Jan Wieck <JanWieck@Yahoo.com> writes:
> Tom Lane wrote:
>> I was about to say "I give up, let's just take out the comparison".
> Which then get's us back to your concern about assuming that HPUX and
> Linux manpages can be taken as "every platform will" and hope all
> kernels will limit the sender for recv() to the connected address.
Well, I'd not have cared to trust just those couple of manpages, but
if it's in the Single Unix Spec then it's more likely that everyone
follows it. Also, I checked my yellowing first edition of Stevens,
and it says the same thing: "only datagrams from this address will be
received by the socket". So I'm thinking that this behavior has been
passed down from the original Berkeley sockets code.
> Since all involved processes are children of the postmaster, we can add
> some other, random number based security signature into the message
> itself. Noone outside will know what that is, it's really hard to guess
> and can be checked with a few int32 compares, not even a function call
> required.
We could do that if we're feeling paranoid, but I'm now leaning to the
view that it's not worth the trouble.
regards, tom lane