Le 4 août 09 à 17:00, Rémi Denis-Courmont a écrit :
On Tuesday 04 August 2009 17:49:25 ext Rémi Després wrote:
Le 4 août 09 à 16:30, Lars Eggert a écrit :
Hi,
On 2009-8-4, at 16:27, Rémi Després wrote:
You seem to have missed that the proposal includes a relaxation of
the constraint that zero-checksum UDP datagrams MAY be accepted by
hosts ion the future, just to avoid unnecessary black holes in case
of v4 to v6 translations.
isn't it a non-starter for a translation based *transition*
mechanism to depend on host changes?
I don't think so, because it's never too late do do some good without
doing any harm.
IMHO, translators SHOULD forward zero-checksum datagrams RATHER THAN
dropping them. No harm expected.
"No harm expected"? I find that generating scary-reading false
positive in my
system logs is harmful.
I don't get the point about "scary-reading false positive".
On the other hand, I still believe what I wrote yesterday is valid:
*If a host may receive receive a zero-checksum UDP datagram destined
to another host, this is acceptable because:
- Probability of this happening is very low (with link layer
checksums etc.).
- It is even more UNLIKELY that the source address and the two ports
of the datagram match those of an existing connection (connection
where, in addition, such datagrams must be accepted).
- If this nevertheless happens, the harm must be limited because of
the application nature. (For example, an application that accepts
queries in unchecked datagrams may receive an extra query without
significant consequence).*
IMHO also, hosts, at their next patch release, SHOULD silently accept
zero-checksum IPv6 datagrams RATHER THAN silently dropping them. No
harm expected.
That will cause silent software failures when running on a "too
old" system.
*IPv4 compatible applications should have no problem with UDP zero-
checksums*, or I am mistaken?
Silent failures are the most harmful kind of failures to me.
And to me too, but I don't see any silent "failure" here.
By doing this, connectivity during the transition period will
eventually be improved, possibly before translators are largely
deployed.
No. By doing this, people will experience odd failure modes and
weird error
reports during the long transition period for IPv6 stacks to accept
checksum-
less packets.
If you could prove your assertion, I would reconsider mines
(sentences between *s .
Harm very much expected.
An example would be helpful to understand what you fear.
Regards,
RD
--------------------------------------------------------------------
IETF IPv6 working group mailing list
[email protected]
Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------
