Re: [Qemu-devel] [PATCH v2 2/6] nbd/server: use errp instead of LOG

[Vladimir Sementsov-Ogievskiy]

07.07.2017 17:40, Eric Blake wrote:
> On 06/21/2017 10:34 AM, Vladimir Sementsov-Ogievskiy wrote:
> > Move to modern errp scheme from just LOGging errors.
> >
> > Signed-off-by: Vladimir Sementsov-Ogievskiy <address@hidden>
> > ---
> >   nbd/server.c | 268 +++++++++++++++++++++++++++++++++++------------------------
> >   1 file changed, 161 insertions(+), 107 deletions(-)
> > -static coroutine_fn int nbd_negotiate(NBDClient *client)
> > +static coroutine_fn int nbd_negotiate(NBDClient *client, Error **errp)
> >   {
> >       char buf[8 + 8 + 8 + 128];
> >       int ret;
> Now that you have this,
>
> >      qio_channel_set_blocking(client->ioc, false, NULL);
> you should probably pass errp through to qio_channel_set_blocking and
> return early if that fails (either here or in one of the followup
> patches) [hmm, I note that most callers of qio_channel_set_blocking()
> ignore errors - outside of NBD, so that's a separate cleanup series]

I've looked through realizations of ioc_klass->io_set_blocking - looks 
like they all are returning zero always.

> > @@ -591,21 +617,23 @@ static coroutine_fn int nbd_negotiate(NBDClient *client)
> >   
> >       if (oldStyle) {
> >           if (client->tlscreds) {
> > -            TRACE("TLS cannot be enabled with oldstyle protocol");
> > +            error_setg(errp, "TLS cannot be enabled with oldstyle protocol");
> >               return -EINVAL;
> This code is unreachable, since earlier we have:
>
> >      oldStyle = client->exp != NULL && !client->tlscreds;
> You could just as easily delete the conditional.  In fact, our code has:
>
> if (oldStyle) {
>      do stuff
> } else {
>      do stuff
> }
>
> if (oldStyle) {
>      do stuff
> } else {
>      do stuff
> }
>
> and it would be simple to merge that into a single if statement.

--
Best regards,
Vladimir