[FASTCGI] Length Incosistency Between FCGI_RECORD and FCGI_PARAMS

[Volkan YAZICI]

Hi,

AFAIU, FCGI_PARAMS are wrapped in FCGI_RECORD data types. (Please
correct me if I am wrong.) And there are four types of FCGI_PARAMS
message formats with variable length headers:

  1. {KEY-LEN[1], VAL-LEN[1], KEY[KEY-LEN], VAL[VAL-LEN]}
  2. {KEY-LEN[1], VAL-LEN[4], KEY[KEY-LEN], VAL[VAL-LEN]}
  3. {KEY-LEN[4], VAL-LEN[1], KEY[KEY-LEN], VAL[VAL-LEN]}
  4. {KEY-LEN[4], VAL-LEN[4], KEY[KEY-LEN], VAL[VAL-LEN]}

But an FCGI_RECORD message header format is of fixed length. Now
consider below message:

  {VERSION    [1],
   TYPE[1],                  // FCGI_PARAMS Type
   REQUEST-ID [2],
   CONTENT-LEN[2],
   PADDING-LEN[1],
   RESERVED   [1],
   CONTENT    [CONTENT-LEN], // {KEY-LEN[4],
                             //  VAL-LEN[4],
                             //  KEY    [KEY-LEN],
                             //  VAL    [VAL-LEN]}
   PADDING    [PADDING-LEN]}

Whole record length is represented by 2 bytes (maximum 2^(2*8) bytes),
but a key+value length might be representing a block of 8 bytes (maximum
2^(4*8) bytes) in size.

  +-----------------------------+
  | ### FCGI_RECORD ########### |
  | +----------------------------------------------+
  | | ### FCGI_PARAMS ############################ |
  | +----------------------------------------------+
  +-----------------------------+

  ^                             ^                  ^
  |---- 2^(2*8)=65536 bytes ----|                  |
  |--------------- 2^(4*8)=4294967296 -------------|

How can this be possible? I have read FastCGI specification[1] over and
over again but couldn't find a clue. Would you mind helping me to
clarify this point please?


Regards.

[1] http://www.fastcgi.com/devkit/doc/fcgi-spec.html