xref: /freebsd/sys/kern/kern_sysctl.c (revision b2d48be1bc7df45ddd13b143a160d0acb5a383c5)
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/rmlock.h>
58 #include <sys/sbuf.h>
59 #include <sys/sx.h>
60 #include <sys/sysproto.h>
61 #include <sys/uio.h>
62 #ifdef KTRACE
63 #include <sys/ktrace.h>
64 #endif
65 
66 #include <net/vnet.h>
67 
68 #include <security/mac/mac_framework.h>
69 
70 #include <vm/vm.h>
71 #include <vm/vm_extern.h>
72 
73 static MALLOC_DEFINE(M_SYSCTL, "sysctl", "sysctl internal magic");
74 static MALLOC_DEFINE(M_SYSCTLOID, "sysctloid", "sysctl dynamic oids");
75 static MALLOC_DEFINE(M_SYSCTLTMP, "sysctltmp", "sysctl temp output buffer");
76 
77 /*
78  * The sysctllock protects the MIB tree.  It also protects sysctl
79  * contexts used with dynamic sysctls.  The sysctl_register_oid() and
80  * sysctl_unregister_oid() routines require the sysctllock to already
81  * be held, so the sysctl_wlock() and sysctl_wunlock() routines are
82  * provided for the few places in the kernel which need to use that
83  * API rather than using the dynamic API.  Use of the dynamic API is
84  * strongly encouraged for most code.
85  *
86  * The sysctlmemlock is used to limit the amount of user memory wired for
87  * sysctl requests.  This is implemented by serializing any userland
88  * sysctl requests larger than a single page via an exclusive lock.
89  */
90 static struct rmlock sysctllock;
91 static struct sx sysctlmemlock;
92 
93 #define	SYSCTL_WLOCK()		rm_wlock(&sysctllock)
94 #define	SYSCTL_WUNLOCK()	rm_wunlock(&sysctllock)
95 #define	SYSCTL_RLOCK(tracker)	rm_rlock(&sysctllock, (tracker))
96 #define	SYSCTL_RUNLOCK(tracker)	rm_runlock(&sysctllock, (tracker))
97 #define	SYSCTL_WLOCKED()	rm_wowned(&sysctllock)
98 #define	SYSCTL_ASSERT_LOCKED()	rm_assert(&sysctllock, RA_LOCKED)
99 #define	SYSCTL_ASSERT_WLOCKED()	rm_assert(&sysctllock, RA_WLOCKED)
100 #define	SYSCTL_ASSERT_RLOCKED()	rm_assert(&sysctllock, RA_RLOCKED)
101 #define	SYSCTL_INIT()		rm_init_flags(&sysctllock, "sysctl lock", \
102 				    RM_SLEEPABLE)
103 #define	SYSCTL_SLEEP(ch, wmesg, timo)					\
104 				rm_sleep(ch, &sysctllock, 0, wmesg, timo)
105 
106 static int sysctl_root(SYSCTL_HANDLER_ARGS);
107 
108 /* Root list */
109 struct sysctl_oid_list sysctl__children = SLIST_HEAD_INITIALIZER(&sysctl__children);
110 
111 static int	sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del,
112 		    int recurse);
113 static int	sysctl_old_kernel(struct sysctl_req *, const void *, size_t);
114 static int	sysctl_new_kernel(struct sysctl_req *, void *, size_t);
115 
116 static struct sysctl_oid *
117 sysctl_find_oidname(const char *name, struct sysctl_oid_list *list)
118 {
119 	struct sysctl_oid *oidp;
120 
121 	SYSCTL_ASSERT_LOCKED();
122 	SLIST_FOREACH(oidp, list, oid_link) {
123 		if (strcmp(oidp->oid_name, name) == 0) {
124 			return (oidp);
125 		}
126 	}
127 	return (NULL);
128 }
129 
130 /*
131  * Initialization of the MIB tree.
132  *
133  * Order by number in each list.
134  */
135 void
136 sysctl_wlock(void)
137 {
138 
139 	SYSCTL_WLOCK();
140 }
141 
142 void
143 sysctl_wunlock(void)
144 {
145 
146 	SYSCTL_WUNLOCK();
147 }
148 
149 static int
150 sysctl_root_handler_locked(struct sysctl_oid *oid, void *arg1, intptr_t arg2,
151     struct sysctl_req *req, struct rm_priotracker *tracker)
152 {
153 	int error;
154 
155 	if (oid->oid_kind & CTLFLAG_DYN)
156 		atomic_add_int(&oid->oid_running, 1);
157 
158 	if (tracker != NULL)
159 		SYSCTL_RUNLOCK(tracker);
160 	else
161 		SYSCTL_WUNLOCK();
162 
163 	if (!(oid->oid_kind & CTLFLAG_MPSAFE))
164 		mtx_lock(&Giant);
165 	error = oid->oid_handler(oid, arg1, arg2, req);
166 	if (!(oid->oid_kind & CTLFLAG_MPSAFE))
167 		mtx_unlock(&Giant);
168 
169 	if (tracker != NULL)
170 		SYSCTL_RLOCK(tracker);
171 	else
172 		SYSCTL_WLOCK();
173 
174 	if (oid->oid_kind & CTLFLAG_DYN) {
175 		if (atomic_fetchadd_int(&oid->oid_running, -1) == 1 &&
176 		    (oid->oid_kind & CTLFLAG_DYING) != 0)
177 			wakeup(&oid->oid_running);
178 	}
179 
180 	return (error);
181 }
182 
183 static void
184 sysctl_load_tunable_by_oid_locked(struct sysctl_oid *oidp)
185 {
186 	struct sysctl_req req;
187 	struct sysctl_oid *curr;
188 	char *penv = NULL;
189 	char path[64];
190 	ssize_t rem = sizeof(path);
191 	ssize_t len;
192 	int val_int;
193 	long val_long;
194 	int64_t val_64;
195 	quad_t val_quad;
196 	int error;
197 
198 	path[--rem] = 0;
199 
200 	for (curr = oidp; curr != NULL; curr = SYSCTL_PARENT(curr)) {
201 		len = strlen(curr->oid_name);
202 		rem -= len;
203 		if (curr != oidp)
204 			rem -= 1;
205 		if (rem < 0) {
206 			printf("OID path exceeds %d bytes\n", (int)sizeof(path));
207 			return;
208 		}
209 		memcpy(path + rem, curr->oid_name, len);
210 		if (curr != oidp)
211 			path[rem + len] = '.';
212 	}
213 
214 	memset(&req, 0, sizeof(req));
215 
216 	req.td = curthread;
217 	req.oldfunc = sysctl_old_kernel;
218 	req.newfunc = sysctl_new_kernel;
219 	req.lock = REQ_UNWIRED;
220 
221 	switch (oidp->oid_kind & CTLTYPE) {
222 	case CTLTYPE_INT:
223 		if (getenv_int(path + rem, &val_int) == 0)
224 			return;
225 		req.newlen = sizeof(val_int);
226 		req.newptr = &val_int;
227 		break;
228 	case CTLTYPE_UINT:
229 		if (getenv_uint(path + rem, (unsigned int *)&val_int) == 0)
230 			return;
231 		req.newlen = sizeof(val_int);
232 		req.newptr = &val_int;
233 		break;
234 	case CTLTYPE_LONG:
235 		if (getenv_long(path + rem, &val_long) == 0)
236 			return;
237 		req.newlen = sizeof(val_long);
238 		req.newptr = &val_long;
239 		break;
240 	case CTLTYPE_ULONG:
241 		if (getenv_ulong(path + rem, (unsigned long *)&val_long) == 0)
242 			return;
243 		req.newlen = sizeof(val_long);
244 		req.newptr = &val_long;
245 		break;
246 	case CTLTYPE_S64:
247 		if (getenv_quad(path + rem, &val_quad) == 0)
248 			return;
249 		val_64 = val_quad;
250 		req.newlen = sizeof(val_64);
251 		req.newptr = &val_64;
252 		break;
253 	case CTLTYPE_U64:
254 		/* XXX there is no getenv_uquad() */
255 		if (getenv_quad(path + rem, &val_quad) == 0)
256 			return;
257 		val_64 = val_quad;
258 		req.newlen = sizeof(val_64);
259 		req.newptr = &val_64;
260 		break;
261 	case CTLTYPE_STRING:
262 		penv = kern_getenv(path + rem);
263 		if (penv == NULL)
264 			return;
265 		req.newlen = strlen(penv);
266 		req.newptr = penv;
267 		break;
268 	default:
269 		return;
270 	}
271 	error = sysctl_root_handler_locked(oidp, oidp->oid_arg1,
272 	    oidp->oid_arg2, &req, NULL);
273 	if (error != 0)
274 		printf("Setting sysctl %s failed: %d\n", path + rem, error);
275 	if (penv != NULL)
276 		freeenv(penv);
277 }
278 
279 void
280 sysctl_register_oid(struct sysctl_oid *oidp)
281 {
282 	struct sysctl_oid_list *parent = oidp->oid_parent;
283 	struct sysctl_oid *p;
284 	struct sysctl_oid *q;
285 	int oid_number;
286 	int timeout = 2;
287 
288 	/*
289 	 * First check if another oid with the same name already
290 	 * exists in the parent's list.
291 	 */
292 	SYSCTL_ASSERT_WLOCKED();
293 	p = sysctl_find_oidname(oidp->oid_name, parent);
294 	if (p != NULL) {
295 		if ((p->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
296 			p->oid_refcnt++;
297 			return;
298 		} else {
299 			printf("can't re-use a leaf (%s)!\n", p->oid_name);
300 			return;
301 		}
302 	}
303 	/* get current OID number */
304 	oid_number = oidp->oid_number;
305 
306 #if (OID_AUTO >= 0)
307 #error "OID_AUTO is expected to be a negative value"
308 #endif
309 	/*
310 	 * Any negative OID number qualifies as OID_AUTO. Valid OID
311 	 * numbers should always be positive.
312 	 *
313 	 * NOTE: DO NOT change the starting value here, change it in
314 	 * <sys/sysctl.h>, and make sure it is at least 256 to
315 	 * accomodate e.g. net.inet.raw as a static sysctl node.
316 	 */
317 	if (oid_number < 0) {
318 		static int newoid;
319 
320 		/*
321 		 * By decrementing the next OID number we spend less
322 		 * time inserting the OIDs into a sorted list.
323 		 */
324 		if (--newoid < CTL_AUTO_START)
325 			newoid = 0x7fffffff;
326 
327 		oid_number = newoid;
328 	}
329 
330 	/*
331 	 * Insert the OID into the parent's list sorted by OID number.
332 	 */
333 retry:
334 	q = NULL;
335 	SLIST_FOREACH(p, parent, oid_link) {
336 		/* check if the current OID number is in use */
337 		if (oid_number == p->oid_number) {
338 			/* get the next valid OID number */
339 			if (oid_number < CTL_AUTO_START ||
340 			    oid_number == 0x7fffffff) {
341 				/* wraparound - restart */
342 				oid_number = CTL_AUTO_START;
343 				/* don't loop forever */
344 				if (!timeout--)
345 					panic("sysctl: Out of OID numbers\n");
346 				goto retry;
347 			} else {
348 				oid_number++;
349 			}
350 		} else if (oid_number < p->oid_number)
351 			break;
352 		q = p;
353 	}
354 	/* check for non-auto OID number collision */
355 	if (oidp->oid_number >= 0 && oidp->oid_number < CTL_AUTO_START &&
356 	    oid_number >= CTL_AUTO_START) {
357 		printf("sysctl: OID number(%d) is already in use for '%s'\n",
358 		    oidp->oid_number, oidp->oid_name);
359 	}
360 	/* update the OID number, if any */
361 	oidp->oid_number = oid_number;
362 	if (q != NULL)
363 		SLIST_INSERT_AFTER(q, oidp, oid_link);
364 	else
365 		SLIST_INSERT_HEAD(parent, oidp, oid_link);
366 
367 	if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE &&
368 #ifdef VIMAGE
369 	    (oidp->oid_kind & CTLFLAG_VNET) == 0 &&
370 #endif
371 	    (oidp->oid_kind & CTLFLAG_TUN) != 0 &&
372 	    (oidp->oid_kind & CTLFLAG_NOFETCH) == 0) {
373 		/* only fetch value once */
374 		oidp->oid_kind |= CTLFLAG_NOFETCH;
375 		/* try to fetch value from kernel environment */
376 		sysctl_load_tunable_by_oid_locked(oidp);
377 	}
378 }
379 
380 void
381 sysctl_unregister_oid(struct sysctl_oid *oidp)
382 {
383 	struct sysctl_oid *p;
384 	int error;
385 
386 	SYSCTL_ASSERT_WLOCKED();
387 	error = ENOENT;
388 	if (oidp->oid_number == OID_AUTO) {
389 		error = EINVAL;
390 	} else {
391 		SLIST_FOREACH(p, oidp->oid_parent, oid_link) {
392 			if (p == oidp) {
393 				SLIST_REMOVE(oidp->oid_parent, oidp,
394 				    sysctl_oid, oid_link);
395 				error = 0;
396 				break;
397 			}
398 		}
399 	}
400 
401 	/*
402 	 * This can happen when a module fails to register and is
403 	 * being unloaded afterwards.  It should not be a panic()
404 	 * for normal use.
405 	 */
406 	if (error)
407 		printf("%s: failed to unregister sysctl\n", __func__);
408 }
409 
410 /* Initialize a new context to keep track of dynamically added sysctls. */
411 int
412 sysctl_ctx_init(struct sysctl_ctx_list *c)
413 {
414 
415 	if (c == NULL) {
416 		return (EINVAL);
417 	}
418 
419 	/*
420 	 * No locking here, the caller is responsible for not adding
421 	 * new nodes to a context until after this function has
422 	 * returned.
423 	 */
424 	TAILQ_INIT(c);
425 	return (0);
426 }
427 
428 /* Free the context, and destroy all dynamic oids registered in this context */
429 int
430 sysctl_ctx_free(struct sysctl_ctx_list *clist)
431 {
432 	struct sysctl_ctx_entry *e, *e1;
433 	int error;
434 
435 	error = 0;
436 	/*
437 	 * First perform a "dry run" to check if it's ok to remove oids.
438 	 * XXX FIXME
439 	 * XXX This algorithm is a hack. But I don't know any
440 	 * XXX better solution for now...
441 	 */
442 	SYSCTL_WLOCK();
443 	TAILQ_FOREACH(e, clist, link) {
444 		error = sysctl_remove_oid_locked(e->entry, 0, 0);
445 		if (error)
446 			break;
447 	}
448 	/*
449 	 * Restore deregistered entries, either from the end,
450 	 * or from the place where error occured.
451 	 * e contains the entry that was not unregistered
452 	 */
453 	if (error)
454 		e1 = TAILQ_PREV(e, sysctl_ctx_list, link);
455 	else
456 		e1 = TAILQ_LAST(clist, sysctl_ctx_list);
457 	while (e1 != NULL) {
458 		sysctl_register_oid(e1->entry);
459 		e1 = TAILQ_PREV(e1, sysctl_ctx_list, link);
460 	}
461 	if (error) {
462 		SYSCTL_WUNLOCK();
463 		return(EBUSY);
464 	}
465 	/* Now really delete the entries */
466 	e = TAILQ_FIRST(clist);
467 	while (e != NULL) {
468 		e1 = TAILQ_NEXT(e, link);
469 		error = sysctl_remove_oid_locked(e->entry, 1, 0);
470 		if (error)
471 			panic("sysctl_remove_oid: corrupt tree, entry: %s",
472 			    e->entry->oid_name);
473 		free(e, M_SYSCTLOID);
474 		e = e1;
475 	}
476 	SYSCTL_WUNLOCK();
477 	return (error);
478 }
479 
480 /* Add an entry to the context */
481 struct sysctl_ctx_entry *
482 sysctl_ctx_entry_add(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
483 {
484 	struct sysctl_ctx_entry *e;
485 
486 	SYSCTL_ASSERT_WLOCKED();
487 	if (clist == NULL || oidp == NULL)
488 		return(NULL);
489 	e = malloc(sizeof(struct sysctl_ctx_entry), M_SYSCTLOID, M_WAITOK);
490 	e->entry = oidp;
491 	TAILQ_INSERT_HEAD(clist, e, link);
492 	return (e);
493 }
494 
495 /* Find an entry in the context */
496 struct sysctl_ctx_entry *
497 sysctl_ctx_entry_find(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
498 {
499 	struct sysctl_ctx_entry *e;
500 
501 	SYSCTL_ASSERT_WLOCKED();
502 	if (clist == NULL || oidp == NULL)
503 		return(NULL);
504 	TAILQ_FOREACH(e, clist, link) {
505 		if(e->entry == oidp)
506 			return(e);
507 	}
508 	return (e);
509 }
510 
511 /*
512  * Delete an entry from the context.
513  * NOTE: this function doesn't free oidp! You have to remove it
514  * with sysctl_remove_oid().
515  */
516 int
517 sysctl_ctx_entry_del(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
518 {
519 	struct sysctl_ctx_entry *e;
520 
521 	if (clist == NULL || oidp == NULL)
522 		return (EINVAL);
523 	SYSCTL_WLOCK();
524 	e = sysctl_ctx_entry_find(clist, oidp);
525 	if (e != NULL) {
526 		TAILQ_REMOVE(clist, e, link);
527 		SYSCTL_WUNLOCK();
528 		free(e, M_SYSCTLOID);
529 		return (0);
530 	} else {
531 		SYSCTL_WUNLOCK();
532 		return (ENOENT);
533 	}
534 }
535 
536 /*
537  * Remove dynamically created sysctl trees.
538  * oidp - top of the tree to be removed
539  * del - if 0 - just deregister, otherwise free up entries as well
540  * recurse - if != 0 traverse the subtree to be deleted
541  */
542 int
543 sysctl_remove_oid(struct sysctl_oid *oidp, int del, int recurse)
544 {
545 	int error;
546 
547 	SYSCTL_WLOCK();
548 	error = sysctl_remove_oid_locked(oidp, del, recurse);
549 	SYSCTL_WUNLOCK();
550 	return (error);
551 }
552 
553 int
554 sysctl_remove_name(struct sysctl_oid *parent, const char *name,
555     int del, int recurse)
556 {
557 	struct sysctl_oid *p, *tmp;
558 	int error;
559 
560 	error = ENOENT;
561 	SYSCTL_WLOCK();
562 	SLIST_FOREACH_SAFE(p, SYSCTL_CHILDREN(parent), oid_link, tmp) {
563 		if (strcmp(p->oid_name, name) == 0) {
564 			error = sysctl_remove_oid_locked(p, del, recurse);
565 			break;
566 		}
567 	}
568 	SYSCTL_WUNLOCK();
569 
570 	return (error);
571 }
572 
573 
574 static int
575 sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del, int recurse)
576 {
577 	struct sysctl_oid *p, *tmp;
578 	int error;
579 
580 	SYSCTL_ASSERT_WLOCKED();
581 	if (oidp == NULL)
582 		return(EINVAL);
583 	if ((oidp->oid_kind & CTLFLAG_DYN) == 0) {
584 		printf("can't remove non-dynamic nodes!\n");
585 		return (EINVAL);
586 	}
587 	/*
588 	 * WARNING: normal method to do this should be through
589 	 * sysctl_ctx_free(). Use recursing as the last resort
590 	 * method to purge your sysctl tree of leftovers...
591 	 * However, if some other code still references these nodes,
592 	 * it will panic.
593 	 */
594 	if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
595 		if (oidp->oid_refcnt == 1) {
596 			SLIST_FOREACH_SAFE(p,
597 			    SYSCTL_CHILDREN(oidp), oid_link, tmp) {
598 				if (!recurse) {
599 					printf("Warning: failed attempt to "
600 					    "remove oid %s with child %s\n",
601 					    oidp->oid_name, p->oid_name);
602 					return (ENOTEMPTY);
603 				}
604 				error = sysctl_remove_oid_locked(p, del,
605 				    recurse);
606 				if (error)
607 					return (error);
608 			}
609 		}
610 	}
611 	if (oidp->oid_refcnt > 1 ) {
612 		oidp->oid_refcnt--;
613 	} else {
614 		if (oidp->oid_refcnt == 0) {
615 			printf("Warning: bad oid_refcnt=%u (%s)!\n",
616 				oidp->oid_refcnt, oidp->oid_name);
617 			return (EINVAL);
618 		}
619 		sysctl_unregister_oid(oidp);
620 		if (del) {
621 			/*
622 			 * Wait for all threads running the handler to drain.
623 			 * This preserves the previous behavior when the
624 			 * sysctl lock was held across a handler invocation,
625 			 * and is necessary for module unload correctness.
626 			 */
627 			while (oidp->oid_running > 0) {
628 				oidp->oid_kind |= CTLFLAG_DYING;
629 				SYSCTL_SLEEP(&oidp->oid_running, "oidrm", 0);
630 			}
631 			if (oidp->oid_descr)
632 				free(__DECONST(char *, oidp->oid_descr),
633 				    M_SYSCTLOID);
634 			free(__DECONST(char *, oidp->oid_name), M_SYSCTLOID);
635 			free(oidp, M_SYSCTLOID);
636 		}
637 	}
638 	return (0);
639 }
640 /*
641  * Create new sysctls at run time.
642  * clist may point to a valid context initialized with sysctl_ctx_init().
643  */
644 struct sysctl_oid *
645 sysctl_add_oid(struct sysctl_ctx_list *clist, struct sysctl_oid_list *parent,
646 	int number, const char *name, int kind, void *arg1, intptr_t arg2,
647 	int (*handler)(SYSCTL_HANDLER_ARGS), const char *fmt, const char *descr)
648 {
649 	struct sysctl_oid *oidp;
650 
651 	/* You have to hook up somewhere.. */
652 	if (parent == NULL)
653 		return(NULL);
654 	/* Check if the node already exists, otherwise create it */
655 	SYSCTL_WLOCK();
656 	oidp = sysctl_find_oidname(name, parent);
657 	if (oidp != NULL) {
658 		if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
659 			oidp->oid_refcnt++;
660 			/* Update the context */
661 			if (clist != NULL)
662 				sysctl_ctx_entry_add(clist, oidp);
663 			SYSCTL_WUNLOCK();
664 			return (oidp);
665 		} else {
666 			SYSCTL_WUNLOCK();
667 			printf("can't re-use a leaf (%s)!\n", name);
668 			return (NULL);
669 		}
670 	}
671 	oidp = malloc(sizeof(struct sysctl_oid), M_SYSCTLOID, M_WAITOK|M_ZERO);
672 	oidp->oid_parent = parent;
673 	SLIST_INIT(&oidp->oid_children);
674 	oidp->oid_number = number;
675 	oidp->oid_refcnt = 1;
676 	oidp->oid_name = strdup(name, M_SYSCTLOID);
677 	oidp->oid_handler = handler;
678 	oidp->oid_kind = CTLFLAG_DYN | kind;
679 	oidp->oid_arg1 = arg1;
680 	oidp->oid_arg2 = arg2;
681 	oidp->oid_fmt = fmt;
682 	if (descr != NULL)
683 		oidp->oid_descr = strdup(descr, M_SYSCTLOID);
684 	/* Update the context, if used */
685 	if (clist != NULL)
686 		sysctl_ctx_entry_add(clist, oidp);
687 	/* Register this oid */
688 	sysctl_register_oid(oidp);
689 	SYSCTL_WUNLOCK();
690 	return (oidp);
691 }
692 
693 /*
694  * Rename an existing oid.
695  */
696 void
697 sysctl_rename_oid(struct sysctl_oid *oidp, const char *name)
698 {
699 	char *newname;
700 	char *oldname;
701 
702 	newname = strdup(name, M_SYSCTLOID);
703 	SYSCTL_WLOCK();
704 	oldname = __DECONST(char *, oidp->oid_name);
705 	oidp->oid_name = newname;
706 	SYSCTL_WUNLOCK();
707 	free(oldname, M_SYSCTLOID);
708 }
709 
710 /*
711  * Reparent an existing oid.
712  */
713 int
714 sysctl_move_oid(struct sysctl_oid *oid, struct sysctl_oid_list *parent)
715 {
716 	struct sysctl_oid *oidp;
717 
718 	SYSCTL_WLOCK();
719 	if (oid->oid_parent == parent) {
720 		SYSCTL_WUNLOCK();
721 		return (0);
722 	}
723 	oidp = sysctl_find_oidname(oid->oid_name, parent);
724 	if (oidp != NULL) {
725 		SYSCTL_WUNLOCK();
726 		return (EEXIST);
727 	}
728 	sysctl_unregister_oid(oid);
729 	oid->oid_parent = parent;
730 	oid->oid_number = OID_AUTO;
731 	sysctl_register_oid(oid);
732 	SYSCTL_WUNLOCK();
733 	return (0);
734 }
735 
736 /*
737  * Register the kernel's oids on startup.
738  */
739 SET_DECLARE(sysctl_set, struct sysctl_oid);
740 
741 static void
742 sysctl_register_all(void *arg)
743 {
744 	struct sysctl_oid **oidp;
745 
746 	sx_init(&sysctlmemlock, "sysctl mem");
747 	SYSCTL_INIT();
748 	SYSCTL_WLOCK();
749 	SET_FOREACH(oidp, sysctl_set)
750 		sysctl_register_oid(*oidp);
751 	SYSCTL_WUNLOCK();
752 }
753 SYSINIT(sysctl, SI_SUB_KMEM, SI_ORDER_FIRST, sysctl_register_all, 0);
754 
755 /*
756  * "Staff-functions"
757  *
758  * These functions implement a presently undocumented interface
759  * used by the sysctl program to walk the tree, and get the type
760  * so it can print the value.
761  * This interface is under work and consideration, and should probably
762  * be killed with a big axe by the first person who can find the time.
763  * (be aware though, that the proper interface isn't as obvious as it
764  * may seem, there are various conflicting requirements.
765  *
766  * {0,0}	printf the entire MIB-tree.
767  * {0,1,...}	return the name of the "..." OID.
768  * {0,2,...}	return the next OID.
769  * {0,3}	return the OID of the name in "new"
770  * {0,4,...}	return the kind & format info for the "..." OID.
771  * {0,5,...}	return the description the "..." OID.
772  */
773 
774 #ifdef SYSCTL_DEBUG
775 static void
776 sysctl_sysctl_debug_dump_node(struct sysctl_oid_list *l, int i)
777 {
778 	int k;
779 	struct sysctl_oid *oidp;
780 
781 	SYSCTL_ASSERT_LOCKED();
782 	SLIST_FOREACH(oidp, l, oid_link) {
783 
784 		for (k=0; k<i; k++)
785 			printf(" ");
786 
787 		printf("%d %s ", oidp->oid_number, oidp->oid_name);
788 
789 		printf("%c%c",
790 			oidp->oid_kind & CTLFLAG_RD ? 'R':' ',
791 			oidp->oid_kind & CTLFLAG_WR ? 'W':' ');
792 
793 		if (oidp->oid_handler)
794 			printf(" *Handler");
795 
796 		switch (oidp->oid_kind & CTLTYPE) {
797 			case CTLTYPE_NODE:
798 				printf(" Node\n");
799 				if (!oidp->oid_handler) {
800 					sysctl_sysctl_debug_dump_node(
801 					    SYSCTL_CHILDREN(oidp), i + 2);
802 				}
803 				break;
804 			case CTLTYPE_INT:    printf(" Int\n"); break;
805 			case CTLTYPE_UINT:   printf(" u_int\n"); break;
806 			case CTLTYPE_LONG:   printf(" Long\n"); break;
807 			case CTLTYPE_ULONG:  printf(" u_long\n"); break;
808 			case CTLTYPE_STRING: printf(" String\n"); break;
809 			case CTLTYPE_U64:    printf(" uint64_t\n"); break;
810 			case CTLTYPE_S64:    printf(" int64_t\n"); break;
811 			case CTLTYPE_OPAQUE: printf(" Opaque/struct\n"); break;
812 			default:	     printf("\n");
813 		}
814 
815 	}
816 }
817 
818 static int
819 sysctl_sysctl_debug(SYSCTL_HANDLER_ARGS)
820 {
821 	struct rm_priotracker tracker;
822 	int error;
823 
824 	error = priv_check(req->td, PRIV_SYSCTL_DEBUG);
825 	if (error)
826 		return (error);
827 	SYSCTL_RLOCK(&tracker);
828 	sysctl_sysctl_debug_dump_node(&sysctl__children, 0);
829 	SYSCTL_RUNLOCK(&tracker);
830 	return (ENOENT);
831 }
832 
833 SYSCTL_PROC(_sysctl, 0, debug, CTLTYPE_STRING|CTLFLAG_RD|CTLFLAG_MPSAFE,
834 	0, 0, sysctl_sysctl_debug, "-", "");
835 #endif
836 
837 static int
838 sysctl_sysctl_name(SYSCTL_HANDLER_ARGS)
839 {
840 	int *name = (int *) arg1;
841 	u_int namelen = arg2;
842 	int error = 0;
843 	struct sysctl_oid *oid;
844 	struct sysctl_oid_list *lsp = &sysctl__children, *lsp2;
845 	struct rm_priotracker tracker;
846 	char buf[10];
847 
848 	SYSCTL_RLOCK(&tracker);
849 	while (namelen) {
850 		if (!lsp) {
851 			snprintf(buf,sizeof(buf),"%d",*name);
852 			if (req->oldidx)
853 				error = SYSCTL_OUT(req, ".", 1);
854 			if (!error)
855 				error = SYSCTL_OUT(req, buf, strlen(buf));
856 			if (error)
857 				goto out;
858 			namelen--;
859 			name++;
860 			continue;
861 		}
862 		lsp2 = 0;
863 		SLIST_FOREACH(oid, lsp, oid_link) {
864 			if (oid->oid_number != *name)
865 				continue;
866 
867 			if (req->oldidx)
868 				error = SYSCTL_OUT(req, ".", 1);
869 			if (!error)
870 				error = SYSCTL_OUT(req, oid->oid_name,
871 					strlen(oid->oid_name));
872 			if (error)
873 				goto out;
874 
875 			namelen--;
876 			name++;
877 
878 			if ((oid->oid_kind & CTLTYPE) != CTLTYPE_NODE)
879 				break;
880 
881 			if (oid->oid_handler)
882 				break;
883 
884 			lsp2 = SYSCTL_CHILDREN(oid);
885 			break;
886 		}
887 		lsp = lsp2;
888 	}
889 	error = SYSCTL_OUT(req, "", 1);
890  out:
891 	SYSCTL_RUNLOCK(&tracker);
892 	return (error);
893 }
894 
895 /*
896  * XXXRW/JA: Shouldn't return name data for nodes that we don't permit in
897  * capability mode.
898  */
899 static SYSCTL_NODE(_sysctl, 1, name, CTLFLAG_RD | CTLFLAG_MPSAFE | CTLFLAG_CAPRD,
900     sysctl_sysctl_name, "");
901 
902 static int
903 sysctl_sysctl_next_ls(struct sysctl_oid_list *lsp, int *name, u_int namelen,
904 	int *next, int *len, int level, struct sysctl_oid **oidpp)
905 {
906 	struct sysctl_oid *oidp;
907 
908 	SYSCTL_ASSERT_LOCKED();
909 	*len = level;
910 	SLIST_FOREACH(oidp, lsp, oid_link) {
911 		*next = oidp->oid_number;
912 		*oidpp = oidp;
913 
914 		if (oidp->oid_kind & CTLFLAG_SKIP)
915 			continue;
916 
917 		if (!namelen) {
918 			if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
919 				return (0);
920 			if (oidp->oid_handler)
921 				/* We really should call the handler here...*/
922 				return (0);
923 			lsp = SYSCTL_CHILDREN(oidp);
924 			if (!sysctl_sysctl_next_ls(lsp, 0, 0, next+1,
925 				len, level+1, oidpp))
926 				return (0);
927 			goto emptynode;
928 		}
929 
930 		if (oidp->oid_number < *name)
931 			continue;
932 
933 		if (oidp->oid_number > *name) {
934 			if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
935 				return (0);
936 			if (oidp->oid_handler)
937 				return (0);
938 			lsp = SYSCTL_CHILDREN(oidp);
939 			if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1,
940 				next+1, len, level+1, oidpp))
941 				return (0);
942 			goto next;
943 		}
944 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
945 			continue;
946 
947 		if (oidp->oid_handler)
948 			continue;
949 
950 		lsp = SYSCTL_CHILDREN(oidp);
951 		if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1, next+1,
952 			len, level+1, oidpp))
953 			return (0);
954 	next:
955 		namelen = 1;
956 	emptynode:
957 		*len = level;
958 	}
959 	return (1);
960 }
961 
962 static int
963 sysctl_sysctl_next(SYSCTL_HANDLER_ARGS)
964 {
965 	int *name = (int *) arg1;
966 	u_int namelen = arg2;
967 	int i, j, error;
968 	struct sysctl_oid *oid;
969 	struct sysctl_oid_list *lsp = &sysctl__children;
970 	struct rm_priotracker tracker;
971 	int newoid[CTL_MAXNAME];
972 
973 	SYSCTL_RLOCK(&tracker);
974 	i = sysctl_sysctl_next_ls(lsp, name, namelen, newoid, &j, 1, &oid);
975 	SYSCTL_RUNLOCK(&tracker);
976 	if (i)
977 		return (ENOENT);
978 	error = SYSCTL_OUT(req, newoid, j * sizeof (int));
979 	return (error);
980 }
981 
982 /*
983  * XXXRW/JA: Shouldn't return next data for nodes that we don't permit in
984  * capability mode.
985  */
986 static SYSCTL_NODE(_sysctl, 2, next, CTLFLAG_RD | CTLFLAG_MPSAFE | CTLFLAG_CAPRD,
987     sysctl_sysctl_next, "");
988 
989 static int
990 name2oid(char *name, int *oid, int *len, struct sysctl_oid **oidpp)
991 {
992 	struct sysctl_oid *oidp;
993 	struct sysctl_oid_list *lsp = &sysctl__children;
994 	char *p;
995 
996 	SYSCTL_ASSERT_LOCKED();
997 
998 	for (*len = 0; *len < CTL_MAXNAME;) {
999 		p = strsep(&name, ".");
1000 
1001 		oidp = SLIST_FIRST(lsp);
1002 		for (;; oidp = SLIST_NEXT(oidp, oid_link)) {
1003 			if (oidp == NULL)
1004 				return (ENOENT);
1005 			if (strcmp(p, oidp->oid_name) == 0)
1006 				break;
1007 		}
1008 		*oid++ = oidp->oid_number;
1009 		(*len)++;
1010 
1011 		if (name == NULL || *name == '\0') {
1012 			if (oidpp)
1013 				*oidpp = oidp;
1014 			return (0);
1015 		}
1016 
1017 		if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1018 			break;
1019 
1020 		if (oidp->oid_handler)
1021 			break;
1022 
1023 		lsp = SYSCTL_CHILDREN(oidp);
1024 	}
1025 	return (ENOENT);
1026 }
1027 
1028 static int
1029 sysctl_sysctl_name2oid(SYSCTL_HANDLER_ARGS)
1030 {
1031 	char *p;
1032 	int error, oid[CTL_MAXNAME], len = 0;
1033 	struct sysctl_oid *op = 0;
1034 	struct rm_priotracker tracker;
1035 
1036 	if (!req->newlen)
1037 		return (ENOENT);
1038 	if (req->newlen >= MAXPATHLEN)	/* XXX arbitrary, undocumented */
1039 		return (ENAMETOOLONG);
1040 
1041 	p = malloc(req->newlen+1, M_SYSCTL, M_WAITOK);
1042 
1043 	error = SYSCTL_IN(req, p, req->newlen);
1044 	if (error) {
1045 		free(p, M_SYSCTL);
1046 		return (error);
1047 	}
1048 
1049 	p [req->newlen] = '\0';
1050 
1051 	SYSCTL_RLOCK(&tracker);
1052 	error = name2oid(p, oid, &len, &op);
1053 	SYSCTL_RUNLOCK(&tracker);
1054 
1055 	free(p, M_SYSCTL);
1056 
1057 	if (error)
1058 		return (error);
1059 
1060 	error = SYSCTL_OUT(req, oid, len * sizeof *oid);
1061 	return (error);
1062 }
1063 
1064 /*
1065  * XXXRW/JA: Shouldn't return name2oid data for nodes that we don't permit in
1066  * capability mode.
1067  */
1068 SYSCTL_PROC(_sysctl, 3, name2oid,
1069     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_MPSAFE
1070     | CTLFLAG_CAPRW, 0, 0, sysctl_sysctl_name2oid, "I", "");
1071 
1072 static int
1073 sysctl_sysctl_oidfmt(SYSCTL_HANDLER_ARGS)
1074 {
1075 	struct sysctl_oid *oid;
1076 	struct rm_priotracker tracker;
1077 	int error;
1078 
1079 	SYSCTL_RLOCK(&tracker);
1080 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1081 	if (error)
1082 		goto out;
1083 
1084 	if (oid->oid_fmt == NULL) {
1085 		error = ENOENT;
1086 		goto out;
1087 	}
1088 	error = SYSCTL_OUT(req, &oid->oid_kind, sizeof(oid->oid_kind));
1089 	if (error)
1090 		goto out;
1091 	error = SYSCTL_OUT(req, oid->oid_fmt, strlen(oid->oid_fmt) + 1);
1092  out:
1093 	SYSCTL_RUNLOCK(&tracker);
1094 	return (error);
1095 }
1096 
1097 
1098 static SYSCTL_NODE(_sysctl, 4, oidfmt, CTLFLAG_RD|CTLFLAG_MPSAFE|CTLFLAG_CAPRD,
1099     sysctl_sysctl_oidfmt, "");
1100 
1101 static int
1102 sysctl_sysctl_oiddescr(SYSCTL_HANDLER_ARGS)
1103 {
1104 	struct sysctl_oid *oid;
1105 	struct rm_priotracker tracker;
1106 	int error;
1107 
1108 	SYSCTL_RLOCK(&tracker);
1109 	error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1110 	if (error)
1111 		goto out;
1112 
1113 	if (oid->oid_descr == NULL) {
1114 		error = ENOENT;
1115 		goto out;
1116 	}
1117 	error = SYSCTL_OUT(req, oid->oid_descr, strlen(oid->oid_descr) + 1);
1118  out:
1119 	SYSCTL_RUNLOCK(&tracker);
1120 	return (error);
1121 }
1122 
1123 static SYSCTL_NODE(_sysctl, 5, oiddescr, CTLFLAG_RD|CTLFLAG_MPSAFE|CTLFLAG_CAPRD,
1124     sysctl_sysctl_oiddescr, "");
1125 
1126 /*
1127  * Default "handler" functions.
1128  */
1129 
1130 /*
1131  * Handle an int, signed or unsigned.
1132  * Two cases:
1133  *     a variable:  point arg1 at it.
1134  *     a constant:  pass it in arg2.
1135  */
1136 
1137 int
1138 sysctl_handle_int(SYSCTL_HANDLER_ARGS)
1139 {
1140 	int tmpout, error = 0;
1141 
1142 	/*
1143 	 * Attempt to get a coherent snapshot by making a copy of the data.
1144 	 */
1145 	if (arg1)
1146 		tmpout = *(int *)arg1;
1147 	else
1148 		tmpout = arg2;
1149 	error = SYSCTL_OUT(req, &tmpout, sizeof(int));
1150 
1151 	if (error || !req->newptr)
1152 		return (error);
1153 
1154 	if (!arg1)
1155 		error = EPERM;
1156 	else
1157 		error = SYSCTL_IN(req, arg1, sizeof(int));
1158 	return (error);
1159 }
1160 
1161 /*
1162  * Based on on sysctl_handle_int() convert milliseconds into ticks.
1163  * Note: this is used by TCP.
1164  */
1165 
1166 int
1167 sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
1168 {
1169 	int error, s, tt;
1170 
1171 	tt = *(int *)arg1;
1172 	s = (int)((int64_t)tt * 1000 / hz);
1173 
1174 	error = sysctl_handle_int(oidp, &s, 0, req);
1175 	if (error || !req->newptr)
1176 		return (error);
1177 
1178 	tt = (int)((int64_t)s * hz / 1000);
1179 	if (tt < 1)
1180 		return (EINVAL);
1181 
1182 	*(int *)arg1 = tt;
1183 	return (0);
1184 }
1185 
1186 
1187 /*
1188  * Handle a long, signed or unsigned.
1189  * Two cases:
1190  *     a variable:  point arg1 at it.
1191  *     a constant:  pass it in arg2.
1192  */
1193 
1194 int
1195 sysctl_handle_long(SYSCTL_HANDLER_ARGS)
1196 {
1197 	int error = 0;
1198 	long tmplong;
1199 #ifdef SCTL_MASK32
1200 	int tmpint;
1201 #endif
1202 
1203 	/*
1204 	 * Attempt to get a coherent snapshot by making a copy of the data.
1205 	 */
1206 	if (arg1)
1207 		tmplong = *(long *)arg1;
1208 	else
1209 		tmplong = arg2;
1210 #ifdef SCTL_MASK32
1211 	if (req->flags & SCTL_MASK32) {
1212 		tmpint = tmplong;
1213 		error = SYSCTL_OUT(req, &tmpint, sizeof(int));
1214 	} else
1215 #endif
1216 		error = SYSCTL_OUT(req, &tmplong, sizeof(long));
1217 
1218 	if (error || !req->newptr)
1219 		return (error);
1220 
1221 	if (!arg1)
1222 		error = EPERM;
1223 #ifdef SCTL_MASK32
1224 	else if (req->flags & SCTL_MASK32) {
1225 		error = SYSCTL_IN(req, &tmpint, sizeof(int));
1226 		*(long *)arg1 = (long)tmpint;
1227 	}
1228 #endif
1229 	else
1230 		error = SYSCTL_IN(req, arg1, sizeof(long));
1231 	return (error);
1232 }
1233 
1234 /*
1235  * Handle a 64 bit int, signed or unsigned.
1236  * Two cases:
1237  *     a variable:  point arg1 at it.
1238  *     a constant:  pass it in arg2.
1239  */
1240 int
1241 sysctl_handle_64(SYSCTL_HANDLER_ARGS)
1242 {
1243 	int error = 0;
1244 	uint64_t tmpout;
1245 
1246 	/*
1247 	 * Attempt to get a coherent snapshot by making a copy of the data.
1248 	 */
1249 	if (arg1)
1250 		tmpout = *(uint64_t *)arg1;
1251 	else
1252 		tmpout = arg2;
1253 	error = SYSCTL_OUT(req, &tmpout, sizeof(uint64_t));
1254 
1255 	if (error || !req->newptr)
1256 		return (error);
1257 
1258 	if (!arg1)
1259 		error = EPERM;
1260 	else
1261 		error = SYSCTL_IN(req, arg1, sizeof(uint64_t));
1262 	return (error);
1263 }
1264 
1265 /*
1266  * Handle our generic '\0' terminated 'C' string.
1267  * Two cases:
1268  * 	a variable string:  point arg1 at it, arg2 is max length.
1269  * 	a constant string:  point arg1 at it, arg2 is zero.
1270  */
1271 
1272 int
1273 sysctl_handle_string(SYSCTL_HANDLER_ARGS)
1274 {
1275 	size_t outlen;
1276 	int error = 0, ro_string = 0;
1277 
1278 	/*
1279 	 * A zero-length buffer indicates a fixed size read-only
1280 	 * string:
1281 	 */
1282 	if (arg2 == 0) {
1283 		arg2 = strlen((char *)arg1) + 1;
1284 		ro_string = 1;
1285 	}
1286 
1287 	if (req->oldptr != NULL) {
1288 		char *tmparg;
1289 
1290 		if (ro_string) {
1291 			tmparg = arg1;
1292 		} else {
1293 			/* try to make a coherent snapshot of the string */
1294 			tmparg = malloc(arg2, M_SYSCTLTMP, M_WAITOK);
1295 			memcpy(tmparg, arg1, arg2);
1296 		}
1297 
1298 		outlen = strnlen(tmparg, arg2 - 1) + 1;
1299 		error = SYSCTL_OUT(req, tmparg, outlen);
1300 
1301 		if (!ro_string)
1302 			free(tmparg, M_SYSCTLTMP);
1303 	} else {
1304 		outlen = strnlen((char *)arg1, arg2 - 1) + 1;
1305 		error = SYSCTL_OUT(req, NULL, outlen);
1306 	}
1307 	if (error || !req->newptr)
1308 		return (error);
1309 
1310 	if ((req->newlen - req->newidx) >= arg2) {
1311 		error = EINVAL;
1312 	} else {
1313 		arg2 = (req->newlen - req->newidx);
1314 		error = SYSCTL_IN(req, arg1, arg2);
1315 		((char *)arg1)[arg2] = '\0';
1316 	}
1317 	return (error);
1318 }
1319 
1320 /*
1321  * Handle any kind of opaque data.
1322  * arg1 points to it, arg2 is the size.
1323  */
1324 
1325 int
1326 sysctl_handle_opaque(SYSCTL_HANDLER_ARGS)
1327 {
1328 	int error, tries;
1329 	u_int generation;
1330 	struct sysctl_req req2;
1331 
1332 	/*
1333 	 * Attempt to get a coherent snapshot, by using the thread
1334 	 * pre-emption counter updated from within mi_switch() to
1335 	 * determine if we were pre-empted during a bcopy() or
1336 	 * copyout(). Make 3 attempts at doing this before giving up.
1337 	 * If we encounter an error, stop immediately.
1338 	 */
1339 	tries = 0;
1340 	req2 = *req;
1341 retry:
1342 	generation = curthread->td_generation;
1343 	error = SYSCTL_OUT(req, arg1, arg2);
1344 	if (error)
1345 		return (error);
1346 	tries++;
1347 	if (generation != curthread->td_generation && tries < 3) {
1348 		*req = req2;
1349 		goto retry;
1350 	}
1351 
1352 	error = SYSCTL_IN(req, arg1, arg2);
1353 
1354 	return (error);
1355 }
1356 
1357 /*
1358  * Transfer functions to/from kernel space.
1359  * XXX: rather untested at this point
1360  */
1361 static int
1362 sysctl_old_kernel(struct sysctl_req *req, const void *p, size_t l)
1363 {
1364 	size_t i = 0;
1365 
1366 	if (req->oldptr) {
1367 		i = l;
1368 		if (req->oldlen <= req->oldidx)
1369 			i = 0;
1370 		else
1371 			if (i > req->oldlen - req->oldidx)
1372 				i = req->oldlen - req->oldidx;
1373 		if (i > 0)
1374 			bcopy(p, (char *)req->oldptr + req->oldidx, i);
1375 	}
1376 	req->oldidx += l;
1377 	if (req->oldptr && i != l)
1378 		return (ENOMEM);
1379 	return (0);
1380 }
1381 
1382 static int
1383 sysctl_new_kernel(struct sysctl_req *req, void *p, size_t l)
1384 {
1385 	if (!req->newptr)
1386 		return (0);
1387 	if (req->newlen - req->newidx < l)
1388 		return (EINVAL);
1389 	bcopy((char *)req->newptr + req->newidx, p, l);
1390 	req->newidx += l;
1391 	return (0);
1392 }
1393 
1394 int
1395 kernel_sysctl(struct thread *td, int *name, u_int namelen, void *old,
1396     size_t *oldlenp, void *new, size_t newlen, size_t *retval, int flags)
1397 {
1398 	int error = 0;
1399 	struct sysctl_req req;
1400 
1401 	bzero(&req, sizeof req);
1402 
1403 	req.td = td;
1404 	req.flags = flags;
1405 
1406 	if (oldlenp) {
1407 		req.oldlen = *oldlenp;
1408 	}
1409 	req.validlen = req.oldlen;
1410 
1411 	if (old) {
1412 		req.oldptr= old;
1413 	}
1414 
1415 	if (new != NULL) {
1416 		req.newlen = newlen;
1417 		req.newptr = new;
1418 	}
1419 
1420 	req.oldfunc = sysctl_old_kernel;
1421 	req.newfunc = sysctl_new_kernel;
1422 	req.lock = REQ_UNWIRED;
1423 
1424 	error = sysctl_root(0, name, namelen, &req);
1425 
1426 	if (req.lock == REQ_WIRED && req.validlen > 0)
1427 		vsunlock(req.oldptr, req.validlen);
1428 
1429 	if (error && error != ENOMEM)
1430 		return (error);
1431 
1432 	if (retval) {
1433 		if (req.oldptr && req.oldidx > req.validlen)
1434 			*retval = req.validlen;
1435 		else
1436 			*retval = req.oldidx;
1437 	}
1438 	return (error);
1439 }
1440 
1441 int
1442 kernel_sysctlbyname(struct thread *td, char *name, void *old, size_t *oldlenp,
1443     void *new, size_t newlen, size_t *retval, int flags)
1444 {
1445         int oid[CTL_MAXNAME];
1446         size_t oidlen, plen;
1447 	int error;
1448 
1449 	oid[0] = 0;		/* sysctl internal magic */
1450 	oid[1] = 3;		/* name2oid */
1451 	oidlen = sizeof(oid);
1452 
1453 	error = kernel_sysctl(td, oid, 2, oid, &oidlen,
1454 	    (void *)name, strlen(name), &plen, flags);
1455 	if (error)
1456 		return (error);
1457 
1458 	error = kernel_sysctl(td, oid, plen / sizeof(int), old, oldlenp,
1459 	    new, newlen, retval, flags);
1460 	return (error);
1461 }
1462 
1463 /*
1464  * Transfer function to/from user space.
1465  */
1466 static int
1467 sysctl_old_user(struct sysctl_req *req, const void *p, size_t l)
1468 {
1469 	size_t i, len, origidx;
1470 	int error;
1471 
1472 	origidx = req->oldidx;
1473 	req->oldidx += l;
1474 	if (req->oldptr == NULL)
1475 		return (0);
1476 	/*
1477 	 * If we have not wired the user supplied buffer and we are currently
1478 	 * holding locks, drop a witness warning, as it's possible that
1479 	 * write operations to the user page can sleep.
1480 	 */
1481 	if (req->lock != REQ_WIRED)
1482 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
1483 		    "sysctl_old_user()");
1484 	i = l;
1485 	len = req->validlen;
1486 	if (len <= origidx)
1487 		i = 0;
1488 	else {
1489 		if (i > len - origidx)
1490 			i = len - origidx;
1491 		if (req->lock == REQ_WIRED) {
1492 			error = copyout_nofault(p, (char *)req->oldptr +
1493 			    origidx, i);
1494 		} else
1495 			error = copyout(p, (char *)req->oldptr + origidx, i);
1496 		if (error != 0)
1497 			return (error);
1498 	}
1499 	if (i < l)
1500 		return (ENOMEM);
1501 	return (0);
1502 }
1503 
1504 static int
1505 sysctl_new_user(struct sysctl_req *req, void *p, size_t l)
1506 {
1507 	int error;
1508 
1509 	if (!req->newptr)
1510 		return (0);
1511 	if (req->newlen - req->newidx < l)
1512 		return (EINVAL);
1513 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
1514 	    "sysctl_new_user()");
1515 	error = copyin((char *)req->newptr + req->newidx, p, l);
1516 	req->newidx += l;
1517 	return (error);
1518 }
1519 
1520 /*
1521  * Wire the user space destination buffer.  If set to a value greater than
1522  * zero, the len parameter limits the maximum amount of wired memory.
1523  */
1524 int
1525 sysctl_wire_old_buffer(struct sysctl_req *req, size_t len)
1526 {
1527 	int ret;
1528 	size_t wiredlen;
1529 
1530 	wiredlen = (len > 0 && len < req->oldlen) ? len : req->oldlen;
1531 	ret = 0;
1532 	if (req->lock != REQ_WIRED && req->oldptr &&
1533 	    req->oldfunc == sysctl_old_user) {
1534 		if (wiredlen != 0) {
1535 			ret = vslock(req->oldptr, wiredlen);
1536 			if (ret != 0) {
1537 				if (ret != ENOMEM)
1538 					return (ret);
1539 				wiredlen = 0;
1540 			}
1541 		}
1542 		req->lock = REQ_WIRED;
1543 		req->validlen = wiredlen;
1544 	}
1545 	return (0);
1546 }
1547 
1548 int
1549 sysctl_find_oid(int *name, u_int namelen, struct sysctl_oid **noid,
1550     int *nindx, struct sysctl_req *req)
1551 {
1552 	struct sysctl_oid_list *lsp;
1553 	struct sysctl_oid *oid;
1554 	int indx;
1555 
1556 	SYSCTL_ASSERT_LOCKED();
1557 	lsp = &sysctl__children;
1558 	indx = 0;
1559 	while (indx < CTL_MAXNAME) {
1560 		SLIST_FOREACH(oid, lsp, oid_link) {
1561 			if (oid->oid_number == name[indx])
1562 				break;
1563 		}
1564 		if (oid == NULL)
1565 			return (ENOENT);
1566 
1567 		indx++;
1568 		if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1569 			if (oid->oid_handler != NULL || indx == namelen) {
1570 				*noid = oid;
1571 				if (nindx != NULL)
1572 					*nindx = indx;
1573 				KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
1574 				    ("%s found DYING node %p", __func__, oid));
1575 				return (0);
1576 			}
1577 			lsp = SYSCTL_CHILDREN(oid);
1578 		} else if (indx == namelen) {
1579 			*noid = oid;
1580 			if (nindx != NULL)
1581 				*nindx = indx;
1582 			KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
1583 			    ("%s found DYING node %p", __func__, oid));
1584 			return (0);
1585 		} else {
1586 			return (ENOTDIR);
1587 		}
1588 	}
1589 	return (ENOENT);
1590 }
1591 
1592 /*
1593  * Traverse our tree, and find the right node, execute whatever it points
1594  * to, and return the resulting error code.
1595  */
1596 
1597 static int
1598 sysctl_root(SYSCTL_HANDLER_ARGS)
1599 {
1600 	struct sysctl_oid *oid;
1601 	struct rm_priotracker tracker;
1602 	int error, indx, lvl;
1603 
1604 	SYSCTL_RLOCK(&tracker);
1605 
1606 	error = sysctl_find_oid(arg1, arg2, &oid, &indx, req);
1607 	if (error)
1608 		goto out;
1609 
1610 	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1611 		/*
1612 		 * You can't call a sysctl when it's a node, but has
1613 		 * no handler.  Inform the user that it's a node.
1614 		 * The indx may or may not be the same as namelen.
1615 		 */
1616 		if (oid->oid_handler == NULL) {
1617 			error = EISDIR;
1618 			goto out;
1619 		}
1620 	}
1621 
1622 	/* Is this sysctl writable? */
1623 	if (req->newptr && !(oid->oid_kind & CTLFLAG_WR)) {
1624 		error = EPERM;
1625 		goto out;
1626 	}
1627 
1628 	KASSERT(req->td != NULL, ("sysctl_root(): req->td == NULL"));
1629 
1630 #ifdef CAPABILITY_MODE
1631 	/*
1632 	 * If the process is in capability mode, then don't permit reading or
1633 	 * writing unless specifically granted for the node.
1634 	 */
1635 	if (IN_CAPABILITY_MODE(req->td)) {
1636 		if ((req->oldptr && !(oid->oid_kind & CTLFLAG_CAPRD)) ||
1637 		    (req->newptr && !(oid->oid_kind & CTLFLAG_CAPWR))) {
1638 			error = EPERM;
1639 			goto out;
1640 		}
1641 	}
1642 #endif
1643 
1644 	/* Is this sysctl sensitive to securelevels? */
1645 	if (req->newptr && (oid->oid_kind & CTLFLAG_SECURE)) {
1646 		lvl = (oid->oid_kind & CTLMASK_SECURE) >> CTLSHIFT_SECURE;
1647 		error = securelevel_gt(req->td->td_ucred, lvl);
1648 		if (error)
1649 			goto out;
1650 	}
1651 
1652 	/* Is this sysctl writable by only privileged users? */
1653 	if (req->newptr && !(oid->oid_kind & CTLFLAG_ANYBODY)) {
1654 		int priv;
1655 
1656 		if (oid->oid_kind & CTLFLAG_PRISON)
1657 			priv = PRIV_SYSCTL_WRITEJAIL;
1658 #ifdef VIMAGE
1659 		else if ((oid->oid_kind & CTLFLAG_VNET) &&
1660 		     prison_owns_vnet(req->td->td_ucred))
1661 			priv = PRIV_SYSCTL_WRITEJAIL;
1662 #endif
1663 		else
1664 			priv = PRIV_SYSCTL_WRITE;
1665 		error = priv_check(req->td, priv);
1666 		if (error)
1667 			goto out;
1668 	}
1669 
1670 	if (!oid->oid_handler) {
1671 		error = EINVAL;
1672 		goto out;
1673 	}
1674 
1675 	if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
1676 		arg1 = (int *)arg1 + indx;
1677 		arg2 -= indx;
1678 	} else {
1679 		arg1 = oid->oid_arg1;
1680 		arg2 = oid->oid_arg2;
1681 	}
1682 #ifdef MAC
1683 	error = mac_system_check_sysctl(req->td->td_ucred, oid, arg1, arg2,
1684 	    req);
1685 	if (error != 0)
1686 		goto out;
1687 #endif
1688 #ifdef VIMAGE
1689 	if ((oid->oid_kind & CTLFLAG_VNET) && arg1 != NULL)
1690 		arg1 = (void *)(curvnet->vnet_data_base + (uintptr_t)arg1);
1691 #endif
1692 	error = sysctl_root_handler_locked(oid, arg1, arg2, req, &tracker);
1693 
1694 	KFAIL_POINT_ERROR(_debug_fail_point, sysctl_running, error);
1695 
1696 out:
1697 	SYSCTL_RUNLOCK(&tracker);
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.oldptr && 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 		error = sysctl_root(0, name, namelen, &req);
1798 		if (error != EAGAIN)
1799 			break;
1800 		kern_yield(PRI_USER);
1801 	}
1802 
1803 	CURVNET_RESTORE();
1804 
1805 	if (req.lock == REQ_WIRED && req.validlen > 0)
1806 		vsunlock(req.oldptr, req.validlen);
1807 	if (memlocked)
1808 		sx_xunlock(&sysctlmemlock);
1809 
1810 	if (error && error != ENOMEM)
1811 		return (error);
1812 
1813 	if (retval) {
1814 		if (req.oldptr && req.oldidx > req.validlen)
1815 			*retval = req.validlen;
1816 		else
1817 			*retval = req.oldidx;
1818 	}
1819 	return (error);
1820 }
1821 
1822 /*
1823  * Drain into a sysctl struct.  The user buffer should be wired if a page
1824  * fault would cause issue.
1825  */
1826 static int
1827 sbuf_sysctl_drain(void *arg, const char *data, int len)
1828 {
1829 	struct sysctl_req *req = arg;
1830 	int error;
1831 
1832 	error = SYSCTL_OUT(req, data, len);
1833 	KASSERT(error >= 0, ("Got unexpected negative value %d", error));
1834 	return (error == 0 ? len : -error);
1835 }
1836 
1837 struct sbuf *
1838 sbuf_new_for_sysctl(struct sbuf *s, char *buf, int length,
1839     struct sysctl_req *req)
1840 {
1841 
1842 	/* Supply a default buffer size if none given. */
1843 	if (buf == NULL && length == 0)
1844 		length = 64;
1845 	s = sbuf_new(s, buf, length, SBUF_FIXEDLEN | SBUF_INCLUDENUL);
1846 	sbuf_set_drain(s, sbuf_sysctl_drain, req);
1847 	return (s);
1848 }
1849