2020-12-09 12:42:57

by Russell Coker

[permalink] [raw]
Subject: udevadm etc

# ls -l /usr/lib/systemd/systemd-udevd
lrwxrwxrwx. 1 root root 12 Dec 3 22:53 /usr/lib/systemd/systemd-udevd -> /
bin/udevadm

In systemd verision 247.1 (which is in Debian/Unstable now) systemd-udevd is a
symlink to udevadm. In systemd version 241 (in Debian/Buster) it's not a
symlink. The systemd changelog doesn't mention this change so I don't know
exactly when it happened.

type_transition init_t udevadm_exec_t:process udevadm_t;
type_transition initrc_t udevadm_exec_t:process udevadm_t;
type_transition sysadm_t udevadm_exec_t:process udevadm_t;
type_transition consolekit_t udev_exec_t:process udev_t;
type_transition devicekit_disk_t udev_exec_t:process udev_t;
type_transition hald_t udev_exec_t:process udev_t;
type_transition hotplug_t udev_exec_t:process udev_t;
type_transition init_t udev_exec_t:process udev_t;
type_transition initrc_t udev_exec_t:process udev_t;
type_transition kernel_t udev_exec_t:process udev_t;
type_transition virtd_t udev_exec_t:process udev_t;

Above is a list of the relevant type_transition rules from refpolicy taken
from git 3 days ago (there don't appear to be any udev changes since then).

I think that the only thing to do is to have init_t and initrc_t run udevadm
in the udev_t domain.

Is it worth having a udevadm_t domain just for running it from sysadm_t or
should we have that run as udev_t too?

--
My Main Blog http://etbe.coker.com.au/
My Documents Blog http://doc.coker.com.au/




2020-12-09 22:03:19

by Chris PeBenito

[permalink] [raw]
Subject: Re: udevadm etc

On 12/9/20 7:38 AM, Russell Coker wrote:
> In systemd verision 247.1 (which is in Debian/Unstable now) systemd-udevd is a
> symlink to udevadm. In systemd version 241 (in Debian/Buster) it's not a
> symlink. The systemd changelog doesn't mention this change so I don't know
> exactly when it happened.
>
> type_transition init_t udevadm_exec_t:process udevadm_t;
> type_transition initrc_t udevadm_exec_t:process udevadm_t;
> type_transition sysadm_t udevadm_exec_t:process udevadm_t;
> type_transition consolekit_t udev_exec_t:process udev_t;
> type_transition devicekit_disk_t udev_exec_t:process udev_t;
> type_transition hald_t udev_exec_t:process udev_t;
> type_transition hotplug_t udev_exec_t:process udev_t;
> type_transition init_t udev_exec_t:process udev_t;
> type_transition initrc_t udev_exec_t:process udev_t;
> type_transition kernel_t udev_exec_t:process udev_t;
> type_transition virtd_t udev_exec_t:process udev_t;
>
> Above is a list of the relevant type_transition rules from refpolicy taken
> from git 3 days ago (there don't appear to be any udev changes since then).
>
> I think that the only thing to do is to have init_t and initrc_t run udevadm
> in the udev_t domain.
>
> Is it worth having a udevadm_t domain just for running it from sysadm_t or
> should we have that run as udev_t too?

The udevadm_t domain is much smaller than udev_t, and critically doesn't have
all the device access udev_t has. Systemd merging the binaries into one doesn't
necessarily mean the policy has to merge too.

--
Chris PeBenito

2020-12-10 00:08:01

by Russell Coker

[permalink] [raw]
Subject: Re: udevadm etc

On Thursday, 10 December 2020 9:00:38 AM AEDT Chris PeBenito wrote:
> On 12/9/20 7:38 AM, Russell Coker wrote:
> > In systemd verision 247.1 (which is in Debian/Unstable now) systemd-udevd
> > is a symlink to udevadm. In systemd version 241 (in Debian/Buster) it's
> > not a symlink. The systemd changelog doesn't mention this change so I
> > don't know exactly when it happened.
> >
> > type_transition init_t udevadm_exec_t:process udevadm_t;
> > type_transition initrc_t udevadm_exec_t:process udevadm_t;
> > type_transition sysadm_t udevadm_exec_t:process udevadm_t;
> > type_transition consolekit_t udev_exec_t:process udev_t;
> > type_transition devicekit_disk_t udev_exec_t:process udev_t;
> > type_transition hald_t udev_exec_t:process udev_t;
> > type_transition hotplug_t udev_exec_t:process udev_t;
> > type_transition init_t udev_exec_t:process udev_t;
> > type_transition initrc_t udev_exec_t:process udev_t;
> > type_transition kernel_t udev_exec_t:process udev_t;
> > type_transition virtd_t udev_exec_t:process udev_t;
> >
> > Above is a list of the relevant type_transition rules from refpolicy taken
> > from git 3 days ago (there don't appear to be any udev changes since
> > then).
> >
> > I think that the only thing to do is to have init_t and initrc_t run
> > udevadm in the udev_t domain.
> >
> > Is it worth having a udevadm_t domain just for running it from sysadm_t or
> > should we have that run as udev_t too?
>
> The udevadm_t domain is much smaller than udev_t, and critically doesn't
> have all the device access udev_t has. Systemd merging the binaries into
> one doesn't necessarily mean the policy has to merge too.

Currently only 3 domains can transition to udevadm_t. 2 of those 3 also
transition to udev_t so they can't use udevadm_t without code changes. The
only remaining domain is sysadm_t.

The question is not "do we have to merge the policy because systemd merges
binaries" but "should we have a separate domain only used for sysadm_t"?

--
My Main Blog http://etbe.coker.com.au/
My Documents Blog http://doc.coker.com.au/



2020-12-10 16:32:32

by Chris PeBenito

[permalink] [raw]
Subject: Re: udevadm etc

On 12/9/20 7:02 PM, Russell Coker wrote:
> On Thursday, 10 December 2020 9:00:38 AM AEDT Chris PeBenito wrote:
>> On 12/9/20 7:38 AM, Russell Coker wrote:
>>> In systemd verision 247.1 (which is in Debian/Unstable now) systemd-udevd
>>> is a symlink to udevadm. In systemd version 241 (in Debian/Buster) it's
>>> not a symlink. The systemd changelog doesn't mention this change so I
>>> don't know exactly when it happened.
>>>
>>> type_transition init_t udevadm_exec_t:process udevadm_t;
>>> type_transition initrc_t udevadm_exec_t:process udevadm_t;
>>> type_transition sysadm_t udevadm_exec_t:process udevadm_t;
>>> type_transition consolekit_t udev_exec_t:process udev_t;
>>> type_transition devicekit_disk_t udev_exec_t:process udev_t;
>>> type_transition hald_t udev_exec_t:process udev_t;
>>> type_transition hotplug_t udev_exec_t:process udev_t;
>>> type_transition init_t udev_exec_t:process udev_t;
>>> type_transition initrc_t udev_exec_t:process udev_t;
>>> type_transition kernel_t udev_exec_t:process udev_t;
>>> type_transition virtd_t udev_exec_t:process udev_t;
>>>
>>> Above is a list of the relevant type_transition rules from refpolicy taken
>>> from git 3 days ago (there don't appear to be any udev changes since
>>> then).
>>>
>>> I think that the only thing to do is to have init_t and initrc_t run
>>> udevadm in the udev_t domain.
>>>
>>> Is it worth having a udevadm_t domain just for running it from sysadm_t or
>>> should we have that run as udev_t too?
>>
>> The udevadm_t domain is much smaller than udev_t, and critically doesn't
>> have all the device access udev_t has. Systemd merging the binaries into
>> one doesn't necessarily mean the policy has to merge too.
>
> Currently only 3 domains can transition to udevadm_t. 2 of those 3 also
> transition to udev_t so they can't use udevadm_t without code changes. The
> only remaining domain is sysadm_t.
>
> The question is not "do we have to merge the policy because systemd merges
> binaries" but "should we have a separate domain only used for sysadm_t"?

Ah, I misunderstood. The alternate question is, should any of these domains
that transition to udev_t actually transition to udevadm_t instead? For a long
time udevadm was labeled udev_exec_t.

I did some cursory searching in devicekit and I see udisks execs udevadm. I
couldn't find any instance of it executing udev, so there is a possibility of
udevadm_t being more useful.

If it ends up with sysadm_t being the only one that only execs udevadm, then I
agree that udevadm_t doesn't provide the hoped benefits.

--
Chris PeBenito