xref: /freebsd/sys/kern/kern_sysctl.c (revision 7431dfd4580e850375fe5478d92ec770344db098)
1 /*-
2  * Copyright (c) 1982, 1986, 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Mike Karels at Berkeley Software Design, Inc.
7  *
8  * Quite extensively rewritten by Poul-Henning Kamp of the FreeBSD
9  * project, to make these variables more userfriendly.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 4. Neither the name of the University nor the names of its contributors
20  *    may be used to endorse or promote products derived from this software
21  *    without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  *
35  *	@(#)kern_sysctl.c	8.4 (Berkeley) 4/14/94
36  */
37 
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
40 
41 #include "opt_capsicum.h"
42 #include "opt_compat.h"
43 #include "opt_ktrace.h"
44 
45 #include <sys/param.h>
46 #include <sys/fail.h>
47 #include <sys/systm.h>
48 #include <sys/capsicum.h>
49 #include <sys/kernel.h>
50 #include <sys/sysctl.h>
51 #include <sys/malloc.h>
52 #include <sys/priv.h>
53 #include <sys/proc.h>
54 #include <sys/jail.h>
55 #include <sys/lock.h>
56 #include <sys/mutex.h>
57 #include <sys/sbuf.h>
58 #include <sys/sx.h>
59 #include <sys/sysproto.h>
60 #include <sys/uio.h>
61 #ifdef KTRACE
62 #include <sys/ktrace.h>
63 #endif
64 
65 #include <net/vnet.h>
66 
67 #include <security/mac/mac_framework.h>
68 
69 #include <vm/vm.h>
70 #include <vm/vm_extern.h>
71 
72 static MALLOC_DEFINE(M_SYSCTL, "sysctl", "sysctl internal magic");
73 static MALLOC_DEFINE(M_SYSCTLOID, "sysctloid", "sysctl dynamic oids");
74 static MALLOC_DEFINE(M_SYSCTLTMP, "sysctltmp", "sysctl temp output buffer");
75 
76 /*
77  * The sysctllock protects the MIB tree.  It also protects sysctl
78  * contexts used with dynamic sysctls.  The sysctl_register_oid() and
79  * sysctl_unregister_oid() routines require the sysctllock to already
80  * be held, so the sysctl_xlock() and sysctl_xunlock() routines are
81  * provided for the few places in the kernel which need to use that
82  * API rather than using the dynamic API.  Use of the dynamic API is
83  * strongly encouraged for most code.
84  *
85  * The sysctlmemlock is used to limit the amount of user memory wired for
86  * sysctl requests.  This is implemented by serializing any userland
87  * sysctl requests larger than a single page via an exclusive lock.
88  */
89 static struct sx sysctllock;
90 static struct sx sysctlmemlock;
91 
92 #define	SYSCTL_XLOCK()		sx_xlock(&sysctllock)
93 #define	SYSCTL_XUNLOCK()	sx_xunlock(&sysctllock)
94 #define	SYSCTL_SLOCK()		sx_slock(&sysctllock)
95 #define	SYSCTL_SUNLOCK()	sx_sunlock(&sysctllock)
96 #define	SYSCTL_XLOCKED()	sx_xlocked(&sysctllock)
97 #define	SYSCTL_ASSERT_LOCKED()	sx_assert(&sysctllock, SA_LOCKED)
98 #define	SYSCTL_ASSERT_XLOCKED()	sx_assert(&sysctllock, SA_XLOCKED)
99 #define	SYSCTL_ASSERT_SLOCKED()	sx_assert(&sysctllock, SA_SLOCKED)
100 #define	SYSCTL_INIT()		sx_init(&sysctllock, "sysctl lock")
101 #define	SYSCTL_SLEEP(ch, wmesg, timo)					\
102 				sx_sleep(ch, &sysctllock, 0, wmesg, timo)
103 
104 static int sysctl_root(SYSCTL_HANDLER_ARGS);
105 
106 /* Root list */
107 struct sysctl_oid_list sysctl__children = SLIST_HEAD_INITIALIZER(&sysctl__children);
108 
109 static int	sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del,
110 		    int recurse);
111 static int	sysctl_old_kernel(struct sysctl_req *, const void *, size_t);
112 static int	sysctl_new_kernel(struct sysctl_req *, void *, size_t);
113 
114 static void
115 sysctl_lock(bool xlock)
116 {
117 
118 	if (xlock)
119 		SYSCTL_XLOCK();
120 	else
121 		SYSCTL_SLOCK();
122 }
123 
124 static bool
125 sysctl_unlock(void)
126 {
127 	bool xlocked;
128 
129 	xlocked = SYSCTL_XLOCKED();
130 	if (xlocked)
131 		SYSCTL_XUNLOCK();
132 	else
133 		SYSCTL_SUNLOCK();
134 	return (xlocked);
135 }
136 
137 static struct sysctl_oid *
138 sysctl_find_oidname(const char *name, struct sysctl_oid_list *list)
139 {
140 	struct sysctl_oid *oidp;
141 
142 	SYSCTL_ASSERT_LOCKED();
143 	SLIST_FOREACH(oidp, list, oid_link) {
144 		if (strcmp(oidp->oid_name, name) == 0) {
145 			return (oidp);
146 		}
147 	}
148 	return (NULL);
149 }
150 
151 /*
152  * Initialization of the MIB tree.
153  *
154  * Order by number in each list.
155  */
156 void
157 sysctl_xlock(void)
158 {
159 
160 	SYSCTL_XLOCK();
161 }
162 
163 void
164 sysctl_xunlock(void)
165 {
166 
167 	SYSCTL_XUNLOCK();
168 }
169 
170 static int
171 sysctl_root_handler_locked(struct sysctl_oid *oid, void *arg1, intptr_t arg2,
172     struct sysctl_req *req)
173 {
174 	int error;
175 	bool xlocked;
176 
177 	atomic_add_int(&oid->oid_running, 1);
178 	xlocked = sysctl_unlock();
179 
180 	if (!(oid->oid_kind & CTLFLAG_MPSAFE))
181 		mtx_lock(&Giant);
182 	error = oid->oid_handler(oid, arg1, arg2, req);
183 	if (!(oid->oid_kind & CTLFLAG_MPSAFE))
184 		mtx_unlock(&Giant);
185 
186 	sysctl_lock(xlocked);
187 	if (atomic_fetchadd_int(&oid->oid_running, -1) == 1 &&
188 	    (oid->oid_kind & CTLFLAG_DYING) != 0)
189 		wakeup(&oid->oid_running);
190 
191 	return (error);
192 }
193 
194 static void
195 sysctl_load_tunable_by_oid_locked(struct sysctl_oid *oidp)
196 {
197 	struct sysctl_req req;
198 	struct sysctl_oid *curr;
199 	char *penv = NULL;
200 	char path[64];
201 	ssize_t rem = sizeof(path);
202 	ssize_t len;
203 	int val_int;
204 	long val_long;
205 	int64_t val_64;
206 	quad_t val_quad;
207 	int error;
208 
209 	path[--rem] = 0;
210 
211 	for (curr = oidp; curr != NULL; curr = SYSCTL_PARENT(curr)) {
212 		len = strlen(curr->oid_name);
213 		rem -= len;
214 		if (curr != oidp)
215 			rem -= 1;
216 		if (rem < 0) {
217 			printf("OID path exceeds %d bytes\n", (int)sizeof(path));
218 			return;
219 		}
220 		memcpy(path + rem, curr->oid_name, len);
221 		if (curr != oidp)
222 			path[rem + len] = '.';
223 	}
224 
225 	memset(&req, 0, sizeof(req));
226 
227 	req.td = curthread;
228 	req.oldfunc = sysctl_old_kernel;
229 	req.newfunc = sysctl_new_kernel;
230 	req.lock = REQ_UNWIRED;
231 
232 	switch (oidp->oid_kind & CTLTYPE) {
233 	case CTLTYPE_INT:
234 		if (getenv_int(path + rem, &val_int) == 0)
235 			return;
236 		req.newlen = sizeof(val_int);
237 		req.newptr = &val_int;
238 		break;
239 	case CTLTYPE_UINT:
240 		if (getenv_uint(path + rem, (unsigned int *)&val_int) == 0)
241 			return;
242 		req.newlen = sizeof(val_int);
243 		req.newptr = &val_int;
244 		break;
245 	case CTLTYPE_LONG:
246 		if (getenv_long(path + rem, &val_long) == 0)
247 			return;
248 		req.newlen = sizeof(val_long);
249 		req.newptr = &val_long;
250 		break;
251 	case CTLTYPE_ULONG:
252 		if (getenv_ulong(path + rem, (unsigned long *)&val_long) == 0)
253 			return;
254 		req.newlen = sizeof(val_long);
255 		req.newptr = &val_long;
256 		break;
257 	case CTLTYPE_S64:
258 		if (getenv_quad(path + rem, &val_quad) == 0)
259 			return;
260 		val_64 = val_quad;
261 		req.newlen = sizeof(val_64);
262 		req.newptr = &val_64;
263 		break;
264 	case CTLTYPE_U64:
265 		/* XXX there is no getenv_uquad() */
266 		if (getenv_quad(path + rem, &val_quad) == 0)
267 			return;
268 		val_64 = val_quad;
269 		req.newlen = sizeof(val_64);
270 		req.newptr = &val_64;
271 		break;
272 	case CTLTYPE_STRING:
273 		penv = kern_getenv(path + rem);
274 		if (penv == NULL)
275 			return;
276 		req.newlen = strlen(penv);
277 		req.newptr = penv;
278 		break;
279 	default:
280 		return;
281 	}
282 	error = sysctl_root_handler_locked(oidp, oidp->oid_arg1,
283 	    oidp->oid_arg2, &req);
284 	if (error != 0)
285 		printf("Setting sysctl %s failed: %d\n", path, error);
286 	if (penv != NULL)
287 		freeenv(penv);
288 }
289 
290 void
291 sysctl_register_oid(struct sysctl_oid *oidp)
292 {
293 	struct sysctl_oid_list *parent = oidp->oid_parent;
294 	struct sysctl_oid *p;
295 	struct sysctl_oid *q;
296 
297 	/*
298 	 * First check if another oid with the same name already
299 	 * exists in the parent's list.
300 	 */
301 	SYSCTL_ASSERT_XLOCKED();
302 	p = sysctl_find_oidname(oidp->oid_name, parent);
303 	if (p != NULL) {
304 		if ((p->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
305 			p->oid_refcnt++;
306 			return;
307 		} else {
308 			printf("can't re-use a leaf (%s)!\n", p->oid_name);
309 			return;
310 		}
311 	}
312 	/*
313 	 * If this oid has a number OID_AUTO, give it a number which
314 	 * is greater than any current oid.
315 	 * NOTE: DO NOT change the starting value here, change it in
316 	 * <sys/sysctl.h>, and make sure it is at least 256 to
317 	 * accomodate e.g. net.inet.raw as a static sysctl node.
318 	 */
319 	if (oidp->oid_number == OID_AUTO) {
320 		static int newoid = CTL_AUTO_START;
321 
322 		oidp->oid_number = newoid++;
323 		if (newoid == 0x7fffffff)
324 			panic("out of oids");
325 	}
326 #if 0
327 	else if (oidp->oid_number >= CTL_AUTO_START) {
328 		/* do not panic; this happens when unregistering sysctl sets */
329 		printf("static sysctl oid too high: %d", oidp->oid_number);
330 	}
331 #endif
332 
333 	/*
334 	 * Insert the oid into the parent's list in order.
335 	 */
336 	q = NULL;
337 	SLIST_FOREACH(p, parent, oid_link) {
338 		if (oidp->oid_number < p->oid_number)
339 			break;
340 		q = p;
341 	}
342 	if (q)
343 		SLIST_INSERT_AFTER(q, oidp, oid_link);
344 	else
345 		SLIST_INSERT_HEAD(parent, oidp, oid_link);
346 
347 	if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE &&
348 #ifdef VIMAGE
349 	    (oidp->oid_kind & CTLFLAG_VNET) == 0 &&
350 #endif
351 	    (oidp->oid_kind & CTLFLAG_TUN) != 0 &&
352 	    (oidp->oid_kind & CTLFLAG_NOFETCH) == 0) {
353 		sysctl_load_tunable_by_oid_locked(oidp);
354 	}
355 }
356 
357 void
358 sysctl_unregister_oid(struct sysctl_oid *oidp)
359 {
360 	struct sysctl_oid *p;
361 	int error;
362 
363 	SYSCTL_ASSERT_XLOCKED();
364 	error = ENOENT;
365 	if (oidp->oid_number == OID_AUTO) {
366 		error = EINVAL;
367 	} else {
368 		SLIST_FOREACH(p, oidp->oid_parent, oid_link) {
369 			if (p == oidp) {
370 				SLIST_REMOVE(oidp->oid_parent, oidp,
371 				    sysctl_oid, oid_link);
372 				error = 0;
373 				break;
374 			}
375 		}
376 	}
377 
378 	/*
379 	 * This can happen when a module fails to register and is
380 	 * being unloaded afterwards.  It should not be a panic()
381 	 * for normal use.
382 	 */
383 	if (error)
384 		printf("%s: failed to unregister sysctl\n", __func__);
385 }
386 
387 /* Initialize a new context to keep track of dynamically added sysctls. */
388 int
389 sysctl_ctx_init(struct sysctl_ctx_list *c)
390 {
391 
392 	if (c == NULL) {
393 		return (EINVAL);
394 	}
395 
396 	/*
397 	 * No locking here, the caller is responsible for not adding
398 	 * new nodes to a context until after this function has
399 	 * returned.
400 	 */
401 	TAILQ_INIT(c);
402 	return (0);
403 }
404 
405 /* Free the context, and destroy all dynamic oids registered in this context */
406 int
407 sysctl_ctx_free(struct sysctl_ctx_list *clist)
408 {
409 	struct sysctl_ctx_entry *e, *e1;
410 	int error;
411 
412 	error = 0;
413 	/*
414 	 * First perform a "dry run" to check if it's ok to remove oids.
415 	 * XXX FIXME
416 	 * XXX This algorithm is a hack. But I don't know any
417 	 * XXX better solution for now...
418 	 */
419 	SYSCTL_XLOCK();
420 	TAILQ_FOREACH(e, clist, link) {
421 		error = sysctl_remove_oid_locked(e->entry, 0, 0);
422 		if (error)
423 			break;
424 	}
425 	/*
426 	 * Restore deregistered entries, either from the end,
427 	 * or from the place where error occured.
428 	 * e contains the entry that was not unregistered
429 	 */
430 	if (error)
431 		e1 = TAILQ_PREV(e, sysctl_ctx_list, link);
432 	else
433 		e1 = TAILQ_LAST(clist, sysctl_ctx_list);
434 	while (e1 != NULL) {
435 		sysctl_register_oid(e1->entry);
436 		e1 = TAILQ_PREV(e1, sysctl_ctx_list, link);
437 	}
438 	if (error) {
439 		SYSCTL_XUNLOCK();
440 		return(EBUSY);
441 	}
442 	/* Now really delete the entries */
443 	e = TAILQ_FIRST(clist);
444 	while (e != NULL) {
445 		e1 = TAILQ_NEXT(e, link);
446 		error = sysctl_remove_oid_locked(e->entry, 1, 0);
447 		if (error)
448 			panic("sysctl_remove_oid: corrupt tree, entry: %s",
449 			    e->entry->oid_name);
450 		free(e, M_SYSCTLOID);
451 		e = e1;
452 	}
453 	SYSCTL_XUNLOCK();
454 	return (error);
455 }
456 
457 /* Add an entry to the context */
458 struct sysctl_ctx_entry *
459 sysctl_ctx_entry_add(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
460 {
461 	struct sysctl_ctx_entry *e;
462 
463 	SYSCTL_ASSERT_XLOCKED();
464 	if (clist == NULL || oidp == NULL)
465 		return(NULL);
466 	e = malloc(sizeof(struct sysctl_ctx_entry), M_SYSCTLOID, M_WAITOK);
467 	e->entry = oidp;
468 	TAILQ_INSERT_HEAD(clist, e, link);
469 	return (e);
470 }
471 
472 /* Find an entry in the context */
473 struct sysctl_ctx_entry *
474 sysctl_ctx_entry_find(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
475 {
476 	struct sysctl_ctx_entry *e;
477 
478 	SYSCTL_ASSERT_XLOCKED();
479 	if (clist == NULL || oidp == NULL)
480 		return(NULL);
481 	TAILQ_FOREACH(e, clist, link) {
482 		if(e->entry == oidp)
483 			return(e);
484 	}
485 	return (e);
486 }
487 
488 /*
489  * Delete an entry from the context.
490  * NOTE: this function doesn't free oidp! You have to remove it
491  * with sysctl_remove_oid().
492  */
493 int
494 sysctl_ctx_entry_del(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
495 {
496 	struct sysctl_ctx_entry *e;
497 
498 	if (clist == NULL || oidp == NULL)
499 		return (EINVAL);
500 	SYSCTL_XLOCK();
501 	e = sysctl_ctx_entry_find(clist, oidp);
502 	if (e != NULL) {
503 		TAILQ_REMOVE(clist, e, link);
504 		SYSCTL_XUNLOCK();
505 		free(e, M_SYSCTLOID);
506 		return (0);
507 	} else {
508 		SYSCTL_XUNLOCK();
509 		return (ENOENT);
510 	}
511 }
512 
513 /*
514  * Remove dynamically created sysctl trees.
515  * oidp - top of the tree to be removed
516  * del - if 0 - just deregister, otherwise free up entries as well
517  * recurse - if != 0 traverse the subtree to be deleted
518  */
519 int
520 sysctl_remove_oid(struct sysctl_oid *oidp, int del, int recurse)
521 {
522 	int error;
523 
524 	SYSCTL_XLOCK();
525 	error = sysctl_remove_oid_locked(oidp, del, recurse);
526 	SYSCTL_XUNLOCK();
527 	return (error);
528 }
529 
530 int
531 sysctl_remove_name(struct sysctl_oid *parent, const char *name,
532     int del, int recurse)
533 {
534 	struct sysctl_oid *p, *tmp;
535 	int error;
536 
537 	error = ENOENT;
538 	SYSCTL_XLOCK();
539 	SLIST_FOREACH_SAFE(p, SYSCTL_CHILDREN(parent), oid_link, tmp) {
540 		if (strcmp(p->oid_name, name) == 0) {
541 			error = sysctl_remove_oid_locked(p, del, recurse);
542 			break;
543 		}
544 	}
545 	SYSCTL_XUNLOCK();
546 
547 	return (error);
548 }
549 
550 
551 static int
552 sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del, int recurse)
553 {
554 	struct sysctl_oid *p, *tmp;
555 	int error;
556 
557 	SYSCTL_ASSERT_XLOCKED();
558 	if (oidp == NULL)
559 		return(EINVAL);
560 	if ((oidp->oid_kind & CTLFLAG_DYN) == 0) {
561 		printf("can't remove non-dynamic nodes!\n");
562 		return (EINVAL);
563 	}
564 	/*
565 	 * WARNING: normal method to do this should be through
566 	 * sysctl_ctx_free(). Use recursing as the last resort
567 	 * method to purge your sysctl tree of leftovers...
568 	 * However, if some other code still references these nodes,
569 	 * it will panic.
570 	 */
571 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
572 		if (oidp->oid_refcnt == 1) {
573 			SLIST_FOREACH_SAFE(p,
574 			    SYSCTL_CHILDREN(oidp), oid_link, tmp) {
575 				if (!recurse) {
576 					printf("Warning: failed attempt to "
577 					    "remove oid %s with child %s\n",
578 					    oidp->oid_name, p->oid_name);
579 					return (ENOTEMPTY);
580 				}
581 				error = sysctl_remove_oid_locked(p, del,
582 				    recurse);
583 				if (error)
584 					return (error);
585 			}
586 		}
587 	}
588 	if (oidp->oid_refcnt > 1 ) {
589 		oidp->oid_refcnt--;
590 	} else {
591 		if (oidp->oid_refcnt == 0) {
592 			printf("Warning: bad oid_refcnt=%u (%s)!\n",
593 				oidp->oid_refcnt, oidp->oid_name);
594 			return (EINVAL);
595 		}
596 		sysctl_unregister_oid(oidp);
597 		if (del) {
598 			/*
599 			 * Wait for all threads running the handler to drain.
600 			 * This preserves the previous behavior when the
601 			 * sysctl lock was held across a handler invocation,
602 			 * and is necessary for module unload correctness.
603 			 */
604 			while (oidp->oid_running > 0) {
605 				oidp->oid_kind |= CTLFLAG_DYING;
606 				SYSCTL_SLEEP(&oidp->oid_running, "oidrm", 0);
607 			}
608 			if (oidp->oid_descr)
609 				free(__DECONST(char *, oidp->oid_descr),
610 				    M_SYSCTLOID);
611 			free(__DECONST(char *, oidp->oid_name), M_SYSCTLOID);
612 			free(oidp, M_SYSCTLOID);
613 		}
614 	}
615 	return (0);
616 }
617 /*
618  * Create new sysctls at run time.
619  * clist may point to a valid context initialized with sysctl_ctx_init().
620  */
621 struct sysctl_oid *
622 sysctl_add_oid(struct sysctl_ctx_list *clist, struct sysctl_oid_list *parent,
623 	int number, const char *name, int kind, void *arg1, intptr_t arg2,
624 	int (*handler)(SYSCTL_HANDLER_ARGS), const char *fmt, const char *descr)
625 {
626 	struct sysctl_oid *oidp;
627 
628 	/* You have to hook up somewhere.. */
629 	if (parent == NULL)
630 		return(NULL);
631 	/* Check if the node already exists, otherwise create it */
632 	SYSCTL_XLOCK();
633 	oidp = sysctl_find_oidname(name, parent);
634 	if (oidp != NULL) {
635 		if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
636 			oidp->oid_refcnt++;
637 			/* Update the context */
638 			if (clist != NULL)
639 				sysctl_ctx_entry_add(clist, oidp);
640 			SYSCTL_XUNLOCK();
641 			return (oidp);
642 		} else {
643 			SYSCTL_XUNLOCK();
644 			printf("can't re-use a leaf (%s)!\n", name);
645 			return (NULL);
646 		}
647 	}
648 	oidp = malloc(sizeof(struct sysctl_oid), M_SYSCTLOID, M_WAITOK|M_ZERO);
649 	oidp->oid_parent = parent;
650 	SLIST_INIT(&oidp->oid_children);
651 	oidp->oid_number = number;
652 	oidp->oid_refcnt = 1;
653 	oidp->oid_name = strdup(name, M_SYSCTLOID);
654 	oidp->oid_handler = handler;
655 	oidp->oid_kind = CTLFLAG_DYN | kind;
656 	oidp->oid_arg1 = arg1;
657 	oidp->oid_arg2 = arg2;
658 	oidp->oid_fmt = fmt;
659 	if (descr != NULL)
660 		oidp->oid_descr = strdup(descr, M_SYSCTLOID);
661 	/* Update the context, if used */
662 	if (clist != NULL)
663 		sysctl_ctx_entry_add(clist, oidp);
664 	/* Register this oid */
665 	sysctl_register_oid(oidp);
666 	SYSCTL_XUNLOCK();
667 	return (oidp);
668 }
669 
670 /*
671  * Rename an existing oid.
672  */
673 void
674 sysctl_rename_oid(struct sysctl_oid *oidp, const char *name)
675 {
676 	char *newname;
677 	char *oldname;
678 
679 	newname = strdup(name, M_SYSCTLOID);
680 	SYSCTL_XLOCK();
681 	oldname = __DECONST(char *, oidp->oid_name);
682 	oidp->oid_name = newname;
683 	SYSCTL_XUNLOCK();
684 	free(oldname, M_SYSCTLOID);
685 }
686 
687 /*
688  * Reparent an existing oid.
689  */
690 int
691 sysctl_move_oid(struct sysctl_oid *oid, struct sysctl_oid_list *parent)
692 {
693 	struct sysctl_oid *oidp;
694 
695 	SYSCTL_XLOCK();
696 	if (oid->oid_parent == parent) {
697 		SYSCTL_XUNLOCK();
698 		return (0);
699 	}
700 	oidp = sysctl_find_oidname(oid->oid_name, parent);
701 	if (oidp != NULL) {
702 		SYSCTL_XUNLOCK();
703 		return (EEXIST);
704 	}
705 	sysctl_unregister_oid(oid);
706 	oid->oid_parent = parent;
707 	oid->oid_number = OID_AUTO;
708 	sysctl_register_oid(oid);
709 	SYSCTL_XUNLOCK();
710 	return (0);
711 }
712 
713 /*
714  * Register the kernel's oids on startup.
715  */
716 SET_DECLARE(sysctl_set, struct sysctl_oid);
717 
718 static void
719 sysctl_register_all(void *arg)
720 {
721 	struct sysctl_oid **oidp;
722 
723 	sx_init(&sysctlmemlock, "sysctl mem");
724 	SYSCTL_INIT();
725 	SYSCTL_XLOCK();
726 	SET_FOREACH(oidp, sysctl_set)
727 		sysctl_register_oid(*oidp);
728 	SYSCTL_XUNLOCK();
729 }
730 SYSINIT(sysctl, SI_SUB_KMEM, SI_ORDER_FIRST, sysctl_register_all, 0);
731 
732 /*
733  * "Staff-functions"
734  *
735  * These functions implement a presently undocumented interface
736  * used by the sysctl program to walk the tree, and get the type
737  * so it can print the value.
738  * This interface is under work and consideration, and should probably
739  * be killed with a big axe by the first person who can find the time.
740  * (be aware though, that the proper interface isn't as obvious as it
741  * may seem, there are various conflicting requirements.
742  *
743  * {0,0}	printf the entire MIB-tree.
744  * {0,1,...}	return the name of the "..." OID.
745  * {0,2,...}	return the next OID.
746  * {0,3}	return the OID of the name in "new"
747  * {0,4,...}	return the kind & format info for the "..." OID.
748  * {0,5,...}	return the description the "..." OID.
749  */
750 
751 #ifdef SYSCTL_DEBUG
752 static void
753 sysctl_sysctl_debug_dump_node(struct sysctl_oid_list *l, int i)
754 {
755 	int k;
756 	struct sysctl_oid *oidp;
757 
758 	SYSCTL_ASSERT_LOCKED();
759 	SLIST_FOREACH(oidp, l, oid_link) {
760 
761 		for (k=0; k<i; k++)
762 			printf(" ");
763 
764 		printf("%d %s ", oidp->oid_number, oidp->oid_name);
765 
766 		printf("%c%c",
767 			oidp->oid_kind & CTLFLAG_RD ? 'R':' ',
768 			oidp->oid_kind & CTLFLAG_WR ? 'W':' ');
769 
770 		if (oidp->oid_handler)
771 			printf(" *Handler");
772 
773 		switch (oidp->oid_kind & CTLTYPE) {
774 			case CTLTYPE_NODE:
775 				printf(" Node\n");
776 				if (!oidp->oid_handler) {
777 					sysctl_sysctl_debug_dump_node(
778 					    SYSCTL_CHILDREN(oidp), i + 2);
779 				}
780 				break;
781 			case CTLTYPE_INT:    printf(" Int\n"); break;
782 			case CTLTYPE_UINT:   printf(" u_int\n"); break;
783 			case CTLTYPE_LONG:   printf(" Long\n"); break;
784 			case CTLTYPE_ULONG:  printf(" u_long\n"); break;
785 			case CTLTYPE_STRING: printf(" String\n"); break;
786 			case CTLTYPE_U64:    printf(" uint64_t\n"); break;
787 			case CTLTYPE_S64:    printf(" int64_t\n"); break;
788 			case CTLTYPE_OPAQUE: printf(" Opaque/struct\n"); break;
789 			default:	     printf("\n");
790 		}
791 
792 	}
793 }
794 
795 static int
796 sysctl_sysctl_debug(SYSCTL_HANDLER_ARGS)
797 {
798 	int error;
799 
800 	error = priv_check(req->td, PRIV_SYSCTL_DEBUG);
801 	if (error)
802 		return (error);
803 	SYSCTL_SLOCK();
804 	sysctl_sysctl_debug_dump_node(&sysctl__children, 0);
805 	SYSCTL_SUNLOCK();
806 	return (ENOENT);
807 }
808 
809 SYSCTL_PROC(_sysctl, 0, debug, CTLTYPE_STRING|CTLFLAG_RD|CTLFLAG_MPSAFE,
810 	0, 0, sysctl_sysctl_debug, "-", "");
811 #endif
812 
813 static int
814 sysctl_sysctl_name(SYSCTL_HANDLER_ARGS)
815 {
816 	int *name = (int *) arg1;
817 	u_int namelen = arg2;
818 	int error = 0;
819 	struct sysctl_oid *oid;
820 	struct sysctl_oid_list *lsp = &sysctl__children, *lsp2;
821 	char buf[10];
822 
823 	SYSCTL_SLOCK();
824 	while (namelen) {
825 		if (!lsp) {
826 			snprintf(buf,sizeof(buf),"%d",*name);
827 			if (req->oldidx)
828 				error = SYSCTL_OUT(req, ".", 1);
829 			if (!error)
830 				error = SYSCTL_OUT(req, buf, strlen(buf));
831 			if (error)
832 				goto out;
833 			namelen--;
834 			name++;
835 			continue;
836 		}
837 		lsp2 = 0;
838 		SLIST_FOREACH(oid, lsp, oid_link) {
839 			if (oid->oid_number != *name)
840 				continue;
841 
842 			if (req->oldidx)
843 				error = SYSCTL_OUT(req, ".", 1);
844 			if (!error)
845 				error = SYSCTL_OUT(req, oid->oid_name,
846 					strlen(oid->oid_name));
847 			if (error)
848 				goto out;
849 
850 			namelen--;
851 			name++;
852 
853 			if ((oid->oid_kind & CTLTYPE) != CTLTYPE_NODE)
854 				break;
855 
856 			if (oid->oid_handler)
857 				break;
858 
859 			lsp2 = SYSCTL_CHILDREN(oid);
860 			break;
861 		}
862 		lsp = lsp2;
863 	}
864 	error = SYSCTL_OUT(req, "", 1);
865  out:
866 	SYSCTL_SUNLOCK();
867 	return (error);
868 }
869 
870 /*
871  * XXXRW/JA: Shouldn't return name data for nodes that we don't permit in
872  * capability mode.
873  */
874 static SYSCTL_NODE(_sysctl, 1, name, CTLFLAG_RD | CTLFLAG_MPSAFE | CTLFLAG_CAPRD,
875     sysctl_sysctl_name, "");
876 
877 static int
878 sysctl_sysctl_next_ls(struct sysctl_oid_list *lsp, int *name, u_int namelen,
879 	int *next, int *len, int level, struct sysctl_oid **oidpp)
880 {
881 	struct sysctl_oid *oidp;
882 
883 	SYSCTL_ASSERT_LOCKED();
884 	*len = level;
885 	SLIST_FOREACH(oidp, lsp, oid_link) {
886 		*next = oidp->oid_number;
887 		*oidpp = oidp;
888 
889 		if (oidp->oid_kind & CTLFLAG_SKIP)
890 			continue;
891 
892 		if (!namelen) {
893 			if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
894 				return (0);
895 			if (oidp->oid_handler)
896 				/* We really should call the handler here...*/
897 				return (0);
898 			lsp = SYSCTL_CHILDREN(oidp);
899 			if (!sysctl_sysctl_next_ls(lsp, 0, 0, next+1,
900 				len, level+1, oidpp))
901 				return (0);
902 			goto emptynode;
903 		}
904 
905 		if (oidp->oid_number < *name)
906 			continue;
907 
908 		if (oidp->oid_number > *name) {
909 			if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
910 				return (0);
911 			if (oidp->oid_handler)
912 				return (0);
913 			lsp = SYSCTL_CHILDREN(oidp);
914 			if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1,
915 				next+1, len, level+1, oidpp))
916 				return (0);
917 			goto next;
918 		}
919 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
920 			continue;
921 
922 		if (oidp->oid_handler)
923 			continue;
924 
925 		lsp = SYSCTL_CHILDREN(oidp);
926 		if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1, next+1,
927 			len, level+1, oidpp))
928 			return (0);
929 	next:
930 		namelen = 1;
931 	emptynode:
932 		*len = level;
933 	}
934 	return (1);
935 }
936 
937 static int
938 sysctl_sysctl_next(SYSCTL_HANDLER_ARGS)
939 {
940 	int *name = (int *) arg1;
941 	u_int namelen = arg2;
942 	int i, j, error;
943 	struct sysctl_oid *oid;
944 	struct sysctl_oid_list *lsp = &sysctl__children;
945 	int newoid[CTL_MAXNAME];
946 
947 	SYSCTL_SLOCK();
948 	i = sysctl_sysctl_next_ls(lsp, name, namelen, newoid, &j, 1, &oid);
949 	SYSCTL_SUNLOCK();
950 	if (i)
951 		return (ENOENT);
952 	error = SYSCTL_OUT(req, newoid, j * sizeof (int));
953 	return (error);
954 }
955 
956 /*
957  * XXXRW/JA: Shouldn't return next data for nodes that we don't permit in
958  * capability mode.
959  */
960 static SYSCTL_NODE(_sysctl, 2, next, CTLFLAG_RD | CTLFLAG_MPSAFE | CTLFLAG_CAPRD,
961     sysctl_sysctl_next, "");
962 
963 static int
964 name2oid(char *name, int *oid, int *len, struct sysctl_oid **oidpp)
965 {
966 	struct sysctl_oid *oidp;
967 	struct sysctl_oid_list *lsp = &sysctl__children;
968 	char *p;
969 
970 	SYSCTL_ASSERT_LOCKED();
971 
972 	for (*len = 0; *len < CTL_MAXNAME;) {
973 		p = strsep(&name, ".");
974 
975 		oidp = SLIST_FIRST(lsp);
976 		for (;; oidp = SLIST_NEXT(oidp, oid_link)) {
977 			if (oidp == NULL)
978 				return (ENOENT);
979 			if (strcmp(p, oidp->oid_name) == 0)
980 				break;
981 		}
982 		*oid++ = oidp->oid_number;
983 		(*len)++;
984 
985 		if (name == NULL || *name == '\0') {
986 			if (oidpp)
987 				*oidpp = oidp;
988 			return (0);
989 		}
990 
991 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
992 			break;
993 
994 		if (oidp->oid_handler)
995 			break;
996 
997 		lsp = SYSCTL_CHILDREN(oidp);
998 	}
999 	return (ENOENT);
1000 }
1001 
1002 static int
1003 sysctl_sysctl_name2oid(SYSCTL_HANDLER_ARGS)
1004 {
1005 	char *p;
1006 	int error, oid[CTL_MAXNAME], len = 0;
1007 	struct sysctl_oid *op = 0;
1008 
1009 	if (!req->newlen)
1010 		return (ENOENT);
1011 	if (req->newlen >= MAXPATHLEN)	/* XXX arbitrary, undocumented */
1012 		return (ENAMETOOLONG);
1013 
1014 	p = malloc(req->newlen+1, M_SYSCTL, M_WAITOK);
1015 
1016 	error = SYSCTL_IN(req, p, req->newlen);
1017 	if (error) {
1018 		free(p, M_SYSCTL);
1019 		return (error);
1020 	}
1021 
1022 	p [req->newlen] = '\0';
1023 
1024 	SYSCTL_SLOCK();
1025 	error = name2oid(p, oid, &len, &op);
1026 	SYSCTL_SUNLOCK();
1027 
1028 	free(p, M_SYSCTL);
1029 
1030 	if (error)
1031 		return (error);
1032 
1033 	error = SYSCTL_OUT(req, oid, len * sizeof *oid);
1034 	return (error);
1035 }
1036 
1037 /*
1038  * XXXRW/JA: Shouldn't return name2oid data for nodes that we don't permit in
1039  * capability mode.
1040  */
1041 SYSCTL_PROC(_sysctl, 3, name2oid,
1042     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_MPSAFE
1043     | CTLFLAG_CAPRW, 0, 0, sysctl_sysctl_name2oid, "I", "");
1044 
1045 static int
1046 sysctl_sysctl_oidfmt(SYSCTL_HANDLER_ARGS)
1047 {
1048 	struct sysctl_oid *oid;
1049 	int error;
1050 
1051 	SYSCTL_SLOCK();
1052 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1053 	if (error)
1054 		goto out;
1055 
1056 	if (oid->oid_fmt == NULL) {
1057 		error = ENOENT;
1058 		goto out;
1059 	}
1060 	error = SYSCTL_OUT(req, &oid->oid_kind, sizeof(oid->oid_kind));
1061 	if (error)
1062 		goto out;
1063 	error = SYSCTL_OUT(req, oid->oid_fmt, strlen(oid->oid_fmt) + 1);
1064  out:
1065 	SYSCTL_SUNLOCK();
1066 	return (error);
1067 }
1068 
1069 
1070 static SYSCTL_NODE(_sysctl, 4, oidfmt, CTLFLAG_RD|CTLFLAG_MPSAFE|CTLFLAG_CAPRD,
1071     sysctl_sysctl_oidfmt, "");
1072 
1073 static int
1074 sysctl_sysctl_oiddescr(SYSCTL_HANDLER_ARGS)
1075 {
1076 	struct sysctl_oid *oid;
1077 	int error;
1078 
1079 	SYSCTL_SLOCK();
1080 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1081 	if (error)
1082 		goto out;
1083 
1084 	if (oid->oid_descr == NULL) {
1085 		error = ENOENT;
1086 		goto out;
1087 	}
1088 	error = SYSCTL_OUT(req, oid->oid_descr, strlen(oid->oid_descr) + 1);
1089  out:
1090 	SYSCTL_SUNLOCK();
1091 	return (error);
1092 }
1093 
1094 static SYSCTL_NODE(_sysctl, 5, oiddescr, CTLFLAG_RD|CTLFLAG_MPSAFE|CTLFLAG_CAPRD,
1095     sysctl_sysctl_oiddescr, "");
1096 
1097 /*
1098  * Default "handler" functions.
1099  */
1100 
1101 /*
1102  * Handle an int, signed or unsigned.
1103  * Two cases:
1104  *     a variable:  point arg1 at it.
1105  *     a constant:  pass it in arg2.
1106  */
1107 
1108 int
1109 sysctl_handle_int(SYSCTL_HANDLER_ARGS)
1110 {
1111 	int tmpout, error = 0;
1112 
1113 	/*
1114 	 * Attempt to get a coherent snapshot by making a copy of the data.
1115 	 */
1116 	if (arg1)
1117 		tmpout = *(int *)arg1;
1118 	else
1119 		tmpout = arg2;
1120 	error = SYSCTL_OUT(req, &tmpout, sizeof(int));
1121 
1122 	if (error || !req->newptr)
1123 		return (error);
1124 
1125 	if (!arg1)
1126 		error = EPERM;
1127 	else
1128 		error = SYSCTL_IN(req, arg1, sizeof(int));
1129 	return (error);
1130 }
1131 
1132 /*
1133  * Based on on sysctl_handle_int() convert milliseconds into ticks.
1134  * Note: this is used by TCP.
1135  */
1136 
1137 int
1138 sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
1139 {
1140 	int error, s, tt;
1141 
1142 	tt = *(int *)arg1;
1143 	s = (int)((int64_t)tt * 1000 / hz);
1144 
1145 	error = sysctl_handle_int(oidp, &s, 0, req);
1146 	if (error || !req->newptr)
1147 		return (error);
1148 
1149 	tt = (int)((int64_t)s * hz / 1000);
1150 	if (tt < 1)
1151 		return (EINVAL);
1152 
1153 	*(int *)arg1 = tt;
1154 	return (0);
1155 }
1156 
1157 
1158 /*
1159  * Handle a long, signed or unsigned.
1160  * Two cases:
1161  *     a variable:  point arg1 at it.
1162  *     a constant:  pass it in arg2.
1163  */
1164 
1165 int
1166 sysctl_handle_long(SYSCTL_HANDLER_ARGS)
1167 {
1168 	int error = 0;
1169 	long tmplong;
1170 #ifdef SCTL_MASK32
1171 	int tmpint;
1172 #endif
1173 
1174 	/*
1175 	 * Attempt to get a coherent snapshot by making a copy of the data.
1176 	 */
1177 	if (arg1)
1178 		tmplong = *(long *)arg1;
1179 	else
1180 		tmplong = arg2;
1181 #ifdef SCTL_MASK32
1182 	if (req->flags & SCTL_MASK32) {
1183 		tmpint = tmplong;
1184 		error = SYSCTL_OUT(req, &tmpint, sizeof(int));
1185 	} else
1186 #endif
1187 		error = SYSCTL_OUT(req, &tmplong, sizeof(long));
1188 
1189 	if (error || !req->newptr)
1190 		return (error);
1191 
1192 	if (!arg1)
1193 		error = EPERM;
1194 #ifdef SCTL_MASK32
1195 	else if (req->flags & SCTL_MASK32) {
1196 		error = SYSCTL_IN(req, &tmpint, sizeof(int));
1197 		*(long *)arg1 = (long)tmpint;
1198 	}
1199 #endif
1200 	else
1201 		error = SYSCTL_IN(req, arg1, sizeof(long));
1202 	return (error);
1203 }
1204 
1205 /*
1206  * Handle a 64 bit int, signed or unsigned.
1207  * Two cases:
1208  *     a variable:  point arg1 at it.
1209  *     a constant:  pass it in arg2.
1210  */
1211 int
1212 sysctl_handle_64(SYSCTL_HANDLER_ARGS)
1213 {
1214 	int error = 0;
1215 	uint64_t tmpout;
1216 
1217 	/*
1218 	 * Attempt to get a coherent snapshot by making a copy of the data.
1219 	 */
1220 	if (arg1)
1221 		tmpout = *(uint64_t *)arg1;
1222 	else
1223 		tmpout = arg2;
1224 	error = SYSCTL_OUT(req, &tmpout, sizeof(uint64_t));
1225 
1226 	if (error || !req->newptr)
1227 		return (error);
1228 
1229 	if (!arg1)
1230 		error = EPERM;
1231 	else
1232 		error = SYSCTL_IN(req, arg1, sizeof(uint64_t));
1233 	return (error);
1234 }
1235 
1236 /*
1237  * Handle our generic '\0' terminated 'C' string.
1238  * Two cases:
1239  * 	a variable string:  point arg1 at it, arg2 is max length.
1240  * 	a constant string:  point arg1 at it, arg2 is zero.
1241  */
1242 
1243 int
1244 sysctl_handle_string(SYSCTL_HANDLER_ARGS)
1245 {
1246 	size_t outlen;
1247 	int error = 0, ro_string = 0;
1248 
1249 	/*
1250 	 * A zero-length buffer indicates a fixed size read-only
1251 	 * string:
1252 	 */
1253 	if (arg2 == 0) {
1254 		arg2 = strlen((char *)arg1) + 1;
1255 		ro_string = 1;
1256 	}
1257 
1258 	if (req->oldptr != NULL) {
1259 		char *tmparg;
1260 
1261 		if (ro_string) {
1262 			tmparg = arg1;
1263 		} else {
1264 			/* try to make a coherent snapshot of the string */
1265 			tmparg = malloc(arg2, M_SYSCTLTMP, M_WAITOK);
1266 			memcpy(tmparg, arg1, arg2);
1267 		}
1268 
1269 		outlen = strnlen(tmparg, arg2 - 1) + 1;
1270 		error = SYSCTL_OUT(req, tmparg, outlen);
1271 
1272 		if (!ro_string)
1273 			free(tmparg, M_SYSCTLTMP);
1274 	} else {
1275 		outlen = strnlen((char *)arg1, arg2 - 1) + 1;
1276 		error = SYSCTL_OUT(req, NULL, outlen);
1277 	}
1278 	if (error || !req->newptr)
1279 		return (error);
1280 
1281 	if ((req->newlen - req->newidx) >= arg2) {
1282 		error = EINVAL;
1283 	} else {
1284 		arg2 = (req->newlen - req->newidx);
1285 		error = SYSCTL_IN(req, arg1, arg2);
1286 		((char *)arg1)[arg2] = '\0';
1287 	}
1288 	return (error);
1289 }
1290 
1291 /*
1292  * Handle any kind of opaque data.
1293  * arg1 points to it, arg2 is the size.
1294  */
1295 
1296 int
1297 sysctl_handle_opaque(SYSCTL_HANDLER_ARGS)
1298 {
1299 	int error, tries;
1300 	u_int generation;
1301 	struct sysctl_req req2;
1302 
1303 	/*
1304 	 * Attempt to get a coherent snapshot, by using the thread
1305 	 * pre-emption counter updated from within mi_switch() to
1306 	 * determine if we were pre-empted during a bcopy() or
1307 	 * copyout(). Make 3 attempts at doing this before giving up.
1308 	 * If we encounter an error, stop immediately.
1309 	 */
1310 	tries = 0;
1311 	req2 = *req;
1312 retry:
1313 	generation = curthread->td_generation;
1314 	error = SYSCTL_OUT(req, arg1, arg2);
1315 	if (error)
1316 		return (error);
1317 	tries++;
1318 	if (generation != curthread->td_generation && tries < 3) {
1319 		*req = req2;
1320 		goto retry;
1321 	}
1322 
1323 	error = SYSCTL_IN(req, arg1, arg2);
1324 
1325 	return (error);
1326 }
1327 
1328 /*
1329  * Transfer functions to/from kernel space.
1330  * XXX: rather untested at this point
1331  */
1332 static int
1333 sysctl_old_kernel(struct sysctl_req *req, const void *p, size_t l)
1334 {
1335 	size_t i = 0;
1336 
1337 	if (req->oldptr) {
1338 		i = l;
1339 		if (req->oldlen <= req->oldidx)
1340 			i = 0;
1341 		else
1342 			if (i > req->oldlen - req->oldidx)
1343 				i = req->oldlen - req->oldidx;
1344 		if (i > 0)
1345 			bcopy(p, (char *)req->oldptr + req->oldidx, i);
1346 	}
1347 	req->oldidx += l;
1348 	if (req->oldptr && i != l)
1349 		return (ENOMEM);
1350 	return (0);
1351 }
1352 
1353 static int
1354 sysctl_new_kernel(struct sysctl_req *req, void *p, size_t l)
1355 {
1356 	if (!req->newptr)
1357 		return (0);
1358 	if (req->newlen - req->newidx < l)
1359 		return (EINVAL);
1360 	bcopy((char *)req->newptr + req->newidx, p, l);
1361 	req->newidx += l;
1362 	return (0);
1363 }
1364 
1365 int
1366 kernel_sysctl(struct thread *td, int *name, u_int namelen, void *old,
1367     size_t *oldlenp, void *new, size_t newlen, size_t *retval, int flags)
1368 {
1369 	int error = 0;
1370 	struct sysctl_req req;
1371 
1372 	bzero(&req, sizeof req);
1373 
1374 	req.td = td;
1375 	req.flags = flags;
1376 
1377 	if (oldlenp) {
1378 		req.oldlen = *oldlenp;
1379 	}
1380 	req.validlen = req.oldlen;
1381 
1382 	if (old) {
1383 		req.oldptr= old;
1384 	}
1385 
1386 	if (new != NULL) {
1387 		req.newlen = newlen;
1388 		req.newptr = new;
1389 	}
1390 
1391 	req.oldfunc = sysctl_old_kernel;
1392 	req.newfunc = sysctl_new_kernel;
1393 	req.lock = REQ_UNWIRED;
1394 
1395 	SYSCTL_SLOCK();
1396 	error = sysctl_root(0, name, namelen, &req);
1397 	SYSCTL_SUNLOCK();
1398 
1399 	if (req.lock == REQ_WIRED && req.validlen > 0)
1400 		vsunlock(req.oldptr, req.validlen);
1401 
1402 	if (error && error != ENOMEM)
1403 		return (error);
1404 
1405 	if (retval) {
1406 		if (req.oldptr && req.oldidx > req.validlen)
1407 			*retval = req.validlen;
1408 		else
1409 			*retval = req.oldidx;
1410 	}
1411 	return (error);
1412 }
1413 
1414 int
1415 kernel_sysctlbyname(struct thread *td, char *name, void *old, size_t *oldlenp,
1416     void *new, size_t newlen, size_t *retval, int flags)
1417 {
1418         int oid[CTL_MAXNAME];
1419         size_t oidlen, plen;
1420 	int error;
1421 
1422 	oid[0] = 0;		/* sysctl internal magic */
1423 	oid[1] = 3;		/* name2oid */
1424 	oidlen = sizeof(oid);
1425 
1426 	error = kernel_sysctl(td, oid, 2, oid, &oidlen,
1427 	    (void *)name, strlen(name), &plen, flags);
1428 	if (error)
1429 		return (error);
1430 
1431 	error = kernel_sysctl(td, oid, plen / sizeof(int), old, oldlenp,
1432 	    new, newlen, retval, flags);
1433 	return (error);
1434 }
1435 
1436 /*
1437  * Transfer function to/from user space.
1438  */
1439 static int
1440 sysctl_old_user(struct sysctl_req *req, const void *p, size_t l)
1441 {
1442 	size_t i, len, origidx;
1443 	int error;
1444 
1445 	origidx = req->oldidx;
1446 	req->oldidx += l;
1447 	if (req->oldptr == NULL)
1448 		return (0);
1449 	/*
1450 	 * If we have not wired the user supplied buffer and we are currently
1451 	 * holding locks, drop a witness warning, as it's possible that
1452 	 * write operations to the user page can sleep.
1453 	 */
1454 	if (req->lock != REQ_WIRED)
1455 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
1456 		    "sysctl_old_user()");
1457 	i = l;
1458 	len = req->validlen;
1459 	if (len <= origidx)
1460 		i = 0;
1461 	else {
1462 		if (i > len - origidx)
1463 			i = len - origidx;
1464 		if (req->lock == REQ_WIRED) {
1465 			error = copyout_nofault(p, (char *)req->oldptr +
1466 			    origidx, i);
1467 		} else
1468 			error = copyout(p, (char *)req->oldptr + origidx, i);
1469 		if (error != 0)
1470 			return (error);
1471 	}
1472 	if (i < l)
1473 		return (ENOMEM);
1474 	return (0);
1475 }
1476 
1477 static int
1478 sysctl_new_user(struct sysctl_req *req, void *p, size_t l)
1479 {
1480 	int error;
1481 
1482 	if (!req->newptr)
1483 		return (0);
1484 	if (req->newlen - req->newidx < l)
1485 		return (EINVAL);
1486 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
1487 	    "sysctl_new_user()");
1488 	error = copyin((char *)req->newptr + req->newidx, p, l);
1489 	req->newidx += l;
1490 	return (error);
1491 }
1492 
1493 /*
1494  * Wire the user space destination buffer.  If set to a value greater than
1495  * zero, the len parameter limits the maximum amount of wired memory.
1496  */
1497 int
1498 sysctl_wire_old_buffer(struct sysctl_req *req, size_t len)
1499 {
1500 	int ret;
1501 	size_t wiredlen;
1502 
1503 	wiredlen = (len > 0 && len < req->oldlen) ? len : req->oldlen;
1504 	ret = 0;
1505 	if (req->lock != REQ_WIRED && req->oldptr &&
1506 	    req->oldfunc == sysctl_old_user) {
1507 		if (wiredlen != 0) {
1508 			ret = vslock(req->oldptr, wiredlen);
1509 			if (ret != 0) {
1510 				if (ret != ENOMEM)
1511 					return (ret);
1512 				wiredlen = 0;
1513 			}
1514 		}
1515 		req->lock = REQ_WIRED;
1516 		req->validlen = wiredlen;
1517 	}
1518 	return (0);
1519 }
1520 
1521 int
1522 sysctl_find_oid(int *name, u_int namelen, struct sysctl_oid **noid,
1523     int *nindx, struct sysctl_req *req)
1524 {
1525 	struct sysctl_oid_list *lsp;
1526 	struct sysctl_oid *oid;
1527 	int indx;
1528 
1529 	SYSCTL_ASSERT_LOCKED();
1530 	lsp = &sysctl__children;
1531 	indx = 0;
1532 	while (indx < CTL_MAXNAME) {
1533 		SLIST_FOREACH(oid, lsp, oid_link) {
1534 			if (oid->oid_number == name[indx])
1535 				break;
1536 		}
1537 		if (oid == NULL)
1538 			return (ENOENT);
1539 
1540 		indx++;
1541 		if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1542 			if (oid->oid_handler != NULL || indx == namelen) {
1543 				*noid = oid;
1544 				if (nindx != NULL)
1545 					*nindx = indx;
1546 				KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
1547 				    ("%s found DYING node %p", __func__, oid));
1548 				return (0);
1549 			}
1550 			lsp = SYSCTL_CHILDREN(oid);
1551 		} else if (indx == namelen) {
1552 			*noid = oid;
1553 			if (nindx != NULL)
1554 				*nindx = indx;
1555 			KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
1556 			    ("%s found DYING node %p", __func__, oid));
1557 			return (0);
1558 		} else {
1559 			return (ENOTDIR);
1560 		}
1561 	}
1562 	return (ENOENT);
1563 }
1564 
1565 /*
1566  * Traverse our tree, and find the right node, execute whatever it points
1567  * to, and return the resulting error code.
1568  */
1569 
1570 static int
1571 sysctl_root(SYSCTL_HANDLER_ARGS)
1572 {
1573 	struct sysctl_oid *oid;
1574 	int error, indx, lvl;
1575 
1576 	SYSCTL_ASSERT_SLOCKED();
1577 
1578 	error = sysctl_find_oid(arg1, arg2, &oid, &indx, req);
1579 	if (error)
1580 		return (error);
1581 
1582 	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1583 		/*
1584 		 * You can't call a sysctl when it's a node, but has
1585 		 * no handler.  Inform the user that it's a node.
1586 		 * The indx may or may not be the same as namelen.
1587 		 */
1588 		if (oid->oid_handler == NULL)
1589 			return (EISDIR);
1590 	}
1591 
1592 	/* Is this sysctl writable? */
1593 	if (req->newptr && !(oid->oid_kind & CTLFLAG_WR))
1594 		return (EPERM);
1595 
1596 	KASSERT(req->td != NULL, ("sysctl_root(): req->td == NULL"));
1597 
1598 #ifdef CAPABILITY_MODE
1599 	/*
1600 	 * If the process is in capability mode, then don't permit reading or
1601 	 * writing unless specifically granted for the node.
1602 	 */
1603 	if (IN_CAPABILITY_MODE(req->td)) {
1604 		if (req->oldptr && !(oid->oid_kind & CTLFLAG_CAPRD))
1605 			return (EPERM);
1606 		if (req->newptr && !(oid->oid_kind & CTLFLAG_CAPWR))
1607 			return (EPERM);
1608 	}
1609 #endif
1610 
1611 	/* Is this sysctl sensitive to securelevels? */
1612 	if (req->newptr && (oid->oid_kind & CTLFLAG_SECURE)) {
1613 		lvl = (oid->oid_kind & CTLMASK_SECURE) >> CTLSHIFT_SECURE;
1614 		error = securelevel_gt(req->td->td_ucred, lvl);
1615 		if (error)
1616 			return (error);
1617 	}
1618 
1619 	/* Is this sysctl writable by only privileged users? */
1620 	if (req->newptr && !(oid->oid_kind & CTLFLAG_ANYBODY)) {
1621 		int priv;
1622 
1623 		if (oid->oid_kind & CTLFLAG_PRISON)
1624 			priv = PRIV_SYSCTL_WRITEJAIL;
1625 #ifdef VIMAGE
1626 		else if ((oid->oid_kind & CTLFLAG_VNET) &&
1627 		     prison_owns_vnet(req->td->td_ucred))
1628 			priv = PRIV_SYSCTL_WRITEJAIL;
1629 #endif
1630 		else
1631 			priv = PRIV_SYSCTL_WRITE;
1632 		error = priv_check(req->td, priv);
1633 		if (error)
1634 			return (error);
1635 	}
1636 
1637 	if (!oid->oid_handler)
1638 		return (EINVAL);
1639 
1640 	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1641 		arg1 = (int *)arg1 + indx;
1642 		arg2 -= indx;
1643 	} else {
1644 		arg1 = oid->oid_arg1;
1645 		arg2 = oid->oid_arg2;
1646 	}
1647 #ifdef MAC
1648 	error = mac_system_check_sysctl(req->td->td_ucred, oid, arg1, arg2,
1649 	    req);
1650 	if (error != 0)
1651 		return (error);
1652 #endif
1653 #ifdef VIMAGE
1654 	if ((oid->oid_kind & CTLFLAG_VNET) && arg1 != NULL)
1655 		arg1 = (void *)(curvnet->vnet_data_base + (uintptr_t)arg1);
1656 #endif
1657 	error = sysctl_root_handler_locked(oid, arg1, arg2, req);
1658 
1659 	KFAIL_POINT_ERROR(_debug_fail_point, sysctl_running, error);
1660 
1661 	return (error);
1662 }
1663 
1664 #ifndef _SYS_SYSPROTO_H_
1665 struct sysctl_args {
1666 	int	*name;
1667 	u_int	namelen;
1668 	void	*old;
1669 	size_t	*oldlenp;
1670 	void	*new;
1671 	size_t	newlen;
1672 };
1673 #endif
1674 int
1675 sys___sysctl(struct thread *td, struct sysctl_args *uap)
1676 {
1677 	int error, i, name[CTL_MAXNAME];
1678 	size_t j;
1679 
1680 	if (uap->namelen > CTL_MAXNAME || uap->namelen < 2)
1681 		return (EINVAL);
1682 
1683  	error = copyin(uap->name, &name, uap->namelen * sizeof(int));
1684  	if (error)
1685 		return (error);
1686 
1687 	error = userland_sysctl(td, name, uap->namelen,
1688 		uap->old, uap->oldlenp, 0,
1689 		uap->new, uap->newlen, &j, 0);
1690 	if (error && error != ENOMEM)
1691 		return (error);
1692 	if (uap->oldlenp) {
1693 		i = copyout(&j, uap->oldlenp, sizeof(j));
1694 		if (i)
1695 			return (i);
1696 	}
1697 	return (error);
1698 }
1699 
1700 /*
1701  * This is used from various compatibility syscalls too.  That's why name
1702  * must be in kernel space.
1703  */
1704 int
1705 userland_sysctl(struct thread *td, int *name, u_int namelen, void *old,
1706     size_t *oldlenp, int inkernel, void *new, size_t newlen, size_t *retval,
1707     int flags)
1708 {
1709 	int error = 0, memlocked;
1710 	struct sysctl_req req;
1711 
1712 	bzero(&req, sizeof req);
1713 
1714 	req.td = td;
1715 	req.flags = flags;
1716 
1717 	if (oldlenp) {
1718 		if (inkernel) {
1719 			req.oldlen = *oldlenp;
1720 		} else {
1721 			error = copyin(oldlenp, &req.oldlen, sizeof(*oldlenp));
1722 			if (error)
1723 				return (error);
1724 		}
1725 	}
1726 	req.validlen = req.oldlen;
1727 
1728 	if (old) {
1729 		if (!useracc(old, req.oldlen, VM_PROT_WRITE))
1730 			return (EFAULT);
1731 		req.oldptr= old;
1732 	}
1733 
1734 	if (new != NULL) {
1735 		if (!useracc(new, newlen, VM_PROT_READ))
1736 			return (EFAULT);
1737 		req.newlen = newlen;
1738 		req.newptr = new;
1739 	}
1740 
1741 	req.oldfunc = sysctl_old_user;
1742 	req.newfunc = sysctl_new_user;
1743 	req.lock = REQ_UNWIRED;
1744 
1745 #ifdef KTRACE
1746 	if (KTRPOINT(curthread, KTR_SYSCTL))
1747 		ktrsysctl(name, namelen);
1748 #endif
1749 
1750 	if (req.oldlen > PAGE_SIZE) {
1751 		memlocked = 1;
1752 		sx_xlock(&sysctlmemlock);
1753 	} else
1754 		memlocked = 0;
1755 	CURVNET_SET(TD_TO_VNET(td));
1756 
1757 	for (;;) {
1758 		req.oldidx = 0;
1759 		req.newidx = 0;
1760 		SYSCTL_SLOCK();
1761 		error = sysctl_root(0, name, namelen, &req);
1762 		SYSCTL_SUNLOCK();
1763 		if (error != EAGAIN)
1764 			break;
1765 		kern_yield(PRI_USER);
1766 	}
1767 
1768 	CURVNET_RESTORE();
1769 
1770 	if (req.lock == REQ_WIRED && req.validlen > 0)
1771 		vsunlock(req.oldptr, req.validlen);
1772 	if (memlocked)
1773 		sx_xunlock(&sysctlmemlock);
1774 
1775 	if (error && error != ENOMEM)
1776 		return (error);
1777 
1778 	if (retval) {
1779 		if (req.oldptr && req.oldidx > req.validlen)
1780 			*retval = req.validlen;
1781 		else
1782 			*retval = req.oldidx;
1783 	}
1784 	return (error);
1785 }
1786 
1787 /*
1788  * Drain into a sysctl struct.  The user buffer should be wired if a page
1789  * fault would cause issue.
1790  */
1791 static int
1792 sbuf_sysctl_drain(void *arg, const char *data, int len)
1793 {
1794 	struct sysctl_req *req = arg;
1795 	int error;
1796 
1797 	error = SYSCTL_OUT(req, data, len);
1798 	KASSERT(error >= 0, ("Got unexpected negative value %d", error));
1799 	return (error == 0 ? len : -error);
1800 }
1801 
1802 struct sbuf *
1803 sbuf_new_for_sysctl(struct sbuf *s, char *buf, int length,
1804     struct sysctl_req *req)
1805 {
1806 
1807 	s = sbuf_new(s, buf, length, SBUF_FIXEDLEN);
1808 	sbuf_set_drain(s, sbuf_sysctl_drain, req);
1809 	return (s);
1810 }
1811