fix(billing): reject saturated pre-consume quota

service/billing.go

@@ -2,6 +2,7 @@ package service
 
 import (
 	"fmt"
+	"net/http"
 
 	"github.com/QuantumNous/new-api/logger"
 	relaycommon "github.com/QuantumNous/new-api/relay/common"
@@ -17,6 +18,23 @@ const (
 // PreConsumeBilling 根据用户计费偏好创建 BillingSession 并执行预扣费。
 // 会话存储在 relayInfo.Billing 上，供后续 Settle / Refund 使用。
 func PreConsumeBilling(c *gin.Context, preConsumedQuota int, relayInfo *relaycommon.RelayInfo) *types.NewAPIError {
+	if relayInfo != nil && relayInfo.QuotaClamp != nil {
+		clamp := relayInfo.QuotaClamp
+		return types.NewErrorWithStatusCode(
+			fmt.Errorf("pre-consume quota is out of range: operation=%s kind=%s value=%g", clamp.Op, clamp.Kind, clamp.Original),
+			types.ErrorCodeModelPriceError,
+			http.StatusBadRequest,
+			types.ErrOptionWithSkipRetry(),
+		)
+	}
+	if preConsumedQuota < 0 {
+		return types.NewErrorWithStatusCode(
+			fmt.Errorf("pre-consume quota cannot be negative: %d", preConsumedQuota),
+			types.ErrorCodeModelPriceError,
+			http.StatusBadRequest,
+			types.ErrOptionWithSkipRetry(),
+		)
+	}
 	session, apiErr := NewBillingSession(c, relayInfo, preConsumedQuota)
 	if apiErr != nil {
 		return apiErr

service/quota_saturation_test.go

@@ -1,10 +1,12 @@
 package service
 
 import (
+	"net/http"
 	"testing"
 
 	"github.com/QuantumNous/new-api/common"
 	relaycommon "github.com/QuantumNous/new-api/relay/common"
+	"github.com/QuantumNous/new-api/types"
 
 	"github.com/gin-gonic/gin"
 	"github.com/stretchr/testify/require"
@@ -72,3 +74,36 @@ func TestAttachQuotaSaturationNoClampNoMarker(t *testing.T) {
 	_, hasAdmin := other["admin_info"]
 	require.False(t, hasAdmin, "no admin_info should be added when there is no clamp")
 }
+
+func TestPreConsumeBillingRejectsSaturatedQuotaBeforeDeduction(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	c, _ := gin.CreateTestContext(nil)
+	info := &relaycommon.RelayInfo{
+		QuotaClamp: &common.QuotaClamp{
+			Op:       "QuotaFromFloat",
+			Kind:     common.QuotaClampOverflow,
+			Original: 1e30,
+			Clamped:  common.MaxQuota,
+		},
+	}
+
+	apiErr := PreConsumeBilling(c, common.MaxQuota, info)
+
+	require.NotNil(t, apiErr)
+	require.Equal(t, types.ErrorCodeModelPriceError, apiErr.GetErrorCode())
+	require.Equal(t, http.StatusBadRequest, apiErr.StatusCode)
+	require.Nil(t, info.Billing)
+}
+
+func TestPreConsumeBillingRejectsNegativeQuotaBeforeDeduction(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	c, _ := gin.CreateTestContext(nil)
+	info := &relaycommon.RelayInfo{}
+
+	apiErr := PreConsumeBilling(c, -1, info)
+
+	require.NotNil(t, apiErr)
+	require.Equal(t, types.ErrorCodeModelPriceError, apiErr.GetErrorCode())
+	require.Equal(t, http.StatusBadRequest, apiErr.StatusCode)
+	require.Nil(t, info.Billing)
+}