Skip to content

GHSA-539m-9xh6-q6rr CVE-2026-73619 PYSEC-2026-3948

GitPython: Incomplete unsafe_git_archive_options denylist omits --add-file / --add-virtual-file, enabling arbitrary file read via Repo.archive()

Medium PyPI
Severity
0.0 / 10
Medium · CVSS 4.0
Exploit chance (EPSS)
0.41%
Chance of exploitation in the next 30 days. Higher than 35.46% of scored vulnerabilities.
Known exploited
No
Not in CISA's catalog
Weakness
CWE-200 Exposure of Sensitive Information to an Unauthorized Actor
CWE-73 External Control of File Name or Path
Fix
3.1.57
Upgrade gitpython to this version or later

Description

Target: gitpython-developers/GitPython Tested: HEAD 07e80555 (2026-07-25), latest release 3.1.55, git version 2.50.1

Summary

Repo.archive() does call the option guard, so this is not a missing-guard report. The guard is present and working; the denylist it consults is incomplete.

# git/repo/base.py:169
unsafe_git_archive_options = [
    # Allows arbitrary command execution through the remote git-upload-archive command.
    "--exec",
    # Writes output to a caller-controlled filesystem path.
    "--output",
    "-o",
]

The comment on --output states the protected class in the project's own words: an option that lets the caller name a filesystem path is unsafe. --output is blocked because it writes to a caller-chosen path.

git archive also accepts --add-file=<path> and --add-virtual-file=<path:content> (both present in current git; verified against git version 2.50.1). --add-file reads a caller-chosen path — including an absolute path outside the repository — and places the bytes into the archive the caller receives. Neither option is in the list, and no other layer references them:

$ grep -rniE "add.file|add_file" git/
git/index/base.py:771:   R"""Add files from the working tree, ...      # unrelated docstring

Net effect: the guard blocks arbitrary file write at this sink while permitting arbitrary file read at the same sink.

Reachability proof (verified at the sink)

poc/poc_addfile.py at HEAD 07e80555. The PoC creates its own out-of-tree canary, so it runs from a clean machine:

-- CONTROL: options the denylist covers (expect BLOCKED) --
  [BLOCKED] output='/tmp/gp_written.tar': --output is not allowed, use `allow_unsafe_options=True` to allow it.
  [BLOCKED] o='/tmp/gp_written.tar': -o is not allowed, use `allow_unsafe_options=True` to allow it.
  [BLOCKED] exec='touch /tmp/gp_exec': --exec is not allowed, use `allow_unsafe_options=True` to allow it.

-- SIBLING OMITTED FROM THE DENYLIST: --add-file (expect ALLOWED) --
  [ALLOWED] add_file='/tmp/gp_canary.txt'  -> archive 10240 bytes
  archive members: ['f.txt', 'gp_canary.txt']
  >>> EXFILTRATED gp_canary.txt: 'secret-canary-12345'
  >>> byte-for-byte match with the out-of-tree file: CONFIRMED

-- also: --add-virtual-file (attacker-chosen name AND content) --
  [ALLOWED] add_virtual_file='pwn.txt:hello'  -> archive 10240 bytes

The three blocked lines are the control: they prove the guard is active on this call path, so the fourth result is a gap in list membership rather than a guard that never ran.

Minimal reproduction:

import io, tarfile
from git import Repo

buf = io.BytesIO()
Repo("/path/to/repo").archive(buf, format="tar", add_file="/etc/passwd")
print(tarfile.open(fileobj=io.BytesIO(buf.getvalue())).getnames())
# ['<repo files>', 'passwd']   <- contents readable by whoever receives the archive

The canary is untracked and lives outside the repository; its contents are recovered from the returned archive and asserted byte-for-byte against the on-disk file. The option is rendered by transform_kwargs into --add-file=<path> and reaches git archive unmodified.

Direct precedent

GHSA-6p8h-3wgx-97gf (High, published 2026-07-22) is the same defect on the sibling list: "Incomplete unsafe_git_clone_options denylist omits --template" — an option absent from one of these denylists, reachable under the same caller-controlled-options precondition, accepted and fixed by adding it. git log shows the archive list itself has already been extended reactively once, in 701ce32f (fix: Guard unsafe git command options, GHSA-956x-8gvw-wg5v), and the --template omission was then fixed separately in ffcb5359.

--add-virtual-file is the same gap pointing the other way

--add-virtual-file=<path:content> lets the caller inject attacker-chosen content under an attacker-chosen name into an archive that downstream consumers will reasonably treat as repository-derived.

Suggested remediation

  1. Preferred — allowlist. Repo.archive() has a small legitimate option surface (format, prefix, worktree_attributes, remote, compression level, plus paths). Accepting those and rejecting the rest means a future git release cannot add another path-taking option that silently reopens this.
  2. Minimum — extend the list with --add-file and --add-virtual-file, and make the membership rule "the option takes a filesystem path or URL" rather than "the option executes a command". The existing comment on --output already implies that rule; applying it consistently is what closes the class instead of this instance.

Scope limits

  • Impact is arbitrary file read at the privileges of the process. Not code execution — I make no such claim here.
  • It requires the embedding application to forward caller-influenced kwargs into Repo.archive(). That is the identical precondition to --output, --exec and --template, all of which this project has treated as reportable.

Disclosure

Reported privately via GitHub private vulnerability reporting. Happy to test a candidate patch against the PoC. No public disclosure until you have shipped a fix and are ready.

Addendum (2026-07-25) — related observation on the same membership question, filed here rather than separately

While auditing the archive denylist, the same class of gap was identified in unsafe_git_clone_options. A second advisory is not being requested, as the issue is lower severity and should inform the fix for the issue above rather than require separate triage. Recording it here to provide the complete picture in one place.

Repo._clone() treats a URL's protocol as a security boundary and applies check_unsafe_protocols() to exactly one input:

clone_url = Git.polish_url(url, expand_vars=False)
if not allow_unsafe_protocols:
    Git.check_unsafe_protocols(clone_url)      # the positional url only

git clone accepts a second URL via --bundle-uri=<uri>, which git dereferences before the main transport runs. That option is absent from unsafe_git_clone_options, so the option guard passes it, and check_unsafe_protocols() never inspects it. A caller-influenced value therefore drives an outbound request from the host:

Repo.clone_from(trusted_url, dest,
                multi_options=["--bundle-uri=http://169.254.169.254/latest/meta-data/"])
# no UnsafeProtocolError, no UnsafeOptionError

Confirmed against a local listener — the request leaves the process:

127.0.0.1 - - [24/Jul/2026 23:07:41] "GET /internal-metadata HTTP/1.1" 404 -

file:///path is likewise accepted without error. Note this is not a tokenisation bypass: multi_options is shlex.split before the check (per c9a26789 / GHSA-x2qx-6953-8485), so the fully-split --bundle-uri=... token is checked and legitimately passes because the option is not on the list.

Why it belongs with this report: both are the membership question rather than the matching logic — is the set of blocked options complete, and does the protocol guard inspect every URL git will dereference? The structural remediation proposed above covers both if extended slightly: prefer an allowlist per command, and route every URL-bearing option through check_unsafe_protocols(), not only the positional URL. Adding --bundle-uri to unsafe_git_clone_options would be the minimal fix.

Affected versions

PyPI

PackageVulnerableFixed in
gitpython
< 3.1.57
3.1.57

How to fix

Move to the fixed version or later with your package manager, update the lockfile, and run your tests. If the package only arrives through another dependency, upgrade that dependency first. If none of its releases pulls a fixed version yet, force the version with your package manager's override.

gitpython PyPI → 3.1.57

pip
pip install --upgrade "gitpython>=3.1.57"
poetry
poetry add "gitpython>=3.1.57"
uv
uv add "gitpython>=3.1.57"

Severity breakdown

CVSS 4.0 0.0

CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:H/VI:N/VA:N/SC:N/SI:N/SA:N/E:X/CR:X/IR:X/AR:X/MAV:X/MAC:X/MAT:X/MPR:X/MUI:X/MVC:X/MVI:X/MVA:X/MSC:X/MSI:X/MSA:X/S:X/AU:X/R:X/V:X/RE:X/U:X

Attack vector
Network
Attack complexity
Low
Attack requirements
None
Privileges required
Low
User interaction
None
Confidentiality (vulnerable system)
High
Integrity (vulnerable system)
None
Availability (vulnerable system)
None
Confidentiality (subsequent systems)
None
Integrity (subsequent systems)
None
Availability (subsequent systems)
None
Exploit maturity
Not defined
Scope
X

CVSS 3.1 6.5

CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:N

Attack vector
Network
Attack complexity
Low
Privileges required
Low
User interaction
None
Scope
Unchanged
Confidentiality
High
Integrity
None
Availability
None

Weakness

CWE-200: Exposure of Sensitive Information to an Unauthorized Actor

The product exposes sensitive information to an actor that is not explicitly authorized to have access to that information.

How to prevent it
  • Architecture and Design: Compartmentalize the system to have "safe" areas where trust boundaries can be unambiguously drawn. Do not allow sensitive data to go outside of the trust boundary and always be careful when interfacing with a compartment outside of the safe area. Ensure that appropriate compartmentalization is built into the system design, and the compartmentalization allows for and reinforces privilege separation functionality. Architects and designers should rely on the principle of least privilege to decide the appropriate time to use privileges and the time to drop privileges.

CWE-73: External Control of File Name or Path

The product allows user input to control or influence paths or file names that are used in filesystem operations.

How to prevent it
  • Architecture and Design: When the set of filenames is limited or known, create a mapping from a set of fixed input values (such as numeric IDs) to the actual filenames, and reject all other inputs. For example, ID 1 could map to "inbox.txt" and ID 2 could map to "profile.txt". Features such as the ESAPI AccessReferenceMap provide this capability.
  • Architecture and Design, Operation: Run your code in a "jail" or similar sandbox environment that enforces strict boundaries between the process and the operating system. This may effectively restrict all access to files within a particular directory. Examples include the Unix chroot jail and AppArmor. In general, managed code may provide some protection. This may not be a feasible solution, and it only limits the impact to the operating system; the rest of your application may still be subject to compromise. Be careful to avoid CWE-243 and other weaknesses related to jails.
  • Architecture and Design: For any security checks that are performed on the client side, ensure that these checks are duplicated on the server side, in order to avoid CWE-602. Attackers can bypass the client-side checks by modifying values after the checks have been performed, or by changing the client to remove the client-side checks entirely. Then, these modified values would be submitted to the server.
  • Implementation: Assume all input is malicious. Use an "accept known good" input validation strategy, i.e., use a list of acceptable inputs that strictly conform to specifications. Reject any input that does not strictly conform to specifications, or transform it into something that does. When performing input validation, consider all potentially relevant properties, including length, type of input, the full range of acceptable values, missing or extra inputs, syntax, consistency across related fields, and conformance to business rules. As an example of business rule logic, "boat" may be syntactically valid because it only contains alphanumeric characters, but it is not valid if the input is only expected to contain colors such as "red" or "blue." Do not rely exclusively on looking for malicious or malformed inputs. This is likely to miss at least one undesirable input, especially if the code's environment changes. This can give attackers enough room to bypass the intended validation. However, denylists can be useful for detecting potential attacks or determining which inputs are so malformed that they should be rejected outright. When validating filenames, use stringent allowlists that limit the character set to be used. If feasible, only allow a single "." character in the filename to avoid weaknesses such as CWE-23, and exclude directory separators such as "/" to avoid CWE-36. Use a list of allowable file extensions, which will help to avoid CWE-434. Do not rely exclusively on a filtering mechanism that removes potentially dangerous characters. This is equivalent to a denylist, which may be incomplete (CWE-184). For example, filtering "/" is insufficient protection if the filesystem also supports the use of "\" as a directory separator. Another possible error could occur when the filtering is applied in a way that still produces dangerous data (CWE-182). For example, if "../" sequences are removed from the ".../...//" string in a sequential fashion, two instances of "../" would b
  • Implementation: Use a built-in path canonicalization function (such as realpath() in C) that produces the canonical version of the pathname, which effectively removes ".." sequences and symbolic links (CWE-23, CWE-59).
  • Installation, Operation: Use OS-level permissions and run as a low-privileged user to limit the scope of any successful attack.
  • Operation, Implementation: If you are using PHP, configure your application so that it does not use register_globals. During implementation, develop your application so that it does not rely on this feature, but be wary of implementing a register_globals emulation that is subject to weaknesses such as CWE-95, CWE-621, and similar issues.
  • Testing: Use tools and techniques that require manual (human) analysis, such as penetration testing, threat modeling, and interactive tools that allow the tester to record and modify an active session. These may be more effective than strictly automated techniques. This is especially the case with weaknesses that are related to design and business rules.

References

Timeline

  1. Published
  2. Last updated by the source
  3. Latest EPSS score
AdvisorySeverityWhy it's related
GHSA-83vg-56qc-22m7
GitPython validates a submodule's ".git" path only after cloning into it (RCE)
High Also affects gitpython
GHSA-gq48-pqfc-9p58
GitPython: Submodule.move() follows an attacker-committed intermediate symlink outside the worktree
Medium Also affects gitpython
GHSA-23mf-xhv8-69c2
GitPython: arbitrary file read via attacker-controlled branch name in `Head.checkout()` (incomplete fix of CVE-2026-78679)
High Also affects gitpython
GHSA-f9j4-qggq-h239
GitPython Remote.pull() refspec bypasses the --upload-pack unsafe-option guard, enabling arbitrary command execution
Critical Also affects gitpython
GHSA-w8jc-g24h-crhw
GitPython command wrappers omit unsafe protocol and option checks
High Also affects gitpython
GHSA-m64x-33q8-m5h7
GitPython parse_actor_and_date() is vulnerable to quadratic backtracking on malformed identity lines
High Also affects gitpython
GHSA-fx3j-rwgx-fr94
GitPython: separate_git_dir leaks env vars via crafted .gitmodules submodule name
High Also affects gitpython
GHSA-59cr-6r3x-644w
GitPython submodule update path traversal can write outside the repository
Medium Also affects gitpython
Is this in your code?

MergeHarbor checks every repository's dependencies against this database, ranks what matters by exploitability and reachability, routes it to the owning team and opens the fix pull request.

Check your repos with MergeHarbor