
2638 				umv = *(const u_long *)p;
2639 				mv = *(const long *)p;
2640 				break;
2641 			case CTLTYPE_S8:
2642 			case CTLTYPE_U8:
2643 				umv = *(const uint8_t *)p;
2644 				mv = *(const int8_t *)p;
2645 				break;
2646 			case CTLTYPE_S16:
2647 			case CTLTYPE_U16:
2648 				umv = *(const uint16_t *)p;
2649 				mv = *(const int16_t *)p;
2650 				break;
2651 			case CTLTYPE_S32:
2652 			case CTLTYPE_U32:
2653 				umv = *(const uint32_t *)p;
2654 				mv = *(const int32_t *)p;
2655 				break;
2656 			case CTLTYPE_S64:
2657 			case CTLTYPE_U64:
2658 				umv = *(const uint64_t *)p;
2659 				mv = *(const int64_t *)p;
2660 				break;
2661 			}
2662 
2663 			db_printf("%s", sep1);
2664 			if (xflag)
2665 				db_printf("%#0*jx", hexlen, umv);
2666 			else if (!sign)
2667 				db_printf("%ju", umv);
2668 			else if (g_ddb_oid->oid_fmt[1] == 'K') {
2669 				/* Kelvins are currently unsupported. */
2670 				error = EOPNOTSUPP;
2671 				goto out;
2672 			} else
2673 				db_printf("%jd", mv);
2674 
2675 			sep1 = " ";
2676 			len -= intlen;
2677 			p += intlen;
2678 		}
2679 		error = 0;
2680 		goto out;
2681 
2682 	case CTLTYPE_OPAQUE:
2683 		/* TODO: Support struct functions. */
2684 
2685 		/* FALLTHROUGH */
2686 	default:
2687 		db_printf("Format:%s Length:%zu Dump:0x",
2688 		    g_ddb_oid->oid_fmt, len);
2689 		while (len-- && (xflag || p < val + 16))
2690 			db_printf("%02x", *p++);
2691 		if (!xflag && len > 16)
2692 			db_printf("...");
2693 		error = 0;
2694 		goto out;
2695 	}
2696 
2697 out:
2698 	req->oldidx += slen;
2699 	return (error);
2700 }
2701 
2702 /*
2703  * Avoid setting new sysctl values from the debugger
2704  */
2705 static int
2706 sysctl_new_ddb(struct sysctl_req *req, void *p, size_t l)
2707 {
2708 
2709 	if (!req->newptr)
2710 		return (0);
2711 
2712 	/* Changing sysctls from the debugger is currently unsupported */
2713 	return (EPERM);
2714 }
2715 
2716 /*
2717  * Run a sysctl handler with the DDB oldfunc and newfunc attached.
2718  * Instead of copying any output to a buffer we'll dump it right to
2719  * the console.
2720  */
2721 static int
2722 db_sysctl(struct sysctl_oid *oidp, int *name, u_int namelen,
2723     void *old, size_t *oldlenp, size_t *retval, int flags)
2724 {
2725 	struct sysctl_req req;
2726 	int error;
2727 
2728 	/* Setup the request */
2729 	bzero(&req, sizeof req);
2730 	req.td = kdb_thread;
2731 	req.oldfunc = sysctl_old_ddb;
2732 	req.newfunc = sysctl_new_ddb;
2733 	req.lock = REQ_UNWIRED;
2734 	if (oldlenp) {
2735 		req.oldlen = *oldlenp;
2736 	}
2737 	req.validlen = req.oldlen;
2738 	if (old) {
2739 		req.oldptr = old;
2740 	}
2741 
2742 	/* Setup our globals for sysctl_old_ddb */
2743 	g_ddb_oid = oidp;
2744 	g_ddb_sysctl_flags = flags;
2745 	g_ddb_sysctl_printed = 0;
2746 
2747 	error = sysctl_root(0, name, namelen, &req);
2748 
2749 	/* Reset globals */
2750 	g_ddb_oid = NULL;
2751 	g_ddb_sysctl_flags = 0;
2752 
2753 	if (retval) {
2754 		if (req.oldptr && req.oldidx > req.validlen)
2755 			*retval = req.validlen;
2756 		else
2757 			*retval = req.oldidx;
2758 	}
2759 	return (error);
2760 }
2761 
2762 /*
2763  * Show a sysctl's name
2764  */
2765 static void
2766 db_show_oid_name(int *oid, size_t nlen)
2767 {
2768 	struct sysctl_oid *oidp;
2769 	int qoid[CTL_MAXNAME+2];
2770 	int error;
2771 
2772 	qoid[0] = 0;
2773 	memcpy(qoid + 2, oid, nlen * sizeof(int));
2774 	qoid[1] = 1;
2775 
2776 	error = sysctl_find_oid(qoid, nlen + 2, &oidp, NULL, NULL);
2777 	if (error)
2778 		db_error("sysctl name oid");
2779 
2780 	error = db_sysctl(oidp, qoid, nlen + 2, NULL, NULL, NULL, 0);
2781 	if (error)
2782 		db_error("sysctl name");
2783 }
2784 
2785 /*
2786  * Check to see if an OID is safe to print from ddb.
2787  */
2788 static bool
2789 db_oid_safe(const struct sysctl_oid *oidp)
2790 {
2791 	for (unsigned int i = 0; i < nitems(db_safe_handlers); ++i) {
2792 		if (oidp->oid_handler == db_safe_handlers[i])
2793 			return (true);
2794 	}
2795 
2796 	return (false);
2797 }
2798 
2799 /*
2800  * Show a sysctl at a specific OID
2801  * Compare to the input handling in show_var from sbin/sysctl/sysctl.c
2802  */
2803 static int
2804 db_show_oid(struct sysctl_oid *oidp, int *oid, size_t nlen, int flags)
2805 {
2806 	int error, xflag, oflag, Nflag, nflag;
2807 	size_t len;
2808 
2809 	xflag = flags & DB_SYSCTL_HEX;
2810 	oflag = flags & DB_SYSCTL_OPAQUE;
2811 	nflag = flags & DB_SYSCTL_VALUE_ONLY;
2812 	Nflag = flags & DB_SYSCTL_NAME_ONLY;
2813 
2814 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_OPAQUE &&
2815 	    (!xflag && !oflag))
2816 		return (0);
2817 
2818 	if (Nflag) {
2819 		db_show_oid_name(oid, nlen);
2820 		error = 0;
2821 		goto out;
2822 	}
2823 
2824 	if (!nflag) {
2825 		db_show_oid_name(oid, nlen);
2826 		db_printf(": ");
2827 	}
2828 
2829 	if ((flags & DB_SYSCTL_SAFE_ONLY) && !db_oid_safe(oidp)) {
2830 		db_printf("Skipping, unsafe to print while recursing.");
2831 		error = 0;
2832 		goto out;
2833 	}
2834 
2835 	/* Try once, and ask about the size */
2836 	len = 0;
2837 	error = db_sysctl(oidp, oid, nlen,
2838 	    NULL, NULL, &len, flags);
2839 	if (error)
2840 		goto out;
2841 
2842 	if (!g_ddb_sysctl_printed)
2843 		/* Lie about the size */
2844 		error = db_sysctl(oidp, oid, nlen,
2845 		    (void *) 1, &len, NULL, flags);
2846 
2847 out:
2848 	db_printf("\n");
2849 	return (error);
2850 }
2851 
2852 /*
2853  * Show all sysctls under a specific OID
2854  * Compare to sysctl_all from sbin/sysctl/sysctl.c
2855  */
2856 static int
2857 db_show_sysctl_all(int *oid, size_t len, int flags)
2858 {
2859 	struct sysctl_oid *oidp;
2860 	int name1[CTL_MAXNAME + 2], name2[CTL_MAXNAME + 2];
2861 	size_t l1, l2;
2862 
2863 	name1[0] = CTL_SYSCTL;
2864 	name1[1] = CTL_SYSCTL_NEXT;
2865 	l1 = 2;
2866 	if (len) {
2867 		memcpy(name1 + 2, oid, len * sizeof(int));
2868 		l1 += len;
2869 	} else {
2870 		name1[2] = CTL_KERN;
2871 		l1++;
2872 	}
2873 	for (;;) {
2874 		int i, error;
2875 
2876 		l2 = sizeof(name2);
2877 		error = kernel_sysctl(kdb_thread, name1, l1,
2878 		    name2, &l2, NULL, 0, &l2, 0);
2879 		if (error != 0) {
2880 			if (error == ENOENT)
2881 				return (0);
2882 			else
2883 				db_error("sysctl(next)");
2884 		}
2885 
2886 		l2 /= sizeof(int);
2887 
2888 		if (l2 < (unsigned int)len)
2889 			return (0);
2890 
2891 		for (i = 0; i < len; i++)
2892 			if (name2[i] != oid[i])
2893 				return (0);
2894 
2895 		/* Find the OID in question */
2896 		error = sysctl_find_oid(name2, l2, &oidp, NULL, NULL);
2897 		if (error)
2898 			return (error);
2899 
2900 		i = db_show_oid(oidp, name2, l2, flags | DB_SYSCTL_SAFE_ONLY);
2901 
2902 		if (db_pager_quit)
2903 			return (0);
2904 
2905 		memcpy(name1+2, name2, l2 * sizeof(int));
2906 		l1 = 2 + l2;
2907 	}
2908 }
2909 
2910 /*
2911  * Show a sysctl by its user facing string
2912  */
2913 static int
2914 db_sysctlbyname(char *name, int flags)
2915 {
2916 	struct sysctl_oid *oidp;
2917 	int oid[CTL_MAXNAME];
2918 	int error, nlen;
2919 
2920 	error = name2oid(name, oid, &nlen, &oidp);
2921 	if (error) {
2922 		return (error);
2923 	}
2924 
2925 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
2926 		db_show_sysctl_all(oid, nlen, flags);
2927 	} else {
2928 		error = db_show_oid(oidp, oid, nlen, flags);
2929 	}
2930 
2931 	return (error);
2932 }
2933 
2934 static void
2935 db_sysctl_cmd_usage(void)
2936 {
2937 	db_printf(
2938 	    " sysctl [/Nnox] <sysctl>					    \n"
2939 	    "								    \n"
2940 	    " <sysctl> The name of the sysctl to show.			    \n"
2941 	    "								    \n"
2942 	    " Show a sysctl by hooking into SYSCTL_IN and SYSCTL_OUT.	    \n"
2943 	    " This will work for most sysctls, but should not be used	    \n"
2944 	    " with sysctls that are known to malloc.			    \n"
2945 	    "								    \n"
2946 	    " While recursing any \"unsafe\" sysctls will be skipped.	    \n"
2947 	    " Call sysctl directly on the sysctl to try printing the	    \n"
2948 	    " skipped sysctl. This is unsafe and may make the ddb	    \n"
2949 	    " session unusable.						    \n"
2950 	    "								    \n"
2951 	    " Arguments:						    \n"
2952 	    "	/N	Display only the name of the sysctl.		    \n"
2953 	    "	/n	Display only the value of the sysctl.		    \n"
2954 	    "	/o	Display opaque values.				    \n"
2955 	    "	/x	Display the sysctl in hex.			    \n"
2956 	    "								    \n"
2957 	    "For example:						    \n"
2958 	    "sysctl vm.v_free_min					    \n"
2959 	    "vn.v_free_min: 12669					    \n"
2960 	    );
2961 }
2962 
2963 /*
2964  * Show a specific sysctl similar to sysctl (8).
2965  */
2966 DB_COMMAND_FLAGS(sysctl, db_sysctl_cmd, CS_OWN)
2967 {
2968 	char name[TOK_STRING_SIZE];
2969 	int error, i, t, flags;
2970 
2971 	/* Parse the modifiers */
2972 	t = db_read_token();
2973 	if (t == tSLASH || t == tMINUS) {
2974 		t = db_read_token();
2975 		if (t != tIDENT) {
2976 			db_printf("Bad modifier\n");
2977 			error = EINVAL;
2978 			goto out;
2979 		}
2980 		db_strcpy(modif, db_tok_string);
2981 	}
2982 	else {
2983 		db_unread_token(t);
2984 		modif[0] = '\0';
2985 	}
2986 
2987 	flags = 0;
2988 	for (i = 0; i < nitems(db_sysctl_modifs); i++) {
2989 		if (strchr(modif, db_sysctl_modifs[i])) {
2990 			flags |= db_sysctl_modif_values[i];
2991 		}
2992 	}
2993 
2994 	/* Parse the sysctl names */
2995 	t = db_read_token();
2996 	if (t != tIDENT) {
2997 		db_printf("Need sysctl name\n");
2998 		error = EINVAL;
2999 		goto out;
3000 	}
3001 
3002 	/* Copy the name into a temporary buffer */
3003 	db_strcpy(name, db_tok_string);
3004 
3005 	/* Ensure there is no trailing cruft */
3006 	t = db_read_token();
3007 	if (t != tEOL) {
3008 		db_printf("Unexpected sysctl argument\n");
3009 		error = EINVAL;
3010 		goto out;
3011 	}
3012 
3013 	error = db_sysctlbyname(name, flags);
3014 	if (error == ENOENT) {
3015 		db_printf("unknown oid: '%s'\n", db_tok_string);
3016 		goto out;
3017 	} else if (error) {
3018 		db_printf("%s: error: %d\n", db_tok_string, error);
3019 		goto out;
3020 	}
3021 
3022 out:
3023 	/* Ensure we eat all of our text */
3024 	db_flush_lex();
3025 
3026 	if (error == EINVAL) {
3027 		db_sysctl_cmd_usage();
3028 	}
3029 }
3030 
3031 #endif /* DDB */
3032