Re: systemd-resolved and single-label DNS entries

[Jeff Layton <jlayton@kernel.org>]

On Tue, 2025-08-26 at 11:07 -0400, Chuck Lever wrote:
> On 8/26/25 10:21 AM, Jeff Layton wrote:
> > On Tue, 2025-08-26 at 14:53 +0200, Daniel Gomez wrote:
> > > On 26/08/2025 12.58, Jeff Layton wrote:
> > > > I'm hitting a problem with NFS testing on kdevops guests. Recently when
> > > > I bring up a guest, it's unable to resolve the name of the NFS server
> > > > via DNS.
> > > 
> > > FYI, nfstests works fine with Debian testing (forky). Both hosts and guests.
> > > 
> > > 
> > 
> > 
> > Thanks. Some questions:
> > 
> > 1/ did your setup get the entries in /etc/hosts? That seems to be what
> > has changed recently. That's not working for some reason.
> > 
> > 2/ Does forky use systemd-resolved? The problem is clearly there with
> > Fedora 42. The DNS server on the virtual bridge works fine:
> > 
> > ----------------8<-----------------
> > $ dig @192.168.122.1 kdevops-nfsd.
> > 
> > ; <<>> DiG 9.18.33 <<>> @192.168.122.1 kdevops-nfsd.
> > ; (1 server found)
> > ;; global options: +cmd
> > ;; Got answer:
> > ;; ->>HEADER<<- opcode: QUERY, status: NOERROR, id: 24955
> > ;; flags: qr aa rd ra; QUERY: 1, ANSWER: 1, AUTHORITY: 0, ADDITIONAL: 1
> > 
> > ;; OPT PSEUDOSECTION:
> > ; EDNS: version: 0, flags:; udp: 1232
> > ;; QUESTION SECTION:
> > ;kdevops-nfsd.                  IN      A
> > 
> > ;; ANSWER SECTION:
> > kdevops-nfsd.           0       IN      A       192.168.122.85
> > 
> > ;; Query time: 0 msec
> > ;; SERVER: 192.168.122.1#53(192.168.122.1) (UDP)
> > ;; WHEN: Tue Aug 26 10:18:47 EDT 2025
> > ;; MSG SIZE  rcvd: 57
> > 
> > $ resolvectl query kdevops-nfsd.
> > kdevops-nfsd.: 'kdevops-nfsd.' not found
> > ----------------8<-----------------
> > 
> > It seems like we're doing something "unseemly" by putting bare names in
> > DNS like this, given that systemd-resolved won't pass them along.
> > 
> > Since systemd-resolved will only resolve bare names without a domain
> > using LLMNR, we should either take steps to make that work, or fix it
> > so that systemd-resolved will forward the requests (by setting up a
> > proper domain name and setting up the search path in the resolver).
> > 
> 
> "won't pass them along" -- does that depend on the settings in
> /etc/nsswitch.conf ?
> 

What's happening in my environment is that this list in the
update_etc_hosts playbook comes up empty:

- name: Build list of hosts                                                                    
  set_fact:       
    ueh_hosts: "{{ ansible_play_hosts_all | difference([inventory_hostname]) }}"
                       
That recently changed in this commit:

commit 1cf0800c9ffced98da393e5d2b1594302d23eb78
Author: Daniel Gomez <da.gomez@samsung.com>
Date:   Thu Jul 24 21:36:14 2025 +0200

    gen_hosts: templates: include localhost in the all group

I couldn't revert that commit cleanly, so I tried out the commit just
before it, and /etc/hosts got updated correctly. I'll have to look at
proper fixes for this, but if anyone has ideas I'm happy to try them
out.
-- 
Jeff Layton <jlayton@kernel.org>