Re: SCSI disk devices

Dan Merillat (
Tue, 13 May 1997 04:22:04 -0400

Chris Arguin writes:

> I would tend to think that using both schemes wouldn't be so bad. I like
> the Solaris approach for it's "elegance", but Naming the partitions
> using the volume label seems much more practical in most cases. The most
> obvious problem with that I can see is that your CD-ROM's label would
> change constantly.

Point being? the device ID won't change, just the label. A userspace
daemon (the userland side of autofs, or perhaps kerneld) could get a
notice (LABEL_CHANGED) and reread /proc and create the new devicename.

I'd be happy if we just had some more information about scsi available
in /proc/scsi ... volume serial, current location...
all I can pull out is:
Host: scsi0 Channel: 00 Id: 00 Lun: 00
Vendor: SEAGATE Model: ST31230N Rev: 0594
Type: Direct-Access ANSI SCSI revision: 02

Quick, what is the device name? Or number, I could create the proper
device if I needed to. Can't tell? Well, there goes the userspace

> Maybe if the kernel created the volume ID names as sym-links, so that
> you could also easily tell where that drive was just by looking at them.

Hehehehe. kernel bloat! And _WHY_ is creating device name policy a
kernel issue?

Seriously, if/when we have 32bit device ID's the device number should
be fixed based on cxtxdxsx (or whatever scheme you prefer) And
enough information should be availible in /proc to let a userspace program
mknod() the proper names (based on volume, serial number, or whatever
name the user wants to give them. There's no reason to import more
names into the kernel.

Of course, major 8 will have to be the device in the order it was assigned
(like we have now) or else way too much will break. Also because
it's a hell of a lot easier to boot off of 08:01 then it is to put in
LILO: linux root=/dev/c1t0l0p1 (or worse, in hex xx:01010001)

- --Dan

------- End of Forwarded Message