Re: per-FIB socket binding

> On 21.12.2024 19:34, Bjoern A. Zeeb wrote:
> > How much use are FIBs still these days?  Half of the original use cases
> > I can think of could easily and better be overcome by using vnet jails
> > with a physical or virtual interface (e.g, vcc) being delegated to the
> > vnet.

Among the other half is simple multihoming. If a host has interfaces on two networks 
then it needs two default routes which means two FIBs. Many of us use ipfw to select a 
FIB for an outbound packet based on its IP source address. Without this the next-next-
hop will wonder why we're failing BCP38 source address validation, and it may be a 
firewall which only passes traffic it expects, and it may be a differently congested network 
that upsets TCP pacing since the payloads take one path and the ACKs take the other. 
Path symmetry is an unalloyed good. Path assymetry is a livable problem but worthy of 
avoidance. Some operators just avoid multi-homing. As an operator I often don't have a 
choice about multi-homing or else my other choices are worse. FIBs make all of this 
better, and easily.

> > I would honestly know who and how FIBs are still in use today or if they
> > should be put on a list to be removed for 16 (I assume I might be
> > surprised).

I have a refactor to propose for the networking stack in 16 but more for performance and 
clarity than for correctness or simplicity. More on that another day. In any case I use FIBs 
everywhere all the time.

On Monday, December 23, 2024 10:29:01 AM UTC Andrey V. Elsukov wrote:
> Some might say that VNET is useless and should be removed instead. We
> have bhyve and old-style jails. Without VNET the kernel code will be
> robust and simple again, and easy for debugging.

I introduced Mr. Zec to NL-Net when he was first looking for sponsors, and I was very 
happy with the results, and even more so when the FreeBSD Foundation got involved and 
more so after that when Mr. Zeeb got involved. I think VNET has earned a place in 
FreeBSD's pantheon but that the form it takes in the source code (all those macros) could 
be simplified.

> FIBs are useful as is, but also can be used with "ipfw setfib" that make
> it irreplaceable.

For my primary FIB use case, ipfw is OK, but I think we need a different default. To that 
end, see Message-ID <[email protected]>.

-- 
Paul Vixie