| 1 | //go:build linux && rootrunner |
| 2 | |
| 3 | package host |
| 4 | |
| 5 | import ( |
| 6 | "context" |
| 7 | "encoding/json" |
| 8 | "errors" |
| 9 | "net" |
| 10 | "os" |
| 11 | "os/exec" |
| 12 | "path/filepath" |
| 13 | "strings" |
| 14 | "testing" |
| 15 | "time" |
| 16 | |
| 17 | "github.com/carlosframework/platform/internal/rootstate" |
| 18 | "golang.org/x/sys/unix" |
| 19 | ) |
| 20 | |
| 21 | const unitMigrationRebootDir = "/var/lib/carlos-e52-migration" |
| 22 | const migrationRebootUnit = "plan14-e52-migration.service" |
| 23 | const migrationRebootSocket = "/run/carlos/plan14-e52.sock" |
| 24 | |
| 25 | func migrationRebootMaterial(t *testing.T) claimUnitMaterializer { |
| 26 | t.Helper() |
| 27 | store, err := rootstate.Open(filepath.Join(unitMigrationRebootDir, "state"), true, false) |
| 28 | if err != nil { |
| 29 | t.Fatal(err) |
| 30 | } |
| 31 | t.Cleanup(func() { store.Close() }) |
| 32 | files := claimUnitFiles{dir: "/etc/systemd/system"} |
| 33 | k := claimPredecessorKernel{ctl: rebootPredecessorCtl, group: claimPredecessorGroup, boot: claimUnitBoot} |
| 34 | return claimUnitMaterializer{state: store, read: files.read, create: files.create, replace: files.replace, predecessor: k.capture, stopPredecessor: k.stop, |
| 35 | socketCapture: captureProductionClaimUnitSocket, socketPark: parkProductionClaimUnitSocket, idle: func(string) error { return nil }} |
| 36 | } |
| 37 | |
| 38 | func TestRootRebootClaimUnitMigrationPhase1(t *testing.T) { |
| 39 | for _, dir := range []string{unitMigrationRebootDir, "/run/carlos", "/run/carlos-control"} { |
| 40 | if err := os.Mkdir(dir, 0755); err != nil { |
| 41 | t.Fatal(err) |
| 42 | } |
| 43 | } |
| 44 | binary, err := os.Executable() |
| 45 | if err != nil { |
| 46 | t.Fatal(err) |
| 47 | } |
| 48 | body, err := os.ReadFile(binary) |
| 49 | if err != nil { |
| 50 | t.Fatal(err) |
| 51 | } |
| 52 | if err := os.WriteFile(filepath.Join(unitMigrationRebootDir, "payload"), body, 0755); err != nil { |
| 53 | t.Fatal(err) |
| 54 | } |
| 55 | legacy := []byte("[Service]\nType=simple\nEnvironment=PLAN14_TENANT_SOCKET=" + migrationRebootSocket + "\nExecStart=" + unitMigrationRebootDir + "/payload plan14-tenant\n") |
| 56 | path := "/etc/systemd/system/" + migrationRebootUnit |
| 57 | f, err := os.OpenFile(path, os.O_CREATE|os.O_EXCL|os.O_WRONLY, 0644) |
| 58 | if err != nil { |
| 59 | t.Fatal(err) |
| 60 | } |
| 61 | _, err = f.Write(legacy) |
| 62 | if err = errors.Join(err, f.Close()); err != nil { |
| 63 | t.Fatal(err) |
| 64 | } |
| 65 | if out, err := rebootPredecessorCtl(t.Context(), "daemon-reload"); err != nil { |
| 66 | t.Fatal(out, err) |
| 67 | } |
| 68 | if out, err := rebootPredecessorCtl(t.Context(), "start", migrationRebootUnit); err != nil { |
| 69 | t.Fatal(out, err) |
| 70 | } |
| 71 | for until := time.Now().Add(5 * time.Second); ; { |
| 72 | conn, err := net.DialTimeout("unix", migrationRebootSocket, time.Second) |
| 73 | if err == nil { |
| 74 | conn.Close() |
| 75 | break |
| 76 | } |
| 77 | if time.Now().After(until) { |
| 78 | t.Fatal(err) |
| 79 | } |
| 80 | time.Sleep(20 * time.Millisecond) |
| 81 | } |
| 82 | _, in, s := plan14UnitBindingFixture(t, migrationRebootUnit, "", migrationRebootSocket) |
| 83 | s.Program, s.PredecessorDigest = strings.Repeat("a", 64), claimUnitDigest(legacy) |
| 84 | b := claimUnitBinding{Schema: 1, Owner: unitBindingOwner(in), Shape: s} |
| 85 | m := migrationRebootMaterial(t) |
| 86 | m.stopPredecessor = func(context.Context, claimUnitPredecessor) error { return os.ErrClosed } |
| 87 | if err := m.install(t.Context(), b); !errors.Is(err, os.ErrClosed) { |
| 88 | t.Fatal("migration intent boundary", err) |
| 89 | } |
| 90 | r, err := m.load(b.Owner.Host) |
| 91 | if err != nil || r.Migration == nil || r.Migration.Socket.Boot == "" || r.Identity != (rootstate.FileIdentity{}) { |
| 92 | t.Fatal("lost durable migration", r, err) |
| 93 | } |
| 94 | raw, _ := json.Marshal(b) |
| 95 | if err := os.WriteFile(filepath.Join(unitMigrationRebootDir, "binding.json"), raw, 0600); err != nil { |
| 96 | t.Fatal(err) |
| 97 | } |
| 98 | old := claimEdgeMaskOwner |
| 99 | defer func() { claimEdgeMaskOwner = old }() |
| 100 | claimEdgeMaskOwner = filepath.Join(unitMigrationRebootDir, "edge-mask") |
| 101 | if _, err := createClaimEdgeMask(t.Context(), r.Migration.Socket.Boot); err != nil { |
| 102 | t.Fatal(err) |
| 103 | } |
| 104 | t.Logf("migration saved boot=%s device=%d mount=%d socket=%d:%d", r.Migration.Definition.Boot, r.Migration.Definition.Device, r.Migration.Definition.Mount, r.Migration.Socket.Device, r.Migration.Socket.Inode) |
| 105 | } |
| 106 | |
| 107 | func TestRootRebootClaimUnitMigrationPhase2(t *testing.T) { |
| 108 | t.Cleanup(func() { |
| 109 | rebootPredecessorCtl(context.Background(), "stop", migrationRebootUnit) |
| 110 | os.Remove("/etc/systemd/system/" + migrationRebootUnit) |
| 111 | os.Remove("/run/systemd/system/carlos-edge.service") |
| 112 | rebootPredecessorCtl(context.Background(), "daemon-reload") |
| 113 | for _, dir := range []string{unitMigrationRebootDir, "/run/carlos", "/run/carlos-control"} { |
| 114 | os.RemoveAll(dir) |
| 115 | } |
| 116 | }) |
| 117 | for _, dir := range []string{"/run/carlos", "/run/carlos-control"} { |
| 118 | if err := os.Mkdir(dir, 0755); err != nil { |
| 119 | t.Fatal(err) |
| 120 | } |
| 121 | } |
| 122 | var b claimUnitBinding |
| 123 | raw, err := os.ReadFile(filepath.Join(unitMigrationRebootDir, "binding.json")) |
| 124 | if err != nil || json.Unmarshal(raw, &b) != nil { |
| 125 | t.Fatal(err) |
| 126 | } |
| 127 | m := migrationRebootMaterial(t) |
| 128 | r, err := m.load(b.Owner.Host) |
| 129 | if err != nil { |
| 130 | t.Fatal(err) |
| 131 | } |
| 132 | boot, err := claimUnitBoot() |
| 133 | if err != nil || boot == r.Migration.Definition.Boot { |
| 134 | t.Fatal("actual reboot required", err) |
| 135 | } |
| 136 | oldMaskOwner := claimEdgeMaskOwner |
| 137 | defer func() { claimEdgeMaskOwner = oldMaskOwner }() |
| 138 | claimEdgeMaskOwner = filepath.Join(unitMigrationRebootDir, "edge-mask") |
| 139 | if err := os.Symlink("/dev/null", "/run/systemd/system/carlos-edge.service"); err != nil { |
| 140 | t.Fatal(err) |
| 141 | } |
| 142 | if out, err := rebootPredecessorCtl(t.Context(), "daemon-reload"); err != nil { |
| 143 | t.Fatal(out, err) |
| 144 | } |
| 145 | if err := recoverLegacyClaimEdgeMask(t.Context()); err != nil { |
| 146 | t.Fatal(err) |
| 147 | } |
| 148 | if target, err := os.Readlink("/run/systemd/system/carlos-edge.service"); err != nil || target != "/dev/null" { |
| 149 | t.Fatal("old boot removed replacement mask", err) |
| 150 | } |
| 151 | if _, err := os.Lstat(claimEdgeMaskOwner); !os.IsNotExist(err) { |
| 152 | t.Fatal("old mask record retained", err) |
| 153 | } |
| 154 | path := "/etc/systemd/system/" + migrationRebootUnit |
| 155 | if err := unix.Mount(path, path, "", unix.MS_BIND, ""); err != nil { |
| 156 | t.Fatal(err) |
| 157 | } |
| 158 | refused := m.install(t.Context(), b) |
| 159 | if err := unix.Unmount(path, 0); err != nil { |
| 160 | t.Fatal(err) |
| 161 | } |
| 162 | if refused == nil { |
| 163 | t.Fatal("substituted definition mount accepted") |
| 164 | } |
| 165 | listener, err := net.ListenUnix("unix", &net.UnixAddr{Name: migrationRebootSocket, Net: "unix"}) |
| 166 | if err != nil { |
| 167 | t.Fatal(err) |
| 168 | } |
| 169 | if err = m.install(t.Context(), b); err == nil { |
| 170 | t.Fatal("post-boot replacement socket removed") |
| 171 | } |
| 172 | if _, err := os.Lstat(migrationRebootSocket); err != nil { |
| 173 | t.Fatal("replacement socket lost", err) |
| 174 | } |
| 175 | listener.Close() |
| 176 | if err := m.install(t.Context(), b); err != nil { |
| 177 | t.Fatal("resume migration after reboot", err) |
| 178 | } |
| 179 | if err := m.verify(b); err != nil { |
| 180 | t.Fatal(err) |
| 181 | } |
| 182 | if err := projectClaimBoundSocket(b.Shape); err != nil { |
| 183 | t.Fatal(err) |
| 184 | } |
| 185 | if target, err := os.Readlink(migrationRebootSocket); err != nil || target != claimSocketTarget(b.Shape) { |
| 186 | t.Fatal("socket projection", target, err) |
| 187 | } |
| 188 | if out, err := exec.Command("systemctl", "daemon-reload").CombinedOutput(); err != nil { |
| 189 | t.Fatal(string(out), err) |
| 190 | } |
| 191 | if _, err := rebootPredecessorCtl(t.Context(), "start", migrationRebootUnit); err == nil { |
| 192 | time.Sleep(100 * time.Millisecond) |
| 193 | pid, _ := rebootPredecessorCtl(t.Context(), "show", "-p", "MainPID", "--value", migrationRebootUnit) |
| 194 | if pid != "0" { |
| 195 | t.Fatal("migration started without reservation", pid) |
| 196 | } |
| 197 | } |
| 198 | } |
| 199 | |
amadan