Re: [PATCH v3] KVM: SEV: Add KVM_SEV_SNP_HV_REPORT_REQ command

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

On 1/24/26 08:40, Thomas Courrege wrote:
> Sorry, i didn't saw the response, i changed the email i use.
> 
> On 21-01-2026 00:45, Tom Lendacky wrote:
>> On 12/15/25 08:14, Thomas Courrege wrote:
>>
>>> +	size_t rsp_size = sizeof(*report_rsp);
>>> +	int ret;
>> The declarations above should be in reverse fir tree order.
>     
> Like that ?
>     struct sev_data_snp_msg_report_rsp *report_rsp;
>     struct sev_data_snp_hv_report_req data;
>     struct kvm_sev_snp_hv_report_req params;
>     struct kvm_sev_info *sev = to_kvm_sev_info(kvm);
>     size_t rsp_size = sizeof(*report_rsp);
>     void __user *u_report;
>     void __user *u_params;
>     int ret;

	struct kvm_sev_info *sev = to_kvm_sev_info(kvm);
	struct sev_data_snp_msg_report_rsp *report_rsp;
	struct kvm_sev_snp_hv_report_req params;
	struct sev_data_snp_hv_report_req data;
	size_t rsp_size = sizeof(*report_rsp);
	void __user *u_report;
	void __user *u_params;
	int ret;

>>> +	if (ret)
>>> +		goto e_free_rsp;
>>> +
>>> +	if (!report_rsp->status)
>>> +		rsp_size += report_rsp->report_size;
>>> +
>>> +	if (params.report_len < rsp_size) {
>>> +		rsp_size = sizeof(*report_rsp);
>>> +		ret = -ENOSPC;
>>> +	}
>> This can be contained within the if above it, right?
>>
>> if (!report_rsp->status) {
>> 	if (params.report_len < (rsp_size + report_rsp->report_size))
>> 		ret = -ENOSPC;
>> 	else
>> 		rsp_size += report_rsp->report_size;
>> }
> 
> This leads to an error in case the user wants to query the report size.
> 
> 
> Using params.report_len = 32, the nested if is true and thus the user get
> 
> back the default rsp_size (= 32), not increased with report_size (= 1184).

But isn't params.report_len set below to the proper value since it wasn't
using rsp_size? The rsp_size variable is only used for the copy_to_user()
for the report itself. Assuming you want to copy what's in 'rsp' no matter
the return code you get, then can't you just do:

if (!report_rsp->status) {
	if (params.report_len < (rsp_size + report_rsp->report_size))
		ret = -ENOSPC;
	else
		rsp_size += report_rsp->report_size;

	params.report_len = sizeof(*report_rsp) + report_rsp->report_size;
}

if (copy_to_user(u_report, report_rsp, rsp_size))
	ret = -EFAULT;

Thanks,
Tom

> 
>>> +
>>> +	if (copy_to_user(u_report, report_rsp, rsp_size))
>>> +		ret = -EFAULT;
>>> +
>>> +	params.report_len = sizeof(*report_rsp) + report_rsp->report_size;
>> I'm not sure if we can rely on report_rsp->report_size being valid if
>> resport_rsp->status is not zero. So maybe just set this to rsp_size.
>>
>> Thanks,
>> Tom
> maybe something like this ? to avoid copying on ENOSPC, where this issue come from
> 
>     if (!report_rsp->status)
>         rsp_size += report_rsp->report_size;
> 
>     if (params.report_len < rsp_size) {
>         ret = -ENOSPC;
>     } else {
>         if (copy_to_user(u_report, report_rsp, rsp_size))
>             ret = -EFAULT;
>     }
> 
>     params.report_len = rsp_size;
> 
> 
> To test this specific case : 
>     https://github.com/Th0rOnDoR/test-length-sev/blob/main/sev_test.c
> 
> Thanks, 
> Thomas