OPINIONATED OMARCHY
TROUBLESHOOTING CORPUS
790
RECORDS
790
AUDITED
0
UNCHECKED
2388
SOURCES
LIVE
network high occasional DISPUTE

Fix total DNS failure after connecting Tailscale

omarchyarchendeavouroscachyoslaptopdesktop
CORRECTED An auditor found the fix wrong or incomplete and rewrote it.
AUDIT NOTE

The cited source was the wrong URL and the corpus copy of it was stale. https://github.com/basecamp/omarchy/issues/6962 is a hard 404, confirmed with curl -o /dev/null -w '%{http_code}'. The issue lives at https://github.com/omacom/omarchy/issues/6962 and returns 200. I read that issue and every comment in full. It supports the record closely: the reporter gives the same two health strings, CorpDNS: true with RouteAll: false, the installer-only --accept-routes, and the flagless widget tailscale up. It is still OPEN, so nothing was fixed upstream. I confirmed the two Omarchy claims myself on this machine rather than trusting the issue: /usr/share/omarchy/bin/omarchy-install-service-tailscale runs sudo tailscale up --accept-routes, and /usr/share/omarchy/shell/plugins/panels/tailscale/Model.js line 73 returns ["tailscale", "up"] with no flags. Both are byte-identical on tag v4.0.3, fetched with gh api, so the record is true on the newest release and not only on the 4.0.0 the reporter ran. The flag defaults hold: Tailscale's CLI reference says --accept-dns 'Defaults to accepting DNS settings' and --accept-routes defaults off on every platform except Windows, iOS, Android and the two macOS variants. What the record misses is the Omarchy DNS stack, which is the thing a reader will chase first and waste time on. omarchy-dns writes a NetworkManager global override at /etc/NetworkManager/conf.d/20-omarchy-dns.conf and a global DNS=1.1.1.1 ... into /etc/systemd/resolved.conf, and NetworkManager.conf(5) states that a valid [global-dns-domain-*] block overrides the servers of active connections. It does NOT apply here: NetworkManager never manages tailscale0, and tailscaled pushes resolvers into systemd-resolved as per-link DNS. I confirmed the live shape with resolvectl status, resolvectl dns and resolvectl domain: Global DNS is Cloudflare, every link has zero DNS servers and -DefaultRoute, /etc/resolv.conf is a symlink to ../run/systemd/resolve/stub-resolv.conf, systemd-resolved is enabled and active, and /etc/systemd/resolved.conf.d/10-disable-multicast.conf from omarchy-settings 4.0.2-1 sets LLMNR=no and MulticastDNS=no. Per systemd-resolved.service(8) a query goes to the servers of the best matching routing domain, so once tailscaled sets ~. on tailscale0 the tailnet resolvers take every name and the global Cloudflare servers are never consulted. That is the mechanism behind 'all DNS breaks', and it also means the symptom needs the tailnet to override local DNS, which the record does not say. I also found a real diagnostic trap by reading current_dns_provider() in /usr/share/omarchy/bin/omarchy-dns: it reads only the NM drop-in then /etc/systemd/resolved.conf, so omarchy dns prints Cloudflare while every lookup is going to a dead tailnet resolver. The danger was half right. It correctly warns about route shadowing but says nothing about the cost of the other remedy: with --accept-dns=false there is no DNS on tailscale0, so queries fall to the global Cloudflare servers omarchy-dns wrote, which sends internal names outside the tunnel in clear text. That belongs in a danger. Tailscale is NOT installed here (pacman -Q tailscale fails), so nothing about tailscaled's runtime behaviour was exercised: I did not observe tailscale0, did not run tailscale in any form, and did not reproduce the widget resetting RouteAll. That last point is the issue reporter's observation, attributed as such in the corrected danger rather than asserted. I set corrected_frequency to occasional because common overstates it: the symptom needs a tailnet that overrides DNS, resolvers that live only behind a peer-advertised subnet route, and auth skipped in the installer, and one upstream report is the whole evidence base.

The cause above was rewritten on 2026-09-11 to match this note; the fix was corrected by the audit itself.
RISK

--accept-routes makes this machine honour every subnet route advertised on the tailnet, which can shadow local LAN addresses. On a home network that overlaps a tailnet subnet, this can black-hole your own router.

--accept-dns=false carries a different cost on Omarchy. With no DNS left on tailscale0, every lookup falls back to the global servers omarchy-dns wrote into /etc/systemd/resolved.conf, which are Cloudflare's 1.1.1.1 and 1.0.0.1 on a stock install. Internal names the tailnet admin meant to keep inside the tunnel are then sent to Cloudflare in clear text, and MagicDNS names stop resolving at all.

Either pref can be undone from the bar. The widget's on and off path runs a flagless tailscale up, confirmed in the shipped Model.js, and a reporter on the cited issue traced a silently reset RouteAll=false to exactly that path. That reset was not reproduced here. Re-check tailscale debug prefs after toggling the widget.

Symptom

Toggling Tailscale on breaks **all** DNS, not just MagicDNS. Nothing resolves anywhere while the Tailscale widget still says "Connected". tailscale status --json health messages include:


Tailscale can't reach the configured DNS servers. Internet connectivity may be affected.
Some peers are advertising routes but --accept-routes is false

Cause

On Linux tailscale up defaults to --accept-dns on and --accept-routes off. Tailscale's CLI reference gives --accept-dns as 'Defaults to accepting DNS settings' and --accept-routes as off on every platform except Windows, iOS, Android and the two macOS variants. When the tailnet pushes nameservers that only exist behind a peer-advertised subnet route, the client takes the nameservers and refuses the routes that reach them, so every lookup fails rather than only MagicDNS names. This needs a tailnet that overrides or split-DNS-points at internal resolvers. A tailnet with no DNS override loses only *.ts.net names, not all of them.

Two Omarchy 4 details make it easy to land in. Both were read out of the shipped files on an omarchy 4.0.2-1 machine and are unchanged on tag v4.0.3. The installer is the only path that passes the flag, and that same command is the login gate, so skipping auth there means the pref is never written:


# /usr/share/omarchy/bin/omarchy-install-service-tailscale
sudo tailscale up --accept-routes

The bar widget then logs in with a flagless tailscale up, which applies the Linux defaults on a fresh node:


// /usr/share/omarchy/shell/plugins/panels/tailscale/Model.js, loginPlan()
return { authUrl: "", command: ["tailscale", "up"] }

The widget parses tailscale status --json and ignores Health, so it keeps reporting Connected.

Omarchy's own DNS override is not the cause, and it is worth ruling out before chasing it. omarchy-dns writes a NetworkManager global override at /etc/NetworkManager/conf.d/20-omarchy-dns.conf and a global DNS= line in /etc/systemd/resolved.conf, and NetworkManager.conf(5) says a valid [global-dns-domain-*] block overrides the servers of active connections. NetworkManager never manages tailscale0, and tailscaled pushes its resolvers straight into systemd-resolved as per-link DNS on that interface, so the override does not reach them. When the tailnet overrides local DNS, tailscaled also gives tailscale0 the route-only domain ~., and per systemd-resolved.service(8) a query is sent to the DNS servers of the best matching routing domain. ~. matches everything, so the tailnet resolvers win over the global Cloudflare servers and take every name to an address the machine cannot reach.

Fix

First establish that this is the tailnet resolver and not Omarchy's own DNS setting. omarchy dns reads only 20-omarchy-dns.conf and /etc/systemd/resolved.conf, so it reports Cloudflare and tells you nothing about tailscale0:


resolvectl status tailscale0
resolvectl domain                 # a ~. on tailscale0 means every lookup goes to the tailnet
tailscale status --json | jq '.Health'
tailscale debug prefs | grep -E 'CorpDNS|RouteAll'

Then pick one. Accept the routes that make the pushed resolvers reachable:


sudo tailscale set --accept-routes=true

Or stop using tailnet DNS and keep the local resolvers:


sudo tailscale set --accept-dns=false
sudo resolvectl flush-caches

sudo is unnecessary once the installer's sudo tailscale set --operator="$USER" has applied.

Do not reach for omarchy dns to fix this. It rewrites /etc/systemd/resolved.conf wholesale and reloads the DNS stack without touching the per-link configuration on tailscale0, which is where the broken resolvers are.

Verify

tailscale debug prefs shows RouteAll: true, or CorpDNS: false if you took the other branch. tailscale status --json | jq '.Health' is empty or null, and resolvectl query archlinux.org succeeds with Tailscale connected. After --accept-dns=false, resolvectl domain no longer shows ~. on tailscale0 and resolvectl status tailscale0 lists no DNS servers.

Sources

tailscale-accept-dns-breaks-all-dns