On Mon, Jan 17, 2011 at 10:21:26AM +0100, Christoph Hellwig wrote: > The actual removable bit looks fine, but I don't think the connection of > the change callback looks sane. What's the rationale for it?
Since we're using bdrv_set_removable(), the user may try to eject the block device from the QEMU monitor. At that point we have a closed BlockDriverState and all operations will (at best) error since there is no medium. The callback removes the USB MSD so that eject is equivalent to removing the device. It's a hack and we could remove it, but then we're left with a weird guest-visible state that you can't get into with a physical USB thumbdrive. I was considering not using bdrv_set_removable() and instead adding a hint to the BlockDriverState which gets checked when constructing the SCSI INQUIRY response. If we take that approach, then QEMU doesn't consider the block device removable in the eject/change medium sense. Stefan
