fix(batch): align variable preflight with policy evaluation

Treat unsupported variable spellings as literals, inspect only condition values, and decode BinaryEquals before checking required context. Keep missing request variables fail-closed.

Fixes #347

Signed-off-by: Feng Ruohang <rh@vonng.com>

cmd/batch-authz-variables_test.go

@@ -0,0 +1,152 @@
+// Copyright (c) 2026 Feng Ruohang
+// SPDX-License-Identifier: AGPL-3.0-or-later
+
+package cmd
+
+import (
+	"encoding/base64"
+	"fmt"
+	"maps"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+
+	"github.com/pgsty/silo-pkg/v3/policy"
+)
+
+// Use explicit condition values to exercise the evaluator, including literal
+// spellings that S3 tag validation does not permit in stored tags.
+func TestBatchPolicyConditionVariableParity(t *testing.T) {
+	f := newBatchAuthzFixture(t)
+	cases := []struct {
+		value, actual string
+		matches       bool
+		requestOnly   bool
+	}{
+		{"${$}100", "${$}100", true, false},
+		{"${$}100", "$100", false, false},
+		{"${*}", "${*}", true, false},
+		{"${?}", "${?}", true, false},
+		{"${custom:unknown}", "${custom:unknown}", true, false},
+		{"${aws:Unknown}", "${aws:Unknown}", true, false},
+		{"${aws:UserAgent", "${aws:UserAgent", true, false},
+		{"${aws:username}", "", true, false},
+		{"${aws:UserAgent}", "test-agent", true, true},
+		{"${prefix${aws:UserAgent}}", "${prefixtest-agent}", true, true},
+	}
+	for i, tc := range cases {
+		for opIndex, op := range []string{
+			"StringEquals", "StringNotEquals", "StringEqualsIgnoreCase", "StringNotEqualsIgnoreCase",
+			"StringLike", "StringNotLike", "ForAnyValue:StringEquals", "ForAllValues:StringEquals", "BinaryEquals",
+		} {
+			for _, effect := range []string{"Allow", "Deny"} {
+				t.Run(fmt.Sprintf("%d/%s/%s", i, op, effect), func(t *testing.T) {
+					name := fmt.Sprintf("var-%d-%d-%s", i, opIndex, effect)
+					conditionValue := tc.value
+					if op == "BinaryEquals" {
+						conditionValue = base64.StdEncoding.EncodeToString([]byte(conditionValue))
+					}
+					statement := fmt.Sprintf(`{"Effect":%q,"Action":["s3:GetObject"],"Resource":["arn:aws:s3:::batch-authz/data/*"],"Condition":{%q:{"s3:ExistingObjectTag/label":%q}}}`, effect, op, conditionValue)
+					if effect == "Deny" {
+						statement += `,{"Effect":"Allow","Action":["s3:GetObject"],"Resource":["arn:aws:s3:::batch-authz/data/*"]}`
+					}
+					cred, sub := batchContextUser(t, f, name, statement)
+					actual := tc.actual
+					if tc.value == "${aws:username}" {
+						actual = cred.AccessKey
+					}
+					values := map[string][]string{"ExistingObjectTag/label": {actual}}
+					ordinaryValues := maps.Clone(values)
+					ordinaryValues["username"] = []string{cred.AccessKey}
+					ordinaryValues["UserAgent"] = []string{"test-agent"}
+					ordinaryAllowed := f.sys.isAllowedLocally(policy.Args{
+						AccountName:     cred.AccessKey,
+						Action:          policy.GetObjectAction,
+						BucketName:      f.bucket,
+						ObjectName:      "data/" + name,
+						ConditionValues: ordinaryValues,
+					})
+					matches := tc.matches
+					if strings.Contains(op, "Not") {
+						matches = !matches
+					}
+					if want := matches == (effect == "Allow"); ordinaryAllowed != want {
+						t.Fatalf("ordinary evaluator allowed=%v want %v", ordinaryAllowed, want)
+					}
+					batchErr := sub.authorize(f.ctx, batchAuthzRequest{
+						action: policy.GetObjectAction, bucket: f.bucket, object: "data/" + name, values: values,
+					})
+					if tc.requestOnly {
+						if !isBatchJobAuthzError(batchErr) {
+							t.Fatalf("missing request context must deny: %v", batchErr)
+						}
+						if (effect == "Deny") != isBatchContextRefusal(batchErr) {
+							t.Fatalf("only a Deny must refuse missing context: %v", batchErr)
+						}
+						return
+					}
+					if (batchErr == nil) != ordinaryAllowed {
+						t.Fatalf("ordinary allowed=%v; batch=%v", ordinaryAllowed, batchErr)
+					}
+				})
+			}
+		}
+	}
+
+	// Condition keys themselves are never expanded.
+	const statement = `{"Effect":"Allow","Action":["s3:GetObject"],"Resource":["arn:aws:s3:::batch-authz/*"],"Condition":{"StringEquals":{"s3:ExistingObjectTag/${aws:UserAgent}":"literal"}}}`
+	cred, sub := batchContextUser(t, f, "literal-condition-key", statement)
+	values := map[string][]string{"ExistingObjectTag/${aws:UserAgent}": {"literal"}}
+	if !f.sys.isAllowedLocally(policy.Args{AccountName: cred.AccessKey, Action: policy.GetObjectAction, BucketName: f.bucket, ObjectName: "data/key", ConditionValues: values}) {
+		t.Fatal("ordinary evaluator did not match the literal condition key")
+	}
+	if err := sub.authorize(f.ctx, batchAuthzRequest{action: policy.GetObjectAction, bucket: f.bucket, object: "data/key", values: values}); err != nil {
+		t.Fatalf("batch treated a condition key as a variable: %v", err)
+	}
+}
+
+// Object names permit these spellings, so also compare real S3 GETs with batch.
+func TestBatchPolicyResourceVariableParity(t *testing.T) {
+	f := newBatchAuthzFixture(t)
+	router := initTestAPIEndPoints(f.obj, []string{"GetObject"})
+	for i, raw := range []string{"${$}100", "${*}", "${?}", "${custom:unknown}", "${prefix${aws:UserAgent}}", "${aws:username}", "${aws:UserAgent}"} {
+		for _, effect := range []string{"Allow", "Deny"} {
+			t.Run(fmt.Sprintf("%d/%s", i, effect), func(t *testing.T) {
+				name := fmt.Sprintf("res-%d-%s", i, effect)
+				statement := fmt.Sprintf(`{"Effect":%q,"Action":["s3:GetObject"],"Resource":[%q]}`, effect, "arn:aws:s3:::batch-authz/"+raw)
+				if effect == "Deny" {
+					statement += `,{"Effect":"Allow","Action":["s3:GetObject"],"Resource":["arn:aws:s3:::batch-authz/*"]}`
+				}
+				cred, sub := batchContextUser(t, f, name, statement)
+				object := strings.ReplaceAll(raw, "${aws:username}", cred.AccessKey)
+				if raw == "${aws:UserAgent}" {
+					object = "test-agent"
+				}
+				_, err := f.obj.PutObject(f.ctx, f.bucket, object, mustGetPutObjReader(t, strings.NewReader("data"), 4, "", ""), ObjectOptions{})
+				mustIAM(t, err)
+				req, err := newTestSignedRequestV4(http.MethodGet, getGetObjectURL("", f.bucket, object), 0, nil, cred.AccessKey, cred.SecretKey, map[string]string{"User-Agent": "test-agent"})
+				mustIAM(t, err)
+				rec := httptest.NewRecorder()
+				router.ServeHTTP(rec, req)
+				wantStatus := http.StatusOK
+				if effect == "Deny" {
+					wantStatus = http.StatusForbidden
+				}
+				if rec.Code != wantStatus {
+					t.Fatalf("S3 read: %d want %d: %s", rec.Code, wantStatus, rec.Body.String())
+				}
+				err = sub.authorize(f.ctx, batchAuthzRequest{action: policy.GetObjectAction, bucket: f.bucket, object: object})
+				if raw == "${aws:UserAgent}" {
+					if !isBatchJobAuthzError(err) || (effect == "Deny") != isBatchContextRefusal(err) {
+						t.Fatalf("batch did not reject missing request context: %v", err)
+					}
+					return
+				}
+				if (err == nil) != (rec.Code == http.StatusOK) {
+					t.Fatalf("resource %q %s: S3=%d batch=%v", raw, effect, rec.Code, err)
+				}
+			})
+		}
+	}
+}

cmd/batch-authz.go

@@ -829,15 +829,15 @@ func batchStatementNeedsRequest(st policy.Statement, args policy.Args, supplied
 	// Policy variables in condition values.
 	if len(st.Conditions) > 0 {
 		data, err := json.Marshal(st.Conditions)
-		if err != nil || batchUnsuppliedVariable(string(data), supplied) {
+		if err != nil || batchConditionsNeedVariables(data, supplied) {
 			return true
 		}
 	}
 	return false
 }
 
-// batchUnsuppliedVariable reports whether s holds a policy variable, ${key},
+// batchUnsuppliedVariable follows Resource.Match: only CommonKeys are expanded,
-// whose value the job does not supply.
+// and an unsupported ${...} sequence (including its nested text) stays literal.
 func batchUnsuppliedVariable(s string, supplied map[string]bool) bool {
 	for {
 		i := strings.Index(s, "${")
@@ -849,13 +849,53 @@ func batchUnsuppliedVariable(s string, supplied map[string]bool) bool {
 		if j < 0 {
 			return false
 		}
-		if !batchKeySupplied(supplied, s[:j]) {
+		if condition.CommonKeysMap[condition.KeyName(s[:j])] && !batchKeySupplied(supplied, s[:j]) {
 			return true
 		}
 		s = s[j+1:]
 	}
 }
 
+// Conditions replace each recognized variable wherever it occurs in a string
+// value, unlike the resource parser. Inspect values only, not condition keys,
+// and decode BinaryEquals as the policy evaluator does before substitution.
+func batchConditionsNeedVariables(data []byte, supplied map[string]bool) bool {
+	var conditions map[string]map[string]condition.ValueSet
+	if err := json.Unmarshal(data, &conditions); err != nil {
+		return true
+	}
+	for name, keys := range conditions {
+		name = strings.TrimPrefix(strings.TrimPrefix(name, "ForAnyValue:"), "ForAllValues:")
+		name = strings.TrimSuffix(name, "IfExists")
+		switch name {
+		case "StringEquals", "StringNotEquals", "StringEqualsIgnoreCase", "StringNotEqualsIgnoreCase", "StringLike", "StringNotLike", "BinaryEquals":
+		default:
+			continue
+		}
+		for _, values := range keys {
+			for value := range values {
+				s, err := value.GetString()
+				if err != nil {
+					return true
+				}
+				if name == "BinaryEquals" {
+					decoded, err := base64.StdEncoding.DecodeString(s)
+					if err != nil {
+						return true
+					}
+					s = string(decoded)
+				}
+				for _, key := range condition.CommonKeys {
+					if strings.Contains(s, key.VarName()) && !batchKeySupplied(supplied, string(key)) {
+						return true
+					}
+				}
+			}
+		}
+	}
+	return false
+}
+
 // batchStatementMayApply matches the resources of a statement, in both forms
 // a bucket request can take. A statement whose resources the evaluator
 // translates (S3 Tables and vectors actions cover S3 data actions) is taken