[PATCH] selinux-policy: update to version v2.0

[Dominick Grift]

Dominick Grift <dominick.grift at defensec.nl> writes:

> Hi, Thank you for feedback. Comments inline below:
>
> Stefan Hellermann <stefan at the2masters.de> writes:
>

<snip>

>> audit(1736704702.290:4): avc:  denied  { associate } for  pid=1010
>> comm="mv" name="sysupgrade.tgz" scontext=sys.id:sys.role:dos.fs
>> tcontext=sys.id:sys.role:xattr.fs tclass=filesystem permissive=1
>
> This is caused by mv'ing the file from a fat filesystem (fat does not
> support extended attributes) to an extended attribute file system. When
> you mv a file you also mv its associated context with it.
>
> This should not be allowed. Instead you should use cp. mv does not make
> much sense anyway cross filesystem.
>

This bothered me so I would like to explain why I object to this.

mv and cp are more complicated than some think. I see this all the time
where people for example use `cp -a` without realizing the consequences.

But regardless of this, coreutils has extensive support for SELinux and
`mv -Z` would have addressed the above challenge. The issue is that
busybox' `mv` does not support -Z and so eventually I will have to draw
the line somewhere anyway. This seems like a good place to start.

>> Sun Jan 12 17:58:25 2025 user.warn kernel: urandom-seed: Seed file not
>> found (/etc/urandom.seed)
>> Sun Jan 12 17:58:25 2025 user.info kernel: procd: - early -
>> Sun Jan 12 17:58:25 2025 kern.notice kernel: audit: type=1400
>> audit(1736704702.590:5): avc:  denied  { write } for  pid=1166
>> comm="mkdir" name="/" dev="tmpfs" ino=1
>> scontext=sys.id:sys.role:hotplug.call.subj
>> tcontext=sys.id:sys.role:tmp.fs tclass=dir permissive=1
>> Sun Jan 12 17:58:25 2025 kern.notice kernel: audit: type=1400
>> audit(1736704702.590:6): avc:  denied  { add_name } for  pid=1166
>> comm="mkdir" name="virtio-ports"
>> scontext=sys.id:sys.role:hotplug.call.subj
>> tcontext=sys.id:sys.role:tmp.fs tclass=dir permissive=1
>> Sun Jan 12 17:58:25 2025 kern.notice kernel: audit: type=1400
>> audit(1736704702.590:7): avc:  denied  { create } for  pid=1166
>> comm="mkdir" name="virtio-ports"
>> scontext=sys.id:sys.role:hotplug.call.subj
>> tcontext=sys.id:sys.role:tmp.fs tclass=dir permissive=1
>> Sun Jan 12 17:58:25 2025 kern.notice kernel: audit: type=1400
>> audit(1736704702.590:8): avc:  denied  { create } for  pid=1167
>> comm="ln" name="org.qemu.guest_agent.0"
>> scontext=sys.id:sys.role:hotplug.call.subj
>> tcontext=sys.id:sys.role:tmp.fs tclass=lnk_file permissive=1
>> Sun Jan 12 17:58:25 2025 user.info kernel: procd: - ubus -
>> Sun Jan 12 17:58:25 2025 user.info kernel: procd: - init -
>
> This seems like an 'exotic hotplug script'. I have an accomodation for
> this. see if this comment helps:
> https://git.defensec.nl/?p=selinux-policy.git;a=blob;f=src/agent/sysagent/hotplugsysagent.cil;h=3987b8540ae537d174a74cceb2c89ce26ef3c813;hb=HEAD#l115

We'll have to see how this will work out practically. I am open to
suggestions for alternative approaches but this seems like a fair
approach.

There are also challenges here. For example in the above event, the
script is trying to create a dir and symlink in /dev. In OpenWrt there
is no (easy) way to make a distinction between devtmpfs and and a common
tmpfs. If I we're to allow this then that would later potentially
present challenges when another script wants to create a dir or symlink in /tmp.

Again, eventually I would have to draw the line somewhere as to what
should be allowed by default and what is to be considered exotic. This
looks like a good place.

Just trying to explain some of the rationale because I am open to better
alternatives. I just don't see any.

-- 
Key fingerprint = FCD2 3660 5D6B 9D27 7FC6  E0FF DA7E 521F 10F6 4098
Dominick Grift
Mastodon: @kcinimod at defensec.nl