Greetings,
* Fabien COELHO (coelho@cri.ensmp.fr) wrote:
> >>The implementations should not have to be in any particular language: Shell,
> >>Perl, Python, C should be possible.
> >
> >I disagree that it makes any sense to pass actual encryption out to a
> >shell script.
>
> Yes, sure. I'm talking about key management. For actual crypto functions,
> ISTM that they should be "replaceable", i.e. if some institution does not
> believe in AES then they could switch to something else.
The proposed implementation makes it pretty straight-forward to add in
other implementations and other crypto algorithms if people wish to do
so.
> >>After giving it more thought during the day, I think that only one
> >>command and a basic protocol is needed. Maybe something as simple as
> >>
> >> /path/to/command --options arguments…
> >
> >This is what's proposed- a command is run to acquire the KEK (as a hex
> >encoded set of bytes, making it possible to use a shell script, which is
> >what the patch does too).
>
> Yep, but that is not what I'm trying to advocate. The "KEK" (if any), would
> stay in the process, not be given back to the database or command using it.
> Then the database/tool would interact with the command to get the actual
> per-file keys when needed.
"When needed" is every single write that we do of any file in the entire
backend. Reaching out to something external every time we go to use the
key really isn't sensible- except, perhaps, in the case where we are
doing full crypto off-loading and we don't actually have to deal with
the KEK or the DEK at all, but that's all on the cluster encryption
side, really, and isn't the focus of this patch and what it's bringing-
all of which is needed anyway.
> >[...] The API to fetch the KEK doesn't care at all about where it's stored
> >or how it's derived or anything like that.
>
> >There's a relatively small change which could be made to have PG request
> >all of the keys that it'll need on startup, if we want to go there as has
> >been suggested elsewhere, but even if we do that, PG needs to be able to
> >do that itself too, otherwise it's not a complete capability and there
> >seems little point in doing something that's just a pass-thru to something
> >else and isn't able to really be used.
>
> I do not understand. Postgres should provide a working implementation,
> whether the functions are directly inside pg or though an external process,
> they need to be there anyway. I'm trying to fix a clean, possibly external
> API so that it is possible to have something different from the choices made
> in the patch.
Right, we need a working implementation that's part of core, that's what
this effort is trying to bring.
Thanks,
Stephen
