issue with database_name parameter

[Shawn M. Emery]

Will Fiveash wrote On 05/11/06 14:54,:
> On Thu, May 11, 2006 at 04:38:52PM -0400, Ken Raeburn wrote:
> 
>>On May 11, 2006, at 15:46, Will Fiveash wrote:
>>
>>>I've noticed that the latest MIT krb code with LDAP support appears to
>>>ignore the database_name if it is located in a [realms] section  
>>>entry in
>>>the kdc.conf (kadmin.local issues an error stating the KDB does not
>>>exist).  If I place the database_name in a [dbmodules] section  
>>>entry in
>>>the krb5.conf then kadmin.local works.
>>
>>Yep, this was an unfortunate effect of the DAL integration.  To fix  
>>it, I'm working on changes so that the KDC will consult both kdc.conf  
>>and krb5.conf, combined, for its configuration information.
> 
> 
> Okay.  BTW, found another bug:
> 
> In krb5_rc_resolve_full() there is:
> 
>     if ((retval = krb5_rc_resolve_type(context, id,type))) {
>         FREE(type);
>         k5_mutex_destroy(&(*id)->lock);
>         FREE(*id);
>         return retval;
>     }
> 
> however in krb5_rc_resolve_type() there is this:
> 
>     for (t = typehead;t && strcmp(t->ops->type,type);t = t->next)
>     ;
>     if (!t) {
>         k5_mutex_unlock(&rc_typelist_lock);
>         return KRB5_RC_TYPE_NOTFOUND;
>     }
> 
> If return KRB5_RC_TYPE_NOTFOUND; is executed the krb5kdc will core dump on:
> 
>         k5_mutex_destroy(&(*id)->lock);
> 
> Note this occurred when I did:
> 
> krb5kdc -n -R FILE:/usr/local/krb_ldap/var/krb5kdc/rcache

This is a known issue that I filed back on 1/9/06:
#3332 rcache mutex access

Shawn.
--