Re: [PATCH v2 2/2] tools/xenstore: simplify xenstored main loop

[Julien Grall <julien@xxxxxxx>]

Hi Juergen,

On 17/05/2021 07:10, Juergen Gross wrote:
> On 14.05.21 19:05, Julien Grall wrote:
> > Hi Juergen,
> >
> > On 14/05/2021 12:56, Juergen Gross wrote:
> > > The main loop of xenstored is rather complicated due to different
> > > handling of socket and ring-page interfaces. Unify that handling by
> > > introducing interface type specific functions can_read() and
> > > can_write().
> > >
> > > Signed-off-by: Juergen Gross <jgross@xxxxxxxx>
> > > ---
> > > V2:
> > > - split off function vector introduction (Julien Grall)
> > > ---
> > >   tools/xenstore/xenstored_core.c   | 77 +++++++++++++++----------------
> > >   tools/xenstore/xenstored_core.h   |  2 +
> > >   tools/xenstore/xenstored_domain.c |  2 +
> > >   3 files changed, 41 insertions(+), 40 deletions(-)
> > >
> > > diff --git a/tools/xenstore/xenstored_core.c 
> > > b/tools/xenstore/xenstored_core.c
> > > index 856f518075..883a1a582a 100644
> > > --- a/tools/xenstore/xenstored_core.c
> > > +++ b/tools/xenstore/xenstored_core.c
> > > @@ -1659,9 +1659,34 @@ static int readfd(struct connection *conn, 
> > > void *data, unsigned int len)
> > >       return rc;
> > >   }
> > > +static bool socket_can_process(struct connection *conn, int mask)
> > > +{
> > > +    if (conn->pollfd_idx == -1)
> > > +        return false;
> > > +
> > > +    if (fds[conn->pollfd_idx].revents & ~(POLLIN | POLLOUT)) {
> > > +        talloc_free(conn);
> > > +        return false;
> > > +    }
> > > +
> > > +    return (fds[conn->pollfd_idx].revents & mask) && !conn->is_ignored;
> > > +}
> > > +
> > > +static bool socket_can_write(struct connection *conn)
> > > +{
> > > +    return socket_can_process(conn, POLLOUT);
> > > +}
> > > +
> > > +static bool socket_can_read(struct connection *conn)
> > > +{
> > > +    return socket_can_process(conn, POLLIN);
> > > +}
> > > +
> > >   const struct interface_funcs socket_funcs = {
> > >       .write = writefd,
> > >       .read = readfd,
> > > +    .can_write = socket_can_write,
> > > +    .can_read = socket_can_read,
> > >   };
> > >   static void accept_connection(int sock)
> > > @@ -2296,47 +2321,19 @@ int main(int argc, char *argv[])
> > >               if (&next->list != &connections)
> > >                   talloc_increase_ref_count(next);
> > > -            if (conn->domain) {
> > > -                if (domain_can_read(conn))
> > > -                    handle_input(conn);
> > > -                if (talloc_free(conn) == 0)
> > > -                    continue;
> > > -
> > > -                talloc_increase_ref_count(conn);
> > > -                if (domain_can_write(conn) &&
> > > -                    !list_empty(&conn->out_list))
> > AFAICT, the check "!list_empty(&conn->out_list)" can be safely removed 
> > because write_messages() will check if the list is empty (list_top() 
> > returns NULL in this case). Is that correct?
> Yes.
Thanks, how about adding in the commit message:

"Take the opportunity to remove the empty list check before calling 
write_messages() because the function is already able to cope with an 
empty list."
I can update the commit message while committing it.

Cheers,

--
Julien Grall