The devfsd programme is a daemon, run by the system boot
scripts, which can provide for intelligent management of device
entries in the Device Filesystem (devfs). It is desirable to start the
daemon at the beginning of the boot scripts, in particular before
filesystem checks.
As part of its setup phase devfsd creates certain symbolic
links which are compiled into the code. These links are required by
/usr/src/linux/Documentation/devices.txt. This behaviour may
change in future revisions.
devfsd will read the special control file .devfsd in a
mounted devfs, listening for the creation and removal of device
entries (this is termed a change operation). For each change
operation, devfsd can take many actions. The daemon will
normally run itself in the background and send messages to syslog.
The opening of the syslog service is automatically delayed until
/dev/log is created.
At startup, before switching to daemon mode, devfsd will scan
the mounted device tree and will generate synthetic REGISTER
events for each leaf node.
OPTIONS
mntpnt
This specifies a mount point for where devfs is mounted. This is
typically "/dev". The daemon will change directory to this mount point
before opening any files.
-v
This option will print the protocol version numbers for devfsd
and the kernel-side protocol version provided by devfs itself. The
programme then exits.
-d
Run devfsd in debug mode. Used for debugging the kernel-side
protocol implemented by devfs. This also prints the protocol version
numbers. In this mode the daemon runs in the foreground.
-t num
Set the trace level to num. If the level is greater than 0 then
output is sent to stderr rather than syslog(3). If the level is
greater than 1 then devfsd will run in trace mode. Higher levels
yield more trace information.
-D mask
Set the debug mask for devfs.
-fg
Run the daemon in the foreground.
-np
Exit after parsing the configuration file and processing synthetic
REGISTER events. Do not poll for events. This is primarily used
for creating compatibility entries without needing a daemon running in
the background.
CONFIGURATION
The configuration file /etc/devfsd.conf configures the devfsd
programme. It is a simple ASCII file with one configuration option per
line. Comment lines must start with a leading '#' character. Comment
lines and blank lines are ignored. Each configuration option is a
keyword followed by zero or more parameters, depending on the
option. The following section details the configuration options. Below
are the permitted keywords:
INCLUDE location
Include the configuration file named by location. Variable
expansion is applied to location (see the section on VARIABLE
EXPANSION below). If this is a directory, then all files (except those
which start with '.') are read. This is recursive (i.e. files which
are in fact subdirectories are also read).
If the first character of location is "+",
then the rest of location is the name of an NIS map to load.
This operation is silently ignored if the NIS domain has not yet been
set (SIGHUP should then be sent after the NIS domain has been
set).
OPTIONAL_INCLUDE location
As above, except that if the file does not exist, it is silently
ignored.
CLEAR_CONFIG
Clear the current configuration.
The config file can also be used to specify actions that should be
taken at specified events. Each line specifies an event and
action. The syntax is:
EVENT devname ACTION [args...]
where EVENT is the event and ACTION is the action to take
when the event occurs. The action is only taken when devname
matches the name of the device entry affected (this does not include
the mount point for the filesystem). This is processed as a regular
expression. Some actions support extra arguments, passed as
args.
Variable expansion is applied to the arguments (see the section on
VARIABLE EXPANSION below). After variable expansion, regular
expression substitution is performed (see the section on REGULAR
EXPRESSION SUBSTITUTION). Note that the following special variable
names are also recognised:
devpath
the full path name of the new device
devname
the name of the new device inside the devfs namespace
mode
the mode of the device entry in octal
uid
the user ID of the process or inode
gid
the group ID of the process or inode
The following events are recognised:
REGISTER
The device entry was registered by a device driver.
UNREGISTER
The device entry was unregistered by a device driver.
ASYNC_OPEN
The inode was opened (the opening process does not wait for a
response).
CLOSE
The file was closed.
LOOKUP
An inode lookup was performed and there was no device entry. This
event is not sent if the initiating process is devfsd or one of its
children.
CHANGE
Some inode attributes where changed.
CREATE
An inode was created by a process.
DELETE
An inode was deleted by a process.
The following actions are recognised:
PERMISSIONS owner_and_group access_mode
owner_and_group specifies the owner and group that the file
should be set to. This must be of the form "user.group" and either
"user" or "group" component may be symbolic or numeric. To specify
that the user or group be left alone, use a numeric value of "-1" for
the respective component. access_mode specifies the access mode
the file should be set to. This must be either an octal value or a
symbolic string of nine characters with the form "rwxrwxrwx". Where
access is not given, a '-' character should be used in place
(e.g. "rw-rw-r--" gives read and write access to the user and group
and only read access to everybody else).
MODLOAD
This action will pass "/dev/$devname" (i.e. "/dev/" prefixed to the
device name) to the module loading facility. In addition, the
/etc/modules.devfs configuration file is used.
EXECUTE path [arg...]
This action allows you to run a programme. path is the pathname
of the programme to run and arg is a set of optional arguments
passed to the programme (maximum 6). path is the first argument
(i.e. argv[0]) passed to the programme.
MFUNCTION path function [arg...]
This action allows you to run a "main"-style function within a shared
object. path is the pathname of the shared object, with the
special path of "GLOBAL" signifying all global symbols (e.g. from
libc). function is the name of the function symbol to run and
arg is a set of optional arguments passed to the function
(maximum 5). path is the first argument (i.e. argv[0]) passed to
the function. The prototype for this function is similar to the
main function of C programmes. The prototype is:
int func (int argc, char **argv);
CFUNCTION path function [arg...]
This action is similar to the MFUNCTION action, except for the
calling convention. The prototype for this function is:
Up to 5 arguments may be passed, each being of type void *. The
following special arguments are recognised:
EVENT
a struct devfsd_notify_struct * is passed
COPY source destination
This action will copy the device type and permissions of source
to destination.
IGNORE
This action causes all subsequent processing for the device to be
ignored.
MKOLDCOMPAT
This action creates an "old" compatibility entry for the device.
MKNEWCOMPAT
This action creates a "new" compatibility entry for the device.
RMOLDCOMPAT
This action removes an "old" compatibility entry for the device.
RMNEWCOMPAT
This action removes a "new" compatibility entry for the device.
VARIABLE EXPANSION
A subset of normal Bourne shell variable expansion is applied to
various expressions. The currently supported subset is:
$variable, ${variable} and ${variable:-word}.
Variables are taken from the environment. The following variable names
are also defined :
hostname
The hostname of the machine
mntpnt
The mount point for devfs
REGULAR EXPRESSION SUBSTITUTION
Sections of the matched regular expression can be included in
an action. Use \0 to refer to the entire regular expression
matched, \1 to refer to the first parenthesized subexpression,
\2 to refer to the second, and so on. (Use \\ to
include an actual backslash.)
SIGNALS
devfsd responds to signals in a variety of ways:
SIGINT
devfsd will exit cleanly.
SIGTERM
devfsd will exit cleanly.
SIGHUP
The configuration file is re-read and any shared objects are
reloaded. Then the mounted device tree is scanned and synthetic
REGISTER events are generated for each leaf node.
EXAMPLES
Create and destroy old compatibility entries:
REGISTER .* MKOLDCOMPAT
UNREGISTER .* RMOLDCOMPAT
Create and destroy new compatibility entries:
REGISTER .* MKNEWCOMPAT
UNREGISTER .* RMNEWCOMPAT
Load modules:
LOOKUP .* MODLOAD
Make and remove a symbolic link:
REGISTER mydir/mydev CFUNCTION GLOBAL symlink $devname mydev
UNREGISTER mydir/mydev CFUNCTION GLOBAL unlink mydev
the configuration file. If this file is missing or has zero size,
devfsd will exit after its setup phase.
/etc/modules.devfs
the generic module configuration file (required for the MODLOAD
action), which is installed with devfsd. This in turn includes
/etc/modules.conf
CAVEATS
Make sure you understand the implications of regular expression
matching. For example, if you had a configuration line such as:
LOOKUP cdrom CFUNCTION GLOBAL symlink ${mntpnt}/cdroms/cdrom0 $devpath
then referencing "/dev/cdrom1" will create the symbolic link
"/dev/cdrom1". Further, referencing "/dev/cdroms/1" would also create
the "/dev/cdroms/1" symbolic link. This is probably not what you want.
If there was not already a "/dev/cdroms" directory, then you would get
a "/dev/cdrom" symbolic link. Definately not what you want!
The correct configuration line would be:
LOOKUP ^cdrom$ CFUNCTION GLOBAL symlink ${mntpnt}/cdroms/cdrom0 $devpath
A mailing list exists for devfs kernel patch and devfsd release
announcements. This list also has a small amount of discussion traffic
(a few messages per week). You may subscribe by sending a message to
majordomo@oss.sgi.com with subscribe devfs in the message
body.
