Roxen.git/ server/ protocols/ http.pike

2008-02-19

2008-02-19 17:09:38 by Martin Stjernholm <mast@lysator.liu.se>

• 34acd02d7149b7bb55e1fd8da8a17d70772e467a (67 lines) (+30/-37) ^{[ Show | Annotate ]}
  Branch: 5.2

Less intrusive browser bug kludge in the parsing of
application/x-www-form-urlencoded data in POST requests - the original
data is no longer touched. This makes the applauncher kludge
unnecessary.

Also added id->misc->content_type_type and
id->misc->content_type_params.

Rev: server/protocols/http.pike:1.550

2:   // Modified by Francesco Chemolli to add throttling capabilities.   // Copyright © 1996 - 2004, Roxen IS.    - constant cvs_version = "$Id: http.pike,v 1.549 2008/02/18 18:51:24 mast Exp $"; + constant cvs_version = "$Id: http.pike,v 1.550 2008/02/19 17:09:38 mast Exp $";   // #define REQUEST_DEBUG   #define MAGIC_ERROR   
866:    misc[linename] = lower_case(contents);    break;    case "content-type": +  // The following is the useless result of an old "fix" (it +  // was lowercased before that); it'd been better to have used +  // the field directly in request_headers instead if the +  // lowercasing messed things up.    misc[linename] = contents; -  + #define WS "%*[ \t]" +  sscanf (contents, WS"%[^ \t/]"WS"/"WS"%[^ \t;]"WS";"WS"%s", +  string ct_type, string ct_subtype, +  misc->content_type_params); +  misc->content_type_type = lower_case (ct_type + "/" + ct_subtype);    break;    case "destination":    if (mixed err = catch {
907:    case "POST":    // See http://www.w3.org/TR/html401/interact/forms.html#h-17.13.4    // for spec of the format of form submissions. -  switch(lower_case((((misc["content-type"]||"")+";")/";")[0]-" ")) +  switch (misc->content_type_type)    { -  default: { -  // Normal form data. -  // -  // FIXME: This assumes only html forms are ever submitted -  // using http POST, thereby possibly precluding other -  // applications of the http protocol. Using -  // "application/x-www-form-urlencoded" above instead of going -  // here by default would fix that, and I don't think even the -  // lowliest client gets the content type wrong. But still, I -  // refrain from introducing a possible interoperability -  // problem without any tangible benefit. /mast +  case "application/x-www-form-urlencoded": { +  // Form data.       string v;    -  // Oh, the joy of supporting all clients is endless. -  if(client && -  sizeof(client) >= 3 && -  client[0] == "Roxen" && -  client[1] == "Application" && -  client[2] == "Launcher") -  { -  // Don't trim whitespace for Application Launcher. It uses a -  // non-standard way of "posting" data on Mac OS. -  } -  else -  { +     // Ok.. This might seem somewhat odd, but IE seems to add a    // (spurious) \r\n to the end of the data, and some versions of    // opera seem to add (spurious) \r\n to the start of the data. -  if (has_prefix (data, "\r\n")) -  data = data[2..], l = misc->len = l - 2; -  if (has_suffix (data, "\r\n")) -  data = data[..sizeof (data) - 3], l = misc->len = l - 2; -  } +  // +  // 4.5 and older trimmed id->data directly. We no longer do +  // that to allow the application to see the unmodified data if +  // the client says it's application/x-www-form-urlencoded but +  // in reality sends something else (some versions of the roxen +  // applauncher did that). (This is safe for correct +  // applauncher/x-www-form-urlencoded data since it doesn't +  // contain any whitespace.) +  v = String.trim_all_whites (data);    -  // Store it temporary in the misc mapping so we can decode it -  // together with the query string in +  // Store it temporarily in the misc mapping so we can decode +  // it together with the query string in    // things_to_do_when_not_sending_from_cache.    //    // Note: This is just kludgy temporary storage. It's nothing    // to rely on and it should not be documented. -  misc->post_form = data; +  misc->post_form = v;    break;    }