[openbeosnetteam] Re: PPP #2

> Ooops, it seems I've click on "send now" to quickly!?
>
> David wrote:
> > Essentially the idea goes like this...
> >
> > When the ppp module starts is basically starts waiting for messages
> > using receive_data (blocking) in a seperate thread. before doing
> > this it loads all the modules that it can use for input/output and
> > protocols.
> >
> > The userland PPP app starts up and sends a message to the control
> > thread, using known codes for the actions and with up to 100
> > bytes of data.
>
> Is this use of BeOS thread builtin port (send_data/receive_data) only
> for controling PPP?
> Or did I miss something here?
>
> In this case, why not use ioctl() instead, which is far more portable and
> don't need an extra thread, most of the time blocked on receive_data()?
>
> We can set specific opcodes for PPP in driver/net_stack_driver.h
>
> enum {
>   NET_IOCTL_BASE = 0xbe230000,
>   NET_SOCKIO_IOCTL_BASE = NET_IOCTL_BASE + 0x100,
>   NET_STACK_IOCTL_BASE = NET_IOCTL_BASE + 0x200,
>   NET_PPP_IOCTL_BASE = NET_IOCTL_BASE + 0x300,
>   NET_SERIAL_PPP_IOCTL_BASE = NET_IOCTL_BASE + 0x400,
>   NET_PPPOE_IOCTL_BASE = NET_IOCTL_BASE + 0x500
> };
>
> As the stack driver only handle NET_STACK_IOCTL_BASE range ioctl() opcodes
himself and pass all others to the core module, this one, instead of
> passing opcodes to his socket-related soo_ioctl() function, could
> try to send it to any net_module he knows, until one of them (or none)
> claims to have handling it:
>
> pseudo-code:
> int core:ioctl(opcode, ...) {
>
>   if (opcode >= NET_SOCKIO_IOCTL_BASE &&
>       opcode < NET_SOCKIO_IOCTL_BASE + 0x100)
>     return soo_ioctl(opcode, ...);
>
>   // it's not a socket ioctl(), so hands it to modules:
>   for( module = net_module_list; module; module = module->next ) {
>     if (module->info.ioctl == NULL)
>       continue;
>
>     rc = module->info.ioctl(opcode, ...);
>     if (rc >= 0)
>         return rc;
>   }
>
>   return B_ERROR;
> }
>
> Just a suggestion, as I fear seeing too many heterogenous ways to control
> each parts or the network stack, when ioctl() plain simple for this...

OK. might be easier...

>
> > ppp user app asks for a new ppp connection for serial2,
> > - ppp control thread scans a list of "match" strings for a match, gets
> > a match on "serial" and so asks the serial_ppp module (who has
registered
> > this match) to create a new device for "2".
>
> I like the way each "ppp link" register with the main PPP module.
> However, make this method even more open, by using a ppp kind specifier
and
> a ppp device:instead of "serial2", a "serial_ppp /dev/ports/serial2".
> For PPPoE, it would be "pppoe tulip0", for PPPoA it could be
> "pppoa /dev/net/speed_touch/0" for example...

Current system doesn't prohibit any of this.

> BTW, don't make assertion on serial device name, you could have
> other entries in /dev/ports/ than only serial1 and serial2.
> And USB modems could publish an usb_acm0 entry here, for example.

I haven't, but at present the serial_ppp device uses just serial ports. :)
I'm not sure if usb modems would use that or a usb_ppp module?

> > - if we get a new serial_ppp device we create a new ppp device and
> > attach the serial_ppp pointer to it
> >
> > This all works already. Next steps are harder :)  Some of the code
> > needs working on, but I think I'm making progress.
>
> This is really cool!
> You rocks, again, David.
>
> OFF-TOPIC: so, how was AOTC? ;-)

Attack of the Clones (took me a minute) - good.

david

Other related posts: