[asterisk-dev] Implementation using libpri and socketpair()

[Paulo Garcia]

Hi,

just for feedback.

I choose to create the fifo kernel module to accomplish the task since I
found more suitable to my needs. If someone wants to check this kernel
module, it is available at
http://dgvchannel.cvs.sourceforge.net/dgvchannel/unstable/fifodev/.

I'm testing it now and probably needs improvements but I think it is very
stable and solve the intermixing problem.

Thanks for all


Paulo Garcia




On 2/8/07, Christian <crich-ml at beronet.com> wrote:
>
> yeah my approach works only if you can modify the "write" as well.
>
> the only way i know which would definitely work is: you need to write a
> kernel module which creates a char device, there each write results in a
> single "copy_from_user" and you won't have intermixing of packets.  then
> you could send the packet either directly to your hardware or you use
> another character device to send it up to your process which will also
> send the exact amount of data, like:
>
> /dev/packetloopIN + /dev/packetloopOUT
>
>
> but that's probably as much work as interpreting the q.921 header ;-)
>
>
>
>
>
>
> Paulo Garcia wrote:
> > Hi Christian,
> >
> > I think I understand what you are saying but the data read comes
> > directly from libpri and I cannot realize how can I put some data
> > before the package, because I saw that into libpri there are two
> > writes to my socket file descriptor but in the read part (inside my
> > program thread) there are only one read with both packets togheter.
> >
> > After I receive the packet from libpri, I can send it directly to my
> > card and the packet arrives at other side perfectly since my hardware
> > already prepared to do it. The only problem is when I get two packets
> > joint togheter.
> >
> > I´m thinking if using another kind of interprocess communication I can
> > avoid this socketpair behaviour.
> >
> > Thanks in advance!
> >
> >
> > Paulo
> >
> >
> >
> >
> > On 2/7/07, * Christian* <crich-ml at beronet.com
> > <mailto:crich-ml at beronet.com>> wrote:
> >
> >     Hi Paulo,
> >
> >     i have had similar issues with a quite different use-case ;-) the
> >     problem is that you need to frame your messages and transmit these
> >     frames instead of the transparent data. you can use hdlc for
> >     example, i
> >     believe there are some simple implementations out there, but that is
> >     sort of overkill. The most simple thing would be to just put the
> >     length
> >     of the frame before each packet, the length could be of maximum 4
> >     bytes,
> >     then do a read of just as much bytes which you know from this
> >     length, like:
> >
> >     uint32 len;
> >     char buf[1024];
> >     read(fd, &len, 4);
> >     read(fd,buf,len);
> >
> >     then you can be sure that you've got only this single frame in buf,
> >     (well if read returns less then len you need to reread of course).
> >
> >     Besides that i have somewhere read that you can set a HDLC
> >     Property via
> >     fcntl to terminals, but i haven't had ever the time to test that
> out.
> >
> >
> >     cheers,
> >
> >     christian
> >
> >
> >     Paulo Garcia wrote:
> >     > Hi,
> >     >
> >     > in my own channel, I'm implementing ISDN support using Libpri. The
> >     > implementation is completely independent from zaptel then I'm
> >     trying
> >     > to do all communications between my hardware and libpri using
> >     > socketpair function.
> >     >
> >     > The idea is to create a socketpair, passing fd[0] to pri_new
> >     function
> >     > and writing to fd[1] to pass data to libpri. So far so good...
> >     >
> >     > To receive data from libpri, I have a thread checking for all file
> >     > descriptors fd[1] using select() function and when I receive
> >     > something, I pass it to my card. This approach also works well but
> >     > sometimes I receive two messages (from libpri) in only one read()
> >     > call. I think this is because of libpri writes two messages so
> fast
> >     > then when the read is called, there are already two messages in
> >     socket
> >     > buffer...
> >     >
> >     > How can I avoid this behavior? Maybe using another way different
> of
> >     > socketpair ? This communication will be used only inside the same
> >     > process, between different threads.
> >     >
> >     > Any path or idea will be appreciated.
> >     >
> >     >
> >     > Regards
> >     >
> >     >
> >     > Paulo Garcia
> >     >
> >
> ------------------------------------------------------------------------
> >     >
> >     > _______________________________________________
> >     > --Bandwidth and Colocation provided by Easynews.com
> >     <http://Easynews.com> --
> >     >
> >     > asterisk-dev mailing list
> >     > To UNSUBSCRIBE or update options visit:
> >     >     http://lists.digium.com/mailman/listinfo/asterisk-dev
> >     >
> >
> >     _______________________________________________
> >     --Bandwidth and Colocation provided by Easynews.com
> >     <http://Easynews.com> --
> >
> >     asterisk-dev mailing list
> >     To UNSUBSCRIBE or update options visit:
> >        http://lists.digium.com/mailman/listinfo/asterisk-dev
> >
> >
> > ------------------------------------------------------------------------
> >
> > _______________________________________________
> > --Bandwidth and Colocation provided by Easynews.com --
> >
> > asterisk-dev mailing list
> > To UNSUBSCRIBE or update options visit:
> >    http://lists.digium.com/mailman/listinfo/asterisk-dev
> >
>
> _______________________________________________
> --Bandwidth and Colocation provided by Easynews.com --
>
> asterisk-dev mailing list
> To UNSUBSCRIBE or update options visit:
>    http://lists.digium.com/mailman/listinfo/asterisk-dev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.digium.com/pipermail/asterisk-dev/attachments/20070314/b499f18f/attachment-0001.htm