This function looks innocent enough but has potential to mess with the
wrong interfaces or do operations all over or just general code drift
that doesn't seem so nice nowadays.
Tested on static LAN with WAN track and dhcp4/6 over there. Also
survices a reboot which is good enough for a first commit. More testing
to come.
Also destroys 6rd/6to4 tunnel devices now although the scrubbing of
the interface should work after this rework. Just trying to stay
consistent with the configure functions which destroy the interface
on each configure run.
Copyright in file since 2015 when work started to remove the PHP module.
A multitude of improvemets followed ever after.
Related PR: https://github.com/opnsense/core/issues/2661
/dev/zfs/pool does not exist so we can do a "zfs/pool" input
in alignment with gmirror and graid inputs. Move the probing
after zfs load to be able to unload again also.
in some situation, like OpenStack, mtu is less then default (1500) and
it cause replication issue
we have noticed on a low used firewall with 2/3k State table size more
then 100k State table size on the backup server
copy mtu from real interface fix this issue
Part one makes it possible to inject branding info from the Makefile
which is not fully complete yet but can always be finished. The new
hash value can thus be used as a unqiue identifier for page resources
that may be subject to caching. By using the git hash it allows us
to have this effect on test commits as well as earch working version
as we don't want to track the changes for each file but still get a
good amount of caching.
Several vendors use EFI BC instead of the EFI x86-64 value for the arch, some thinkpads for example. We noticed this while we tried to use PXE on all our machines and that some EFI systems with 64-bit didn't work and some did. We then found the old commit that change the if clause from 00:07 to 00:09 and tested around. Thus we found out that some machines use 7 and some 9, so IMHO it's correct to redirect both cases to the 64bit syslinux filename. This commit doesn't break old configs, just adds the additional option.
This simplifies the logic for the interface redirect target
as we use :0 and push it directly to the kernel to resolve.
From pf.conf:
Tables can also be used for the redirect address of nat and
rdr rules and in the routing options of filter rules, but only
for round-robin pools.
