ykxmz/_build/test_deploy_sh.sh

#!/bin/bash
# ════════════════════════════════════════════════════════════════
# deploy.sh 的实测脚本
# ════════════════════════════════════════════════════════════════
#
# 这台开发机是 Windows（Git Bash），没有 systemctl / ip / iptables，
# 所以测试的做法是：
#   1. 造一个沙箱目录，用 deploy.sh 的 --prefix 把所有写入关在里面
#   2. 在 PATH 最前面放一批「桩命令」，把 Linux 专有的东西接住
#      （systemctl / ip / iptables / nmcli / cloudflared / curl …）
#   3. 同一个沙箱**跑两遍**，验证第二遍不报错、不把已有东西冲掉
#
# 验证的是：语法、参数解析、文件部署、unit 生成、幂等性。
# **不能**验证的：真实的 systemd 行为、iptables 真的有没有生效、
# 隧道能不能连上 —— 那些得在真的 Linux 机器上跑。
#
# 用法: bash _build/test_deploy_sh.sh
# ════════════════════════════════════════════════════════════════
set -u

REPO=$(cd "$(dirname "$0")/.." && pwd)
DEPLOY="$REPO/relay/deploy.sh"
TUNE="$REPO/relay/deploy-tune.sh"

PASS=0
FAIL=0
ok()   { PASS=$((PASS + 1)); printf "    [PASS] %s\n" "$1"; }
bad()  { FAIL=$((FAIL + 1)); printf "    [FAIL] %s\n" "$1"; }
check(){ if [ "$2" = "$3" ]; then ok "$1"; else bad "$1（期望 $3，实际 $2）"; fi; }
exists(){ if [ -e "$2" ]; then ok "$1"; else bad "$1（找不到 $2）"; fi; }
noexist(){ if [ ! -e "$2" ]; then ok "$1"; else bad "$1（不该存在 $2）"; fi; }

echo "============================================================"
echo "  deploy.sh 实测"
echo "============================================================"
echo "  仓库:   $REPO"
echo "  脚本:   $DEPLOY"

[ -f "$DEPLOY" ] || { echo "  找不到 deploy.sh"; exit 1; }
[ -f "$TUNE" ] || { echo "  找不到 deploy-tune.sh"; exit 1; }

# ════════════════════════════════════════════════════════════════
echo
echo "── 1. 语法检查（bash -n）──"
for f in "$DEPLOY" "$TUNE"; do
    if bash -n "$f" 2>/tmp/synerr; then
        ok "$(basename "$f") 语法正确"
    else
        bad "$(basename "$f") 语法错误: $(head -3 /tmp/synerr)"
    fi
done

# ════════════════════════════════════════════════════════════════
echo
echo "── 2. 换行符 / BOM（.sh 必须 LF + 无 BOM）──"
python - "$DEPLOY" "$TUNE" <<'PYEOF'
import sys
from pathlib import Path
bad = 0
for p in sys.argv[1:]:
    raw = Path(p).read_bytes()
    crlf = raw.count(b"\r\n")
    bom = raw.startswith(b"\xef\xbb\xbf")
    name = Path(p).name
    if crlf == 0 and not bom:
        print("    [PASS] %s 是 LF 且无 BOM" % name)
    else:
        print("    [FAIL] %s CRLF=%d BOM=%s" % (name, crlf, bom))
        bad += 1
sys.exit(1 if bad else 0)
PYEOF
if [ $? -eq 0 ]; then PASS=$((PASS + 2)); else FAIL=$((FAIL + 2)); fi

# ════════════════════════════════════════════════════════════════
echo
echo "── 3. --help ──"
HELP=$(bash "$DEPLOY" --help 2>&1)
RC=$?
check "--help 退出码为 0" "$RC" "0"
if echo "$HELP" | grep -q '用法'; then ok "--help 有「用法」段"; else bad "--help 没输出用法"; fi
if echo "$HELP" | grep -q -- '--prefix'; then ok "--help 列出了 --prefix"; else bad "--help 缺 --prefix"; fi
if echo "$HELP" | grep -q 'TUNNEL_TOKEN'; then ok "--help 提到了 TUNNEL_TOKEN"; else bad "--help 没提 TUNNEL_TOKEN"; fi
# --help 不能有副作用
if echo "$HELP" | grep -qi '装好了'; then bad "--help 竟然执行了部署"; else ok "--help 没有副作用"; fi

echo
echo "── 3b. 非法参数要报错退出 ──"
bash "$DEPLOY" --no-such-flag >/dev/null 2>&1
check "未知参数退出码为 2" "$?" "2"
bash "$DEPLOY" --protocol bogus >/dev/null 2>&1
check "非法协议退出码为 2" "$?" "2"

# ════════════════════════════════════════════════════════════════
echo
echo "── 4. 搭沙箱（桩命令 + 临时安装根）──"
TMP=$(mktemp -d)
STUB="$TMP/stub"
SANDBOX="$TMP/root"
STUB_LOG="$TMP/stub.log"
mkdir -p "$STUB"
: > "$STUB_LOG"
export STUB_LOG

REAL_PY=$(command -v python 2>/dev/null || true)

mk() {  # mk <名字> <<'EOF' ...
    local name="$1"
    cat > "$STUB/$name"
    chmod +x "$STUB/$name"
}

mk systemctl <<'EOF'
#!/bin/bash
# 桩：会**记住**哪些服务被启动过。
# 如果这里 is-active 一律返回 active，那么"无变化就不该重启"这类
# 断言永远测不出来（net-tune.sh 会以为隧道在跑，然后去重启它）。
echo "systemctl $*" >> "$STUB_LOG"
STATE="$STUB_LOG.state"
touch "$STATE"

# ⚠ 服务名要自己从参数里挑出来。
#   直接取 $2 是错的：`systemctl is-active --quiet foo` 里 $2 是 --quiet，
#   不是服务名 —— 这样挑出来的是 "--quiet"，grep 永远不中，
#   于是所有服务都被当成"没在跑"。
SVC=""
for a in "$@"; do
    case "$a" in
        -*) continue ;;
        is-active|is-enabled|start|stop|restart|reload|enable|disable|status|daemon-reload|mask|unmask)
            continue ;;
        *) SVC="$a" ;;
    esac
done

case "$1" in
    is-active)
        if [ -n "$SVC" ] && grep -qxF "$SVC" "$STATE" 2>/dev/null; then
            echo "active"; exit 0
        fi
        echo "inactive"; exit 3 ;;
    is-enabled) echo "enabled"; exit 0 ;;
    start|restart)
        [ -n "$SVC" ] && { grep -qxF "$SVC" "$STATE" 2>/dev/null \
            || echo "$SVC" >> "$STATE"; }
        exit 0 ;;
    stop)
        if [ -n "$SVC" ]; then
            grep -vxF "$SVC" "$STATE" > "$STATE.t" 2>/dev/null
            mv "$STATE.t" "$STATE" 2>/dev/null
        fi
        exit 0 ;;
    *) exit 0 ;;
esac
EOF

mk ip <<'EOF'
#!/bin/bash
echo "ip $*" >> "$STUB_LOG"
# deploy-tune.sh 用它找默认路由的网卡
if [ "$1" = "-4" ]; then
    echo "default via 192.168.1.1 dev wlan0 proto dhcp metric 600"
fi
exit 0
EOF

mk iptables <<'EOF'
#!/bin/bash
# 桩：真的维护一张规则表。
# 如果这里偷懒一律返回 0，那么 deploy-tune.sh 里
# 「-C 查存在 -> -D 删除」的循环会永远转下去（真踩过这个坑）。
echo "iptables $*" >> "$STUB_LOG"
RULES="$STUB_LOG.rules"
touch "$RULES"

# -L 用来数规则条数
if [ "${1:-}" = "-t" ] && [ "${3:-}" = "-L" ]; then
    while read -r r; do
        [ -n "$r" ] && echo "TCPMSS  tcp --  $r"
    done < "$RULES"
    exit 0
fi

chain=""; mss=""; action=""
prev=""
for a in "$@"; do
    case "$prev" in
        -C|-D|-A) chain="$a" ;;
        --set-mss) mss="$a" ;;
    esac
    case "$a" in
        -C|-D|-A) [ -z "$action" ] && action="$a" ;;
    esac
    prev="$a"
done
key="$chain:$mss"

case "$action" in
    -C) grep -qxF "$key" "$RULES" 2>/dev/null && exit 0 || exit 1 ;;
    -A) grep -qxF "$key" "$RULES" 2>/dev/null || echo "$key" >> "$RULES"; exit 0 ;;
    -D)
        if grep -qxF "$key" "$RULES" 2>/dev/null; then
            grep -vxF "$key" "$RULES" > "$RULES.tmp" 2>/dev/null
            mv "$RULES.tmp" "$RULES"
            exit 0
        fi
        exit 1 ;;
esac
exit 0
EOF

mk nmcli <<'EOF'
#!/bin/bash
echo "nmcli $*" >> "$STUB_LOG"
exit 0
EOF

mk getent <<'EOF'
#!/bin/bash
exit 0
EOF

mk free <<'EOF'
#!/bin/bash
echo "              total        used        free"
echo "Mem:            460         131         144"
EOF

mk cloudflared <<'EOF'
#!/bin/bash
if [ "$1" = "--version" ]; then
    echo "cloudflared version 2026.9.3 (stub)"
    exit 0
fi
echo "cloudflared $*" >> "$STUB_LOG"
exit 0
EOF

mk curl <<'EOF'
#!/bin/bash
echo "curl $*" >> "$STUB_LOG"
# 健康检查要返回 {"ok": true}
for a in "$@"; do
    case "$a" in
        *v1/health*) echo '{"ok": true, "version": "stub"}'; exit 0 ;;
    esac
done
# -o FILE 的下载：写点东西假装成功
prev=""
for a in "$@"; do
    if [ "$prev" = "-o" ]; then echo "stub-binary" > "$a"; exit 0; fi
    prev="$a"
done
exit 0
EOF

# python3 在很多 Windows 上是应用商店的占位符，用真的 python 转一手
if [ -n "$REAL_PY" ]; then
    mk python3 <<EOF
#!/bin/bash
exec "$REAL_PY" "\$@"
EOF
fi

export PATH="$STUB:$PATH"
echo "  沙箱:   $TMP"
echo "  安装根: $SANDBOX"
echo "  桩命令: $(ls "$STUB" | tr '\n' ' ')"

# 桩还在不在？每节开头都查一次。
# 桩目录一旦丢了，后面所有"没启用 / 没生成 / 退出码不对"的断言
# 都会变成假失败，而且报错信息完全指不到真正的原因 ——
# 实测踩过：桩被删掉之后，deploy.sh 找不到 cloudflared 就去
# 真的下载 GitHub，整个测试卡死十分钟。
stub_ok() {
    local where="$1"
    if [ -d "$STUB" ] && [ "$(command -v systemctl 2>/dev/null)" = "$STUB/systemctl" ]; then
        ok "桩完好（$where）"
    else
        bad "桩丢了或 PATH 被改（$where）:
         STUB 存在吗: $([ -d "$STUB" ] && echo 是 || echo 否)
         systemctl 解析到: $(command -v systemctl 2>/dev/null || echo 空)
         PATH 前 80 字符: ${PATH:0:80}"
    fi
}

# ════════════════════════════════════════════════════════════════
echo
echo "── 5. 第一次部署（--skip-tunnel，不碰隧道）──"
OUT1="$TMP/run1.log"
bash "$DEPLOY" --prefix "$SANDBOX" --skip-tunnel --port 8080 > "$OUT1" 2>&1
RC1=$?
check "第一次退出码为 0" "$RC1" "0"
if [ "$RC1" != "0" ]; then
    echo "    ── 输出末尾 ──"
    tail -20 "$OUT1" | sed 's/^/      /'
fi

REL="$SANDBOX/root/lanremote-relay"
UNIT="$SANDBOX/etc/systemd/system"

# 把第一遍的桩日志留一份 —— 第二遍会被清空，
# 但"第一次装的时候确实做了 daemon-reload"这事得留证据。
cp "$STUB_LOG" "$STUB_LOG.run1" 2>/dev/null || true

exists "relay.py 已部署"        "$REL/relay.py"
exists "admin.py 已部署"        "$REL/admin.py"
exists "landing.py 已部署"      "$REL/landing.py"
exists "showcase.py 已部署"     "$REL/showcase.py"
exists "favicon.ico 已部署"     "$REL/favicon.ico"
exists "deploy-tune.sh 已部署"  "$REL/deploy-tune.sh"
exists "watchdog.sh 已部署"     "$REL/watchdog.sh"
exists "site_shots 已部署"      "$REL/site_shots/pc-main.png"
exists "data 目录已建"          "$REL/data"
exists "downloads 目录已建"     "$REL/downloads"

for u in lanremote-relay lanremote-nettune lanremote-tunnel; do
    exists "$u.service 已生成" "$UNIT/$u.service"
done
exists "备用 http2 单元已生成" "$UNIT/lanremote-tunnel-h2.service"
exists "看门狗 timer 已生成"   "$UNIT/lanremote-watchdog.timer"

echo
echo "     生成的 relay unit 里应该有的关键项:"
for kw in "RELAY_POLL=60" "RELAY_PORT=8080" "python3"; do
    if grep -q "$kw" "$UNIT/lanremote-relay.service" 2>/dev/null; then
        ok "relay unit 含 $kw"
    else
        bad "relay unit 缺 $kw"
    fi
done
echo
echo "     两个隧道单元的内容（默认主 quic+8 / 副本 http2+2）:"
# 主隧道
if grep -q -- '--protocol quic' "$UNIT/lanremote-tunnel.service" 2>/dev/null; then
    ok "主隧道单元用 quic"
else
    bad "主隧道单元没写 quic"
fi
if grep -q -- '--ha-connections 8' "$UNIT/lanremote-tunnel.service" 2>/dev/null; then
    ok "主隧道 8 条连接（实测最优配比）"
else
    bad "主隧道连接数不是 8"
fi
# 副本隧道
if grep -q -- '--protocol http2' "$UNIT/lanremote-tunnel-h2.service" 2>/dev/null; then
    ok "副本隧道单元用 http2"
else
    bad "副本隧道单元没写 http2"
fi
if grep -q -- '--ha-connections 2' "$UNIT/lanremote-tunnel-h2.service" 2>/dev/null; then
    ok "副本隧道 2 条连接（实测最优配比）"
else
    bad "副本隧道连接数不是 2"
fi
# ★ metrics 端口必须显式钉死且不重复
#   两边都用默认 20241 的话，停掉主实例再启副实例时副实例会抢到 20241，
#   所有按端口读指标的脚本就读错实例了，而且不报错。
if grep -q -- '--metrics 127.0.0.1:20241' "$UNIT/lanremote-tunnel.service" 2>/dev/null; then
    ok "主隧道 metrics 钉死 20241"
else
    bad "主隧道没显式指定 metrics 端口"
fi
if grep -q -- '--metrics 127.0.0.1:20242' "$UNIT/lanremote-tunnel-h2.service" 2>/dev/null; then
    ok "副本隧道 metrics 钉死 20242"
else
    bad "副本隧道没显式指定 metrics 端口"
fi
if grep -q 'ExecStart=/bin/bash' "$UNIT/lanremote-nettune.service" 2>/dev/null; then
    ok "nettune 只有一个入口脚本"
else
    bad "nettune 单元不对"
fi
# Token 不能出现在 unit 里（unit 是 644 的）
if grep -q 'TUNNEL_TOKEN=' "$UNIT/lanremote-tunnel.service" 2>/dev/null; then
    bad "unit 里竟然有 TUNNEL_TOKEN（会泄露）"
else
    ok "unit 里没有明文 Token"
fi

# 便捷命令
for c in lanremote-url lanremote-key lanremote-status; do
    exists "便捷命令 $c 已生成" "$SANDBOX/usr/local/bin/$c"
done

# ════════════════════════════════════════════════════════════════
echo
echo "── 6. 幂等性：放几个「不许被冲掉」的东西，再跑第二遍 ──"

# 设备数据、管理密码、已有的公网地址 —— 这些重跑时必须原样保留
echo "device-sentinel"   > "$REL/data/devices.json"
echo "key-sentinel"      > "$REL/data/admin_key.txt"
echo "pw-sentinel"       > "$REL/data/admin_password.txt"
echo "https://keep.me"   > "$REL/public_url.txt"
# allowNew 是管理员在管理页上拨的开关，存在 relay_state.json 里。
# 用户重跑脚本升级时**绝对不能**把它拨回去 —— 那是安全设置。
cat > "$REL/data/relay_state.json" <<'EOF'
{
  "allowNew": false,
  "codes": [
    {"deviceId": "desktop-sentinel-abc123", "code": "789113"}
  ]
}
EOF
STATE_BEFORE=$(md5sum "$REL/data/relay_state.json" | awk '{print $1}')
# 手工改过的 config.yml 也不能被无脑覆盖
mkdir -p "$SANDBOX/root/.cloudflared"
cat > "$SANDBOX/root/.cloudflared/config.yml" <<'EOF'
# 手工改过的配置，重跑不许动
tunnel: 11111111-2222-3333-4444-555555555555
ingress:
  - hostname: hand-edited.example.com
    service: http://127.0.0.1:8080
  - service: http_status:404
EOF
BEFORE_HASH=$(md5sum "$SANDBOX/root/.cloudflared/config.yml" | awk '{print $1}')
# 记下部署文件的指纹，第二遍后应该一致（内容相同）
BEFORE_RELAY=$(md5sum "$REL/relay.py" | awk '{print $1}')
# 清空桩日志，好判断第二遍**有没有多此一举地重启服务**
: > "$STUB_LOG"

OUT2="$TMP/run2.log"
bash "$DEPLOY" --prefix "$SANDBOX" --skip-tunnel --port 8080 > "$OUT2" 2>&1
RC2=$?
check "第二次退出码为 0" "$RC2" "0"
if [ "$RC2" != "0" ]; then
    echo "    ── 输出末尾 ──"
    tail -20 "$OUT2" | sed 's/^/      /'
fi

check "设备数据没被冲掉"  "$(cat "$REL/data/devices.json" 2>/dev/null)"      "device-sentinel"
check "管理钥匙没被冲掉"  "$(cat "$REL/data/admin_key.txt" 2>/dev/null)"     "key-sentinel"
check "管理密码没被冲掉"  "$(cat "$REL/data/admin_password.txt" 2>/dev/null)" "pw-sentinel"
check "公网地址没被冲掉"  "$(cat "$REL/public_url.txt" 2>/dev/null)"         "https://keep.me"
# ★ allowNew 必须原样保留
STATE_AFTER=$(md5sum "$REL/data/relay_state.json" | awk '{print $1}')
check "relay_state.json 没被改（allowNew 保住了）" "$STATE_AFTER" "$STATE_BEFORE"
if grep -q '"allowNew": false' "$REL/data/relay_state.json" 2>/dev/null; then
    ok "allowNew 仍然是 false（没被拨回 true）"
else
    bad "allowNew 被改了"
fi
AFTER_HASH=$(md5sum "$SANDBOX/root/.cloudflared/config.yml" | awk '{print $1}')
check "手工 config.yml 没被覆盖" "$AFTER_HASH" "$BEFORE_HASH"
AFTER_RELAY=$(md5sum "$REL/relay.py" | awk '{print $1}')
check "relay.py 内容稳定" "$AFTER_RELAY" "$BEFORE_RELAY"

# 两次输出里的「警告」数量应该差不多 —— 第二遍不应该冒出新的错误
W1=$(grep -c '✗' "$OUT1" 2>/dev/null || echo 0)
W2=$(grep -c '✗' "$OUT2" 2>/dev/null || echo 0)
check "第二遍没有新增 ✗ 错误" "$W2" "$W1"

# ────────────────────────────────────────────────────────────────
echo
echo "── 6b. 没变化时不该重启服务（别打断在线连接）──"
# 第二遍什么都没变，桩日志里就不该出现 restart。
# 用户是在**已经跑着的机器上**重跑本脚本升级的，
# 每次无条件 restart 会清掉内存里的设备注册和会话，
# 正在遥控的连接要几秒才重新接上。
if grep -q 'systemctl restart lanremote-relay' "$STUB_LOG" 2>/dev/null; then
    bad "第二遍无变化却重启了中转（会打断在线连接）"
    grep 'restart' "$STUB_LOG" | head -3 | sed 's/^/      /'
else
    ok "第二遍没有重启中转（无变化）"
fi
if grep -q 'systemctl restart lanremote-tunnel' "$STUB_LOG" 2>/dev/null; then
    bad "第二遍无变化却重启了隧道"
else
    ok "第二遍没有重启隧道（无变化）"
fi
if grep -q 'systemctl daemon-reload' "$STUB_LOG" 2>/dev/null; then
    bad "第二遍无变化却做了 daemon-reload"
else
    ok "第二遍没有多余的 daemon-reload"
fi
if grep -q '内容没变' "$OUT2" 2>/dev/null; then
    ok "输出里明确说了「内容没变、保持运行」"
else
    bad "输出没说明为什么不重启"
fi

# ════════════════════════════════════════════════════════════════
echo
echo "── 7. 凭据模式：给出域名时要能生成 config.yml ──"
CFD="$SANDBOX/root/.cloudflared"
rm -f "$CFD/config.yml"
echo '{"TunnelID":"abcdef01-2345-6789-abcd-ef0123456789"}' \
    > "$CFD/abcdef01-2345-6789-abcd-ef0123456789.json"

OUT3="$TMP/run3.log"
bash "$DEPLOY" --prefix "$SANDBOX" --hostname relay.test.example \
     > "$OUT3" 2>&1
RC3=$?
check "凭据模式退出码为 0" "$RC3" "0"
exists "config.yml 已生成" "$CFD/config.yml"
if grep -q 'abcdef01-2345-6789-abcd-ef0123456789' "$CFD/config.yml" 2>/dev/null; then
    ok "config.yml 里的隧道 ID 取自凭据文件名"
else
    bad "config.yml 隧道 ID 不对"
fi
if grep -q 'relay.test.example' "$CFD/config.yml" 2>/dev/null; then
    ok "config.yml 写入了域名"
else
    bad "config.yml 缺域名"
fi
if grep -q 'http_status:404' "$CFD/config.yml" 2>/dev/null; then
    ok "config.yml 有兜底 404 规则（没有它 cloudflared 起不来）"
else
    bad "config.yml 缺兜底规则"
fi
if grep -q '127.0.0.1:8080' "$CFD/config.yml" 2>/dev/null; then
    ok "config.yml 指向本地中转端口"
else
    bad "config.yml 回源地址不对"
fi

# ────────────────────────────────────────────────────────────────
echo
echo "── 7c. 默认：两个隧道单元都要启用（协议级故障隔离）──"
stub_ok "7c"
# 真机实测（40 次/组）：quic 8 + h2 2 拿到 100% 可用率 + 最低 P90/P99。
# 单独跑一个协议都不如两个一起跑：
#   QUIC 握手要 6+ 个包往返，TCP SYN 只要 3 个且激进重传，
#   约 5% 丢包下 (1-p)^6 ≈ 72%（不是"运营商封 UDP"，那是错的测量）
#   -> quic 拨号成功率只有约 72%，需要 http2 兜住失败的那部分。
: > "$STUB_LOG"
printf 'lanremote-relay\n' > "$STUB_LOG.state"
bash "$DEPLOY" --prefix "$SANDBOX" > "$TMP/run7c.log" 2>&1
check "默认（不传 --protocol）退出码为 0" "$?" "0"
if grep -qxF 'systemctl enable lanremote-tunnel' "$STUB_LOG" 2>/dev/null; then
    ok "默认启用了主隧道"
else
    bad "默认没启用主隧道"
fi
if grep -qxF 'systemctl enable lanremote-tunnel-h2' "$STUB_LOG" 2>/dev/null; then
    ok "默认启用了副本隧道"
else
    bad "默认没启用副本隧道（双跑才对）"
fi
if grep -qxF 'systemctl disable --now lanremote-tunnel' "$STUB_LOG" 2>/dev/null \
   || grep -qxF 'systemctl disable --now lanremote-tunnel-h2' "$STUB_LOG" 2>/dev/null; then
    bad "默认把某个隧道关掉了（应该两个都跑）"
else
    ok "默认没有关掉任何一个隧道"
fi

echo
echo "── 7d. --protocol quic：只跑主隧道 ──"
stub_ok "7d"
: > "$STUB_LOG"
printf 'lanremote-relay\n' > "$STUB_LOG.state"
bash "$DEPLOY" --prefix "$SANDBOX" --protocol quic \
     > "$TMP/run7d.log" 2>&1
check "--protocol quic 退出码为 0" "$?" "0"
if grep -qxF 'systemctl enable lanremote-tunnel' "$STUB_LOG" 2>/dev/null; then
    ok "quic 模式启用了主隧道"
else
    bad "quic 模式没启用主隧道"
fi
if grep -qxF 'systemctl disable --now lanremote-tunnel-h2' "$STUB_LOG" 2>/dev/null; then
    ok "quic 模式停用了副本隧道"
else
    bad "quic 模式没停用副本隧道"
fi
if grep -q '只跑 quic' "$TMP/run7d.log" 2>/dev/null; then
    ok "输出说明了为什么停用副本"
else
    bad "输出没说清楚"
fi

echo
echo "── 7e. --protocol http2：只跑副本隧道 ──"
stub_ok "7e"
: > "$STUB_LOG"
printf 'lanremote-relay\n' > "$STUB_LOG.state"
bash "$DEPLOY" --prefix "$SANDBOX" --protocol http2 \
     > "$TMP/run7e.log" 2>&1
check "--protocol http2 退出码为 0" "$?" "0"
if grep -qxF 'systemctl enable lanremote-tunnel-h2' "$STUB_LOG" 2>/dev/null; then
    ok "http2 模式启用了副本隧道"
else
    bad "http2 模式没启用副本隧道"
fi
if grep -qxF 'systemctl disable --now lanremote-tunnel' "$STUB_LOG" 2>/dev/null; then
    ok "http2 模式停用了主隧道"
else
    bad "http2 模式没停用主隧道"
fi

echo
echo "── 7f. --protocol 的非法值要报错 ──"
stub_ok "7f"
bash "$DEPLOY" --prefix "$SANDBOX" --protocol udp >/dev/null 2>&1
check "非法协议退出码为 2" "$?" "2"
bash "$DEPLOY" --prefix "$SANDBOX" --protocol both >/dev/null 2>&1
check "both 是合法值" "$?" "0"

echo
echo "── 7g. --port 换端口时 config.yml 必须跟着改 ──"
stub_ok "7g"
# 只改 unit 里的 RELAY_PORT、不改 config.yml 的话，
# 隧道会回源到老端口（没人听），现象是"隧道连上了但一访问就 502"。
if grep -q '127.0.0.1:8080' "$CFD/config.yml" 2>/dev/null; then
    ok "换端口前 config.yml 是 8080"
else
    bad "换端口前的基准不对"
fi
: > "$STUB_LOG"
printf 'lanremote-relay\n' > "$STUB_LOG.state"
bash "$DEPLOY" --prefix "$SANDBOX" --port 9090 > "$TMP/run7g.log" 2>&1
check "--port 9090 退出码为 0" "$?" "0"
if grep -q '127.0.0.1:9090' "$CFD/config.yml" 2>/dev/null; then
    ok "config.yml 的回源端口跟到了 9090"
else
    bad "config.yml 没跟着换端口（隧道会回源到错端口）: $(grep -m1 'service:' "$CFD/config.yml")"
fi
if grep -q 'RELAY_PORT=9090' "$UNIT/lanremote-relay.service" 2>/dev/null; then
    ok "relay unit 的端口也换成了 9090"
else
    bad "relay unit 端口没换"
fi
# 换回 8080，后面的断言还要用
bash "$DEPLOY" --prefix "$SANDBOX" --port 8080 >/dev/null 2>&1
if grep -q '127.0.0.1:8080' "$CFD/config.yml" 2>/dev/null; then
    ok "换回 8080 也跟上了"
else
    bad "换回 8080 没跟上"
fi

echo
echo "── 7b. 再跑一次（不给 --hostname），config.yml 不该被改 ──"
stub_ok "7b"
H3=$(md5sum "$CFD/config.yml" | awk '{print $1}')
bash "$DEPLOY" --prefix "$SANDBOX" > "$TMP/run4.log" 2>&1
RC4=$?
check "再次运行退出码为 0" "$RC4" "0"
H4=$(md5sum "$CFD/config.yml" | awk '{print $1}')
check "没给 --hostname 时 config.yml 保持不变" "$H4" "$H3"

# ════════════════════════════════════════════════════════════════
echo
echo "── 8. Token 模式 ──"
OUT5="$TMP/run5.log"
TEST_TOKEN="eyJhIjoiZmFrZS10b2tlbi1mb3ItdGVzdCJ9"
rm -f "$CFD/tunnel.env"
TUNNEL_TOKEN="$TEST_TOKEN" \
    bash "$DEPLOY" --prefix "$SANDBOX" > "$OUT5" 2>&1
RC5=$?
check "Token 模式退出码为 0" "$RC5" "0"
exists "Token 写进了 tunnel.env" "$CFD/tunnel.env"
check "tunnel.env 内容正确" \
      "$(cat "$CFD/tunnel.env" 2>/dev/null)" "TUNNEL_TOKEN=$TEST_TOKEN"
if grep -q "$TEST_TOKEN" "$UNIT/lanremote-tunnel.service" 2>/dev/null; then
    bad "unit 里出现了明文 Token"
else
    ok "unit 里没有明文 Token（只有 EnvironmentFile 引用）"
fi
if grep -q 'EnvironmentFile' "$UNIT/lanremote-tunnel.service" 2>/dev/null; then
    ok "unit 用了 EnvironmentFile 读 Token"
else
    bad "unit 缺 EnvironmentFile"
fi
if grep -q -- '--token' "$UNIT/lanremote-tunnel.service" 2>/dev/null; then
    ok "unit 的 ExecStart 里用 --token \${TUNNEL_TOKEN}"
else
    bad "unit 的 ExecStart 没走 Token 模式"
fi

echo
echo "── 8b. tunnel.env 权限应该是 600 ──"
PERM=$(stat -c '%a' "$CFD/tunnel.env" 2>/dev/null || echo "?")
# Git Bash 上 chmod 的语义和 Linux 不完全一样，这里只做提示
echo "      tunnel.env 权限: $PERM（Linux 上应为 600）"

# ════════════════════════════════════════════════════════════════
echo
echo "── 9. 缺文件时要明确报错，不能静默 ──"
BROKEN="$TMP/broken"
mkdir -p "$BROKEN"
cp "$REPO/relay/relay.py" "$BROKEN/"
# 故意少给 admin.py / landing.py / showcase.py / favicon.ico / site_shots
OUT6="$TMP/run6.log"
bash "$BROKEN/../broken/relay.py" --help >/dev/null 2>&1   # 只是占位，别在意
python - "$DEPLOY" "$BROKEN" <<'PYEOF' > "$TMP/run6.log" 2>&1
import shutil, subprocess, sys
from pathlib import Path
deploy, broken = sys.argv[1], sys.argv[2]
# 把 deploy.sh 拷到「残缺的包」里再跑
shutil.copy2(deploy, Path(broken) / "deploy.sh")
r = subprocess.run(["bash", str(Path(broken) / "deploy.sh"),
                    "--prefix", broken + "/sandbox", "--skip-tunnel"],
                   capture_output=True, text=True, encoding="utf-8",
                   errors="replace")
print(r.stdout or "")
print(r.stderr or "")
sys.exit(r.returncode)
PYEOF
RC6=$?
if [ "$RC6" != "0" ]; then
    ok "文件不全时明确失败（退出码 $RC6）"
else
    bad "文件不全竟然还成功了"
fi
if grep -q '程序文件不全' "$TMP/run6.log" 2>/dev/null; then
    ok "报错信息说清了是「程序文件不全」"
else
    bad "报错信息不够明确: $(tail -3 "$TMP/run6.log" 2>/dev/null | tr '\n' ' ')"
fi

# ════════════════════════════════════════════════════════════════
echo
echo "── 10. 桩命令确实被调用过（证明流程真的走了）──"
for kw in "systemctl enable" "iptables" ; do
    if grep -q "$kw" "$STUB_LOG" 2>/dev/null; then
        ok "调用过 $kw"
    else
        bad "没调用过 $kw"
    fi
done
# daemon-reload 只在 unit 真变了的时候才该发生，
# 所以拿第一遍的日志来验（第一遍是全新安装，unit 肯定是新的）
if grep -q 'systemctl daemon-reload' "$STUB_LOG.run1" 2>/dev/null; then
    ok "第一次安装做了 daemon-reload"
else
    bad "第一次安装没做 daemon-reload"
fi

# ════════════════════════════════════════════════════════════════
echo
echo "── 11. 脚本里不许有写死的域名 ──"
# 这个中转是自用的。脚本里烧一个域名进去，别人一照抄就指到别人机器上了。
for f in "$DEPLOY" "$TUNE" "$REPO/relay/net-tune.sh"; do
    base=$(basename "$f")
    if grep -qE 'relay\.[a-z0-9-]+\.(top|com|net|org|cn)' "$f" 2>/dev/null; then
        bad "$base 里有写死的真实域名: $(grep -oE 'relay\.[a-z0-9-]+\.(top|com|net|org|cn)' "$f" | head -1)"
    else
        ok "$base 里没有写死的域名"
    fi
done
# 示例里的占位符必须是「你的域名」这种明显是占位的写法
if grep -q '你的域名' "$DEPLOY" 2>/dev/null; then
    ok "示例用的是「你的域名」占位符"
else
    bad "示例没看到占位符"
fi

# ════════════════════════════════════════════════════════════════
echo
echo "── 12. 有凭据但没给域名 → 不许乱猜，必须明确提示 ──"
stub_ok "12"
CFD2="$TMP/nohost/root/.cloudflared"
SANDBOX2="$TMP/nohost"
mkdir -p "$CFD2"
echo '{"TunnelID":"deadbeef-0000-1111-2222-333344445555"}' \
    > "$CFD2/deadbeef-0000-1111-2222-333344445555.json"
OUT7="$TMP/run7.log"
: > "$STUB_LOG"
# 这是台"新机器"：中转在跑，隧道从没配过。
# 把桩的服务状态重置成这个样子 —— 否则前面几节把隧道标成 active 了，
# net-tune.sh 会以为它在跑然后去重启，断言就测不准。
printf 'lanremote-relay\n' > "$STUB_LOG.state"
bash "$DEPLOY" --prefix "$SANDBOX2" > "$OUT7" 2>&1
RC7=$?
check "没域名时退出码为 0（中转照装，只是隧道不启）" "$RC7" "0"
if [ -f "$CFD2/config.yml" ]; then
    bad "没域名却凭空生成了 config.yml（内容是: $(head -3 "$CFD2/config.yml" | tr '\n' ' '))"
else
    ok "没域名时没有生成 config.yml（没有乱猜）"
fi
if grep -q '需要你提供公网域名' "$OUT7" 2>/dev/null; then
    ok "明确提示了「需要你提供公网域名」"
else
    bad "提示不够明确"
fi
if grep -q -- '--hostname' "$OUT7" 2>/dev/null; then
    ok "提示里给了怎么补域名"
else
    bad "提示里没说怎么补"
fi
if grep -q 'TUNNEL_TOKEN' "$OUT7" 2>/dev/null; then
    ok "提示里给了 Token 模式这个替代办法"
else
    bad "提示里没提 Token 模式"
fi
if grep -q 'systemctl restart lanremote-tunnel' "$STUB_LOG" 2>/dev/null; then
    bad "隧道没配好却还是启动了隧道服务"
else
    ok "隧道没配好时没有去启动它"
fi
# 但中转本身要照常起来
if grep -q 'systemctl enable lanremote-relay' "$STUB_LOG" 2>/dev/null; then
    ok "中转仍然正常安装并启用"
else
    bad "中转没被启用"
fi

# ════════════════════════════════════════════════════════════════
echo
echo "============================================================"
echo "  结果: 通过 $PASS 项，失败 $FAIL 项"
echo "============================================================"
if [ "$FAIL" = "0" ]; then
    echo "  沙箱目录（想自己看就留着）: $TMP"
    rm -rf "$TMP"
    exit 0
else
    echo "  沙箱目录保留供排查: $TMP"
    echo
    echo "  ── 第一次部署输出（末尾 25 行）──"
    tail -25 "$OUT1" 2>/dev/null | sed 's/^/    /'
    exit 1
fi