Re: [Qemu-devel] [PATCH v4 29/38] blockdev: Add blockdev-insert-medium

[Max Reitz]

On 10.09.2015 03:12, Wen Congyang wrote:
> On 09/09/2015 08:59 PM, Max Reitz wrote:
>> On 09.09.2015 12:01, Wen Congyang wrote:
>>> On 09/09/2015 05:20 AM, Max Reitz wrote:
>>>> On 08.09.2015 11:13, Wen Congyang wrote:
>>>>> On 07/21/2015 01:45 AM, Max Reitz wrote:
>>>>>> And a helper function for that, which directly takes a pointer to the
>>>>>> BDS to be inserted instead of its node-name (which will be used for
>>>>>> implementing 'change' using blockdev-insert-medium).
>>>>>>
>>>>>> Signed-off-by: Max Reitz <address@hidden>
>>>>>> ---
>>>>>>  blockdev.c           | 48 
>>>>>> ++++++++++++++++++++++++++++++++++++++++++++++++
>>>>>>  qapi/block-core.json | 17 +++++++++++++++++
>>>>>>  qmp-commands.hx      | 37 +++++++++++++++++++++++++++++++++++++
>>>>>>  3 files changed, 102 insertions(+)
>>>>>>
>>>>>> diff --git a/blockdev.c b/blockdev.c
>>>>>> index 481760a..a80d0e2 100644
>>>>>> --- a/blockdev.c
>>>>>> +++ b/blockdev.c
>>>>>> @@ -2164,6 +2164,54 @@ void qmp_blockdev_remove_medium(const char 
>>>>>> *device, Error **errp)
>>>>>>      }
>>>>>>  }
>>>>>>  
>>>>>> +static void qmp_blockdev_insert_anon_medium(const char *device,
>>>>>> +                                            BlockDriverState *bs, Error 
>>>>>> **errp)
>>>>>> +{
>>>>>> +    BlockBackend *blk;
>>>>>> +    bool has_device;
>>>>>> +
>>>>>> +    blk = blk_by_name(device);
>>>>>> +    if (!blk) {
>>>>>> +        error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
>>>>>> +                  "Device '%s' not found", device);
>>>>>> +        return;
>>>>>> +    }
>>>>>> +
>>>>>> +    /* For BBs without a device, we can exchange the BDS tree at will */
>>>>>> +    has_device = blk_get_attached_dev(blk);
>>>>>> +
>>>>>> +    if (has_device && !blk_dev_has_removable_media(blk)) {
>>>>>> +        error_setg(errp, "Device '%s' is not removable", device);
>>>>>> +        return;
>>>>>> +    }
>>>>>> +
>>>>>> +    if (has_device && !blk_dev_is_tray_open(blk)) {
>>>>>> +        error_setg(errp, "Tray of device '%s' is not open", device);
>>>>>> +        return;
>>>>>> +    }
>>>>>> +
>>>>>> +    if (blk_bs(blk)) {
>>>>>> +        error_setg(errp, "There already is a medium in device '%s'", 
>>>>>> device);
>>>>>> +        return;
>>>>>> +    }
>>>>>> +
>>>>>> +    blk_insert_bs(blk, bs);
>>>>>> +}
>>>>>> +
>>>>>> +void qmp_blockdev_insert_medium(const char *device, const char 
>>>>>> *node_name,
>>>>>> +                                Error **errp)
>>>>>> +{
>>>>>> +    BlockDriverState *bs;
>>>>>> +
>>>>>> +    bs = bdrv_find_node(node_name);
>>>>>> +    if (!bs) {
>>>>>> +        error_setg(errp, "Node '%s' not found", node_name);
>>>>>> +        return;
>>>>>> +    }
>>>>>
>>>>> Hmm, it is OK if the bs is not top BDS?
>>>>
>>>> I think so, yes. Generally, there's probably no reason to do that, but I
>>>> don't know why we should not allow that case. For instance, you might
>>>> want to make a backing file available read-only somewhere.
>>>>
>>>> It should be impossible to make it available writable, and it should not
>>>> be allowed to start a block-commit operation while the backing file can
>>>> be accessed by the guest, but this should be achieved using op blockers.
>>>>
>>>> What we need for this to work are fine-grained op blockers, I think. But
>>>> working around that for now by only allowing to insert top BDS won't
>>>> work, since you can still start block jobs which target top BDS, too
>>>> (e.g. blockdev-backup can write to a BDS/BB that is visible to the guest).
>>>>
>>>> All in all, I think it's fine to insert non-top BDS, but we should
>>>> definitely worry about which exact BDS one can insert once we have
>>>> fine-grained op blockers.
>>>
>>> A BDS can be written by its parent, its block backend, a block job..
>>> So I think we should have some way to avoid more than two sources writing
>>> to it, otherwise the data may be corrupted.
>>
>> Yes, and that would be op blockers.
>>
>> As I said, using blockdev-backup you can write to a BB that can be
>> written to by the guest as well. I think this is a bug, but it is a bug
>> that needs to be fixed by having better op blockers in place, which Jeff
>> Cody is working on.
>>
>> Regarding this series, I don't consider this to be too big of an issue.
>> Yes, if you are working with floppy disks, you can have the case of a
>> block job and the guest writing to the BDS at the same time. But I can't
>> really imagine who would use floppy disks and block jobs at the same
>> time (people who still use floppy disks for their VMs don't strike me as
>> the kind of people who use the management features of qemu, especially
>> not for those floppy disks).
>>
>> Other than that, this function (blockdev-insert-medium) can only be used
>> for optical ROM devices (I don't think we have CD/DVD-RW support, do
>> we?), so it's much less of an issue there.
>>
>> So all in all I don't consider this too big of an issue here. If others
>> think different, then I would delay this part of the series (which
>> overhauls the "change" command) until we have fine-grained op blockers.
> 
> In most cases, the user uses this command to change CD/DVD media, so it is OK.
> But IIRC scsi disk can also be changed. So we can mark this command as 
> experimental
> (the command name can be x-blockdev-insert-medium).

I'd rather delay this part than mark it experimental. But then again,
seeing that we have cases like this already (i.e. blockdev-backup) and
nobody seems to be complaining, I still think it should be fine.

Max

signature.asc
Description: OpenPGP digital signature