Poking around bdrv_is_inserted()

[Markus Armbruster]

bdrv_is_inserted() returns false when:

    /**
     * Return TRUE if the media is present
     */
    bool bdrv_is_inserted(BlockDriverState *bs)
    {
        BlockDriver *drv = bs->drv;
        BdrvChild *child;

        if (!drv) {
            return false;

1. @bs has no driver (this is how we represent "no medium").

        }
        if (drv->bdrv_is_inserted) {
            return drv->bdrv_is_inserted(bs);

2. Its driver's ->bdrv_is_inserted() returns false.  This is how
passthrough block backends signal "host device has no medium".  Right
now, the only user is "host_cdrom".

        }
        QLIST_FOREACH(child, &bs->children, next) {
            if (!bdrv_is_inserted(child->bs)) {
                return false;

3. Any of its children has no medium.  Common use looking through
filters, which have a single child.

            }
        }
        return true;
    }

Makes sense.

Now look at the uses of QERR_DEVICE_HAS_NO_MEDIUM.

* external_snapshot_prepare() in blockdev.c:

    if (!bdrv_is_inserted(state->old_bs)) {
        error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
        goto out;
    }

  where @device is the device name, i.e. BlockdevSnapshot member @node
  or BlockdevSnapshotSync member @device.  Uh-oh: the latter can be
  null.  If we can reach the error_setg() then, we crash on some
  systems.

* bdrv_snapshot_delete() and bdrv_snapshot_load_tmp() in
  block/snaphot.c:

    if (!drv) {
        error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, bdrv_get_device_name(bs));
        return -ENOMEDIUM;
    }

  where @drv is bs->drv.

  Why do we check only for 1. here instead of calling
  bdrv_is_inserted()?