Merge pull request #218 from BaaRai/feat/secrets-phase3b-evidence

secrets: encrypt the evidence snapshot heads at rest (phase 3b)

db/evidence_secrets_test.go

@@ -0,0 +1,245 @@
+package db
+
+import (
+	"database/sql"
+	"encoding/base64"
+	"encoding/json"
+	"fmt"
+	"strings"
+	"testing"
+)
+
+const (
+	evCookie = "session=ev-cookie-secret"
+	evBearer = "Bearer ev-auth-secret"
+)
+
+// evidenceFixture records a finding. With legacy=false the snapshot goes through
+// the normal insert path; with legacy=true it is written in clear by raw SQL.
+func evidenceFixture(t *testing.T, d *DB, tag string, legacy bool) (*Task, int64, TrafficEvidenceSnapshot) {
+	t.Helper()
+	task, err := d.CreateTask("evidence heads at rest "+tag, "goal", nil, 0, 0)
+	if err != nil {
+		t.Fatal(err)
+	}
+	s := TrafficEvidenceSnapshot{SourceTrafficID: "t-sec-ev-" + tag, CapturedAt: 1, URL: "http://ev.example/" + tag, Method: "GET", Status: 200,
+		ReqHead:  "GET /" + tag + " HTTP/1.1\nCookie: " + evCookie + "\nAuthorization: " + evBearer + "\n",
+		RespHead: "HTTP/1.1 200 OK\nSet-Cookie: " + evCookie + "\n", ReqHash: "rh-" + tag, RespHash: "ph-" + tag}
+	s.ID = TrafficSnapshotID(s)
+	t.Cleanup(func() {
+		d.Exec(`DELETE FROM task_archives WHERE task_id=$1`, task.ID)
+		d.DeleteTaskCascade(task.ID, false, true) // findings and their bindings too
+		if _, err := d.Exec(`DELETE FROM traffic_evidence_snapshots WHERE id=$1`, s.ID); err != nil {
+			t.Errorf("cleanup snapshot: %v", err)
+		}
+	})
+	f, err := d.Exploration(task.ExplorationID).RecordFinding(t.Context(), RecordFindingInput{TaskID: task.ID, ExplorationID: task.ExplorationID, Summary: "evidence " + tag, Severity: "low"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if legacy {
+		_, err = d.Exec(`INSERT INTO traffic_evidence_snapshots(id,source_traffic_id,captured_at,url,method,status,req_head,resp_head,req_hash,resp_hash,req_len,resp_len)
+VALUES($1,$2,1,$3,'GET',200,$4,$5,$6,$7,0,0)`, s.ID, s.SourceTrafficID, s.URL, s.ReqHead, s.RespHead, s.ReqHash, s.RespHash)
+		if err == nil {
+			_, err = d.Exec(`INSERT INTO finding_traffic_bindings(finding_id,snapshot_id,position) VALUES($1,$2,0)`, f.FindingID, s.ID)
+		}
+	} else {
+		err = d.WithEvidenceTx(t.Context(), func(tx *sql.Tx) error {
+			return AddFindingTrafficTx(tx, f.FindingID, []PreparedTrafficEvidence{{Ref: TrafficRef{TrafficID: s.SourceTrafficID, Role: "proof"}, Snapshot: s}})
+		})
+	}
+	if err != nil {
+		t.Fatal(err)
+	}
+	return task, f.FindingID, s
+}
+
+// rawHeads reads the stored heads; both must be sealed and hide the secrets.
+func rawHeads(t *testing.T, d *DB, id string) (req, resp string) {
+	t.Helper()
+	if err := d.QueryRow(`SELECT req_head,resp_head FROM traffic_evidence_snapshots WHERE id=$1`, id).Scan(&req, &resp); err != nil {
+		t.Fatal(err)
+	}
+	for _, v := range []string{req, resp} {
+		if !isEncrypted(v) || strings.Contains(v, evCookie) || strings.Contains(v, evBearer) {
+			t.Fatalf("head stored in clear: %q", v)
+		}
+	}
+	return req, resp
+}
+
+// assertClearHeads checks the application read and that the stored value is
+// sealed exactly once.
+func assertClearHeads(t *testing.T, d *DB, findingID int64, want TrafficEvidenceSnapshot) {
+	t.Helper()
+	got, err := d.GetFindingTraffic(t.Context(), findingID)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(got.Bindings) != 1 || got.Bindings[0].Snapshot.ReqHead != want.ReqHead || got.Bindings[0].Snapshot.RespHead != want.RespHead || got.Bindings[0].SnapshotID != want.ID {
+		t.Fatalf("read back: %+v", got.Bindings)
+	}
+	req, resp := rawHeads(t, d, want.ID)
+	box := activeSecretBox.Load()
+	if p, err := box.decrypt(colEvReqHead.aad(), req); err != nil || p != want.ReqHead {
+		t.Fatalf("req_head not sealed once: %q %v", p, err)
+	}
+	if p, err := box.decrypt(colEvRespHead.aad(), resp); err != nil || p != want.RespHead {
+		t.Fatalf("resp_head not sealed once: %q %v", p, err)
+	}
+}
+
+func TestEvidenceHeadsSealedAtInsert(t *testing.T) {
+	d, err := Open(testDSN(t))
+	if err != nil {
+		t.Skipf("postgres unavailable (%v) — skipping", err)
+	}
+	t.Cleanup(func() { d.Close() })
+	_, fid, s := evidenceFixture(t, d, "insert", false)
+	assertClearHeads(t, d, fid, s)
+	var url, method string
+	if err := d.QueryRow(`SELECT url,method FROM traffic_evidence_snapshots WHERE id=$1`, s.ID).Scan(&url, &method); err != nil || url != s.URL || method != "GET" {
+		t.Fatalf("metadata must stay in clear: %q %q %v", url, method, err)
+	}
+}
+
+func TestEvidenceHeadsBackfillAndKeyChecks(t *testing.T) {
+	d, err := Open(testDSN(t))
+	if err != nil {
+		t.Skipf("postgres unavailable (%v) — skipping", err)
+	}
+	t.Cleanup(func() { d.Close() })
+	_, fid, s := evidenceFixture(t, d, "legacy", true)
+	t.Setenv(secretKeyEnv, base64.StdEncoding.EncodeToString(testSecretKey))
+	if err := d.InitSecrets(t.TempDir()); err != nil {
+		t.Fatal(err)
+	}
+	assertClearHeads(t, d, fid, s)
+	req, resp := rawHeads(t, d, s.ID)
+	if err := d.InitSecrets(t.TempDir()); err != nil {
+		t.Fatal(err)
+	}
+	if r2, p2 := rawHeads(t, d, s.ID); r2 != req || p2 != resp {
+		t.Fatal("second backfill must leave the ciphertexts untouched")
+	}
+
+	// A head sealed under another key: the read and the startup check fail and
+	// name the column. The startup check samples one sealed row per column, so
+	// this row must be the only sealed snapshot.
+	var sealedRows int
+	if err := d.QueryRow(`SELECT count(*) FROM traffic_evidence_snapshots WHERE req_head LIKE 'enc:v1:%' OR resp_head LIKE 'enc:v1:%'`).Scan(&sealedRows); err != nil || sealedRows != 1 {
+		t.Fatalf("want only this fixture's snapshot sealed, got %d (%v)", sealedRows, err)
+	}
+	other, _ := newSecretBox([]byte("ffffffffffffffffffffffffffffffff"))
+	for _, c := range []secretColumn{colEvReqHead, colEvRespHead} {
+		bad, _ := other.encrypt(c.aad(), "GET / HTTP/1.1\nCookie: other\n")
+		if _, err := d.Exec(`UPDATE traffic_evidence_snapshots SET `+c.col+`=$1 WHERE id=$2`, bad, s.ID); err != nil {
+			t.Fatal(err)
+		}
+		if _, err := d.GetFindingTraffic(t.Context(), fid); err == nil || !strings.Contains(err.Error(), c.aad()) {
+			t.Fatalf("want a read failure naming %s, got %v", c.aad(), err)
+		}
+		if err := d.InitSecrets(t.TempDir()); err == nil || !strings.Contains(err.Error(), "secret key check") || !strings.Contains(err.Error(), c.aad()) {
+			t.Fatalf("want a key-check failure naming %s, got %v", c.aad(), err)
+		}
+		if _, err := d.Exec(`UPDATE traffic_evidence_snapshots SET req_head=$1,resp_head=$2 WHERE id=$3`, req, resp, s.ID); err != nil {
+			t.Fatal(err)
+		}
+	}
+	assertClearHeads(t, d, fid, s)
+
+	// No key with sealed heads: refuse, and each head counts. One head stays
+	// sealed so the refusal never turns into a new key.
+	t.Setenv(secretKeyEnv, "")
+	dir := t.TempDir()
+	withHeads := encryptedCount(t, d, dir)
+	for _, c := range []secretColumn{colEvReqHead, colEvRespHead} {
+		if _, err := d.Exec(`UPDATE traffic_evidence_snapshots SET req_head=$1,resp_head=$2 WHERE id=$3`, req, resp, s.ID); err != nil {
+			t.Fatal(err)
+		}
+		if _, err := d.Exec(`UPDATE traffic_evidence_snapshots SET `+c.col+`='' WHERE id=$1`, s.ID); err != nil {
+			t.Fatal(err)
+		}
+		if n := encryptedCount(t, d, dir); withHeads-n != 1 {
+			t.Fatalf("no-key refusal counted %d then %d encrypted values; %s must count", withHeads, n, c.aad())
+		}
+	}
+}
+
+// Export keeps the heads sealed; import opens them to check the snapshot ID,
+// then the insert seals them once. A legacy clear archive is sealed on import.
+func TestEvidenceHeadsArchiveRoundTrip(t *testing.T) {
+	d, err := Open(testDSN(t))
+	if err != nil {
+		t.Skipf("postgres unavailable (%v) — skipping", err)
+	}
+	t.Cleanup(func() { d.Close() })
+	if err = d.EnsureLLMRecordsTable(); err != nil {
+		t.Fatal(err)
+	}
+	if err = d.EnsureLLMUsageTable(); err != nil {
+		t.Fatal(err)
+	}
+	for _, legacy := range []bool{false, true} {
+		t.Run(fmt.Sprint("legacy=", legacy), func(t *testing.T) {
+			task, fid, s := evidenceFixture(t, d, fmt.Sprint("archive-", legacy), false)
+			if err := d.SetPaused(task.ID, true); err != nil {
+				t.Fatal(err)
+			}
+			archive, err := d.QueueTaskArchive(task.ID)
+			if err != nil {
+				t.Fatal(err)
+			}
+			if _, err = d.ClaimTaskArchiveJob(t.Context()); err != nil {
+				t.Fatal(err)
+			}
+			snapshot, err := d.SnapshotTaskArchive(task.ID)
+			if err != nil {
+				t.Fatal(err)
+			}
+			exported := string(snapshot.Tables["traffic_evidence_snapshots"])
+			if !strings.Contains(exported, secretPrefix) || strings.Contains(exported, evCookie) || strings.Contains(exported, evBearer) {
+				t.Fatalf("archive must keep the heads sealed: %s", exported)
+			}
+			if err = d.CompleteTaskArchive(archive.ID, snapshot, "/tmp/evidence-heads-test.tar.zst", "fixture", 1, 1); err != nil {
+				t.Fatal(err)
+			}
+			var rows []map[string]any
+			if err = json.Unmarshal(snapshot.Tables["traffic_evidence_snapshots"], &rows); err != nil || len(rows) != 1 {
+				t.Fatal(rows, err)
+			}
+			if legacy { // an archive written before the heads were sealed
+				rows[0]["req_head"], rows[0]["resp_head"] = s.ReqHead, s.RespHead
+				snapshot.Tables["traffic_evidence_snapshots"], _ = json.Marshal(rows)
+			} else { // a head sealed under another key fails the import, naming the column
+				other, _ := newSecretBox([]byte("ffffffffffffffffffffffffffffffff"))
+				bad, _ := other.encrypt(colEvReqHead.aad(), s.ReqHead)
+				good := rows[0]["req_head"]
+				rows[0]["req_head"] = bad
+				raw, _ := json.Marshal(rows)
+				if _, err := ArchiveEvidenceSnapshots(&TaskArchiveSnapshot{FormatVersion: snapshot.FormatVersion, Tables: map[string]json.RawMessage{"traffic_evidence_snapshots": raw}}); err == nil || !strings.Contains(err.Error(), colEvReqHead.aad()) {
+					t.Fatalf("want an import failure naming %s, got %v", colEvReqHead.aad(), err)
+				}
+				rows[0]["req_head"] = good
+			}
+			list, err := ArchiveEvidenceSnapshots(snapshot)
+			if err != nil || len(list) != 1 || list[0].ReqHead != s.ReqHead || list[0].RespHead != s.RespHead {
+				t.Fatalf("archive snapshots: %+v %v", list, err)
+			}
+			// The hot row is gone (GC), so the restore inserts it again.
+			if _, err = d.Exec(`DELETE FROM traffic_evidence_snapshots WHERE id=$1`, s.ID); err != nil {
+				t.Fatal(err)
+			}
+			if _, err = d.QueueTaskArchiveRestore(archive.ID); err != nil {
+				t.Fatal(err)
+			}
+			if _, err = d.ClaimTaskArchiveJob(t.Context()); err != nil {
+				t.Fatal(err)
+			}
+			if _, err = d.RestoreTaskArchive(archive.ID, snapshot, 0); err != nil {
+				t.Fatal(err)
+			}
+			assertClearHeads(t, d, fid, s)
+		})
+	}
+}

db/finding_traffic.go

@@ -195,10 +195,29 @@ func InsertEvidenceSnapshotTx(tx *sql.Tx, s TrafficEvidenceSnapshot) error {
 	if s.CreatedAt.IsZero() {
 		s.CreatedAt = time.Now().UTC()
 	}
-	_, err := tx.Exec(`INSERT INTO traffic_evidence_snapshots
+	// The ID hashes the clear heads; only the stored heads are sealed.
+	reqHead, err := sealSecret(colEvReqHead, s.ReqHead)
+	if err != nil {
+		return err
+	}
+	respHead, err := sealSecret(colEvRespHead, s.RespHead)
+	if err != nil {
+		return err
+	}
+	_, err = tx.Exec(`INSERT INTO traffic_evidence_snapshots
 (id,source_traffic_id,captured_at,url,method,status,content_type,req_head,resp_head,req_hash,resp_hash,req_len,resp_len,created_at)
 VALUES($1,$2,$3,$4,$5,$6,$7,$8,$9,$10,$11,$12,$13,$14) ON CONFLICT(id) DO NOTHING`,
-		s.ID, s.SourceTrafficID, s.CapturedAt, s.URL, s.Method, s.Status, s.ContentType, s.ReqHead, s.RespHead, s.ReqHash, s.RespHash, s.ReqLen, s.RespLen, s.CreatedAt)
+		s.ID, s.SourceTrafficID, s.CapturedAt, s.URL, s.Method, s.Status, s.ContentType, reqHead, respHead, s.ReqHash, s.RespHash, s.ReqLen, s.RespLen, s.CreatedAt)
+	return err
+}
+
+// openHeads decrypts the heads of a snapshot read from the table or an archive.
+// Clear legacy values pass through; a value that fails to decrypt is an error.
+func (s *TrafficEvidenceSnapshot) openHeads() (err error) {
+	if s.ReqHead, err = openSecret(colEvReqHead, s.ReqHead); err != nil {
+		return err
+	}
+	s.RespHead, err = openSecret(colEvRespHead, s.RespHead)
 	return err
 }
 
@@ -263,6 +282,9 @@ FROM finding_traffic_bindings b JOIN traffic_evidence_snapshots s ON s.id=b.snap
 		if err := json.Unmarshal(raw, &b.Snapshot); err != nil {
 			return nil, err
 		}
+		if err := b.Snapshot.openHeads(); err != nil {
+			return nil, err
+		}
 		out.Bindings = append(out.Bindings, b)
 	}
 	return out, rows.Err()

db/finding_traffic_archive.go

@@ -15,8 +15,17 @@ func ArchiveEvidenceSnapshots(snapshot *TaskArchiveSnapshot) ([]TrafficEvidenceS
 	if snapshot.FormatVersion < 3 {
 		return nil, ErrTaskArchiveFormatMismatch
 	}
-	err := json.Unmarshal(snapshot.Tables["traffic_evidence_snapshots"], &out)
-	return out, err
+	// The archive keeps the heads as stored (sealed). They are opened here so the
+	// restore can check each content-addressed ID; the insert then seals them once.
+	if err := json.Unmarshal(snapshot.Tables["traffic_evidence_snapshots"], &out); err != nil {
+		return nil, err
+	}
+	for i := range out {
+		if err := out[i].openHeads(); err != nil {
+			return nil, err
+		}
+	}
+	return out, nil
 }
 
 func restoreFindingTrafficTx(tx *sql.Tx, snapshot *TaskArchiveSnapshot) error {

db/secrets.go

@@ -81,7 +81,10 @@ var (
 	colCredPass    = secretColumn{"task_credentials", "password", false, ""}
 	colNotifConfig = secretColumn{"notification_channels", "config", true, ""}
 	colToolExec    = secretColumn{"tools", "exec", true, ""} // system=false rows: headers, proxy, body, code, command
-	secretColumns  = append([]secretColumn{colLLMKey, colLLMProxy, colMCPEnv, colCredPass, colNotifConfig, colToolExec}, secretSettingColumns()...)
+	// Evidence head blocks carry the captured Cookie and Authorization headers.
+	colEvReqHead  = secretColumn{"traffic_evidence_snapshots", "req_head", false, ""}
+	colEvRespHead = secretColumn{"traffic_evidence_snapshots", "resp_head", false, ""}
+	secretColumns = append([]secretColumn{colLLMKey, colLLMProxy, colMCPEnv, colCredPass, colNotifConfig, colToolExec, colEvReqHead, colEvRespHead}, secretSettingColumns()...)
 )
 
 // secretSettingKeys are the settings stored encrypted, AAD "settings.<key>".

evidence/testmain_test.go

@@ -34,5 +34,10 @@ func runEvidenceSuite(m *testing.M) int {
 		panic(err)
 	}
 	defer conn.ExecContext(context.Background(), `SELECT pg_advisory_unlock(7337741002)`)
+	// Snapshot inserts seal the heads: load a fixed test key (env, no key file).
+	os.Setenv("ARTEX_SECRET_KEY", "MDEyMzQ1Njc4OWFiY2RlZjAxMjM0NTY3ODlhYmNkZWY=")
+	if err = pg.InitSecrets(""); err != nil {
+		panic(err)
+	}
 	return m.Run()
 }

server/finding_traffic_test.go

@@ -172,6 +172,11 @@ func TestFindingTrafficAPIAndExport(t *testing.T) {
 	if w = req("GET", fmt.Sprintf("%s/%d", base, first), ""); w.Code != 200 || !strings.Contains(w.Body.String(), `"binary":true`) {
 		t.Fatal(w.Code, w.Body)
 	}
+	// The viewer reads the heads in clear; the table stores them sealed.
+	var rawHead string
+	if err := s.m.pg.QueryRow(`SELECT req_head FROM traffic_evidence_snapshots WHERE id=$1`, list.Bindings[0].SnapshotID).Scan(&rawHead); err != nil || !strings.HasPrefix(rawHead, "enc:v1:") || !strings.Contains(w.Body.String(), `"req_head":"POST / HTTP/1.1"`) {
+		t.Fatalf("head at rest %q (%v), viewer %s", rawHead, err, w.Body)
+	}
 	if w = req("GET", fmt.Sprintf("%s/%d/body?side=response&offset=20001", base, first), ""); w.Code != 422 {
 		t.Fatal(w.Code, w.Body)
 	}