UpdateServerConfig was at 31; SonarCloud threshold is 20. The deep
nesting came from wrapping the whole body in the mgmtCacheResolver
nil-check and stacking five for-loops + conditionals inside it.
Invert to an early return when the resolver is nil, then split the
remaining work into two focused helpers:
- registerMgmtCacheHandlers: handler-chain wiring (pool-root
subdomain wrapper + exact-match registration) with the toZone-
canonicalized membership check.
- reconcileMgmtPoolRoots: extraDomains refcounting against the
current pool-root set, plus the mgmtPoolRoots tracking map.
Pure refactor — behaviour is identical.
resolveOnDemand deduplicated the cache write but not the upstream
lookup — a burst of concurrent queries for a freshly-learned pool-root
subdomain (A + AAAA from a single getaddrinfo, or multiple peer
workers dialing the same foreign relay) each hit the bypass resolver
independently.
Wrap the lookupRecords call in refreshGroup.Do with an 'ondemand:'-
prefixed key so concurrent callers collapse into one request; the
prefix namespaces this key off scheduleRefresh's keyspace so the two
paths don't collide.
Error and empty-result paths preserved; the cache write still runs
under the existing TOCTOU-safe exists check + mutex so a winning
refresh isn't clobbered.
GetPoolRootDomains returns what the extractor stored; GetCachedDomains
strips the trailing dot from question names. Today those happen to
produce the same domain.Domain string because both sources run through
domain.FromString, but relying on that parity is fragile.
Run both sides of the poolRootSet membership check through toZone so
the comparison is independent of each source's canonical form.
Prevents a future extractor or cache-entry path from silently routing
a pool-root into exactDomains, where a second registerHandler call
with the bare resolver would clobber the subdomainMatchHandler
registration (AddHandler replaces entries with the same pattern +
priority).
The subdomain-matching handler chain entry only helps when the OS DNS
manager routes the query to the daemon's listener. On systems using
systemd-resolved (Ubuntu 20.04 / 22.04, most modern Debian/RHEL-family
distros), the daemon delegates a *specific set* of match domains to
the wt0 link via D-Bus. Any FQDN outside that set is answered from
the host's global upstream DNS, never touching the handler chain.
With only 'netbird.cloud' and the in-addr.arpa reverse zone advertised
as match domains, a relay instance FQDN like
streamline-de-fra1-0.relay.netbird.io is resolved globally. When an
exit-node default route is active on wt0 but the peer is not yet
live, that global lookup times out on the local stub (127.0.0.53)
because systemd-resolved's uplink DNS is unreachable through the
overlay, and the foreign-relay dial fails exactly the way the
ENOKEY path did on file-manager hosts.
Bump the pool-root domains (ServerDomains.Relay entries) into
extraDomains so applyHostConfig adds them as match-only domains
alongside 'netbird.cloud'. systemd-resolved then delegates
*.relay.netbird.io to the wt0 link, queries hit the daemon's DNS
listener, and the on-demand resolve path via the bypass resolver
runs as intended.
Tracking map mgmtPoolRoots isolates this refcount from the
RegisterHandler path so updates across successive mgmt syncs
increment/decrement only the changed set.
When an exit-node peer's network-map installs a 0.0.0.0/0 default route
on the overlay interface before that peer's WireGuard key material is
active, any UDP socket dialing an off-link address is routed into wt0
and the kernel returns ENOKEY.
Two places needed fixing:
1. The mgmt cache refresh path. It reactively refreshes the
control-plane FQDNs advertised by the mgmt (api/signal/stun/turn/
the Relay pool root) after the daemon has installed its own
resolv.conf pointing at the overlay listener. Previously the
refresh dial followed the chain's upstream handler, which followed
the overlay default route and deadlocked on ENOKEY.
2. Foreign relay FQDN resolution. When a remote peer is homed on a
different relay instance than us, we need to resolve a streamline-*
subdomain that is not in the cache. That lookup went through the
same overlay-routed upstream and failed identically, deadlocking
the exit-node test whenever the relay LB put the two peers on
different instances.
Fix both by giving the mgmt cache a dedicated net.Resolver that dials
the original pre-NetBird system nameservers through nbnet.NewDialer.
The dialer marks the socket as control-plane (SO_MARK on Linux,
IP_BOUND_IF on darwin, IP_UNICAST_IF on Windows); the routemanager's
policy rules keep those sockets on the underlay regardless of the
overlay default.
Pool-root domains (the Relay entries in ServerDomains) now register
through a subdomain-matching wrapper so that instance subdomains like
streamline-de-fra1-0.relay.netbird.io also hit the mgmt cache handler.
On cache miss under a pool root, ServeDNS resolves the FQDN on demand
through the bypass resolver, caches the result, and returns it.
Pool-root membership is derived dynamically from mgmt-advertised
ServerDomains.Relay[] — no hardcoded domain lists, no protocol change.
No hardcoded fallback nameservers: if the host had no original system
resolver at all, the bypass resolver stays nil and the stale-while-
revalidate cache keeps serving. The general upstream forwarder and
the user DNS path are unchanged.
- Add GetSelectedClientRoutes() to the route manager that filters through FilterSelectedExitNodes, returning only active routes instead of all management routes
- Use GetSelectedClientRoutes() in the DNS route checker so deselected exit nodes' 0.0.0.0/0 no longer matches upstream DNS IPs — this prevented the resolver from switching
away from the utun-bound socket after exit node deselection
- Initialize iOS DNS server with host DNS fallback addresses (1.1.1.1:53, 1.0.0.1:53) and a permanent root zone handler, matching Android's behavior — without this, unmatched
DNS queries arriving via the 0.0.0.0/0 tunnel route had no handler and were silently dropped
Replace fmt.Sprintf("%s:%d", ip, port) with net.JoinHostPort() to
properly handle IPv6 addresses that need bracket wrapping (e.g.,
[2606:4700:4700::1111]:53 instead of 2606:4700:4700::1111:53).
Without this fix, configuring IPv6 nameservers causes "too many colons
in address" errors because Go's net.Dial cannot parse the malformed
address string.
Fixes#5601
Related to #4074
Co-authored-by: easonysliu <easonysliu@tencent.com>
* Fix DNS probe thread safety and avoid blocking engine sync
Refactor ProbeAvailability to prevent blocking the engine's sync mutex
during slow DNS probes. The probe now derives its context from the
server's own context (s.ctx) instead of accepting one from the caller,
and uses a mutex to ensure only one probe runs at a time — new calls
cancel the previous probe before starting. Also fixes a data race in
Stop() when accessing probeCancel without the probe mutex.
* Ensure DNS probe thread safety by locking critical sections
Add proper locking to prevent data races when accessing shared resources during DNS probe execution and Stop(). Update handlers snapshot logic to avoid conflicts with concurrent writers.
* Rename context and remove redundant cancellation
* Cancel first and lock
* Add locking to ensure thread safety when reactivating upstream servers
* [client] Batch macOS DNS domains across multiple scutil keys to avoid truncation
scutil has undocumented limits: 99-element cap on d.add arrays and ~2048
byte value buffer for SupplementalMatchDomains. Users with 60+ domains
hit silent domain loss. This applies the same batching approach used on
Windows (nrptMaxDomainsPerRule=50), splitting domains into indexed
resolver keys (NetBird-Match-0, NetBird-Match-1, etc.) with 50-element
and 1500-byte limits per key.
* check for all keys on getRemovableKeysWithDefaults
* use multi error
* Optimize Windows DNS performance with domain batching and batch mode
Implement two-layer optimization to reduce Windows NRPT registry operations:
1. Domain Batching (host_windows.go):
- Batch domains per NRPT
- Reduces NRPT rules by ~97% (e.g., 184 domains: 184 rules → 4 rules)
- Modified addDNSMatchPolicy() to create batched NRPT entries
- Added comprehensive tests in host_windows_test.go
2. Batch Mode (server.go):
- Added BeginBatch/EndBatch methods to defer DNS updates
- Modified RegisterHandler/DeregisterHandler to skip applyHostConfig in batch mode
- Protected all applyHostConfig() calls with batch mode checks
- Updated route manager to wrap route operations with batch calls
* Update tests
* Fix log line
* Fix NRPT rule index to ensure cleanup covers partially created rules
* Ensure NRPT entry count updates even on errors to improve cleanup reliability
* Switch DNS batch mode logging from Info to Debug level
* Fix batch mode to not suppress critical DNS config updates
Batch mode should only defer applyHostConfig() for RegisterHandler/
DeregisterHandler operations. Management updates and upstream nameserver
failures (deactivate/reactivate callbacks) need immediate DNS config
updates regardless of batch mode to ensure timely failover.
Without this fix, if a nameserver goes down during a route update,
the system DNS config won't be updated until EndBatch(), potentially
delaying failover by several seconds.
Or if you prefer a shorter version:
Fix batch mode to allow immediate DNS updates for critical paths
Batch mode now only affects RegisterHandler/DeregisterHandler.
Management updates and nameserver failures always trigger immediate
DNS config updates to ensure timely failover.
* Add DNS batch cancellation to rollback partial changes on errors
Introduces CancelBatch() method to the DNS server interface to handle error
scenarios during batch operations. When route updates fail partway through, the DNS
server can now discard accumulated changes instead of applying partial state. This
prevents leaving the DNS configuration in an inconsistent state when route manager
operations encounter errors.
The changes add error-aware batch handling to prevent partial DNS configuration
updates when route operations fail, which improves system reliability.
Per RFC 4592 section 2.2.1, wildcards should only match when the queried
name does not exist in the zone. Previously, if host.example.com had an
A record and *.example.com had an AAAA record, querying AAAA for
host.example.com would incorrectly return the wildcard AAAA instead of
NODATA.
Now the resolver checks if the domain exists (with any record type)
before falling back to wildcard matching, returning proper NODATA
responses for existing names without the requested record type.
* **New Features**
* Wildcard DNS fallback for eligible query types (excluding NS/SOA): attempts wildcard records when no exact match, rewrites wildcard names back to the original query, and rotates responses; preserves CNAME resolution.
* **Tests**
* Vastly expanded coverage for wildcard behaviors, precedence, multi-record round‑robin, multi-type chains, multi-hop and cross-zone scenarios, and edge cases (NXDOMAIN/NODATA, fallthrough).
* **Chores**
* CI lint config updated to ignore an additional codespell entry.
Adds a per-instance uint64 hash to DefaultServer to detect identical merged host DNS configs (including extra domains). applyHostConfig computes and compares the hash, skips applying if unchanged, treats hash errors as a fail-safe (proceed to apply), and updates the stored hash only after successful hashing and apply.
in some cases iOS and macOS may be locked when looking for management domains during network changes
This change introduce an additional timeout on top of the context call
Two new boolean flags, SearchDomainDisabled and SkipPTRProcess, are added to CustomZone and its protobuf; they are propagated through the engine to DNS host logic. Host matching now uses SearchDomainDisabled directly, and PTR collection skips zones with SkipPTRProcess; reverse zones are initialized with SearchDomainDisabled: true.
Upgrade Go toolchain and golang.org/x/* deps to 1.24.10, standardize GitHub Actions to derive Go version from go.mod and adjust checkout ordering, raise WASM size limit to 55 MB, update FreeBSD tarball and gomobile refs, fix a few format-string/logging calls, treat usernames ending with $ as system accounts, and add Windows tests.
- Move `util/grpc` and `util/net` to `client` so `internal` packages can be accessed
- Add methods to return the next best interface after the NetBird interface.
- Use `IP_UNICAST_IF` sock opt to force the outgoing interface for the NetBird `net.Dialer` and `net.ListenerConfig` to avoid routing loops. The interface is picked by the new route lookup method.
- Some refactoring to avoid import cycles
- Old behavior is available through `NB_USE_LEGACY_ROUTING=true` env var
Zoltán Papp