FreeLists / openbeosnetteam / [openbeosnetteam] Re: Ping & timers...

["David Reid" <dreid@xxxxxxxxxxxx>]

Philippe,

Here's your follow up question for 10pts...

How do we timeout a select event? If I call select() for 10seconds, who is
responsible for saying after 10secs that there has been no event...

Also, every time we call select for a socket with more than 1 "flag" set we
get multiple notifications, so this needs to be borne in mind... Hmm,
thinking...

david

----- Original Message -----
From: <philippe.houdoin@xxxxxxx>
To: <openbeosnetteam@xxxxxxxxxxxxx>
Sent: Tuesday, April 02, 2002 3:04 PM
Subject: [openbeosnetteam] Re: Ping & timers...


> > Any progress on how we use the variables passed into select yet?
>
> Oh, sorry for not answering this point more quicker.
>
> Yes, the solution is to call the BeOS kernel notify_select_event(),
> with the corresponding selectsync and ref.
> Those are the ones passed by kernel to /dev/net/socket
> net_socket_[de]select() hooks.
>
> I should have commit this fix long ago already, but can't find
> both time and where I could store the selectsync / ref for each
> select event (read, write, exception).
> The best way /me think is to add:
>
>   struct {
>     selectsync * sync;
>     uint32       ref;
>   } selectinfo[3];
>
> in struct socket definition.
> But this imply that socket structure will be no more an opaque thing
> for the socket driver, which is bad (tm).
>
> Anyway, in driver/socket.c:
>
> static status_t net_socket_select
>   (
>   void *       cookie,
>   uint8        event,
>   uint32       ref,
>   selectsync * sync
>   )
> {
>   struct socket * so = (struct socket *) cookie;
>
>   so->selectinfo[event].sync = sync;
>   so->selectinfo[event].ref  = ref;
>   return B_OK;
> }
>
>
> static status_t net_socket_deselect
>   (
>   void *       cookie,
>   uint8        event,
>   selectsync * sync
>   )
> {
>   struct socket * so = (struct socket *) cookie;
>
>   so->selectinfo[event].sync = NULL;
>   so->selectinfo[event].ref  = 0;
>   return B_OK;
> }
>
> And, last but not least, *somewhere*, you will need to
> call notify_select_event() when a select event raise on the socket:
>
> // these are missing from KernelExport.h ...
> #define  B_SELECT_READ       1
> #define  B_SELECT_WRITE      2
> #define  B_SELECT_EXCEPTION  3
> extern void notify_select_event(selectsync *sync, uint32 ref);
>
> if (so->selectinfo[event].sync)
>   notify_select_event(so->selectinfo[event].sync,
>                       so->selectinfo[event].ref);
>
> Problem is "where"?
>
> I must confess I'm lost in the current higher level of socket layer.
> Should the notify take place in sowakeup()?
> Who wait on the read/write sockbuf sb_pop semaphore?
> Is there is already some "exception" handling somewhere, where
> a B_SELECT_EXCEPTION should be notified too?
>
> -Philippe, lost in no_more_than_six_letters_struct_fields_names land ;-)
>
>