PyPI · githacker
GitHacker: Path traversal in ref/hash parsing enables existence oracle and hex-fragment exfiltration via malicious .git server
GitHacker through 1.1.7 did not validate path segments parsed from attacker-controlled .git/HEAD before joining them onto its output directory. A malicious server could coerce GitHacker into reading arbitrary local files. Contents do not stream back wholesale, but the recovery loop turns any 40-character hex substring into an outbound HTTP GET — an existence oracle for arbitrary paths plus hex-fragment exfiltration of file contents.
GitHacker rebuilds a remote .git/ by fetching files into temp_dst. Two functions derived filesystem paths from server-controlled content:
add_head_file_tasks reads the downloaded .git/HEAD, parses ref: <ref-path>, and joins the raw ref-path onto temp_dst/.git/logs/ before reading the resulting file.add_hashes_parsed scans any file it reads for 40-character hex substrings and emits GET .git/objects/<sha[0:2]>/<sha[2:]> for each one — onto the attacker's server and into the local output tree.Pre-fix, add_head_file_tasks did not validate the ref segments. A malicious .git/HEAD of
ref: ../../../../../../etc/passwd
caused add_head_file_tasks to traverse out of temp_dst and read /etc/passwd. The bytes flowed into add_hashes_parsed, which emitted one outbound HTTP request per 40-char-hex match — observable on the attacker's logs.
PR #65 originally classified this as arbitrary local file read. Joint analysis during coordinated disclosure narrowed the primitive: file contents do not stream back wholesale because the only egress channel is the 40-char-hex regex. In practice an attacker can:
/etc/shadow, /root/.ssh/id_rsa, /home/<user>/.git-credentials, build artifacts under /tmp/).Not exploitable without victim action: the attacker must persuade the victim to run GitHacker against a URL they control. Project guidance has always been to run GitHacker inside a disposable container.
A working write-side primitive (attacker drops content outside temp_dst via add_folder / add_task) does not reproduce against the shipped 1.1.7 source. Empirical testing by the reporter against GitHacker-1.1.7.tar.gz (25 traversal-style payloads including ....//, %2e%2e%2f, layered foo/../../, NUL bytes, backslash variants, absolute paths) yielded 0/25 escapes. Two structural reasons:
add_folder anchors every derived path on self.url + '.git/', so the first path component after the url_length strip is always .git. os.path.join's absolute-path short-circuit never fires.str.replace("..", "") is greedy non-overlapping; ....// collapses to //, ......// to ///, etc. No literal .. survives into add_task.5f2a8ba is still the correct fix for the read-side primitive and additionally hardens add_task as defense in depth against future regressions — for example, if a later caller removes the .git/ anchor in add_folder or wires a new server-controlled segment source into add_task.
Commit 5f2a8ba introduces _is_safe_path_segment as a single trust boundary: every segment about to be joined onto temp_dst or appended to an outgoing URL is validated against an allowlist before add_task accepts it. Empty / . / .. / separators / NUL / control characters are rejected; the brittle replace("..", "") filter is removed.
The fix also tightens adjacent surfaces preemptively:
add_folder switches to urlparse-based scheme + netloc + path comparison.construct_url_from_path_components percent-encodes every segment.PR #65's two-layer defense (allowlist regex + os.path.realpath() confinement) was consolidated onto the allowlist applied at queue-time, removing the TOCTOU window an after-the-fact realpath() check leaves open and the per-call-site drift risk. PR #65 was closed in favour of the broader fix.
Regression tests in tests/test_ref_validation.py (commit 16fcd81) pin the PoC and six bypass variants (extra-depth, mid-path, NUL, absolute path, leading-dot, .lock-suffix).
Reported and patched-prototyped by Zac Wang (@7a6163) in #65. Zac refined the impact framing from "arbitrary file read" to "existence oracle + hex-fragment exfiltration" and verified the absence of a write-side primitive against the shipped 1.1.7 sdist with a 25-payload harness.
Patched in 1.1.8 (commit 5f2a8ba; tests 16fcd81).
Run GitHacker inside a disposable container. Do not point GitHacker at any URL whose contents are not under your control.
Is your project exposed to this? Stateward checks every dependency on every pull request and flags it only if your code actually reaches it.
Check my repoSources: CISA KEV (public domain), OSV.dev & GitHub Advisory Database (CC-BY-4.0), FIRST EPSS, NVD/CWE (public domain). Served live from the Stateward advisory database.