[libvirt] [PATCH 4/5] qemu: Enable secure boot

Thu Aug 4 14:39:56 UTC 2016

On Thu, Aug 04, 2016 at 03:45:39PM +0200, Laszlo Ersek wrote:
> On 08/04/16 15:14, Pavel Hrdina wrote:
> > On Wed, Jul 27, 2016 at 05:11:59PM +0200, Laszlo Ersek wrote:
> >> On 07/27/16 10:43, Michal Privoznik wrote:
> >>> In qemu, enabling this feature boils down to adding the following
> >>> onto the command line:
> >>>
> >>>   -global driver=cfi.pflash01,property=secure,value=on
> >>>
> >>> However, there are some constraints resulting from the
> >>> implementation. For instance, System Management Mode (SMM) is
> >>> required to be enabled, the machine type must be q35-2.5 or
> > 
> > s/q35-2.5/q35-2.4/
> > 
> >>> later, and the guest should be x86_64. While technically it is
> >>> possible to have 32 bit guests with secure boot, some non-trivial
> >>> CPU flags tuning is required (for instance lm and nx flags must
> >>> be prohibited). Given complexity of our CPU driver, this is not
> >>> trivial. Therefore I've chosen to forbid 32 bit guests for now.
> >>> If there's ever need, we can refine the check later.
> >>>
> >>> Signed-off-by: Michal Privoznik <mprivozn at redhat.com>
> >>> ---
> >>>  src/qemu/qemu_command.c                            |  7 ++++++
> >>>  src/qemu/qemu_domain.c                             | 27 ++++++++++++++++++++
> >>>  .../qemuxml2argv-bios-nvram-secure.args            | 29 ++++++++++++++++++++++
> >>>  tests/qemuxml2argvtest.c                           |  7 ++++++
> >>>  4 files changed, 70 insertions(+)
> >>>  create mode 100644 tests/qemuxml2argvdata/qemuxml2argv-bios-nvram-secure.args
> >>
> >> This patch looks almost complete to me (it causes all necessary QEMU
> >> options to appear, directly or indirectly (= via requiring SMM)).
> >> However, can you also enforce that the Q35 machtype has version 2.5 or
> >> later? Technically, "pc-q35-2.4" exists too, and it's not good enough
> >> (according to the instructions I wrote up in OvmfPkg/README earlier). I
> >> certainly never tested it.
> >>
> >> Thanks,
> >> Laszlo
> > 
> > I've tested it and it seems to work also with "pc-q35-2.4".  I've installed
> > Fedora 24 inside a guest and I can see "Secure boot enabled" in dmesg output.
> > Unless Laszlo has some more information about secure boot and why it shouldn't
> > work with "pc-q35-2.4" this patch can be pushed as is.
> > 
> > ACK
> > 
> 
> The Secure Boot feature and the SMM driver stack are orthogonal
> build-time options in OVMF. You may enable both, you may enable neither,
> and you may enable only one of them as well.
> 
> However, the Secure Boot feature is not actually secure without the SMM
> driver stack. Meaning, the software interfaces that relate to Secure
> Boot will be available, and -- once certificates have been enrolled -- a
> "well behaved" guest will see Secure Boot enabled. However, a malicious
> guest can directly tamper with the pflash chip that stores the
> authenticated UEFI variables. In other words, without the SMM driver
> stack, a malicious guest can subvert / circumvent Secure Boot.
> 
> Therefore, for "production environments", it makes sense to refer *only*
> to the combination
> 
>   Secure Boot Feature + SMM Driver Stack
> 
> as "secure". This is the meaning of "secure" that you can see in the
> commit messages and the new XML tags here.
> 
> Therefore, whether or not your test results provide actual information,
> depends on the following question: did your OVMF build include the SMM
> Driver Stack *as well*, or only the Secure Boot feature?
> 
> Because, if solely the latter was built into your OVMF binary, that
> would suffice for the
> 
>   Secure boot enabled
> 
> message to appear in the guest dmesg.
> 
> However, per se the message doesn't prove that the SMM driver stack was
> built into the binary. Consequently, the message also doesn't prove that
> pc-q35-2.4 provides everything that's needed for SMM.
> 
> Where did you get your OVMF binary?

I have two OVMF binaries, one from Fedora 24 rpm package:

    edk2-ovmf-20160418gita8c39ba-4.fc24.noarch.rpm

and the second one I've built myself from upstream edk2 repository using

    "./build.sh -a X64 -b DEBUG -p OvmfPkg/OvmfPkgX64.dsc -D SECURE_BOOT_ENABLE -D SMM_REQUIRE"

> ... If you want to verify the presence of the SMM driver stack, please
> add the following to your domain XML (note the QEMU namespace
> declaration in the root element):
> 
>   <domain
>    type='kvm'
>    xmlns:qemu='http://libvirt.org/schemas/domain/qemu/1.0'
>   >
>     [...]
> 
>     <qemu:commandline>
>       <qemu:arg value='-global'/>
>       <qemu:arg value='isa-debugcon.iobase=0x402'/>
>       <qemu:arg value='-debugcon'/>
>       <qemu:arg value='file:/tmp/GUEST_NAME.log'/>
>     </qemu:commandline>
> </domain>
> 
> Then look for the following string in /tmp/GUEST_NAME.log:
> 
>   SMM CPU Module exit from SMRAM with EFI_SUCCESS
> 
> If you see it, your OVMF binary contains the SMM driver stack, and then
> the message in the dmesg is meaningful as evidence. If you don't see it,
> then the message in the dmesg is worthless, as evidence for pc-q35-2.4.

I'm using upstream qemu 2.6.0 and with the OVMF binary from Fedora 24 package
called "OVMF_CODE.secboot.fd" and machine type "pc-q35-2.4" I'm able to see that
line in that log file and also "Secure boot enabled" in dmesg log.

The line "SMM CPU Module exit from SMRAM with EFI_SUCCESS" is also in that log
file with the built OVMF binary.

So based on the testing with OVMF binary from the fedora package it should be
good enough to use also "pc-q35-2.4".

Pavel