With the introduction of an SVSM, Linux will be running at a non-zero
VMPL. Any request for an attestation report at a higher priviledge VMPL
than what Linux is currently running will result in an error. Allow for
the privlevel_floor attribute to be updated dynamically so that the
attribute may be set dynamically.
Signed-off-by: Tom Lendacky <[email protected]>
---
drivers/virt/coco/sev-guest/sev-guest.c | 5 ++++-
include/linux/tsm.h | 2 +-
2 files changed, 5 insertions(+), 2 deletions(-)
diff --git a/drivers/virt/coco/sev-guest/sev-guest.c b/drivers/virt/coco/sev-guest/sev-guest.c
index 1ff897913bf4..bba6531cb606 100644
--- a/drivers/virt/coco/sev-guest/sev-guest.c
+++ b/drivers/virt/coco/sev-guest/sev-guest.c
@@ -885,7 +885,7 @@ static int sev_report_new(struct tsm_report *report, void *data)
return 0;
}
-static const struct tsm_ops sev_tsm_ops = {
+static struct tsm_ops sev_tsm_ops = {
.name = KBUILD_MODNAME,
.report_new = sev_report_new,
};
@@ -972,6 +972,9 @@ static int __init sev_guest_probe(struct platform_device *pdev)
snp_dev->input.resp_gpa = __pa(snp_dev->response);
snp_dev->input.data_gpa = __pa(snp_dev->certs_data);
+ /* Set the privlevel_floor attribute based on the current VMPL */
+ sev_tsm_ops.privlevel_floor = snp_get_vmpl();
+
ret = tsm_register(&sev_tsm_ops, snp_dev, &tsm_report_extra_type);
if (ret)
goto e_free_cert_data;
diff --git a/include/linux/tsm.h b/include/linux/tsm.h
index de8324a2223c..50c5769657d8 100644
--- a/include/linux/tsm.h
+++ b/include/linux/tsm.h
@@ -54,7 +54,7 @@ struct tsm_report {
*/
struct tsm_ops {
const char *name;
- const unsigned int privlevel_floor;
+ unsigned int privlevel_floor;
int (*report_new)(struct tsm_report *report, void *data);
};
--
2.43.2
Tom Lendacky wrote:
> With the introduction of an SVSM, Linux will be running at a non-zero
> VMPL. Any request for an attestation report at a higher priviledge VMPL
> than what Linux is currently running will result in an error. Allow for
> the privlevel_floor attribute to be updated dynamically so that the
> attribute may be set dynamically.
>
> Signed-off-by: Tom Lendacky <[email protected]>
> ---
> drivers/virt/coco/sev-guest/sev-guest.c | 5 ++++-
> include/linux/tsm.h | 2 +-
> 2 files changed, 5 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/virt/coco/sev-guest/sev-guest.c b/drivers/virt/coco/sev-guest/sev-guest.c
> index 1ff897913bf4..bba6531cb606 100644
> --- a/drivers/virt/coco/sev-guest/sev-guest.c
> +++ b/drivers/virt/coco/sev-guest/sev-guest.c
> @@ -885,7 +885,7 @@ static int sev_report_new(struct tsm_report *report, void *data)
> return 0;
> }
>
> -static const struct tsm_ops sev_tsm_ops = {
> +static struct tsm_ops sev_tsm_ops = {
> .name = KBUILD_MODNAME,
> .report_new = sev_report_new,
> };
> @@ -972,6 +972,9 @@ static int __init sev_guest_probe(struct platform_device *pdev)
> snp_dev->input.resp_gpa = __pa(snp_dev->response);
> snp_dev->input.data_gpa = __pa(snp_dev->certs_data);
>
> + /* Set the privlevel_floor attribute based on the current VMPL */
> + sev_tsm_ops.privlevel_floor = snp_get_vmpl();
Why is this not vmpck_id?
On 4/15/24 23:55, Dan Williams wrote:
> Tom Lendacky wrote:
>> With the introduction of an SVSM, Linux will be running at a non-zero
>> VMPL. Any request for an attestation report at a higher priviledge VMPL
>> than what Linux is currently running will result in an error. Allow for
>> the privlevel_floor attribute to be updated dynamically so that the
>> attribute may be set dynamically.
>>
>> Signed-off-by: Tom Lendacky <[email protected]>
>> ---
>> drivers/virt/coco/sev-guest/sev-guest.c | 5 ++++-
>> include/linux/tsm.h | 2 +-
>> 2 files changed, 5 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/virt/coco/sev-guest/sev-guest.c b/drivers/virt/coco/sev-guest/sev-guest.c
>> index 1ff897913bf4..bba6531cb606 100644
>> --- a/drivers/virt/coco/sev-guest/sev-guest.c
>> +++ b/drivers/virt/coco/sev-guest/sev-guest.c
>> @@ -885,7 +885,7 @@ static int sev_report_new(struct tsm_report *report, void *data)
>> return 0;
>> }
>>
>> -static const struct tsm_ops sev_tsm_ops = {
>> +static struct tsm_ops sev_tsm_ops = {
>> .name = KBUILD_MODNAME,
>> .report_new = sev_report_new,
>> };
>> @@ -972,6 +972,9 @@ static int __init sev_guest_probe(struct platform_device *pdev)
>> snp_dev->input.resp_gpa = __pa(snp_dev->response);
>> snp_dev->input.data_gpa = __pa(snp_dev->certs_data);
>>
>> + /* Set the privlevel_floor attribute based on the current VMPL */
>> + sev_tsm_ops.privlevel_floor = snp_get_vmpl();
>
> Why is this not vmpck_id?
Good catch, this probably should be pulled out separately and submitted
as a Fixes: against the current support. If you think it's important
enough, I can do that and put this at the beginning of the series. Or I
can just modify this to use the vmpck_id value. Any preference?
Thanks,
Tom
Tom Lendacky wrote:
> On 4/15/24 23:55, Dan Williams wrote:
> > Tom Lendacky wrote:
> >> With the introduction of an SVSM, Linux will be running at a non-zero
> >> VMPL. Any request for an attestation report at a higher priviledge VMPL
> >> than what Linux is currently running will result in an error. Allow for
> >> the privlevel_floor attribute to be updated dynamically so that the
> >> attribute may be set dynamically.
> >>
> >> Signed-off-by: Tom Lendacky <[email protected]>
> >> ---
> >> drivers/virt/coco/sev-guest/sev-guest.c | 5 ++++-
> >> include/linux/tsm.h | 2 +-
> >> 2 files changed, 5 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/drivers/virt/coco/sev-guest/sev-guest.c b/drivers/virt/coco/sev-guest/sev-guest.c
> >> index 1ff897913bf4..bba6531cb606 100644
> >> --- a/drivers/virt/coco/sev-guest/sev-guest.c
> >> +++ b/drivers/virt/coco/sev-guest/sev-guest.c
> >> @@ -885,7 +885,7 @@ static int sev_report_new(struct tsm_report *report, void *data)
> >> return 0;
> >> }
> >>
> >> -static const struct tsm_ops sev_tsm_ops = {
> >> +static struct tsm_ops sev_tsm_ops = {
> >> .name = KBUILD_MODNAME,
> >> .report_new = sev_report_new,
> >> };
> >> @@ -972,6 +972,9 @@ static int __init sev_guest_probe(struct platform_device *pdev)
> >> snp_dev->input.resp_gpa = __pa(snp_dev->response);
> >> snp_dev->input.data_gpa = __pa(snp_dev->certs_data);
> >>
> >> + /* Set the privlevel_floor attribute based on the current VMPL */
> >> + sev_tsm_ops.privlevel_floor = snp_get_vmpl();
> >
> > Why is this not vmpck_id?
>
> Good catch, this probably should be pulled out separately and submitted
> as a Fixes: against the current support. If you think it's important
> enough, I can do that and put this at the beginning of the series. Or I
> can just modify this to use the vmpck_id value. Any preference?
I dunno, you tell me. What breaks if privlevel_floor is mismatched vs
vmpl and/or vmpck_id? If it warrants a "Fixes:" it should probably be
broken out.
However, I *guess* it is just adding some sanity checking precision to
userspace requests and makes some input validation not catch errors when
userspace tries to generate reports from the wrong level, right? I.e.
privlevel_floor may be lower than expected, but userspace should not be
depending on that since the report generation will fail.
On 4/16/24 10:57, Dan Williams wrote:
> Tom Lendacky wrote:
>> On 4/15/24 23:55, Dan Williams wrote:
>>> Tom Lendacky wrote:
>>>> With the introduction of an SVSM, Linux will be running at a non-zero
>>>> VMPL. Any request for an attestation report at a higher priviledge VMPL
>>>> than what Linux is currently running will result in an error. Allow for
>>>> the privlevel_floor attribute to be updated dynamically so that the
>>>> attribute may be set dynamically.
>>>>
>>>> Signed-off-by: Tom Lendacky <[email protected]>
>>>> ---
>>>> drivers/virt/coco/sev-guest/sev-guest.c | 5 ++++-
>>>> include/linux/tsm.h | 2 +-
>>>> 2 files changed, 5 insertions(+), 2 deletions(-)
>>>>
>>>> diff --git a/drivers/virt/coco/sev-guest/sev-guest.c b/drivers/virt/coco/sev-guest/sev-guest.c
>>>> index 1ff897913bf4..bba6531cb606 100644
>>>> --- a/drivers/virt/coco/sev-guest/sev-guest.c
>>>> +++ b/drivers/virt/coco/sev-guest/sev-guest.c
>>>> @@ -885,7 +885,7 @@ static int sev_report_new(struct tsm_report *report, void *data)
>>>> return 0;
>>>> }
>>>>
>>>> -static const struct tsm_ops sev_tsm_ops = {
>>>> +static struct tsm_ops sev_tsm_ops = {
>>>> .name = KBUILD_MODNAME,
>>>> .report_new = sev_report_new,
>>>> };
>>>> @@ -972,6 +972,9 @@ static int __init sev_guest_probe(struct platform_device *pdev)
>>>> snp_dev->input.resp_gpa = __pa(snp_dev->response);
>>>> snp_dev->input.data_gpa = __pa(snp_dev->certs_data);
>>>>
>>>> + /* Set the privlevel_floor attribute based on the current VMPL */
>>>> + sev_tsm_ops.privlevel_floor = snp_get_vmpl();
>>>
>>> Why is this not vmpck_id?
>>
>> Good catch, this probably should be pulled out separately and submitted
>> as a Fixes: against the current support. If you think it's important
>> enough, I can do that and put this at the beginning of the series. Or I
>> can just modify this to use the vmpck_id value. Any preference?
>
> I dunno, you tell me. What breaks if privlevel_floor is mismatched vs
> vmpl and/or vmpck_id? If it warrants a "Fixes:" it should probably be
> broken out.
>
> However, I *guess* it is just adding some sanity checking precision to
> userspace requests and makes some input validation not catch errors when
> userspace tries to generate reports from the wrong level, right? I.e.
> privlevel_floor may be lower than expected, but userspace should not be
> depending on that since the report generation will fail.
Yeah, it just results in a different type of error. If the VMPL
specified by the user is numerically lower than the vmpck_id, then the
request will fail with a specific return code value. With the change to
privlevel_floor to use vmpck_id, then you would just get the error that
much sooner when trying to set a value that is lower than the floor.
Since I don't think the vmpck_id module parameter is a common case
today, let's just leave that change in this patch.
Thanks,
Tom