PT-2026-128163
CVE-2026-99170
·
تاريخ النشر
2026-10-28
·
تاريخ التحديث
2026-10-28
الوصف من
الدرجة
N/A
غير متوفرلا تتوفر حاليًا تقييمات أو مقاييس للخطورة. سنحدّث المعلومات المقابلة في الصفحة فور توفرها.
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//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---
وجدت خطأً في الوصف؟ لديك ما تضيفه؟ لا تتردد في مراسلتنا 👾