| 1 | //go:build unix |
| 2 | |
| 3 | package hostguard |
| 4 | |
| 5 | import ( |
| 6 | "context" |
| 7 | "errors" |
| 8 | "testing" |
| 9 | "time" |
| 10 | |
| 11 | "github.com/carlosframework/platform/internal/hostclaim" |
| 12 | ) |
| 13 | |
| 14 | func fenceFixture(t *testing.T) (*guardFixture, FenceTarget, Captured) { |
| 15 | t.Helper() |
| 16 | f := newGuardFixture(t) |
| 17 | f.obs.Claim.Holder.Session = 7 |
| 18 | f.obs.Claim.Op.Kind = hostclaim.KindLanding |
| 19 | f.obs.Claim.Op.ID = "operation-1" |
| 20 | f.obs.Claim.Op.Phase = "probing" |
| 21 | if err := f.g.Activate(t.Context(), f.obs.Host, "uk", nil); err != nil { |
| 22 | t.Fatal(err) |
| 23 | } |
| 24 | captured, err := Capture(f.g.DataDir, f.obs.Host) |
| 25 | if err != nil { |
| 26 | t.Fatal(err) |
| 27 | } |
| 28 | target := FenceTarget{FenceID: "fence-1", Host: f.obs.Host, Incarnation: "inc", Stamp: 1, Session: 7, |
| 29 | Op: "operation-1", Phase: "probing", Box: "uk", Enrollment: "enrollment-1", |
| 30 | BootID: "boot-1", ControllerID: "controller-1", ClaimETag: "one", DirectoryRevision: "directory-9"} |
| 31 | return f, target, captured |
| 32 | } |
| 33 | |
| 34 | func TestBeginFenceDeniesOnlyExactCurrentAuthority(t *testing.T) { |
| 35 | f, target, captured := fenceFixture(t) |
| 36 | bad := target |
| 37 | bad.ClaimETag = "stale" |
| 38 | if _, err := f.g.BeginFence(t.Context(), bad); !errors.Is(err, ErrFenceTarget) { |
| 39 | t.Fatalf("stale begin = %v", err) |
| 40 | } |
| 41 | if _, err := f.g.BeginFence(t.Context(), target); err != nil { |
| 42 | t.Fatal(err) |
| 43 | } |
| 44 | if _, err := f.g.BeginFence(t.Context(), target); err != nil { // exact replay |
| 45 | t.Fatal(err) |
| 46 | } |
| 47 | retry := target |
| 48 | retry.FenceID = "new-id-must-not-replace-durable-id" |
| 49 | got, err := f.g.BeginFence(t.Context(), retry) |
| 50 | if err != nil || got != target { |
| 51 | t.Fatalf("begin retry = %+v, %v; want original %+v", got, err, target) |
| 52 | } |
| 53 | called := false |
| 54 | err = f.g.With(t.Context(), target.Host, captured, allow, func() error { called = true; return nil }) |
| 55 | if !errors.Is(err, ErrFenced) || called { |
| 56 | t.Fatalf("guard under fence = %v, called=%v", err, called) |
| 57 | } |
| 58 | f.obs.Claim.Op.Phase = "next-phase" |
| 59 | if err := f.g.With(t.Context(), target.Host, captured, allow, func() error { called = true; return nil }); !errors.Is(err, ErrFenced) { |
| 60 | t.Fatalf("phase change bypassed fence: %v", err) |
| 61 | } |
| 62 | |
| 63 | // A later writer session is a different authority tuple. The old deny |
| 64 | // remains durable for cleanup but cannot deny the successor session. |
| 65 | f.obs.Claim.Holder.Session = 8 |
| 66 | if err := f.g.With(t.Context(), target.Host, captured, allow, func() error { called = true; return nil }); err != nil || !called { |
| 67 | t.Fatalf("new session guard = %v, called=%v", err, called) |
| 68 | } |
| 69 | } |
| 70 | |
| 71 | func TestBeginFenceCannotOverwriteUnsettledTarget(t *testing.T) { |
| 72 | f, first, _ := fenceFixture(t) |
| 73 | if _, err := f.g.BeginFence(t.Context(), first); err != nil { |
| 74 | t.Fatal(err) |
| 75 | } |
| 76 | second := first |
| 77 | second.FenceID = "fence-2" |
| 78 | second.Phase = "next-phase" |
| 79 | second.ClaimETag = "two" |
| 80 | second.DirectoryRevision = "directory-10" |
| 81 | f.obs.ETag = "two" |
| 82 | f.obs.Claim.Op.Phase = "next-phase" |
| 83 | if _, err := f.g.BeginFence(t.Context(), second); !errors.Is(err, ErrFenceActive) { |
| 84 | t.Fatalf("unsettled overwrite = %v", err) |
| 85 | } |
| 86 | at := time.Date(2026, 10, 6, 14, 30, 0, 0, time.UTC) |
| 87 | if err := f.g.CompleteFence(t.Context(), first, []byte("first receipt"), at, func(context.Context, FenceTarget, []byte, time.Time) error { return nil }); err != nil { |
| 88 | t.Fatal(err) |
| 89 | } |
| 90 | if _, err := f.g.BeginFence(t.Context(), second); err != nil { |
| 91 | t.Fatalf("completed predecessor blocked successor fence: %v", err) |
| 92 | } |
| 93 | if err := f.g.CleanupWith(t.Context(), first, func(context.Context, FenceReceipt) error { return nil }, func() error { return nil }); err != nil { |
| 94 | t.Fatal(err) |
| 95 | } |
| 96 | r, err := readRecord(f.g.DataDir, first.Host) |
| 97 | if err != nil || r.ActiveFence == nil || *r.ActiveFence != second || len(r.FenceJournal) != 1 { |
| 98 | t.Fatalf("predecessor cleanup damaged successor fence: %+v, %v", r, err) |
| 99 | } |
| 100 | } |
| 101 | |
| 102 | func TestCompleteFenceVerificationFailureAndImmutableReplay(t *testing.T) { |
| 103 | f, target, captured := fenceFixture(t) |
| 104 | if _, err := f.g.BeginFence(t.Context(), target); err != nil { |
| 105 | t.Fatal(err) |
| 106 | } |
| 107 | failure := errors.New("root rejected evidence") |
| 108 | at := time.Date(2026, 10, 6, 14, 0, 0, 0, time.UTC) |
| 109 | if err := f.g.CompleteFence(t.Context(), target, []byte("receipt"), at, func(context.Context, FenceTarget, []byte, time.Time) error { return failure }); !errors.Is(err, failure) { |
| 110 | t.Fatalf("failed verification = %v", err) |
| 111 | } |
| 112 | r, err := readRecord(f.g.DataDir, target.Host) |
| 113 | if err != nil || len(r.FenceJournal) != 0 || r.ActiveFence == nil { |
| 114 | t.Fatalf("failed verification persisted evidence: %+v, %v", r, err) |
| 115 | } |
| 116 | verify := func(ctx context.Context, got FenceTarget, _ []byte, _ time.Time) error { |
| 117 | // Verification is outside the flock: reacquiring it completes rather |
| 118 | // than deadlocking. The installed deny is visible to the nested guard. |
| 119 | return f.g.With(ctx, got.Host, captured, allow, func() error { return nil }) |
| 120 | } |
| 121 | if err := f.g.CompleteFence(t.Context(), target, []byte("receipt"), at, verify); !errors.Is(err, ErrFenced) { |
| 122 | t.Fatalf("nested verifier = %v", err) |
| 123 | } |
| 124 | verify = func(context.Context, FenceTarget, []byte, time.Time) error { return nil } |
| 125 | if err := f.g.CompleteFence(t.Context(), target, []byte("receipt"), at, verify); err != nil { |
| 126 | t.Fatal(err) |
| 127 | } |
| 128 | if err := f.g.CompleteFence(t.Context(), target, []byte("receipt"), at, verify); err != nil { |
| 129 | t.Fatalf("exact replay = %v", err) |
| 130 | } |
| 131 | if err := f.g.CompleteFence(t.Context(), target, []byte("changed"), at, verify); !errors.Is(err, ErrReceiptChanged) { |
| 132 | t.Fatalf("changed replay = %v", err) |
| 133 | } |
| 134 | } |
| 135 | |
| 136 | func TestCleanupUsesJournalDuringControlOutage(t *testing.T) { |
| 137 | f, target, _ := fenceFixture(t) |
| 138 | if _, err := f.g.BeginFence(t.Context(), target); err != nil { |
| 139 | t.Fatal(err) |
| 140 | } |
| 141 | at := time.Date(2026, 10, 6, 15, 0, 0, 0, time.UTC) |
| 142 | if err := f.g.CompleteFence(t.Context(), target, []byte{0, 1, 2, 255}, at, func(context.Context, FenceTarget, []byte, time.Time) error { return nil }); err != nil { |
| 143 | t.Fatal(err) |
| 144 | } |
| 145 | f.err = errors.New("control store unavailable") |
| 146 | verified, cleaned := false, false |
| 147 | err := f.g.CleanupWith(t.Context(), target, func(_ context.Context, receipt FenceReceipt) error { |
| 148 | verified = string(receipt.Bytes) == string([]byte{0, 1, 2, 255}) && receipt.EvidenceAt.Equal(at) |
| 149 | return nil |
| 150 | }, func() error { cleaned = true; return nil }) |
| 151 | if err != nil || !verified || !cleaned { |
| 152 | t.Fatalf("cleanup = %v verified=%v cleaned=%v", err, verified, cleaned) |
| 153 | } |
| 154 | r, err := readRecord(f.g.DataDir, target.Host) |
| 155 | if err != nil || r.ActiveFence == nil || len(r.FenceJournal) != 1 { |
| 156 | t.Fatalf("cleaned record = %+v, %v", r, err) |
| 157 | } |
| 158 | f.err = nil |
| 159 | called := false |
| 160 | if err := f.g.With(t.Context(), target.Host, hostguardCaptured(target), allow, func() error { called = true; return nil }); !errors.Is(err, ErrFenced) || called { |
| 161 | t.Fatalf("cleanup reopened fenced session: %v called=%v", err, called) |
| 162 | } |
| 163 | } |
| 164 | |
| 165 | func TestCleanupOldFencePreservesSuccessor(t *testing.T) { |
| 166 | f, target, _ := fenceFixture(t) |
| 167 | if _, err := f.g.BeginFence(t.Context(), target); err != nil { |
| 168 | t.Fatal(err) |
| 169 | } |
| 170 | at := time.Date(2026, 10, 6, 16, 0, 0, 0, time.UTC) |
| 171 | if err := f.g.CompleteFence(t.Context(), target, []byte("root receipt"), at, func(context.Context, FenceTarget, []byte, time.Time) error { return nil }); err != nil { |
| 172 | t.Fatal(err) |
| 173 | } |
| 174 | f.obs.ETag = "two" |
| 175 | f.obs.Claim.Holder.Incarnation = "successor" |
| 176 | f.obs.Claim.Holder.Stamp = 2 |
| 177 | f.obs.Claim.Holder.Session = 8 |
| 178 | f.obs.Claim.Op = hostclaim.Op{Kind: hostclaim.KindServing, ID: "operation-2", Phase: "serving", Exec: "uk"} |
| 179 | if err := f.g.Activate(t.Context(), target.Host, "uk", nil); err != nil { |
| 180 | t.Fatal(err) |
| 181 | } |
| 182 | if err := f.g.CleanupWith(t.Context(), target, func(context.Context, FenceReceipt) error { return nil }, func() error { return nil }); err != nil { |
| 183 | t.Fatal(err) |
| 184 | } |
| 185 | r, err := readRecord(f.g.DataDir, target.Host) |
| 186 | if err != nil || r.Incarnation != "successor" || r.ActiveStamp != 2 || r.ActiveFence == nil || len(r.FenceJournal) != 1 { |
| 187 | t.Fatalf("successor after cleanup = %+v, %v", r, err) |
| 188 | } |
| 189 | } |
| 190 | |
| 191 | func hostguardCaptured(target FenceTarget) Captured { |
| 192 | return Captured{Incarnation: target.Incarnation, Stamp: target.Stamp} |
| 193 | } |
| 194 | |
| 195 | func TestBeginFenceResumesAfterUnrelatedClaimCAS(t *testing.T) { |
| 196 | f, target, _ := fenceFixture(t) |
| 197 | target.RosterRevision = 9 |
| 198 | if _, err := f.g.BeginFence(t.Context(), target); err != nil { |
| 199 | t.Fatal(err) |
| 200 | } |
| 201 | f.obs.ETag = "bookkeeping" |
| 202 | if _, err := f.g.BeginFence(t.Context(), target); !errors.Is(err, ErrFenceTarget) { |
| 203 | t.Fatalf("stale request: %v", err) |
| 204 | } |
| 205 | fresh := target |
| 206 | fresh.ClaimETag = f.obs.ETag |
| 207 | fresh.FenceID = "retry" |
| 208 | // Roster binding B2: unrelated roster writes between the request and its |
| 209 | // retry move the receipt numeral but not the binding. Retry authority |
| 210 | // ignores the numeral, and both lookups return the persisted target with |
| 211 | // its ORIGINAL numeral. |
| 212 | fresh.RosterRevision = target.RosterRevision + 4 |
| 213 | got, err := f.g.BeginFence(t.Context(), fresh) |
| 214 | if err != nil || got != target || got.RosterRevision != 9 { |
| 215 | t.Fatalf("resume=%+v %v", got, err) |
| 216 | } |
| 217 | existing, err := f.g.ExistingFence(t.Context(), fresh) |
| 218 | if err != nil || existing != target || existing.RosterRevision != 9 { |
| 219 | t.Fatalf("existing=%+v %v", existing, err) |
| 220 | } |
| 221 | // Exact stored-target comparisons stay exact: the retry's numeral names no |
| 222 | // persisted fence. |
| 223 | if err := f.g.ResumeFence(t.Context(), fresh); !errors.Is(err, ErrFenceTarget) { |
| 224 | t.Fatalf("resume with a changed numeral: %v", err) |
| 225 | } |
| 226 | at := time.Unix(1, 0).UTC() |
| 227 | if err := f.g.CompleteFence(t.Context(), got, []byte("original receipt"), at, func(context.Context, FenceTarget, []byte, time.Time) error { return nil }); err != nil { |
| 228 | t.Fatal(err) |
| 229 | } |
| 230 | receipt, err := f.g.FenceReceiptFor(t.Context(), target) |
| 231 | if err != nil || string(receipt.Bytes) != "original receipt" || receipt.Target != target { |
| 232 | t.Fatalf("receipt=%+v %v", receipt, err) |
| 233 | } |
| 234 | if _, err := f.g.FenceReceiptFor(t.Context(), fresh); !errors.Is(err, ErrNoFenceReceipt) { |
| 235 | t.Fatalf("receipt for a changed numeral: %v", err) |
| 236 | } |
| 237 | for _, change := range []func(*FenceTarget){func(v *FenceTarget) { v.Session++ }, func(v *FenceTarget) { v.Op = "another" }, func(v *FenceTarget) { v.ControllerID = "another" }, func(v *FenceTarget) { v.DirectoryRevision = "another-binding" }} { |
| 238 | g, old, _ := fenceFixture(t) |
| 239 | old.RosterRevision = 9 |
| 240 | if _, err := g.g.BeginFence(t.Context(), old); err != nil { |
| 241 | t.Fatal(err) |
| 242 | } |
| 243 | fresh := old |
| 244 | change(&fresh) |
| 245 | fresh.RosterRevision++ |
| 246 | if _, err := g.g.BeginFence(t.Context(), fresh); err == nil { |
| 247 | t.Fatal("changed authority resumed") |
| 248 | } |
| 249 | if _, err := g.g.ExistingFence(t.Context(), fresh); !errors.Is(err, ErrFenceTarget) { |
| 250 | t.Fatalf("changed authority found: %v", err) |
| 251 | } |
| 252 | } |
| 253 | } |
| 254 | |