Re: [RFC PATCH v3 1/9] vsock: SO_RCVLOWAT transport set callback

[Stefano Garzarella]

On Tue, Aug 09, 2022 at 09:45:47AM +0000, Arseniy Krasnov wrote:
> On 09.08.2022 12:37, Arseniy Krasnov wrote:
> > On 08.08.2022 13:30, Stefano Garzarella wrote:
> > > On Mon, Aug 8, 2022 at 12:23 PM Stefano Garzarella <sgarzare@xxxxxxxxxx> wrote:
> > > > On Wed, Aug 03, 2022 at 01:51:05PM +0000, Arseniy Krasnov wrote:
> > > > > This adds transport specific callback for SO_RCVLOWAT, because in some
> > > > > transports it may be difficult to know current available number of bytes
> > > > > ready to read. Thus, when SO_RCVLOWAT is set, transport may reject it.
> > > > >
> > > > > Signed-off-by: Arseniy Krasnov <AVKrasnov@xxxxxxxxxxxxxx>
> > > > > ---
> > > > > include/net/af_vsock.h   |  1 +
> > > > > net/vmw_vsock/af_vsock.c | 25 +++++++++++++++++++++++++
> > > > > 2 files changed, 26 insertions(+)
> > > > >
> > > > > diff --git a/include/net/af_vsock.h b/include/net/af_vsock.h
> > > > > index f742e50207fb..eae5874bae35 100644
> > > > > --- a/include/net/af_vsock.h
> > > > > +++ b/include/net/af_vsock.h
> > > > > @@ -134,6 +134,7 @@ struct vsock_transport {
> > > > >       u64 (*stream_rcvhiwat)(struct vsock_sock *);
> > > > >       bool (*stream_is_active)(struct vsock_sock *);
> > > > >       bool (*stream_allow)(u32 cid, u32 port);
> > > > > +      int (*set_rcvlowat)(struct vsock_sock *, int);
> > > >
> > > > checkpatch suggests to add identifier names. For some we put them in,
> > > > for others we didn't, but I suggest putting them in for the new ones
> > > > because I think it's clearer too.
> > > >
> > > > WARNING: function definition argument 'struct vsock_sock *' should also
> > > > have an identifier name
> > > > #25: FILE: include/net/af_vsock.h:137:
> > > > +       int (*set_rcvlowat)(struct vsock_sock *, int);
> > > >
> > > > WARNING: function definition argument 'int' should also have an identifier name
> > > > #25: FILE: include/net/af_vsock.h:137:
> > > > +       int (*set_rcvlowat)(struct vsock_sock *, int);
> > > >
> > > > total: 0 errors, 2 warnings, 0 checks, 44 lines checked
> > > >
> > > > >       /* SEQ_PACKET. */
> > > > >       ssize_t (*seqpacket_dequeue)(struct vsock_sock *vsk, struct msghdr *msg,
> > > > > diff --git a/net/vmw_vsock/af_vsock.c b/net/vmw_vsock/af_vsock.c
> > > > > index f04abf662ec6..016ad5ff78b7 100644
> > > > > --- a/net/vmw_vsock/af_vsock.c
> > > > > +++ b/net/vmw_vsock/af_vsock.c
> > > > > @@ -2129,6 +2129,30 @@ vsock_connectible_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
> > > > >       return err;
> > > > > }
> > > > >
> > > > > +static int vsock_set_rcvlowat(struct sock *sk, int val)
> > > > > +{
> > > > > +      const struct vsock_transport *transport;
> > > > > +      struct vsock_sock *vsk;
> > > > > +      int err = 0;
> > > > > +
> > > > > +      vsk = vsock_sk(sk);
> > > > > +
> > > > > +      if (val > vsk->buffer_size)
> > > > > +              return -EINVAL;
> > > > > +
> > > > > +      transport = vsk->transport;
> > > > > +
> > > > > +      if (!transport)
> > > > > +              return -EOPNOTSUPP;
> > > >
> > > > I don't know whether it is better in this case to write it in
> > > > sk->sk_rcvlowat, maybe we can return EOPNOTSUPP only when the trasport
> > > > is assigned and set_rcvlowat is not defined. This is because usually the
> > > > options are set just after creation, when the transport is practically
> > > > unassigned.
> > > >
> > > > I mean something like this:
> > > >
> > > >          if (transport) {
> > > >                  if (transport->set_rcvlowat)
> > > >                          return transport->set_rcvlowat(vsk, val);
> > > >                  else
> > > >                          return -EOPNOTSUPP;
> > > >          }
> > > >
> > > >          WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
> > > >
> > > >          return 0;
> > >
> > > Since hv_sock implements `set_rcvlowat` to return EOPNOTSUPP. maybe we
> > > can just do the following:
> > >
> > >         if (transport && transport->set_rcvlowat)
> > >                 return transport->set_rcvlowat(vsk, val);
> > >
> > >         WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
> > >         return 0;
> > >
> > > That is, the default behavior is to set sk->sk_rcvlowat, but for
> > > transports that want a different behavior, they need to define
> > > set_rcvlowat() (like hv_sock).
> > Hm ok, i see. I've implemented logic when non-empty transport is required, because hyperv transport
> > forbids to set SO_RCVLOWAT, so user needs to call this setsockopt AFTER transport is assigned(to check
> > that transport allows it. Not after socket creation as You mentioned above). Otherwise there is no sense
> > in such callback - it will be never used. Also in code above - for hyperv we will have different behavior
> > depends on when set_rcvlowat is called: before or after transport assignment. Is it ok?
> sorry, i mean: for hyperv, if user sets sk_rcvlowat before transport is assigned, it sees 0 - success, but in fact
> hyperv transport forbids this option.

I see, but I think it's better to set it and not respect in hyperv (as 
we've practically done until now with all transports) than to prevent 
the setting until we assign a transport.

At most when we use hyperv anyway we get notified per byte, so we should 
just get more notifications than we expect.

Thanks,
Stefano