On Fri, Jul 03, 2015 at 08:42:15AM +0800, Fam Zheng wrote:
> On Thu, 07/02 14:30, Stefan Hajnoczi wrote:
> > On Tue, Jun 30, 2015 at 10:29:20AM +0800, Fam Zheng wrote:
> > > It returns true as long as there is another attached port. This is not
> > > strictly necessary because even if there is only one port (the sender),
> > > net_hub_port_receive could succeed with a NOP. So always deliver the
> > > packets, instead of queuing them.
> > >
> > > This fixes the possible hanging issue after net layer changed how
> > > can_read is handled. That is, if net_hub_port_can_receive returned
> > > false, the peer would disable the queue until it's explicitly flushed
> > > (for example, a call to qemu_flush_queued_packets() in net_hub_add_port,
> > > where net_hub_port_can_receive() would become true). This patch avoids
> > > that complication.
> > >
> > > Signed-off-by: Fam Zheng <f...@redhat.com>
> > > ---
> > > net/hub.c | 20 --------------------
> > > 1 file changed, 20 deletions(-)
> >
> > Hmm...I misread the hub code:
> >
> > net_hub_port_can_receive() returns true if *any* port can receive.
> >
> > net_hub_receive_iov() always accepts packets. It never discards or
> > queues.
> >
> > So in order to move to the semantics you want, let's drop
> > net_hub_port_can_receive() *and* change net_hub_receive_iov():
> >
> > static ssize_t net_hub_receive(NetHub *hub, NetHubPort *source_port,
> > const uint8_t *buf, size_t len)
> > {
> > NetHubPort *port;
> >
> > QLIST_FOREACH(port, &hub->ports, next) {
> > if (port == source_port) {
> > continue;
> > }
> >
> > /* No need for a callback because net_hub_flush() is called
> > * when the peer flushes the queue anyway.
> > *
> > * Note that packets are duplicated if there are multiple
> > * ports and some of them accepted a packet before a later
> > * port queued it. Live with it, the network tolerates
> > * duplicates.
> > */
> > if (qemu_send_packet(&port->nc, buf, len) == 0) {
> > return 0;
>
> I'm not sure if it is a good idea to return 0 from either net_hub_port_receive
> or net_hub_receive. Because that way if one out of ten ports is down, other
> ones will no longer get more packets from tap.You are right, in order to carry over existing semantics it needs to return 0 when all ports returned 0. That way .can_receive has been eliminated but behavior is identical except for duplicate packets, which can happen on the network anyway.
pgpJXMcSt0jcp.pgp
Description: PGP signature
