Re: [Qemu-devel] [PATCH v2 04/37] hw/usb-storage: Check whether BB is inserted

[Max Reitz <mreitz@redhat.com>]

On 2015-03-04 at 09:41, Max Reitz wrote:
> On 2015-03-04 at 09:39, Kevin Wolf wrote:
>> Am 04.03.2015 um 15:24 hat Max Reitz geschrieben:
>>> On 2015-03-04 at 09:20, Kevin Wolf wrote:
>>>> Am 04.03.2015 um 15:07 hat Max Reitz geschrieben:
>>>>> On 2015-03-04 at 09:02, Kevin Wolf wrote:
>>>>>> Am 09.02.2015 um 18:11 hat Max Reitz geschrieben:
>>>>>>> Only call bdrv_key_required() on the BlockDriverState if the
>>>>>>> BlockBackend has an inserted medium.
>>>>>>>
>>>>>>> Signed-off-by: Max Reitz <mreitz@redhat.com>
>>>>>>> Reviewed-by: Eric Blake <eblake@redhat.com>
>>>>>>> ---
>>>>>>>   hw/usb/dev-storage.c | 2 +-
>>>>>>>   1 file changed, 1 insertion(+), 1 deletion(-)
>>>>>>>
>>>>>>> diff --git a/hw/usb/dev-storage.c b/hw/usb/dev-storage.c
>>>>>>> index 4539733..3123baf 100644
>>>>>>> --- a/hw/usb/dev-storage.c
>>>>>>> +++ b/hw/usb/dev-storage.c
>>>>>>> @@ -638,7 +638,7 @@ static void 
>>>>>>> usb_msd_realize_storage(USBDevice *dev, Error **errp)
>>>>>>>       usb_msd_handle_reset(dev);
>>>>>>>       s->scsi_dev = scsi_dev;
>>>>>>> -    if (bdrv_key_required(blk_bs(blk))) {
>>>>>>> +    if (blk_is_inserted(blk) && bdrv_key_required(blk_bs(blk))) {
>>>>>>>           if (cur_mon) {
>>>>>>>               monitor_read_bdrv_key_start(cur_mon, blk_bs(blk),
>>>>>>> usb_msd_password_cb, s);
>>>>>> Why would bdrv_key_required() ever return true when no medium is
>>>>>> inserted? Sounds like a bug to me, like not resetting state 
>>>>>> correctly on
>>>>>> bdrv_close() of an encrypted image.
>>>>> The point is that blk_bs(blk) might be NULL.
>>>> This is not what blk_is_inserted() is checking. It happens to protect
>>>> you against segfaults because it's robust against using NULL, but with
>>>> an existing BDS, checking whether there is a medium inserted (in the
>>>> physical device for passthrough drivers) doesn't make sense.
>>> Not right now it's not. See patch 6.
>> Patch 6 looks unrelated, at least in v2. But if you're trying to say
>> that I looked at the wrong version, you're right: It doesn't protect you
>> against segfaults at this point yet (which is okay, because blk->bs
>> can't be NULL yet), it only performs the misguided inserted check.
>
> Oops, yes, I meant patch 7.
>
>> Doesn't answer my initial question or make that check any better.
>
> The answer to your initial question is: bdrv_key_required() assumes a 
> non-NULL BDS pointer is passed (which is reasonable). Therefore, it 
> crashes when "no medium is inserted" in the sense of !blk_bs(blk).
>
> How it makes the check better: I'll move it after patch 7.

And after realizing that patch 8 breaks this (because you might have a 
non-inserted host CD-ROM backing an encrypted qcow2 file), I'll use 
blk_bs(blk) instead of blk_is_inserted(blk).

Max