freebsd-skq/sys/kern/kern_sysctl.c

1783 lines
40 KiB
C
Raw Normal View History

1994-05-24 10:09:53 +00:00
/*-
* Copyright (c) 1982, 1986, 1989, 1993
* The Regents of the University of California. All rights reserved.
*
* This code is derived from software contributed to Berkeley by
* Mike Karels at Berkeley Software Design, Inc.
*
* Quite extensively rewritten by Poul-Henning Kamp of the FreeBSD
* project, to make these variables more userfriendly.
*
1994-05-24 10:09:53 +00:00
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 4. Neither the name of the University nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
* @(#)kern_sysctl.c 8.4 (Berkeley) 4/14/94
*/
2003-06-11 00:56:59 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include "opt_capsicum.h"
#include "opt_compat.h"
#include "opt_ktrace.h"
1994-05-24 10:09:53 +00:00
#include <sys/param.h>
#include <sys/fail.h>
#include <sys/systm.h>
#include <sys/capsicum.h>
1994-05-24 10:09:53 +00:00
#include <sys/kernel.h>
#include <sys/sysctl.h>
#include <sys/malloc.h>
#include <sys/priv.h>
#include <sys/proc.h>
#include <sys/jail.h>
#include <sys/lock.h>
#include <sys/mutex.h>
#include <sys/sbuf.h>
#include <sys/sx.h>
#include <sys/sysproto.h>
#include <sys/uio.h>
#ifdef KTRACE
#include <sys/ktrace.h>
#endif
Build on Jeff Roberson's linker-set based dynamic per-CPU allocator (DPCPU), as suggested by Peter Wemm, and implement a new per-virtual network stack memory allocator. Modify vnet to use the allocator instead of monolithic global container structures (vinet, ...). This change solves many binary compatibility problems associated with VIMAGE, and restores ELF symbols for virtualized global variables. Each virtualized global variable exists as a "reference copy", and also once per virtual network stack. Virtualized global variables are tagged at compile-time, placing the in a special linker set, which is loaded into a contiguous region of kernel memory. Virtualized global variables in the base kernel are linked as normal, but those in modules are copied and relocated to a reserved portion of the kernel's vnet region with the help of a the kernel linker. Virtualized global variables exist in per-vnet memory set up when the network stack instance is created, and are initialized statically from the reference copy. Run-time access occurs via an accessor macro, which converts from the current vnet and requested symbol to a per-vnet address. When "options VIMAGE" is not compiled into the kernel, normal global ELF symbols will be used instead and indirection is avoided. This change restores static initialization for network stack global variables, restores support for non-global symbols and types, eliminates the need for many subsystem constructors, eliminates large per-subsystem structures that caused many binary compatibility issues both for monitoring applications (netstat) and kernel modules, removes the per-function INIT_VNET_*() macros throughout the stack, eliminates the need for vnet_symmap ksym(2) munging, and eliminates duplicate definitions of virtualized globals under VIMAGE_GLOBALS. Bump __FreeBSD_version and update UPDATING. Portions submitted by: bz Reviewed by: bz, zec Discussed with: gnn, jamie, jeff, jhb, julian, sam Suggested by: peter Approved by: re (kensmith)
2009-07-14 22:48:30 +00:00
#include <net/vnet.h>
#include <security/mac/mac_framework.h>
#include <vm/vm.h>
#include <vm/vm_extern.h>
static MALLOC_DEFINE(M_SYSCTL, "sysctl", "sysctl internal magic");
static MALLOC_DEFINE(M_SYSCTLOID, "sysctloid", "sysctl dynamic oids");
static MALLOC_DEFINE(M_SYSCTLTMP, "sysctltmp", "sysctl temp output buffer");
/*
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
* The sysctllock protects the MIB tree. It also protects sysctl
* contexts used with dynamic sysctls. The sysctl_register_oid() and
* sysctl_unregister_oid() routines require the sysctllock to already
* be held, so the sysctl_lock() and sysctl_unlock() routines are
* provided for the few places in the kernel which need to use that
* API rather than using the dynamic API. Use of the dynamic API is
* strongly encouraged for most code.
*
* The sysctlmemlock is used to limit the amount of user memory wired for
* sysctl requests. This is implemented by serializing any userland
* sysctl requests larger than a single page via an exclusive lock.
*/
static struct sx sysctllock;
static struct sx sysctlmemlock;
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
#define SYSCTL_XLOCK() sx_xlock(&sysctllock)
#define SYSCTL_XUNLOCK() sx_xunlock(&sysctllock)
#define SYSCTL_ASSERT_XLOCKED() sx_assert(&sysctllock, SA_XLOCKED)
#define SYSCTL_INIT() sx_init(&sysctllock, "sysctl lock")
#define SYSCTL_SLEEP(ch, wmesg, timo) \
sx_sleep(ch, &sysctllock, 0, wmesg, timo)
static int sysctl_root(SYSCTL_HANDLER_ARGS);
/* Root list */
struct sysctl_oid_list sysctl__children = SLIST_HEAD_INITIALIZER(&sysctl__children);
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
static int sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del,
int recurse);
static int sysctl_old_kernel(struct sysctl_req *, const void *, size_t);
static int sysctl_new_kernel(struct sysctl_req *, void *, size_t);
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
static struct sysctl_oid *
sysctl_find_oidname(const char *name, struct sysctl_oid_list *list)
{
struct sysctl_oid *oidp;
SYSCTL_ASSERT_XLOCKED();
SLIST_FOREACH(oidp, list, oid_link) {
if (strcmp(oidp->oid_name, name) == 0) {
return (oidp);
}
}
return (NULL);
}
/*
* Initialization of the MIB tree.
*
* Order by number in each list.
*/
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
void
sysctl_lock(void)
{
SYSCTL_XLOCK();
}
void
sysctl_unlock(void)
{
SYSCTL_XUNLOCK();
}
static int
sysctl_root_handler_locked(struct sysctl_oid *oid, void *arg1, intptr_t arg2,
struct sysctl_req *req)
{
int error;
oid->oid_running++;
SYSCTL_XUNLOCK();
if (!(oid->oid_kind & CTLFLAG_MPSAFE))
mtx_lock(&Giant);
error = oid->oid_handler(oid, arg1, arg2, req);
if (!(oid->oid_kind & CTLFLAG_MPSAFE))
mtx_unlock(&Giant);
SYSCTL_XLOCK();
oid->oid_running--;
if (oid->oid_running == 0 && (oid->oid_kind & CTLFLAG_DYING) != 0)
wakeup(&oid->oid_running);
return (error);
}
static void
sysctl_load_tunable_by_oid_locked(struct sysctl_oid *oidp)
{
struct sysctl_req req;
struct sysctl_oid *curr;
char *penv = NULL;
char path[64];
ssize_t rem = sizeof(path);
ssize_t len;
int val_int;
long val_long;
int64_t val_64;
quad_t val_quad;
int error;
path[--rem] = 0;
for (curr = oidp; curr != NULL; curr = SYSCTL_PARENT(curr)) {
len = strlen(curr->oid_name);
rem -= len;
if (curr != oidp)
rem -= 1;
if (rem < 0) {
printf("OID path exceeds %d bytes\n", (int)sizeof(path));
return;
}
memcpy(path + rem, curr->oid_name, len);
if (curr != oidp)
path[rem + len] = '.';
}
memset(&req, 0, sizeof(req));
req.td = curthread;
req.oldfunc = sysctl_old_kernel;
req.newfunc = sysctl_new_kernel;
req.lock = REQ_UNWIRED;
switch (oidp->oid_kind & CTLTYPE) {
case CTLTYPE_INT:
if (getenv_int(path + rem, &val_int) == 0)
return;
req.newlen = sizeof(val_int);
req.newptr = &val_int;
break;
case CTLTYPE_UINT:
if (getenv_uint(path + rem, (unsigned int *)&val_int) == 0)
return;
req.newlen = sizeof(val_int);
req.newptr = &val_int;
break;
case CTLTYPE_LONG:
if (getenv_long(path + rem, &val_long) == 0)
return;
req.newlen = sizeof(val_long);
req.newptr = &val_long;
break;
case CTLTYPE_ULONG:
if (getenv_ulong(path + rem, (unsigned long *)&val_long) == 0)
return;
req.newlen = sizeof(val_long);
req.newptr = &val_long;
break;
case CTLTYPE_S64:
if (getenv_quad(path + rem, &val_quad) == 0)
return;
val_64 = val_quad;
req.newlen = sizeof(val_64);
req.newptr = &val_64;
break;
case CTLTYPE_U64:
/* XXX there is no getenv_uquad() */
if (getenv_quad(path + rem, &val_quad) == 0)
return;
val_64 = val_quad;
req.newlen = sizeof(val_64);
req.newptr = &val_64;
break;
case CTLTYPE_STRING:
penv = getenv(path + rem);
if (penv == NULL)
return;
req.newlen = strlen(penv);
req.newptr = penv;
break;
default:
return;
}
error = sysctl_root_handler_locked(oidp, oidp->oid_arg1,
oidp->oid_arg2, &req);
if (error != 0) {
printf("Setting sysctl '%s' to '%s' failed: %d\n",
path, penv, error);
}
freeenv(penv);
}
void
sysctl_register_oid(struct sysctl_oid *oidp)
{
struct sysctl_oid_list *parent = oidp->oid_parent;
struct sysctl_oid *p;
struct sysctl_oid *q;
/*
* First check if another oid with the same name already
* exists in the parent's list.
*/
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
SYSCTL_ASSERT_XLOCKED();
p = sysctl_find_oidname(oidp->oid_name, parent);
if (p != NULL) {
if ((p->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
p->oid_refcnt++;
return;
} else {
printf("can't re-use a leaf (%s)!\n", p->oid_name);
return;
}
}
/*
* If this oid has a number OID_AUTO, give it a number which
* is greater than any current oid.
* NOTE: DO NOT change the starting value here, change it in
* <sys/sysctl.h>, and make sure it is at least 256 to
* accomodate e.g. net.inet.raw as a static sysctl node.
*/
if (oidp->oid_number == OID_AUTO) {
static int newoid = CTL_AUTO_START;
oidp->oid_number = newoid++;
if (newoid == 0x7fffffff)
panic("out of oids");
}
#if 0
else if (oidp->oid_number >= CTL_AUTO_START) {
/* do not panic; this happens when unregistering sysctl sets */
printf("static sysctl oid too high: %d", oidp->oid_number);
}
#endif
/*
* Insert the oid into the parent's list in order.
*/
q = NULL;
SLIST_FOREACH(p, parent, oid_link) {
if (oidp->oid_number < p->oid_number)
break;
q = p;
}
if (q)
SLIST_INSERT_AFTER(q, oidp, oid_link);
else
SLIST_INSERT_HEAD(parent, oidp, oid_link);
if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE &&
#ifdef VIMAGE
(oidp->oid_kind & CTLFLAG_VNET) == 0 &&
#endif
(oidp->oid_kind & CTLFLAG_TUN) != 0 &&
(oidp->oid_kind & CTLFLAG_NOFETCH) == 0) {
sysctl_load_tunable_by_oid_locked(oidp);
}
}
void
sysctl_unregister_oid(struct sysctl_oid *oidp)
{
struct sysctl_oid *p;
int error;
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
SYSCTL_ASSERT_XLOCKED();
error = ENOENT;
if (oidp->oid_number == OID_AUTO) {
error = EINVAL;
} else {
SLIST_FOREACH(p, oidp->oid_parent, oid_link) {
if (p == oidp) {
SLIST_REMOVE(oidp->oid_parent, oidp,
sysctl_oid, oid_link);
error = 0;
break;
}
}
}
/*
* This can happen when a module fails to register and is
* being unloaded afterwards. It should not be a panic()
* for normal use.
*/
if (error)
printf("%s: failed to unregister sysctl\n", __func__);
}
/* Initialize a new context to keep track of dynamically added sysctls. */
int
sysctl_ctx_init(struct sysctl_ctx_list *c)
{
if (c == NULL) {
return (EINVAL);
}
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
/*
* No locking here, the caller is responsible for not adding
* new nodes to a context until after this function has
* returned.
*/
TAILQ_INIT(c);
return (0);
}
/* Free the context, and destroy all dynamic oids registered in this context */
int
sysctl_ctx_free(struct sysctl_ctx_list *clist)
{
struct sysctl_ctx_entry *e, *e1;
int error;
error = 0;
/*
* First perform a "dry run" to check if it's ok to remove oids.
* XXX FIXME
* XXX This algorithm is a hack. But I don't know any
* XXX better solution for now...
*/
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
SYSCTL_XLOCK();
TAILQ_FOREACH(e, clist, link) {
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
error = sysctl_remove_oid_locked(e->entry, 0, 0);
if (error)
break;
}
/*
* Restore deregistered entries, either from the end,
* or from the place where error occured.
* e contains the entry that was not unregistered
*/
if (error)
e1 = TAILQ_PREV(e, sysctl_ctx_list, link);
else
e1 = TAILQ_LAST(clist, sysctl_ctx_list);
while (e1 != NULL) {
sysctl_register_oid(e1->entry);
e1 = TAILQ_PREV(e1, sysctl_ctx_list, link);
}
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
if (error) {
SYSCTL_XUNLOCK();
return(EBUSY);
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
}
/* Now really delete the entries */
e = TAILQ_FIRST(clist);
while (e != NULL) {
e1 = TAILQ_NEXT(e, link);
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
error = sysctl_remove_oid_locked(e->entry, 1, 0);
if (error)
panic("sysctl_remove_oid: corrupt tree, entry: %s",
e->entry->oid_name);
free(e, M_SYSCTLOID);
e = e1;
}
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
SYSCTL_XUNLOCK();
return (error);
}
/* Add an entry to the context */
struct sysctl_ctx_entry *
sysctl_ctx_entry_add(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
{
struct sysctl_ctx_entry *e;
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
SYSCTL_ASSERT_XLOCKED();
if (clist == NULL || oidp == NULL)
return(NULL);
e = malloc(sizeof(struct sysctl_ctx_entry), M_SYSCTLOID, M_WAITOK);
e->entry = oidp;
TAILQ_INSERT_HEAD(clist, e, link);
return (e);
}
/* Find an entry in the context */
struct sysctl_ctx_entry *
sysctl_ctx_entry_find(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
{
struct sysctl_ctx_entry *e;
SYSCTL_ASSERT_XLOCKED();
if (clist == NULL || oidp == NULL)
return(NULL);
TAILQ_FOREACH(e, clist, link) {
if(e->entry == oidp)
return(e);
}
return (e);
}
/*
* Delete an entry from the context.
* NOTE: this function doesn't free oidp! You have to remove it
* with sysctl_remove_oid().
*/
int
sysctl_ctx_entry_del(struct sysctl_ctx_list *clist, struct sysctl_oid *oidp)
{
struct sysctl_ctx_entry *e;
if (clist == NULL || oidp == NULL)
return (EINVAL);
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
SYSCTL_XLOCK();
e = sysctl_ctx_entry_find(clist, oidp);
if (e != NULL) {
TAILQ_REMOVE(clist, e, link);
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
SYSCTL_XUNLOCK();
free(e, M_SYSCTLOID);
return (0);
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
} else {
SYSCTL_XUNLOCK();
return (ENOENT);
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
}
}
/*
* Remove dynamically created sysctl trees.
* oidp - top of the tree to be removed
* del - if 0 - just deregister, otherwise free up entries as well
* recurse - if != 0 traverse the subtree to be deleted
*/
int
sysctl_remove_oid(struct sysctl_oid *oidp, int del, int recurse)
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
{
int error;
SYSCTL_XLOCK();
error = sysctl_remove_oid_locked(oidp, del, recurse);
SYSCTL_XUNLOCK();
return (error);
}
int
sysctl_remove_name(struct sysctl_oid *parent, const char *name,
int del, int recurse)
{
struct sysctl_oid *p, *tmp;
int error;
error = ENOENT;
SYSCTL_XLOCK();
SLIST_FOREACH_SAFE(p, SYSCTL_CHILDREN(parent), oid_link, tmp) {
if (strcmp(p->oid_name, name) == 0) {
error = sysctl_remove_oid_locked(p, del, recurse);
break;
}
}
SYSCTL_XUNLOCK();
return (error);
}
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
static int
sysctl_remove_oid_locked(struct sysctl_oid *oidp, int del, int recurse)
{
struct sysctl_oid *p, *tmp;
int error;
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
SYSCTL_ASSERT_XLOCKED();
if (oidp == NULL)
return(EINVAL);
if ((oidp->oid_kind & CTLFLAG_DYN) == 0) {
printf("can't remove non-dynamic nodes!\n");
return (EINVAL);
}
/*
* WARNING: normal method to do this should be through
* sysctl_ctx_free(). Use recursing as the last resort
* method to purge your sysctl tree of leftovers...
* However, if some other code still references these nodes,
* it will panic.
*/
if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
if (oidp->oid_refcnt == 1) {
SLIST_FOREACH_SAFE(p,
SYSCTL_CHILDREN(oidp), oid_link, tmp) {
if (!recurse) {
printf("Warning: failed attempt to "
"remove oid %s with child %s\n",
oidp->oid_name, p->oid_name);
return (ENOTEMPTY);
}
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
error = sysctl_remove_oid_locked(p, del,
recurse);
if (error)
return (error);
}
}
}
if (oidp->oid_refcnt > 1 ) {
oidp->oid_refcnt--;
} else {
if (oidp->oid_refcnt == 0) {
printf("Warning: bad oid_refcnt=%u (%s)!\n",
oidp->oid_refcnt, oidp->oid_name);
return (EINVAL);
}
sysctl_unregister_oid(oidp);
if (del) {
/*
* Wait for all threads running the handler to drain.
* This preserves the previous behavior when the
* sysctl lock was held across a handler invocation,
* and is necessary for module unload correctness.
*/
while (oidp->oid_running > 0) {
oidp->oid_kind |= CTLFLAG_DYING;
SYSCTL_SLEEP(&oidp->oid_running, "oidrm", 0);
}
if (oidp->oid_descr)
free(__DECONST(char *, oidp->oid_descr),
M_SYSCTLOID);
free(__DECONST(char *, oidp->oid_name), M_SYSCTLOID);
free(oidp, M_SYSCTLOID);
}
}
return (0);
}
/*
* Create new sysctls at run time.
* clist may point to a valid context initialized with sysctl_ctx_init().
*/
struct sysctl_oid *
sysctl_add_oid(struct sysctl_ctx_list *clist, struct sysctl_oid_list *parent,
int number, const char *name, int kind, void *arg1, intptr_t arg2,
int (*handler)(SYSCTL_HANDLER_ARGS), const char *fmt, const char *descr)
{
struct sysctl_oid *oidp;
/* You have to hook up somewhere.. */
if (parent == NULL)
return(NULL);
/* Check if the node already exists, otherwise create it */
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
SYSCTL_XLOCK();
oidp = sysctl_find_oidname(name, parent);
if (oidp != NULL) {
if ((oidp->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
oidp->oid_refcnt++;
/* Update the context */
if (clist != NULL)
sysctl_ctx_entry_add(clist, oidp);
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
SYSCTL_XUNLOCK();
return (oidp);
} else {
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
SYSCTL_XUNLOCK();
printf("can't re-use a leaf (%s)!\n", name);
return (NULL);
}
}
oidp = malloc(sizeof(struct sysctl_oid), M_SYSCTLOID, M_WAITOK|M_ZERO);
oidp->oid_parent = parent;
SLIST_INIT(&oidp->oid_children);
oidp->oid_number = number;
oidp->oid_refcnt = 1;
oidp->oid_name = strdup(name, M_SYSCTLOID);
oidp->oid_handler = handler;
oidp->oid_kind = CTLFLAG_DYN | kind;
oidp->oid_arg1 = arg1;
oidp->oid_arg2 = arg2;
oidp->oid_fmt = fmt;
if (descr != NULL)
oidp->oid_descr = strdup(descr, M_SYSCTLOID);
/* Update the context, if used */
if (clist != NULL)
sysctl_ctx_entry_add(clist, oidp);
/* Register this oid */
sysctl_register_oid(oidp);
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
SYSCTL_XUNLOCK();
return (oidp);
}
/*
* Rename an existing oid.
*/
void
sysctl_rename_oid(struct sysctl_oid *oidp, const char *name)
{
char *newname;
char *oldname;
newname = strdup(name, M_SYSCTLOID);
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
SYSCTL_XLOCK();
oldname = __DECONST(char *, oidp->oid_name);
oidp->oid_name = newname;
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
SYSCTL_XUNLOCK();
free(oldname, M_SYSCTLOID);
}
/*
* Reparent an existing oid.
*/
int
sysctl_move_oid(struct sysctl_oid *oid, struct sysctl_oid_list *parent)
{
struct sysctl_oid *oidp;
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
SYSCTL_XLOCK();
if (oid->oid_parent == parent) {
SYSCTL_XUNLOCK();
return (0);
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
}
oidp = sysctl_find_oidname(oid->oid_name, parent);
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
if (oidp != NULL) {
SYSCTL_XUNLOCK();
return (EEXIST);
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
}
sysctl_unregister_oid(oid);
oid->oid_parent = parent;
oid->oid_number = OID_AUTO;
sysctl_register_oid(oid);
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
SYSCTL_XUNLOCK();
return (0);
}
/*
* Register the kernel's oids on startup.
*/
SET_DECLARE(sysctl_set, struct sysctl_oid);
static void
sysctl_register_all(void *arg)
{
struct sysctl_oid **oidp;
sx_init(&sysctlmemlock, "sysctl mem");
SYSCTL_INIT();
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
SYSCTL_XLOCK();
SET_FOREACH(oidp, sysctl_set)
sysctl_register_oid(*oidp);
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
SYSCTL_XUNLOCK();
}
SYSINIT(sysctl, SI_SUB_KMEM, SI_ORDER_FIRST, sysctl_register_all, 0);
/*
* "Staff-functions"
*
* These functions implement a presently undocumented interface
* used by the sysctl program to walk the tree, and get the type
* so it can print the value.
* This interface is under work and consideration, and should probably
* be killed with a big axe by the first person who can find the time.
* (be aware though, that the proper interface isn't as obvious as it
* may seem, there are various conflicting requirements.
*
* {0,0} printf the entire MIB-tree.
* {0,1,...} return the name of the "..." OID.
* {0,2,...} return the next OID.
* {0,3} return the OID of the name in "new"
* {0,4,...} return the kind & format info for the "..." OID.
* {0,5,...} return the description the "..." OID.
*/
#ifdef SYSCTL_DEBUG
static void
sysctl_sysctl_debug_dump_node(struct sysctl_oid_list *l, int i)
{
int k;
struct sysctl_oid *oidp;
SYSCTL_ASSERT_XLOCKED();
SLIST_FOREACH(oidp, l, oid_link) {
for (k=0; k<i; k++)
printf(" ");
printf("%d %s ", oidp->oid_number, oidp->oid_name);
printf("%c%c",
oidp->oid_kind & CTLFLAG_RD ? 'R':' ',
oidp->oid_kind & CTLFLAG_WR ? 'W':' ');
if (oidp->oid_handler)
printf(" *Handler");
switch (oidp->oid_kind & CTLTYPE) {
case CTLTYPE_NODE:
printf(" Node\n");
if (!oidp->oid_handler) {
sysctl_sysctl_debug_dump_node(
SYSCTL_CHILDREN(oidp), i + 2);
}
break;
case CTLTYPE_INT: printf(" Int\n"); break;
case CTLTYPE_UINT: printf(" u_int\n"); break;
case CTLTYPE_LONG: printf(" Long\n"); break;
case CTLTYPE_ULONG: printf(" u_long\n"); break;
case CTLTYPE_STRING: printf(" String\n"); break;
case CTLTYPE_U64: printf(" uint64_t\n"); break;
case CTLTYPE_S64: printf(" int64_t\n"); break;
case CTLTYPE_OPAQUE: printf(" Opaque/struct\n"); break;
default: printf("\n");
}
}
}
static int
sysctl_sysctl_debug(SYSCTL_HANDLER_ARGS)
{
int error;
error = priv_check(req->td, PRIV_SYSCTL_DEBUG);
if (error)
return (error);
SYSCTL_XLOCK();
sysctl_sysctl_debug_dump_node(&sysctl__children, 0);
SYSCTL_XUNLOCK();
return (ENOENT);
}
SYSCTL_PROC(_sysctl, 0, debug, CTLTYPE_STRING|CTLFLAG_RD,
0, 0, sysctl_sysctl_debug, "-", "");
#endif
static int
sysctl_sysctl_name(SYSCTL_HANDLER_ARGS)
{
int *name = (int *) arg1;
u_int namelen = arg2;
int error = 0;
struct sysctl_oid *oid;
struct sysctl_oid_list *lsp = &sysctl__children, *lsp2;
char buf[10];
SYSCTL_XLOCK();
while (namelen) {
if (!lsp) {
snprintf(buf,sizeof(buf),"%d",*name);
if (req->oldidx)
error = SYSCTL_OUT(req, ".", 1);
if (!error)
error = SYSCTL_OUT(req, buf, strlen(buf));
if (error)
goto out;
namelen--;
name++;
continue;
}
lsp2 = 0;
SLIST_FOREACH(oid, lsp, oid_link) {
if (oid->oid_number != *name)
continue;
if (req->oldidx)
error = SYSCTL_OUT(req, ".", 1);
if (!error)
error = SYSCTL_OUT(req, oid->oid_name,
strlen(oid->oid_name));
if (error)
goto out;
namelen--;
name++;
if ((oid->oid_kind & CTLTYPE) != CTLTYPE_NODE)
break;
if (oid->oid_handler)
break;
lsp2 = SYSCTL_CHILDREN(oid);
break;
}
lsp = lsp2;
}
error = SYSCTL_OUT(req, "", 1);
out:
SYSCTL_XUNLOCK();
return (error);
}
/*
* XXXRW/JA: Shouldn't return name data for nodes that we don't permit in
* capability mode.
*/
static SYSCTL_NODE(_sysctl, 1, name, CTLFLAG_RD | CTLFLAG_CAPRD,
sysctl_sysctl_name, "");
static int
sysctl_sysctl_next_ls(struct sysctl_oid_list *lsp, int *name, u_int namelen,
int *next, int *len, int level, struct sysctl_oid **oidpp)
{
struct sysctl_oid *oidp;
SYSCTL_ASSERT_XLOCKED();
*len = level;
SLIST_FOREACH(oidp, lsp, oid_link) {
*next = oidp->oid_number;
*oidpp = oidp;
if (oidp->oid_kind & CTLFLAG_SKIP)
continue;
if (!namelen) {
if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
return (0);
if (oidp->oid_handler)
/* We really should call the handler here...*/
return (0);
lsp = SYSCTL_CHILDREN(oidp);
if (!sysctl_sysctl_next_ls(lsp, 0, 0, next+1,
len, level+1, oidpp))
return (0);
goto emptynode;
}
if (oidp->oid_number < *name)
continue;
if (oidp->oid_number > *name) {
if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
return (0);
if (oidp->oid_handler)
return (0);
lsp = SYSCTL_CHILDREN(oidp);
if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1,
next+1, len, level+1, oidpp))
return (0);
goto next;
}
if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
continue;
if (oidp->oid_handler)
continue;
lsp = SYSCTL_CHILDREN(oidp);
if (!sysctl_sysctl_next_ls(lsp, name+1, namelen-1, next+1,
len, level+1, oidpp))
return (0);
next:
namelen = 1;
emptynode:
*len = level;
}
return (1);
}
static int
sysctl_sysctl_next(SYSCTL_HANDLER_ARGS)
{
int *name = (int *) arg1;
u_int namelen = arg2;
int i, j, error;
struct sysctl_oid *oid;
struct sysctl_oid_list *lsp = &sysctl__children;
int newoid[CTL_MAXNAME];
SYSCTL_XLOCK();
i = sysctl_sysctl_next_ls(lsp, name, namelen, newoid, &j, 1, &oid);
SYSCTL_XUNLOCK();
if (i)
return (ENOENT);
error = SYSCTL_OUT(req, newoid, j * sizeof (int));
return (error);
}
/*
* XXXRW/JA: Shouldn't return next data for nodes that we don't permit in
* capability mode.
*/
static SYSCTL_NODE(_sysctl, 2, next, CTLFLAG_RD | CTLFLAG_CAPRD,
sysctl_sysctl_next, "");
static int
name2oid(char *name, int *oid, int *len, struct sysctl_oid **oidpp)
{
struct sysctl_oid *oidp;
struct sysctl_oid_list *lsp = &sysctl__children;
char *p;
SYSCTL_ASSERT_XLOCKED();
for (*len = 0; *len < CTL_MAXNAME;) {
p = strsep(&name, ".");
oidp = SLIST_FIRST(lsp);
for (;; oidp = SLIST_NEXT(oidp, oid_link)) {
if (oidp == NULL)
return (ENOENT);
if (strcmp(p, oidp->oid_name) == 0)
break;
}
*oid++ = oidp->oid_number;
(*len)++;
if (name == NULL || *name == '\0') {
if (oidpp)
*oidpp = oidp;
return (0);
}
if ((oidp->oid_kind & CTLTYPE) != CTLTYPE_NODE)
break;
if (oidp->oid_handler)
break;
lsp = SYSCTL_CHILDREN(oidp);
}
return (ENOENT);
}
static int
sysctl_sysctl_name2oid(SYSCTL_HANDLER_ARGS)
{
char *p;
int error, oid[CTL_MAXNAME], len = 0;
struct sysctl_oid *op = 0;
if (!req->newlen)
return (ENOENT);
1999-03-30 09:00:45 +00:00
if (req->newlen >= MAXPATHLEN) /* XXX arbitrary, undocumented */
return (ENAMETOOLONG);
p = malloc(req->newlen+1, M_SYSCTL, M_WAITOK);
error = SYSCTL_IN(req, p, req->newlen);
if (error) {
free(p, M_SYSCTL);
return (error);
}
p [req->newlen] = '\0';
SYSCTL_XLOCK();
error = name2oid(p, oid, &len, &op);
SYSCTL_XUNLOCK();
free(p, M_SYSCTL);
if (error)
return (error);
error = SYSCTL_OUT(req, oid, len * sizeof *oid);
return (error);
}
/*
* XXXRW/JA: Shouldn't return name2oid data for nodes that we don't permit in
* capability mode.
*/
SYSCTL_PROC(_sysctl, 3, name2oid,
CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_MPSAFE
| CTLFLAG_CAPRW, 0, 0, sysctl_sysctl_name2oid, "I", "");
static int
sysctl_sysctl_oidfmt(SYSCTL_HANDLER_ARGS)
{
struct sysctl_oid *oid;
int error;
SYSCTL_XLOCK();
error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
if (error)
goto out;
if (oid->oid_fmt == NULL) {
error = ENOENT;
goto out;
}
error = SYSCTL_OUT(req, &oid->oid_kind, sizeof(oid->oid_kind));
if (error)
goto out;
error = SYSCTL_OUT(req, oid->oid_fmt, strlen(oid->oid_fmt) + 1);
out:
SYSCTL_XUNLOCK();
return (error);
}
static SYSCTL_NODE(_sysctl, 4, oidfmt, CTLFLAG_RD|CTLFLAG_MPSAFE|CTLFLAG_CAPRD,
sysctl_sysctl_oidfmt, "");
static int
sysctl_sysctl_oiddescr(SYSCTL_HANDLER_ARGS)
{
struct sysctl_oid *oid;
int error;
SYSCTL_XLOCK();
error = sysctl_find_oid(arg1, arg2, &oid, NULL, req);
if (error)
goto out;
if (oid->oid_descr == NULL) {
error = ENOENT;
goto out;
}
error = SYSCTL_OUT(req, oid->oid_descr, strlen(oid->oid_descr) + 1);
out:
SYSCTL_XUNLOCK();
return (error);
}
static SYSCTL_NODE(_sysctl, 5, oiddescr, CTLFLAG_RD|CTLFLAG_CAPRD,
sysctl_sysctl_oiddescr, "");
/*
* Default "handler" functions.
*/
/*
* Handle an int, signed or unsigned.
* Two cases:
* a variable: point arg1 at it.
* a constant: pass it in arg2.
*/
int
sysctl_handle_int(SYSCTL_HANDLER_ARGS)
{
int tmpout, error = 0;
/*
* Attempt to get a coherent snapshot by making a copy of the data.
*/
if (arg1)
tmpout = *(int *)arg1;
else
tmpout = arg2;
error = SYSCTL_OUT(req, &tmpout, sizeof(int));
if (error || !req->newptr)
return (error);
if (!arg1)
error = EPERM;
else
error = SYSCTL_IN(req, arg1, sizeof(int));
return (error);
}
/*
* Based on on sysctl_handle_int() convert milliseconds into ticks.
Build on Jeff Roberson's linker-set based dynamic per-CPU allocator (DPCPU), as suggested by Peter Wemm, and implement a new per-virtual network stack memory allocator. Modify vnet to use the allocator instead of monolithic global container structures (vinet, ...). This change solves many binary compatibility problems associated with VIMAGE, and restores ELF symbols for virtualized global variables. Each virtualized global variable exists as a "reference copy", and also once per virtual network stack. Virtualized global variables are tagged at compile-time, placing the in a special linker set, which is loaded into a contiguous region of kernel memory. Virtualized global variables in the base kernel are linked as normal, but those in modules are copied and relocated to a reserved portion of the kernel's vnet region with the help of a the kernel linker. Virtualized global variables exist in per-vnet memory set up when the network stack instance is created, and are initialized statically from the reference copy. Run-time access occurs via an accessor macro, which converts from the current vnet and requested symbol to a per-vnet address. When "options VIMAGE" is not compiled into the kernel, normal global ELF symbols will be used instead and indirection is avoided. This change restores static initialization for network stack global variables, restores support for non-global symbols and types, eliminates the need for many subsystem constructors, eliminates large per-subsystem structures that caused many binary compatibility issues both for monitoring applications (netstat) and kernel modules, removes the per-function INIT_VNET_*() macros throughout the stack, eliminates the need for vnet_symmap ksym(2) munging, and eliminates duplicate definitions of virtualized globals under VIMAGE_GLOBALS. Bump __FreeBSD_version and update UPDATING. Portions submitted by: bz Reviewed by: bz, zec Discussed with: gnn, jamie, jeff, jhb, julian, sam Suggested by: peter Approved by: re (kensmith)
2009-07-14 22:48:30 +00:00
* Note: this is used by TCP.
*/
int
sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
{
int error, s, tt;
Permit buiding kernels with options VIMAGE, restricted to only a single active network stack instance. Turning on options VIMAGE at compile time yields the following changes relative to default kernel build: 1) V_ accessor macros for virtualized variables resolve to structure fields via base pointers, instead of being resolved as fields in global structs or plain global variables. As an example, V_ifnet becomes: options VIMAGE: ((struct vnet_net *) vnet_net)->_ifnet default build: vnet_net_0._ifnet options VIMAGE_GLOBALS: ifnet 2) INIT_VNET_* macros will declare and set up base pointers to be used by V_ accessor macros, instead of resolving to whitespace: INIT_VNET_NET(ifp->if_vnet); becomes struct vnet_net *vnet_net = (ifp->if_vnet)->mod_data[VNET_MOD_NET]; 3) Memory for vnet modules registered via vnet_mod_register() is now allocated at run time in sys/kern/kern_vimage.c, instead of per vnet module structs being declared as globals. If required, vnet modules can now request the framework to provide them with allocated bzeroed memory by filling in the vmi_size field in their vmi_modinfo structures. 4) structs socket, ifnet, inpcbinfo, tcpcb and syncache_head are extended to hold a pointer to the parent vnet. options VIMAGE builds will fill in those fields as required. 5) curvnet is introduced as a new global variable in options VIMAGE builds, always pointing to the default and only struct vnet. 6) struct sysctl_oid has been extended with additional two fields to store major and minor virtualization module identifiers, oid_v_subs and oid_v_mod. SYSCTL_V_* family of macros will fill in those fields accordingly, and store the offset in the appropriate vnet container struct in oid_arg1. In sysctl handlers dealing with virtualized sysctls, the SYSCTL_RESOLVE_V_ARG1() macro will compute the address of the target variable and make it available in arg1 variable for further processing. Unused fields in structs vnet_inet, vnet_inet6 and vnet_ipfw have been deleted. Reviewed by: bz, rwatson Approved by: julian (mentor)
2009-04-30 13:36:26 +00:00
tt = *(int *)arg1;
s = (int)((int64_t)tt * 1000 / hz);
error = sysctl_handle_int(oidp, &s, 0, req);
if (error || !req->newptr)
return (error);
tt = (int)((int64_t)s * hz / 1000);
if (tt < 1)
return (EINVAL);
Permit buiding kernels with options VIMAGE, restricted to only a single active network stack instance. Turning on options VIMAGE at compile time yields the following changes relative to default kernel build: 1) V_ accessor macros for virtualized variables resolve to structure fields via base pointers, instead of being resolved as fields in global structs or plain global variables. As an example, V_ifnet becomes: options VIMAGE: ((struct vnet_net *) vnet_net)->_ifnet default build: vnet_net_0._ifnet options VIMAGE_GLOBALS: ifnet 2) INIT_VNET_* macros will declare and set up base pointers to be used by V_ accessor macros, instead of resolving to whitespace: INIT_VNET_NET(ifp->if_vnet); becomes struct vnet_net *vnet_net = (ifp->if_vnet)->mod_data[VNET_MOD_NET]; 3) Memory for vnet modules registered via vnet_mod_register() is now allocated at run time in sys/kern/kern_vimage.c, instead of per vnet module structs being declared as globals. If required, vnet modules can now request the framework to provide them with allocated bzeroed memory by filling in the vmi_size field in their vmi_modinfo structures. 4) structs socket, ifnet, inpcbinfo, tcpcb and syncache_head are extended to hold a pointer to the parent vnet. options VIMAGE builds will fill in those fields as required. 5) curvnet is introduced as a new global variable in options VIMAGE builds, always pointing to the default and only struct vnet. 6) struct sysctl_oid has been extended with additional two fields to store major and minor virtualization module identifiers, oid_v_subs and oid_v_mod. SYSCTL_V_* family of macros will fill in those fields accordingly, and store the offset in the appropriate vnet container struct in oid_arg1. In sysctl handlers dealing with virtualized sysctls, the SYSCTL_RESOLVE_V_ARG1() macro will compute the address of the target variable and make it available in arg1 variable for further processing. Unused fields in structs vnet_inet, vnet_inet6 and vnet_ipfw have been deleted. Reviewed by: bz, rwatson Approved by: julian (mentor)
2009-04-30 13:36:26 +00:00
*(int *)arg1 = tt;
return (0);
}
/*
2013-02-11 23:05:10 +00:00
* Handle a long, signed or unsigned.
* Two cases:
* a variable: point arg1 at it.
* a constant: pass it in arg2.
*/
int
sysctl_handle_long(SYSCTL_HANDLER_ARGS)
{
int error = 0;
long tmplong;
#ifdef SCTL_MASK32
int tmpint;
#endif
/*
* Attempt to get a coherent snapshot by making a copy of the data.
*/
if (arg1)
tmplong = *(long *)arg1;
else
tmplong = arg2;
#ifdef SCTL_MASK32
if (req->flags & SCTL_MASK32) {
tmpint = tmplong;
error = SYSCTL_OUT(req, &tmpint, sizeof(int));
} else
#endif
error = SYSCTL_OUT(req, &tmplong, sizeof(long));
if (error || !req->newptr)
return (error);
if (!arg1)
error = EPERM;
#ifdef SCTL_MASK32
else if (req->flags & SCTL_MASK32) {
error = SYSCTL_IN(req, &tmpint, sizeof(int));
*(long *)arg1 = (long)tmpint;
}
#endif
else
error = SYSCTL_IN(req, arg1, sizeof(long));
return (error);
}
/*
2013-02-11 23:05:10 +00:00
* Handle a 64 bit int, signed or unsigned.
* Two cases:
* a variable: point arg1 at it.
* a constant: pass it in arg2.
*/
int
sysctl_handle_64(SYSCTL_HANDLER_ARGS)
{
int error = 0;
uint64_t tmpout;
/*
* Attempt to get a coherent snapshot by making a copy of the data.
*/
if (arg1)
tmpout = *(uint64_t *)arg1;
else
tmpout = arg2;
error = SYSCTL_OUT(req, &tmpout, sizeof(uint64_t));
if (error || !req->newptr)
return (error);
if (!arg1)
error = EPERM;
else
error = SYSCTL_IN(req, arg1, sizeof(uint64_t));
return (error);
}
/*
* Handle our generic '\0' terminated 'C' string.
* Two cases:
* a variable string: point arg1 at it, arg2 is max length.
* a constant string: point arg1 at it, arg2 is zero.
*/
int
sysctl_handle_string(SYSCTL_HANDLER_ARGS)
{
size_t outlen;
int error = 0, ro_string = 0;
/*
* A zero-length buffer indicates a fixed size read-only
* string:
*/
if (arg2 == 0) {
arg2 = strlen((char *)arg1) + 1;
ro_string = 1;
}
if (req->oldptr != NULL) {
char *tmparg;
if (ro_string) {
tmparg = arg1;
} else {
/* try to make a coherent snapshot of the string */
tmparg = malloc(arg2, M_SYSCTLTMP, M_WAITOK);
memcpy(tmparg, arg1, arg2);
}
outlen = strnlen(tmparg, arg2 - 1) + 1;
error = SYSCTL_OUT(req, tmparg, outlen);
if (!ro_string)
free(tmparg, M_SYSCTLTMP);
} else {
outlen = strnlen((char *)arg1, arg2 - 1) + 1;
error = SYSCTL_OUT(req, NULL, outlen);
}
1999-03-30 09:00:45 +00:00
if (error || !req->newptr)
return (error);
1999-03-30 09:00:45 +00:00
if ((req->newlen - req->newidx) >= arg2) {
error = EINVAL;
} else {
arg2 = (req->newlen - req->newidx);
error = SYSCTL_IN(req, arg1, arg2);
((char *)arg1)[arg2] = '\0';
}
return (error);
}
/*
* Handle any kind of opaque data.
* arg1 points to it, arg2 is the size.
*/
int
sysctl_handle_opaque(SYSCTL_HANDLER_ARGS)
{
int error, tries;
u_int generation;
struct sysctl_req req2;
/*
* Attempt to get a coherent snapshot, by using the thread
* pre-emption counter updated from within mi_switch() to
* determine if we were pre-empted during a bcopy() or
* copyout(). Make 3 attempts at doing this before giving up.
* If we encounter an error, stop immediately.
*/
tries = 0;
req2 = *req;
retry:
generation = curthread->td_generation;
error = SYSCTL_OUT(req, arg1, arg2);
if (error)
return (error);
tries++;
if (generation != curthread->td_generation && tries < 3) {
*req = req2;
goto retry;
}
error = SYSCTL_IN(req, arg1, arg2);
return (error);
}
/*
* Transfer functions to/from kernel space.
* XXX: rather untested at this point
*/
static int
sysctl_old_kernel(struct sysctl_req *req, const void *p, size_t l)
{
size_t i = 0;
if (req->oldptr) {
i = l;
if (req->oldlen <= req->oldidx)
i = 0;
else
if (i > req->oldlen - req->oldidx)
i = req->oldlen - req->oldidx;
if (i > 0)
bcopy(p, (char *)req->oldptr + req->oldidx, i);
}
req->oldidx += l;
if (req->oldptr && i != l)
return (ENOMEM);
return (0);
}
static int
sysctl_new_kernel(struct sysctl_req *req, void *p, size_t l)
{
if (!req->newptr)
return (0);
if (req->newlen - req->newidx < l)
return (EINVAL);
bcopy((char *)req->newptr + req->newidx, p, l);
req->newidx += l;
return (0);
}
int
kernel_sysctl(struct thread *td, int *name, u_int namelen, void *old,
size_t *oldlenp, void *new, size_t newlen, size_t *retval, int flags)
{
int error = 0;
struct sysctl_req req;
bzero(&req, sizeof req);
req.td = td;
req.flags = flags;
if (oldlenp) {
req.oldlen = *oldlenp;
}
req.validlen = req.oldlen;
if (old) {
req.oldptr= old;
}
if (new != NULL) {
req.newlen = newlen;
req.newptr = new;
}
req.oldfunc = sysctl_old_kernel;
req.newfunc = sysctl_new_kernel;
req.lock = REQ_UNWIRED;
SYSCTL_XLOCK();
error = sysctl_root(0, name, namelen, &req);
SYSCTL_XUNLOCK();
if (req.lock == REQ_WIRED && req.validlen > 0)
vsunlock(req.oldptr, req.validlen);
if (error && error != ENOMEM)
return (error);
if (retval) {
if (req.oldptr && req.oldidx > req.validlen)
*retval = req.validlen;
else
*retval = req.oldidx;
}
return (error);
}
int
kernel_sysctlbyname(struct thread *td, char *name, void *old, size_t *oldlenp,
void *new, size_t newlen, size_t *retval, int flags)
{
int oid[CTL_MAXNAME];
2001-06-22 19:54:38 +00:00
size_t oidlen, plen;
int error;
oid[0] = 0; /* sysctl internal magic */
oid[1] = 3; /* name2oid */
oidlen = sizeof(oid);
error = kernel_sysctl(td, oid, 2, oid, &oidlen,
(void *)name, strlen(name), &plen, flags);
if (error)
return (error);
error = kernel_sysctl(td, oid, plen / sizeof(int), old, oldlenp,
new, newlen, retval, flags);
return (error);
}
/*
* Transfer function to/from user space.
*/
static int
sysctl_old_user(struct sysctl_req *req, const void *p, size_t l)
{
size_t i, len, origidx;
int error;
origidx = req->oldidx;
req->oldidx += l;
if (req->oldptr == NULL)
return (0);
/*
* If we have not wired the user supplied buffer and we are currently
* holding locks, drop a witness warning, as it's possible that
* write operations to the user page can sleep.
*/
if (req->lock != REQ_WIRED)
WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
"sysctl_old_user()");
i = l;
len = req->validlen;
if (len <= origidx)
i = 0;
else {
if (i > len - origidx)
i = len - origidx;
if (req->lock == REQ_WIRED) {
error = copyout_nofault(p, (char *)req->oldptr +
origidx, i);
} else
error = copyout(p, (char *)req->oldptr + origidx, i);
if (error != 0)
return (error);
}
if (i < l)
return (ENOMEM);
return (0);
}
static int
sysctl_new_user(struct sysctl_req *req, void *p, size_t l)
{
1995-11-14 09:26:17 +00:00
int error;
if (!req->newptr)
return (0);
if (req->newlen - req->newidx < l)
return (EINVAL);
WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
"sysctl_new_user()");
error = copyin((char *)req->newptr + req->newidx, p, l);
req->newidx += l;
return (error);
}
/*
* Wire the user space destination buffer. If set to a value greater than
* zero, the len parameter limits the maximum amount of wired memory.
*/
int
sysctl_wire_old_buffer(struct sysctl_req *req, size_t len)
{
int ret;
size_t wiredlen;
wiredlen = (len > 0 && len < req->oldlen) ? len : req->oldlen;
ret = 0;
if (req->lock != REQ_WIRED && req->oldptr &&
req->oldfunc == sysctl_old_user) {
if (wiredlen != 0) {
ret = vslock(req->oldptr, wiredlen);
if (ret != 0) {
if (ret != ENOMEM)
return (ret);
wiredlen = 0;
}
}
req->lock = REQ_WIRED;
req->validlen = wiredlen;
}
return (0);
}
int
sysctl_find_oid(int *name, u_int namelen, struct sysctl_oid **noid,
int *nindx, struct sysctl_req *req)
{
struct sysctl_oid_list *lsp;
struct sysctl_oid *oid;
int indx;
SYSCTL_ASSERT_XLOCKED();
lsp = &sysctl__children;
indx = 0;
while (indx < CTL_MAXNAME) {
SLIST_FOREACH(oid, lsp, oid_link) {
if (oid->oid_number == name[indx])
break;
}
if (oid == NULL)
return (ENOENT);
indx++;
if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
if (oid->oid_handler != NULL || indx == namelen) {
*noid = oid;
if (nindx != NULL)
*nindx = indx;
KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
("%s found DYING node %p", __func__, oid));
return (0);
}
lsp = SYSCTL_CHILDREN(oid);
} else if (indx == namelen) {
*noid = oid;
if (nindx != NULL)
*nindx = indx;
KASSERT((oid->oid_kind & CTLFLAG_DYING) == 0,
("%s found DYING node %p", __func__, oid));
return (0);
} else {
return (ENOTDIR);
}
}
return (ENOENT);
}
/*
* Traverse our tree, and find the right node, execute whatever it points
* to, and return the resulting error code.
*/
static int
sysctl_root(SYSCTL_HANDLER_ARGS)
{
struct sysctl_oid *oid;
int error, indx, lvl;
SYSCTL_ASSERT_XLOCKED();
error = sysctl_find_oid(arg1, arg2, &oid, &indx, req);
if (error)
return (error);
if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
/*
* You can't call a sysctl when it's a node, but has
* no handler. Inform the user that it's a node.
* The indx may or may not be the same as namelen.
*/
if (oid->oid_handler == NULL)
return (EISDIR);
}
/* Is this sysctl writable? */
if (req->newptr && !(oid->oid_kind & CTLFLAG_WR))
return (EPERM);
KASSERT(req->td != NULL, ("sysctl_root(): req->td == NULL"));
#ifdef CAPABILITY_MODE
/*
* If the process is in capability mode, then don't permit reading or
* writing unless specifically granted for the node.
*/
if (IN_CAPABILITY_MODE(req->td)) {
if (req->oldptr && !(oid->oid_kind & CTLFLAG_CAPRD))
return (EPERM);
if (req->newptr && !(oid->oid_kind & CTLFLAG_CAPWR))
return (EPERM);
}
#endif
/* Is this sysctl sensitive to securelevels? */
if (req->newptr && (oid->oid_kind & CTLFLAG_SECURE)) {
lvl = (oid->oid_kind & CTLMASK_SECURE) >> CTLSHIFT_SECURE;
error = securelevel_gt(req->td->td_ucred, lvl);
if (error)
return (error);
}
/* Is this sysctl writable by only privileged users? */
if (req->newptr && !(oid->oid_kind & CTLFLAG_ANYBODY)) {
int priv;
if (oid->oid_kind & CTLFLAG_PRISON)
priv = PRIV_SYSCTL_WRITEJAIL;
#ifdef VIMAGE
else if ((oid->oid_kind & CTLFLAG_VNET) &&
prison_owns_vnet(req->td->td_ucred))
priv = PRIV_SYSCTL_WRITEJAIL;
#endif
else
priv = PRIV_SYSCTL_WRITE;
error = priv_check(req->td, priv);
if (error)
return (error);
}
if (!oid->oid_handler)
return (EINVAL);
if ((oid->oid_kind & CTLTYPE) == CTLTYPE_NODE) {
2004-07-28 06:42:41 +00:00
arg1 = (int *)arg1 + indx;
arg2 -= indx;
} else {
arg1 = oid->oid_arg1;
arg2 = oid->oid_arg2;
}
#ifdef MAC
error = mac_system_check_sysctl(req->td->td_ucred, oid, arg1, arg2,
req);
if (error != 0)
return (error);
#endif
#ifdef VIMAGE
if ((oid->oid_kind & CTLFLAG_VNET) && arg1 != NULL)
arg1 = (void *)(curvnet->vnet_data_base + (uintptr_t)arg1);
#endif
error = sysctl_root_handler_locked(oid, arg1, arg2, req);
KFAIL_POINT_ERROR(_debug_fail_point, sysctl_running, error);
return (error);
}
#ifndef _SYS_SYSPROTO_H_
struct sysctl_args {
int *name;
u_int namelen;
void *old;
size_t *oldlenp;
void *new;
size_t newlen;
};
#endif
1994-05-24 10:09:53 +00:00
int
sys___sysctl(struct thread *td, struct sysctl_args *uap)
1994-05-24 10:09:53 +00:00
{
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
int error, i, name[CTL_MAXNAME];
size_t j;
1994-05-24 10:09:53 +00:00
if (uap->namelen > CTL_MAXNAME || uap->namelen < 2)
return (EINVAL);
error = copyin(uap->name, &name, uap->namelen * sizeof(int));
if (error)
1994-05-24 10:09:53 +00:00
return (error);
error = userland_sysctl(td, name, uap->namelen,
uap->old, uap->oldlenp, 0,
uap->new, uap->newlen, &j, 0);
if (error && error != ENOMEM)
return (error);
if (uap->oldlenp) {
Expand the scope of the sysctllock sx lock to protect the sysctl tree itself. Back in 1.1 of kern_sysctl.c the sysctl() routine wired the "old" userland buffer for most sysctls (everything except kern.vnode.*). I think to prevent issues with wiring too much memory it used a 'memlock' to serialize all sysctl(2) invocations, meaning that only one user buffer could be wired at a time. In 5.0 the 'memlock' was converted to an sx lock and renamed to 'sysctl lock'. However, it still only served the purpose of serializing sysctls to avoid wiring too much memory and didn't actually protect the sysctl tree as its name suggested. These changes expand the lock to actually protect the tree. Later on in 5.0, sysctl was changed to not wire buffers for requests by default (sysctl_handle_opaque() will still wire buffers larger than a single page, however). As a result, user buffers are no longer wired as often. However, many sysctl handlers still wire user buffers, so it is still desirable to serialize userland sysctl requests. Kernel sysctl requests are allowed to run in parallel, however. - Expose sysctl_lock()/sysctl_unlock() routines to exclusively lock the sysctl tree for a few places outside of kern_sysctl.c that manipulate the sysctl tree directly including the kernel linker and vfs_register(). - sysctl_register() and sysctl_unregister() require the caller to lock the sysctl lock using sysctl_lock() and sysctl_unlock(). The rest of the public sysctl API manage the locking internally. - Add a locked variant of sysctl_remove_oid() for internal use so that external uses of the API do not need to be aware of locking requirements. - The kernel linker no longer needs Giant when manipulating the sysctl tree. - Add a missing break to the loop in vfs_register() so that we stop looking at the sysctl MIB once we have changed it. MFC after: 1 month
2009-02-06 14:51:32 +00:00
i = copyout(&j, uap->oldlenp, sizeof(j));
if (i)
return (i);
}
return (error);
}
/*
* This is used from various compatibility syscalls too. That's why name
* must be in kernel space.
*/
int
userland_sysctl(struct thread *td, int *name, u_int namelen, void *old,
size_t *oldlenp, int inkernel, void *new, size_t newlen, size_t *retval,
int flags)
{
int error = 0, memlocked;
struct sysctl_req req;
bzero(&req, sizeof req);
req.td = td;
req.flags = flags;
1995-11-14 09:26:17 +00:00
if (oldlenp) {
if (inkernel) {
req.oldlen = *oldlenp;
} else {
error = copyin(oldlenp, &req.oldlen, sizeof(*oldlenp));
if (error)
return (error);
}
}
req.validlen = req.oldlen;
if (old) {
if (!useracc(old, req.oldlen, VM_PROT_WRITE))
return (EFAULT);
req.oldptr= old;
}
if (new != NULL) {
if (!useracc(new, newlen, VM_PROT_READ))
return (EFAULT);
req.newlen = newlen;
req.newptr = new;
}
req.oldfunc = sysctl_old_user;
req.newfunc = sysctl_new_user;
req.lock = REQ_UNWIRED;
#ifdef KTRACE
if (KTRPOINT(curthread, KTR_SYSCTL))
ktrsysctl(name, namelen);
#endif
if (req.oldlen > PAGE_SIZE) {
memlocked = 1;
sx_xlock(&sysctlmemlock);
} else
memlocked = 0;
CURVNET_SET(TD_TO_VNET(td));
for (;;) {
req.oldidx = 0;
req.newidx = 0;
SYSCTL_XLOCK();
error = sysctl_root(0, name, namelen, &req);
SYSCTL_XUNLOCK();
if (error != EAGAIN)
break;
kern_yield(PRI_USER);
}
CURVNET_RESTORE();
if (req.lock == REQ_WIRED && req.validlen > 0)
vsunlock(req.oldptr, req.validlen);
if (memlocked)
sx_xunlock(&sysctlmemlock);
if (error && error != ENOMEM)
return (error);
if (retval) {
if (req.oldptr && req.oldidx > req.validlen)
*retval = req.validlen;
else
*retval = req.oldidx;
}
return (error);
1994-05-24 10:09:53 +00:00
}
/*
* Drain into a sysctl struct. The user buffer should be wired if a page
* fault would cause issue.
*/
static int
sbuf_sysctl_drain(void *arg, const char *data, int len)
{
struct sysctl_req *req = arg;
int error;
error = SYSCTL_OUT(req, data, len);
KASSERT(error >= 0, ("Got unexpected negative value %d", error));
return (error == 0 ? len : -error);
}
struct sbuf *
sbuf_new_for_sysctl(struct sbuf *s, char *buf, int length,
struct sysctl_req *req)
{
s = sbuf_new(s, buf, length, SBUF_FIXEDLEN);
sbuf_set_drain(s, sbuf_sysctl_drain, req);
return (s);
}