CVE-2026-99170

namespace: a mount that copy_tree() leaves out can't take over a lock

Published Oct 28, 2026 CVSS 3.1 · 8.8 High EPSS < 1% KEV · Not listed
Description
In the Linux kernel, the following vulnerability has been resolved:

namespace: a mount that copy_tree() leaves out can't take over a lock

Since commit c62a4766937e ("move_mount: transfer MNT_LOCKED") a mount
can be moved beneath a locked mount. The lock moves from the old cover
to the new mount which takes over the job of hiding what's under it and
the old cover can be unmounted. That works as long as the new mount
stays. But the caller picks the new mount and copy_tree() leaves a
mount of a mount namespace file out of every copy made without
CL_COPY_MNT_NS_FILE, together with everything mounted on it, and never
looks at MNT_LOCKED. unshare(CLONE_NEWNS) and mount propagation make
such copies.

So a task that is root in its own user namespace gets rid of a locked
cover on a file without a race and without touching the filesystem
below it. It makes a child with a mount namespace of its own, binds the
child's /proc/<pid>/ns/mnt beneath the cover with MOVE_MOUNT_BENEATH,
which hands the lock to that bind, and calls unshare(CLONE_NEWNS). The
new mount namespace has nothing on the file. That's the usual way to
mask a single file, a bind of /dev/null or of an empty file. With a
bind of an ordinary file the file stays covered.

copy_tree() already refuses to leave out a locked unbindable mount with
-EPERM since commit df7342b24018 ("mount: Don't allow copying
MNT_UNBINDABLE|MNT_LOCKED mounts"). Do the same for a locked mount of a
mount namespace file. That also covers a tree copied with
CL_COPY_MNT_NS_FILE and attached into a mount namespace of another user
namespace where lock_mnt_tree() locks it. Note that unshare(CLONE_NEWNS)
and clone(CLONE_NEWNS) fail with EPERM from such a mount namespace from
then on. The mount can't be left out and it can't be unmounted.

And refuse to move a mount of a mount namespace file beneath a locked
mount in the first place. Otherwise the task ends up with a mount it
can't unmount and an unshare() that fails for good.

Chris also found a second mount that goes away by itself, a clone of an
automounted mount that shrink_submounts() took along with its parent.
Commit f78ddcc87224 ("namespace: never expire a locked mount") took
care of that one.

Chris's reproducer is appended. Mode nsfile reads SECRET after the
unshare() on 7.3-rc7, with this patch the move_mount() fails with
EINVAL and the file stays covered. nsfile-control uses an ordinary file
and reads MASK both ways.

// SPDX-License-Identifier: GPL-2.0
/*
* Get rid of a locked cover with MOVE_MOUNT_BENEATH.
*
* The "admin" (root) makes a tmpfs W:
* W/D file, holds SECRET, covered by a bind of a file with MASK
* W/Dd/inner file, holds SECRET, Dd is covered by a bind of a directory
* whose "inner" holds MASK
* W/own directory of uid 1000 with the file "mine" and the
* directory "minedir", whose "inner" holds MINE
* W/dbg debugfs, mode 755. dbg/tracing is an automount point
* The "attacker" is uid 1000. It enters a new user namespace with a new mount
* namespace, so W and both covers are locked there.
*
* usage: prog MODE
* plain put a bind of own/mine beneath the cover of D, unmount the cover
* nsfile put a bind of the mount namespace file of a child beneath the
* cover of D, then unshare(CLONE_NEWNS)
* nsfile-control the same with a bind of own/mine
* shrink put a bind of the automounted tracefs beneath the cover of Dd,
* unmount the cover, put a tmpfs beneath W, call umount(W), which fails
* shrink-control the same with a bind of own/minedir
* The environment variable AUTOMOUNT_DIR names another automount point for "shrink".
*/
#define _GNU_SOURCE
#include <errno.h>
#include <fcntl.h>
#include <grp.h>
#include <sched.h>
#include <signal.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>

---truncated---

Analysis

AI-generated

Technical impact

The vulnerability originates in the Linux kernel’s handling of mount namespace locks during copy_tree() operations. When a copy is performed without including CL_COPY_MNT_NS_FILE, the function may omit a mount of a mount‑namespace file, allowing a new mount to be placed under a previously locked mount. A process that is root in its own user namespace can exploit this to move the missing mount beneath the lock, take over the lock, and unmount the covering mount without a race. The attacker can then access or modify files that were previously protected, effectively escalating privileges locally. This flaw is related to improper handling of lock states, enabling unauthorized removal of a mounting guard.

Affected Systems

All Linux kernel versions that lack the fixes introduced by commits c62a4766937e, df7342b24018, and f78ddcc87224 are affected. Common distribution kernels released before the merge of these commits remain vulnerable. The issue specifically impacts operations involving mount namespace manipulation such as bind mounts, MOVE_MOUNT_BENEATH, and unshare(CLONE_NEWNS) or clone(CLONE_NEWNS). Kernels that have incorporated these commits are no longer vulnerable.

Risk and Exploitability

The CVSS score of 8.8 classifies the flaw as high severity. The EPSS score indicates a very low probability of exploitation at present (< 1%), and it is not listed in the CISA KEV catalog. The exploit requires local root within a user namespace and the ability to perform mount operations; it does not provide remote code execution or system‑wide compromise. The most likely attack surface is an environment that allows untrusted code to create or manipulate user namespaces, such as container runtimes or applications that spawn namespaces (inferred from the described requirements). Organizations that run such workloads should consider this risk.

Remediation

No solution or workaround provided in the CVE record.

OpenCVE Recommended Actions

  • Apply a Linux kernel version that includes the patch commits c62a4766937e, df7342b24018, and f78ddcc87224, which correct lock handling during mount namespace operations.
  • If a kernel upgrade is not possible, enforce restrictions on CL_COPY_MNT_NS_FILE and prohibit binding of mount namespace files under locked mounts using seccomp or policy controls.
  • Limit or prevent the use of unshare(CLONE_NEWNS) and clone(CLONE_NEWNS) in untrusted code paths to mitigate the creation of privileged namespaces that could exploit the flaw.

Generated by OpenCVE AI on October 31, 2026 at 13:36 UTC.

Tracking

Sign in to view the affected projects.

Advisories

No advisories available
History
Type Values Removed Values Added
Metrics cvssV3_1

{'score': 5.5, 'vector': 'CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:N'}

cvssV3_1

{'score': 8.8, 'vector': 'CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:C/C:H/I:H/A:H'}

Type Values Removed Values Added
Weaknesses CWE-269
CWE-362
Type Values Removed Values Added
Weaknesses CWE-281
References
Metrics threat_severity

None

cvssV3_1

{'score': 5.5, 'vector': 'CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:N'}

threat_severity

Moderate

Type Values Removed Values Added
First Time appeared Linux linux
Vendors & Products Linux linux
Type Values Removed Values Added
Weaknesses CWE-269
CWE-362
Type Values Removed Values Added
Description In the Linux kernel, the following vulnerability has been resolved: namespace: a mount that copy_tree() leaves out can't take over a lock Since commit c62a4766937e ("move_mount: transfer MNT_LOCKED") a mount can be moved beneath a locked mount. The lock moves from the old cover to the new mount which takes over the job of hiding what's under it and the old cover can be unmounted. That works as long as the new mount stays. But the caller picks the new mount and copy_tree() leaves a mount of a mount namespace file out of every copy made without CL_COPY_MNT_NS_FILE, together with everything mounted on it, and never looks at MNT_LOCKED. unshare(CLONE_NEWNS) and mount propagation make such copies. So a task that is root in its own user namespace gets rid of a locked cover on a file without a race and without touching the filesystem below it. It makes a child with a mount namespace of its own, binds the child's /proc/<pid>/ns/mnt beneath the cover with MOVE_MOUNT_BENEATH, which hands the lock to that bind, and calls unshare(CLONE_NEWNS). The new mount namespace has nothing on the file. That's the usual way to mask a single file, a bind of /dev/null or of an empty file. With a bind of an ordinary file the file stays covered. copy_tree() already refuses to leave out a locked unbindable mount with -EPERM since commit df7342b24018 ("mount: Don't allow copying MNT_UNBINDABLE|MNT_LOCKED mounts"). Do the same for a locked mount of a mount namespace file. That also covers a tree copied with CL_COPY_MNT_NS_FILE and attached into a mount namespace of another user namespace where lock_mnt_tree() locks it. Note that unshare(CLONE_NEWNS) and clone(CLONE_NEWNS) fail with EPERM from such a mount namespace from then on. The mount can't be left out and it can't be unmounted. And refuse to move a mount of a mount namespace file beneath a locked mount in the first place. Otherwise the task ends up with a mount it can't unmount and an unshare() that fails for good. Chris also found a second mount that goes away by itself, a clone of an automounted mount that shrink_submounts() took along with its parent. Commit f78ddcc87224 ("namespace: never expire a locked mount") took care of that one. Chris's reproducer is appended. Mode nsfile reads SECRET after the unshare() on 7.3-rc7, with this patch the move_mount() fails with EINVAL and the file stays covered. nsfile-control uses an ordinary file and reads MASK both ways. // SPDX-License-Identifier: GPL-2.0 /* * Get rid of a locked cover with MOVE_MOUNT_BENEATH. * * The "admin" (root) makes a tmpfs W: * W/D file, holds SECRET, covered by a bind of a file with MASK * W/Dd/inner file, holds SECRET, Dd is covered by a bind of a directory * whose "inner" holds MASK * W/own directory of uid 1000 with the file "mine" and the * directory "minedir", whose "inner" holds MINE * W/dbg debugfs, mode 755. dbg/tracing is an automount point * The "attacker" is uid 1000. It enters a new user namespace with a new mount * namespace, so W and both covers are locked there. * * usage: prog MODE * plain put a bind of own/mine beneath the cover of D, unmount the cover * nsfile put a bind of the mount namespace file of a child beneath the * cover of D, then unshare(CLONE_NEWNS) * nsfile-control the same with a bind of own/mine * shrink put a bind of the automounted tracefs beneath the cover of Dd, * unmount the cover, put a tmpfs beneath W, call umount(W), which fails * shrink-control the same with a bind of own/minedir * The environment variable AUTOMOUNT_DIR names another automount point for "shrink". */ #define _GNU_SOURCE #include <errno.h> #include <fcntl.h> #include <grp.h> #include <sched.h> #include <signal.h> #include <stdio.h> #include <stdlib.h> #include <string.h> #include <unistd.h> ---truncated---
Title namespace: a mount that copy_tree() leaves out can't take over a lock
First Time appeared Linux
Linux linux Kernel
CPEs cpe:2.3:o:linux:linux_kernel:*:*:*:*:*:*:*:*
Vendors & Products Linux
Linux linux Kernel
References
Weaknesses
  • CWE-281

    Improper Preservation of Permissions

Sources

Status
PUBLISHED
Assigner
Linux
Published
Updated
Oct 31, 2026, 12:44 PM
Reserved
Sep 25, 2026, 10:25 AM
Status
Received
Published
Oct 28, 2026, 11:17 AM
Modified
Oct 31, 2026, 1:17 PM
Severity
Moderate
Public date
Oct 28, 2026, 12:00 AM