Fedify security updates: 2.0.36, 2.1.32, 2.2.21, 2.3.16, and 2.4.6
If you use an affected Fedify release, update now. These patches address six security vulnerabilities affecting Fedify and related packages. They fix two content-spoofing issues in backfill, an embedded-object origin-check bypass, and three denial-of-service issues caused by missing network deadlines.
The patched releases for the vocabulary, WebFinger, activity delivery, and NodeInfo issues are 2.0.36, 2.1.32, 2.2.21, 2.3.16, and 2.4.6. Update to the corresponding release on your minor line. The two @fedify/backfill issues affect the 2.3 and 2.4 lines and are fixed in 2.3.16 and 2.4.6. The security policy covers the latest two minor lines, 2.3 and 2.4. If you are still on a 1.x release, move to a supported 2.x release.
Update directly installed copies of the affected packages as well as @fedify/fedify, and check your lockfile for older transitive copies. If you generate custom vocabulary classes with @fedify/vocab-tools, update the generator, regenerate the classes, and rebuild your application before deploying. See the vocabulary documentation.
Embedded-object origin-check bypass with an anonymous parent (GHSA-r388-57j3-5fgp, medium, CVSS 3.1 5.3)
An attacker-controlled document can make a vocabulary accessor return a forged embedded object that claims to belong to another server, without fetching it from that server. The bypass requires a parent object without an id and a property repeated under its compact name and full IRI. A second decoding path skipped the origin check for that anonymous parent.
This can reach inbox listeners through a signed Create or Announce from an actor on the attacker's own server. It also affects backfill and application code that parses remote JSON-LD and reads embedded objects through vocabulary accessors. If the application stores or displays the returned content under its claimed ID, the result can be content spoofing or cache poisoning. A valid signature from the attacker's actor does not establish ownership of the forged object's ID.
Affected: @fedify/fedify, @fedify/vocab, and @fedify/vocab-tools on the 2.0–2.4 lines before the patched release above. The vocabulary bundled with @fedify/fedify 1.9 and 1.10 is also affected; no patched 1.x release is planned. Earlier 1.x releases lack this origin protection altogether and are not a safe downgrade.
Action needed: upgrade the vocabulary package, including any directly installed copy. If you use generated custom vocabulary classes, update @fedify/vocab-tools, regenerate the classes, and rebuild your application.
Until you can upgrade, fetch an accessor-returned object again by its ID with lookupObject() under the default origin policy before storing or displaying it, and use only the fetched copy. Apply this to nested objects and backfilled items too. Setting crossOrigin: "throw" does not close this bypass. Rejecting only top-level inbox activities without an ID is also insufficient: an activity with an ID can contain an anonymous nested object.
WebFinger lookups can stall without a deadline (GHSA-9pmv-w936-hj3f, high, CVSS 3.1 7.5)
A malicious WebFinger endpoint can hold requests open or respond slowly across redirects. Fedify limited the number of redirects but imposed no built-in deadline for the lookup. Repeated or concurrent lookups can occupy calling operations and reduce availability. This affects applications resolving attacker-controlled handles and can also be reached when re-fetching an unverified inbound activity falls back to WebFinger after a document-fetch failure. Reported by zx (manus-pi).
Affected: @fedify/webfinger on the 2.0–2.4 lines before the patched release above.
The patched implementation applies one 10-second deadline across DNS resolution, redirects, requests, and response-body reading for each WebFinger lookup. For non-portable actors, the initial lookup and alias verification share a single 10-second budget, with a limit of 100 raw subject/alias candidates.
Action needed: upgrade @fedify/webfinger and @fedify/vocab, including directly installed copies. A caller's cancellation signal can shorten the WebFinger deadline, but cannot extend it.
This does not put a 10-second limit on every lookupObject() call. In particular, the default document loaders on the 2.0–2.3 lines still have no built-in deadline, so a document fetch before or after WebFinger resolution can stall independently. Applications on those lines should enforce a separate deadline in their document loader or move to the patched 2.4 line.
Outbound activity delivery can occupy workers indefinitely (GHSA-fq5h-gjhp-g75x, high, CVSS 3.1 7.5)
A malicious destination inbox can accept a delivery and then stall the response or a redirect hop. Neither RSA-signed nor unsigned delivery had a total deadline. Concurrent deliveries to such inboxes can occupy workers and connections and delay delivery to other servers. Ed25519-only configurations are not the only ones affected. Reported by zx (manus-pi).
Affected: @fedify/fedify before the patched release on each branch above. The confirmed affected range for the unsigned redirect helper starts at 2.0.28; the signed-path issue predates that helper, so older releases should not be treated as safe.
The patched implementation gives each delivery attempt one 30-second deadline across connection setup, redirects, HTTP-signature retries, and response consumption. It does not reset the budget at each hop. A timeout before the final response is a retryable delivery failure. If an error response has already arrived, its status is preserved even if reading its body times out, so permanent failures are not turned into generic timeout retries.
Action needed: upgrade @fedify/fedify on both request-serving processes and delivery workers.
NodeInfo lookups can stall without a deadline (GHSA-64hq-2hgr-q4pm, medium, CVSS 3.1 5.3)
A malicious NodeInfo endpoint can delay discovery, redirects, or the document response and keep the calling operation waiting. The redirect-count limit did not bound the elapsed time. This affects application features that explicitly query attacker-influenced server URLs and the fedify nodeinfo command; inbound federation does not automatically trigger NodeInfo lookups. Reported by zx (manus-pi).
Affected: @fedify/fedify before the patched release on each branch above.
The patched client uses one 10-second default deadline across discovery and document fetching, including redirects and body reading. It also detects redirect cycles. A timeout or cancellation returns undefined.
Action needed: upgrade @fedify/fedify, and update any separately installed @fedify/cli used to query remote instances. See the CLI installation instructions. The API's timeout option is in milliseconds; timeout: null disables the deadline. Keep a bounded timeout for untrusted destinations. The CLI's -T/--timeout option is in seconds and defaults to 10.
On Node.js, NodeInfo DNS lookups are also limited to two concurrent OS resolver calls per process. An OS lookup already in progress cannot be cancelled and continues to occupy its slot until it finishes, even if the calling operation has timed out.
Backfill trusts forged cross-origin embedded content (GHSA-x58x-xx7h-7rh4, high, CVSS 3.1 7.5)
@fedify/backfill passed crossOrigin: "trust" to vocabulary accessors, bypassing their default origin checks. An attacker-controlled document could embed a Note, collection item, reply, or activity object with another server's ID, and backfill could return that embedded copy instead of fetching the genuine object. An application that persists results by ID could overwrite genuine content or poison its cache.
Affected: @fedify/backfill 2.3.x before 2.3.16 and 2.4.x before 2.4.6.
The patched implementation uses the default origin policy rather than trusting cross-origin embedded content. This may require additional fetches and can change which items are returned.
Action needed: upgrade @fedify/backfill to 2.3.16 or 2.4.6 and update @fedify/vocab to the matching patched version. The vocabulary fix above is needed as well; the backfill change alone does not close the anonymous-parent bypass.
Until you can upgrade, re-fetch backfilled items by ID with lookupObject() under the default origin policy before persisting them, and use only the fetched copies. Backfill results remain unverified remote data after the update; an origin check is not proof of authorship.
Backfill accepts forged IDs from custom document loaders (GHSA-h3jv-px69-2gpj, high, CVSS 3.1 7.5)
A custom BackfillDocumentLoader that fetches remote JSON and parses it directly with fromJsonLd() can lose the URL from which it retrieved the document. Before the fix, backfill could accept that parsed object's claimed ID even when it belonged to another origin. The attacker could supply a forged collection or post that the application then ingests under a victim server's ID. Reported by @z9mb1.
Affected: @fedify/backfill on the 2.3 and 2.4 lines before 2.3.16 and 2.4.6, respectively, when used with a custom loader that does not enforce retrieval-origin checks.
The patched implementation rejects a loader result whose ID origin differs from the requested IRI before caching or traversing it. Because BackfillDocumentLoader does not expose the final response URL, this check can also reject legitimate cross-origin redirects.
Action needed: upgrade @fedify/backfill and review your custom loaders against the backfill documentation. Prefer lookupObject() with the default origin policy to directly parsing fetched JSON. If you provide a custom JSON-LD document loader, preserve the final response URL in documentUrl so lookupObject() can perform its origin check. Do not turn untrusted remote JSON into vocabulary constructor values: constructor-supplied objects are treated as trusted local data.
Thanks to manus-pi for reporting issues. If anything unclear, comment below. Thanks.