[CVE-2022-26488] Escalation of privilege via Windows Installer

[zooba]

BPO	46948
Nosy	@gpshead, @pfmoore, @tjguk, @ned-deily, @ambv, @zware, @zooba, @pablogsal, @miss-islington
PRs	bpo-46948: Fix CVE-2022-26488 by ensuring the Windows Installer correctly uses the install path during repair #31726 [3.10] bpo-46948: Fix CVE-2022-26488 by ensuring the Windows Installer correctly uses the install path during repair #31727 [3.9] bpo-46948: Fix CVE-2022-26488 by ensuring the Windows Installer correctly uses the install path during repair #31728 [3.8] bpo-46948: Fix CVE-2022-26488 by ensuring the Windows Installer correctly uses the install path during repair #31729 [3.7] bpo-46948: Fix CVE-2022-26488 by ensuring the Windows Installer correctly uses the install path during repair #31730 bpo-46948: Fix launcher installer build failure due to first part of fix #31920 [3.10] bpo-46948: Fix launcher installer build failure due to first part of fix (GH-31920) #31922 [3.9] bpo-46948: Fix launcher installer build failure due to first part of fix (GH-31920) #31923 [3.8] bpo-46948: Fix launcher installer build failure due to first part of fix (GH-31920) #31924 [3.7] bpo-46948: Fix launcher installer build failure due to first part of fix (GH-31920) #31925

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = 'https://github.com/zooba'
closed_at = <Date 2022-03-16.12:23:48.554>
created_at = <Date 2022-03-07.16:33:17.993>
labels = ['type-security', '3.8', '3.9', '3.10', '3.11', '3.7', 'release-blocker', 'OS-windows']
title = '[CVE-2022-26488] Escalation of privilege via Windows Installer'
updated_at = <Date 2022-03-16.12:23:48.554>
user = 'https://github.com/zooba'

bugs.python.org fields:

activity = <Date 2022-03-16.12:23:48.554>
actor = 'steve.dower'
assignee = 'steve.dower'
closed = True
closed_date = <Date 2022-03-16.12:23:48.554>
closer = 'steve.dower'
components = ['Windows']
creation = <Date 2022-03-07.16:33:17.993>
creator = 'steve.dower'
dependencies = []
files = []
hgrepos = []
issue_num = 46948
keywords = ['patch']
message_count = 16.0
messages = ['414673', '414678', '414679', '414681', '414682', '414683', '414685', '414711', '414733', '414752', '415306', '415309', '415310', '415314', '415317', '415331']
nosy_count = 9.0
nosy_names = ['gregory.p.smith', 'paul.moore', 'tim.golden', 'ned.deily', 'lukasz.langa', 'zach.ware', 'steve.dower', 'pablogsal', 'miss-islington']
pr_nums = ['31726', '31727', '31728', '31729', '31730', '31920', '31922', '31923', '31924', '31925']
priority = 'release blocker'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = 'security'
url = 'https://bugs.python.org/issue46948'
versions = ['Python 3.7', 'Python 3.8', 'Python 3.9', 'Python 3.10', 'Python 3.11']

[zooba]

CVE-2022-26488 is an escalation of privilege vulnerability in the Windows installer for the following releases of CPython:

• 3.11.0a6 and earlier
• 3.10.2 and earlier
• 3.9.10 and earlier
• 3.8.12 and earlier
• All end-of-life releases of 3.5, 3.6 and 3.7

The vulnerability exists when installed for all users, and when the "Add Python to PATH" option has been selected. A local user without administrative permissions can trigger a repair operation that adds incorrect additional paths to the system PATH variable, and then use search path hijacking to achieve escalation of privilege. Per-user installs (the default) are also affected, but cannot be used for escalation of privilege.

Besides updating, this vulnerability may be mitigated by modifying an existing install to disable the "Add Python to PATH" or "Add Python to environment variables" option. Manually adding the install directory to PATH is not affected.

Thanks to the Lockheed Martin Red Team for detecting and reporting the issue to the Python Security Response Team.

[pablogsal]

The 3.11.0a6 release is ongoing. I assume is ok to not block this release on this issue, given that an alpha is inherently unsafe

[zooba]

Yeah, this is fine to still be in alpha 6. Very unlikely that anyone is making it a system-wide default anyway, and certainly not in secure/production systems.

[zooba]

New changeset 136842c by Steve Dower in branch '3.10':
bpo-46948: Fix CVE-2022-26488 by ensuring the Windows Installer correctly uses the install path during repair (GH-31727)
136842c

[zooba]

New changeset 77446d2 by Steve Dower in branch 'main':
bpo-46948: Fix CVE-2022-26488 by ensuring the Windows Installer correctly uses the install path during repair (GH-31726)
77446d2

[zooba]

New changeset 101a1be by Steve Dower in branch '3.9':
bpo-46948: Fix CVE-2022-26488 by ensuring the Windows Installer correctly uses the install path during repair (GH-31728)
101a1be

[ned-deily]

New changeset 9747627 by Steve Dower in branch '3.7':
bpo-46948: Fix CVE-2022-26488 by ensuring the Windows Installer correctly uses the install path during repair (GH-31730)
9747627

[gpshead]

Is there anything on our end we can do to prevent this kind of issue in the future?

Am I wrong to see this as just fixing our package to avoid a design flaw in Windows OS level package management?

Certainly other packages in the world must run into similar problems.

[ambv]

New changeset cff1b78 by Steve Dower in branch '3.8':
bpo-46948: Fix CVE-2022-26488 by ensuring the Windows Installer correctly uses the install path during repair (GH-31729)
cff1b78

[zooba]

Is there anything on our end we can do to prevent this kind of issue in the future?

Probably not, I think it's just a lesson learned about the capabilities of the MSI format and its integration with Windows (well, we could hurry up moving everyone to the Windows Store, which doesn't have this issue, but that seems unlikely ;) )

Similar issues have been reported to the Windows Installer team (e.g. CVE-2021-41379, CVE-2021-26415) that could have been fixed by disabling the unelevated repair function, but weren't. So I think it just has to become a known thing for people building MSIs that a "repair" can be run by non-elevated users, and install-time variables may not be preserved for the repair. (In our case, that means actually searching for the existing install rather than trusting the variable our bundle normally provides to the MSI.)

[zooba]

The fix for this regressed the installer for the py.exe launcher, which breaks our release builds.

I'm patching it now. It's going under the same issue number because it will be needed for anyone applying this patch directly and then building the installer themselves.

[zooba]

New changeset 7088120 by Steve Dower in branch 'main':
bpo-46948: Fix launcher installer build failure due to first part of fix (GH-31920)
7088120

[miss-islington]

New changeset 58d30b9 by Miss Islington (bot) in branch '3.10':
bpo-46948: Fix launcher installer build failure due to first part of fix (GH-31920)
58d30b9

[miss-islington]

New changeset 70eb9db by Miss Islington (bot) in branch '3.9':
bpo-46948: Fix launcher installer build failure due to first part of fix (GH-31920)
70eb9db

[ned-deily]

New changeset 4a1d65f by Miss Islington (bot) in branch '3.7':
bpo-46948: Fix launcher installer build failure due to first part of fix (GH-31920) (GH-31925)
4a1d65f

[ambv]

New changeset 2b97cfd by Miss Islington (bot) in branch '3.8':
bpo-46948: Fix launcher installer build failure due to first part of fix (GH-31920) (GH-31924)
2b97cfd