Re: [Qemu-devel] [PATCH v1 12/21] io: implement socket watch for win32 using WSAEventSelect+select

[Paolo Bonzini <pbonzini@redhat.com>]

On 09/03/2016 18:28, Daniel P. Berrange wrote:
> From: Paolo Bonzini <pbonzini@redhat.com>

Reviewing my own patch looks weird. :)

> On Win32 we cannot directly poll on socket handles. Instead we
> create a Win32 event object and associate the socket handle with
> the event. When the event signals readyness we then have to
> use select to determine which events are ready. Creating Win32
> events is moderately heavyweight, so we don't want todo it
> every time we create a GSource, so this associates a single
> event with a QIOChannel.
> 
> Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
> ---
>  include/io/channel.h |   3 ++
>  io/channel-socket.c  |   9 ++++
>  io/channel-watch.c   | 136 ++++++++++++++++++++++++++++++++++++++++++++++++++-
>  io/channel.c         |  14 ++++++
>  4 files changed, 161 insertions(+), 1 deletion(-)
> 
> diff --git a/include/io/channel.h b/include/io/channel.h
> index 0a1f1ce..20b973a 100644
> --- a/include/io/channel.h
> +++ b/include/io/channel.h
> @@ -78,6 +78,9 @@ typedef gboolean (*QIOChannelFunc)(QIOChannel *ioc,
>  struct QIOChannel {
>      Object parent;
>      unsigned int features; /* bitmask of QIOChannelFeatures */
> +#ifdef _WIN32
> +    HANDLE event; /* For use with GSource on Win23 */

Even s390 would have Win24 and Win31 but not Win23. :)

> +#endif
>  };
>  
>  /**
> diff --git a/io/channel-socket.c b/io/channel-socket.c
> index 6f7f594..ff49853 100644
> --- a/io/channel-socket.c
> +++ b/io/channel-socket.c
> @@ -55,6 +55,10 @@ qio_channel_socket_new(void)
>      ioc = QIO_CHANNEL(sioc);
>      ioc->features |= (1 << QIO_CHANNEL_FEATURE_SHUTDOWN);
>  
> +#ifdef WIN32
> +    ioc->event = CreateEvent(NULL, FALSE, FALSE, NULL);
> +#endif
> +
>      trace_qio_channel_socket_new(sioc);
>  
>      return sioc;
> @@ -341,6 +345,11 @@ qio_channel_socket_accept(QIOChannelSocket *ioc,
>      cioc->remoteAddrLen = sizeof(ioc->remoteAddr);
>      cioc->localAddrLen = sizeof(ioc->localAddr);
>  
> +#ifdef WIN32
> +    ((QIOChannel *)cioc)->event = CreateEvent(NULL, FALSE, FALSE, NULL);
> +#endif

QIO_CHANNEL(cioc)->event?

> +    WSAEventSelect(ssource->socket, NULL, 0);

This should probably be moved in qio_channel_socket_finalize.

> 
> +    /* WSAEnumNetworkEvents is edge-triggered, so we need a separate
> +     * call to select to find which events are actually available.
> +     * However, if there were no reported events at the time of the last
> +     * call, and no new events since then, we know that the socket is
> +     * quiescent.
> +     */
> +    if (!ssource->revents && !ev.lNetworkEvents) {
> +        return 0;
> +    }
> +

This is unfortunately unsafe, because:

1) WSAEventSelect clears all pending events (so the next
WSAEnumNetworkEvents returns no FD_READ, unless you have read all data
in the buffer with recv)

2) setting a socket to non-blocking should call WSAEventSelect (see
below), but cannot e.g. set ->revents to ~0 for all existing GSource.

It's probably possible to add a generation count or something like that,
but for now please revert this part (I had made it a separate commit in
my prototype because I wasn't sure if it was okay).

> +    ssource->condition = condition;
> +    ssource->socket = socket;
> +    ssource->revents = 0;
> +    ssource->fd.fd = (gintptr)ioc->event;
> +    ssource->fd.events = G_IO_IN;
> +
> +    g_source_add_poll(source, &ssource->fd);
> +    WSAEventSelect(ssource->socket, ioc->event,
> +                   FD_READ | FD_ACCEPT | FD_CLOSE |
> +                   FD_CONNECT | FD_WRITE | FD_OOB);

This should be moved where the socket is made non-blocking in
qio_channel_socket_set_blocking (because qemu_qemu_set_nonblock also
calls WSAEventSelect).

It's probably worth adding a comment that
qio_channel_socket_source_check only works in non-blocking mode for Windows.

Thanks,

Paolo