The Perl Toolchain Summit needs more sponsors. If your company depends on Perl, please support this very important event.
 / 
Net-IMP-0.51
River stage one • 1 direct dependent • 1 total dependent
++
/ Net::IMP

NAME

Net::IMP - Inspection and Modification Protocol

SYNOPSIS

    package mySessionLog;
    use base 'Net::IMP::Base';
    use Net::IMP;

    # creates factory object
    sub new_factory {
        my ($class,%args) = @_;
        ... create factory object ...
        ... $factory->new_analyzer calls later $class->new_analyzer ...
    }

    # creates new analyzer object, gets %args from factory
    sub new_analyzer {
        my ($class,%args) = @_;
        ... handle private %args ...
        my $self = $class->SUPER::new_analyzer( %args );
        # prepass everything forever in both directions
        $self->add_results(
            [ IMP_PREPASS, 0, IMP_MAXOFFSET ],  # for dir client->server
            [ IMP_PREPASS, 1, IMP_MAXOFFSET ];  # for dir server->client
        );
        return $self;
    }

    # new data for analysis, $offset should only be set if there are gaps
    # (e.g. when we PASSed data with offset in the future)
    sub data {
        my ($self,$dir,$data,$offset) = @_;
        ... log data ...
    }

    package main;
    if (my @err = mySessionLog->validate_cfg(%config)) {
        die "@err"
    }
    my $factory = mySessionLog->new_factory(%config);
    # calls mySessionLog->new_analyzer
    my $analyzer = $factory->new_analyzer(...);
    $analyzer->set_callback(\&imp_cb);

    $analyzer->data(0,'data from dir 0');
    .... will call imp_cb as soon as results are there ...
    $analyzer->data(0,''); # eof from dir 0

    # callback for results
    sub imp_cb {
        for my $rv (@_) {
            my $rtype = shift(@$rv);
            if ( $rtype == IMP_PASS ) ...
            ...
        }
    }

DESCRIPTION

IMP is a protocol for inspection, modification and rejection of data between two sides (client and server) using an analyzer implementing this interface.

Basics

IMP is an asynchronous protocol, usually used together with callbacks.

  • Using the data method, data from the input stream gets put into the analyzer.

  • The analyzer processes the data and generates results. It might be possible, that it needs more data before generating a result or that it can only results for part of the data and needs more data for more results.

    Each result contains a result type. Most results also contain direction of the input stream which caused the result and the offset of this stream. The offset is the position in the input stream, up to which data got used in generating the result, e.g. a result of IMP_PASS means that data up to the offset got used in the result and thus data up to this offset can be passed.

  • The results get usually propagated with a callback set with set_callback. If no callback is set, results can be polled with the poll_results method.

Usage of Terms

Factory

The factory object is used to create analyzers within a specific context.

Context

The context is the environment where the analyzer executes. E.g. when analyzing TCP connections, a new context is created for each TCP connection.

Analyzer

The analyzer is the object which does the analysis of the data within a specific context. It will be created by the factory for a given context.

Methods

The following API is defined.

$class->new_factory(%args) => $factory

This creates a new factory object which is later used to create the context. In the default implementation, an argument rtypes => [ IMP_PASS, IMP_PREPASS... ] can be given where the caller can specify the response types it supports. This will be checked against the list returned by $class->USED_RTYPES(%args) and if the class uses response types not implemented by the caller it will croak.

Also an argument dtypes => [ IMP_DATA_STREAM, IMP_DATA_PACKET...] can be given to specify the list of data types supported by the caller. All dtypes not supported by the implementation will be removed from the given list, maintaining the order of entries in the list. The supported dtypes are determined using $class->supported_dtypes(%args).

$factory->new_analyzer(%args) => $self|undef

Creates a new analyzer object. %args from this call will be merged with %args from the new_factory call and will be used to create the context for the analysis. The details for %args depend on the analyzed protocol and the requirements of the analyzer, but usually these are things like source and destination ip and port, URL, mime type etc.

With a key of cb the callback can already be set here as <[$code,@args]> instead of later with set_callback.

The factory might decide based on the given context information, that no analysis is needed. In this case it will return undef, otherwise the new analyzer object.

$self->set_callback($code,@args)

Sets or changes the callback of the analyzer object. If results are outstanding, they might be delivered to this callback before the method returns.

$code is a coderef while @args are additional user specified arguments which should be used in the callback (typically object reference or similar). The callback is called with $code->(@args,@results) whenever new results are available.

If $code is undef, an existing callback will be removed.

If no callback is given, the results need to be polled with poll_results.

$self->data($dir,$data,$offset,[$type])

Forwards new data to the analyzer. $dir is the direction, e.g. 0 from client and 1 from server. $data are the data. $data of '' means end of data.

$offset is the current position (octet) in the data stream. It must be set to a value greater than 0 after data got omitted as a result of PASS or PASS_PATTERN, so that the analyzer can resynchronize the internal position value with the original position in the data stream. In any other case it should be set to 0.

$type is the type of the data. If not given, IMP_DATA_STREAM will be assumed. There are two global data type definitions:

IMP_DATA_STREAM

This is for streaming data, e.g. chunks from these datatypes can be concatinated and analyzed together, parts can be replaced etc.

IMP_DATA_PACKET

This is for generic packetized data, where each chunk (e.g. call to data) contains a single packet, which should be analyzed as a separate entity. This means no concatinating with previous or future chunks and no replacing of only parts of the packet. Also, any offsets given in calls to data or in the results need to be at packet boundary (or IMP_MAX_OFFSET).

All other data types are considered subtypes of IMP_DATA_PACKET and thus share the same restrictions.

Results will be delivered through the callback or via poll_results.

$self->poll_results => @results

Returns outstanding results. If a callback is attached, no results will be delivered this way.

($factory|class)->supported_dtypes($types,%args)

This method is used withing new_factory to get the list of data types supported by the application. It can also be used later to restrict the types used by the factory to the given types and to associate the type string identifier with the typ number inside the factory based on the values given by the caller.

%args are the arguments from new_analyzer. They are only given when called with class.

$types is a \@list of available types. Each type is a dualvar (see dualvar in Scalar::Util) consisting of a string identifier and a number. Currently defined are:

IMP_DATA_STREAM - (0,'stream')
IMP_DATA_PACKET - (1,'packet')

All other types must have a number>1.

This method returns the list of supported types (dualvars again).

($factory|class)->USED_RTYPES(%args)

This function returns the list of the IMP return codes, which are used by the IMP module. %args are the arguments from new_factory, in case the list of return codes depends on the configuration. They are only given when called with class.

Net::IMP->set_debug

This is just a convinient way to call Net::IMP::Debug->set_debug. See Net::IMP::Debug for more information.

Results

The results returned inside the callback or via poll_results can be of the following kind:

[ IMP_PASS, $dir, $offset ]

Accept all data up to $offset in the data stream for direction $dir.

If $offset specifies data which were not yet seen by the analyzer, these data don't need to be forwarded to analyzer. If they were still forwarded to the analyzer (because they were already on the way, unstoppable) the analyzer just throws them away until $offset is reached. This feature is useful for ignoring whole subcontexts (like MIME content based on a Content-length header).

A special case is a $offset of IMP_MAXOFFSET, in this case the analyzer is not interested in further information about the connection.

[ IMP_PASS_PATTERN, $dir, $regex, $len ]

This is the same as IMP_PASS, except a pattern will be given instead of an offset. All data up to but not including the pattern don't need to be forwarded to the analyzer. Because $regex might be complex the analyzer has to specify how many octets the $regex might match at most, so that the caller can adjust its buffer.

Because there might be data already on the way to the analyzer, the analyzer needs to check all incoming data without explicit offset if they match the pattern. If it gets data with explicit offset, that means, that the pattern was matched inside the client at the specified position. In this case it should remove all data it got before (even if they included offset already) and resync at the specified offset.

For better performance the analyzer should check any data it has already in the buffer if they already contain the pattern. In this case the issue can be dealt internally and there is no need to send this reply to the caller.

If the caller receives this reply, it should check all data it has still in the buffer (e.g. which were not passed) wether they contain the pattern. If the caller finds the pattern, it should call data with an explicit offset, so that the analyzer can resynchronize the position in the data stream.

[ IMP_PREPASS, $dir, $offset ]

This is similar to IMP_PASS. If <$offset> specifies data, which were already forwarded to the analyzer, they get accepted. If it specified not yet forwarded data, they get accepted also up to $offset, but contrary to IMP_PASS they get also forwarded to the analyzer.

Thus data can be forwarded before they get inspected, but they get inspected nevertheless. This might be known good data, but inspection is needed to maintain the state or to log the data.

Or it might be potentially bad data, but a low latency is required and small amounts of bad data are accepted. In this case the window for bad data might be set small enough to allow high latency while limiting impact of malicious data. This can be done through continues updates of $offset.

[ IMP_DENY, $dir, $reason ]

Deny any more data on this context. If $reason is given, it should be used to construct a message to the client.

Deny results by closing the context in a way visible to the client (e.g. closing the connection with RST).

[ IMP_DROP ]

Deny any more data on this context and close the context. The preferred way for closing the context is to be not visible to the client (e.g just drop any more packets of an UDP connection).

[ IMP_REPLACE, $dir, $offset, $data ]

Ignore the original data up to $offset, instead send $data. $offset needs be be in the range of the data the analyzer got through data method.

[ IMP_TOSENDER, $dir, $data ]

Send data back to the sender. This might be used to reject data, e.g. replace them with nothing and send an error message back to the sender. This can be useful to reject single commands in SMTP, FTP...

[ IMP_LOG, $dir, $offset, $len, $level, $msg ]

This contains a log message $msg which is about data in direction $dir starting with $offset and $len octets long. $level might specify a log level like debug, info, warn... .

The caller should just log the information in this case.

$level is one of LOG_IMP_*, which are similar to syslog levels, e.g. IMP_LOG_DEBUG, IMP_LOG_INFO,... These level constants can be imported with use Net::IMP ':log'.

[ IMP_PORT_OPEN|IMP_PORT_CLOSE, $dir, $offset, ... ]

Some protocols like FTP, SIP, H.323 dynamically allocate ports. These results detect when port allocation/destruction is done and should provide enough information for the caller to open/close the ports and track the data through additional analyzers.

TODO: details will be specified when this feature is needed.

[ IMP_ACCTFIELD, $key, $value ]

This specifies a tuple which should be used for accounting (like name of logfile, URL...)

TODO

  • sample integration into relayd

  • optimizing initial setup

    Optimizing initial setup, so that IMP_PREPASS IMP_MAXOFFSET could be set w/o getting data first.

  • define level in IMP_LOG

  • specify IMP_PORT_*

    Specify IMP_PORT_* and have sample implementation which uses it.

  • behavior on EOF

    There is currently no way for the analyzer to issue a IMP_REPLACE on read-shutdown on one side, because the IMP client will forward the shutdown once all buffers are empty. It might be possible solution to require the analyzer to explicitly acknowledge the processing of the shutdown by sending an IMP_PASS with an offset after the connection end.

AUTHOR

Steffen Ullrich <sullr@cpan.org>

Thanks to everybody who helped with time, ideas, reviews or bug reports, notably Alexander Bluhm and others at genua.de.

COPYRIGHT

Copyright by Steffen Ullrich.

This module is free software; you can redistribute it and/or modify it under the same terms as Perl itself.