release: v1.2.2 — 双域名配置与 Host 隔离修复
GitHub Release / release (push) Canceled after 0s
Publish Docker image / build (linux/amd64, ubuntu-latest, amd64) (push) Canceled after 0s
Publish Docker image / build (linux/arm64, ubuntu-24.04-arm, arm64) (push) Canceled after 0s
Publish Docker image / merge (push) Canceled after 0s
ci / ci (22, ubuntu-latest) (push) Canceled after 0s

修复设置页误清空管理域、配置与生效 URL 混淆及第三 Host 绕过隔离;加固 Settings API 与中间件,并更新文档站。
This commit is contained in:
O96u committed 2026-08-30 10:57:26 +08:00
1 parent 6c6766221b
commit c9c34515dd
36 files changed
+2578 -180

No files matched your search

+236
View File
@@ -0,0 +1,236 @@
# Cloudflare & preferred-edge deployment
In dual-domain mode, PicHost distinguishes **site** vs **image** traffic from the request **Host** (and `X-Forwarded-Host` from a trusted reverse proxy). Cloudflare orange cloud, preferred-edge IPs, Workers, and Pages are all fine — as long as the hostname identity reaching PicHost stays correct.
> **One-line rule:** The network path may change; the hostname identity must not — `admin → admin`, `img → img`, never `img → admin`.
See [Dual-domain separation](./domain-separation.md) for middleware behavior and [Reverse proxy](./reverse-proxy.md) for general proxy requirements.
---
## Quick checklist
| # | Check | Expected |
| - | ----- | -------- |
| 1 | Browser hostname | `admin.*` or `img.*`; do not mix IP / `pages.dev` |
| 2 | Pages / Worker in the middle | Image host must **not** `fetch(admin…)` |
| 3 | Reverse-proxy upstream | Both hostnames → `http://127.0.0.1:6892` |
| 4 | Nginx `proxy_set_header Host` | `$host` (preserve client hostname) |
| 5 | PicHost settings | `SITE_BASE_URL` / `IMAGE_BASE_URL` match `server_name` |
| 6 | Port `6892` | Reachable only from localhost or internal proxy |
---
## Recommended layout
Both hostnames go through Cloudflare (proxied) and **independently** reverse-proxy to one PicHost instance:
```mermaid
flowchart TB
internet[Internet]
cf[Cloudflare proxied]
admin[admin.example.com]
img[img.example.com]
panel[1Panel / Nginx]
app[PicHost :6892]
internet --> cf
cf --> admin
cf --> img
admin --> panel
img --> panel
panel --> app
```
```env
SITE_BASE_URL=https://admin.example.com
IMAGE_BASE_URL=https://img.example.com
```
```text
admin.example.com → http://127.0.0.1:6892
img.example.com → http://127.0.0.1:6892
```
You do **not** need separate ports for “frontend / backend / images” — one PicHost instance on port `6892` is enough.
### Site hostname ≠ frontend domain
| Hostname | Serves |
| -------- | ------ |
| `admin.example.com` | Admin UI, login, settings, gallery, API, upload |
| `img.example.com` | Image delivery, hotlinks, CDN |
Use **site / image hostname**, not “frontend / backend”. PicHost does **not** require `frontend :3000 + backend :6892` in production.
---
## Orange cloud & preferred edge
**DNS Proxied (orange cloud)** is normal and recommended:
```text
admin.example.com, img.example.com
A/CNAME → origin → Proxied
```
Opening `admin.example.com` should yield `Host: admin.example.com` at PicHost; `img.example.com` should yield `Host: img.example.com`.
**Preferred-edge** setups (alternate IPs, entry hostnames) are also fine. Preferred routing may change the path to Cloudflare’s edge, but must **not** rewrite the image hostname’s origin identity to the site hostname:
```text
admin → CF → admin.example.com → origin → PicHost
img → CF → img.example.com → origin → PicHost
```
---
## 6. Typical bad topologies
These patterns break Host isolation even when the browser still shows the image hostname:
### ❌ Bad 1: Image host fetches site host
```mermaid
flowchart LR
user[User on img.example.com]
worker[Pages / Worker]
admin[fetch admin.example.com]
app[PicHost]
user --> worker --> admin --> app
```
The last hop has `Host: admin.example.com`, so PicHost treats it as site traffic — `https://img.example.com/` may show the admin UI. This is **Host rewrite**, not a cache bug.
### ❌ Bad 2: Preferred proxy upstreams to site host
```text
img.example.com → preferred hostname → admin.example.com → origin
```
Even for routing only, using the site hostname as upstream breaks image isolation.
### ❌ Bad 3: Public exposure of port 6892
```text
any host / IP:6892 → PicHost
```
Users can bypass Cloudflare and Nginx `server_name` rules. Port `6892` should be reachable only from `127.0.0.1` or an internal reverse proxy.
### ✅ Correct pattern
```text
admin → admin → PicHost
img → img → PicHost
```
Do **not** use Pages / Workers as a **full-site reverse proxy** for PicHost (extra `pages.dev` / `workers.dev` entry points). PicHost needs `node-server`, SQLite, `sharp`, and local `data/` — it **cannot** run as a Node app on Pages/Workers. Keep Docker/VPS for the app; use orange-cloud DNS and optional R2 on CF.
---
## Workers / Pages guidance
If you only need headers, auth, Referer rules, logging, or cache — and PicHost remains your origin:
| Prefer | Avoid |
| ------ | ----- |
| `img.example.com` → **Worker Route** → existing `img` origin | `img` → Worker → `fetch(admin.example.com)` |
Cloudflare recommends [Routes](https://developers.cloudflare.com/workers/configuration/routing/routes/) when an external origin already exists, rather than [Custom Domains](https://developers.cloudflare.com/workers/configuration/routing/custom-domains/) used to chain cross-host `fetch()`.
Code like this on the image hostname is a smoking gun:
```js
fetch(`https://admin.example.com${url.pathname}`)
```
---
## Nginx / 1Panel layout
Two `server` blocks, same upstream. Critical line: `proxy_set_header Host $host;`
```nginx
server {
server_name admin.example.com;
location / {
proxy_pass http://127.0.0.1:6892;
proxy_set_header Host $host;
proxy_set_header X-Forwarded-Proto $scheme;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
client_max_body_size 12m;
}
}
server {
server_name img.example.com;
location / {
proxy_pass http://127.0.0.1:6892;
proxy_set_header Host $host;
proxy_set_header X-Forwarded-Proto $scheme;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
}
}
# Reject bare IP and unknown hostnames (strongly recommended)
server {
listen 80 default_server;
listen 443 ssl default_server;
server_name _;
return 444;
}
```
With orange cloud, restrict the origin firewall to **[Cloudflare IP ranges](https://www.cloudflare.com/ips/)** only.
---
## Two-layer Host protection
Use reverse proxy and application layers **together**:
| Layer | Role |
| ----- | ---- |
| **Nginx default server** | Reject bare IP, `pages.dev`, and other undeclared hostnames |
| **PicHost middleware (v1.2.x+)** | When dual-domain is on, unconfigured hosts get 404 at the app (`localhost` / `127.0.0.1` exempt in development) |
Even with app-layer blocking, keep a default server at the proxy to reduce junk traffic to PicHost.
---
## X-Forwarded-Host
PicHost may consult `X-Forwarded-Host` behind a reverse proxy, but that header is **forgeable** by clients. Therefore:
- Do not expose port `6892` publicly
- Only trusted proxies should reach the app
- Prefer keeping `Host` correct from the first hop
---
## Troubleshooting: admin UI on image hostname?
**Inspect the proxy chain first — do not purge CF cache first.**
Cache issues usually mean stale JS/CSS/images or occasional 404s. If `/`, `/login`, and `/settings` work **reliably** on the image host, the Host was almost certainly rewritten.
```text
① Browser hostname?
② Pages / Worker / preferred proxy in the middle?
③ What URL does the proxy fetch?
④ Does it fetch admin.example.com?
⑤ Which Host does 1Panel / PicHost see?
```
---
## Related
- [Dual-domain separation](./domain-separation.md)
- [Reverse proxy](./reverse-proxy.md)
- [Environment variables](./configuration.md)
- [FAQ](./faq.md)