xref: /freebsd/sys/kern/kern_sysctl.c (revision 7ec2f6bce5d28e6662c29e63f6ab6b7ef57d98b2)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Mike Karels at Berkeley Software Design, Inc.
9  *
10  * Quite extensively rewritten by Poul-Henning Kamp of the FreeBSD
11  * project, to make these variables more userfriendly.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. Neither the name of the University nor the names of its contributors
22  *    may be used to endorse or promote products derived from this software
23  *    without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  *
37  *	@(#)kern_sysctl.c	8.4 (Berkeley) 4/14/94
38  */
39 
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
42 
43 #include "opt_capsicum.h"
44 #include "opt_ddb.h"
45 #include "opt_ktrace.h"
46 #include "opt_sysctl.h"
47 
48 #include <sys/param.h>
49 #include <sys/fail.h>
50 #include <sys/systm.h>
51 #include <sys/capsicum.h>
52 #include <sys/kernel.h>
53 #include <sys/limits.h>
54 #include <sys/sysctl.h>
55 #include <sys/malloc.h>
56 #include <sys/priv.h>
57 #include <sys/proc.h>
58 #include <sys/jail.h>
59 #include <sys/kdb.h>
60 #include <sys/lock.h>
61 #include <sys/mutex.h>
62 #include <sys/rmlock.h>
63 #include <sys/sbuf.h>
64 #include <sys/sx.h>
65 #include <sys/sysproto.h>
66 #include <sys/uio.h>
67 #ifdef KTRACE
68 #include <sys/ktrace.h>
69 #endif
70 
71 #ifdef DDB
72 #include <ddb/ddb.h>
73 #include <ddb/db_lex.h>
74 #endif
75 
76 #include <net/vnet.h>
77 
78 #include <security/mac/mac_framework.h>
79 
80 #include <vm/vm.h>
81 #include <vm/vm_extern.h>
82 
83 static MALLOC_DEFINE(M_SYSCTL, "sysctl", "sysctl internal magic");
84 static MALLOC_DEFINE(M_SYSCTLOID, "sysctloid", "sysctl dynamic oids");
85 static MALLOC_DEFINE(M_SYSCTLTMP, "sysctltmp", "sysctl temp output buffer");
86 
87 /*
88  * The sysctllock protects the MIB tree.  It also protects sysctl
89  * contexts used with dynamic sysctls.  The sysctl_register_oid() and
90  * sysctl_unregister_oid() routines require the sysctllock to already
91  * be held, so the sysctl_wlock() and sysctl_wunlock() routines are
92  * provided for the few places in the kernel which need to use that
93  * API rather than using the dynamic API.  Use of the dynamic API is
94  * strongly encouraged for most code.
95  *
96  * The sysctlmemlock is used to limit the amount of user memory wired for
97  * sysctl requests.  This is implemented by serializing any userland
98  * sysctl requests larger than a single page via an exclusive lock.
99  *
100  * The sysctlstringlock is used to protect concurrent access to writable
101  * string nodes in sysctl_handle_string().
102  */
103 static struct rmlock sysctllock;
104 static struct sx __exclusive_cache_line sysctlmemlock;
105 static struct sx sysctlstringlock;
106 
107 #define	SYSCTL_WLOCK()		rm_wlock(&sysctllock)
108 #define	SYSCTL_WUNLOCK()	rm_wunlock(&sysctllock)
109 #define	SYSCTL_RLOCK(tracker)	rm_rlock(&sysctllock, (tracker))
110 #define	SYSCTL_RUNLOCK(tracker)	rm_runlock(&sysctllock, (tracker))
111 #define	SYSCTL_WLOCKED()	rm_wowned(&sysctllock)
112 #define	SYSCTL_ASSERT_LOCKED()	rm_assert(&sysctllock, RA_LOCKED)
113 #define	SYSCTL_ASSERT_WLOCKED()	rm_assert(&sysctllock, RA_WLOCKED)
114 #define	SYSCTL_ASSERT_RLOCKED()	rm_assert(&sysctllock, RA_RLOCKED)
115 #define	SYSCTL_INIT()		rm_init_flags(&sysctllock, "sysctl lock", \
116 				    RM_SLEEPABLE)
117 #define	SYSCTL_SLEEP(ch, wmesg, timo)					\
118 				rm_sleep(ch, &sysctllock, 0, wmesg, timo)
119 
120 static int sysctl_root(SYSCTL_HANDLER_ARGS);
121 
122 /* Root list */
123 struct sysctl_oid_list sysctl__children = SLIST_HEAD_INITIALIZER(&sysctl__children);
124 
125 static int	sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del,
126 		    int recurse);
127 static int	sysctl_old_kernel(struct sysctl_req *, const void *, size_t);
128 static int	sysctl_new_kernel(struct sysctl_req *, void *, size_t);
129 
130 static struct sysctl_oid *
131 sysctl_find_oidname(const char *name, struct sysctl_oid_list *list)
132 {
133 	struct sysctl_oid *oidp;
134 
135 	SYSCTL_ASSERT_LOCKED();
136 	SLIST_FOREACH(oidp, list, oid_link) {
137 		if (strcmp(oidp->oid_name, name) == 0) {
138 			return (oidp);
139 		}
140 	}
141 	return (NULL);
142 }
143 
144 /*
145  * Initialization of the MIB tree.
146  *
147  * Order by number in each list.
148  */
149 void
150 sysctl_wlock(void)
151 {
152 
153 	SYSCTL_WLOCK();
154 }
155 
156 void
157 sysctl_wunlock(void)
158 {
159 
160 	SYSCTL_WUNLOCK();
161 }
162 
163 static int
164 sysctl_root_handler_locked(struct sysctl_oid *oid, void *arg1, intmax_t arg2,
165     struct sysctl_req *req, struct rm_priotracker *tracker)
166 {
167 	int error;
168 
169 	if (oid->oid_kind & CTLFLAG_DYN)
170 		atomic_add_int(&oid->oid_running, 1);
171 
172 	if (tracker != NULL)
173 		SYSCTL_RUNLOCK(tracker);
174 	else
175 		SYSCTL_WUNLOCK();
176 
177 	/*
178 	 * Treat set CTLFLAG_NEEDGIANT and unset CTLFLAG_MPSAFE flags the same,
179 	 * untill we're ready to remove all traces of Giant from sysctl(9).
180 	 */
181 	if ((oid->oid_kind & CTLFLAG_NEEDGIANT) ||
182 	    (!(oid->oid_kind & CTLFLAG_MPSAFE)))
183 		mtx_lock(&Giant);
184 	error = oid->oid_handler(oid, arg1, arg2, req);
185 	if ((oid->oid_kind & CTLFLAG_NEEDGIANT) ||
186 	    (!(oid->oid_kind & CTLFLAG_MPSAFE)))
187 		mtx_unlock(&Giant);
188 
189 	KFAIL_POINT_ERROR(_debug_fail_point, sysctl_running, error);
190 
191 	if (tracker != NULL)
192 		SYSCTL_RLOCK(tracker);
193 	else
194 		SYSCTL_WLOCK();
195 
196 	if (oid->oid_kind & CTLFLAG_DYN) {
197 		if (atomic_fetchadd_int(&oid->oid_running, -1) == 1 &&
198 		    (oid->oid_kind & CTLFLAG_DYING) != 0)
199 			wakeup(&oid->oid_running);
200 	}
201 
202 	return (error);
203 }
204 
205 static void
206 sysctl_load_tunable_by_oid_locked(struct sysctl_oid *oidp)
207 {
208 	struct sysctl_req req;
209 	struct sysctl_oid *curr;
210 	char *penv = NULL;
211 	char path[96];
212 	ssize_t rem = sizeof(path);
213 	ssize_t len;
214 	uint8_t data[512] __aligned(sizeof(uint64_t));
215 	int size;
216 	int error;
217 
218 	path[--rem] = 0;
219 
220 	for (curr = oidp; curr != NULL; curr = SYSCTL_PARENT(curr)) {
221 		len = strlen(curr->oid_name);
222 		rem -= len;
223 		if (curr != oidp)
224 			rem -= 1;
225 		if (rem < 0) {
226 			printf("OID path exceeds %d bytes\n", (int)sizeof(path));
227 			return;
228 		}
229 		memcpy(path + rem, curr->oid_name, len);
230 		if (curr != oidp)
231 			path[rem + len] = '.';
232 	}
233 
234 	memset(&req, 0, sizeof(req));
235 
236 	req.td = curthread;
237 	req.oldfunc = sysctl_old_kernel;
238 	req.newfunc = sysctl_new_kernel;
239 	req.lock = REQ_UNWIRED;
240 
241 	switch (oidp->oid_kind & CTLTYPE) {
242 	case CTLTYPE_INT:
243 		if (getenv_array(path + rem, data, sizeof(data), &size,
244 		    sizeof(int), GETENV_SIGNED) == 0)
245 			return;
246 		req.newlen = size;
247 		req.newptr = data;
248 		break;
249 	case CTLTYPE_UINT:
250 		if (getenv_array(path + rem, data, sizeof(data), &size,
251 		    sizeof(int), GETENV_UNSIGNED) == 0)
252 			return;
253 		req.newlen = size;
254 		req.newptr = data;
255 		break;
256 	case CTLTYPE_LONG:
257 		if (getenv_array(path + rem, data, sizeof(data), &size,
258 		    sizeof(long), GETENV_SIGNED) == 0)
259 			return;
260 		req.newlen = size;
261 		req.newptr = data;
262 		break;
263 	case CTLTYPE_ULONG:
264 		if (getenv_array(path + rem, data, sizeof(data), &size,
265 		    sizeof(long), GETENV_UNSIGNED) == 0)
266 			return;
267 		req.newlen = size;
268 		req.newptr = data;
269 		break;
270 	case CTLTYPE_S8:
271 		if (getenv_array(path + rem, data, sizeof(data), &size,
272 		    sizeof(int8_t), GETENV_SIGNED) == 0)
273 			return;
274 		req.newlen = size;
275 		req.newptr = data;
276 		break;
277 	case CTLTYPE_S16:
278 		if (getenv_array(path + rem, data, sizeof(data), &size,
279 		    sizeof(int16_t), GETENV_SIGNED) == 0)
280 			return;
281 		req.newlen = size;
282 		req.newptr = data;
283 		break;
284 	case CTLTYPE_S32:
285 		if (getenv_array(path + rem, data, sizeof(data), &size,
286 		    sizeof(int32_t), GETENV_SIGNED) == 0)
287 			return;
288 		req.newlen = size;
289 		req.newptr = data;
290 		break;
291 	case CTLTYPE_S64:
292 		if (getenv_array(path + rem, data, sizeof(data), &size,
293 		    sizeof(int64_t), GETENV_SIGNED) == 0)
294 			return;
295 		req.newlen = size;
296 		req.newptr = data;
297 		break;
298 	case CTLTYPE_U8:
299 		if (getenv_array(path + rem, data, sizeof(data), &size,
300 		    sizeof(uint8_t), GETENV_UNSIGNED) == 0)
301 			return;
302 		req.newlen = size;
303 		req.newptr = data;
304 		break;
305 	case CTLTYPE_U16:
306 		if (getenv_array(path + rem, data, sizeof(data), &size,
307 		    sizeof(uint16_t), GETENV_UNSIGNED) == 0)
308 			return;
309 		req.newlen = size;
310 		req.newptr = data;
311 		break;
312 	case CTLTYPE_U32:
313 		if (getenv_array(path + rem, data, sizeof(data), &size,
314 		    sizeof(uint32_t), GETENV_UNSIGNED) == 0)
315 			return;
316 		req.newlen = size;
317 		req.newptr = data;
318 		break;
319 	case CTLTYPE_U64:
320 		if (getenv_array(path + rem, data, sizeof(data), &size,
321 		    sizeof(uint64_t), GETENV_UNSIGNED) == 0)
322 			return;
323 		req.newlen = size;
324 		req.newptr = data;
325 		break;
326 	case CTLTYPE_STRING:
327 		penv = kern_getenv(path + rem);
328 		if (penv == NULL)
329 			return;
330 		req.newlen = strlen(penv);
331 		req.newptr = penv;
332 		break;
333 	default:
334 		return;
335 	}
336 	error = sysctl_root_handler_locked(oidp, oidp->oid_arg1,
337 	    oidp->oid_arg2, &req, NULL);
338 	if (error != 0)
339 		printf("Setting sysctl %s failed: %d\n", path + rem, error);
340 	if (penv != NULL)
341 		freeenv(penv);
342 }
343 
344 /*
345  * Locate the path to a given oid.  Returns the length of the resulting path,
346  * or -1 if the oid was not found.  nodes must have room for CTL_MAXNAME
347  * elements and be NULL initialized.
348  */
349 static int
350 sysctl_search_oid(struct sysctl_oid **nodes, struct sysctl_oid *needle)
351 {
352 	int indx;
353 
354 	SYSCTL_ASSERT_LOCKED();
355 	indx = 0;
356 	while (indx < CTL_MAXNAME && indx >= 0) {
357 		if (nodes[indx] == NULL && indx == 0)
358 			nodes[indx] = SLIST_FIRST(&sysctl__children);
359 		else if (nodes[indx] == NULL)
360 			nodes[indx] = SLIST_FIRST(&nodes[indx - 1]->oid_children);
361 		else
362 			nodes[indx] = SLIST_NEXT(nodes[indx], oid_link);
363 
364 		if (nodes[indx] == needle)
365 			return (indx + 1);
366 
367 		if (nodes[indx] == NULL) {
368 			indx--;
369 			continue;
370 		}
371 
372 		if ((nodes[indx]->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
373 			indx++;
374 			continue;
375 		}
376 	}
377 	return (-1);
378 }
379 
380 static void
381 sysctl_warn_reuse(const char *func, struct sysctl_oid *leaf)
382 {
383 	struct sysctl_oid *nodes[CTL_MAXNAME];
384 	char buf[128];
385 	struct sbuf sb;
386 	int rc, i;
387 
388 	(void)sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN | SBUF_INCLUDENUL);
389 	sbuf_set_drain(&sb, sbuf_printf_drain, NULL);
390 
391 	sbuf_printf(&sb, "%s: can't re-use a leaf (", __func__);
392 
393 	memset(nodes, 0, sizeof(nodes));
394 	rc = sysctl_search_oid(nodes, leaf);
395 	if (rc > 0) {
396 		for (i = 0; i < rc; i++)
397 			sbuf_printf(&sb, "%s%.*s", nodes[i]->oid_name,
398 			    i != (rc - 1), ".");
399 	} else {
400 		sbuf_printf(&sb, "%s", leaf->oid_name);
401 	}
402 	sbuf_printf(&sb, ")!\n");
403 
404 	(void)sbuf_finish(&sb);
405 }
406 
407 #ifdef SYSCTL_DEBUG
408 static int
409 sysctl_reuse_test(SYSCTL_HANDLER_ARGS)
410 {
411 	struct rm_priotracker tracker;
412 
413 	SYSCTL_RLOCK(&tracker);
414 	sysctl_warn_reuse(__func__, oidp);
415 	SYSCTL_RUNLOCK(&tracker);
416 	return (0);
417 }
418 SYSCTL_PROC(_sysctl, OID_AUTO, reuse_test,
419     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 0, sysctl_reuse_test, "-",
420     "");
421 #endif
422 
423 void
424 sysctl_register_oid(struct sysctl_oid *oidp)
425 {
426 	struct sysctl_oid_list *parent = oidp->oid_parent;
427 	struct sysctl_oid *p;
428 	struct sysctl_oid *q;
429 	int oid_number;
430 	int timeout = 2;
431 
432 	/*
433 	 * First check if another oid with the same name already
434 	 * exists in the parent's list.
435 	 */
436 	SYSCTL_ASSERT_WLOCKED();
437 	p = sysctl_find_oidname(oidp->oid_name, parent);
438 	if (p != NULL) {
439 		if ((p->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
440 			p->oid_refcnt++;
441 			return;
442 		} else {
443 			sysctl_warn_reuse(__func__, p);
444 			return;
445 		}
446 	}
447 	/* get current OID number */
448 	oid_number = oidp->oid_number;
449 
450 #if (OID_AUTO >= 0)
451 #error "OID_AUTO is expected to be a negative value"
452 #endif
453 	/*
454 	 * Any negative OID number qualifies as OID_AUTO. Valid OID
455 	 * numbers should always be positive.
456 	 *
457 	 * NOTE: DO NOT change the starting value here, change it in
458 	 * <sys/sysctl.h>, and make sure it is at least 256 to
459 	 * accommodate e.g. net.inet.raw as a static sysctl node.
460 	 */
461 	if (oid_number < 0) {
462 		static int newoid;
463 
464 		/*
465 		 * By decrementing the next OID number we spend less
466 		 * time inserting the OIDs into a sorted list.
467 		 */
468 		if (--newoid < CTL_AUTO_START)
469 			newoid = 0x7fffffff;
470 
471 		oid_number = newoid;
472 	}
473 
474 	/*
475 	 * Insert the OID into the parent's list sorted by OID number.
476 	 */
477 retry:
478 	q = NULL;
479 	SLIST_FOREACH(p, parent, oid_link) {
480 		/* check if the current OID number is in use */
481 		if (oid_number == p->oid_number) {
482 			/* get the next valid OID number */
483 			if (oid_number < CTL_AUTO_START ||
484 			    oid_number == 0x7fffffff) {
485 				/* wraparound - restart */
486 				oid_number = CTL_AUTO_START;
487 				/* don't loop forever */
488 				if (!timeout--)
489 					panic("sysctl: Out of OID numbers\n");
490 				goto retry;
491 			} else {
492 				oid_number++;
493 			}
494 		} else if (oid_number < p->oid_number)
495 			break;
496 		q = p;
497 	}
498 	/* check for non-auto OID number collision */
499 	if (oidp->oid_number >= 0 && oidp->oid_number < CTL_AUTO_START &&
500 	    oid_number >= CTL_AUTO_START) {
501 		printf("sysctl: OID number(%d) is already in use for '%s'\n",
502 		    oidp->oid_number, oidp->oid_name);
503 	}
504 	/* update the OID number, if any */
505 	oidp->oid_number = oid_number;
506 	if (q != NULL)
507 		SLIST_INSERT_AFTER(q, oidp, oid_link);
508 	else
509 		SLIST_INSERT_HEAD(parent, oidp, oid_link);
510 
511 	if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE &&
512 #ifdef VIMAGE
513 	    (oidp->oid_kind & CTLFLAG_VNET) == 0 &&
514 #endif
515 	    (oidp->oid_kind & CTLFLAG_TUN) != 0 &&
516 	    (oidp->oid_kind & CTLFLAG_NOFETCH) == 0) {
517 		/* only fetch value once */
518 		oidp->oid_kind |= CTLFLAG_NOFETCH;
519 		/* try to fetch value from kernel environment */
520 		sysctl_load_tunable_by_oid_locked(oidp);
521 	}
522 }
523 
524 void
525 sysctl_register_disabled_oid(struct sysctl_oid *oidp)
526 {
527 
528 	/*
529 	 * Mark the leaf as dormant if it's not to be immediately enabled.
530 	 * We do not disable nodes as they can be shared between modules
531 	 * and it is always safe to access a node.
532 	 */
533 	KASSERT((oidp->oid_kind & CTLFLAG_DORMANT) == 0,
534 	    ("internal flag is set in oid_kind"));
535 	if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
536 		oidp->oid_kind |= CTLFLAG_DORMANT;
537 	sysctl_register_oid(oidp);
538 }
539 
540 void
541 sysctl_enable_oid(struct sysctl_oid *oidp)
542 {
543 
544 	SYSCTL_ASSERT_WLOCKED();
545 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
546 		KASSERT((oidp->oid_kind & CTLFLAG_DORMANT) == 0,
547 		    ("sysctl node is marked as dormant"));
548 		return;
549 	}
550 	KASSERT((oidp->oid_kind & CTLFLAG_DORMANT) != 0,
551 	    ("enabling already enabled sysctl oid"));
552 	oidp->oid_kind &= ~CTLFLAG_DORMANT;
553 }
554 
555 void
556 sysctl_unregister_oid(struct sysctl_oid *oidp)
557 {
558 	struct sysctl_oid *p;
559 	int error;
560 
561 	SYSCTL_ASSERT_WLOCKED();
562 	if (oidp->oid_number == OID_AUTO) {
563 		error = EINVAL;
564 	} else {
565 		error = ENOENT;
566 		SLIST_FOREACH(p, oidp->oid_parent, oid_link) {
567 			if (p == oidp) {
568 				SLIST_REMOVE(oidp->oid_parent, oidp,
569 				    sysctl_oid, oid_link);
570 				error = 0;
571 				break;
572 			}
573 		}
574 	}
575 
576 	/*
577 	 * This can happen when a module fails to register and is
578 	 * being unloaded afterwards.  It should not be a panic()
579 	 * for normal use.
580 	 */
581 	if (error) {
582 		printf("%s: failed(%d) to unregister sysctl(%s)\n",
583 		    __func__, error, oidp->oid_name);
584 	}
585 }
586 
587 /* Initialize a new context to keep track of dynamically added sysctls. */
588 int
589 sysctl_ctx_init(struct sysctl_ctx_list *c)
590 {
591 
592 	if (c == NULL) {
593 		return (EINVAL);
594 	}
595 
596 	/*
597 	 * No locking here, the caller is responsible for not adding
598 	 * new nodes to a context until after this function has
599 	 * returned.
600 	 */
601 	TAILQ_INIT(c);
602 	return (0);
603 }
604 
605 /* Free the context, and destroy all dynamic oids registered in this context */
606 int
607 sysctl_ctx_free(struct sysctl_ctx_list *clist)
608 {
609 	struct sysctl_ctx_entry *e, *e1;
610 	int error;
611 
612 	error = 0;
613 	/*
614 	 * First perform a "dry run" to check if it's ok to remove oids.
615 	 * XXX FIXME
616 	 * XXX This algorithm is a hack. But I don't know any
617 	 * XXX better solution for now...
618 	 */
619 	SYSCTL_WLOCK();
620 	TAILQ_FOREACH(e, clist, link) {
621 		error = sysctl_remove_oid_locked(e->entry, 0, 0);
622 		if (error)
623 			break;
624 	}
625 	/*
626 	 * Restore deregistered entries, either from the end,
627 	 * or from the place where error occurred.
628 	 * e contains the entry that was not unregistered
629 	 */
630 	if (error)
631 		e1 = TAILQ_PREV(e, sysctl_ctx_list, link);
632 	else
633 		e1 = TAILQ_LAST(clist, sysctl_ctx_list);
634 	while (e1 != NULL) {
635 		sysctl_register_oid(e1->entry);
636 		e1 = TAILQ_PREV(e1, sysctl_ctx_list, link);
637 	}
638 	if (error) {
639 		SYSCTL_WUNLOCK();
640 		return(EBUSY);
641 	}
642 	/* Now really delete the entries */
643 	e = TAILQ_FIRST(clist);
644 	while (e != NULL) {
645 		e1 = TAILQ_NEXT(e, link);
646 		error = sysctl_remove_oid_locked(e->entry, 1, 0);
647 		if (error)
648 			panic("sysctl_remove_oid: corrupt tree, entry: %s",
649 			    e->entry->oid_name);
650 		free(e, M_SYSCTLOID);
651 		e = e1;
652 	}
653 	SYSCTL_WUNLOCK();
654 	return (error);
655 }
656 
657 /* Add an entry to the context */
658 struct sysctl_ctx_entry *
659 sysctl_ctx_entry_add(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
660 {
661 	struct sysctl_ctx_entry *e;
662 
663 	SYSCTL_ASSERT_WLOCKED();
664 	if (clist == NULL || oidp == NULL)
665 		return(NULL);
666 	e = malloc(sizeof(struct sysctl_ctx_entry), M_SYSCTLOID, M_WAITOK);
667 	e->entry = oidp;
668 	TAILQ_INSERT_HEAD(clist, e, link);
669 	return (e);
670 }
671 
672 /* Find an entry in the context */
673 struct sysctl_ctx_entry *
674 sysctl_ctx_entry_find(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
675 {
676 	struct sysctl_ctx_entry *e;
677 
678 	SYSCTL_ASSERT_WLOCKED();
679 	if (clist == NULL || oidp == NULL)
680 		return(NULL);
681 	TAILQ_FOREACH(e, clist, link) {
682 		if(e->entry == oidp)
683 			return(e);
684 	}
685 	return (e);
686 }
687 
688 /*
689  * Delete an entry from the context.
690  * NOTE: this function doesn't free oidp! You have to remove it
691  * with sysctl_remove_oid().
692  */
693 int
694 sysctl_ctx_entry_del(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
695 {
696 	struct sysctl_ctx_entry *e;
697 
698 	if (clist == NULL || oidp == NULL)
699 		return (EINVAL);
700 	SYSCTL_WLOCK();
701 	e = sysctl_ctx_entry_find(clist, oidp);
702 	if (e != NULL) {
703 		TAILQ_REMOVE(clist, e, link);
704 		SYSCTL_WUNLOCK();
705 		free(e, M_SYSCTLOID);
706 		return (0);
707 	} else {
708 		SYSCTL_WUNLOCK();
709 		return (ENOENT);
710 	}
711 }
712 
713 /*
714  * Remove dynamically created sysctl trees.
715  * oidp - top of the tree to be removed
716  * del - if 0 - just deregister, otherwise free up entries as well
717  * recurse - if != 0 traverse the subtree to be deleted
718  */
719 int
720 sysctl_remove_oid(struct sysctl_oid *oidp, int del, int recurse)
721 {
722 	int error;
723 
724 	SYSCTL_WLOCK();
725 	error = sysctl_remove_oid_locked(oidp, del, recurse);
726 	SYSCTL_WUNLOCK();
727 	return (error);
728 }
729 
730 int
731 sysctl_remove_name(struct sysctl_oid *parent, const char *name,
732     int del, int recurse)
733 {
734 	struct sysctl_oid *p, *tmp;
735 	int error;
736 
737 	error = ENOENT;
738 	SYSCTL_WLOCK();
739 	SLIST_FOREACH_SAFE(p, SYSCTL_CHILDREN(parent), oid_link, tmp) {
740 		if (strcmp(p->oid_name, name) == 0) {
741 			error = sysctl_remove_oid_locked(p, del, recurse);
742 			break;
743 		}
744 	}
745 	SYSCTL_WUNLOCK();
746 
747 	return (error);
748 }
749 
750 static int
751 sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del, int recurse)
752 {
753 	struct sysctl_oid *p, *tmp;
754 	int error;
755 
756 	SYSCTL_ASSERT_WLOCKED();
757 	if (oidp == NULL)
758 		return(EINVAL);
759 	if ((oidp->oid_kind & CTLFLAG_DYN) == 0) {
760 		printf("Warning: can't remove non-dynamic nodes (%s)!\n",
761 		    oidp->oid_name);
762 		return (EINVAL);
763 	}
764 	/*
765 	 * WARNING: normal method to do this should be through
766 	 * sysctl_ctx_free(). Use recursing as the last resort
767 	 * method to purge your sysctl tree of leftovers...
768 	 * However, if some other code still references these nodes,
769 	 * it will panic.
770 	 */
771 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
772 		if (oidp->oid_refcnt == 1) {
773 			SLIST_FOREACH_SAFE(p,
774 			    SYSCTL_CHILDREN(oidp), oid_link, tmp) {
775 				if (!recurse) {
776 					printf("Warning: failed attempt to "
777 					    "remove oid %s with child %s\n",
778 					    oidp->oid_name, p->oid_name);
779 					return (ENOTEMPTY);
780 				}
781 				error = sysctl_remove_oid_locked(p, del,
782 				    recurse);
783 				if (error)
784 					return (error);
785 			}
786 		}
787 	}
788 	if (oidp->oid_refcnt > 1 ) {
789 		oidp->oid_refcnt--;
790 	} else {
791 		if (oidp->oid_refcnt == 0) {
792 			printf("Warning: bad oid_refcnt=%u (%s)!\n",
793 				oidp->oid_refcnt, oidp->oid_name);
794 			return (EINVAL);
795 		}
796 		sysctl_unregister_oid(oidp);
797 		if (del) {
798 			/*
799 			 * Wait for all threads running the handler to drain.
800 			 * This preserves the previous behavior when the
801 			 * sysctl lock was held across a handler invocation,
802 			 * and is necessary for module unload correctness.
803 			 */
804 			while (oidp->oid_running > 0) {
805 				oidp->oid_kind |= CTLFLAG_DYING;
806 				SYSCTL_SLEEP(&oidp->oid_running, "oidrm", 0);
807 			}
808 			if (oidp->oid_descr)
809 				free(__DECONST(char *, oidp->oid_descr),
810 				    M_SYSCTLOID);
811 			if (oidp->oid_label)
812 				free(__DECONST(char *, oidp->oid_label),
813 				    M_SYSCTLOID);
814 			free(__DECONST(char *, oidp->oid_name), M_SYSCTLOID);
815 			free(oidp, M_SYSCTLOID);
816 		}
817 	}
818 	return (0);
819 }
820 /*
821  * Create new sysctls at run time.
822  * clist may point to a valid context initialized with sysctl_ctx_init().
823  */
824 struct sysctl_oid *
825 sysctl_add_oid(struct sysctl_ctx_list *clist, struct sysctl_oid_list *parent,
826 	int number, const char *name, int kind, void *arg1, intmax_t arg2,
827 	int (*handler)(SYSCTL_HANDLER_ARGS), const char *fmt, const char *descr,
828 	const char *label)
829 {
830 	struct sysctl_oid *oidp;
831 
832 	/* You have to hook up somewhere.. */
833 	if (parent == NULL)
834 		return(NULL);
835 	/* Check if the node already exists, otherwise create it */
836 	SYSCTL_WLOCK();
837 	oidp = sysctl_find_oidname(name, parent);
838 	if (oidp != NULL) {
839 		if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
840 			oidp->oid_refcnt++;
841 			/* Update the context */
842 			if (clist != NULL)
843 				sysctl_ctx_entry_add(clist, oidp);
844 			SYSCTL_WUNLOCK();
845 			return (oidp);
846 		} else {
847 			sysctl_warn_reuse(__func__, oidp);
848 			SYSCTL_WUNLOCK();
849 			return (NULL);
850 		}
851 	}
852 	oidp = malloc(sizeof(struct sysctl_oid), M_SYSCTLOID, M_WAITOK|M_ZERO);
853 	oidp->oid_parent = parent;
854 	SLIST_INIT(&oidp->oid_children);
855 	oidp->oid_number = number;
856 	oidp->oid_refcnt = 1;
857 	oidp->oid_name = strdup(name, M_SYSCTLOID);
858 	oidp->oid_handler = handler;
859 	oidp->oid_kind = CTLFLAG_DYN | kind;
860 	oidp->oid_arg1 = arg1;
861 	oidp->oid_arg2 = arg2;
862 	oidp->oid_fmt = fmt;
863 	if (descr != NULL)
864 		oidp->oid_descr = strdup(descr, M_SYSCTLOID);
865 	if (label != NULL)
866 		oidp->oid_label = strdup(label, M_SYSCTLOID);
867 	/* Update the context, if used */
868 	if (clist != NULL)
869 		sysctl_ctx_entry_add(clist, oidp);
870 	/* Register this oid */
871 	sysctl_register_oid(oidp);
872 	SYSCTL_WUNLOCK();
873 	return (oidp);
874 }
875 
876 /*
877  * Rename an existing oid.
878  */
879 void
880 sysctl_rename_oid(struct sysctl_oid *oidp, const char *name)
881 {
882 	char *newname;
883 	char *oldname;
884 
885 	newname = strdup(name, M_SYSCTLOID);
886 	SYSCTL_WLOCK();
887 	oldname = __DECONST(char *, oidp->oid_name);
888 	oidp->oid_name = newname;
889 	SYSCTL_WUNLOCK();
890 	free(oldname, M_SYSCTLOID);
891 }
892 
893 /*
894  * Reparent an existing oid.
895  */
896 int
897 sysctl_move_oid(struct sysctl_oid *oid, struct sysctl_oid_list *parent)
898 {
899 	struct sysctl_oid *oidp;
900 
901 	SYSCTL_WLOCK();
902 	if (oid->oid_parent == parent) {
903 		SYSCTL_WUNLOCK();
904 		return (0);
905 	}
906 	oidp = sysctl_find_oidname(oid->oid_name, parent);
907 	if (oidp != NULL) {
908 		SYSCTL_WUNLOCK();
909 		return (EEXIST);
910 	}
911 	sysctl_unregister_oid(oid);
912 	oid->oid_parent = parent;
913 	oid->oid_number = OID_AUTO;
914 	sysctl_register_oid(oid);
915 	SYSCTL_WUNLOCK();
916 	return (0);
917 }
918 
919 /*
920  * Register the kernel's oids on startup.
921  */
922 SET_DECLARE(sysctl_set, struct sysctl_oid);
923 
924 static void
925 sysctl_register_all(void *arg)
926 {
927 	struct sysctl_oid **oidp;
928 
929 	sx_init(&sysctlmemlock, "sysctl mem");
930 	sx_init(&sysctlstringlock, "sysctl string handler");
931 	SYSCTL_INIT();
932 	SYSCTL_WLOCK();
933 	SET_FOREACH(oidp, sysctl_set)
934 		sysctl_register_oid(*oidp);
935 	SYSCTL_WUNLOCK();
936 }
937 SYSINIT(sysctl, SI_SUB_KMEM, SI_ORDER_FIRST, sysctl_register_all, NULL);
938 
939 /*
940  * "Staff-functions"
941  *
942  * These functions implement a presently undocumented interface
943  * used by the sysctl program to walk the tree, and get the type
944  * so it can print the value.
945  * This interface is under work and consideration, and should probably
946  * be killed with a big axe by the first person who can find the time.
947  * (be aware though, that the proper interface isn't as obvious as it
948  * may seem, there are various conflicting requirements.
949  *
950  * {CTL_SYSCTL, CTL_SYSCTL_DEBUG}		printf the entire MIB-tree.
951  * {CTL_SYSCTL, CTL_SYSCTL_NAME, ...}		return the name of the "..."
952  *						OID.
953  * {CTL_SYSCTL, CTL_SYSCTL_NEXT, ...}		return the next OID, honoring
954  *						CTLFLAG_SKIP.
955  * {CTL_SYSCTL, CTL_SYSCTL_NAME2OID}		return the OID of the name in
956  *						"new"
957  * {CTL_SYSCTL, CTL_SYSCTL_OIDFMT, ...}		return the kind & format info
958  *						for the "..." OID.
959  * {CTL_SYSCTL, CTL_SYSCTL_OIDDESCR, ...}	return the description of the
960  *						"..." OID.
961  * {CTL_SYSCTL, CTL_SYSCTL_OIDLABEL, ...}	return the aggregation label of
962  *						the "..." OID.
963  * {CTL_SYSCTL, CTL_SYSCTL_NEXTNOSKIP, ...}	return the next OID, ignoring
964  *						CTLFLAG_SKIP.
965  */
966 
967 #ifdef SYSCTL_DEBUG
968 static void
969 sysctl_sysctl_debug_dump_node(struct sysctl_oid_list *l, int i)
970 {
971 	int k;
972 	struct sysctl_oid *oidp;
973 
974 	SYSCTL_ASSERT_LOCKED();
975 	SLIST_FOREACH(oidp, l, oid_link) {
976 		for (k=0; k<i; k++)
977 			printf(" ");
978 
979 		printf("%d %s ", oidp->oid_number, oidp->oid_name);
980 
981 		printf("%c%c",
982 			oidp->oid_kind & CTLFLAG_RD ? 'R':' ',
983 			oidp->oid_kind & CTLFLAG_WR ? 'W':' ');
984 
985 		if (oidp->oid_handler)
986 			printf(" *Handler");
987 
988 		switch (oidp->oid_kind & CTLTYPE) {
989 			case CTLTYPE_NODE:
990 				printf(" Node\n");
991 				if (!oidp->oid_handler) {
992 					sysctl_sysctl_debug_dump_node(
993 					    SYSCTL_CHILDREN(oidp), i + 2);
994 				}
995 				break;
996 			case CTLTYPE_INT:    printf(" Int\n"); break;
997 			case CTLTYPE_UINT:   printf(" u_int\n"); break;
998 			case CTLTYPE_LONG:   printf(" Long\n"); break;
999 			case CTLTYPE_ULONG:  printf(" u_long\n"); break;
1000 			case CTLTYPE_STRING: printf(" String\n"); break;
1001 			case CTLTYPE_S8:     printf(" int8_t\n"); break;
1002 			case CTLTYPE_S16:    printf(" int16_t\n"); break;
1003 			case CTLTYPE_S32:    printf(" int32_t\n"); break;
1004 			case CTLTYPE_S64:    printf(" int64_t\n"); break;
1005 			case CTLTYPE_U8:     printf(" uint8_t\n"); break;
1006 			case CTLTYPE_U16:    printf(" uint16_t\n"); break;
1007 			case CTLTYPE_U32:    printf(" uint32_t\n"); break;
1008 			case CTLTYPE_U64:    printf(" uint64_t\n"); break;
1009 			case CTLTYPE_OPAQUE: printf(" Opaque/struct\n"); break;
1010 			default:	     printf("\n");
1011 		}
1012 	}
1013 }
1014 
1015 static int
1016 sysctl_sysctl_debug(SYSCTL_HANDLER_ARGS)
1017 {
1018 	struct rm_priotracker tracker;
1019 	int error;
1020 
1021 	error = priv_check(req->td, PRIV_SYSCTL_DEBUG);
1022 	if (error)
1023 		return (error);
1024 	SYSCTL_RLOCK(&tracker);
1025 	sysctl_sysctl_debug_dump_node(&sysctl__children, 0);
1026 	SYSCTL_RUNLOCK(&tracker);
1027 	return (ENOENT);
1028 }
1029 
1030 SYSCTL_PROC(_sysctl, CTL_SYSCTL_DEBUG, debug, CTLTYPE_STRING | CTLFLAG_RD |
1031     CTLFLAG_MPSAFE, 0, 0, sysctl_sysctl_debug, "-", "");
1032 #endif
1033 
1034 static int
1035 sysctl_sysctl_name(SYSCTL_HANDLER_ARGS)
1036 {
1037 	int *name = (int *) arg1;
1038 	u_int namelen = arg2;
1039 	int error;
1040 	struct sysctl_oid *oid;
1041 	struct sysctl_oid_list *lsp = &sysctl__children, *lsp2;
1042 	struct rm_priotracker tracker;
1043 	char buf[10];
1044 
1045 	error = sysctl_wire_old_buffer(req, 0);
1046 	if (error)
1047 		return (error);
1048 
1049 	SYSCTL_RLOCK(&tracker);
1050 	while (namelen) {
1051 		if (!lsp) {
1052 			snprintf(buf,sizeof(buf),"%d",*name);
1053 			if (req->oldidx)
1054 				error = SYSCTL_OUT(req, ".", 1);
1055 			if (!error)
1056 				error = SYSCTL_OUT(req, buf, strlen(buf));
1057 			if (error)
1058 				goto out;
1059 			namelen--;
1060 			name++;
1061 			continue;
1062 		}
1063 		lsp2 = NULL;
1064 		SLIST_FOREACH(oid, lsp, oid_link) {
1065 			if (oid->oid_number != *name)
1066 				continue;
1067 
1068 			if (req->oldidx)
1069 				error = SYSCTL_OUT(req, ".", 1);
1070 			if (!error)
1071 				error = SYSCTL_OUT(req, oid->oid_name,
1072 					strlen(oid->oid_name));
1073 			if (error)
1074 				goto out;
1075 
1076 			namelen--;
1077 			name++;
1078 
1079 			if ((oid->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1080 				break;
1081 
1082 			if (oid->oid_handler)
1083 				break;
1084 
1085 			lsp2 = SYSCTL_CHILDREN(oid);
1086 			break;
1087 		}
1088 		lsp = lsp2;
1089 	}
1090 	error = SYSCTL_OUT(req, "", 1);
1091  out:
1092 	SYSCTL_RUNLOCK(&tracker);
1093 	return (error);
1094 }
1095 
1096 /*
1097  * XXXRW/JA: Shouldn't return name data for nodes that we don't permit in
1098  * capability mode.
1099  */
1100 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_NAME, name, CTLFLAG_RD |
1101     CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_name, "");
1102 
1103 static int
1104 sysctl_sysctl_next_ls(struct sysctl_oid_list *lsp, int *name, u_int namelen,
1105     int *next, int *len, int level, struct sysctl_oid **oidpp, bool honor_skip)
1106 {
1107 	struct sysctl_oid *oidp;
1108 
1109 	SYSCTL_ASSERT_LOCKED();
1110 	*len = level;
1111 	SLIST_FOREACH(oidp, lsp, oid_link) {
1112 		*next = oidp->oid_number;
1113 		*oidpp = oidp;
1114 
1115 		if ((oidp->oid_kind & CTLFLAG_DORMANT) != 0)
1116 			continue;
1117 
1118 		if (honor_skip && (oidp->oid_kind & CTLFLAG_SKIP) != 0)
1119 			continue;
1120 
1121 		if (!namelen) {
1122 			if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1123 				return (0);
1124 			if (oidp->oid_handler)
1125 				/* We really should call the handler here...*/
1126 				return (0);
1127 			lsp = SYSCTL_CHILDREN(oidp);
1128 			if (!sysctl_sysctl_next_ls(lsp, 0, 0, next+1,
1129 				len, level+1, oidpp, honor_skip))
1130 				return (0);
1131 			goto emptynode;
1132 		}
1133 
1134 		if (oidp->oid_number < *name)
1135 			continue;
1136 
1137 		if (oidp->oid_number > *name) {
1138 			if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1139 				return (0);
1140 			if (oidp->oid_handler)
1141 				return (0);
1142 			lsp = SYSCTL_CHILDREN(oidp);
1143 			if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1,
1144 				next+1, len, level+1, oidpp, honor_skip))
1145 				return (0);
1146 			goto next;
1147 		}
1148 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1149 			continue;
1150 
1151 		if (oidp->oid_handler)
1152 			continue;
1153 
1154 		lsp = SYSCTL_CHILDREN(oidp);
1155 		if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1, next+1,
1156 			len, level+1, oidpp, honor_skip))
1157 			return (0);
1158 	next:
1159 		namelen = 1;
1160 	emptynode:
1161 		*len = level;
1162 	}
1163 	return (ENOENT);
1164 }
1165 
1166 static int
1167 sysctl_sysctl_next(SYSCTL_HANDLER_ARGS)
1168 {
1169 	int *name = (int *) arg1;
1170 	u_int namelen = arg2;
1171 	int len, error;
1172 	struct sysctl_oid *oid;
1173 	struct sysctl_oid_list *lsp = &sysctl__children;
1174 	struct rm_priotracker tracker;
1175 	int next[CTL_MAXNAME];
1176 
1177 	SYSCTL_RLOCK(&tracker);
1178 	error = sysctl_sysctl_next_ls(lsp, name, namelen, next, &len, 1, &oid,
1179 	    oidp->oid_number == CTL_SYSCTL_NEXT);
1180 	SYSCTL_RUNLOCK(&tracker);
1181 	if (error)
1182 		return (error);
1183 	error = SYSCTL_OUT(req, next, len * sizeof (int));
1184 	return (error);
1185 }
1186 
1187 /*
1188  * XXXRW/JA: Shouldn't return next data for nodes that we don't permit in
1189  * capability mode.
1190  */
1191 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_NEXT, next, CTLFLAG_RD |
1192     CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_next, "");
1193 
1194 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_NEXTNOSKIP, nextnoskip, CTLFLAG_RD |
1195     CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_next, "");
1196 
1197 static int
1198 name2oid(char *name, int *oid, int *len, struct sysctl_oid **oidpp)
1199 {
1200 	struct sysctl_oid *oidp;
1201 	struct sysctl_oid_list *lsp = &sysctl__children;
1202 	char *p;
1203 
1204 	SYSCTL_ASSERT_LOCKED();
1205 
1206 	for (*len = 0; *len < CTL_MAXNAME;) {
1207 		p = strsep(&name, ".");
1208 
1209 		oidp = SLIST_FIRST(lsp);
1210 		for (;; oidp = SLIST_NEXT(oidp, oid_link)) {
1211 			if (oidp == NULL)
1212 				return (ENOENT);
1213 			if (strcmp(p, oidp->oid_name) == 0)
1214 				break;
1215 		}
1216 		*oid++ = oidp->oid_number;
1217 		(*len)++;
1218 
1219 		if (name == NULL || *name == '\0') {
1220 			if (oidpp)
1221 				*oidpp = oidp;
1222 			return (0);
1223 		}
1224 
1225 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1226 			break;
1227 
1228 		if (oidp->oid_handler)
1229 			break;
1230 
1231 		lsp = SYSCTL_CHILDREN(oidp);
1232 	}
1233 	return (ENOENT);
1234 }
1235 
1236 static int
1237 sysctl_sysctl_name2oid(SYSCTL_HANDLER_ARGS)
1238 {
1239 	char *p;
1240 	int error, oid[CTL_MAXNAME], len = 0;
1241 	struct sysctl_oid *op = NULL;
1242 	struct rm_priotracker tracker;
1243 	char buf[32];
1244 
1245 	if (!req->newlen)
1246 		return (ENOENT);
1247 	if (req->newlen >= MAXPATHLEN)	/* XXX arbitrary, undocumented */
1248 		return (ENAMETOOLONG);
1249 
1250 	p = buf;
1251 	if (req->newlen >= sizeof(buf))
1252 		p = malloc(req->newlen+1, M_SYSCTL, M_WAITOK);
1253 
1254 	error = SYSCTL_IN(req, p, req->newlen);
1255 	if (error) {
1256 		if (p != buf)
1257 			free(p, M_SYSCTL);
1258 		return (error);
1259 	}
1260 
1261 	p [req->newlen] = '\0';
1262 
1263 	SYSCTL_RLOCK(&tracker);
1264 	error = name2oid(p, oid, &len, &op);
1265 	SYSCTL_RUNLOCK(&tracker);
1266 
1267 	if (p != buf)
1268 		free(p, M_SYSCTL);
1269 
1270 	if (error)
1271 		return (error);
1272 
1273 	error = SYSCTL_OUT(req, oid, len * sizeof *oid);
1274 	return (error);
1275 }
1276 
1277 /*
1278  * XXXRW/JA: Shouldn't return name2oid data for nodes that we don't permit in
1279  * capability mode.
1280  */
1281 SYSCTL_PROC(_sysctl, CTL_SYSCTL_NAME2OID, name2oid, CTLTYPE_INT | CTLFLAG_RW |
1282     CTLFLAG_ANYBODY | CTLFLAG_MPSAFE | CTLFLAG_CAPRW, 0, 0,
1283     sysctl_sysctl_name2oid, "I", "");
1284 
1285 static int
1286 sysctl_sysctl_oidfmt(SYSCTL_HANDLER_ARGS)
1287 {
1288 	struct sysctl_oid *oid;
1289 	struct rm_priotracker tracker;
1290 	int error;
1291 
1292 	error = sysctl_wire_old_buffer(req, 0);
1293 	if (error)
1294 		return (error);
1295 
1296 	SYSCTL_RLOCK(&tracker);
1297 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1298 	if (error)
1299 		goto out;
1300 
1301 	if (oid->oid_fmt == NULL) {
1302 		error = ENOENT;
1303 		goto out;
1304 	}
1305 	error = SYSCTL_OUT(req, &oid->oid_kind, sizeof(oid->oid_kind));
1306 	if (error)
1307 		goto out;
1308 	error = SYSCTL_OUT(req, oid->oid_fmt, strlen(oid->oid_fmt) + 1);
1309  out:
1310 	SYSCTL_RUNLOCK(&tracker);
1311 	return (error);
1312 }
1313 
1314 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_OIDFMT, oidfmt, CTLFLAG_RD |
1315     CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_oidfmt, "");
1316 
1317 static int
1318 sysctl_sysctl_oiddescr(SYSCTL_HANDLER_ARGS)
1319 {
1320 	struct sysctl_oid *oid;
1321 	struct rm_priotracker tracker;
1322 	int error;
1323 
1324 	error = sysctl_wire_old_buffer(req, 0);
1325 	if (error)
1326 		return (error);
1327 
1328 	SYSCTL_RLOCK(&tracker);
1329 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1330 	if (error)
1331 		goto out;
1332 
1333 	if (oid->oid_descr == NULL) {
1334 		error = ENOENT;
1335 		goto out;
1336 	}
1337 	error = SYSCTL_OUT(req, oid->oid_descr, strlen(oid->oid_descr) + 1);
1338  out:
1339 	SYSCTL_RUNLOCK(&tracker);
1340 	return (error);
1341 }
1342 
1343 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_OIDDESCR, oiddescr, CTLFLAG_RD |
1344     CTLFLAG_MPSAFE|CTLFLAG_CAPRD, sysctl_sysctl_oiddescr, "");
1345 
1346 static int
1347 sysctl_sysctl_oidlabel(SYSCTL_HANDLER_ARGS)
1348 {
1349 	struct sysctl_oid *oid;
1350 	struct rm_priotracker tracker;
1351 	int error;
1352 
1353 	error = sysctl_wire_old_buffer(req, 0);
1354 	if (error)
1355 		return (error);
1356 
1357 	SYSCTL_RLOCK(&tracker);
1358 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1359 	if (error)
1360 		goto out;
1361 
1362 	if (oid->oid_label == NULL) {
1363 		error = ENOENT;
1364 		goto out;
1365 	}
1366 	error = SYSCTL_OUT(req, oid->oid_label, strlen(oid->oid_label) + 1);
1367  out:
1368 	SYSCTL_RUNLOCK(&tracker);
1369 	return (error);
1370 }
1371 
1372 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_OIDLABEL, oidlabel, CTLFLAG_RD |
1373     CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_oidlabel, "");
1374 
1375 /*
1376  * Default "handler" functions.
1377  */
1378 
1379 /*
1380  * Handle a bool.
1381  * Two cases:
1382  *     a variable:  point arg1 at it.
1383  *     a constant:  pass it in arg2.
1384  */
1385 
1386 int
1387 sysctl_handle_bool(SYSCTL_HANDLER_ARGS)
1388 {
1389 	uint8_t temp;
1390 	int error;
1391 
1392 	/*
1393 	 * Attempt to get a coherent snapshot by making a copy of the data.
1394 	 */
1395 	if (arg1)
1396 		temp = *(bool *)arg1 ? 1 : 0;
1397 	else
1398 		temp = arg2 ? 1 : 0;
1399 
1400 	error = SYSCTL_OUT(req, &temp, sizeof(temp));
1401 	if (error || !req->newptr)
1402 		return (error);
1403 
1404 	if (!arg1)
1405 		error = EPERM;
1406 	else {
1407 		error = SYSCTL_IN(req, &temp, sizeof(temp));
1408 		if (!error)
1409 			*(bool *)arg1 = temp ? 1 : 0;
1410 	}
1411 	return (error);
1412 }
1413 
1414 /*
1415  * Handle an int8_t, signed or unsigned.
1416  * Two cases:
1417  *     a variable:  point arg1 at it.
1418  *     a constant:  pass it in arg2.
1419  */
1420 
1421 int
1422 sysctl_handle_8(SYSCTL_HANDLER_ARGS)
1423 {
1424 	int8_t tmpout;
1425 	int error = 0;
1426 
1427 	/*
1428 	 * Attempt to get a coherent snapshot by making a copy of the data.
1429 	 */
1430 	if (arg1)
1431 		tmpout = *(int8_t *)arg1;
1432 	else
1433 		tmpout = arg2;
1434 	error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout));
1435 
1436 	if (error || !req->newptr)
1437 		return (error);
1438 
1439 	if (!arg1)
1440 		error = EPERM;
1441 	else
1442 		error = SYSCTL_IN(req, arg1, sizeof(tmpout));
1443 	return (error);
1444 }
1445 
1446 /*
1447  * Handle an int16_t, signed or unsigned.
1448  * Two cases:
1449  *     a variable:  point arg1 at it.
1450  *     a constant:  pass it in arg2.
1451  */
1452 
1453 int
1454 sysctl_handle_16(SYSCTL_HANDLER_ARGS)
1455 {
1456 	int16_t tmpout;
1457 	int error = 0;
1458 
1459 	/*
1460 	 * Attempt to get a coherent snapshot by making a copy of the data.
1461 	 */
1462 	if (arg1)
1463 		tmpout = *(int16_t *)arg1;
1464 	else
1465 		tmpout = arg2;
1466 	error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout));
1467 
1468 	if (error || !req->newptr)
1469 		return (error);
1470 
1471 	if (!arg1)
1472 		error = EPERM;
1473 	else
1474 		error = SYSCTL_IN(req, arg1, sizeof(tmpout));
1475 	return (error);
1476 }
1477 
1478 /*
1479  * Handle an int32_t, signed or unsigned.
1480  * Two cases:
1481  *     a variable:  point arg1 at it.
1482  *     a constant:  pass it in arg2.
1483  */
1484 
1485 int
1486 sysctl_handle_32(SYSCTL_HANDLER_ARGS)
1487 {
1488 	int32_t tmpout;
1489 	int error = 0;
1490 
1491 	/*
1492 	 * Attempt to get a coherent snapshot by making a copy of the data.
1493 	 */
1494 	if (arg1)
1495 		tmpout = *(int32_t *)arg1;
1496 	else
1497 		tmpout = arg2;
1498 	error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout));
1499 
1500 	if (error || !req->newptr)
1501 		return (error);
1502 
1503 	if (!arg1)
1504 		error = EPERM;
1505 	else
1506 		error = SYSCTL_IN(req, arg1, sizeof(tmpout));
1507 	return (error);
1508 }
1509 
1510 /*
1511  * Handle an int, signed or unsigned.
1512  * Two cases:
1513  *     a variable:  point arg1 at it.
1514  *     a constant:  pass it in arg2.
1515  */
1516 
1517 int
1518 sysctl_handle_int(SYSCTL_HANDLER_ARGS)
1519 {
1520 	int tmpout, error = 0;
1521 
1522 	/*
1523 	 * Attempt to get a coherent snapshot by making a copy of the data.
1524 	 */
1525 	if (arg1)
1526 		tmpout = *(int *)arg1;
1527 	else
1528 		tmpout = arg2;
1529 	error = SYSCTL_OUT(req, &tmpout, sizeof(int));
1530 
1531 	if (error || !req->newptr)
1532 		return (error);
1533 
1534 	if (!arg1)
1535 		error = EPERM;
1536 	else
1537 		error = SYSCTL_IN(req, arg1, sizeof(int));
1538 	return (error);
1539 }
1540 
1541 /*
1542  * Based on on sysctl_handle_int() convert milliseconds into ticks.
1543  * Note: this is used by TCP.
1544  */
1545 
1546 int
1547 sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
1548 {
1549 	int error, s, tt;
1550 
1551 	tt = *(int *)arg1;
1552 	s = (int)((int64_t)tt * 1000 / hz);
1553 
1554 	error = sysctl_handle_int(oidp, &s, 0, req);
1555 	if (error || !req->newptr)
1556 		return (error);
1557 
1558 	tt = (int)((int64_t)s * hz / 1000);
1559 	if (tt < 1)
1560 		return (EINVAL);
1561 
1562 	*(int *)arg1 = tt;
1563 	return (0);
1564 }
1565 
1566 /*
1567  * Handle a long, 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_long(SYSCTL_HANDLER_ARGS)
1575 {
1576 	int error = 0;
1577 	long tmplong;
1578 #ifdef SCTL_MASK32
1579 	int tmpint;
1580 #endif
1581 
1582 	/*
1583 	 * Attempt to get a coherent snapshot by making a copy of the data.
1584 	 */
1585 	if (arg1)
1586 		tmplong = *(long *)arg1;
1587 	else
1588 		tmplong = arg2;
1589 #ifdef SCTL_MASK32
1590 	if (req->flags & SCTL_MASK32) {
1591 		tmpint = tmplong;
1592 		error = SYSCTL_OUT(req, &tmpint, sizeof(int));
1593 	} else
1594 #endif
1595 		error = SYSCTL_OUT(req, &tmplong, sizeof(long));
1596 
1597 	if (error || !req->newptr)
1598 		return (error);
1599 
1600 	if (!arg1)
1601 		error = EPERM;
1602 #ifdef SCTL_MASK32
1603 	else if (req->flags & SCTL_MASK32) {
1604 		error = SYSCTL_IN(req, &tmpint, sizeof(int));
1605 		*(long *)arg1 = (long)tmpint;
1606 	}
1607 #endif
1608 	else
1609 		error = SYSCTL_IN(req, arg1, sizeof(long));
1610 	return (error);
1611 }
1612 
1613 /*
1614  * Handle a 64 bit int, signed or unsigned.
1615  * Two cases:
1616  *     a variable:  point arg1 at it.
1617  *     a constant:  pass it in arg2.
1618  */
1619 int
1620 sysctl_handle_64(SYSCTL_HANDLER_ARGS)
1621 {
1622 	int error = 0;
1623 	uint64_t tmpout;
1624 
1625 	/*
1626 	 * Attempt to get a coherent snapshot by making a copy of the data.
1627 	 */
1628 	if (arg1)
1629 		tmpout = *(uint64_t *)arg1;
1630 	else
1631 		tmpout = arg2;
1632 	error = SYSCTL_OUT(req, &tmpout, sizeof(uint64_t));
1633 
1634 	if (error || !req->newptr)
1635 		return (error);
1636 
1637 	if (!arg1)
1638 		error = EPERM;
1639 	else
1640 		error = SYSCTL_IN(req, arg1, sizeof(uint64_t));
1641 	return (error);
1642 }
1643 
1644 /*
1645  * Handle our generic '\0' terminated 'C' string.
1646  * Two cases:
1647  * 	a variable string:  point arg1 at it, arg2 is max length.
1648  * 	a constant string:  point arg1 at it, arg2 is zero.
1649  */
1650 
1651 int
1652 sysctl_handle_string(SYSCTL_HANDLER_ARGS)
1653 {
1654 	char *tmparg;
1655 	size_t outlen;
1656 	int error = 0, ro_string = 0;
1657 
1658 	/*
1659 	 * If the sysctl isn't writable and isn't a preallocated tunable that
1660 	 * can be modified by kenv(2), microoptimise and treat it as a
1661 	 * read-only string.
1662 	 * A zero-length buffer indicates a fixed size read-only
1663 	 * string.  In ddb, don't worry about trying to make a malloced
1664 	 * snapshot.
1665 	 */
1666 	if ((oidp->oid_kind & (CTLFLAG_WR | CTLFLAG_TUN)) == 0 ||
1667 	    arg2 == 0 || kdb_active) {
1668 		arg2 = strlen((char *)arg1) + 1;
1669 		ro_string = 1;
1670 	}
1671 
1672 	if (req->oldptr != NULL) {
1673 		if (ro_string) {
1674 			tmparg = arg1;
1675 			outlen = strlen(tmparg) + 1;
1676 		} else {
1677 			tmparg = malloc(arg2, M_SYSCTLTMP, M_WAITOK);
1678 			sx_slock(&sysctlstringlock);
1679 			memcpy(tmparg, arg1, arg2);
1680 			sx_sunlock(&sysctlstringlock);
1681 			outlen = strlen(tmparg) + 1;
1682 		}
1683 
1684 		error = SYSCTL_OUT(req, tmparg, outlen);
1685 
1686 		if (!ro_string)
1687 			free(tmparg, M_SYSCTLTMP);
1688 	} else {
1689 		if (!ro_string)
1690 			sx_slock(&sysctlstringlock);
1691 		outlen = strlen((char *)arg1) + 1;
1692 		if (!ro_string)
1693 			sx_sunlock(&sysctlstringlock);
1694 		error = SYSCTL_OUT(req, NULL, outlen);
1695 	}
1696 	if (error || !req->newptr)
1697 		return (error);
1698 
1699 	if (req->newlen - req->newidx >= arg2 ||
1700 	    req->newlen - req->newidx < 0) {
1701 		error = EINVAL;
1702 	} else if (req->newlen - req->newidx == 0) {
1703 		sx_xlock(&sysctlstringlock);
1704 		((char *)arg1)[0] = '\0';
1705 		sx_xunlock(&sysctlstringlock);
1706 	} else {
1707 		arg2 = req->newlen - req->newidx;
1708 		tmparg = malloc(arg2, M_SYSCTLTMP, M_WAITOK);
1709 
1710 		error = SYSCTL_IN(req, tmparg, arg2);
1711 		if (error) {
1712 			free(tmparg, M_SYSCTLTMP);
1713 			return (error);
1714 		}
1715 
1716 		sx_xlock(&sysctlstringlock);
1717 		memcpy(arg1, tmparg, arg2);
1718 		((char *)arg1)[arg2] = '\0';
1719 		sx_xunlock(&sysctlstringlock);
1720 		free(tmparg, M_SYSCTLTMP);
1721 		req->newidx += arg2;
1722 	}
1723 	return (error);
1724 }
1725 
1726 /*
1727  * Handle any kind of opaque data.
1728  * arg1 points to it, arg2 is the size.
1729  */
1730 
1731 int
1732 sysctl_handle_opaque(SYSCTL_HANDLER_ARGS)
1733 {
1734 	int error, tries;
1735 	u_int generation;
1736 	struct sysctl_req req2;
1737 
1738 	/*
1739 	 * Attempt to get a coherent snapshot, by using the thread
1740 	 * pre-emption counter updated from within mi_switch() to
1741 	 * determine if we were pre-empted during a bcopy() or
1742 	 * copyout(). Make 3 attempts at doing this before giving up.
1743 	 * If we encounter an error, stop immediately.
1744 	 */
1745 	tries = 0;
1746 	req2 = *req;
1747 retry:
1748 	generation = curthread->td_generation;
1749 	error = SYSCTL_OUT(req, arg1, arg2);
1750 	if (error)
1751 		return (error);
1752 	tries++;
1753 	if (generation != curthread->td_generation && tries < 3) {
1754 		*req = req2;
1755 		goto retry;
1756 	}
1757 
1758 	error = SYSCTL_IN(req, arg1, arg2);
1759 
1760 	return (error);
1761 }
1762 
1763 /*
1764  * Based on on sysctl_handle_int() convert microseconds to a sbintime.
1765  */
1766 int
1767 sysctl_usec_to_sbintime(SYSCTL_HANDLER_ARGS)
1768 {
1769 	int error;
1770 	int64_t tt;
1771 	sbintime_t sb;
1772 
1773 	tt = *(int64_t *)arg1;
1774 	sb = sbttous(tt);
1775 
1776 	error = sysctl_handle_64(oidp, &sb, 0, req);
1777 	if (error || !req->newptr)
1778 		return (error);
1779 
1780 	tt = ustosbt(sb);
1781 	*(int64_t *)arg1 = tt;
1782 
1783 	return (0);
1784 }
1785 
1786 /*
1787  * Based on on sysctl_handle_int() convert milliseconds to a sbintime.
1788  */
1789 int
1790 sysctl_msec_to_sbintime(SYSCTL_HANDLER_ARGS)
1791 {
1792 	int error;
1793 	int64_t tt;
1794 	sbintime_t sb;
1795 
1796 	tt = *(int64_t *)arg1;
1797 	sb = sbttoms(tt);
1798 
1799 	error = sysctl_handle_64(oidp, &sb, 0, req);
1800 	if (error || !req->newptr)
1801 		return (error);
1802 
1803 	tt = mstosbt(sb);
1804 	*(int64_t *)arg1 = tt;
1805 
1806 	return (0);
1807 }
1808 
1809 /*
1810  * Convert seconds to a struct timeval.  Intended for use with
1811  * intervals and thus does not permit negative seconds.
1812  */
1813 int
1814 sysctl_sec_to_timeval(SYSCTL_HANDLER_ARGS)
1815 {
1816 	struct timeval *tv;
1817 	int error, secs;
1818 
1819 	tv = arg1;
1820 	secs = tv->tv_sec;
1821 
1822 	error = sysctl_handle_int(oidp, &secs, 0, req);
1823 	if (error || req->newptr == NULL)
1824 		return (error);
1825 
1826 	if (secs < 0)
1827 		return (EINVAL);
1828 	tv->tv_sec = secs;
1829 
1830 	return (0);
1831 }
1832 
1833 /*
1834  * Transfer functions to/from kernel space.
1835  * XXX: rather untested at this point
1836  */
1837 static int
1838 sysctl_old_kernel(struct sysctl_req *req, const void *p, size_t l)
1839 {
1840 	size_t i = 0;
1841 
1842 	if (req->oldptr) {
1843 		i = l;
1844 		if (req->oldlen <= req->oldidx)
1845 			i = 0;
1846 		else
1847 			if (i > req->oldlen - req->oldidx)
1848 				i = req->oldlen - req->oldidx;
1849 		if (i > 0)
1850 			bcopy(p, (char *)req->oldptr + req->oldidx, i);
1851 	}
1852 	req->oldidx += l;
1853 	if (req->oldptr && i != l)
1854 		return (ENOMEM);
1855 	return (0);
1856 }
1857 
1858 static int
1859 sysctl_new_kernel(struct sysctl_req *req, void *p, size_t l)
1860 {
1861 	if (!req->newptr)
1862 		return (0);
1863 	if (req->newlen - req->newidx < l)
1864 		return (EINVAL);
1865 	bcopy((const char *)req->newptr + req->newidx, p, l);
1866 	req->newidx += l;
1867 	return (0);
1868 }
1869 
1870 int
1871 kernel_sysctl(struct thread *td, int *name, u_int namelen, void *old,
1872     size_t *oldlenp, void *new, size_t newlen, size_t *retval, int flags)
1873 {
1874 	int error = 0;
1875 	struct sysctl_req req;
1876 
1877 	bzero(&req, sizeof req);
1878 
1879 	req.td = td;
1880 	req.flags = flags;
1881 
1882 	if (oldlenp) {
1883 		req.oldlen = *oldlenp;
1884 	}
1885 	req.validlen = req.oldlen;
1886 
1887 	if (old) {
1888 		req.oldptr= old;
1889 	}
1890 
1891 	if (new != NULL) {
1892 		req.newlen = newlen;
1893 		req.newptr = new;
1894 	}
1895 
1896 	req.oldfunc = sysctl_old_kernel;
1897 	req.newfunc = sysctl_new_kernel;
1898 	req.lock = REQ_UNWIRED;
1899 
1900 	error = sysctl_root(0, name, namelen, &req);
1901 
1902 	if (req.lock == REQ_WIRED && req.validlen > 0)
1903 		vsunlock(req.oldptr, req.validlen);
1904 
1905 	if (error && error != ENOMEM)
1906 		return (error);
1907 
1908 	if (retval) {
1909 		if (req.oldptr && req.oldidx > req.validlen)
1910 			*retval = req.validlen;
1911 		else
1912 			*retval = req.oldidx;
1913 	}
1914 	return (error);
1915 }
1916 
1917 int
1918 kernel_sysctlbyname(struct thread *td, char *name, void *old, size_t *oldlenp,
1919     void *new, size_t newlen, size_t *retval, int flags)
1920 {
1921         int oid[CTL_MAXNAME];
1922         size_t oidlen, plen;
1923 	int error;
1924 
1925 	oid[0] = CTL_SYSCTL;
1926 	oid[1] = CTL_SYSCTL_NAME2OID;
1927 	oidlen = sizeof(oid);
1928 
1929 	error = kernel_sysctl(td, oid, 2, oid, &oidlen,
1930 	    (void *)name, strlen(name), &plen, flags);
1931 	if (error)
1932 		return (error);
1933 
1934 	error = kernel_sysctl(td, oid, plen / sizeof(int), old, oldlenp,
1935 	    new, newlen, retval, flags);
1936 	return (error);
1937 }
1938 
1939 /*
1940  * Transfer function to/from user space.
1941  */
1942 static int
1943 sysctl_old_user(struct sysctl_req *req, const void *p, size_t l)
1944 {
1945 	size_t i, len, origidx;
1946 	int error;
1947 
1948 	origidx = req->oldidx;
1949 	req->oldidx += l;
1950 	if (req->oldptr == NULL)
1951 		return (0);
1952 	/*
1953 	 * If we have not wired the user supplied buffer and we are currently
1954 	 * holding locks, drop a witness warning, as it's possible that
1955 	 * write operations to the user page can sleep.
1956 	 */
1957 	if (req->lock != REQ_WIRED)
1958 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
1959 		    "sysctl_old_user()");
1960 	i = l;
1961 	len = req->validlen;
1962 	if (len <= origidx)
1963 		i = 0;
1964 	else {
1965 		if (i > len - origidx)
1966 			i = len - origidx;
1967 		if (req->lock == REQ_WIRED) {
1968 			error = copyout_nofault(p, (char *)req->oldptr +
1969 			    origidx, i);
1970 		} else
1971 			error = copyout(p, (char *)req->oldptr + origidx, i);
1972 		if (error != 0)
1973 			return (error);
1974 	}
1975 	if (i < l)
1976 		return (ENOMEM);
1977 	return (0);
1978 }
1979 
1980 static int
1981 sysctl_new_user(struct sysctl_req *req, void *p, size_t l)
1982 {
1983 	int error;
1984 
1985 	if (!req->newptr)
1986 		return (0);
1987 	if (req->newlen - req->newidx < l)
1988 		return (EINVAL);
1989 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
1990 	    "sysctl_new_user()");
1991 	error = copyin((const char *)req->newptr + req->newidx, p, l);
1992 	req->newidx += l;
1993 	return (error);
1994 }
1995 
1996 /*
1997  * Wire the user space destination buffer.  If set to a value greater than
1998  * zero, the len parameter limits the maximum amount of wired memory.
1999  */
2000 int
2001 sysctl_wire_old_buffer(struct sysctl_req *req, size_t len)
2002 {
2003 	int ret;
2004 	size_t wiredlen;
2005 
2006 	wiredlen = (len > 0 && len < req->oldlen) ? len : req->oldlen;
2007 	ret = 0;
2008 	if (req->lock != REQ_WIRED && req->oldptr &&
2009 	    req->oldfunc == sysctl_old_user) {
2010 		if (wiredlen != 0) {
2011 			ret = vslock(req->oldptr, wiredlen);
2012 			if (ret != 0) {
2013 				if (ret != ENOMEM)
2014 					return (ret);
2015 				wiredlen = 0;
2016 			}
2017 		}
2018 		req->lock = REQ_WIRED;
2019 		req->validlen = wiredlen;
2020 	}
2021 	return (0);
2022 }
2023 
2024 int
2025 sysctl_find_oid(int *name, u_int namelen, struct sysctl_oid **noid,
2026     int *nindx, struct sysctl_req *req)
2027 {
2028 	struct sysctl_oid_list *lsp;
2029 	struct sysctl_oid *oid;
2030 	int indx;
2031 
2032 	SYSCTL_ASSERT_LOCKED();
2033 	lsp = &sysctl__children;
2034 	indx = 0;
2035 	while (indx < CTL_MAXNAME) {
2036 		SLIST_FOREACH(oid, lsp, oid_link) {
2037 			if (oid->oid_number == name[indx])
2038 				break;
2039 		}
2040 		if (oid == NULL)
2041 			return (ENOENT);
2042 
2043 		indx++;
2044 		if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
2045 			if (oid->oid_handler != NULL || indx == namelen) {
2046 				*noid = oid;
2047 				if (nindx != NULL)
2048 					*nindx = indx;
2049 				KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
2050 				    ("%s found DYING node %p", __func__, oid));
2051 				return (0);
2052 			}
2053 			lsp = SYSCTL_CHILDREN(oid);
2054 		} else if (indx == namelen) {
2055 			if ((oid->oid_kind & CTLFLAG_DORMANT) != 0)
2056 				return (ENOENT);
2057 			*noid = oid;
2058 			if (nindx != NULL)
2059 				*nindx = indx;
2060 			KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
2061 			    ("%s found DYING node %p", __func__, oid));
2062 			return (0);
2063 		} else {
2064 			return (ENOTDIR);
2065 		}
2066 	}
2067 	return (ENOENT);
2068 }
2069 
2070 /*
2071  * Traverse our tree, and find the right node, execute whatever it points
2072  * to, and return the resulting error code.
2073  */
2074 
2075 static int
2076 sysctl_root(SYSCTL_HANDLER_ARGS)
2077 {
2078 	struct sysctl_oid *oid;
2079 	struct rm_priotracker tracker;
2080 	int error, indx, lvl;
2081 
2082 	SYSCTL_RLOCK(&tracker);
2083 
2084 	error = sysctl_find_oid(arg1, arg2, &oid, &indx, req);
2085 	if (error)
2086 		goto out;
2087 
2088 	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
2089 		/*
2090 		 * You can't call a sysctl when it's a node, but has
2091 		 * no handler.  Inform the user that it's a node.
2092 		 * The indx may or may not be the same as namelen.
2093 		 */
2094 		if (oid->oid_handler == NULL) {
2095 			error = EISDIR;
2096 			goto out;
2097 		}
2098 	}
2099 
2100 	/* Is this sysctl writable? */
2101 	if (req->newptr && !(oid->oid_kind & CTLFLAG_WR)) {
2102 		error = EPERM;
2103 		goto out;
2104 	}
2105 
2106 	KASSERT(req->td != NULL, ("sysctl_root(): req->td == NULL"));
2107 
2108 #ifdef CAPABILITY_MODE
2109 	/*
2110 	 * If the process is in capability mode, then don't permit reading or
2111 	 * writing unless specifically granted for the node.
2112 	 */
2113 	if (IN_CAPABILITY_MODE(req->td)) {
2114 		if ((req->oldptr && !(oid->oid_kind & CTLFLAG_CAPRD)) ||
2115 		    (req->newptr && !(oid->oid_kind & CTLFLAG_CAPWR))) {
2116 			error = EPERM;
2117 			goto out;
2118 		}
2119 	}
2120 #endif
2121 
2122 	/* Is this sysctl sensitive to securelevels? */
2123 	if (req->newptr && (oid->oid_kind & CTLFLAG_SECURE)) {
2124 		lvl = (oid->oid_kind & CTLMASK_SECURE) >> CTLSHIFT_SECURE;
2125 		error = securelevel_gt(req->td->td_ucred, lvl);
2126 		if (error)
2127 			goto out;
2128 	}
2129 
2130 	/* Is this sysctl writable by only privileged users? */
2131 	if (req->newptr && !(oid->oid_kind & CTLFLAG_ANYBODY)) {
2132 		int priv;
2133 
2134 		if (oid->oid_kind & CTLFLAG_PRISON)
2135 			priv = PRIV_SYSCTL_WRITEJAIL;
2136 #ifdef VIMAGE
2137 		else if ((oid->oid_kind & CTLFLAG_VNET) &&
2138 		     prison_owns_vnet(req->td->td_ucred))
2139 			priv = PRIV_SYSCTL_WRITEJAIL;
2140 #endif
2141 		else
2142 			priv = PRIV_SYSCTL_WRITE;
2143 		error = priv_check(req->td, priv);
2144 		if (error)
2145 			goto out;
2146 	}
2147 
2148 	if (!oid->oid_handler) {
2149 		error = EINVAL;
2150 		goto out;
2151 	}
2152 
2153 	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
2154 		arg1 = (int *)arg1 + indx;
2155 		arg2 -= indx;
2156 	} else {
2157 		arg1 = oid->oid_arg1;
2158 		arg2 = oid->oid_arg2;
2159 	}
2160 #ifdef MAC
2161 	error = mac_system_check_sysctl(req->td->td_ucred, oid, arg1, arg2,
2162 	    req);
2163 	if (error != 0)
2164 		goto out;
2165 #endif
2166 #ifdef VIMAGE
2167 	if ((oid->oid_kind & CTLFLAG_VNET) && arg1 != NULL)
2168 		arg1 = (void *)(curvnet->vnet_data_base + (uintptr_t)arg1);
2169 #endif
2170 	error = sysctl_root_handler_locked(oid, arg1, arg2, req, &tracker);
2171 
2172 out:
2173 	SYSCTL_RUNLOCK(&tracker);
2174 	return (error);
2175 }
2176 
2177 #ifndef _SYS_SYSPROTO_H_
2178 struct sysctl_args {
2179 	int	*name;
2180 	u_int	namelen;
2181 	void	*old;
2182 	size_t	*oldlenp;
2183 	void	*new;
2184 	size_t	newlen;
2185 };
2186 #endif
2187 int
2188 sys___sysctl(struct thread *td, struct sysctl_args *uap)
2189 {
2190 	int error, i, name[CTL_MAXNAME];
2191 	size_t j;
2192 
2193 	if (uap->namelen > CTL_MAXNAME || uap->namelen < 2)
2194 		return (EINVAL);
2195 
2196  	error = copyin(uap->name, &name, uap->namelen * sizeof(int));
2197  	if (error)
2198 		return (error);
2199 
2200 	error = userland_sysctl(td, name, uap->namelen,
2201 		uap->old, uap->oldlenp, 0,
2202 		uap->new, uap->newlen, &j, 0);
2203 	if (error && error != ENOMEM)
2204 		return (error);
2205 	if (uap->oldlenp) {
2206 		i = copyout(&j, uap->oldlenp, sizeof(j));
2207 		if (i)
2208 			return (i);
2209 	}
2210 	return (error);
2211 }
2212 
2213 int
2214 kern___sysctlbyname(struct thread *td, const char *oname, size_t namelen,
2215     void *old, size_t *oldlenp, void *new, size_t newlen, size_t *retval,
2216     int flags, bool inkernel)
2217 {
2218 	int oid[CTL_MAXNAME];
2219 	char namebuf[16];
2220 	char *name;
2221 	size_t oidlen;
2222 	int error;
2223 
2224 	if (namelen > MAXPATHLEN || namelen == 0)
2225 		return (EINVAL);
2226 	name = namebuf;
2227 	if (namelen > sizeof(namebuf))
2228 		name = malloc(namelen, M_SYSCTL, M_WAITOK);
2229 	error = copyin(oname, name, namelen);
2230 	if (error != 0)
2231 		goto out;
2232 
2233 	oid[0] = CTL_SYSCTL;
2234 	oid[1] = CTL_SYSCTL_NAME2OID;
2235 	oidlen = sizeof(oid);
2236 	error = kernel_sysctl(td, oid, 2, oid, &oidlen, (void *)name, namelen,
2237 	    retval, flags);
2238 	if (error != 0)
2239 		goto out;
2240 	error = userland_sysctl(td, oid, *retval / sizeof(int), old, oldlenp,
2241 	    inkernel, new, newlen, retval, flags);
2242 
2243 out:
2244 	if (namelen > sizeof(namebuf))
2245 		free(name, M_SYSCTL);
2246 	return (error);
2247 }
2248 
2249 #ifndef	_SYS_SYSPROTO_H_
2250 struct __sysctlbyname_args {
2251 	const char	*name;
2252 	size_t	namelen;
2253 	void	*old;
2254 	size_t	*oldlenp;
2255 	void	*new;
2256 	size_t	newlen;
2257 };
2258 #endif
2259 int
2260 sys___sysctlbyname(struct thread *td, struct __sysctlbyname_args *uap)
2261 {
2262 	size_t rv;
2263 	int error;
2264 
2265 	error = kern___sysctlbyname(td, uap->name, uap->namelen, uap->old,
2266 	    uap->oldlenp, uap->new, uap->newlen, &rv, 0, 0);
2267 	if (error != 0)
2268 		return (error);
2269 	if (uap->oldlenp != NULL)
2270 		error = copyout(&rv, uap->oldlenp, sizeof(rv));
2271 
2272 	return (error);
2273 }
2274 
2275 /*
2276  * This is used from various compatibility syscalls too.  That's why name
2277  * must be in kernel space.
2278  */
2279 int
2280 userland_sysctl(struct thread *td, int *name, u_int namelen, void *old,
2281     size_t *oldlenp, int inkernel, const void *new, size_t newlen,
2282     size_t *retval, int flags)
2283 {
2284 	int error = 0, memlocked;
2285 	struct sysctl_req req;
2286 
2287 	bzero(&req, sizeof req);
2288 
2289 	req.td = td;
2290 	req.flags = flags;
2291 
2292 	if (oldlenp) {
2293 		if (inkernel) {
2294 			req.oldlen = *oldlenp;
2295 		} else {
2296 			error = copyin(oldlenp, &req.oldlen, sizeof(*oldlenp));
2297 			if (error)
2298 				return (error);
2299 		}
2300 	}
2301 	req.validlen = req.oldlen;
2302 	req.oldptr = old;
2303 
2304 	if (new != NULL) {
2305 		req.newlen = newlen;
2306 		req.newptr = new;
2307 	}
2308 
2309 	req.oldfunc = sysctl_old_user;
2310 	req.newfunc = sysctl_new_user;
2311 	req.lock = REQ_UNWIRED;
2312 
2313 #ifdef KTRACE
2314 	if (KTRPOINT(curthread, KTR_SYSCTL))
2315 		ktrsysctl(name, namelen);
2316 #endif
2317 	memlocked = 0;
2318 	if (req.oldptr && req.oldlen > 4 * PAGE_SIZE) {
2319 		memlocked = 1;
2320 		sx_xlock(&sysctlmemlock);
2321 	}
2322 	CURVNET_SET(TD_TO_VNET(td));
2323 
2324 	for (;;) {
2325 		req.oldidx = 0;
2326 		req.newidx = 0;
2327 		error = sysctl_root(0, name, namelen, &req);
2328 		if (error != EAGAIN)
2329 			break;
2330 		kern_yield(PRI_USER);
2331 	}
2332 
2333 	CURVNET_RESTORE();
2334 
2335 	if (req.lock == REQ_WIRED && req.validlen > 0)
2336 		vsunlock(req.oldptr, req.validlen);
2337 	if (memlocked)
2338 		sx_xunlock(&sysctlmemlock);
2339 
2340 	if (error && error != ENOMEM)
2341 		return (error);
2342 
2343 	if (retval) {
2344 		if (req.oldptr && req.oldidx > req.validlen)
2345 			*retval = req.validlen;
2346 		else
2347 			*retval = req.oldidx;
2348 	}
2349 	return (error);
2350 }
2351 
2352 /*
2353  * Drain into a sysctl struct.  The user buffer should be wired if a page
2354  * fault would cause issue.
2355  */
2356 static int
2357 sbuf_sysctl_drain(void *arg, const char *data, int len)
2358 {
2359 	struct sysctl_req *req = arg;
2360 	int error;
2361 
2362 	error = SYSCTL_OUT(req, data, len);
2363 	KASSERT(error >= 0, ("Got unexpected negative value %d", error));
2364 	return (error == 0 ? len : -error);
2365 }
2366 
2367 struct sbuf *
2368 sbuf_new_for_sysctl(struct sbuf *s, char *buf, int length,
2369     struct sysctl_req *req)
2370 {
2371 
2372 	/* Supply a default buffer size if none given. */
2373 	if (buf == NULL && length == 0)
2374 		length = 64;
2375 	s = sbuf_new(s, buf, length, SBUF_FIXEDLEN | SBUF_INCLUDENUL);
2376 	sbuf_set_drain(s, sbuf_sysctl_drain, req);
2377 	return (s);
2378 }
2379 
2380 #ifdef DDB
2381 
2382 /* The current OID the debugger is working with */
2383 static struct sysctl_oid *g_ddb_oid;
2384 
2385 /* The current flags specified by the user */
2386 static int g_ddb_sysctl_flags;
2387 
2388 /* Check to see if the last sysctl printed */
2389 static int g_ddb_sysctl_printed;
2390 
2391 static const int ctl_sign[CTLTYPE+1] = {
2392 	[CTLTYPE_INT] = 1,
2393 	[CTLTYPE_LONG] = 1,
2394 	[CTLTYPE_S8] = 1,
2395 	[CTLTYPE_S16] = 1,
2396 	[CTLTYPE_S32] = 1,
2397 	[CTLTYPE_S64] = 1,
2398 };
2399 
2400 static const int ctl_size[CTLTYPE+1] = {
2401 	[CTLTYPE_INT] = sizeof(int),
2402 	[CTLTYPE_UINT] = sizeof(u_int),
2403 	[CTLTYPE_LONG] = sizeof(long),
2404 	[CTLTYPE_ULONG] = sizeof(u_long),
2405 	[CTLTYPE_S8] = sizeof(int8_t),
2406 	[CTLTYPE_S16] = sizeof(int16_t),
2407 	[CTLTYPE_S32] = sizeof(int32_t),
2408 	[CTLTYPE_S64] = sizeof(int64_t),
2409 	[CTLTYPE_U8] = sizeof(uint8_t),
2410 	[CTLTYPE_U16] = sizeof(uint16_t),
2411 	[CTLTYPE_U32] = sizeof(uint32_t),
2412 	[CTLTYPE_U64] = sizeof(uint64_t),
2413 };
2414 
2415 #define DB_SYSCTL_NAME_ONLY	0x001	/* Compare with -N */
2416 #define DB_SYSCTL_VALUE_ONLY	0x002	/* Compare with -n */
2417 #define DB_SYSCTL_OPAQUE	0x004	/* Compare with -o */
2418 #define DB_SYSCTL_HEX		0x008	/* Compare with -x */
2419 
2420 #define DB_SYSCTL_SAFE_ONLY	0x100	/* Only simple types */
2421 
2422 static const char db_sysctl_modifs[] = {
2423 	'N', 'n', 'o', 'x',
2424 };
2425 
2426 static const int db_sysctl_modif_values[] = {
2427 	DB_SYSCTL_NAME_ONLY, DB_SYSCTL_VALUE_ONLY,
2428 	DB_SYSCTL_OPAQUE, DB_SYSCTL_HEX,
2429 };
2430 
2431 /* Handlers considered safe to print while recursing */
2432 static int (* const db_safe_handlers[])(SYSCTL_HANDLER_ARGS) = {
2433 	sysctl_handle_bool,
2434 	sysctl_handle_8,
2435 	sysctl_handle_16,
2436 	sysctl_handle_32,
2437 	sysctl_handle_64,
2438 	sysctl_handle_int,
2439 	sysctl_handle_long,
2440 	sysctl_handle_string,
2441 	sysctl_handle_opaque,
2442 };
2443 
2444 /*
2445  * Use in place of sysctl_old_kernel to print sysctl values.
2446  *
2447  * Compare to the output handling in show_var from sbin/sysctl/sysctl.c
2448  */
2449 static int
2450 sysctl_old_ddb(struct sysctl_req *req, const void *ptr, size_t len)
2451 {
2452 	const u_char *val, *p;
2453 	const char *sep1;
2454 	size_t intlen, slen;
2455 	uintmax_t umv;
2456 	intmax_t mv;
2457 	int sign, ctltype, hexlen, xflag, error;
2458 
2459 	/* Suppress false-positive GCC uninitialized variable warnings */
2460 	mv = 0;
2461 	umv = 0;
2462 
2463 	slen = len;
2464 	val = p = ptr;
2465 
2466 	if (ptr == NULL) {
2467 		error = 0;
2468 		goto out;
2469 	}
2470 
2471 	/* We are going to print */
2472 	g_ddb_sysctl_printed = 1;
2473 
2474 	xflag = g_ddb_sysctl_flags & DB_SYSCTL_HEX;
2475 
2476 	ctltype = (g_ddb_oid->oid_kind & CTLTYPE);
2477 	sign = ctl_sign[ctltype];
2478 	intlen = ctl_size[ctltype];
2479 
2480 	switch (ctltype) {
2481 	case CTLTYPE_NODE:
2482 	case CTLTYPE_STRING:
2483 		db_printf("%.*s", (int) len, (const char *) p);
2484 		error = 0;
2485 		goto out;
2486 
2487 	case CTLTYPE_INT:
2488 	case CTLTYPE_UINT:
2489 	case CTLTYPE_LONG:
2490 	case CTLTYPE_ULONG:
2491 	case CTLTYPE_S8:
2492 	case CTLTYPE_S16:
2493 	case CTLTYPE_S32:
2494 	case CTLTYPE_S64:
2495 	case CTLTYPE_U8:
2496 	case CTLTYPE_U16:
2497 	case CTLTYPE_U32:
2498 	case CTLTYPE_U64:
2499 		hexlen = 2 + (intlen * CHAR_BIT + 3) / 4;
2500 		sep1 = "";
2501 		while (len >= intlen) {
2502 			switch (ctltype) {
2503 			case CTLTYPE_INT:
2504 			case CTLTYPE_UINT:
2505 				umv = *(const u_int *)p;
2506 				mv = *(const int *)p;
2507 				break;
2508 			case CTLTYPE_LONG:
2509 			case CTLTYPE_ULONG:
2510 				umv = *(const u_long *)p;
2511 				mv = *(const long *)p;
2512 				break;
2513 			case CTLTYPE_S8:
2514 			case CTLTYPE_U8:
2515 				umv = *(const uint8_t *)p;
2516 				mv = *(const int8_t *)p;
2517 				break;
2518 			case CTLTYPE_S16:
2519 			case CTLTYPE_U16:
2520 				umv = *(const uint16_t *)p;
2521 				mv = *(const int16_t *)p;
2522 				break;
2523 			case CTLTYPE_S32:
2524 			case CTLTYPE_U32:
2525 				umv = *(const uint32_t *)p;
2526 				mv = *(const int32_t *)p;
2527 				break;
2528 			case CTLTYPE_S64:
2529 			case CTLTYPE_U64:
2530 				umv = *(const uint64_t *)p;
2531 				mv = *(const int64_t *)p;
2532 				break;
2533 			}
2534 
2535 			db_printf("%s", sep1);
2536 			if (xflag)
2537 				db_printf("%#0*jx", hexlen, umv);
2538 			else if (!sign)
2539 				db_printf("%ju", umv);
2540 			else if (g_ddb_oid->oid_fmt[1] == 'K') {
2541 				/* Kelvins are currently unsupported. */
2542 				error = EOPNOTSUPP;
2543 				goto out;
2544 			} else
2545 				db_printf("%jd", mv);
2546 
2547 			sep1 = " ";
2548 			len -= intlen;
2549 			p += intlen;
2550 		}
2551 		error = 0;
2552 		goto out;
2553 
2554 	case CTLTYPE_OPAQUE:
2555 		/* TODO: Support struct functions. */
2556 
2557 		/* FALLTHROUGH */
2558 	default:
2559 		db_printf("Format:%s Length:%zu Dump:0x",
2560 		    g_ddb_oid->oid_fmt, len);
2561 		while (len-- && (xflag || p < val + 16))
2562 			db_printf("%02x", *p++);
2563 		if (!xflag && len > 16)
2564 			db_printf("...");
2565 		error = 0;
2566 		goto out;
2567 	}
2568 
2569 out:
2570 	req->oldidx += slen;
2571 	return (error);
2572 }
2573 
2574 /*
2575  * Avoid setting new sysctl values from the debugger
2576  */
2577 static int
2578 sysctl_new_ddb(struct sysctl_req *req, void *p, size_t l)
2579 {
2580 
2581 	if (!req->newptr)
2582 		return (0);
2583 
2584 	/* Changing sysctls from the debugger is currently unsupported */
2585 	return (EPERM);
2586 }
2587 
2588 /*
2589  * Run a sysctl handler with the DDB oldfunc and newfunc attached.
2590  * Instead of copying any output to a buffer we'll dump it right to
2591  * the console.
2592  */
2593 static int
2594 db_sysctl(struct sysctl_oid *oidp, int *name, u_int namelen,
2595     void *old, size_t *oldlenp, size_t *retval, int flags)
2596 {
2597 	struct sysctl_req req;
2598 	int error;
2599 
2600 	/* Setup the request */
2601 	bzero(&req, sizeof req);
2602 	req.td = kdb_thread;
2603 	req.oldfunc = sysctl_old_ddb;
2604 	req.newfunc = sysctl_new_ddb;
2605 	req.lock = REQ_UNWIRED;
2606 	if (oldlenp) {
2607 		req.oldlen = *oldlenp;
2608 	}
2609 	req.validlen = req.oldlen;
2610 	if (old) {
2611 		req.oldptr = old;
2612 	}
2613 
2614 	/* Setup our globals for sysctl_old_ddb */
2615 	g_ddb_oid = oidp;
2616 	g_ddb_sysctl_flags = flags;
2617 	g_ddb_sysctl_printed = 0;
2618 
2619 	error = sysctl_root(0, name, namelen, &req);
2620 
2621 	/* Reset globals */
2622 	g_ddb_oid = NULL;
2623 	g_ddb_sysctl_flags = 0;
2624 
2625 	if (retval) {
2626 		if (req.oldptr && req.oldidx > req.validlen)
2627 			*retval = req.validlen;
2628 		else
2629 			*retval = req.oldidx;
2630 	}
2631 	return (error);
2632 }
2633 
2634 /*
2635  * Show a sysctl's name
2636  */
2637 static void
2638 db_show_oid_name(int *oid, size_t nlen)
2639 {
2640 	struct sysctl_oid *oidp;
2641 	int qoid[CTL_MAXNAME+2];
2642 	int error;
2643 
2644 	qoid[0] = 0;
2645 	memcpy(qoid + 2, oid, nlen * sizeof(int));
2646 	qoid[1] = 1;
2647 
2648 	error = sysctl_find_oid(qoid, nlen + 2, &oidp, NULL, NULL);
2649 	if (error)
2650 		db_error("sysctl name oid");
2651 
2652 	error = db_sysctl(oidp, qoid, nlen + 2, NULL, NULL, NULL, 0);
2653 	if (error)
2654 		db_error("sysctl name");
2655 }
2656 
2657 /*
2658  * Check to see if an OID is safe to print from ddb.
2659  */
2660 static bool
2661 db_oid_safe(const struct sysctl_oid *oidp)
2662 {
2663 	for (unsigned int i = 0; i < nitems(db_safe_handlers); ++i) {
2664 		if (oidp->oid_handler == db_safe_handlers[i])
2665 			return (true);
2666 	}
2667 
2668 	return (false);
2669 }
2670 
2671 /*
2672  * Show a sysctl at a specific OID
2673  * Compare to the input handling in show_var from sbin/sysctl/sysctl.c
2674  */
2675 static int
2676 db_show_oid(struct sysctl_oid *oidp, int *oid, size_t nlen, int flags)
2677 {
2678 	int error, xflag, oflag, Nflag, nflag;
2679 	size_t len;
2680 
2681 	xflag = flags & DB_SYSCTL_HEX;
2682 	oflag = flags & DB_SYSCTL_OPAQUE;
2683 	nflag = flags & DB_SYSCTL_VALUE_ONLY;
2684 	Nflag = flags & DB_SYSCTL_NAME_ONLY;
2685 
2686 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_OPAQUE &&
2687 	    (!xflag && !oflag))
2688 		return (0);
2689 
2690 	if (Nflag) {
2691 		db_show_oid_name(oid, nlen);
2692 		error = 0;
2693 		goto out;
2694 	}
2695 
2696 	if (!nflag) {
2697 		db_show_oid_name(oid, nlen);
2698 		db_printf(": ");
2699 	}
2700 
2701 	if ((flags & DB_SYSCTL_SAFE_ONLY) && !db_oid_safe(oidp)) {
2702 		db_printf("Skipping, unsafe to print while recursing.");
2703 		error = 0;
2704 		goto out;
2705 	}
2706 
2707 	/* Try once, and ask about the size */
2708 	len = 0;
2709 	error = db_sysctl(oidp, oid, nlen,
2710 	    NULL, NULL, &len, flags);
2711 	if (error)
2712 		goto out;
2713 
2714 	if (!g_ddb_sysctl_printed)
2715 		/* Lie about the size */
2716 		error = db_sysctl(oidp, oid, nlen,
2717 		    (void *) 1, &len, NULL, flags);
2718 
2719 out:
2720 	db_printf("\n");
2721 	return (error);
2722 }
2723 
2724 /*
2725  * Show all sysctls under a specific OID
2726  * Compare to sysctl_all from sbin/sysctl/sysctl.c
2727  */
2728 static int
2729 db_show_sysctl_all(int *oid, size_t len, int flags)
2730 {
2731 	struct sysctl_oid *oidp;
2732 	int name1[CTL_MAXNAME + 2], name2[CTL_MAXNAME + 2];
2733 	size_t l1, l2;
2734 
2735 	name1[0] = CTL_SYSCTL;
2736 	name1[1] = CTL_SYSCTL_NEXT;
2737 	l1 = 2;
2738 	if (len) {
2739 		memcpy(name1 + 2, oid, len * sizeof(int));
2740 		l1 += len;
2741 	} else {
2742 		name1[2] = CTL_KERN;
2743 		l1++;
2744 	}
2745 	for (;;) {
2746 		int i, error;
2747 
2748 		l2 = sizeof(name2);
2749 		error = kernel_sysctl(kdb_thread, name1, l1,
2750 		    name2, &l2, NULL, 0, &l2, 0);
2751 		if (error != 0) {
2752 			if (error == ENOENT)
2753 				return (0);
2754 			else
2755 				db_error("sysctl(next)");
2756 		}
2757 
2758 		l2 /= sizeof(int);
2759 
2760 		if (l2 < (unsigned int)len)
2761 			return (0);
2762 
2763 		for (i = 0; i < len; i++)
2764 			if (name2[i] != oid[i])
2765 				return (0);
2766 
2767 		/* Find the OID in question */
2768 		error = sysctl_find_oid(name2, l2, &oidp, NULL, NULL);
2769 		if (error)
2770 			return (error);
2771 
2772 		i = db_show_oid(oidp, name2, l2, flags | DB_SYSCTL_SAFE_ONLY);
2773 
2774 		if (db_pager_quit)
2775 			return (0);
2776 
2777 		memcpy(name1+2, name2, l2 * sizeof(int));
2778 		l1 = 2 + l2;
2779 	}
2780 }
2781 
2782 /*
2783  * Show a sysctl by its user facing string
2784  */
2785 static int
2786 db_sysctlbyname(char *name, int flags)
2787 {
2788 	struct sysctl_oid *oidp;
2789 	int oid[CTL_MAXNAME];
2790 	int error, nlen;
2791 
2792 	error = name2oid(name, oid, &nlen, &oidp);
2793 	if (error) {
2794 		return (error);
2795 	}
2796 
2797 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
2798 		db_show_sysctl_all(oid, nlen, flags);
2799 	} else {
2800 		error = db_show_oid(oidp, oid, nlen, flags);
2801 	}
2802 
2803 	return (error);
2804 }
2805 
2806 static void
2807 db_sysctl_cmd_usage(void)
2808 {
2809 	db_printf(
2810 	    " sysctl [/Nnox] <sysctl>					    \n"
2811 	    "								    \n"
2812 	    " <sysctl> The name of the sysctl to show.			    \n"
2813 	    "								    \n"
2814 	    " Show a sysctl by hooking into SYSCTL_IN and SYSCTL_OUT.	    \n"
2815 	    " This will work for most sysctls, but should not be used	    \n"
2816 	    " with sysctls that are known to malloc.			    \n"
2817 	    "								    \n"
2818 	    " While recursing any \"unsafe\" sysctls will be skipped.	    \n"
2819 	    " Call sysctl directly on the sysctl to try printing the	    \n"
2820 	    " skipped sysctl. This is unsafe and may make the ddb	    \n"
2821 	    " session unusable.						    \n"
2822 	    "								    \n"
2823 	    " Arguments:						    \n"
2824 	    "	/N	Display only the name of the sysctl.		    \n"
2825 	    "	/n	Display only the value of the sysctl.		    \n"
2826 	    "	/o	Display opaque values.				    \n"
2827 	    "	/x	Display the sysctl in hex.			    \n"
2828 	    "								    \n"
2829 	    "For example:						    \n"
2830 	    "sysctl vm.v_free_min					    \n"
2831 	    "vn.v_free_min: 12669					    \n"
2832 	    );
2833 }
2834 
2835 /*
2836  * Show a specific sysctl similar to sysctl (8).
2837  */
2838 DB_FUNC(sysctl, db_sysctl_cmd, db_cmd_table, CS_OWN, NULL)
2839 {
2840 	char name[TOK_STRING_SIZE];
2841 	int error, i, t, flags;
2842 
2843 	/* Parse the modifiers */
2844 	t = db_read_token();
2845 	if (t == tSLASH || t == tMINUS) {
2846 		t = db_read_token();
2847 		if (t != tIDENT) {
2848 			db_printf("Bad modifier\n");
2849 			error = EINVAL;
2850 			goto out;
2851 		}
2852 		db_strcpy(modif, db_tok_string);
2853 	}
2854 	else {
2855 		db_unread_token(t);
2856 		modif[0] = '\0';
2857 	}
2858 
2859 	flags = 0;
2860 	for (i = 0; i < nitems(db_sysctl_modifs); i++) {
2861 		if (strchr(modif, db_sysctl_modifs[i])) {
2862 			flags |= db_sysctl_modif_values[i];
2863 		}
2864 	}
2865 
2866 	/* Parse the sysctl names */
2867 	t = db_read_token();
2868 	if (t != tIDENT) {
2869 		db_printf("Need sysctl name\n");
2870 		error = EINVAL;
2871 		goto out;
2872 	}
2873 
2874 	/* Copy the name into a temporary buffer */
2875 	db_strcpy(name, db_tok_string);
2876 
2877 	/* Ensure there is no trailing cruft */
2878 	t = db_read_token();
2879 	if (t != tEOL) {
2880 		db_printf("Unexpected sysctl argument\n");
2881 		error = EINVAL;
2882 		goto out;
2883 	}
2884 
2885 	error = db_sysctlbyname(name, flags);
2886 	if (error == ENOENT) {
2887 		db_printf("unknown oid: '%s'\n", db_tok_string);
2888 		goto out;
2889 	} else if (error) {
2890 		db_printf("%s: error: %d\n", db_tok_string, error);
2891 		goto out;
2892 	}
2893 
2894 out:
2895 	/* Ensure we eat all of our text */
2896 	db_flush_lex();
2897 
2898 	if (error == EINVAL) {
2899 		db_sysctl_cmd_usage();
2900 	}
2901 }
2902 
2903 #endif /* DDB */
2904