fix(network-legacy): write fallback teamd config to /etc/teamd

[prabhakarpujeri]

Problem

Booting with team=<master>:<slaves>[:<runner>] on the kernel command line and no host ifcfg TEAM_CONFIG relies on the fallback configuration that modules.d/35network-legacy/parse-team.sh generates. Commit e4483e5 ("Configure the runner for team interfaces") changed the output path of that generated config from /etc/teamd/<master>.conf to /tmp/<master>.conf — but the consumer was never updated: ifup.sh:380 still launches

teamd -d -U -n -N -t <master> -f /etc/teamd/<master>.conf

So the fallback config is written to a location nobody reads. With a real teamd this fails as:

Failed to get absolute path of "/etc/teamd/team0.conf": No such file or directory

→ teamd exits, the team master is never created, and network boot over a team is broken for the cmdline-only (no host config) case. Setups whose host TEAM_CONFIG was installed into /etc/teamd/ at image build time (module-setup.sh) are unaffected — which is presumably why this went unnoticed since 2020.

Fix

Restore writing the fallback config to /etc/teamd/<master>.conf, matching the consumer in ifup.sh and the pre-e4483e59 behavior. One line changed. The /tmp/team.<master>.info / /tmp/team.<master>.up state files correctly stay in /tmp (they are consumed by ifup.sh directly).

Verification

No CI coverage for this path (the network-legacy BOND/BRIDGE/VLAN tests covering team were removed in 88406ce as always-skipped), so verified manually end-to-end with the real parse-team.sh and the real teamd 1.32 on Fedora 44 in private mount/network namespaces:

===== BEFORE (current master) =====
WARN: Team master team0 specified ... Using activebackup.
--- teamd invocation (same flags as ifup.sh:380):
>>> teamd DEAD, team0 ABSENT. teamd log:
Failed to get absolute path of "/etc/teamd/team0.conf": No such file or directory

===== AFTER (this PR) =====
WARN: Team master team0 specified ... Using activebackup.
>>> teamd RUNNING, team0 device EXISTS. teamdctl state:
setup:
  runner: activebackup
4: team0: <BROADCAST,MULTICAST> mtu 1500 qdisc noop state DOWN ...

(team0 is DOWN only because no ports are enslaved in this minimal test — ifup.sh enslaves the working slaves right after starting teamd. The failing part — loading the config and creating the device with the right runner — works.)

The generated JSON {"runner": {"name": "activebackup"}, "link_watch": {"name": "ethtool"}} is identical in both runs; only its location changes.

[devkontrol]

CC @aafeijoo-suse @Nowa-Ammerlaan - can you please help with the maintenance of network-legacy and help to review this PR ? Thanks !

[devkontrol]

Context

Part of a bug-review pass over the tree (this is finding H4 there). Same pass produced #2599, #2600 and the analysis behind #2603.

What is the relevance of this section for this PR ?

[prabhakarpujeri]

You're right — that paragraph was leftover from our internal review notes, removed from the description. Thanks for the review.

[Nowa-Ammerlaan]

CC @aafeijoo-suse @Nowa-Ammerlaan - can you please help with the maintenance of network-legacy and help to review this PR ? Thanks !

I have no knowledge of teamd, sorry.

[aafeijoo-suse]

I have no knowledge of teamd, sorry.

Ditto. Ask @trumbaut why he changed it before assuming the LLM is right.

[prabhakarpujeri]

Fair point — @trumbaut, since e4483e5 moved the fallback path from /etc/teamd/<master>.conf to /tmp/<master>.conf, could you share the original reason? If there was a rationale (e.g. something about /etc not yet being populated), I'd like to capture it here.

For context, the regression evidence posted above is empirical, not just reading code: running the real parse-team.sh and the real teamd 1.32 in sandboxed mount+network namespaces — before: teamd exits with Failed to get absolute path "/etc/teamd/team0.conf": No such file or directory, after: teamd loads the config and teamdctl reports the activebackup runner. The /tmp/team.<master>.info / /tmp/team.<master>.up state files intentionally stay in /tmp; only the config the daemon is launched with moves back to the pre-e4483e59 location.

[devkontrol]

CC @dracut-ng/gentoo-maint @dracut-ng/opensuse-maint

[mwilck]

openSUSE doesn't support teamd any more.

[devkontrol]

Booting with team=<master>:<slaves>[:<runner>]

openSUSE doesn't support teamd any more.

Perhaps we should be removing this code than instead of fixing it ? I also do not think we have teamd in any of the CI containers.

[mwilck]

Perhaps we should be removing this code than instead of fixing it ?

I was just talking about openSUSE. I can't speak for other distros. But it seems that the libteam project has been inactive for some time.

[devkontrol]

Perhaps we should be removing this code than instead of fixing it ?

I was just talking about openSUSE. I can't speak for other distros. But it seems that the libteam project has been inactive for some time.

Thanks @mwilck . Somehow we need to make a decision on this for the purpose of what is maintained upstream. Many distributions already gave up on network-legacy dracut module completely.

Non-systemd distributions (including Gentoo) and and openSUSE are the top use-cases left as far as I can tell. Many non-systemd distributions do not package teamd (Alpine, Void).

Would be good to hear from Gentoo on this (CC @ConiKost)

[chewi]

Gentoo still has a package for it. It doesn't have a dedicated maintainer, but there's probably little to do on it. I'm not sure if the reporter is actually using it or just fishing for issues with AI. Not that the fix is unwelcome. It's straightforward enough, although I wonder if it should be fixed the other way, with ifup.sh updated to use /tmp instead. Or maybe /run would be better if teamd continues to run after switching root. I don't know anything about teamd either.