ldb_dn_copy(): should we take a talloc_reference?

[Jeremy Allison]

On Thu, Feb 08, 2018 at 06:37:29AM +0100, Volker Lendecke via samba-technical wrote:
> On Thu, Feb 08, 2018 at 04:26:22PM +1300, Andrew Bartlett via samba-technical wrote:
> > * On one hand, I would say that the answer is to make the struct ldb_dn
> > take a talloc_reference() to the ldb and assign the element in
> > ldb_dn_copy().  (We would then also need to take the reference in
> > ldb_dn_new()).  On the other, I can hear the howls of protest already. 
> 
> Just very recently I've been in source4/nbt_server.c and libcli/nbt/.
> This is old code doing significant amounts of talloc_reference. This
> gave me significant amounts of grief. I do not claim I even remotely
> understand what you're talking about, but "talloc_reference" is really
> a trigger word for me, re-confirmed by very recent experience.
> 
> Just. Don't. Do. It.
> 
> Please find other ways to deal with your problem without
> talloc_reference.

Even sick with flu, I have to add my +1 to this.

Please do *NOT* add any more talloc_reference() calls.

It's probably too late to end up removing all talloc_reference()
from Samba, but given infinite resources that's where I'd want
us to end up.

It's at least as dangerous as strcpy(), which I managed to
ban a long time ago, but harder to find bugs with as the
problems are not as immediately apparent.

If you have to reference count, add an explicit reference
counter.