Hi Boris,
On 28/06/2021 22:28, Borislav Petkov wrote:
> Just a couple of notes below:
>
> On Mon, Jun 28, 2021 at 06:34:28PM +0000, Dov Murik wrote:
>> Confidential computing hardware such as AMD SEV (Secure Encrypted
>> Virtualization) allows guest owners to inject secrets into the VMs
>> memory without the host/hypervisor being able to read them. In SEV,
>> secret injection is performed early in the VM launch process, before the
>> guest starts running.
>>
>> Support for secret injection is already available in OVMF (in its AmdSev
>> package; see edk2 commit 01726b6d23d4 "OvmfPkg/AmdSev: Expose the Sev
>> Secret area using a configuration table" [1]), but the secrets were not
>> available in the guest kernel.
>>
>> The patch series copies the secrets from the EFI-provided memory to
>> kernel reserved memory, and optionally exposes them to userspace via
>> securityfs using a new sev_secret kernel module.
>>
>> The first patch in efi/libstub copies the secret area from the EFI
>> memory to specially allocated memory; the second patch reserves that
>> memory block; and the third patch introduces the new sev_secret module
>> that exposes the content of the secret entries as securityfs files, and
>> allows clearing out secrets with a file unlink interface.
>>
>> This has been tested with AMD SEV guests, but the kernel side of
>> handling the secret area has no SEV-specific dependencies, and therefore
>> should be usable for any confidential computing hardware that can
>> publish the secret area via the standard EFI config table entry.
>>
>> Here is a simple example for usage of the sev_secret module in a guest to which
>> secrets were injected during launch:
>
> That's all fine and good but I miss the "why" in this explanation. I.e.,
> a proper use case of a guest owner providing those sekrits to the guest
> would be good.
>
OK, I'll add it in the cover letter; something along the lines of:
An example would be a guest performing computations on encrypted files.
The Guest Owner provides the decryption key (= secret) using the secret
injection mechanism. The guest application reads the secret from the
sev_secret filesystem and proceeds to decrypt the files into memory and
then performs the needed computations on the content.
Host can't read the files from the disk image because they are
encrypted. Host can't read the decryption key because it is passed using
the secret injection mechanism (= secure channel). Host can't read the
decrypted content from memory because it's a confidential
(memory-encrypted) guest.
>>
>> # modprobe sev_secret
>> # ls -la /sys/kernel/security/sev_secret/
>
> So that sysfs URL becomes an ABI. Shouldn't this be:
>
> /sys/kernel/security/coco/
>
> instead which stands for "confidential computing" and contains all kinds
> of protected guest things. TDX might wanna do something similar there,
> for example.
>
On one hand, I agree. This entire series has no SEV-specific code (but
I tested it only on SEV).
On the other hand, secret injection mechanisms in SEV-SNP and TDX are
different beasts than the one used here (SEV). In SEV the secrets must
be injected at VM launch time; so when OVMF runs and kernel efistub runs
the secrets are already there (patches 1+2). However, in SNP there's no
secret injection at launch; (/me hand-waving) the guest can securely
talk with the PSP hardware, check the attestation, and if OK then
securely contact some Guest Owner secret provider to get the required
secrets. Not sure it makes sense for the kernel to be part of this
"getting secrets from secret provider and exposing them in securityfs".
So maybe for regular SEV we'll use this sev_secret module to get one
secret which will allow the guest to contact to the Guest Owner secret
provider (and from here continue like SNP or TDX). Brijesh (AMD) also
suggested collapsing the proposed sev_secret module into the new
sev-guest module ("[PATCH Part1 RFC v3 22/22] virt: Add SEV-SNP guest
driver", sent 2021-06-02), and the logic suggested here will be used
when SNP is not active.
Or taking a step back: Maybe the kernel should not try to unify
SEV/SEV-SNP/TDX/PEF/s390x-SE. Each should have its own API. A
userspace process will have to understand what is available and get the
required info to run the application in a confidential environment.
Or maybe we can find an API that fits all these confidential computing
mechanisms and expose a unified API that hides the underlying
implementation.
(I'm not really sure - that's the reason this is an RFC series.)
>> total 0
>> drwxr-xr-x 2 root root 0 Jun 28 11:54 .
>> drwxr-xr-x 3 root root 0 Jun 28 11:54 ..
>> -r--r----- 1 root root 0 Jun 28 11:54 736870e5-84f0-4973-92ec-06879ce3da0b
>> -r--r----- 1 root root 0 Jun 28 11:54 83c83f7f-1356-4975-8b7e-d3a0b54312c6
>> -r--r----- 1 root root 0 Jun 28 11:54 9553f55d-3da2-43ee-ab5d-ff17f78864d2
>> -r--r----- 1 root root 0 Jun 28 11:54 e6f5a162-d67f-4750-a67c-5d065f2a9910
>>
>> # xxd /sys/kernel/security/sev_secret/e6f5a162-d67f-4750-a67c-5d065f2a9910
>> 00000000: 7468 6573 652d 6172 652d 7468 652d 6b61 these-are-the-ka
>> 00000010: 7461 2d73 6563 7265 7473 0001 0203 0405 ta-secrets......
>> 00000020: 0607 ..
>>
>> # rm /sys/kernel/security/sev_secret/e6f5a162-d67f-4750-a67c-5d065f2a9910
>>
>> # ls -la /sys/kernel/security/sev_secret/
>> total 0
>> drwxr-xr-x 2 root root 0 Jun 28 11:55 .
>> drwxr-xr-x 3 root root 0 Jun 28 11:54 ..
>> -r--r----- 1 root root 0 Jun 28 11:54 736870e5-84f0-4973-92ec-06879ce3da0b
>> -r--r----- 1 root root 0 Jun 28 11:54 83c83f7f-1356-4975-8b7e-d3a0b54312c6
>> -r--r----- 1 root root 0 Jun 28 11:54 9553f55d-3da2-43ee-ab5d-ff17f78864d2
>>
>>
>> [1] https://github.com/tianocore/edk2/commit/01726b6d23d4
>>
>> v2 changes:
>> - Add unlink support in sev_secret securityfs.
>>
>>
>> Dov Murik (3):
>> efi/libstub: Copy confidential computing secret area
>> efi: Reserve confidential computing secret area
>> virt: Add sev_secret module to expose confidential computing secrets
>>
>> drivers/firmware/efi/Makefile | 2 +-
>> drivers/firmware/efi/confidential-computing.c | 41 +++
>> drivers/firmware/efi/efi.c | 5 +
>> drivers/firmware/efi/libstub/Makefile | 3 +-
>> .../efi/libstub/confidential-computing.c | 68 ++++
>> drivers/firmware/efi/libstub/efi-stub.c | 2 +
>> drivers/firmware/efi/libstub/efistub.h | 2 +
>> drivers/firmware/efi/libstub/x86-stub.c | 2 +
>> drivers/virt/Kconfig | 2 +
>> drivers/virt/Makefile | 1 +
>> drivers/virt/sev_secret/Kconfig | 11 +
>> drivers/virt/sev_secret/Makefile | 2 +
>> drivers/virt/sev_secret/sev_secret.c | 298 ++++++++++++++++++
>> include/linux/efi.h | 11 +
>> 14 files changed, 448 insertions(+), 2 deletions(-)
>> create mode 100644 drivers/firmware/efi/confidential-computing.c
>> create mode 100644 drivers/firmware/efi/libstub/confidential-computing.c
>> create mode 100644 drivers/virt/sev_secret/Kconfig
>> create mode 100644 drivers/virt/sev_secret/Makefile
>> create mode 100644 drivers/virt/sev_secret/sev_secret.c
>
> Those "confidential-computing.c" filenames are too long. I'd vote for
> coco.c.
>
When I wrote this I didn't yet encounter "coco" as an abbreviation. Now
there's a linux-coco mailing list, but I saw no other mentions of it in
the kernel (as an abbreviation for confidential computing).
I agree that the full term is too long; I considered conf-comp (but in
my mind "conf" is short for "configuration"). I used it in one place:
"ConfCompSecret" in patch 2/3.
If as a community we settle on coco / CoCo / COCO then I agree these
should be renamed.
(in QEMU they use CGS = Confidential Guest Support [1].)
[1]
https://lore.kernel.org/qemu-devel/20210202041315.196530-3-david@xxxxxxxxxxxxxxxxxxxxx/
> Same for your naming: efi_copy_confidential_computing_secret_area() -
> that is a wow and doesn't look like kernel code to me. :)
>
> Another example why it is too long:
>
> + {LINUX_EFI_CONFIDENTIAL_COMPUTING_SECRET_AREA_GUID,
> + &efi.confidential_computing_secret,
> + "ConfCompSecret"},
>
> I'd do
>
> { LINUX_EFI_COCO_SECRET_AREA_GUID, &efi.coco_secret, "ConfCompSecret" },
>
>
I agree.
Thanks,
-Dov
