CVE-2024-41672: DuckDB: sniff_csv provides filesystem access even when enable_external_access is disabled

Summary Content in filesystem is accessible for reading using sniffcsv, even with enableexternalaccess=false.

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

PoC SQL SET enableexternalaccess=false; SET lockconfiguration=true;

SELECT Columns FROM sniffcsv('/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 readcsv('/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 sniffcsv 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 disabledfilesystems setting: SQL SET disabledfilesystems='LocalFileSystem'; SET lockconfiguration=true;

SELECT Columns FROM sniffcsv('/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 (https://github.com/duckdb/duckdb/pull/13133), and will be released with the next DuckDB release.