MikroORM has SQL injection via runtime-controlled identifiers and JSON-path keys

Summary

MikroORM's identifier-quoting helper (Platform.quoteIdentifier and the postgres/mssql overrides) and its JSON-path emitters (Platform.getSearchJsonPropertyKey, quoteJsonKey) did not properly escape characters that delimit the SQL identifier or string-literal context they emit into. When application code passes attacker-influenced strings to public ORM APIs that expect an identifier or a JSON-property filter, an attacker can break out of the quoted context and inject arbitrary SQL.

Affected APIs

The vulnerability is reachable when application code passes an attacker-influenced string to any of the following documented APIs:

• Multi-tenant schema option — em.fork({ schema }), qb.withSchema(name), wrap(entity).setSchema(name), em.create(Cls, data, { schema }). The schema name is concatenated into the SQL identifier and never had its dialect quote character escaped.
• em.find / qb.where JSON-property filters — em.find(Entity, { jsonCol: { [userKey]: value } }). The user-supplied JSON sub-keys cannot be validated against any metadata (JSON columns are schemaless by design), and were spliced into the SQL string literal of the JSON path expression without escaping.
• qb.where / qb.orderBy / qb.groupBy / qb.having / qb.select keys — keys containing . or :: bypassed the structured-where metadata validator in CriteriaNode, then flowed through the same broken quoteIdentifier. Apps that forwarded raw filter keys from request input were already broken on authorization grounds (e.g. { isAdmin: true }); the SQL injection here is a defence-in-depth failure on top of that.

The vulnerability does not affect documented escape-hatch APIs (raw(), the sql tagged template, qb.raw(), em.raw()) — those are documented as accepting raw SQL and are the application's responsibility to sanitize.

Impact

• Confidentiality — read from any table/schema the database user has access to (cross-tenant data leak in multi-tenant deployments).
• Integrity — on dialects supporting multi-statement queries (MSSQL, MySQL with multi-statement enabled), execute additional arbitrary SQL statements (data modification, in-database privilege escalation).
• Availability — DROP/TRUNCATE via injected statements where the database user has the privilege.

Affected dialects

All SQL dialects supported by MikroORM. The identifier-quoting bug exists in every dialect's quoteIdentifier (the dialect's own quote character — backtick, double quote, or right bracket — was not doubled when embedded in the identifier). The JSON-path bug exists in all dialects' getSearchJsonPropertyKey/quoteJsonKey.

MongoDB driver is not affected (no SQL).

Patches

• v7: upgrade to 7.0.14 or later.
• v6: upgrade to 6.6.14 or later.

Patches:

• Identifier quoting: #7653 (master) / #7654 (6.x)
• JSON-path keys: #7656 (master) / #7657 (6.x)

Workarounds

If users cannot upgrade immediately:

• For multi-tenant apps using em.fork({ schema }) / wrap().setSchema() / qb.withSchema(): validate the schema name against a strict allowlist (e.g. ^[A-Za-z_][\w$]*$) before passing it to MikroORM.
• For applications that pass where / orderBy filters from request input: validate every key against the entity's known properties before constructing the filter; do not pass keys containing . or :: from user input.
• For applications that allow filtering on JSON columns from request input: validate every JSON sub-key against an allowlist (or against ^[a-zA-Z_][\w]*$) before passing it to em.find.

Credits

Reported and patched by Martin Adámek (project maintainer) during an internal security review.