xref: /freebsd/sys/kern/kern_sysctl.c (revision a831d7d1c538b93ffec095657d9a6e6e778db195)
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 	if (oid->oid_kind & CTLFLAG_DYN)
178 		atomic_add_int(&oid->oid_running, 1);
179 	xlocked = sysctl_unlock();
180 
181 	if (!(oid->oid_kind & CTLFLAG_MPSAFE))
182 		mtx_lock(&Giant);
183 	error = oid->oid_handler(oid, arg1, arg2, req);
184 	if (!(oid->oid_kind & CTLFLAG_MPSAFE))
185 		mtx_unlock(&Giant);
186 
187 	sysctl_lock(xlocked);
188 	if (oid->oid_kind & CTLFLAG_DYN) {
189 		if (atomic_fetchadd_int(&oid->oid_running, -1) == 1 &&
190 		    (oid->oid_kind & CTLFLAG_DYING) != 0)
191 			wakeup(&oid->oid_running);
192 	}
193 
194 	return (error);
195 }
196 
197 static void
198 sysctl_load_tunable_by_oid_locked(struct sysctl_oid *oidp)
199 {
200 	struct sysctl_req req;
201 	struct sysctl_oid *curr;
202 	char *penv = NULL;
203 	char path[64];
204 	ssize_t rem = sizeof(path);
205 	ssize_t len;
206 	int val_int;
207 	long val_long;
208 	int64_t val_64;
209 	quad_t val_quad;
210 	int error;
211 
212 	path[--rem] = 0;
213 
214 	for (curr = oidp; curr != NULL; curr = SYSCTL_PARENT(curr)) {
215 		len = strlen(curr->oid_name);
216 		rem -= len;
217 		if (curr != oidp)
218 			rem -= 1;
219 		if (rem < 0) {
220 			printf("OID path exceeds %d bytes\n", (int)sizeof(path));
221 			return;
222 		}
223 		memcpy(path + rem, curr->oid_name, len);
224 		if (curr != oidp)
225 			path[rem + len] = '.';
226 	}
227 
228 	memset(&req, 0, sizeof(req));
229 
230 	req.td = curthread;
231 	req.oldfunc = sysctl_old_kernel;
232 	req.newfunc = sysctl_new_kernel;
233 	req.lock = REQ_UNWIRED;
234 
235 	switch (oidp->oid_kind & CTLTYPE) {
236 	case CTLTYPE_INT:
237 		if (getenv_int(path + rem, &val_int) == 0)
238 			return;
239 		req.newlen = sizeof(val_int);
240 		req.newptr = &val_int;
241 		break;
242 	case CTLTYPE_UINT:
243 		if (getenv_uint(path + rem, (unsigned int *)&val_int) == 0)
244 			return;
245 		req.newlen = sizeof(val_int);
246 		req.newptr = &val_int;
247 		break;
248 	case CTLTYPE_LONG:
249 		if (getenv_long(path + rem, &val_long) == 0)
250 			return;
251 		req.newlen = sizeof(val_long);
252 		req.newptr = &val_long;
253 		break;
254 	case CTLTYPE_ULONG:
255 		if (getenv_ulong(path + rem, (unsigned long *)&val_long) == 0)
256 			return;
257 		req.newlen = sizeof(val_long);
258 		req.newptr = &val_long;
259 		break;
260 	case CTLTYPE_S64:
261 		if (getenv_quad(path + rem, &val_quad) == 0)
262 			return;
263 		val_64 = val_quad;
264 		req.newlen = sizeof(val_64);
265 		req.newptr = &val_64;
266 		break;
267 	case CTLTYPE_U64:
268 		/* XXX there is no getenv_uquad() */
269 		if (getenv_quad(path + rem, &val_quad) == 0)
270 			return;
271 		val_64 = val_quad;
272 		req.newlen = sizeof(val_64);
273 		req.newptr = &val_64;
274 		break;
275 	case CTLTYPE_STRING:
276 		penv = kern_getenv(path + rem);
277 		if (penv == NULL)
278 			return;
279 		req.newlen = strlen(penv);
280 		req.newptr = penv;
281 		break;
282 	default:
283 		return;
284 	}
285 	error = sysctl_root_handler_locked(oidp, oidp->oid_arg1,
286 	    oidp->oid_arg2, &req);
287 	if (error != 0)
288 		printf("Setting sysctl %s failed: %d\n", path, error);
289 	if (penv != NULL)
290 		freeenv(penv);
291 }
292 
293 void
294 sysctl_register_oid(struct sysctl_oid *oidp)
295 {
296 	struct sysctl_oid_list *parent = oidp->oid_parent;
297 	struct sysctl_oid *p;
298 	struct sysctl_oid *q;
299 
300 	/*
301 	 * First check if another oid with the same name already
302 	 * exists in the parent's list.
303 	 */
304 	SYSCTL_ASSERT_XLOCKED();
305 	p = sysctl_find_oidname(oidp->oid_name, parent);
306 	if (p != NULL) {
307 		if ((p->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
308 			p->oid_refcnt++;
309 			return;
310 		} else {
311 			printf("can't re-use a leaf (%s)!\n", p->oid_name);
312 			return;
313 		}
314 	}
315 	/*
316 	 * If this oid has a number OID_AUTO, give it a number which
317 	 * is greater than any current oid.
318 	 * NOTE: DO NOT change the starting value here, change it in
319 	 * <sys/sysctl.h>, and make sure it is at least 256 to
320 	 * accomodate e.g. net.inet.raw as a static sysctl node.
321 	 */
322 	if (oidp->oid_number == OID_AUTO) {
323 		static int newoid = CTL_AUTO_START;
324 
325 		oidp->oid_number = newoid++;
326 		if (newoid == 0x7fffffff)
327 			panic("out of oids");
328 	}
329 #if 0
330 	else if (oidp->oid_number >= CTL_AUTO_START) {
331 		/* do not panic; this happens when unregistering sysctl sets */
332 		printf("static sysctl oid too high: %d", oidp->oid_number);
333 	}
334 #endif
335 
336 	/*
337 	 * Insert the oid into the parent's list in order.
338 	 */
339 	q = NULL;
340 	SLIST_FOREACH(p, parent, oid_link) {
341 		if (oidp->oid_number < p->oid_number)
342 			break;
343 		q = p;
344 	}
345 	if (q)
346 		SLIST_INSERT_AFTER(q, oidp, oid_link);
347 	else
348 		SLIST_INSERT_HEAD(parent, oidp, oid_link);
349 
350 	if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE &&
351 #ifdef VIMAGE
352 	    (oidp->oid_kind & CTLFLAG_VNET) == 0 &&
353 #endif
354 	    (oidp->oid_kind & CTLFLAG_TUN) != 0 &&
355 	    (oidp->oid_kind & CTLFLAG_NOFETCH) == 0) {
356 		sysctl_load_tunable_by_oid_locked(oidp);
357 	}
358 }
359 
360 void
361 sysctl_unregister_oid(struct sysctl_oid *oidp)
362 {
363 	struct sysctl_oid *p;
364 	int error;
365 
366 	SYSCTL_ASSERT_XLOCKED();
367 	error = ENOENT;
368 	if (oidp->oid_number == OID_AUTO) {
369 		error = EINVAL;
370 	} else {
371 		SLIST_FOREACH(p, oidp->oid_parent, oid_link) {
372 			if (p == oidp) {
373 				SLIST_REMOVE(oidp->oid_parent, oidp,
374 				    sysctl_oid, oid_link);
375 				error = 0;
376 				break;
377 			}
378 		}
379 	}
380 
381 	/*
382 	 * This can happen when a module fails to register and is
383 	 * being unloaded afterwards.  It should not be a panic()
384 	 * for normal use.
385 	 */
386 	if (error)
387 		printf("%s: failed to unregister sysctl\n", __func__);
388 }
389 
390 /* Initialize a new context to keep track of dynamically added sysctls. */
391 int
392 sysctl_ctx_init(struct sysctl_ctx_list *c)
393 {
394 
395 	if (c == NULL) {
396 		return (EINVAL);
397 	}
398 
399 	/*
400 	 * No locking here, the caller is responsible for not adding
401 	 * new nodes to a context until after this function has
402 	 * returned.
403 	 */
404 	TAILQ_INIT(c);
405 	return (0);
406 }
407 
408 /* Free the context, and destroy all dynamic oids registered in this context */
409 int
410 sysctl_ctx_free(struct sysctl_ctx_list *clist)
411 {
412 	struct sysctl_ctx_entry *e, *e1;
413 	int error;
414 
415 	error = 0;
416 	/*
417 	 * First perform a "dry run" to check if it's ok to remove oids.
418 	 * XXX FIXME
419 	 * XXX This algorithm is a hack. But I don't know any
420 	 * XXX better solution for now...
421 	 */
422 	SYSCTL_XLOCK();
423 	TAILQ_FOREACH(e, clist, link) {
424 		error = sysctl_remove_oid_locked(e->entry, 0, 0);
425 		if (error)
426 			break;
427 	}
428 	/*
429 	 * Restore deregistered entries, either from the end,
430 	 * or from the place where error occured.
431 	 * e contains the entry that was not unregistered
432 	 */
433 	if (error)
434 		e1 = TAILQ_PREV(e, sysctl_ctx_list, link);
435 	else
436 		e1 = TAILQ_LAST(clist, sysctl_ctx_list);
437 	while (e1 != NULL) {
438 		sysctl_register_oid(e1->entry);
439 		e1 = TAILQ_PREV(e1, sysctl_ctx_list, link);
440 	}
441 	if (error) {
442 		SYSCTL_XUNLOCK();
443 		return(EBUSY);
444 	}
445 	/* Now really delete the entries */
446 	e = TAILQ_FIRST(clist);
447 	while (e != NULL) {
448 		e1 = TAILQ_NEXT(e, link);
449 		error = sysctl_remove_oid_locked(e->entry, 1, 0);
450 		if (error)
451 			panic("sysctl_remove_oid: corrupt tree, entry: %s",
452 			    e->entry->oid_name);
453 		free(e, M_SYSCTLOID);
454 		e = e1;
455 	}
456 	SYSCTL_XUNLOCK();
457 	return (error);
458 }
459 
460 /* Add an entry to the context */
461 struct sysctl_ctx_entry *
462 sysctl_ctx_entry_add(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
463 {
464 	struct sysctl_ctx_entry *e;
465 
466 	SYSCTL_ASSERT_XLOCKED();
467 	if (clist == NULL || oidp == NULL)
468 		return(NULL);
469 	e = malloc(sizeof(struct sysctl_ctx_entry), M_SYSCTLOID, M_WAITOK);
470 	e->entry = oidp;
471 	TAILQ_INSERT_HEAD(clist, e, link);
472 	return (e);
473 }
474 
475 /* Find an entry in the context */
476 struct sysctl_ctx_entry *
477 sysctl_ctx_entry_find(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
478 {
479 	struct sysctl_ctx_entry *e;
480 
481 	SYSCTL_ASSERT_XLOCKED();
482 	if (clist == NULL || oidp == NULL)
483 		return(NULL);
484 	TAILQ_FOREACH(e, clist, link) {
485 		if(e->entry == oidp)
486 			return(e);
487 	}
488 	return (e);
489 }
490 
491 /*
492  * Delete an entry from the context.
493  * NOTE: this function doesn't free oidp! You have to remove it
494  * with sysctl_remove_oid().
495  */
496 int
497 sysctl_ctx_entry_del(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
498 {
499 	struct sysctl_ctx_entry *e;
500 
501 	if (clist == NULL || oidp == NULL)
502 		return (EINVAL);
503 	SYSCTL_XLOCK();
504 	e = sysctl_ctx_entry_find(clist, oidp);
505 	if (e != NULL) {
506 		TAILQ_REMOVE(clist, e, link);
507 		SYSCTL_XUNLOCK();
508 		free(e, M_SYSCTLOID);
509 		return (0);
510 	} else {
511 		SYSCTL_XUNLOCK();
512 		return (ENOENT);
513 	}
514 }
515 
516 /*
517  * Remove dynamically created sysctl trees.
518  * oidp - top of the tree to be removed
519  * del - if 0 - just deregister, otherwise free up entries as well
520  * recurse - if != 0 traverse the subtree to be deleted
521  */
522 int
523 sysctl_remove_oid(struct sysctl_oid *oidp, int del, int recurse)
524 {
525 	int error;
526 
527 	SYSCTL_XLOCK();
528 	error = sysctl_remove_oid_locked(oidp, del, recurse);
529 	SYSCTL_XUNLOCK();
530 	return (error);
531 }
532 
533 int
534 sysctl_remove_name(struct sysctl_oid *parent, const char *name,
535     int del, int recurse)
536 {
537 	struct sysctl_oid *p, *tmp;
538 	int error;
539 
540 	error = ENOENT;
541 	SYSCTL_XLOCK();
542 	SLIST_FOREACH_SAFE(p, SYSCTL_CHILDREN(parent), oid_link, tmp) {
543 		if (strcmp(p->oid_name, name) == 0) {
544 			error = sysctl_remove_oid_locked(p, del, recurse);
545 			break;
546 		}
547 	}
548 	SYSCTL_XUNLOCK();
549 
550 	return (error);
551 }
552 
553 
554 static int
555 sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del, int recurse)
556 {
557 	struct sysctl_oid *p, *tmp;
558 	int error;
559 
560 	SYSCTL_ASSERT_XLOCKED();
561 	if (oidp == NULL)
562 		return(EINVAL);
563 	if ((oidp->oid_kind & CTLFLAG_DYN) == 0) {
564 		printf("can't remove non-dynamic nodes!\n");
565 		return (EINVAL);
566 	}
567 	/*
568 	 * WARNING: normal method to do this should be through
569 	 * sysctl_ctx_free(). Use recursing as the last resort
570 	 * method to purge your sysctl tree of leftovers...
571 	 * However, if some other code still references these nodes,
572 	 * it will panic.
573 	 */
574 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
575 		if (oidp->oid_refcnt == 1) {
576 			SLIST_FOREACH_SAFE(p,
577 			    SYSCTL_CHILDREN(oidp), oid_link, tmp) {
578 				if (!recurse) {
579 					printf("Warning: failed attempt to "
580 					    "remove oid %s with child %s\n",
581 					    oidp->oid_name, p->oid_name);
582 					return (ENOTEMPTY);
583 				}
584 				error = sysctl_remove_oid_locked(p, del,
585 				    recurse);
586 				if (error)
587 					return (error);
588 			}
589 		}
590 	}
591 	if (oidp->oid_refcnt > 1 ) {
592 		oidp->oid_refcnt--;
593 	} else {
594 		if (oidp->oid_refcnt == 0) {
595 			printf("Warning: bad oid_refcnt=%u (%s)!\n",
596 				oidp->oid_refcnt, oidp->oid_name);
597 			return (EINVAL);
598 		}
599 		sysctl_unregister_oid(oidp);
600 		if (del) {
601 			/*
602 			 * Wait for all threads running the handler to drain.
603 			 * This preserves the previous behavior when the
604 			 * sysctl lock was held across a handler invocation,
605 			 * and is necessary for module unload correctness.
606 			 */
607 			while (oidp->oid_running > 0) {
608 				oidp->oid_kind |= CTLFLAG_DYING;
609 				SYSCTL_SLEEP(&oidp->oid_running, "oidrm", 0);
610 			}
611 			if (oidp->oid_descr)
612 				free(__DECONST(char *, oidp->oid_descr),
613 				    M_SYSCTLOID);
614 			free(__DECONST(char *, oidp->oid_name), M_SYSCTLOID);
615 			free(oidp, M_SYSCTLOID);
616 		}
617 	}
618 	return (0);
619 }
620 /*
621  * Create new sysctls at run time.
622  * clist may point to a valid context initialized with sysctl_ctx_init().
623  */
624 struct sysctl_oid *
625 sysctl_add_oid(struct sysctl_ctx_list *clist, struct sysctl_oid_list *parent,
626 	int number, const char *name, int kind, void *arg1, intptr_t arg2,
627 	int (*handler)(SYSCTL_HANDLER_ARGS), const char *fmt, const char *descr)
628 {
629 	struct sysctl_oid *oidp;
630 
631 	/* You have to hook up somewhere.. */
632 	if (parent == NULL)
633 		return(NULL);
634 	/* Check if the node already exists, otherwise create it */
635 	SYSCTL_XLOCK();
636 	oidp = sysctl_find_oidname(name, parent);
637 	if (oidp != NULL) {
638 		if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
639 			oidp->oid_refcnt++;
640 			/* Update the context */
641 			if (clist != NULL)
642 				sysctl_ctx_entry_add(clist, oidp);
643 			SYSCTL_XUNLOCK();
644 			return (oidp);
645 		} else {
646 			SYSCTL_XUNLOCK();
647 			printf("can't re-use a leaf (%s)!\n", name);
648 			return (NULL);
649 		}
650 	}
651 	oidp = malloc(sizeof(struct sysctl_oid), M_SYSCTLOID, M_WAITOK|M_ZERO);
652 	oidp->oid_parent = parent;
653 	SLIST_INIT(&oidp->oid_children);
654 	oidp->oid_number = number;
655 	oidp->oid_refcnt = 1;
656 	oidp->oid_name = strdup(name, M_SYSCTLOID);
657 	oidp->oid_handler = handler;
658 	oidp->oid_kind = CTLFLAG_DYN | kind;
659 	oidp->oid_arg1 = arg1;
660 	oidp->oid_arg2 = arg2;
661 	oidp->oid_fmt = fmt;
662 	if (descr != NULL)
663 		oidp->oid_descr = strdup(descr, M_SYSCTLOID);
664 	/* Update the context, if used */
665 	if (clist != NULL)
666 		sysctl_ctx_entry_add(clist, oidp);
667 	/* Register this oid */
668 	sysctl_register_oid(oidp);
669 	SYSCTL_XUNLOCK();
670 	return (oidp);
671 }
672 
673 /*
674  * Rename an existing oid.
675  */
676 void
677 sysctl_rename_oid(struct sysctl_oid *oidp, const char *name)
678 {
679 	char *newname;
680 	char *oldname;
681 
682 	newname = strdup(name, M_SYSCTLOID);
683 	SYSCTL_XLOCK();
684 	oldname = __DECONST(char *, oidp->oid_name);
685 	oidp->oid_name = newname;
686 	SYSCTL_XUNLOCK();
687 	free(oldname, M_SYSCTLOID);
688 }
689 
690 /*
691  * Reparent an existing oid.
692  */
693 int
694 sysctl_move_oid(struct sysctl_oid *oid, struct sysctl_oid_list *parent)
695 {
696 	struct sysctl_oid *oidp;
697 
698 	SYSCTL_XLOCK();
699 	if (oid->oid_parent == parent) {
700 		SYSCTL_XUNLOCK();
701 		return (0);
702 	}
703 	oidp = sysctl_find_oidname(oid->oid_name, parent);
704 	if (oidp != NULL) {
705 		SYSCTL_XUNLOCK();
706 		return (EEXIST);
707 	}
708 	sysctl_unregister_oid(oid);
709 	oid->oid_parent = parent;
710 	oid->oid_number = OID_AUTO;
711 	sysctl_register_oid(oid);
712 	SYSCTL_XUNLOCK();
713 	return (0);
714 }
715 
716 /*
717  * Register the kernel's oids on startup.
718  */
719 SET_DECLARE(sysctl_set, struct sysctl_oid);
720 
721 static void
722 sysctl_register_all(void *arg)
723 {
724 	struct sysctl_oid **oidp;
725 
726 	sx_init(&sysctlmemlock, "sysctl mem");
727 	SYSCTL_INIT();
728 	SYSCTL_XLOCK();
729 	SET_FOREACH(oidp, sysctl_set)
730 		sysctl_register_oid(*oidp);
731 	SYSCTL_XUNLOCK();
732 }
733 SYSINIT(sysctl, SI_SUB_KMEM, SI_ORDER_FIRST, sysctl_register_all, 0);
734 
735 /*
736  * "Staff-functions"
737  *
738  * These functions implement a presently undocumented interface
739  * used by the sysctl program to walk the tree, and get the type
740  * so it can print the value.
741  * This interface is under work and consideration, and should probably
742  * be killed with a big axe by the first person who can find the time.
743  * (be aware though, that the proper interface isn't as obvious as it
744  * may seem, there are various conflicting requirements.
745  *
746  * {0,0}	printf the entire MIB-tree.
747  * {0,1,...}	return the name of the "..." OID.
748  * {0,2,...}	return the next OID.
749  * {0,3}	return the OID of the name in "new"
750  * {0,4,...}	return the kind & format info for the "..." OID.
751  * {0,5,...}	return the description the "..." OID.
752  */
753 
754 #ifdef SYSCTL_DEBUG
755 static void
756 sysctl_sysctl_debug_dump_node(struct sysctl_oid_list *l, int i)
757 {
758 	int k;
759 	struct sysctl_oid *oidp;
760 
761 	SYSCTL_ASSERT_LOCKED();
762 	SLIST_FOREACH(oidp, l, oid_link) {
763 
764 		for (k=0; k<i; k++)
765 			printf(" ");
766 
767 		printf("%d %s ", oidp->oid_number, oidp->oid_name);
768 
769 		printf("%c%c",
770 			oidp->oid_kind & CTLFLAG_RD ? 'R':' ',
771 			oidp->oid_kind & CTLFLAG_WR ? 'W':' ');
772 
773 		if (oidp->oid_handler)
774 			printf(" *Handler");
775 
776 		switch (oidp->oid_kind & CTLTYPE) {
777 			case CTLTYPE_NODE:
778 				printf(" Node\n");
779 				if (!oidp->oid_handler) {
780 					sysctl_sysctl_debug_dump_node(
781 					    SYSCTL_CHILDREN(oidp), i + 2);
782 				}
783 				break;
784 			case CTLTYPE_INT:    printf(" Int\n"); break;
785 			case CTLTYPE_UINT:   printf(" u_int\n"); break;
786 			case CTLTYPE_LONG:   printf(" Long\n"); break;
787 			case CTLTYPE_ULONG:  printf(" u_long\n"); break;
788 			case CTLTYPE_STRING: printf(" String\n"); break;
789 			case CTLTYPE_U64:    printf(" uint64_t\n"); break;
790 			case CTLTYPE_S64:    printf(" int64_t\n"); break;
791 			case CTLTYPE_OPAQUE: printf(" Opaque/struct\n"); break;
792 			default:	     printf("\n");
793 		}
794 
795 	}
796 }
797 
798 static int
799 sysctl_sysctl_debug(SYSCTL_HANDLER_ARGS)
800 {
801 	int error;
802 
803 	error = priv_check(req->td, PRIV_SYSCTL_DEBUG);
804 	if (error)
805 		return (error);
806 	SYSCTL_SLOCK();
807 	sysctl_sysctl_debug_dump_node(&sysctl__children, 0);
808 	SYSCTL_SUNLOCK();
809 	return (ENOENT);
810 }
811 
812 SYSCTL_PROC(_sysctl, 0, debug, CTLTYPE_STRING|CTLFLAG_RD|CTLFLAG_MPSAFE,
813 	0, 0, sysctl_sysctl_debug, "-", "");
814 #endif
815 
816 static int
817 sysctl_sysctl_name(SYSCTL_HANDLER_ARGS)
818 {
819 	int *name = (int *) arg1;
820 	u_int namelen = arg2;
821 	int error = 0;
822 	struct sysctl_oid *oid;
823 	struct sysctl_oid_list *lsp = &sysctl__children, *lsp2;
824 	char buf[10];
825 
826 	SYSCTL_SLOCK();
827 	while (namelen) {
828 		if (!lsp) {
829 			snprintf(buf,sizeof(buf),"%d",*name);
830 			if (req->oldidx)
831 				error = SYSCTL_OUT(req, ".", 1);
832 			if (!error)
833 				error = SYSCTL_OUT(req, buf, strlen(buf));
834 			if (error)
835 				goto out;
836 			namelen--;
837 			name++;
838 			continue;
839 		}
840 		lsp2 = 0;
841 		SLIST_FOREACH(oid, lsp, oid_link) {
842 			if (oid->oid_number != *name)
843 				continue;
844 
845 			if (req->oldidx)
846 				error = SYSCTL_OUT(req, ".", 1);
847 			if (!error)
848 				error = SYSCTL_OUT(req, oid->oid_name,
849 					strlen(oid->oid_name));
850 			if (error)
851 				goto out;
852 
853 			namelen--;
854 			name++;
855 
856 			if ((oid->oid_kind & CTLTYPE) != CTLTYPE_NODE)
857 				break;
858 
859 			if (oid->oid_handler)
860 				break;
861 
862 			lsp2 = SYSCTL_CHILDREN(oid);
863 			break;
864 		}
865 		lsp = lsp2;
866 	}
867 	error = SYSCTL_OUT(req, "", 1);
868  out:
869 	SYSCTL_SUNLOCK();
870 	return (error);
871 }
872 
873 /*
874  * XXXRW/JA: Shouldn't return name data for nodes that we don't permit in
875  * capability mode.
876  */
877 static SYSCTL_NODE(_sysctl, 1, name, CTLFLAG_RD | CTLFLAG_MPSAFE | CTLFLAG_CAPRD,
878     sysctl_sysctl_name, "");
879 
880 static int
881 sysctl_sysctl_next_ls(struct sysctl_oid_list *lsp, int *name, u_int namelen,
882 	int *next, int *len, int level, struct sysctl_oid **oidpp)
883 {
884 	struct sysctl_oid *oidp;
885 
886 	SYSCTL_ASSERT_LOCKED();
887 	*len = level;
888 	SLIST_FOREACH(oidp, lsp, oid_link) {
889 		*next = oidp->oid_number;
890 		*oidpp = oidp;
891 
892 		if (oidp->oid_kind & CTLFLAG_SKIP)
893 			continue;
894 
895 		if (!namelen) {
896 			if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
897 				return (0);
898 			if (oidp->oid_handler)
899 				/* We really should call the handler here...*/
900 				return (0);
901 			lsp = SYSCTL_CHILDREN(oidp);
902 			if (!sysctl_sysctl_next_ls(lsp, 0, 0, next+1,
903 				len, level+1, oidpp))
904 				return (0);
905 			goto emptynode;
906 		}
907 
908 		if (oidp->oid_number < *name)
909 			continue;
910 
911 		if (oidp->oid_number > *name) {
912 			if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
913 				return (0);
914 			if (oidp->oid_handler)
915 				return (0);
916 			lsp = SYSCTL_CHILDREN(oidp);
917 			if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1,
918 				next+1, len, level+1, oidpp))
919 				return (0);
920 			goto next;
921 		}
922 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
923 			continue;
924 
925 		if (oidp->oid_handler)
926 			continue;
927 
928 		lsp = SYSCTL_CHILDREN(oidp);
929 		if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1, next+1,
930 			len, level+1, oidpp))
931 			return (0);
932 	next:
933 		namelen = 1;
934 	emptynode:
935 		*len = level;
936 	}
937 	return (1);
938 }
939 
940 static int
941 sysctl_sysctl_next(SYSCTL_HANDLER_ARGS)
942 {
943 	int *name = (int *) arg1;
944 	u_int namelen = arg2;
945 	int i, j, error;
946 	struct sysctl_oid *oid;
947 	struct sysctl_oid_list *lsp = &sysctl__children;
948 	int newoid[CTL_MAXNAME];
949 
950 	SYSCTL_SLOCK();
951 	i = sysctl_sysctl_next_ls(lsp, name, namelen, newoid, &j, 1, &oid);
952 	SYSCTL_SUNLOCK();
953 	if (i)
954 		return (ENOENT);
955 	error = SYSCTL_OUT(req, newoid, j * sizeof (int));
956 	return (error);
957 }
958 
959 /*
960  * XXXRW/JA: Shouldn't return next data for nodes that we don't permit in
961  * capability mode.
962  */
963 static SYSCTL_NODE(_sysctl, 2, next, CTLFLAG_RD | CTLFLAG_MPSAFE | CTLFLAG_CAPRD,
964     sysctl_sysctl_next, "");
965 
966 static int
967 name2oid(char *name, int *oid, int *len, struct sysctl_oid **oidpp)
968 {
969 	struct sysctl_oid *oidp;
970 	struct sysctl_oid_list *lsp = &sysctl__children;
971 	char *p;
972 
973 	SYSCTL_ASSERT_LOCKED();
974 
975 	for (*len = 0; *len < CTL_MAXNAME;) {
976 		p = strsep(&name, ".");
977 
978 		oidp = SLIST_FIRST(lsp);
979 		for (;; oidp = SLIST_NEXT(oidp, oid_link)) {
980 			if (oidp == NULL)
981 				return (ENOENT);
982 			if (strcmp(p, oidp->oid_name) == 0)
983 				break;
984 		}
985 		*oid++ = oidp->oid_number;
986 		(*len)++;
987 
988 		if (name == NULL || *name == '\0') {
989 			if (oidpp)
990 				*oidpp = oidp;
991 			return (0);
992 		}
993 
994 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
995 			break;
996 
997 		if (oidp->oid_handler)
998 			break;
999 
1000 		lsp = SYSCTL_CHILDREN(oidp);
1001 	}
1002 	return (ENOENT);
1003 }
1004 
1005 static int
1006 sysctl_sysctl_name2oid(SYSCTL_HANDLER_ARGS)
1007 {
1008 	char *p;
1009 	int error, oid[CTL_MAXNAME], len = 0;
1010 	struct sysctl_oid *op = 0;
1011 
1012 	if (!req->newlen)
1013 		return (ENOENT);
1014 	if (req->newlen >= MAXPATHLEN)	/* XXX arbitrary, undocumented */
1015 		return (ENAMETOOLONG);
1016 
1017 	p = malloc(req->newlen+1, M_SYSCTL, M_WAITOK);
1018 
1019 	error = SYSCTL_IN(req, p, req->newlen);
1020 	if (error) {
1021 		free(p, M_SYSCTL);
1022 		return (error);
1023 	}
1024 
1025 	p [req->newlen] = '\0';
1026 
1027 	SYSCTL_SLOCK();
1028 	error = name2oid(p, oid, &len, &op);
1029 	SYSCTL_SUNLOCK();
1030 
1031 	free(p, M_SYSCTL);
1032 
1033 	if (error)
1034 		return (error);
1035 
1036 	error = SYSCTL_OUT(req, oid, len * sizeof *oid);
1037 	return (error);
1038 }
1039 
1040 /*
1041  * XXXRW/JA: Shouldn't return name2oid data for nodes that we don't permit in
1042  * capability mode.
1043  */
1044 SYSCTL_PROC(_sysctl, 3, name2oid,
1045     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_MPSAFE
1046     | CTLFLAG_CAPRW, 0, 0, sysctl_sysctl_name2oid, "I", "");
1047 
1048 static int
1049 sysctl_sysctl_oidfmt(SYSCTL_HANDLER_ARGS)
1050 {
1051 	struct sysctl_oid *oid;
1052 	int error;
1053 
1054 	SYSCTL_SLOCK();
1055 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1056 	if (error)
1057 		goto out;
1058 
1059 	if (oid->oid_fmt == NULL) {
1060 		error = ENOENT;
1061 		goto out;
1062 	}
1063 	error = SYSCTL_OUT(req, &oid->oid_kind, sizeof(oid->oid_kind));
1064 	if (error)
1065 		goto out;
1066 	error = SYSCTL_OUT(req, oid->oid_fmt, strlen(oid->oid_fmt) + 1);
1067  out:
1068 	SYSCTL_SUNLOCK();
1069 	return (error);
1070 }
1071 
1072 
1073 static SYSCTL_NODE(_sysctl, 4, oidfmt, CTLFLAG_RD|CTLFLAG_MPSAFE|CTLFLAG_CAPRD,
1074     sysctl_sysctl_oidfmt, "");
1075 
1076 static int
1077 sysctl_sysctl_oiddescr(SYSCTL_HANDLER_ARGS)
1078 {
1079 	struct sysctl_oid *oid;
1080 	int error;
1081 
1082 	SYSCTL_SLOCK();
1083 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1084 	if (error)
1085 		goto out;
1086 
1087 	if (oid->oid_descr == NULL) {
1088 		error = ENOENT;
1089 		goto out;
1090 	}
1091 	error = SYSCTL_OUT(req, oid->oid_descr, strlen(oid->oid_descr) + 1);
1092  out:
1093 	SYSCTL_SUNLOCK();
1094 	return (error);
1095 }
1096 
1097 static SYSCTL_NODE(_sysctl, 5, oiddescr, CTLFLAG_RD|CTLFLAG_MPSAFE|CTLFLAG_CAPRD,
1098     sysctl_sysctl_oiddescr, "");
1099 
1100 /*
1101  * Default "handler" functions.
1102  */
1103 
1104 /*
1105  * Handle an int, signed or unsigned.
1106  * Two cases:
1107  *     a variable:  point arg1 at it.
1108  *     a constant:  pass it in arg2.
1109  */
1110 
1111 int
1112 sysctl_handle_int(SYSCTL_HANDLER_ARGS)
1113 {
1114 	int tmpout, error = 0;
1115 
1116 	/*
1117 	 * Attempt to get a coherent snapshot by making a copy of the data.
1118 	 */
1119 	if (arg1)
1120 		tmpout = *(int *)arg1;
1121 	else
1122 		tmpout = arg2;
1123 	error = SYSCTL_OUT(req, &tmpout, sizeof(int));
1124 
1125 	if (error || !req->newptr)
1126 		return (error);
1127 
1128 	if (!arg1)
1129 		error = EPERM;
1130 	else
1131 		error = SYSCTL_IN(req, arg1, sizeof(int));
1132 	return (error);
1133 }
1134 
1135 /*
1136  * Based on on sysctl_handle_int() convert milliseconds into ticks.
1137  * Note: this is used by TCP.
1138  */
1139 
1140 int
1141 sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
1142 {
1143 	int error, s, tt;
1144 
1145 	tt = *(int *)arg1;
1146 	s = (int)((int64_t)tt * 1000 / hz);
1147 
1148 	error = sysctl_handle_int(oidp, &s, 0, req);
1149 	if (error || !req->newptr)
1150 		return (error);
1151 
1152 	tt = (int)((int64_t)s * hz / 1000);
1153 	if (tt < 1)
1154 		return (EINVAL);
1155 
1156 	*(int *)arg1 = tt;
1157 	return (0);
1158 }
1159 
1160 
1161 /*
1162  * Handle a long, signed or unsigned.
1163  * Two cases:
1164  *     a variable:  point arg1 at it.
1165  *     a constant:  pass it in arg2.
1166  */
1167 
1168 int
1169 sysctl_handle_long(SYSCTL_HANDLER_ARGS)
1170 {
1171 	int error = 0;
1172 	long tmplong;
1173 #ifdef SCTL_MASK32
1174 	int tmpint;
1175 #endif
1176 
1177 	/*
1178 	 * Attempt to get a coherent snapshot by making a copy of the data.
1179 	 */
1180 	if (arg1)
1181 		tmplong = *(long *)arg1;
1182 	else
1183 		tmplong = arg2;
1184 #ifdef SCTL_MASK32
1185 	if (req->flags & SCTL_MASK32) {
1186 		tmpint = tmplong;
1187 		error = SYSCTL_OUT(req, &tmpint, sizeof(int));
1188 	} else
1189 #endif
1190 		error = SYSCTL_OUT(req, &tmplong, sizeof(long));
1191 
1192 	if (error || !req->newptr)
1193 		return (error);
1194 
1195 	if (!arg1)
1196 		error = EPERM;
1197 #ifdef SCTL_MASK32
1198 	else if (req->flags & SCTL_MASK32) {
1199 		error = SYSCTL_IN(req, &tmpint, sizeof(int));
1200 		*(long *)arg1 = (long)tmpint;
1201 	}
1202 #endif
1203 	else
1204 		error = SYSCTL_IN(req, arg1, sizeof(long));
1205 	return (error);
1206 }
1207 
1208 /*
1209  * Handle a 64 bit int, signed or unsigned.
1210  * Two cases:
1211  *     a variable:  point arg1 at it.
1212  *     a constant:  pass it in arg2.
1213  */
1214 int
1215 sysctl_handle_64(SYSCTL_HANDLER_ARGS)
1216 {
1217 	int error = 0;
1218 	uint64_t tmpout;
1219 
1220 	/*
1221 	 * Attempt to get a coherent snapshot by making a copy of the data.
1222 	 */
1223 	if (arg1)
1224 		tmpout = *(uint64_t *)arg1;
1225 	else
1226 		tmpout = arg2;
1227 	error = SYSCTL_OUT(req, &tmpout, sizeof(uint64_t));
1228 
1229 	if (error || !req->newptr)
1230 		return (error);
1231 
1232 	if (!arg1)
1233 		error = EPERM;
1234 	else
1235 		error = SYSCTL_IN(req, arg1, sizeof(uint64_t));
1236 	return (error);
1237 }
1238 
1239 /*
1240  * Handle our generic '\0' terminated 'C' string.
1241  * Two cases:
1242  * 	a variable string:  point arg1 at it, arg2 is max length.
1243  * 	a constant string:  point arg1 at it, arg2 is zero.
1244  */
1245 
1246 int
1247 sysctl_handle_string(SYSCTL_HANDLER_ARGS)
1248 {
1249 	size_t outlen;
1250 	int error = 0, ro_string = 0;
1251 
1252 	/*
1253 	 * A zero-length buffer indicates a fixed size read-only
1254 	 * string:
1255 	 */
1256 	if (arg2 == 0) {
1257 		arg2 = strlen((char *)arg1) + 1;
1258 		ro_string = 1;
1259 	}
1260 
1261 	if (req->oldptr != NULL) {
1262 		char *tmparg;
1263 
1264 		if (ro_string) {
1265 			tmparg = arg1;
1266 		} else {
1267 			/* try to make a coherent snapshot of the string */
1268 			tmparg = malloc(arg2, M_SYSCTLTMP, M_WAITOK);
1269 			memcpy(tmparg, arg1, arg2);
1270 		}
1271 
1272 		outlen = strnlen(tmparg, arg2 - 1) + 1;
1273 		error = SYSCTL_OUT(req, tmparg, outlen);
1274 
1275 		if (!ro_string)
1276 			free(tmparg, M_SYSCTLTMP);
1277 	} else {
1278 		outlen = strnlen((char *)arg1, arg2 - 1) + 1;
1279 		error = SYSCTL_OUT(req, NULL, outlen);
1280 	}
1281 	if (error || !req->newptr)
1282 		return (error);
1283 
1284 	if ((req->newlen - req->newidx) >= arg2) {
1285 		error = EINVAL;
1286 	} else {
1287 		arg2 = (req->newlen - req->newidx);
1288 		error = SYSCTL_IN(req, arg1, arg2);
1289 		((char *)arg1)[arg2] = '\0';
1290 	}
1291 	return (error);
1292 }
1293 
1294 /*
1295  * Handle any kind of opaque data.
1296  * arg1 points to it, arg2 is the size.
1297  */
1298 
1299 int
1300 sysctl_handle_opaque(SYSCTL_HANDLER_ARGS)
1301 {
1302 	int error, tries;
1303 	u_int generation;
1304 	struct sysctl_req req2;
1305 
1306 	/*
1307 	 * Attempt to get a coherent snapshot, by using the thread
1308 	 * pre-emption counter updated from within mi_switch() to
1309 	 * determine if we were pre-empted during a bcopy() or
1310 	 * copyout(). Make 3 attempts at doing this before giving up.
1311 	 * If we encounter an error, stop immediately.
1312 	 */
1313 	tries = 0;
1314 	req2 = *req;
1315 retry:
1316 	generation = curthread->td_generation;
1317 	error = SYSCTL_OUT(req, arg1, arg2);
1318 	if (error)
1319 		return (error);
1320 	tries++;
1321 	if (generation != curthread->td_generation && tries < 3) {
1322 		*req = req2;
1323 		goto retry;
1324 	}
1325 
1326 	error = SYSCTL_IN(req, arg1, arg2);
1327 
1328 	return (error);
1329 }
1330 
1331 /*
1332  * Transfer functions to/from kernel space.
1333  * XXX: rather untested at this point
1334  */
1335 static int
1336 sysctl_old_kernel(struct sysctl_req *req, const void *p, size_t l)
1337 {
1338 	size_t i = 0;
1339 
1340 	if (req->oldptr) {
1341 		i = l;
1342 		if (req->oldlen <= req->oldidx)
1343 			i = 0;
1344 		else
1345 			if (i > req->oldlen - req->oldidx)
1346 				i = req->oldlen - req->oldidx;
1347 		if (i > 0)
1348 			bcopy(p, (char *)req->oldptr + req->oldidx, i);
1349 	}
1350 	req->oldidx += l;
1351 	if (req->oldptr && i != l)
1352 		return (ENOMEM);
1353 	return (0);
1354 }
1355 
1356 static int
1357 sysctl_new_kernel(struct sysctl_req *req, void *p, size_t l)
1358 {
1359 	if (!req->newptr)
1360 		return (0);
1361 	if (req->newlen - req->newidx < l)
1362 		return (EINVAL);
1363 	bcopy((char *)req->newptr + req->newidx, p, l);
1364 	req->newidx += l;
1365 	return (0);
1366 }
1367 
1368 int
1369 kernel_sysctl(struct thread *td, int *name, u_int namelen, void *old,
1370     size_t *oldlenp, void *new, size_t newlen, size_t *retval, int flags)
1371 {
1372 	int error = 0;
1373 	struct sysctl_req req;
1374 
1375 	bzero(&req, sizeof req);
1376 
1377 	req.td = td;
1378 	req.flags = flags;
1379 
1380 	if (oldlenp) {
1381 		req.oldlen = *oldlenp;
1382 	}
1383 	req.validlen = req.oldlen;
1384 
1385 	if (old) {
1386 		req.oldptr= old;
1387 	}
1388 
1389 	if (new != NULL) {
1390 		req.newlen = newlen;
1391 		req.newptr = new;
1392 	}
1393 
1394 	req.oldfunc = sysctl_old_kernel;
1395 	req.newfunc = sysctl_new_kernel;
1396 	req.lock = REQ_UNWIRED;
1397 
1398 	SYSCTL_SLOCK();
1399 	error = sysctl_root(0, name, namelen, &req);
1400 	SYSCTL_SUNLOCK();
1401 
1402 	if (req.lock == REQ_WIRED && req.validlen > 0)
1403 		vsunlock(req.oldptr, req.validlen);
1404 
1405 	if (error && error != ENOMEM)
1406 		return (error);
1407 
1408 	if (retval) {
1409 		if (req.oldptr && req.oldidx > req.validlen)
1410 			*retval = req.validlen;
1411 		else
1412 			*retval = req.oldidx;
1413 	}
1414 	return (error);
1415 }
1416 
1417 int
1418 kernel_sysctlbyname(struct thread *td, char *name, void *old, size_t *oldlenp,
1419     void *new, size_t newlen, size_t *retval, int flags)
1420 {
1421         int oid[CTL_MAXNAME];
1422         size_t oidlen, plen;
1423 	int error;
1424 
1425 	oid[0] = 0;		/* sysctl internal magic */
1426 	oid[1] = 3;		/* name2oid */
1427 	oidlen = sizeof(oid);
1428 
1429 	error = kernel_sysctl(td, oid, 2, oid, &oidlen,
1430 	    (void *)name, strlen(name), &plen, flags);
1431 	if (error)
1432 		return (error);
1433 
1434 	error = kernel_sysctl(td, oid, plen / sizeof(int), old, oldlenp,
1435 	    new, newlen, retval, flags);
1436 	return (error);
1437 }
1438 
1439 /*
1440  * Transfer function to/from user space.
1441  */
1442 static int
1443 sysctl_old_user(struct sysctl_req *req, const void *p, size_t l)
1444 {
1445 	size_t i, len, origidx;
1446 	int error;
1447 
1448 	origidx = req->oldidx;
1449 	req->oldidx += l;
1450 	if (req->oldptr == NULL)
1451 		return (0);
1452 	/*
1453 	 * If we have not wired the user supplied buffer and we are currently
1454 	 * holding locks, drop a witness warning, as it's possible that
1455 	 * write operations to the user page can sleep.
1456 	 */
1457 	if (req->lock != REQ_WIRED)
1458 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
1459 		    "sysctl_old_user()");
1460 	i = l;
1461 	len = req->validlen;
1462 	if (len <= origidx)
1463 		i = 0;
1464 	else {
1465 		if (i > len - origidx)
1466 			i = len - origidx;
1467 		if (req->lock == REQ_WIRED) {
1468 			error = copyout_nofault(p, (char *)req->oldptr +
1469 			    origidx, i);
1470 		} else
1471 			error = copyout(p, (char *)req->oldptr + origidx, i);
1472 		if (error != 0)
1473 			return (error);
1474 	}
1475 	if (i < l)
1476 		return (ENOMEM);
1477 	return (0);
1478 }
1479 
1480 static int
1481 sysctl_new_user(struct sysctl_req *req, void *p, size_t l)
1482 {
1483 	int error;
1484 
1485 	if (!req->newptr)
1486 		return (0);
1487 	if (req->newlen - req->newidx < l)
1488 		return (EINVAL);
1489 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
1490 	    "sysctl_new_user()");
1491 	error = copyin((char *)req->newptr + req->newidx, p, l);
1492 	req->newidx += l;
1493 	return (error);
1494 }
1495 
1496 /*
1497  * Wire the user space destination buffer.  If set to a value greater than
1498  * zero, the len parameter limits the maximum amount of wired memory.
1499  */
1500 int
1501 sysctl_wire_old_buffer(struct sysctl_req *req, size_t len)
1502 {
1503 	int ret;
1504 	size_t wiredlen;
1505 
1506 	wiredlen = (len > 0 && len < req->oldlen) ? len : req->oldlen;
1507 	ret = 0;
1508 	if (req->lock != REQ_WIRED && req->oldptr &&
1509 	    req->oldfunc == sysctl_old_user) {
1510 		if (wiredlen != 0) {
1511 			ret = vslock(req->oldptr, wiredlen);
1512 			if (ret != 0) {
1513 				if (ret != ENOMEM)
1514 					return (ret);
1515 				wiredlen = 0;
1516 			}
1517 		}
1518 		req->lock = REQ_WIRED;
1519 		req->validlen = wiredlen;
1520 	}
1521 	return (0);
1522 }
1523 
1524 int
1525 sysctl_find_oid(int *name, u_int namelen, struct sysctl_oid **noid,
1526     int *nindx, struct sysctl_req *req)
1527 {
1528 	struct sysctl_oid_list *lsp;
1529 	struct sysctl_oid *oid;
1530 	int indx;
1531 
1532 	SYSCTL_ASSERT_LOCKED();
1533 	lsp = &sysctl__children;
1534 	indx = 0;
1535 	while (indx < CTL_MAXNAME) {
1536 		SLIST_FOREACH(oid, lsp, oid_link) {
1537 			if (oid->oid_number == name[indx])
1538 				break;
1539 		}
1540 		if (oid == NULL)
1541 			return (ENOENT);
1542 
1543 		indx++;
1544 		if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1545 			if (oid->oid_handler != NULL || indx == namelen) {
1546 				*noid = oid;
1547 				if (nindx != NULL)
1548 					*nindx = indx;
1549 				KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
1550 				    ("%s found DYING node %p", __func__, oid));
1551 				return (0);
1552 			}
1553 			lsp = SYSCTL_CHILDREN(oid);
1554 		} else if (indx == namelen) {
1555 			*noid = oid;
1556 			if (nindx != NULL)
1557 				*nindx = indx;
1558 			KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
1559 			    ("%s found DYING node %p", __func__, oid));
1560 			return (0);
1561 		} else {
1562 			return (ENOTDIR);
1563 		}
1564 	}
1565 	return (ENOENT);
1566 }
1567 
1568 /*
1569  * Traverse our tree, and find the right node, execute whatever it points
1570  * to, and return the resulting error code.
1571  */
1572 
1573 static int
1574 sysctl_root(SYSCTL_HANDLER_ARGS)
1575 {
1576 	struct sysctl_oid *oid;
1577 	int error, indx, lvl;
1578 
1579 	SYSCTL_ASSERT_SLOCKED();
1580 
1581 	error = sysctl_find_oid(arg1, arg2, &oid, &indx, req);
1582 	if (error)
1583 		return (error);
1584 
1585 	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1586 		/*
1587 		 * You can't call a sysctl when it's a node, but has
1588 		 * no handler.  Inform the user that it's a node.
1589 		 * The indx may or may not be the same as namelen.
1590 		 */
1591 		if (oid->oid_handler == NULL)
1592 			return (EISDIR);
1593 	}
1594 
1595 	/* Is this sysctl writable? */
1596 	if (req->newptr && !(oid->oid_kind & CTLFLAG_WR))
1597 		return (EPERM);
1598 
1599 	KASSERT(req->td != NULL, ("sysctl_root(): req->td == NULL"));
1600 
1601 #ifdef CAPABILITY_MODE
1602 	/*
1603 	 * If the process is in capability mode, then don't permit reading or
1604 	 * writing unless specifically granted for the node.
1605 	 */
1606 	if (IN_CAPABILITY_MODE(req->td)) {
1607 		if (req->oldptr && !(oid->oid_kind & CTLFLAG_CAPRD))
1608 			return (EPERM);
1609 		if (req->newptr && !(oid->oid_kind & CTLFLAG_CAPWR))
1610 			return (EPERM);
1611 	}
1612 #endif
1613 
1614 	/* Is this sysctl sensitive to securelevels? */
1615 	if (req->newptr && (oid->oid_kind & CTLFLAG_SECURE)) {
1616 		lvl = (oid->oid_kind & CTLMASK_SECURE) >> CTLSHIFT_SECURE;
1617 		error = securelevel_gt(req->td->td_ucred, lvl);
1618 		if (error)
1619 			return (error);
1620 	}
1621 
1622 	/* Is this sysctl writable by only privileged users? */
1623 	if (req->newptr && !(oid->oid_kind & CTLFLAG_ANYBODY)) {
1624 		int priv;
1625 
1626 		if (oid->oid_kind & CTLFLAG_PRISON)
1627 			priv = PRIV_SYSCTL_WRITEJAIL;
1628 #ifdef VIMAGE
1629 		else if ((oid->oid_kind & CTLFLAG_VNET) &&
1630 		     prison_owns_vnet(req->td->td_ucred))
1631 			priv = PRIV_SYSCTL_WRITEJAIL;
1632 #endif
1633 		else
1634 			priv = PRIV_SYSCTL_WRITE;
1635 		error = priv_check(req->td, priv);
1636 		if (error)
1637 			return (error);
1638 	}
1639 
1640 	if (!oid->oid_handler)
1641 		return (EINVAL);
1642 
1643 	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1644 		arg1 = (int *)arg1 + indx;
1645 		arg2 -= indx;
1646 	} else {
1647 		arg1 = oid->oid_arg1;
1648 		arg2 = oid->oid_arg2;
1649 	}
1650 #ifdef MAC
1651 	error = mac_system_check_sysctl(req->td->td_ucred, oid, arg1, arg2,
1652 	    req);
1653 	if (error != 0)
1654 		return (error);
1655 #endif
1656 #ifdef VIMAGE
1657 	if ((oid->oid_kind & CTLFLAG_VNET) && arg1 != NULL)
1658 		arg1 = (void *)(curvnet->vnet_data_base + (uintptr_t)arg1);
1659 #endif
1660 	error = sysctl_root_handler_locked(oid, arg1, arg2, req);
1661 
1662 	KFAIL_POINT_ERROR(_debug_fail_point, sysctl_running, error);
1663 
1664 	return (error);
1665 }
1666 
1667 #ifndef _SYS_SYSPROTO_H_
1668 struct sysctl_args {
1669 	int	*name;
1670 	u_int	namelen;
1671 	void	*old;
1672 	size_t	*oldlenp;
1673 	void	*new;
1674 	size_t	newlen;
1675 };
1676 #endif
1677 int
1678 sys___sysctl(struct thread *td, struct sysctl_args *uap)
1679 {
1680 	int error, i, name[CTL_MAXNAME];
1681 	size_t j;
1682 
1683 	if (uap->namelen > CTL_MAXNAME || uap->namelen < 2)
1684 		return (EINVAL);
1685 
1686  	error = copyin(uap->name, &name, uap->namelen * sizeof(int));
1687  	if (error)
1688 		return (error);
1689 
1690 	error = userland_sysctl(td, name, uap->namelen,
1691 		uap->old, uap->oldlenp, 0,
1692 		uap->new, uap->newlen, &j, 0);
1693 	if (error && error != ENOMEM)
1694 		return (error);
1695 	if (uap->oldlenp) {
1696 		i = copyout(&j, uap->oldlenp, sizeof(j));
1697 		if (i)
1698 			return (i);
1699 	}
1700 	return (error);
1701 }
1702 
1703 /*
1704  * This is used from various compatibility syscalls too.  That's why name
1705  * must be in kernel space.
1706  */
1707 int
1708 userland_sysctl(struct thread *td, int *name, u_int namelen, void *old,
1709     size_t *oldlenp, int inkernel, void *new, size_t newlen, size_t *retval,
1710     int flags)
1711 {
1712 	int error = 0, memlocked;
1713 	struct sysctl_req req;
1714 
1715 	bzero(&req, sizeof req);
1716 
1717 	req.td = td;
1718 	req.flags = flags;
1719 
1720 	if (oldlenp) {
1721 		if (inkernel) {
1722 			req.oldlen = *oldlenp;
1723 		} else {
1724 			error = copyin(oldlenp, &req.oldlen, sizeof(*oldlenp));
1725 			if (error)
1726 				return (error);
1727 		}
1728 	}
1729 	req.validlen = req.oldlen;
1730 
1731 	if (old) {
1732 		if (!useracc(old, req.oldlen, VM_PROT_WRITE))
1733 			return (EFAULT);
1734 		req.oldptr= old;
1735 	}
1736 
1737 	if (new != NULL) {
1738 		if (!useracc(new, newlen, VM_PROT_READ))
1739 			return (EFAULT);
1740 		req.newlen = newlen;
1741 		req.newptr = new;
1742 	}
1743 
1744 	req.oldfunc = sysctl_old_user;
1745 	req.newfunc = sysctl_new_user;
1746 	req.lock = REQ_UNWIRED;
1747 
1748 #ifdef KTRACE
1749 	if (KTRPOINT(curthread, KTR_SYSCTL))
1750 		ktrsysctl(name, namelen);
1751 #endif
1752 
1753 	if (req.oldlen > PAGE_SIZE) {
1754 		memlocked = 1;
1755 		sx_xlock(&sysctlmemlock);
1756 	} else
1757 		memlocked = 0;
1758 	CURVNET_SET(TD_TO_VNET(td));
1759 
1760 	for (;;) {
1761 		req.oldidx = 0;
1762 		req.newidx = 0;
1763 		SYSCTL_SLOCK();
1764 		error = sysctl_root(0, name, namelen, &req);
1765 		SYSCTL_SUNLOCK();
1766 		if (error != EAGAIN)
1767 			break;
1768 		kern_yield(PRI_USER);
1769 	}
1770 
1771 	CURVNET_RESTORE();
1772 
1773 	if (req.lock == REQ_WIRED && req.validlen > 0)
1774 		vsunlock(req.oldptr, req.validlen);
1775 	if (memlocked)
1776 		sx_xunlock(&sysctlmemlock);
1777 
1778 	if (error && error != ENOMEM)
1779 		return (error);
1780 
1781 	if (retval) {
1782 		if (req.oldptr && req.oldidx > req.validlen)
1783 			*retval = req.validlen;
1784 		else
1785 			*retval = req.oldidx;
1786 	}
1787 	return (error);
1788 }
1789 
1790 /*
1791  * Drain into a sysctl struct.  The user buffer should be wired if a page
1792  * fault would cause issue.
1793  */
1794 static int
1795 sbuf_sysctl_drain(void *arg, const char *data, int len)
1796 {
1797 	struct sysctl_req *req = arg;
1798 	int error;
1799 
1800 	error = SYSCTL_OUT(req, data, len);
1801 	KASSERT(error >= 0, ("Got unexpected negative value %d", error));
1802 	return (error == 0 ? len : -error);
1803 }
1804 
1805 struct sbuf *
1806 sbuf_new_for_sysctl(struct sbuf *s, char *buf, int length,
1807     struct sysctl_req *req)
1808 {
1809 
1810 	s = sbuf_new(s, buf, length, SBUF_FIXEDLEN);
1811 	sbuf_set_drain(s, sbuf_sysctl_drain, req);
1812 	return (s);
1813 }
1814