Executive brief
DBHub is a database client tool that can be run in HTTP server mode for remote access and web-based clients. When started with HTTP transport, it exposes an unauthenticated endpoint vulnerable to DNS rebinding attacks. An attacker can trick a victim's browser into executing arbitrary SQL queries against the victim's database, allowing unauthorized data exfiltration or modification without authentication.
Technical details
DBHub's HTTP server attempts to prevent cross-origin browser attacks by comparing the `Origin` and `Host` header hostnames, then reflecting the validated origin into `Access-Control-Allow-Origin`. However, this check is insufficient against DNS rebinding: after an attacker-controlled hostname rebinds to the victim's DBHub server IP address, both headers contain the attacker's hostname, so the equality check passes. The vulnerability is in `/src/server.ts` middleware that does not enforce an explicit allowed-hosts policy. The HTTP `/mcp` endpoint accepts unauthenticated JSON-RPC method calls with no CSRF protection. Attack vector is network and requires victim browser interaction (visiting attacker-controlled website), but no authentication credentials. A successful exploit grants access to all MCP tools (e.g., `execute_sql`) with permissions inherited from the DBHub server process, enabling SQL injection, data exfiltration, and potentially data modification depending on configured permissions.
Affected products
- Bytebase DBHub 0.21.2
Timeline
- 2026-09-24: disclosed: Advisory published; PoC provided
- 2026-01-01: other: CVE-2026-61742 assigned