On Mon, 4 Nov 2013, Steve Holme wrote:

> > I can't see anything immediately obvious from the test data but I will
> > also investigate over the next 30 minutes or so.
>
> The only thing niggling me is whether the shell out to ntlm auth needs
> the CRLF that would have been on the end of the base64 string or
> even just a LF to make the swrite() work correctly?

I re-read the winbind code again, kind of butchered fake_ntlm together to
get it compiling under Windows (took out all the environment stuff,
hardcoded the path to test1310 and removed the logmsg() calls) so I could
see what it does and understand now that the type-1 and type-3 message
requests are pumped into ntlm_auth and whether it needs an EOL character or
not.

At the end of all that... it did... so have pushed another fix.

It is also worth noting that the other pointer to needing this was the other
call to ntlm_wb_response() in curl_ntlm_wb.c:346 so I have performed the
same at line 359 - prior to the Curl_base64_decode() changes the type-2
message would have contained a EOL character (possibly in the form CRLF ??)
as that would have been present in the header received from the HTTP server.

I will keep another eye (I must be running out of eyes by now!) out on the
auto builds but this could explain why the tests just hung, as fake_ntlm
would have never responded with a reply which mean that the winbind code
would possibly just sit around waiting :(

Kind Regards

Steve
-------------------------------------------------------------------
List admin: http://cool.haxx.se/list/listinfo/curl-library
Etiquette: http://curl.haxx.se/mail/etiquette.html
Received on 2013-11-04