Re: A future for the SSH File Transfer Protocol?

[Darren J Moffat <Darren.Moffat%Sun.COM@localhost>]

Damien Miller wrote:

> Why are we discussing requirements on somethings that is already
> specified and widely deployed?

Because there is discussion on the ietf-ssh list about extending and
chaning some of the functionality.

In particular how the SFTP protocol deals with the identity of the user.

Currently a uint32_t is used to send a "UID" which is fine for UNIX
since that is what most systems use as the uid, however it is not fine
for NT.  In parallel with this it has been proposed that the long
output goes away in favour of the client having to build it themselves
from the information it gets from the server.

To that end there has been proposals to change this area of
the protocol and augment it with a method for the client to translate
the opaque uid into a name suitable for display.  The rational being
that the mapping maybe different on the client than on the server.

My concern was that this is starting to get into the area and issues of
NFS and NIS: NFS does transfer NIS does name id mappings.

I fully agree that SCP is not suitable and neither is FTP over SSH but
before we go changing a lot of things I think it would be a good idea
to document where the end goal is so that we know when we are done and
we know if we are changing things because people really need that
functionality.

On the whol SFTP is a decent protocol but there are a few little warts
that need to be tidied up I just wanted to make sure that in tidying
these up SFTP didn't turn into a file sharing protocol and starting
becoming a poor mans NFS or CIFS.

I have no problem with documenting the protocol as it stands just now
just conern about where it seemed to be heading.