Re: [libvirt] [PATCH 03/16] Add callback to virNetClient to be invoked on EOF

On Thu, Jul 19, 2012 at 04:18:33PM +0800, Daniel Veillard wrote: > > @@ -1636,17 +1657,21 @@ void virNetClientIncomingEvent(virNetSocketPtr sock, > > VIR_DEBUG("%s : VIR_EVENT_HANDLE_HANGUP or " > > "VIR_EVENT_HANDLE_ERROR encountered", __FUNCTION__); > > virNetSocketRemoveIOCallback(sock); > > - goto done; > > + goto eof; > > } > > > > if (events & VIR_EVENT_HANDLE_WRITABLE) { > > - if (virNetClientIOHandleOutput(client) < 0) > > + if (virNetClientIOHandleOutput(client) < 0) { > > virNetSocketRemoveIOCallback(sock); > > + goto eof; > > + } > > } > > > > if (events & VIR_EVENT_HANDLE_READABLE) { > > - if (virNetClientIOHandleInput(client) < 0) > > + if (virNetClientIOHandleInput(client) < 0) { > > virNetSocketRemoveIOCallback(sock); > > + goto eof; > > + } > > } > > > What about the case ? > if (client->haveTheBuck || client->wantClose) ? > sure it's induced locally instead of getting raised by an eof If either of those conditions are true, the virNetClientIncomingEvent method should not be invoked - we just have a check for those at the top as a sanity check. In the haveTheBuck==true case, this callback will be activated again once the buck is released, so we'll deal with EOF callbacks correctly. In the wantClose==true case, we might miss out invoking the EOF callback - I suspect this is only a problem if the app code explicitly asked for a close itself, or if the keepalive code triggers. I'll check into this and if there are any fixes required, I'll do them as a followup patch.