Re: [Qemu-devel] Re: [RFC][PATCH v5 08/21] virtagent: add agent_viewfile qmp/hmp command

[Michael Roth]

On 12/06/2010 05:20 PM, Michael Roth wrote:
> On 12/06/2010 04:08 PM, Adam Litke wrote:
> > On Fri, 2010-12-03 at 12:03 -0600, Michael Roth wrote:
> > > Utilize the getfile RPC to provide a means to view text files in the
> > > guest. Getfile can handle binary files as well but we don't advertise
> > > that here due to the special handling requiring to store it and provide
> > > it back to the user (base64 encoding it for instance). Hence the
> > > otherwise confusing "viewfile" as opposed to "getfile".
> >
> > What happens to the monitor if you use this to view a binary file?
>
> At the very least we probably get a lot of truncated files from the
> binary->string conversion via monitor_printf(). Im not sure how the
> qobject/json layer would deal with things.
>
> > Retrieving binary files progmatically using the QMP interface is a valid
> > use case right?
>
> For getfile (the RPC), but not for viewfile (HMP/QMP). It's doable, but
> we'd *have to* pass this data to the user as base64-encoded data at the
> QMP level. At the HMP level I think we're good either way, since we
> could just base64 decode in the print function.
>
> So in the case of QMP we'd be pushing complexity to the user in exchange
> for not having a seperate plain-text-only interface.
>
> Either way seems reasonable, but I'd been planning on adding a seperate
> `agent_copyfile <remote_path> <local_path>` command for dealing with
> binary data, and making viewfile quick and easy for plain text (both for
> HMP and QMP).

Although, agent_copyfile doesn't seem like the right approach looking at 
things like future libvirt integration. So we will most likely end up 
with a QMP command that passes base64-encoded binary data to the 
end-user for binary data, which we can provide a pretty-printing HMP 
function to decode. We'd need to take care to differentiate the HMP 
command from the QMP one however, else we'd have users tempted to do 
something like:

echo "agent_getfile /remotepath/rand.bin" | socat stdin 
unix-connect:monitor.sock > /localpath/rand.bin

to avoid having to decode the data. Would documenting the HMP 
counterpart as being reliable only for plain-text be sufficient? Or 
Should be have QMP:agent_getfile() and HMP:agent_viewfile()?