xref: /freebsd/sys/kern/kern_sysctl.c (revision 3e5645b78f476816ca3b5acc28b29bbafbb9c444)
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 + rem, 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 	int oid_number;
300 	int timeout = 2;
301 
302 	/*
303 	 * First check if another oid with the same name already
304 	 * exists in the parent's list.
305 	 */
306 	SYSCTL_ASSERT_XLOCKED();
307 	p = sysctl_find_oidname(oidp->oid_name, parent);
308 	if (p != NULL) {
309 		if ((p->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
310 			p->oid_refcnt++;
311 			return;
312 		} else {
313 			printf("can't re-use a leaf (%s)!\n", p->oid_name);
314 			return;
315 		}
316 	}
317 	/* get current OID number */
318 	oid_number = oidp->oid_number;
319 
320 #if (OID_AUTO >= 0)
321 #error "OID_AUTO is expected to be a negative value"
322 #endif
323 	/*
324 	 * Any negative OID number qualifies as OID_AUTO. Valid OID
325 	 * numbers should always be positive.
326 	 *
327 	 * NOTE: DO NOT change the starting value here, change it in
328 	 * <sys/sysctl.h>, and make sure it is at least 256 to
329 	 * accomodate e.g. net.inet.raw as a static sysctl node.
330 	 */
331 	if (oid_number < 0) {
332 		static int newoid;
333 
334 		/*
335 		 * By decrementing the next OID number we spend less
336 		 * time inserting the OIDs into a sorted list.
337 		 */
338 		if (--newoid < CTL_AUTO_START)
339 			newoid = 0x7fffffff;
340 
341 		oid_number = newoid;
342 	}
343 
344 	/*
345 	 * Insert the OID into the parent's list sorted by OID number.
346 	 */
347 retry:
348 	q = NULL;
349 	SLIST_FOREACH(p, parent, oid_link) {
350 		/* check if the current OID number is in use */
351 		if (oid_number == p->oid_number) {
352 			/* get the next valid OID number */
353 			if (oid_number < CTL_AUTO_START ||
354 			    oid_number == 0x7fffffff) {
355 				/* wraparound - restart */
356 				oid_number = CTL_AUTO_START;
357 				/* don't loop forever */
358 				if (!timeout--)
359 					panic("sysctl: Out of OID numbers\n");
360 				goto retry;
361 			} else {
362 				oid_number++;
363 			}
364 		} else if (oid_number < p->oid_number)
365 			break;
366 		q = p;
367 	}
368 	/* check for non-auto OID number collision */
369 	if (oidp->oid_number >= 0 && oidp->oid_number < CTL_AUTO_START &&
370 	    oid_number >= CTL_AUTO_START) {
371 		printf("sysctl: OID number(%d) is already in use for '%s'\n",
372 		    oidp->oid_number, oidp->oid_name);
373 	}
374 	/* update the OID number, if any */
375 	oidp->oid_number = oid_number;
376 	if (q != NULL)
377 		SLIST_INSERT_AFTER(q, oidp, oid_link);
378 	else
379 		SLIST_INSERT_HEAD(parent, oidp, oid_link);
380 
381 	if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE &&
382 #ifdef VIMAGE
383 	    (oidp->oid_kind & CTLFLAG_VNET) == 0 &&
384 #endif
385 	    (oidp->oid_kind & CTLFLAG_TUN) != 0 &&
386 	    (oidp->oid_kind & CTLFLAG_NOFETCH) == 0) {
387 		/* only fetch value once */
388 		oidp->oid_kind |= CTLFLAG_NOFETCH;
389 		/* try to fetch value from kernel environment */
390 		sysctl_load_tunable_by_oid_locked(oidp);
391 	}
392 }
393 
394 void
395 sysctl_unregister_oid(struct sysctl_oid *oidp)
396 {
397 	struct sysctl_oid *p;
398 	int error;
399 
400 	SYSCTL_ASSERT_XLOCKED();
401 	error = ENOENT;
402 	if (oidp->oid_number == OID_AUTO) {
403 		error = EINVAL;
404 	} else {
405 		SLIST_FOREACH(p, oidp->oid_parent, oid_link) {
406 			if (p == oidp) {
407 				SLIST_REMOVE(oidp->oid_parent, oidp,
408 				    sysctl_oid, oid_link);
409 				error = 0;
410 				break;
411 			}
412 		}
413 	}
414 
415 	/*
416 	 * This can happen when a module fails to register and is
417 	 * being unloaded afterwards.  It should not be a panic()
418 	 * for normal use.
419 	 */
420 	if (error)
421 		printf("%s: failed to unregister sysctl\n", __func__);
422 }
423 
424 /* Initialize a new context to keep track of dynamically added sysctls. */
425 int
426 sysctl_ctx_init(struct sysctl_ctx_list *c)
427 {
428 
429 	if (c == NULL) {
430 		return (EINVAL);
431 	}
432 
433 	/*
434 	 * No locking here, the caller is responsible for not adding
435 	 * new nodes to a context until after this function has
436 	 * returned.
437 	 */
438 	TAILQ_INIT(c);
439 	return (0);
440 }
441 
442 /* Free the context, and destroy all dynamic oids registered in this context */
443 int
444 sysctl_ctx_free(struct sysctl_ctx_list *clist)
445 {
446 	struct sysctl_ctx_entry *e, *e1;
447 	int error;
448 
449 	error = 0;
450 	/*
451 	 * First perform a "dry run" to check if it's ok to remove oids.
452 	 * XXX FIXME
453 	 * XXX This algorithm is a hack. But I don't know any
454 	 * XXX better solution for now...
455 	 */
456 	SYSCTL_XLOCK();
457 	TAILQ_FOREACH(e, clist, link) {
458 		error = sysctl_remove_oid_locked(e->entry, 0, 0);
459 		if (error)
460 			break;
461 	}
462 	/*
463 	 * Restore deregistered entries, either from the end,
464 	 * or from the place where error occured.
465 	 * e contains the entry that was not unregistered
466 	 */
467 	if (error)
468 		e1 = TAILQ_PREV(e, sysctl_ctx_list, link);
469 	else
470 		e1 = TAILQ_LAST(clist, sysctl_ctx_list);
471 	while (e1 != NULL) {
472 		sysctl_register_oid(e1->entry);
473 		e1 = TAILQ_PREV(e1, sysctl_ctx_list, link);
474 	}
475 	if (error) {
476 		SYSCTL_XUNLOCK();
477 		return(EBUSY);
478 	}
479 	/* Now really delete the entries */
480 	e = TAILQ_FIRST(clist);
481 	while (e != NULL) {
482 		e1 = TAILQ_NEXT(e, link);
483 		error = sysctl_remove_oid_locked(e->entry, 1, 0);
484 		if (error)
485 			panic("sysctl_remove_oid: corrupt tree, entry: %s",
486 			    e->entry->oid_name);
487 		free(e, M_SYSCTLOID);
488 		e = e1;
489 	}
490 	SYSCTL_XUNLOCK();
491 	return (error);
492 }
493 
494 /* Add an entry to the context */
495 struct sysctl_ctx_entry *
496 sysctl_ctx_entry_add(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
497 {
498 	struct sysctl_ctx_entry *e;
499 
500 	SYSCTL_ASSERT_XLOCKED();
501 	if (clist == NULL || oidp == NULL)
502 		return(NULL);
503 	e = malloc(sizeof(struct sysctl_ctx_entry), M_SYSCTLOID, M_WAITOK);
504 	e->entry = oidp;
505 	TAILQ_INSERT_HEAD(clist, e, link);
506 	return (e);
507 }
508 
509 /* Find an entry in the context */
510 struct sysctl_ctx_entry *
511 sysctl_ctx_entry_find(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
512 {
513 	struct sysctl_ctx_entry *e;
514 
515 	SYSCTL_ASSERT_XLOCKED();
516 	if (clist == NULL || oidp == NULL)
517 		return(NULL);
518 	TAILQ_FOREACH(e, clist, link) {
519 		if(e->entry == oidp)
520 			return(e);
521 	}
522 	return (e);
523 }
524 
525 /*
526  * Delete an entry from the context.
527  * NOTE: this function doesn't free oidp! You have to remove it
528  * with sysctl_remove_oid().
529  */
530 int
531 sysctl_ctx_entry_del(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
532 {
533 	struct sysctl_ctx_entry *e;
534 
535 	if (clist == NULL || oidp == NULL)
536 		return (EINVAL);
537 	SYSCTL_XLOCK();
538 	e = sysctl_ctx_entry_find(clist, oidp);
539 	if (e != NULL) {
540 		TAILQ_REMOVE(clist, e, link);
541 		SYSCTL_XUNLOCK();
542 		free(e, M_SYSCTLOID);
543 		return (0);
544 	} else {
545 		SYSCTL_XUNLOCK();
546 		return (ENOENT);
547 	}
548 }
549 
550 /*
551  * Remove dynamically created sysctl trees.
552  * oidp - top of the tree to be removed
553  * del - if 0 - just deregister, otherwise free up entries as well
554  * recurse - if != 0 traverse the subtree to be deleted
555  */
556 int
557 sysctl_remove_oid(struct sysctl_oid *oidp, int del, int recurse)
558 {
559 	int error;
560 
561 	SYSCTL_XLOCK();
562 	error = sysctl_remove_oid_locked(oidp, del, recurse);
563 	SYSCTL_XUNLOCK();
564 	return (error);
565 }
566 
567 int
568 sysctl_remove_name(struct sysctl_oid *parent, const char *name,
569     int del, int recurse)
570 {
571 	struct sysctl_oid *p, *tmp;
572 	int error;
573 
574 	error = ENOENT;
575 	SYSCTL_XLOCK();
576 	SLIST_FOREACH_SAFE(p, SYSCTL_CHILDREN(parent), oid_link, tmp) {
577 		if (strcmp(p->oid_name, name) == 0) {
578 			error = sysctl_remove_oid_locked(p, del, recurse);
579 			break;
580 		}
581 	}
582 	SYSCTL_XUNLOCK();
583 
584 	return (error);
585 }
586 
587 
588 static int
589 sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del, int recurse)
590 {
591 	struct sysctl_oid *p, *tmp;
592 	int error;
593 
594 	SYSCTL_ASSERT_XLOCKED();
595 	if (oidp == NULL)
596 		return(EINVAL);
597 	if ((oidp->oid_kind & CTLFLAG_DYN) == 0) {
598 		printf("can't remove non-dynamic nodes!\n");
599 		return (EINVAL);
600 	}
601 	/*
602 	 * WARNING: normal method to do this should be through
603 	 * sysctl_ctx_free(). Use recursing as the last resort
604 	 * method to purge your sysctl tree of leftovers...
605 	 * However, if some other code still references these nodes,
606 	 * it will panic.
607 	 */
608 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
609 		if (oidp->oid_refcnt == 1) {
610 			SLIST_FOREACH_SAFE(p,
611 			    SYSCTL_CHILDREN(oidp), oid_link, tmp) {
612 				if (!recurse) {
613 					printf("Warning: failed attempt to "
614 					    "remove oid %s with child %s\n",
615 					    oidp->oid_name, p->oid_name);
616 					return (ENOTEMPTY);
617 				}
618 				error = sysctl_remove_oid_locked(p, del,
619 				    recurse);
620 				if (error)
621 					return (error);
622 			}
623 		}
624 	}
625 	if (oidp->oid_refcnt > 1 ) {
626 		oidp->oid_refcnt--;
627 	} else {
628 		if (oidp->oid_refcnt == 0) {
629 			printf("Warning: bad oid_refcnt=%u (%s)!\n",
630 				oidp->oid_refcnt, oidp->oid_name);
631 			return (EINVAL);
632 		}
633 		sysctl_unregister_oid(oidp);
634 		if (del) {
635 			/*
636 			 * Wait for all threads running the handler to drain.
637 			 * This preserves the previous behavior when the
638 			 * sysctl lock was held across a handler invocation,
639 			 * and is necessary for module unload correctness.
640 			 */
641 			while (oidp->oid_running > 0) {
642 				oidp->oid_kind |= CTLFLAG_DYING;
643 				SYSCTL_SLEEP(&oidp->oid_running, "oidrm", 0);
644 			}
645 			if (oidp->oid_descr)
646 				free(__DECONST(char *, oidp->oid_descr),
647 				    M_SYSCTLOID);
648 			free(__DECONST(char *, oidp->oid_name), M_SYSCTLOID);
649 			free(oidp, M_SYSCTLOID);
650 		}
651 	}
652 	return (0);
653 }
654 /*
655  * Create new sysctls at run time.
656  * clist may point to a valid context initialized with sysctl_ctx_init().
657  */
658 struct sysctl_oid *
659 sysctl_add_oid(struct sysctl_ctx_list *clist, struct sysctl_oid_list *parent,
660 	int number, const char *name, int kind, void *arg1, intptr_t arg2,
661 	int (*handler)(SYSCTL_HANDLER_ARGS), const char *fmt, const char *descr)
662 {
663 	struct sysctl_oid *oidp;
664 
665 	/* You have to hook up somewhere.. */
666 	if (parent == NULL)
667 		return(NULL);
668 	/* Check if the node already exists, otherwise create it */
669 	SYSCTL_XLOCK();
670 	oidp = sysctl_find_oidname(name, parent);
671 	if (oidp != NULL) {
672 		if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
673 			oidp->oid_refcnt++;
674 			/* Update the context */
675 			if (clist != NULL)
676 				sysctl_ctx_entry_add(clist, oidp);
677 			SYSCTL_XUNLOCK();
678 			return (oidp);
679 		} else {
680 			SYSCTL_XUNLOCK();
681 			printf("can't re-use a leaf (%s)!\n", name);
682 			return (NULL);
683 		}
684 	}
685 	oidp = malloc(sizeof(struct sysctl_oid), M_SYSCTLOID, M_WAITOK|M_ZERO);
686 	oidp->oid_parent = parent;
687 	SLIST_INIT(&oidp->oid_children);
688 	oidp->oid_number = number;
689 	oidp->oid_refcnt = 1;
690 	oidp->oid_name = strdup(name, M_SYSCTLOID);
691 	oidp->oid_handler = handler;
692 	oidp->oid_kind = CTLFLAG_DYN | kind;
693 	oidp->oid_arg1 = arg1;
694 	oidp->oid_arg2 = arg2;
695 	oidp->oid_fmt = fmt;
696 	if (descr != NULL)
697 		oidp->oid_descr = strdup(descr, M_SYSCTLOID);
698 	/* Update the context, if used */
699 	if (clist != NULL)
700 		sysctl_ctx_entry_add(clist, oidp);
701 	/* Register this oid */
702 	sysctl_register_oid(oidp);
703 	SYSCTL_XUNLOCK();
704 	return (oidp);
705 }
706 
707 /*
708  * Rename an existing oid.
709  */
710 void
711 sysctl_rename_oid(struct sysctl_oid *oidp, const char *name)
712 {
713 	char *newname;
714 	char *oldname;
715 
716 	newname = strdup(name, M_SYSCTLOID);
717 	SYSCTL_XLOCK();
718 	oldname = __DECONST(char *, oidp->oid_name);
719 	oidp->oid_name = newname;
720 	SYSCTL_XUNLOCK();
721 	free(oldname, M_SYSCTLOID);
722 }
723 
724 /*
725  * Reparent an existing oid.
726  */
727 int
728 sysctl_move_oid(struct sysctl_oid *oid, struct sysctl_oid_list *parent)
729 {
730 	struct sysctl_oid *oidp;
731 
732 	SYSCTL_XLOCK();
733 	if (oid->oid_parent == parent) {
734 		SYSCTL_XUNLOCK();
735 		return (0);
736 	}
737 	oidp = sysctl_find_oidname(oid->oid_name, parent);
738 	if (oidp != NULL) {
739 		SYSCTL_XUNLOCK();
740 		return (EEXIST);
741 	}
742 	sysctl_unregister_oid(oid);
743 	oid->oid_parent = parent;
744 	oid->oid_number = OID_AUTO;
745 	sysctl_register_oid(oid);
746 	SYSCTL_XUNLOCK();
747 	return (0);
748 }
749 
750 /*
751  * Register the kernel's oids on startup.
752  */
753 SET_DECLARE(sysctl_set, struct sysctl_oid);
754 
755 static void
756 sysctl_register_all(void *arg)
757 {
758 	struct sysctl_oid **oidp;
759 
760 	sx_init(&sysctlmemlock, "sysctl mem");
761 	SYSCTL_INIT();
762 	SYSCTL_XLOCK();
763 	SET_FOREACH(oidp, sysctl_set)
764 		sysctl_register_oid(*oidp);
765 	SYSCTL_XUNLOCK();
766 }
767 SYSINIT(sysctl, SI_SUB_KMEM, SI_ORDER_FIRST, sysctl_register_all, 0);
768 
769 /*
770  * "Staff-functions"
771  *
772  * These functions implement a presently undocumented interface
773  * used by the sysctl program to walk the tree, and get the type
774  * so it can print the value.
775  * This interface is under work and consideration, and should probably
776  * be killed with a big axe by the first person who can find the time.
777  * (be aware though, that the proper interface isn't as obvious as it
778  * may seem, there are various conflicting requirements.
779  *
780  * {0,0}	printf the entire MIB-tree.
781  * {0,1,...}	return the name of the "..." OID.
782  * {0,2,...}	return the next OID.
783  * {0,3}	return the OID of the name in "new"
784  * {0,4,...}	return the kind & format info for the "..." OID.
785  * {0,5,...}	return the description the "..." OID.
786  */
787 
788 #ifdef SYSCTL_DEBUG
789 static void
790 sysctl_sysctl_debug_dump_node(struct sysctl_oid_list *l, int i)
791 {
792 	int k;
793 	struct sysctl_oid *oidp;
794 
795 	SYSCTL_ASSERT_LOCKED();
796 	SLIST_FOREACH(oidp, l, oid_link) {
797 
798 		for (k=0; k<i; k++)
799 			printf(" ");
800 
801 		printf("%d %s ", oidp->oid_number, oidp->oid_name);
802 
803 		printf("%c%c",
804 			oidp->oid_kind & CTLFLAG_RD ? 'R':' ',
805 			oidp->oid_kind & CTLFLAG_WR ? 'W':' ');
806 
807 		if (oidp->oid_handler)
808 			printf(" *Handler");
809 
810 		switch (oidp->oid_kind & CTLTYPE) {
811 			case CTLTYPE_NODE:
812 				printf(" Node\n");
813 				if (!oidp->oid_handler) {
814 					sysctl_sysctl_debug_dump_node(
815 					    SYSCTL_CHILDREN(oidp), i + 2);
816 				}
817 				break;
818 			case CTLTYPE_INT:    printf(" Int\n"); break;
819 			case CTLTYPE_UINT:   printf(" u_int\n"); break;
820 			case CTLTYPE_LONG:   printf(" Long\n"); break;
821 			case CTLTYPE_ULONG:  printf(" u_long\n"); break;
822 			case CTLTYPE_STRING: printf(" String\n"); break;
823 			case CTLTYPE_U64:    printf(" uint64_t\n"); break;
824 			case CTLTYPE_S64:    printf(" int64_t\n"); break;
825 			case CTLTYPE_OPAQUE: printf(" Opaque/struct\n"); break;
826 			default:	     printf("\n");
827 		}
828 
829 	}
830 }
831 
832 static int
833 sysctl_sysctl_debug(SYSCTL_HANDLER_ARGS)
834 {
835 	int error;
836 
837 	error = priv_check(req->td, PRIV_SYSCTL_DEBUG);
838 	if (error)
839 		return (error);
840 	SYSCTL_SLOCK();
841 	sysctl_sysctl_debug_dump_node(&sysctl__children, 0);
842 	SYSCTL_SUNLOCK();
843 	return (ENOENT);
844 }
845 
846 SYSCTL_PROC(_sysctl, 0, debug, CTLTYPE_STRING|CTLFLAG_RD|CTLFLAG_MPSAFE,
847 	0, 0, sysctl_sysctl_debug, "-", "");
848 #endif
849 
850 static int
851 sysctl_sysctl_name(SYSCTL_HANDLER_ARGS)
852 {
853 	int *name = (int *) arg1;
854 	u_int namelen = arg2;
855 	int error = 0;
856 	struct sysctl_oid *oid;
857 	struct sysctl_oid_list *lsp = &sysctl__children, *lsp2;
858 	char buf[10];
859 
860 	SYSCTL_SLOCK();
861 	while (namelen) {
862 		if (!lsp) {
863 			snprintf(buf,sizeof(buf),"%d",*name);
864 			if (req->oldidx)
865 				error = SYSCTL_OUT(req, ".", 1);
866 			if (!error)
867 				error = SYSCTL_OUT(req, buf, strlen(buf));
868 			if (error)
869 				goto out;
870 			namelen--;
871 			name++;
872 			continue;
873 		}
874 		lsp2 = 0;
875 		SLIST_FOREACH(oid, lsp, oid_link) {
876 			if (oid->oid_number != *name)
877 				continue;
878 
879 			if (req->oldidx)
880 				error = SYSCTL_OUT(req, ".", 1);
881 			if (!error)
882 				error = SYSCTL_OUT(req, oid->oid_name,
883 					strlen(oid->oid_name));
884 			if (error)
885 				goto out;
886 
887 			namelen--;
888 			name++;
889 
890 			if ((oid->oid_kind & CTLTYPE) != CTLTYPE_NODE)
891 				break;
892 
893 			if (oid->oid_handler)
894 				break;
895 
896 			lsp2 = SYSCTL_CHILDREN(oid);
897 			break;
898 		}
899 		lsp = lsp2;
900 	}
901 	error = SYSCTL_OUT(req, "", 1);
902  out:
903 	SYSCTL_SUNLOCK();
904 	return (error);
905 }
906 
907 /*
908  * XXXRW/JA: Shouldn't return name data for nodes that we don't permit in
909  * capability mode.
910  */
911 static SYSCTL_NODE(_sysctl, 1, name, CTLFLAG_RD | CTLFLAG_MPSAFE | CTLFLAG_CAPRD,
912     sysctl_sysctl_name, "");
913 
914 static int
915 sysctl_sysctl_next_ls(struct sysctl_oid_list *lsp, int *name, u_int namelen,
916 	int *next, int *len, int level, struct sysctl_oid **oidpp)
917 {
918 	struct sysctl_oid *oidp;
919 
920 	SYSCTL_ASSERT_LOCKED();
921 	*len = level;
922 	SLIST_FOREACH(oidp, lsp, oid_link) {
923 		*next = oidp->oid_number;
924 		*oidpp = oidp;
925 
926 		if (oidp->oid_kind & CTLFLAG_SKIP)
927 			continue;
928 
929 		if (!namelen) {
930 			if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
931 				return (0);
932 			if (oidp->oid_handler)
933 				/* We really should call the handler here...*/
934 				return (0);
935 			lsp = SYSCTL_CHILDREN(oidp);
936 			if (!sysctl_sysctl_next_ls(lsp, 0, 0, next+1,
937 				len, level+1, oidpp))
938 				return (0);
939 			goto emptynode;
940 		}
941 
942 		if (oidp->oid_number < *name)
943 			continue;
944 
945 		if (oidp->oid_number > *name) {
946 			if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
947 				return (0);
948 			if (oidp->oid_handler)
949 				return (0);
950 			lsp = SYSCTL_CHILDREN(oidp);
951 			if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1,
952 				next+1, len, level+1, oidpp))
953 				return (0);
954 			goto next;
955 		}
956 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
957 			continue;
958 
959 		if (oidp->oid_handler)
960 			continue;
961 
962 		lsp = SYSCTL_CHILDREN(oidp);
963 		if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1, next+1,
964 			len, level+1, oidpp))
965 			return (0);
966 	next:
967 		namelen = 1;
968 	emptynode:
969 		*len = level;
970 	}
971 	return (1);
972 }
973 
974 static int
975 sysctl_sysctl_next(SYSCTL_HANDLER_ARGS)
976 {
977 	int *name = (int *) arg1;
978 	u_int namelen = arg2;
979 	int i, j, error;
980 	struct sysctl_oid *oid;
981 	struct sysctl_oid_list *lsp = &sysctl__children;
982 	int newoid[CTL_MAXNAME];
983 
984 	SYSCTL_SLOCK();
985 	i = sysctl_sysctl_next_ls(lsp, name, namelen, newoid, &j, 1, &oid);
986 	SYSCTL_SUNLOCK();
987 	if (i)
988 		return (ENOENT);
989 	error = SYSCTL_OUT(req, newoid, j * sizeof (int));
990 	return (error);
991 }
992 
993 /*
994  * XXXRW/JA: Shouldn't return next data for nodes that we don't permit in
995  * capability mode.
996  */
997 static SYSCTL_NODE(_sysctl, 2, next, CTLFLAG_RD | CTLFLAG_MPSAFE | CTLFLAG_CAPRD,
998     sysctl_sysctl_next, "");
999 
1000 static int
1001 name2oid(char *name, int *oid, int *len, struct sysctl_oid **oidpp)
1002 {
1003 	struct sysctl_oid *oidp;
1004 	struct sysctl_oid_list *lsp = &sysctl__children;
1005 	char *p;
1006 
1007 	SYSCTL_ASSERT_LOCKED();
1008 
1009 	for (*len = 0; *len < CTL_MAXNAME;) {
1010 		p = strsep(&name, ".");
1011 
1012 		oidp = SLIST_FIRST(lsp);
1013 		for (;; oidp = SLIST_NEXT(oidp, oid_link)) {
1014 			if (oidp == NULL)
1015 				return (ENOENT);
1016 			if (strcmp(p, oidp->oid_name) == 0)
1017 				break;
1018 		}
1019 		*oid++ = oidp->oid_number;
1020 		(*len)++;
1021 
1022 		if (name == NULL || *name == '\0') {
1023 			if (oidpp)
1024 				*oidpp = oidp;
1025 			return (0);
1026 		}
1027 
1028 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1029 			break;
1030 
1031 		if (oidp->oid_handler)
1032 			break;
1033 
1034 		lsp = SYSCTL_CHILDREN(oidp);
1035 	}
1036 	return (ENOENT);
1037 }
1038 
1039 static int
1040 sysctl_sysctl_name2oid(SYSCTL_HANDLER_ARGS)
1041 {
1042 	char *p;
1043 	int error, oid[CTL_MAXNAME], len = 0;
1044 	struct sysctl_oid *op = 0;
1045 
1046 	if (!req->newlen)
1047 		return (ENOENT);
1048 	if (req->newlen >= MAXPATHLEN)	/* XXX arbitrary, undocumented */
1049 		return (ENAMETOOLONG);
1050 
1051 	p = malloc(req->newlen+1, M_SYSCTL, M_WAITOK);
1052 
1053 	error = SYSCTL_IN(req, p, req->newlen);
1054 	if (error) {
1055 		free(p, M_SYSCTL);
1056 		return (error);
1057 	}
1058 
1059 	p [req->newlen] = '\0';
1060 
1061 	SYSCTL_SLOCK();
1062 	error = name2oid(p, oid, &len, &op);
1063 	SYSCTL_SUNLOCK();
1064 
1065 	free(p, M_SYSCTL);
1066 
1067 	if (error)
1068 		return (error);
1069 
1070 	error = SYSCTL_OUT(req, oid, len * sizeof *oid);
1071 	return (error);
1072 }
1073 
1074 /*
1075  * XXXRW/JA: Shouldn't return name2oid data for nodes that we don't permit in
1076  * capability mode.
1077  */
1078 SYSCTL_PROC(_sysctl, 3, name2oid,
1079     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_MPSAFE
1080     | CTLFLAG_CAPRW, 0, 0, sysctl_sysctl_name2oid, "I", "");
1081 
1082 static int
1083 sysctl_sysctl_oidfmt(SYSCTL_HANDLER_ARGS)
1084 {
1085 	struct sysctl_oid *oid;
1086 	int error;
1087 
1088 	SYSCTL_SLOCK();
1089 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1090 	if (error)
1091 		goto out;
1092 
1093 	if (oid->oid_fmt == NULL) {
1094 		error = ENOENT;
1095 		goto out;
1096 	}
1097 	error = SYSCTL_OUT(req, &oid->oid_kind, sizeof(oid->oid_kind));
1098 	if (error)
1099 		goto out;
1100 	error = SYSCTL_OUT(req, oid->oid_fmt, strlen(oid->oid_fmt) + 1);
1101  out:
1102 	SYSCTL_SUNLOCK();
1103 	return (error);
1104 }
1105 
1106 
1107 static SYSCTL_NODE(_sysctl, 4, oidfmt, CTLFLAG_RD|CTLFLAG_MPSAFE|CTLFLAG_CAPRD,
1108     sysctl_sysctl_oidfmt, "");
1109 
1110 static int
1111 sysctl_sysctl_oiddescr(SYSCTL_HANDLER_ARGS)
1112 {
1113 	struct sysctl_oid *oid;
1114 	int error;
1115 
1116 	SYSCTL_SLOCK();
1117 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1118 	if (error)
1119 		goto out;
1120 
1121 	if (oid->oid_descr == NULL) {
1122 		error = ENOENT;
1123 		goto out;
1124 	}
1125 	error = SYSCTL_OUT(req, oid->oid_descr, strlen(oid->oid_descr) + 1);
1126  out:
1127 	SYSCTL_SUNLOCK();
1128 	return (error);
1129 }
1130 
1131 static SYSCTL_NODE(_sysctl, 5, oiddescr, CTLFLAG_RD|CTLFLAG_MPSAFE|CTLFLAG_CAPRD,
1132     sysctl_sysctl_oiddescr, "");
1133 
1134 /*
1135  * Default "handler" functions.
1136  */
1137 
1138 /*
1139  * Handle an int, signed or unsigned.
1140  * Two cases:
1141  *     a variable:  point arg1 at it.
1142  *     a constant:  pass it in arg2.
1143  */
1144 
1145 int
1146 sysctl_handle_int(SYSCTL_HANDLER_ARGS)
1147 {
1148 	int tmpout, error = 0;
1149 
1150 	/*
1151 	 * Attempt to get a coherent snapshot by making a copy of the data.
1152 	 */
1153 	if (arg1)
1154 		tmpout = *(int *)arg1;
1155 	else
1156 		tmpout = arg2;
1157 	error = SYSCTL_OUT(req, &tmpout, sizeof(int));
1158 
1159 	if (error || !req->newptr)
1160 		return (error);
1161 
1162 	if (!arg1)
1163 		error = EPERM;
1164 	else
1165 		error = SYSCTL_IN(req, arg1, sizeof(int));
1166 	return (error);
1167 }
1168 
1169 /*
1170  * Based on on sysctl_handle_int() convert milliseconds into ticks.
1171  * Note: this is used by TCP.
1172  */
1173 
1174 int
1175 sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
1176 {
1177 	int error, s, tt;
1178 
1179 	tt = *(int *)arg1;
1180 	s = (int)((int64_t)tt * 1000 / hz);
1181 
1182 	error = sysctl_handle_int(oidp, &s, 0, req);
1183 	if (error || !req->newptr)
1184 		return (error);
1185 
1186 	tt = (int)((int64_t)s * hz / 1000);
1187 	if (tt < 1)
1188 		return (EINVAL);
1189 
1190 	*(int *)arg1 = tt;
1191 	return (0);
1192 }
1193 
1194 
1195 /*
1196  * Handle a long, signed or unsigned.
1197  * Two cases:
1198  *     a variable:  point arg1 at it.
1199  *     a constant:  pass it in arg2.
1200  */
1201 
1202 int
1203 sysctl_handle_long(SYSCTL_HANDLER_ARGS)
1204 {
1205 	int error = 0;
1206 	long tmplong;
1207 #ifdef SCTL_MASK32
1208 	int tmpint;
1209 #endif
1210 
1211 	/*
1212 	 * Attempt to get a coherent snapshot by making a copy of the data.
1213 	 */
1214 	if (arg1)
1215 		tmplong = *(long *)arg1;
1216 	else
1217 		tmplong = arg2;
1218 #ifdef SCTL_MASK32
1219 	if (req->flags & SCTL_MASK32) {
1220 		tmpint = tmplong;
1221 		error = SYSCTL_OUT(req, &tmpint, sizeof(int));
1222 	} else
1223 #endif
1224 		error = SYSCTL_OUT(req, &tmplong, sizeof(long));
1225 
1226 	if (error || !req->newptr)
1227 		return (error);
1228 
1229 	if (!arg1)
1230 		error = EPERM;
1231 #ifdef SCTL_MASK32
1232 	else if (req->flags & SCTL_MASK32) {
1233 		error = SYSCTL_IN(req, &tmpint, sizeof(int));
1234 		*(long *)arg1 = (long)tmpint;
1235 	}
1236 #endif
1237 	else
1238 		error = SYSCTL_IN(req, arg1, sizeof(long));
1239 	return (error);
1240 }
1241 
1242 /*
1243  * Handle a 64 bit int, signed or unsigned.
1244  * Two cases:
1245  *     a variable:  point arg1 at it.
1246  *     a constant:  pass it in arg2.
1247  */
1248 int
1249 sysctl_handle_64(SYSCTL_HANDLER_ARGS)
1250 {
1251 	int error = 0;
1252 	uint64_t tmpout;
1253 
1254 	/*
1255 	 * Attempt to get a coherent snapshot by making a copy of the data.
1256 	 */
1257 	if (arg1)
1258 		tmpout = *(uint64_t *)arg1;
1259 	else
1260 		tmpout = arg2;
1261 	error = SYSCTL_OUT(req, &tmpout, sizeof(uint64_t));
1262 
1263 	if (error || !req->newptr)
1264 		return (error);
1265 
1266 	if (!arg1)
1267 		error = EPERM;
1268 	else
1269 		error = SYSCTL_IN(req, arg1, sizeof(uint64_t));
1270 	return (error);
1271 }
1272 
1273 /*
1274  * Handle our generic '\0' terminated 'C' string.
1275  * Two cases:
1276  * 	a variable string:  point arg1 at it, arg2 is max length.
1277  * 	a constant string:  point arg1 at it, arg2 is zero.
1278  */
1279 
1280 int
1281 sysctl_handle_string(SYSCTL_HANDLER_ARGS)
1282 {
1283 	size_t outlen;
1284 	int error = 0, ro_string = 0;
1285 
1286 	/*
1287 	 * A zero-length buffer indicates a fixed size read-only
1288 	 * string:
1289 	 */
1290 	if (arg2 == 0) {
1291 		arg2 = strlen((char *)arg1) + 1;
1292 		ro_string = 1;
1293 	}
1294 
1295 	if (req->oldptr != NULL) {
1296 		char *tmparg;
1297 
1298 		if (ro_string) {
1299 			tmparg = arg1;
1300 		} else {
1301 			/* try to make a coherent snapshot of the string */
1302 			tmparg = malloc(arg2, M_SYSCTLTMP, M_WAITOK);
1303 			memcpy(tmparg, arg1, arg2);
1304 		}
1305 
1306 		outlen = strnlen(tmparg, arg2 - 1) + 1;
1307 		error = SYSCTL_OUT(req, tmparg, outlen);
1308 
1309 		if (!ro_string)
1310 			free(tmparg, M_SYSCTLTMP);
1311 	} else {
1312 		outlen = strnlen((char *)arg1, arg2 - 1) + 1;
1313 		error = SYSCTL_OUT(req, NULL, outlen);
1314 	}
1315 	if (error || !req->newptr)
1316 		return (error);
1317 
1318 	if ((req->newlen - req->newidx) >= arg2) {
1319 		error = EINVAL;
1320 	} else {
1321 		arg2 = (req->newlen - req->newidx);
1322 		error = SYSCTL_IN(req, arg1, arg2);
1323 		((char *)arg1)[arg2] = '\0';
1324 	}
1325 	return (error);
1326 }
1327 
1328 /*
1329  * Handle any kind of opaque data.
1330  * arg1 points to it, arg2 is the size.
1331  */
1332 
1333 int
1334 sysctl_handle_opaque(SYSCTL_HANDLER_ARGS)
1335 {
1336 	int error, tries;
1337 	u_int generation;
1338 	struct sysctl_req req2;
1339 
1340 	/*
1341 	 * Attempt to get a coherent snapshot, by using the thread
1342 	 * pre-emption counter updated from within mi_switch() to
1343 	 * determine if we were pre-empted during a bcopy() or
1344 	 * copyout(). Make 3 attempts at doing this before giving up.
1345 	 * If we encounter an error, stop immediately.
1346 	 */
1347 	tries = 0;
1348 	req2 = *req;
1349 retry:
1350 	generation = curthread->td_generation;
1351 	error = SYSCTL_OUT(req, arg1, arg2);
1352 	if (error)
1353 		return (error);
1354 	tries++;
1355 	if (generation != curthread->td_generation && tries < 3) {
1356 		*req = req2;
1357 		goto retry;
1358 	}
1359 
1360 	error = SYSCTL_IN(req, arg1, arg2);
1361 
1362 	return (error);
1363 }
1364 
1365 /*
1366  * Transfer functions to/from kernel space.
1367  * XXX: rather untested at this point
1368  */
1369 static int
1370 sysctl_old_kernel(struct sysctl_req *req, const void *p, size_t l)
1371 {
1372 	size_t i = 0;
1373 
1374 	if (req->oldptr) {
1375 		i = l;
1376 		if (req->oldlen <= req->oldidx)
1377 			i = 0;
1378 		else
1379 			if (i > req->oldlen - req->oldidx)
1380 				i = req->oldlen - req->oldidx;
1381 		if (i > 0)
1382 			bcopy(p, (char *)req->oldptr + req->oldidx, i);
1383 	}
1384 	req->oldidx += l;
1385 	if (req->oldptr && i != l)
1386 		return (ENOMEM);
1387 	return (0);
1388 }
1389 
1390 static int
1391 sysctl_new_kernel(struct sysctl_req *req, void *p, size_t l)
1392 {
1393 	if (!req->newptr)
1394 		return (0);
1395 	if (req->newlen - req->newidx < l)
1396 		return (EINVAL);
1397 	bcopy((char *)req->newptr + req->newidx, p, l);
1398 	req->newidx += l;
1399 	return (0);
1400 }
1401 
1402 int
1403 kernel_sysctl(struct thread *td, int *name, u_int namelen, void *old,
1404     size_t *oldlenp, void *new, size_t newlen, size_t *retval, int flags)
1405 {
1406 	int error = 0;
1407 	struct sysctl_req req;
1408 
1409 	bzero(&req, sizeof req);
1410 
1411 	req.td = td;
1412 	req.flags = flags;
1413 
1414 	if (oldlenp) {
1415 		req.oldlen = *oldlenp;
1416 	}
1417 	req.validlen = req.oldlen;
1418 
1419 	if (old) {
1420 		req.oldptr= old;
1421 	}
1422 
1423 	if (new != NULL) {
1424 		req.newlen = newlen;
1425 		req.newptr = new;
1426 	}
1427 
1428 	req.oldfunc = sysctl_old_kernel;
1429 	req.newfunc = sysctl_new_kernel;
1430 	req.lock = REQ_UNWIRED;
1431 
1432 	SYSCTL_SLOCK();
1433 	error = sysctl_root(0, name, namelen, &req);
1434 	SYSCTL_SUNLOCK();
1435 
1436 	if (req.lock == REQ_WIRED && req.validlen > 0)
1437 		vsunlock(req.oldptr, req.validlen);
1438 
1439 	if (error && error != ENOMEM)
1440 		return (error);
1441 
1442 	if (retval) {
1443 		if (req.oldptr && req.oldidx > req.validlen)
1444 			*retval = req.validlen;
1445 		else
1446 			*retval = req.oldidx;
1447 	}
1448 	return (error);
1449 }
1450 
1451 int
1452 kernel_sysctlbyname(struct thread *td, char *name, void *old, size_t *oldlenp,
1453     void *new, size_t newlen, size_t *retval, int flags)
1454 {
1455         int oid[CTL_MAXNAME];
1456         size_t oidlen, plen;
1457 	int error;
1458 
1459 	oid[0] = 0;		/* sysctl internal magic */
1460 	oid[1] = 3;		/* name2oid */
1461 	oidlen = sizeof(oid);
1462 
1463 	error = kernel_sysctl(td, oid, 2, oid, &oidlen,
1464 	    (void *)name, strlen(name), &plen, flags);
1465 	if (error)
1466 		return (error);
1467 
1468 	error = kernel_sysctl(td, oid, plen / sizeof(int), old, oldlenp,
1469 	    new, newlen, retval, flags);
1470 	return (error);
1471 }
1472 
1473 /*
1474  * Transfer function to/from user space.
1475  */
1476 static int
1477 sysctl_old_user(struct sysctl_req *req, const void *p, size_t l)
1478 {
1479 	size_t i, len, origidx;
1480 	int error;
1481 
1482 	origidx = req->oldidx;
1483 	req->oldidx += l;
1484 	if (req->oldptr == NULL)
1485 		return (0);
1486 	/*
1487 	 * If we have not wired the user supplied buffer and we are currently
1488 	 * holding locks, drop a witness warning, as it's possible that
1489 	 * write operations to the user page can sleep.
1490 	 */
1491 	if (req->lock != REQ_WIRED)
1492 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
1493 		    "sysctl_old_user()");
1494 	i = l;
1495 	len = req->validlen;
1496 	if (len <= origidx)
1497 		i = 0;
1498 	else {
1499 		if (i > len - origidx)
1500 			i = len - origidx;
1501 		if (req->lock == REQ_WIRED) {
1502 			error = copyout_nofault(p, (char *)req->oldptr +
1503 			    origidx, i);
1504 		} else
1505 			error = copyout(p, (char *)req->oldptr + origidx, i);
1506 		if (error != 0)
1507 			return (error);
1508 	}
1509 	if (i < l)
1510 		return (ENOMEM);
1511 	return (0);
1512 }
1513 
1514 static int
1515 sysctl_new_user(struct sysctl_req *req, void *p, size_t l)
1516 {
1517 	int error;
1518 
1519 	if (!req->newptr)
1520 		return (0);
1521 	if (req->newlen - req->newidx < l)
1522 		return (EINVAL);
1523 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
1524 	    "sysctl_new_user()");
1525 	error = copyin((char *)req->newptr + req->newidx, p, l);
1526 	req->newidx += l;
1527 	return (error);
1528 }
1529 
1530 /*
1531  * Wire the user space destination buffer.  If set to a value greater than
1532  * zero, the len parameter limits the maximum amount of wired memory.
1533  */
1534 int
1535 sysctl_wire_old_buffer(struct sysctl_req *req, size_t len)
1536 {
1537 	int ret;
1538 	size_t wiredlen;
1539 
1540 	wiredlen = (len > 0 && len < req->oldlen) ? len : req->oldlen;
1541 	ret = 0;
1542 	if (req->lock != REQ_WIRED && req->oldptr &&
1543 	    req->oldfunc == sysctl_old_user) {
1544 		if (wiredlen != 0) {
1545 			ret = vslock(req->oldptr, wiredlen);
1546 			if (ret != 0) {
1547 				if (ret != ENOMEM)
1548 					return (ret);
1549 				wiredlen = 0;
1550 			}
1551 		}
1552 		req->lock = REQ_WIRED;
1553 		req->validlen = wiredlen;
1554 	}
1555 	return (0);
1556 }
1557 
1558 int
1559 sysctl_find_oid(int *name, u_int namelen, struct sysctl_oid **noid,
1560     int *nindx, struct sysctl_req *req)
1561 {
1562 	struct sysctl_oid_list *lsp;
1563 	struct sysctl_oid *oid;
1564 	int indx;
1565 
1566 	SYSCTL_ASSERT_LOCKED();
1567 	lsp = &sysctl__children;
1568 	indx = 0;
1569 	while (indx < CTL_MAXNAME) {
1570 		SLIST_FOREACH(oid, lsp, oid_link) {
1571 			if (oid->oid_number == name[indx])
1572 				break;
1573 		}
1574 		if (oid == NULL)
1575 			return (ENOENT);
1576 
1577 		indx++;
1578 		if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1579 			if (oid->oid_handler != NULL || indx == namelen) {
1580 				*noid = oid;
1581 				if (nindx != NULL)
1582 					*nindx = indx;
1583 				KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
1584 				    ("%s found DYING node %p", __func__, oid));
1585 				return (0);
1586 			}
1587 			lsp = SYSCTL_CHILDREN(oid);
1588 		} else if (indx == namelen) {
1589 			*noid = oid;
1590 			if (nindx != NULL)
1591 				*nindx = indx;
1592 			KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
1593 			    ("%s found DYING node %p", __func__, oid));
1594 			return (0);
1595 		} else {
1596 			return (ENOTDIR);
1597 		}
1598 	}
1599 	return (ENOENT);
1600 }
1601 
1602 /*
1603  * Traverse our tree, and find the right node, execute whatever it points
1604  * to, and return the resulting error code.
1605  */
1606 
1607 static int
1608 sysctl_root(SYSCTL_HANDLER_ARGS)
1609 {
1610 	struct sysctl_oid *oid;
1611 	int error, indx, lvl;
1612 
1613 	SYSCTL_ASSERT_SLOCKED();
1614 
1615 	error = sysctl_find_oid(arg1, arg2, &oid, &indx, req);
1616 	if (error)
1617 		return (error);
1618 
1619 	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1620 		/*
1621 		 * You can't call a sysctl when it's a node, but has
1622 		 * no handler.  Inform the user that it's a node.
1623 		 * The indx may or may not be the same as namelen.
1624 		 */
1625 		if (oid->oid_handler == NULL)
1626 			return (EISDIR);
1627 	}
1628 
1629 	/* Is this sysctl writable? */
1630 	if (req->newptr && !(oid->oid_kind & CTLFLAG_WR))
1631 		return (EPERM);
1632 
1633 	KASSERT(req->td != NULL, ("sysctl_root(): req->td == NULL"));
1634 
1635 #ifdef CAPABILITY_MODE
1636 	/*
1637 	 * If the process is in capability mode, then don't permit reading or
1638 	 * writing unless specifically granted for the node.
1639 	 */
1640 	if (IN_CAPABILITY_MODE(req->td)) {
1641 		if (req->oldptr && !(oid->oid_kind & CTLFLAG_CAPRD))
1642 			return (EPERM);
1643 		if (req->newptr && !(oid->oid_kind & CTLFLAG_CAPWR))
1644 			return (EPERM);
1645 	}
1646 #endif
1647 
1648 	/* Is this sysctl sensitive to securelevels? */
1649 	if (req->newptr && (oid->oid_kind & CTLFLAG_SECURE)) {
1650 		lvl = (oid->oid_kind & CTLMASK_SECURE) >> CTLSHIFT_SECURE;
1651 		error = securelevel_gt(req->td->td_ucred, lvl);
1652 		if (error)
1653 			return (error);
1654 	}
1655 
1656 	/* Is this sysctl writable by only privileged users? */
1657 	if (req->newptr && !(oid->oid_kind & CTLFLAG_ANYBODY)) {
1658 		int priv;
1659 
1660 		if (oid->oid_kind & CTLFLAG_PRISON)
1661 			priv = PRIV_SYSCTL_WRITEJAIL;
1662 #ifdef VIMAGE
1663 		else if ((oid->oid_kind & CTLFLAG_VNET) &&
1664 		     prison_owns_vnet(req->td->td_ucred))
1665 			priv = PRIV_SYSCTL_WRITEJAIL;
1666 #endif
1667 		else
1668 			priv = PRIV_SYSCTL_WRITE;
1669 		error = priv_check(req->td, priv);
1670 		if (error)
1671 			return (error);
1672 	}
1673 
1674 	if (!oid->oid_handler)
1675 		return (EINVAL);
1676 
1677 	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1678 		arg1 = (int *)arg1 + indx;
1679 		arg2 -= indx;
1680 	} else {
1681 		arg1 = oid->oid_arg1;
1682 		arg2 = oid->oid_arg2;
1683 	}
1684 #ifdef MAC
1685 	error = mac_system_check_sysctl(req->td->td_ucred, oid, arg1, arg2,
1686 	    req);
1687 	if (error != 0)
1688 		return (error);
1689 #endif
1690 #ifdef VIMAGE
1691 	if ((oid->oid_kind & CTLFLAG_VNET) && arg1 != NULL)
1692 		arg1 = (void *)(curvnet->vnet_data_base + (uintptr_t)arg1);
1693 #endif
1694 	error = sysctl_root_handler_locked(oid, arg1, arg2, req);
1695 
1696 	KFAIL_POINT_ERROR(_debug_fail_point, sysctl_running, error);
1697 
1698 	return (error);
1699 }
1700 
1701 #ifndef _SYS_SYSPROTO_H_
1702 struct sysctl_args {
1703 	int	*name;
1704 	u_int	namelen;
1705 	void	*old;
1706 	size_t	*oldlenp;
1707 	void	*new;
1708 	size_t	newlen;
1709 };
1710 #endif
1711 int
1712 sys___sysctl(struct thread *td, struct sysctl_args *uap)
1713 {
1714 	int error, i, name[CTL_MAXNAME];
1715 	size_t j;
1716 
1717 	if (uap->namelen > CTL_MAXNAME || uap->namelen < 2)
1718 		return (EINVAL);
1719 
1720  	error = copyin(uap->name, &name, uap->namelen * sizeof(int));
1721  	if (error)
1722 		return (error);
1723 
1724 	error = userland_sysctl(td, name, uap->namelen,
1725 		uap->old, uap->oldlenp, 0,
1726 		uap->new, uap->newlen, &j, 0);
1727 	if (error && error != ENOMEM)
1728 		return (error);
1729 	if (uap->oldlenp) {
1730 		i = copyout(&j, uap->oldlenp, sizeof(j));
1731 		if (i)
1732 			return (i);
1733 	}
1734 	return (error);
1735 }
1736 
1737 /*
1738  * This is used from various compatibility syscalls too.  That's why name
1739  * must be in kernel space.
1740  */
1741 int
1742 userland_sysctl(struct thread *td, int *name, u_int namelen, void *old,
1743     size_t *oldlenp, int inkernel, void *new, size_t newlen, size_t *retval,
1744     int flags)
1745 {
1746 	int error = 0, memlocked;
1747 	struct sysctl_req req;
1748 
1749 	bzero(&req, sizeof req);
1750 
1751 	req.td = td;
1752 	req.flags = flags;
1753 
1754 	if (oldlenp) {
1755 		if (inkernel) {
1756 			req.oldlen = *oldlenp;
1757 		} else {
1758 			error = copyin(oldlenp, &req.oldlen, sizeof(*oldlenp));
1759 			if (error)
1760 				return (error);
1761 		}
1762 	}
1763 	req.validlen = req.oldlen;
1764 
1765 	if (old) {
1766 		if (!useracc(old, req.oldlen, VM_PROT_WRITE))
1767 			return (EFAULT);
1768 		req.oldptr= old;
1769 	}
1770 
1771 	if (new != NULL) {
1772 		if (!useracc(new, newlen, VM_PROT_READ))
1773 			return (EFAULT);
1774 		req.newlen = newlen;
1775 		req.newptr = new;
1776 	}
1777 
1778 	req.oldfunc = sysctl_old_user;
1779 	req.newfunc = sysctl_new_user;
1780 	req.lock = REQ_UNWIRED;
1781 
1782 #ifdef KTRACE
1783 	if (KTRPOINT(curthread, KTR_SYSCTL))
1784 		ktrsysctl(name, namelen);
1785 #endif
1786 
1787 	if (req.oldlen > PAGE_SIZE) {
1788 		memlocked = 1;
1789 		sx_xlock(&sysctlmemlock);
1790 	} else
1791 		memlocked = 0;
1792 	CURVNET_SET(TD_TO_VNET(td));
1793 
1794 	for (;;) {
1795 		req.oldidx = 0;
1796 		req.newidx = 0;
1797 		SYSCTL_SLOCK();
1798 		error = sysctl_root(0, name, namelen, &req);
1799 		SYSCTL_SUNLOCK();
1800 		if (error != EAGAIN)
1801 			break;
1802 		kern_yield(PRI_USER);
1803 	}
1804 
1805 	CURVNET_RESTORE();
1806 
1807 	if (req.lock == REQ_WIRED && req.validlen > 0)
1808 		vsunlock(req.oldptr, req.validlen);
1809 	if (memlocked)
1810 		sx_xunlock(&sysctlmemlock);
1811 
1812 	if (error && error != ENOMEM)
1813 		return (error);
1814 
1815 	if (retval) {
1816 		if (req.oldptr && req.oldidx > req.validlen)
1817 			*retval = req.validlen;
1818 		else
1819 			*retval = req.oldidx;
1820 	}
1821 	return (error);
1822 }
1823 
1824 /*
1825  * Drain into a sysctl struct.  The user buffer should be wired if a page
1826  * fault would cause issue.
1827  */
1828 static int
1829 sbuf_sysctl_drain(void *arg, const char *data, int len)
1830 {
1831 	struct sysctl_req *req = arg;
1832 	int error;
1833 
1834 	error = SYSCTL_OUT(req, data, len);
1835 	KASSERT(error >= 0, ("Got unexpected negative value %d", error));
1836 	return (error == 0 ? len : -error);
1837 }
1838 
1839 struct sbuf *
1840 sbuf_new_for_sysctl(struct sbuf *s, char *buf, int length,
1841     struct sysctl_req *req)
1842 {
1843 
1844 	/* Supply a default buffer size if none given. */
1845 	if (buf == NULL && length == 0)
1846 		length = 64;
1847 	s = sbuf_new(s, buf, length, SBUF_FIXEDLEN | SBUF_INCLUDENUL);
1848 	sbuf_set_drain(s, sbuf_sysctl_drain, req);
1849 	return (s);
1850 }
1851