Hello,
On Tue, 20 Mar 2001, Joseph Mack wrote:
> > By default TCP replies RST to the client when there is no
> > existing socket. LVS does not keep info for already expired connections
> > and this is the reason ICMP is replied. Of course, if we implement
> > TCP RST replies we can reply TCP RST instead of ICMP.
> >
> > The other question is how the client interpretes the ICMP
> > replies. Linux at least allows the application to listen for such
> > ICMP replies.
>
> > In the other case (which is default) they are reported
> ^^^^^^^^^^ ^^^^
> > after a TCP timeout (as soft errors).
>
> Julian,
>
>
> what is "the other case" (non Linux) and why is it default?
By default if the application does not listen for ICMP
errors they are reported after a TCP timeout and according to the
TCP state. But the application can register for these ICMP errors and by
this way to detect them immediately as they are received for the socket.
May be it is not recommended to use this information from untrusted
sources. The ICMP errors are reported immediately for some TCP (SYN)
states.
> what is "they"?
>
> Thanks Joe
Regards
--
Julian Anastasov <ja@xxxxxx>
|