>>      Note that a similar problem with disks has been solved elegantly by
>> referencing disks by their uuid or label in /etc/fstab. Maybe
>> /etc/network/interface could specify the MAC address as a hook. This would
>> only suppose that the hotplugger creates a symlink to the interface in some
>> /dev/net/by-address/  subdirectory. With this solution, it is up to the admin
>> to decide if  s?he wants a simple configuration based on interface name
>> (eth0) or a secured one alla "Address=a0:d3:c1:9d:a5:86".
> It turns out that something like this already exists.  The ifrename command
> renames interfaces using a /etc/iftab file.  But it is not installed
> by default.

     The problem is the rename, because of the race condition. In 
comparison, here is a citation of the mount man page:

The device indication.
               Most  devices  are  indicated by a file name (of a block 
special
               device), like /dev/sda1, but there are other 
possibilities.  For
               example,  in  the  case  of  an  NFS mount, device may 
look like
               knuth.cwi.nl:/dir.  It is possible to indicate a block  
special
               device using its volume LABEL or UUID (see the -L and -U 
options
               below).

     See the option to use uuid? Mount doesn't require that /dev/sda1 be 
*renamed*; it just can find the block special device itself, given the 
uuid. The randomness of disk naming is solved without messing with those 
names. It doesn't seem a problem to offer a similar option in the ip (or 
ifconfig) command.

     The advantage of supporting an option like 
"hwaddr=a0:d3:c1:9d:a5:86" is that the admin is free to specify 
interfaces by names or by MAC address. Of course, there is now the 
possibility to change the MAC address of an interface, but this is a 
case of severe hacking where the admin has to understand what s?he does.

     Didier