Peter Memishian wrote:
> I can't answer that but we do support it.
And the system will weather the issue, right?
> It sounds like you're in favour of just noting this is flawed
> and moving on as there doesn't seem to be anything useful
> that can be done here, either way, yes?
Yes. It seems fairly obvious to me that changing the interface index of
an actively used interface will lead to confusion. For instance, an
application may have previously mapped a given interface name into an
interface index, and then used one of a number of ioctls to bind itself to
that index. If someone reassociates that interface index with a different
interface, the application won't know and it will use the wrong interface.
...
> I can't answer that but we do support it. I found references
> in 1999/225 and 1997/184. I could quite easily see someone
> using "ifconfig hme0 index #" on a production system if they
> are transitioning interfaces with DR.
They don't need to do that; RCM will take care of everything for them
via SIOCSLIFOINDEX. That said, the IPMP rearchitecture will obsolete
this problem entirely since applications will no longer bind to the
underlying interface indices.
Well, I've never heard of an application binding to an
interface index (it is bad enough if they chose an interface
name rather than IP address) but it wouldn't surprise me
if someone wrote a program to function in that way. I'd
only hope that they tested their application against use
of ifconfig to change the index before releasing it.
I've summarised the original problem in 6430851.
Darren
_______________________________________________
networking-discuss mailing list
[email protected]
