What is *supposed* to happen with automated nftables accept rules?
So I'm working with a libvirt issue that partly turned out, after many hours, to be "libvirt is adding rules to make its VMs work but it can't override my default reject in my main nftables rule set so nothing works for the VMs".
There are many many posts with libvirt specifically about problems like this with nftables. I assume similar things happen with docker and anything else that needs to be like "yeah I know you want to reject traffic but this system I'm managing needs to work".
The "problem" is that I have "reject with icmpx type port-unreachable" in my INPUT chain. Which, by the way, I got straight from /etc/nftables/main.nft in the nftables RPM on Fedora 41 (which is what I am running). 30 years of sysadminning leads me to believe that default-deny anything that isn't explicitely accepted is the correct move. :)
So my question is, what *should* happen here? As far as I can tell, there is absolutely nothing the libvirt tooling can do to override my reject. I can't jump between tables so I can't do like "jump libvert_inp" in my chain. What's the right move? Options I've thought of; I'm hoping there's something better:
- I try to add a very generic blanket accept for "traffic that's probably libvirt related"; I dunno can I add a rule for virbr0 if libvirt hasn't set it up yet?, that seems unlikely to work, and I can't see what else I could do for such a rule
- libvirt and I both move to firewalld which I gather doesn't have this problem?
- It's just expected that I have to copy the libvirt rules in details into my chain so the accepts happen in the right place
Surely this comes up regularly and there must be a standard response?, but I definitely couldn't find anything on the nftables wiki.
Hi,
Quoted 5 lines · On 21. februára 2025 15:40:51 UTC, [email protected] wrote:
So my question is, what *should* happen here? As far as I can tell, there is absolutely nothing the libvirt tooling can do to override my reject. I can't jump between tables so I can't do like "jump
libvert_inp" in my chain. What's the right move? Options I've thought of; I'm hoping there's something better:
Basicaly, you can "overide" reject in two steps:
1. mark packets accepted by libvirt (or generaly in any other rule)
2. exlude these marks from your reject
You must check in libvirt docs, If it is able to set some mark...
regards
Signature · 2 lines
Slavko https://www.slavino.sk/
A message that is not in the archive (sent off-list or not yet received): CAPR8QA+LdZOQB+-i6Uf68S-jwBYBswyWDX33=D+o8vSYW47OLA@mail.gmail.com
libvirt doesn't have that currently, but that's the sort of better solution I had in mind; I'll talk to them about it. Thanks!
Quoted 30 lines · Robin Powell
libvirt doesn't have that currently, but that's the sort of better solution I had in mind; I'll talk to them about it.
Quoted 25 lines · Slavko
Hi,
Quoted 5 lines · On 21. februára 2025 15:40:51 UTC, [email protected] wrote:
So my question is, what *should* happen here? As far as I can tell, there is absolutely nothing the libvirt tooling can do to override my reject. I can't jump between tables so I can't do like "jump
libvert_inp" in my chain. What's the right move? Options I've thought of; I'm hoping there's something better:Basicaly, you can "overide" reject in two steps:
1. mark packets accepted by libvirt (or generaly in any other rule)
2. exlude these marks from your rejectYou must check in libvirt docs, If it is able to set some mark...
regards
--
Slavko
https://www.slavino.sk/
Quoted 3 lines · On 2/21/25 10:40, [email protected] wrote:
So my question is, what *should* happen here? As far as I can tell, there is absolutely nothing the libvirt tooling can do to override my reject.
The libvirt *tooling* can't do anything about it, but the libvirt *documentation* can. Rather than changing your firewall rules without telling you (!!!), libvirt should just document what network traffic requirements it has, and let you update your firewall appropriately yourself.
For example, [1] is the documentation from OpenShift explaining what sort of node-to-node traffic needs to be allowed, so people creating their own firewalls (via any technology) can avoid blocking critical cluster traffic.
-- Dan