FalkorDB EUVD-2026-95473

The attacker must be able to issue Redis replication commands to the target so that FalkorDB processes an attacker-supplied RDB stream - in practice this means an instance reachable with no password set (the default, unauthenticated case cited in the description) or where the attacker can otherwise act as/inject a replication source. The specific trigger is an RDB deleted-nodes (or deleted-edges) buffer whose byte length is not a multiple of sizeof(NodeID)/sizeof(EdgeID), which bypasses the release-build-removed ASSERT in RdbLoadDeletedNodes/RdbLoadDeletedEdges. Limiting factors: instances configured with authentication (requirepass) or network-restricted so that untrusted peers cannot initiate replication are not exposed to unauthenticated exploitation, and only FalkorDB versions before 4.18.1 are affected; reliable code execution (versus a crash) additionally depends on heap layout and is not guaranteed.

This is a genuinely high-priority memory-corruption flaw, not a high-CVSS-but-low-risk paper cut. The vendor-assigned CVSS 4.0 vector (AV:N/AC:L/AT:N/PR:N/UI:N) describes a network-reachable, low-complexity, unauthenticated issue with high confidentiality, integrity and availability impact (9.3), and the technical root cause is credible: a CWE-415 double free / use-after-free caused by a validation check implemented with ASSERT(), which is compiled out of release builds, so a mis-sized deleted-nodes buffer is freed, then read and freed again. The patch diff confirms the mechanism precisely and shows the fix replacing ASSERT with a real runtime check (RedisModule_Assert plus a length-and-count validation). The one important caveat is the PR:N rating: the description scopes the attacker to one 'who can issue Redis replication commands (for example, against an instance with no password configured)', so true unauthenticated exploitation specifically assumes a default/passwordless or otherwise replication-reachable deployment - on a properly authenticated instance this becomes an authenticated or privileged operation, which the raw CVSS score does not reflect. No EPSS score, no CISA KEV listing, and no public exploit code are present in the input; absence of these means there is no evidence of active or at-scale exploitation at time of analysis, so prioritization should rest on the strong technical impact plus exposure of passwordless instances rather than on observed exploitation. Reaching reliable RCE from a double free is also non-trivial in practice, so denial of service (process crash) is the more certain outcome while code execution is plausible but unproven.

An attacker who can reach a FalkorDB instance that has no password configured issues Redis replication commands to feed the server a crafted RDB stream in which the deleted-nodes buffer length is not a multiple of sizeof(NodeID). Because the alignment check uses ASSERT() (removed in release builds), RdbLoadDeletedNodes frees the buffer and keeps going, reading and freeing it again, corrupting the heap to crash the redis-server process or potentially execute arbitrary code. No public proof-of-concept is identified at time of analysis, but the fixing PR #1843 documents the exact flawed code path.