[krbdev.mit.edu #5442] memory leak in gss_accept_sec_context()

[Public Submitter via RT]

[guest - Sun Oct 28 05:11:06 2007]:

> [guest - Tue Jul  3 17:17:57 2007]:
> 
> > Actually, reviewing my patch a little bit, it seems that I released
> > the cred which then goes on to be used in the error case.  A not
> > terribly elegant fix to this would be to duplicate the release code
> > above the fail: label and leave the existing code at the tail of the
> > function...
> 
> I've encountered the same leak, i'm using the following patch right
>    now:
> Would be really nice is this would be fixed in an future krb5 release
>    :)
> 

>         if (decode_krb5_ap_req(&ap_req, &request)) {
>  	   krb5_free_context(context);
> +	   if (!verifier_cred_handle && cred_handle)
> +	       krb5_gss_release_cred(minor_status, &cred_handle);
>  	   return (major_status);
>         }

Noticed today that this patch destroys minor_status: In any error case,
gss_accept_sec_context() returns now 'No Error' :(. So the additional
gss_release_cred() should use some temporary OM_uint32 for there
minor_status.