fix(sip-call): require native PJSIP module reload confirmation

2026-10-08 15:35:27 +08:00
parent 6bf305284a
commit 0a64b383c7
5 changed files with 107 additions and 7 deletions
@@ -107,7 +107,7 @@
 | 线路 | SIP 服务端（历史登记，非已验证线路） | 主叫标识 | 被叫前缀 |
 | --- | --- | --- | --- |
 | 数企 | `61.132.228.221:5060` | `BD93205882` | `7089` |
-| 中鼎 | `60.171.24.90:5060` | `mbkq` | 无 |
+| 中鼎 | `222.186.130.228:5060` | `mbkq` | 无 |
 | 百应 | `160.202.254.79:5060` | `KQ91526` | `mka755108` |


@@ -29,7 +29,7 @@ chmod 600 sip-xx.env

 ## 自动配置与退出

-取得该 Asterisk 配置目录的独占测试锁，确认空闲后保存原 `pjsip.conf` 到本次私有目录的 `pjsip.conf.before`。临时增加独立 include/endpoint/AOR，reload 后核对实际服务地址、主叫、codec、transport 和 context。通过才执行一次原生 ARI Dial。
+取得该 Asterisk 配置目录的独占测试锁，确认空闲后保存原 `pjsip.conf` 到本次私有目录的 `pjsip.conf.before`。临时增加独立 include/endpoint/AOR，使用原生 `module reload res_pjsip.so` 并核对明确的成功响应，不依赖可选的 `pjsip reload` 别名，不把 CLI 退出码 0 当作加载成功；随后核对实际服务地址、主叫、codec、transport 和 context。通过才执行一次原生 ARI Dial。

 收到真正的 `Up` 和 `StasisStart` 才记录接通；按 `HOLD_SECONDS` 保持后挂断。观察到通道结束或挂断后的 ARI 404 才记录结束确认。正常结束后恢复原 `pjsip.conf` 原字节、删除本次专用配置、reload 并确认 endpoint 已移除，不改 `go-sip-managed.conf` 或业务 SIP 加载代次。

@@ -3,6 +3,7 @@ package sipcall
 import (
 	"bytes"
 	"context"
+	"crypto/sha256"
 	"errors"
 	"fmt"
 	"os"
@@ -88,7 +89,7 @@ func applyConfiguration(ctx context.Context, c Config, cli cliFunc) (restore fun
 		if e := os.Remove(managed); e != nil {
 			return e
 		}
-		if _, e := cli(cleanupCtx, "pjsip reload"); e != nil {
+		if e := reloadPJSIP(cleanupCtx, cli); e != nil {
 			return e
 		}
 		out, e := cli(cleanupCtx, "pjsip show endpoint "+c.Endpoint())
@@ -106,7 +107,7 @@ func applyConfiguration(ctx context.Context, c Config, cli cliFunc) (restore fun
 	if err = atomicConfig(basePath, appended); err != nil {
 		return restore, err
 	}
-	if _, err = cli(ctx, "pjsip reload"); err != nil {
+	if err = reloadPJSIP(ctx, cli); err != nil {
 		return restore, err
 	}
 	endpoint, err := cli(ctx, "pjsip show endpoint "+c.Endpoint())
@@ -140,6 +141,21 @@ func applyConfiguration(ctx context.Context, c Config, cli cliFunc) (restore fun
 	return restore, nil
 }

+// Asterisk CLI errors can exit zero. Confirm the native module response before
+// checking loaded objects; pjsip reload is an optional alias, not a native command.
+func reloadPJSIP(ctx context.Context, cli cliFunc) error {
+	out, err := cli(ctx, "module reload res_pjsip.so")
+	if err != nil {
+		return err
+	}
+	for _, line := range strings.Split(string(out), "\n") {
+		if strings.TrimSpace(line) == "Module 'res_pjsip.so' reloaded successfully." {
+			return nil
+		}
+	}
+	return fmt.Errorf("native PJSIP reload not confirmed (output_sha256=%x)", sha256.Sum256(out))
+}
+
 func atomicConfig(path string, data []byte) error {
 	f, err := os.CreateTemp(filepath.Dir(path), ".sip-call-write-*")
 	if err != nil {
@@ -18,8 +18,8 @@ func configCLI(c Config, basePath string) cliFunc {
 			return []byte("0 active channels\n0 active calls\n"), nil
 		case "pjsip show transports":
 			return []byte("Transport: go-sip-udp udp 0 0 0.0.0.0:5060\n"), nil
-		case "pjsip reload":
-			return []byte("Module reloaded\n"), nil
+		case "module reload res_pjsip.so":
+			return []byte("Module 'res_pjsip.so' reloaded successfully.\n"), nil
 		case "pjsip show endpoint " + c.Endpoint():
 			data, err := os.ReadFile(basePath)
 			if err != nil {
@@ -116,7 +116,7 @@ func TestConfigurationFailsClosedOnRuntimeMismatch(t *testing.T) {
 				if failure == "caller" && strings.HasPrefix(s, "pjsip show endpoint") {
 					return bytes.ReplaceAll(out, []byte("BD1234"), []byte("WRONG")), nil
 				}
-				if failure == "reload" && s == "pjsip reload" {
+				if failure == "reload" && s == "module reload res_pjsip.so" {
 					return nil, errors.New("reload failed")
 				}
 				if failure == "aor" && strings.HasPrefix(s, "pjsip show aor") {
@@ -0,0 +1,84 @@
+package sipcall
+
+import (
+	"bytes"
+	"context"
+	"fmt"
+	"os"
+	"strings"
+	"testing"
+)
+
+func TestNativePJSIPModuleReloadIsUsedAndConfirmed(t *testing.T) {
+	c, path, base := stageFixture(t)
+	fixture := configCLI(c, path)
+	var reloads int
+	cli := func(ctx context.Context, command string) ([]byte, error) {
+		switch command {
+		case "pjsip reload":
+			return []byte("No such command 'pjsip reload' (type 'core show help pjsip reload' for other possible commands)\n"), nil
+		case "module reload res_pjsip.so":
+			reloads++
+			return []byte("Module 'res_pjsip.so' reloaded successfully.\n"), nil
+		default:
+			return fixture(ctx, command)
+		}
+	}
+	restore, err := applyConfiguration(context.Background(), c, cli)
+	if err != nil {
+		t.Fatalf("supported native reload failed: %v", err)
+	}
+	if err := restore(); err != nil {
+		t.Fatal(err)
+	}
+	if reloads != 2 {
+		t.Fatalf("expected native module reload for apply and restore, got %d", reloads)
+	}
+	data, err := os.ReadFile(path)
+	if err != nil || !bytes.Equal(data, base) {
+		t.Fatal("original configuration not restored")
+	}
+}
+
+func TestPJSIPReloadRejectsZeroExitWithoutModuleConfirmation(t *testing.T) {
+	for _, output := range []string{
+		"No such command 'module reload res_pjsip.so'\n",
+		"Module 'res_pjsip.so' could not be reloaded.\n",
+		"Module 'res_other.so' reloaded successfully.\n",
+		"",
+	} {
+		for _, phase := range []string{"apply", "restore"} {
+			t.Run(fmt.Sprintf("%s/%q", phase, output), func(t *testing.T) {
+				c, path, _ := stageFixture(t)
+				fixture := configCLI(c, path)
+				var reloads, runtimeReads int
+				cli := func(ctx context.Context, command string) ([]byte, error) {
+					if command == "module reload res_pjsip.so" {
+						reloads++
+						if phase == "apply" || reloads == 2 {
+							return []byte(output), nil
+						}
+						return []byte("Module 'res_pjsip.so' reloaded successfully.\n"), nil
+					}
+					if strings.HasPrefix(command, "pjsip show endpoint ") {
+						runtimeReads++
+					}
+					return fixture(ctx, command)
+				}
+				restore, err := applyConfiguration(context.Background(), c, cli)
+				if phase == "apply" {
+					if err == nil || !strings.Contains(err.Error(), "reload not confirmed") || runtimeReads != 0 {
+						t.Fatalf("unconfirmed reload must fail before readback: error=%v reads=%d", err, runtimeReads)
+					}
+					return
+				}
+				if err != nil {
+					t.Fatal(err)
+				}
+				if err = restore(); err == nil || !strings.Contains(err.Error(), "reload not confirmed") {
+					t.Fatalf("unconfirmed restoration reload accepted: %v", err)
+				}
+			})
+		}
+	}
+}