Re: [REGRESSION] cdrom drive doesn't detect removal

[Kay Sievers]

On Thu, Sep 30, 2010 at 22:47, Florian Mickler <florian@xxxxxxxxxxx> wrote:
> On Thu, 30 Sep 2010 22:32:33 +0200
> Kay Sievers <kay.sievers@xxxxxxxx> wrote:
>
>> On Thu, Sep 30, 2010 at 22:14, Florian Mickler <florian@xxxxxxxxxxx> wrote:
>> > On Thu, 30 Sep 2010 21:27:52 +0200 Kay Sievers <kay.sievers@xxxxxxxx> wrote:
>>
>> >> I don't think anything needs to be or can really be fixed in the
>> >> kernel. We need a working O_EXCL, and before Tejun's patch it was
>> >> broken. That the broken behavior seemed to have some wanted effects on
>> >> some boxes can't the reason to leave O_EXCL broken.
>> >
>> > Ack on this on the technical side.
>> >
>> > But we have to look from the user side at this, as bug tracking is
>> > mostly a tool to get feedback from the users. So if we close this bug,
>> > we loose valuable feedback on the functioning of the kernel.
>> >
>> > So, I can't really close the report if it is not resolved in any way.
>>
>> As said, it can't be resolved. That it seemed to work was a buggy
>> O_EXCL, which let stuff through the O_EXCL barrier, which was never
>> supposed to reach the drive. I can't even reproduce that on older
>> kernels. It seems like a specific timing problem.
>
> Yes I see this from the changelog of Tejuns commit. It shure looks like
> it all works just fine...
>
> Aren't cdrom devices physically locked while mounted? How could this be
> a problem then? Is it about force-opening the device?

Yes, udisks doesn't support force-unlocked cdrom doors. This all can
not happen otherwise.

> And what are the symptoms when the cd is removed while mounted? Are
> there horrible timeouts when accessing the file system?

It did even crash the entire kernel sometimes in the past, I didn't
try the last 2 years.

> I don't think we pay rent in the bugzilla yet... So this can just stay
> open, until the necessary infrastructure is in place..
> But I'm wondering if it is futile or not.

Sure, keep it open if you like. I just say that it's not a regression
in the sense that it can be fixed by rolling back to an earlier state.
Such functionality will need a new feature to be implemented. It was
only a bug that let O_EXCL pass the barrier, and that same bug broke
physical media, and we need to fix it.

Kay
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/