Re: [libvirt PATCH 01/13] security: add SELinux policy for virt

On Thu, Aug 19, 2021 at 05:23:48PM +0200, Vit Mojzis wrote: > On 10. 08. 21 18:35, Daniel P. Berrangé wrote: >> On Tue, Aug 10, 2021 at 10:39:23AM +0200, Pavel Hrdina wrote: >>> On Fri, Aug 06, 2021 at 06:47:58PM +0100, Daniel P. Berrangé wrote: >>>> From: Nikola Knazekova <nknazeko(a)redhat.com&gt; >>>> >>>> SELinux policy was created for: >>>> >>>> Hypervisor drivers: >>>> - virtqemud (QEMU/KVM) >>>> - virtlxcd (LXC) >>>> - virtvboxd (VirtualBox) >>>> >>>> Secondary drivers: >>>> - virtstoraged (host storage mgmt) >>>> - virtnetworkd (virtual network mgmt) >>>> - virtinterface (network interface mgmt) >>>> - virtnodedevd (physical device mgmt) >>>> - virtsecretd (security credential mgmt) >>>> - virtnwfilterd (ip[6]tables/ebtables mgmt) >>>> - virtproxyd (proxy daemon) >>>> >>>> SELinux policy for virtvxz and virtxend has not been created yet, >>>> because I wasn't able to reproduce AVC messages. These drivers >>>> run in unconfined_domain until the AVC messages are reproduced >>>> internally and policy for these drivers is made. >>>> >>>> Signed-off-by: Nikola Knazekova <nknazeko(a)redhat.com&gt; >>>> --- >>>> src/security/selinux/virt.fc | 111 ++ >>>> src/security/selinux/virt.if | 1984 ++++++++++++++++++++++++++++++++ >>>> src/security/selinux/virt.te | 2078 ++++++++++++++++++++++++++++++++++ >>>> 3 files changed, 4173 insertions(+) >>>> create mode 100644 src/security/selinux/virt.fc >>>> create mode 100644 src/security/selinux/virt.if >>>> create mode 100644 src/security/selinux/virt.te >>>> >>>> diff --git a/src/security/selinux/virt.fc b/src/security/selinux/virt.fc >>>> new file mode 100644 >>>> index 0000000000..554e1094d9 >>>> --- /dev/null >>>> +++ b/src/security/selinux/virt.fc >>>> @@ -0,0 +1,111 @@ >>>> +HOME_DIR/\.libvirt(/.*)? gen_context(system_u:object_r:virt_home_t,s0) >>>> +HOME_DIR/\.libvirt/qemu(/.*)? gen_context(system_u:object_r:svirt_home_t,s0) >>>> +HOME_DIR/\.cache/libvirt(/.*)? gen_context(system_u:object_r:virt_home_t,s0) >>>> +HOME_DIR/\.cache/libvirt/qemu(/.*)? gen_context(system_u:object_r:svirt_home_t,s0) >>>> +HOME_DIR/\.config/libvirt(/.*)? gen_context(system_u:object_r:virt_home_t,s0) >>>> +HOME_DIR/\.config/libvirt/qemu(/.*)? gen_context(system_u:object_r:svirt_home_t,s0) >>>> +HOME_DIR/VirtualMachines(/.*)? gen_context(system_u:object_r:virt_home_t,s0) >>>> +HOME_DIR/VirtualMachines/isos(/.*)? gen_context(system_u:object_r:virt_content_t,s0) >>> These two doesn't look like libvirt selinux bits, more like virt-manager >>> or some other tool. >> Rationale is largely lost in the mists of time to be honest. $HOME/VirtualMachines >> does make sense for desktop virt use case I think, while the below rules make >> sense as a direct translation of libvirt's system paths. >> >> I think its ok to have both really >> >>>> +HOME_DIR/\.local/share/libvirt/images(/.*)? gen_context(system_u:object_r:svirt_home_t,s0) >>>> +HOME_DIR/\.local/share/libvirt/boot(/.*)? gen_context(system_u:object_r:svirt_home_t,s0) >>>> +/var/lib/libvirt(/.*)? gen_context(system_u:object_r:virt_var_lib_t,s0) >>>> +/var/lib/libvirt/boot(/.*)? gen_context(system_u:object_r:virt_content_t,s0) >>>> +/var/lib/libvirt/images(/.*)? gen_context(system_u:object_r:virt_image_t,s0) >>>> +/var/lib/libvirt/isos(/.*)? gen_context(system_u:object_r:virt_content_t,s0) >>>> +/var/lib/libvirt/lockd(/.*)? gen_context(system_u:object_r:virt_var_lockd_t,s0) >>>> +/var/lib/libvirt/qemu(/.*)? gen_context(system_u:object_r:qemu_var_run_t,s0-mls_systemhigh) >>>> + >>>> +/var/log/log(/.*)? gen_context(system_u:object_r:virt_log_t,s0) >>> Based on commit from selinux-policy 63ead48cf8 this seems vdsm related. >>> I don't think that we use this directory in libvirt. >> Yeah, that's dubious. > Good point, we'll move it out of virt policy. >>>> +/var/log/libvirt(/.*)? gen_context(system_u:object_r:virt_log_t,s0) >>>> +/var/run/libvirtd\.pid -- gen_context(system_u:object_r:virt_var_run_t,s0) >>>> +# Avoid calling m4's "interface" by using en empty string >>>> +/var/run/libvirt/interfac(e)(/.*)? gen_context(system_u:object_r:virtinterfaced_var_run_t,s0) >>>> +/var/run/libvirt/nodedev(/.*)? gen_context(system_u:object_r:virtnodedevd_var_run_t,s0) >>>> +/var/run/libvirt/nwfilter(/.*)? gen_context(system_u:object_r:virtnwfilterd_var_run_t,s0) >>>> +/var/run/libvirt/secrets(/.*)? gen_context(system_u:object_r:virtsecretd_var_run_t,s0) >>>> +/var/run/libvirt/storage(/.*)? gen_context(system_u:object_r:virtstoraged_var_run_t,s0) >>>> + >>>> +/var/run/virtlogd\.pid -- gen_context(system_u:object_r:virtlogd_var_run_t,s0) >>>> +/var/run/virtlxcd\.pid -- gen_context(system_u:object_r:virt_lxc_var_run_t,s0) >>>> +/var/run/virtqemud\.pid -- gen_context(system_u:object_r:virtqemud_var_run_t,s0) >>>> +/var/run/virtvboxd\.pid -- gen_context(system_u:object_r:virtvboxd_var_run_t,s0) >>>> +/var/run/virtproxyd\.pid -- gen_context(system_u:object_r:virtproxyd_var_run_t,s0) >>>> +/var/run/virtinterfaced\.pid -- gen_context(system_u:object_r:virtinterfaced_var_run_t,s0) >>>> +/var/run/virtnetworkd\.pid -- gen_context(system_u:object_r:virtnetworkd_var_run_t,s0) >>>> +/var/run/virtnodedevd\.pid -- gen_context(system_u:object_r:virtnodedevd_var_run_t,s0) >>>> +/var/run/virtnwfilterd\.pid -- gen_context(system_u:object_r:virtnwfilterd_var_run_t,s0) >>>> +/var/run/virtnwfilterd-binding\.pid -- gen_context(system_u:object_r:virtnwfilterd_var_run_t,s0) >>>> +/var/run/virtsecretd\.pid -- gen_context(system_u:object_r:virtsecretd_var_run_t,s0) >>>> +/var/run/virtstoraged\.pid -- gen_context(system_u:object_r:virtstoraged_var_run_t,s0) >>> [...] >>> >>> I was not able to figure out on which selinux policy is this one based >>> on as the upstream for rawhide from <https://github.com/fedora-selinux/selinux-policy.git> >>> is a bit different. There are some cosmetics changes but I see two major >>> differences: >>> >>> - the upstream policy doesn't have split-daemon bits compared to >>> this one, I checked it and it looks reasonable but I'm not that >>> familiar with selinux policy >> Now I compare the two, I see there's a bunch of stuff in the current >> fedora virt.te that doesn't exist in this virt.te. > This policy has been mostly rewritten by Nikola Knazekova to work with the > split daemon configuration. > > It's only been tested using libvirt-tck and by running some VM by hand so we > could use your help running other tests you have available. > > There has actually been some progress on the policy since I last updated the > PR. The latest version is available here: > https://github.com/5umm3r15/selinux-policy/blob/libvirt-selinux/policy/mo... > > https://github.com/5umm3r15/selinux-policy/blob/libvirt-selinux/policy/mo... > > https://github.com/5umm3r15/selinux-policy/blob/libvirt-selinux/policy/mo... > > The MLS parts of the policy are still not 100% since we are not sure about > some access that is taking place during testing with libvirt-tck. > > Please see: > https://gitlab.com/libvirt/libvirt-tck/-/merge_requests/8#note_601463944 Doh I missed those questions. >> So if we deploy this, then its likely to break stuff that's in the >> current virt.te that we've omitted. >> >> I should have checked this more closely before re-sendig, as I just >> blindly assumed that the differences to fedora-selinux had been >> eliminated after my original review comments :-( >> >>> - the upstream policy has important `system.token` issue fix that >>> we've seen recently introduced by upstream commit <1f761d0bbd> >> My view for pulling any SElinux policy into libvirt.git is that we need >> to untangle the current fedora selinux virt policy first to remove all >> the non-libvirt pieces. It should then be a direct copy into libvirt.git >> with no modifications. >> >> So I don't think this is mergable as it exists now. > The policy has been split to virt and virt_supplementary (https://github.com/5umm3r15/selinux-policy/blob/libvirt-selinux/policy/mo...), > where virt_supplementary has the bits that are non-libvirt (this part will > stay in selinux-policy repo). IIUC, the repo 5umm3r15/selinux-policy is not the main repo used for Fedora policy.

If I merged this policy in libvirt now, and we deployed it on Fedora we would regress becaue virt_supplementary doesn't exist in any current Fedora / rawhide IIUC.

Is there an ETA for merging the virt/virt_supplementary stuff into the official Fedora policy, and updating rawhide / RHEL-9, so that we can in turn merge & release the libvirt parts ?

Regards, Daniel