++ed by:

1 non-PAUSE user(s).

Steffen Ullrich

NAME

Net::Inspect::L7::HTTP - guesses and handles HTTP traffic

SYNOPSIS

 my $req = Net::Inspect::L7::HTTP::Request::Simple->new(..);
 my $http = Net::Inspect::L7::HTTP->new($req);
 my $guess = Net::Inspect::L5::GuessProtocol->new;
 $guess->attach($http);
 ...

DESCRIPTION

This class extracts HTTP requests from TCP connections. It provides all hooks required for Net::Inspect::L4::TCP and is usually used together with it. It provides the guess_protocol hook so it can be used with Net::Inspect::L5::GuessProtocol.

Attached flow is usually a Net::Inspect::L7::HTTP::Request::* object.

Hooks provided:

guess_protocol($guess,$dir,$data,$eof,$time,$meta)
new_connection($meta,%args)

This returns an object for the connection. With $args{header_maxsize} the maximum size of the message headers can be given, that is:

  $args{header_maxsize}[0] - request header, default 64k
  $args{header_maxsize}[1] - response header, default 16k
  $args{header_maxsize}[2] - chunked header, default 2k
$connection->in($dir,$data,$eof,$time)

Processes new data and returns number of bytes processed. Any data not processed must be sent again with the next call.

$data are the data as string. In some cases $data can be [ 'gap' => $len ], e.g. only the information, that there would be $len bytes of data w/o submitting the data. These should only be submitted in request and response bodies and only if the attached layer can handle these gaps in the in_request_body and in_response_body methods.

Gaps on other places are not allowed, because all other data are needed for interpreting the placement of request, response and data inside the connection.

$connection->fatal($reason,$dir,$time)

Hooks called:

new_request(\%meta,$conn)

This should return an request object. The reference to the connection object is given in case the request object likes to call fatal to end the connection.

The function should not get hold of $conn, e.g. only store a weak reference, otherwise memory might leak.

$request->in_request_header($header,$time,\%hdr_meta)

Called when the full request header is read. $header is the string of the header.

%hdr_meta contains information extracted from the header:

method - method of request
url - url, as given in request
version - version of HTTP spoken in request
info - first line of request (method url version)
fields - (key => \@values) hash of header fields
junk - invalid data found in header fields part
content_length - length of request body
chunked - true if body uses transfer encoding chunked
upgrade - contains hash when protocol upgrade was requested

Currently this hash contains the key websocket with the value of the sec-websocket-key if a valid request for a Websocket upgrade was detected.

expect - contains hash for expectations from Expect header

Currently the only possible key is 100-continue.

$request->in_response_header($header,$time,\%hdr_meta)

Called when the full response header is read. $header is the string of the header.

%hdr_meta contains information extracted from the header:

If a keep-alive connection was closed by the server after the request was transmitted but after the response header was sent this function is called once with an empty header to signal this (normal) condition.

version - version of HTTP spoken in response
code - status code from response
reason - reason given for response code
fields - (key => \@values) hash of header fields
junk - invalid data found in header fields part
content_length - length of request body if known, else undef
chunked - true if body uses transfer encoding chunked
upgrade - new protocol when switching protocols, e.g. 'websocket'
preliminary - true if this is a preliminary response
$request->in_request_body($data,$eobody,$time)

Called for a chunk of data of the request body. $eobody is true if this is the last chunk of the request body. If the request body is empty the method will be called once with ''. This function will not be called for CONNECT requests because these are special.

$data can be [ 'gap' => $len ] if the input to this layer were gaps.

$request->in_response_body($data,$eobody,$time)

Called for a chunk of data of the response body. $eof is true if this is the last chunk of the connection. $eobody is true if this is the last chunk of the response body. If the response body is empty the method will be called once with ''.

This function will not be called for CONNECT requests or for successful protocol upgrades caused by a valid reply of 101 Switching Protocols. In this case instead in_data or a special protocol callback will be used to handle any future data.

$data can be [ 'gap' => $len ] if the input to this layer were gaps.

$request->in_chunk_header($dir,$header,$time)

will be called with the chunk header for chunked encoding. Usually one is not interested in the chunk framing, only in the content so that this method will be empty. Will be called before the chunk data.

$request->in_chunk_trailer($dir,$trailer,$time)

will be called with the chunk trailer for chunked encoding. Usually one is not interested in the chunk framing, only in the content so that this method will be empty. Will be called after in_response_body/in_request_body got called with eof true.

$request->in_data($dir,$data,$eof,$time)

Will be called for any data after successful upgrade with a CONNECT request, unless the request object specifically handled these upgrades by implementing a upgrade_CONNECT or upgrade_ANY method. $dir is 0 for data from client, 1 for data from server.

$request->in_junk($dir,$data,$eof,$time)

Will be called for legally ignored junk (empty lines) in front of request or response body. $dir is 0 for data from client, 1 for data from server.

$request->fatal($reason,$dir,$time)

will be called on fatal errors, mostly protocol iregularities.

Methods suitable for overwriting:

new_request(\%meta)

default implementation will just call new_request from the attached flow

Helpful methods

$connection->dump_state

collects the state of the open connections. If defined wantarray it will return a message, otherwise output it via xdebug

$connection->offset(@dir)

returns the current offset(s) in the data stream, that is the position behind the within the in_* methods forwarded data.

$connection->gap_offset(@dir)

If the next bytes of the input stream are not needed to interpret the HTTP protocol (i.e. plain body data) this gives the offsets up to which data are "gapable". If no gaps are possible at the current state 0 will be returned. If everything can be gaps (usually because end of body is caused by end of connection) -1 will be returned.

$connection->gap_diff(@dir)

This is similar to gap_offset but will return the difference from the current position, i.e. how large the next gap can be. -1 again means an unlimited gap.

$connection->set_gap_diff($dir,$diff)

This function is used internally by upgrade handlers to change the idea how large the next gap can be. If $diff is defined it is assumed that the next $diff bytes could be skipped without loosing information necessary to keep maintain the proper connection state. If $diff is not defined it will be assumed that no data can be skipped.

$connection->open_requests(@index)

in array context returns the objects for the open requests, in scalar context the number of open requests. If index is given only the specified objects will be returned, e.g. index -1 is the object currently receiving response data while index 0 specifies the object currently receiving request data (both are the same unless pipelining is used)

Protocol Upgrades

Protocol upgrades are usually done by the server responding with a status code of 101 Switching Protocols to a request containing a Upgrade header. A different kind of upgrade is done with the CONNECT request.

These kind of upgrades are handled in a generic way by calling the connect_$method method of the request object. $method is the lower case name of the new protocol as set in the Upgrade header of the response. For upgrades done with the CONNECT method "CONNECT" will be used instead. The upgrade handler is called as $request_object->upgrade_$method($self,$request,$response). See Net::Inspect::L7::HTTP::WebSocket for a more detailed description of this function and its arguments.

If there is no method specific function it will try upgrade_ANY. This call is similar to upgrade_$method but with an additional argument for the method, e.g. $request_object->upgrade_$method($self,$request,$response,$method). If this function is also not defined it will use a built-in handler calling in_data for the CONNECT method. For any other methods the connection will be considered bad because it does not know how to handle the remaining data.

exportable utility functions and constants

METHODS_WITHOUT_RQBODY

This constant is an array reference of all request methods which will not have a request body, i.e. which have an implicit and non-changeble content-length of 0.

METHODS_WITH_RQBODY

This constant is an array reference of all request methods which must have a specified request body, even if the content-lenth is explicitly set to 0.

Methods which are not in METHODS_WITH_RQBODY or METHODS_WITHOUT_RQBODY might have a request body, that is if no content-length is explicitly given (or chunked transfer encoding is used) it is assumed that they don't have a body.

METHODS_WITHOUT_RPBODY

This constant is an array reference of all request methods which don't require a response body, i.e. which have an implicit and non-changeble content-length of 0.

CODE_WITHOUT_RPBODY

This constant is an array reference of all response codes which will not have a response body, i.e. which have an implicit and non-changeble content-length of 0.

parse_hdrfields($header,\%fields) -> $bad_lines

This function parses the given message header (without request or status line!) and extracts the key:value pairs into %fields. Each key in %fields is the lower-case representation of the key from the HTTP message and the value in %fields is a list with all values, i.e. a list with a single element if the specific key was only used once the header, but with multiple elements if the key was used multiple times. Any continuation lines will be transformed into a single line.

It will return any remaining data in $header which could not be interpreted as proper key:value pairs. If the message contains no errors it will thus return ''.

parse_reqhdr($string,\%header,[$external_length]) -> $bad_header

This will parse the given $string as a request header and extract information into \%header. These information then later will be given to in_request_header. See there for more details about the contents of the hash.

If $external_length is true it will not complain if a content-length is required but not defined.

parse_rsphdr($string,\%request,\%header) -> $bad_header

This will parse the given $string as a response header and extract information into \%header. These information then later will be given to in_request_header. See there for more details about the contents of the hash.

%request contains information about the request. One might simple use the hash filled by parse_reqhdr here. If not at least the information about method, expect and upgrade must be provided because they are needed to interpret the response correctly.