Libguestfs - May 2022 - nbdkit: how to require use of filter

Hi,

Is there a way for a nbdkit plugin to *require* use of a nbdkit filter
(specifically the nbdkit-blocksize-filter) programmatically?

Thanks!
-Nikolaus
-- 
GPG Fingerprint: ED31 791B 2C5C 1613 AF38 8B8A D113 FCAC 3C4E 599F

             ?Time flies like an arrow, fruit flies like a Banana.?

On Mon, May 02, 2022 at 10:25:44AM +0100, Nikolaus Rath
wrote:
> Hi,
> 
> Is there a way for a nbdkit plugin to *require* use of a nbdkit filter
> (specifically the nbdkit-blocksize-filter) programmatically?
At present, no.  What are you envisioning - enforcing that the plugin
doesn't want to deal with sub-sector offsets and lengths?  Is there
any other filter that would be useful to require, or can we focus more
on the question about insisting on aligned requests?  Also, is your
plugin mostly worried about enforcement of minimum block size (for
example, no requests smaller than 512-byte alignment, which the
blocksize filter gives with read-modify-write), or is it also worried
about other features that the blocksize filter can add (such as
splitting larger requests)?

The best you can do in the current code is provide the .block_size
callback to at least inform modern NBD clients about your preferences
(libnbd defaults to refusing to send requests over the wire that don't
obey the server's advertised constraints).

I did ask the question when we first added .block_size whether we
should also provide a way for plugins to request the internal nbdkit
server code to enforce limits (use of the nbdkit-blocksize-filter with
read-modify-write is one way to obey underlying limits; but another
way is to just up-front fail unaligned requests rather than passing
them into the plugin).  We could either do this by enhancing the
existing .block_size() function to add in a special return value to
insist on strict enforcement (right now, we only document returning -1
on error and 0 when successfully populating limits; we could add a
documented return of 1), or by adding yet another callback (for
example, if .strict_block_size is absent or returns 0, .block_size
values are advisory to the client, but unaligned values reach the
plugin; if .strict_block_size returns 1, .block_size values are
enforced by nbdkit before reaching the plugin).

Thoughts about that approach?

-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org

On Mon, May 02, 2022 at 10:25:44AM +0100, Nikolaus Rath
wrote:
> Hi,
> 
> Is there a way for a nbdkit plugin to *require* use of a nbdkit filter
> (specifically the nbdkit-blocksize-filter) programmatically?
No it's not possible in general, you just have to document it.

However as it's *blocksize* filter that you're talking about, I assume
you mean that your filter has blocksize constraints, eg. that it
cannot handle sub-sector requests.  In this case it is possible to
tell clients that they must not issue requests smaller than a certain
size (and also hint about optimal and maximum sizes).  You can do this
by writing a .block_size callback.  For example:

https://gitlab.com/nbdkit/nbdkit/-/blob/1f5dc0679e570ca2fa149ceaa52922d70462b62c/plugins/vddk/vddk.c#L851

Not all clients obey these (although these clients are buggy!)  In
particular nbdcopy does not obey the minimum constraint, which
actually causes a bug for us in virt-v2v.  There's a long thread about
it on qemu-devel of which the most important message is:

https://www.mail-archive.com/qemu-devel at nongnu.org/msg865252.html

I would advertise a block size constraint in your plugin, and if you
find a client that doesn't request & then obey this constraint, get
the author of the client to fix it :-)

Rich.

-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
Read my programming and virtualization blog: http://rwmj.wordpress.com
virt-top is 'top' for virtual machines.  Tiny program with many
powerful monitoring features, net stats, disk stats, logging, etc.
http://people.redhat.com/~rjones/virt-top