On Fri, 2004-08-06 at 17:14, Martin Mares wrote:
Hello!
I see always the same answers from Linux people who don't know anyrthing than
their belly button :-(
Chek Solaris to see that your statements are wrong.
Well, so could you please enlighten the Linux people and say in a couple
of words how it could be done?
Have a nice fortnight
I can offer reasons as to why it cannot :-)
The path_to_inst stuff assumes a simple physical bus topology. It is
completely unsuited to e.g. a fibre-channel fabric. It is also unsuited
to iSCSI - my disks aren't attached to eth0, they're attached to
whichever interface has a route to the server. It's also worthless for
USB. The controller, bus and target are meaningless - the target number
is dynamically assigned at plugin so giving a name to controller 0, bus
0 target 3 is utterly pointless.
Linux and/or associated drivers has mechanisms to handle consistent
naming for a number of these (WWPN-binding for FC, similar device
binding of the unique ids for iSCSI, the hotplug infrastructure for usb
etc.). All of these map devices to consistent device names in /dev. The
"Unix" way of accessing devices has always been via names in /dev. I
completely fail to understand why Joerg wants to try to force a naming
model that is both alien to the native systems (I want /dev/cdrw on
Linux; I probably want D: on Windows or something like that), and is
inadequate to the task if you move beyond the simple world of parallel
SCSI.