Peter Memishian wrote:
> I don't think it can happen.
I think Darren's talking about SIOCSLIFINDEX.
Yup. This would appear to allow the SNMP rule Jim cited
to be broken, too, unless I'm missing something. The
original problem I mentioned is not something that is
ever likely to happen but it is out there.
I think there needs to be two seperate indexes - one used
for the physical interface (phyint_ifindex) and another
used for ill's (and handles SIOCSLIFINDEX) so we can keep
the phyint_ifindex private and obeying the SNMP rule.
To get a wrap around with our current numbering of
ifindex requires a continuous plumb/unplumb rate of
over 100 per second for an entire year. The only sane
scenario I can think of backing that would be something
PPP related.
However this doesn't solve the original problem (b_prev
holding the ifindex) as there are still edge situations
that could result in the packet going out the wrong NIC.
A more sensible/sane approach would be to ditch using
b_prev like this and attach the mblk's in a chain to
either the IRE or ILL.
Thoughts? Comments?
Darren
_______________________________________________
networking-discuss mailing list
[email protected]
