On Tue, Aug 26, 2014 at 12:07:54PM -0400, Stefan Berger wrote:
"Kevin O'Connor" kevin@koconnor.net wrote on 08/26/2014 11:19:14 AM:
If this is the intent, can't we just pass a flag (via fw_cfg) from QEMU command line to SeaBIOS to force a clear? That is, the guest software can't manipulate the QEMU command line (or its fw_cfg entries) and so the ability to set a flag there is proof of physical presence. (Access to the virtual machine disk images and virtual machine command line is as close to "physical" as one can get.)
One would need at least a flag to indicate that the BIOS automatically give up ownership of the TPM. Giving up ownership also means that the device automatically becomes disabled and deactivated. The BIOS would then presumably automatically have to enabled and activate the TPM again without user interaction.
Off the top of my head, I would think one could use a single multi-purpose state indicator (eg, "TPM is enabled, active, ownable", "TPM is enabled, active, non-ownable", "TPM is disabled, deactivated, cleared, unownable", etc.). I'd guess there are several permutations that wouldn't make sense and the total list could be simplified.
The other aspect is that this extension propagates all the way into higher layers: libvirt would need an API and command line tool extension just to set this flag and presumably use the QEMU monitor with a new command to indicate it. You want to be able to do this in a cloud environment, you need another API and/or GUI support in your cloud stack for doing just this... I doesn't seem to become a lot easier this way.
Not easier. But I don't think adding this menu to SeaBIOS is the solution either. As before, for the bulk of users it's just cryptic, and for those rare users that do need it, it is not in a place they want it.
On coreboot, a similar solution could be accomplished by setting a flag in CBFS (the flash). Granted, one doesn't need to be physically present to reprogram the flash, but if one can reprogram the flash, they could just as easily reprogram SeaBIOS anyway.
I am not so familiar with how CBFS is handled. Is it at least access-restricted to root? I guess one would need a tool to write the above flag(s) into the flash at the right position.
Yes, at a minimum it would require super-user access. (Some hardware platforms introduce additional restrictions.)
-Kevin
"Kevin O'Connor" kevin@koconnor.net wrote on 08/26/2014 12:49:57 PM:
From: "Kevin O'Connor" kevin@koconnor.net To: Stefan Berger/Watson/IBM@IBMUS Cc: seabios@seabios.org, Stefan Berger stefanb@linux.vnet.ibm.com Date: 08/26/2014 12:50 PM Subject: Re: [SeaBIOS] [PATCH v8 0/8] Add TPM support to SeaBIOS
On Tue, Aug 26, 2014 at 12:07:54PM -0400, Stefan Berger wrote:
"Kevin O'Connor" kevin@koconnor.net wrote on 08/26/2014 11:19:14 AM:
If this is the intent, can't we just pass a flag (via fw_cfg) from QEMU command line to SeaBIOS to force a clear? That is, the guest software can't manipulate the QEMU command line (or its fw_cfg entries) and so the ability to set a flag there is proof of physical presence. (Access to the virtual machine disk images and virtual machine command line is as close to "physical" as one can get.)
One would need at least a flag to indicate that the BIOS automatically
give up ownership of the TPM. Giving up ownership also means that the device automatically becomes disabled and deactivated. The BIOS would then presumably automatically have to enabled and activate the TPM again without user interaction.
Off the top of my head, I would think one could use a single multi-purpose state indicator (eg, "TPM is enabled, active, ownable", "TPM is enabled, active, non-ownable", "TPM is disabled, deactivated, cleared, unownable", etc.). I'd guess there are several permutations that wouldn't make sense and the total list could be simplified.
The other aspect is that this extension propagates all the way into
higher
layers: libvirt would need an API and command line tool extension just to set this flag and presumably use the QEMU
monitor with a new command to indicate it. You want to be able to do this in a cloud environment, you need
another
API and/or GUI support in your cloud stack for doing just this... I doesn't seem to become a lot easier this way.
Not easier. But I don't think adding this menu to SeaBIOS is the solution either. As before, for the bulk of users it's just cryptic, and for those rare users that do need it, it is not in a place they want it.
Let me approach this all the way from the cloud user perspective:
One can add a TPM to a VM via an attribute to the image one wants to deploy. So having a TPM attached to a VM would be an option. Libvirt can then create a slightly different XML that instructs QEMU to show the SeaBIOS menu IF a TPM is attached. If no TPM is attached, no menu is shown, so no changes here. I find this a useable solution that helps those that want a TPM to be attached to the VM and leaves things as they are for those that don't.
OpenStack for example allows you to interact with the VGA console that QEMU presents and where you have access to the SeaBIOS menu. If one needs to interact with the menu for example to release ownership of the TPM, then this would be an occasion a user presses the F11 button to get to the TPM menu and control it, otherwise the user can wait those 2 seconds for the auto-boot to start.
Having to interact with cloud CLI tools or libvirt tools for setting the flags for what the BIOS is supposed to do with the TPM the next time the VM is warm-rebooted (cold boot presumably would not work) doesn't sound that attractive. Besides that root gets too much control over VMs with attached TPMs running on a system if libvirt was to be extended with this kind of support. You could then set the above mentioned flags and see what happens when the VM gives up ownership the next time it boots...
Stefan
-
Kevin O'Connor -
Stefan Berger