[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [patch] merge mechglue-branch into HEAD

On Sun, 18 Jun 2006 00:36:09 -0400
Michael B Allen <mba2000@ioplex.com> wrote:
> > I used Martin Rex's gsstest to test the mechglue.
> I just tried this and there were many instances of the following 3 errors:
>   ERROR: gss_acquire_cred returned BOGUS value lifetime!
>   ERROR: transfer_via_child(): Child returned accumulated error code rc=1
>   ERROR: gss_wrap_size_limit(integ): max_in==0  for  req_output_size=2028 ??
> If I recall correctly the gssapi code never tried to handle the lifetime
> parameter at all. The second error looks like it's just an artifact of
> another error but that's just a guess.
> Also, running it under valgrind didn't look good. Unfortunately I cannot
> get filenames and line numbers even with ./configure --enable-debug. There
> were so many different leaks that I cannot help but suspect that the
> leaking is an artifact of failures within gsstest. Anyway, I doubt
> they could be attributed to the port. There just weren't that many
> code changes.

The latest HEAD yields pretty much exactly the same results except for
the lifetime errors. So the lifetime thing is a bug but the other stuff
isn't new. I suppose there must be a lot of memory attributed to the
globals for the krb5_context, ccache and so on. But it would be nice to
check those by either hacking gsstest to call a GSSAPI_KRB5_DEINIT routine
and/or somehow get valgrind to cough up symbol names and line numbers.


Michael B Allen
PHP Extension for SSO w/ Windows Group Authorization