On Thu, 2002-08-01 at 15:13, Christoph Hellwig wrote:
> On Thu, Aug 01, 2002 at 02:45:37PM -0500, Luciano Chavez wrote:
> > Umm, where did you see O_NDELAY?
>
> I meant O_NONBLOCK, sorry.
>
I see. Cut and paste from one of the other EVMS utilities brought that.
I don't see that the flag is necessary. Good catch.
> > Yes, evms does appear in /proc/devices. What dynamic major detection?
> > Are you referring to the way device-mapper obtains major numbers? We
> > only have one, 117.
>
> I refer to the way of checking the major number that the actually running
> kernel uses. this is important for drivers such a dm or xvm that don't
> have a static major but also helps other in case it e.g. acquires another
> major number.
>
Understood. I don't see a problem with the extra check but figure the
first one will always succeed for an EVMS volume.
> > Do you really want to fail mkfs.xfs if you can't get the stripe info?
>
> is there any reason such an ioctl should fail for a valid evms configuration?
> if no then yes, I want it to fail. else we should just return 0.
>
No, it shouldn't fail. EVMS 1.1.0 introduces the
EVMS_GET_VOL_STRIPE_INFO IOCtl and the permanent 117 major. The previous
stable release 1.0.1 still used the experimental major of 63 so the
current code should never get far enough to invoke the IOCtl since the
major on the device isn't recognized to be EVMS.
That being the case, I suppose it is OK to fail the mkfs if the IOCtl
fails.
--
regards,
Luciano Chavez
lnx1138@xxxxxxxxxx
http://evms.sourceforge.net
|