Dear developers,

I am new to security research and am exploring how LLMs can be leveraged to
detect software vulnerabilities.

The two vulnerabilities described here were initially discovered by an
agent and were subsequently *manually verified by me*, confirming that they
are technically exploitable. However, I am not fully certain whether they
should be considered *real-world vulnerabilities* or merely *toy /
theoretical vulnerabilities* with limited practical impact.

Detailed technical descriptions, including the discovery process and
verification steps, are provided in the accompanying Markdown files.
If these turn out to be false positives, I sincerely apologize for any
inconvenience this may have caused.

*Issue 1 (symlink-based traversal):*

The second issue concerns path traversal during BusyBox tar extraction via
pre-existing symlink components in the destination directory. Although
archive paths are sanitized, file writes can be redirected outside the
extraction root. This behavior is reproducible, but its security relevance
is unclear to me.

*Issue 2 (hard link handling):*

The first issue involves BusyBox tar extraction of hard link entries, where
the hard link target is not sanitized and may reference paths outside the
extraction directory. I have reproduced this behavior locally but am unsure
whether it should be considered a real security vulnerability or intended
behavior.

--
Best wishes,
Zhicheng Chen
Ph.D. student at Texas A&M University

Attachment: path traversal.md
Description: Binary data

Attachment: path traversal2.md
Description: Binary data

_______________________________________________
busybox mailing list
[email protected]
https://lists.busybox.net/mailman/listinfo/busybox

Reply via email to