Re: SCSI device numbering (was: Re: Ideas for v2.1

Erik Corry (erik@kroete2.freinet.de)
Wed, 3 Jul 96 09:07 MET DST


Leonard N. Zubkoff (lnz@dandelion.com) wrote:
: Furthermore, in the event that a device or host adapter is added or removed
: from the system, the name of any other device should not change unless it has
: been moved or its host adapter has been moved.

There is a need for the name of the device (ie the minor number) to stay
constant when devices are added and removed (switched on and off), but
I think it is too much to ask to keep the names the same when _adapters_
are added/removed. If you require that, the task becomes too difficult
and is then not implemented (the situtation up till now).

If I add or remove a controller, that's a conscious decision that involves
switching off the machine. I expect to have to do some reconfiguring. But
if I switch off the external tape drive because it has a noisy fan and the
machine for some reason reboots itself (perhaps due to a parity error) it's
a little sad to find the reboot failed because the number of SCSI devices
has changed. That's not the same as saying I want to be able to completely
rebuild the system with new controllers etc. and still boot without
editing /etc/fstab.

A lot of the huge minor number suggestions I've seen here actually make
a common situation more difficult: I reconfigure a SCSI adapter
because I've added a sound card. If I have a minor number that's
dependent on the IRQ or I/O address, then my disks will all change
name. If the adapters are simply numbered in the order they are found,
then the minor numbers can't change for the common case that I only
have one SCSI adapter. This isn't a case of small-system chauvenism:
I don't think the suggestion helps the large systems either. I think
the only thing that helps a large system is mounting by volume labels,
which has little to do with the minor number change.

So the suggestions that make sense are:

1) Minor number encodes stably the device's position on the SCSI bus, but not
which bus it's on.
2) The ability to give volume labels to mount (and as the root device perhaps)

and what doesn't make sense is:

3) Attempting to stably encode the identity of the SCSI adapter in the
minor number.

IMHO

: [Yet another /proc/dev suggestion]

Leonard, it's been done to death. The only argument that's going to help
this cause is (maybe) "I've got the patches and lots of people are using
it and they love it".

P.S. on the mounting-by-volume-labels idea: What if two volumes have
the same label? This could be done deliberately as part of an attack,
for example putting a trojan volume in a removable SCSI drive.

-- 
It's not so much an afterlife, more a sort of apres-vie.  --DNA
--
Erik Corry ehcorry@inet.uni-c.dk http://inet.uni-c.dk/~ehcorry/  +45 86166287
Configuration/programmering af Internet/TCPIP/Unix/News/Mail/WWW/net-sikkerhed..