Singluarity ineffectively applies selinux / apparmor LSM process labels

Description

Impact

Native Mode (default)

Singularity's default native runtime allows users to apply restrictions to container processes using the apparmor or selinux Linux Security Modules (LSMs), via the --security selinux:<label> or --security apparmor:<profile> flags.

LSM labels are written to process or thread attrs/exec under /proc. If a user relies on LSM restrictions to prevent malicious operations then, under certain circumstances, an attacker can redirect the LSM label write operation so that it is ineffective. This requires:

  • The attacker to cause the user to run a malicious container image that redirects the mount of /proc to the destination of a shared mount, either known to be configured on the target system, or that will be specified by the user when running the container.
  • Control of the content of the shared mount, for example through another malicious container which also binds it, or as a user with relevant permissions on the host system it is bound from.

Note that Singularity does not attempt to prevent damaging operations, or container escape, from containers that are started as the host root user. When a non-root user starts a container any LSM writes to /proc are performed as that user. For these reasons, the denial-of-service and container escape attacks detailed in runc CVE-2025-52881 are not relevant. Processes running in non-root containers are subject to the standard permissions for the non-root account used, and cannot escalate privilege, even when intended container-specific LSM labels are not correctly applied.

In addition, a bug in the detection of selinux support in Singularity's default setuid flow means that --security selinux:<label> flags may not be applied, even in the absence of an attack - but in this case a warning message is emitted, indicating that selinux is unavailable. This warning may be may be overlooked, mis-interpreted, or not seen when singularity is run from a script or other tool. Failure to apply requested restrictions should result in a fatal error, rather than a warning message.

OCI-Mode

Singularity's OCI-mode is unaffected as it does not currently support applying LSM restrictions via the --security flag.

Patches

Ineffective write of selinux process labels is addressed via an update to the containers/selinux dependency in https://github.com/sylabs/singularity/pull/3850. This update brings in the upstream fix for CVE-2025-52881 in this dependency.

Ineffective write of apparmor process labels is addressed in commit 5af3e79.

Failure to detect apparmor / selinux support, when --security flags are provided, is made an error rather than a warning in commit 2788296.

Workarounds

There are no known workarounds, other than to define system-wide apparmor / selinux policy for Singularity itself. This would apply to all containers, not just those run with the --security flags. Additionally, restrictions that are reasonable to apply to container processes may impact the functionality of Singularity.

References

Related vulnerabilities in runc:

Basic information

Type
reviewed
Severity
medium
Advisory on GitHub
Open advisory ↗
Repository advisory
Open repository advisory ↗
Source code
Browse source ↗
Published (advisory)
2025-12-02 21:07:02 UTC
Updated
2025-12-09 19:51:29 UTC
GitHub reviewed
2025-12-02 21:07:02 UTC
NVD published
2025-12-02

EPSS Score

Score Percentile
0.02% 4.00%

CVSS Scores

Base score Version Severity Vector
4.5 3.1
CVSS:3.1/AV:L/AC:H/PR:N/UI:R/S:U/C:L/I:L/A:L Click to expand
Attack vector (AV:L)
They already need access on the box, or another person has to do something wrong; it’s not a remote drive-by.
Attack complexity (AC:H)
Even with access, the exploit needs extra luck, timing, or a fussy environment to actually work.
Privileges required (PR:N)
No account or special rights needed—anonymous or random user is enough.
User interaction (UI:R)
A real person has to do something—click, install, enable—otherwise it doesn’t land.
Scope (S:U)
Damage stays in the same “trust bubble” as the broken component—no big spill into unrelated systems.
Confidentiality (C:L)
Some sensitive info could get out, but not a total data dump.
Integrity (I:L)
Attackers could change some data, but it’s limited—not everything goes.
Availability (A:L)
Might cause slowdowns, glitches, or partial disruption—not a full brick.

Identifiers

CWEs

CWE id Name
CWE-61 UNIX Symbolic Link (Symlink) Following
CWE-706 Use of Incorrectly-Resolved Name or Reference

Affected packages (2)

Vulnerable version ranges and first patched releases as published by GitHub.

Ecosystem Package Vulnerable range First patched Vulnerable functions
go github.com/sylabs/singularity/v4 >= 4.2.0-rc.1, < 4.3.5 4.3.5
go github.com/sylabs/singularity/v4 < 4.1.11 4.1.11

References

cvelogic Threat Intelligence