[PATCH v4 2/7] qemu: tpm: Allow offline migration with TPM_EMULATOR only with shared storage

[Stefan Berger]

On 11/7/22 03:31, Michal Prívozník wrote:
> On 10/24/22 12:28, Stefan Berger wrote:
>> Allow migration with TPM_EMULATOR (swtpm) only if shared storage has been
>> set up for the TPM state directory.
>>
>> Signed-off-by: Stefan Berger <stefanb at linux.ibm.com>
>> ---
>>   src/qemu/qemu_migration.c |  6 ++++++
>>   src/qemu/qemu_tpm.c       | 28 ++++++++++++++++++++++++++++
>>   src/qemu/qemu_tpm.h       |  5 +++++
>>   3 files changed, 39 insertions(+)
>>
>> diff --git a/src/qemu/qemu_migration.c b/src/qemu/qemu_migration.c
>> index 33105cf07b..16bf7ac178 100644
>> --- a/src/qemu/qemu_migration.c
>> +++ b/src/qemu/qemu_migration.c
>> @@ -38,6 +38,7 @@
>>   #include "qemu_security.h"
>>   #include "qemu_slirp.h"
>>   #include "qemu_block.h"
>> +#include "qemu_tpm.h"
>>   
>>   #include "domain_audit.h"
>>   #include "virlog.h"
>> @@ -2579,6 +2580,11 @@ qemuMigrationSrcBeginPhase(virQEMUDriver *driver,
>>                              _("tunnelled offline migration does not make sense"));
>>               return NULL;
>>           }
>> +        if (qemuTPMHasSharedStorage(driver, vm->def) != 1) {
>> +            virReportError(VIR_ERR_OPERATION_INVALID, "%s",
>> +                           _("offline migration requires TPM state directory to be on shared storage"));
> 
> Does it though? We don't have such requirement for say domain disks. I
> believe the point of --offline is to merely just get domain XML an
> define it on the destination. That's why --offline requires --persistent
> and why a lot of checks are skipped when --offline.
> 
> I mean, for disks we just assume users will copy them onto destination
> (or use a shared FS). We can assume the same for TPM, can't we? This
> would also allow us to simplify qemuTPMHasSharedStorage() because it no
> longer needs to call qemuTPMEmulatorInitPaths() because for running
> guests the paths were already initialized.
> 
> And if we chose to keep this check in, then I think it should live in
> qemuMigrationSrcIsAllowed() which is the places where similar checks are
> performed.

I don't mind dropping the patch then.

   Stefan

> 
> Michal
>