Recommended value in CONFIG_LSM option on SELinux system?

Recommended value in CONFIG_LSM option on SELinux system?

[Nicolas Iooss]

Hello,

Recently there was a bug in Arch Linux where SELinux was no longer
enabled after booting [1], because the default kernel configuration
changed recently [2]:

-CONFIG_LSM="lockdown,yama"
+CONFIG_LSM="lockdown,yama,bpf"

By doing so, setting "security=selinux" on the kernel command line
seemed to break the system, because reading /proc/$PID/attr/current
resulted in "Invalid argument" errors. Replacing "security=selinux"
with "lsm=selinux,lockdown,yama,bpf" fixed the issue and everything is
now fine, but now I am wondering: how should CONFIG_LSM (and option
"lsm" on the kernel command line) be set, on a system which is using
SELinux?

Such information is lacking from the documentation [3] [4]. Therefore
I took a look at Fedora [5] and RHEL [6]:

* Fedora uses CONFIG_LSM="yama,loadpin,safesetid,integrity,selinux,smack,tomoyo,apparmor",
which was the default value until Linux 5.4 [7].
* RHEL uses CONFIG_LSM="yama,integrity,selinux".

It seems to be strange to have an "outdated" configuration value in
the configuration file, but this could be fine if the new modules are
not expected to be used without the kernel being booted with a
"lsm=..." option.

But there is something that I did not understand: setting
"lsm=selinux,lockdown,yama,bpf" worked, /sys/kernel/security/lsm
showed "capability,selinux,lockdown,yama,bpf", but this violated what
the documentation stated [3]:
"A list of the active security modules can be found by reading
/sys/kernel/security/lsm. This is a comma separated list, and will
always include the capability module. The list reflects the order in
which checks are made. The capability module will always be first,
followed by any “minor” modules (e.g. Yama) and then the one “major”
module (e.g. SELinux) if there is one configured."

Is "lsm=selinux,lockdown,yama,bpf" really problematic?

TL;DR: It would be very helpful if there were some clear guidelines
which were documented in the kernel documentation about how to
configure CONFIG_LSM on SELinux systems.

Thanks,
Nicolas

[1] https://github.com/archlinuxhardened/selinux/issues/81
[2] https://github.com/archlinux/svntogit-packages/commit/69cb8c2d2884181e799e67b09d67fcf7944d8408
[3] https://www.kernel.org/doc/html/v5.11-rc7/admin-guide/LSM/index.html
[4] https://www.kernel.org/doc/html/v5.11-rc7/admin-guide/LSM/SELinux.html
[5] https://src.fedoraproject.org/rpms/kernel/blob/dd9f5d552f96c5171a0f04170dbca7e74e8d13c7/f/kernel-x86_64-fedora.config#_3232
[6] https://src.fedoraproject.org/rpms/kernel/blob/dd9f5d552f96c5171a0f04170dbca7e74e8d13c7/f/kernel-x86_64-rhel.config#_2834
[7] commit https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=000d388ed3bbed745f366ce71b2bb7c2ee70f449

Re: Recommended value in CONFIG_LSM option on SELinux system?

[Casey Schaufler]

On 2/8/2021 12:35 PM, Nicolas Iooss wrote:
> Hello,
>
> Recently there was a bug in Arch Linux where SELinux was no longer
> enabled after booting [1], because the default kernel configuration
> changed recently [2]:
>
> -CONFIG_LSM="lockdown,yama"
> +CONFIG_LSM="lockdown,yama,bpf"

Neither of these settings will enable SELinux by default.

> By doing so, setting "security=selinux" on the kernel command line
> seemed to break the system, because reading /proc/$PID/attr/current
> resulted in "Invalid argument" errors.

Is this in addition to an "lsm=" specification on the command line?

> Replacing "security=selinux"
> with "lsm=selinux,lockdown,yama,bpf" fixed the issue and everything is
> now fine, but now I am wondering: how should CONFIG_LSM (and option
> "lsm" on the kernel command line) be set, on a system which is using
> SELinux?

CONFIG_SECURITY_SELINUX=y
CONFIG_DEFAULT_SECURITY_SELINUX=y
CONFIG_LSM="lockdown,yama,selinux"

> Such information is lacking from the documentation [3] [4]. Therefore
> I took a look at Fedora [5] and RHEL [6]:
>
> * Fedora uses CONFIG_LSM="yama,loadpin,safesetid,integrity,selinux,smack,tomoyo,apparmor",
> which was the default value until Linux 5.4 [7].
> * RHEL uses CONFIG_LSM="yama,integrity,selinux".
>
> It seems to be strange to have an "outdated" configuration value in
> the configuration file, but this could be fine if the new modules are
> not expected to be used without the kernel being booted with a
> "lsm=..." option.

Keeping the "outdated" configuration values is necessary for
compatibility. We never intended that specifying either of
security= or lsm= on the boot line be required. Because there
is no way to maintain the old behavior of security=selinux
while allowing security=lockdown,yama,selinux we had to introduce
lsm=. 

> But there is something that I did not understand: setting
> "lsm=selinux,lockdown,yama,bpf" worked, /sys/kernel/security/lsm
> showed "capability,selinux,lockdown,yama,bpf", but this violated what
> the documentation stated [3]:
> "A list of the active security modules can be found by reading
> /sys/kernel/security/lsm. This is a comma separated list, and will
> always include the capability module. The list reflects the order in
> which checks are made. The capability module will always be first,
> followed by any “minor” modules (e.g. Yama) and then the one “major”
> module (e.g. SELinux) if there is one configured."
>
> Is "lsm=selinux,lockdown,yama,bpf" really problematic?

The documentation is out of date regarding the "major" module
having to be last. That was true before the lsm= option was introduced.

>
> TL;DR: It would be very helpful if there were some clear guidelines
> which were documented in the kernel documentation about how to
> configure CONFIG_LSM on SELinux systems.

Thanks for the feedback. We are at a mid-point in the development of
module stacking. It's not too late to make things better based on your
experience.

>
> Thanks,
> Nicolas
>
> [1] https://github.com/archlinuxhardened/selinux/issues/81
> [2] https://github.com/archlinux/svntogit-packages/commit/69cb8c2d2884181e799e67b09d67fcf7944d8408
> [3] https://www.kernel.org/doc/html/v5.11-rc7/admin-guide/LSM/index.html
> [4] https://www.kernel.org/doc/html/v5.11-rc7/admin-guide/LSM/SELinux.html
> [5] https://src.fedoraproject.org/rpms/kernel/blob/dd9f5d552f96c5171a0f04170dbca7e74e8d13c7/f/kernel-x86_64-fedora.config#_3232
> [6] https://src.fedoraproject.org/rpms/kernel/blob/dd9f5d552f96c5171a0f04170dbca7e74e8d13c7/f/kernel-x86_64-rhel.config#_2834
> [7] commit https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=000d388ed3bbed745f366ce71b2bb7c2ee70f449
>

Re: Recommended value in CONFIG_LSM option on SELinux system?

[Nicolas Iooss]

On Mon, Feb 8, 2021 at 10:38 PM Casey Schaufler <casey@schaufler-ca.com> wrote:
>
> On 2/8/2021 12:35 PM, Nicolas Iooss wrote:
> > Hello,
> >
> > Recently there was a bug in Arch Linux where SELinux was no longer
> > enabled after booting [1], because the default kernel configuration
> > changed recently [2]:
> >
> > -CONFIG_LSM="lockdown,yama"
> > +CONFIG_LSM="lockdown,yama,bpf"
>
> Neither of these settings will enable SELinux by default.
>
> > By doing so, setting "security=selinux" on the kernel command line
> > seemed to break the system, because reading /proc/$PID/attr/current
> > resulted in "Invalid argument" errors.
>
> Is this in addition to an "lsm=" specification on the command line?

My description of the issue was not very clear, my apologies. Here is
another tentative.

For some years, Arch Linux developers have been compiling their
official kernel (https://archlinux.org/packages/core/x86_64/linux/ ;
https://github.com/archlinux/svntogit-packages/tree/packages/linux/trunk)
with CONFIG_SECURITY_SELINUX=y even though SELinux was disabled by
default. In order to use SELinux, users were required to install some
packages (such as systemd with SELinux support) and to add "selinux=1
security=selinux" to their kernel command line. Last week, the
official Arch Linux kernel was updated to use
CONFIG_LSM="lockdown,yama,bpf". Booting a system with this kernel and
"selinux=1 security=selinux" was broken (D-Bus refused to start, as
well as all network services, because systemd failed to get some
important SELinux contexts through /proc/$PID/attr/... files). But
using "selinux=1 lsm=selinux,lockdown,yama,bpf" fixed this issue
(before, there was no lsm= command line).

> > Replacing "security=selinux"
> > with "lsm=selinux,lockdown,yama,bpf" fixed the issue and everything is
> > now fine, but now I am wondering: how should CONFIG_LSM (and option
> > "lsm" on the kernel command line) be set, on a system which is using
> > SELinux?
>
> CONFIG_SECURITY_SELINUX=y
> CONFIG_DEFAULT_SECURITY_SELINUX=y
> CONFIG_LSM="lockdown,yama,selinux"
>
> > Such information is lacking from the documentation [3] [4]. Therefore
> > I took a look at Fedora [5] and RHEL [6]:
> >
> > * Fedora uses CONFIG_LSM="yama,loadpin,safesetid,integrity,selinux,smack,tomoyo,apparmor",
> > which was the default value until Linux 5.4 [7].
> > * RHEL uses CONFIG_LSM="yama,integrity,selinux".
> >
> > It seems to be strange to have an "outdated" configuration value in
> > the configuration file, but this could be fine if the new modules are
> > not expected to be used without the kernel being booted with a
> > "lsm=..." option.
>
> Keeping the "outdated" configuration values is necessary for
> compatibility. We never intended that specifying either of
> security= or lsm= on the boot line be required. Because there
> is no way to maintain the old behavior of security=selinux
> while allowing security=lockdown,yama,selinux we had to introduce
> lsm=.
>
> > But there is something that I did not understand: setting
> > "lsm=selinux,lockdown,yama,bpf" worked, /sys/kernel/security/lsm
> > showed "capability,selinux,lockdown,yama,bpf", but this violated what
> > the documentation stated [3]:
> > "A list of the active security modules can be found by reading
> > /sys/kernel/security/lsm. This is a comma separated list, and will
> > always include the capability module. The list reflects the order in
> > which checks are made. The capability module will always be first,
> > followed by any “minor” modules (e.g. Yama) and then the one “major”
> > module (e.g. SELinux) if there is one configured."
> >
> > Is "lsm=selinux,lockdown,yama,bpf" really problematic?
>
> The documentation is out of date regarding the "major" module
> having to be last. That was true before the lsm= option was introduced.
>
> >
> > TL;DR: It would be very helpful if there were some clear guidelines
> > which were documented in the kernel documentation about how to
> > configure CONFIG_LSM on SELinux systems.
>
> Thanks for the feedback. We are at a mid-point in the development of
> module stacking. It's not too late to make things better based on your
> experience.

Thanks for your quick reply! Anyway it seems that my issue was caused
by an incompatibility between selinux and bpf LSM. On my test system
(running Linux 5.10.11):

* Using "lsm=selinux,lockdown,yama,bpf" works (SELinux is functional,
the system boots fine)
* Using "lsm=lockdown,yama,selinux,bpf" works too
* Using "lsm=lockdown,yama,bpf,selinux" does not work
(/proc/$PID/attr/current raises "Invalid argument")

So it is important that "selinux" comes before "bpf" in CONFIG_LSM
(and "lsm" parameter). This fact seems to be known, as "bpf" was added
last in the default values of CONFIG_LSM options
(https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/security/Kconfig?h=v5.11-rc7#n280).
Is this documented somewhere? It is neither in
https://www.kernel.org/doc/html/v5.11-rc7/bpf/bpf_lsm.html nor
https://www.kernel.org/doc/html/v5.11-rc7/admin-guide/LSM/index.html.

Nicolas

> >
> > Thanks,
> > Nicolas
> >
> > [1] https://github.com/archlinuxhardened/selinux/issues/81
> > [2] https://github.com/archlinux/svntogit-packages/commit/69cb8c2d2884181e799e67b09d67fcf7944d8408
> > [3] https://www.kernel.org/doc/html/v5.11-rc7/admin-guide/LSM/index.html
> > [4] https://www.kernel.org/doc/html/v5.11-rc7/admin-guide/LSM/SELinux.html
> > [5] https://src.fedoraproject.org/rpms/kernel/blob/dd9f5d552f96c5171a0f04170dbca7e74e8d13c7/f/kernel-x86_64-fedora.config#_3232
> > [6] https://src.fedoraproject.org/rpms/kernel/blob/dd9f5d552f96c5171a0f04170dbca7e74e8d13c7/f/kernel-x86_64-rhel.config#_2834
> > [7] commit https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=000d388ed3bbed745f366ce71b2bb7c2ee70f449
> >
>

Re: Recommended value in CONFIG_LSM option on SELinux system?

[Casey Schaufler]

On 2/8/2021 10:52 PM, Nicolas Iooss wrote:
> On Mon, Feb 8, 2021 at 10:38 PM Casey Schaufler <casey@schaufler-ca.com> wrote:
>> On 2/8/2021 12:35 PM, Nicolas Iooss wrote:
>>> Hello,
>>>
>>> Recently there was a bug in Arch Linux where SELinux was no longer
>>> enabled after booting [1], because the default kernel configuration
>>> changed recently [2]:
>>>
>>> -CONFIG_LSM="lockdown,yama"
>>> +CONFIG_LSM="lockdown,yama,bpf"
>> Neither of these settings will enable SELinux by default.
>>
>>> By doing so, setting "security=selinux" on the kernel command line
>>> seemed to break the system, because reading /proc/$PID/attr/current
>>> resulted in "Invalid argument" errors.
>> Is this in addition to an "lsm=" specification on the command line?
> My description of the issue was not very clear, my apologies. Here is
> another tentative.
>
> For some years, Arch Linux developers have been compiling their
> official kernel (https://archlinux.org/packages/core/x86_64/linux/ ;
> https://github.com/archlinux/svntogit-packages/tree/packages/linux/trunk)
> with CONFIG_SECURITY_SELINUX=y even though SELinux was disabled by
> default. 

That would imply you do not have CONFIG_DEFAULT_SECURITY_SELINUX=y
That's completely reasonable.

> In order to use SELinux, users were required to install some
> packages (such as systemd with SELinux support) and to add "selinux=1
> security=selinux" to their kernel command line. 

Sensible.

> Last week, the
> official Arch Linux kernel was updated to use
> CONFIG_LSM="lockdown,yama,bpf".

This is a list of the modules to be active at boot if
they are available.

>  Booting a system with this kernel and
> "selinux=1 security=selinux" was broken (D-Bus refused to start, as
> well as all network services, because systemd failed to get some
> important SELinux contexts through /proc/$PID/attr/... files). 

Because selinux isn't in the CONFIG_LSM list it is not
going to available from the command line.

> But
> using "selinux=1 lsm=selinux,lockdown,yama,bpf" fixed this issue
> (before, there was no lsm= command line).

The security= and lsm= command line options should not be used
together. The lsm= option is the complete security module list.
The security= option only changes the "major" module for backward
compatibility.

>
>>> Replacing "security=selinux"
>>> with "lsm=selinux,lockdown,yama,bpf" fixed the issue and everything is
>>> now fine, but now I am wondering: how should CONFIG_LSM (and option
>>> "lsm" on the kernel command line) be set, on a system which is using
>>> SELinux?
>> CONFIG_SECURITY_SELINUX=y
>> CONFIG_DEFAULT_SECURITY_SELINUX=y
>> CONFIG_LSM="lockdown,yama,selinux"
>>
>>> Such information is lacking from the documentation [3] [4]. Therefore
>>> I took a look at Fedora [5] and RHEL [6]:
>>>
>>> * Fedora uses CONFIG_LSM="yama,loadpin,safesetid,integrity,selinux,smack,tomoyo,apparmor",
>>> which was the default value until Linux 5.4 [7].
>>> * RHEL uses CONFIG_LSM="yama,integrity,selinux".
>>>
>>> It seems to be strange to have an "outdated" configuration value in
>>> the configuration file, but this could be fine if the new modules are
>>> not expected to be used without the kernel being booted with a
>>> "lsm=..." option.
>> Keeping the "outdated" configuration values is necessary for
>> compatibility. We never intended that specifying either of
>> security= or lsm= on the boot line be required. Because there
>> is no way to maintain the old behavior of security=selinux
>> while allowing security=lockdown,yama,selinux we had to introduce
>> lsm=.
>>
>>> But there is something that I did not understand: setting
>>> "lsm=selinux,lockdown,yama,bpf" worked, /sys/kernel/security/lsm
>>> showed "capability,selinux,lockdown,yama,bpf", but this violated what
>>> the documentation stated [3]:
>>> "A list of the active security modules can be found by reading
>>> /sys/kernel/security/lsm. This is a comma separated list, and will
>>> always include the capability module. The list reflects the order in
>>> which checks are made. The capability module will always be first,
>>> followed by any “minor” modules (e.g. Yama) and then the one “major”
>>> module (e.g. SELinux) if there is one configured."
>>>
>>> Is "lsm=selinux,lockdown,yama,bpf" really problematic?
>> The documentation is out of date regarding the "major" module
>> having to be last. That was true before the lsm= option was introduced.
>>
>>> TL;DR: It would be very helpful if there were some clear guidelines
>>> which were documented in the kernel documentation about how to
>>> configure CONFIG_LSM on SELinux systems.
>> Thanks for the feedback. We are at a mid-point in the development of
>> module stacking. It's not too late to make things better based on your
>> experience.
> Thanks for your quick reply! Anyway it seems that my issue was caused
> by an incompatibility between selinux and bpf LSM. On my test system
> (running Linux 5.10.11):
>
> * Using "lsm=selinux,lockdown,yama,bpf" works (SELinux is functional,
> the system boots fine)
> * Using "lsm=lockdown,yama,selinux,bpf" works too
> * Using "lsm=lockdown,yama,bpf,selinux" does not work
> (/proc/$PID/attr/current raises "Invalid argument")

You can't (currently) use SELinux and BPF at the same time.
This is because the infrastructure does not support multiple
secid<->secctx translation hooks. You get the first one in the
list. BPF provides all hooks, so the SELinux hooks aren't
reached and the secid to secctx translation fails in the
"bpf,selinux" case. The "selinux,bpf" case appears to work,
but the BPF hook for security_secid_to_secctx() will never get
called. This may be acceptable for your use case. Patches to
allow for multiple callers of these hooks are under review.

> So it is important that "selinux" comes before "bpf" in CONFIG_LSM
> (and "lsm" parameter). This fact seems to be known, as "bpf" was added
> last in the default values of CONFIG_LSM options
> (https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/security/Kconfig?h=v5.11-rc7#n280).
> Is this documented somewhere? It is neither in
> https://www.kernel.org/doc/html/v5.11-rc7/bpf/bpf_lsm.html nor
> https://www.kernel.org/doc/html/v5.11-rc7/admin-guide/LSM/index.html.



>
> Nicolas
>
>>> Thanks,
>>> Nicolas
>>>
>>> [1] https://github.com/archlinuxhardened/selinux/issues/81
>>> [2] https://github.com/archlinux/svntogit-packages/commit/69cb8c2d2884181e799e67b09d67fcf7944d8408
>>> [3] https://www.kernel.org/doc/html/v5.11-rc7/admin-guide/LSM/index.html
>>> [4] https://www.kernel.org/doc/html/v5.11-rc7/admin-guide/LSM/SELinux.html
>>> [5] https://src.fedoraproject.org/rpms/kernel/blob/dd9f5d552f96c5171a0f04170dbca7e74e8d13c7/f/kernel-x86_64-fedora.config#_3232
>>> [6] https://src.fedoraproject.org/rpms/kernel/blob/dd9f5d552f96c5171a0f04170dbca7e74e8d13c7/f/kernel-x86_64-rhel.config#_2834
>>> [7] commit https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=000d388ed3bbed745f366ce71b2bb7c2ee70f449
>>>