Message-ID: <17e319ee-60dc-4d2b-8fd4-0b6960bfa3a7@infradead.org>
Date: Sun, 31 Mar 2024 08:48:11 -0700
From: Randy Dunlap <rdunlap@...radead.org>
To: "Chang S. Bae" <chang.seok.bae@...el.com>, linux-kernel@...r.kernel.org,
 linux-crypto@...r.kernel.org, dm-devel@...hat.com
Cc: ebiggers@...nel.org, luto@...nel.org, dave.hansen@...ux.intel.com,
 tglx@...utronix.de, bp@...en8.de, mingo@...nel.org, x86@...nel.org,
 herbert@...dor.apana.org.au, ardb@...nel.org, elliott@....com,
 dan.j.williams@...el.com, bernie.keany@...el.com,
 charishma1.gairuboyina@...el.com, Bagas Sanjaya <bagasdotme@...il.com>
Subject: Re: [PATCH v9 01/14] Documentation/x86: Document Key Locker

Hi,


On 3/28/24 18:53, Chang S. Bae wrote:
> Document the overview of the feature along with relevant consideration
> when provisioning dm-crypt volumes with AES-KL instead of AES-NI.
> 
> Signed-off-by: Chang S. Bae <chang.seok.bae@...el.com>
> Reviewed-by: Dan Williams <dan.j.williams@...el.com>
> Reviewed-by: Bagas Sanjaya <bagasdotme@...il.com>
> Cc: Randy Dunlap <rdunlap@...radead.org>
> ---
> Changes from v8:
> * Change wording of documentation slightly. (Randy Dunlap and Bagas
>   Sanjaya)
> 
> Changes from v6:
> * Rebase on the upstream -- commit ff61f0791ce9 ("docs: move x86
>   documentation into Documentation/arch/"). (Nathan Huckleberry)
> * Remove a duplicated sentence -- 'But there is no AES-KL instruction
>   to process a 192-bit key.'
> * Update the text for clarity and readability:
>   - Clarify the error code and exemplify the backup failure
>   - Use 'wrapping key' instead of less readable 'IWKey'
> 
> Changes from v5:
> * Fix a typo: 'feature feature' -> 'feature'
> 
> Changes from RFC v2:
> * Add as a new patch.
> 
> The preview is available here:
>   https://htmlpreview.github.io/?https://github.com/intel-staging/keylocker/kdoc/arch/x86/keylocker.html
> ---
>  Documentation/arch/x86/index.rst     |  1 +
>  Documentation/arch/x86/keylocker.rst | 96 ++++++++++++++++++++++++++++
>  2 files changed, 97 insertions(+)
>  create mode 100644 Documentation/arch/x86/keylocker.rst
> 
> diff --git a/Documentation/arch/x86/index.rst b/Documentation/arch/x86/index.rst
> index 8ac64d7de4dc..669c239c009f 100644
> --- a/Documentation/arch/x86/index.rst
> +++ b/Documentation/arch/x86/index.rst
> @@ -43,3 +43,4 @@ x86-specific Documentation
>     features
>     elf_auxvec
>     xstate
> +   keylocker
> diff --git a/Documentation/arch/x86/keylocker.rst b/Documentation/arch/x86/keylocker.rst
> new file mode 100644
> index 000000000000..b28addb8eaf4
> --- /dev/null
> +++ b/Documentation/arch/x86/keylocker.rst
> @@ -0,0 +1,96 @@
> +.. SPDX-License-Identifier: GPL-2.0
> +
> +==============
> +x86 Key Locker
> +==============
> +
> +Introduction
> +============
> +
> +Key Locker is a CPU feature to reduce key exfiltration opportunities
> +while maintaining a programming interface similar to AES-NI. It
> +converts the AES key into an encoded form, called the 'key handle'.
> +The key handle is a wrapped version of the clear-text key where the
> +wrapping key has limited exposure. Once converted, all subsequent data
> +encryption using new AES instructions (AES-KL) uses this key handle,
> +reducing the exposure of private key material in memory.
> +
> +CPU-internal Wrapping Key
> +=========================
> +
> +The CPU-internal wrapping key is an entity in a software-invisible CPU
> +state. On every system boot, a new key is loaded. So the key handle that
> +was encoded by the old wrapping key is no longer usable on system shutdown
> +or reboot.
> +
> +And the key may be lost on the following exceptional situation upon wakeup:
> +
> +Wrapping Key Restore Failure
> +----------------------------
> +
> +The CPU state is volatile with the ACPI S3/4 sleep states. When the system
> +supports those states, the key has to be backed up so that it is restored
> +on wake up. The kernel saves the key in non-volatile media.
> +
> +Upon the event of a wrapping key restore failure upon resume from suspend,
> +all established key handles become invalid. In flight dm-crypt operations

                                               In-flight

> +receive error results from pending operations. In the likely scenario that
> +dm-crypt is hosting the root filesystem the recovery is identical to if a
> +storage controller failed to resume from suspend or reboot. If the volume
> +impacted by a wrapping key restore failure is a data volume then it is
> +possible that I/O errors on that volume do not bring down the rest of the
> +system. However, a reboot is still required because the kernel will have
> +soft-disabled Key Locker. Upon the failure, the crypto library code will
> +return -ENODEV on every AES-KL function call. The Key Locker implementation
> +only loads a new wrapping key at initial boot, not any time after like
> +resume from suspend.
> +
> +Use Case and Non-use Cases
> +==========================
> +
> +Bare metal disk encryption is the only intended use case.
> +
> +Userspace usage is not supported because there is no ABI provided to
> +communicate and coordinate wrapping-key restore failure to userspace. For
> +now, key restore failures are only coordinated with kernel users. But the
> +kernel can not prevent userspace from using the feature's AES instructions
> +('AES-KL') when the feature has been enabled. So, the lack of userspace
> +support is only documented, not actively enforced.
> +
> +Key Locker is not expected to be advertised to guest VMs and the kernel
> +implementation ignores it even if the VMM enumerates the capability. The
> +expectation is that a guest VM wants private wrapping key state, but the
> +architecture does not provide that. An emulation of that capability, by
> +caching per-VM wrapping keys in memory, defeats the purpose of Key Locker.
> +The backup / restore facility is also not performant enough to be suitable
> +for guest VM context switches.
> +
> +AES Instruction Set
> +===================
> +
> +The feature accompanies a new AES instruction set. This instruction set is
> +analogous to AES-NI. A set of AES-NI instructions can be mapped to an
> +AES-KL instruction. For example, AESENC128KL is responsible for ten rounds
> +of transformation, which is equivalent to nine times AESENC and one
> +AESENCLAST in AES-NI.
> +
> +But they have some notable differences:
> +
> +* AES-KL provides a secure data transformation using an encrypted key.
> +
> +* If an invalid key handle is provided, e.g. a corrupted one or a handle
> +  restriction failure, the instruction fails with setting RFLAGS.ZF. The
> +  crypto library implementation includes the flag check to return -EINVAL.
> +  Note that this flag is also set if the wrapping key is changed, e.g.,
> +  because of the backup error.
> +
> +* AES-KL implements support for 128-bit and 256-bit keys, but there is no
> +  AES-KL instruction to process an 192-bit key. The AES-KL cipher
> +  implementation logs a warning message with a 192-bit key and then falls
> +  back to AES-NI. So, this 192-bit key-size limitation is only documented,
> +  not enforced. It means the key will remain in clear-text in memory. This
> +  is to meet Linux crypto-cipher expectation that each implementation must
> +  support all the AES-compliant key sizes.
> +
> +* Some AES-KL hardware implementation may have noticeable performance
> +  overhead when compared with AES-NI instructions.

Reviewed-by: Randy Dunlap <rdunlap@...radead.org>

thanks.
-- 
#Randy

Powered by blists - more mailing lists