[Ocfs2-tools-devel] [RFC] changing default block device detection method and more

[Fabio Massimo Di Nitto]

Joel Becker wrote:
> On Tue, Nov 27, 2007 at 07:33:05AM +0100, Fabio Massimo Di Nitto wrote:
>> note that I already applied a "udev filter" in /dev scanning to skip .static.
>>
>> +      if (strcmp(namelist[n]->d_name, ".") &&
>> +         strcmp(namelist[n]->d_name, "..") &&
>> +         strcmp(namelist[n]->d_name, ".static") && <----
>> +         strcmp(namelist[n]->d_name, ".udev"))
> 
> 	Checking on things, wouldn't it be best as
> "if (namelist[n]->d_name[0] == ".")"?  Everyone does magic things
> (.static, .udev, .initramfs, .udev.tbd, etc) as hidden files.  Makes us
> pretty future proof.

Yes indeed this is a good idea.

> 
>> I think I have a solution that I believe is the best compromise.
>>
>> We start scanning /sys/block for disks (this will solve directly the filtering
>> problem at the source by skipping tapes, cd-rom's no matter how they are
>> connected to the system or called in /dev).
> 
> 	Yeah, <sysfspath>/removable is a nice start - if 0 we don't even
> worry about checking <sysfspath>/device/media.

I need to check this one. IIRC there are some devices that do not report
correctly the removable bit and some devices have /device/media others have
device/type (IDE vs SCSI IIRC) with different values to determine what they are.
I have it all mapped in another piece of code already. Just need to turn it into C.

# get what kind of media it is.
media=$(cat /sys/block/$block/device/media) 2>/dev/null || media=$(cat
/sys/block/$block/device/type) 2>/dev/null
case $media in
 0|e) media=disk ;;
 5) media=cdrom ;;
esac

>> We gather the major/minor for each device and by simply doing a stat in /dev
>> (and filtering .static as in the previous code) we can associate maj/min to the
> 
> 	We're scanning recursively in /dev, not just listing the
> toplevel, right?

Yes of course we are recursive.

> 
>> name in /dev. If we find no devices we can either wait a few secs or just skip
>> it (thought luck).
> 
> 	Yup.  I'd say wait a second.  The stat(2) will be cached on the
> second pass.

Ok. This can lead to a problem tho that we saw for a while on raid devices (I
think it has been solved in recent kernels) where the device in /dev appears
before the device is ready to be used. We will need to sleep 1 or 2 also in the
scanning device code if we fail.

> 
>> It is slightly more expensive because there might be tons of entries in /sys but
>> none of them will involve a disk access and we will do a real check only on
>> devices we care. Stat of /dev is cheap as an ls (you can't make it cheaper than
>> that I believe).
> 
> 	It depends on what you are looking for.  If you want to iterate
> on devices, you have to do the whole scan anyway (eg the partition list
> in ocfs2console).  If you want a specific device, /dev/disk/by-* is what
> you want.  (I have to look at our use cases again).

Ok.

> 
>> The code will be a bit more complex as we need to work on 2 lists rather than
>> one, but I doubt this is an issue at all.
>>
>> How does it sound?
> 
> 	Sounds like a nice approach.  Are you going to try to make it
> code usable in ocfs2-tools and gfs2-tools?

Yes I can do the code, just bear in mind I will go away for a few weeks soon.

Fabio

-- 
I'm going to make him an offer he can't refuse.