Ubuntu Manpages

SYSCTL_DECL,

#include <sys/types.h>
#include <sys/sysctl.h>

SYSCTL_DECL(name);

SYSCTL_INT(parent, nbr, name, access, ptr, val, descr);

SYSCTL_LONG(parent, nbr, name, access, ptr, val, descr);

SYSCTL_NODE(parent, nbr, name, access, handler, descr);

SYSCTL_OPAQUE(parent, nbr, name, access, ptr, len, fmt, descr);

SYSCTL_PROC(parent, nbr, name, access, ptr, arg, handler, fmt, descr);

SYSCTL_QUAD(parent, nbr, name, access, ptr, val, descr);

SYSCTL_STRING(parent, nbr, name, access, arg, len, descr);

SYSCTL_STRUCT(parent, nbr, name, access, ptr, type, descr);

SYSCTL_UINT(parent, nbr, name, access, ptr, val, descr);

SYSCTL_ULONG(parent, nbr, name, access, ptr, val, descr);

SYSCTL_UQUAD(parent, nbr, name, access, ptr, val, descr);

The SYSCTL kernel interfaces allow code to statically declare sysctl(8) MIB entries, which will be initialized when the kernel module containing the declaration is initialized. When the module is unloaded, the sysctl will be automatically destroyed.

Sysctl nodes are created in a hierarchical tree, with all static nodes being represented by named C data structures; in order to create a new node under an existing node in the tree, the structure representing the desired parent node must be declared in the current context using ().

New nodes are declared using one of (), (), (), (), (), (), (), (), (), (), and (). Each macro accepts a parent name, as declared using SYSCTL_DECL(), an OID number, typically OID_AUTO, a node name, a set of control and access flags, and a description. Depending on the macro, a pointer to a variable supporting the MIB entry, a size, a value, and a function pointer implementing the MIB entry may also be present.

For most of the above macros, declaring a type as part of the access flags is not necessary — however, when declaring a sysctl implemented by a function, including a type in the access mask is required:

This is a node intended to be a parent for other nodes.
This is a signed integer.
This is a nul-terminated string stored in a character array.
This is a 64-bit signed integer.
This is an opaque data structure.
Alias for CTLTYPE_OPAQUE.
This is an unsigned integer.
This is a signed long.
This is an unsigned long.
This is a 64-bit unsigned integer.

All sysctl types except for new node declarations require one of the following flags to be set indicating the read and write disposition of the sysctl:

This is a read-only sysctl.
This is a read-only sysctl which can be set by a system tunable.
This is a writable sysctl.
This sysctl is readable and writable.
This sysctl is readable and writable and can also be set by a system tunable.

Additionally, any of the following optional flags may also be specified:

Any user or process can write to this sysctl.
This sysctl can be written to only if the effective securelevel of the process is ≤ 0.
This sysctl can be written to by processes in jail(2).
When iterating the sysctl name space, do not list this sysctl.
Advisory flag that a system tunable also exists for this variable.

When creating new sysctls, careful attention should be paid to the security implications of the monitoring or management interface being created. Most sysctls present in the kernel are read-only or writable only by the superuser. Sysctls exporting extensive information on system data structures and operation, especially those implemented using procedures, will wish to implement access control to limit the undesired exposure of information about other processes, network connections, etc.

The following top level sysctl name spaces are commonly used:

compat
Compatibility layer information.
debug
Debugging information. Various name spaces exist under debug.
hw
Hardware and device driver information.
kern
Kernel behavior tuning; generally deprecated in favor of more specific name spaces.
machdep
Machine-dependent configuration parameters.
net
Network subsystem. Various protocols have name spaces under net.
regression
Regression test configuration and information.
security
Security and security-policy configuration and information.
sysctl
Reserved name space for the implementation of sysctl.
user
Configuration settings relating to user application behavior. Generally, configuring applications using kernel sysctls is discouraged.
vfs
Virtual file system configuration and information.
vm
Virtual memory subsystem configuration and information.

Sample use of SYSCTL_DECL() to declare the security sysctl tree for use by new nodes:

SYSCTL_DECL(_security);

Examples of integer, opaque, string, and procedure sysctls follow:

/*
 * Example of a constant integer value.  Notice that the control
 * flags are CTLFLAG_RD, the variable pointer is NULL, and the
 * value is declared.
 */
SYSCTL_INT(_debug_sizeof, OID_AUTO, bio, CTLFLAG_RD, NULL,
    sizeof(struct bio), "sizeof(struct bio)");

/*
 * Example of a variable integer value.  Notice that the control
 * flags are CTLFLAG_RW, the variable pointer is set, and the
 * value is 0.
 */
static int	doingcache = 1;		/* 1 => enable the cache */
SYSCTL_INT(_debug, OID_AUTO, vfscache, CTLFLAG_RW, &doingcache, 0,
    "Enable name cache");

/*
 * Example of a variable string value.  Notice that the control
 * flags are CTLFLAG_RW, that the variable pointer and string
 * size are set.  Unlike newer sysctls, this older sysctl uses a
 * static oid number.
 */
char kernelname[MAXPATHLEN] = "/kernel";	/* XXX bloat */
SYSCTL_STRING(_kern, KERN_BOOTFILE, bootfile, CTLFLAG_RW,
    kernelname, sizeof(kernelname), "Name of kernel file booted");

/*
 * Example of an opaque data type exported by sysctl.  Notice that
 * the variable pointer and size are provided, as well as a format
 * string for sysctl(8).
 */
static l_fp pps_freq;	/* scaled frequence offset (ns/s) */
SYSCTL_OPAQUE(_kern_ntp_pll, OID_AUTO, pps_freq, CTLFLAG_RD,
    &pps_freq, sizeof(pps_freq), "I", "");

/*
 * Example of a procedure based sysctl exporting string
 * information.  Notice that the data type is declared, the NULL
 * variable pointer and 0 size, the function pointer, and the
 * format string for sysctl(8).
 */
SYSCTL_PROC(_kern_timecounter, OID_AUTO, hardware, CTLTYPE_STRING |
    CTLFLAG_RW, NULL, 0, sysctl_kern_timecounter_hardware, "A",
    "");

When adding, modifying, or removing sysctl names, it is important to be aware that these interfaces may be used by users, libraries, applications, or documentation (such as published books), and are implicitly published application interfaces. As with other application interfaces, caution must be taken not to break existing applications, and to think about future use of new name spaces so as to avoid the need to rename or remove interfaces that might be depended on in the future.

The semantics chosen for a new sysctl should be as clear as possible, and the name of the sysctl must closely reflect its semantics. Therefore the sysctl name deserves a fair amount of consideration. It should be short but yet representative of the sysctl meaning. If the name consists of several words, they should be separated by underscore characters, as in compute_summary_at_mount. Underscore characters may be omitted only if the name consists of not more than two words, each being not longer than four characters, as in bootfile. For boolean sysctls, negative logic should be totally avoided. That is, do not use names like no_foobar or foobar_disable. They are confusing and lead to configuration errors. Use positive logic instead: foobar, foobar_enable.

A temporary sysctl node that should not be relied upon must be designated as such by a leading underscore character in its name. For example: _dirty_hack.

sysctl(3), sysctl(8), sysctl_add_oid(9), sysctl_ctx_free(9), sysctl_ctx_init(9), sysctl_remove_oid(9)

The sysctl(8) utility first appeared in 4.4BSD.

The sysctl implementation originally found in BSD has been extensively rewritten by Poul-Henning Kamp in order to add support for name lookups, name space iteration, and dynamic addition of MIB nodes.

This man page was written by Robert N. M. Watson.