On 10/10/18 4:14 PM, unman wrote:
On Tue, Oct 09, 2018 at 09:18:22PM +0300, Ivan Mitev wrote:

On 10/9/18 7:44 PM, mfreemon wrote:
On 10/8/18 10:56 AM, mfreemon wrote:
On 10/2/18 2:25 AM, Ivan Mitev wrote:
On 10/2/18 1:32 AM, Chris Laprise wrote:
On 10/01/2018 05:48 PM, mfreemon wrote:
On 1/11/18 3:01 PM, Chris Laprise wrote:
   > On 01/10/2018 03:47 PM, Connor Page wrote:
   >> The official templates use nftables so shouldn’t be mixed with
iptables. I didn’t have time to learn about nftables, so just removed
nftables package from debian 9 template. YMMV.
   > Hmmm, I was just thinking how Qubes' own guest scripts still use
   > iptables even in fedora-26.
   > IIUC, iptables and nft are two different interfaces
to netfilter. I
   > don't know if it really matters, at least for the R4.0 window. I'd
   > prefer to put the syntax change (for docs) off until
a later release.

I was recently thrown by the mix of both nftables and iptables in R4.

The qubes docs don't clarify much.  The qubes firewall scripts use
nft. Most of the discussion on the qubes website documentation is
about iptables, but there are also a few mentions of nft.  The upgrade
instructions (going from R3.2 to R4) did not mention converting rules
from iptables to nftables.  It looks like other related projects (one
example is qubes-tunnel) is using iptables.

Just reading a few things and trying to come up to speed, I get the
impression that nftables and iptables should not both by used at the
same time.  Even if technically possible (i.e. both sets of rules
applied correctly), it strikes me as not a great idea to maintain
packet filtering rules in two different ways.

What is the best practice recommendation on this (for R4, Fedora 28
template)?  Are we to be using, exclusively, nftables in R4?

The last I read about this (for 4.0) is that nftables is used in Fedora
Qubes code, but Debian Qubes is still using iptables. That
still appears
to be the case since nftables is not installed in my
debian-9 templates.

I've submitted qubes-tunnel to Qubes with iptables commands only, with
the intention to transition to nftables (or that other new interface in
Linux, name escapes me just now) for Qubes 4.1. Someone who is just
starting a project might be better off going with nftables.

... until yet another packet filtering mechanism replaces nftables (in
that case, bpfilter [1]).

I understand the rationale behind using nftables [2] but given how it is
widespread (hint: close to 0 even amongst seasoned sysadmins) IMHO it
wasn't worth it. The OP's post confirms there's quite some confusion
about how it interacts with iptables, and the official documentation is
far from helpful.
I'm quite proficient with iptables and networking in general but it took
me half an hour to understand how to tweak Qubes' nftables rules last
time I wanted to change something in the firewall, while I would have
done that task in less than one minute with iptables. I could have spent
a few hours learning nftables to improve the official doc but at my age
I prefer to spend time learning tech that significantly improves things
(eg. Qubes OS over standard linux distribution) over loosing time
learning stuff that is only marginally better.
Anyway - I digress :)

[1] https://old.lwn.net/Articles/747551/

I'm concerned about the confusion and unnecessary complexity here.

Network packet filtering is certainly (one of) those features that
software such Qubes needs to be solid on (in both design approach
and implementation detail).

Is the Qubes team confident in the current situation, such that
users of Qubes should not be concerned?

nb.  This is not meant to be a criticism at all.  I very much
appreciate the hard (and complicated) work going into Qubes.  I'm
just looking to understand the current situation better so as to
judge whether my concern is warranted or not.

As an example:  I'm wanting to enable some specific network traffic
between two qubes.  The docs say to use iptables 
  qubes-firewall-user-script also specifies iptables rules.  But
qvm-firewall implements the rules it manages using nftables.  So the
firewall VMs have both iptables rules and nftables rules in effect.  And
these are different sets of rules.  It's not that the iptables command
and the nft command are just two user interfaces showing the same packet
filtering rules.  They are different packet filtering rules.  This seems
like a receipt for disaster.

Is this the wrong forum for this discussion?  Should this be on
qubes-devel, or an issue in qubes-issues at

You'll definitely get more visibility on qubes-devel.

FWIW I'm not concerned about the complexity itself: I trust the Qubes devs
not to mess up.
IMHO the problem is that people proficient with iptables are not willing to
spend time learning yet another packet filter tool when iptables works for
99.99% of the cases (+, as others pointed out, nftables is still not feature
complete wrt. iptables). For those users - an overwhelming majority - Qubes'
nftables firewall is a black box that is difficult to

I think this is the problem. I remember stalwarts hanging on to ipchains
for similar reasons. (I speak as someone who has clung on to iptables for
far too long.)
It seems to me that the few features lacking in nftables are only of
interest to people who are fully capable of learning a new tool. The
extras that nft brings completely outweigh the deficiencies.
nft provides tools to translate your iptables rules in to the new
syntax, so there's really no excuse for not diving in. Even if you have
minimal time, you can write your iptables rules and then translate them
to nft.

I don't think that's a good reason. Ipchains (and even ipfwadm before) had major deficiencies - a crippling one being their lack of state. Iptables addressed those issues and that's why everybody quickly switched to it: the benefits it provided for people serious about firewalling infinitely outweighed the time they'd have to spend learning the tool.

FWIW I've been involved in huge/complicated firewalls the only issue we had with iptables was when restoring thousands of rules: it took a bit of time; and even then, the increasing power of PCs made it increasingly negligible over time. Ipset was later included in the kernel and solved our "issue".

So, as an advanced user of iptables I don't have/see any issue that would be solved by nftables. Sure, one could translate its iptables rules to nftables but the problem is learning how to use nftables later on. For instance, why did it just take me more than 10 minutes to find out how to simply list rules ?

`man nft` ?  {list | flush} ruleset [family] ;

`nft list` -> error.

How/where am I supposed to find out that I have to type `nft list tables` (which I found searching on a random post on the web). And how are "tables" related to a "ruleset" ? etc.

Also, with bpfilter advertised to replace nftables/iptables (maybe someday), people will be reluctant to learn nftables.

Just to be clear, I'm not arguing with you nor any of the devs - I'm just stating my experience, which is pretty much the same among all the "network" guys I know.

Qubes tries to provide a straightforward experience for relatively
inexperienced users, and the nft/iptables mix per distribution is a
compromise to that end.

Probably, but it would be interesting to understand what exactly nftables provides that iptables can't. Marek's post in the issue I've linked to in another post mentioned something about whonix.

The docs need to be updated to provide nft rules throughout.

^ this. + sample usage of nftables.

You received this message because you are subscribed to the Google Groups 
"qubes-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to qubes-users+unsubscr...@googlegroups.com.
To post to this group, send email to qubes-users@googlegroups.com.
To view this discussion on the web visit 
For more options, visit https://groups.google.com/d/optout.

Reply via email to