On Tue, Mar 20, 2018 at 9:50 PM, Jamie Strandboge <jamie(a)canonical.com>
wrote:
On Tue, 2018-03-20 at 13:08 +0100, Christian Ehrhardt wrote:
> Recent changes have made implementing this mandatory to hot add any
> memory.
> Implementing this in apparmor fixes this as well as allows hot-add of
> nvdimm
> tpye memory with an nvdimmPath set generating a AppArmor rule for
> that
> path.
>
> Example hot adding:
> <memory model='nvdimm'>
> <source>
> <path>/tmp/nvdimm-test</path>
> </source>
> <target>
> <size unit='KiB'>524288</size>
> <node>0</node>
> </target>
> </memory>
> Creates now:
> "/tmp/nvdimm-test" rwk,
>
> Signed-off-by: Christian Ehrhardt <christian.ehrhardt(a)canonical.com>
> ---
> src/security/security_apparmor.c | 50
> ++++++++++++++++++++++++++++++++++++++++
> 1 file changed, 50 insertions(+)
>
> diff --git a/src/security/security_apparmor.c
> b/src/security/security_apparmor.c
> index a989992..4ae1e3d 100644
> --- a/src/security/security_apparmor.c
> +++ b/src/security/security_apparmor.c
> @@ -718,6 +718,53 @@
> AppArmorRestoreSecurityDiskLabel(virSecurityManagerPtr mgr,
>
> /* Called when hotplugging */
> static int
> +AppArmorSetMemoryLabel(virSecurityManagerPtr mgr,
> + virDomainDefPtr def,
> + virDomainMemoryDefPtr mem)
> +{
> + virSecurityLabelDefPtr secdef;
> +
> + switch ((virDomainMemoryModel) mem->model) {
> + case VIR_DOMAIN_MEMORY_MODEL_NVDIMM:
> + secdef = virDomainDefGetSecurityLabelDef(def,
> SECURITY_APPARMOR_NAME);
> + if (!secdef || !secdef->relabel)
> + return 0;
> +
> + if (!virFileExists(mem->nvdimmPath)) {
> + virReportError(VIR_ERR_INTERNAL_ERROR,
> + _("%s: \'%s\' does not exist"),
> + __func__, mem->nvdimmPath);
> + return -1;
> + }
> +
> + return reload_profile(mgr, def, mem->nvdimmPath, true);
> + break;
> + case VIR_DOMAIN_MEMORY_MODEL_NONE:
> + case VIR_DOMAIN_MEMORY_MODEL_DIMM:
> + case VIR_DOMAIN_MEMORY_MODEL_LAST:
> + break;
> + }
> +
> + return 0;
> +}
> +
> +
> +static int
> +AppArmorRestoreMemoryLabel(virSecurityManagerPtr mgr,
> + virDomainDefPtr def,
> + virDomainMemoryDefPtr mem
> ATTRIBUTE_UNUSED)
> +{
> + virSecurityLabelDefPtr secdef =
> + virDomainDefGetSecurityLabelDef(def,
> SECURITY_APPARMOR_NAME);
> +
> + if (!secdef || !secdef->relabel)
> + return 0;
> +
This secdef check is done already in reload_profile.
I was following other examples, but you are right - I'll drop it in a V2
> + return reload_profile(mgr, def, NULL, false);
Calling this with NULL means that virt-aa-helper will be called and
remove the mem file, but will it also remove anything else that was
hotplugged and not present in 'xml = virDomainDefFormat(def, ...)'?
I've not looked at this in a while, so maybe it's ok (it seems like
this is what the various other AppArmorRestore* functions do after
all).
Yeah I was following other examples that were present in the file already.
It is a complex case to follow "what will happen"
A test case for hotplugging all these different devices and
hotunplugging each would be good to show if hotunplugging one
inadvertently removes rules for other devices that are still
hotplugged.
I agree, first I ensured I actually test the requested code.
Disks should be going through AppArmorRestoreSecurityImageLabel
I ensured with GDB that they do on detach.
reload_profile (mgr=0x7f96741060e0, def=0x7f967418b480, fn=0x0,
append=false) at ../../../src/security/security_apparmor.c:287
That way we can be sure that attach/detach disks is a valid test for the
reload_profile call with NULL.
Test with two disks
$ qemu-img create /var/lib/libvirt/images/hp-test.qcow 10M
$ qemu-img create /var/lib/libvirt/images/hp-test2.qcow 10M
root@b:~# cat hotplug-disk
cat: hotplug-disk: No such file or directory
root@b:~# cat hotplug-disk.xml
<!-- qemu-img create /var/lib/libvirt/images/hp-test.qcow 10M -->
<disk type='file' device='disk'>
<driver name='qemu' type='raw'/>
<source file='/var/lib/libvirt/images/hp-test.qcow'/>
<target dev='vdy' bus='virtio'/>
</disk>
root@b:~# cat hotplug-disk2.xml
<!-- qemu-img create /var/lib/libvirt/images/hp-test.qcow 10M -->
<disk type='file' device='disk'>
<driver name='qemu' type='raw'/>
<source file='/var/lib/libvirt/images/hp-test2.qcow'/>
<target dev='vdz' bus='virtio'/>
</disk>
root@b:~# cat /etc/apparmor.d/libvirt/libvirt-$(virsh dominfo b-test | awk
'/^UUID:/ {print $2}').files | grep hp
root@b:~# virsh attach-device b-test hotplug-disk.xml --live
Device attached successfully
root@b:~# cat /etc/apparmor.d/libvirt/libvirt-$(virsh dominfo b-test | awk
'/^UUID:/ {print $2}').files | grep hp
"/var/lib/libvirt/images/hp-test.qcow" rwk,
root@b:~# virsh attach-device b-test hotplug-disk2.xml --live
Device attached successfully
root@b:~# cat /etc/apparmor.d/libvirt/libvirt-$(virsh dominfo b-test | awk
'/^UUID:/ {print $2}').files | grep hp
"/var/lib/libvirt/images/hp-test.qcow" rwk,
"/var/lib/libvirt/images/hp-test2.qcow" rwk,
root@b:~# virsh detach-device b-test hotplug-disk.xml --live
Device detached successfully
root@b:~# cat /etc/apparmor.d/libvirt/libvirt-$(virsh dominfo b-test | awk
'/^UUID:/ {print $2}').files | grep hp
"/var/lib/libvirt/images/hp-test2.qcow" rwk,
The one not detached remains in the profile as it should.
So it seems safe against the mentioned "accidental" removal of the other
hotplugged devices.
I can't see how to do so in a unit test atm, maybe down the road in a dep8
or qa test.
But the test above should reassure you that it is not a generic problem
currently.
> +}
> +
> +/* Called when hotplugging */
> +static int
> AppArmorSetSecurityImageLabel(virSecurityManagerPtr mgr,
> virDomainDefPtr def,
> virStorageSourcePtr src)
> @@ -1115,6 +1162,9 @@ virSecurityDriver virAppArmorSecurityDriver = {
> .domainSetSecurityImageLabel =
> AppArmorSetSecurityImageLabel,
> .domainRestoreSecurityImageLabel =
> AppArmorRestoreSecurityImageLabel,
>
> + .domainSetSecurityMemoryLabel = AppArmorSetMemoryLabel,
> + .domainRestoreSecurityMemoryLabel =
> AppArmorRestoreMemoryLabel,
> +
> .domainSetSecurityDaemonSocketLabel =
> AppArmorSetSecurityDaemonSocketLabel,
> .domainSetSecuritySocketLabel =
> AppArmorSetSecuritySocketLabel,
> .domainClearSecuritySocketLabel =
> AppArmorClearSecuritySocketLabel,
--
Jamie Strandboge |
http://www.canonical.com
--
Christian Ehrhardt
Software Engineer, Ubuntu Server
Canonical Ltd