Title: Critical MongoDB Use-After-Free (CVE-2026-118383): Unauthenticated Two-Request Crash Set-Up, Fixes Out for Seven Branches
Summary:
MongoDB shipped fixes for CVE-2026-118383, a critical use-after-free in the primary catch-up process of MongoDB Server, announced November 10 and rated 9.5 on the CVSS 4.0 scale — with possible impacts of unexpected server termination or, under certain conditions, arbitrary code execution in the server process. The flaw affects replica set members, including the shard and config server members of sharded clusters; a standalone mongod does not perform catch-up and is not affected. Datadog Security Labs, which published a technical analysis on November 11, reproduced the crash: an unauthenticated client sends two ping requests carrying a readConcern.afterOpTime value, and because vulnerable servers accept ping before login, enabling authentication does not stop them. Each request plants a pending “waiter” notification that does not expire — harmless on its own. Damage requires a second step outside the attacker’s control: the same member has to take an election while still trailing its peers. As that lagging primary-to-be finishes catching up, its end-of-catch-up cleanup routine executes a second time, dereferences memory it had already released, and the process goes down before it can serve a single write. Neither the replication lag nor the election is something an attacker can arrange, and they cannot see when either occurs; given both preconditions, though, the outcome is deterministic rather than a matter of timing. In Datadog’s lab the vulnerable 8.0.34 died with a SIGSEGV and backtrace pointing at CatchupState::abort_inlock(), leaving the test cluster without a primary, while the fixed 8.0.37 rejected both requests as Unauthorized; the researchers did not achieve code execution, and the public proof of concept’s reliable mode requires a test-only setting a remote attacker cannot enable. First fixed releases per branch: 9.0.7, 8.3.16, 8.0.37, 7.0.48, 6.0.30, 5.0.35 and 4.4.32; every earlier version in each branch is affected. Operators should put the fixed release on every replica set member — containerized instances and any nodes that sat out a maintenance window alike — and confirm the running version on each one afterwards; Atlas clusters receive patched builds automatically. The 8.2 branch, whose newest release is 8.2.12, is not among the affected versions listed; per MongoDB’s lifecycle schedule (as reported by THN), that branch’s support ended on July 31, so it is no longer patched — yet the PoC author reported the bug still reproduced on 8.2.12 in a lab test. Until you can upgrade, restrict network access to MongoDB to the systems that need it; Datadog also documents detection signatures in structured logs — command ID 51803 for unauthenticated expired afterOpTime pings, catch-up entry (21359) and double catch-up exit (21363) events, and the CatchupState::abort_inlock() backtrace. MongoDB said it found the flaw internally and was not aware of any exploitation attempts at announcement.
References: