Re: [Qemu-devel] RFC configurable block formats

[Anthony Liguori <anthony@codemonkey.ws>]

Markus Armbruster wrote:
> We have code for a quite a few block formats.  A quick grep shows bochs,
> cloop, cow, dmg, ftp, ftps, host_cdrom, host_device, host_floppy, http,
> https, nbd, parallels, qcow, qcow2, raw, tftp, vdi, vmdk, vpc, vvfat.
> Only formats ftp, ftps, http, https, tftp are optional (see configure
> --enable-curl).
>
> While I trust that all of these formats are useful at least for some
> people in some circumstances, some of them are of a kind that friends
> don't let friends use in production.
>
> In short, I'd like to be able to configure block formats.  Simple
> enough, eh?  Except there's a catch: I'd like to be able to include more
> formats in tools like qemu-img than in qemu proper.  That lets me
> restrict qemu proper, where a faulty block driver has the greatest
> potential for mischief, to the formats I trust and need, and still keep
> useful capability for other formats in qemu-img.
>
> Thus, I'd like to be able to configure a block driver off for
> everything, or on for utility programs and off for qemu proper, or on
> for everything.
>
> A naive implementation of this idea simply links only those block
> drivers into a program that have been configured for it.  Unfortunately,
> this leads to an unwanted difference in behavior between the different
> programs when the format is probed.
>
> Probing gives every block driver a chance to "score" the image, and
> picks the one with the highest score (I'm simplifying, but it'll do to
> illustrate the problem).  If two programs have different sets of
> drivers, probing may yield different results.  I don't like that.
>
> Say I configure crusty old qcow (note lack of 2) for utility programs
> only.  Then I don't want qemu to silently treat qcow images as raw, I
> want it to tell me it can't do qcow.  To be precise:
>
> If a format is configured off, no program shall recognize it.
>
> Else all programs shall recognize it, but
>
>     if it is configured on for utility programs, off for qemu proper,
>     then recognizing it in qemu proper shall be an error.
>
> If you agree this is useful, I'd be willing to code it.
>   

I'd rather see something like a driver white list/black list for qemu 
proper.  The list would be used to exclude block formats and could be 
extended to support read-only formats vs. read-write formats.  For 
instance, --enable-block-formats='qcow2 raw'.  It avoids polluting the 
block interface with knowledge of the distinction between a "utility" 
program and qemu proper.

Regards,

Anthony Liguori