Description

An unauthenticated party able to reach the port of a MongoDB Connector for BI (mongosqld) instance may generate enough routine connection log activity to exhaust the storage backing the configured log path. When a log write or log rotation operation subsequently fails, the resulting error is not handled and the shared mongosqld process ends, ending service for all connected SQL clients. The process continues to end on startup until an operator restores available storage, and the diagnostic message explaining the condition is not recorded.

Executive Summary

Technical Details

Mitigations

References

Additional Metadata

{
  "score": "7.5",
  "pubDate": "2026-08-28T22:16:54.383Z",
  "pubdate": "2026-08-28T22:16:54.383Z",
  "executiveSummary": "A critical availability vulnerability exists in the MongoDB Connector for BI (mongosqld) that allows an unauthenticated, network-adjacent attacker to trigger a permanent Denial of Service (DoS) condition.\nThe vulnerability stems from improper error handling during log write or log rotation operations when the underlying storage media reaches capacity.\nAn attacker can exploit this by generating a high volume of connection requests, forcing excessive log activity that exhausts the available disk space.\nOnce storage is depleted, the mongosqld process terminates abruptly due to unhandled exceptions, resulting in a total service outage for all SQL clients.\nBecause the service fails to start until disk space is reclaimed by an operator, the impact is a persistent disruption of database connectivity.\nThis vulnerability is particularly severe because it does not require authentication and can be leveraged by any party capable of reaching the mongosqld service port.\nThe lack of diagnostic logging upon crash complicates incident response, as the primary indicator of failure is hidden from standard system logs.",
  "technicalDetails": "The root cause of this vulnerability lies in a lack of robust error handling within the mongosqld logging subsystem. When the application attempts to perform write operations or file rotations for its configured log path, it assumes the operation will succeed. If the underlying storage filesystem is full, the OS-level write operation fails, and the application fails to handle the resulting I/O error state gracefully.\nThe exploitation flow begins with an unauthenticated attacker identifying an exposed mongosqld instance. The attacker then initiates a flood of connection attempts to the service port. Each connection request generates entries in the mongosqld logs. By programmatically sustaining this high-frequency connection activity, the attacker forces the logging subsystem to consume disk space rapidly until the storage volume reaches 100% capacity.\nOnce the disk capacity is exhausted, the next attempt to log a connection or rotate the log file triggers an unhandled exception. This crash is fatal to the mongosqld process, causing it to terminate immediately and drop all existing SQL client sessions. Furthermore, because the error condition prevents the log write, the application cannot record the diagnostic information or the stack trace identifying the cause of the failure, effectively masking the root cause from system administrators.\nA significant post-exploitation impact is the creation of a 'fail-at-startup' loop. Upon process termination, the mongosqld service attempts to restart; however, because the disk remains full, the process initialization sequence—which may involve its own logging or file system checks—immediately fails again. This ensures that the service remains unavailable even after a reboot or automated restart, requiring manual intervention to clear space on the partition.\nThe vulnerability is accessible to any attacker with network connectivity to the mongosqld port. No prior knowledge of the database, credentials, or elevated privileges is required to execute this denial-of-service attack. The vulnerability persists across versions of the product where this error-handling logic remains unaddressed, posing a significant risk to the availability of the BI integration layer."
}