POE::Filter::HTTPD(3) parse simple HTTP requests, and serialize HTTP::Response

SYNOPSIS


#!perl
use warnings;
use strict;
use POE qw(Component::Server::TCP Filter::HTTPD);
use HTTP::Response;
POE::Component::Server::TCP->new(
Port => 8088,
ClientFilter => 'POE::Filter::HTTPD', ### <-- HERE WE ARE!
ClientInput => sub {
my $request = $_[ARG0];
# It's a response for the client if there was a problem.
if ($request->isa("HTTP::Response")) {
my $response = $request;
$request = $response->request;
warn "ERROR: ", $request->message if $request;
$_[HEAP]{client}->put($response);
$_[KERNEL]->yield("shutdown");
return;
}
my $request_fields = '';
$request->headers()->scan(
sub {
my ($header, $value) = @_;
$request_fields .= (
"<tr><td>$header</td><td>$value</td></tr>"
);
}
);
my $response = HTTP::Response->new(200);
$response->push_header( 'Content-type', 'text/html' );
$response->content(
"<html><head><title>Your Request</title></head>" .
"<body>Details about your request:" .
"<table border='1'>$request_fields</table>" .
"</body></html>"
);
$_[HEAP]{client}->put($response);
$_[KERNEL]->yield("shutdown");
}
);
print "Aim your browser at port 8088 of this host.\n";
POE::Kernel->run();
exit;

DESCRIPTION

POE::Filter::HTTPD interprets input streams as HTTP 0.9, 1.0 or 1.1 requests. It returns a HTTP::Request objects upon successfully parsing a request.

On failure, it returns an HTTP::Response object describing the failure. The intention is that application code will notice the HTTP::Response and send it back without further processing. The erroneous request object is sometimes available via the ``$r->request'' in HTTP::Response method. This is illustrated in the ``SYNOPSIS''.

For output, POE::Filter::HTTPD accepts HTTP::Response objects and returns their corresponding streams.

Please see HTTP::Request and HTTP::Response for details about how to use these objects.

HTTP headers are not allowed to have UTF-8 characters; they must be ISO-8859-1. POE::Filter::HTTPD will convert all UTF-8 into the MIME encoded equivalent. It uses utf8::is_utf8 for detection-8 and Email::MIME::RFC2047::Encoder for convertion. If utf8 is not installed, no conversion happens. If Email::MIME::RFC2047::Encoder is not installed, utf8::downgrade is used instead. In this last case, you will see a warning if you try to send UTF-8 headers.

PUBLIC FILTER METHODS

POE::Filter::HTTPD implements the basic POE::Filter interface.

new

new() accepts a list of named parameters.

"MaxBuffer" sets the maximum amount of data the filter will hold in memory. Defaults to 512 MB (536870912 octets). Because POE::Filter::HTTPD copies all data into memory, setting this number to high would allow a malicious HTTPD client to fill all server memory and swap.

"MaxContent" sets the maximum size of the content of an HTTP request. Defaults to 1 MB (1038336 octets). Because POE::Filter::HTTPD copies all data into memory, setting this number to high would allow a malicious HTTPD client to fill all server memory and swap. Ignored if ``Streaming'' is set.

"Streaming" turns on request streaming mode. Defaults to off. In streaming mode this filter will return either an HTTP::Request object or a block of content. The HTTP::Request object's content will return empty. The blocks of content will be parts of the request's body, up to Content-Length in size. You distinguish between request objects and content blocks using "Scalar::Util/bless" (See ``Streaming request'' below). This option supersedes ``MaxContent''.

CAVEATS

Some versions of libwww are known to generate invalid HTTP. For example, this code (adapted from the HTTP::Request::Common documentation) will cause an error in a POE::Filter::HTTPD daemon:

NOTE: Using this test with libwww-perl/5.834 showed that it added the proper HTTP/1.1 data! We're not sure which version of LWP fixed this. This example is valid for older LWP installations, beware!

  use HTTP::Request::Common;
  use LWP::UserAgent;
  my $ua = LWP::UserAgent->new();
  $ua->request(POST 'http://example.com', [ foo => 'bar' ]);

By default, HTTP::Request is HTTP version agnostic. It makes no attempt to add an HTTP version header unless you specifically declare a protocol using "$request->protocol('HTTP/1.0')".

According to the HTTP 1.0 RFC (1945), when faced with no HTTP version header, the parser is to default to HTTP/0.9. POE::Filter::HTTPD follows this convention. In the transaction detailed above, the Filter::HTTPD based daemon will return a 400 error since POST is not a valid HTTP/0.9 request type.

Upon handling a request error, it is most expedient and reliable to respond with the error and shut down the connection. Invalid HTTP requests may corrupt the request stream. For example, the absence of a Content-Length header signals that a request has no content. Requests with content but without that header will be broken into a content-less request and invalid data. The invalid data may also appear to be a request! Hilarity will ensue, possibly repeatedly, until the filter can find the next valid request. By shutting down the connection on the first sign of error, the client can retry its request with a clean connection and filter.

Streaming Request

Normally POE::Filter::HTTPD reads the entire request content into memory before returning the HTTP::Request to your code. In streaming mode, it will return the content separately, as unblessed scalars. The content may be split up into blocks of varying sizes, depending on OS and transport constraints. Your code can distinguish the request object from the content blocks using ``blessed'' in Scalar::Util.

    use Scalar::Util;
    use POE::Wheel::ReadWrite;
    use POE::Filter:HTTPD;
    $heap->{wheel} = POE::Wheel::ReadWrite->new( 
                        InputEvent => 'http_input',
                        Filter => POE::Filter::HTTPD->new( Streaming => 1 ),
                        # ....
                );
    sub http_input_handler
    {
        my( $heap, $req_or_data ) = @_[ HEAP, ARG0 ];
        if( blessed $req_or_data ) {
            my $request = $req_or_data;
            if( $request->isa( 'HTTP::Response') ) {
                # HTTP error
                $heap->{wheel}->put( $request );
            }
            else {
                # HTTP request
                # ....
            }
        }
        else {
            my $data = $req_or_data;
            # ....
        }
    }

You may trivally create a DoS bug if you hold all content in memory but do not impose a maximum Content-Length. An attacker could send "Content-Length: 1099511627776" (aka 1 TB) and keep sending data until all your system's memory and swap is filled.

Content-Length has been sanitized by POE::Filter::HTTPD so checking it is trivial :

    if( $request->headers( 'Content-Length' ) > 1024*1024 ) {
        my $resp = HTTP::Response->new( RC_REQUEST_ENTITY_TOO_LARGE ), 
                                             "So much content!" ) 
        $heap->{wheel}->put( $resp );
        return;
    }

If you want to handle large amounts of data, you should save the content to a file before processing it. You still need to check Content-Length or an attacker might fill up the partition.

    use File::Temp qw(tempfile);
    if( blessed $_[ARG0] ) {
        $heap->{request} = $_[ARG0];
        if( $heap->{request}->method eq 'GET' ) {
            handle_get( $heap );
            delete $heap->{request};
            return;
        }
        my( $fh, $file ) = tempfile( "httpd-XXXXXXXX", TMPDIR=>1 );
        $heap->{content_file} = $file;
        $heap->{content_fh} = $fh;
        $heap->{content_size} = 0;
    }
    else {
        return unless $heap->{request};
        $heap->{content_size} += length( $_[ARG0] );
        $heap->{content_fh}->print( $_[ARG0] );
        if( $heap->{content_size} >= $heap->{request}->headers( 'content-length' ) ) {
            delete $heap->{content_fh};
            delete $heap->{content_size};
            # Now we can parse $heap->{content_file}
            if( $heap->{request}->method eq 'POST' ) {
                handle_post( $heap );
            }
            else {
                # error ...
            }
        }
    }
    sub handle_post
    {
        my( $heap ) = @_;
        # Now we have to load and parse $heap->{content_file}            
        # Next 6 lines make the data available to CGI->init
        local $ENV{REQUEST_METHOD} = 'POST';
        local $CGI::PERLEX = $CGI::PERLEX = "CGI-PerlEx/Fake";
        local $ENV{CONTENT_TYPE} = $heap->{req}->header( 'content-type' );
        local $ENV{CONTENT_LENGTH} = $heap->{req}->header( 'content-length' );
        my $keep = IO::File->new( "<&STDIN" ) or die "Unable to reopen STDIN: $!";
        open STDIN, "<$heap->{content_file}" or die "Reopening STDIN failed: $!";
        my $qcgi = CGI->new();
        # cleanup
        open STDIN, "<&".$keep->fileno or die "Unable to reopen $keep: $!";
        undef $keep;
        unlink delete $heap->{content_file};
        # now use $q as you would normaly
        my $file = $q->upload( 'field_name' );
        
        # ....
    }
    sub handle_get
    {
        my( $heap ) = @_;
        # 4 lines to get data into CGI->init
        local $ENV{REQUEST_METHOD} = 'GET';
        local $CGI::PERLEX = $CGI::PERLEX = "CGI-PerlEx/Fake";   
        local $ENV{CONTENT_TYPE} = $heap->{req}->header( 'content-type' );
        local $ENV{'QUERY_STRING'} = $heap->{req}->uri->query;
        my $q = CGI->new();
        # now use $q as you would normaly
        # ....
    }

Streaming Response

It is possible to use POE::Filter::HTTPD for streaming content, but an application can use it to send headers and then switch to POE::Filter::Stream.

From the input handler (the InputEvent handler if you're using wheels, or the ClientInput handler for POE::Component::Server::TCP):

  my $response = HTTP::Response->new(200);
  $response->push_header('Content-type', 'audio/x-mpeg');
  $_[HEAP]{client}->put($response);
  $_[HEAP]{client}->set_output_filter(POE::Filter::Stream->new());

Then the output-flushed handler (FlushEvent for POE::Wheel::ReadWrite, or ClientFlushed for POE::Component::Server::TCP) can put() chunks of the stream as needed.

  my $bytes_read = sysread(
    $_[HEAP]{file_to_stream}, my $buffer = '', 4096
  );
  if ($bytes_read) {
    $_[HEAP]{client}->put($buffer);
  }
  else {
    delete $_[HEAP]{file_to_stream};
    $_[KERNEL]->yield("shutdown");
  }

BUGS

Many aspects of HTTP 1.0 and higher are not supported, such as keep-alive. A simple I/O filter can't support keep-alive, for example. A number of more feature-rich POE HTTP servers are on the CPAN. See <http://search.cpan.org/search?query=POE+http+server&mode=dist>

AUTHORS & COPYRIGHTS

POE::Filter::HTTPD was contributed by Artur Bergman. Documentation is provided by Rocco Caputo.

Please see POE for more information about authors and contributors.