[Libguestfs] [libnbd PATCH 4/4] states: Use RESYNC to handle more structured reply server bugs

[Richard W.M. Jones]

On Fri, Aug 12, 2022 at 08:07:25AM -0500, Eric Blake wrote:
> Plus a typo in the subject for 1/4.  With everything fixed, the series
> is now in as 185195d..0883029.

Thanks - as mentioned before I'll give this a good "going over" with
fuzzing next week.

> Another leniency issue I am exploring is whether structured reply mode
> can cope with a server that mistakenly sends a simply reply to
> NBD_CMD_READ,

I think in the spirit of NBD having at least a simple core protocol we
should definitely try to cope with home-brew servers and suchlike that
are not quite conforming.  I often run into more obscure NBD servers
[see addendum] and wonder how good they are.

> and whether the client can ignore replies for an unknown
> cookie.

This one sees like something we'd want to opt into though.  Couldn't
it indicate quite serious desynchronization and/or data loss?

Rich.

Addendum - lesser known NBD servers of note:
- VMware's
- Veritas Netbackup
- https://github.com/vi/nbdserve
- https://github.com/amccurry/nbd-server
- https://github.com/takahiro-hirofuchi/xnbd
- https://bitbucket.org/belczyk/bnbd/wiki/Home
- https://github.com/xapi-project/nbd
- https://docs.rs/nbd/latest/nbd/

-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
Read my programming and virtualization blog: http://rwmj.wordpress.com
nbdkit - Flexible, fast NBD server with plugins
https://gitlab.com/nbdkit/nbdkit