On Thu, 11 Dec 2003 19:55:41 -0500
"Theodore Ts'o" <tytso(_at_)mit(_dot_)edu> wrote:
On Thu, Dec 11, 2003 at 10:10:44PM +0100, Anthony G. Atkielski wrote:
The dumb authors, I think, are those who built Linux implementations
that doggedly attempt to negotiate ECN and are unprepared for cases
where it does not work, even though it's unreasonable to assume that the
entire world is equipped to handle ECN or that all other hosts will
cheerfully ignore the setting of bits that are supposed to be zero. In
this context, Linux is beginning to remind me of Netscape in the early
days.
What Linux implemented was specifically what was specified by RFC
3168, no more no less. This RFC was in fact designed to deal with
hosts that were not equipped for ECN. The issue is whether or not
intermediate hosts are justified in dropping packets just because some
bits that were reserved for future use are no longer zero. I would
argue no.
In summary, "Be liberal in what you accept, and conservative in what you send."
?
Firewalls could be considered to be performing QA for defined protocol fields.
I agree that reserved fields shouldn't be "QA"'ed for their default values. I
can't remember exactly where I saw the definition, I've understood reserved
fields to mean "could change in the future, don't rely on this default value".
In fact, while many or most hosts do not support ECN, very few errant
firewalls and/or load balancing boxes were dropping packets that
support ECN. Firmware updates have been available for over two years
to fix those firewalls are broken, including no doubt the ones used by
ISOC. It's just that the ISOC firewall admins simply haven't had the
wit to upgrade their firewall firmware. Pretty much all of the
commercial websites were fixed a long, long, LONG time ago.
- Ted
This problem doesn't seem to be that uncommon, I can't seem to access one of
the fathers of the Internet's home pages with ECN switched on either.
Regards,
Mark.