Re: [PATCH v5 07/10] crypto: ccp: Add preferred access checking method

[Tom Lendacky <thomas.lendacky@amd.com>]

On 11/12/24 13:47, Dionna Amalie Glaze wrote:
> On Mon, Nov 11, 2024 at 2:46 PM Tom Lendacky <thomas.lendacky@amd.com> wrote:
>>
>> On 11/7/24 17:24, Dionna Glaze wrote:
>>> sev_issue_cmd_external_user is the only function that checks permissions
>>> before performing its task. With the new GCTX API, it's important to
>>> establish permission once and have that determination dominate later API
>>> uses. This is implicitly how ccp has been used by dominating uses of
>>> sev_do_cmd by a successful sev_issue_cmd_external_user call.
>>>
>>> Consider sev_issue_cmd_external_user deprecated by
>>> checking if a held file descriptor passes file_is_sev, similar to the
>>> file_is_kvm function.
>>>
>>> This also fixes the header comment that the bad file error code is
>>> -%EINVAL when in fact it is -%EBADF.
>>
>> Same comment as before. This commit merely creates a helper function, so
>> this commit message is not appropriate.
>>
> 
> Is this a meta-comment about how the commit presupposes being in a
> series with a goal, but should have a self-contained commit message? I
> don't know what "same comment as before" you're referring to.

I made the same comment in your previous series.

> How about this:
> 
> crypto: ccp: Add file_is_sev to identify access
> 
> Access to the ccp driver only needs to be determined once, so

once per KVM ioctl invocation

> sev_issue_cmd_external_user called in a loop (e.g. for
> SNP_LAUNCH_UPDATE) does more than it needs to.
> 
> The file_is_sev function allows the caller to determine access before using
> sev_do_cmd or other API methods multiple times without extra access
> checking.
> 
> This also fixes the header comment that the bad file error code is
> -%EINVAL when in fact it is -%EBADF.

Yes, I like this better.

Thanks,
Tom

> 
> 
> 
>