Re: [PATCH V3 2/4] crypto: ccp - Move SEV_INIT retry for corrupted data

[Peter Gonda]

On Tue, Nov 9, 2021 at 10:31 AM Sean Christopherson <seanjc@xxxxxxxxxx> wrote:
>
> On Tue, Nov 09, 2021, Peter Gonda wrote:
> > On Tue, Nov 9, 2021 at 9:31 AM Sean Christopherson <seanjc@xxxxxxxxxx> wrote:
> > >
> > > On Tue, Nov 02, 2021, Peter Gonda wrote:
> > > > This change moves the data corrupted retry of SEV_INIT into the
> > >
> > > Use imperative mood.
> >
> > Will do for next revision
> >
> > >
> > > > __sev_platform_init_locked() function. This is for upcoming INIT_EX
> > > > support as well as helping direct callers of
> > > > __sev_platform_init_locked() which currently do not support the
> > > > retry.
> > > >
> > > > Signed-off-by: Peter Gonda <pgonda@xxxxxxxxxx>
> > > > Reviewed-by: Marc Orr <marcorr@xxxxxxxxxx>
> > > > Acked-by: David Rientjes <rientjes@xxxxxxxxxx>
> > > > Acked-by: Tom Lendacky <thomas.lendacky@xxxxxxx>
> > > > Cc: Tom Lendacky <thomas.lendacky@xxxxxxx>
> > > > Cc: Brijesh Singh <brijesh.singh@xxxxxxx>
> > > > Cc: Marc Orr <marcorr@xxxxxxxxxx>
> > > > Cc: Joerg Roedel <jroedel@xxxxxxx>
> > > > Cc: Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx>
> > > > Cc: David Rientjes <rientjes@xxxxxxxxxx>
> > > > Cc: John Allen <john.allen@xxxxxxx>
> > > > Cc: "David S. Miller" <davem@xxxxxxxxxxxxx>
> > > > Cc: Paolo Bonzini <pbonzini@xxxxxxxxxx>
> > > > Cc: linux-crypto@xxxxxxxxxxxxxxx
> > > > Cc: linux-kernel@xxxxxxxxxxxxxxx
> > > > ---
> > > >  drivers/crypto/ccp/sev-dev.c | 24 ++++++++++++------------
> > > >  1 file changed, 12 insertions(+), 12 deletions(-)
> > > >
> > > > diff --git a/drivers/crypto/ccp/sev-dev.c b/drivers/crypto/ccp/sev-dev.c
> > > > index ec89a82ba267..e4bc833949a0 100644
> > > > --- a/drivers/crypto/ccp/sev-dev.c
> > > > +++ b/drivers/crypto/ccp/sev-dev.c
> > > > @@ -267,6 +267,18 @@ static int __sev_platform_init_locked(int *error)
> > > >       }
> > > >
> > > >       rc = __sev_do_cmd_locked(SEV_CMD_INIT, &data, error);
> > > > +     if (rc && *error == SEV_RET_SECURE_DATA_INVALID) {
> > >
> > > There are no guarantees that @error is non-NULL as this is reachable via an
> > > exported function, sev_platform_init().  Which ties in with my complaints in the
> > > previous patch that the API is a bit of a mess.
> >
> > That seems like a bug from the caller right? Is it typical that we
> > sanity-check the caller in these instances?
>
> sev-dev.c needs to make up its mind.  __sev_do_cmd_locked() very clearly allows
> a NULL @error, ergo all of the wrappers for sev_do_cmd() support a NULL @error.
>
> > For example the same comment could be made here:
> > https://elixir.bootlin.com/linux/latest/source/drivers/crypto/ccp/sev-dev.c#L336
> >
> > ```
> > static int sev_get_platform_state(int *state, int *error)
> > {
> > struct sev_user_data_status data;
> > int rc;
> >
> > rc = __sev_do_cmd_locked(SEV_CMD_PLATFORM_STATUS, &data, error);
> > if (rc)
> > return rc;
> >
> > *state = data.state;  <--- State could be null.
>
> No, because this is an internal helper and all call sites can be easily audited.
>
> > return rc;
> > }
> > ```
> >
> > Example outside of this driver:
> > https://elixir.bootlin.com/linux/v5.15.1/source/arch/x86/kvm/x86.c#L468
> >
> > ```
> > int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
> > {
> > enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
> > enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);  <---
> > msr_info could be null here
> > u64 reserved_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu) | 0x2ff |
> > (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
> >
> > if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
> > return 1;
> > if (!msr_info->host_initiated) {
> > if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
> > return 1;
> > if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
> > return 1;
> > }
> >
> > kvm_lapic_set_base(vcpu, msr_info->data);
> > kvm_recalculate_apic_map(vcpu->kvm);
> > return 0;
> > }
> > EXPORT_SYMBOL_GPL(kvm_set_apic_base);
> > ```
>
> The difference is that KVM has consistent expecations for a set of functions,
> whereas sev-dev.c does not.   Yes, KVM will explode if @msr_info is NULL, and
> there are undoubtedly a bajillion flows in the kernel that would do the same,
> but unlike the functions declared in include/linux/psp-sev.h() the requirements
> on the caller are fairly obvious.  E.g. why should this be illegal from a caller's
> perspective?
>
>         sev_platform_init(NULL);
>         sev_platform_status(&status, NULL);

Ack. I'll store a intermediate error in __sev_platform_init_locked and
export to @error if its not null.