After the lpd server has done
its initialization, it will attempt to bind to the lpd listening port specified by the lpd_port value. This value has the
format [ipaddr%]port. If the ipaddr is specified then the lpd server
binds to the interface with the specified address otherwise
it binds to all interfaces. The port value can be a number or
name the name of a service; The port corresponding to the
service name is used. The printer
services port is 515. If the port binding operation is
successful and the server has not been request to run in foreground mode by the -F command line option, then a child process
is forked and the parent process will exit. The child process
then takes steps to disconnect itself from the control
terminal of the process that started it.
The main lpd process will then
start a queue checking process that
will check all of the spool queues used by the server for
queues that have pending jobs. This process sends a message
to the main lpd process requesting
that it start a service process for this queue.
The lpd process will then sit
in a loop waiting for one of the following events:
An incoming connection request. If the maximum number
of children has not been exceeded, then a new process
will be forked to handle this connection.
A child process exiting. The server will check to see
if there is a pending request to start a server for a
queue that could not be accommodated due to too many
processes running.
A request to start a service process for a queue. If
the number of active processes is less than the maximum
allowed a service process will be started, otherwise the
request will be placed on a list for service when the
number of processes active decreases.
A timeout alarm for the queue rescanning operation.
This is discussed in the next section in detail.
When connection is accepted by the lpd spooler, the following steps are taken
to processes the job.
First, a timeout is established for the transfer of
the information from client to the lpd server. This is done to prevent a
denial of service attack by processes that do not close
connections in a timely manner.
A single line is read into an internal buffer. This
line must be terminated with a NEWLINE character.
The input line is parsed and the actions required are
determined.
If the activity requires access to the spool queue
information, then the current directory of the process is
changed to the spool directory. This allows all file
accesses to then be relative to this directory.
If the processing requires starting a spool queue
server process, a message is sent to the main lpd server process to start a spool
queue server process. By having all the processes serving
spool queues children of the main server process it is
possible to monitor and limit the total number of active
processes. This is important on systems with a very large
number of queues.
After the processing of the original request has been
completed, the process with then check to see if the
Spool Queue for the printer should be processed.