NAME

sysctl — get or set kernel state

SYNOPSIS

sysctl

[-AdeMnq] [-r | -x] [name ...]

sysctl

[-nq] [-r | -x] -w name[?]=value ...

sysctl

[-en] [-r | -x] -a

sysctl

[-nq] [-r | -x] -f file

DESCRIPTION

The sysctl utility retrieves kernel state and allows processes with appropriate privilege to set kernel state. The state to be retrieved or set is described using a ``Management Information Base'' (``MIB'') style name, described as a dotted set of components. The ‘/’ character may also be used as a separator and a leading separator character is accepted. If name specifies a non-leaf node in the MIB, all the nodes underneath name will be printed.

The following options are available:

-A

List all the known MIB names including tables, unless any MIB arguments or -f file are given. Those with string or integer values will be printed as with the -a flag; for table or structure values that sysctl is not able to print, the name of the utility to retrieve them is given. Errors in retrieving or setting values will be directed to stdout instead of stderr.

-a

List all the currently available string or integer values. The use of a solitary separator character (either ‘.’ or ‘/’) by itself has the same effect. Any given name arguments are ignored if this option is specified.

-d

Descriptions of each of the nodes selected will be printed instead of their values.

-e

Separate the name and the value of the variable(s) with ‘=’. This is useful for producing output which can be fed back to the sysctl utility. This option is ignored if -n is specified or a variable is being set.

-f

Specifies the name of a file to read and process. Blank lines and comments (beginning with ‘#’) are ignored. Line continuations with ‘\’ are permitted. Remaining lines are processed similarly to command line arguments of the form name or name=value. The -w flag is implied by -f. Any name arguments are ignored.

-M

Makes sysctl print the MIB instead of any of the actual values contained in the MIB. This causes the entire MIB to be printed unless specific MIB arguments or -f file are also given.

-n

Specifies that the printing of the field name should be suppressed and that only its value should be output. This flag is useful for setting shell variables. For example, to save the pagesize in variable psize, use:

set psize=`sysctl -n hw.pagesize`

-q

Used to indicate that nothing should be printed for reads or writes unless an error is detected. For reads, not finding a variable does not print an error, but exits with an error code. This is useful just for testing that a variable exists.

-r

Raw output form. Values printed are in their raw binary forms as retrieved directly from the kernel. Some additional nodes that sysctl cannot print directly can be retrieved with this flag. This option conflicts with the -x option.

-w

Sets the MIB style name given to the value given. The MIB style name and value must be separated by ‘=’ with no whitespace. To prevent an error if the MIB style name does not exist (as would be the case with optional kernel components), one can separate the MIB style name and the value with ‘?=’. Only integral and string values can be set via this method.

-x

Makes sysctl print the requested value in a hexadecimal representation instead of its regular form. If specified more than once, the output for each value resembles that of hexdump(1) when given the -C flag. This option conflicts with the -r option.

The ‘proc’ top-level MIB has a special semantic: it represent per-process values and as such may differ from one process to another. The second-level name is the pid of the process (in decimal form), or the special word ‘curproc’. For variables below ‘proc.<pid>.rlimit’, the integer value may be replaced with the string ‘unlimited’ if it matches the magic value used to disable a limit.

The information available from sysctl consists of integers, strings, and tables. The tabular information can only be retrieved by special purpose programs such as ps, systat, and netstat. See sysctl(7) for description of available MIBs.

CREATION AND DELETION

New nodes are allowed to be created by the superuser when the kernel is running at security level 0. These new nodes may refer to existing kernel data or to new data that is only instrumented by sysctl(3) itself.

The syntax for creating new nodes is “//create=new.node.path” followed by one or more of the following attributes separated by commas. The use of a double separator (both ‘/’ and ‘.’ can be used as separators) as the prefix tells sysctl that the first series of tokens is not a MIB name, but a command. It is recommended that the double separator preceding the command not be the same as the separator used in naming the MIB entry so as to avoid possible parse conflicts. The “value” assigned, if one is given, must be last.

• type=<T> where T must be one of “node”, “int”, “string”, “quad”, or “struct”. If the type is omitted, the “node” type is assumed.
• size=<S> here, S asserts the size of the new node. Nodes of type “node” should not have a size set. The size may be omitted for nodes of types “int” or “quad”. If the size is omitted for a node of type “string”, the size will be determined by the length of the given value, or by the kernel for kernel strings. Nodes of type “struct” must have their size explicitly set.
• addr=<A> or symbol=<A> The kernel address of the data being instrumented. If “symbol” is used, the symbol must be globally visible to the in-kernel ksyms(4) driver.
• n=<N> The MIB number to be assigned to the new node. If no number is specified, the kernel will assign a value.
• flags=<F> A concatenated string of single letters that govern the behavior of the node. Flags currently available are:

  a

  Allow anyone to write to the node, if it is writable.

  h

  “Hidden”. sysctl must be invoked with -A or the hidden node must be specifically requested in order to see it

  i

  “Immediate”. Makes the node store data in itself, rather than allocating new space for it. This is the default for nodes of type “int” and “quad”. This is the opposite of owning data.

  o

  “Own”. When the node is created, separate space will be allocated to store the data to be instrumented. This is the default for nodes of type “string” and “struct” where it is not possible to guarantee sufficient space to store the data in the node itself.

  p

  “Private”. Nodes that are marked private, and children of nodes so marked, are only viewable by the superuser. Be aware that the immediate data that some nodes may store is not necessarily protected by this.

  x

  “Hexadecimal”. Make sysctl default to hexadecimal display of the retrieved value

  r

  “Read-only”. The data instrumented by the given node is read-only. Note that other mechanisms may still exist for changing the data. This is the default for nodes that instrument data.

  w

  “Writable”. The data instrumented by the given node is writable at any time. This is the default for nodes that can have children.

• value=<V> An initial starting value for a new node that does not reference existing kernel data. Initial values can only be assigned for nodes of the “int”, “quad”, and “string” types.

New nodes must fit the following set of criteria:

• If the new node is to address an existing kernel object, only one of the “symbol” or “addr” arguments may be given.
• The size for a “struct” type node must be specified; no initial value is expected or permitted.
• Either the size or the initial value for a “string” node must be given.
• The node which will be the parent of the new node must be writable.

If any of the given parameters describes an invalid configuration, sysctl will emit a diagnostic message to the standard error and exit.

Descriptions can be added by the super-user to any node that does not have one, provided that the node is not marked with the “PERMANENT” flag. The syntax is similar to the syntax for creating new nodes with the exception of the keyword that follows the double separator at the start of the command: “//describe=new.node.path=new node description”. Once a description has been added, it cannot be changed or removed.

When destroying nodes, only the path to the node is necessary, i.e., “//destroy=old.node.path”. No other parameters are expected or permitted. Nodes being destroyed must have no children, and their parent must be writable. Nodes that are marked with the “PERMANENT” flag (as assigned by the kernel) may not be deleted.

In all cases, the initial ‘=’ that follows the command (eg, “create”, “destroy”, or “describe”) may be replaced with another instance of the separator character, provided that the same separator character is used for the length of the name specification.

FILES

/etc/sysctl.conf

sysctl variables set at boot time

EXAMPLES

For example, to retrieve the maximum number of processes allowed in the system, one would use the following request:

sysctl kern.maxproc

To set the maximum number of processes allowed in the system to 1000, one would use the following request:

sysctl -w kern.maxproc=1000

Information about the system clock rate may be obtained with:

sysctl kern.clockrate

Information about the load average history may be obtained with:

sysctl vm.loadavg

To view the values of the per-process variables of the current shell, the request:

sysctl proc.$$

can be used if the shell interpreter replaces $$ with its pid (this is true for most shells).

To redirect core dumps to the /var/tmp/<username> directory,

sysctl -w proc.$$.corename=/var/tmp/%u/%n.core

should be used.

sysctl -w proc.curproc.corename=/var/tmp/%u/%n.core

changes the value for the sysctl process itself, and will not have the desired effect.

To create the root of a new sub-tree called “local” add some children to the new node, and some descriptions:

sysctl -w //create=local 
sysctl -w //describe=local=my local sysctl tree 
sysctl -w //create=local.esm_debug,type=int,symbol=esm_debug,flags=w 
sysctl -w //describe=local.esm_debug=esm driver debug knob 
sysctl -w //create=local.audiodebug,type=int,symbol=audiodebug,flags=w 
sysctl -w //describe=local.audiodebug=generic audio debug knob

Note that the children are made writable so that the two debug settings in question can be tuned arbitrarily.

To destroy that same subtree:

sysctl -w //destroy=local.esm_debug 
sysctl -w //destroy=local.audiodebug 
sysctl -w //destroy=local

SEE ALSO

sysctl(3), ksyms(4), sysctl(7)

HISTORY

sysctl first appeared in 4.4BSD.