mapper: skip peers with invalid names instead of failing the map

A peer whose GivenName fails GetFQDN aborted the whole map for every node
that could see it. Drop and log it; SSH policy errors degrade too.

Fixes #3346
This commit is contained in:
Kristoffer Dalby
2026-06-26 14:19:19 +00:00
parent d528686f14
commit 08956d51a4
2 changed files with 113 additions and 2 deletions
+24 -2
View File
@@ -9,6 +9,8 @@ import (
"github.com/juanfont/headscale/hscontrol/policy"
policyv2 "github.com/juanfont/headscale/hscontrol/policy/v2"
"github.com/juanfont/headscale/hscontrol/types"
"github.com/juanfont/headscale/hscontrol/util/zlog/zf"
"github.com/rs/zerolog/log"
"tailscale.com/tailcfg"
"tailscale.com/types/views"
"tailscale.com/util/multierr"
@@ -152,7 +154,15 @@ func (b *MapResponseBuilder) WithSSHPolicy() *MapResponseBuilder {
sshPolicy, err := b.mapper.state.SSHPolicy(node)
if err != nil {
b.addError(err)
// SSH policy is optional for a node to function. Rather than fail the
// whole map (leaving the node unable to connect), log and continue
// without it; the node still receives a usable netmap.
log.Warn().Caller().
Err(err).
Uint64(zf.NodeID, node.ID().Uint64()).
Str(zf.NodeHostname, node.Hostname()).
Msg("building map response: skipping SSH policy for node; node will receive a map without SSH rules")
return b
}
@@ -284,7 +294,19 @@ func (b *MapResponseBuilder) buildTailPeers(peers views.Slice[types.NodeView]) (
return b.mapper.state.RoutesForPeer(node, peer, matchers)
}, b.mapper.cfg, allCapMaps[peer.ID()])
if err != nil {
return nil, err
// One peer with invalid data (e.g. an empty or over-long
// GivenName that fails GetFQDN) must not blank out the map for
// every node that can see it. Drop the offending peer, log it
// with the identity an operator needs to fix it, and keep
// building from the remaining valid peers.
log.Warn().Caller().
Err(err).
Uint64(zf.NodeID, peer.ID().Uint64()).
Str(zf.NodeHostname, peer.Hostname()).
Uint64("map.viewer.node.id", b.nodeID.Uint64()).
Msgf("dropping peer %d from map response: invalid node data; fix with `headscale nodes rename %d <name>`", peer.ID(), peer.ID())
continue
}
// [tailcfg.Node.CapMap] on a peer carries the small set of
+89
View File
@@ -3,6 +3,7 @@ package mapper
import (
"fmt"
"net/netip"
"strings"
"testing"
"github.com/google/go-cmp/cmp"
@@ -536,6 +537,94 @@ func TestBuildFromChangeVisibilityMatchesFullMap(t *testing.T) {
}
}
// TestFullMapResponseSurvivesPeerWithInvalidName proves a single node with an
// FQDN-invalid GivenName must not break map generation for its peers.
//
// A node whose stored GivenName is empty (ErrNodeHasNoGivenName) or yields an
// FQDN longer than MaxHostnameLength (ErrHostnameTooLong) makes GetFQDN, and
// therefore TailNode, return an error. buildTailPeers used to abort the entire
// peer list on the first such error, so MapResponseBuilder.Build() failed for
// every node that could see the bad peer; on the initial-connection path that
// surfaced as "PollNetMap: ... unexpected EOF" and the "Unable to connect to
// the Tailscale coordination server" health warning. A legacy DB row loads
// verbatim (NewNodeStore reads db.ListNodes() without re-sanitising names), so
// the bad peer persists across restart. The build for an unaffected viewer
// must succeed: the bad peer is dropped, valid peers and self survive.
func TestFullMapResponseSurvivesPeerWithInvalidName(t *testing.T) {
for _, tt := range []struct {
name string
badName string
}{
{"empty given name", ""},
{"over-long fqdn", strings.Repeat("a", types.MaxHostnameLength+1)},
} {
t.Run(tt.name, func(t *testing.T) {
tmp := t.TempDir()
p4 := netip.MustParsePrefix("100.64.0.0/10")
p6 := netip.MustParsePrefix("fd7a:115c:a1e0::/48")
cfg := &types.Config{
Database: types.DatabaseConfig{
Type: types.DatabaseSqlite,
Sqlite: types.SqliteConfig{Path: tmp + "/h.db"},
},
PrefixV4: &p4,
PrefixV6: &p6,
IPAllocation: types.IPAllocationStrategySequential,
BaseDomain: "headscale.test",
Policy: types.PolicyConfig{Mode: types.PolicyModeDB},
DERP: types.DERPConfig{
DERPMap: &tailcfg.DERPMap{
Regions: map[int]*tailcfg.DERPRegion{999: {RegionID: 999}},
},
},
Tuning: types.Tuning{
NodeStoreBatchSize: state.TestBatchSize,
NodeStoreBatchTimeout: state.TestBatchTimeout,
},
}
database, err := db.NewHeadscaleDatabase(cfg)
require.NoError(t, err)
user := database.CreateUserForTest("u1")
n1 := database.CreateRegisteredNodeForTest(user, "n1") // viewer, valid
bad := database.CreateRegisteredNodeForTest(user, "bad") // peer, name corrupted below
good := database.CreateRegisteredNodeForTest(user, "good") // peer, valid control
// Simulate a legacy/corrupt row that v29 loads verbatim.
require.NoError(t, database.DB.
Model(&types.Node{}).
Where("id = ?", bad.ID).
Update("given_name", tt.badName).Error)
require.NoError(t, database.Close())
s, err := state.NewState(cfg)
require.NoError(t, err)
t.Cleanup(func() { _ = s.Close() })
// Allow-all so n1 sees both peers; the bad one must still be dropped.
_, err = s.SetPolicy([]byte(`{"acls":[{"action":"accept","src":["*"],"dst":["*:*"]}]}`))
require.NoError(t, err)
m := &mapper{state: s, cfg: cfg}
capVer := tailcfg.CurrentCapabilityVersion
resp, err := m.fullMapResponse(n1.ID, capVer)
require.NoError(t, err, "n1's map must build despite a peer with an invalid name")
require.NotNil(t, resp)
require.NotNil(t, resp.Node, "n1 must receive its own self node")
peers := map[tailcfg.NodeID]bool{}
for _, p := range resp.Peers {
peers[p.ID] = true
}
assert.False(t, peers[bad.ID.NodeID()], "the peer with an invalid name must be dropped")
assert.True(t, peers[good.ID.NodeID()], "valid peers must remain in the map")
})
}
}
// TestGenerateDNSConfigNilHostinfoNoPanic proves generateDNSConfig does not
// panic when a node's Hostinfo is nil (e.g. a legacy DB row with a NULL
// host_info column). addNextDNSMetadata dereferenced node.Hostinfo().OS()