Description
### Impact
Scrapy was following redirects regardless of the URL protocol, so redirects were working for `data://`, `file://`, `ftp://`, `s3://`, and any other scheme defined in the `DOWNLOAD_HANDLERS` setting.
However, HTTP redirects should only work between URLs that use the `http://` or `https://` schemes.
A malicious actor, given write access to the start requests (e.g. ability to define `start_urls`) of a spider and read access to the spider output, could exploit this vulnerability to:
- Redirect to any local file using the `file://` scheme to read its contents.
- Redirect to an `ftp://` URL of a malicious FTP server to obtain the FTP username and password configured in the spider or project.
- Redirect to any `s3://` URL to read its content using the S3 credentials configured in the spider or project.
For `file://` and `s3://`, how the spider implements its parsing of input data into an output item determines what data would be vulnerable. A spider that always outputs the entire contents of a response would be completely vulnerable, while a spider that extracted only fragments from the response could significantly limit vulnerable data.
### Patches
Upgrade to Scrapy 2.11.2.
### Workarounds
Replace the built-in retry middlewares (`RedirectMiddleware` and `MetaRefreshMiddleware`) with custom ones that implement the fix from Scrapy 2.11.2, and verify that they work as intended.
### References
This security issue was reported by @mvsantos at https://github.com/scrapy/scrapy/issues/457.
Affected packages
| Ecosystem | Package | Affected | Fixed | Record |
|---|---|---|---|---|
| PyPI | scrapy | < 2.11.2 | 2.11.2 | GHSA-23j4-mw76-5v7h |