> It is probably better to not use LPF currently, because the version in
> 2.2.1 is buggy. For DHCP's need user space filtering should be fast
> enough anyways (remember that you can bind on a link layer protocol for
> packet sockets)
This feature has been present since Linux version 1. The reason we
didn't use it from the start is because your statement, "For DHCP's
need user space filtering should be fast enough" is hopelessly
optimistic. Whether or not user space filtering is fast enough
depends on how many packets the machine on which the DHCP server is
running receives, not on the usual packet rate the DHCP protocol
requires.
If you run just the DHCP server on a machine, on a relatively
broadcast-quiet network, then sure, the overhead is limited. If you
want to run the DHCP service from a busy NFS server, the kernel's
going to copy every single NFS mobygram into user space for the DHCP
server to examine. Here, the overhead is somewhat horrifying. It
might still work, but nobody could ever claim that it'd be a sensible
way to go.
If the lpf implementation is not reliable, this is yet another reason
why the old API, which worked just fine, should be reinstated.
_MelloN_
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/