2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1989, 1993
5 * The Regents of the University of California. All rights reserved.
7 * This code is derived from software contributed to Berkeley by
8 * Mike Karels at Berkeley Software Design, Inc.
10 * Quite extensively rewritten by Poul-Henning Kamp of the FreeBSD
11 * project, to make these variables more userfriendly.
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 * 3. Neither the name of the University nor the names of its contributors
22 * may be used to endorse or promote products derived from this software
23 * without specific prior written permission.
25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * @(#)kern_sysctl.c 8.4 (Berkeley) 4/14/94
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
43 #include "opt_capsicum.h"
45 #include "opt_ktrace.h"
46 #include "opt_sysctl.h"
48 #include <sys/param.h>
50 #include <sys/systm.h>
51 #include <sys/capsicum.h>
52 #include <sys/kernel.h>
53 #include <sys/limits.h>
54 #include <sys/sysctl.h>
55 #include <sys/malloc.h>
61 #include <sys/mutex.h>
62 #include <sys/rmlock.h>
65 #include <sys/sysproto.h>
68 #include <sys/ktrace.h>
73 #include <ddb/db_lex.h>
78 #include <security/mac/mac_framework.h>
81 #include <vm/vm_extern.h>
83 static MALLOC_DEFINE(M_SYSCTL, "sysctl", "sysctl internal magic");
84 static MALLOC_DEFINE(M_SYSCTLOID, "sysctloid", "sysctl dynamic oids");
85 static MALLOC_DEFINE(M_SYSCTLTMP, "sysctltmp", "sysctl temp output buffer");
88 * The sysctllock protects the MIB tree. It also protects sysctl
89 * contexts used with dynamic sysctls. The sysctl_register_oid() and
90 * sysctl_unregister_oid() routines require the sysctllock to already
91 * be held, so the sysctl_wlock() and sysctl_wunlock() routines are
92 * provided for the few places in the kernel which need to use that
93 * API rather than using the dynamic API. Use of the dynamic API is
94 * strongly encouraged for most code.
96 * The sysctlmemlock is used to limit the amount of user memory wired for
97 * sysctl requests. This is implemented by serializing any userland
98 * sysctl requests larger than a single page via an exclusive lock.
100 * The sysctlstringlock is used to protect concurrent access to writable
101 * string nodes in sysctl_handle_string().
103 static struct rmlock sysctllock;
104 static struct sx __exclusive_cache_line sysctlmemlock;
105 static struct sx sysctlstringlock;
107 #define SYSCTL_WLOCK() rm_wlock(&sysctllock)
108 #define SYSCTL_WUNLOCK() rm_wunlock(&sysctllock)
109 #define SYSCTL_RLOCK(tracker) rm_rlock(&sysctllock, (tracker))
110 #define SYSCTL_RUNLOCK(tracker) rm_runlock(&sysctllock, (tracker))
111 #define SYSCTL_WLOCKED() rm_wowned(&sysctllock)
112 #define SYSCTL_ASSERT_LOCKED() rm_assert(&sysctllock, RA_LOCKED)
113 #define SYSCTL_ASSERT_WLOCKED() rm_assert(&sysctllock, RA_WLOCKED)
114 #define SYSCTL_ASSERT_RLOCKED() rm_assert(&sysctllock, RA_RLOCKED)
115 #define SYSCTL_INIT() rm_init_flags(&sysctllock, "sysctl lock", \
117 #define SYSCTL_SLEEP(ch, wmesg, timo) \
118 rm_sleep(ch, &sysctllock, 0, wmesg, timo)
120 static int sysctl_root(SYSCTL_HANDLER_ARGS);
123 struct sysctl_oid_list sysctl__children = SLIST_HEAD_INITIALIZER(&sysctl__children);
125 static int sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del,
127 static int sysctl_old_kernel(struct sysctl_req *, const void *, size_t);
128 static int sysctl_new_kernel(struct sysctl_req *, void *, size_t);
130 static struct sysctl_oid *
131 sysctl_find_oidname(const char *name, struct sysctl_oid_list *list)
133 struct sysctl_oid *oidp;
135 SYSCTL_ASSERT_LOCKED();
136 SLIST_FOREACH(oidp, list, oid_link) {
137 if (strcmp(oidp->oid_name, name) == 0) {
145 * Initialization of the MIB tree.
147 * Order by number in each list.
164 sysctl_root_handler_locked(struct sysctl_oid *oid, void *arg1, intmax_t arg2,
165 struct sysctl_req *req, struct rm_priotracker *tracker)
169 if (oid->oid_kind & CTLFLAG_DYN)
170 atomic_add_int(&oid->oid_running, 1);
173 SYSCTL_RUNLOCK(tracker);
178 * Treat set CTLFLAG_NEEDGIANT and unset CTLFLAG_MPSAFE flags the same,
179 * untill we're ready to remove all traces of Giant from sysctl(9).
181 if ((oid->oid_kind & CTLFLAG_NEEDGIANT) ||
182 (!(oid->oid_kind & CTLFLAG_MPSAFE)))
184 error = oid->oid_handler(oid, arg1, arg2, req);
185 if ((oid->oid_kind & CTLFLAG_NEEDGIANT) ||
186 (!(oid->oid_kind & CTLFLAG_MPSAFE)))
189 KFAIL_POINT_ERROR(_debug_fail_point, sysctl_running, error);
192 SYSCTL_RLOCK(tracker);
196 if (oid->oid_kind & CTLFLAG_DYN) {
197 if (atomic_fetchadd_int(&oid->oid_running, -1) == 1 &&
198 (oid->oid_kind & CTLFLAG_DYING) != 0)
199 wakeup(&oid->oid_running);
206 sysctl_load_tunable_by_oid_locked(struct sysctl_oid *oidp)
208 struct sysctl_req req;
209 struct sysctl_oid *curr;
212 ssize_t rem = sizeof(path);
214 uint8_t data[512] __aligned(sizeof(uint64_t));
220 for (curr = oidp; curr != NULL; curr = SYSCTL_PARENT(curr)) {
221 len = strlen(curr->oid_name);
226 printf("OID path exceeds %d bytes\n", (int)sizeof(path));
229 memcpy(path + rem, curr->oid_name, len);
231 path[rem + len] = '.';
234 memset(&req, 0, sizeof(req));
237 req.oldfunc = sysctl_old_kernel;
238 req.newfunc = sysctl_new_kernel;
239 req.lock = REQ_UNWIRED;
241 switch (oidp->oid_kind & CTLTYPE) {
243 if (getenv_array(path + rem, data, sizeof(data), &size,
244 sizeof(int), GETENV_SIGNED) == 0)
250 if (getenv_array(path + rem, data, sizeof(data), &size,
251 sizeof(int), GETENV_UNSIGNED) == 0)
257 if (getenv_array(path + rem, data, sizeof(data), &size,
258 sizeof(long), GETENV_SIGNED) == 0)
264 if (getenv_array(path + rem, data, sizeof(data), &size,
265 sizeof(long), GETENV_UNSIGNED) == 0)
271 if (getenv_array(path + rem, data, sizeof(data), &size,
272 sizeof(int8_t), GETENV_SIGNED) == 0)
278 if (getenv_array(path + rem, data, sizeof(data), &size,
279 sizeof(int16_t), GETENV_SIGNED) == 0)
285 if (getenv_array(path + rem, data, sizeof(data), &size,
286 sizeof(int32_t), GETENV_SIGNED) == 0)
292 if (getenv_array(path + rem, data, sizeof(data), &size,
293 sizeof(int64_t), GETENV_SIGNED) == 0)
299 if (getenv_array(path + rem, data, sizeof(data), &size,
300 sizeof(uint8_t), GETENV_UNSIGNED) == 0)
306 if (getenv_array(path + rem, data, sizeof(data), &size,
307 sizeof(uint16_t), GETENV_UNSIGNED) == 0)
313 if (getenv_array(path + rem, data, sizeof(data), &size,
314 sizeof(uint32_t), GETENV_UNSIGNED) == 0)
320 if (getenv_array(path + rem, data, sizeof(data), &size,
321 sizeof(uint64_t), GETENV_UNSIGNED) == 0)
327 penv = kern_getenv(path + rem);
330 req.newlen = strlen(penv);
336 error = sysctl_root_handler_locked(oidp, oidp->oid_arg1,
337 oidp->oid_arg2, &req, NULL);
339 printf("Setting sysctl %s failed: %d\n", path + rem, error);
345 * Locate the path to a given oid. Returns the length of the resulting path,
346 * or -1 if the oid was not found. nodes must have room for CTL_MAXNAME
347 * elements and be NULL initialized.
350 sysctl_search_oid(struct sysctl_oid **nodes, struct sysctl_oid *needle)
354 SYSCTL_ASSERT_LOCKED();
356 while (indx < CTL_MAXNAME && indx >= 0) {
357 if (nodes[indx] == NULL && indx == 0)
358 nodes[indx] = SLIST_FIRST(&sysctl__children);
359 else if (nodes[indx] == NULL)
360 nodes[indx] = SLIST_FIRST(&nodes[indx - 1]->oid_children);
362 nodes[indx] = SLIST_NEXT(nodes[indx], oid_link);
364 if (nodes[indx] == needle)
367 if (nodes[indx] == NULL) {
372 if ((nodes[indx]->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
381 sysctl_warn_reuse(const char *func, struct sysctl_oid *leaf)
383 struct sysctl_oid *nodes[CTL_MAXNAME];
388 (void)sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN | SBUF_INCLUDENUL);
389 sbuf_set_drain(&sb, sbuf_printf_drain, NULL);
391 sbuf_printf(&sb, "%s: can't re-use a leaf (", __func__);
393 memset(nodes, 0, sizeof(nodes));
394 rc = sysctl_search_oid(nodes, leaf);
396 for (i = 0; i < rc; i++)
397 sbuf_printf(&sb, "%s%.*s", nodes[i]->oid_name,
400 sbuf_printf(&sb, "%s", leaf->oid_name);
402 sbuf_printf(&sb, ")!\n");
404 (void)sbuf_finish(&sb);
409 sysctl_reuse_test(SYSCTL_HANDLER_ARGS)
411 struct rm_priotracker tracker;
413 SYSCTL_RLOCK(&tracker);
414 sysctl_warn_reuse(__func__, oidp);
415 SYSCTL_RUNLOCK(&tracker);
418 SYSCTL_PROC(_sysctl, OID_AUTO, reuse_test,
419 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 0, sysctl_reuse_test, "-",
424 sysctl_register_oid(struct sysctl_oid *oidp)
426 struct sysctl_oid_list *parent = oidp->oid_parent;
427 struct sysctl_oid *p;
428 struct sysctl_oid *q;
433 * First check if another oid with the same name already
434 * exists in the parent's list.
436 SYSCTL_ASSERT_WLOCKED();
437 p = sysctl_find_oidname(oidp->oid_name, parent);
439 if ((p->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
443 sysctl_warn_reuse(__func__, p);
447 /* get current OID number */
448 oid_number = oidp->oid_number;
451 #error "OID_AUTO is expected to be a negative value"
454 * Any negative OID number qualifies as OID_AUTO. Valid OID
455 * numbers should always be positive.
457 * NOTE: DO NOT change the starting value here, change it in
458 * <sys/sysctl.h>, and make sure it is at least 256 to
459 * accommodate e.g. net.inet.raw as a static sysctl node.
461 if (oid_number < 0) {
465 * By decrementing the next OID number we spend less
466 * time inserting the OIDs into a sorted list.
468 if (--newoid < CTL_AUTO_START)
475 * Insert the OID into the parent's list sorted by OID number.
479 SLIST_FOREACH(p, parent, oid_link) {
480 /* check if the current OID number is in use */
481 if (oid_number == p->oid_number) {
482 /* get the next valid OID number */
483 if (oid_number < CTL_AUTO_START ||
484 oid_number == 0x7fffffff) {
485 /* wraparound - restart */
486 oid_number = CTL_AUTO_START;
487 /* don't loop forever */
489 panic("sysctl: Out of OID numbers\n");
494 } else if (oid_number < p->oid_number)
498 /* check for non-auto OID number collision */
499 if (oidp->oid_number >= 0 && oidp->oid_number < CTL_AUTO_START &&
500 oid_number >= CTL_AUTO_START) {
501 printf("sysctl: OID number(%d) is already in use for '%s'\n",
502 oidp->oid_number, oidp->oid_name);
504 /* update the OID number, if any */
505 oidp->oid_number = oid_number;
507 SLIST_INSERT_AFTER(q, oidp, oid_link);
509 SLIST_INSERT_HEAD(parent, oidp, oid_link);
511 if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE &&
513 (oidp->oid_kind & CTLFLAG_VNET) == 0 &&
515 (oidp->oid_kind & CTLFLAG_TUN) != 0 &&
516 (oidp->oid_kind & CTLFLAG_NOFETCH) == 0) {
517 /* only fetch value once */
518 oidp->oid_kind |= CTLFLAG_NOFETCH;
519 /* try to fetch value from kernel environment */
520 sysctl_load_tunable_by_oid_locked(oidp);
525 sysctl_register_disabled_oid(struct sysctl_oid *oidp)
529 * Mark the leaf as dormant if it's not to be immediately enabled.
530 * We do not disable nodes as they can be shared between modules
531 * and it is always safe to access a node.
533 KASSERT((oidp->oid_kind & CTLFLAG_DORMANT) == 0,
534 ("internal flag is set in oid_kind"));
535 if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
536 oidp->oid_kind |= CTLFLAG_DORMANT;
537 sysctl_register_oid(oidp);
541 sysctl_enable_oid(struct sysctl_oid *oidp)
544 SYSCTL_ASSERT_WLOCKED();
545 if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
546 KASSERT((oidp->oid_kind & CTLFLAG_DORMANT) == 0,
547 ("sysctl node is marked as dormant"));
550 KASSERT((oidp->oid_kind & CTLFLAG_DORMANT) != 0,
551 ("enabling already enabled sysctl oid"));
552 oidp->oid_kind &= ~CTLFLAG_DORMANT;
556 sysctl_unregister_oid(struct sysctl_oid *oidp)
558 struct sysctl_oid *p;
561 SYSCTL_ASSERT_WLOCKED();
562 if (oidp->oid_number == OID_AUTO) {
566 SLIST_FOREACH(p, oidp->oid_parent, oid_link) {
568 SLIST_REMOVE(oidp->oid_parent, oidp,
569 sysctl_oid, oid_link);
577 * This can happen when a module fails to register and is
578 * being unloaded afterwards. It should not be a panic()
582 printf("%s: failed(%d) to unregister sysctl(%s)\n",
583 __func__, error, oidp->oid_name);
587 /* Initialize a new context to keep track of dynamically added sysctls. */
589 sysctl_ctx_init(struct sysctl_ctx_list *c)
597 * No locking here, the caller is responsible for not adding
598 * new nodes to a context until after this function has
605 /* Free the context, and destroy all dynamic oids registered in this context */
607 sysctl_ctx_free(struct sysctl_ctx_list *clist)
609 struct sysctl_ctx_entry *e, *e1;
614 * First perform a "dry run" to check if it's ok to remove oids.
616 * XXX This algorithm is a hack. But I don't know any
617 * XXX better solution for now...
620 TAILQ_FOREACH(e, clist, link) {
621 error = sysctl_remove_oid_locked(e->entry, 0, 0);
626 * Restore deregistered entries, either from the end,
627 * or from the place where error occurred.
628 * e contains the entry that was not unregistered
631 e1 = TAILQ_PREV(e, sysctl_ctx_list, link);
633 e1 = TAILQ_LAST(clist, sysctl_ctx_list);
635 sysctl_register_oid(e1->entry);
636 e1 = TAILQ_PREV(e1, sysctl_ctx_list, link);
642 /* Now really delete the entries */
643 e = TAILQ_FIRST(clist);
645 e1 = TAILQ_NEXT(e, link);
646 error = sysctl_remove_oid_locked(e->entry, 1, 0);
648 panic("sysctl_remove_oid: corrupt tree, entry: %s",
650 free(e, M_SYSCTLOID);
657 /* Add an entry to the context */
658 struct sysctl_ctx_entry *
659 sysctl_ctx_entry_add(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
661 struct sysctl_ctx_entry *e;
663 SYSCTL_ASSERT_WLOCKED();
664 if (clist == NULL || oidp == NULL)
666 e = malloc(sizeof(struct sysctl_ctx_entry), M_SYSCTLOID, M_WAITOK);
668 TAILQ_INSERT_HEAD(clist, e, link);
672 /* Find an entry in the context */
673 struct sysctl_ctx_entry *
674 sysctl_ctx_entry_find(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
676 struct sysctl_ctx_entry *e;
678 SYSCTL_ASSERT_WLOCKED();
679 if (clist == NULL || oidp == NULL)
681 TAILQ_FOREACH(e, clist, link) {
689 * Delete an entry from the context.
690 * NOTE: this function doesn't free oidp! You have to remove it
691 * with sysctl_remove_oid().
694 sysctl_ctx_entry_del(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
696 struct sysctl_ctx_entry *e;
698 if (clist == NULL || oidp == NULL)
701 e = sysctl_ctx_entry_find(clist, oidp);
703 TAILQ_REMOVE(clist, e, link);
705 free(e, M_SYSCTLOID);
714 * Remove dynamically created sysctl trees.
715 * oidp - top of the tree to be removed
716 * del - if 0 - just deregister, otherwise free up entries as well
717 * recurse - if != 0 traverse the subtree to be deleted
720 sysctl_remove_oid(struct sysctl_oid *oidp, int del, int recurse)
725 error = sysctl_remove_oid_locked(oidp, del, recurse);
731 sysctl_remove_name(struct sysctl_oid *parent, const char *name,
732 int del, int recurse)
734 struct sysctl_oid *p, *tmp;
739 SLIST_FOREACH_SAFE(p, SYSCTL_CHILDREN(parent), oid_link, tmp) {
740 if (strcmp(p->oid_name, name) == 0) {
741 error = sysctl_remove_oid_locked(p, del, recurse);
751 sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del, int recurse)
753 struct sysctl_oid *p, *tmp;
756 SYSCTL_ASSERT_WLOCKED();
759 if ((oidp->oid_kind & CTLFLAG_DYN) == 0) {
760 printf("Warning: can't remove non-dynamic nodes (%s)!\n",
765 * WARNING: normal method to do this should be through
766 * sysctl_ctx_free(). Use recursing as the last resort
767 * method to purge your sysctl tree of leftovers...
768 * However, if some other code still references these nodes,
771 if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
772 if (oidp->oid_refcnt == 1) {
773 SLIST_FOREACH_SAFE(p,
774 SYSCTL_CHILDREN(oidp), oid_link, tmp) {
776 printf("Warning: failed attempt to "
777 "remove oid %s with child %s\n",
778 oidp->oid_name, p->oid_name);
781 error = sysctl_remove_oid_locked(p, del,
788 if (oidp->oid_refcnt > 1 ) {
791 if (oidp->oid_refcnt == 0) {
792 printf("Warning: bad oid_refcnt=%u (%s)!\n",
793 oidp->oid_refcnt, oidp->oid_name);
796 sysctl_unregister_oid(oidp);
799 * Wait for all threads running the handler to drain.
800 * This preserves the previous behavior when the
801 * sysctl lock was held across a handler invocation,
802 * and is necessary for module unload correctness.
804 while (oidp->oid_running > 0) {
805 oidp->oid_kind |= CTLFLAG_DYING;
806 SYSCTL_SLEEP(&oidp->oid_running, "oidrm", 0);
809 free(__DECONST(char *, oidp->oid_descr),
812 free(__DECONST(char *, oidp->oid_label),
814 free(__DECONST(char *, oidp->oid_name), M_SYSCTLOID);
815 free(oidp, M_SYSCTLOID);
821 * Create new sysctls at run time.
822 * clist may point to a valid context initialized with sysctl_ctx_init().
825 sysctl_add_oid(struct sysctl_ctx_list *clist, struct sysctl_oid_list *parent,
826 int number, const char *name, int kind, void *arg1, intmax_t arg2,
827 int (*handler)(SYSCTL_HANDLER_ARGS), const char *fmt, const char *descr,
830 struct sysctl_oid *oidp;
832 /* You have to hook up somewhere.. */
835 /* Check if the node already exists, otherwise create it */
837 oidp = sysctl_find_oidname(name, parent);
839 if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
841 /* Update the context */
843 sysctl_ctx_entry_add(clist, oidp);
847 sysctl_warn_reuse(__func__, oidp);
852 oidp = malloc(sizeof(struct sysctl_oid), M_SYSCTLOID, M_WAITOK|M_ZERO);
853 oidp->oid_parent = parent;
854 SLIST_INIT(&oidp->oid_children);
855 oidp->oid_number = number;
856 oidp->oid_refcnt = 1;
857 oidp->oid_name = strdup(name, M_SYSCTLOID);
858 oidp->oid_handler = handler;
859 oidp->oid_kind = CTLFLAG_DYN | kind;
860 oidp->oid_arg1 = arg1;
861 oidp->oid_arg2 = arg2;
864 oidp->oid_descr = strdup(descr, M_SYSCTLOID);
866 oidp->oid_label = strdup(label, M_SYSCTLOID);
867 /* Update the context, if used */
869 sysctl_ctx_entry_add(clist, oidp);
870 /* Register this oid */
871 sysctl_register_oid(oidp);
877 * Rename an existing oid.
880 sysctl_rename_oid(struct sysctl_oid *oidp, const char *name)
885 newname = strdup(name, M_SYSCTLOID);
887 oldname = __DECONST(char *, oidp->oid_name);
888 oidp->oid_name = newname;
890 free(oldname, M_SYSCTLOID);
894 * Reparent an existing oid.
897 sysctl_move_oid(struct sysctl_oid *oid, struct sysctl_oid_list *parent)
899 struct sysctl_oid *oidp;
902 if (oid->oid_parent == parent) {
906 oidp = sysctl_find_oidname(oid->oid_name, parent);
911 sysctl_unregister_oid(oid);
912 oid->oid_parent = parent;
913 oid->oid_number = OID_AUTO;
914 sysctl_register_oid(oid);
920 * Register the kernel's oids on startup.
922 SET_DECLARE(sysctl_set, struct sysctl_oid);
925 sysctl_register_all(void *arg)
927 struct sysctl_oid **oidp;
929 sx_init(&sysctlmemlock, "sysctl mem");
930 sx_init(&sysctlstringlock, "sysctl string handler");
933 SET_FOREACH(oidp, sysctl_set)
934 sysctl_register_oid(*oidp);
937 SYSINIT(sysctl, SI_SUB_KMEM, SI_ORDER_FIRST, sysctl_register_all, NULL);
942 * These functions implement a presently undocumented interface
943 * used by the sysctl program to walk the tree, and get the type
944 * so it can print the value.
945 * This interface is under work and consideration, and should probably
946 * be killed with a big axe by the first person who can find the time.
947 * (be aware though, that the proper interface isn't as obvious as it
948 * may seem, there are various conflicting requirements.
950 * {CTL_SYSCTL, CTL_SYSCTL_DEBUG} printf the entire MIB-tree.
951 * {CTL_SYSCTL, CTL_SYSCTL_NAME, ...} return the name of the "..."
953 * {CTL_SYSCTL, CTL_SYSCTL_NEXT, ...} return the next OID.
954 * {CTL_SYSCTL, CTL_SYSCTL_NAME2OID} return the OID of the name in
956 * {CTL_SYSCTL, CTL_SYSCTL_OIDFMT, ...} return the kind & format info
958 * {CTL_SYSCTL, CTL_SYSCTL_OIDDESCR, ...} return the description of the
960 * {CTL_SYSCTL, CTL_SYSCTL_OIDLABEL, ...} return the aggregation label of
966 sysctl_sysctl_debug_dump_node(struct sysctl_oid_list *l, int i)
969 struct sysctl_oid *oidp;
971 SYSCTL_ASSERT_LOCKED();
972 SLIST_FOREACH(oidp, l, oid_link) {
976 printf("%d %s ", oidp->oid_number, oidp->oid_name);
979 oidp->oid_kind & CTLFLAG_RD ? 'R':' ',
980 oidp->oid_kind & CTLFLAG_WR ? 'W':' ');
982 if (oidp->oid_handler)
985 switch (oidp->oid_kind & CTLTYPE) {
988 if (!oidp->oid_handler) {
989 sysctl_sysctl_debug_dump_node(
990 SYSCTL_CHILDREN(oidp), i + 2);
993 case CTLTYPE_INT: printf(" Int\n"); break;
994 case CTLTYPE_UINT: printf(" u_int\n"); break;
995 case CTLTYPE_LONG: printf(" Long\n"); break;
996 case CTLTYPE_ULONG: printf(" u_long\n"); break;
997 case CTLTYPE_STRING: printf(" String\n"); break;
998 case CTLTYPE_S8: printf(" int8_t\n"); break;
999 case CTLTYPE_S16: printf(" int16_t\n"); break;
1000 case CTLTYPE_S32: printf(" int32_t\n"); break;
1001 case CTLTYPE_S64: printf(" int64_t\n"); break;
1002 case CTLTYPE_U8: printf(" uint8_t\n"); break;
1003 case CTLTYPE_U16: printf(" uint16_t\n"); break;
1004 case CTLTYPE_U32: printf(" uint32_t\n"); break;
1005 case CTLTYPE_U64: printf(" uint64_t\n"); break;
1006 case CTLTYPE_OPAQUE: printf(" Opaque/struct\n"); break;
1007 default: printf("\n");
1013 sysctl_sysctl_debug(SYSCTL_HANDLER_ARGS)
1015 struct rm_priotracker tracker;
1018 error = priv_check(req->td, PRIV_SYSCTL_DEBUG);
1021 SYSCTL_RLOCK(&tracker);
1022 sysctl_sysctl_debug_dump_node(&sysctl__children, 0);
1023 SYSCTL_RUNLOCK(&tracker);
1027 SYSCTL_PROC(_sysctl, CTL_SYSCTL_DEBUG, debug, CTLTYPE_STRING | CTLFLAG_RD |
1028 CTLFLAG_MPSAFE, 0, 0, sysctl_sysctl_debug, "-", "");
1032 sysctl_sysctl_name(SYSCTL_HANDLER_ARGS)
1034 int *name = (int *) arg1;
1035 u_int namelen = arg2;
1037 struct sysctl_oid *oid;
1038 struct sysctl_oid_list *lsp = &sysctl__children, *lsp2;
1039 struct rm_priotracker tracker;
1042 error = sysctl_wire_old_buffer(req, 0);
1046 SYSCTL_RLOCK(&tracker);
1049 snprintf(buf,sizeof(buf),"%d",*name);
1051 error = SYSCTL_OUT(req, ".", 1);
1053 error = SYSCTL_OUT(req, buf, strlen(buf));
1061 SLIST_FOREACH(oid, lsp, oid_link) {
1062 if (oid->oid_number != *name)
1066 error = SYSCTL_OUT(req, ".", 1);
1068 error = SYSCTL_OUT(req, oid->oid_name,
1069 strlen(oid->oid_name));
1076 if ((oid->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1079 if (oid->oid_handler)
1082 lsp2 = SYSCTL_CHILDREN(oid);
1087 error = SYSCTL_OUT(req, "", 1);
1089 SYSCTL_RUNLOCK(&tracker);
1094 * XXXRW/JA: Shouldn't return name data for nodes that we don't permit in
1097 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_NAME, name, CTLFLAG_RD |
1098 CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_name, "");
1101 sysctl_sysctl_next_ls(struct sysctl_oid_list *lsp, int *name, u_int namelen,
1102 int *next, int *len, int level, struct sysctl_oid **oidpp)
1104 struct sysctl_oid *oidp;
1106 SYSCTL_ASSERT_LOCKED();
1108 SLIST_FOREACH(oidp, lsp, oid_link) {
1109 *next = oidp->oid_number;
1112 if ((oidp->oid_kind & (CTLFLAG_SKIP | CTLFLAG_DORMANT)) != 0)
1116 if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1118 if (oidp->oid_handler)
1119 /* We really should call the handler here...*/
1121 lsp = SYSCTL_CHILDREN(oidp);
1122 if (!sysctl_sysctl_next_ls(lsp, 0, 0, next+1,
1123 len, level+1, oidpp))
1128 if (oidp->oid_number < *name)
1131 if (oidp->oid_number > *name) {
1132 if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1134 if (oidp->oid_handler)
1136 lsp = SYSCTL_CHILDREN(oidp);
1137 if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1,
1138 next+1, len, level+1, oidpp))
1142 if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1145 if (oidp->oid_handler)
1148 lsp = SYSCTL_CHILDREN(oidp);
1149 if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1, next+1,
1150 len, level+1, oidpp))
1161 sysctl_sysctl_next(SYSCTL_HANDLER_ARGS)
1163 int *name = (int *) arg1;
1164 u_int namelen = arg2;
1166 struct sysctl_oid *oid;
1167 struct sysctl_oid_list *lsp = &sysctl__children;
1168 struct rm_priotracker tracker;
1169 int newoid[CTL_MAXNAME];
1171 SYSCTL_RLOCK(&tracker);
1172 i = sysctl_sysctl_next_ls(lsp, name, namelen, newoid, &j, 1, &oid);
1173 SYSCTL_RUNLOCK(&tracker);
1176 error = SYSCTL_OUT(req, newoid, j * sizeof (int));
1181 * XXXRW/JA: Shouldn't return next data for nodes that we don't permit in
1184 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_NEXT, next, CTLFLAG_RD |
1185 CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_next, "");
1188 name2oid(char *name, int *oid, int *len, struct sysctl_oid **oidpp)
1190 struct sysctl_oid *oidp;
1191 struct sysctl_oid_list *lsp = &sysctl__children;
1194 SYSCTL_ASSERT_LOCKED();
1196 for (*len = 0; *len < CTL_MAXNAME;) {
1197 p = strsep(&name, ".");
1199 oidp = SLIST_FIRST(lsp);
1200 for (;; oidp = SLIST_NEXT(oidp, oid_link)) {
1203 if (strcmp(p, oidp->oid_name) == 0)
1206 *oid++ = oidp->oid_number;
1209 if (name == NULL || *name == '\0') {
1215 if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
1218 if (oidp->oid_handler)
1221 lsp = SYSCTL_CHILDREN(oidp);
1227 sysctl_sysctl_name2oid(SYSCTL_HANDLER_ARGS)
1230 int error, oid[CTL_MAXNAME], len = 0;
1231 struct sysctl_oid *op = NULL;
1232 struct rm_priotracker tracker;
1237 if (req->newlen >= MAXPATHLEN) /* XXX arbitrary, undocumented */
1238 return (ENAMETOOLONG);
1241 if (req->newlen >= sizeof(buf))
1242 p = malloc(req->newlen+1, M_SYSCTL, M_WAITOK);
1244 error = SYSCTL_IN(req, p, req->newlen);
1251 p [req->newlen] = '\0';
1253 SYSCTL_RLOCK(&tracker);
1254 error = name2oid(p, oid, &len, &op);
1255 SYSCTL_RUNLOCK(&tracker);
1263 error = SYSCTL_OUT(req, oid, len * sizeof *oid);
1268 * XXXRW/JA: Shouldn't return name2oid data for nodes that we don't permit in
1271 SYSCTL_PROC(_sysctl, CTL_SYSCTL_NAME2OID, name2oid, CTLTYPE_INT | CTLFLAG_RW |
1272 CTLFLAG_ANYBODY | CTLFLAG_MPSAFE | CTLFLAG_CAPRW, 0, 0,
1273 sysctl_sysctl_name2oid, "I", "");
1276 sysctl_sysctl_oidfmt(SYSCTL_HANDLER_ARGS)
1278 struct sysctl_oid *oid;
1279 struct rm_priotracker tracker;
1282 error = sysctl_wire_old_buffer(req, 0);
1286 SYSCTL_RLOCK(&tracker);
1287 error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1291 if (oid->oid_fmt == NULL) {
1295 error = SYSCTL_OUT(req, &oid->oid_kind, sizeof(oid->oid_kind));
1298 error = SYSCTL_OUT(req, oid->oid_fmt, strlen(oid->oid_fmt) + 1);
1300 SYSCTL_RUNLOCK(&tracker);
1304 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_OIDFMT, oidfmt, CTLFLAG_RD |
1305 CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_oidfmt, "");
1308 sysctl_sysctl_oiddescr(SYSCTL_HANDLER_ARGS)
1310 struct sysctl_oid *oid;
1311 struct rm_priotracker tracker;
1314 error = sysctl_wire_old_buffer(req, 0);
1318 SYSCTL_RLOCK(&tracker);
1319 error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1323 if (oid->oid_descr == NULL) {
1327 error = SYSCTL_OUT(req, oid->oid_descr, strlen(oid->oid_descr) + 1);
1329 SYSCTL_RUNLOCK(&tracker);
1333 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_OIDDESCR, oiddescr, CTLFLAG_RD |
1334 CTLFLAG_MPSAFE|CTLFLAG_CAPRD, sysctl_sysctl_oiddescr, "");
1337 sysctl_sysctl_oidlabel(SYSCTL_HANDLER_ARGS)
1339 struct sysctl_oid *oid;
1340 struct rm_priotracker tracker;
1343 error = sysctl_wire_old_buffer(req, 0);
1347 SYSCTL_RLOCK(&tracker);
1348 error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
1352 if (oid->oid_label == NULL) {
1356 error = SYSCTL_OUT(req, oid->oid_label, strlen(oid->oid_label) + 1);
1358 SYSCTL_RUNLOCK(&tracker);
1362 static SYSCTL_NODE(_sysctl, CTL_SYSCTL_OIDLABEL, oidlabel, CTLFLAG_RD |
1363 CTLFLAG_MPSAFE | CTLFLAG_CAPRD, sysctl_sysctl_oidlabel, "");
1366 * Default "handler" functions.
1372 * a variable: point arg1 at it.
1373 * a constant: pass it in arg2.
1377 sysctl_handle_bool(SYSCTL_HANDLER_ARGS)
1383 * Attempt to get a coherent snapshot by making a copy of the data.
1386 temp = *(bool *)arg1 ? 1 : 0;
1388 temp = arg2 ? 1 : 0;
1390 error = SYSCTL_OUT(req, &temp, sizeof(temp));
1391 if (error || !req->newptr)
1397 error = SYSCTL_IN(req, &temp, sizeof(temp));
1399 *(bool *)arg1 = temp ? 1 : 0;
1405 * Handle an int8_t, signed or unsigned.
1407 * a variable: point arg1 at it.
1408 * a constant: pass it in arg2.
1412 sysctl_handle_8(SYSCTL_HANDLER_ARGS)
1418 * Attempt to get a coherent snapshot by making a copy of the data.
1421 tmpout = *(int8_t *)arg1;
1424 error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout));
1426 if (error || !req->newptr)
1432 error = SYSCTL_IN(req, arg1, sizeof(tmpout));
1437 * Handle an int16_t, signed or unsigned.
1439 * a variable: point arg1 at it.
1440 * a constant: pass it in arg2.
1444 sysctl_handle_16(SYSCTL_HANDLER_ARGS)
1450 * Attempt to get a coherent snapshot by making a copy of the data.
1453 tmpout = *(int16_t *)arg1;
1456 error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout));
1458 if (error || !req->newptr)
1464 error = SYSCTL_IN(req, arg1, sizeof(tmpout));
1469 * Handle an int32_t, signed or unsigned.
1471 * a variable: point arg1 at it.
1472 * a constant: pass it in arg2.
1476 sysctl_handle_32(SYSCTL_HANDLER_ARGS)
1482 * Attempt to get a coherent snapshot by making a copy of the data.
1485 tmpout = *(int32_t *)arg1;
1488 error = SYSCTL_OUT(req, &tmpout, sizeof(tmpout));
1490 if (error || !req->newptr)
1496 error = SYSCTL_IN(req, arg1, sizeof(tmpout));
1501 * Handle an int, signed or unsigned.
1503 * a variable: point arg1 at it.
1504 * a constant: pass it in arg2.
1508 sysctl_handle_int(SYSCTL_HANDLER_ARGS)
1510 int tmpout, error = 0;
1513 * Attempt to get a coherent snapshot by making a copy of the data.
1516 tmpout = *(int *)arg1;
1519 error = SYSCTL_OUT(req, &tmpout, sizeof(int));
1521 if (error || !req->newptr)
1527 error = SYSCTL_IN(req, arg1, sizeof(int));
1532 * Based on on sysctl_handle_int() convert milliseconds into ticks.
1533 * Note: this is used by TCP.
1537 sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
1542 s = (int)((int64_t)tt * 1000 / hz);
1544 error = sysctl_handle_int(oidp, &s, 0, req);
1545 if (error || !req->newptr)
1548 tt = (int)((int64_t)s * hz / 1000);
1557 * Handle a long, signed or unsigned.
1559 * a variable: point arg1 at it.
1560 * a constant: pass it in arg2.
1564 sysctl_handle_long(SYSCTL_HANDLER_ARGS)
1573 * Attempt to get a coherent snapshot by making a copy of the data.
1576 tmplong = *(long *)arg1;
1580 if (req->flags & SCTL_MASK32) {
1582 error = SYSCTL_OUT(req, &tmpint, sizeof(int));
1585 error = SYSCTL_OUT(req, &tmplong, sizeof(long));
1587 if (error || !req->newptr)
1593 else if (req->flags & SCTL_MASK32) {
1594 error = SYSCTL_IN(req, &tmpint, sizeof(int));
1595 *(long *)arg1 = (long)tmpint;
1599 error = SYSCTL_IN(req, arg1, sizeof(long));
1604 * Handle a 64 bit int, signed or unsigned.
1606 * a variable: point arg1 at it.
1607 * a constant: pass it in arg2.
1610 sysctl_handle_64(SYSCTL_HANDLER_ARGS)
1616 * Attempt to get a coherent snapshot by making a copy of the data.
1619 tmpout = *(uint64_t *)arg1;
1622 error = SYSCTL_OUT(req, &tmpout, sizeof(uint64_t));
1624 if (error || !req->newptr)
1630 error = SYSCTL_IN(req, arg1, sizeof(uint64_t));
1635 * Handle our generic '\0' terminated 'C' string.
1637 * a variable string: point arg1 at it, arg2 is max length.
1638 * a constant string: point arg1 at it, arg2 is zero.
1642 sysctl_handle_string(SYSCTL_HANDLER_ARGS)
1646 int error = 0, ro_string = 0;
1649 * If the sysctl isn't writable and isn't a preallocated tunable that
1650 * can be modified by kenv(2), microoptimise and treat it as a
1652 * A zero-length buffer indicates a fixed size read-only
1653 * string. In ddb, don't worry about trying to make a malloced
1656 if ((oidp->oid_kind & (CTLFLAG_WR | CTLFLAG_TUN)) == 0 ||
1657 arg2 == 0 || kdb_active) {
1658 arg2 = strlen((char *)arg1) + 1;
1662 if (req->oldptr != NULL) {
1665 outlen = strlen(tmparg) + 1;
1667 tmparg = malloc(arg2, M_SYSCTLTMP, M_WAITOK);
1668 sx_slock(&sysctlstringlock);
1669 memcpy(tmparg, arg1, arg2);
1670 sx_sunlock(&sysctlstringlock);
1671 outlen = strlen(tmparg) + 1;
1674 error = SYSCTL_OUT(req, tmparg, outlen);
1677 free(tmparg, M_SYSCTLTMP);
1680 sx_slock(&sysctlstringlock);
1681 outlen = strlen((char *)arg1) + 1;
1683 sx_sunlock(&sysctlstringlock);
1684 error = SYSCTL_OUT(req, NULL, outlen);
1686 if (error || !req->newptr)
1689 if (req->newlen - req->newidx >= arg2 ||
1690 req->newlen - req->newidx < 0) {
1692 } else if (req->newlen - req->newidx == 0) {
1693 sx_xlock(&sysctlstringlock);
1694 ((char *)arg1)[0] = '\0';
1695 sx_xunlock(&sysctlstringlock);
1697 arg2 = req->newlen - req->newidx;
1698 tmparg = malloc(arg2, M_SYSCTLTMP, M_WAITOK);
1700 error = SYSCTL_IN(req, tmparg, arg2);
1702 free(tmparg, M_SYSCTLTMP);
1706 sx_xlock(&sysctlstringlock);
1707 memcpy(arg1, tmparg, arg2);
1708 ((char *)arg1)[arg2] = '\0';
1709 sx_xunlock(&sysctlstringlock);
1710 free(tmparg, M_SYSCTLTMP);
1711 req->newidx += arg2;
1717 * Handle any kind of opaque data.
1718 * arg1 points to it, arg2 is the size.
1722 sysctl_handle_opaque(SYSCTL_HANDLER_ARGS)
1726 struct sysctl_req req2;
1729 * Attempt to get a coherent snapshot, by using the thread
1730 * pre-emption counter updated from within mi_switch() to
1731 * determine if we were pre-empted during a bcopy() or
1732 * copyout(). Make 3 attempts at doing this before giving up.
1733 * If we encounter an error, stop immediately.
1738 generation = curthread->td_generation;
1739 error = SYSCTL_OUT(req, arg1, arg2);
1743 if (generation != curthread->td_generation && tries < 3) {
1748 error = SYSCTL_IN(req, arg1, arg2);
1754 * Based on on sysctl_handle_int() convert microseconds to a sbintime.
1757 sysctl_usec_to_sbintime(SYSCTL_HANDLER_ARGS)
1763 tt = *(int64_t *)arg1;
1766 error = sysctl_handle_64(oidp, &sb, 0, req);
1767 if (error || !req->newptr)
1771 *(int64_t *)arg1 = tt;
1777 * Based on on sysctl_handle_int() convert milliseconds to a sbintime.
1780 sysctl_msec_to_sbintime(SYSCTL_HANDLER_ARGS)
1786 tt = *(int64_t *)arg1;
1789 error = sysctl_handle_64(oidp, &sb, 0, req);
1790 if (error || !req->newptr)
1794 *(int64_t *)arg1 = tt;
1800 * Convert seconds to a struct timeval. Intended for use with
1801 * intervals and thus does not permit negative seconds.
1804 sysctl_sec_to_timeval(SYSCTL_HANDLER_ARGS)
1812 error = sysctl_handle_int(oidp, &secs, 0, req);
1813 if (error || req->newptr == NULL)
1824 * Transfer functions to/from kernel space.
1825 * XXX: rather untested at this point
1828 sysctl_old_kernel(struct sysctl_req *req, const void *p, size_t l)
1834 if (req->oldlen <= req->oldidx)
1837 if (i > req->oldlen - req->oldidx)
1838 i = req->oldlen - req->oldidx;
1840 bcopy(p, (char *)req->oldptr + req->oldidx, i);
1843 if (req->oldptr && i != l)
1849 sysctl_new_kernel(struct sysctl_req *req, void *p, size_t l)
1853 if (req->newlen - req->newidx < l)
1855 bcopy((const char *)req->newptr + req->newidx, p, l);
1861 kernel_sysctl(struct thread *td, int *name, u_int namelen, void *old,
1862 size_t *oldlenp, void *new, size_t newlen, size_t *retval, int flags)
1865 struct sysctl_req req;
1867 bzero(&req, sizeof req);
1873 req.oldlen = *oldlenp;
1875 req.validlen = req.oldlen;
1882 req.newlen = newlen;
1886 req.oldfunc = sysctl_old_kernel;
1887 req.newfunc = sysctl_new_kernel;
1888 req.lock = REQ_UNWIRED;
1890 error = sysctl_root(0, name, namelen, &req);
1892 if (req.lock == REQ_WIRED && req.validlen > 0)
1893 vsunlock(req.oldptr, req.validlen);
1895 if (error && error != ENOMEM)
1899 if (req.oldptr && req.oldidx > req.validlen)
1900 *retval = req.validlen;
1902 *retval = req.oldidx;
1908 kernel_sysctlbyname(struct thread *td, char *name, void *old, size_t *oldlenp,
1909 void *new, size_t newlen, size_t *retval, int flags)
1911 int oid[CTL_MAXNAME];
1912 size_t oidlen, plen;
1915 oid[0] = CTL_SYSCTL;
1916 oid[1] = CTL_SYSCTL_NAME2OID;
1917 oidlen = sizeof(oid);
1919 error = kernel_sysctl(td, oid, 2, oid, &oidlen,
1920 (void *)name, strlen(name), &plen, flags);
1924 error = kernel_sysctl(td, oid, plen / sizeof(int), old, oldlenp,
1925 new, newlen, retval, flags);
1930 * Transfer function to/from user space.
1933 sysctl_old_user(struct sysctl_req *req, const void *p, size_t l)
1935 size_t i, len, origidx;
1938 origidx = req->oldidx;
1940 if (req->oldptr == NULL)
1943 * If we have not wired the user supplied buffer and we are currently
1944 * holding locks, drop a witness warning, as it's possible that
1945 * write operations to the user page can sleep.
1947 if (req->lock != REQ_WIRED)
1948 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
1949 "sysctl_old_user()");
1951 len = req->validlen;
1955 if (i > len - origidx)
1957 if (req->lock == REQ_WIRED) {
1958 error = copyout_nofault(p, (char *)req->oldptr +
1961 error = copyout(p, (char *)req->oldptr + origidx, i);
1971 sysctl_new_user(struct sysctl_req *req, void *p, size_t l)
1977 if (req->newlen - req->newidx < l)
1979 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
1980 "sysctl_new_user()");
1981 error = copyin((const char *)req->newptr + req->newidx, p, l);
1987 * Wire the user space destination buffer. If set to a value greater than
1988 * zero, the len parameter limits the maximum amount of wired memory.
1991 sysctl_wire_old_buffer(struct sysctl_req *req, size_t len)
1996 wiredlen = (len > 0 && len < req->oldlen) ? len : req->oldlen;
1998 if (req->lock != REQ_WIRED && req->oldptr &&
1999 req->oldfunc == sysctl_old_user) {
2000 if (wiredlen != 0) {
2001 ret = vslock(req->oldptr, wiredlen);
2008 req->lock = REQ_WIRED;
2009 req->validlen = wiredlen;
2015 sysctl_find_oid(int *name, u_int namelen, struct sysctl_oid **noid,
2016 int *nindx, struct sysctl_req *req)
2018 struct sysctl_oid_list *lsp;
2019 struct sysctl_oid *oid;
2022 SYSCTL_ASSERT_LOCKED();
2023 lsp = &sysctl__children;
2025 while (indx < CTL_MAXNAME) {
2026 SLIST_FOREACH(oid, lsp, oid_link) {
2027 if (oid->oid_number == name[indx])
2034 if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
2035 if (oid->oid_handler != NULL || indx == namelen) {
2039 KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
2040 ("%s found DYING node %p", __func__, oid));
2043 lsp = SYSCTL_CHILDREN(oid);
2044 } else if (indx == namelen) {
2045 if ((oid->oid_kind & CTLFLAG_DORMANT) != 0)
2050 KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
2051 ("%s found DYING node %p", __func__, oid));
2061 * Traverse our tree, and find the right node, execute whatever it points
2062 * to, and return the resulting error code.
2066 sysctl_root(SYSCTL_HANDLER_ARGS)
2068 struct sysctl_oid *oid;
2069 struct rm_priotracker tracker;
2070 int error, indx, lvl;
2072 SYSCTL_RLOCK(&tracker);
2074 error = sysctl_find_oid(arg1, arg2, &oid, &indx, req);
2078 if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
2080 * You can't call a sysctl when it's a node, but has
2081 * no handler. Inform the user that it's a node.
2082 * The indx may or may not be the same as namelen.
2084 if (oid->oid_handler == NULL) {
2090 /* Is this sysctl writable? */
2091 if (req->newptr && !(oid->oid_kind & CTLFLAG_WR)) {
2096 KASSERT(req->td != NULL, ("sysctl_root(): req->td == NULL"));
2098 #ifdef CAPABILITY_MODE
2100 * If the process is in capability mode, then don't permit reading or
2101 * writing unless specifically granted for the node.
2103 if (IN_CAPABILITY_MODE(req->td)) {
2104 if ((req->oldptr && !(oid->oid_kind & CTLFLAG_CAPRD)) ||
2105 (req->newptr && !(oid->oid_kind & CTLFLAG_CAPWR))) {
2112 /* Is this sysctl sensitive to securelevels? */
2113 if (req->newptr && (oid->oid_kind & CTLFLAG_SECURE)) {
2114 lvl = (oid->oid_kind & CTLMASK_SECURE) >> CTLSHIFT_SECURE;
2115 error = securelevel_gt(req->td->td_ucred, lvl);
2120 /* Is this sysctl writable by only privileged users? */
2121 if (req->newptr && !(oid->oid_kind & CTLFLAG_ANYBODY)) {
2124 if (oid->oid_kind & CTLFLAG_PRISON)
2125 priv = PRIV_SYSCTL_WRITEJAIL;
2127 else if ((oid->oid_kind & CTLFLAG_VNET) &&
2128 prison_owns_vnet(req->td->td_ucred))
2129 priv = PRIV_SYSCTL_WRITEJAIL;
2132 priv = PRIV_SYSCTL_WRITE;
2133 error = priv_check(req->td, priv);
2138 if (!oid->oid_handler) {
2143 if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
2144 arg1 = (int *)arg1 + indx;
2147 arg1 = oid->oid_arg1;
2148 arg2 = oid->oid_arg2;
2151 error = mac_system_check_sysctl(req->td->td_ucred, oid, arg1, arg2,
2157 if ((oid->oid_kind & CTLFLAG_VNET) && arg1 != NULL)
2158 arg1 = (void *)(curvnet->vnet_data_base + (uintptr_t)arg1);
2160 error = sysctl_root_handler_locked(oid, arg1, arg2, req, &tracker);
2163 SYSCTL_RUNLOCK(&tracker);
2167 #ifndef _SYS_SYSPROTO_H_
2168 struct sysctl_args {
2178 sys___sysctl(struct thread *td, struct sysctl_args *uap)
2180 int error, i, name[CTL_MAXNAME];
2183 if (uap->namelen > CTL_MAXNAME || uap->namelen < 2)
2186 error = copyin(uap->name, &name, uap->namelen * sizeof(int));
2190 error = userland_sysctl(td, name, uap->namelen,
2191 uap->old, uap->oldlenp, 0,
2192 uap->new, uap->newlen, &j, 0);
2193 if (error && error != ENOMEM)
2196 i = copyout(&j, uap->oldlenp, sizeof(j));
2204 kern___sysctlbyname(struct thread *td, const char *oname, size_t namelen,
2205 void *old, size_t *oldlenp, void *new, size_t newlen, size_t *retval,
2206 int flags, bool inkernel)
2208 int oid[CTL_MAXNAME];
2214 if (namelen > MAXPATHLEN || namelen == 0)
2217 if (namelen > sizeof(namebuf))
2218 name = malloc(namelen, M_SYSCTL, M_WAITOK);
2219 error = copyin(oname, name, namelen);
2223 oid[0] = CTL_SYSCTL;
2224 oid[1] = CTL_SYSCTL_NAME2OID;
2225 oidlen = sizeof(oid);
2226 error = kernel_sysctl(td, oid, 2, oid, &oidlen, (void *)name, namelen,
2230 error = userland_sysctl(td, oid, *retval / sizeof(int), old, oldlenp,
2231 inkernel, new, newlen, retval, flags);
2234 if (namelen > sizeof(namebuf))
2235 free(name, M_SYSCTL);
2239 #ifndef _SYS_SYSPROTO_H_
2240 struct __sysctlbyname_args {
2250 sys___sysctlbyname(struct thread *td, struct __sysctlbyname_args *uap)
2255 error = kern___sysctlbyname(td, uap->name, uap->namelen, uap->old,
2256 uap->oldlenp, uap->new, uap->newlen, &rv, 0, 0);
2259 if (uap->oldlenp != NULL)
2260 error = copyout(&rv, uap->oldlenp, sizeof(rv));
2266 * This is used from various compatibility syscalls too. That's why name
2267 * must be in kernel space.
2270 userland_sysctl(struct thread *td, int *name, u_int namelen, void *old,
2271 size_t *oldlenp, int inkernel, const void *new, size_t newlen,
2272 size_t *retval, int flags)
2274 int error = 0, memlocked;
2275 struct sysctl_req req;
2277 bzero(&req, sizeof req);
2284 req.oldlen = *oldlenp;
2286 error = copyin(oldlenp, &req.oldlen, sizeof(*oldlenp));
2291 req.validlen = req.oldlen;
2295 req.newlen = newlen;
2299 req.oldfunc = sysctl_old_user;
2300 req.newfunc = sysctl_new_user;
2301 req.lock = REQ_UNWIRED;
2304 if (KTRPOINT(curthread, KTR_SYSCTL))
2305 ktrsysctl(name, namelen);
2308 if (req.oldptr && req.oldlen > 4 * PAGE_SIZE) {
2310 sx_xlock(&sysctlmemlock);
2312 CURVNET_SET(TD_TO_VNET(td));
2317 error = sysctl_root(0, name, namelen, &req);
2318 if (error != EAGAIN)
2320 kern_yield(PRI_USER);
2325 if (req.lock == REQ_WIRED && req.validlen > 0)
2326 vsunlock(req.oldptr, req.validlen);
2328 sx_xunlock(&sysctlmemlock);
2330 if (error && error != ENOMEM)
2334 if (req.oldptr && req.oldidx > req.validlen)
2335 *retval = req.validlen;
2337 *retval = req.oldidx;
2343 * Drain into a sysctl struct. The user buffer should be wired if a page
2344 * fault would cause issue.
2347 sbuf_sysctl_drain(void *arg, const char *data, int len)
2349 struct sysctl_req *req = arg;
2352 error = SYSCTL_OUT(req, data, len);
2353 KASSERT(error >= 0, ("Got unexpected negative value %d", error));
2354 return (error == 0 ? len : -error);
2358 sbuf_new_for_sysctl(struct sbuf *s, char *buf, int length,
2359 struct sysctl_req *req)
2362 /* Supply a default buffer size if none given. */
2363 if (buf == NULL && length == 0)
2365 s = sbuf_new(s, buf, length, SBUF_FIXEDLEN | SBUF_INCLUDENUL);
2366 sbuf_set_drain(s, sbuf_sysctl_drain, req);
2372 /* The current OID the debugger is working with */
2373 static struct sysctl_oid *g_ddb_oid;
2375 /* The current flags specified by the user */
2376 static int g_ddb_sysctl_flags;
2378 /* Check to see if the last sysctl printed */
2379 static int g_ddb_sysctl_printed;
2381 static const int ctl_sign[CTLTYPE+1] = {
2390 static const int ctl_size[CTLTYPE+1] = {
2391 [CTLTYPE_INT] = sizeof(int),
2392 [CTLTYPE_UINT] = sizeof(u_int),
2393 [CTLTYPE_LONG] = sizeof(long),
2394 [CTLTYPE_ULONG] = sizeof(u_long),
2395 [CTLTYPE_S8] = sizeof(int8_t),
2396 [CTLTYPE_S16] = sizeof(int16_t),
2397 [CTLTYPE_S32] = sizeof(int32_t),
2398 [CTLTYPE_S64] = sizeof(int64_t),
2399 [CTLTYPE_U8] = sizeof(uint8_t),
2400 [CTLTYPE_U16] = sizeof(uint16_t),
2401 [CTLTYPE_U32] = sizeof(uint32_t),
2402 [CTLTYPE_U64] = sizeof(uint64_t),
2405 #define DB_SYSCTL_NAME_ONLY 0x001 /* Compare with -N */
2406 #define DB_SYSCTL_VALUE_ONLY 0x002 /* Compare with -n */
2407 #define DB_SYSCTL_OPAQUE 0x004 /* Compare with -o */
2408 #define DB_SYSCTL_HEX 0x008 /* Compare with -x */
2410 #define DB_SYSCTL_SAFE_ONLY 0x100 /* Only simple types */
2412 static const char db_sysctl_modifs[] = {
2416 static const int db_sysctl_modif_values[] = {
2417 DB_SYSCTL_NAME_ONLY, DB_SYSCTL_VALUE_ONLY,
2418 DB_SYSCTL_OPAQUE, DB_SYSCTL_HEX,
2421 /* Handlers considered safe to print while recursing */
2422 static int (* const db_safe_handlers[])(SYSCTL_HANDLER_ARGS) = {
2430 sysctl_handle_string,
2431 sysctl_handle_opaque,
2435 * Use in place of sysctl_old_kernel to print sysctl values.
2437 * Compare to the output handling in show_var from sbin/sysctl/sysctl.c
2440 sysctl_old_ddb(struct sysctl_req *req, const void *ptr, size_t len)
2442 const u_char *val, *p;
2444 size_t intlen, slen;
2447 int sign, ctltype, hexlen, xflag, error;
2449 /* Suppress false-positive GCC uninitialized variable warnings */
2461 /* We are going to print */
2462 g_ddb_sysctl_printed = 1;
2464 xflag = g_ddb_sysctl_flags & DB_SYSCTL_HEX;
2466 ctltype = (g_ddb_oid->oid_kind & CTLTYPE);
2467 sign = ctl_sign[ctltype];
2468 intlen = ctl_size[ctltype];
2472 case CTLTYPE_STRING:
2473 db_printf("%.*s", (int) len, (const char *) p);
2489 hexlen = 2 + (intlen * CHAR_BIT + 3) / 4;
2491 while (len >= intlen) {
2495 umv = *(const u_int *)p;
2496 mv = *(const int *)p;
2500 umv = *(const u_long *)p;
2501 mv = *(const long *)p;
2505 umv = *(const uint8_t *)p;
2506 mv = *(const int8_t *)p;
2510 umv = *(const uint16_t *)p;
2511 mv = *(const int16_t *)p;
2515 umv = *(const uint32_t *)p;
2516 mv = *(const int32_t *)p;
2520 umv = *(const uint64_t *)p;
2521 mv = *(const int64_t *)p;
2525 db_printf("%s", sep1);
2527 db_printf("%#0*jx", hexlen, umv);
2529 db_printf("%ju", umv);
2530 else if (g_ddb_oid->oid_fmt[1] == 'K') {
2531 /* Kelvins are currently unsupported. */
2535 db_printf("%jd", mv);
2544 case CTLTYPE_OPAQUE:
2545 /* TODO: Support struct functions. */
2549 db_printf("Format:%s Length:%zu Dump:0x",
2550 g_ddb_oid->oid_fmt, len);
2551 while (len-- && (xflag || p < val + 16))
2552 db_printf("%02x", *p++);
2553 if (!xflag && len > 16)
2560 req->oldidx += slen;
2565 * Avoid setting new sysctl values from the debugger
2568 sysctl_new_ddb(struct sysctl_req *req, void *p, size_t l)
2574 /* Changing sysctls from the debugger is currently unsupported */
2579 * Run a sysctl handler with the DDB oldfunc and newfunc attached.
2580 * Instead of copying any output to a buffer we'll dump it right to
2584 db_sysctl(struct sysctl_oid *oidp, int *name, u_int namelen,
2585 void *old, size_t *oldlenp, size_t *retval, int flags)
2587 struct sysctl_req req;
2590 /* Setup the request */
2591 bzero(&req, sizeof req);
2592 req.td = kdb_thread;
2593 req.oldfunc = sysctl_old_ddb;
2594 req.newfunc = sysctl_new_ddb;
2595 req.lock = REQ_UNWIRED;
2597 req.oldlen = *oldlenp;
2599 req.validlen = req.oldlen;
2604 /* Setup our globals for sysctl_old_ddb */
2606 g_ddb_sysctl_flags = flags;
2607 g_ddb_sysctl_printed = 0;
2609 error = sysctl_root(0, name, namelen, &req);
2613 g_ddb_sysctl_flags = 0;
2616 if (req.oldptr && req.oldidx > req.validlen)
2617 *retval = req.validlen;
2619 *retval = req.oldidx;
2625 * Show a sysctl's name
2628 db_show_oid_name(int *oid, size_t nlen)
2630 struct sysctl_oid *oidp;
2631 int qoid[CTL_MAXNAME+2];
2635 memcpy(qoid + 2, oid, nlen * sizeof(int));
2638 error = sysctl_find_oid(qoid, nlen + 2, &oidp, NULL, NULL);
2640 db_error("sysctl name oid");
2642 error = db_sysctl(oidp, qoid, nlen + 2, NULL, NULL, NULL, 0);
2644 db_error("sysctl name");
2648 * Check to see if an OID is safe to print from ddb.
2651 db_oid_safe(const struct sysctl_oid *oidp)
2653 for (unsigned int i = 0; i < nitems(db_safe_handlers); ++i) {
2654 if (oidp->oid_handler == db_safe_handlers[i])
2662 * Show a sysctl at a specific OID
2663 * Compare to the input handling in show_var from sbin/sysctl/sysctl.c
2666 db_show_oid(struct sysctl_oid *oidp, int *oid, size_t nlen, int flags)
2668 int error, xflag, oflag, Nflag, nflag;
2671 xflag = flags & DB_SYSCTL_HEX;
2672 oflag = flags & DB_SYSCTL_OPAQUE;
2673 nflag = flags & DB_SYSCTL_VALUE_ONLY;
2674 Nflag = flags & DB_SYSCTL_NAME_ONLY;
2676 if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_OPAQUE &&
2681 db_show_oid_name(oid, nlen);
2687 db_show_oid_name(oid, nlen);
2691 if ((flags & DB_SYSCTL_SAFE_ONLY) && !db_oid_safe(oidp)) {
2692 db_printf("Skipping, unsafe to print while recursing.");
2697 /* Try once, and ask about the size */
2699 error = db_sysctl(oidp, oid, nlen,
2700 NULL, NULL, &len, flags);
2704 if (!g_ddb_sysctl_printed)
2705 /* Lie about the size */
2706 error = db_sysctl(oidp, oid, nlen,
2707 (void *) 1, &len, NULL, flags);
2715 * Show all sysctls under a specific OID
2716 * Compare to sysctl_all from sbin/sysctl/sysctl.c
2719 db_show_sysctl_all(int *oid, size_t len, int flags)
2721 struct sysctl_oid *oidp;
2722 int name1[CTL_MAXNAME + 2], name2[CTL_MAXNAME + 2];
2725 name1[0] = CTL_SYSCTL;
2726 name1[1] = CTL_SYSCTL_NEXT;
2729 memcpy(name1+2, oid, len * sizeof(int));
2739 error = kernel_sysctl(kdb_thread, name1, l1,
2740 name2, &l2, NULL, 0, &l2, 0);
2742 if (error == ENOENT)
2745 db_error("sysctl(getnext)");
2750 if (l2 < (unsigned int)len)
2753 for (i = 0; i < len; i++)
2754 if (name2[i] != oid[i])
2757 /* Find the OID in question */
2758 error = sysctl_find_oid(name2, l2, &oidp, NULL, NULL);
2762 i = db_show_oid(oidp, name2, l2, flags | DB_SYSCTL_SAFE_ONLY);
2767 memcpy(name1+2, name2, l2 * sizeof(int));
2773 * Show a sysctl by its user facing string
2776 db_sysctlbyname(char *name, int flags)
2778 struct sysctl_oid *oidp;
2779 int oid[CTL_MAXNAME];
2782 error = name2oid(name, oid, &nlen, &oidp);
2787 if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
2788 db_show_sysctl_all(oid, nlen, flags);
2790 error = db_show_oid(oidp, oid, nlen, flags);
2797 db_sysctl_cmd_usage(void)
2800 " sysctl [/Nnox] <sysctl> \n"
2802 " <sysctl> The name of the sysctl to show. \n"
2804 " Show a sysctl by hooking into SYSCTL_IN and SYSCTL_OUT. \n"
2805 " This will work for most sysctls, but should not be used \n"
2806 " with sysctls that are known to malloc. \n"
2808 " While recursing any \"unsafe\" sysctls will be skipped. \n"
2809 " Call sysctl directly on the sysctl to try printing the \n"
2810 " skipped sysctl. This is unsafe and may make the ddb \n"
2811 " session unusable. \n"
2814 " /N Display only the name of the sysctl. \n"
2815 " /n Display only the value of the sysctl. \n"
2816 " /o Display opaque values. \n"
2817 " /x Display the sysctl in hex. \n"
2820 "sysctl vm.v_free_min \n"
2821 "vn.v_free_min: 12669 \n"
2826 * Show a specific sysctl similar to sysctl (8).
2828 DB_FUNC(sysctl, db_sysctl_cmd, db_cmd_table, CS_OWN, NULL)
2830 char name[TOK_STRING_SIZE];
2831 int error, i, t, flags;
2833 /* Parse the modifiers */
2834 t = db_read_token();
2835 if (t == tSLASH || t == tMINUS) {
2836 t = db_read_token();
2838 db_printf("Bad modifier\n");
2842 db_strcpy(modif, db_tok_string);
2850 for (i = 0; i < nitems(db_sysctl_modifs); i++) {
2851 if (strchr(modif, db_sysctl_modifs[i])) {
2852 flags |= db_sysctl_modif_values[i];
2856 /* Parse the sysctl names */
2857 t = db_read_token();
2859 db_printf("Need sysctl name\n");
2864 /* Copy the name into a temporary buffer */
2865 db_strcpy(name, db_tok_string);
2867 /* Ensure there is no trailing cruft */
2868 t = db_read_token();
2870 db_printf("Unexpected sysctl argument\n");
2875 error = db_sysctlbyname(name, flags);
2876 if (error == ENOENT) {
2877 db_printf("unknown oid: '%s'\n", db_tok_string);
2880 db_printf("%s: error: %d\n", db_tok_string, error);
2885 /* Ensure we eat all of our text */
2888 if (error == EINVAL) {
2889 db_sysctl_cmd_usage();