interface renaming

Brooks Davis brooks at one-eyed-alien.net
Mon Nov 3 07:44:01 PST 2003


On Mon, Nov 03, 2003 at 09:52:41AM -0500, Josh Elsasser wrote:
> On Mon, Nov 03, 2003 at 09:27:22AM +0100, Jeroen Ruigrok/asmodai wrote:
> > -On [20031101 19:52], Matthew Dillon (dillon at xxxxxxxxxxxxxxxxxxxx) wrote:
> > >    I think for DragonFly we should leave the naming as it is.  For all intents
> > >    and purposes ethernet interfaces are 'hidden' and 'abstracted' from the
> > >    user be virtue of the route table already, we don't need to abstract them
> > >    further by naming them all the same!
> > 
> > Amen.
> > 
> > It might just be a feeling, but to me it just feels so wrong to have
> > only ethN devices.   I actually like the xlN, fxpN, and so on.
> > 
> > We might ened to check it it makes sense for all devices though.
> 
> I don't think anyone really wants an ethX naming scheme.  What I want
> is just the ability to rename or alias an interface name.  I can't
> count the number of times that I've upgraded or swapped out a network
> card (or motherboard with onboard ethernet) and had to fix all my
> configs.  I would love the ability to just call my interfaces "int"
> and "ext" or something in rc.conf, ipf, and other configs for
> miscellaneous daemons and scripts.

It's is certainly not my intent to implement the Linux stupidity of eth#
in FreeBSD.  I don't know where people keep getting the idea that I want
this.  I might implement it as a tunable option if I get board on a
plane since it's won't take very long and some people do want it, but I
won't let that become the default.

I mostly want to be able to rename interfaces so that changing a nic
requires 1 configuration change not many.  There are also some other
changes I want to make to device cloning that require that names be free
form strings, which is actually why I started down this road in the
first place.  In many ways, making names free form is the most important
reason for the change as there are several pseudo devices for which the
<driver><unit> naming scheme is unsound, ef(4) being the worst.

In the end, I don't really care if DF takes this path or remains the
only BSD under the old scheme (BSDI doesn't count).  FreeBSD did fine
for the 6 years it lagged Net/OpenBSD.  Only the desire for some
features that could definitely be described as sugar coating[0] resulted
in this work being done.  I would suggest that whether or not you adopt
the if_xname change, you import versions of if_printf() and maybe
if_initname() to reduce diffs to 5.x.

-- Brooks

[0] Sugar coating or not, a number of people have asked for them and
most of the cost is a tree sweep and one small API/ABI breakage.

-- 
Any statement of the form "X is the one, true Y" is FALSE.
PGP fingerprint 655D 519C 26A7 82E7 2529  9BF0 5D8E 8BE9 F238 1AD4
Attachment:
pgp00002.pgp
-------------- next part --------------
A non-text attachment was scrubbed...
Name: pgp00002.pgp
Type: application/octet-stream
Size: 189 bytes
Desc: "Description: PGP signature"
URL: <http://lists.dragonflybsd.org/pipermail/kernel/attachments/20031103/a7033e48/attachment-0020.obj>


More information about the Kernel mailing list