RPC::pServer(3) Perl extension for writing pRPC servers

SYNOPSIS


use RPC::pServer;

$sock = IO::Socket::INET->new('LocalPort' => 9000,
'Proto' => 'tcp',
'Listen' = 5,
'Reuse' => 1);
$connection = new RPC::pServer('sock' => $sock,
'configFile' => $file,
'funcTable' => $funcTableRef,
# other attributes #
);
while ($running) {
$connection->Loop();
if ($connection->error) {
# Do something
}
}

DESCRIPTION

pRPC (Perl RPC) is a package that simplifies the writing of Perl based client/server applications. RPC::pServer is the package used on the server side, and you guess what Net::pRPC::Client is for. See Net::pRPC::Client(3) for this part.

pRPC works by defining a set of of functions that may be executed by the client. For example, the server might offer a function ``multiply'' to the client. Now a function call

    @result = $con->Call('multiply', $a, $b);

on the client will be mapped to a corresponding call

    multiply($con, $data, $a, $b);

on the server. (See the funcTable description below for $data.) The function call's result will be returned to the client and stored in the array @result. Simple, eh? :-)

Server methods

new
The server constructor. Unlike the usual constructors, this one will in general not return immediately, at least not for a server running as a daemon. Instead it will return if a connection is established with a connection object as result. The result will be an error string or the connection object, thus you will typically do a

    $con = RPC::pServer->new ( ...);
    if (!ref($con)) {
        print "Error $con.\n";
    } else {
        # Accept connection
        ...
    }
Accept
Deny
After a connection is established, the server should call either of these methods. If he calls Accept(), he should continue with calling the Loop() method for processing the clients requests.
Loop
When a connection is established, the Loop method must be called. It will process the client's requests. If Loop() returns FALSE, an error occurred. It is the main programs task to decide what to do in that case.
Encrypt
This method can be used to get or set the cipher attribute, thus the encryption mode. If the method is passed an argument, the argument will be used as the new encryption mode. ('undef' for no encryption.) In either case the current encryption mode will be returned. Example:

    # Get the current encryption mode
    $mode = $server->Encrypt();
    # Currently disable encryption
    $server->Encrypt(undef);
    # Switch back to the old mode
    $server->Encrypt($mode);

Server attributes

Server attributes will typically be supplied with the "new" constructor.
configFile
RPC::pServer has a builtin authorization mechanism based on a configuration file. If you want to use this mechanism, just supply the name of a configuration file with the attribute configFile and the server will accept or deny connections based on the configuration file.

The authorization scheme is host based, but you may add user based functionality with the user and password attributes. See ``CONFIGURATION FILE'' below.

client
This attribute is useful in conjunction with the configFile. If the server has authorized a client by using the config file, he will create a hash ref with all the client attributes and store a reference to this hash under the key client. Thus you can easily extend the configuration file for your own purposes, at least as long as host based configuration is sufficient for you.
sock
An object of type IO::Socket, if this program is running as a daemon. An accept() call will be executed on this socket in order to wait for connections. See IO::Socket(3).

An inetd based server should leave this attribute empty: The method will use STDIN and STDOUT instead.

Note: The latter is not yet functionable, I first need to work out how to create an object of type IO::socket for an inetd based server's STDIN and STDOUT. It seems this is currently not supported by IO::Socket.

cipher
This attribute can be used to add encryption quite easily. pRPC is not bound to a certain encryption method, but to a block encryption API. The attribute is an object supporting the methods blocksize, encrypt and decrypt. For example, the modules Crypt::DES and Crypt::IDEA support such an interface.

Do not modify this attribute directly, use the encrypt method instead! However, it is legal to pass the attribute to the constructor.

Example:

    use Crypt::DES;
    $crypt = DES->new(pack("H*", "0123456789abcdef"));
    $client->Encrypt($crypt);
    # or, to stop encryption
    $client->Encrypt(undef);

You might prefer encryption being client dependent, so there is the additional possibility to setup encryption in the server configuration file. See ``CONFIGURATION FILE''. Client encryption definitions take precedence over the cipher attribute.

However, you can set or remove encryption on the fly (putting "undef" as attribute value will stop encryption), but you have to be sure, that both sides change the encryption mode.

funcTable
This attribute is a hash reference. The hash keys are the names of methods, that the client may execute on the server. The hash values are hash references (again). The RPC::pServer module will use the key 'code' only: It contains a code reference to the function performing the clients function call. The first argument of the function call will be the connection object itself, the second will be the 'funcTable' value. You are free to use this hash reference in any way you want, the exception being the 'code' key. The function must return a list: In case of errors the results will be the values 0, followed by a textual error message. In case of success, it ought to return nonzero, followed by the result list being sent to the client.
stderr
a value of TRUE will enable logging messages to STDERR, the default is using syslog(); if the stderr attribute is FALSE, you might call openlog() to configure the application name and facility. See Sys::Syslog(3).
debug
this will cause the server to log debugging information about client requests using the Log method. A value of 0 disables debugging.
application
version
user
password
it is part of the pRPC authorization process, that the client must obeye a login procedure where he will pass an application name, a protocol version and optionally a user name and password. These are not used by pRPC, but when the new method returns with a connection object, the main program may use these for additional authorization.

These attributes are read-only.

io
this attribute is a Storable object created for communication with the client. You may use this, for example, when you want to change the encryption mode with Storable::Encrypt(). See Storable(3).

CONFIGURATION FILE

The server configuration file is currently not much more than a collection of client names or ip numbers that should be permitted or denied to connect to the server. Any client is represented by a definition like the following:

        accept .*\.neckar-alb\.de
            encryption    DES
            key           063fde7982defabc
            encryptModule Crypt::DES
        deny .*

In other words a client definition begins with either "accept pattern" or "deny pattern", followed by some client attributes, each of the attributes being on a separate line, followed by the attribute value. The "pattern" is a perl regular expression matching either the clients host name or IP number. In particular this means that you have to escape dots, for example a client with IP number 194.77.118.1 is represented by the pattern "194\.77\.118\.1".

Currently known attributes are:

encryption
key
encryptionModule
These will be used for creating an encryption object which is used for communication with the client, see Storable(3) for details. The object is created with a sequence like

        use $encryptionModule;
        $cipher = $encryption->new(pack("H*", $key));

encryptionModule defaults to encryption, the reason why we need both is the brain damaged design of the Crypt::IDEA and Crypt::DES modules, which use different module and package names without any obvious reason.

You may add any other attribute you want, thus extending your authorization file. The RPC::pServer module will simply ignore them, but your main program will find them in the client attribute of the RPC::pServer object. This can be used for additional client dependent configuration.

PREDEFINED FUNCTIONS

RPC::pServer offers some predefined methods which are designed for ease in work with objects. In short they allow creation of objects on the server, passing handles to the client and working with these handles in a fashion similar to the use of the true objects.

The handle functions need to share some common data, namely a hash array of object handles (keys) and objects (values). The problem is, how to allocate these variables. By keeping a multithreaded environment in mind, we suggest to store the hash on the stack of the server's main loop.

The handle functions get access to this loop, by looking into the 'handles' attribute of the respective entry in the 'funcTables' hash. See above for a description of the 'funcTables' hash.

See below for an example of using the handle functions.

NewHandle
This method can be inserted into the servers function table. The client may call this function to create objects and receive handles to the objects. The corresponding entry in the function table must have a key classes: This is a list reference with class names. The client is restricted to create objects of these classes only.

The NewHandle function expects, that the constructor returns an object in case of success or 'undef' otherwise. Note, that this isn't true in all cases, for example the RPC::pServer and Net::pRPC::Client classes behave different. In that cases you have to write your own constructor with a special error handling. The StoreHandle method below will help you. Constructors with a different name than new are another example when you need StoreHandle directly.

StoreHandle
After you have created an object on behave of the clients request, you'd like to store it for later use. This is what StoreHandle does for you. It returns an object handle which may be passed back to the client. The client can pass the objects back to the server for use in CallMethod or UseHandle.
NewHandle
The NewHandle is mainly a wrapper for StoreHandle. It creates an object of the given class, passes it to StoreHandle and returns the result. The NewHandle method is designed for direct use within the servers function table.
UseHandle
This is the counterpart of StoreHandle: It gets an object handle, passed by the client, as argument and returns the corresponding object, if any. An 'undef' value will be returned for an invalid handle.
CallMethod
This function receives an object handle as argument and the name of a method being executed. The method will be invoked on the corresponding object and the result will be returned.

A special method is 'DESTROY', valid for any object handle. It disposes the object, the handle becomes invalid.

All handle functions are demonstrated in the following example.

EXAMPLE

Enough wasted time, spread the example, not the word. :-) Let's write a simple server, say a spreadsheet server. Of course we are not interested in the details of the spreadsheet part (which could well be implemented in a separate program), the spreadsheet example is chosen, because it is obvious, that such a server is dealing with complex data structures. For example, a ``sum'' method should be able to add over complete rows, columns or even rectangular regions of the spreadsheet. And another thing, obviously a spread- sheet could easily be represented by perl data structures: The spreadsheet itself could be a list of lists, the elements of the latter lists being hash references, each describing one column. You see, such a spreadsheet is an ideal object for the Storable(3) class. But now, for something completely different:

    #!/usr/local/bin/perl -wT # Note the -T switch! I mean it!
    use 5.0004;               # Yes, this really *is* required.
    use strict;               # Always a good choice.
    use IO::Socket();
    use RPC::pServer;
    # Constants
    $MY_APPLICATION = "Test Application";
    $MY_VERSION = 1.0;
    # Functions that the clients may execute; for simplicity
    # these aren't designed in an object oriented manner.
    # Function returning a simple scalar
    sub sum ($$$$$) {
        my($con, $data, $spreadsheet, $from, $to) = @_;
        # Example: $from = A3, $to = B5
        my($sum) = SpreadSheet::Sum($spreadsheet, $from, $to);
        return (1, $sum);
    }
    # Function returning another spreadsheet, thus a complex object
    sub double ($$$$$) {
        my($con, $data, $spreadsheet, $from, $to);
        # Doubles the region given by $from and $to, returns
        # a spreadsheet
        my($newSheet) = SpreadSheet::Double($spreadsheet, $from, $to);
        (1, $newSheet);
    }
    # Quit function; showing the use of $data
    sub quit ($$) {
        my($con, $data) = @_;
        $$data = 0;   # Tell the server's Loop() method, that we
                      # are done.
        (1, "Bye!");
    }
    # Now we give the handle functions a try. First of all, a
    # spreadsheet constructor:
    sub spreadsheet ($$$$) {
        my ($con, $data, $rows, $cols) = @_;
        my ($sheet, $handle);
        if (!defined($sheet = SpreadSheet::Empty($rows, $cols))) {
            $con->error = "Cannot create spreadsheet";
            return (0, $con->error);
        }
        if (!defined($handle = StoreHandle($con, $data, $sheet))) {
            return (0, $con->error); # StoreHandle stored error message
        }
        (1, $handle);
    }
    # Now a similar function to "double", except that a spreadsheet
    # is doubled, which is stored locally on the server and not
    # remotely on the client
    sub rdouble ($$$$$) {
        my($con, $data, $sHandle, $from, $to);
        my($spreadsheet) = UseHandle($con, $data, $sHandle);
        if (!defined($spreadsheet)) {
            return (0, $con->error); # UseHandle stored an error message
        }
        # Doubles the region given by $from and $to, returns
        # a spreadsheet
        my($newSheet) = SpreadSheet::Double($spreadsheet, $from, $to);
        my($handle);
        if (!defined($handle = StoreHandle($con, $data, $newSheet))) {
            return (0, $con->error); # StoreHandle stored error message
        }
        (1, $newSheet);
    }
    # This function is called for any valid connection to a client
    # In a loop it processes the clients requests.
    #
    # Note, that we are using local data only, thus we should be
    # safe in a multithreaded environment. (Of course, noone knows
    # about the spreadsheet functions ... ;-)
    sub Server ($) {
        my($con) = shift;
        my($con, $configFile, %funcTable);
        my($running) = 1;
        my(%handles) = (); # Note: handle hash is on the local stack
        # First, create the servers function table. Note the
        # references to the handle hash in entries that access
        # the handle functions.
        %funcTable = {
            'sum'         => { 'code' => &sum },
            'double'      => { 'code' => &list },
            'quit'        => { 'code' => &quit,
                               'data' = \$running }
            'spreadsheet' => { 'code' => \&spreadsheet,
                               'handles' => \%handles },
            'rdouble'     => { 'code' => \&rdouble,
                               'handles' = \%handles },
            # An alternative to the 'spreadsheet' entry above;
            'NewHandle'   => { 'code' => \&RPC::pServer::NewHandle,
                               'handles' => \%handles,
                               'classes' => [ 'Spreadsheet' ] },
            # Give client access to *all* (!) spreadsheet methods
            'CallMethod'  => { 'code' => \&RPC::pServer::CallMethod,
                               'handles' => \%handles }
        };
        $con->{'funcTable'} = \%funcTable;
        while($running) {
            if (!$con->Loop()) {
                $con->Log('err', "Exiting.\n"); # Error already logged
                exit 10;
            }
        }
        $con->Log('notice', "Client quits.\n");
        exit 0;
    }
    # Avoid Zombie ball ...
    sub childGone { my $pid = wait; $SIG{CHLD} = \&childGone; }
    # Now for main
    {
        my ($iAmDaemon, $sock);
        # Process command line arguments ...
        ...
        # If running as daemon: Create a socket object.
        if ($iAmDaemon) {
            $sock = IO::Socket->new('Proto' => 'tcp',
                                    'Listen' => SOMAXCONN,
                                    'LocalPort' => 'wellKnownPort(42)',
                                    'LocalAddr' => Socket::INADDR_ANY
                                   );
        } else {
            $sock = undef; # Let RPC::pServer create a sock object
        }
        while (1) {
            # Wait for a client establishing a connection
            my $con = RPC::pServer('sock' => $sock,
                                   'configFile' => 'testapp.conf');
            if (!ref($con)) {
                print STDERR "Cannot create server: $con\n";
            } else {
                if ($con->{'application'} ne $MY_APPLICATION) {
                    # Whatever this client wants to connect to:
                    # It's not us :-)
                    $con->Deny("This is a $MY_APPLICATION server. Go away");
                } elsif ($con->{'version'} > $MY_VERSION) {
                    # We are running an old version of the protocol :-(
                    $con->Deny("Sorry, but this is version $MY_VERSION");
                } elsif (!IsAuthorizedUser($con->{'user'},
                                           $con->{'password'})) {
                    $con->Deny("Access denied");
                } else {
                    # Ok, we accept the client. Spawn a child and let
                    # the child do anything else.
                    my $pid = fork();
                    if (!defined($pid)) {
                        $con->Deny("Cannot fork: $!");
                    } elsif ($pid == 0) {
                        # I am the child
                        $con->Accept("Welcome to the pleasure dome ...");
                        Server();
                    }
                }
            }
        }                       
    }

SECURITY

It has to be said: pRPC based servers are a potential security problem! I did my best to avoid security problems, but it is more than likely, that I missed something. Security was a design goal, but not *the* design goal. (A well known problem ...)

I highly recommend the following design principles:

Protection against trusted users

perlsec
Read the perl security FAQ ("perldoc perlsec") and use the "-T" switch.
taintperl
Use the "-T" switch. I mean it!
Verify data
Never untaint strings withouth verification, better verify twice. For example the CallMethod function first checks, whether an object handle is in a a proper format (currently integer numbers, but don't rely on that, it could change). If it is, then it will still be verified, that an object with the given handle exists.
Be restrictive
Think twice, before you give a client access to a function. In particular, think twice, before you give a client access to objects via the handle methods: If a client can coerce CallMethod() on an object, he has access to *all* methods of that object!
perlsec
And just in case I forgot it: Read the "perlsec" man page. :-)

Protection against untrusted users

Host based authorization
pRPC has a builtin host based authorization scheme; use it! See ``CONFIGURATION FILE''.
User based authorization
pRPC has no builtin user based authorization scheme; that doesn't mean, that you should not implement one.
Encryption
Using encryption with pRPC is extremely easy. There is absolutely no reason for communicating unencrypted with the clients. Even more: I recommend two phase encryption: The first phase is the login phase, where to use a host based key. As soon as the user has authorized, you should switch to a user based key. See the DBD::pNET agent for an example.

AUTHOR

Jochen Wiedmann, [email protected]