Re: [Qemu-devel] Re: RFC: blockdev_add & friends, brief rationale, QMP docs

[Kevin Wolf <kwolf@redhat.com>]

Am 31.05.2010 15:48, schrieb Luiz Capitulino:
> On Mon, 31 May 2010 13:05:37 +0200
> Markus Armbruster <armbru@redhat.com> wrote:
> 
>> Avi Kivity <avi@redhat.com> writes:
>>
>>> On 05/28/2010 10:24 PM, Luiz Capitulino wrote:
>>>>
>>>>> If a password is needed, we should throw an error and let the QMP client
>>>>> set the password and try again.
>>>>>      
>>>>   It's what we do today, a password should be set with block_passwd before
>>>> issuing the change command. Otherwise an error is throw.
>>>>    
>>>
>>> Is the password some kind of global or per-monitor property?  In that
>>> case it doesn't work with parallel execution of commands; better to
>>> have a password field (or assign IDs to passwords and require a
>>> passwordid=... argument).
>>
>> It sets the password in the host BlockDriverState.  Which must already
>> exist, i.e. you do it after blockdev_add.
>>
>> What happens if the guest device accesses the host drive before the key
>> is set?
> 
>  It's supposed to fail, right Kevin?

I don't think it's a situation that is even supposed to happen. Which is
exactly why I proposed an additional field to avoid it in the first place.

As far as I know in the old monitor the user would be prompted for the
password and as long as he doesn't enter it the VM is stopped, so we
don't get in the same situation there. But if you enter a wrong password
(I'm not sure if it's the same), it just seems to read garbage.

Kevin