Summary

Content in filesystem is accessible for reading using sniff_csv, even with enable_external_access=false.

Details

During a pentest, a security researcher was able to access environment variable data and other system data by using the sniff_csv function, even though we set enable_external_access to false.

PoC

SET enable_external_access=false;
SET lock_configuration=true;

SELECT Columns FROM sniff_csv('/proc/self/environ');

Impact

Provides an attacker with access to filesystem even when access is expected to be disabled and other similar functions do NOT provide access.

For example select * from read_csv('/proc/self/environ') fails with a permission error.

There seems to be two vectors to this vulnerability:

1. Access to files that should otherwise not be allowed. (We expect Permission Error: Scanning CSV files is disabled through configuration and not to provide any access to the file or even acknowledge that it exists).
2. The content from a (non-csv?) file can be read (e.g. /etc/hosts, proc/self/environ, etc) even though that doesn't seem to be the intent of the sniff_csv function (my understanding is it's intending to provide information about the shape of the data, but not provide the data itself).

Workaround

It is possible to work around the issue by disabling the local file system using the disabled_filesystems setting:

SET disabled_filesystems='LocalFileSystem';
SET lock_configuration=true;

SELECT Columns FROM sniff_csv('/proc/self/environ');

will result in Permission Error: File system LocalFileSystem has been disabled by configuration.

Fix

A fix has been merged into the main branch (duckdb/duckdb#13133), and will be released with the next DuckDB release.

References

• GHSA-w2gf-jxc9-pf2q
• https://nvd.nist.gov/vuln/detail/CVE-2024-41672
• duckdb/duckdb#13133
• duckdb/duckdb@c9b7c98
• https://github.com/pypa/advisory-database/tree/main/vulns/duckdb/PYSEC-2024-203.yaml