systemd ignores RootDirectory option in .service units
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
lxd (Ubuntu) |
Invalid
|
Undecided
|
Unassigned | ||
Bionic |
Fix Released
|
Undecided
|
Unassigned | ||
systemd (Ubuntu) |
Confirmed
|
Undecided
|
Unassigned | ||
Bionic |
Confirmed
|
Undecided
|
Unassigned | ||
Focal |
Confirmed
|
Undecided
|
Unassigned | ||
Impish |
Confirmed
|
Undecided
|
Unassigned |
Bug Description
[Impact]
Ubuntu carries a patch on top of systemd [a] to silence
namespace set up failures. This is meant as a workaround
for a bug in the LXD version shipped in Ubuntu 18.04.
Masking namespace set up failures creates a false sense of
security for the user/admin.
As mentioned in comment #1, systemd upstream explains that silencing
this kind of error is dangerous and should be avoided.
Backporting the LXD fix [b] to Ubuntu 18.04 would allow namespaces
to work inside containers. This is the goal of this SRU.
Ultimately, once LXD in Ubuntu 18.04 includes the fix [b], it would
be possible to drop the Ubuntu-specific patch for systemd [a]. This
is however *not an immediate concern for this SRU*.
[Test Plan]
1) Create a 18.04 VM:
$ lxc launch images:ubuntu/18.04 lp1959047 --vm
$ sleep 30 # give it time to boot
1.5) Enable bionic-proposed:
$ echo "deb http://
2) Install and initialize LXD in it:
$ lxc exec lp1959047 -- apt-get update
$ lxc exec lp1959047 -- apt-get install -y lxd
$ lxc exec lp1959047 -- lxd init --auto
3) Create a Jammy container and enable systemd debugging:
$ lxc exec lp1959047 -- lxc init images:ubuntu/22.04 c1
$ lxc exec lp1959047 -- lxc config set c1 raw.lxc 'lxc.init.cmd = /sbin/init systemd.
$ lxc exec lp1959047 -- lxc start c1
4) Check if namespace set up failures are logged:
$ lxc exec lp1959047 -- lxc exec c1 -- journalctl -b0 --grep 'Failed to set up namespace'
Mar 24 23:29:19 c1 systemd[99]: systemd-
Mar 24 23:29:19 c1 systemd[132]: systemd-
Mar 24 23:29:19 c1 systemd[131]: systemd-
Mar 24 23:29:20 c1 systemd[136]: systemd-
Mar 24 23:29:20 c1 systemd[128]: e2scrub_
Mar 24 23:29:23 c1 systemd[243]: systemd-
If LXD in Ubuntu 18.04 has the patch, the "Failed to set up namespace" messages wouldn't be there.
[Where problems could occur]
The LXD fix changes the Apparmor profile used for containers. This essentially
loosen the mount restrictions applied to containers.
Weakening the Apparmor profile could make it easier for a process in the container
to do damage that would have otherwise been blocked. On the other hand, this
allows making use of namespaces/
Upstream LXD has the fix since 2019 which make it less likely to run into
problems with the backport.
The backported fix was also tested manually to ensure LXD still behaved normally
and that it avoided the namespace set up failures in Jammy containers.
[a]: https:/
[b]: https:/
[Initial bug description]
The version of systemd (249.5-2ubuntu4) currently packaged for the Ubuntu development version (22.04 Jammy Jellyfish) totally ignores the RootDirectory= option in systemd service files. With RootDirectory, systemd should start the service after calling chroot() on the supplied directory.
To test/reproduce, create a test service file with the following contents:
# /etc/systemd/
[Unit]
Description=LSB Release Information
[Service]
Type=simple
RootDirectory=
ExecStartPre=
ExecStart=
You should have a chroot environment in the specified RootDirectory, even though you can still deduce if systemd attempted to chroot or not from the resulting error message.
In my example, I installed an end-of-life Ubuntu 14.04 Trusty Tahr in the chroot environment. On systems NOT affected by the problem, I get the following result when I start this test service. This is what I'd expect.
Jan 25 20:40:40 dolly systemd[1]: Starting LSB Release Information...
Jan 25 20:40:40 dolly pwd[361]: /
Jan 25 20:40:40 dolly systemd[1]: Started LSB Release Information.
Jan 25 20:40:40 dolly lsb_release[362]: No LSB modules are available.
Jan 25 20:40:40 dolly lsb_release[362]: Distributor ID: Ubuntu
Jan 25 20:40:40 dolly lsb_release[362]: Description: Ubuntu 14.04 LTS
Jan 25 20:40:40 dolly lsb_release[362]: Release: 14.04
Jan 25 20:40:40 dolly lsb_release[362]: Codename: trusty
Jan 25 20:40:40 dolly systemd[1]: lsb-release.
On the problematic system, however, I get the following result.
Jan 25 21:21:08 savelog systemd[1]: Starting LSB Release Information...
Jan 25 21:21:08 savelog systemd[1]: Started LSB Release Information.
Jan 25 21:21:08 savelog pwd[81114]: /
Jan 25 21:21:08 savelog lsb_release[81115]: No LSB modules are available.
Jan 25 21:21:08 savelog lsb_release[81115]: Distributor ID: Ubuntu
Jan 25 21:21:08 savelog lsb_release[81115]: Description: Ubuntu Jammy Jellyfish (development branch)
Jan 25 21:21:08 savelog lsb_release[81115]: Release: 22.04
Jan 25 21:21:08 savelog lsb_release[81115]: Codename: jammy
Jan 25 21:21:08 savelog systemd[1]: lsb-release.
It totally run the service on the host's root filesystem, it didn't care even the slightest that a RootDirectory is specified.
Tested on the following releases / systemd versions:
Ubuntu 18.04.6 Bionic Beaver – ISSUE NOT PRESENT
systemd 237
+PAM +AUDIT +SELINUX +IMA +APPARMOR +SMACK +SYSVINIT +UTMP +LIBCRYPTSETUP +GCRYPT +GNUTLS +ACL +XZ +LZ4 +SECCOMP +BLKID +ELFUTILS +KMOD -IDN2 +IDN -PCRE2 default-
Ubuntu 20.04.3 Focal Fossa – ISSUE NOT PRESENT
systemd 245 (245.4-4ubuntu3.15)
+PAM +AUDIT +SELINUX +IMA +APPARMOR +SMACK +SYSVINIT +UTMP +LIBCRYPTSETUP +GCRYPT +GNUTLS +ACL +XZ +LZ4 +SECCOMP +BLKID +ELFUTILS +KMOD +IDN2 -IDN +PCRE2 default-
Ubuntu 21.10 Impish Indri – ISSUE NOT PRESENT
systemd 248 (248.3-1ubuntu8.2)
+PAM +AUDIT +SELINUX +APPARMOR +IMA +SMACK +SECCOMP +GCRYPT +GNUTLS -OPENSSL +ACL +BLKID +CURL +ELFUTILS -FIDO2 +IDN2 -IDN +IPTC +KMOD +LIBCRYPTSETUP -LIBFDISK +PCRE2 -PWQUALITY -P11KIT -QRENCODE +BZIP2 +LZ4 +XZ +ZLIB +ZSTD -XKBCOMMON +UTMP +SYSVINIT default-
Ubuntu 22.04 Jammy Jellyfish (development branch) – ISSUE PRESENT
systemd 249 (249.5-2ubuntu4)
+PAM +AUDIT +SELINUX +APPARMOR +IMA +SMACK +SECCOMP +GCRYPT +GNUTLS -OPENSSL +ACL +BLKID +CURL +ELFUTILS -FIDO2 +IDN2 -IDN +IPTC +KMOD +LIBCRYPTSETUP -LIBFDISK +PCRE2 -PWQUALITY -P11KIT -QRENCODE +BZIP2 +LZ4 +XZ +ZLIB +ZSTD -XKBCOMMON +UTMP +SYSVINIT default-
Note that the problem is produced under an LXC container; since systemd detects virtualization, it might change how it behaves.
It's either a bug or an intentional change I don't understand yet (i.e. the RootDirectory option has deprecated and is about to be replaced with something else, or there are additional conditions to be met before RootDirectory is considered), but I think in the latter case I should at least get a warning that there is a change in configuration. I imagine suddenly everyone's existing service units utilizing RootDirectory silently stop working without any information regarding why.
description: | updated |
This is caused by https:/ /git.launchpad. net/ubuntu/ +source/ systemd/ tree/debian/ patches/ debian/ UBUNTU- Revert- namespace- be-more- careful- when-handling- namespacin. patch?h= ubuntu/ jammy
From upstream's point of view, ignoring sandboxing options requested by unit owners is quite dangerous. It can result in programs running completely unconstrained.
See: https:/ /github. com/systemd/ systemd/ issues/ 22760