On 08/07/2015 13:01, Kevin Wolf wrote:
> Am 08.07.2015 um 12:47 hat Laurent Vivier geschrieben:
>>
>>
>> On 08/07/2015 12:31, Kevin Wolf wrote:
>>> Am 02.07.2015 um 16:18 hat Laurent Vivier geschrieben:
>>>>
>>>>
>>>> On 02/07/2015 16:03, Paolo Bonzini wrote:
>>>>>
>>>>>
>>>>> On 02/07/2015 15:58, Laurent Vivier wrote:
>>>>>> Since any /dev entry can be treated as a raw disk image, it is worth
>>>>>> noting which devices can be accessed when and how. /dev/rdisk nodes are
>>>>>> character-special devices, but are "raw" in the BSD sense and force
>>>>>> block-aligned I/O. They are closer to the physical disk than the buffer
>>>>>> cache. /dev/disk nodes, on the other hand, are buffered block-special
>>>>>> devices and are used primarily by the kernel's filesystem code.
>>>>>
>>>>> So the right thing to do would not be just to set need_alignment, but to
>>>>> probe it like we do on Linux for BDRV_O_NO_CACHE.
>>>>>
>>>>> I'm okay with doing the simple thing, but it needs a comment for
>>>>> non-BSDers.
>>>>
>>>> So, what we have to do, in our case, for MacOS X cdrom, is something like:
>>>>
>>>> ... GetBSDPath ...
>>>> ...
>>>> if (flags & BDRV_O_NOCACHE) {
>>>> strcat(bsdPath, "r");
>>>> }
>>>> ...
>>>
>>> I would avoid such magic. What we could do is rejecting /dev/rdisk nodes
>>> without BDRV_O_NOCACHE.
>>
>> It's not how it works...
>>
>> Look in hdev_open().
>>
>> If user provides /dev/cdrom on the command line, in the case of MacOS X,
>> QEMU searches for a cdrom drive in the system and set filename to
>> /dev/rdiskX according to the result.
>
> Oh, we're already playing such games... I guess you're right then.
>
> It even seems to be not only for '/dev/cdrom', but for everything
> starting with this string. Does anyone know what's the reason for that?
At least 10 years old reason:
commit 3b0d4f61c917c4612b561d75b33a11f4da00738b
Author: bellard <bellard@c046a42c-6fe2-441c-8c8c-71466251a162>
Date: Sun Oct 30 18:30:10 2005 +0000
OS X: support for the built in CD-ROM drive (Mike Kronenberg)
>
> Also, I guess before doing strcat() on bsdPath, we should check the
> buffer length...
>
>> Perhaps this part should be removed.
>>
>> But if we just want to correct the bug, we must not set filename to
>> /dev/rdiskX if NOCACHE is not set but to /dev/diskX
>>
>> It's the aim of this change.
>
> Yes, that looks right.
>
> Kevin
>
