iptables-nft-default
Summary
Make iptables-nft the preferred iptables variant.
Also see Changes/firewalld_default_to_nftables.
Owner
- Name: Phil Sutter
- Email: psutter@redhat.com
Current status
- Targeted release: Fedora 32
- Last updated: 2019-11-19
- Tracker bug: #1774046
- Release notes tracker: #420
Detailed Description
iptables-nft package provides alternative implementations of
iptables, ip6tables, ebtables and arptables and associated save and restore
commands. These use nftables internally while providing the same look'n'feel as
the original tools. Users may choose between both implementations using
alternatives tool.
Upstream considers the traditional implementations legacy and therefore renamed
the binaries adding '-legacy' suffix. In Fedora, same has been done to
arptables and ebtables packages, namely renaming them
to arptables-legacy and ebtables-legacy. Legacy
iptables and ip6tables remain in
iptables package, which in fact is the only one other packages
depend upon.
To change the status quo, two measures are planned:
Raise priority of nft-variants in alternatives
Currently, legacy variants are installed with priority 10 and nft variants with priority 5. This must be changed as otherwise installing iptables-legacy in a system with iptables-nft installed would change the active alternative (since they are in automatic mode by default).
On the other hand, existing systems using legacy variants should not be changed by a system update. Therefore nft variants' priorities should be chosen to match legacy ones.
Rename iptables package
New name should be iptables-legacy which aligns with ebtables and arptables and reflects upstream status. To resolve dependencies, Provides: iptables statement will be added to iptables-nft package. This should automatically change the default variant to nft.
Benefit to Fedora
- RHEL8 ships nft-variants exclusively, make Fedora align with that by default while still providing the option to fall back to legacy tools.
- New features and improvements are likely to hit nft-variants due to the possibility nftables backend allows for. Although at this point some legacy features (e.g. ebtables among match) are still missing, others are already there (like, e.g. xtables-monitor tool) or are being upstreamed right now (improved tool performance when dealing with large rulesets).
Scope
- Proposal owners:
Changes are rather simple: Rename iptables package, add Provides: line to iptables-nft package, change priorities used when calling alternatives.
- Other developers:
The changed tools may cause regressions among packages using them and it affects only new installations (or those manually switched over). So while no explicit effort is required from them, they should be made aware of the change so they take a possible regression in iptables into account, quickly test against legacy variant and file a ticket (or complain to the right person) if that fixes the problem.
- Release engineering: #8934
- Policies and guidelines: No change required
- Trademark approval: N/A (not needed for this Change)
Upgrade/compatibility impact
Due to the package rename and Provides: line, upgrades will pull
in iptables-nft package. But due to the equal alternatives
priorities, existing choices won't be changed and so existing installations
shouldn't be harmed (apart from forced installation of
iptables-nft package).
Sadly, there are a few known issues, like e.g. missing support for ebtables
broute table or among match and a few iptables targets/matches. Users depending
on such features are advised to install iptables-legacy package
and switch variants using alternatives.
How To Test
Any users of iptables/ebtables/arptables should switch to nft-variants using alternatives tool (if necessary) and check that everything works as before. Any issues should be reported despite the known compatibility issues described above since knowledge about who uses the missing features is valuable information for both up- and downstream.
User Experience
Ideally look'n'feel shouldn't change. Since iptables-nft does not need a lock file anymore, no problems with stale xtables-lock or parallel iptables calls in different mount namespaces are expected anymore. Given the changes currently being upstreamed, users dealing with large rulesets should see a performance increase when manipulating the ruleset (lower run-times of iptables or iptables-restore, packet processing speed should not really change).
Dependencies
Other packages depending on iptables:
- NFStest
- clatd
- ctdb
- fail2ban-server
- firewalld
- fwsnort
- iptstate
- libvirt-daemon-driver-network
- libvirt-daemon-driver-nwfilter
- moby-engine
- nfacct
- origin
- podman
- psad
- python3-ipatests
- ravada
- rkt
- shorewall
- shorewall-init
- shorewall-lite
- shorewall6
- shorewall6-lite
- sshuttle
- sslsplit
- ufw
Since nft-variants are supposed to be drop-in replacements, no outside contribution is needed in order to perform this change.
Contingency Plan
- Contingency mechanism: Nothing needs to be done, the change should be atomic.
- Contingency deadline: N/A
- Blocks release? No