Re: Memory leak with ldap_first_attribute()/ldap_next_attribute()?

[kunkee@neosoft.com (Randy Kunkee)]

> You need to do a ldap_memfree on each attribute returned by
> ldap_first_attribute or ldap_next_attribute
> 
> -Steve

No, no, this is wrong.  You need to free the result, and do nothing with the
ber values until the end.

Ldap_msgfree(LDAPMessage *lm) frees everything associated with a result,
including all the attribute values -- they are not copied when you walk
through the attributes and you should not try to free anything within a
result.

However, I stand corrected, and agree that the code

    if (ber)
       ber_free(ber,0);
    }

*is* needed (I was recalling an old version of ldap_next_attribute, which
used to free ber when it hit the end of the loop).

There is one unlikely case in ldap_first_attribute which appears to be
a leak:

        if ( ber_pvt_ber_remaining( ber ) == 0 ) {
                assert( len == 0 );
                return NULL;
        }
        assert( len != 0 );

This occurs before the *berout = ber, so it should probably have a
ber_free(ber, 0) inserted before the return.  Perhaps the assert()
makes it moot.

I'm probably overlapping replies with Kurt.

Randy


> 
> >>> Bertrand Croq <bcroq@virtual-net.fr> 04-Oct-00 11:06:48 AM >>>
> Hi,
>   after some hours of debugging, I have been able to write a simple code
> that produces memory leaks on my system (OpenLDAP 2.0.4 on Linux Debian
> 2.2 (x86)):
> 
> {
> [... ldap_search() ...]
> for (a=ldap_first_attribute(ld, e, &ber);
>       a!=NULL;
>       a=ldap_next_attribute(ld, e, ber)) {
>   /* nothing in the body */
> }
> if (ber)
>   ber_free(ber,0);
> }
> 
>   When I comment out this loop or when ldap_next_attribute() is not
> called (when there is no attribute), I don't get any memory leak... This
> looks like a problem already posted to this mailing list ("Possible
> memory leak in libldap.a/liblber.a" by Ashley Neal Hornbeck on 21 Jan
> 1999). Does anybody knows what/where the problem can be?
> 
> Thanks.
> -- 
> Bertrand Croq - VIRTUAL NET (http://www.virtual-net.fr)
> 80, avenue des Buttes de Coesmes - 35700 RENNES
> tel: +33 2 23 21 06 30 - fax: +33 2 99 38 16 85