Udev issues in today's coming rawhide

Harald Hoyer harald at redhat.com
Thu Mar 2 15:55:35 UTC 2006


Arjan van de Ven wrote:
> On Thu, 2006-03-02 at 10:13 -0500, John Thacker wrote:
> 
>>On Thu, Mar 02, 2006 at 03:39:21PM +0100, Arjan van de Ven wrote:
>>
>>>better would be to use UUID's, or at least vendor identification strings
>>>etc from the device. We need to NOT tie these device names to underlying
>>>accidental device numbers. That is a major major step back.
>>
>>Wouldn't vendor identification strings, at least by themselves, still 
>>cause problems when people have two identical model devices?  That's
>>fairly rare for CDROM devices but not unheard of.  From what I understand
>>of the output of udevinfo, it doesn't look like a UUID is exported
>>as an environment variable the way that the vendor id strings are.
>>I don't claim to understand everything, though, and of course it could
>>be changed as well.
>>
>>My understanding anyway is that the old method still somewhat depended
>>the underlying accidental device numbers, at least when it came to
>>deciding which was /dev/cdrom and which was /dev/cdrom1.
> 
> 
> but so does the new one! For example on one of my test boxes, it depends
> on if I have a USB stick inserted.. if it is at boot, sda is my usb
> stick and sdb is my sata disk, and sdc my cd writer.
> if it's not, sda is my disk and sdb my cd writer.
> Now add USB cd writers to the mix and it's clear that such device names
> are not persistent at all and utterly useless for identification.
> 
> (and since USB goes via my kvm, this device order is potentially
> different each time I switch my kvm to this machine)
> 
> while you can have 2 of the same type, the problem is at least less than
> before, not bigger than before as is the case with the proposed
> "solution".
> 
> 

Arjan, if you want to have IDs... use:
/dev/disk/by-id/




More information about the fedora-devel-list mailing list