On Wed, 2008-04-23 at 15:27 -0700, Cathy Zhou wrote: > Cathy Zhou wrote: > >>>> One hole in this approach exists when dlmgmtd isn't running and network > >>>> devices attach (does this happen during reconfigure boot?). What do we > >>>> do for that case? Perhaps we need to re-think whether links are created > >>>> at all if devices attach and dlmgmtd isn't running... > >>>> > >>> dlmgmtd is usually running before network attaches, except during the > >>> netinstall or netboot, that a primary device is usually plumbed by the > >>> kernel > >>> at very early phase before any daemon can be run. Note that a link cannot > >>> be > >>> renamed when it is plumbed. How to rename this primary device to a vanity > >>> name > >>> is another problem that we need to resolve. > >> Do we care about what datalinks are called during a netboot/netinstall? > >> > > I am not familiar with the Indiana boot process yet. If no datalink > > configuration or other configuration based on datalink names needs to be > > done > > during boot, then I guess we could defer the rename to the subsequent boot. > > > Another issue that just comes to my mind when I read the thread on Indiana > alias: is it acceptable that the the numbering scheme of netN names are > completely arbitrary? Say we have 4 interfaces: bge0, bge1, ce0, ce1. Is it > acceptable to have mappings like: > > net0(bge1) net1(ce0) net2(ce1) net3(bge0)
Goot point. Maybe we can use the devname as a naming "hint" somehow? What would be the preferred naming scheme in the scenario you just posed? In the order in which devices have registered, would it be: net1(bge1) net2(ce0) net3(ce1) net0(bge0) ? I wonder if this logic be codified in a sane manner. -Seb
