Libguestfs - May 2020 - Re: Provide NBD via Browser over Websockets

On 5/29/20 8:50 AM, Daniel P. Berrangé wrote:
>>> (2) You need to persuade qemu's NBD client to read from a
WebSocket.
>>> I didn't really know anything about WebSockets until today but
it
>>> seems as if they are a full-duplex protocol layered on top of HTTP
[a].
>>> Is there a WebSocket proxy that turns WS into plain TCP (a bit like
>>> stunnel)?  Google suggests [b].
>>>
>>> [a] https://en.wikipedia.org/wiki/WebSocket#Protocol_handshake
>>> [b] https://github.com/novnc/websockify
>>
>> qemu already knows how to connect as a client to websockets; Dan
Berrange
>> knows more about that setup.  I suspect it would not be too difficult
to
>> teach the qemu NBD client code to use a WebSocket instead of a Unix or
TCP
>> socket as its data source.
> 
> Actually the inverse. The QIOChannelWebsocket impl is only the server
> side of the problem, as used by QEMU's VNC server. We've never
implemented
> the client side. There is nothing especially stopping us doing that - just
> needs someone motivated with time to work on it.
In the meantime, you may still be able to set up something like:

local machine:
iso -> NBD server -> Unix socket -> websockify -> WebSocket

remote machine:
WebSocket -> websockify -> Unix socket -> qemu NBD client

Adding websocket client support into qemu would reduce the length of the 
chain slightly (for less data copying) by getting rid of a websockify 
proxy middleman, but would not necessarily improve performance (it's 
hard to say where the latency bottlenecks will be in the chain).

-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3226
Virtualization:  qemu.org | libvirt.org

On Fri, May 29, 2020 at 08:58:06AM -0500, Eric Blake
wrote:
> On 5/29/20 8:50 AM, Daniel P. Berrangé wrote:
> 
> >>>(2) You need to persuade qemu's NBD client to read from a
WebSocket.
> >>>I didn't really know anything about WebSockets until today
but it
> >>>seems as if they are a full-duplex protocol layered on top of
HTTP [a].
> >>>Is there a WebSocket proxy that turns WS into plain TCP (a bit
like
> >>>stunnel)?  Google suggests [b].
> >>>
> >>>[a] https://en.wikipedia.org/wiki/WebSocket#Protocol_handshake
> >>>[b] https://github.com/novnc/websockify
> >>
> >>qemu already knows how to connect as a client to websockets; Dan
Berrange
> >>knows more about that setup.  I suspect it would not be too
difficult to
> >>teach the qemu NBD client code to use a WebSocket instead of a Unix
or TCP
> >>socket as its data source.
> >
> >Actually the inverse. The QIOChannelWebsocket impl is only the server
> >side of the problem, as used by QEMU's VNC server. We've never
implemented
> >the client side. There is nothing especially stopping us doing that -
just
> >needs someone motivated with time to work on it.
> 
> In the meantime, you may still be able to set up something like:
> 
> local machine:
> iso -> NBD server -> Unix socket -> websockify -> WebSocket
I guess the idea is to have a zero-install solution for the browser.
As I said in the email earlier this is very common for IPMI-type
remote access to blade servers and in my experience is implemented
using a Java applet and a proprietary protocol terminated at the BMC
(which then emulates a virtual CDROM to the server).  There are some
HP blade servers on Red Hat's internal Beaker instance where you can
play with this.  For qemu we wouldn't need to invent a new protocol
when NBD is available and already implemented (albeit not yet on top
of WebSockets).

The NBD server must run inside the browser and therefore be either
written from scratch in Javascript, or an existing server
cross-compiled to WASM (if that is possible - I don't really know).
> remote machine:
> WebSocket -> websockify -> Unix socket -> qemu NBD client
> 
> Adding websocket client support into qemu would reduce the length of
> the chain slightly (for less data copying) by getting rid of a
> websockify proxy middleman, but would not necessarily improve
> performance (it's hard to say where the latency bottlenecks will be
> in the chain).
Rich.

-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
Read my programming and virtualization blog: http://rwmj.wordpress.com
libguestfs lets you edit virtual machines.  Supports shell scripting,
bindings from many languages.  http://libguestfs.org

On Fri, 29 May 2020, Richard W.M. Jones wrote:
> On Fri, May 29, 2020 at 08:58:06AM -0500, Eric Blake wrote:
> > On 5/29/20 8:50 AM, Daniel P. Berrangé wrote:
> > 
> > >>>(2) You need to persuade qemu's NBD client to read
from a WebSocket.
> > >>>I didn't really know anything about WebSockets until
today but it
> > >>>seems as if they are a full-duplex protocol layered on top
of HTTP [a].
> > >>>Is there a WebSocket proxy that turns WS into plain TCP (a
bit like
> > >>>stunnel)?  Google suggests [b].
> > >>>
> > >>>[a]
https://en.wikipedia.org/wiki/WebSocket#Protocol_handshake
> > >>>[b] https://github.com/novnc/websockify
> > >>
> > >>qemu already knows how to connect as a client to websockets;
Dan Berrange
> > >>knows more about that setup.  I suspect it would not be too
difficult to
> > >>teach the qemu NBD client code to use a WebSocket instead of a
Unix or TCP
> > >>socket as its data source.
> > >
> > >Actually the inverse. The QIOChannelWebsocket impl is only the
server
> > >side of the problem, as used by QEMU's VNC server. We've
never implemented
> > >the client side. There is nothing especially stopping us doing
that - just
> > >needs someone motivated with time to work on it.
> > 
> > In the meantime, you may still be able to set up something like:
> > 
> > local machine:
> > iso -> NBD server -> Unix socket -> websockify ->
WebSocket
> 
> I guess the idea is to have a zero-install solution for the browser.
> As I said in the email earlier this is very common for IPMI-type
> remote access to blade servers and in my experience is implemented
> using a Java applet and a proprietary protocol terminated at the BMC
> (which then emulates a virtual CDROM to the server).  There are some
> HP blade servers on Red Hat's internal Beaker instance where you can
> play with this.  For qemu we wouldn't need to invent a new protocol
> when NBD is available and already implemented (albeit not yet on top
> of WebSockets).
> 
> The NBD server must run inside the browser and therefore be either
> written from scratch in Javascript, or an existing server
> cross-compiled to WASM (if that is possible - I don't really know).
Interesting idea about WASM.  I'll see if I can build one of the simple 
nbd servers that are around.  Not sure how to link it to the JS file IO, 
however.

--
Eric Wheeler

> > remote machine:
> > WebSocket -> websockify -> Unix socket -> qemu NBD client
> > 
> > Adding websocket client support into qemu would reduce the length of
> > the chain slightly (for less data copying) by getting rid of a
> > websockify proxy middleman, but would not necessarily improve
> > performance (it's hard to say where the latency bottlenecks will
be
> > in the chain).
> 
> Rich.
> 
> -- 
> Richard Jones, Virtualization Group, Red Hat
http://people.redhat.com/~rjones
> Read my programming and virtualization blog: http://rwmj.wordpress.com
> libguestfs lets you edit virtual machines.  Supports shell scripting,
> bindings from many languages.  http://libguestfs.org
> 
>