[PATCH 6/6] msi: Make MsiProcessMessage() RPC-compatible.

[Hans Leidekker]

On Tue, 2018-04-17 at 09:28 -0500, Zebediah Figura wrote:
> On 17/04/18 09:17, Hans Leidekker wrote:
> > On Mon, 2018-04-16 at 20:20 -0500, Zebediah Figura wrote:
> > > +UINT unmarshal_record(const struct wire_record *in, MSIHANDLE *out)
> > > +{
> > > +    MSIRECORD *rec;
> > > +    unsigned int i;
> > > +    UINT r;
> > > +
> > > +    rec = MSI_CreateRecord(in->count);
> > > +    if (!rec) return ERROR_OUTOFMEMORY;
> > > +
> > > +    for (i = 0; i <= in->count; i++)
> > > +    {
> > > +        switch (in->fields[i].type)
> > > +        {
> > > +        case MSIFIELD_NULL:
> > > +            break;
> > > +        case MSIFIELD_INT:
> > > +        case MSIFIELD_INTPTR:
> > > +            r = MSI_RecordSetInteger(rec, i, in->fields[i].u.iVal);
> > > +            break;
> > 
> > MSIFIELD_INTPTR is used to store a pointer in MsiViewFetch. We should probably
> > get rid of that but while we have it I think it would be better to print a
> > message here and return an error.
> > 
> Unless I'm misunderstanding the code, it's necessary for MsiViewModify
> to work, and I have tests showing that function works across custom actions.

The pointer would get truncated in a 64-bit process if you store it like that.
Since this field type is only used to validate the record in MsiViewModify I
think it would be better to get rid of it and find some other way to validate
the record.