Re: [virtio-comment] Re: [RFC PATCH v1 0/1] virtio-vsock: introduce SEQPACKET description

[Stefano Garzarella <sgarzare@redhat.com>]

On Tue, Mar 16, 2021 at 09:50:48AM -0400, Michael S. Tsirkin wrote:
> On Mon, Feb 22, 2021 at 11:16:54AM +0100, Stefano Garzarella wrote:
> > On Thu, Feb 18, 2021 at 09:07:12AM +0300, Arseny Krasnov wrote:
> > > This patchset adds description of SOCK_SEQPACKET socket's type
> > > of virtio vsock.
> > >
> > > Here is implementation:
> > > https://lkml.org/lkml/2021/2/18/24
> > >
> > > Arseny Krasnov(1):
> > >  virtio-vsock: add SOCK_SEQPACKET description
> > >
> > > virtio-vsock.tex | 40 +++++++++++++++++++++++++++++++++++++---
> > > 1 files changed, 37 insertions(+), 3 deletions(-)
> > >
> > > TODO:
> > > - for messages identification I use header's field called 'msg_cnt'.
> > >  May be it is better to call it 'msg_id' or 'msg_num', because it
> > >  is used as ID, but implemented as counter.
> >
> > If we use it only as an identifier, I think 'msg_id' is preferable and we
> > shouldn't mention in the specs that it's a counter, since it's just a
> > possible implementation.
> > Instead if we use the counter for some control, for example if we lost a
> > packet, then maybe msg_cnt is better.
> > But since the channel shouldn't lose packets, I don't think this is the
> > case.
> >
> > >
> > > - in current version of specification, some values of headers' fields
> > >  still unnamed, for example type of socket (stream or seqpacket), then
> > >  shutdown flags. Spec says that for stream socket value of 'type'
> > >  must be 1. For receive shutdown bit 0 is set in 'flags', for send
> > >  shutdown bit 1 is set in 'flags'. But in Linux these unnamed ones and
> > >  zeroes are implemented as enums, so may be it will be ok to add such
> > >  enums in specification (as 'enum virtio_vsock_event_id').
> >
> > Since we have an enumerate for VIRTIO_VSOCK_OP_* values, I think we can add
> > enums for socket type and maybe 'flags'.
>
> We really need to switch enums to defines, for consistency.

I fully agree, indeed at least in the virtio-vsock part, we use the 
enumerate as they were many defines, always redefining the assigned 
value.
Using defines, we are forced to always define the value and I think 
that's fair!

Thanks,
Stefano