Table of Contents

1) Introduction

qSnapper is a GUI frontend for the snapper utility for managing Btrfs filesystem snapshots. In April we received a review request for qSnapper, because it contains a privileged D-Bus service and Polkit policies.

Normally, Btrfs snapshots can only be managed with root privileges. qSnapper aims to provide more user-friendly and fine-grained access to snapshot management features, based on a privileged daemon which utilizes D-Bus and Polkit. Our review of the service uncovered various security issues, which led to a longer coordinated disclosure to allow upstream to develop bugfixes, which are part of upstream release 1.3.3.

The full details of the security issues will be covered in the sections below. The original review was performed on qSnapper release 1.1.3. Since bigger upstream changes appeared in the meantime this report is based on qSnapper release 1.3.2, however.

2) Overview of the qSnapper D-Bus Service

The qSnapper project contains a daemon named qsnapper-dbus-service, which runs with full root privileges. The daemon provides the D-Bus interface “com.presire.qsnapper.Operations” on the system bus; some of its methods are provided to arbitrary users in the system without authentication, while the majority of them is protected by Polkit authentication checks. The implementation of the various D-Bus methods offered by the service is found in snapshotoperations.cpp. All of the security issues outlined below are located in this compilation unit.

3) Security Issues

3.1) Weak Polkit Authentication Check is Subject to Race Condition (CVE-2026-41045)

The SnapshotOperations::checkAuthorization() function uses Polkit’s UnixProcess subject in an unsafe way to authenticate clients. The code obtains the client’s PID from the active D-Bus connection, which leads to a race condition. At the time the Polkit daemon (polkitd) checks the provided PID, the process in question can already have been replaced by another, privileged process. This is a well-known class of Polkit authentication bypasses which was assigned CVE-2013-4288.

In effect this means that clients may be able to bypass all Polkit authentication checks performed by qSnapper, although the attack is somewhat complex and might need multiple attempts to succeed. There is nothing stopping a local attacker from doing that, however. In combination with the configuration path traversal in D-Bus methods like WriteSnapperConfig() described below, this can be used to achieve a full local root exploit.

To fix this, we suggested to use Polkit’s SystemBusName subject instead. With this subject, the D-Bus daemon obtains the client’s UID in a race-free fashion from the UNIX domain socket the client used to connect, and polkitd cannot be fooled by recycled PIDs.

3.2) Path Traversal via configName Parameter (CVE-2026-41046)

All the D-Bus methods accepting a configName parameter (that is nearly all of them) suffer from a path traversal vulnerability. This string parameter is directly passed to snapper::Snapper(), which is part of the libsnapper API, without checking its safety. libsnapper expects a basename here to look up a configuration file in /etc/snapper/configs. In the context of qSnapper this string can contain additional path components like ../../path/to/crafted.cfg. This allows clients to use arbitrary configuration files as input to libsnapper.

When looking at the writeSnapperConfig() D-Bus method, which requires admin authentication, this path traversal allows to write largely attacker-controlled data to arbitrary locations. Combined with the Polkit authentication bypass described earlier in section 3.1, this would allow for a full local root exploit.

In case of other D-Bus methods which use the “list-snapshots” Polkit action, the path traversal is also accessible to locally logged-in users without any authentication. Luckily the libsnapper configuration file format offers no features that lead to a trivial local root exploit in this context. The impact of the vulnerability can be any of the following, however:

  • local Denial-of-Service (DoS): by pointing libsnapper to a special file like /dev/zero, the daemon will consume the maximum amount of memory and be killed by the kernel. By pointing libsnapper to a named FIFO special file, the daemon will block indefinitely.
  • information leak from parsing of private files: by pointing libsnapper to a private file like /etc/shadow, libsnapper will parse the sensitive data from the file as configuration data. If any logging features are active and the logs can be accessed by unprivileged users, then private data from such a file might leak into the unprivileged context. We are not aware of such an information leak in default installations, however.
  • by providing crafted configuration data further attacks can be carried out:
    • SUBVOLUME can be pointed to arbitrary mounts in the system. Luckily libsnapper is conservative in its file operations here, and quite strictly verifies the validity of this path; it verifies if it is really a mount point and is backed by the expected block device and file system etc.
    • ALLOW_USERS and ALLOW_GROUPS can be set to arbitrary users and groups, allowing them officially to control the Snapper configuration.
    • The SYNC_ACL setting causes libsnapper to apply ACL entries to the .snapshots directory of the volume path, granting the users and groups from ALLOW_USERS and ALLOW_GROUPS to access the snapshot data. Thus this is a major local information leak.

libsnapper is using different backends depending on the file system type. Different code is used for each backend, which means that the exploitability of some aspects of the vulnerability depends on which backend is used. On openSUSE only the Btrfs and LVM thin volume backends are compiled-in by default, which only offer little additional attack surface.

To fix the issue, we suggested to upstream to verify the configName parameter in all cases and reject it if any / or .. components are found in it.

3.3) Information Leak via “diff” Methods (CVE-2026-41047)

The Polkit action “list-snapshots” is allowed for locally logged-in users without authentication, and is used by multiple qSnapper D-Bus methods which are likely considered “read-only” operations not harmful to the system. In version 1.1.3 of qSnapper the following methods are relying on this Polkit action:

  • ListSnapshots()
  • GetFileChanges()
  • GetFileDiffAndDetails()

In version 1.3.2 of qSnapper these additional methods are using the action:

  • GetFileChangesBetween()
  • GetFileDiffBetween()

These methods allow to obtain information about metadata changes of arbitrary files between snapshots or between snapshots and the live filesystem. The newer GetFileDiffBetween() method even offers full file content diff output. These are major local information leaks, since unprivileged users can get a diff e.g. of the /etc/shadow file between different system states, or changes from root’s home directory which could leak sensitive data like passwords or private keys.

To fix this, we suggested to upstream that all methods providing non-public information get restricted by auth_admin Polkit checks. Alternatively, a check could be implemented whether the files to be diffed are normally accessible to the caller by using checks similar to the access() system call; such an approach would likely be rather complex and error-prone, however.

3.4) Caching of Authentication allows Authentication Bypass (CVE-2026-41048, CVE-2026-41049)

In version 1.2.1 of qSnapper an m_authenticated flag was introduced to the daemon’s code, which caches authentication once a client has passed certain Polkit action authentication checks. This cached authentication is shared between the methods DeleteSnapshot(), RestoreFiles() and RestoreFilesDirect(). There are two issues with this approach:

  • DeleteSnapshot() uses the “delete-snapshot” action, while the other methods use the “rollback-snapshot” action. If a caller can authenticate for “delete-snapshot” then it is also implicitly authenticated for “rollback-snapshot” and vice versa. This is not how Polkit is intended to be used. If a system administrator decides to relax the authentication requirement for “delete-snapshot”, then this automatically implies the “rollback-snapshot” action now, which are two very different impacts to the system. We assigned CVE-2026-41048 for this aspect of the issue.
  • The implementation of the caching logic assumes that a single interactive client is talking to the daemon. In fact, arbitrary users can invoke these methods. If a legitimate client is authenticated for one of the affected actions once, then arbitrary other users like nobody are now also considered authenticated and Polkit will not be invoked anymore. This authentication bypass allows for a full local root exploit in the context of the RestoreFiles() methods, as outlined below in section 3.6. We assigned CVE-2026-41049 for this aspect of the issue.

We suggested to upstream to drop this form of authentication caching, or alternatively implement an authentication cache tied to each user and each Polkit action in question to avoid the issues.

3.5) Defense-in-Depth Issues when Restoring Arbitrary Files

This section discusses Defense-in-Depth issues which remain once all the more tangible issues discussed above are fixed. The restoreFiles() D-Bus method accepts arbitrary filePaths as parameter to be restored from a version found in a snapshot. Clients need to authenticate as an administrator to do this, so there is no direct attack surface. Let’s consider the path /var/lib/colord/mapping.db to be restored, however:

# On openSUSE Tumbleweed this directory is owned and writable by `colord`
$ ls -lh /var/lib/colord
drwxr-xr-x 3 colord colord 4.0K Mar 6 2024 .

At the time the qSnapper service performs the chown() system call, the colord service user can stage a symlink attack in the colord directory and cause arbitrary files in the system to receive colord ownership, allowing a colord to root exploit. A similar issue affects the copyRegularFile() function, where a file on the live file system is newly created.

This issue is hard to fix, given the many different file system situations that the restore operation can end up in. The basic algorithm to address the problem would need to use low-level system calls like openat(), fchown() and fstatat() to safely traverse file system paths and inspect any symbolic links on the way. Given the complexity of the bugfix and limited impact, we agreed with the upstream author that this fix does not need to be prepared during the coordinated disclosure period but can be developed in the open afterwards.

Given the amount of front-line security issues already found in qSnapper in the course of this review, we refrained from assigning a dedicated CVE for these Defense-in-Depth issues in this case.

3.6) Other Issues

This section covers a number of other issues that are not fully-fledged security issues, but still represent weaknesses or possible hardenings for the qSnapper implementation.

  • the D-Bus method RestoreFiles() implements complex file operations to rollback changes of a file path to the state found in a snapshot. These file operations allow authenticated clients to achieve nearly arbitrary file operations:
    • recursive deletion of arbitrary paths
    • copy of arbitrary files to arbitrary locations
    • change of ownership of files to arbitrary users
    • change of the file mode to arbitrary values

    This is because the method is not verifying whether the claimed file exists in the snapshot in the first place and whether the filePaths contain any ../ directory components to escape the snapshot directory. While this method is protected by Polkit’s auth_admin setting, any D-Bus method should be written conservatively and in such a way that the intended purpose of the method cannot be bypassed. The very similar method RestoreFilesDirect() has the same issues and also contains a lot of duplicate code which should be merged with the other method’s code.

  • the Quit() method can be called by anyone in the system without authentication. This is kind of a local DoS against an active qSnapper daemon in use by other users in the system.
  • the logfile created in /var/log/qSnapper will be world-readable, which could leak sensitive information. For hardening purposes it could be made private to root.

4) Upstream Bugfixes

The upstream author closely followed our suggestions to fix the security issues described in this report. Version 1.3.3 of qSnapper contains the following bugfixes:

  • commit a6caf538 addresses the following issues:
    • the Polkit UnixProcess subject is replaced by the SystemBusName subject to avoid the race condition during authentication (issue 3.1).
    • the configName parameter in all D-Bus methods is now carefully verified and rejected if it allows path traversal (issue 3.2).
    • the authentication caching is removed (issue 3.4).
    • the RestoreFiles methods reject path traversal attempts (see section 3.6).
  • commit f375b74c addresses the following issues:
    • a new view-diff Polkit action is introduced to generate file diffs which requires auth_admin authentication to prevent information leaks (issue 3.3).
    • the unauthenticated Quit D-Bus method was dropped (see section 3.6).
    • log files are now created without world-readable bit (see section 3.6).

The more complex defense-in-depth issues outlined in section 3.5 will be addressed by upstream in the open at a later time.

5) Disclosure Process

Once we established contact with the upstream author, we had a very constructive discussion of the coordinated disclosure process. Upstream quickly acknowledged the issues, opted for coordinated disclosure and communicated a time schedule when bugfixes were expected to be available for review and when a bugfix release of qSnapper would be published. When we agreed on the final bugfixes in the middle of May, we agreed on choosing an earlier publication date than originally intended.

We want to thank the upstream author for handling this report in a very professional fashion and for greatly improving the security and robustness of qSnapper.

6) Timeline

2026-04-09 Lacking any other point of contact, we created a public GitHub issue asking for a security contact.
2026-04-13 The upstream developer responded in the issue providing an email address for us to reach out to.
2026-04-13 We reached out to the developer privately by email providing a detailed report and offering coordinated disclosure.
2026-04-14 Due to issues with mail servers our email did not reach the upstream developer. Upstream offered us another email address to try.
2026-04-15 We received a reply from the upstream developer confirming the issues and requesting coordinated disclosure. Upstream accepted our offer to assign CVEs for the issues. A preliminary disclosure date of 2026-07-07 was established.
2026-04-17 We assigned CVEs and shared them with upstream.
2026-05-08 We received a tarball containing the current release candidate code of qSnapper.
2026-05-11 We asked the upstream author if individual bugfixes could be supplied instead.
2026-05-12 We received the individual patches we asked for.
2026-05-12 We provided feedback on the current state of the patches. Apart from minor details and hardening suggestions the patches were already in good shape.
2026-05-14 We received updated patches for review and found no further issues. Given the good progress on the bugfixes, an earlier coordinated release date of 2026-05-26 was agreed upon.
2026-05-26 As planned, upstream released qSnapper 1.3.3 containing the bugfixes.
2026-05-26 Publication of this report.

7) References