Implement expression for matching on source IP? about nftables HOT 7 OPEN

Hey! Here’s a few pointers:

1. You can work in a network namespace to play around with nft without any consequences for your host system:

# ip netns add nattest
# ip netns exec nattest /bin/bash

2. You can use nft’s --debug all flag to get a bunch more details about what’s going on. For this particular case:

# nft --debug all add rule nat postrouting ip saddr 192.168.69.2 masquerade
[…]
ip nat postrouting 
  [ payload load 4b @ network header + 12 => reg 1 ]
  [ cmp eq reg 1 0x0245a8c0 ]
  [ masq ]
[…]

These are the expressions you’ll need to model with your Go code.

3. Check out

   nftables/nftables_test.go

   Line 77 in 9ac63cb

   func TestConfigureNAT(t *testing.T) {

   — you’ll want to duplicate that, run nft through strace, stuff the resulting byte slices into the test and then work on the Go portion until it matches. I’ve done this to verify that all the building blocks are in place, and will commit it for illustration purposes in a sec.

Hope that helps,

from nftables.

Thank you so much!

Would you be open to a PR where I implement an abstraction for building these rules? Something like:

builder := nftables.ExprBuilder{Chain: chain}
if err := builder.FilterSourceAddr(srcIP); err != nil {
  // ... Errors such as ipv6 address /w ipv4 table family
}
builder.ActionCounter(&counter)
builder.ActionMasquerade()
conn.AddRule(&Rule{ ... Exprs: builder.Expr() ... })

from nftables.

Disclaimer: I haven’t thought much about how a good abstraction layer would look like.

To me, an important question that I can’t answer yet, is whether the ideal abstraction would implement nft(8)’s config/command line syntax (and how stable those are), or whether something else makes more sense.

Intuitively, I’d gravitate towards implementing nft’s syntax: that way, users could just copy their already-existing configuration files. What do you think?

from nftables.

For the sake of minimizing the learning curve, I agree we should keep the concepts as similar to nft & nft's representation as possible.

However, I do think the abstraction needs to be embodied in types; any approach which does string parsing is probably re-inventing the wheel, and loosing the benefits of a type system.

Maybe:

builder.Table(Table{
  Name: ...
  Family: ...
  Chain: &Chain{
    ...
    Rules: []builder.Rules{
      builder.FilterSaddr(saddr),
      builder.Masquerade(),
    },
  },
}).Build()

Thoughts?

from nftables.

However, I do think the abstraction needs to be embodied in types; any approach which does string parsing is probably re-inventing the wheel, and loosing the benefits of a type system.

I’m not sure about that — neither for nor against, I just can’t tell. Given that, I think the best course of action is to develop your abstraction in a separate repository and see how well it works out in practice.

from nftables.

I'll give it a try and report back.

from nftables.

I tried a few things but complexity always crept in to the stage where the APIs were roughly equivalent.

The best thing we can do is probably a bunch of examples & maybe a paragraph on using nft --debug all to work out what to do.

from nftables.