Public SourceMage Discussion List

[Pieter Lenaerts <e-type AT sourcemage.org>]

On Wed, Jun 21, 2006 at 12:59:24PM +0200, Thomas Orgis wrote:
> Am Tue, 20 Jun 2006 11:24:28 +0200
> schrieb Pieter Lenaerts <e-type AT sourcemage.org>:
> 
> > 2. this led us to think about a bus and id variable pair, uniquely
> > identifying the network devices. netconf could present the user with a
> > list of devices from /sys/class/net/ and automagically fill the variable
> > pair.
> 
> Erm, I didn't attend that discussion but do you mean pci device ids? Or
> the pci bus address? Both together may identify a card, one alone not.

this is why we're thinking of a variable pair

> We shouldn't get confused with multiple identical nics in one machine
> and also the config shouldn't break without need when a card (maybe the
> only one!) is moved to another slot.

pci address is not the only candidate for the id variable on the
pci bus. usb bus address is unusable as well. we just have to be
carefull like you say about what we pick.

static dev compatibility is another thing we have to consider for
example.

> I've learned that the MAC is _the_ way to identify a NIC - as long as
> one hasn't fiddled with it before.

the point is that we want to fiddle with it.
mac is volatile in this story, which makes it unusable as unique
identifier.

> I think it would be enough to separate the fields for setting and getting
> the mac (MAC, DEVICE_MAC) and stay with the mac as identifier.

just evaluate some of the use cases and you'll see right away that
they _can_ _not_ do the job once the mac has been changed.

you _can_ _not_ use a volatile field as an identifier.

> So only who really intends to does change the mac of the device -
> I think that someone _really_ should know what he is up to, anyway,
> and should be prepared for any slight discomfort.

personnally I think we should either get device renaming and setting
macs out of netconf or make it robust so that it does what it seems to
provide.

"slight discomforts" and "oh, it only works under circumstances X, Y
and Z" are crap, so either our stuff does a good job or it doesn't say
it can do the job.

> I don't see the network scripts in trouble with changing device name and
> mac since they should know what _they_ do and if the user interferes he
> know what he does

no.
netconf has to work, regardless of what the user did to his device
settings.

--
Pieter Lenaerts
Source Mage GNU/Linux
http://www.sourcemage.org - "Linux so advanced it may well be magic"

Attachment: pgpSOl2YLU7cv.pgp
Description: PGP signature