服务端 zip 加一键部署脚本(cloudflared 安装 + 部署 + 公网,全自动)

用户要求:服务器端文件里放一个 .sh,双击就能检查 cloudflared 装没装、
自动部署、配好公网,以后全自动。

== 新增 ==

relay/deploy.sh        一键部署主脚本,7 步全自动:
                       环境检查 -> 装 cloudflared -> 部署程序 -> 网络调优
                       -> 写 4 个 systemd 服务 -> 配隧道 -> 启动自检打公网地址
relay/deploy-tune.sh   网络调优,可单独重跑:
                       keepalive 30/10/3 + 路径 MTU 探测 + MSS 1360
                       + 接口 MTU 1400 + mask networkd-wait-online
                       每一项都写明了实测依据

用法:
    bash deploy.sh                           复用已有凭据,全自动
    TUNNEL_TOKEN=eyJh... bash deploy.sh      Token 模式
    bash deploy.sh --hostname relay.你的域名  换域名
    bash deploy.sh --skip-tunnel             只装中转
    bash deploy.sh --help                    全部参数

重跑 = 升级;已有数据/配对码/管理密码/凭据/手改过的 config.yml 都不动。

== 自动化到什么程度 ==

* cloudflared 不在就自动下载装上(并校验能执行)
* 隧道支持两种模式:已有凭据文件、或 TUNNEL_TOKEN
* **不传 --hostname 且没有 Token 时明确提示要填**,不会静默用一个写死的值
  (用户明确说过:这个中转站是他自己用的,不对外开放,
   绝不能把私人域名烧成默认值)
* deploy-tune.sh 的 MTU 取 1400:路径 MTU 实测 1472,
  而 MSS 钳制只管 TCP、QUIC 走 UDP 完全不受保护;
  接口 MTU 降到 1400 后 QUIC 建齐 10 条连接从 28 秒缩到 12 秒

== 顺带修掉两个真 bug ==

1. write_tunnel_unit 写的文件名漏了 .service 后缀 ——
   文件写下去了但 systemd 完全看不见,enable 还会安静地失败。
2. 发布包一直缺 site_shots/(老 bug):
   package_remote.py 用 iterdir() + is_file() 收文件,目录进不去。
   新机器装完官网截图全是 404。只有从 zip 解压真跑一遍才暴露。
   同时把 deploy.sh 里 site_shots 的检查降级成警告(缺截图不该让部署失败)。

另外给 deploy-tune.sh 清旧 MSS 规则的循环加了 10 次上限 ——
原写法在 -D 报成功但规则仍在时会死循环把整个部署卡死。

== 验证 ==

* _build/test_deploy_sh.sh        沙箱 65 项,全通过
* _build/test_zip_deploy.sh       从发布 zip 解压再跑,30 项,全通过
* _build/test_deploy_real.sh      **真机 Debian 上实测**(新增):
    zip 解出 17 个文件 -> 语法/LF/无 BOM 通过
    --help 退出码 0 且无副作用、未知参数退出码 2
    沙箱安装成功、幂等重跑哨兵不被冲
    现有三个中转服务全程 active,自检 200
This commit is contained in:
xmz7891
2026-10-07 17:28:09 +08:00
parent d4f335b8f9
commit facb156aca
10 changed files with 2331 additions and 89 deletions
+494
View File
@@ -0,0 +1,494 @@
#!/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
echo "systemctl $*" >> "$STUB_LOG"
case "$1" in
is-active) echo "active"; exit 0 ;;
is-enabled) echo "enabled"; 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' ' ')"
# ════════════════════════════════════════════════════════════════
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"
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 $kw"
fi
done
if grep -q 'http2' "$UNIT/lanremote-tunnel-h2.service" 2>/dev/null; then
ok "h2 单元用的是 http2"
else
bad "h2 单元没写 http2"
fi
if grep -q 'quic' "$UNIT/lanremote-tunnel.service" 2>/dev/null; then
ok "主隧道单元用的是 quic"
else
bad "主隧道单元没写 quic"
fi
if grep -q 'ha-connections 10' "$UNIT/lanremote-tunnel.service" 2>/dev/null; then
ok "隧道单元含 ha-connections 10"
else
bad "隧道单元缺 ha-connections 10"
fi
if grep -q 'ExecStart=/bin/bash' "$UNIT/lanremote-nettune.service" 2>/dev/null; then
ok "nettune 顺序执行两个调优脚本"
else
bad "nettune 单元不对"
fi
# Token 不能出现在 unit 里(unit 是 644 的)
if grep -q 'TUNNEL_TOKEN=' "$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" > "$REL/public_url.txt"
# 手工改过的 config.yml 也不能被无脑覆盖
mkdir -p "$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 $1}')
OUT2="$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 "$REL/public_url.txt" 2>/dev/null)" "https://keep.me"
AFTER_HASH=$(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" "$W1"
# ════════════════════════════════════════════════════════════════
echo
echo "── 7. 凭据模式:给出域名时要能生成 config.yml ──"
CFD="$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' "$CFD/config.yml" 2>/dev/null; then
ok "config.yml 指向本地中转端口"
else
bad "config.yml 回源地址不对"
fi
echo
echo "── 7b. 再跑一次(不给 --hostname),config.yml 不该被改 ──"
H3=$(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 daemon-reload" "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
# ════════════════════════════════════════════════════════════════
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