fix(plugins): 出站主机授权两处真缺陷 —— 已启动 worker 看不到新授权、授权不落盘

生产实证:channels 插件的 auth/state 在主机已授权后仍被拒
(auth/state 失败: host "copilot.tencent.com" is not an allowed outbound destination),
而 /api/plugins/channels/kernel 明明显示 hosts_granted=26。

缺陷一(授权对已启动 worker 无效):
  httpAllowlist 是不可变快照;bootPluginState 在 worker 启动时把指针绑进 Lua
  闭包,闭包与 worker 同寿。GrantExtraHosts 之后 ps.allowHTTP 被**整个换新**
  (newHTTPAllowlist 返回新实例),已启动的 worker 仍持旧指针 → 授权接口返回
  ok、设置页显示已批,请求却继续被拒。分片 VM(plugin_shard.go)同样中招。
  修法:allowlist 改为带 RWMutex 的**可变共享对象**,GrantExtraHosts 与
  rebuildOutboundAllowlistLocked 原地 add —— 所有闭包立即看到新授权,worker
  无需重建。rebuild 从"替换实例"改为"原地折叠",并在注释里写明为什么。

缺陷二(授权不持久化):
  .kernel-perms.json 只写 actions;hosts 授权只存在运行态 map,重启即失。
  设置页看不出异常(actions 有持久化,读起来像"都持久化了")。
  修法:文件行改为 jsonPluginPolicy{actions,all_read,hosts}(旧文件向后兼容,
  纯 actions 行照常解析);Load 恢复 extraHosts;Save 把每个插件的 host 授权
  与 actions 一起写入(支持"只批了主机、没批动作"的行)。

回归测试(internal/lua/plugin_http_test.go):
- TestGrantExtraHostsVisibleToAlreadyBootedWorker:按事故顺序构造(先启动
  worker → 再授权),断言同一批 worker 立即可用新授权 + 旧引用不再失效。
- TestHostGrantsPersistAcrossRestart:授权落盘 + 新 registry 读回 + 插件重载
  后授权保留。

go test ./internal/lua/ 全绿;build/vet 通过。
This commit is contained in:
JianFeeeee
2026-10-13 06:57:28 +08:00
parent 84a9fccd9a
commit a07449d32e
5 changed files with 229 additions and 11 deletions

View File

@ -43,6 +43,7 @@ import (
"net/http"
"net/url"
"strings"
"sync"
"time"
golua "github.com/aarzilli/golua/lua"
@ -80,19 +81,39 @@ var httpClient = &http.Client{
const httpGlobal = "__llmsproxy_http"
// httpAllowlist is the set of hosts plugins may contact: the host of every
// configured source's base_url.
// configured source's base_url, plus every operator-granted extra host.
//
// It is a LIVE, shared object, not a snapshot. This matters: bootPluginState
// binds one *httpAllowlist into every Lua closure at state-boot time, and those
// closures live for the plugin's lifetime. Granting an extra host after a
// worker booted must therefore MUTATE the same object the closures hold —
// replacing ps.allowHTTP with a fresh instance would leave every already-booted
// worker checking the STALE list, and a grant would look accepted (the API
// answers ok, the UI shows it) while requests kept getting refused with
// "not an allowed outbound destination". That exact failure shipped once.
//
// Mutations are rare (operator approves a host); lookups are per request.
type httpAllowlist struct {
mu sync.RWMutex
hosts map[string]bool
}
func newHTTPAllowlist(hosts []string) *httpAllowlist {
m := make(map[string]bool, len(hosts))
a := &httpAllowlist{hosts: make(map[string]bool, len(hosts))}
a.add(hosts)
return a
}
// add merges hosts into the live allowlist. Called by rebuild (source hosts
// change on hot reload) and by GrantExtraHosts (operator approves a host).
func (a *httpAllowlist) add(hosts []string) {
a.mu.Lock()
defer a.mu.Unlock()
for _, h := range hosts {
if h != "" {
m[strings.ToLower(h)] = true
a.hosts[strings.ToLower(h)] = true
}
}
return &httpAllowlist{hosts: m}
}
// allows reports whether host:port may be contacted. Ports 80/443 are implied
@ -103,6 +124,8 @@ func (a *httpAllowlist) allows(host string, port string) bool {
if host == "" {
return false
}
a.mu.RLock()
defer a.mu.RUnlock()
if a.hosts[host] {
return true
}