Compare commits

...
Sign in to create a new pull request.

1 commit

Author SHA1 Message Date
Lumir Balhar
40c9b4419c Security fixes for CVE-2026-13346 and CVE-2026-8643 2026-08-26 08:52:14 +02:00
3 changed files with 488 additions and 0 deletions

341
CVE-2026-13346.patch Normal file
View file

@ -0,0 +1,341 @@
From 6686b964762255a933e68939bb49710216ae7bb9 Mon Sep 17 00:00:00 2001
From: Damian Shaw <damian.peter.shaw@gmail.com>
Date: Fri, 14 Aug 2026 09:49:19 +0000
Subject: [PATCH 2/2] Fix Link.filename double URL decode - path traversal
(CVE-2026-13346)
Upstream PRs: https://github.com/pypa/pip/pull/14110
- Add PathComponent newtype to enforce single-component filenames
- Remove double urllib.parse.unquote() call in Link.filename
- Add join_within_directory() preventing path escape at download sites
- Update download.py and prepare.py call sites
- Backport test coverage from upstream commits 1, 3, 4 of PR #14110
Co-Authored-By: Lumir Balhar <lbalhar@redhat.com>
---
src/pip/_internal/models/link.py | 61 ++++++++++---
src/pip/_internal/network/download.py | 20 +++--
src/pip/_internal/operations/prepare.py | 6 +-
tests/unit/test_link.py | 113 +++++++++++++++++++++++-
4 files changed, 179 insertions(+), 21 deletions(-)
diff --git a/src/pip/_internal/models/link.py b/src/pip/_internal/models/link.py
index 200ec34..e6a0c87 100644
--- a/src/pip/_internal/models/link.py
+++ b/src/pip/_internal/models/link.py
@@ -14,6 +14,7 @@ from dataclasses import dataclass
from typing import (
Any,
NamedTuple,
+ NewType,
)
from pip._internal.exceptions import InvalidEggFragment
@@ -31,6 +32,47 @@ from pip._internal.utils.urls import path_to_url, url_to_path
logger = logging.getLogger(__name__)
+# A single path component: percent-decoded once and reduced to a basename, so it
+# contains no path separator and is not a ``.`` or ``..`` reference. The empty
+# string means "no component".
+PathComponent = NewType("PathComponent", str)
+
+
+def _to_path_component(name: str) -> PathComponent:
+ """Reduce ``name`` to a single path component, or ``""`` if it has none.
+
+ ``os.path.basename`` drops any directory part, drive letter, or separator;
+ a ``.``, ``..``, or empty result is not a component and becomes ``""``.
+ """
+ name = os.path.basename(name)
+ if name in ("", os.curdir, os.pardir):
+ return PathComponent("")
+
+ return PathComponent(name)
+
+
+def as_path_component(name: str) -> PathComponent:
+ """Like ``_to_path_component`` but reject the empty result.
+
+ Use where a file is about to be written, so a missing name is an error
+ rather than a silent fallback to the directory itself.
+ """
+ component = _to_path_component(name)
+ if not component:
+ raise ValueError(f"Unexpected file name derived from URL: {name!r}")
+
+ return component
+
+
+def join_within_directory(directory: str, component: PathComponent) -> str:
+ """Join a single path ``component`` onto ``directory``.
+
+ ``component`` is a :data:`PathComponent`, so by type it has no separator and
+ is not a ``.`` or ``..`` reference; the result can never escape ``directory``.
+ """
+ return os.path.join(directory, component)
+
+
# Order matters, earlier hashes have a precedence over later hashes for what
# we will pick to use.
_SUPPORTED_HASHES = ("sha512", "sha384", "sha256", "sha224", "sha1", "md5")
@@ -423,18 +465,13 @@ class Link:
return redact_auth_from_url(self.url)
@property
- def filename(self) -> str:
- path = self.path.rstrip("/")
- name = posixpath.basename(path)
- if not name:
- # Make sure we don't leak auth information if the netloc
- # includes a username and password.
- netloc, user_pass = split_auth_from_netloc(self.netloc)
- return netloc
-
- name = urllib.parse.unquote(name)
- assert name, f"URL {self._url!r} produced no filename"
- return name
+ def filename(self) -> PathComponent:
+ name = _to_path_component(posixpath.basename(self.path.rstrip("/")))
+ if name:
+ return name
+
+ # No component in the path; fall back to the netloc, dropping any auth.
+ return _to_path_component(split_auth_from_netloc(self.netloc)[0])
@property
def file_path(self) -> str:
diff --git a/src/pip/_internal/network/download.py b/src/pip/_internal/network/download.py
index 2696642..fa71c75 100644
--- a/src/pip/_internal/network/download.py
+++ b/src/pip/_internal/network/download.py
@@ -20,7 +20,12 @@ from pip._vendor.urllib3.exceptions import ReadTimeoutError
from pip._internal.cli.progress_bars import BarType, get_download_progress_renderer
from pip._internal.exceptions import IncompleteDownloadError, NetworkConnectionError
from pip._internal.models.index import PyPI
-from pip._internal.models.link import Link
+from pip._internal.models.link import (
+ Link,
+ PathComponent,
+ as_path_component,
+ join_within_directory,
+)
from pip._internal.network.cache import SafeFileCache, is_from_cache
from pip._internal.network.session import CacheControlAdapter, PipSession
from pip._internal.network.utils import HEADERS, raise_for_status, response_chunks
@@ -117,11 +122,14 @@ def parse_content_disposition(content_disposition: str, default_filename: str) -
return filename or default_filename
-def _get_http_response_filename(resp: Response, link: Link) -> str:
+def _get_http_response_filename(resp: Response, link: Link) -> PathComponent:
"""Get an ideal filename from the given HTTP response, falling back to
the link filename if not provided.
+
+ The result is validated as a single path component, so it can be joined onto
+ a download directory without escaping it.
"""
- filename = link.filename # fallback
+ filename: str = link.filename # fallback
# Have a look at the Content-Disposition header for a better guess
content_disposition = resp.headers.get("content-disposition")
if content_disposition:
@@ -135,7 +143,7 @@ def _get_http_response_filename(resp: Response, link: Link) -> str:
ext = os.path.splitext(resp.url)[1]
if ext:
filename += ext
- return filename
+ return as_path_component(filename)
@dataclass
@@ -188,7 +196,9 @@ class Downloader:
resp = self._http_get(link)
download_size = _get_http_response_size(resp)
- filepath = os.path.join(location, _get_http_response_filename(resp, link))
+ filepath = join_within_directory(
+ location, _get_http_response_filename(resp, link)
+ )
with open(filepath, "wb") as content_file:
download = _FileDownload(link, content_file, download_size)
self._process_response(download, resp)
diff --git a/src/pip/_internal/operations/prepare.py b/src/pip/_internal/operations/prepare.py
index 67f9ee9..d260d15 100644
--- a/src/pip/_internal/operations/prepare.py
+++ b/src/pip/_internal/operations/prepare.py
@@ -29,7 +29,7 @@ from pip._internal.exceptions import (
from pip._internal.index.package_finder import PackageFinder
from pip._internal.metadata import BaseDistribution, get_metadata_distribution
from pip._internal.models.direct_url import ArchiveInfo
-from pip._internal.models.link import Link
+from pip._internal.models.link import Link, join_within_directory
from pip._internal.models.wheel import Wheel
from pip._internal.network.download import Downloader
from pip._internal.network.lazy_wheel import (
@@ -201,7 +201,7 @@ def _check_download_dir(
"""Check download_dir for previously downloaded file with correct hash
If a correct file is found return its path else None
"""
- download_path = os.path.join(download_dir, link.filename)
+ download_path = join_within_directory(download_dir, link.filename)
if not os.path.exists(download_path):
return None
@@ -683,7 +683,7 @@ class RequirementPreparer:
# No distribution was downloaded for this requirement.
return
- download_location = os.path.join(self.download_dir, link.filename)
+ download_location = join_within_directory(self.download_dir, link.filename)
if not os.path.exists(download_location):
shutil.copy(req.local_file_path, download_location)
download_path = display_path(download_location)
diff --git a/tests/unit/test_link.py b/tests/unit/test_link.py
index c49f854..bc8cb8a 100644
--- a/tests/unit/test_link.py
+++ b/tests/unit/test_link.py
@@ -1,9 +1,17 @@
from __future__ import annotations
+import os
+import posixpath
+
import pytest
from pip._internal.exceptions import InvalidEggFragment, PipError
-from pip._internal.models.link import Link, links_equivalent
+from pip._internal.models.link import (
+ Link,
+ as_path_component,
+ join_within_directory,
+ links_equivalent,
+)
from pip._internal.utils.hashes import Hashes
@@ -29,6 +37,13 @@ class TestLink:
("https://example.com/path/page.html", "page.html"),
# Test a quoted character.
("https://example.com/path/page%231.html", "page#1.html"),
+ # A doubly-encoded separator must stay encoded: the path is decoded
+ # exactly once, so the file name keeps its literal "%2F" instead of
+ # collapsing into a "/".
+ (
+ "https://example.com/a%252Fb.whl",
+ "a%2Fb.whl",
+ ),
(
"http://yo/myproject-1.0%2Bfoobar.0-py2.py3-none-any.whl",
"myproject-1.0+foobar.0-py2.py3-none-any.whl",
@@ -49,6 +64,52 @@ class TestLink:
link = Link(url)
assert link.filename == expected
+ @pytest.mark.parametrize(
+ "url",
+ [
+ "https://example.com/a%252Fb.whl",
+ "https://example.com/%252e%252e%252fb.whl",
+ ],
+ )
+ def test_filename_decoded_once_stays_single_component(self, url: str) -> None:
+ # The path is decoded exactly once, so an encoded separator stays
+ # encoded and the file name remains a single path component rather
+ # than collapsing into a "/"-separated path.
+ filename = Link(url).filename
+ assert not posixpath.isabs(filename)
+ assert posixpath.basename(filename) == filename
+
+ @pytest.mark.parametrize(
+ "url",
+ [
+ "https://example.com/..",
+ "https://example.com/.",
+ "https://example.com/foo/%2e%2e",
+ ],
+ )
+ def test_filename_parent_reference_falls_back_to_netloc(self, url: str) -> None:
+ # A path that is only a "." or ".." reference has no usable file name,
+ # so filename falls back to the netloc rather than handing back a
+ # traversal component that could escape a download directory.
+ assert Link(url).filename == "example.com"
+
+ @pytest.mark.parametrize(
+ "url",
+ [
+ # A path-less URL whose authority looks like a traversal: the netloc
+ # fallback must still reduce to a single path component.
+ "http://..\\..\\..\\evil.whl",
+ "http://../",
+ "http://..",
+ ],
+ )
+ def test_filename_is_always_a_path_component(self, url: str) -> None:
+ # filename must never carry a separator or parent reference, so joining
+ # it onto a directory can never escape that directory.
+ name = Link(url).filename
+ assert os.path.basename(name) == name
+ assert name not in (os.curdir, os.pardir)
+
def test_splitext(self) -> None:
assert ("wheel", ".whl") == Link("http://yo/wheel.whl").splitext()
@@ -244,3 +305,53 @@ def test_links_equivalent(url1: str, url2: str) -> None:
)
def test_links_equivalent_false(url1: str, url2: str) -> None:
assert not links_equivalent(Link(url1), Link(url2))
+
+
+@pytest.mark.parametrize(
+ "name",
+ [
+ "wheel.whl",
+ "myproject-1.0+foobar.0-py2.py3-none-any.whl",
+ # A literal "%2F" is a normal file name, not a separator.
+ "a%2Fb.whl",
+ ],
+)
+def test_as_path_component_keeps_plain_name(name: str) -> None:
+ assert as_path_component(name) == name
+
+
+@pytest.mark.parametrize(
+ "name",
+ [
+ os.path.join(os.sep, "abs", "pkg.whl"),
+ os.path.join("..", "pkg.whl"),
+ os.path.join("nested", "pkg.whl"),
+ ],
+)
+def test_as_path_component_reduces_to_basename(name: str) -> None:
+ # A name carrying directory components is reduced to its basename, so the
+ # result always stays inside the directory it is later joined onto.
+ assert as_path_component(name) == os.path.basename(name)
+
+
+@pytest.mark.parametrize("name", ["", ".", "..", "/", os.path.join("sub", "..")])
+def test_as_path_component_rejects_empty_or_parent_reference(name: str) -> None:
+ with pytest.raises(ValueError):
+ as_path_component(name)
+
+
+@pytest.mark.parametrize(
+ "name",
+ [
+ "pkg.whl",
+ # A literal "%2F" is a normal file name, not a separator.
+ "a%2Fb.whl",
+ ],
+)
+def test_join_within_directory_stays_inside(name: str) -> None:
+ # The component is joined onto the directory as its final element, so the
+ # result stays inside the directory.
+ directory = os.path.join("base", "downloads")
+ joined = join_within_directory(directory, as_path_component(name))
+ assert joined == os.path.join(directory, name)
+ assert os.path.basename(joined) == name
--
2.55.0

139
CVE-2026-8643.patch Normal file
View file

@ -0,0 +1,139 @@
From c840c212830bb8e9dab56de6c57916b111858aec Mon Sep 17 00:00:00 2001
From: Damian Shaw <damian.peter.shaw@gmail.com>
Date: Mon, 18 May 2026 23:04:43 -0400
Subject: [PATCH 1/2] Reject entry point names that escape scripts dir
---
src/pip/_internal/operations/install/wheel.py | 26 +++++++-
tests/unit/test_wheel.py | 64 +++++++++++++++++++
2 files changed, 87 insertions(+), 3 deletions(-)
diff --git a/src/pip/_internal/operations/install/wheel.py b/src/pip/_internal/operations/install/wheel.py
index 40097d6..231e400 100644
--- a/src/pip/_internal/operations/install/wheel.py
+++ b/src/pip/_internal/operations/install/wheel.py
@@ -397,11 +397,31 @@ class MissingCallableSuffix(InstallationError):
)
-def _raise_for_invalid_entrypoint(specification: str) -> None:
+def _script_within_dir(name: str, scripts_dir: str) -> bool:
+ """Return whether script ``name`` resolves to a path inside the ``scripts_dir``.
+
+ distlib joins the entry point name onto the scripts directory, so a name
+ with path separators or ``..`` components can resolve elsewhere.
+ """
+ root = os.path.normpath(scripts_dir)
+ dest = os.path.normpath(os.path.join(scripts_dir, name))
+ return dest.startswith(root + os.sep)
+
+
+def _raise_for_invalid_entrypoint(specification: str, scripts_dir: str) -> None:
entry = get_export_entry(specification)
- if entry is not None and entry.suffix is None:
+ if entry is None:
+ return
+
+ if entry.suffix is None:
raise MissingCallableSuffix(str(entry))
+ if not _script_within_dir(entry.name, scripts_dir):
+ raise InstallationError(
+ f"Invalid script entry point name {entry.name!r}: the script "
+ f"would be installed outside the scripts directory ({scripts_dir})."
+ )
+
class PipScriptMaker(ScriptMaker):
# Override distlib's default script template with one that
@@ -419,7 +439,7 @@ class PipScriptMaker(ScriptMaker):
def make(
self, specification: str, options: dict[str, Any] | None = None
) -> list[str]:
- _raise_for_invalid_entrypoint(specification)
+ _raise_for_invalid_entrypoint(specification, self.target_dir)
return super().make(specification, options)
diff --git a/tests/unit/test_wheel.py b/tests/unit/test_wheel.py
index e0ac649..c1aafe5 100644
--- a/tests/unit/test_wheel.py
+++ b/tests/unit/test_wheel.py
@@ -462,6 +462,32 @@ class TestInstallUnpackedWheel:
assert os.path.basename(wheel_path) in exc_text
assert entrypoint in exc_text
+ @pytest.mark.parametrize("bad_name", ["../../outside", "..", "."])
+ @pytest.mark.parametrize("entry_point_type", ["console_scripts", "gui_scripts"])
+ def test_wheel_install_rejects_entry_point_path_traversal(
+ self, data: TestData, tmpdir: Path, bad_name: str, entry_point_type: str
+ ) -> None:
+ """An entry point name with separators or ``..`` must not install a
+ script outside the scripts directory.
+ """
+ self.prep(data, tmpdir)
+ wheel_path = make_wheel(
+ "simple",
+ "0.1.0",
+ entry_points={entry_point_type: [f"{bad_name} = simple:main"]},
+ ).save_to_dir(tmpdir)
+ with pytest.raises(InstallationError) as e:
+ wheel.install_wheel(
+ "simple",
+ str(wheel_path),
+ scheme=self.scheme,
+ req_description="simple",
+ )
+
+ assert "outside the scripts directory" in str(e.value)
+ # Nothing was written outside the install destination.
+ assert not os.path.exists(os.path.join(str(tmpdir), "outside"))
+
class TestMessageAboutScriptsNotOnPATH:
tilde_warning_msg = (
@@ -665,3 +691,41 @@ def test_get_console_script_specs_replaces_python_version(
"not_pip_or_easy_install-99 = whatever",
"not_pip_or_easy_install-99.88 = whatever",
]
+
+
+@pytest.mark.parametrize(
+ "name, within",
+ [
+ ("pip", True),
+ ("pip3.13", True),
+ ("foo-bar.baz", True),
+ ("...", True), # a literal filename, not a path component
+ ("sub/script", True), # in-tree subdirectory
+ ("a/../b", True),
+ ("sub\\script", True), # backslash stays in-tree on POSIX and Windows
+ (" ../../inside", True), # distlib keeps a leading space; resolves in-tree
+ ("../outside", False),
+ ("../../outside", False),
+ ("a/../../outside", False),
+ ("/etc/cron.d/outside", False), # absolute path; os.path.join drops the root
+ # "." and ".." pass PyPI's [\w.-]+ name check but must be rejected here.
+ (".", False),
+ ("..", False),
+ ("", False),
+ ],
+)
+def test_script_within_dir(name: str, within: bool) -> None:
+ assert wheel._script_within_dir(name, "/srv/env/bin") is within
+
+
+def test_script_within_dir_allows_doubled_slash_root() -> None:
+ # A scripts directory can have a doubled leading slash
+ assert wheel._script_within_dir("pip", "//srv/env/bin") is True
+ assert wheel._script_within_dir("../outside", "//srv/env/bin") is False
+
+
+@pytest.mark.skipif(not WINDOWS, reason="drive letters only matter on Windows")
+def test_script_within_dir_rejects_other_drive() -> None:
+ # Validate that a script on a different drive is rejected,
+ # and doesn't throw an error
+ assert wheel._script_within_dir("D:\\outside", "C:\\env\\bin") is False
--
2.55.0

View file

@ -101,6 +101,14 @@ Patch: downstream-remove-pytest-subket.patch
# Upstream fix: https://github.com/urllib3/urllib3/commit/f05b1329126d5be6de501f9d1e3e36738bc08857
Patch: urllib3-CVE-2025-50181.patch
# CVE-2026-8643: entry point path traversal in console_scripts/gui_scripts
# Upstream fix: https://github.com/pypa/pip/pull/14000
Patch: CVE-2026-8643.patch
# CVE-2026-13346: Link.filename double URL decode allows path traversal
# Upstream fix: https://github.com/pypa/pip/pull/14110
Patch: CVE-2026-13346.patch
# Remove -s from Python shebang - ensure that packages installed with pip
# to user locations are seen by pip itself
%undefine _py3_shebang_s